78559 THE WORLD BANK BANGLADESH Development Series Bangladesh Poverty Assessment: Assessing a Decade of Progress in Reducing Poverty, 2000-2010 Bangladesh Development Series Paper No. 31 The World Bank Office, Dhaka June 2013 A joint report of the Human Development & Poverty Reduction and Economic Management Sectors South Asia Region, World Bank www.worldbank.org.bd/bds Document of the World Bank The World Bank World Bank Office Dhaka Plot – E-32, Agargaon, Sher-e-Bangla Nagar, Dhaka- 1207, Bangladesh Tel: 880-2-8159001-28 Fax: 880-2-8159029-30 www.worldbank.org.bd World Bank 1818 H Street, N.W. Washington DC 20433, USA Tel: 1-202-4731000 Fax: 1-202-477-66391 www.worldbank.org All Bangladesh Development Series (BDS) publications are downloadable at: www.worldbank.org.bd/bds Standard Disclaimer: All rights reserved. The findings, interpretation, and conclusions expressed herein are those of the authors and do not necessarily reflect the views of the Board of Executive Directors of the World Bank or the governments they represent. Copying and/or formatting portions or all of this work without prior permission may be in violation of applicable law. The World Bank encourages dissemination of its work and will normally grant permission promptly. Any queries in this regard should be addressed to the Office of the Publisher, World Bank, 1818 H. Street NW, Washington, DC 20433, email: pubrights@worldbank.org. Copyright Statement: The material in this publication is copyrighted. The World Bank encourages dissemination of this work and will normally grant permission to reproduce portion of the work promptly. Design Cover designed and published by Progressive Printers. Table of Contents Executive Summary: Bangladesh Poverty Assessment ......................................................................... xii Introduction ................................................................................................................................................. 1 Part I: Poverty Patterns and the Living Conditions of the Poor .............................................. 5 1. Poverty, Growth and Inequality ........................................................................................................ 6 2. Profiling the Poor: Characteristics of the Poor and Determinants of Poverty ................................ 22 Part II: Understanding the Drivers of Poverty Reduction over the 2000-2010 Decade ....... 43 3. Decomposing Poverty: Understanding key drivers of poverty reduction ....................................... 44 4. Labor Income: Linking the Labor Market and Poverty Reduction ................................................. 57 5. Demographic Transition and Poverty: Fertility, Mortality, and Living Arrangements .................. 74 Part III: Seasonal Poverty, Food Commodity Price Shocks, and the Role of Safety Nets and Microfinance................................................................................................................................ 81 6. Understanding Seasonal Extreme Poverty in the Northwestern Region of Bangladesh ................. 82 7. The Impact of Food Price Shocks on Wages, Welfare, and Policy Responses............................... 94 8. Safety Nets and Vulnerabilities..................................................................................................... 108 9. Microfinance Expansion and Poverty Reduction .......................................................................... 121 Part IV: Revisiting the East-West Divide ............................................................................... 129 10. Changing Poverty Patterns Across Regions.................................................................................. 130 Supplemental Content .............................................................................................................. 141 iii List of Figures Figure 1-1: Poverty Trends ........................................................................................................................... 8 Figure 1-2: Distribution of Per-capita Real Expenditure by Survey Year .................................................. 10 Figure 1-3: Poverty Headcount Rate under Different Adult Equivalence Assumptions ............................ 10 Figure 1-4: Growth Incidence Curve .......................................................................................................... 11 Figure 1-5: Growth and Redistribution Components of Changes in Poverty ............................................. 12 Figure 1-6: Lorenz Curve and Gini Coefficient .......................................................................................... 13 Figure 1-7: Relative and Absolute Differences in Real per-capita Consumption ....................................... 13 Figure 1-8: Datt and Ravallion (1992) Growth Decomposition Method .................................................... 15 Figure 1-9: Annual Growth of Real Gross Domestic Product, Population ................................................. 18 Figure 1-10: GDP Sectoral Decomposition ................................................................................................ 18 Figure 1-11: Demographic Changes and the Labor Force .......................................................................... 19 Figure 1-12: South Asia Trends .................................................................................................................. 20 Figure 2-1: Improvements in Households’ Assets Ownership: 2000-2010 ................................................ 23 Figure 2-2: Percentage Change in Poverty Headcount (2005-2010) by Land Ownership.......................... 25 Figure 2-3: Gross and Net Enrollment Rates .............................................................................................. 27 Figure 2-4: Percent of Households Consuming a Particular Food Group in 2005 and 2010 ...................... 29 Figure 2-5: Fraction of Households Consuming Food Items from a Specific Group ................................. 29 Figure 2-6: Average Household Dietary Diversity Score by Consumption Deciles................................... 30 Figure 2-7: Frequency Distribution of HDDS in 2005 and 2010................................................................ 30 Figure 2-8: Cumulative Distribution of HDDS in 2005 and 2010 .............................................................. 30 Figure 2-9: Stochastic Dominance of HDDS by Poverty Status in 2005.................................................... 31 Figure 2-10: Stochastic Dominance of HDDS by Poverty Status in 2010.................................................. 31 Figure 2-11: Cumulative Frequency Distribution - Dietary Diversity Secured .......................................... 31 Figure 2-12: Cumulative Frequency Distribution - Dietary Diversity Secured .......................................... 31 Figure 2-13: Decline in Poverty Rates between 2005 and 2010 by PSUs .................................................. 32 Figure 2-14: Annual Changes in DDS between 2005 and 2010 by PSUs .................................................. 32 Figure 2-15: Prevalence of Under-weight (0-59 months) weight (0-59 months) ....................................... 33 Figure 2-16: Medical Expenditures, by category ........................................................................................ 39 Figure 2-17: Sources of Payments for Medical Expenditures .................................................................... 40 Figure 2-18: Distribution of Poor Workers (share of poor workers) .......................................................... 40 Figure 2-19: Poverty Headcount by Type of Worker, 2000-2010 (percent poor) ...................................... 41 Figure 3-1: Demographic Changes ............................................................................................................. 45 Figure 3-2: Demographic Characteristics ................................................................................................... 45 Figure 3-3: Changes in the Structure of Working Age Population ............................................................. 46 Figure 3-4: Changes in the Structure of Employed Population .................................................................. 47 Figure 3-5: Non-Labor Income Growth ...................................................................................................... 47 Figure 3-6: Change in Household Consumption-to-income ratio ............................................................... 48 Figure 3-7: National and Domestic Savings (% of GDP at Current Market Prices) ................................... 48 Figure 3-8: Decomposing Consumption Per-Capita ................................................................................... 49 Figure 3-9: Contribution to Poverty Reduction 2000-2010 ........................................................................ 49 Figure 3-10: Cumulative Contributions to Poverty Reduction, 2000 – 2010 ............................................. 53 Figure 4-1: Sectoral Employment ............................................................................................................... 58 Figure 4-2: Decomposition of GDP per-capita and Output per Worker, 2000 - 2010 ................................ 59 Figure 4-3: Demographic Changes ............................................................................................................. 60 Figure 4-4: Labor Force Participation (percent of working age population) .............................................. 61 Figure 4-5: Probability of Female Labor Force Participation ..................................................................... 61 Figure 4-6: Unemployed and Underemployed Population (share of working age population) .................. 62 Figure 4-7: Employment by Occupational Type (percent of total employed) ............................................ 63 Figure 4-8: Education Level of Working Age Population Currently not in School ................................... 63 iv Figure 4-9: Educational Attainment Across the Distribution ..................................................................... 64 Figure 4-10: Average Income Growth by Education Level ........................................................................ 64 Figure 4-11: Business Constraints .............................................................................................................. 65 Figure 4-12: Household Non-Agricultural Enterprises ............................................................................... 67 Figure 4-13: Labor Income Growth Incidence Curves ............................................................................... 68 Figure 4-14: Rural versus Urban Real Income Growth .............................................................................. 69 Figure 4-15: Decomposing Differences in Wages ...................................................................................... 70 Figure 5-1: Births per Woman Across Time and by Division .................................................................... 74 Figure 5-2: Bangladesh: Population Pyramids............................................................................................ 75 Figure 5-3: Change in Population by Age Groups and Divisions, 2000-2010 ........................................... 76 Figure 5-4: Prevalence of Contraceptive Use over Time ............................................................................ 76 Figure 5-5: Change in Household Type by Division, 2000-2010 ............................................................... 78 Figure 5-6: Change in the Share of Elderly Living Arrangements, 2000-2010 .......................................... 79 Figure 6-1: Income and Expenditure by Season over Time ........................................................................ 84 Figure 6-2: Trends of Per-capita Household Expenditure in the Monga areas 2008/09 ............................. 84 Figure 6-3: Calorie Intake per-capita per-day by Consumption Quintiles .................................................. 84 Figure 6-4: Hidden Instability Behind Employment Ratios in Monga Areas ............................................. 87 Figure 6-5: Employment Categories Among the Richest and the Poorest 20 Percent (%) – Male Only.... 88 Figure 6-6: Change in Income Composition over Time by Region, 2000–10 ............................................ 90 Figure 6-7: Loans and Credit ...................................................................................................................... 91 Figure 6-8: Percentage of Households that have Access to Social Safety Nets (Monga Region) .............. 92 Figure 7-1: World Bank Food Price Index.................................................................................................. 94 Figure 7-2: Trends in Production and Consumption of Rice in Bangladesh .............................................. 95 Figure 7-3: Cereal Consumption in Bangladesh ........................................................................................ 95 Figure 7-4: Exposure to Food Price Shock ................................................................................................. 96 Figure 7-5: Real Rural and Urban Wages in Bangladesh ........................................................................... 99 Figure 7-6: Real Wages ............................................................................................................................ 100 Figure 7-7: Real Rural Wages Using Alternative Cost of Living Indexes ................................................ 101 Figure 8-1: Social Protection Expenditure as a Percentage of GDP, 1998-2013 ...................................... 110 Figure 8-2: Component-Wise Allocations of Social Protection Budget, 2012-13 .................................... 111 Figure 8-3: Proportion of households with access to safety nets .............................................................. 112 Figure 8-4: SN Coverage and Poverty by Division (2010) ....................................................................... 112 Figure 8-5: Distribution of Beneficiaries Across Consumption Quintiles ................................................ 116 Figure 8-6: Coping Mechanisms for Idiosyncratic Shocks ....................................................................... 117 Figure 8-7: Coping Mechanisms for Covariate Shocks ............................................................................ 117 Figure 9-1: Trend of Microfinance Members in Bangladesh.................................................................... 122 Figure 9-2: Trend in Loans Disbursed by the MFIs in Bangladesh ......................................................... 123 Figure 9-3: Trend of Savings Mobilized by the MFIs in Bangladesh....................................................... 123 Figure 10-1: Integrated (IR) vs. Less Integrated (LIR) Regions in Bangladesh ....................................... 130 Figure 10-2: Poverty Headcount Across Regions ..................................................................................... 130 Figure 10-3: Contributions to Poverty Reduction ..................................................................................... 131 Figure 10-4: Log of Per-capita Real Expenditure by Region ................................................................... 132 Figure 10-5: Log of Daily Earnings ......................................................................................................... 133 Figure 10-6: Hours Worked Per Year and by Region ............................................................................... 133 Figure 10-7: Change in Population by Age Groups and Divisions ........................................................... 134 Figure 10-8: Change in the Share of Elderly Living Arrangements ......................................................... 135 Figure 10-9: Children’s Access to Health Professionals.......................................................................... 136 Figure 10-10: Children’s Access to Education (6-10 years old) ............................................................... 137 Figure 10-11: Children’s Access to Electricity (11-15 years old) ........................................................... 138 Figure 10-12: Children’s Access to Sanitation (0-5 years old) ................................................................. 139 v List of Tables Table 1.1: Poverty Headcount Rates ............................................................................................................. 7 Table 1.2: Percentage Change in Poverty Headcount Rates ......................................................................... 7 Table 1.3: Depth and Severity of Poverty ..................................................................................................... 8 Table 1.4: Mean Real Per-capita Monthly Consumption............................................................................ 11 Table 1.5: Growth Elasticity Estimates – Datt and Ravallion (1992) Method ........................................... 16 Table 1.6: Growth Elasticity ....................................................................................................................... 17 Table 1.7: Predicted versus Actual Poverty Estimates for 2010 ................................................................. 17 Table 1.8: Poverty Headcount Projections .................................................................................................. 17 Table 2.1: Trends in Basic Assets and Amenities ....................................................................................... 22 Table 2.2: Trends of Poverty by Land Ownership in Rural Areas.............................................................. 25 Table 2.3: Demographic Characteristics of Households ............................................................................. 26 Table 2.4: Education of Household Head and Poverty ............................................................................... 27 Table 2.5: Years of Completed Education .................................................................................................. 28 Table 2.6: Percentage of Population with Moderate and Severe Deficiency in Calorie Intake .................. 33 Table 2.7: Selected Health Indicators for Adults and Children: years 2005 and 2010 ............................... 34 Table 2.8: Selected Health Indicators for Urban and Rural: years 2005 and 2010 ..................................... 36 Table 2.9: Changes in Selected Health Indicators ...................................................................................... 37 Table 2.10: Child Immunization Rates - 2005 and 2010 ............................................................................ 38 Table 4.1: Overview of the Labor Market .................................................................................................. 58 Table 5.1: Infant Mortality Rate over Time ................................................................................................ 75 Table 5.2: Distribution of Households, by Type and Mean Household Size .............................................. 78 Table 5.3: Living Arrangements of Persons Aged 55 and Older ................................................................ 79 Table 6.1: Comparing Rangpur to the Rest of the Country ........................................................................ 82 Table 6.2: Trends of Poverty by Land Ownership in Rangpur ................................................................... 85 Table 6.3: Trends in Basic Assets and Amenities, Rangpur Division ........................................................ 85 Table 6.4: Measures of Employment Diversity within a Household .......................................................... 88 Table 6.5: Temporary Migration: ≤ 1 month over 1-year survey period (%) (age ≥ 12) ............................ 89 Table 6.6: Received Remittance (%) .......................................................................................................... 90 Table 7.1: Welfare Impact of a Rice Price Increase by Poverty ................................................................. 97 Table 7.2: Public Distribution System of Food Grains (PFDS) (‘000 MT) .............................................. 105 Table 8.1: Expenditures on Largest Safety Net Programs, 2008 – 2012 (BDT crore, real values) .......... 111 Table 8.2: Access to SNs by Per-capita Expenditure Quintile 2005, 2010.............................................. 113 Table 8.3: Transfer Amounts of Major SN Programs (monthly) .............................................................. 113 Table 8.4: Performance of SN Transfers .................................................................................................. 114 Table 8.5: Leakage of In-kind and Cash Transfers (monthly) .................................................................. 114 Table 9.1: Moderate and Extreme Poverty Headcounts in Bangladesh, CBN Method ............................ 121 Table 9.2: Population with Moderate and Severe Deficiency in Calorie Intake (%) ................................ 122 Table 9.3: Growth of Microfinance Clientele in Bangladesh (%) ............................................................ 123 Table 9.4: Household Distribution by Microcredit Program Participation: 1991-2011 ........................... 124 Table 9.5: Microcredit Program Participation Rate Among Households: 1991-2011 .............................. 125 Table 9.6: Household Average Cumulative Borrowing from Microcredit Programs over Time ............. 125 Table 9.7: Income, Expenditure, and Poverty by Microcredit Participation Status .................................. 126 Table 9.8: Income, Expenditure and Poverty by Level of Microcredit Participation ............................... 127 vi List of Boxes Box 1-1: Bangladesh Poverty Measurement ................................................................................................. 6 Box 1-2: Poverty Measures: Poverty Headcount, Depth, and Severity ........................................................ 9 Box 1-3: Datt and Ravallion (1992) Growth Decomposition Method ........................................................ 14 Box 2-1: Bold Ideas, Changing Lives ......................................................................................................... 24 Box 3-1: Decomposing Poverty Based on Bourguignon et al. (2008) ........................................................ 52 Box 4-1: Rising Real Wages: Is Bangladesh Reaching the Lewis Turning Point? ..................................... 71 Box 6-1: Rangpur – Distinct Features of a Lagging Region ....................................................................... 82 Box 6-2: Why Seasonality Should Matter in Poverty Assessment ............................................................. 83 Box 6-3: The Monga Phenomenon: Evidence from the InM Survey ......................................................... 86 Box 6-4: Small Incentives, Large Improvements ....................................................................................... 93 Box 7-1: Safety Net Programs Linked with the PFDS ............................................................................. 103 Box 8-1: Issues with Data Comparability Between 2005 and 2010 ......................................................... 108 Box 8-2: Characteristics of Key Safety Net Programs.............................................................................. 109 Box 8-3: Employment Generation Program for the Poorest: .................................................................... 115 vii Foreword Throughout the 2000-2010 decade, Bangladesh experienced steady and strong GDP growth of nearly 6 percent per year on average. During this period, poverty rates also demonstrated an impressive steady improvement, falling by about 1.7 percentage points per year. Using data from the Household Income and Expenditure Survey (HIES), a joint product of the Bangladesh Bureau of Statistics and the World Bank, this report shows that, while 49 percent of Bangladeshis were poor in 2000, this percentage had dropped to 31.5 by 2010. In addition, Bangladesh also achieved better health outcomes, improved living conditions for the poor, lower childhood mortality, increased under-five vaccination rates for all children, increased literacy rates, and improved safety net coverage. These are impressive achievements! The analysis of the 2000-2010 period presented in this report finds that poverty reduction was closely linked to growth in labor income and demographic changes. It also shows that the sources of income growth varied significantly between the first and the second halves of the 2000-2010 decade. Differential rates of poverty reduction along an “East-West Divide” characterized the first half (World Bank 2008a), but regional differences were significantly reduced in the second half. Nevertheless, while overall improvement in wellbeing was strong across all regions, poverty continues to be a substantial and stubborn problem in Bangladesh, where about 47 million people still live in poverty and 26 million people in extreme poverty. Moreover, poverty in rural areas continues to be relatively more pervasive and extreme than in urban areas, whereas urban areas remain relatively more unequal. Overall, the report’s findings suggest that sustained poverty reduction moving forward will require coordinated multi-sectoral action. To maintain steady growth in jobs, it will be necessary to promote investments to raise agricultural productivity and also to promote growth in the demand for salaried work in manufacturing and services (e.g., through improving infrastructure, supporting entrepreneurship by promoting the rule of law, and strengthening the transparency and accountability of both the public and private sectors). With regard to demographic changes, a “demographic opportunity” will be realized only if there are policies in place to harness it. Thus, Bangladesh will need both to cater more aggressively to the needs of the growing youth cohorts in the coming years and to begin preparations for aging-out of the demographic dividend. Lastly, Bangladesh can make better use of social safety net coverage and human capital formation to sustain poverty reduction. There remains much potential for improving the linkage between safety nets and poverty reduction through improved design, targeting and timing of safety net responses. In addition, education remains a key pathway to breaking free from the intergenerational transmission of poverty, and there is similarly substantial scope for improving the quality of, and expanding access to, schooling. A continuation of the policies and programs that have proven successful, combined with improving safety nets and placing more emphasis on skills development and child nutrition, could prove a powerful formula for increasing human capital and further poverty reduction in the future. Ernesto May Jesko Hentschel Johannes Zutt Sector Director Sector Director Country Director Poverty Reduction and Human Development The World Bank Office, Dhaka Economic Management South Asia Region Bangladesh South Asia Region viii Acknowledgements The core team preparing this report consisted of Lea Gimenez (World Bank, SASEP), Faizuddin Ahmed (World Bank, SASEP), Iffath Sharif (World Bank, SASSP) and Dean Jolliffe (World Bank, DECPI). Lea Gimenez played a key role in all aspects of preparing this report, including the writing, analysis, and coordinating role of synthesizing the background papers. Faizuddin Ahmed provided core support and guidance on the appropriate use of the Bangladesh Household Income and Expenditure Survey, and served as an advisor to authors of the background papers ensuring appropriate use of the data. Iffath Sharif and Dean Jolliffe, co-Team Leaders of the report, are responsible for the overall content and direction of the report, the validity of the measures and estimates provided in the report, the appropriateness of the methodological approaches and the consistency of these approaches across the numerous sections of this report. A key element of this report, which guides the structure of the report, is the analysis that decomposes the reductions in poverty into several components and assigns relative magnitudes of importance to each. This work was led by Gabriela Inchauste (World Bank, PRMPR) and Sergio Olivieri (World Bank, PRMPR). They worked closely with the core team to ensure that the underlying assumptions in the construction of the variables made sense in the context of Bangladesh, and the findings from this work helped the team to focus on those factors which played the largest role in reducing poverty. The work from the poverty decompositions helped shape the call for analytical work on specific topics, many of which resulted in background papers to be released as a separate volume. Work on these background papers, as well as some core work for this report, has been generously supported the UK Department of International Development (DfID) through the Joint Technical Assistance Programme (JOTAP). This support has been critical for the successful completion of the analysis that underpins the findings in this report. The background papers and analysis reflects the work of more than twenty analysts, including World Bank staff, academics, consultants, and researchers from the International Food Policy Research Institute (IFPRI). The core team worked with each of the teams to ensure that the constructed analytical data files were consistent across teams and that the messages from one team were consistent with findings from another. The background paper linking labor and poverty reductions was produced by Gabriela Inchauste, Sergio Olivieri and Tonmoy Islam (University of Wisconsin). The background analysis on Bangladesh’s demographic transition was led by Maitreyi Das (World Bank, SDV) with support from Ieva Zumbyte (Consultant). The section on seasonal extreme poverty benefited from background papers written by Shahidur Khandker (World Bank, DECAR) and Hussain Samad (World Bank, SASDE) & Nobuo Yoshida (World Bank, PRMPR) and Gaurav Khanna (University of Michigan). The chapter on food prices and poverty draws from two papers -- one background paper written by Hanan Jacoby (World Bank, DECAR) and Basab Dasgupta (World Bank, UDRUR) and the other paper was written by a team from IFPRI consisting of Xiaobo Zhang, Shahidur Rashid, Kaikaus Ahmad, Valerie Muller, Hak Lim Lee, Solomon Lemma, Saika Belal, and Akhter Ahmed. The chapter on safety nets and vulnerabilities was supported by analysis from Maitreyi Das and Celine Ferre (Consultant). The background material on microfinance and poverty reduction was written by Shahidur Khandker (World Bank, DECAR) and Gayatri Koolwal (World Bank, DECAR). Background papers by Wahid Abdallah (BRAC University) on the inequality of opportunity and another by Forhad Shilpi (World Bank, DECAR) with support from Ieva Zumbyte on spatial differences in poverty support the chapter on the East-West divide. The profile of poverty benefited from a background paper on dietary diversity by Atonu Rabbani (University of Dhaka), education statistics from Syed Rashed Al-Zayed (World Bank, SASED) and Mrittika Shamsuddin (Georgetown University), ix and material on remittances and migration from Sonia Plaza (World Bank, PRMED). Hossain Zillur Rahman (Power and Participation Research Center - PPRC) helped with the introduction. Throughout the development of this report, the team received feedback from several sources. In particular, the team received guidance from Wahiduddin Mahmud (Dhaka University), Mustafa Mujeri (Bangladesh Institute for Development Studies - BIDS), Binayak Sen (BIDS), Mahabub Hossain (Bangladesh Rural Advancement Committee). Hossain Zillur Rahman, Riti Ibrahim (Bangladesh Bureau of Statistics – BBS), Mustafizur Rahman (Center for Policy Dialogue - CPD), Debapriya Bhattacharya (CPD), Baqui Khalily (PKSF), Quazi Mesbahuddin Ahmed (PKSF), Fazlul Kader (PKSF), Akhter Ahmed (IFPRI) Arifur Rahman (DfID), Iftekhar Hossain (DfID), Ken Simler (World Bank, ECSP3), Ambar Narayan (World Bank, PRMPR) Zahid Hussain (World Bank, SASEP), Manohar Sharma (World Bank, EASPW), and Nobuo Yoshida. The team is also grateful for feedback from participants from summer 2012 workshops held at the World Bank, including participants from DfID, World Food Programme, Institute of Microfinance, University of Ulster (UK), Australian Agency for International Development, BIDS, PKSF, and PRCC. We would also like to extend our thanks to Shamsul Alam, Planning Commission, Government of Bangladesh for providing feedback on preliminary findings and for arranging to have staff from his unit participate in a workshop held at the World Bank offices. We benefitted immensely from all of the comments and guidance received. Mehar Akhtar Khan (World Bank, SASEP) and Rita Soni (World Bank, SASEP) provided support in assembling our team of consultants and in arranging meetings and workshops to discuss findings with key stakeholders and Government counterparts. The report also greatly benefited from the detailed editing work provided by Rhea Bhatta (Lehigh University). Finally, the team gratefully acknowledges guidance from Johannes Zutt (World Bank, Bangladesh Country Director), Christine Kimes (World Bank, SACBD), Andras Horvai (World Bank, SACBA), Ernesto May (World Bank, SASPM), Vinaya Swaroop (World Bank, SASEP), Pablo Gottret (World Bank, SASSP), Jesko S. Hentschel (World Bank, SASHD), Sanjay Kathuria (World Bank, SASEP), and Salman Zaidi (World Bank, SASEP). x GOVERNMENT FISCAL YEAR July 1 – June 30 CURRENT EQUIVALENTS Currency Unit = Bangladeshi Taka (Tk) US$1 = Tk 78.01 (April 2013) LIST OF ABBREVIATIONS AE Adult Equivalence ASA Association of Social Advancement BBS Bangladesh Bureau of Statistics BES Bangladesh Enterprise Survey BIDS Bangladesh Institute of BNPI Basic Need Price Index Development Studies BRDB Bangladesh Rural Development CBN Cost of Basic Needs CCT Conditional Cash Transfer CDF Cumulative Density Function CDF Credit and Development Forum CFW Cash-For-Work CPI Consumer Price Index DAM Department of Agricultural Marketing DDS Dietary Diversity Score DHS Demographic Health Survey EGPP Employment Generation Program for the Poorest FAO Food and Agriculture FCPI Food Consumer Price Index FDI Foreign Direct Investment FFE Food for Education FFW Food For Work FGT Foster Greer Thorbecke FPC Fair Price Card FPI Food Price Index GB Grameen Bank GCPI General Consumer Price Index GDP Gross Domestic Product GoB Government of Bangladesh GR Gratuitous Relief HIES Household Income and Expenditure Surveys HOI Human Opportunity Index InM Institute of Microfinance IR Integrated Regions LGRD Ministry of Local Government and Rural Development LIR Less Integrated Regions LFS Labor Force Survey MDG Millennium Development Goal MFI Micro-Finance Institutions MoF Ministry of Finance NGO Non-Governmental Organization NIPORT National Institute of Population Research OECD Organization for Economic Cooperation and and Training Development OMS Open Market Sales PDBF Palli Daridra Bimochan Foundation PFDS Public Food Distribution System PKSF Palli Karma Shahayak Foundation PMT Proxy Means Tests SME Small Micro-Enterprise SN Social Safety Net SNA System of National Account TFR Total Fertility Rate Tk Taka – Local Currency in Bangladesh TR Test Relief UN United Nations USAID United States Agency for International USDA United States Department of Agriculture Development VGD Vulnerable Group Development VGF Vulnerable Group Feeding WDI World Development Index WFP World Food Program Vice President: Isabel M. Guerrero, SARVP Country Director: Johannes Zutt, SACBD Sector Director: Ernesto May, SASPM & Jesko Hentschel, SASHD Sector Manager: Vinaya Swaroop, SASEP & Pablo Gottret, SASSP Task Team Leaders: Dean Jolliffe, DECPI & Iffath Sharif, SASSP xi Executive Summary: Bangladesh Poverty Assessment Poverty Trends and the Characteristics of the Poor 1. Bangladesh experienced a uniform and steady decline in consumption-based poverty rates between 2000 and 2010. Poverty rates demonstrated an impressive and steady improvement during this period with an average decline of 1.74 percentage points per year. Poverty declined 1.78 percentage points per year between 2000 and 2005, and the analogous decline for the 2005-2010 period was 1.70 percentage points. This suggests that the series of shocks that affected Bangladesh in 2007/2008 did not significantly slow down the speed of poverty reduction.1 2. Over the decade, there was a persistent decline in the number of poor people—from nearly 63 million in 2000, to 55 million in 2005, and then 47 million in 2010. Despite population growth, the population of poor households declined by 26 percent in 10 years. The number of extreme poor people also declined from 44 million in 2000 to 34.6 million in 2005, and down to 26 million in 2010 – a massive 41 percent decrease. The clear narrative that emerges is that the steady decline in poverty rates has been sufficient to ensure substantial reductions over the decade in the number of people living in poverty and extreme poverty. 3. Bangladesh is on track for reaching both of the poverty Millennium Development Goals (MDG). The depth of poverty was nearly halved over the 2000-2010 period, allowing Bangladesh to attain the depth-of-poverty MDG target of 8 percent at least five years in advance of 2015. Similarly, the poverty projections based on the last three national HIES surveys suggest that Bangladesh will achieve the first MDG goal of halving the poverty headcount to 28.5 percent sometime in 2013. 4. While the rate of poverty reduction was comparable across urban and rural areas, extreme poverty continues to be a rural phenomenon. While poverty rates at the national-level fell by 36 percent over the decade, rural areas in 2010 had only attained the decade old poverty level of urban areas. The poverty rate in rural areas in 2010 was 35.2 percent, precisely the same rate for urban areas in 2000 (Table 1). With a 39.5 percent reduction in poverty, urban areas have achieved remarkable progress,2 but extreme poverty continues as a significant rural phenomenon. In 2010, 21.1 percent of the rural population was extremely poor, compared to 37.9 percent in 2000 (Table 1). This represents a substantially smaller decline in extreme poverty (44 percent) in rural areas than in urban areas (61 percent). As further evidence of the difference, in 2010, the large majority (60 percent) of the poor in rural areas were also extremely poor compared to 36 percent in urban areas. Table 1: Poverty Headcount Rates Poverty Extreme Poverty 2000 2005 2010 2000 2005 2010 National 48.9 40.0 31.5 34.3 25.1 17.6 Urban 35.2 28.4 21.3 19.9 14.6 7.7 Rural 52.3 43.8 35.2 37.9 28.6 21.1 Source: All estimates are CBN based on HIES 2005, updated for 2010, and back-casted for 2000. 2010 update: survey-based food prices, and nonfood allowance re-estimated using ‘upper’ poverty lines. Official Poverty Lines estimated for HIES (2000, 2005, and 2010). 1 Chapter 7 of this report investigates the welfare effects of the 2008 food price shock. 2 For more details on the urbanization process that has taken place over the decade, refer to the World Bank 2012a. xii Figure 1: Poverty Headcount Across Regions 60.0% 56.7% 53.1% 52.0% 51.2% 50.0% 45.7% 46.7% 45.1% 45.7% 42.4% 39.4% 40.0% 34.0% 35.6% 33.8% 32.0% 30.5% 32.1% 26.2% 28.1% 2000 30.0% 2005 20.0% 2010 10.0% 0.0% Barisal Chittagong Dhaka Khulna Rajshahi Sylhet Source: HIES 2000, 2005, and 2010. 5. Poverty patterns have dramatically changed across regions between 2000 and 2010. Poverty estimates in 2005 highlighted the need for creating economic opportunities for narrowing the development gap between the East and the West of Bangladesh. While the East was rapidly improving and benefiting from its geographical proximity to growth poles, the Western region of Bangladesh had been lagging behind. In particular, the poverty headcount figures from the years 2000 and 2005 revealed that while there were large declines in poverty in all of the Eastern most divisions (Chittagong, Dhaka and Sylhet), their Western counterparts (Barisal, Khulna and Rajshahi) had remained practically stagnant (Figure 1). 6. While the first half of the decade was characterized by uneven poverty reduction, the second half provided the lagging regions with an opportunity to catch up. The 2010 poverty estimates describe a changed Bangladesh. Comparing the new poverty headcount figures from 2010 to those from 2005 reveals a reversal of poverty patterns. Not only did the Western divisions (Barisal, Khulna and Rajshahi) experience larger reductions in poverty (as measured by the poverty headcount ratio) they also managed to reach levels of poverty that are closer to those of their Eastern counterparts (Chittagong, Dhaka and Sylhet). Dhaka, the division with the lowest poverty rate in 2005, did not experience much change in its poverty headcount. An important exception to this pattern is the Northern part of the country, specifically Rangpur, which has over 42 percent of its population living below the upper poverty line. 7. This remarkable reduction in poverty was coupled with an equally impressive improvement in the living conditions of the poor. During the first part of the decade all households saw large improvements in the quality of material used in the construction of their homes and in their access to basic services. Between 2000 and 2005, a large number of households saw improvements in terms of the materials used in the constructions of their homes (that is, more households were in homes with walls and roofs made of corrugated iron, steel, and cement) as well as in their access to services (that is, more households in homes with sanitary latrines and electricity). Between 2005 and 2010, while the poor continued to improve the quality of their homes, the largest improvements for all households were in terms of the amenities households own, such as television sets and cellular phones. 8. Significant changes took place in the 2000-2010 period in terms of the relationship between land-ownership and poverty reduction. Between 2000 and 2005, reductions in the rate of poverty were greater among households with relatively larger landholdings (that is, those having more than 1.5 acres of land), but then between 2005 and 2010, poverty reductions were greater for those households who had relatively smaller landholdings (Figure 2). In other words, landless households who are traditionally the most deprived experienced larger reduction in poverty at the end of the decade relative to households with xiii larger landholdings. It’s important to emphasize that this relationship is about the change in poverty rates over time. Land ownership and the level of poverty were negatively correlated over the entire decade. Figure 2: Percentage Change in Poverty Headcount (2005-2010) by Land Ownership in Rural Areas 0 Landless <0.05 acre Functionally Marginal 0.5-1.5 Small 1.5-2.5 acres Medium/large: 2.5 -0.05 Percentage change in poverty landless 0.05-0.5 acres acres or more -0.1 acre -0.11 headcount 2005-2010 -0.15 -0.2 -0.20 -0.18 -0.25 -0.26 -0.24 -0.3 -0.29 -0.29 -0.29 -0.33 -0.35 -0.38 -0.4 2000-05 2005-10 Source: HIES 2000, 2005, and 2010. 9. The demographic characteristics of the poor did not change dramatically. The analysis shows the standard correlations: households that are poor are (i) larger in size; (ii) have higher dependency ratios; (iii) have higher number of children; (iv) have household heads that are nearly three years younger on average than heads of non-poor households; and (v) have household heads that have little or no education. 10. Even though a more educated population helped to reduce poverty, improvements in the educational structure of the workforce accounted for a relatively small amount of the observed poverty reduction over the 2000-2010 period. Most of the education related poverty reduction occurred in the non-farm sector during the early part of the decade. While having a household head with no education is highly correlated with poverty (72 percent of the poor households are headed by an individual with no education), those households headed by an individual with no education (or less than primary education) saw the largest reduction in poverty over the decade. These patterns are consistent with the findings in the report on the increase in the relative price of unskilled labor.3 In particular, the report finds that higher food prices permeated the economy by increasing the wages of agricultural workers. Thus, it was primarily price shocks, and not increases in education levels, that increased low- skill wages of agricultural workers and thereby reduced poverty. 11. Health and nutrition outcomes tell a bittersweet story for Bangladesh. In the 2005-2010 period, Bangladesh witnessed significant improvements in terms of access to health care, the most notable achievement being the significant decline in the immunization gap between the poor and non-poor. With regard to nutrition, however, the evidence suggests that change is slow to come. Bangladesh is unlikely to meet the Millennium Development Goal of reducing moderate food deficiency (access to fewer than 2,122 kilocalories per person per-day) to 24 percent. In 2010, an estimated 38 percent of the population experienced moderate food deficiency, dropping just 6 percentage points over the last decade. Similarly, low dietary diversity, as measured by the Household Dietary Diversity Score (HDDS), was a persistent problem in Bangladesh even as the country experienced significant declines in inter-regional differences 3 This has also been described as an increase in returns to endowments or characteristics rather than changes in these endowments. xiv in poverty rates. Figures 3 and 4 show that areas with higher poverty rates in 2005 also registered higher poverty declines between 2005 and 2010 but that no such relationship is found between the 2005 HDDS and the poverty changes between 2005 and 2010. Figure 3: Decline in Poverty Rates between 2005 and Figure 4: Annual Changes in HDDS between 2005 2010 by PSUs and 2010 by PSUs 0.12 y = 0.174x - 0.0427 0.20 y = -0.0029x + 0.0478 Anual Percentage point Decline between Annual Change between 2005 and 2010 R² = 0.5904 R² = 0.0005 0.10 0.15 0.08 0.10 2005 and 2010 0.06 0.04 0.05 0.02 0.00 4.5 5 5.5 6 6.5 7 7.5 0.00 15% 25% 35% 45% 55% 65% -0.05 -0.02 -0.04 -0.10 Poverty Rate in 2005 Dietary Diversity Score in 2005 Source: Rabanni (2012). Data source: HIES 2005 and 2010. Source: Rabanni (2012). Data source: HIES 2005 and 2010. 12. Despite significant improvements in access to health and education services, large regional differences in access to electricity and sanitary facilities remain. In addition to faster poverty reduction, the West was also able to make larger gains in the human opportunity index of access to health and education services relative to the East. This index, which measures the level of access to services conditional on certain exogenous circumstances, reveals that children from less integrated Western regions had relatively better access to professional health services and education whereas children from integrated Eastern regions had better access to electricity and sanitation. Access to electricity and sanitary facilities was dismally low in rural areas relative to urban areas irrespective of their level of integration. This latter finding suggests inequities in access to these services within the integrated Eastern regions that are biased towards urban areas. Drivers of Poverty Reduction 13. In the 2000-2010 period, growth in consumption expenditure rather than its redistribution was the main driver of poverty reduction. Growth incidence curves (GIC) display growth rates at each percentile of the consumption distribution and are a useful tool to examine whether growth was shared across the spectrum of rich and poor. Between 2000 and 2005, the GIC reveals that the growth in per- capita consumption largely benefited both the rich and the poor equally, though there is some evidence that the extreme rich and the extreme poor grew at a faster than average pace (Figure 5). The story was different in the 2005-2010 period where growth was more ‘pro-poor’. In particular, the increase in per- capita consumption was higher than average for those below the 80th percentile and above the 5th percentile of the per-capita consumption distribution. Growth over the decade was, on average, slightly pro-poor. In terms of inequality of real per-capita consumption, as measured by the Gini index, there was also a modest decline in inequality at the national-level. The ‘pro-poor’ growth of the second half of the decade had an equalizing effect. xv Figure 5: Growth Incidence Curve 2000-2005 2005-2010 2000-2010 Growth Incidence Curve - Bangladesh Growth Incidence Curve - Bangladesh Growth Incidence Curve - Bangladesh 3 5 2.5 2 4 2 Growth Rate Growth Rate Growth Rate 1 3 1.5 0 2 1 -1 -2 .5 1 0 20 40 60 80 100 0 20 40 60 80 100 0 20 40 60 80 100 Percentiles Percentiles Percentiles GIC Growth rate in mean 95% CI GIC Growth rate in mean 95% CI GIC Growth rate in mean 95% CI Note: The distribution represents the change over time in consumption expenditure. The base is the national poverty line for 2005. Source: HIES 2000, 2005, and 2010. 14. Changes in both income and the share of adult population largely contributed in reducing both the prevalence as well as the severity of poverty. To better understand the drivers of poverty reduction, the analysis decomposes the reductions in poverty into specific components of consumption (Paes de Barros et al., 2006). In the first part of the analysis, these components consist of the household’s marginal propensity to consume, the real income per-capita and the number of household members (see Figure 6, part A). This basic decomposition helps to determine the contribution of each of the components to changes in real per-capita consumption over the decade, and thus their relative contribution to the observed poverty reduction over the same period. The results (see Figure 7-A) show that over the 2000- 2010 decade, total income per adult explained 57 percent of the observed poverty reduction, while changes linked to the dependency ratio explained about 43 percent of the reduction. Figure 6: Decomposing Consumption Per-Capita Part B Propensity to consume & measurement error Shared of Share of adults occupied adults Consumption Labor income per Income per-capita per-capita adult Income per adult Labor income per Non Labor occupied adults Number of income per adult members n Part A Source: Inchauste and Olivieri (2012). Poverty and Labor income Labor income was the single most important contributor to poverty reduction. The analysis next focuses on decomposing changes in consumption (and poverty) resulting from changes in income per- capita through labor, non-labor, and household composition effects (Figure 7-B). Labor income was the dominant factor in higher incomes and lower poverty rates (both for the poor and extreme poor). Non- xvi labor income comprised of domestic and international remittances and other transfers reduced poverty but had a relatively smaller role. Demographics also played a positive role in poverty reduction. In particular, a higher share of adults in the household helped to increase income per-capita and thereby significantly reduce both the prevalence as well as severity of poverty. Figure 7: Contribution to Poverty Reduction 2000-2010 A. Changes in income and the share of adult B. Labor income was the single most important population both played large roles in poverty contributor to poverty reduction. reduction over the decade. 250 Other Non-labor Percentage of total poverty reduction 105 200 64.0 Transfers 85 64.9 63.2 150 152.0 100 103.9 126.1 Capital 65 45 50 44.4 49.8 Labor Income 58.1 66.8 40.6 25 49.3 0 5 -50 Occupation share -14.2 -21.3 -15 -30.8 -100 Adult population -35 -150 Head count Gap Severity Head Gap Severity Consumption Income Consumption Income Ratio Adult population Income count Ratio Source: HIES 2000, 2005, and 2010. 15. Both labor income and the increase in the share of the adult population were the two most important contributors to the regional poverty reduction patterns as well. However, they accounted for a larger share of the overall poverty reduction in the West relative Figure 8: Contributions to Poverty Reduction to the East (Figure 8). Inferences Unexplained based on the linkage between labor 170 Other Non-labor income and poverty reduction at the 120 national-level (and similarly, the 70 Transfers declining dependency ratio and Capital poverty reduction) are therefore all the 20 Labor income more relevant for the West. -30 Occupation share Interestingly, changes in the -80 consumption-to-income ratio were Adult population National East West correlated with poverty reductions in the East, but had adverse effects on Source: HIES 2000, 2005, and 2010. poverty in the West. A potential interpretation of this is that relative to 2000, in 2010 Western households were consuming less of their income (that is, saving more) which would translate to higher poverty rates for given level of income. Transfers, remittances, and the share of occupied adults played a more significant role in poverty reduction in the East relative to the West. 16. Growth in labor income was mostly driven by increases in farm income. Using a more extensive decomposition methodology proposed by Bourguignon et al. (2008), the importance of changes in earnings and consumption on account of changes in educational attainment, age, gender, occupation, sector, and geographical distribution of the labor force are distinguished. The analysis suggests that the growth in labor income was driven by higher returns to individual and household endowments, pointing xvii to an increase in the relative price of labor and higher productivity as the main contributors to poverty reduction. In particular, returns to farm and non-farm endowments accounted for 64 percent of poverty reduction, with the large majority of this due to returns to farm endowments. The reduction in poverty associated with higher returns to endowments was concentrated in rural agricultural farm households. Given the large size of the agricultural sector, this pattern helps to explain the impressive reduction in poverty over the decade. 17. There were qualitative differences in how growth in labor income occurred over the decade. During the first part of the decade, the increase in wages in the non-farm sector was the most important factor contributing to poverty reduction. Additionally, there were also three ‘poverty reducing’ shifts taking place: (i) workers moving away from agriculture and towards manufacturing and services, (ii) workers moving away from daily and self-employed work and towards salaried jobs, and (iii) an increase in the level of education of the workforce. During the second half of the decade, more of the poverty reduction occurred in the farm sector. In particular, the farm sector experienced a significant increase in labor income, which was not associated with higher levels of education or changes in occupation. This was largely due to an increase in rural real wages, which reduced poverty and the gap between the urban and rural wages. 18. The purchasing power of Figure 9: Rural versus Urban Real Income Growth workers from urban areas fell relative to that of workers from 15% rural areas. Dividing workers by occupations and region, it is clear that 2000-05 2005-10 2000-10 10% most of the income growth taking place in the second half of the decade 5% came from the rural self-employed working in their own farms (Figure 9). Day laborers from rural areas also saw 0% their incomes grow, albeit at a more moderate rate. Relative to rural areas, -5% the growth pattern of income was less favorable in urban areas. While day -10% laborers and salaried workers from Rural Urban Rural Urban Rural Urban Rural Urban urban areas saw their incomes stagnate Day Laborer Nonfarm Self- Salaried Farm Self- over the decade, non-farm self- Employed Employed employed workers experienced a decline in their real incomes. Source: HIES 2000, 2005, and 2010. 19. Urban areas had strong growth in employment during the first part of the decade, but this growth was partially reversed during the second half of the decade. Between 1999 and 2005, the urban population grew from 21 percent to 24 percent of the total population; however, it declined by one percentage point in 2010. Concomitant with these population patterns, the urban labor market expanded rapidly during the first part of the decade, only to decline slightly in the latter part. Between 2005 and 2010, labor income originating from urban areas declined from 56 to 43 percent, returning to its 2000 level. 20. Increasing the labor force participation of women remains a challenge. The labor force participation rate of men remained relatively stable at around 80 percent during the 2000s.4 The labor 4 Labor force participation is measured as a proportion of the working age population, defined as people ages 15- 64. xviii force participation rate of women, on the other hand, increased from 25 percent to about 35 percent between 2000 and 2010. As in most developing countries, female labor force participation in Bangladesh increases at the higher end of the income distribution among the women who are also likely to have higher levels of educational attainment. Low wages (relative to men) partially explain women’s low rates of participation. 21. Early marriage and early motherhood are substantial constraints to increasing female labor force participation. In 2010, around one third of women between the ages of 20 and 24 were married by the time they were 15 years of age, this despite the legal age of marriage being 18 years. Although the relationship between marriage and the likelihood of working for women declined substantially over the last decade, the probability that a woman participates in the labor market falls by 22 percent when she is married, making marriage a substantial deterrent to female labor force participation. Similarly, holding all else equal, women with young children were also less likely to participate in the labor force. These findings suggest that postponing marriage, which would also delay early motherhood, could increase women’s welfare over the course of their lives.5 22. On the demand side, there was a significant expansion in job creation in industry and services throughout the decade. This helped usher in important structural changes, including: (i) a gradual decline in the size of the agricultural sector and a rise of the services and industry sectors; (ii) employment growth in urban areas (although only for the first half of the decade); and (iii) a movement away from agriculture, toward industry and services that coincided with movements away from daily and self-employed income and towards salaried employment. Figure 10: Business Constraints A. Major Business Constraints B. Share of firms expected to give gifts Electricity to get things done Corruption in meetings with tax officials Access to finance Tax administration to get a water connection Inadequately educated workforce Tax rates to get an import license Courts system to get an electrical connection Practices of informal competitors Crime, theft and disorder to get an operating license Customs and trade regulations Business licensing and permits to secure government contract Transportation to get a construction permit Labor regulations 0 10 20 30 40 50 60 70 80 90 0 10 20 30 40 50 60 70 80 90 Bangladesh South Asia All Bangladesh South Asia All Source: World Bank Business Enterprise Survey, Bangladesh, 2007. 5 Delayed motherhood is also likely to have a positive effect on the health and nutrition of infants and young children. xix 23. Access to electricity and finance, the prevalence of corruption, and the lack of an adequately trained workforce are significant bottlenecks for further business development. These constraints appear to be much more significant in Bangladesh relative to the rest of the South Asia region (Figure 10- A). About 78 percent of firms identified access to electricity as a major constraint. The second biggest constraint as reported by business owners was corruption, with 55 percent of firms identifying it as a major constraint. Further, 85 percent of firms reported having to give gifts to public officials “to get things done”, compared to 37 percent for the rest of the South Asia region (Figure 10-B). Third, access to finance was identified as a major constraint by 43 percent of firms. Lastly, the lack of an adequately trained labor force was reported as a major constraint by 25 percent of firms. The impact of corruption on the business climate cannot be overemphasized, as it directly affects investment decisions and therefore the potential for further job creation. 24. The agricultural sector also continued to be important, employing nearly half of the workforce and providing over 45 percent of total household income. Although the agricultural sector had been shrinking over the past decade, it experienced an upswing in income growth during the last half of the decade. The share of agriculture in total employment declined from 51 to 48 percent between 2000 and 2005, and further to 47 percent in 2010 (Figure 11-A). However while total labor income from agriculture grew at only 1.7 percent annually between 2000 and 2005, between 2005 and 2010 it grew by an astounding 9.8 percent a year (Figure 11-B). Nevertheless, it is important to note that unlike the trend observed in the agriculture sector, real incomes in the industry and services sectors declined in the second half of the decade. In the case of the industry sector, the sectoral decline in real income that occurred in the second part of the decade more than offset the gains that took place in the first part of the decade. Figure 11: Sectoral Employment A. Sectoral Composition of Employment B. Annual Growth in Income by Sector (percent of total employed) (annualized growth rate of total income) 60 12% 51.3 2003 2005 2010 10% 2000-05 50 46.6 8% 2005-10 40 34.7 35.2 6% 2000-10 4% 30 2% 18.2 20 14.0 0% 10 -2% -4% 0 -6% Agriculture Industry Services Agriculture Industry Services Source: Labor Force Surveys. Source: HIES 2000, 2005, and 2010. 25. Demographic trends over the past decade suggest potential opportunities as well as challenges for future growth and poverty reduction through job creation. Population growth slowed down considerably over the last thirty years, declining from an average of 2.7 percent per year in the 1980s to an average of 1.4 percent per year in the 2000’s. Despite the slowdown, Bangladesh added 19 million people to its total population, a 15 percent increase between 2000 and 2010. The working age population expanded more rapidly than the total population, growing at an average rate of 2.3 percent between 2000 and 2010, a 25 percent increase over this period. The ‘bulge’ among the 5-14 year-olds in the 2010 population indicates that this trend will continue over the next decade. While a growing labor xx force can be an asset for income generation and growth, absorbing such a large wave of new entrants every year poses a major challenge for the labor market. Poverty and the Demographic Transition 26. Changes in the demographic composition of the population, in particular a lower dependency ratio due to an increase in the adult population, played a central role in reducing poverty over the decade. Given that it was mostly young people entering the workforce (that is, those who generally earn less than their more experienced counterparts) the effect of having a larger workforce was countered by lower returns to these young workers. Nevertheless, the ensuing lower dependency rate was a large contributor to poverty reduction. Similarly, the growing share of women in the labor force also contributed to poverty reduction, with the marginal effect of this change being larger in the first half of the decade, particularly for the non-farm sector. The decomposition analysis of the determinants of poverty decline reveals that demographic factors, particularly, age, gender and the regional composition of the workforce contributed to at least 25 percent of the observed poverty decline. Understanding the factors behind the observed demographic changes is therefore important to understand how demographic transition has contributed to poverty reduction. 27. The lower dependency ratio was largely driven by reduction in fertility, whereas the changes in living arrangements were driven in part by labor mobility and changing social norms. Bangladesh’s dramatic fertility decline is a result of an aggressive, supply-driven family planning program in place since the 70s that provided door-step delivery of contraceptives to women who had traditionally been in purdah. The family planning contraceptive program went hand in hand with improved family planning service delivery through clinical services, community-based distribution, social marketing, and a committed band of workers (mostly female). Contraceptive prevalence in Bangladesh rose from a mere 8 percent in 1975 to over 61 percent in 2011. The decline in fertility has been a major factor in the overall improvement in maternal and child health. Therefore, though the primary driving force for the family planning program was to reduce the pace of population growth, it had very positive spillovers on maternal and child health outcomes as well. 28. Bangladesh will soon be at the Figure 12: Births per Woman Across Time replacement level6 of fertility. Coinciding with the extensive family planning programs, 10 6.3 5.1 Births per woman Bangladesh halved its fertility rate between 1971 4.3 3.4 3.3 3.3 3.0 5 2.7 2.3 and 2004, going from more than 6 children per woman to about 3. The latest total fertility rate 0 (TFR) estimates show continued, steady decline from 2.7 children per women in 2007 to 2.3 in 2011, suggesting that Bangladesh will soon be at replacement level (Figure 12). Source: Bangladesh DHS 2011. 29. Bangladesh has managed to reverse the infamous gender inequality in infant mortality that characterizes most South Asian countries. Contrary to what is found in India and contrary to Bangladesh in previous decades, boys now have higher infant mortality rates than do girls, which is in keeping with the biological norm. However, neonatal mortality (death in the first 28 days) has fallen much more slowly than mortality above one month of age. 6 Replacement fertility is the total fertility rate at which women would have only enough children to replace themselves and their partner. The replacement fertility rate is roughly 2.1 births per woman for most industrialized countries. xxi 30. Changes in fertility and mortality have important consequences for the age structure of the population. Reflecting the decline in fertility and mortality, the change in population pyramids for the years 2000 and 2010 show a lower proportion of children and an increase in the number of working age adults (Figure 13). While there was an increase in the share of the elderly, the increase in the share of working age adults by and large outpaced the increase in the share of the elderly. These trends combined with the decrease in the share of children yielded a lower overall household dependency ratio at the end of the decade. Figure 13: Change in Population by Age Groups and Divisions, 2000-2010 4% Percentage point 2% 0% -2% Barisal Chittagon Dhaka Khulna Rajshahi Sylhet -4% [0-5] [6-10] [11-15] [16-20] [21-25] [26-30] [31-35] [36-40] [41-45] [46-50] [51-55] [56-60] [61-65] [66-70] [71-75] [76-80] [81-85] [85+] Source: HIES 2000, 2005, and 2010. 31. The patterns of changes in the age composition of the population for the 2000-2010 period are consistent with the changes in the regional poverty patterns described in Figure 1. While the East was experiencing larger reductions in poverty in the first half of the decade, it also was experiencing a relatively larger decline in the share of under-15 year olds in the total population relative to the West (Figure 14). Similarly, during the first part of the 2000-2010 decade, the increase in the share of the working-age population (15-64 age group) was larger in the East relative to the West. Because both of these demographic patterns mean that dependency ratios were declining at a faster rate in the East during the first part of the decade, one would expect poverty rates to be declining faster as well. In the latter half of the decade, between 2005 and 2010, both of these demographic patterns reversed and correspondingly, the West experienced relatively more reduction in poverty. The demographic trends are very informative in explaining both the diverging (2000-2005) and the converging (2005-2010) poverty patterns over the decade and across regions. Figure 14: Change in Population by Age Groups and Divisions, 2000-2005 and 2005-2010 3 2 Percentage point 1 0 -1 -2 -3 2000-05 2005-10 2000-05 2005-10 East West [0-14] -2.65 -1.74 -1.42 -2.33 [15-64] 2.55 1.32 2.09 1.77 [64+] 0.11 0.41 -0.68 0.57 Source: own estimates using HIES 2000, 2005, and 2010. xxii 32. In addition to changes in household size and composition, change in the organizational structure of households was a third important change in household structure over the decade. In particular, while nuclear households continued to be the predominant type of living arrangements, there was a decline in the proportion of joint and extended households and a commensurate increase in the proportion of single, semi-single and nuclear households. The decade also gave way to the breakdown of the traditional pattern of co-residence with sons. The proportion of elderly living with their sons dropped consistently across the country over the 2000-2010 decade, whereas the proportion of the elderly living with ‘others’ increased significantly. 33. These demographic shifts imply that Bangladesh needs to aggressively cater to the needs of the various age cohorts in the coming years in order to continue on a path of poverty reduction. First, the swelling youth cohorts underscore the need for the Government of Bangladesh (GoB) to invest in skills and education (including general, vocational and technical schooling) for the youth entering the labor force. Linking this labor to productive employment either via overseas migration or in growing sectors of the economy will be critical to fully realize a “demographic dividend” which is likely to last another twenty odd years. With increasing numbers of the elderly surviving longer and tending to live on their own, Bangladesh will need to develop a strategy for handing the aging population. Over the next twenty years, Bangladesh will need to consider programs and policies to protect the elderly in a manner that is fiscally sustainable and culturally appropriate. Another emerging challenge (and opportunity) resulting from the demographic transition is to provide the working-age population with incentives and opportunities to save for retirement and to insure against risks. A cushion against shocks is particularly important given the large number of vulnerable people who live just above the poverty (in 2010, 52 million non-poor people consumed less than 1.5 times the value of the poverty line, a reminder that a large number of people remain very vulnerable to shock-induced poverty). Shocks and Coping Mechanisms of the Poor 34. Poor households living in the poorest regions are less able to cope with shocks . Household survey data show that households in Bangladesh still rely on coping mechanisms that are likely to have negative welfare consequences or that are very costly. Rural households are particularly vulnerable. For example, while urban households a more likely to rely on savings relative to rural households, rural households are more likely to deplete their assets or to use high-interest loans from money lenders relative to their urban counterparts. When faced with climate shocks, an overwhelming majority of households are not able to effectively cope with the shock. This is consistent with the notion that when dealing with climatic or covariate shocks households are unable to rely on community-based coping instruments. 35. Seasonality of consumption still prevails in Rangpur. The austere seasonal poverty and hunger, or Monga, observed in Rangpur, is well known (see Box 1). Due to its severity, this seasonal phenomenon which lasts about 3 months, from September to November, has important consequences for livelihoods and well-being. Even as the depth of poverty declined by nearly a third and the severity of poverty decreased by more than 34 percent over the 2005-2010 period, Rangpur continued to lag substantially behind other regions. In 2010, the prevalence of poverty in Rangpur was higher than the nation’s average in 2005, suggesting that Rangpur is at least five years behind in terms of poverty reduction relative to the rest of the country. 36. The seasonal deprivation that households experience during the lean season increases their employment diversity significantly. In particular, the analysis presented in this report reveals that employment diversity within a household is the highest in the Monga season. While diversification of employment can improve resilience against shocks, the empirical evidence suggests that the intra- household employment diversification that prevails in rural Bangladesh appears to have negative effects on household consumption and increases the probability of falling into poverty. xxiii Box 1: Rangpur—Distinct Features of a Lagging Region Some of the major reasons for Rangpur’s comparative disadvantage are:  Inadequate investment in infrastructure, including electricity, resulting in a non-diversified rural economy and limited opportunities for off-farm employment.  Low crop yields due to poor soil quality (for example, soil salinity).  A high proportion of landless households that depend on wage-labor income.  Low wage rates for both male and female agricultural day laborers.  Risk of floods and river erosion.  Livelihood vulnerability of people living in char areas, consisting of reclaimed land from rivers and including tiny island-like fragments.  Poor inflows of remittances from migrant family members working in the country or abroad. Source: Background paper by Khandker and Samad (2012a) prepared for this Poverty Assessment Report. 37. Among the poor living in regions affected by seasonal deprivation, seasonal migration is used as a key coping strategy, whereas access to remittances does not constitute an important source of income. In many areas of rural Bangladesh, remittance incomes from family members working abroad represents a significant proportion of household income, as well as a substantial source of fund inflows into the local economy. Rangpur is an exception however. For example, 18 percent of individuals aged 12 years or older in the Monga region migrated temporarily compared to only 7 percent of that same age category in the Non-Monga region. Among the bottom 20 percent of the population, around 22 percent migrated compared to only 9 percent of households from the richest 20 percent of the population. Remittances however comprised only 4.7 percent of total household income, compared to 14.8 percent for rural households in other regions. 38. Access to credit remains fairly limited and is generally insufficient to smooth seasonal consumption patterns. The analysis presented in this report shows that the poorest 20 percent of the population in Monga areas is most vulnerable to seasonal deprivation and, therefore, has relatively high rates of loan application and approval. Nevertheless, the poorest 20 percent of the population from Monga areas has a lower loan approval rate relative to the average-income-level household residing in Monga areas as well as to their counterparts residing in non-Monga areas. 39. Social safety net expenditures were Figure 15: Percentage of Households that have Access not well targeted towards those most in need to Social Safety Nets in the Monga Region of income support to avert seasonal food deprivations. The safety net coverage in the Poorest (Bottom 20%) Monga areas was around 35 percent. The coverage rate for the extreme poor was nearly 2nd Quintile 44 percent, implying that 56 percent of the 3rd Quintile extremely poor households were left behind 4th Quintile with no access to any safety net programs. Richest (5th Quintile) Moreover, 27 percent of the richest quintile received some support from safety nets in the Source: Monga 2008/09 data Monga region (Figure 15). 40. Poor households are also vulnerable in the short run to adverse food price shocks. Despite having increased its food grain production over the past years, from 25 million metric tons in 2000 to around 30 million metric tons in 2008-09, Bangladesh continues to rely on food grains imports to meet its domestic demand. As a result, Bangladesh remains vulnerable to both local and global price spikes. Since xxiv 2005 prices of major food crops have been surging in the international market: between 2005 and 2008, rice prices rose by 25 percent, wheat prices by 70 percent and maize prices by 80 percent. 41. The nutritional intake of most Bangladeshis is negatively affected by staple food price shocks. For the average Bangladeshi household, about 74 percent of calories Figure 16: Cereal Consumption in Bangladesh consumed come from cereals (Figure 16). Energy Consumption by Food Groups Moreover, rice alone represents more Milk and milk than 40 percent of total consumption for Pulses/legumes/ nuts products Oil/fats 8% the poorest households. As a result, the Fish and seafood 2% 1% Sugar/honey nutritional intake of most every 3% 2% Eggs Bangladeshi is likely to be harmed by 1% food commodity price shocks, and the deleterious impact of the shock on Meat, poultry, offal nutrition is likely to affect poor children 1% Fruits the most. As one example, coinciding 1% with both the beginning of the food Vegetables commodity price spike and devastating 4% Cereals 74% floods that hit South Asia in 2007 Root and tubers 3% (leaving nearly 7 million people marooned),7 there was a dramatic increase in the number of wasted 0 to 59 Source: Background paper by Rabbani (2012) commissioned for this Poverty Assessment Report. months old children between 2004 and 2007. 42. Indeed the short-term effect of the 2007-2008 food price increase affected the poor disproportionately. Given that the poor spend most of their income on food, they bore the brunt of higher food prices in the immediate aftermath of the shock. In the medium term, after wages adjusted to reflect the price shock, the impact of the food price shock largely equalized along the wealth distribution. In the long term, after the price shock permeated to other sectors, particularly the service sector (the output of which is disproportionately consumed by the rich), the short term effect of the shock appears to have been moderately reversed: the population at the higher end of the income distribution was relatively more affected than those at the lower end of the distribution. 43. Coinciding with the staple-food price shocks, real rural wages escalated in the second half of the 2000s. The monthly wage data suggests that the food price shock hit households immediately between 2007 and 2008; by 2009, wages began adjusting to the new price regime, bouncing back to their pre-shock level (Figure 17). Interestingly, while the wage adjustment took place in both urban and rural areas, the urban wage increase meant that wages went back to their near pre-price shock levels, whereas the rural wage increase more than compensated for its earlier decline. Moreover, the real wage patterns imply that urban wages fluctuated more than rural wages. These wage patterns are consistent with the findings from the decomposition analysis which suggest that the growth in income experienced by rural day laborers over the 2005-2010 period contributed to poverty reduction. 44. The exogenous price increase, and not necessarily an increase in their labor productivity (measured in physical terms), allowed rural agricultural workers to experience an increase in their wages. This finding suggests an explanation as to why the largest contribution to poverty reduction over the last decade was the increase in returns to endowments or characteristics, rather than changes in these 7 See http://news.bbc.co.uk/2/hi/south_asia/6927389 (retrieved on November 10, 2012). xxv endowments.8 Moreover, the fact that poverty reduction over the second part of the decade was highly concentrated among rural agricultural farm households provides further support to this hypothesis, as it is these households that were most likely to benefit from the price shock. The evidence presented in the report suggests diversifying away from agriculture is not necessarily essential to obtain poverty reduction. In fact the data show that while over the first part of the decade the reduction in poverty was due to diversification, over the second part of the decade, poverty reduction was the result of an increase in the returns to farming. Figure 17: Real Rural and Urban Wages in Bangladesh 14.0 Urban and Rural Wage adjusted by coarse rice price 12.0 Pre-food price shock Food price shock Wage adjustment period 10.0 8.0 6.0 Rising real wages? 4.0 Urban_Wage Rural_Wage Note: The monthly wage data is obtained from Monthly Statistical Bulletin, Department of Agricultural Marketing (DAM), Bangladesh. The real wages are deflated by coarse rice prices. The rural sample excludes the rural wage from the mega cities. Urban wage are for the unskilled workers (for example, helpers in the construction sites, carpentries, and other sectors). The base year is set to 2009/2010. Source: Background paper prepared by IFPRI (2012) for this Poverty Assessment Report. 45. Thus the scaling up of operations such as the Open Market Sales program in urban areas in the immediate aftermath of the 2007-08 price shocks was an appropriate Government response. The program was designed to sell rice at a subsidized price of Tk 25/kg up to 5 kg per household using designated dealers. In rural areas, given the existing problems of leakage in the food-based safety nets, the step to set up a self-targeted, cash-based workfare program (then known as the Hundred Day Employment Program but now known as the Employment Generation Program for the Poorest) was also a step in the right direction. The analysis presented in this report suggests two key lessons regarding the use of public safety nets to buffer the negative impact of price shocks: (i) the importance of the timeliness of the assistance to cushion the immediate impact of a crisis, and (ii) the importance of proper targeting of the assistance to those who need it the most. 46. Bangladesh has a wide spectrum of social safety net programs that can potentially mitigate the impact of shocks. The Government provided an average of 12 percent of total annual public expenditures (about 1.5 percent of GDP) for social protection from 2000-2008. Since then, the budgetary allocations were increased in response to the 2008 global food and energy price crises to about 14 percent of the total budget on average over 2009-2012, and reaching as high as 2.6 percent of GDP in FY11. Such increases in allocation were reflected in the HIES which showed that the proportion of households covered by safety net programs also rose from 13 percent in 2005 to 24 percent in 2010, bringing the total number of households reached by social assistance programs to roughly eight million. However even though the safety net coverage of the poor has improved it nevertheless remains low: in 2010, a third of the poor received at least one social assistance program as compared to 21 percent in 2005. 8 In particular, returns to farm and non-farm endowments amounted to almost 65 percent of the observed reduction in poverty, which occurred primarily in the farm sector. xxvi 47. Dramatic improvements in the geographic coverage of safety net programs took place between 2005 and 2010. The vast majority of safety net beneficiaries reside in the country side: 7.2 million recipient households live in rural areas, which is about ten times the number of urban recipient households. The population coverage of safety net programs, which varies significantly by region, is now positively correlated with division level poverty rates. For example, Barisal, with the highest poverty rate (39 percent), has the second highest safety net coverage among all divisions (34 percent of the population receiving benefits from at least one program). In contrast, Chittagong and Dhaka, which have the lowest poverty rates in the country (respectively 26 and 31 percent), have the lowest safety net coverage (19.4 and 18.8 percent). This distribution of SN beneficiaries across divisions represents a significant improvement over time: in 2005, the coverage of safety nets was negatively correlated with divisional level poverty rates. 48. However while participation in safety nets Table 2: Access to SNs by Per-capita remains progressive there are larger inclusion errors Expenditure Quintile 2005, 2010 in 2010 compared to 2005. In 2010, the poorest quintile Quintile 2010 2005 has the largest proportion of households covered by at 1 39 24 least one safety net program (39 percent). Nonetheless, 10 2 32 16 percent of those in the richest quintile were also covered by at least one safety net program. Furthermore, 3 25 14 expansion of safety nets occurred faster within the richest 4 20 8 quintiles: the proportion of beneficiaries in the top 40 5 10 4 percent of the income distribution increased by 150 Total 24.6 12.6 percent over the past five years, while the poorest quintile Source: HIES 2005 and 2010. increased coverage by 62.5 percent (Table 2). 49. The targeting performance of safety nets did not improve even though there was a substantial expansion in cash-based safety nets over a five-year period. In 2005, the bulk of public transfers in Bangladesh were allocated to more expensive and “leaky” food transfer programs. In 2010, the data shows that the Government has boosted cash transfer programs and thereby their share of total program spending (Figure 18). The increased emphasis on cash transfers reflects a recognition of their greater cost effectiveness and presumption of their lower risk of misappropriation. Nevertheless, food transfer programs remain an important pillar of Bangladesh’s food security strategy, and serve a secondary role in turning over the country’s emergency public food grain stocks. Efforts to further phase out food-based transfer programs is therefore a nontrivial process and will require careful long term planning. Figure 18. Proportion of households with access to safety nets 2005 2010 4% 6% 87% 12% 76% 24% 6% 15% 3% 2% No Safety Net Cash only No Safety Net Cash only Kind only Cash and kind Kind only Cash and kind Source: HIES 2005 and 2010. xxvii 50. The adequacy of actual benefit amounts received by safety-net beneficiaries, as reported by households, still remains very low. Transfers to poor households, whether in kind or in cash, should be sufficient in size to have some substantial positive effect on their well-being. Over the 2005-2010 period safety-net benefit amounts increased only marginally in real terms, and remained small relative to the needs of a typical poor beneficiary. The value of safety net benefits to poor accounted for only about ten percent of consumption for poor beneficiaries. 51. Poor targeting of resources and inadequate transfers make the current allocations for safety nets inadequate to contribute significantly to poverty reduction. With the current average benefit amounts thinly spread, simulations suggest safety net transfers could potentially be reducing the level of poverty by 1.5 percentage points. That is, in the absence of SN transfers, poverty rates in Bangladesh would be 33 percent (up from 31.5 percent). If however the average SN transfer was given to the poorest families, poverty rates in Bangladesh would be 27.2 percent, or a reduction of 4.3 percentage points.9 52. The empirical evidence suggests that access to microfinance is more effective than safety nets in helping households cope with shocks. In 2009, the two most common shocks experienced by households were health and climate-related shocks. More than 50 percent of Bangladeshi households reported experiencing one or more shocks over a one-year recall period. Rural households were more likely to experience shocks (and also face a higher number of shocks on average) relative to urban households. Less than 2 percent of households reported safety nets as one of the top four coping mechanisms used to face a shock. The relative importance of savings and borrowing compared to safety nets in dealing with shocks is not surprising given the coverage and reach of microfinance organizations operating in Bangladesh and the relatively low transfer amounts and coverage of safety nets to the poor. 53. Since the mid-1990s, the Figure 19: Trend of Microfinance Membership in Bangladesh microfinance sector in Bangladesh 40 grew at a phenomenal rate. 30 Active members 20 Microfinance institutions providing loans (millions) 10 in Bangladesh had about 34 million 0 members in 2010. Figure 19 shows that 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 the membership in any sort of Microfinance Institutions (MFI) grew from about 8 million in 1996 to over 34 Source: CDF (1996-2007) InM & CDF (2008-2010) Grameen Bank (2010) million by 2010. 54. Regular microcredit program participants do better in terms of exiting extreme poverty than irregular participants and non-participants. A 3-round panel household survey conducted in 1991/92, 1998/99 and 2011 (spanning over a 20-year period) shows that poverty as well as extreme poverty reduction was steady for all households (i.e. those who participated regularly or long-term participants, those who participated irregularly or short-term participants, as well as those who never participated or non-participants in microcredit programs). This finding is consistent with the overall poverty reduction patterns observed at the national-level. However, even when the prevalence of extreme 9 These results were obtained the following way: SN transfers were subtracted from recipient households. Households were then ranked in increasing order of per-capita consumption. Starting from the poorest, all households were given the average SN transfer of Tk 488 per month until exhaustion of the budget that was allocated in the initial (“real”) scenario (Tk 3.81 billion per month). Poverty rates were then obtained using that modified consumption measure. Since transfer amounts reported by households can be underestimated, a similar exercise was conducted using an average household transfer of Tk 700 (instead of Tk 488), that is a 40 percent increase. The simulated poverty rates were then 24.8, or a reduction of poverty rates of 6.8 percentage points. xxviii poverty was about the same for all three groups of households in 1991/92, long-term participants in microcredit programs did consistently better in reducing extreme poverty rates over time relative to short- term participants and non-participants. 55. An important role for microfinance is in smoothing consumption. While the existing evidence of a causal pathway between microfinance participation and changes in household-level welfare is mixed, the existence of a causal pathway between the microcredit and consumption smoothing is more widely accepted. Using cross-sectional data from nearly 1800 Bangladeshi households, a recent study shows that household consumption variability lessened by about 50 percent among Grameen Bank borrowers;10 that is, when faced with income shocks, households with access to microfinance experienced the shock with half the severity of those without access. Consistent with the literature, the findings from this report suggest that microcredit program participation is positively associated with a household’s ability to smooth consumption. (Research also suggests that poor households value and use financial services more for managing household consumption and cash flow.11) Given the expectation that Bangladesh will continue to be subject to severe, adverse economic shocks, microfinance will continue to play an important role in helping the poor to buffer themselves from these shocks and to help mitigate extreme poverty. Policy Implications 56. Regional comparison suggests scope for further improvement in poverty alleviation. Despite its remarkable poverty reduction patterns over 2000-2010, Bangladesh continues to have the highest rate of poverty relative to its South Asian counterparts. Bangladesh’s five-year cumulative rate of GDP growth for the 2005-2010 period was below the South Asian average growth rate (admittedly, a rate primarily driven by India’s high growth pace). By the end of the decade, the per-capita GDP (in constant 2000 USD) in Bangladesh was the second lowest in the region, followed only by Nepal. The low real per- capita GDP explains why Bangladesh also continues to have the highest rate of poverty in the region. Nevertheless, even as Bangladesh’s high poverty levels must be underscored, both the HIES and the national accounts suggest that, unlike the 1991-2000 period, growth in Bangladesh over the last decade was characterized by two positive trends: a high speed of poverty reduction and a stabilized level of inequality relative to other countries in the region. 57. Understanding the evolution of poverty patterns, the main drivers of poverty reduction and vulnerabilities faced by the poor are particularly important to inform the policies for the next decade. The analysis presented in the report shows that the two most important contributors to poverty reduction observed in Bangladesh over the 2000-2010 decade were: (i) the growth of labor income, which increased mostly due to an improvement in the returns to existing endowments (fueled by the rising food prices and a tightening of the rural labor market), rather than due to an improvement in household endowment levels; and (ii) demographic changes, particularly lower dependency ratios resulting from the lower fertility rates. Public and private transfers also contributed to poverty reduction, but did so to a lesser extent. The poverty reduction policy recommendations that follow from the extensive analysis undertaken in this report are separated into three categories: (i) policies aimed at boosting income generation; (ii) policies aimed at managing the demographic transition; and (iii) policies aimed at strengthening the safety net system. 10 Murdoch, J. 1998. “Does Microfinance Really Help the Poor? New Evidence from Flagship Programs in Bangladesh”. Princeton University Working Paper No. 198 11 Daryl Collins, Jonathan Morduch, Stuart Rutherford, and Orlanda Ruthven, 2009. Portfolios of the Poor, Princeton: Princeton University Press, 2009. xxix Income generation 58. Increase the productivity of workers. To the extent that real rural wages continue to rise (either driven by rising commodity prices, the tightening of the labor market, or productivity increases), this will almost certainly result in continued reductions in poverty, particularly so in the farm sector where a large share of the poor live. This will also mean a further narrowing of the East-West gap, as the majority of the farm daily and self-employed workers are located in the West. While this is good news for the rural poor, it is also important to consider the implications of the source of such growth in wages. Growth in wages driven by price shocks and adjusting relative prices, means poverty reduction in Bangladesh will be vulnerable to external price fluctuations given that farming continues to be the employment sector where most of the poor earn their livelihoods. To reduce the reliance on externally induced poverty reductions, it will be prudent of the Government to continue investments in improved agricultural productivity, primary schooling and human capital formation focused on low-skilled workers. Measures to improve the stock of human capital of those with the lowest stock will lead to reduction in poverty through improved productivity. 59. Ensure that the demand for salaried work in the manufacturing and service sectors continues to grow. The type of poverty reduction observed in the first part of the decade on account of movements toward non-farm work calls for policies to ensure that the demand for salaried work continues to grow resiliently. Because this type of work necessitates a constant supply of skilled workers, continued efforts to expand the quantity and quality of education can go a long way in ensuring sustainable and inclusive rather than diverging growth. Similarly, there is an evident need for active support to entrepreneurship via the promotion of the rule of law (that is, strengthening transparency and accountability). Similarly, the demand side analysis of the labor market shows that improving infrastructure and promoting human capital investments have a great potential to generate the right investment environment for firms to continue creating jobs at the needed pace. Since low wages and a large labor force are the country’s major comparative advantages, wage growth in Bangladesh is bounded by wages prevailing in competing foreign direct investment (FDI) destinations (such as China, India, Turkey, Vietnam, etc.) which also happen to have better business environment and infrastructure than Bangladesh. It is important therefore, that low wages are not the only competitive advantage that Bangladesh can offer to investors as it comes at the expense of the wellbeing of the poor. Rather investments in infrastructure and appropriate labor laws are just as important as investments in raising worker productivity. Demography 60. Provide continued support to family planning programs (with a strong focus on the less well off) and on delaying early marriage. Despite the reduction in household size, on average, poor households continued to have more children than the non-poor households and their household heads were both younger by more than two years and less educated than the non-poor household heads. Because high dependency ratios and low education are both positively associated with poverty, this finding suggests that family planning programs should now refocus on the less well off, as well as on delaying early marriage. For example, instrumenting a holistic approach to address the problem of early marriage and its implications for human capital formation is a possible avenue for achieving further poverty reduction. Well-enforced age-of-consent laws, in combination with educational stipends that reduce the opportunity cost for children transitioning from primary to secondary school, could help these children to both successfully complete their secondary education and avoid early marriage and its potential negative links. Yet, it must be underscored that "changing attitudes is the strategy that underpins all other efforts to xxx end early marriage."12 In fact, changing attitudes is perhaps the most important and the most challenging aspect of averting early marriage and its consequences on well-being. It begins with a dialogue that takes place at the community level—involving parents, educators, religious leaders, decision-makers, and young adolescents for whom the dialogue has the most immediate impact. A policy blend involving changes in the law, the right incentives and a multi-sectoral dialogue aimed at changing attitudes towards early marriage could prove a powerful prescription for attaining further poverty reduction and improved human capital outcomes for women and their children.13 61. Aggressively improve the skills of the swelling youth cohorts to ensure they are gainfully employed in the coming years. Bangladesh will need to focus more attention to the skill development of a rapidly expanding labor force, including the policies aimed at enhancing opportunities for overseas migration. This will be essentially to ease the labor market pressures caused by the demographic transition. Given the trends in female education outcomes and low rates of female labor force participation, a focus on creating ‘female-friendly’ jobs and work environments and labor policies will also help to facilitate a higher level of entry by young female graduates into the labor force. A ‘demographic dividend’ is only as good as the policies in place to reap its potential benefits, and a strategy on job creation to address this window of opportunity of about 20 more years will be critical. 62. Begin preparations for aging out of the ‘demographic dividend’. With an increasing number of the elderly surviving longer and tending to live alone, Bangladesh may find itself at the cusp of an aging challenge in twenty years. The positive aspect of this is that Bangladesh has some time to prepare for it and to put in place programs and policies that can protect its elderly in a manner that is also fiscally sustainable and culturally appropriate. An emerging challenge resulting from the demographic transition is to also provide for those who have the means, some way to save for retirement and insure against risks. Social insurance coverage can be expanded in Bangladesh using the country’s unique advantage of a large and vibrant micro-finance network that provides limited contributory pension savings plans and health insurance to their poor member households. The Government can help expand this coverage by establishing the appropriate regulatory framework. Furthermore the Government can provide the right incentives to banks and insurance companies to offer pensions or social insurance to formal sector workers who currently have no access to these services. Safety Nets 63. Improve the linkage between safety nets and poverty reduction. The analysis of the report suggests that, to be more effective, safety net programs need to be: (i) better timed to more adequately address short-term needs, (ii) better targeted ensure that benefits are primarily received by the poor, and (iii) better tailored to meet the specific needs of the poor. This implies a strong focus on current safety nets to improve their design, and strengthen the institutional and administrative capacity of programs. It also means consolidating the numerous programs to adequately address poverty and mitigate vulnerability to poverty in a sustainable way. Consolidation of safety net programs in Bangladesh along these three principles would improve efficiency and establish a solid foundation for increasing investments in safety net programs with increased benefit levels. 64. Address persistent seasonal shocks and global commodity price volatility more effectively. Further deepening of micro-lending and safety nets in terms of both coverage and improved targeting will help poor households smooth their consumption to mitigate seasonality of income, particularly in the ecologically vulnerable regions such as Rangpur. In order to buffer the negative impact of future food price shocks, the Government must carefully design well-timed and well-managed policies that provide 12 UNICEF. 2001. 13 Field and Ambrus 2008; Brown 2012; Bates et al. 2007. xxxi targeted assistance to those that need it most. This will mean further improving the performance and scope of large seasonal employment generation programs for the poor such as Food for Works, Test Relief and EGPP. Unlike price controls and export barriers, these policies are likely to be more operationally challenging, yet are also proven to be more effective and to have a less deleterious long- term impact on the rest of the economy. Moreover, as the Bangladeshi diet is known to be low in micronutrients and diversity, conditional cash transfer programs, which have been shown to increase the diversity and caloric content of the food consumption of participant households, can be supplemented with blanket distribution of micronutrients sachets. The latter is a relatively cheap way to effectively fight micronutrient deficiency and the related morbidities. 65. Link cash based transfers to human capital formation among the poor. Safety nets have gradually shifted from food transfers to cash transfers in recognition of the fact that the latter are more cost effective. Linking this larger pool of unconditional cash allowances to human development outcomes is likely to improve the quality of social protection expenditures. Particular emphasis needs to be placed on programs and policies that focus on: (i) early childhood development in a way that integrates health and nutrition services, early stimulation and learning, and pre-school education, and (ii) building the skills and improving the employability of the poor youth. xxxii Introduction 1. Bangladesh began its journey as an independent nation in 1971 with an extraordinarily bleak prognosis. Devastated by war, overpopulated, perennially vulnerable to natural calamities, and with few natural resources, prospects for a future beyond mere survival had appeared as forbiddingly difficult for the new nation. In 1973, the International Economic Association, during its annual meeting in the capital city of Dhaka, expressed such pessimism when it described Bangladesh as a ‘test-case of development’. Such pessimism was not unfounded. 2. During the first two years of the country’s independent existence, the growth rate of GDP was negative, -6.83 percent in 1971 and -14.74 percent in 1972, as a consequence of the destruction and displacement wrought by the war of liberation. Other economic indicators were equally dismal. The poverty rate hovered around 70 percent.14 The economy was overwhelmingly dependent on rain-fed agriculture powered only by age-old technology, cows, and ploughs. Agricultural laborers, who comprised the largest group within the labor force (61 percent as per the 1981 census), found employment, on average, for only 122 days per year (mid-1970s) and faced virtually stagnant real wages.15 These rural poor constituted the majority of the nearly one million deaths that occurred during the famine of 1974.16 3. Other indicators were also stark during the early years.17 The population growth rate was 2.6 percent, and life expectancy was 42 years. A woman, on average, bore seven children; and of 1,000 live births, 239 babies did not survive beyond the age of five. With respect to education, the literacy rate was only 24 percent in 1974, with female literacy at an abysmal 15 percent.18 Primary enrollment was at a rate of 60 percent, with female enrollment at 41 percent. Secondary enrollment was less than one-third of primary enrollment at 18 percent, and female secondary enrollment at only 8 percent.19 Furthermore, rural and urban areas were socially, culturally, and infrastructurally isolated from one another, and the urban population constituted less than eight percent of the entire population.20 Rural isolation was so prevalent that a typical journey from the capital city to a village in outlying districts would take several days and involve inconvenient and uncertain methods of transportation. 4. The country’s beginning could hardly have been bleaker. Yet, forty years later, Bangladesh has succeeded in forging an entirely different script, triumphing over the statistics of despair in ways and to an extent rarely found in the text-books of developmental transformation. Bangladesh’s story merits recognition and praise, even if an analyst’s search for explanations remains a work-in-progress. 5. In recent decades, the country has experienced vast improvements across many indicators. As of 2010, poverty was down to 31.5 percent.21 The introduction of irrigation, mechanization, and modern farming practices has fundamentally altered the agricultural landscape. Despite reductions in the quantity 14 S.R. Osmani, Notes on Some Recent Estimates of Rural Poverty in Bangladesh, The Bangladesh Development Studies, Vol XVIII, No. 3, September, 1990 15 Mahabub Hossain, 1984, ‘Agricultural Development in Bangladesh: A Historical Perspective’ in The Bangladesh Development Studies, Vol XII No. 4, December, 1984 p.43 16 M. Alamgir (assisted by Salimullah), Famine, 1974: Political Economy of Mass Starvation in Bangladesh , BIDS, Dhaka, 1977 17 World Bank, Bangladesh: Promoting Higher Growth and Human Development Vol 1, Report No. 6616-BD, 1987 18 Abu Abdullah (ed), ‘Public and Private Social Provisioning’ in Modernization at Bay: Structure and Change in Bangladesh, UPL, Dhaka, 1991, p.72-73. 19 Ibid. 20 Hossain Zillur Rahman, 2012, Urban Bangladesh: Challenges of Transition, PPRC, Dhaka 21 BBS, HIES 2010 of land devoted to farming, food production has more than tripled to 35 million metric tons (MT). The broader economy has also undergone deep structural transformations. Its key manufacturing industry, ready-made garments (RMG), has turned into a $20 billion plus export sector since the late 1970s, when the sector sported a base of just a few million dollars. Services, migration-linked remittances, and manufacturing are among the important factors contributing to current GDP growth rates in excess of six percent, a rate far exceeding the anemic growth rates experienced during the first decade of the country’s independence. 6. Infrastructural transformation has also been fundamental to economic growth. The rural-urban divide has given way to a rural-urban continuum.22 Through a system of feeder roads, many of the most remote villages are now connected to the national road system, thereby facilitating participation in an integrated national economy rather than only isolated local economies. 7. The introduction of safety net programs has also played a critical role in the country’s progress. In 1974, the state was incapable of preventing and dealing with the effects of a devastating famine. Less than twenty-five years later, in 1998, when a large flood engulfed the whole country, the institutional capacity of the state demonstrated substantial progress. Through safety net instruments, such as Vulnerable-Group Feeding (VGF) cards and innovations in health epidemic prevention, such as oral rehydration salts (ORS), the rural economy was able to rebound from the flood within a matter of weeks and with negligible casualties. An extensive system of safety net programs has virtually eliminated post- disaster secondary cycles of death and hunger. 8. At a relatively low level of economic development, achievements with respect to social indicators have also been impressive. The population growth rate has been reduced to 1.1 percent. The average number of children born per woman has declined from 6.9 to 2.6. Life expectancy has risen to 69 years, from the 1971 base of 42 years. Under-five mortality has fallen from the base year rate of 239 to 48 per 1,000 live births. The primary net enrollment ratio has also risen to 77 percent with gender gaps rapidly narrowing. 9. Despite making impressive economic and social improvements in recent decades, Bangladesh still faces many formidable challenges. While great progress has been made toward reducing hunger, malnutrition still persists as an entrenched problem. Enrollment successes are compromised by serious shortfalls in completion rates. Newer vulnerabilities also loom: urban poverty; climate change; economic price shocks. Notwithstanding such factors, the fact remains that this country, whose best prospect was initially deemed to have been mere survival, is today a legitimate contender for middle income aspirations. 10. Over the course of the four decades since independence, significant changes have occurred for many critical segments of society, including farmers, the business community, youth, women, and the broad ranks of the poor themselves. During this time period, a largely illiterate peasantry has embraced modern technology and, in the process, has fundamentally transformed Bangladeshi agriculture. After its devastating experience with the 1974 famine, the State prioritized safety nets through the Food-for-Work program (FFW) and other subsequent programs, including an expansive NGO sector, as key instruments to redress poverty and women’s economic empowerment. During 1979-1981, young entrepreneurs, sensing a unique economic opportunity, transformed the RMG sector into a globally competitive manufacturing industry, which serves as one of the country’s key drivers of growth. Successive governments have also contributed to the country’s incline by increasing emphasis on female education, institutionalizing safety nets, and, more recently, focusing on information technology. 22 Hossain Zillur Rahman (ed), 2012, Bangladesh Urban Dynamics, PPRC, Dhaka. 2 11. Several policy reforms, particularly those starting in the late 1980s, also played an important role in the country’s growth. Focused on trade liberalization, the lifting of regulatory burden on agriculture, banking and telecommunications reforms, and fiscal responsibility, these reforms served to unleash a multitude of private sector initiatives, which sharply accelerated the growth process from the 1990s. The international community, too, played an extraordinarily supportive role through provision of aid and, later, through preferential market access. 12. The purpose of this report is to document some of the aforementioned achievements over the 2000-2010 decade and to illustrate their collective impact on poverty in Bangladesh. Analysis is undertaken to identify which factors contributed to the rapid decline in poverty over time. The main limitation of this report is that the analysis is based on a limited number of data sources, which do not cover all aspects of the poverty reduction process. Nevertheless, to the extent possible, the analysis covers the key drivers of poverty reduction over what has been a remarkable decade for Bangladesh. 13. The report is organized into four parts. Part I focuses on explaining poverty patterns observed over the 2000-2010 period, noting qualitative differences between the first and second half of the decade. The analysis in Chapter 1 offers poverty projections based on survey data from this period. Chapter 2 describes some key characteristics of the poor. Using poverty decomposition methodology, Part II identifies the main drivers of the poverty reduction experienced over the last decade. Chapter 3 shows that the two most important contributors to poverty reduction over the 2000-2010 period were the growth of labor income and the declining dependency ratio. The remaining two chapters in this section focus on labor income and demographic factors to understand their respective linkages to poverty. 14. The past few years have underscored the importance of global factors affecting country-level outcomes. However, the series of shocks that affected Bangladesh in 2007-2008 did not significantly slow down the speed of poverty reduction. In Chapters 6 and 7 of Part III, the report attempts to uncover some of the reasons underlying Bangladesh’s resilience to these global shocks as well as the way in which poor households cope with seasonal shocks, which are a permanent feature of some rural parts of the country, namely Rangpur. Chapters 8 and 9 explore the role of safety nets and microfinance in helping households deal with shocks and poverty. 15. In Part IV, Chapter 10 revisits one of the key findings of the World Bank Poverty Assessment of 2005 (published in 2008). The poverty headcount figures from the first part of the decade revealed that while poverty had decreased in both rural and urban areas, the reduction had been highly unbalanced with respect to region, favoring the Eastern part of the country over the West. The same figures for the latter part of the decade indicate that East-West poverty differences had significantly diminished. This section looks at some of the factors that help to explain the convergence in poverty rates across regions. 16. Poverty has many psychological consequences, such as the deflation of spirit and the abandonment of hope. The analysis in this report shows that Bangladesh has and continues to signal resolve despite persistent problems. Although great strides have been made in redressing poverty, the analysis in the report points to the necessity of even greater strides to end extreme poverty, to enhance shared prosperity, and to promote environmentally sustainable development over the next decade. As in the past, success is likely to come from a coalition of efforts, consisting not just of citizens but also of those who lead and those who formulate and implement policies. The report offers insight into some of these strategic policy decisions. 3 Part I: Poverty Patterns and the Living Conditions of the Poor 1. Poverty, Growth and Inequality 17. Over the last decade, Bangladesh has experienced steady and strong GDP growth, averaging a rate of six percent per year. The Bangladesh Bureau of Statistics (BBS) reports that poverty rates have also demonstrated steady improvement during this period, with an average decline of 1.74 percentage points per year (BBS 2011), a rate of decline that outperforms a majority of countries (Newman et al. 2008). The primary goal of this chapter is to analyze changes in poverty incidence between the last three rounds of the Household Income and Expenditure Survey (HIES), covering the 2000-2010 period. The chapter presents a national and regional (urban/rural) analysis of poverty and consumption trends. Box 1-1: Bangladesh Poverty Measurement As suggested by Ravallion (2001), poverty lines (PLs) in Bangladesh are periodically updated using a price index (as done in 1995/96 and 2000 for both food and non-food PLs, and in 2010 for food PL only) or re-estimated using the cost of basic needs (CBN) method (as done in 1991/92 and 2005). The expectation is that either of the two methods (i.e. price index or CBN method) guarantees constancy in terms of real expenditure (or any measure of welfare), and thus provides a good measure of poverty and price changes over time. Year 200023 2005 2010 Food PL Updated from 1991/92 Re-estimated (CBN) Updated from 2005 Non-food PL Updated from 1991/92 Re-estimated (CBN) Re-estimated (CBN) Under the first method, the food basket’s quantities are fixed and only market price is updated using an appropriate price index. Under the second method, or CBN method, calculation of the poverty line entails estimation of the average level of per-capita expenditure at which individuals are expected to meet their basic food and non-food related needs. The CBN method is implemented in three steps. In the first step, the cost of a fixed food bundle is computed. In the case of Bangladesh, this bundle consists of eleven food items: rice, wheat, pulses, milk, oil, meat, fresh water fish, potato, other vegetables, sugar, and fruits. This food bundle provides the minimal nutritional requirements for a diet corresponding to 2,122 kcal per-day per person. In the second step, two different non-food allowances for non-food consumption are computed: the lower non-food allowance and the upper non-food allowance. The former reflects the median amount spent on non-food items by households whose total consumption is approximately equal to their food-poverty line;24 the latter corresponds to the amount spent on non-food items by households whose food consumption is approximately equal to their food-poverty line. In the third step, the food and non-food allowances are added together. The sum of food and upper non-food allowances constitute the upper poverty line. In 2010, poverty lines were adjusted by: (i) updating the 2005 food poverty lines with food inflation rates, calculated from unit values of the HIES 2005 and HIES 2010 data; and (ii) re-estimating non-food poverty lines using HIES 2010 data to adjust for changes in non-food/food rates (for more details, see Appendix 1: Inflation Annex). The poverty estimates have been made public by the government through the Preliminary Report on Household Income and Expenditure Survey – 2010 (BBS 2011). 23 The 2005 poverty lines were also back-casted to 2000. Please see Appendix 1: Inflation Annex Tables IA-4 to IA-5. 24 The rationale behind this calculation is that the non-food budgets of these households are set to just afford the bare essentials. 6 Table 1.1: Poverty Headcount Rates Poverty Extreme Poverty 2000 2005 2010 2000 2005 2010 National 48.9 40.0 31.5 34.3 25.1 17.6 Urban 35.2 28.4 21.3 19.9 14.6 7.7 Rural 52.3 43.8 35.2 37.9 28.6 21.1 Source: All estimates are CBN based on HIES 2005, updated for 2010, and back-casted for 2000. 2010 update: survey-based food prices and non-food allowance re-estimated using “upper” poverty lines. Official Poverty Lines estimated for HIES (2000, 2005, and 2010). 18. Prior to this poverty assessment exercise, the World Bank played an important role in supporting the BBS in the development of the HIES questionnaire, training its staff, data processing, and analysis. Based on the HIES data, the BBS produced official 2010 poverty estimates for Bangladesh (BBS 2011). The poverty estimates are based on a Cost of Basic Needs (CBN) methodology and were derived by adjusting existing poverty lines to reflect changes in the cost of meeting basic needs, as indicated by the HIES 2010 data (for more details on the CBN methodology, see Box 1-1). 19. This chapter, and most of the report, will utilize the 2010 round of the HIES. The 2010 survey is the third round in a series of national household surveys conducted by the BBS to estimate poverty levels in Bangladesh.25 As temporal comparisons are crucial to understanding how the poverty reduction process has qualitatively changed over time, this chapter also takes advantage of the 2000 and 2005 rounds of the HIES. Section 1 presents poverty, growth, and distributional trends at national-, rural-, and urban-levels and weights Bangladesh’s outcomes against the outcomes of its South Asian counterparts. In Section 2, poverty trends are simulated under alternative growth scenarios. Section 3 takes a closer look at regional trends in poverty reduction, in particular, focusing on growth and inequality trends at the regional level. 1.1 Poverty, growth, and inequality in recent years 20. Table 1.1 shows the Cost of Basic Needs (CBN) upper and lower poverty estimates for Bangladesh based on HIES data from 2000, 2005, and 2010. From 2000 to 2010, Bangladesh experienced a uniform and steady decline in poverty rates. Poverty rates demonstrated impressive and steady improvement during this period, with an average decline of 1.74 percentage points per year. During the 2000-2005 period, the average decline in poverty rates was 1.78 percentage points per year; the analogous decline for the 2005-2010 period was 1.7 percentage points. In 2000, 49 percent of the population was poor; by 2010, this number dropped to 31.5 percent. This reduction in the national-level poverty rate suggests that the series of shocks affecting Bangladesh in 2007-2008 did not significantly slow down the speed of poverty reduction.26 Table 1.2: Percentage Change in Poverty Headcount Rates Poverty Extreme Poverty 2005-2000 2010-2005 2010-2000 2005-2000 2010-2005 2010-2000 National -18% -21% -36% -27% -30% -49% Urban -19% -25% -39% -27% -47% -61% Rural -16% -20% -33% -25% -26% -44% Source: HIES 2000, 2005 and 2005. 25 Prior to 2000, these household surveys, previously known as the Household Expenditure Survey (HES), only collected data on expenditure and date back to 1973-74. 26 Chapter 7 investigates the welfare effects of the 2008 commodity food price shock. 7 Trends in poverty – national, rural, and urban 21. The national poverty headcount rate Figure 1-1: Poverty Trends decreased by 17.4 percentage points over the period from 2000 to 2010. Across urban and Poverty Headcount rural areas, the rate of poverty reduction was comparable; in 2010, 35.2 (21.3) percent of National Urban Rural the rural (urban) population was poor, 52.3 compared to 52.3 (35.2) percent in 2000 48.9 (Table 1.1 and Figure 1-1). While the changes 43.8 in poverty rates represent an outstanding 35.6 40 percent reduction over a ten-year span at the 35.2 35.2 national-level (Table 1.2), rural areas had only 28.4 31.5 attained the decade-old poverty rate of urban areas in 2010. In general, the percentage 21.3 change in poverty headcount rates for the 2000 2005 2010 2000-2010 period was larger in urban areas (39 percent) relative to rural areas (33 Source: HIES 2000, 2005, and 2010. percent), and the gap in the speed of poverty reduction during the 2000-2005 period between rural and urban areas (3 percentage points) widened over the 2005-2010 period (5 percentage points). 22. Extreme poverty continues to be a rural phenomenon.27 In terms of extreme poverty, the national poverty headcount decreased by 16.7 percentage points over the 2000-2010 period. In 2010, 21.1 (7.7) percent of the rural (urban) population was extremely poor, compared to 37.9 (19.9) percent in 2000 (Table 1.1). That is, in 2010, 60 (36) percent of the poor in rural (urban) areas were also extremely poor. Furthermore, the rate of extreme poverty decline was 26 (47) percent in rural (urban) areas between 2005 and 2010, compared to 25 (27) percent between 2000 and 2005. Depth and severity of poverty Table 1.3: Depth and Severity of Poverty 23. The ratio of the depth of poverty to 60 49 headcount (6.5/31.5) in 2010 indicates that, on 40 average, the poor fell nearly 21 percent short 40 32 of the poverty threshold (i.e. the poor consume at a level equal to only 79 percent of 20 13 9 the cost of basic needs). The same ratio was 5 7 3 2 26 percent in 2000 and 23 percent in 2005. At 0 the national-level, the depth of poverty was 2000 2005 2010 nearly halved over the 2000-2010 period Poverty headcount Poverty depth Severity (Table 1.3). This rapid decline in the depth of poverty allowed Bangladesh to attain its Millennium Development Goal (MDG) target Poverty Depth Severity about five years ahead of schedule. (The 2000 2005 2010 2000 2005 2010 depth of poverty had been 16 percent during National 12.8 9 6.5 4.6 2.9 2 the 1990s, and the goal was to reduce this to 8 Urban 9 6.5 4.3 3.3 2.1 1.3 percent by 2015.)28 Rural 13.7 9.8 7.4 4.9 3.1 2.2 Source: HIES 2000, 2005, and 2010. 27 For more details on the urbanization process taking place over the decade, refer to the World Bank (2012). 28 Box 1-2 includes a definition of the three poverty measures presented in Table 1.3. 8 Box 1-2: Poverty Measures: Poverty Headcount, Depth, and Severity The poverty headcount index measures the proportion of the population that is poor. This measure does not indicate how poor the poor are. The poverty depth index (also known as the poverty gap index) measures the extent to which individuals fall below the poverty line (poverty gaps) as a proportion of the poverty line. The sum of these poverty gaps gives the minimum cost of eliminating poverty, if transfers were perfectly targeted. The measure does not reflect changes in inequality among the poor. The severity of poverty index (also known as the poverty gap square index) averages the squares of the poverty gaps relative to the poverty line. This measure is one of the Foster-Greer-Thorbecke (FGT) class of poverty measures that allows varying weights to be placed on the income (or expenditure) level of the poorest members in society. Source: Haughton and Khandker (2009). 24. The decline in poverty depth was larger in urban areas (52 percent) relative to rural areas (46 percent). The difference in poverty depth reduction between urban and rural areas widened over the decade. Like the poverty headcount rate, the difference in the speed of poverty depth reduction between rural and urban areas that existed in the 2000-2005 period (less than 0.5 percent) widened over the 2005- 2010 period (10 percent). A similar pattern is observed for the severity measure. 25. Overall, significant improvements occurred with respect to the incidence of poverty, the severity of poverty, as well as the depth and inequality of poverty among the poor over the last decade. A clear narrative emerges: over the last decade, poverty has continued to decline in both rural and urban areas in Bangladesh. In general, fewer people are below the poverty line, and variation in the severity of poverty among the poor has significantly narrowed, primarily due to decreasing numbers of individuals who are extremely poor. Nevertheless, poverty in rural areas continues to be relatively more pervasive and extreme, and the gap in the speed of poverty reduction between urban and rural areas has, in fact, widened over that last five years. Sensitivity of poverty trends to methods of measurements 26. In Figure 1-2.A, we observe that the distribution of per-capita real expenditure has shifted down and to the right for both the 2000-2005 and 2005-2010 periods. These shifts suggest that real per-capita expenditure has increased for the entire population. According to the cumulative distribution of per-capita real expenditures (Figure 1-2.B), the poverty rate in 2005 is below that of 2000, regardless of how high the poverty line is set. The same is true for the year 2010 relative to both 2005 and 2000.29 Hence, irrespective of the poverty line level, the official poverty estimates indicate that poverty has declined in 2005 relative to 2000 and in 2010 relative to 2005. While these reductions in poverty indicate a positive trend, individuals who are no longer classified as poor may nevertheless be vulnerable to poverty. For instance, the percentage of non-poor people consuming less than 1.5 times the national poverty line was 28 percent (or about 36 million people) in 2000. By 2010, about 35 percent of the population consumed more than the poverty line and less than 1.5 times the national poverty line (52 million non-poor people who consume less than 150 percent of the poverty line). 29 This holds true for the relevant range of the poverty line. In other words, first order stochastic dominance holds for the year 2000 relative to 2005. A similar pattern is observed for 2005 relative to 2010, yet the first order stochastic dominance does not hold at high levels of real per-capita expenditure. 9 Figure 1-2: Distribution of Per-capita Real Expenditure by Survey Year A. Density B. Cumulative Distribution .0015 1 National Poverty Line 1.5 x National Poverty Line (861.6 TK) National Poverty Line 2005 .8 (861.6 TK) .001 .6 1.5 x National Poverty Line Density Density .4 .0005 .2 0 0 0 500 1000 1500 2000 2500 0 500 1000 1500 2000 2500 Real per capita expenditures Real per capita expenditures 2000, =1081TK 2005, =1210TK 2010, =1297TK 2000, =1081TK 2005, =1210TK 2010, =1297TK Note: The vertical lines represent the mean real per-capita expenditure for each survey year (µ). The base is the national poverty line for 2005. Source: HIES 2000, 2005, and 2010. 27. Next, the analysis considers whether the poverty decline is robust to household composition and household size. Under the official poverty headcount, every individual, young or old, in every household, large or small, is assumed to have the same needs. In reality, however, the needs of children are likely to be very different from the needs of adults, and the costs faced by individuals in large households are likely to be lower due to the sharing of public goods. For these reasons, the robustness of the poverty trends must be evaluated under alternative adult equivalents scales30 and economies of scale assumptions. The term adult equivalence (AE) simply refers to the notion that every household member counts as some fraction of an adult member. Accordingly, Figure 1-3: Poverty Headcount Rate under Different Adult the total household consumption is divided Equivalence Assumptions by the AE rather than the number of 60.0% household members.31 50.0% A Poverty headcount 40.0% 28. Figure 1-3 shows poverty headcount 30.0% B trends for four different scenarios. The same 20.0% C poverty reduction trends are observed when 10.0% D accounting for variations in household 0.0% composition and size. In particular, when the 2000 2005 2010 economies of scale parameter is set to 0.9 s=1, t=1 48.9% 40.0% 31.5% (see Figure 1-3, lines C and D), a slight s=.7, t=1 34.9% 26.3% 20.0% decrease in the speed of poverty reduction s=1, t=.9 33.8% 25.4% 18.7% occurs over the 2005-2010 period relative to s=.7, t=.9 21.7% 14.7% 10.6% the 2000-2005 period. This decrease is likely due to the larger decline in average Note: A=Number of adults; C=Number of children; s=expenditure on a child household size that took place during the relative to an adult; t=economies of scale parameter. Source: HIES 2000, 2005, and 2010. second part of the decade.32 30 The adult equivalent (AE) is given by the following formula: AE = (A+sC). For example, assume that the expenditure for every child is, on average, one-half of that of an adult. Then, the adult equivalent of a household that has five members, three of which are children and two of which are adults is: AE = 2 + ½ 3 = 3.5. 31 The AE formula can be modified to account for economies of scale. In particular, the formula becomes: AE = (A+sC)t, where the parameter t 1 captures economies of scale. 32 Household composition fell by 5.8 percent in the 2000-2005 period (from 5.2 in 2000 to 4.9 in 2005) and by 8 percent in the 2005-2010 period (from 4.9 in 2005 to 4.5 in 2010). 10 Nevertheless, the decline in poverty is robust to all four scenarios and for all survey years, suggesting that poverty in Bangladesh has significantly declined over the last decade. Consumption Growth and Distributional Changes 29. This section explores trends in consumption growth and distributional changes taking place over the period from 2000 to 2010. Table 1.4 shows that average real per-capita consumption increased by 20 percent over the last decade, 60 percent of which took place over the first part of the decade. While real per-capita consumption for the year 2010 remained about 26 percent lower in rural areas relative to urban areas, the average annual growth in real per-capita consumption was twice as large in rural areas (2.1 percent) relative to urban areas (0.9 percent) throughout the decade. Table 1.4: Mean Real Per-capita Monthly Consumption Per-capita Consumption Cumulative Change (%) Average Annual Growth (%) 2000 2005 2010 2000-2005 2005-2010 2000-2010 2000-2005 2005-2010 2000-2010 National 1081 1210 1297 11.9% 7.2% 20.0% 2.4% 1.4% 2.0% Urban 1464 1535 1600 4.8% 4.2% 9.3% 1.0% 0.8% 0.9% Rural 985 1103 1190 12.0% 7.8% 20.8% 2.4% 1.6% 2.1% Note: The base is the national poverty line for 2005. Source: HIES 2000, 2005, and 2010. 30. During the 2000-2005 period, the increase in per-capita consumption benefited both the rich and the poor, particularly for those in the upper (the extremely rich) and lower (the extremely poor) tails of the consumption distribution relative to the 40th to 80th percentiles (Figure 1-4). The “pro-poor” growth rate of per-capita consumption over this period (2.27 percent) was virtually equal to the mean growth rate of per-capita consumption (2.28 percent).33 Figure 1-4: Growth Incidence Curve 2000-2005 2005-2010 2000-2010 Growth Incidence Curve - Bangladesh Growth Incidence Curve - Bangladesh Growth Incidence Curve - Bangladesh 3 5 2.5 2 4 2 Growth Rate Growth Rate Growth Rate 1 3 1.5 0 2 1 -1 1 -2 .5 0 20 40 60 80 100 0 20 40 60 80 100 0 20 40 60 80 100 Percentiles Percentiles Percentiles GIC Growth rate in mean 95% CI GIC Growth rate in mean 95% CI GIC Growth rate in mean 95% CI Growth Rate Mean 2.28 1.41 1.84 Median 2.13 1.87 2.00 Percentile 2.22 1.62 1.92 Pro-poor 2.27 1.76 2.01 Note: The base is the national poverty line for 2005. Source: HIES 2000, 2005, and 2010. 33 Here, “pro-poor” is defined as growth that reduces poverty. A more precise definition is provided by Ravallion and Chen (2003): “Pro-poor growth is the ordinary growth rate in the mean scaled up or down the ratio of the actual change in the Watts index to the change implied by distribution-neutral growth”. 11 31. For the 2005-2010 period, growth was relatively more “pro-poor”. In particular, the increase in per-capita consumption was higher than average for those in the 10th to 80th percentiles relative to those in the upper and lower tails of the consumption distribution. Those below the 70th percentile of the per- capita consumption distribution experienced the largest increases in per-capita consumption. The “pro- poor” growth rate of per-capita consumption over the second half of the decade (1.76 percent) was higher than the mean growth rate of per-capita consumption (1.41 percent). The same was true for the “pro-poor” growth rate over the decade (2.01 percent) relative to the mean growth rate (1.84 percent). 32. Figure 1-5 presents the Datt and Ravallion (1992) decomposition of changes in poverty headcount into its growth and redistribution components.34 In the 2000-2005 period, the reduction in the poverty headcount ratio was fully explained by the growth component. Furthermore, the redistribution component had a negative effect on poverty headcount. However, during the second half of the decade, the redistribution component complemented the growth component. This decomposition suggests stark differences in the underlying components of poverty decline between the first and the second halves of the decade. Over the 2000-2010 period, both the growth and redistribution components moved in the same direction, with the former being the predominant driving force for poverty reduction. Figure 1-5: Growth and Redistribution Components of Changes in Poverty 0.00 -0.05 -0.07 -0.10 -0.07 -0.09 -0.09 -0.09 -0.09 -0.15 -0.14 -0.17 -0.17 -0.20 Nation Rural (2000- Urban (2000- Nation Rural (2005- Urban (2005- Nation Rural (2000- Urban (2000- (2000-2005) 2005) 2005) (2005-2010) 2010) 2010) (2000-2010) 2010) 2010) Poverty Reduction -0.09 -0.09 -0.07 -0.09 -0.09 -0.07 -0.17 -0.17 -0.14 Redistribution 0.01 0.02 0.01 -0.02 -0.02 -0.02 -0.02 0.00 -0.01 Growth -0.10 -0.10 -0.08 -0.06 -0.07 -0.05 -0.16 -0.17 -0.13 Note: The results are obtained by taking the average of the two decompositions – with 2000 and 2005 as base years. Source: HIES 2000, 2005, and 2010. 33. In terms of inequality, as measured by the Gini index of real per-capita consumption, Bangladesh experienced a modest decline in inequality at the national-level. According to the Gini index, inequality in rural areas, where more than 70 percent of the population resides, increased in the first part of the decade and then decreased in the latter part of the decade. The net inequality change in rural areas during the 2000-2010 period was slightly positive, indicating an increase in inequality (bottom of Figure 1-6) 34. In urban areas, inequality remained higher relative to rural areas throughout the decade. However, it also trended downwards, as depicted by the changing shape of the Lorenz curves corresponding to urban areas (Figure 1-6), even as the proportion of the urban population increased. Over 34 See Appendix 4. 12 the 2000-2010 decade, the net change in inequality indicates a decrease in inequality in urban areas. Overall, however, inequality remained relatively stable over the decade. Figure 1-6: Lorenz Curve and Gini Coefficient 2000 2005 2010 Lorenz curve Lorenz curve Lorenz curve Bangladesh Bangladesh Bangladesh 1 1 1 .1 .2 .3 .4 .5 .6 .7 .8 .9 .1 .2 .3 .4 .5 .6 .7 .8 .9 .1 .2 .3 .4 .5 .6 .7 .8 .9 0 0 0 0 .1 .2 .3 .4 .5 .6 .7 .8 .9 1 0 .1 .2 .3 .4 .5 .6 .7 .8 .9 1 0 .1 .2 .3 .4 .5 .6 .7 .8 .9 1 Cumulative population proportion Cumulative population proportion Cumulative population proportion Rural Urban Line of perfect equality Rural Urban Line of perfect equality Rural Urban Line of perfect equality Gini National 0.307 0.310 0.299 Rural 0.271 0.278 0.273 Urban 0.368 0.353 0.330 Pop-share Rural 0.799 0.753 0.737 Urban 0.201 0.247 0.263 Source: HIES 2000, 2005, 2010 Relative versus Absolute Inequality 35. The relative measures of inequality, such as the Gini index and the Lorenz curve (see Figure 1- 6), tell us that relative inequality in Bangladesh was stagnant over the last decade. This finding suggests that real per-capita consumption expenditures increased proportionally across different groups. Also of interest is the size of differences in per-capita consumption levels between different groups in a given year. Like the Gini index and the Lorenz curve, Figure 1-7.A shows that the ratio between different percentiles of real per-capita consumption remained virtually unchanged from 2000 to 2010. 36. Given that inequality was large in 2000 and per-capita real expenditure continued to grow, we expect that a proportional increase in real per-capita consumption increases absolute inequality. This notion is depicted in Figure 1-7.B: over the decade, absolute inequality has continued to increase uniformly across different percentiles of the real per-capita consumption distribution. Figure 1-7: Relative and Absolute Differences in Real per-capita Consumption A. Ratio of percentiles of real per-capita B. Absolute differences between percentiles expenditure: 2000, 2005 and 2010 of real per-capita expenditure: 600 2000, 2005 and 2010 2.50 Real Taka (Base 2005) 500 Real Taka (Base 2005) 2.00 400 1.50 300 1.00 200 0.50 100 0.00 0 p90/p10 p90/p50 p75/p25 p75/p50 p50/p25 p50/p10 p90-p10 p90-p50 p75-p25 p75-p50 p50-p25 p50-p10 2000 2.05 1.39 1.47 1.20 1.22 1.47 2000 456 249 246 131 116 206 2005 1.99 1.38 1.45 1.20 1.21 1.44 2005 493 272 266 141 125 220 2010 2.03 1.38 1.47 1.20 1.22 1.47 2010 543 292 295 154 141 251 Source: HIES 2000, 2005, and 2010. 13 1.2 Projecting recent trends in growth, inequality, and poverty into the future 37. Among other important targets, the poverty MDG for Bangladesh stipulates that the proportion of people living in extreme poverty that prevailed in 1990 (57 percent) must be reduced by at least one-half by the year 2015. That is, between 2010 and 2015, Bangladesh must reduce its poverty level by an average of 0.6 percent per annum, equivalent to a cumulative reduction of 3 percentage points over the course of this period. Assuming population growth continues to decline at the same rate as during the 2000-2010 period, achieving the poverty MDG implies lifting over 4.7 million people out of poverty. Given Bangladesh’s performance in poverty reduction over the last two decades, can we expect the proportion of the country’s population living in poverty to be 28.5 percent or less by 2015? In this section, we use data from the last three HIES surveys to estimate Bangladesh's net elasticity of poverty reduction to growth in per-capita expenditure. This elasticity estimate is then used to project the poverty headcount index into the future. The method methodology used for this exercise is the Datt and Ravallion (1992) decomposition method (see Box 1-3). As the net elasticity of poverty is estimated with respect to growth in per-capita expenditure, the GDP growth rates have to be converted into growth in per-capita consumption.35 In particular, we estimate the impact of changes in the income to consumption ratio (income is measured as real per-capita GDP and consumption is measured as the growth in HIES- based per-capita consumption) on poverty.36 Box 1-3: Datt and Ravallion (1992) Growth Decomposition Method The net elasticity of poverty to growth, or the percentage decrease in poverty resulting from a one percent change in growth rate while allowing inequality to vary, is given by: . where is referred to as the direct effect, or growth component, and is referred to as the indirect effect, or distribution component. The direct effect indicates by how much poverty would change as a result of a one percent growth rate and in the absence of changes in the distribution of real per-capita consumption expenditure (i.e. holding inequality constant). The indirect effect captures the interaction between the elasticity of inequality to growth, , and the elasticity of poverty to inequality, holding real consumption growth constant, . The indirect effect measures the change in poverty resulting from a change in inequality while holding growth constant (i.e. holding the mean of real per-capita consumption expenditures constant).37 To obtain the direct and the indirect effects, the Datt and Ravallion (1992) growth decomposition method is used first. Under this method, a hypothetical distribution of real per-capita consumption is generated under the assumption that consumption increases uniformly and at the average growth rate across the population. To obtain the direct and indirect components of poverty reduction, two types of comparisons are made: 35 The estimates from HIES show that real per-capita consumption increased, on average, by 2.4 percent per year between 2000 and 2005, data from World Development Indicators (WDI) indicate that during this period, real per-capita Gross Domestic Product (GDP) increased by about 4.3 percent per year. 36 Our projections use the ratio of the HIES-based annual growth rate of per-capita consumption expenditure to the annual real GDP growth rate for two time periods, 2000-2010 and 2005-2010. 37 If inequality increases with growth ( > 0), some of the impact of growth on poverty will be eliminated due to the associated increase in inequality. 14  First, to obtain the growth (or direct component), the hypothetical distribution is measured against the actual distribution at the base year. Under both the hypothetical and the original distributions, individuals’ relative positions are the same (inequality is held constant).  Next, the hypothetical distribution is measured against the actual distribution at the end of the period. Under both the hypothetical distribution and the actual end of period distribution, individuals’ relative positions change, yet the average real per-capita consumption expenditure level is held constant.  Then, to obtain the indirect component, the percentage change in poverty resulting from distributional changes (i.e. the difference in the poverty headcount ratio under the hypothetical distribution and the actual end of period distribution) is divided by the percentage change in mean real per-capita consumption expenditure. 38. The results of the Datt and Ravallion (1992) decomposition for the 2000-2005, 2005-2010, and 2000-2010 periods are presented in Figure 1-8. Consider Figure 1-8.A and Figure 1-8.B: the areas between the actual per-capita consumption distribution and the hypothetical distribution represent individuals who have moved-up the consumption distribution as a result of growth in real per-capita Figure 1-8: Datt and Ravallion (1992) Growth Decomposition Method 2000-2005 2005-2010 A B .0015 .0015 Growth (direct) Component .001 .001 .0005 .0005 0 0 0 1000 2000 3000 0 1000 2000 3000 real per-capita expenditure real per-capita expenditure Hypothetical Actual Hypothetical Actual C D .0015 .0015 Percentage change in poverty resulting from distributional changes .001 .001 .0005 .0005 0 0 0 500 1000 1500 2000 2500 0 500 1000 1500 2000 2500 real per-capita expenditure real per-capita expenditure Hypothetical Actual Hypothetical Actual Source: HIES 2000, 2005, and 2010. 15 consumption. This area was larger in the 2000-2005 period relative to the 2005-2010 period. On the other hand, when considering Figure 1-8.C and Figure1-8.D, the areas between the actual per-capita consumption and hypothetical distributions represent people who have moved-up the consumption distribution as a result of the redistribution effect, as opposed to growth in consumption. The area between the distributions was larger for the 2005-2010 period relative to the earlier half of the decade. 39. Overall, growth was the driving force for poverty reduction during the first part of the decade (Figure 1-8.A), whereas in the latter part of the decade, redistribution was also an important contributor (Figure 1-8.D). Comparing Figure 1-8.A and B to Figure 1-8.C and D, the overall poverty reduction was mainly the result of growth rather than redistribution, during the 2000-2010 period. The parameter estimates corresponding to these decompositions are presented in Table 1.5 below. Datt and Ravallion (1992) Growth Decomposition Method 40. Gross elasticity of poverty to consumption growth ( ): For the 2000-2005 Table 1.5: Growth Elasticity Estimates – Datt and period, without changes in inequality (as Ravallion (1992) Method measured by the Gini index), a one percent increase in per-capita real expenditure results in Time Period a 1.89 percent decline in the headcount index of Parameter 2000-2005 2005-2010 2000-2010 poverty (Table 1.5). At a base-year national -1.89 -1.30 -1.55 poverty headcount of 48.9 percent, this reduction 0.05 -0.27 -0.10 implies an outstanding 0.92 percentage point -1.84 -1.58 -1.64 decline per annum in the poverty headcount Source: HIES 2000, 2005, and 2010. (48.9 –1.89/100 = –0.92). For the 2005-2010 period, the estimated implies that a one percent increase in per capita real expenditure yields a more modest –1.30 percent decline in the headcount index of poverty. This reduction implies a 0.52 percentage point decline per annum at the base-year national poverty headcount of 40 percent (40 –1.30/100 = –0.52). Finally, the average gross elasticity for the decade is –1.55, which translates into a 0.76 percentage point decline per annum in the poverty headcount (48.9 –1.55/100= –0.76). 41. The elasticities of poverty to inequality and inequality to growth ( ): For the 2000-2005 period, the impact of redistribution, or the indirect effect, is an increase in poverty. A one percent increase in per-capita real expenditure implies a 0.05 percent increase in the headcount index of poverty, which translates to a 0.02 percentage point increase per annum at a base-year national poverty headcount of 48.9 percent (48.9 0.05/100 = 0.02). For the 2005-2010 period, the analogous effect implies that a one percent increase in per-capita real expenditure results in a 0.27 percent decline in the headcount index of poverty; or, at a base-year national poverty headcount of 40 percent, a 0.11 percentage point reduction per annum in the poverty headcount (40 –0.27/100= –0.11). Finally, the average indirect effect for the decade is –0.10, which translates into a 0.05 percentage point decline per annum in the poverty headcount (48.9 –0.10/100 = –0.05). 42. The net elasticity of poverty to growth ( ): For the 2000-2005 period, the estimated net impact of growth on poverty ( ) is –1.84. Given the base-year poverty headcount of 48.9 percent, a one percent increase in real per-capita consumption results in a 0.90 percentage point decline in the headcount index of poverty (48.9 –1.84/100 = –0.90). For the 2005-2010 period, the estimated net impact of growth on poverty is –1.58. At a base-year poverty headcount of 40 percent, a one percent increase in real per-capita consumption yields a 0.63 percentage point reduction in the headcount index of poverty (40 –1.58/100 = –0.63). Over the entire period, the average net elasticity of poverty to growth is –1.64. Taking 2000 as 16 the base year, this implies a 0.80 percentage point decline per annum in the headcount index of poverty (48.9 –1.64/100 = –0.80). 43. Alternatively, the net elasticity of poverty to growth (λ) can be estimated using the regression method. Under this method, the Table 1.6: Growth Elasticity gross elasticity of poverty to consumption growth is obtained by Estimates – Regression Method regressing the growth rate of poverty on the growth rates of real per- Time Period capita consumption (the corresponding parameter is γ); and the Parame 2000- 2000-2010 elasticity of poverty to inequality is obtained by regressing the ter 2005 -2.06 -2.50 poverty growth rate on the growth rate of the Gini coefficient of inequality (the corresponding parameter is β). Similarly, the 0.61 0.65 elasticity of inequality to growth is obtained by regressing the -1.46 -1.85 growth rate of the Gini coefficient of inequality on the growth rates Source: HIES 2000, 2005, and 2010. of real per-capita consumption (the corresponding parameter is δ). Parameter estimates using the regression method are presented in Table 1.6. 44. Using the 2005 Table 1.7: Predicted versus Actual Poverty Estimates for 2010 poverty headcount as the Datt and Ravallion (1990) Regression Method base, poverty headcount projections for 2010 are Data from 2000-2005 2000-2010 2000-2005 2000-2010 1 generated under four Predicted 30.4 31.4 32.2 30.4 different scenarios. These Actual 31.5 31.5 31.5 31.5 projections are presented in Table 1.7. Overall, the Difference 1.1 0.1 -0.7 1.1 projections obtained from Note: 1Prediction for the year 2010 using poverty headcount from 2005 as the baseline. the application of the Datt Source: HIES 2000, 2005, and 2010. and Ravallion (1992) method to the 2000-2010 HIES data perform better than projections from the alternative scenarios. Therefore, this is the preferred method for projecting the 2015 poverty estimates for Bangladesh. 45. Five year poverty estimates are projected by applying the elasticity of poverty to growth, estimated using both the preferred method (i.e. Datt and Ravallion 1992) and the regression method, to the baseline poverty level of 2010 (31.5). Six alternate Table 1.8: Poverty Headcount Projections scenarios are considered. The first four scenarios RM DR RM DR DR DR correspond to the parameters HIES period (parameters) 2000- 2000- 2000- 2000- 2000- 2000- presented in Table 1.5 and 2005 2005 2010 2010 2010 2010 1 Table 1.6 and are applied to Assumed GPD Growth 5.8 5.8 5.8 5.8 4.8 3.8 the ratio of average real GDP Net elasticity -1.46 -1.84 -1.85 -1.64 -1.64 -1.64 growth to the HIES-implied 2010 31.50 31.50 31.50 31.50 31.50 31.50 average real per-capita 2011 30.05 29.67 29.67 29.87 30.15 30.43 consumption growth. The remaining two scenarios 2012 28.67 27.95 27.94 28.32 28.86 29.40 correspond to the elasticity 2013 27.35 26.33 26.31 26.86 27.62 28.40 parameters presented in the 2014 26.10 24.80 24.78 25.47 26.44 27.44 last column of Table 1.5 2015 24.90 23.36 23.34 24.15 25.31 26.51 (obtained using the Datt and Target – 2015 Estimate 3.60 5.14 5.16 4.35 3.19 1.99 Ravallion (1992) method Note: 1Estimates use the real GDP growth over Per-capita real expenditure growth. applied to the HIES data for Source: HIES 2000, 2005, and 2010. RM = Regression method. DR = Datt and Ravallion. the 2000-2010 period), which 17 are applied to the income/consumption ratio, assuming more pessimistic real GDP growth scenarios of 4.8 percent and 3.8 percent, respectively. Estimates for each scenario are presented in Table1.8. The projected figures suggest Bangladesh will achieve its poverty MDG goal of halving the 1990 poverty rate at some point before the end of 2013. Under all scenarios, the 2015 poverty headcount is below the MDG target of 28.5 for 2015. Even under the most pessimistic scenario, assuming a 3.8 percent GDP growth rate, the poverty headcount projection still overshoots the MDG target by two percentage points. Growth in Per-capita Gross Domestic Product and Demographics Trends 46. Over the last decade, Figure 1-9: Annual Growth of Real Gross Domestic Product, Bangladesh experienced steady Population and strong GDP growth, averaging a rate of 5.8 percent per year (Figure 1-9). Average 7 Annual Percentage Growth GDP growth was higher during 6 the second part of the decade (6.2 5 percent) relative to the first part 4 of the decade (5.5 percent). This 3 pattern also persisted for per- 2 capita real GDP growth, which 1 averaged 4.3 percent over the 0 decade, 3.8 percent in the first 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 part of the decade, and 5 percent in the later part of the decade. Real GDP growth Population growth Population growth in largest city Per-capita real GDP growth 47. The rising trends in average and per-capita GDP growth coincided with steady Source: World Development Indicators 2011. declines in population growth, which averaged 1.4 percent over the decade, 1.7 percent in the first part of the decade, and 1.1 percent in the later part of the decade. Additionally, we observe that the gap between the growth rates of per-capita real GDP and real GDP narrows; the gap was 1.8 percentage points at the beginning of the decade and 1.2 percentage points at the end of the decade (Figure 1-9). 48. In terms of changes in the sectoral composition of GDP over Figure 1-10: GDP Sectoral Decomposition the decade, agriculture shrank as a share of GDP, whereas the industry 50 (mainly manufacturing) and service Percent of GDP 40 sectors grew as a share of GDP (Figure 1-10). These compositional 30 changes suggest that the Bangladeshi 20 economy is continuing its transition from a predominantly agriculture- 10 00 01 02 03 04 05 06 07 08 09 10 based economy into a manufacture- Year (2000-2010 period) based economy. The changes also Agriculture Manufacturing hint at occupational changes within Industry (Manuf.& Other Ind.) Services the labor force as potentially important factors contributing to the growth and poverty reduction Source: World Development Indicators 2011. observed over the course of the decade. 18 Figure 1-11: Demographic Changes and the Labor Force Average household size Shares of Adults and Occupied Employment as a share of adults (%) working age population 68.4 63.4 5.2 48.4 50.1 82.9 81.6 82.3 4.5 11.1 12.1 14.8 Average household size Share of adults per Occupied adults (as a share household of number of adults) Female Male 2000 2010 2000 2010 2000 2005 2010 Source: HIES 2000, 2005, and 2010. 49. Given the deceleration in the rate of population growth, understanding its demographic and economic implications is important. First, we observe a significant decline in average household size between 2000 and 2010. Second, the deceleration in the rate of population growth has been substantial enough to already start bearing expected outcomes; for example, the youth bulge observed in earlier periods has now reached working age. Third, we observe a steady increase in labor force participation for women, who have also significantly benefited from growth of the salaried manufacture-based economy as well as increased access to education (see Figure 1-11). The net effect of these demographic changes also appears to be a significant driver for the observed poverty reduction in Bangladesh over the decade. Chapter 3 explores the relationship between demographic changes and poverty reduction, and Chapter 5 investigates the demographic changes taking place over the decade in greater detail. Growth and Poverty Reduction Relative to South Asia 50. In spite of two devastating natural disasters that occurred in 2007 and the subsequent food price shock of 2008, Bangladesh’s real GDP growth was greater in the 2005-2010 period relative to the 2000- 2005 period. Furthermore, the country’s real GDP growth was also stable relative to many of its South Asian counterparts (Figure 1-12, top panel). Nevertheless, Bangladesh’s five-year cumulative rate of GDP growth fell below the South Asian average growth rate – a rate which is primarily driven by India’s growth pace. By the end of the decade, with the exception of Nepal, Bangladesh’s per-capita GDP (in constant 2000 US$) was the lowest in the region (Figure 1-12, second panel). This relatively low real GDP growth partially explains why Bangladesh also continues to have the highest poverty rate in the region. 51. The decline in the population growth rate that characterized the South Asian region, particularly during the latter part of the 2000-2010 period, was relatively larger in Bangladesh (Figure 1-12, third panel). With the exception of Sri Lanka, the population growth rate in Bangladesh was also the lowest in the region. Associated with the steep decline in population growth, we also observe a relatively steep reduction in the poverty headcount (Figure 1-12, fourth panel). Hence, while Bangladesh’s high poverty levels must be underscored, both the HIES and the national accounts suggest that, unlike the 1991-2000 period, growth in Bangladesh over the last decade was characterized by two positive trends: a high speed of poverty reduction and a stabilized level of inequality relative to other countries in the region. 19 Figure 1-12: South Asia Trends Annual % GDP growth (constant 40.00% 35.00% Bangladesh 30.00% Nepal 25.00% 20.00% India US$) 15.00% Pakistan 10.00% Sri Lanka 5.00% South Asia 0.00% 2000 2005 2010 1400 1200 Bangladesh (constant 2000 US$) GDP per-capita 1000 Nepal 800 India 600 Pakistan Sri Lanka 400 South Asia 200 2000 2005 2010 2.5 2.3 2.1 Bangladesh Population growth 1.9 Nepal (annual %) 1.7 India 1.5 1.3 Pakistan 1.1 Sri Lanka 0.9 South Asia 0.7 2000 2005 2010 Poverty headcount ratio at national 50 45 40 Bangladesh 35 poverty line Nepal 30 India 25 Pakistan 20 15 Sri Lanka 10 2000 2002 2004 2005 2006 2007 2010 Source: World Development Indicators 2011. 20 Summary and Roadmap 52. Poverty estimates based on the 2010 HIES show that the proportion of poor has substantially declined over the period from 2000 to 2010. As of 2010, poverty headcount rates, based on both upper and lower poverty lines estimated using the Cost of Basic Needs (CBN) method, indicate that the proportions of poor and extremely poor are 31.5 percent and 17.6 percent, respectively. Over the 2000 to 2010 period, the rate of decline in poverty has been consistently around 1.8 percentage points per year. The percentage decline in poverty was higher in urban areas (25 percent) than in rural areas (20 percent). With respect to extreme poverty, the decline is especially impressive in urban areas, where extreme poverty is down to a single-digit figure of eight percent. 53. In general, fewer Bangladeshis are below the poverty line, and variation in the severity of poverty among the poor has significantly narrowed, primarily due to decreasing numbers of individuals who are extremely poor. At the national-level, the depth of poverty was reduced by nearly one-half over the 2000- 2010 period, allowing Bangladesh to attain its MDG target of halving the depth of poverty from 16 percent to 8 percent at least five years earlier than expected. Poverty projections based on the last three HIES surveys suggest that Bangladesh will achieve its MDG goal of halving its poverty headcount to 28.5 percent sometime before the publication of this poverty assessment. While these trends are encouraging, poverty in rural areas continues to be relatively more pervasive and extreme, and the gap in the speed of poverty reduction between urban and rural areas has, in fact, widened over that last five years. 54. This chapter shows that, in the 2000-2010 period, growth rather than redistribution served as the main driver of poverty reduction. Nevertheless, redistribution was also an important contributor to poverty reduction during the second part of the decade. Analysis of Bangladeshi’s expenditure patterns partially explains this distinction between the two five-year periods. In the first part of the decade, growth favored those at the tails of the real per-capita expenditure distribution (i.e., the poorest and the affluent) more than those at the center (or middle class). In the second part of the decade, this trend reversed; in particular, growth benefited those above the 15th and below the 80th percentiles of the distribution. The remaining chapters of this Poverty Assessment report will explore candidate hypotheses to explain this dramatic change in poverty patterns that took place in Bangladesh in the 2000-2010 period. 55. The rest of this report is divided as follows. The remainder of Part I provides a detailed account of the changes in characteristics and living conditions of poor households over the decade. Part II decomposes the distributional changes in income occurring over the 2000-2010 period into the various components driving these changes in order to answer the following questions: What are the factors behind the observed poverty and distributional changes? Can some of the poverty reduction be attributed to higher employment, higher productivity, or higher remittances and transfers? Was the reduction in poverty linked to changes in the sectoral composition of employment, changes in the characteristics of human capital, or simply higher returns to those characteristics? Part III explores Bangladesh’s vulnerabilities, focusing on both seasonal deprivation and the welfare effects of commodity price shocks. This section also examines the extent to which safety nets and microfinance have helped the poor to improve their well-being and to cope with negative shocks. Part IV re-examines the East-West divide, which was one of the main findings from the previous Poverty Assessment report published in 2008 (World Bank 2008a). The analysis contrasts regional patterns during the first and the second part of the decade. 21 2. Profiling the Poor: Characteristics of the Poor and Determinants of Poverty 1. Since its independence, the 2000-2010 decade is perhaps the most remarkable period in Bangladesh’s history with respect to growth and poverty reduction. Chapter 1 shows that while growth, rather than redistribution, was the main driver of poverty reduction over the decade, redistribution was nevertheless an important driver in second part of the decade. This chapter provides a detailed account of the characteristics and living conditions of poor households as well as how these factors have changed over the decade. Using data from all three rounds of the HIES, the chapter profiles poor households by analyzing the dynamics of a series of indicators measuring alternative dimensions of well-being, including asset ownership, demographic composition, educational attainment, nutrition, health, and the labor market. 2.1. Welfare Indicators 2. This section profiles households’ welfare changes by analyzing a range of characteristics associated with living conditions. Since higher income and greater consumption are typically correlated with better living conditions, observing similar improvement patterns for the alternative welfare dimensions provides support for the findings discussed in the preceding chapter. 3. Table 2.1 presents basic asset and amenity indicators for each HIES survey year, 2000, 2005, and 2010. All non-consumption indicators of welfare show significant improvements between 2000 and 2010 for both the general population and the poor. In the first part of the decade, housing conditions improved dramatically; particularly significant were the improvements in the proportion of households having walls and roofs made of corrugated iron sheets and cement as well as the proportion of all households using sanitary latrine facilities. Improvements for roofs and safe latrine facilities were more substantial among the poorest households (i.e. bottom deciles). In the second part of the decade, the largest improvements occurred with respect to amenities owned; in particular, improvements in the proportion of all households having access to a phone line, electricity services as well as – with the exception of the poorest households - owning a television set were most significant. Moreover, poorer households also continued to experience large improvements in the quality of walls of their homes. 4. Figure 2-1.A illustrates the improvements in asset indicators between 2000 and 2010 along seven key dimensions of welfare: livestock ownership, type of wall and roof of dwelling, safe latrine access, electricity access, TV, and phone ownership. With the exception of livestock ownership changes between 2005 and 2010, the boundaries of the hexagon for 2010 lie outside those corresponding to 2005 and 2000, indicating improvements along the six other dimensions. Similar to the entire population, households in the bottom 3 deciles of the real per-capita consumption distribution experienced relatively large Table 2.1: Trends in Basic Assets and Amenities All households Bottom 5 deciles Bottom 3 deciles 2000 2005 2010 2000 2005 2010 2000 2005 2010 Livestock ownership (%) 35.2 40.3 39.8 33.6 43.3 41.7 31.6 42.5 42.9 Wall of dwelling* 37.7 55.2 63.6 21.4 39.5 53.1 17.4 33.9 47.5 Roof of dwelling* 76.4 89.9 91.9 68.1 84.2 88.4 64.5 81.6 86.4 Safe latrine use (%) 52.0 69.3 75.1 35.2 55.6 64.7 29.4 50.0 59.1 Electricity connection (%) 31.2 44.2 55.2 14.6 25.4 39.1 10.0 20.2 28.5 TV ownership (%) 15.8 26.5 35.8 3.6 10.1 18.1 1.8 6.7 10.8 Phone ownership (%) 1.5 12.2 63.9 0.1 1.5 46.2 0.0 0.9 36.3 Source: HIES 2000, 2005, and 2010. * Percent with cement / CI sheet. 22 improvements in asset ownership along the same six dimensions of welfare between 2000 and 2005 as well as between 2005 and 2010 (Figure 2-1.B) (the bottom 3 deciles of the real per-capita consumption distribution roughly represent the extremely poor in 2000, when the poverty rate was 49 percent). Particularly significant was the increase in the proportion of households owning a cell phone. While during the first part of the decade, only the more affluent benefited from the expansion of mobile phone networks, as the cell phone industry penetrated most of the country during the second part of the decade, the poor also began to enjoy the benefits associated with access to cell phones (see Box 2-1). 5. Figure 2-1.C displays percentage point changes in each welfare measure over the relevant period for all households. The cumulative change for the decade is given by the green bars, which equals the sum of the blue and red bars. Similar to trends for real per-capita consumption over the period, the majority of improvements for nearly all measures occurred during the first part of the decade. A notable exception is phone ownership (Table 2.1), which also experienced the most significant improvement over the course of the decade. Other categories that experienced relatively large were electricity connections and walls of dwellings. Figure 2-1: Improvements in Households’ Assets Ownership: 2000-2010 A. All Households B. Bottom 3 Deciles Livestock Livestock ownership ownership (%) (%) 100 100 Wall of Wall of Phone Phone dwelling (% dwelling ownership ownership with cement / (% with (%) 50 CI sheet) (%) 50 cement /… TV 0 Roof of TV 0 Roof of dwelling (% dwelling ownership ownership with cement / (% with (%) (%) CI sheet) cement /… Electricity Electricity Safe latrine Safe latrine connection connection use (%) use (%) (%) (%) 2010 2005 2000 2010 2005 2000 C. Trends in Assets Ownership and Basic Amenities – All households (% point changes) 70 60 2000-2005 2005-2010 2000-2010 50 40 30 20 10 0 -10 with cement / with cement / connection (%) Real Per Capita Safe latrine use ownership (%) ownership (%) TV ownership dwelling (% dwelling (% Consumption CI sheet) CI sheet) Electricity Livestock Roof of Wall of Phone (%) (%) Source: Source: HIES 2000, HIES 2005, and 2000, 2010 2005, . and 2010. 23 8os .Z...l: lleld Jdeu, a ...... ..._ ~to mobile 1ilellllolo&Y improvw IIIII flml:lioaiq ofm,.,... 1D11 IIIII ...U-'beiD& of ill playm. A~ IIU4y b)' JllPlU llDdJ dial IIIII majority of BenaJed""d .._,. Ule mobile ,.._ to CCIIIIIeet wllh .,_. and tilden (Minllm ct 11.2012), IUI4 more 111m 40 pnc11t ofdac filmumllllllkll a;n.in .-l.ll1lllplllalloverllu: pblml: (ICC allo Blab 2011). Similarly, Abr (2008) :lladl tbiC pia 1l'l4m equ.ippecl witllmobile lieciiDoloC)' bid 'bellllr _ , 1111111 IIIIWillmwl time, wbUlb ~ dum to mm Z9 pcnlCIIl hisbm piUfila 111m lndcn with 110 mobile - . Apart from &dl~ tnmrerti-, saviD8 time b)' Rducing tnnd clilbiJKlcl,. IDII _ . . cmploymcat Oft'OIIII!Ii•,II!Obi'- tlecluJolov IIIIo baa politMo impacll OD od!lr "im•lioiJI of'ftll-beiiJc, IUdlu cbild beallb, ......,;OJ!,-~ cqualily (Wadd Benlr.2012d). ,_.t Table 2.lllllowl a NOtllt, 1118Uift i.Ju:leMe iD ceDplJoJM 4>"1'HvilJ 1111111111 tiM poor. ln. 2005, about oae of tlu: paar lwl. cell piJmal, 11114 by 2010, - dum -thiJd of~ in the boaum 30 pcnlCIIl of the popubdiaB W P-- Mobile plgN I!OC oaly cnae . , 'buaiuu oppoJtuailiet 1bl IIIII paar, IIIey IIIIo bliDa - to ldlrmatloa, m..._, hlallb, and other~ t o - dJe moll~ ft!Jil8IUI ofB'1118'""""h PW:iq modln cell p11ota in a. ..-or- fium a. paarm~ bouleboldll in nmDk: , . . _ bu 111c: poladiiJI to c:n:M: wm. wiD 'bala. llitnatiooo tbiC I!OC oalJ ....... iJ»u lllll1iq ICiivi- 8lld pro1ill but allo improft tlu: .-lilJ of li& fiB' .upardo:ipmta J.Z Ltmd Ownenhlp ud Poverty 6. Being highly c:amJamd. with poverty atatua, land ownmhip is often ll8ed u a tuptiDg criterion Car IIDii-plJVCrtf pro~ iD Bmgl•desh. Tabk 2.2 p1 ="'poverty 1Jalda by llllld owmuhip iD ruD1 arees. For all tbree ll11rWY years, poverty rates were Dt:gltivdy Mlated to the aize ofbmdholdings.ln other wurdl, ~povczty 111ta ate 1181MXlia&cd wilh lllllll1er laudhohlillp, aud W:e vena. 25 Table 2.2: Trends of Poverty by Land Ownership in Rural Areas Poverty Rate Population Distribution Land size 2000 2005 2010 2000 2005 2010 Landless <0.05 acre 63.5 56.8 45.6 48.0 45.8 50.9 Functionally landless 0.05-0.5 acre 59.7 48.8 34.6 13.0 15.9 15.9 Marginal 0.5-1.5 acres 47.2 35.1 25.0 17.5 18.8 18.0 Small 1.5-2.5 acres 35.4 23.7 16.8 9.2 8.8 6.8 Medium/large: 2.5 acres or more 20.7 12.8 9.7 12.4 10.7 8.4 Source: HIES 2000, 2005, and 2010. 7. Moreover, in the 2000-2005 period, the absolute decline in poverty in rural areas was a linear function of land ownership size (Figure 2-2, blue dotted line). Interestingly, with the exception of the landless, the latter relationship was somewhat reversed in the later part of the decade (Figure 2-2, green long-dashed line). Like the Datt and Ravallion (1992) decomposition presented in Chapter 1 (see Figure 1-5), as well as the trends in basic amenities over the decade discussed above (see Figure 2-1.C), Figure 2-2 suggests that the relationship between poverty reduction and land ownership changed between the first and the second part of the decade. In particular, the figure reveals that, while poverty declines were larger among households with relatively larger landholdings (i.e. more than 1.5 acres of land) during the 2000-2005 period, the 2005-2010 period brought larger poverty declines to households having relatively smaller landholdings (i.e. less than 2.5 acres of land). 8. Overall, the pattern for the 2005-2010 period is consistent with the redistributive nature of poverty reduction that characterized the second half of the decade, as discussed in Chapter 1. Because the landless and small land owners are generally the most deprived households, this pattern is also consistent with the finding that the increase in demand for unskilled workers, coupled with the increase in rural wages, were important poverty reducing factors over the course of the decade (see Chapters 4 and 7). Figure 2-2: Percentage Change in Poverty Headcount (2005-2010) by Land Ownership 0 Landless <0.05 acre Functionally Marginal 0.5-1.5 Small 1.5-2.5 acres Medium/large: 2.5 -0.05 Percentage change in poverty landless 0.05-0.5 acres acres or more acre headcount 2005-2010 -0.1 -0.11 -0.15 -0.18 -0.2 -0.20 -0.25 -0.24 -0.26 -0.3 -0.29 -0.29 -0.29 -0.33 -0.35 -0.38 -0.4 2000-05 2005-10 Note: Data corresponds to rural area household only. Source: HIES 2000, 2005, and 2010. 25 Table 2.3: Demographic Characteristics of Households All households Poor households Non-poor households Demographics 2000 2005 2010 2000 2005 2010 2000 2005 2010 Household Size 5.18 4.85 4.50 5.4 5.2 4.97 5.0 4.6 4.31 Dependency Ratio 0.77 0.69 0.65 0.99 0.91 0.88 0.60 0.57 0.57 Number of children 2.1 1.8 1.6 2.5 2.3 2.1 1.6 1.5 1.3 Number of Male Adults 1.6 1.5 1.4 1.4 1.4 1.4 1.7 1.6 1.5 Number of Female Adults 1.5 1.5 1.5 1.5 1.5 1.5 1.6 1.6 1.5 Head female 0.09 0.10 0.14 0.08 0.08 0.12 0.09 0.12 0.15 Age of head (years) 44.5 45.3 45.7 43.2 43.5 43.9 45.6 46.4 46.5 Source: HIES 2000, 2005, and 2010 2.2. Demographic Composition of Poor Households 9. Table 2.3 shows that poor households are larger in size and have higher dependency ratios. Similarly, higher dependency ratios are attributable to poor households having more children and fewer adults. Consistent with these notions, Chapter 3 demonstrates that the aggregate change in the demographic composition of households accounted for more than 25 percent of the overall poverty reduction occurring between 2000 and 2010 (see also Chapter 5). Apart from age/gender composition, poor households are also less likely to be headed by a female, and household heads are nearly three years younger, on average, than heads of non-poor households. The age difference between poor and non-poor household heads is consistent with the idea that the poor, compared to the non-poor, are more likely to marry and give birth to their first child at a younger age.38 10. Much of the age gap observed between poor and non-poor household heads is reconciled by the idea that the intermingling of financial struggles and cultural norms forces parents to marry their young daughters (Burket et al. 2006). However, early marriage is not only associated with poverty, it is also linked to a large array of poor welfare outcomes and to the intergenerational transmission of such outcomes. Therefore, providing vulnerable children with alternatives to early marriage is imperative and warrants attention. 2.3. Poverty and Education 11. Decomposition analysis (Chapter 3) shows that a more educated and experienced population helped to reduce poverty, particularly in the non-farm sector. Yet, improvements in the educational structure of the workforce accounted for only five percent of the observed poverty reduction over the 2000-2010 period (Figure 3-10 and Table A3-10). Moreover, most of the education-related poverty reduction occurred in the non-farm sector, having a negligible impact on farm workers, during the early part of the decade (Table A3-7 and A3-8). The finding that education explains a relatively small portion of the observed poverty reduction is surprising. To help to reconcile this finding, this section takes a closer look at trends in educational outcomes for the 2000-2010 period. 38 According to NIPORT (2009), in 2007, the median age at first marriage among women 20-49 years old in the highest wealth quintile was 19.7, whereas girls in the lowest wealth quintile married at a median age of 17.3, a difference of more than two years. Similarly, 42 percent of adolescents in the lowest wealth quintile had begun childbearing, compared to only 20 percent of adolescents in the highest wealth quintile. 26 Poverty and the educational attainment of household heads 12. Panel A of Table 2.4 Table 2.4: Education of Household Head and Poverty shows poverty rates by A. Poverty Rate B. Population Distribution education level for household 2000 2005 2010 2000 2005 2010 heads. A significant pattern No Education 63.2 54.7 42.8 57.3 53.5 52.6 emerges from this table. First, Primary 40.3 35.1 29.7 15.4 15.5 16.0 between 2000 and 2005, as well Secondary 30 21.4 17.3 19.9 22.1 22.1 as between 2005 and 2010, the Higher Secondary 8.8 8.5 7.2 5.9 3.6 3.7 poverty rate for households Graduate and above 3.1 4.3 3.1 1.6 5.3 5.6 headed by someone with Source: HIES 2000, 2005, and 2010. primary education or less significantly declined. Additionally, while a household head with no education is highly correlated with poverty (unsurprisingly, given that 72 percent of poor households are headed by an individual with no education), households headed by an individual with no education (or less than primary education) experienced the largest reduction in poverty over the decade. 13. These patterns are consistent with findings from both the decomposition analysis of Chapter 3 as well as Jacoby and Dasgupta (2012). In particular, while our decomposition analysis points to an increase in the relative price of unskilled labor39, analysis of the impact of the 2008 food price spike provides a credible explanation for the relative price increase: higher commodity food prices permeated the economy by increasing the wages of agricultural workers (Chapter 7). In other words, the exogenous price increase created by the shock, and not necessarily increased labor productivity or higher education, caused agricultural workers to experience an increase in wages. Another striking fact is that the distribution of educational attainment of household heads has not significantly changed over the decade. In fact, it remained practically unchanged between 2005 and 2010. This fact warrants the question, how does the education of the poor compare to that of the entire population? 14. Since 2005, both gross and net enrollment rates rose across all education levels (Figure 2-3). Although it did not significantly change in the 2000-2005 period, the gross primary enrollment rate saw a substantial rise in the 2005-2010 period. According to HIES 2010, primary gross enrollment in Figure 2-3: Gross and Net Enrollment Rates Gross Enrollment Rates by Education Net Enrollment Rates by Education Level Level in Bangladesh (2000-10) in Bangladesh (2000-10) 2000 2005 2010 2000 2005 2010 120.0 101.3 100.0 91.1 76.9 100.0 80.0 65.4 80.0 62.6 60.0 49.8 52.1 44.8 37.6 60.0 33.7 40.0 40.0 19.8 20.0 10.7 5.8 8.6 20.0 6.5 9.8 0.0 0.0 Primary Secondary Higher Tertiary Primary Secondary Higher Tertiary Secondary Secondary Source: HIES 2005 and 2010. 39 This has also been described as an increase in returns to endowments or characteristics rather than changes in these endowments (Chapter 4). 27 Bangladesh stands at 101 percent, secondary gross enrollment at 63 percent, and higher secondary enrollment at 45 percent. Between 2005 and 2010, the attrition rate also substantially improved for all education levels, particularly for higher secondary education, which saw the gross enrollment rate nearly double. Net enrollment follows a similar pattern, suggesting that more people are attending school at the appropriate age. 15. In both 2005 and 2010, for the poor and non-poor alike, the female enrollment rate was higher at all education levels (primary, secondary, and higher secondary) than the male enrollment rate (Table A2- 1). Moreover, irrespective of gender, primary enrollment was higher in rural areas relative to urban areas. At higher education levels (secondary and higher secondary levels), however, this regional pattern was reversed. Furthermore, as seen in Table A2-1, the gross enrollment rate increases with the level of consumption of households, and despite a decline in the gross enrollment rate among wealthier households, the gap in the rate between the rich and poor widens with years of education. For example, in 2010, the difference in the gross enrollment rates between the poorest and richest quintiles was 10 percentage points for primary, 45 percentage points for secondary and 70 percentage points for higher secondary. 16. Between 2005 and 2010, Table 2.5: Years of Completed Education average years of education for individuals aged 16 to 40 increased Age 2005 2010 by about half a year (from 5.06 to Category 5.5 years) while completion rates Both Male Female Both Male Female improved for all cohorts (Table 16-20 6.0 5.9 6.2 6.6 6.4 6.8 2.5). The improvement in years of 21-25 5.6 6.2 5.2 6.0 6.2 5.9 educational attainment among the 26-30 4.6 5.4 3.9 5.1 5.5 4.7 adult population is consistent with 31-35 3.8 4.5 3.1 4.2 4.8 3.7 the increase in education that took 36-40 3.4 4.2 2.7 3.7 4.3 3.1 place primarily over the latter half Poor 2.6 3.0 2.3 3.0 3.2 2.8 of the decade. On average, the Non-poor 6.1 6.6 5.7 6.2 6.5 5.9 increase in women’s education Total 4.9 5.3 4.4 5.3 5.6 5.0 levels (from 4.55 to 5.22 years) Source: HIES 2005 and 2010. exceeded the increase experienced by men (from 5.6 to 5.83 years). The difference in educational attainment between the poor and non-poor slightly declined, although the gap, at 3.49 years of education, is still substantial. 1.4. Poverty and Nutrition 17. This section examines nutritional diversity, as measured by dietary diversity scores (DDS), of the poor relative to the non-poor. 40, 41 For our analysis, we define DDS as the number of distinct food groups consumed by a household during the week prior to being surveyed for the HIES. Each food group represents a special class of nutrients, and a higher DDS indicates greater diversity of food intake and better quality diets. Thus, this measure can be used as a relatively simple indicator for the micronutrient adequacy of households’ diets (Steyn et al. 2006). 18. At the household-level, higher DDS are negatively correlated with poverty (a correlation that holds true for Bangladesh, see Figure 2-6). Households’ dietary patterns as well as diet quality are 40 This section draws from a background paper by Rabbani (2012), commissioned for this poverty assessment report. 41 The HDDS was developed by USAID’s Food and Nutrition Technical Assistance Project (FANTA). For details on the construction of the DDS, refer to Appendix 2. 28 important inputs into the production of anthropometric outcomes, so either measure can be used as a proxy for such outcomes. For example, Rah et al. (2010) find that good dietary diversity is strongly negatively associated with stunting among children aged less than five years. Similarly, low dietary diversity is often associated with higher prevalence of infections (Waterlow 1994). In the absence of anthropometric measures, this section uses DDS to shed light on the nutritional status of Bangladeshi households. 19. Figure 2-4 compares the dietary patterns of Bangladeshi households in 2005 Figure 2-4: Percent of Households Consuming a Particular Food Group in 2005 and 2010 and 2010. Overall, the figure reveals that nearly all households consume from the 100% 98% 2005 2010 97% “cereals” and “oil/fats” food groups. While 90% 80% 82% 70% many households consume roots and tubers 70% 61% 65% (for example, potatoes), starchy cereals 60% 50% 36% 36% (namely, rice) remain the main source of 40% 29% 31% 30% 19% energy for most households on any given day. 20% 10% With regard to protein, households primarily 10% 0% rely on fish; in 2005, about 63 percent of households consumed fish on any given day compared to 61 percent in 2010. In general, households’ consumption of meat products (about 10 percent in both years) and eggs Note: The trace amount of consumption for each food item was not counted (about 14 percent and 19 percent in 2005 and toward a food group. 2010, respectively) was relatively low. Pulses Source: Rabbani (2012). Data source: HIES 2005 and 2010. and similar food items provide only limited respite from low protein consumption of meat products. While 80 percent of households, on average, consumed vegetables,42 consumption of fruits was relatively low in both years (about 25 percent of households). Figure 2-5: Fraction of Households Consuming Food Items from a Specific Group A. HIES 2005 (N = 139,505)… B. HIES 2010 (N = 171,360)… Non-poor Poor Severely poor Non-poor Poor Severely Poor Cereals Root and Cereals Miscellaneous tubers Root and Miscellaneous tubers Sugar/honey Vegetables Sugar/honey Vegetables Oil/fats Fruits Oil/fats Fruits Milk and milk Meat, poultry, Milk and milk Meat, poultry, products offal products offal Pulses/legume Pulses/legume Eggs Eggs s/nuts s/nuts Fish and Fish and seafood seafood Notes: Households are stratified into three groups according to poverty status: “Non-poor”: households with per-capita consumption levels above the upper poverty line; “Poor”: households with per-capita consumption levels between the lower and the upper poverty lines; “Severely poor”: households with per-capita consumption levels below the lower poverty line. Source: Rabanni (2012). 42 In rural Bangladesh, vegetable consumption sometimes reflects non-market appropriation, such as gathering through kitchen gardening, particularly for poor households. 29 20. Stratifying households into groups based on poverty status (i.e., “poor” if the household falls between the lower and upper poverty line; “extremely poor” if it falls below the lower poverty line), dietary consumption patterns are compared in Figure 2-5. This figure reveals some striking patterns. First, the dietary components that distinguish the diets of the poor from the non-poor were milk, sugar products, and eggs. Consumption of eggs among the non-poor was about 20 percent compared with 8 percent among extremely poor households. Similarly, consumption of milk products was 38 percent among non- poor households and 9 percent among the extremely poor. Second, fish and seafood consumption was also more prevalent among the non-poor. While fish remained the dominant source of protein for all three groups (about 65 percent of households consumed fish), only one-half of extremely poor households consumed fish on a daily basis, indicating a 15 percentage point gap in consumption between the poor and the non-poor. 21. Consumption of meat products was relatively low for the overall population, and especially low among extremely poor households. Only three percent of households reported consuming meat on a given day in 2005 or 2010; for non-poor households, the rate was about 13 percent; for poor households, about 6 percent. For three food groups, cereals, oil/fat, and vegetables, no discernible differences exist between the three groups. On any given day, nearly all households consumed food items from the cereals and oil/fat groups, and four out of five households consumed at least one vegetable food item. 22. Figure 2-6 shows average household Figure 2-6: Average Household Dietary Diversity DDS by consumption decile. The figure Score by Consumption Deciles demonstrates a positive relationship 7.5 between consumption and DDS. While 2005 2010 7.0 higher consumption does not necessarily 6.5 Average DDS translate into greater dietary diversity, 6.0 consumption expenditure is nevertheless 5.5 positively related to better diversity. In 5.0 particular, for both 2005 and 2010, 4.5 households in the lowest deciles reported 4.0 1 2 3 4 5 6 7 8 9 10 dietary diversity scores ranging from 4.5 to 5, suggesting that the diets of relatively Consumption Decile poor households barely meet basic energy Source: Rabbani (2012). Data source: HIES 2005 and 2010. requirements. The estimates suggest that Figure 2-7: Frequency Distribution of HDDS in 2005 Figure 2-8: Cumulative Distribution of HDDS in and 2010 2005 and 2010 35.0% 1.0 0.9 30.0% 0.8 0.7 25.0% 2005 non-poor 0.6 CDF 20.0% 2005 all poor 0.5 2005 0.4 15.0% 2010 non-poor 0.3 2010 10.0% 2010 all poor 0.2 0.1 5.0% 0.0 3 4 5 6 7 8 9 10 11 0.0% DDS ≤3 4 5 6 7 8 9 10 11 Note: Total number of observations: 139,505 for 2005 and 171,360 for Note: Total number of observations: 139,505 for 2005 and 171,360 for 2010. 2010. Source: Rabbani (2012). Data source: HIES 2005 and 2010. Source: Rabbani (2012). Data source: HIES 2005 and 2010. 30 poorer households primarily consumed rice and fats, which were likely complemented with some vegetables and, to a lesser extent, fish. Households in the top decile, however, consumed as many as seven food groups, on average, suggesting that their diets were more nutritionally diverse. The distributions of households’ DDS indicate that overall dietary diversity remained fairly static between 2005 and 2010 (Figure 2-7, Figure 2-8). Average household DDS was 5.65 in 2005 and 5.70 in 2010. Moreover, the cumulative distribution functions for 2005 and 2010 show no evidence of stochastic dominance for one distribution over the other, suggesting that no significant welfare changes related to dietary diversity occurred over this period (Figure 2-9). Figure 2-9: Stochastic Dominance of HDDS by Figure 2-10: Stochastic Dominance of HDDS by Poverty Status in 2005 Poverty Status in 2010 1.0 1.0 Cumulative Distribution Cumulative Distribution 0.8 0.8 Non-poor Non-poor 0.6 0.6 Moderately Moderately 0.4 0.4 Poor Poor 0.2 0.2 Ultra-poor Ultra-poor 0.0 0.0 3 4 5 6 7 8 9 10 11 3 4 5 6 7 8 9 10 11 DDS DDS Source: Rabbani (2012). Data source: HIES 2005. Source: Rabbani (2012). Data source: HIES 2010. 23. While both the level of DDS and its distribution have remained somewhat unchanged over the 2005 to 2010 period, Figure 2-9 and Figure 2-10 show that the cumulative distribution functions (CDFs) differ significantly by poverty status. In particular, the DDS-CDF corresponding to the non-poor stochastically dominates that of the poor, which, in turn, stochastically dominates that of the extremely poor. As expected, non-poor households have greater dietary diversity relative to poor households. 24. Figure 2-11 and Figure 2-12 show the distribution of the risk of having a low DDS at the household-level, rather than the household diversity score on a given day. A diversity score exceeding Figure 2-11: Cumulative Frequency Distribution - Figure 2-12: Cumulative Frequency Distribution - Dietary Diversity Secured Dietary Diversity Secured Number of Day(s) in 2005 Number of Day(s) in 2010 Non-poor Moderately Poor Ultra-poor Non-poor Moderately Poor Ultra-poor 1.0 1.0 0.9 0.9 0.8 0.8 0.7 0.7 0.6 0.6 0.5 0.5 0.4 0.4 0.3 0.3 0.2 0.2 0.1 0.1 0.0 0.0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 Source: Rabbani (2012). Data source: HIES 2005. Source: Rabbani (2012). Data source: HIES 2010. 31 four implies that a household consumes beyond the basic set of food items. This measure of diversity security follows a pattern similar to the DDS. For both years, about 50 percent of households are unable to secure a DDS of five or more during at least seven of the 14 days when the HIES consumption data is collected. When considering only the non-poor, the picture improves significantly; 16 percent and 14 percent, in 2005 and 2010, respectively, face a relatively high risk of having a low DDS. 25. Lastly, Figure 2-13 (Figure 2-14) explores the relationship between 2005 levels of poverty (DDS) and changes in poverty between 2005 and 2010. For each survey year, poverty rates and DDSs are stratified according to 16 primary sampling units (PSUs), corresponding to the urban, rural, and metropolitan statistical area of each division. The PSU-level change in poverty rates between 2005 and 2010 is plotted against the corresponding PSU-level poverty prevailing in 2005 (Figure 2-13). This figure shows a positive relationship between poverty rates in 2005 and changes in poverty between 2005 and 2010, suggesting that PSUs with greater poverty rates in 2005 also registered higher poverty declines between 2005 and 2010. Interestingly, Figure 2-14 reveals no such relationship between the 2005 DDS and poverty changes between 2005 and 2010. Figure 2-13: Decline in Poverty Rates between 2005 Figure 2-14: Annual Changes in DDS between 2005 and 2010 by PSUs and 2010 by PSUs 0.12 0.20 Annual Change between 2005 and 0.10 Anual Percentage point Decline 0.15 y = 0.174x - 0.0427 between 2005 and 2010 0.08 R² = 0.5904 y = -0.0029x + 0.0478 0.10 R² = 0.0005 0.06 2010 0.04 0.05 0.02 0.00 0.00 4.5 5 5.5 6 6.5 7 7.5 15% 25% 35% 45% 55% 65% -0.05 -0.02 -0.04 -0.10 Poverty Rate in 2005 Dietary Diversity Score in 2005 Source: Rabanni (2012). Data source: HIES 2005 and 2010. Source: Rabanni (2012). Data source: HIES 2005 and 2010. Anthropometric Outcomes 26. Our analysis of poverty trends, presented in Chapter 1, shows that Bangladesh has reached the MDG 1.A: 1.2 of halving the poverty gap to eight percent by 2015 and is on track for reaching the MDG 1.A: 1.1 of halving the poverty rate to 29 percent by 2015. Similarly, evidence from the Demographic Health Survey (DHS) 2011 indicates a five percentage point decline in the prevalence of underweight children under five years, from 41 percent in 2007 to 36 percent in 2011 (Figure 2-15.A). If the reduction in underweight prevalence continues at about one-half of this rate, the MDG 1.8, which stipulates underweight prevalence to be reduced to 33 percent, is likely to be met by 2015 as well. In fact, children, aged five months or younger, residing in urban areas had already achieved this goal by 2007. Like poverty, however, childhood underweight, a measure of overall nutritional health, is significantly more prevalent in rural areas. Moreover, improvements in the childhood underweight outcome have not been evenly shared across the wealth distribution and, in fact, continue to display a strong negative wealth gradient. 32 Figure 2-15: Prevalence of Under-weight (0-59 months) weight (0-59 months) A. Weight-for-age (under-weight) A. Weight-for-height (wasting) 2004 2007 2011 2004 2007 2011 25 70 20 18 18 60 50 16 16 16 50 42 39 14 14 36 36 15 12 40 28 28 30 21 10 20 10 5 0 Lowest Second Middle Fourth Highest Nation Urban Rural 0 Lowest Second Middle Fourth Highest Nation Urban Rural Note: “Weight-for-age is a composite index of weight-for-height and height-for-age, and thus does not distinguish between acute malnutrition (wasting) and chronic malnutrition (stunting). A child can be underweight for his age because he is stunted, because he is wasted or both. Weight- for-age is a good overall indicator of a population’s nutritional health” (DHS 2011). Source: DHS 2004, 2007, and 2011. 27. Figure 2-15.B displays trends in the weight-for-height (wasting) measure.43 This anthropometric measure is highly responsive to short-term fluctuations in various external factors, such as food intake, disease, and changes in the environment, and, therefore, is an indicator for current nutritional status. Coinciding with the beginning of the food commodity price spike as well as the devastating floods that hit South Asia in 2007, leaving nearly seven million people marooned,44 the number of wasted children, aged 0 to 59 months, dramatically increased between 2004 and 2007. However, while the worsening of wasting most strongly affected children residing in rural areas between 2004 and 2007, the same rural children also experienced the improvements that occurred in this indicator between 2007 and 2011. 28. The significant variability displayed by this indicator over the 2004 to 2011 period is also consistent with the effects of the commodity food price shock on household welfare. As the food price shock hit households in 2007, its immediate effect was a spike in staple foods and, subsequently, a likely decline in food consumption among those unable to insure against the shock. Yet, as shown in Chapter 7, the less affluent, particularly those engaged in agricultural activities, were able to reap benefits, namely in the form of higher wages for agricultural workers, from the price shock’s spillover effects. 29. Even in the face of outstanding national-level achievements with respect to poverty reduction and the prevalence of childhood underweight, achieving every hunger- and nutrition-related MDG is a challenging undertaking (Table 2.6). While Bangladesh is likely to meet the first calorie target, MDG 1.9a Table 2.6: Percentage of Population with Moderate and Severe Deficiency in Calorie Intake Year Moderate deficiency (MDG 1.C: 1.9) Severe deficiency (MDG 1.C: 1.9a) (< 2,122 kilocalories/person/day) (< 1,805 kilocalories/person/day) Rural Urban National Rural Urban National 2000 42.3 52.5 44.3 18.7 25.0 20.0 2005 39.5 43.2 40.4 17.9 24.4 19.5 2010 36.8 42.7 38.4 14.9 19.7 16.1 Source: HIES 2000, 2005, and 2010. 43 According to the DHS 2011, a child who is below minus two standard deviations from the reference median for weight-for-height is considered to be too thin for his or her height, i.e. wasted. 44 See http://news.bbc.co.uk/2/hi/south_asia/6927389 (retrieved on November 10, 2012). 33 (i.e. 14 percent or less of the population experiencing severe food deficiency), it is unlikely to meet the second calorie target, MDG 1.9 (i.e. 24 percent or less of the population experiencing moderate food deficiency), even if growth over the next three years continues at a steady pace. Moreover, if growth deteriorates, then so too will the gains in poverty and hunger reduction. 2.5. Poverty and Health Outcomes Poverty and Selected Health Indicators for Adults and Children 30. In 2005, irrespective of poverty status, approximately two out of every ten individuals reported experiencing a symptom of illness or injury over the last 30 days (Table 2.7). While the non-poor had a slightly higher prevalence of symptoms or illness relative to the poor, the difference in prevalence between the two groups is not statistically significant, neither among children nor adults. Even as the poor are often more likely to experience health shocks, the finding that the non-poor have a slightly higher prevalence of self-reported illness is not surprising. Research shows that the poor tend to underreport illness, which arises from the “forgetting of entire episodes of acute illness” (Das et al. 2012). 31. In 2005, about 88 (82) percent of non-poor (poor) adults reporting an illness or injury episode over the last 30 days sought help to treat their malady; the difference between the poor and non-poor is a statistically significant six percentage point. A similar pattern was observed between poor and non-poor Table 2.7: Selected Health Indicators for Adults and Children: years 2005 and 2010 2005 Adults Children Non- Poor Diff. (p- Non- Poor Diff. (p- poor value) poor value) Symptom of illness/ injury (last 30 days) 0.19 0.18 0.11 0.19 0.18 0.07 Sought help to treat symptom of illness/ injury 0.88 0.82 0.00 0.88 0.80 0.00 Type of help sought to treat symptom of illness/ injury: Pharmacy 0.28 0.31 0.17 0.29 0.35 0.01 Doctor 0.41 0.35 0.00 0.40 0.34 0.01 Reasons for not seeking help when sick Too costly 0.15 0.34 0.00 0.14 0.35 0.00 Not serious 0.74 0.52 0.00 0.76 0.51 0.00 2010 Adults Children Non- Poor Diff. (p- Non- Poor Diff. (p- poor value) poor value) Symptom of illness/ injury (last 30 days) 0.19 0.18 0.12 0.20 0.18 0.10 Sought help to treat symptom of illness/ injury 0.92 0.91 0.12 0.93 0.92 0.52 Type of help sought to treat symptom of illness/ injury: Pharmacy 0.35 0.43 0.00 0.42 0.50 0.00 Doctor 0.57 0.46 0.00 0.51 0.38 0.00 Reasons for not seeking help when sick Too costly 0.11 0.35 0.00 0.03 0.25 0.00 Not serious 0.55 0.40 0.02 0.74 0.55 0.03 Source: HIES 2005 and 2010. 34 children: a statistically significant eight percentage point difference. By 2010, however, the large, income-related gap in treatment became smaller and statistically insignificant. Nevertheless, by 2010, significant differences remained with regard to the type of care sought by the poor and the non-poor. In 2005 (2010), when ill or injured, poor children were six (eight) percentage points more likely to see a pharmacist compared to non-poor children. Similarly, in 2005, ill or injured poor children were six percentage points less likely to see a doctor relative to non-poor children, and this difference more than doubled by 2010. A similar pattern was observed for adults. 32. Moreover, the main reason for not seeking help when sick varied significantly between the poor and non-poor. In 2005, thirty-four percent of poor adults cited high cost as the reason for not seeking help, compared to only 15 percent of non-poor adults, a difference of 19 percentage points between the two groups. Non-poor adults, on the other hand, were 22 percentage points more likely to not seek help because they consider their illness or injury not serious. An analogous pattern was observed among poor and non-poor children. By 2010, the differences between the poor and non-poor became more pronounced. Among the poor, 35 (25) percent of adults (children) claimed high cost as the main reason for not seeking help, compared to only 11 (3) percent of non-poor adults (children). Poverty and Selected Health Indicators by Location 33. Table 2.8 reports the same health outcomes as Table 2.7 but by location (i.e. urban versus rural). In rural areas, both in 2005 and 2010, the non-poor were more likely to self-report an illness or injury than the poor, with the differences being statistically significant in both years. Interestingly, in 2010, the urban poor were three percentage points more likely than the urban non-poor to self-report an illness or injury, perhaps suggesting that the urban poor have better access to health services relative to the rural poor. Irrespective of location, the non-poor were also about six to seven percentage points more likely than the poor to seek help when ill or injured; however, these differences became smaller and statistically insignificant by 2010. In 2005, the poor residing in urban areas were 11 percentage points more likely than the non-poor to seek help from a pharmacist when ill or injured. Irrespective of location, the non- poor were also more likely than the poor to seek help from a doctor when ill or injured. Nevertheless, as 49 percent of the non-poor from urban areas sought help from a doctor, they were the most likely to receive proper help; 11 percentage points more likely than both the poor in urban areas and the non-poor in rural areas. The rural poor were the worst off in the sense that only one-third sought a doctor’s help when ill or injured. 34. By 2010, irrespective of location, the poor were eight to nine percentage points more likely than the non-poor to seek help from a pharmacist when ill or injured. As in 2005, the non-poor remained more likely than the poor to seek help from a doctor when ill or injured in 2010. Sixty percent of the urban non- poor sought help from a doctor, a value that is 14 percentage points greater than the urban poor and six percentage points higher than the rural non-poor. While a larger proportion of the rural poor sought help from a doctor in 2010 relative to 2005, the rural poor still remained the worst off along this dimension since only 42 percent reported seeking help from a doctor. 35. When considering reasons for not seeking help when injured or ill, large differences existed across income groups and across locations in 2005. Across income groups, the poor were significantly more likely than the non-poor to cite high cost as a barrier for seeking help. Regardless of location, the non-poor were 22 to 25 percentage points more likely than the poor to report not seeking help because their illness or injury was not serious. Large differences were also present among the poor depending upon location. For example, the poor from rural areas were 10 percentage points more likely to cite high cost as a barrier for seeking help than the poor from urban areas. Similarly, the urban poor were 10 percentage points more likely than the rural poor to report that their illness or injury was not serious enough to warrant seeking help. 35 Table 2.8: Selected Health Indicators for Urban and Rural: years 2005 and 2010 2005 Urban Rural Non- Poor Diff. Non- Poor Diff. poor (p-value) poor (p-value) Symptom of illness/ injury (last 30 days) 0.16 0.17 0.83 0.20 0.18 0.00 Sought help to treat symptom of illness/ injury 0.89 0.82 0.00 0.88 0.81 0.00 Type of help sought to treat symptom of illness/ injury: Pharmacy 0.20 0.31 0.00 0.31 0.33 0.37 Doctor 0.49 0.38 0.00 0.38 0.33 0.04 Reasons for not seeking help when sick Too costly 0.05 0.26 0.00 0.17 0.36 0.00 Not serious 0.85 0.60 0.00 0.72 0.50 0.00 2010 Urban Rural Non- Poor Diff. Non- Poor Diff. poor (p-value) poor (p-value) Symptom of illness/ injury (last 30 days) 0.13 0.16 0.00 0.23 0.19 0.00 Sought help to treat symptom of illness/ injury 0.94 0.92 0.17 0.92 0.91 0.37 Type of help sought to treat symptom of illness/ injury: Pharmacy 0.34 0.43 0.03 0.39 0.47 0.00 Doctor 0.60 0.46 0.00 0.54 0.42 0.00 Reasons for not seeking help when sick Too costly 0.07 0.30 0.01 0.08 0.31 0.00 Not serious 0.76 0.57 0.07 0.59 0.45 0.04 Source: HIES 2005 and 2010. 36. In 2010, the income-related and regional patterns that prevailed were similar to those in 2005. Interestingly, for both the poor and non-poor, the location-related difference associated with citing high cost as the reason for not seeking treatment considerably decreased; in other words, the rural poor look more similar to the urban poor and the rural non-poor more similar to their urban counterparts (see last two rows of Table 2.8). Difference in Selected Health Indicators over Time 37. Between 2005 and 2010, significant changes occurred with respect to the selected health indicators (Table 2.9). Non-poor Bangladeshis from urban (rural) areas were less (more) likely to self- report illness or injuries in 2010 relative to 2005. Yet, irrespective of poverty status and location, more Bangladeshis sought help to treat their symptoms or injuries in 2010. With the exception of poor children and the urban poor, the poor and non-poor were more likely to seek help from both pharmacists and doctors. In 2010, as compared to 2005, with respect to reasons for not seeking help when sick, non-poor adults were less likely to claim their illness or injury as not serious, and non-poor children were less likely to report the cost of care was too high. When considering location, only the rural non-poor were less likely to claim that the cost of care was too high or that their illness or injury was not serious as their main reasons for not seeking help. 36 Table 2.9: Changes in Selected Health Indicators = YES whenever the group mean has changed significantly between 2005 and 2010 Adult Child Non- Non- Poor Poor poor poor Symptom of illness/ injury (last 30 days) Sought help to treat symptom of illness/injury YES YES YES YES Type of help sought to treat symptom of illness/injury: Pharmacy YES YES YES YES Doctor YES YES YES Reasons for not seeking help when sick Too costly YES Not serious YES Urban Rural Non- Non- Poor Poor poor poor Symptom of illness/ injury (last 30 days) YES YES Sought help to treat symptom of illness/injury YES YES YES YES Type of help sought to treat symptom of illness/injury: Pharmacy YES YES YES YES Doctor YES YES YES Reasons for not seeking help when sick Too costly YES Not serious YES Note: YES = difference (i.e. Mean2010 – Mean2005) is statistically significant at the 5 percent level. Source: HIES 2005 and 2010. Poverty and Child Immunization 38. Between 2005 and 2010, significant improvements occurred with regard to immunization rates for children between the ages of one and five years (Table 2.10). While immunization rates were already relatively high in Bangladesh in 2005, the proportion of children immunized increased between 2005 and 2010, irrespective of income. For the poor, this increase was statistically significant (see last four rows of Table 2.10). Moreover, irrespective of gender and income level, the number of children fully-immunized significantly increased. Between the two years, the gap between the percentage of poor and non-poor fully-immunized narrowed from five percentage points to one percentage point for all children; it remained significant for girls but became statistically insignificant for boys. Similarly, the number of children with an immunization card dramatically increased between 2005 and 2010, and the large differences in immunization rates that existed between the poor and non-poor in 2005 became statistically insignificant by 2010. 37 Table 2.10: Child Immunization Rates - 2005 and 2010 2005 Children (≤5 years old) Boys (≤ 5 years old) Girls (≤ 5 years old) Non- Poor Diff. Non- Poor Diff. Non- Poor Diff. poor (p-val.) poor (p-val.) poor (p-val.) Immunized 0.98 0.96 0.98 0.00 0.96 0.06 0.98 0.95 0.00 Fully-immunized 0.82 0.77 0.82 0.00 0.78 0.05 0.82 0.77 0.00 Has immun. Card 0.68 0.56 0.67 0.00 0.56 0.00 0.68 0.57 0.00 BCG 0.96 0.93 0.96 0.00 0.93 0.00 0.96 0.92 0.00 DPT-1 0.97 0.94 0.97 0.00 0.95 0.01 0.97 0.94 0.00 DPT-2 0.95 0.91 0.95 0.00 0.91 0.00 0.94 0.90 0.00 DPT-3 0.91 0.86 0.90 0.00 0.86 0.00 0.91 0.86 0.00 Polio-1 0.95 0.93 0.96 0.00 0.93 0.00 0.95 0.92 0.03 Polio-2 0.93 0.89 0.93 0.00 0.89 0.00 0.93 0.88 0.00 Polio-3 0.87 0.84 0.87 0.03 0.85 0.26 0.87 0.84 0.04 Measles 0.84 0.80 0.83 0.00 0.81 0.18 0.84 0.79 0.00 2010 Children (<5 years old) Boys (< 5 years old) Girls (< 5 years old) Non- Poor Diff. Non- Poor Diff. Non- Poor Diff. poor (p-val.) poor (p-val.) poor (p-val.) Immunized 0.98 0.98 0.77 0.98 0.99 0.67 0.99 0.98 0.34 Fully-immunized 0.96 0.95 0.02 0.96 0.95 0.35 0.97 0.94 0.00 Has immun. Card 0.80 0.77 0.11 0.80 0.79 0.41 0.79 0.76 0.08 BCG 0.98 0.98 0.47 0.98 0.98 0.88 0.99 0.98 0.19 DPT-1 0.98 0.98 0.30 0.98 0.98 0.96 0.99 0.97 0.09 DPT-2 0.98 0.97 0.16 0.98 0.98 0.95 0.98 0.97 0.03 DPT-3 0.98 0.97 0.12 0.97 0.97 0.97 0.98 0.96 0.02 Polio-1 0.98 0.98 0.22 0.98 0.98 1.00 0.98 0.97 0.04 Polio-2 0.98 0.97 0.25 0.98 0.98 0.82 0.98 0.97 0.05 Polio-3 0.98 0.97 0.04 0.98 0.97 0.31 0.98 0.96 0.03 Measles 0.97 0.96 0.06 0.97 0.97 0.74 0.97 0.95 0.01 =YES whenever the group mean has changed significantly between 2005 and 2010 Immunized YES YES YES Fully-immunized YES YES YES YES YES YES Note: YES = difference (i.e. Mean2010 – Mean2005) is statistically significant at the 5 percent level. Source: HIES 2005 and 2010. 39. When considering immunization-specific differences, a similar pattern is evident; the statistically significant differences that existed between the poor and non-poor in 2005 virtually disappeared for all children in 2010. However, looking at the same immunization-specific measures by gender reveals that poor girls still continued to lag behind their non-poor counterparts in 2010. Specifically, while the large differences in immunization rates that existed between poor and non-poor girls in 2005 decreased, the differences were still statistically significant for DPT-2 and 3, all three Polio vaccines, and Measles in 2010. 38 40. Bangladesh’s Expanded Programme on Immunization (EPI) has significantly contributed to reducing mortality and morbidity from vaccine-preventable diseases.45 As part of EPI, a major national polio and measles immunization campaign took place between 2005 and 2006. UNICEF reports that this campaign was successful at reaching about 35 million children, resulting in a drastic reduction in measles rates.46 The campaign was repeated in 2010, targeting about 20 million children who were born after the last campaign. Similar vaccination campaigns and initiatives have been undertaken by EPI in collaboration with various international organizations and donors.47 HIES estimates, showing high immunization rates and significant improvements in immunization outcomes between 2005 and 2010 support the conjecture that EPI has been one of the most successful public health interventions in the country. More importantly, the fact that Bangladesh is on track for achieving the MDG 4 for under-five mortality (48 deaths per 1,000 live-births by the year 2015, see NIPORT 2011) is further evidence in support of this assessment. Poverty and Medical Expenditures 41. The largest component of medical expenditures is medicines for both the poor and non-poor, regardless of location and survey year (Figure 2-16). However, some interesting differences emerge between the relative distributions of medical expenditures between the two groups. In particular, the non- poor allocate a relatively larger proportion of their budgets to consultations and test exams. These differences are consistent with the pattern we observe in Table 2.8, which shows that while the non-poor, relative to the poor, are more likely to visit a doctor when ill, the poor are more likely to visit a pharmacist when compared to the non-poor. Finally, another interesting observation is that the non-poor from urban areas devote nearly twice as large a proportion of medical expenditures to consultations (13 percent) than the non-poor from rural areas (seven percent). Figure 2-16: Medical Expenditures, by category 2005 2010 100% 100% 50% 86% 50% 86% 85% 80% 71% 82% 80% 75% 0% 0% Rural, non- Urban, non- Rural, poor Urban, poor Rural, non- Urban, non- Rural, poor Urban, poor poor poor poor poor Medicines Consultation Transportation Tests Medicines Consultation Transportation Tests Hospital Tips Maternity Hospital Tips Maternity Source: HIES 2005 and 2010. 42. In terms of payment sources, more than 40 (13) percent of rural households used their savings to pay for medical expenditures in 2005 (2010), irrespective of poverty status, whereas less than 24 (12) percent of urban households used their savings. Similarly, a larger proportion of rural households, relative to urban households, paid for medical expenditures by depleting their belongings and assets or by making other arrangements, which included borrowing, mortgaging their homes, or receiving assistance from 45 http://www.unicef.org/bangladesh/health_nutrition_468.htm 46 http://www.unicef.org/infobycountry/bangladesh_52966.html 47 http://www.unicef.org/media/files/Measlescampaign.PDF 39 relatives (Figure 2-17). While urban and rural households continued to differ in how they financed their medical expenditures, rural households looked more similar to urban households in 2010 as compared to 2005. Figure 2-17: Sources of Payments for Medical Expenditures 2005 2010 76% 88% 82% 88% 81% 86% 83% 87% 44% 41% 13% 22% 10% 17% 24% 11% 17% 6% 8% 12% 4% 13% 13% 10% 8% 3% 2% 3% 2% 1% 2% 1% Rural, non-poor Urban, non- Rural, poor Urban, poor Rural, non-poor Urban, non-poor Rural, poor Urban, poor poor Income Savings Sold belongings/assets Other arrangements Income Savings Sold belongings/assets Other arrangements Source: HIES 2005 and 2010. 2.6. Poverty and the Labor Market 43. Lack of labor income is generally associated with higher risk of poverty.48 Therefore, this section analyzes changes in the distribution of poor workers across employment sectors, occupational type, and over time. Poverty and employment sectors 44. Nearly one-half of poor workers are employed by the agricultural sector, while another one- quarter work in the services sector (Figure 2-18.A). However, the share of poor workers in agriculture slightly decreased between 2005 and 2010, while the manufacturing sector’s share of poor workers Figure 2-18: Distribution of Poor Workers (share of poor workers) A. By Sector of Employment B. By Occupational Type 100 100 13.1 14.1 17.5 23.3 25.8 26.5 80 24.3 21.6 18.6 60 16.6 14.0 12.8 12.3 16.9 19.3 50 40 48.6 51.5 51.6 55.4 51.0 47.5 20 0 0 2000 2005 2010 2000 2005 2010 Daily workers Self-employed non-farm Agriculture Manufacturing Industry Services Self-employed in farm Salaried Source: HIES 2000, 2005, and 2010. 48 See, for example, Davis (2008), who showed that some of the most important causes of economic hardships in Bangladesh are dowry and too few earning members in the household. Such findings underscore the importance of adequate returns to labor for poverty reduction. 40 increased. This shift to manufacturing employment has coincided with an Figure 2-19: Poverty Headcount by Type of Worker, 2000- 2010 (percent poor) increase in the share of the poor who are Moderate Poverty salaried, while the share of self-employed in agriculture has declined (Figure 2- 80% 67% 2000 18.B). These findings are consistent with 57% 2005 60% other empirical work that finds that the 45% 2010 39% 37% development of the nonfarm sector has led 40% 36% 26% 24% 27%24% to an increase in nonfarm income and a 23% 22% major reduction in rural poverty in 20% Bangladesh (Nargis and Hossain 2006). 0% Moreover, these changes coincide with the Daily Farm Self-employed Non-farm Self Employed Salaried growth of labor income that occurred in these sectors and occupations over the Source: HIES 2000, 2005, and 2010. decade, as documented in Chapter 4. Nevertheless, one-half of poor workers are still employed as daily workers. While the poverty rate for these type of workers has considerably declined over the decade, it still remains high (39 percent) relative to the national average (see Figure 2-19). Non-poor workers tend to be salaried employees or self- employed outside of agriculture, where poverty rates are 22 percent and 24 percent, respectively. Interestingly, poverty headcount rates among the agricultural self-employed declined from 45 percent in 2000 to 23 percent in 2010. 2.7. Conclusions 45. This chapter profiles characteristics of the poor in Bangladesh by evaluating several alternative measures of well-being. Like the Datt and Ravallion (1992) decomposition results presented in Chapter 1, the analysis of assets, home amenities, and land ownership highlights significant differences between the first and second halves of the decade. 46. First, during the 2000-2005 period, many households experienced substantial improvements in overall housing conditions. For example, a large number of households saw improvements in the quality of materials used to construct their homes (i.e. walls and roofs made of corrugated iron, steel, and cement) as well as their access to services (i.e. sanitary latrines and electricity). During the 2005-2010 period, while the poor continued to improve the quality of their homes, the largest improvements for all households occurred with respect to amenities owned, such as television sets and access to cellular phones. 47. Second, the analysis of the relationship between land ownership and poverty reveals that, while in the 2000-2005 period, poverty declines were larger among households with relatively larger landholdings (i.e. more than 1.5 acres of land), the 2005-2010 period brought larger poverty declines to households having relatively smaller landholdings, which are more likely to be deprived households. 48. In terms of demographic characteristics and educational attainment, our analysis shows the standard correlations: households that are poor tend to (1) be larger in size; (2) have higher dependency ratios; (3) have a greater number of children; (4) be headed by individuals who are nearly three years younger than non-poor households; and (5) be headed by individuals with no education. 49. Early marriage is often associated with poverty and lower socio economic status, hence the finding that heads of poor households are generally younger and less educated is not surprising. Still, even though poverty and lack of both education and alternatives may drive individuals to marry at a young age, early marriage is also likely to affect human capital formation and, thus, perpetuate the 41 intergenerational transmission of poverty. For example, using data from rural Bangladesh, Field and Ambrus (2008) show that each additional year that marriage is delayed is associated with 0.22 additional years of schooling and 5.6 percent higher literacy. In turn, literature using data from other countries shows that higher parental education improves infant health, reduces parity, increases use of prenatal care, and helps to buffer negative health shocks experienced by infants, findings which suggest clear pathways between higher education and intergenerational transmission of welfare (Currie and Moretti 2003; Chou et al. 2010). Policies aimed at breaking the intergenerational transmission of poverty and low human capital should be designed to target vulnerable groups, such as teenagers from poor households, who are at higher risk for school drop-out and early marriage. 50. Analysis of health and nutritional outcomes tells a mixed story for Bangladesh. In the 2005-2010 period, Bangladesh witnessed significant improvements in access to health care, the most notable achievement was the significant reduction in the differences in immunization rates between the poor and non-poor. With regard to nutrition, however, the evidence suggests that low dietary diversity is a persistent problem in Bangladesh, even as the country experienced significant declines in inter-regional differences in poverty rates. For example, even in the face of outstanding achievements in reducing poverty and childhood underweight at the national-level, achieving all of the hunger-related MDGs remains a daunting undertaking for Bangladesh. 51. Finally, with regard to poverty by sectors of employment, most poor workers are still employed in the agricultural sector, while another one-quarter is employed in the services sector. Nonetheless, the share of poor workers in agriculture slightly dropped between 2005 and 2010, while the manufacturing sector saw its share increase. This shift has coincided with an increase in the share of poor who are salaried employees and a reduction in the share of self-employed in agriculture.49 49 Labor market trends will be further examined in Chapter 4. 42 Part II: Understanding the Drivers of Poverty Reduction over the 2000-2010 Decade 3. Decomposing Poverty: Understanding key drivers of poverty reduction 1. As a first step toward understanding the underlying dynamics of poverty reduction in Bangladesh over the last decade, this chapter decomposes distributional changes in income occurring between the years 2000, 2005, and 2010. We then examine the various components driving these distributional changes. In particular, this chapter answers the following questions: What are the factors affecting the observed poverty and distributional changes? Is some of the reduction attributable to higher employment, higher productivity, or higher remittances? Was the reduction in poverty linked to changes in the sectoral composition of employment, changes in the characteristics of human capital, or simply higher returns to these characteristics? 2. Although these decompositions do not allow for the identification of causal effects, they help to focus attention to the quantitative elements that are most important to describing changes in poverty. Thus, the components provide a road map for the remaining chapters of the report. The insights provided by a deeper understanding of income and poverty dynamics can contribute to the expansion of the evidence base for Bangladesh’s policy-making going forward. 3. The rest of the chapter is organized as follows. Section 1 provides descriptive statistics for the main income component used by the HIES survey in all three rounds (2000, 2005, and 2010) and discusses potential sources of the observed distributional changes in income occurring over the period. Section 2 gives an introduction to the model of consumption and earnings underlying the poverty decompositions and explores the main decomposition results. Section 3 concludes. 3.1. Drivers of Poverty Reduction 4. Before undertaking decomposition analysis, we first examine the trends for each underlying component of income. Table A3-1 shows that real per-capita income increased by 54 percent between 2005 and 2010, which considerably exceeds the increase of 19 percent between 2000 and 2005. The table also reveals that these gains were realized by the entire spectrum of the income distribution, even the poorest. Income for the bottom decile, the poorest 10 percent of the population, increased by 29 percent between 2005 and 2010; the corresponding gain was only seven percent between 2000 and 2005. 5. The fifth and tenth columns of Table A3-1 suggest that while the percentage change in remittances between 2000 and 2005 was negative for only the second, fourth, and fifth deciles of per- capita income, the growth in remittances was positive and substantial, averaging 106 percent, across all deciles of per-capita income in the 2005-2010 period. In fact, over both the 2000-2005 and 2005-2010 periods, the 34 percent increase in income, resulting from the 106 percent increase in international remittances, was the major driver of non-labor income growth. 6. Remittances have been rapidly growing and, particularly over the last five years, this growth has been shared across the entire income distribution, suggesting that remittance income may have been a key driver in poverty reduction. The results in Table A3-2, however, temper this interpretation. The table reports changes in the relative contributions of each component of income. The percentage changes in the income share of foreign remittances are impressive for both periods, especially the later period (see fourth and eighth columns of Table A3-2 for the 2000-2005 and 2005-2010 periods, respectively). Yet, the share of non-labor income was only about 26 percent of total income in 2000 and actually dropped to 18 percent in 2010, while the share of labor income grew from 74 percent in 2000 to 82 percent in 2010. 44 7. Furthermore, the percentage changes in the share of non-labor sources of per-capita income were negative for both periods (-9 percent and -22 percent for the 2000-2005 and 2005-2010 periods, respectively), and the negative changes were particularly notable for the lower deciles of the income distribution (Table A3-2). Trends in the income components suggest that much of the improvement experienced in the lower half of the distribution has come from increases in the shares of labor income over time. Non-labor income shares have actually declined for the lower half of the distribution. 8. An alternative explanation for the observed poverty reduction Figure 3-1: Demographic Changes is given by demographic changes. Population Growth and Population ages 15-64 On the one hand, population (annual percentage change and percent of total population) growth has considerably slowed 2.00 1.85 64.1 65.0 down. On the other hand, the youth 1.80 64.0 bulge observed in earlier periods % of Total Population Annual growth rate % 1.60 1.40 63.0 has now reached working age 1.40 61.4 62.0 1.20 1.12 61.0 (Figure 3-1).50 These simultaneous 1.00 58.7 60.0 changes are evidenced by the 0.80 59.0 observed decrease in average 0.60 58.0 household size as well as the 0.40 57.0 increase in the share of adults per 0.20 56.0 household (Figure 3-2 and Table 0.00 55.0 A3-3). A greater number of adults per household generally implies lower dependency rates and, under Population ages 15-64 Population growth the assumption that the labor market is able to absorb these Source: WDI, 2011. potential new entrants, the possibility of higher consumption per-capita. Figure 3-2: Demographic Characteristics A. Household sizes have declined… B...while the share of adults has increased 70.0 6.0 5.2 68.4 4.8 5.0 4.5 68.0 66.3 4.0 3.13 3.07 2.94 66.0 3.0 2.0 64.0 63.4 1.0 62.0 0.0 2000 2005 2010 60.0 2000 2005 2010 Average household size Number of adults per household 2000 2005 2010 Source: Own estimates based on HIES 2000, 2005, and 2010. 50 Despite this deceleration, Bangladesh has added 19 million people to its total population, a 15 percent increase between 2000 and 2010. See Chapter 4 for further details. 45 9. Moreover, as described in greater detail in Chapter 4, simple summary statistics reveal that, despite the observed population growth, labor force participation remained relatively stable and the employment-to-population ratio increased over the decade, particularly for women between 2005 and 2010 (Figure 4-4 and Table A3-3).51 Figure 3-3: Changes in the Structure of Working Age Population A. Education levels (percent change of working B. Urban Population as a Share of Total (percent age population) change working age population) 40.0 40.0 28.4 28.5 30.0 30.0 19.2 22.6 20.0 10.9 7.7 20.0 10.0 5.4 10.0 4.8 0.0 0.0 -10.0 -5.3 -5.0 -1.8 -10.0 -20.0 -13.1 -6.3 -7.9 -17.6 -20.0 Illiterate & Complete primary Higher secondary Incomplete primary & lower secondary & Terciary Rural (percent of total) Urban (percent of total) 2000 vs 2005 2005 vs 2010 2000 vs 2010 2000 vs 2005 2005 vs 2010 2000 vs 2010 Source: Own estimates based on HIES 2000, 2005, and 2010. 10. At the household-level, the share of occupied (working) adults increased (Table A3-3), pointing to a potential increase in consumption attributable to higher employment. An important question then becomes: how important was the effect of higher employment on the observed decline in poverty during the past decade? In addition to higher employment rates, another possibility is that the income derived from working also increased. As the growth incidence curves show (Figure 3-3), labor income per-capita has grown throughout the decade, growing at a faster rate in the latter half of the decade. The top of the income distribution saw labor incomes grow faster than the bottom of the distribution. Nevertheless, growth across the income distribution was substantial, suggesting that labor income had an important influence on moving people out of poverty. 11. Another potential contributor to the observed poverty reduction is the shift in the educational structure of the population. Over the last ten years, improvements in the educational composition of the workforce are evidenced by a reduction in the percentage of illiterates as well as an increase in the share having completed primary and lower secondary school (Figure 3-4.A). Furthermore, we observe an urbanization process, evidenced by large population shifts from rural to urban regions occurring during the first part of the decade (Figure 3-4.B). During the second half of the decade, however, educational improvements and urbanization slowed down. 12. Changes in the occupational structure of the labor force also occurred, as workers moved away from daily work and self-employed activities and instead toward salaried employment, where labor incomes and productivity are likely to be higher (Figure 3-4.A).52 In addition, consistent with trends for 51 In order to maintain comparability and consistency across HIES rounds and with the previous poverty assessment (World Bank 2008a), we follow the same definitions for labor force participation and employment. These definitions are based on the income section of the questionnaire. For instance, a person is classified as employed either if he (she) self-identifies as being an unpaid family worker or if any wage or self-employed income is recorded. See Paci and Sasin (2008), World Bank – Bangladesh (2008), and Sasin (2009). 52 See Chapter 4 for further details. 46 Figure 3-4: Changes in the Structure of Employed Population A. Occupational Structure B. Economic Sector (percent change in employed population) (percent change in employed population) 30.0 25.7 30.0 20.9 20.0 20.0 14.5 11.6 12.6 10.0 10.0 6.5 5.6 3.7 2.7 0.3 0.0 0.0 -3.6 -3.3 -2.4 -10.0 -6.5 -10.0 -5.5 -8.8 -10.0 -20.0 -14.9 Daily workers Self-employed Salaried -20.0 Agriculture Manufacturing Services 2000 vs 2005 2005 vs 2010 2000 vs 2010 2000 vs 2005 2005 vs 2010 2000 vs 2010 Source: Own estimates based on HIES 2000, 2005, and 2010. indicators of macroeconomic activity (see Figure 1-10), employment shifted away from agriculture and toward manufacturing and services, although this shift decelerated during the second half of the decade (Figure 3-4.B). 13. With regard to transfers, Figure 3-5 shows that both public and private transfers have Figure 3-5: Non-Labor Income Growth grown steadily, the former increasing by 65 Subsidies and Other Social Transfers and percent over the decade. Although the substantial International Remittances (percent of GDP) increase in transfers suggests that the growth in 13.0 Social transfers 10.8 non-labor income serves as a partial explanation 11.0 for some of the observed reduction in poverty, the International 9.0 7.2 benefits from such transfers are not sharply remittances % of GDP focused on the poor. For example, as shown later 7.0 in the report, poor targeting of public resources 5.0 4.2 3.9 significantly reduces the contribution of current 2.9 safety net allocations to the poverty reduction 3.0 2.4 effort (see Chapter 8). Moreover, significant 1.0 constraints, such as the high out-of-pocket cost of 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 migration and the need to rely on informal sources of finance, highly skew access to migration opportunities and international remittances in Source: WDI, 2011. favor of upper income groups (World Bank 2012a). 14. Finally, Figure 3-6 shows that, on average, the consumption-to-income ratio increased in the first half of the decade and then fell in the second half.53 The same pattern was observed for the lowest decile of the income distribution. Over the decade, the ratio declined on average and, for the first decile, which is of particular relevance for the poverty decompositions. While the change in this ratio may be due, in part, to measurement error, the overall change is consistent with an increase in savings rates observed in National Accounts. (Figure 3-7). 53 The consumption-to-income ratio reflects both the average propensity to consume as well as measurement error in both the income and consumption variables. Consumption-to-income ratios are presented in Table A3-4. 47 Figure 3-6: Change in Household Consumption-to- Figure 3-7: National and Domestic Savings (% of income ratio GDP at Current Market Prices) 20.00 35 10.00 30 - 25 (10.00) 20 (20.00) -15.20 -21.78 15 Gross Domestic Savings (30.00) Gross National Savings -33.67 10 (40.00) 2000-01 2001-02 2002-03 2003-04 2004-05 2005-06 2006-07 2007-08 2008-09 2009-10 2010-11 2000 vs 2005 2005 vs 2010 2000 vs 2010 Average 1st Decile 10th Decile Source: HIES 2000, 2005, and 2010. Source: BBS and Bangladesh Bank. 15. To the extent that households consumed a lower share of their income at the end of the decade when compared to the beginning of the decade, the observed changes in consumption may have been less dramatic had the consumption-to-income ratio remained constant. Since poverty is measured, in part, by consumption, poverty rates would have been lower in the later period than they actually were had the consumption-to-income ratio remained constant. 3.2. Decomposing Consumption Expenditures 16. In contrast to methods that focus on aggregate summary statistics, the micro-decomposition methods used in this chapter generate series of simulations of entire counterfactual distributions to account for the contributions of demographics, labor incomes, and non-labor incomes to poverty reduction. Underlying the decomposition is a simple model of household consumption. In particular, consumption per-capita in household h is defined by: [ ] (1) where n is the number of people in household h, is the consumption-to-income ratio, and y represents the total income of household h.54 17. First, we decompose the contribution of changes in real per capita consumption expenditures to poverty reduction following the Paes de Barros et al. (2006) methodology. In particular, poverty reduction is divided into the three components driving distributional changes in consumption expenditures: a household’s marginal propensity to consume; real income per capita; and the number of household members (see Figure 3-8, part A). This basic decomposition helps to estimate the relative contributions of each of the three components to changes in real per-capita consumption and, consequently, to the observed poverty reduction over the decade. Results from this simple decomposition are presented in Figure 3-9.A. 54 In the empirical specification, income is further disaggregated into its components. Please refer to Appendix 4. 48 Figure 3-8: Decomposing Consumption Per-Capita Part B Propensity to consume & measurement error Share of occupied adults Share of adults Consumption per- Labor income per Income per-capita capita adult Income per adult Labor income per Non Labor occupied adult Number of income per adult members n Part A Source: Inchauste and Olivieri (2012). 18. Changes in income and the share of the adult population played almost equal roles in reducing both the prevalence as well as the severity of poverty. Total income per adult was the most significant factor explaining the observed poverty reduction over the 2000-2010 decade. The change in demographic factors is the second more important factor in explaining the reduction in poverty. The Paes de Barros decomposition estimates the overall role of demography to be greater than what we find in the proceeding analysis, but both analyses find demography to be one of the top two factors that explain the reduction in poverty. Additionally, the Paes de Barros decomposition indicates that demographic changes were relatively more important factors in reducing the severity of poverty (relative to the prevalence of poverty). Figure 3-9: Contribution to Poverty Reduction 2000-2010 A. Changes in income and the share of adult B. Labor income was the single most important population both played large roles in poverty contributor to poverty reduction. reduction over the decade 250 Other Non-labor 115 200 64.0 Percentage of total poverty Transfers 63.2 150 64.9 152.0 100 103.9 126.1 Capital 65 reduction 50 66.8 40.6 44.4 49.8 Labor Income 49.3 58.1 0 15 -50 Occupation share -14.2 -21.3 -100 -30.8 Adult population -150 -35 Head count Gap Severity Head Gap Severity Consumption Income Consumption Income Ratio Adult population Income count Ratio Source: HIES 2000, 2005, and 2010. 19. Lastly, the negative contribution associated with changes in the consumption-to-income ratio suggests that if this ratio had remained constant over the period, poverty reduction could have been even larger than actually observed. Next, as depicted in part B of the diagram presented in Figure 3-8, the 49 contributions of changes in the components of income to poverty reduction are analyzed. The results from this decomposition, presented in Figure 3-9.B, show that labor income was the single most important contributor to poverty reduction. Demographics, primarily a declining dependency ratio, also played a large role in poverty reduction. In particular, a greater share of adults and occupied adults in the household helped families to reduce both the prevalence as well as the severity of poverty by more than 40 percent. 20. Evidence for the declining dependency ratio and the overall decline in the growth rate of fertility are important and robust elements of the story to explain poverty reduction over the decade. Nonetheless, the precise level of the estimated effect is sensitive to assumptions about adult equivalence scale adjustments and household economies of scale. Assumptions about the relationship between basic needs and the composition and size of the family are implicit in any analysis of poverty based on household survey data. In this report, and in the official definition of poverty for Bangladesh, the use of per-capita consumption as the measure of individual well-being assumes no economies of scale in household consumption nor does it adjust for differences in needs across children and adults.55 21. The question of whether or not economies of scale in household consumption exist can be addressed by a basic question -- does a family of four require twice as much as a two-person family to meet basic needs? As one example, fixed costs, like shelter, might not change (or might not change in proportion) with changes in household size, so we expect some economies of scale. Alternatively, food consumption (or consumption of ‘rivalrous’ goods) is directly proportionate to household size and has limited scope for economies of scale.56 Vleminckx and Smeeding (2003) and Buhmann et al. (1988) provide examples of how both adjustments can affect poverty analysis. Based on existing literature, we assume that standard adjustments for economies of scale and adult equivalence only modestly reduce the importance of declining household size and dependency ratio in the measured poverty reduction. However, a caution against this assumption is that existing literature is based on whether or not the adjustments affect the measure of poverty, and not whether they affect the difference over time in measured poverty. For our decomposition, the sensitivity to family size and composition could be quite different from the literature cited above. 22. Transfers, which include domestic and international remittances, played a relatively smaller role in reducing poverty. The HIES data indicate that the largest share of remittances is received by those who are well–off; specifically, about 80 percent of the value of remittances are received by the top 30 percent of individuals of the consumption distribution (see Appendix Figure D3-2 for details). This statistic alone signals that remittances are likely to play a relatively small role in poverty reduction. In the next part of this section, we consider results using a different methodological approach based on Bourguignon et al. (2008). Following this method, the role of remittances is larger than what we find using the Paes de Barros decomposition, but it is still significantly less important than labor income and household composition. 23. However, the aforementioned finding has several associated caveats. One potential caveat is that remittances may play a significant role in the overall growth of the economy, despite its small role in poverty reduction. Remittances have been rapidly increasing (see Figure 3.6) and are a significant income source for the non-poor, and the World Bank (2012a) suggests that they are an important income source for continued economic growth. A second caveat is that the poverty decompositions only capture the direct effects of remittances. Although remittances do not appear to have a large, direct role in poverty reduction, they may have significant secondary effects. As one example of a secondary effect, remittances 55 Banks and Johnson (1994) and Jenkins and Cowell (1994) discuss how adult-equivalence adjustments can affect poverty measurement. 56 For a detailed discussion of poverty and family size, see Lanjouw and Ravallion (1995). 50 that are received by someone who is not poor and then used to build a house will not have any direct effect on poverty. But, construction of the house would likely employ low-skill, poor laborers, and this employment would have a positive effect on poverty reduction. Further, if large inflows of remittances are distributed to the relatively well-off in a particular village, it may increase demand for low-skill labor, driving up the market wage (and thereby improving the well-being of all daily laborers). 24. Changes in other non-labor income appear to have had a negative effect on poverty over the decade, driven primarily by changes in the “Other non-labor” component (which includes gratuity, retirement, and other cash or in-kind receipts). Similarly, a change in the ratio of consumption-to-income (Figure 3.9) also negatively contributed to poverty reduction. One interpretation of this last finding is that a decline in consumption, relative to income, could indicate an increase in savings, which could establish the foundation for poverty reduction in the future. 25. In our next step, we utilize the Bourguignon et al. (2008) decomposition methodology in order to distinguish the importance of changes in earnings and consumption attributable to changes in educational attainment, age, gender, occupational, sectoral, and geographical distributions of the labor force. 57 The steps underpinning this decomposition method are outlined in Box 3-1. 26. During the last decade, the most important contributor to poverty reduction was the growth in labor income, predominantly in the form of farm income. This growth in labor income was driven by greater returns to individual and household endowments (Tables A3-11 and Figure 3-10), which point to an increase in the relative price of labor and higher productivity as the main contributors to poverty reduction. In particular, returns to farm and non-farm endowments accounted for 64 percent of the reduction in poverty, with 47 percent occurring in the farm sector. The fact that the reduction in poverty was concentrated among rural, agriculture-based households may be counterintuitive at first, as conventional wisdom calls for diversification to reduce the risk of falling into poverty. However, this puzzle is partially explained by the fact that the increase in returns was likely the result of an exogenous price shock (see Chapter 7). As described below, demographic changes also helped to reduce poverty, accounting for about 25 percent of the reduction in poverty between 2000 and 2010. Finally, non-labor income, in the form of international remittances, accounts for about 11 percent of the observed decline in poverty. 27. Beyond these broad results, the decompositions highlight significant differences in the contributors to poverty reduction between the first and the second halves of the decade. In the 2000-2005 period, the observed reduction in poverty was related to employment diversification; whereas in the 2005- 2010 period, poverty reduction was associated with increasing returns to farming. In particular, during the first half of the decade, the contribution attributable to higher returns to non-farm endowments were twice as large as the contribution of returns to farm endowments. In contrast, in the second half of the decade, returns to non-farm endowments amounted to only one-tenth of the contribution of higher returns to farm endowments. The results hold when analyzing marginal contributions (i.e. changing only one factor at a time) as well as cumulative contributions (simultaneously changing all factors). 28. During the first part of the decade, the higher returns to non-farm endowments coincided with significant sectoral shifts in employment, in particular, from agriculture to manufacturing and services. These shifts suggest that growth in the manufacturing and services sectors attracted workers to activities with higher returns, thereby having a greater impact on individuals in non-farm households. Relative to workers from farm households, workers from non-farm households likely had more suitable 57 The details pertaining to the derivation of the final empirical specification are presented in Appendix 4. 51 characteristics, networks, or proximity to the types of jobs the economy was creating; these differences likely explain why workers from non-farm households benefited more. Box 3-1: Decomposing Poverty Based on Bourguignon et al. (2008) In order to quantify the contribution of each income component to poverty reduction, the decomposition is implemented in four stages. First, the determinants of occupational choice, sectoral choice, and education level are separately estimated for each period. Tables A3-5 and A3-6 report both the actual shares for each survey year along with the simulated shares for household heads and for other family members, respectively. 58 Overall, the simulated shares are close to the true shares, indicating that the underlying specifications of these models can be used to simulate shifts in the composition and structure of the labor force. 59 As a second step, labor income is separated into farm and non-farm income in order to estimate the earnings equation for each period, estimated separately for household heads and for other household members. Regressions are run for each of four groups: (1) salaried; (2) self-employed; (3) daily workers; and (4) net farm revenue for farm households. Tables A3-7 and A3-8 present results for individuals engaged in non-farm activities. Table A3-9 presents results corresponding to net revenue for farm households.60 In the third step, coefficients from the second step regressions are used to simulate counterfactual distributions by changing one variable at a time and by observing the effect of each change on the distribution. The methodology for obtaining these counterfactuals is detailed in Appendix 4. Fourth, the counterfactual poverty rates are compared to the observed poverty rates in order to quantify the impact of each element on poverty reduction. Since applying the first period parameters to the last period data will yield results that are different from applying the last period parameters to the first period data, the counterfactuals are calculated in both directions for every pair of years, and the average counterfactual is reported. In effect, we estimate the marginal contributions of each factor since the parameter estimates are obtained by changing one element at a time, while leaving all other elements constant. Alternatively, given that changes in multiple factors could have interaction effects, the cumulative effect of these decompositions is also computed. This effect is calculated following the methodology proposed by Bourguignon et al. (2008), which entails: (1) sequentially calculating the effects of changes in age and gender, followed by changes in geographical, educational, occupational, and sectoral structure of the population on poverty; and (2) using the results from (1) to sequentially calculate changes in farm and non-farm earnings (on account of changes in the returns to these characteristics), changes in non-labor incomes, and changes in the consumption-to- income ratio.61 29. While the shift to manufacturing and services continued, the growth in agriculture gained momentum, averaging a 4.3 percent growth rate during the second half of the decade (as opposed to only 58 Tables A3-13 and A3-14 (in the Tables Annex) present multinomial logit regressions for occupational choice for household heads and for other members, respectively. Given the considerable diversification of income sources common to rural households (see Davis et al. 2010), we estimate the sectoral choice model for the secondary occupation of individuals in farm households. Results of these regressions are available upon request. Similarly, the educational and sectoral choice models are available upon request. 59 P-values of Pearson Chi-squared tests confirm that no simulated distribution is statistically different from the actual distribution. 60 Net revenue was calculated using the information available on total revenue from agricultural production and the cost of inputs from the detailed household enterprise modules included in the surveys. 61 For more details on this decomposition approach, as well as its potential caveats, refer to the background paper by Inchauste and Olivieri (2012), commissioned for this poverty assessment report. 52 2.7 percent as in the first half of the decade).62 Moreover, an increase in the relative price of commodities (see Chapter 7), as opposed to an increase in productivity, may potentially explain the observed increase in the value of farm labor during the second half of the decade. Demographic and Regional Effects 30. Strong growth in the working-age population and, consequently, lower dependency rates (see Chapter 5) also impacted poverty reduction throughout the decade. As shown in Figure 3-10 and Tables A3-11 and A3-12, changes in the age, gender, and regional composition of the workforce accounted for 25 percent of the observed decline in poverty between 2000 and 2010. Given that mostly young people, who generally earn less than their more experienced counterparts, entered the workforce, the increase in size of the workforce was countered by lower returns to work for young workers (Table A3-11). Nevertheless, the ensuing lower dependency rate was a large contributor to poverty reduction. Similarly, the growing share of women in the labor force also contributed to poverty reduction, and the marginal effect of this change was larger in the first half of the decade, particularly in the non-farm sector (Table A3-11). Figure 3-10: Cumulative Contributions to Poverty Reduction, 2000 – 2010 (percentage change in poverty) -9 Other -8 Residual Non-farm 0.4 Residual Farm 2.9 Sector 5.1 Education 8 Occupation 11.2 International Remittances 16.9 Returns non-farm 25.4 Demographics 47.1 Returns Farm -20 -10 0 10 20 30 40 50 Source: HIES 2000, 2005, and 2010. 31. In terms of regional composition, the share of the working-age population living in Barisal declined by 14 percent between 2000 and 2010. Similarly, Chittagong and Sylhet experienced small declines in their work-forces, particularly in the second half of the decade. In contrast, Dhaka and Rajshahi experienced small increases in their working-age population shares. How did these changes contribute to poverty reduction? As discussed in Chapter 10, the effect of demographic changes on poverty reduction was slightly larger in the West, in conjunction with an increase in the working-age population. Growth in labor income was the most important contributor to poverty reduction in both 62 Although this increase in growth was not enough to increase agriculture’s share of total GDP, it was large enough to maintain the same share as the first five years of the decade. 53 regions, but slightly more important in the West, which is consistent with the fact that agriculture is the primary economic activity in the West.63 32. Together, the effects of demographic and regional changes can at least partially explain some degree of poverty convergence between the East and West. Table A3-11 shows that the earnings penalty for living outside of Dhaka fell over the last decade. This reduction in the earnings penalty points to an increase in the relative price of labor and/or higher productivity outside of Dhaka as significant poverty- reducing factors. This effect largely reflects the effect for farm households, where the increase in returns to labor accounts for 10.5 percent of the reduction in poverty throughout the decade. Higher returns to non-farm work in cities outside of Dhaka contributed an additional five percent to overall poverty reduction over the course of the decade, particularly during the first part of the decade. Education and Experience 33. A more educated and experienced population helped to reduce poverty, particularly in the non- farm sector. Improvements in educational attainment of the workforce accounted for five percent of the observed poverty reduction between 2000 and 2010 (Figure 3-10 and Table A3-12). However, the poverty reduction attributable to higher educational attainment in the non-farm sector primarily occurred during the early part of the decade (Table A3-10). Interestingly, while the large increase in the share of the population that completed primary school (Table A3-3) led to a slight reduction in poverty in the non- farm sector, it also drove down the premium for completing primary school for nearly all, excepting salaried, workers. As a result, the poverty-reducing impact of a more educated workforce was partially offset by a decline in the returns to primary education (Table A3-11). 34. Since the baseline group (omitted category) is comprised of individuals with no schooling, these results indicate an increase in demand for unskilled workers over the course of the decade, which served as an important poverty-reducing factor (see coefficient corresponding to the constant in Table A3-11). This increase in demand affected both farm and non-farm daily and self-employed workers (Tables A3-7 to A3-9). During the same period, salaried workers saw their returns to education increase as well (see Tables A3-7 and A3-8). Overall, the results point to an important increase in the relative price of unskilled labor, which may, at least in part, have been driven by higher food prices, particularly in the case of agricultural workers.64 Occupation 35. As workers sought to benefit from better work opportunities, changes in the occupational structure of non-farm workers also played an important role in poverty reduction. In the non-farm sector, the shift from unpaid, daily, and self-employed work toward salaried employment accounted for 9.3 percent of the reduction in poverty over the decade (Table A3-10).65 In the farm sector, most households are diversified, often participating simultaneously in farm and non-farm activities, so the secondary occupation of household members is important.66 To account for this diversification, we employ an 63 The East region includes Dhaka, Chittagong, and Sylhet; the West region includes Barisal, Khulna, and Rajshahi. The majority of daily and self-employed in farm workers are located in the West while salaried and self-employed, non-farm workers are in the East. 64 See Chapter 7 for a more detailed discussion of the impact of changes in food prices on agricultural wages. 65 The cumulative effects for the entire decade are larger because they include all occupational changes, including the choice between self-employment, daily, and salaried work for non-farm workers and the choice to have a secondary occupation for farm workers. 66 See Davis et al. (2010). 54 occupational choice model for the secondary occupation and use counterfactual simulations in order to understand how changes in secondary occupations contributed to poverty reduction. The results show that the share of farmers with a secondary occupation fell from 30 percent to 10 percent between 2000 and 2010. This decline may be due in part to a substantial increase in the returns to farming. However, it also potentially reflects the idea of greater migration as a form of diversification, a hypothesis that is substantiated by the growing number of farm households headed by a woman, which increased from under 10 percent in 2000 to almost 22 percent in 2010. Lower diversification among farm household members accounts for a slightly higher poverty rate than the rate that would have prevailed had diversification remained constant at the end of the decade. Sector of Work 36. Changes in the sectoral composition of employment also affected poverty reduction. In particular, non-farm households shifted away from agriculture and into manufacturing and services; this shift was associated with a five percent reduction in poverty over the first half of the decade (Table A3-10). In contrast, as the returns to agriculture increased in the second half of the decade, movements away from agriculture led to a slight increase in poverty. Moreover, the returns to work in manufacturing and services substantially declined over the decade. Although the shift into the manufacturing and service sectors accounted for three percent of the observed poverty reduction between 2000 and 2010 (Table A3- 10), the gain was entirely erased by a reduction in the manufacturing and service sector wage premium (Table A3-11). Farm Net Revenue and Rural Assets 37. Greater returns to farm households’ characteristics and endowments imply an increase in the real value of output per worker. This increase holds for all farm households, regardless of characteristics, as expressed by the 45 percent contribution to poverty reduction captured by the constant (Table A3-11). However, the fraction of this increase attributable to an increase in real productivity resulting from higher capital investments, as opposed to from an increase in relative prices, cannot be distinguished. Nevertheless, given that this period was characterized by an increase in commodity prices, the effect of this price spike on the real value of agricultural production likely was an important driver of the observed increase in agricultural returns. 38. In addition to the overall increase in returns to farming experienced by farm households, the most important contributor to poverty reduction was the increase in returns to land, accounting for 42 percent of the reduction in poverty (Table A3-11). This increase is perhaps due to the fact that land is becoming scarcer.67 Still, the contribution of returns to land to poverty reduction was only slightly offset by lower returns to irrigation and a greater number of workers. Non-labor Income 39. While much of the reduction in poverty was due to labor income growth, non-labor income also played a role, albeit relatively smaller. On the one hand, the increase in international remittances over the decade was associated with an 11 percent decline in poverty, most of which occurred during the second part of the decade.68 On the other hand, declines in domestic transfers as well as other non-labor factors (i.e. capital and other transfers) were associated with a slight increase in the poverty rate (Table A3-10). 67 Between 2000 and 2010, average land size declined from 0.8 to 0.6 acres per-capita. 68 This pattern coincides with the stronger growth in international remittances taking place over the second half of the decade (Table A3-1) but could also be influenced by better micro-data collection over time. For instance, the last round of HIES (2010) includes a new module on migration and remittances which has likely improved 55 3.3. Conclusions 40. The objective of this chapter is to account for the contributions of demographic, labor, and non- labor income factors to the reductions in poverty observed over the last decade. Results from the decomposition exercises show that the most important contributor to poverty reduction was the growth of labor income. Similar to the findings presented in Chapter 1, the analysis presented in this chapter also reveals stark contrasts in the nature of poverty reduction experienced in the first relative to the second part of the decade. 41. During the first part of the decade, the increase in non-farm wages was the most important factor contributing to poverty reduction. Paralleling this wage increase, three “poverty-reducing” shifts also took place: (1) workers moved away from agriculture and toward manufacturing and service sector employment; (2) workers moved away from daily and self-employed work and toward salaried jobs; and (3) an increase in the level of education of the workforce. In contrast, during the second half of the decade, poverty reduction occurred primarily in the farm sector. In particular, the farm sector experienced a significant increase in labor income, which was not associated with higher education or changes in occupation; rather, it was associated with a decline in the earnings’ penalty for living outside of Dhaka. 42. Apart from labor income growth, changes in the demographic composition of the population, in particular, lower dependency ratios on account of the larger adult population, also helped to reduce poverty over the decade. Likewise, international remittances played a significant, albeit smaller role in reducing poverty. Experience from the region indicates that well-targeted public transfers to the poor can be very effective in reducing poverty, particularly extreme poverty (IFPRI 2007), and this evidence suggests that Bangladesh can achieve further poverty reduction through improved design of the system of safety net programs. the quality of remittance data. However, this raises a problem of comparability between this measure and previous rounds of the survey. 56 4. Labor Income: Linking the Labor Market and Poverty Reduction 1. The decomposition results from Chapter 3 reveal that the growth of labor income is the most important contributor to poverty reduction over the last decade. In order to interpret this finding in a broader context, this chapter provides an overview of the main changes that took place in the labor market over the decade. The goal of this chapter is to gain a better understanding of the transmission channels between growth and poverty reduction. 2. A comprehensive overview of labor market linkages to poverty reduction necessitates analyzing data on growth, employment, productivity, and poverty. Yet, no single data source encompassing both micro- and macro-level data exists. To overcome this data challenge, the analysis presented in this chapter combines data from several sources. Information on economic growth is derived from the System of National Accounts (SNA). Employment and labor income data come from both HIES and the Labor Force Surveys (LFS). Finally, poverty estimates are derived using consumption data from HIES. In order to validate the usage of various data sources, an assessment of their comparability and compatibility, both across surveys and over time, is presented in the Data Annex for Chapter 4. Overall, the exercise suggests that the surveys provide a relatively strong basis for the analysis undertaken in this chapter. 3. The remainder of this chapter is organized as follows. Section 1 aims to establish selected links between economic growth and the labor market. Section 2 then presents information on the labor supply in Bangladesh by giving an overview of the labor force, including demographic and labor force participation trends, occupational types, and educational attainment. Section 3 examines labor demand by discussing the structural changes in economic activity that occurred during the past decade. Using data from enterprise and microenterprise surveys, the section also considers constraints to employment growth. Section 4 analyzes trends in earnings and wages and attempts to understand earnings and wage differences by gender and type of work. Finally, Section 5 summarizes the main findings and concludes. 4.1 Linking Economic Growth to the Labor Market 4. As mentioned in Chapter 1, Bangladesh experienced steady and strong GDP growth, averaging a rate of 5.8 percent per year, over the period from 2000 to 2010. Agriculture fell as a share of GDP, whereas the industry (mainly manufacturing) and service sectors grew (Figure 1-10). These changes suggest that Bangladesh’s economy continues its transition from a predominantly agriculture-based economy to a manufacture-based economy. 5. Consistent with the growing industry and service sectors, several important structural changes occurred in the labor market between 2000 and 2010. The main processes were: (i) a gradual decline of agriculture and a rise of services; (ii) employment growth in urban areas; and (iii) a movement away from agriculture, toward industry and services. 6. Although the agricultural sector labor force has been shrinking over the past decade, the sector has experienced an upswing in income growth during the last half of the decade. The share of agriculture in total employment declined from 51 percent to 48 percent between 2000 and 2005, and dropped to 47 percent in 2010 (Figure 4-1.A). While total labor income from agriculture grew at only 1.7 percent annually between 2000 and 2005, it grew by an astounding 9.8 percent a year between 2005 and 2010 (Figure 4-1.B). This income growth for agricultural labor is a strong reminder that the sector continues to play an important role in Bangladesh’s economy, employing nearly one-half of the country’s work-force and providing over 45 percent of total household income. Figure 4-1: Sectoral Employment A. Sectoral Composition of Employment B. Annual Growth in Income by Sector (percent of total employed) (annualized growth rate of total income) 60 15.0% 51.3 2000-05 46.6 2003 2005 2010 50 10.0% 2005-10 40 34.7 35.2 2000-10 30 5.0% 18.2 20 14.0 0.0% 10 0 -5.0% Agriculture Industry Services Agriculture Industry Services Source: Labor Force Surveys. Source: HIES 2000, 2005, and 2010. 7. The industry sector’s share grew from 14 percent to 18 percent of total employment between 2000 and 2010, while the share of employment in the service sector remained constant at about 35 percent (Table A4-1). Unlike the agriculture sector, real incomes in the industry and services sectors declined in the second half of the decade. For the industry sector, the sectoral decline in real income that occurred in the second part of the decade more than offset the gains that took place in the first part of the decade (Figure 4-1.B). 8. In terms of the Table 4.1: Overview of the Labor Market geographic composition of employment growth, and as 2000 2003 2005 2010 Real Growthˠ described in Chapter 3, urban Population (million)* 133.4 137.4 149 1.63 areas experienced strong growth % urban * 23.4 23.6 22.9 -0.60 during the first part of the decade, * Working Age Population (million) 75 79.6 90 2.49 but this trend was partially % total* 56.3 58.0 60.0 0.68 reversed during the second half of * the decade (Table 4.1). Between Employment (million) 40.6 45.5 51.5 2.51 1999 and 2005, the urban Employment Rate* 54.1 57.1 57.2 0.04 population share grew from 21 Unemployment Rate * 2.38 2.47 1.66 -7.64 percent to 24 percent of the total Underemployment Rate *1 6.19 7.87 7.21 -1.74 population. However, the urban Hours Worked per Week* 38.63 44.21 46.07 0.83 share declined by one percentage Mean Earnings per Worker** 3060 3367 4141 4.23 point in 2010. The urban labor Labor income (% of Total) ** 75.51 78.33 84.7 1.58 market rapidly expanded during *2 the first part of the decade and Literacy Rate 51.26 61.78 61.54 -0.08 then slightly declined thereafter. Years of Education * 4.16 4.78 4.74 -0.17 In 2010, about 43 percent of labor Note: ˠAnnualized real growth, 2000-2010. income originated from urban Source: From Labor Force Surveys. From HIES; Working less than 20 hours a week. * ** 1 areas (Inchauste 2012), a decline from the 56 percent observed in 2005, but about the same share as observed in 2000. 9. In terms of regional dynamics, the proportion of the working age population increased by about one percent, on average, per year for both the East and West (Table A4-2). Notable exceptions were the 58 Chittagong and Sylhet divisions, which experienced growth of the working age population at rates slightly higher and lower, respectively, than average. Growth, Productivity, and Employment 10. To better understand the links between growth and the labor market, we examine the relative importance of labor productivity and employment growth in explaining GDP growth. A simple decomposition of per-capita GDP shows that increasing productivity (value added per worker) accounted for 71 percent of the growth in GDP per-capita over the course of the decade (Figure 4-2.A). As described earlier, GDP growth in Bangladesh was accompanied by a large increase in the labor force (9.8 million new workers between 2000 and 2010). Nevertheless, the employment rate (the share of employed among the working-age population) did not significantly change, a testament to the strength of the demand for labor, which was able to keep up with this large increase in labor supply. However, the contribution of employment to GDP per-capita growth was only eight percent. More important was the increase in the working age population as a share of the total population (i.e. the fall in the dependency ratio), which accounted for around 21 percent of the growth in per-capita GDP. 11. If productivity growth accounts for most of the increase in GDP per-capita, we need to understand where this growth occurred. As shown in Figure 4-2.B, productivity growth, measured as output per worker, has been especially strong in the industry sector. About 24 percent of the increase in total productivity is associated with the inter-sectoral mobility of workers, taking place primarily during the first part of the decade. As described later in the chapter, the most important channel was the outflow of workers from low-productivity, daily wage jobs in agriculture to jobs in services – a phenomenon related to rural-urban migration and the expansion of non-farm employment. In contrast, the service sector’s contribution to growth in output per worker increased substantially during the second half of the decade. Figure 4-2: Decomposition of GDP per-capita and Output per Worker, 2000 - 2010 A. Contribution to Changes in GDP per-capita B. Contribution to growth in Output per worker, by sector Share of working age 21% Agriculture 18% population Industry 35% Employment (rate) 8% Services 23% Output per worker 71% Inter-sectoral shift 24% Source: HIES 2000, 2005, 2010 and WDI. 12. The rest of this chapter seeks to answer several questions. What are the implications for economic and employment growth going forward? Will labor supply continue to expand at the same rate as has been observed in the past decade? Is the skill set of the work-force and the regulatory environment such 59 that productivity and employment growth will continue? What are the constraints to better job creation moving forward? 4.2 Supply Side: An Overview of the Labor Force 13. Even with the rapid expansion of its working-age population, Bangladesh’s labor market was able to absorb new entrants at an impressive pace. Similarly encouraging were the improvements in the educational attainment of the workforce (Table 4.1), especially since higher education (secondary education or more) is likely to play an important role as Bangladesh continues to develop and grow (Mankiw, Romer and Weil 1992, Barro 1991). Still, while these accomplishments must be underscored, persistent challenges remain; among them, low rates of female labor force participation and relatively high levels of underemployment. These issues are discussed in more detail below. Demographic Trends 14. Demographic trends in Bangladesh create opportunities as well as challenges for future growth and poverty reduction. Population growth has considerably slowed down over the last thirty years, declining from an average of 2.7 percent per year in the 1980s to an average of 1.4 percent per year in the 2000s. Despite the slowdown, Bangladesh has added 19 million people to its total population, a 15 percent increase between 2000 and 2010. The working-age population has been expanding more rapidly than the total population, growing at an average rate of 2.3 percent per year, indicating a 25 percent increase in the working-age population between 2000 and 2010 (Figure 4-3.A). The “bulge” among five to fourteen year-olds in the 2010 population (Figure 4-3.B) indicates that this working-age population trend will continue over the next decade. While a growing labor force can be an asset for income generation and growth, absorbing such a large wave of new entrants every year poses a major challenge for the labor market. Figure 4-3: Demographic Changes A. Population Growth and Working Age Population B. Population Pyramid (ages 15-64) 2.00 1.85 64.1 65 64 60-64 Female Male Annual growth rate % 1.40 63 % of Total Population 1.50 50-54 61.4 62 Age in Years 1.12 61 40-44 1.00 58.7 60 59 30-34 58 0.50 20-24 57 56 0.00 55 10-14 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 0-4 -10.0 -5.0 0.0 5.0 10.0 Population ages 15-64 Population growth Millions of People Source: United Nations World Population Prospects as reported in WDI. 15. The United Nations (UN) projects that the annual growth rate of the working-age population will increase to 2.2 percent until 2015, even as annual population growth slows down to a rate of about 1.3 percent. The relatively faster growth in the working age population will add an estimated 11 million people to the country’s potential workforce between 2010 and 2015. By 2020, the growth rate of the working-age population is expected to decline to 1.8 percent per year, adding another 10 million workers. 60 Figure 4-4: Labor Force Participation (percent of working age population) A. Labor Force Participation B. Female Labor Force Participation Rate across the distribution 100 70% 2003 82.2 79.6 2003 60% 2005 80 2005 50% 2010 2010 54.1 57.2 40% 60 30% 40 34.8 20% 25.3 10% 20 0% 1 2 3 4 5 6 7 8 9 10 0 Total Men Women Deciles of per-capita family income Source: Bangladesh Labor Force Surveys. Source: HIES 2000, 2005, and 2010. Labor Force Participation 16. Labor force participation of men remained relatively stable at around 80 percent during the 2000s.69 The labor force participation rate for women, on the other hand, increased from 25 percent to about 35 percent between 2000 and 2010 (Figure 4-4.A). In spite of this increase, women’s labor force participation in Bangladesh continues to be very low according to international standards.70 17. As in most developing countries, female labor force participation in Bangladesh increases at the higher end of the income distribution (Figure 4-4.B), among women who are also more likely to have Figure 4-5: Probability of Female Labor Force Participation A. ..falls and then rises with education B. ..is lower for married women with young children 100% 2003 2005 2010 80% 0% 2003 2005 2010 60% -10% 40% -20% 20% 0% -30% -20% -40% -40% -50% 1-5 years 6-10 years 10th 12th More than Number of Children under age 6 Married completed completed 12 Years Source: Estimates based on LFS Surveys. Results from a Probit regression of labor force participation for women. The reference case is unmarried, illiterate women with no children. See Table A4-4 for complete results. 69 Labor force participation is measured as a proportion of the working-age population, defined as people ages 15- 64. 70 The female labor force participation rate calculated using the last three rounds of the LFS is about 15 percentage points higher than the rate obtained using the HIES data. Notwithstanding this discrepancy between the two surveys, women’s participation in the Bangladeshi labor market remains low by international standards. 61 higher levels of educational attainment. Consistent with the findings reported in Khandker (1987),71 the relationship between education and female labor force participation is non-linear in Bangladesh. In particular, women with some primary education were less likely to participate in the labor force than illiterate women, whereas women with more than 12 years of education were most likely to participate (Figure 4-5). As shown later in the chapter, lower wages (relative to men) partially explain women’s low rates of participation. 18. Although the relationship between marriage and the likelihood of working substantially attenuated over the last decade, the probability that a woman participates in the labor market falls by 22 percent when she is married. Therefore, marriage is the most important determinant of female labor force participation (Figure 4-5.B and Table A4-3). Similarly, holding all else equal, women with young children are also less likely to participate in the labor force. These findings suggest that postponing marriage could substantially increase women’s welfare, especially in terms of potential lifetime earnings. Unemployment and Underemployment 19. The unemployment rate is slightly higher for men relative to woman, yet it is by no means alarming for either group. In fact, a bigger concern in Bangladesh is the relatively high level of underemployment.72 The Bangladesh Labor Force Surveys (LFS) show that while unemployment was less than three percent between 2003 and 2010, underemployment remained at an average of around seven percent during this period, falling only slightly between 2005 and 2010 (Figure 4-6.A and Figure 4-6.B). Figure 4-6: Unemployed and Underemployed Population (share of working age population) A. Unemployment Rate B. Underemployment Rate 4.0 2003 2005 2010 15.0 2003 2005 2010 3.0 10.0 2.0 1.0 5.0 0.0 Total Men Women 0.0 Total Men Women Source: Bangladesh Labor Force Surveys. Occupation types and characteristics 20. In 2010, labor income constituted nearly 85 percent of total household income and was comprised of three main components: wage employment (25 percent), self-employed earnings (56 percent), and daily labor (19 percent). During the same year, workers were engaged in five major activities: (i) daily wage labor; (ii) salaried jobs; (iii) non-agriculture self-employment; (iv) farming or “self-employment in agriculture”; and (v) unpaid family work. 71 Khandker (1987) investigates time-use of rural married women in Bangladesh. 72 Underemployment is defined as the proportion of the working-age population employed for less than 20 hours a week. 62 21. Daily wage labor, accounting for Figure 4-7: Employment by Occupational Type (percent of nearly 20 percent of all workers throughout total employed) the decade, consists of daily wage employment in agriculture (recruited mostly 50% from the rural landless) and outside of 2003 agriculture (Figure 4-7). On the other hand, 40% 2005 the share of salaried jobs increased from 2010 about 15 percent of total employment in 2000 30% to about 18 percent in 2010. Salaried workers consist of two distinct groups, government 20% workers and non-government workers. The 10% former has declined from about five percent of all workers in 2003 to three percent in 0% 2010. In contrast, the share of private sector Salaried Self-employed Day-laborer Unpaid Family workers has increased from two percent to Worker eleven percent over the same period. Wages Source: Labor Force Surveys. of salaried employees have remained fairly constant, while day-laborers have seen increases over time. 22. The non-agriculture, self-employed group consists of individual, own-account workers and employers. The share of non-farm, self-employed workers increased from about 22 percent of all workers in 2003 to 24 percent in 2010. In contrast, the self-employed in agriculture accounted for about 16 percent of all workers, down from 23 percent in 2003. Relative to other workers, self-employed workers have enjoyed wage increases as well as higher wages between 2000 and 2010. 23. Finally, unpaid family workers comprised nearly 22 percent of the working population. Because women often are the ones who work for their families without monetary compensation, the share of unpaid work is likely to continue growing as these workers join the work-force. To the extent that unpaid family work is highly substitutable with outside employment, the female work force represents potential as a growing and untapped resource going forward (Khandker 1987). Educational Make-up of the Labor Force 24. The rapid growth in school enrollment, particularly for females, had an important effect on characteristics of the Bangladeshi labor force (Figure 4-8). Although 39 percent of workers were illiterate, Figure 4-8: Education Level of Working Age Population Currently not in School A. Males B. Females 60 60 2003 2003 50 2005 50 2005 40 2010 40 2010 30 30 20 20 10 10 0 0 Bachelors or 1-5 Grade 6-8 Grade Illiterate 9-10 Grade SSC/HSC Bachelors or 1-5 Grade 6-8 Grade 9-10 Grade Illiterate SSC/HSC more more Source: Bangladesh Labor Force Surveys. 63 a steady decline in the share of illiterate workers corresponded with an increase in the share of workers with primary education and beyond. With respect to gender, the share of illiterate female workers declined from 55 percent in 2003 to 42 percent in 2010, whereas the share of illiterate male workers declined from 43 percent in 2003 to 35 percent in 2010. Figure 4-9: Educational Attainment Across the Distribution A. Educational Composition of the Workforce by B. Percentage Increase in Years of Education of the Decile Workforce by Decile 100 60% 2000-2005 2000-2010 40% 50 20% 0 0% 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10 Deciles of the Consumption Distribution Deciles of the Consumption Distribution Illiterate 1-5 years 5-9 years 9-12 years Above 12 Years Source: Bangladesh Labor Force Surveys and HIES. 25. As in most countries, educational attainment is highly correlated with income, i.e. higher income groups tend to have higher levels of education (Figure 4-9.A). Over the last decade, improvements in the educational attainment of the Bangladeshi labor force occurred across the entire population. In fact, the largest improvements in education were experienced by those in the lower and middle income groups (Figure 4-9.B). Improvements in the educational level of the workforce also coincided with an increase in income earned by individuals with lower levels of education. 26. As shown in Figure 4-10, Figure 4-10: Average Income Growth by Education Level average income for both men and women with fewer than 10 years of 20% 2000-05 education significantly grew between 15% 2005-10 2005 and 2010. However, the income 10% 2000-10 of those with greater than 10 years of 5% education remained relatively 0% constant. These patterns in average -5% income growth are consistent with the Men Women Men Women Men Women results presented in Chapter 3, which Illiterate to Less than 5th to 10 Grade Greater than 10th Grade show that those with relatively lower 5th Grade levels of education experienced the highest returns to their skills. Source: Bangladesh HIES 2000, 2005, and 2010. In 2005 Taka. 4.3 The Demand Side: Understanding the Constraints to Employment 27. Despite improvements in the average skill set of workers, we must consider whether these skills are appropriate, given the needs of the market, and whether firm expansion is likely to create the level of employment required to meet current demographic trends. This section examines the business environment in Bangladesh; in particular, whether investment and productivity growth are sustainable in order to facilitate the creation of quality employment opportunities. To achieve this purpose, we use the 2007 Bangladesh Enterprise Survey (BES), which collects information on over 1,400 firms across the country. The BES is a nationally representative, firm-level survey conducted periodically by the World 64 Bank in the developing world. It collects data from non-agricultural, privately-owned firms that employ at least five individuals. In the survey, firms are asked to respond to a series of questions related to the investment environment as well as their balance sheets and income statements. Because the survey is nationally representative, it is ideal for obtaining a broader picture of the investment environment in Bangladesh.73 Since a large share of the workforce is employed by microenterprises, the analysis is complemented by looking at a sample of microenterprises surveyed as part of the 2010 HIES. 28. The BES finds that business and regulatory requirements are not insurmountable. Labor, business, and trade regulations pose lower constraints than in the rest of the region and other developing countries. However, as shown in Figure 4-11.A, reliable access to electricity, corruption, and access to finance are much more problematic in Bangladesh as compared to the region and to the rest of the developing world. In particular, seventy-eight percent of firms identified access to electricity as a major constraint, and in 2007, over 50 percent reported owning a generator, which supplied about one-quarter of firms’ energy needs. The second biggest constraint was corruption, identified by 55 percent of firms; 85 percent of firms reported having to give gifts to public officials “to get things done”, compared to 37 percent for the rest of the region (Figure 4-11.B). The impact of corruption on the business climate cannot be overemphasized, as it directly affects investment decisions and, therefore, the potential for employment generation, as discussed below. 29. Third, access to finance was identified as a major constraint by 43 percent of firms, and interestingly, 25 percent of firms reported that lack of an adequately trained labor force was a major constraint. Although the country’s workforce has improved in terms of average education, these Figure 4-11: Business Constraints A. Major Business Constraints B. Share of firms expected to give gifts Electricity to get things done Corruption in meetings with tax officials Access to finance Tax administration to get a water connection Inadequately educated workforce Tax rates to get an import license Courts system to get an electrical connection Practices of informal competitors Crime, theft and disorder to get an operating license Customs and trade regulations Business licensing and permits to secure government contract Transportation to get a construction permit Labor regulations 0 10 20 30 40 50 60 70 80 90 0 10 20 30 40 50 60 70 80 90 Bangladesh South Asia All Bangladesh South Asia All Source: World Bank Business Enterprise Survey, Bangladesh, 2007. 73 For more details on the Bangladesh Enterprise Survey, please see the Data Annex for Chapter 4. 65 improvements do not necessarily imply that firms are able to find workers who possess the appropriate skill sets. The statistics in Figure 4.11 give a sense of the current business environment in Bangladesh. However, going forward, the most important factor for job creation and productivity growth is whether or not firms are willing to invest. The next section explores investment decisions in greater depth. Demand Side Investment Decisions: To Invest or Not to Invest? 30. After the economic liberalization of the 1990s, the private sector played a vital role in the growth and development of Bangladesh’s economy. Private sector investment was largely responsible for increasing the productive capacity of the economy, particularly in the later part of the decade. With public investment falling from 7.4 percent of GDP in 2000 to 5 percent of GDP in 2010, private investment, facilitated by a stable and more market-friendly environment, surged from 10 percent in the early 1990s to over 19 percent of GDP at present. Yet, very little is known about the factors that influence Bangladeshi firms’ decisions about levels of physical and human capital investment. Understanding the factors that affect the investment decisions of Bangladeshi firms can help policymakers generate an investment climate that facilitates inclusive growth. 31. Table A4-6 shows (probit) regression results for a firm’s first-stage decision of whether or not to invest.74 The dependent variable equals 1 if the firm invested and 0 otherwise. Bangladeshi firms face several investment constraints. The results show that a firm’s perception of the business climate is strongly correlated with the firm’s probability to invest. For example, while having products spoil en- route to markets or facing power outages are negatively correlated with the firm’s likelihood to invest, the presence of facilities that enhance productivity (e.g. in-house training, research staff in firms, use of computers and internet facilities) is positively correlated with investment. Although the estimated coefficient is not always significant, the unpredictability of law and order is negatively correlated with the probability to invest. Lastly, firm ownership type is also a significant predictor of firms’ investment decisions; in particular, having an owner who is also manager is positively associated with the likelihood of investing. The estimates are robust when employing regional fixed effects. 32. Table A4-7 presents regression results for the second-stage investment level decision, i.e. how much to invest.75 Conditional on investing, a firm’s investment level is clearly influenced by the investment environment and bottlenecks. For example, informal payments made to government officials and extortions are negatively related to a firm’s level of investment. In particular, a one percent increase in the ratio of informal payments to sales reduces investment by roughly eight percent. Although the value of the ratio of informal payments to sales is not large (ranging from zero to ten percent), the corresponding coefficient estimate is, nevertheless, sizable. Similarly, the belief that security is a threat is also negatively associated with a firm’s investment level. A one point decrease in the security measure (meaning firms believe security has improved) is associated with a 31 percent increase in capital deepening. 33. With regard to infrastructure, transportation problems are negatively correlated with the level of investment made by firms. A one point improvement in transportation is associated with a 20 percent 74 Not all firms made an investment in 2006; therefore, a sample selection model is employed to account for the selectivity of firms’ investment decisions. The empirical specification used is described in detail in Appendix 6. The dependent variable in the second-stage is the natural log of fixed investment per worker, a measure of capital deepening. Using full information maximum likelihood (FIML), coefficients for the firm’s propensity to invest and its level of investment are estimated. While the majority of control variables are categorical (see descriptive statistics in Table A4-5), these variables still offer insight into the investment decisions made as well as the investment atmosphere faced by firms in Bangladesh. 75 Second stage coefficients are estimated using the full information maximum likelihood (FIML) estimator. 66 increase in firms’ investment level. Since the hours of power outages in a day are positively correlated, and the number of generators owned by the firm is negatively associated with investment, the estimates suggest that firms are more inclined to invest in generators to ensure an uninterrupted power supply. Unlike electricity outages, water shortages are negatively related to a firm’s investment level. 34. Turning to human capital, the results indicate that firms with research staff, in-house training, and a more educated workforce have higher levels of investment, suggesting growth-boosting complementarities between human and non-human capital investment. Lastly, exporting firms do not appear to invest more than firms that supply only the domestic market. 35. Identifying the factors that affect a firm’s decision of whether or not to invest as well as its level of investment is of utmost importance in order to guarantee continued investment and employment growth in the future. The impact of corruption on investment decisions is especially problematic, given its relatively large negative impact on the expansion of firms. Greater investment can lead to labor productivity growth that is critical for the continued growth of labor incomes, which is a powerful force for poverty reduction, as seen in Chapter 3. Productivity and Constraints to Growth in Microenterprises 36. While the aforementioned results consider relatively large firms, a substantial portion of the workforce is employed by small firms and microenterprises. In fact, most self-employed, non-agriculture workers are microenterprises, scattered throughout the country. A sample of microenterprises, collected as part of the HIES in 2010, reveals that over 70 percent of these firms are in the service industry and over 90 percent employ between one and three workers. Despite a considerable increase in the urban share of such firms (growing from 27 percent in 2000 to 34 percent in 2010), microenterprises were located primarily in rural areas, with Dhaka and Rajshahi having the highest concentrations (37 percent and 26 percent, respectively). Households that run micro-firms are relatively well-off, given that only a small and declining share of these firms is run by poor households (Figure 4-12.A). Figure 4-12: Household Non-Agricultural Enterprises A. Share of Total Enterprises B. Constraints faced by Household Enterprises 100 100 2000 80 2005 2010 60 40 50 20 0 0 Rural Industry Poor Agriculture Non -poor Services Urban 2000 2005 2010 No problems Financial Tech. knowledge Water, Power & equipm Lack customers Other Source: HIES 2000, 2005, and 2010. 37. Broadly speaking, the existing constraints for microenterprises seem to have eased throughout the decade. However, over one-third of microenterprise owners cited lack of financing as a major constraint (Figure 4-12.B). In 2010, micro-firms’ primary source of financing was own savings, accounting for over 81 percent of total financing, a rate similar to that observed in 2000. Lack of financing was especially severe for rural, non-farm firms in agriculture, and from the Khulna, Rajshahi, and Sylhet divisions. To 67 the extent that microcredit loans help to increase income, production, and employment, especially in the non-farm sector, as documented in Khandker et al. (1998) and Khander and Samad (2012), opportunities for the continued growth of microenterprises exist.76 4.4 Trends and Patterns in Earnings and Wages 38. Section 2 described demographic trends as well as characteristics of the labor force, and found that the level of education is increasing among a rapidly growing workforce. Section 3 then described some demand-side constraints to job creation, for small, medium, and large firms as well as microenterprises. More generally, the interaction of supply and demand conditions in the labor market determines wages and labor income. In this section, trends and patterns in earnings across population subgroups are explored. 39. Average earnings per Figure 4-13: Labor Income Growth Incidence Curves worker (derived from HIES) 2000 vs. 2005, 2005 vs. 2010 and 2000 vs. 2010 increased by 40 percent between 2000 and 2010, implying annual growth of 3.2 percent over the decade. In contrast, real income per-capita grew at 4.1 percent per year, owing to an increase in the share of employed workers. Labor income grew faster during the second half of the decade, with slightly higher growth rates near the higher end of the income distribution (Figure 4-13). This uneven growth further increased inequality, as measured by the rise in the Gini coefficient for family labor income, from 0.462 in 2000 to Source: HIES 2000, 2005, and 2010. 0.476 in 2010. 40. However, Table A4-8 indicates substantial variation in the growth of labor income across sectors and regions. Average labor income increased by 5.43 percent in the West compared to 3.45 percent in the East. With respect to sectors, following a broad-based increase in the first half of the decade, average real incomes in the industry and service sectors fell in the second half of the decade (declining 3.7 percent and 1.8 percent per year, respectively, in each sector; see Figure 4-1.B). In contrast, the agriculture sector experienced a dramatic increase in real income, growing at an average rate of 9.8 percent per year. These counteracting sectoral patterns resulted in the narrowing of the East-West divide during the second half of the decade (Table A4-8). 41. Dividing workers by occupation type and region, a large portion of the income growth taking place in the second half of the decade came from rural, self-employed workers in agriculture (Figure 4- 14). While growth was relatively moderate, day laborers from rural areas also saw incomes grow. Relative to rural areas, the growth pattern of income was less favorable in urban areas. While incomes stagnated for urban day laborers and salaried workers, non-farm, self-employed workers experienced a 76 For more on microfinance in Bangladesh, see Chapter 9. 68 decline in real incomes over the decade. These trends suggest that the purchasing power of urban workers fell relative to that of rural workers. The link between higher rural wages and higher food prices is analyzed in Chapter 7. What Explains Differences in Figure 4-14: Rural versus Urban Real Income Growth Earnings? 42. Earnings functions for the 15% Bangladeshi labor market are estimated in order to understand the 2000-05 2005-10 2000-10 10% factors that account for differences in wage levels and wage growth rates. As expected, the main factors include 5% education, experience, gender, sector, occupation, and geographic location 0% (Tables A4-4 and A4-9). -5% 43. The return to one additional year of education yields, on average, -10% about a two percent wage premium Rural Urban Rural Urban Rural Urban Rural Urban (Table A4-10).77 While this return to Day Laborer Nonfarm Self- Salaried Farm Self- education may appear relatively low, Employed Employed it primarily reflects the fact that significant changes have not occurred Source: HIES 2000, 2005, and 2010. in the level of schooling (both in terms of enrollment and attainment) but real wages increased due to food price shocks. We also note substantial variation across subgroups. For instance, the return to one more year of education is as low as one percent for male daily workers, but as high as 12 percent for female public sector workers. Salaried workers tend to have higher returns to education than non-farm, self-employed workers, particularly for women. Educational returns are larger, on average, for women, reflecting the relative scarcity of educated women, and smaller in rural areas, where production techniques do not reward formal education. 44. The educational wage premium increases with the level of education for both genders, though not as drastically for women (Table A4-10). For example, the income of women with five to ten years of schooling is not significantly different from those with less than five years of schooling, implying that women with relatively low levels of human capital earn about the same as those who are illiterate. This finding suggests that low returns to working negatively affect the decision of women with low levels of education to participate in the labor force. Even though the skill premium has been declining over time, women with more than 10 years of schooling have significantly higher incomes. Analogously, men with five years of schooling or more earn significantly greater incomes than those with less than five years of schooling. Moreover, we also note important differences in the returns to education between the East and West. In particular, while the returns to education are greater for men in the West, they are higher for women in the East (Table A4-11). 77 This estimate is lower than what has been previously seen for Bangladesh (World Bank 2008a), which we believe is linked to two reasons. First this estimate aims to adjust for sample selection bias, a correction which typically reduces the estimated return (selection bias is positive). Second, these estimates include rural workers and, frequently, returns to education are only estimated for urban workers in the formal labor sector (which tends to have greater returns to skills). 69 45. Women continue to be at a disadvantage in the labor market, earning as much as 17 percent less than men in 2010. However, a portion of this difference in earnings is likely due to differences in education levels and other characteristics. In order to extrapolate the factors underlying gender income differentials, we utilize the Oaxaca-Blinder decomposition method (see Appendix 5). This approach is typically used in order to disentangle the share of the wage gap attributable to characteristics from the share attributable to an unexplained component (which may be due, in part, to discrimination). In the context of Bangladesh, the most problematic assumption of this approach is the failure to recognize the potential for deep segmentation between the kinds of jobs that men and women perform. Construction of counterfactuals, i.e. female workers’ earnings if rewarded at male prices, may be meaningless if female workers cannot be matched to male workers with similar characteristics. As described in Ñopo (2008), failure to recognize this problem implies overestimation of the unexplained component of the wage gap. 46. In order to take this issue into account, the wage decomposition is undertaken after matching male and female workers that have similar characteristics. Figure 4-15.A shows that, even once these differences are taken into account, a large portion of the gender wage gap still remains unexplained in 2010. Moreover, in contrast to 2000, when up to nine percent of the wage gap could be explained by differences in human capital characteristics (education, age, industry, and sector of employment), by 2005, the average characteristics of women were higher than those of men. The primary reason why women earned less than men is largely unexplained and not attributable to the idea that women lack qualification requirements; perhaps the gap is attributable to discrimination against women in labor markets. By 2010, while the unexplained portion of the wage gap shrank, it continued to account for over 70 percent of the total wage gap. The unexplained portion of the gender wage differential is itself a powerful reason to explain the low rates of female labor force participation. Figure 4-15: Decomposing Differences in Wages A. Male – Female Wage Gap B. Public – Private Wage Gap 110 100 60 50 10 2000 2005 2010 0 -40 2000 2005 2010 Differences in the characteristics of matched workers Differences in the characteristics of matched workers Differences due to inability to match workers Differences due to inability to match workers Unexplained Unexplained Source: HIES 2000, 2005, and 2010. The method outlined in Ñopo (2008) was used to decompose the gender wage gap . Public sector wage workers earn substantial premiums in the labor market. Both men and women in the public sector earn higher incomes than their private sector counterparts. Although these premiums declined between 2000 and 2005, and have since stagnated, in 2010, a public sector worker still earned wages over 14 percent greater than a private sector worker, irrespective of other benefits associated with public sector employment. To explore the wage differential between public and private sector workers, we undertake a decomposition method similar to the one used to describe the gender wage differential. As Figure 4-15.B shows, one-half of the wage difference between public and private workers cannot be explained by differences in observable characteristics.78 78 See Tables Annex, Table A4-12 and A4-13 for the full set of results. 70 Box 4-1: Rising Real Wages: Is Bangladesh Reaching the Lewis Turning Point? In a background paper commissioned for this poverty assessment, Zhang et al. (2012) argue that the observed real wage increase in Bangladesh (see Chapter 7) is partly the result of more ample job opportunities emerging in the non-farm sector. The most notorious example of such job opportunities is given by the garment sector, which transformed from a virtually inconsequential sector, in terms of job creation and exports, a decade ago into an industry that currently employs more than three million workers and accounts for the largest share of exports in the Bangladeshi economy (see Table 2 in Kabeer and Mahmud 2004; see also Khundker 2002; Hossain et al. 2012). The booming manufacturing sector has, in fact, attracted millions of surplus workers, mostly women, from rural areas. The authors argue that initially, as laborers moved out of the agricultural sector, the impact of this internal migration on rural wages was minimal due to the presence of surplus labor. Over time, however, the supply of seemingly unlimited labor shrank, shifting the labor market’s terms of trade in favor of workers, a chain of events known as the “Lewis turning point” (Lewis 1954), or dual sector model. Using data from several sources, the authors show that wage growth has accelerated since the second half of 2000-2010 period, particularly in rural areas and more recently in the manufacturing sector (See Tables A7-1 to A7-2). The authors also provide evidence that the combination of more abundant job opportunities in the formal sector and rising wages has likely improved the earnings of the poor, thereby helping them break away from poverty. In particular, they show that, in the 2005-2010 period, areas exhibiting larger labor force shifts toward the non-agricultural sector witnessed more dramatic poverty reduction, suggesting that the rapid poverty reduction witnessed in the second part of the decade may be partly due to a combination of increasing non-farm job opportunities and rising real wages. Overall, their findings are largely consistent with both the prediction of the dual sector model – that rising rural wages precede rising urban wages – as well as the finding from our micro-decomposition analysis (Chapter 3) – that the growth of labor income in rural areas was the most important driver of poverty reduction. Given rising labor costs in China and India, Bangladesh’s comparative advantage in labor-intensive industries is going to become more pronounced. Therefore, the trend of rising wages is likely to continue, which is good news for the poor. 47. Location matters in determining wages. Earnings regressions for men indicate regional premiums that vary between -31 percent (Chittagong) to 24 percent (Sylhet), relative to the reference district (Dhaka). In general, over the 2000-2005 period, urban workers experienced a 50 percent wage premium over rural workers. Over the later part of the decade, this urban premium fell due to large income increases in the agriculture sector (see Box 4-1). 4.5 Summary of Main Findings and Conclusions 48. Economic growth in Bangladesh coincided with a large decline in poverty, mediated through labor market activity. Indeed, the growth in per-capita income throughout the decade seems to have largely resulted from an increase in productivity growth, which is partly due to a shift toward more productive occupations. Most astonishing is that, despite tremendous growth in the working-age population and an increase in female labor force participation, employment rates rose during the first part of the decade and have since remained stable. Furthermore, higher employment rates have been accompanied by growth in workers’ real incomes. 49. With respect to labor supply, UN population projections show that an estimated 11 million people will be added to the working-age population between 2010 and 2015. By 2020, the growth rate of the working-age population predicts the addition of another 10 million workers. The rapidly expanding pool of potential workers poses an enormous challenge in terms of ensuring sufficient work opportunities for Bangladeshis. Moreover, despite minor improvements over the last decade, female labor force participation remains low, partly due to their substantially lower wages when compared to males who are 71 characteristically similar. However, as discussed earlier in the chapter, marriage is also an important determinant of female labor force participation. This finding suggests that policies that seek to delay marriage and improve education for young girls will lead to a higher likelihood of productive female employment. 50. With respect to labor demand, job creation expanded throughout the last decade. This expansion helped to usher important structural changes, including: (i) a gradual decline of agriculture and a rise of services; (ii) employment growth in urban areas; and (iii) a movement away from agriculture and, toward industry and services that coincided with movements away from daily and self-employed income and toward salaried employment. Although labor supply in the agricultural sector shrank, the sector experienced an upswing in income growth during the last half of the decade. The poverty reduction experienced during this period reinforces the notion that agriculture continues to play an important role, employing nearly one-half of the country’s workforce. Movements into non-farm work were most pronounced during the first half of the decade, during which workers were absorbed into salaried and non-farm, self-employed work. Employment by small, medium, and large firms expanded over the last decade. Going forward, however, the key bottlenecks to investment growth, including the reliability of electricity, corruption, and a workforce that is able to meet the demands of the market, must be tackled. Movement into self-employed work, on the other hand, led to growth in the number of microenterprises throughout the country. These enterprises are very small, concentrated in rural areas, and largely participating in the service industry. Individuals working in such firms are typically not poor, but micro- firm owners cite lack of financing as their main constraint, particularly in rural areas. 51. Rapid school enrollment led to an improvement in the average years of education of the labor force. However, the returns to education have declined over time. This decline is not surprising; as skilled workers became more abundant, the price for such workers fell, potentially because the growth in supply exceeded the growth in demand for these workers. However, part of the decline in returns may be attributable to a decline in the quality of education. 52. Increases in earnings were slightly greater at the higher end of the income distribution, leading to a small increase in inequality over time. Growth in earnings, however, substantially varied across job types, sectors, and regions; growth was highest among salaried employees in the private sector and in the services sector. After a broad-based increase in the first half of the decade, average real incomes declined in the industry and service sectors in the second half of the decade. In contrast, during the same time period, real incomes in agriculture dramatically increased. Final Remarks 53. Results from the micro-decomposition analysis presented in Chapter 3 and the labor market analysis presented in this chapter point to potential challenges and opportunities for continuing poverty reduction going forward. First, the type of poverty reduction observed in the first part of the decade, on account of movements toward non-farm work, calls for policies to ensure that demand for salaried work continues to grow resiliently. Because this type of work necessitates a constant supply of skilled workers, continued efforts to expand the quantity and quality of education will be crucial to sustain inclusive growth. 66. Second, to the extent that rural wages continue to rise (either driven by rising commodity prices or demand-push inflation), poverty reduction is likely to continue, given that a large share of the poor live in the farm sector, and the East-West poverty gap is likely to continue narrowing, since the majority of farm daily and self-employed workers are located in the West. However, dependence on rising prices and/or demand-pull driven wages is neither sustainable nor practical for attaining further poverty reduction. On the one hand, poverty reduction driven by “price spikes” is a reminder that growth in 72 Bangladesh continues to be vulnerable to price fluctuations and that farming continues to be the predominant employment sector for the poor to earn their livelihoods.79 On the other hand, since low wages and a large labor force are the country’s major comparative advantages, wage growth in Bangladesh is bounded by wages prevailing in competing foreign direct investment (FDI) destinations (such as China, India, Turkey, Vietnam, etc.) which also happen to have better business environment and infrastructure than Bangladesh. It is important therefore, that low wages are not the only competitive advantage that Bangladesh can offer to investors as it comes at the expense of the wellbeing of the poor. Rather investments in infrastructure and appropriate labor laws are just as important as investments in raising worker productivity. 54. Third, to the extent that the working-age population continues to grow, the economy can continue to reap dividends from this demographic change so long as the labor market is able to accommodate the growing workforce. With regard to this point, labor demand analysis shows that (i) promoting the rule of law by strengthening transparency and accountability, (ii) improving infrastructure, and (iii) promoting human capital investments have a great potential to generate the appropriate investment environment in order for firms to continue creating jobs at the pace required to meet the expanding labor supply. 79 These issues will be analyzed in Chapter 7. 73 5. Demographic Transition and Poverty: Fertility, Mortality, and Living Arrangements 1. Poverty in Bangladesh fell by almost 33 percent during the first decade of the millennium. Chapter 2 points to important demographic changes that took place in Bangladesh over the 2000-2010 period; most notably, decreases in household size and dependency ratios and an increase in the number of female-headed households, particularly among the poor. Chapter 3 decomposes the determinants of poverty decline and reveals that demographic factors, particularly, age, gender, and regional composition of the workforce, significantly contributed to the observed poverty decline. This chapter further explores the findings from Chapter 3 by examining the factors underlying the observed demographic changes. It takes a historical look at transitions in fertility and mortality as well as changes in living arrangements, driven in part by labor mobility and changing social norms. 2. In the demography literature, two different demographic transition theories prevail. The first is the classical demographic transition theory, which charts a linear change from high-fertility and high- mortality regimes to lower fertility and mortality rates; a transition which ultimately leads to a “perfect” level of fertility, where each woman has 2.1 children. This “replacement level” of fertility is considered optimal as it “replaces” one death in the “exiting cohort”. The second demographic transition theory, put forward by Van de Kaa (1987), was developed in the context of the extremely low fertility rates witnessed in 20th century Europe, where, even today, many countries are still seeking to increase fertility rates. The idea of demographic transition is generally anchored in discussions of fertility and mortality, determined by marriage patterns, labor market issues, and family formation. This chapter illustrates that in the case of Bangladesh, family formation and living arrangements have had an independent impact on poverty reduction and, thus, reflect a third type of demographic transition. 5.1 Bangladesh’s Fertility and Mortality Transition 3. Since the 1970s, Bangladesh has experienced one of the most striking declines in fertility compared to countries with comparable income levels. Between 1975 and 2004, Bangladesh halved its fertility rate from a high of 6.3 children per woman. While some anxiety persisted about a plateau in the fertility decline between 1993/94 and 2004, the last two Demographic and Health Surveys (DHS) provide evidence for a robust fertility decline since 2004. Figure 5-1 (left panel) shows a decline in the total fertility rate (TFR) over time, from 2.7 in 2007 to 2.3 in 2011, which indicates that Bangladesh will soon be at the replacement level. The decline between 2007 and 2011, in particular, has been faster than Figure 5-1: Births per Woman Across Time and by Division 4 3.7 Births per woman 3.2 3.1 Births per woman 8 2.8 2.8 2.8 6.3 3 2.3 2.2 2 2.4 2.1 Births per woman 6 5.1 1.9 4.3 2 3.4 3.3 3.3 3.0 4 2.7 2.3 1 2 0 0 2007 BDHS 2011 BDHS Note: Infant deaths per 1,000 live births based on medium variant 1950-2010. Source: NIPORT 2009 and 2012. 74 between 2004 and 2007 and averages around 15 percent nationally and 17 percent in urban areas, with Dhaka and Sylhet experiencing the sharpest declines (though Sylhet started from a high level in 2007). In Khulna, the TFR is now below the replacement level of fertility, at 1.9 children per woman, and Rajshahi is at the replacement level (Figure 5-1, right). 4. Fertility decline in Bangladesh has been accompanied by sharp reductions in infant and child mortality, rises in life expectancy, and improvements in sex ratios. In the early 1950s, East Pakistan (now Bangladesh) had substantially higher infant mortality and lower life expectancy than West Pakistan, and somewhat higher life expectancy than the broader South-Central Asia region. Today, Bangladesh is on par with its Asian counterparts with respect to life expectancy and ahead with respect to infant mortality. The speed of the changes in Bangladesh has been remarkable, as shown in Table 5.1. Sex ratios have also improved; they are now better in Bangladesh than in India and Pakistan but have yet to catch up with Nepal, Sri Lanka, and the countries of South East Asia. Table 5.1: Infant Mortality Rate over Time Male Female 1950-55 2000-2005 2005-2010 1950-55 2000-2005 2005-2010 Bangladesh 199.6 63 52 201.4 59.5 46 India 166.1 60.9 52 165.2 64.2 54 Nepal 211.7 64.5 39 210 64.4 39 Pakistan 168.1 73.4 74 169.1 77.5 68 Note: Infant death per 1,000 live births based on medium variant 1950-2010. Source: UN World Population Prospects 2011. 5. South Asia is notorious for gender inequality, which is manifested most starkly in the survival disadvantage of female children, but Bangladesh has reversed this discrepancy as well. Contrary to India, and contrary to previous research on Bangladesh from past decades,80 boys now have higher infant mortality rates than girls, which is in line with the biological norm. However, neonatal mortality (death within the first 28 days) has fallen much more slowly than mortality for ages above one month (World Bank 2008b). 5.2 Changing Age Structure of Bangladesh 6. Fertility and mortality transitions impact the age structure of the population. Of particular importance is the change in dependency ratios. Reflecting the decline in fertility and mortality, population Figure 5-2: Bangladesh: Population Pyramids 2000 2010 80 For more details, see Bhuiya and Streatfield 1991, Chen et al. 1981, Fauveau et al. 1991, Mitra et al. 2000, and Stanton and Clemens 1988. Source: US Census, http://www.census.gov/population/international/data/idb/region.php , accessed on November 3, 2012, 75 pyramids for the years 2000 and 2010 (Figure 5-2) show increases in the numbers of elderly and working- age adults as well as a reduction in the number of children. 7. These trends persist at the regional level. Figure 5-3 shows declines in the shares of infants (0-5 age group) and children (6-15 age group) across all divisions. Chittagong, Rajshahi, and Khulna witnessed the largest drops in the shares of under-15 year olds in the total population. At the same time, the working-age population (16-64 age group) share consistently increased across all divisions in 2010, compared to 2000, with the exception of Barisal, where it marginally declined, and Sylhet, which showed no change. Figure 5-3: Change in Population by Age Groups and Divisions, 2000-2010 4% Percentage point 2% 0% -2% Barisal Chittagon Dhaka Khulna Rajshahi Sylhet -4% [0-5] [6-10] [11-15] [16-20] [21-25] [26-30] [31-35] [36-40] [41-45] [46-50] [51-55] [56-60] [61-65] [66-70] [71-75] [76-80] [81-85] [85+] Source: HIES 2000, 2005, and 2010. 8. Similar patterns are observed for shares of older age groups in the population. However, the increase in the share of working-age adults outpaced the increase in the share of the elderly. These trends, combined with the decrease in the share of children, yielded a lower overall household dependency ratio at the end of the decade. 5.3 How Did the Changes in Fertility and Mortality Come About? 9. Bangladesh’s dramatic fertility Figure 5-4: Prevalence of Contraceptive Use over Time decline was driven by a response to the Malthusian hype about overpopulation in the early 1970s. This response led to an aggressive, supply-driven family planning program that provided door-step delivery of contraceptives to women who had traditionally been in purdah.81 The program went hand-in-hand with improved family planning service delivery through clinical services, community-based distribution, social marketing, and a committed band of (especially female) workers (Larson and Mitra 1992, Andaleeb 1996, World Bank Source: NIPORT 2012. 81 Purdah is the practice of physical segregation of the sexes and the requirement for women to cover their bodies and conceal their form. 76 2008b). Contraceptive prevalence in Bangladesh rose from a mere eight percent in 1975 to over 61 percent in 2011 (Figure 5-4). Several studies have made the link between access to microcredit and reduced fertility (Hashemi, Schuler, and Riley 1996). Studies around the world show that as the number of births per woman falls, maternal and child health outcomes improve. In Bangladesh, too, fertility decline has been a major factor in the overall improvement of maternal and child health. Therefore, though the primary driving force for the family planning program was to reduce the pace of population growth, it had the additional benefits of positively impacting maternal and child health outcomes. 10. Other interventions, such as the successful immunization program of the 1980s and 1990s, also generated substantial gains. In 1993/94, just over 46 percent of children under the age of one were fully immunized. By 2011, over 82 percent of infants younger than one year had received all of their immunizations (NIPORT 2012). These immunization numbers are in sharp contrast to neighboring India, where only 44 percent of children aged 12 to 23 months are fully vaccinated (IIPS 2007). Bangladesh’s urban-rural gap in child immunization has also narrowed, and only two percent of all 12-23 month olds had never been vaccinated.82 Improvements in vaccination coverage among children aged 12-23 months have occurred in all divisions except Barisal, where coverage has actually declined from 90 percent in 2007 to 83 percent in 2011. Across divisions, the highest levels of vaccination coverage for children are found in Rangpur and Khulna (92 percent and 94 percent, respectively), while children in Sylhet Division have the lowest coverage (80 percent). 11. Non-health sector programs, such as the total sanitation drive, also had a pivotal impact on better health outcomes. Sanitary toilets are now commonplace, even in rural areas. The drive against diarrhea (partially an attempt to curtail epidemics that follow routine cyclones and floods experienced by Bangladesh) has made common the use of oral rehydration and clean drinking water, which help to reduce the incidence of infant mortality. Increased girls’ education means that, compared to earlier cohorts, present generation mothers are better able to care for themselves and their children. Finally, the big push for rural infrastructure has led to a vast increase in the network of roads, bringing medical centers and hospitals closer to people (World Bank 2008b). 12. Nevertheless, Bangladesh’s demographic transition was also an unintended consequence of various policies and events, with the historical famine of 1974 serving as a catalyst in at least two ways. Since 1975, when famine prevention emerged as a key priority, the state has invested in rural infrastructure. Initially, it funded off-season, female employment programs and, later, durable rural roads, which have had an enormous impact on women’s physical mobility and access to health care, education, markets, and information. The significance of roads in human development outcomes is evidenced in several empirical findings83 and from focus groups conducted for this report. Famine relief efforts also “unleash(ed) and legitimize(d) the NGO process which has played such an important role in social change in Bangladesh.”84 5.4 Changing Family Formations in Bangladesh: The Third Force in the Demographic Transition 13. The Bangladeshi family has been relatively stable over time, with slow changes occurring with respect to marriage and the family. Age at marriage remains low at 15-18 years and around one-third of 82 In 1999, children aged 12-23 months living in urban areas of Bangladesh were 2.4 times more likely to be fully immunized (receiving all BCG, DPT1 – DPT3, and Polio1 – Polio3) than same-aged children in rural areas. As of 2011, the urban-rural gap has narrowed to 0.7 percentage points, with 86.5 percent coverage in urban areas and 85.8 percent in rural areas (NIPORT 2012). 83 World Bank (2005b) 84 Rahman, Hossain Zillur (personal communication, 2007) 77 women who are aged 20-24 were married by the age of 15 (UNICEF 2011). Divorce is also uncommon, and women generally tend to stay in a marital relationship. In the DHS 2012, just over six percent of ever- married women in the 15-49 age-group reported being divorced, separated, or widowed (NIPORT 2012). Intergenerational residence is customary, and extended families are common. The anthropological literature is replete with discussions of family and community support in economic and social terms (see for instance, Cain et al. 1979; Amin 1998). Table 5.2: Distribution of Households, by Type and Mean Household Size 2000 2005 2010 Household Mean Mean Mean type household % household % household % size size size Single 1.00 1.61% 1.00 2.09% 1.0 2.41% Nuclear 4.65 55.08% 4.40 57.91% 4.1 57.02% Semi-single 3.60 8.27% 3.45 8.39% 3.3 9.73% Extended 5.73 21.67% 5.38 19.05% 5.2 19.20% Joint 7.93 13.37% 7.68 12.56% 7.0 11.65% Total 5.18 100.00% 4.85 100.00% 4.5 100.00% Note: Following Amin (1998), we classify a single household as one with just one member; a nuclear household as one with a married couple and unmarried children; a semi-single household as one with a single parent and children or a household head and relatives, but no married couple; an extended household as one with only one married couple and children and/or relatives; joint households are extended but contain multiple couples. Source: Own estimates using HIES 2000, 2005, and 2010. 14. Since 2000, however, household structure has changed in terms of both average size and organization. This section classifies households according to definitions developed by Amin (1998): a single household as containing just one member; a nuclear household as a married couple and unmarried children; a semi-single household as a single parent and children or a household head and relatives, but no married couple; an extended household as only one married couple and children and/or relatives; joint households as extended but containing multiple couples. These classifications highlight declines in the proportions of joint and extended households and commensurate increases in the proportions of single, semi-single, and nuclear households; though, nuclear households continue to be the predominant organizational structure (Table 5.2). Figure 5-5: Change in Household Type by Division, 2000-2010 8.0% Single Nuclear Semi-single Extended Joint 6.0% 4.0% Percentage point 2.0% 0.0% Barisal Chittagong Dhaka Khulna Rajshani Sylhet -2.0% -4.0% -6.0% Source: Own estimates using HIES 2000, 2005, and 2010. 78 15. However, national averages mask differential changes that have taken place in some divisions. Figure 5-5 illustrates regional differences in household structure (see also Chapter 10). This figure shows that Khulna experienced the most dramatic changes, with a sharp increase in nuclear families and commensurate declines in extended and joint families. Barisal, Chittagong, and Dhaka have also seen changes, though not of the same magnitude, similar to Khulna; while Rajshahi has remained relatively unchanged. Living Arrangements of the Elderly Table 5.3: Living Arrangements of Persons Aged 55 and Older 2000 2005 2010 Living arrangement Urban Rural Urban Rural Urban Rural With married son(s) 47% 49% 47% 47% 44% 42% With unmarried, mature son(s)a 28% 21% 29% 22% 24% 20% With married daughter 6% 4% 3% 3% 5% 3% With unmarried, mature daughter 5% 4% 6% 5% 7% 6% Alone 1% 2% 1% 3% 2% 4% Other 13% 19% 14% 20% 19% 25% Totalb 100% 100% 100% 100% 100% 100% a Mature sons are defined as those aged 15 or older. b Others includes: - Households consisting of only two people: husband and wife, head and his father/mother (most common case). - Elderly is the household’s head and lives with spouse of son/daughter and grandchildren (the second most typical case among all these) - Elderly is a father/mother-in-law who lives with his/her son’s/daughter’s family, but the son/daughter (usually the son) is not identified/present in the household, e.g. elderly living with grandchildren. Source: Own estimates using HIES 2000, 2005, and 2010. 16. Changes in household structure often occur when living arrangements change for the elderly. This change in living arrangements contributes to a change in the dependency ratio of households and has implications for intergenerational transfers. In Bangladesh, as in other South Asian countries, the elderly tend to rely on their sons for support; about 64 percent live with their sons, whether married or unmarried. Customarily, Bangladeshi parents do not co-reside with their daughters. The World Bank Gender Norms Survey 2006 (World Bank 2008b) found that while younger women reported they would consider living with their daughters at old age, men tended to be much more conservative in their responses. In keeping with this cultural context, the HIES (2010) finds that only about six percent of all elderly in Bangladesh live with their daughters, either married or unmarried (Table 5.3). The share is slightly lower in rural areas, which tend to be more conservative. Figure 5-6: Change in the Share of Elderly Living Arrangements, 2000-2010 15% With married son(s) 10% With unmarried, mature son(s) Percentage point 5% With married daughter 0% With unmarried, mature daugther -5% Alone Barisal Chittagong Dhaka Rajshani Khulna Sylhet -10% Other -15% Source: Own estimates using HIES 2000, 2005, and 2010. 79 17. However, in the first decade of the millennium, the proportion of elderly living with their sons also dropped across all divisions, though regional variations are very marked in this regard as well (Figure 5-6). Chittagong, Khulna, Dhaka, and Rajshahi saw the largest drops in older persons who live with their sons and commensurate increases in those living with “others”. While the proportion of elderly who live alone increased in Dhaka, Barisal, and Rajshahi, this increase may be a reflection of greater numbers staying in the workforce beyond age 55 in these regions. On the other hand, the increase in the proportion living with “others” is more instructive and shows the waning of traditional patterns of co-residence with sons. 5.6 Conclusion 18. Since the 1970s, an aggressive, supply-driven family planning program provided door-step delivery of contraceptives to women who had traditionally been in purdah. As a result, between 1971 and 2004, Bangladesh more than halved its fertility rate from a high of 6.3 children per woman. The last two Demographic and Health Surveys (DHS) provide evidence of a robust fertility decline since 2004 and suggest that Bangladesh will soon be at the replacement fertility level. Parallel to the sharp declines in fertility, Bangladesh also experienced impressive changes with respect to other indicators: reductions in infant and child mortality; rise in life expectancy; and improvement in sex ratios. Reflecting the decline in both fertility and mortality, population pyramids for the years 2000 and 2010 (Figure 5-2) show increased numbers of the elderly and working-age adults and, lower numbers of children, changes which translate into an overall decline in the dependency ratio. 19. While Bangladesh’s sharp declines in fertility and mortality are well-recognized trends, this chapter argues that family formation and living arrangements also play a part in the country’s demographic transition. More importantly, changes in these factors have implications for poverty reduction and social change. The analysis shows that since 2000, household structure has changed, not only in terms of average size but also in terms of organization. For example, while nuclear households continue to be the predominant type of living arrangement, the data show declines in the proportions of joint and extended households and commensurate increases in the proportions of single, semi-single, and nuclear households. Moreover, the proportion of elderly living with their sons dropped across all divisions in the first decade of the millennium, while the proportion of the elderly living with “others” significantly increased, showing a breakdown of the traditional patterns of co-residence with sons. 20. Such changes in family formations have numerous implications. First, in coming years, Bangladesh will need to cater more aggressively to the needs of the swelling youth cohorts. Bangladesh’s “demographic dividend” is only as effective as the policies set in place to harness it and is expected to end by around 2040 (Matin 2012). Second, with increasing numbers of the elderly surviving longer and tending to live on their own, Bangladesh may be at the cusp of an aging crisis within the next twenty years. On the positive side, Bangladesh has at least twenty years to prepare and to put programs and policies into place that can protect its elderly in both a fiscally sustainable and culturally appropriate manner. Third, related to the increasing number of female-headed households among the poor, Bangladesh will likely have to target social protection programs toward such households and provide greater opportunities for female household heads. Finally, the changes in family formation seem largely to be driven by labor mobility and internal migration. These processes and their implications for poverty reduction are not well understood. Better analysis of migration patterns and their implications will be in order as Bangladesh moves to the next stage of its poverty reduction efforts. 80 Part III: Seasonal Poverty, Food Commodity Price Shocks, and the Role of Safety Nets and Microfinance 81 6. Understanding Seasonal Extreme Poverty in the Northwestern Region of Bangladesh 7. As demonstrated in previous chapters, the HIES data show that Bangladesh has achieved remarkable progress in poverty reduction over the decade and is on-track for attaining the MDG goal of halving poverty by 2015. The focus of this chapter is Rangpur, a division which continues to exhibit both high prevalence of poverty as well as marked seasonality of income and consumption. 8. In 2010, Rangpur division had the highest prevalence of poverty (42 percent) and extreme poverty (28 percent) relative to other divisions, exceeding the rest of the country by more than 12 percentage points in both categories (Table 6.1). In 2010, the ratio of poverty depth to headcount (10/42) indicates that, on average, Rangpur’s poor fell 24 percent short of the poverty threshold; in other words, the poor only consume at a level equal to 76 percent of the cost of basic needs. For the rest of the country, the analogous ratio was 20 percent. Table 6.1: Comparing Rangpur to the Rest of the Country Rangpur All divisions except Rangpur 2005 2010 Change 2005 2010 Change Poverty 0.56 0.42 -0.14 0.38 0.30 -0.08 Extreme poverty 0.38 0.28 -0.10 0.22 0.16 -0.06 Poverty depth 0.14 0.10 -0.04 0.08 0.06 -0.02 Severity of poverty 0.05 0.03 -0.02 0.03 0.02 -0.01 Food poverty 0.53 0.40 -0.13 0.40 0.33 -0.07 Source: HIES 2005 and 2010. 9. Even as the depth of poverty declined by nearly one-third, and the severity of poverty decreased by more than 34 percent over the 2005-2010 period, Rangpur continues to substantially lag behind other regions (see Box 6.1). In 2010, Rangpur’s prevalence of poverty was higher than the nation’s average in 2005, suggesting that Rangpur is five years behind in poverty reduction relative to the rest of the country. Moreover, Rangpur also suffers from persistent seasonal shocks. The austere seasonal poverty and hunger observed in Rangpur is known as Monga. Due to its severity, this seasonal phenomenon, which lasts about three months, from September to November, and precedes the Aman harvest, is likely to have important consequences for livelihoods and well-being (Box 6-2). Seasonal hunger arising from Box 6-1: Rangpur – Distinct Features of a Lagging Region Some of the major reasons for Rangpur’s comparative disadvantage are:  Inadequate investment in infrastructure, including electricity, resulting in a non-diversified, rural economy and limited opportunities for off-farm employment;  Low crop yields due to poor soil quality (e.g., soil salinity);  A high proportion of landless households that depend on wage-labor income;  Low wage rates for both male and female agricultural day laborers;  Risk of floods and river erosion;  Vulnerability of the livelihood of people living in char areas, consisting of reclaimed land from rivers, including tiny island-like fragments; and  Poor inflows of remittances from migrant family members working in the country or abroad. Source: Khandker and Samad (2012a), background paper prepared for this Poverty Assessment Report. 82 agricultural seasonality is often intensified by crop failure, poor harvest, and extreme weather conditions. In addition to agricultural seasonality, the underlying differences in agro-climate and ecological endowments, as well as local economic diversity found in the Rangpur region also influence the seasonality of income and consumption. 10. The results in Table 6.1 highlight that, though outstanding, the observed progress in poverty reduction in Rangpur over the last decade has been insufficient to reduce marked regional differences between it and the rest of the country. One possible explanation is that the type of seasonal poverty experienced in Rangpur is more severe than observed in other regions of the country. This chapter, therefore, is devoted to rigorously examining the extent of seasonal fluctuations in consumption and calorie intake as well as the various household coping mechanisms. Specifically, the chapter reviews both informal (i.e. substitution of non-food expenditure for food expenditure and employment diversification) and formal coping mechanisms (i.e. access to credit and social safety-nets) practiced by households living in Monga areas, as well as the implications of these practices on household well-being. 11. The analysis utilizes all three rounds of the HIES data (2000, 2005, and 2010) as well as a rich, multi-topic seasonal panel survey (hereafter referred to as Monga survey), which was conducted between 2008 and 2009. The latter is comprised of three survey rounds, starting with the Monga season (October to November 2008) and two post-Monga rounds (February to March 2009 and June to July 2009). The survey covered only the Rangpur division, which is the epicenter of Monga. While most of the Monga data was collected from Rangpur’s Monga regions,85 a non-Monga district, Bogra, was also surveyed and used as a comparison group. Box 6-2: Why Seasonality Should Matter in Poverty Assessment Seasonal hunger, induced by agricultural seasonality, is often a characteristic feature of rural poverty. Thanks to agricultural diversification made possible via technological breakthroughs in many parts of the developing world, the severity of seasonal stress and adversities has been reduced. Nevertheless, in certain agricultural settings of Africa and Asia, such as Bangladesh, the seasonality of hunger has been persistent for a number of reasons. First, as major food crops, such as wheat, rice, and maize, are seasonal, the seasonal stress becomes particularly acute in the event of a crop failure or a poor harvest. Second, seasonality is pronounced in economically depressed and ecologically vulnerable areas. An irregular occurrence, such as floods or failure of the monsoon rains, can magnify the adverse seasonal consequences with irreversible effects on livelihood. Unfortunately, seasonal hunger does not get the attention it deserves in discourses on food insecurity. Part of this lack of coverage is due to the fact that seasonal hunger is “missing” in the system of official data collection and analysis that averages and annualizes poverty numbers (IDS 2009). While no direct account of how many people suffer from seasonal hunger is available, it is estimated that more than 80 percent of the world’s poor live in rural areas and are dependent on agriculture for their livelihood. More so than in the past, seasonality issues deserve greater attention because of the new rising threat to global food security and the livelihood of the poor from climate changes and the associated extreme weather conditions that may make seasonal shocks more frequent, severe, and unpredictable. In economically depressed or ecologically vulnerable areas, the interlocking of poverty, seasonality, and shocks can turn a crop failure or a poor harvest into a famine. Understanding seasonal hunger is a step toward averting famine (Devereux, Vatila, and Swan 2008). Source: Khandker and Samad (2012a), background paper prepared for this Poverty Assessment Report. 85 The five districts included in the Monga region are Gaibandha, Lalmonirhat, Kurigram, Nilphamari, and Rangpur. 83 12 The rest of the chapter is organized as follows. Section 1 describes households’ ability to smooth consumption over the seasons; Section 2 analyzes the use of intra-household employment diversification as an informal coping mechanism to avert poverty and deprivation; Section 3 provides an overview of the role of formal coping mechanisms, such as access to credit markets and social safety nets, in mitigating the effects of Monga; Section 4 concludes. 6.1 An Analysis of Seasonal Deprivation Figure 6-1: Income and Expenditure by Season over Time 13. Using HIES data from 2000, 2005, and 2010, Khandker and Samad (2012a) show that Rangpur exhibits distinct patterns of income and consumption seasonality relative to the rest of the country. In particular, Monga areas experience significant drops in income and consumption during the Monga season, and these drops are persistent throughout all survey years Although monthly expenditure is less than monthly income for some seasons, during the Monga period, income in Rangpur falls short of consumption for all three survey Source: Khandker and Samad (2012a). Data source: HIES 2000, 2005, and 2010. years, and the reduction is strikingly more pronounced compared to the rest of the Figure 6-2: Trends of Per-capita Household Expenditure country (Figure 6-1). in the Monga areas 2008/09 14. The Monga survey used in this 7000 Round 1 (Monga) chapter confirms that such seasonal Round 2 Per-Capita Household Round 3 deprivation continued to prevail in Monga 5000 areas in 2008-09 (Figure 6-2). During the Exnenditure Monga season, per-capita household 3000 expenditure significantly declined at all income levels and was, on average, 5.3 1000 percent lower than in round three (June to Poorest 2nd 3rd 4th Richest July 2009) and 1.9 percent lower than in -1000 20% 20% round two (February to March 2009). Source: Mahadevan et al. (2012). Data source: Monga 2008/09 data. 15. Likewise, the poorest 20 percent of Figure 6-3: Calorie Intake per-capita per-day by the population witnessed a nearly 10 percent Consumption Quintiles reduction in food consumption during the (All Surveyed Areas) Monga season, and the poorest 40 percent of the population could not smooth calorie intake 3000 Calorie Intake per-capita over the three seasons (Figure 6-3). The 2500 inability to stabilize caloric intake was 2000 per-day starkest among the poorest 20 percent; their 1500 already low level of per-capita per-day calorie 1000 intake (1,500 kcal) dropped by around 15 500 percent (about 1,350 kcal) during the Monga 0 Poorest 2nd 3rd 4th Richest season. Similar evidence of seasonality in Oct - Nov 2008 Feb - Mar 2009 Jun - Jul 2009 food consumption is presented in Box 6-3. Source: World Bank staff estimation using the Monga survey. 84 Table 6.2: Trends of Poverty by Land Ownership in Rangpur Poverty Rate Population Distribution Land size 2000 2005 2010 2000 2005 2010 Landless < 0.05 acre 86.9 77.8 63 41.1 43.2 47.4 Functionally landless 0.05-0.5 acre 76.2 64.4 45.5 14.1 16.3 16.1 Marginal 0.5-1.5 acres 58.5 47.7 29.4 18.7 17.7 17.8 Small 1.5-2.5 acres 54.7 31.7 18.2 8 9.4 7.7 Medium/large: 2.5 acres or more 28.6 8.5 6 18.1 13.3 10.9 Source: HIES 2000, 2005, and 2010. 16. Selected welfare indicators, presented in Table 6.2 and Table 6.3, suggest some of the possible reasons underlying the problem of seasonality in Rangpur. Table 6.2 presents poverty by land ownership in Rangpur. Similar to the national trend (see Table 2.2), this table shows that poverty rates were negatively related to the size of landholdings in all three survey years. Yet, the fact that the 2010 poverty estimate for Rangpur’s landless (63 percent) was the same as the 2000 poverty estimates for the nation’s landless highlights Rangpur’s comparative disadvantage in land ownership. 17. Table 6.3 presents means for the basic assets and amenities indicators for each of the HIES survey years, 2000, 2005, and 2010. Similar to the nation (see Table 2.1), all non-consumption welfare indicators show significant improvements for the general population, with even starker improvements for the poor, between 2000 and 2010. 18. While households in Rangpur are more likely to own livestock (the average value of which is also higher relative to the nation), the quality of their homes is significantly lower, and the amenities they enjoy are dramatically worse when compared to the entire nation. In 2010, at the national-level, 75 percent (61 percent) of all households (households in the bottom 3 deciles of the real per-capita consumption distribution) had access to safe latrines; the same estimates were 63 percent (45 percent) for Rangpur. In the same year, 55 percent (33 percent) of all households (households in the bottom 3 deciles) had access to electricity, while the analogous estimates were 30 percent (14 percent) for Rangpur. Similarly striking differences exist between Rangpur and the nation when considering TV and phone ownership. 19. Rangpur’s comparative disadvantage is also evident from trends in monthly agricultural real wage rates, as reported in Khandker and Mahmud (2012). While wage rates have shown an overall upward trend at the national level, they have consistently lagged in Rangpur, demonstrating that lack of Table 6.3: Trends in Basic Assets and Amenities, Rangpur Division All households Bottom 5 deciles Bottom 3 deciles 2000 2005 2010 2000 2005 2010 2000 2005 2010 Livestock ownership (%) 54.9 60.8 57.2 43.6 59.6 54.9 38.7 54.4 51.4 Wall of dwelling* 18.9 31.2 41.6 6.1 22.8 33.1 4.7 17.1 29.9 Roof of dwelling* 60.5 85.8 92.9 47.7 82.9 91.2 41.7 80.8 89.8 Safe latrine use (%) 22.6 40.8 63.4 6.9 29.6 51.2 5.1 24.8 44.7 Electricity connection (%) 13.1 21.1 30.1 2.6 11.5 19.3 1.1 8.3 13.7 TV ownership (%) 6.3 13.6 20.7 1.3 6.1 9.2 0.7 3.3 5.5 Phone ownership (%) 0.2 3.7 42.1 0.0 0.5 27.2 0.0 0.1 20.1 Source: HIES 2000, 2005, and 2010. * percent with cement/CI sheet. 85 real wage and employment growth are prime factors in the persistence of year-round and seasonal poverty. 6.2 Intra-household Employment Diversification as an Informal Coping Mechanism Employment in areas affected by Seasonal Deprivation 20. Contrary to the common belief that many Monga area residents suffer from a lack of employment opportunities during the Monga season, the Monga survey shows that employment ratios (number of employed compared to number of working-age people) did not significantly change across seasons. Employment ratios remained between 83.5 percent and 85.5 percent in Monga areas, whereas they ranged between 77 percent and 85 percent in non-Monga areas. Furthermore, employment ratios were higher in Box 6-3: The Monga Phenomenon: Evidence from the InM Survey A baseline survey, administered by InM in 2006, covered nearly half a million rural households from 23 upazilas (sub-districts). These households constitute roughly 60 percent of all households in the survey villages (for details, see Khandker and Mahmud 2012). The InM survey was carried out as part of an effort by the Palli Karma Shahayak Foundation (PKSF), the only wholesale outlet of microfinance in Bangladesh, to identify the extreme poor who are vulnerable to Monga and to design and implement appropriate interventions to mitigate Monga. In the InM survey, the extreme poor are identified as households that satisfy at least one of the following criteria: (a) have less than 50 decimals (0.5 acres) of land; (b) have a monthly income of Tk 1,500 (equivalent to US$22) or less; or (c) sell labor for daily wage. Of the three criteria, landownership was found to be pervasive, applying to 98 percent of the surveyed households. The InM survey categorizes households into three groups based on their food deprivation status: starvation (no meals on some days); meal rationing (skipping one or two meals or consuming less than adequate at each meal on some days); and full meals (full meals, usually three times a day). This information was recorded for all households for Monga and non-Monga periods. The figure shows the extent of seasonal hunger in the Rangpur region during Monga and non-Monga seasons. Both the extent and the seasonality of hunger become strikingly evident from the estimates presented, Distribution of Rural Households by Food particularly when considering the fact that the lean Deprivation Status in Monga and non-Monga season of 2006 was no different from that of a normal Seasons in the Greater Rangpur Region, 2006 agricultural year in Rangpur. The situation would have been much worse if the region experienced abnormal floods or other natural calamities in that year. Some 47 50.8 47.3 48.3 percent of households experienced starvation during the Monga season, whereas only 4.4 percent had full meals. 40.6 Roughly 50 percent of the extreme poor had to ration food in both seasons, indicating a persistent type of food insecurity in the Rangpur region. Even during the non- Monga season, about nine percent of households experienced starvation—an indication of the dire 8.6 situation among the very poor in the region. 4.4 The extent of seasonality is evidenced by the substantial changes occurring in the starvation and full Starvation Meal Rationing Full Meal meal categories between seasons. In particular, Monga season Non-monga season starvation among rural households in the Rangpur region rises from nine percent in the non-Monga period to a Source: InM baseline survey, 2006. staggering 47 percent in the lean season, i.e., the Monga period. About the same proportion of households also have to ration food during both the Monga and non-Monga periods. Source: Khandker and Mahmud (2012). 86 the Monga region than in the non-Monga region, particularly during the Monga season (round one). However, these aggregate ratios may mask underlying labor market vulnerabilities. For example, while employment ratios for the entire workforce remained constant, further analysis reveals significant instability in terms of employment status. 21. Figure 6-4 shows that the estimated number of individuals employed during the Monga season is similar to the number of individuals employed in round two, at around 4 million to 4.2 million. However, of the total employed population in the Monga season (4 million), only 3.6 million remained employed in round two, with 10 percent of workers, or 0.4 million workers, rendered inactive or unemployed by the second round. Similarly, out of the 4.2 million employed in round two, some 0.6 million workers were unemployed or inactive in the Monga season. In total, as many as one million potential workers did not have a stable employment status moving from the Monga season to round two. Figure 6-4: Hidden Instability Behind Employment Ratios in Monga Areas Employed in the Monga Season Employed in Round 2 (in millions) (in millions) Employed in both 0.41 Employed in both rounds rounds 0.6 Unemployed or Inactive in Round 2 Unemployed or Inactive in the Monga season 1 million individuals have an unstable employment status. Source: Mahadevan et al. (2012). Data source: Monga 2008/09 data. 22. Regarding employment dynamics by sectors, a large proportion of non-agricultural workers in round one switched (back) to the agricultural sector by round two. This sectoral shift may be interpreted as follows: in the absence of agricultural opportunities during the Monga season, several people take up opportunities in the non-agricultural sector; but by round two, they switch back to agriculture. Furthermore, compared with the non-agricultural sector, the most stable group was the self-employed in agriculture, with 94 percent retaining the same employment status between the first and second rounds. One possible explanation is that the self-employed in agriculture are land-owners, comparatively richer, and, thus, better able to cope with the lean season. In contrast, those employed in labor are more prone to coping with the Monga season by way of switching sectors. 23. Employment ratios did not significantly change across consumption quintiles, but employment pattern trends considerably differed between the rich and poor. For example, among men, the percentage of the poorest involved in agricultural labor substantially increased from 42.6 percent to 51.2 percent between round one and round two, while the percentage of the poorest involved in non-farm sector jobs decreased from 50 percent to 40 percent between the two rounds (Figure 6-5). These patterns imply some degree of job substitution, with 10 percent of the extreme poor moving from agricultural labor to non- farm sector jobs in the Monga season. The richest quintile, on the other hand, did not experience these movements. 87 24. This analysis confirms that Figure 6-5: Employment Categories Among the Richest and the employment diversification in Poorest 20 Percent (%) – Male Only Monga areas was prevalent among 100 male workers and that it was significantly changing in response to 50 a lack of agricultural labor jobs. However, these observations say 0 very little about whether different Round 1 Round 2 Round 3 Round 1 Round 2 Round 3 households chose different jobs or Poorest Richest whether each household adopted Labor/salaried job : agri Self-employment: agri Labor/salaried job : non-agri Self-employment: non-agri intra-household employment diversification. Source: Mahadevan et al. (2012). Data source: Monga 2008/09 data. Welfare Effects of Employment Diversity and Dynamics 25. Multivariate regression analysis suggests that the seasonal deprivation experienced by households during the Monga season significantly increased employment diversity. In particular, it reveals that employment diversity was highest during the Monga season (according to four out of six employment indicators used to measure employment diversity within a household), whereas employment diversity declined following the Monga season (indicated by five out of six indicators) (Table 6.4). Table 6.4: Measures of Employment Diversity within a Household Variable Means Percent changes Description name Round 1 Round 2 Round 3 R1→R2 R1→R3 soccu Same occupation as head 0.12 0.16 0.13 26% 5% ssect Same sector as head 0.56 0.57 0.54 2% -3% sagri Member and head in Agriculture 0.49 0.53 0.52 7% 5% bself Member and head in self employed 0.48 0.49 0.53 3% 12% ssectbse Both in same sector, Both Self Employed 0.13 0.12 0.12 -9% -3% bagrise Both agriculture and Self Employed 0.25 0.28 0.28 11% 13% Note: Definitions of these variables follow Bandyopadhyay and Skoufias (2012): (i) soccu – a household member has the same occupation as the household head; (ii) ssect – a household member works in the same sector as the household head; (iii) sagri – a household member and the household head both work in the agricultural sector; (iv) bself – a household member and the household head both are self-employed; (v) ssectbse – a household member and the household head work in the same sector and are both self-employed; (vi) bagrise – a household member and the household head both work in the agricultural sector and are self-employed. Source: Mahadevan et al. (2012). 26. Employment diversification and the switching of employment status reduce opportunities to exploit economies of scale. However, diversification of employment can improve resilience against shocks. As a result, the welfare and growth implications of employment diversification are not clear. Frequently switching employment status deters household members from building skills and knowledge through learning-by-doing but minimizes the risk of loss due to shocks like extreme weather. Therefore, the question of whether or not employment diversification and frequent switching of employment status have negative impacts on household welfare and growth is empirical. 88 27. Bandyopadhyay and Skoufias (2012) provide empirical evidence that intra-household employment diversification prevails in rural Bangladesh and also appears to have negative effects on household consumption. Using HIES 2010 data, their analysis shows that areas with higher rainfall variability and those affected by the severe flood of 1998 exhibit a significantly higher level of intra- household employment diversification. Moreover, the authors show that such employment diversification negatively affected household welfare and increased the probability of falling into poverty. 6.3 Coping with Household Vulnerability 28. Rangpur’s employment patterns appear to underlie its persistent seasonality. The major source of seasonality is the decline in farm wage employment during the Monga season. Compared to casual laborers, the self-employed are better able to protect themselves from lack of employment. While seasonality is fairly predictable, if and how vulnerable households are able to cope with seasonal shocks is of utmost importance. Additional unpredictable shocks, such as floods or droughts, exacerbate the adverse impact by manifold and may bring about irreversible adverse effects on poverty. A household's decision regarding which coping mechanism to choose is likely to depend on a combination of household level factors as well as the availability of informal and formal coping mechanisms. The previous section focused on a potentially important informal coping mechanism: “employment diversification.” This section discusses three additional coping options that allow for the occurrence of income transfer; migration, access to credit, and social safety nets are likely candidates in rural Rangpur. This section discusses three additional coping options: migration, access to credit, and access to safety nets. Migration and Access to Remittances Table 6.5: Temporary Migration: ≤ 1 month over 1- year survey period (%) (age ≥ 12) 29. According to the Monga survey, a Region Total significant share of the population living in the Monga 18.0 Monga region temporarily migrated, particularly Non-Monga 6.9 among the poorest quintiles (Table 6.5). For Cons. Quintile Total example, eighteen percent of individuals aged 12 years or older temporarily migrated in the Monga Poorest 21.8 region, compared to only seven percent in the non- 2nd 19.7 rd Monga region. Among the lowest consumption 3 19.2 quintile, around 22 percent temporarily migrated 4th 13.9 for less than one month, compared to only nine Richest 9.2 percent of households in the richest quintile. Source: Basu et al. 2010. 30. While seasonal migration appears to be a more prevalent coping mechanism among the poor, the share of households who receive remittances during the Monga season is lower in Monga areas (11.8 percent) compared to non-Monga areas (15.9 percent) (Table 6.6). For remittances received on an annual basis, this discrepancy increases: only 12.3 percent of households in Monga regions receive remittances compared to 17.3 percent of households in non-Monga areas. 31. Breaking down access to remittances across the expenditure distribution, we find that both the poorest and the richest populations are likely to receive remittances during the Monga season. However, on a yearly basis, only 12.8 percent of the poorest quintile receives remittances compared to 18 percent for the richest quintile. 32. The HIES data further supports that access to remittances does not constitute an important coping mechanism against seasonal shocks. In many areas of rural Bangladesh, remittance income from family members working abroad represents a significant proportion of household income as well as a substantial source of fund inflows into the local economy. However, the exception to this pattern is Rangpur, where 89 remittances comprised only 4.7 percent of total household income, compared to 14.8 percent for rural households in other regions in 2010 (Figure 6-6). Table 6.6: Received Remittance (%) Survey round Region Oct - Nov 2008 Feb - Mar 2009 Jun - Jul 2009 Total Monga 11.8 15.2 9.9 12.3 Non-Monga 15.9 18.4 17.7 17.3 Total 12.4 15.8 11.2 13.1 Cons. quintile Oct - Nov 2008 Feb - Mar 2009 Jun - Jul 2009 Total Poorest 13.6 15.8 8.3 12.8 2nd 11.0 15.5 9.6 12.0 3rd 8.5 13.8 8.8 10.5 4th 14.2 13.8 9.3 12.3 Richest 14.7 20.1 19.2 18.1 Total 12.4 15.8 11.2 13.1 Source: Basu et al. 2010. Figure 6-6: Change in Income Composition over Time by Region, 2000–10 A. Rangpur Source: Khandker and Samad 2012a. Data Source: HIES data 2000, 2005, and 2010. 33. Further analysis using the HIES 2010 data reveals that more than 70 percent of remittances benefit households that are at the top of the consumption distribution. The distribution of remittances is 90 skewed to the right and has become even more skewed over the course of the decade. The implication of this change in distribution is that, while remittances have grown incredibly quickly and to very large levels, the vast majority of the value of remittances is received by non-poor households and, therefore, has a limited role in directly reducing poverty.86 34. One potential concern is that the HIES data appear to underestimate the total value of remittance income. To address this concern, we scale up the HIES data based on estimates of remittances from macro-economic data. However, even when the HIES remittance data are substantially scaled up, they do not significantly change the results. Under this exercise, the contribution of remittances to poverty reduction does increase, but it continues to be significantly lower than the contribution of labor income. Access to Credit Markets 35. Access to credit, along with savings, is a widely utilized coping mechanism against covariate shocks in Bangladesh. Analysis from a nationally representative survey, conducted during September to October 2009, shows that a little over one-third of the population, and one-quarter of the poor population, coped with covariate shocks using the help of credit (Santos et al. 2011). Residents of Monga areas are no different: according to the Monga survey, a higher rate of loan approval occurs during the Monga season as compared to the rest of the year. Loan coverage in Monga areas is at 43.5 percent, which is relatively high, in the Monga season. Nevertheless, the evidence suggests that poor households are not necessarily able to utilize this coping mechanism to smooth consumption. A key reason is that access to credit remains fairly limited. Figure 6-7: Loans and Credit A. Percentage of Poor Households that Have Access to B. Shares of Access to Credit by Quintile of and Have Been Approved for Loans in the Monga Household Consumption in the Monga Region Season 25 Not applied/approved Approved for a loan 20 80 15 60.32 10 60 52.27 47.73 5 39.68 0 40 Poorest 2nd 3rd 4th Richest 20 (Bottom Quintile Quintile Quintile (5th 20%) Quintile) 0 Poor in Monga areas Poor in Non-Monga areas At least one member of the HH requested a loan Note: “Poor households” refers to those falling in the bottom 20th percentile of per-capita consumption. Source: Mahadevan et al. (2012). Data Source: Monga 2008/09. 36. The Monga survey reveals that the bottom 20 percent of the population in Monga areas is most vulnerable to seasonal deprivation and, therefore, has among the highest rates of loan applications and approval. Yet, as observed from Figure 6-7.A and B, the bottom 20 percent has a lower loan approval rate (by as much as 8 percent) than average households in the same areas as well as their counterparts residing in non-Monga regions. Many reasons may exist for this discrepancy; for instance, lack of sufficient formal or informal lending organizations, lack of assets or land to offer as collateral, and inability to pay 86 Although remittances do not appear to have a large, direct role in poverty reduction, they may have significant secondary effects. However, the analysis of such effects is beyond the scope of this report. 91 back the loans. Irrespective of cause, available credit is insufficient to smooth the consumption patterns of Monga area residents. 37. Of households that did not request loans before round one, around one-third report not applying because they did “not need loans”. Among households that did not apply for loans, 23 percent report that they did not because of high interest rates; 14 percent are afraid of defaulting on the loan and losing their collateral; 10 percent are afraid of being unable to pay back the loan and incurring additional fees. Social Safety Nets 38. An alternative formal coping Figure 6-8: Percentage of Households that have Access mechanism for residents of the Monga region is to Social Safety Nets (Monga Region) access to social safety nets, either through the government or through NGOs. These 12.17 Poorest (Bottom 20%) interventions provide a range of assistance in the form of cash, food, assets, and income 2nd Quintile opportunities. 28.37 14.5 3rd Quintile 39. Safety net coverage is around 35%. 21.59 4th Quintile 23.36 However, although the coverage of the extremely poor is nearly 44 percent, only 28 Richest (5th Quintile) percent of safety net beneficiaries belong to the bottom expenditure quintile (Figure 6-8). Source: Mahadevan et al. (2012). Data Source: Monga 2008/09. Moreover, 12 percent of beneficiaries belong to the richest quintile. Thus, estimates suggest that safety net expenditures, seeking to address seasonal food deprivations, are not necessarily being targeted toward those who are most in need of income support. 40. Using data from a 2006-2007 household survey of the greater Rangpur region, Khandker et al. (2011) find some evidence that the accrued benefits received from safety nets have a positive effect on mitigating seasonal food deprivation. Among recipients of safety net benefits, the authors find a 4.4 percentage point reduction in Monga-time starvation. Given that safety net programs, where present, are effective in helping to mitigate seasonal starvation, Khandker et al. (2012) recommend for programs to be deepened in terms of both coverage and size. They also recommend that government-supported social safety nets should be well-coordinated with those of NGOs in order to enhance effectiveness. 6.4 Conclusion 41. The analysis presented in this chapter provides useful insights into the Monga phenomenon and its implications for the well-being of Monga area residents. First, the analysis shows that Monga continued to have a negative impact on consumption smoothing in 2008-2009. While the extremely poor in non-Monga areas were able to successfully smooth both food consumption and calorie intake, the effect of the Monga phenomenon was particularly severe among the extremely poor living in Monga areas. 42. Second, the analysis reveals that many in the Monga areas were actively shifting resources from non-food items to food items. The rate of reduction in non-food expenditure was nearly 40 percent during the Monga season. Because such large fluctuations in food consumption can have severe, long-term implications for the social and economic activities of households in Monga areas, the focus of policy must be to mitigate such large fluctuations. 92 43. Third, the analysis also shows that many in the Monga areas diversified employment within their household as an ex-ante and ex-post coping mechanism. Like Bandyopadhyay and Skoufias (2012), the analysis presented here suggests that excess diversification and job switching could have negative impacts on household welfare. 44. Fourth, the very limited information on migration, available from the Monga survey, suggests scope for increased migration as a source of coping against seasonality in the Rangpur region. Further analysis on ways to support increased migration as a source of increased income for the Rangpur region is important (see Box 6-4). 45. Fifth, with regard to formal coping mechanisms, the Monga survey shows that approximately 40 percent of the extremely poor had access to loans and almost 45 percent had access to social safety nets in Monga areas. These numbers imply that a majority of the extremely poor who live in Monga areas remain largely unable to resort to formal coping mechanisms to help smooth consumption during the Monga season. While large public safety net programs, such as the Food for Work and Vulnerable Group Feeding programs (discussed in chapters 7 and 8), are in place to fight seasonal shocks, their coverage for the poor was still limited in 2008/09. Since the Monga survey has limited information on migration, remittance, and informal community lending, these topics constitute the most notable omissions from this chapter. Further studies on these topics are warranted in order to generate a more accurate portrait of the effects of seasonality on the lives of the Bangladeshi people. Nevertheless, the information presented in this chapter is a reminder that far greater progress is required to eradicate the effects of the Monga phenomenon. Overall, the findings from the present analysis and previous analyses call for the deepening of micro- lending and safety nets, in terms of coverage and size as well as improvements in targeting the most vulnerable. 46. Finally, this chapter underscores that, despite remarkable reduction in overall poverty levels over the last decade, Rangpur remains a pocket of extreme poverty that has not sufficiently benefited from pro- poor growth to catch up with other regions. For the next five years, anti-poverty policies must be categorically tailored to the problems of Rangpur in order to address these marked regional differences in poverty across Bangladesh. Box 6-4: Small Incentives, Large Improvements Using a randomized experiment in which Monga households were incentivized to send a seasonal migrant to an urban area, the authors found that a small incentive led to a large increase in the number of seasonal migrants, that the migration was successful (in terms of improving consumption by around 30 percent), and that households, given the incentive in one year, continued to be more likely to migrate in future years. The paper argues that the results are consistent with a simple (rational) model of a poverty trap, where households that are close to subsistence are unable to learn whether or not migration is successful due to a small possibility that migrating will turn out badly, leaving household consumption below subsistence. The model helps to understand the type of situation in which we would expect incentives and insurance policies to lead to the sorts of long-term effects observed in this experiment. Implications emanating from this study suggest that programs to support migration for households who are close to subsistence may constitute one set of policy responses to increase the migration rate from Monga areas. Source: Bryan, Chowdhury and Mobarak (2009). 93 7. The Impact of Food Price Shocks on Wages, Welfare, and Policy Responses 1. Since 2005, prices of major food crops have been surging in the international market. The World Bank food price index (FPI) reached its first significant peak, driven mostly by major grain prices, during the first quarter of 2008. By mid-2008, the index reached an inflection point, and then sharply dropped in the immediate aftermath. During the following two years, the index entered a relatively stable period, although its level remained elevated compared to the pre-2008 period. Starting in mid-2010, the index began to rise again and continued to rise until early 2012 (Figure 7-1). Figure 7-1: World Bank Food Price Index 280 Food Grains 260 240 220 200 180 160 140 120 100 2012m03 2007M05 2011M01 2007M01 2007M03 2007M07 2007M09 2007M11 2008M01 2008M03 2008M05 2008M07 2008M09 2008M11 2009M01 2009M03 2009M05 2009M07 2009M09 2009M11 2010M01 2010M03 2010M05 2010M07 2010M09 2010M11 2011M03 2011M05 2011M07 2011M09 2011M11 2012M01 Note: The Global Food Price Index weighs export prices, in nominal U.S. dollars, around the world of five food commodities (cereal, oils/fat, sugar, meat, and dairy), 2005 FPI = 100. Source: World Bank. Source: World Bank. 2. The price hike in 2011 was significant, though not as steep as the one witnessed in the 2008 period. Between 2005 and 2008, rice prices rose by 25 percent, wheat prices rose by 70 percent, and maize prices rose by 80 percent (Ivanic and Martin 2008). Nevertheless, the Food and Agriculture Organization (FAO) and the Organization for Economic Co-operation and Development (OECD) warn that crop prices are expected to continue rising (Mulat et al. 2009), caused primarily by higher oil prices and shortages in food commodity markets. Thus, food price volatility is an emerging source of vulnerability for low income countries like Bangladesh, where food constitutes a large portion of consumption. 3. This chapter investigates how food price volatility affected the poor in Bangladesh over the last decade. The goal of the chapter is to answer the following questions: What makes Bangladesh vulnerable to food price shocks? What is the overall effect of price shocks on the Bangladeshi economy and people? Which particular sub-groups are most affected; how so and to what extent? How quickly and persistently do wages adjust? The chapter also provides a brief overview of policy measures adopted by the Government of Bangladesh to avert the deleterious effects of large food price fluctuations during 2008. 7.1 What makes Bangladesh vulnerable to food price shocks? Prices and Food security 4. Rising international food commodity prices have different pass-through effects on domestic prices, which depend upon a country’s economic fundamentals. At the macro-level, a country’s vulnerability depends on whether the country is a net exporter or importer of food commodities as well as the country’s degree of dependency on food commodity imports. Bangladesh is a largely rural economy 94 where agriculture contributes about 22 percent to total Gross Domestic Product Figure 7-2: Trends in Production and Consumption of Rice in Bangladesh (GDP) and accommodates around 48.1 31000 percent of the labor force. The number of Production(in '000MT) agricultural farm households accounts for 30000 51.33 percent of total households Consumption (in '000MT) 29000 (Agriculture Census, 2008).87 Moreover, in 28000 the absence of a negative production shock (such as a weather shock), 27000 Bangladesh is nearly self-sufficient in rice 26000 production (Figure 7-2). The country has 25000 increased its food grain production over the 24000 past several years, from 25 million metric tons in 2000-01 to around 30 million metric tons in 2008-09. Since agriculture is imperative to the country’s economy, Source: USDA data Base. Bangladeshis could financially benefit from higher global commodity prices (World Bank 2012c). Yet, since only 37 percent of Bangladesh’s total area is arable land, of which almost 30 percent is vulnerable to natural disasters (Khan and Wadud 2012) that adversely affect grain production, Bangladesh continues to rely on food grains imports to meet domestic demand. 5. At the micro-level, households’ vulnerability to food price shocks is a function of: (i) the ability of households to buffer against the price increase; (ii) whether households are net buyers or sellers; and (iii) whether or not affordable food substitutes exist. FAO (2009) reports that approximately 12.3 million people in Bangladesh became food insecure following the 2008 food price shock while about 34.7 million people became under-nourished as a result of higher food prices. In addition, while the average daily calorie intake was only 2,318 according to HIES 2010, approximately 24 (57) million Bangladeshis, or 16 (38) percent of Bangladesh’s population (see Table 2.6), is severely (moderately) food deficient; that is, they cannot afford an average daily intake of more than 1,805 (2,122) calories. Thus, higher food prices Figure 7-3: Cereal Consumption in Bangladesh A. Energy Consumption by Food Groups B. Rice Consumption as a Share of Total Expenditure (%) by consumption quintiles Pulses/legu Milk and Oil/fats 36.1 Fish and mes/nuts milk 8% Sugar/honey seafood 2% products 2% 3% 1% 25 Eggs Share of rice 1% 20.7 Meat, poultry, offal 13 1% Fruits 1% 5.2 Vegetables 4% Root and Cereals tubers 74% 33.39 29.06 24.28 8.2 5.07 3% 2005 1 2 3 4 5 Source: Background paper by Rabbani (2012) commissioned for Source: HIES 2010. this poverty assessment. 87 Khan and Wadud (2012): Food Security and Case of Bangladesh, available at http://www.foodsecurityportal.org/bangladesh/food-security-and-case-bangladesh-report-prepared-department- agricultural-marketing 95 can have severe, adverse effects on the well-being of the poor. 6. Moreover, Bangladeshis consume about 74 percent of their calories from cereals (Figure 7-3.A), and rice accounts for more than 40 percent (23 percent) of total consumption for households at the bottom (top) quintiles (Figure 7-3.B). The nutritional intake of every Bangladeshi, therefore, is likely to be negatively affected by a food commodity price shock. 7.2 Household Exposure to Food Price Increase in Rural Bangladesh: what is the overall effect of price shocks? 7. Using HIES data from 2000, 2005, and 2010, and exploiting the significant food price fluctuations taking place over the 2000-2010 period,88 we measure the exposure of Bangladesh’s rural poor to recent global food price shocks. This novel method for measuring “household exposure” to food price shocks helps to identify the magnitude of and the ways in which food price shocks affect the poor. First, the analysis uses Deaton’s partial equilibrium model in order to estimate the short-term effect of the food price shock. Second, in order to capture the medium-term adjustment occurring in the labor market, the analysis takes into account the effect of changes in food prices on rural wages. Lastly, in order to account for price adjustments taking place throughout the economy, the analysis considers price changes affecting three sectors - agriculture, manufacturing, and service (for more details on the underlying empirical model, please refer to Appendix 7). Figure 7-4: Exposure to Food Price Shock A. Short- and medium-term effects B. Short-, medium-, and long-term effects Source: Background paper by Jacoby and Dasgupta (2012) commissioned for this poverty assessment. Data source: HIES 2000, 2005, and 2010. 8. Assuming a fixed price for services, our analysis reveals that both the short- and medium-term effects of a food commodity price spike yield the expected results. In the short-term, since they spend most of their income on food, the poor bore the brunt of higher food prices (dashed blue line in Figure 7- 4.A). In the medium-term, after wages have adjusted in response to the shock, the impact is largely equalized along the wealth distribution (solid black line in Figure 7-4.A). In the long-term (assuming a 88 Background paper by Jacoby and Dasgupta (2012) commissioned for the Poverty Assessment. Since the objective of this study is to assess the impact of international prices on rural households’ welfare, in order to measure price fluctuations, the authors use the average farm-gate prices (in USD) of major crops (rice, wheat, and jute), as reported in the World Bank’s ‘Distortions to Agricultural Incentives’ database. 96 variable price for services), when the price shock has permeated into other sectors, particularly the service sector (the output of which is disproportionately consumed by the rich), the standard result was reversed (solid black line in Figure 7-4.B). In other words, while the poor were disproportionately affected by the food price increase in the short-term, the higher end of the income distribution was more affected in the long-term. Welfare Impact of Commodity Price Shocks: Who is affected and when? 9. In 2010, as much as 77 percent of rural households were considered net buyers of rice. The commodity food price shock resulted in a nearly 10 percent decline in consumption among rural households. Additionally, in the short-run, households in the two lowest consumption quintiles, for whom rice comprised about 34 percent and 30 percent, respectively, of consumption budgets, were the most adversely affected by the food price increase (Figure 7.4, dashed blue line; Table 7.1.A). In particular, the short-term impacts of the food price shock were 24 percent and 17 percent respective declines in per- capita monthly consumption for households in the two lowest consumption quintiles, whereas the same shock resulted in a one percent increase in monthly per-capita consumption for the highest consumption quintile. Table 7.1.A: Welfare Impact of a Rice Price Increase by Poverty Net benefit ratio Rice consumption Rice Production Net Without With Household as % of total as % of total buyers Wage wage With wage GE category expenditure expenditure of Rice response response response Effect Rural 24.70 15.80 76.99 0.17 -0.10 0.06 -0.10 Quintile 1 33.75 35.95 61.76 0.36 -0.24 0.06 -0.08 29.74 31.86 61.95 0.26 -0.17 0.04 -0.10 Quintile 2 Quintile 3 25.73 27.79 61.47 0.19 -0.11 0.04 -0.11 Quintile 4 21.52 23.58 59.44 0.12 -0.05 0.07 -0.09 Quintile 5 15.16 17.29 60.81 0.06 -0.01 0.07 -0.12 Extremely poor 33.57 11.39 87.67 0.32 -0.22 0.06 -0.08 Poor 31.86 11.50 86.02 0.29 -0.20 0.04 -0.11 Non poor 20.82 18.12 72.10 0.11 -0.04 0.07 -0.10 Source: Background paper by Jacoby and Dasgupta (2012) commissioned for this poverty assessment. Data source: HIES 2000, 2005, and 2010. 10. When considering poverty status, the average short-term impact of the shock on the extremely poor, who experienced a 22 percent decline in consumption, is similar to the experience of the lowest consumption quintile, while the poor experienced a 20 percent decline (or about two percentage points lower in absolute terms). The average effect on the non-poor is a four percent decline in consumption which, while significant, is relatively less severe than the impacts on the poor and extremely poor. Overall, the results for the short-term welfare impacts of the commodity food price shock yields the expected results: the poor are most adversely affected. Moreover, while the results of the household exposure model by Jacoby and Dasgupta (2012) applies only to rural households, the fact that nearly 95 percent of urban household are net buyers of rice suggests that the urban poor were likely to be the most negatively impacted during the food price shock. 11. When accounting for the medium-run effect of the shock, the changes in wages, in response to the price increase, more than compensated households residing in rural areas, irrespective of consumption level. For rural-area households, the average medium-run effect was a 17 percent increase in monthly per- 97 capita consumption. Interestingly, the extremely poor experienced a relatively greater benefit from the wage increase compared to the poor (32 percent versus 29 percent). Nevertheless, the wage increase resulted in significant medium-run gains for households in the third, fourth, and fifth consumption quintiles as well the non-poor, who are more likely to be land-owners and self-employed in agriculture (see Figure 2.2). Lastly, when the price shock’s spillover effects on the service sector are taken into consideration (last column of Table 7-1.A), the results suggest that the negative impact of the shock is largest for households in the highest consumption quintiles, the better off. The extremely poor experienced a relatively milder impact compared to the poor and non-poor. Overall, however, the long- run impact of the food price shock was largely equalized along the wealth distribution. 12. As might be expected, the results suggest a welfare transfer from net buyers, who suffered a 21 percent decline in consumption in the short-run, to net sellers, who experienced a 28 percent increase in consumption (Table 7.1.B). This finding is consistent with results for the various occupational categories. In particular, with the exception of self-employed workers in agriculture (who more likely to be net sellers), daily-workers and self-employed, non-agricultural workers are negatively impacted by the shock in the short-run. The self-employed in agriculture, who are more likely to own land and produce their own food, benefit in the short-run as well as the medium-run but are marginally worse off in the long-run. In contrast, the self-employed in non-agriculture realized no benefit from the medium-run wage increase and were negatively impacted in the short-, medium-, and long-run. Table 7.1.B: Welfare Impact of a Rice Price Increase Across Different Groups Net benefit ratio Rice Rice Net Without With With consumption as Production as buyers Wage wage wage GE Household category % of total % of total of Rice response response response Effect expenditure expenditure Ag. daily labor 30.11 9.96 84.26 0.37 -0.18 0.12 -0.03 Non Ag. daily labor 26.80 6.91 89.66 0.45 -0.20 0.17 0.03 Self-employed in ag. 24.55 32.72 53.05 0.08 0.06 0.15 -0.02 Self-employed in non-ag. 23.26 11.42 81.54 0.05 -0.12 -0.08 -0.24 Net buyer of rice 25.47 4.64 0.20 -0.21 -0.04 -0.20 Net seller of rice 22.13 53.13 0.09 0.28 0.39 0.21 Source: Background paper by Jacoby and Dasgupta (2012) commissioned for this poverty assessment. Data source: HIES 2000, 2005, and 2010. 7.3 Food Price Increases and Wage Adjustments Over Time 13. Given the importance of medium-term wage adjustments in cushioning the impact of a price shock, understanding the pace and behavior of wage adjustments becomes crucial, particularly for designing policy interventions to mitigate the adverse short-run impacts. In order to investigate real wage patterns for the average Bangladeshi worker over the last decade, this section utilizes two data sources: (i) the Monthly Statistical Bulletin; and (ii) the HIES. Evidence from the Monthly Statistical Bulletin 14. The first data source used is monthly wage data from the Monthly Statistical Bulletin (Department of Agricultural Marketing, Bangladesh), which collects both rural and urban wage information. Urban wages are compiled for unskilled workers, which includes carpenters as well as helpers and daily laborers at construction sites, in seven major cities. Rural wages are drawn from data from 25 districts. Figure 7-5 plots average monthly urban and rural real wages for the 2001-2011 period. 98 Figure 7-5: Real Rural and Urban Wages in Bangladesh A 2.4 Urban and Rural wage deflated by CPI (Base = December 2010) 2.2 2.0 Taka in 100 1.8 1.6 1.4 1.2 1.0 Urban_Wage Rural_Average B Urban and Rural wage deflated by FCPI 2.60 (Base=December 2010) 2.40 2.20 2.00 1.80 Taka in 100 1.60 1.40 1.20 1.00 Urban Wage Rural Wage C 14.0 Urban and Rural Wage adjusted by coarse rice price 12.0 Pre-food price shock Food price shock Wage adjustment period 10.0 8.0 6.0 Rising real wages? 4.0 Urban_Wage Rural_Wage Note: The monthly wage data is obtained from Monthly Statistical Bulletin, Department of Agricultural Marketing (DAM), Bangladesh. Real wages are deflated by either the general consumer price index (CPI), food consumer price index (FCPI), or coarse rice prices. The rural sample excludes rural wages from mega cities. Urban wages are for unskilled workers (e.g., helpers in construction sites, carpentries, and other sectors). The base year is set to 2009/2010. Source: Background paper prepared by Zhang et al. (2012) for this poverty assessment report. 15. In Figure 7-5.A, both rural and urban nominal wages are deflated using the general Consumer Price Index (CPI). The figure demonstrates that the gap between rural and urban real wages has narrowed throughout the sample period, particularly since 2008. Moreover, the rate of growth of rural real wages 99 has outstripped the growth rate of urban real wages. In particular, while rural real wages witnessed a rapid increase, urban real wages were stagnant throughout most of the period, beginning to rise only in early 2008. As food still accounts for the largest share of expenditure for most Bangladeshis, we also examine the trends of real wages implied by the food consumer price index (FCPI) as opposed to the general CPI. Figure 7-5.B depicts urban and rural real wages deflated by the FCPI. Patterns are very similar to those observed using the CPI. Rural real wages were relatively stagnant prior to the mid-2000s. However, while the growth of rural real wages accelerated since 2005, urban real wages stagnated and even declined between 2003 and 2007, only to bounce back to levels observed in the early 2000s by late 2008. 16. Rice is the main staple food in Bangladesh. It constitutes nearly two-thirds of total per-capita caloric intake (Bhuiyan et al. 2002). Among the three types of rice that are produced and consumed in Bangladesh (coarse, medium, and fine rice), coarse rice is consumed by the majority of households, particularly the poor. Thus, the price of coarse rice serves as another suitable inflation indicator. Figure 7- 5.C plots patterns of real wages deflated by local coarse rice prices. Rising food prices in 2007 and 2008 had a significant negative impact on real wages in both rural and urban areas. In 2001, workers in cities could buy about 10 kg of rice using one day’s wage, while a rural worker’s daily wage could afford about 6 kg of rice. However, at the height of the food price crisis in 2007, daily wages in urban and rural areas were worth only about 5.5 kg and 4 kg of rice, respectively. By 2011, the purchasing power of daily wages had bounced back into the neighborhood of 8 kg of rice. Comparing the initial and final data points within the sample period, the data reveal that urban real wages have actually plummeted towards the end of the decade whereas rural real wages have increased. Evidence from HIES Data 17. The key advantages of monthly wage data are high frequency and long-term coverage. However, an important drawback of the Monthly Statistical Bulletin is that the sample covers only a small group of professions, mainly unskilled laborers. To check the robustness of the results presented above, the section makes use of the nationally representative Household Income and Expenditure Surveys (HIES). HIES provides wider coverage but is only conducted every five years. The yearly wage trend is estimated using the latest four waves of HIES – 1995, 2000, 2005, and 2010. Figure 7-6: Real Wages A. Real Wage in Bangladesh B. Real Wage in Bangladesh (peak season) (lean season) 250 250 Daily Wage (taka) 200 200 Daily Wages (taka) 150 150 100 100 50 50 0 0 1990 1995 2000 2005 2010 2015 1990 1995 2000 2005 2010 2015 Male Female Male Female Note: Computed by authors based on Bangladesh Household, Income and Expenditure Survey (HIES) in 1995, 2000, 2005, and 2010 data. Rural consumer price index (CPI) is used to adjust real wages. Source: Zhang et al. (2012). 100 18. Table A7-1 and Figure 7-6 present average rural wages in peak and lean seasons for males and females according to the HIES community questionnaires. In Panel B of Table A7-1, nominal wages are deflated using the general CPI in order to compute real wages (with 2010 as the base year). Figure 7-6.A and Figure 7-6.B display trends in real wages for males and females during both agricultural peak and lean seasons for the years 1995 to 2010. Both figures demonstrate that real wages experienced moderate growth, ranging from 4.2 percent for male workers in peak seasons to 7.2 percent for female workers in lean seasons over the five-year period from 1995 to 2000. In the subsequent period, 2000 to 2005, real wages declined for both male and female workers regardless of season. Since 2005, real wages have dramatically improved, experiencing annual growth rates in the neighborhood of 10 percent. In particular, the pace of female real wage growth has overtaken the pace of its male counterpart and, consequently, the male-to-female wage gap narrowed. 19. Since the CPI is only computed at the national-, urban- and rural-levels, it potentially masks spatial differences in price levels. To explore this further, two alternative deflators are used: the Basic Need Price Index (BNPI) and coarse rice prices. Panel C of Table A7-1 presents estimates using the BNPI, which measures inflation by changes in the national upper poverty line over the 2000-2010 period.89 First, real wages are computed at the rural-division-level using the region-specific BNPI. Then, national rural real wages are derived based on division-level real wages. The real wage trends implied by the BNPI are displayed in Figure 7-7. Deflating nominal wages using the BNPI, the magnitude of the real wage rate increase observed in Figure 7-6.B is significantly attenuated. Furthermore, the basic trends continue to hold; real wages grew much faster between 2005 and 2010 compared to the 2000 to 2005 period. 20. Panel D of Table A7-1 and Figure 7-7: Real Rural Wages Using Alternative Cost of Figure 7-7 present average real wage Living Indexes estimates using coarse rice prices as the inflation indicator. Nominal wages are 180 first deflated using the coarse rice price index. Then, national average real wages are computed using local real wages. The 160 estimates show that real wages grew Daily wage Rice much faster in 2005-2010 than in the (taka) earlier period, 2000-2005. In fact, 140 BNPI estimates using the coarse rice price index suggest that real wages dropped between 2000 and 2005 and that the real wage 120 growth rate in the 2005-2010 period is 1995 2000 2005 2010 2015 slower than reported in Panel B. Perhaps this discrepancy between estimates using Note: Computed by authors using data from the Bangladesh Household, Income and Expenditure Survey (HIES) in 1995, 2000, 2005, and 2010. Regional BNPI the two different inflation indicators is and community-level rice prices are used as cost of living indexes, separately. attributable to the rapid rise in rice prices Source: Zhang et al. (2012). relative to other items in the consumption basket during the peak of the food price crisis in 2007 and 2008. 21. However, regardless of the type of adjustment, the observed wage patterns using the HIES data is consistent with patterns observed using the Monthly Wage Bulletin; rural real wages have dramatically increased over the latter half of the decade. 22. To explore whether or not urban wages exhibited patterns similar to rural wages, Table A7-2 presents nominal and real daily wages in cities. While HIES collects a rural community module, no such 89 See Appendix 1: Inflation Annex for a detailed description of the BNPI. 101 module is collected for urban areas. As a result, the urban wage data is obtained from the information provided by interviewed households. Even in cities, a significant portion of people work in the informal sector and earn daily wages. For those earning monthly salaries in the formal sector, salaries are converted into daily wages using the average number of working days per month. For those reporting incomes from both salaries and daily wages, the average daily wage is computed based on monthly salaries derived from formal jobs and daily wages earned in informal jobs, each of which are weighted by the respective number of working days. As in Table A7-1, the reported real wage estimates are computed using both the CPI and BNPI. 23. Panel A of Table A7-2 reports urban nominal wages in 2000, 2005, and 2010. Female wages more than doubled between 2000 and 2005. By comparison, male wages increased by only 43 percent during the same period. In the later period of 2005-2010, both male and female wages surged by more than 55 percent. Interestingly, nominal wage patterns do not follow those of real wages, suggesting that inflation fluctuates substantially from year to year. In Panel B, nominal wages are deflated using the CPI. After the price adjustment, the rapid wage growth suggested by the patterns of nominal wages becomes less remarkable. Female real wages rose by 67 percent during 2000 to 2005 but by only three percent during 2005 to 2010. The growth rate of real wages for males was only 13 percent and 10 percent between 2000-2005 and 2005-2010, respectively. 24. Panel C of Table A7-2 presents real wage estimates adjusted using the BNPI. Since the BNPI implies a higher inflation rate than the CPI during the 2005-2010 period, the observed drop in real wages is expected for this period. Nevertheless, the finding that female real wages quickly caught up to their male counterparts in the period of 2000-2005 continues to hold. Interestingly, the surge in female real wages between 2000 and 2005 resembles the pattern seen for female wages using the CPI. 25. Since women account for a small share of the labor force, national patterns of wages more closely resemble patterns observed for males. On the whole, real wage patterns derived from the HIES are largely consistent with those observed from the monthly wage data. 26. The evidence presented in this section points to a striking pattern. The escalation of real wages did not occur until the second half of the 2000s. Unlike the rapid growth in real wages in rural areas, the growth of urban real wages was lackluster. The monthly wage data suggests that the food price shock between 2007 and 2008 immediately hit households; by 2009, wages began to adjust to the new price regime and bounced back to their pre-shock level (Figure 7-5.C). Interestingly, while the wage adjustment took place in both urban and rural areas, the urban wage increase only brought wages back to pre-price- shock levels whereas the rural wage increase more than compensated for the wage decline. Moreover, the real wage pattern implies that urban wages fluctuated more than rural wages. Finally, the evidence also suggests that Bangladesh has entered an era of rising wages. Whether the increase in wages will continue or level off after reaching its pre-shock level remains to be seen. Given recent global experiences, food price volatility is a new source of vulnerability, and countries like Bangladesh need to ensure that appropriate policy initiatives are in place in order to mitigate its potentially adverse impacts, especially in the short-term. 7.4 Policy Measures Adopted to Deal with the 2008 Food Price Crisis 27. Historically, rice price stabilization, alongside the ability to secure self-sufficiency in rice consumption, have been important policy considerations for Bangladesh. Major shortfalls in domestic rice production led to famines in the 1970s. Since the liberalization of imports in the 1990s, the private sector rice trade has played a large role in maintaining stable prices as well as increasing investments in agricultural technology (improved seeds, increased fertilizer use, and irrigation), factors that have allowed Bangladesh to nearly achieve self-sufficiency in rice. The 2007-2008 period was an exception. Despite 102 being the key supplier of rice, India banned rice exports in early 2008, which dramatically pushed up domestic rice prices in Bangladesh. Between November 2007 and April 2008, average wholesale rice prices were 45 percent higher in real terms than one year earlier (Dorosh and Rashid 2012). The Government of Bangladesh scrambled to find alternative suppliers in countries like Thailand, Vietnam, Myanmar, China, and Pakistan. In the end, given India’s agreement to export fixed quantities, Bangladesh’s total rice imports reached 1.7 million tons, which arrived largely after the crisis was over, in 2007-2008. 28. By drawing down public stocks and increasing public sector imports, the Government injected 702,000 tons of rice into Bangladeshi markets through its Public Food Distribution System (PFDS) during the 2007-2008 period. By the end of April 2008, public rice stocks were only 220,000 tons and wheat stocks were 120,000 tons. The Government was able to inject these stocks into local markets by expanding its safety net system, which is comprised of large food-based programs that are linked to the PFDS and are progressively targeted. The allocation for safety nets increased to 2.3 percent of GDP in 2008-2009, compared to 1.7 percent in 2007-2008. The largest increase in program expenditures was for Open Market Sales (OMS), a program that provided sales of a fixed amount of coarse rice per-day at a subsidized price in urban areas, and Food for Work (FFW), a public works program that provided rice to laborers in rural areas in return for working to develop local infrastructure. Aside from increased expenditures on these and other food-based safety net programs that are specifically designed to be amplified during times of crisis, the Government allocated an additional $125 million for the new cash- based workfare program targeted to the poorest, a program that is currently known as the Employment Generation Program for the Poorest (EGPP). While these programs suffered from weak administrative capacity, the Government was still able to temporarily scale up program benefits in response to the food price shock of 2008.90 29. Despite relatively bountiful harvests, rice prices in Bangladesh have steadily increased since January 2010. In December 2010, the average wholesale price of coarse rice was 36 percent higher than in December 2009. In rural Bangladesh, food price inflation was over 11 percent in 2010. Under this new regime of price volatility, the availability of and access to rice thus remains a key policy issue that has implications for the management of the PFDS in Bangladesh. Box 7-1: Safety Net Programs Linked with the PFDS Open Market Sales (OMS): This is a program designed to sell coarse rice at a subsidized price of Tk 25/kg for up to 5 kg per household using designated dealers. The government procures rice from both local and international markets and provides it to designated dealers at a price of Tk 23.50/kg, yielding a commission of Tk 1.5/kg. Since the 2008 price crisis, the government has increased the number of OMS dealers in Dhaka city to 150 as of January 2011, and has extended its coverage to all divisional cities, 54 district towns, and 379 upazila (sub-districts) involving some 2975 dealers. Fair Price Card (FPC): From an operational point of view, this program is similar to the OMS. The only difference is that it is a targeted program in which the beneficiary must have a Fair Price card. A Fair Price cardholder gets 20 kg of rice every month at a price of Tk 24/kg. A committee, composed of government officials, people’s representatives, and school teachers, selects beneficiaries from among the poor who do not benefit from any other social safety net program. Recently, the government has increased the number of cards available to over two million to cover ultra-poor households, village defense police, and third- and fourth-class employees of public and private organizations. Vulnerable Group Feeding (VGF): This is a program designed to provide food grains to selected poor households, 90 These programs are discussed in greater detail in Chapter 8. 103 which receive 10 kg to 20 kg of rice per month during a period of distress. The goal of the program is to ensure food security and to smooth consumption for beneficiaries. Vulnerable Group Development (VGD): This program assists women by providing a provision of 30 kg of rice or wheat per month. In collaboration with NGOs, this program also trains beneficiaries to develop skills that facilitate the undertaking of income-generating activities. Test Relief (TR): This program is used to generate short-term employment during the rainy season to support activities like cleaning ponds and bushes and making minor repairs to rural roads, schools, mosques, and madrashas. The key difference from the FFW program is that the entire allotment for each project can be monetized and utilized for the purposes of meeting wage and non-wage costs. Workers are paid up to eight kg rice or wheat, or the cash equivalent, for every day of work up to seven hours. Gratuitous Relief (GR): Administered by Union Parishads, this program provides immediate food grants of about 10 to 20 kg of rice to a household for a period of up to 6 months following any disaster. Along with cash benefits, other in-kind transfers, such as blankets and tin, are also provided depending on the nature of the disaster. Food for Work (FFW): The FFW program has been operating in Bangladesh since 1975. It aims to create food- wage employment during the slack season, mostly in construction and the maintenance of rural roads, river embankments, and irrigation channels. A major objective of the program is to provide income to the rural poor during the slack period, when the unemployment rate in rural areas increases. Wage payments are made in-kind (that is, in rice or wheat) rather than in cash. Such a practice is thought to stabilize food grain prices in the market and to improve food consumption and the nutrition of participating households. This program operates in rural areas and employs beneficiaries who must be fit and willing to work on local infrastructure projects in the area. Payment is made based on the amount of earth works completed by each participant. The Public Food Distribution System 30. The stability of rice prices, which is a highly contentious issue, has been a key government policy since Bangladesh’s independence. At the heart of this policy is the PFDS, which was set up in the wake of the Bengal famine of 1943. The PFDS has three main objectives: (i) maintain security stocks against emergencies, such as the food price crisis in 2008 and other weather-related shocks; (ii) maintain stability of food prices; and (iii) maintain food security for the poor population. Essentially, the PFDS seeks to ensure sufficient food imports and domestic procurement of stocks, to maintain the quality of the emergency stocks, and to distribute the remaining stock among various segments of the population. The targeted distribution channels include various social safety schemes (see Box 7-1 for a description of these programs) and special groups, such as the military, police, and low-paid public servants. 31. During fiscal year 2007-2008, a total of 1.3 million metric tons of food grains, comprised of 1.08 million metric tons of rice and 248.5 thousand metric tons of wheat, were distributed through the PFDS channels. For 2008-09, the food grain distribution target was set at 2.9 million metric tons (WFP 2010). In 2009-2010, the Government distributed 1.31 million metric tons. The 2010-11 target for public distribution increased to 2.29 million tons, and in 2011-12, it was 2.1 million tons. Currently, public stocks total 2.77 million, of which 1.6 million tons are expected to be distributed through the various safety net programs (see Table 7.2). Recent trends in the public procurement of grains suggest a renewed focus on ensuring a higher level of public stocks, a focus which is a consequence of hikes in international rice prices and disruptions of rice imports in 2007-08. 32. However, increases in stock levels imply either increased distribution through the various outlet or losses in the quality of the grains. While the direct costs of increased distribution are clearly shown in 104 government accounts, the deterioration of the quality of PFDS food grain, which has not been taken into consideration, has an adverse impact on the nutrient value of the food. This adverse impact is borne by safety net beneficiaries who are forced to consume sub-quality food grains. Indirect costs or leakage of food distribution is also of concern but is not unique to Bangladesh’s food-based programs. In other words, benefits might not reach intended beneficiaries and may instead be lost or captured by program officials. This problem worsens as the number of intermediaries involved in the benefit delivery process increases. Benefit amounts may also be less than stipulated because of unauthorized payments by the beneficiary to program officials. For Bangladesh, the actual level of beneficiary underpayment varies across programs, from about seven percent for the VGD program to more than 20 percent in the case of the FFW (Ahmed et al. 2004). However, leakages and short payments are not entirely due to corruption. Often, program officials use part of the benefits to recover the cost of bagging and transporting the food. Therefore, maintaining good quality storage, effective stock management, and minimizing leakage are important short-term considerations for maintaining an efficient PFDS. Table 7.2: Public Distribution System of Food Grains (PFDS) (‘000 MT) Channels of PFDS FY 11 (Actual) FY 12 (Actual) Budget FY13 Rice Wheat Total Rice Wheat Total Rice Wheat Total Essential Priorities 155 98 253 157 102 259 175 130 305 Other Priorities 16 5 21 16 5 21 20 10 30 Large Employers 0 17 17 0 15 15 0 22 22 Open Market Sales/FPC 1039 147 1186 340 223 563 500 290 790 Sub-total 1210 267 1477 513 345 858 695 452 1147 Food For Works 8 120 128 262 64 326 300 100 400 Test Relief 1 176 177 262 64 326 100 300 400 VGF 114 0 114 159 1 160 400 0 400 VGD 142 122 264 115 168 283 98 173 271 Gratuitous Relief (GR) 33 0 33 50 0 50 80 0 80 Others 61 37 98 51 42 93 45 30 75 Sub-total 359 455 814 899 339 1238 1023 603 1626 Total 1569 722 2291 1412 684 2096 1718 1055 2773 Source: Bangladesh Food Situation Report, April-June2012, Food Price Monitoring Unit (FPMU), Ministry of Food., Food Division, MoDM&R. 7.5 Conclusion 33. Understanding the mechanisms through which commodity food prices affect the economy of a country and the welfare of its people is paramount for designing the appropriate policies to deal with food price shocks. The empirical analysis presented here suggests that over the decade, higher food prices have had short-, medium-, and long-term effects on the economy, with each time horizon exhibiting qualitatively distinct features. 34. The short-term effect, which is driven by the food price elasticity of consumption expenditures, implies the standard result; in the short-run, the poor were hit the hardest as they spent most of their income on food and were the least likely to be able to cushion the price increase (see also Chapter 6). The medium-term effect arises from induced wage responses to food commodity price changes. While the medium-term allowed higher food prices to permeate the economy by increasing wages of agricultural workers and benefiting net sellers, the wage increase was not large enough to mitigate the short-run negative effect experienced by the poor. In other words, the consequence of the medium-term effect was significant attenuation of the degree of the negative short-term effect. However, in the long-run, as higher 105 prices permeated to the non-agricultural sector, individuals and households at the higher end of the consumption distribution were most negatively affected. 35. The findings presented in this chapter suggest an explanation for why the largest contributor to poverty reduction over the last decade was the increase in returns to endowments or characteristics, rather than changes in these endowments (Chapter 3).91 That is, the exogenous price increase, and not necessarily an increase in labor productivity (measured in physical terms), allowed agricultural workers to experience an increase in wages. Moreover, the fact that poverty reduction over the second part of the decade was highly concentrated among rural agricultural households provides further support for this hypothesis, since these households were most likely to benefit from the price shock. This hypothesis and the evidence presented defy conventional wisdom, which asserts the need to diversify away from agriculture in order to reduce poverty. In particular, the data show that while over the first part of the decade, the reduction in poverty was due to diversification, over the second part of the decade, poverty reduction was the result of an increase in the returns to farming. 36. The combined effect of the increase in agricultural wages (medium-term effect) and the increase in non-agricultural sector prices (long-term effect) also provides empirical support to the equalizing growth across the consumption distribution that was only observed over the second part of the decade (Chapters 1 and 2). Thus, the finding that the recent poverty headcount rate for households with heads who are self-employed in agriculture is comparable to that of households with heads who are salaried workers is not surprising. 37. The analysis of real wage trends also suggests that while the food price shock hit households between 2007 and 2008, wages began to adjust to the new price regime, eventually returning to their pre- shock level by 2009 (Figure 7-5.C). While the wage adjustment took place in both urban and rural areas, the urban wage increase brought wages nearly back to their pre-price shock level whereas the rural wage increase more than compensated for the decline. The analysis suggests that Bangladesh may have entered a period of rising wages. Whether or not Bangladesh will continue to experience rising wages will depend on whether the country has reached the Lewis turning point (see Box 4-1). 38. While this is good news for the poor, the implications of rising wages are of utmost importance. In particular, while extremely low wages are demoralizing to workers and families who often live on incomes far below the subsistence level, they have nevertheless helped Bangladesh compete for foreign direct investment against countries like China, India, and Turkey (Zhang et al. 2011; IHS Global Insight 2010; Sincavage et al. 2010). Because of poor infrastructure relative to its FDI competitors, low wages and a large labor force serve as Bangladesh’s major comparative advantages. Thus, if it seeks to remain competitive in an environment of rising wages, Bangladesh must continue to make investments in roads and other forms of infrastructure. 39. Overall, the general equilibrium model presented in this chapter, as well as the analysis of real wage patterns over the 2000-2010 decade, provide plausible explanations for the significant improvements in pro-poor growth observed during the second part of the decade. However, the fact that the poor were most severely affected by the most immediate and powerful effects of the price shock should be underscored. Even though poor farmers, particularly net sellers, most likely benefited from the price shock in the medium-term, the poor living in urban areas were probably the most adversely affected since they faced higher food prices and did not benefit from higher agricultural wages. 91 In particular, returns to farm and non-farm endowments amounted to almost 65 percent of the observed reduction in poverty, which occurred primarily in the farm sector. 106 40. The government’s response of scaling up operations, such as the OMS, in urban areas in the immediate aftermath of the 2007-08 price shock was appropriate. Given the problem of leakages in safety nets in rural areas, in programs such as FFW, the set-up of EGPP, a self-targeted, cash-based workfare program, was also a step in the right direction. Recent evaluations of the program (discussed in Chapter 8) show that it operates with much lower levels of leakage and is able to better select the poorest vulnerable population than in previous years. Presented in this chapter are two key lessons regarding government responses to buffer the negative impacts of price shocks: (i) the timeliness of response to cushion the immediate impact of a crisis; and (ii) proper targeting of the assistance to the poor who most need help. 41. While managing the PFDS has its own costs and challenges, a vast network of programs that allow for the distribution of food to vulnerable populations can also offer solutions to other challenges and problems, such as under-nutrition. As discussed in Chapter 2, the Bangladeshi diet is notoriously low in micronutrients and diversity. Food-based programs could potentially serve as outlets for extensive distribution of fortified foods or micronutrient sachets. The latter is a relatively cheap and effective way to fight micronutrient deficiency and related morbidities (Menon et al. 2007). 107 8. Safety Nets and Vulnerabilities 1. Previous chapters showed that much of the poverty reduction achieved in Bangladesh between 2000 and 2010 can be explained by changes in the population structure and increases in labor income. Over the same period, the allocation of safety net programs dramatically increased. The government provided an average of 12 percent of total annual public expenditures (about 1.5 percent of GDP) for social protection during the period from 2000 to 2008. Since 2008, in response to global food and energy price crises, the allocation was increased to about 14 percent, on average, of the total budget over 2009- 2012, reaching as high as 2.64 percent of GDP in FY11. Such increases in allocation are reflected in the HIES, which shows that the proportion of households covered by a safety net program rose from 13 percent in 2005 to 24 percent in 2010, bringing the total number of households reached by social assistance measures to roughly eight million. This chapter explores some of the ways in which increased investments in social protection may not have been as effective as they could have been in reducing poverty. Based on the analysis, the chapter suggests possible ways to improve the quality of expenditures on safety nets to enhance the focus on poverty reduction. 2. Programs that involve the direct transfer of resources to the poor have an important role to play. They must ensure that the poor are able to meet their basic needs, cope with the impact of economic shocks, and invest in human capital formation in order to be able to benefit from and participate in the growth process. Social Safety Net (SN) programs are indeed important vectors of poverty reduction, as their primary objectives are to reduce the deprivations associated with extreme poverty and mitigate the risk of households falling into (or further into) poverty as a result of a shock – either at the household- or community-level. By mitigating the impact of shocks, well-functioning safety nets have the potential to reduce short-term vulnerability and also improve long-term growth prospects among the poor by preventing inter-generational transmission of poverty. With these benefits in mind, the Government of Bangladesh (GoB) has instituted a number of SN programs to provide cash and in-kind transfers to the poor, and expenditures on these programs have been steadily growing since the mid-1990s. Box 8-1: Issues with Data Comparability Between 2005 and 2010 Between the 2005 and 2010 rounds, the HIES safety net modules differed on three dimensions. First, the safety net module information was collected at different levels between the two years: it was administered at the household-level in 2005 but at the individual-level in 2010a. Second, the main question on safety nets was changed between the two rounds: in 2005, each household was asked to list all safety net programs received by the family, while in 2010, each individual was requested to identify the primary safety net program providing a benefit. Third, the 2005 questionnaire provides information on only ten programs while the 2010 questionnaire lists 30 different types of programs, many of which did not exist in 2005. These three differences in the module between the two rounds are likely to introduce measurement error when measuring changes in access to safety net programs between 2005 and 2010. To minimize this potential measurement error and to ensure comparability across the two rounds, two different variables of access to safety nets were computed. - First, overall access to safety nets is computed at the household-level. In 2005, a household is considered to be a safety net recipient if it receives benefits from at least one program. In 2010, a household is considered to be a safety net recipient if at least one household member receives a SN transfer b. - Second, household-level access to safety net programs is disaggregated for some major programs, such as Vulnerable Group Feeding, Food For Work, Test Relief, Vulnerable Group Development, Gratuitous Relief, and Old Age allowance. a : only household members aged five and above. b : programs that are not targeted to the poor or implemented by non-governmental entities are excluded. 108 3. Using data from the HIES 2005 and 2010, this chapter examines the extent to which safety nets have helped the poor to improve their well-being and cope with negative shocks (see Box 8-1). The analysis will focus on describing changes in the coverage of SNs and their ability to address the high degree of household vulnerability to various shocks. Section 1 presents an overview of various types of safety net programs and trends in SN expenditures. Section 2 discusses the performance of major safety net programs with respect to targeting, coverage, and leakage. Section 3 summarizes the findings on vulnerability, including the nature and the occurrence of idiosyncratic and covariate shocks, and the role of safety nets as a coping mechanism. Box 8-2: Characteristics of Key Safety Net Programs Infrastructure-building programs: Food-for-work (FFW) and Test Relief (TR) distribute food grains (rice and wheat) as wage payments to both male and female workers via labor-intensive public works programs. The Rural Employment Opportunity for Public Asset (REOPA) program, a follow-up to the Rural Maintenance Program (RMP), provides cash wages and training for income-generating activities to participating female beneficiaries. In 2008, the Government of Bangladesh introduced the Employment Generation for Hard Core Poor (later known as Employment Generation Program for the Poorest). This type of program requires participants to engage in physical work for building and maintaining rural infrastructure. They are generally self-targeted because the poor are usually the only people willing to take on onerous, low-paying jobs requiring manual labor. Training programs: The Vulnerable Group Development (VGD) program exclusively targets poor women and provides a monthly food ration for 24 months. Although it was introduced as a relief program in the mid-1970s, it has evolved over time to integrate food security with development objectives. The development package includes training on income-generating activities; awareness-raising for social, legal, health, and nutrition issues; and basic literacy and innumeracy. Beneficiaries of VGD programs are selected by Government administrative structures. Education programs: The Food for Education (FFE) program, established in the early 1990s, distributed monthly food grain rations to poor households that sent their children to primary schools. FFE was terminated in 2002 and was replaced by the cash-based Primary Education Stipend (PES) program. The School Feeding (SF) program distributes micronutrient-fortified energy biscuits to primary school children. Education-based programs have common development objectives that seek to improve the nutrition of children and promote school enrollment, attendance, and reducing dropouts. In addition, the SF program aims to improve students’ attention spans and learning capacities by reducing short-term hunger and micronutrient deficiency. The Government of Bangladesh also pioneered conditional cash transfers and provided cash assistance to girls in secondary schools through the Female Secondary School Assistance Program (FSSAP) conditional on their attendance. The program was redesigned in 2008 to include boys as well as girls from poor families and was renamed the Secondary Education Access and Quality Enhance Program. Relief programs: These programs are designed to mitigate the consequences of disasters like floods, cyclones, and other natural calamities. Currently, only two such programs exist: Vulnerable Group Feeding (VGF) and Gratuitous Relief (GR). Unlike other programs, these programs have no pre-set criteria or conditionality for participation. They are relief programs that seek to help the poor cope and smooth consumption during times of natural disaster. Programs for disadvantaged groups: These programs are unconditional cash transfers and include the Old-Age Allowance Scheme; Allowance for Widowed, Deserted, and Destitute Women; Honorarium Program for Insolvent Freedom Fighters; Fund for Housing for the Distressed; Fund for Rehabilitation of Acid Burnt Women and Physically Handicapped; and Allowance for the Distressed and Disabled Persons. Source: Ahmed et al. 2010. “Income growth, safety nets, and public food distribution.” Background paper prepared for Bangladesh Food Security Investment Forum held during May 25-26, 2010. 109 8.1 The Evolution of the Safety Net Landscape 4. Bangladesh has a long history of implementing a wide spectrum of social safety net programs, which include both cash and in-kind programs, as part of its social protection system (See Box 8-2). Early efforts in social protection were rolled out as emergency relief measures in response to a cyclone or a famine crisis. These efforts instilled a culture of experimentation and innovation: the nature of the crisis determined the nature of instruments and also the number of actors involved in the roll-out of social protection measures. The large numbers of NGOs, which provide an extensive safety net system for the poor in Bangladesh, were also part of these relief efforts. One such example is BRAC (formerly known as Bangladesh Rural Advancement Committee), which was established in 1972 and grew into the world’s largest NGO over time. The current structure of Bangladesh’s rich social protection landscape, in which a large number of programs are poorly coordinated and poorly harmonized, is largely determined by these initial conditions created by both public and private efforts for social protection services. 5. The administrative structure and the methods of implementation for some of these safety net programs have gone through substantive changes over the last thirty years, from being mostly relief- oriented into focusing on poverty reduction and employment generation. For example, food price subsidies were replaced by targeted food distribution. Partnerships were forged NGOs were forged to implement various training and microfinance programs. The government has shown remarkable willingness to evaluate program effectiveness, confront shortcomings, and cancel or modify programs in order to improve performance. For example, the high costs and levels of leakage found in the Palli rationing program influenced the government to abolish and replace it with an innovative conditional cash program, known as Food for Education (FFE), in 1993. As another example, the Employment Generation Program for the Poorest (EGPP), the country’s first cash-based workfare program initiated in 2008, has already gone through numerous iterations of design changes (see Box 8-3). Bangladesh’s willingness and ability to reform safety net programs represent commendable dynamism in safety net policy-making. 6. In the previous budget year Figure 8-1: Social Protection Expenditure as a Percentage of GDP, (FY11-12), the Ministry of Finance 1998-2013 (MOF) financed 99 programs under the caption of “social 3 protection and social empowerment”, which together 2.5 % of GDP constituted about 14 percent of the country’s budget and 2.4 percent of 2 GDP (Figure 8-1). These programs, which vary in size, can be grouped 1.5 into a number of broader categories: food-based 1 FY98 FY99 FY00 FY01 FY02 FY03 FY04 FY05 FY06 FY07 FY08 FY09 FY10 FY11 FY12 FY13 emergency/seasonal relief and public works programs; pensions program; transfers linked to health Note: Budgeted figures are reported for FY13. and education; and cash allowance Source: Bangladesh Ministry of Finance. programs for special groups (Figure 8-2). Emergency/seasonal relief and public works and pension programs comprise relativley larger portions of the budget. A substantial number of smaller programs offer cash allowances to vulnerable groups and other special groups, such as indigenous populations, widows, the disabled, and freedom fighters. 7. Agricultural programs, aimed at supporting farmers, micro-credit, and rural-employment-type programs, and programs addressing climate change are also included in the social protection budget. A 110 sizeable portion of resources (just over seven percent) are spent on numerous small- and medium-sized programs categorized as “others”. Despite variety in the mix of programs, however, social protection expenditures are skewed toward a few large programs. Figure 8-2: Component-Wise Allocations of Social Protection Budget, 2012-13 1543.74 Health and Education 2845.02 957.37 Agriculture Pension 7807.37 Micro credit and Employment Generation 5944.99 Freedom Fighters, Disabled, & Vulnerable groups Climate Change, Environment and Natural Disasters Food Security/vulnerability 1206.54 1006.59 663.61 Others Source: Bangladesh Ministry of Finance. 8. The ten largest programs constitute over 70 percent of the total social assistance budget (see Table 8.1).92 These programs are implemented by six agencies, namely the (i) Ministry of Social Welfare, (ii) Ministry of Food, (iii) Ministry of Disaster Management and Relief, (iv) Ministry of Women and Children’s Affairs, (v) Ministry of Primary and Mass Education, and (vi) Ministry of Secondary Education. A host of other Ministries implement the remaining allocation spread across numerous programs. Table 8.1: Expenditures on Largest Safety Net Programs, 2008 – 2012 (BDT crore, real values) 2008 2009 2010 2011 2012 Old age allowance 472.44 604.48 614.48 553.06 531.72 Open Market Sales 472.81 799.97 820.69 1147 1047.32 Vulnerable Group Feeding 1171.28 818.78 1059.26 841.38 806.95 Vulnerable Group Development 574.80 444.16 440.23 468.42 481.49 Test Relief Food 803.53 670.04 657.85 693.54 693.86 Food for Work 814.12 692.28 685.35 792.03 858.77 Employment Generation Program for the Poorest 729.13 803.07 689.66 620.71 716.12 Stipend for Primary students 384.25 428.99 517.24 546.22 566.33 Secondary and Higher Secondary Level Stipend Program 261.11 357.31 467.10 393.60 337.17 Note: All of these programs had allocations of more than Tk 500 crore in 2012. Source: Bangladesh Ministry of Finance. 9. Social safety net programs have gradually shifted away from food transfers and toward cash transfers. In 2005, the bulk of Bangladesh’s public transfers were allocated to more expensive and “leaky” food transfer programs. Over time, the Government has boosted cash transfer programs and, 92 This figure includes the pension allocation for retired government officials. 111 consequently, their share of total program spending (Figure 8-3). The Food-for-Education program was transformed into a cash-based stipend program, and Cash-For-Work is gradually being incorporated into the FFW program. The EGPP, introduced in 2008, is entirely cash-based. The increased emphasis on cash Figure 8-3: Proportion of households with access to safety nets 2005 2010 4% 6% 87% 12% 76% 24% 6% 15% 3% 2% No Safety Net Cash only No Safety Net Cash only Kind only Cash and kind Kind only Cash and kind Source: HIES 2005 and 2010. transfers reflects recognition of its greater cost-effectiveness and lower risk of misappropriation. Nevertheless, food transfer programs remain an important pillar of Bangladesh’s food security strategy and serve a secondary role in turning over the country’s emergency grain supplies, as illustrated in Chapter 7. 8.2 Safety Nets Coverage, transfer adequacy, and targeting efficiency 10. According to the HIES 2010, SN programs continue to focus on rural areas, as in Figure 8-4: SN Coverage and Poverty by Division (2010) the earlier part of the decade. The vast majority of SN beneficiaries reside in the country-side: 7.2 million recipient families live in rural areas, which is about ten times the number of urban household recipients. Coverage of safety net programs, which varies significantly by region, is closely correlated with division-level poverty rates (see Figure 8-4). For example, Barisal, with the highest poverty rate (39 percent), has the second highest coverage of safety nets among all divisions (34 percent). In contrast, Chittagong and Dhaka, which have the lowest poverty rates in the country (26 percent and 31 percent, respectively), have the lowest coverage of safety nets (19.4 and 18.8 Source: HIES 2010. percent). 11. Similarly, extreme poverty and food poverty are closely correlated with poverty rates at the division-level. Barisal, with the highest proportion of extreme poor (27 percent) and individuals below the food poverty line (10 percent), has the second highest rate of SN coverage. Due to higher poverty rates, rural areas also record higher coverage rates, 29.9 percent as opposed to 9.4 percent in urban centers. These coverage figures have significantly improved over time; in 2005, the coverage of safety nets was 112 negatively correlated with region-level poverty rates. 12. Participation in safety net programs remains Table 8.2: Access to SNs by Per-capita progressive but displays larger inclusion errors in Expenditure Quintile 2005, 2010 2010 compared to 2005 (Table 8.2). In 2010, the poorest quintile had the largest proportion of Quintile 2010 2005 households covered by at least one safety net program 1 39 24 (39 percent). The richest quintile, on the other hand, 2 32 16 accounted for only 10 percent of SN recipients. 3 25 14 However, expansion of safety nets occurred faster 4 20 8 within the richest quintiles: the proportion of beneficiaries in the top 40 percent of the income 5 10 4 distribution increased by 150 percent over the five- Total 24.6 12.6 year period, while the poorest quintile increased Source: HIES 2005 and 2010. coverage by only 62.5 percent. During the same period, coverage of the poorest rose by 63 percent. 13. The adequacy of actual benefits received by SN beneficiaries, as reported by households, still remains very low (Table 8.3). Transfers, whether in-kind or in-cash, are an important income source for recipients, especially among the poorest quintile. Over the 2005-2010 period, SN benefit amounts have only marginally increased in real terms, and these allowances remain small relative to the needs of a typical poor beneficiary. The most generous transfer is for the old-age allowance, with an average transfer of Tk 501 per month, which represents between 28 percent and 44 percent of the poverty line. The average amount of rice given out by the GR program is 15.7 kg of rice, or Tk 378, per beneficiary, which represents between one-fifth and one-third of the poverty line. Similarly, the benefit from the VGF (14.3 kg of rice) is between 17 percent and 26 percent of the poverty line. Table 8.3: Transfer Amounts of Major SN Programs (monthly) Amount received Amount entitled (Taka) Fraction of poverty line (%)* (Taka) VGF 343 (14.3kg) 274 (11.4kg) 17-26 Gratuitous Relief 429 (17.9kg) 378 (15.7kg) 21-33 Old-age allowance 574 501 28-44 Primary school 158 143 8-12 Secondary school 155 154 8-12 *Entitled amount as a fraction of the upper poverty line. Two figures are reported, as the poverty line is location specific. Source: HIES 2010. 14. To assess the relative performance of Bangladesh’s safety net programs, the following indicators are compared between 2005 and 2010: (i) coverage of the poor (share of households in the population below the upper poverty line receiving benefits); (ii) leakage (proportion of non-poor recipient households); (iii) targeting efficiency (share of total program spending accruing to the poor); and (iv) generosity (share of total consumption accounted for by the transfers). 15. The estimations suggest that even though SN coverage of the poor has improved, it nevertheless remains low: one-third of the poor received assistance from at least one social program in 2010 compared to 21 percent in 2005 (Table 8.4). These results also suggest large scale inclusion errors among SN programs. The proportion of SN program recipients who are non-poor increased from 44 percent in 2005 to almost 60 percent in 2010. Similarly, the share of total program spending accruing to the poor dropped from 52.6 percent to 35.3 percent within the five-year period. Average transfer 113 adequacy is also low and worsened over the period: poor households’ share of transfers, in real terms, as a share of total consumption levels has fallen from 22 percent to 11 percent. 16. One possibility is that inefficient SN implementation leads to leakages in the total Table 8.4: Performance of SN Transfers amount of beneficiary entitlements. Several 2005 2010 reasons could explain why allocated resources are Coverage of the poor (%) 20.9 34.4 diverted away from their intended uses. First, poor program design and implementation might cause Leakage (%) 44.3 59.8 part of the transfer to leak to non-poor Targeting efficiency (%) 52.6 35.3 beneficiaries, as shown in Table 8.4. The second Generosity (%) 13.2 8.8 challenge is to prevent the benefit amount from Generosity (for poor) (%) 22.2 10.6 being captured by program officials prior to Source: HIES 2005 and 2010. reaching intended beneficiaries. The extent of this problem increases with the number of intermediaries involved in the benefit delivery process, as in the case of food-based safety nets. Third, benefit amounts may be less than stipulated because of unauthorized payments, made by the beneficiary, to program officials. 17. According to the HIES 2010, recipients receive, on average, 88 percent of expected cash transfers93 and 89.8 percent of in-kind transfers to which they are entitled (see Table 8.5). At the program- level, there is significant variation in the percentage of leakage. IFPRI estimates that the beneficiary-level leakage of transfers can range between two percent and 13.6 percent (Ahmed et al. 2003). However, we earlier note that leakages and short payments are not entirely attributable to corruption. Often, program officials use a portion of benefits to recover the costs of bagging and transporting food. In the case of public works, part of the allocation is frequently used toward the non-wage costs of implementing small infrastructure projects. Table 8.5: Leakage of In-kind and Cash Transfers (monthly) Cash transfer In-kind Share of cash Share of in-kind received (Tk) transfer received (Tk) transfer received (%) transfer received (%) Barisal 497 160 93.5 90.2 Chittagong 195 331 84.0 90.6 Dhaka 350 82 95.7 92.3 Khulna 295 148 78.7 90.9 Rajshahi 264 151 86.1 87.0 Sylhet 316 50 91.0 96.5 Bangladesh 301 160 88.0 89.8 Source: HIES 2010. 18. On the one hand, spending on safety net programs increased between 2005 and 2010; however, little improvement occurred with respect to the adequacy of resources going to the poor. On the other hand, despite improved geographic resource allocation, there was no improvement in the targeting of resources at the household level. Thus, in spite of the increase in the number of cash-based programs, the potential benefits associated with cash transfers, i.e. reduced leakage, were missed. 19. These findings may be attributed, at least in part, to the manner in which safety net programs are 93 Both amounts, entitled and received, are estimated by the households. Thus, the amount entitled may differ from the “true” entitlement as per program eligibility rules. 114 implemented. Lack of coordination and weak administrative capacity, found among most SN programs, present formidable constraints for ensuring the participation of poor households. More than one dozen ministries implement thirty major programs, many of which are similar in design and undermined by weak administrative capacity. Multiple and ineffective targeting mechanisms, along with large numbers of intermediaries in the system (especially for food-based transfers), increase leakage. Box 8-3: Employment Generation Program for the Poorest: An example of a second generation smart safety net In response to the 2008 global food price crisis, the Government of Bangladesh (GoB) launched the Employment Generation Program for the Poorest (EGPP) in 2009. The program seeks to help poor households to cope with vulnerability by providing short-term employment opportunities on local infrastructure projects. Using lessons from similar workfare programs, such as FFW and TR, the EGPP introduced substantial innovations, including: changes to program targeting by concentrating on geographic areas with higher poverty rates and tightening of selection criteria for beneficiaries; the introduction of a quota for female participants; the introduction of cash wage payments through the banking system; and improved program monitoring systems. Implemented by the Ministry of Disaster Management and Relief, the program is heavily focused on results and evidence-based policy-making. Evaluation studies conducted by IFPRI (2012) and the World Bank (2012) find that EGPP performs well with respect to the beneficiary selection process and self-targeting features and experiences low levels of leakage. Program reforms enacted to enhance female participation, with a target of 33 percent, are becoming increasingly successful, with more than 30 percent participation overall, despite wide variability throughout the country. Focus group discussions reveal that a light workload in close proximity to home not only offers an income source for poor women but also an alternative to migration for those who are physically unable to migrate. However, women’s program participation can be hampered by lack of demand, political capture, the presence of ghost workers, and social barriers. These problems are more likely to be prevalent in less poor areas. In general, beneficiaries are satisfied with the move toward payments through bank branches, reporting substantially lower levels of leakage than under the previous payments system. The evaluation also found that distance to banks remained a problem for some beneficiaries, and lack of an attendance register makes it easier to employ ghost workers. Greater project flexibility, in terms of scheme choices, was found to facilitate the creation of useful community assets, but project management is susceptible to political interference, and staff shortages contribute to inadequate monitoring. Early results from the evaluation work are already being used to improve the design and administration of the EGPP in order to enhance outcomes. For instance, the early evaluation work suggested that problems with beneficiary selection were greatest in lower poverty areas, so in order to improve targeting and reduce leakages, the GoB revised the geographic targeting formula by increasing allocations to the poorest areas (from 50 percent to 60 percent) and reducing allocations for the richest areas. To address these challenges with respect to payments to beneficiaries, a payments pilot is underway which seeks to electronically link attendance to payments and make electronic payments via debit cash cards and mobile phones. Results from these pilots are expected to help policymakers make informed decisions regarding payment options that are most accessible for EGPP beneficiaries, who are less likely to own a mobile phone or, prior to the program, have a bank account. 20. Even though current average benefit amounts are thinly spread, simulations suggest that safety net transfers could potentially reduce the level of poverty by 1.5 percentage points if allocated to the poorest segment of the population. In the absence of SN transfers, poverty rates in Bangladesh would be 33 percent (up from 31.5 percent). However, if an amount equivalent to the average SN transfer was given to the poorest families, the poverty rate in Bangladesh would be 27.2 percent, or a 4.3 percentage point reduction.94 94 These results were obtained in the following way. First, SN transfers were subtracted from recipient households. Households were then ranked in increasing order of per-capita income. Starting from the poorest, all households were given the average SN transfer of Tk 488 per month until the budget allocated in the initial 115 21. Since these simulations assign the same transfer amount to every poor household, irrespective of demographic characteristics, the aforementioned numbers do not necessarily reflect reality. Nevertheless, these figures paint a rough picture of the poverty rate that would prevail if resources were targeted to the poorest. These approximate calculations suggest that the GoB could reduce current poverty rates by 14 percent by disbursing the same transfer amount to poor households. Poor resource targeting makes current safety net allocations inadequate to effectively contribute to poverty reduction efforts.95 22. The above analysis suggests on the need for improving the administrative capacity of SNs, particularly the beneficiary identification process. To establish a better targeting system for SNs, the GoB has initiated the development of a database of poor households based on Proxy Means Tests (PMT). The rationale for this initiative is to allow key safety net programs to adopt a more coordinated approach to targeting beneficiaries. 23. The PMT method of targeting, Figure 8-5: Distribution of Beneficiaries Across Consumption which has been adopted in many parts of Quintiles the world, uses household data to identify 100 key characteristics of the poor, which are VGF TR VGD GR Old Age PMT (20% cut off) then used to develop a database of household-level “poverty scores”, which 50 are used to select beneficiaries. A World Bank technical note shows that using a PMT-based targeting system predicts poor households fairly well and potentially 0 1 2 3 4 5 outperforms the current targeting performance of a number of key SNs (see Source: Sharif et al. (2012). Figure 8-5).96 Data Source: HIES 2010. 8.3 Profile of safety net beneficiaries and their ability to cope with shocks 24. Using multivariate probit regression models, we profile descriptive characteristics of SN beneficiaries. The advantage to using these models is the ability to isolate the marginal effect of a particular characteristic instead of the compound effects of various collinear characteristics (e.g. a lower education level may be associated with a higher probability of being in a vulnerable occupation, but the interest is to separate the effects of education and occupation type). Table A8-1 reports various probit regressions, analyzing the marginal effects of the probability of receiving social transfers as a function of different individual- and household-level characteristics. The analysis was carried out at the individual- level and separately for different groups: (a) recipients of any transfer in 2005 (column 1); and (b) recipients of any transfer in 2010 (column 2). 25. The results are consistent with the earlier finding that the SN coverage is geographically well- targeted and progressive. Individuals living in the poorest divisions are most likely to receive a SN (“real”) scenario (Tk 3.81 billion per month) was exhausted. Poverty rates were then obtained using the modified income measure. 95 Since transfer amounts appear to be underestimated, a similar exercise was conducted using an average household transfer of Tk 700 (instead of Tk 488), a 40 percent increase. The total SN budget was accordingly raised by 40 percent to Tk 5.4 billion. The simulated poverty rates were then 24.8 percent, or a reduction of 6.8 percentage points. 96 See Sharif et al. (2012), Developing a Poverty Registry for Bangladesh: Proxy Means Tests Formula Using HIES 2012, World Bank Technical Note. 116 transfer. Relative to Dhaka residents, residents of all other regions have a higher probability of receiving “any type of assistance”, with the poorest region, Barisal, experiencing a marginal effect of 33 percent. 26. Several characteristics increase the probability of receiving safety net transfers, including individuals who: are more vulnerable; are from poorer households; are not working or work as daily laborers (as opposed to salaried workers, employers, and self-employed); are chronically ill; have lower education (none or primary); and have large families or a greater number of dependents. However, some vulnerable groups are less likely to receive social assistance: single parents (adults aged 18 to 59 years old living in households with no other adult and at least one child) and large households. 27. However, analysis from another nationally representative household survey suggests that participation in SNs does not necessarily help households to cope with shocks. According to data collected during September to October 2009, Bangladeshi households faced a wide variety of shocks over the course of one year. More than 50 percent of Bangladeshi households experienced one or more shocks over a one-year recall period. Rural households were more likely to experience shocks and also face a greater number of shocks, and, thus, were more vulnerable. The most common types of shocks experienced by Bangladeshi households are idiosyncratic; health-related shocks (severe disease or death of a household member) are most common (24 percent of all respondents) followed by climatic and environmental shocks (see Santos et al. 2011). 28. Conditional on experiencing a shock, households’ most common coping mechanisms in response to idiosyncratic shocks included the use of savings and loans, help from friends, and depleting assets (see Figure 8-6). Savings and loans, the two most commonly used mechanisms, are mainly used to deal with health shocks; borrowing is also extensively used to cope with asset shocks. The use of savings was reported by 26 percent to 44 percent of households, and the use of loans was reported by 31 percent to 46 percent of households. On the other hand, public assistance, or any form of formal social protection appears to play a negligible role in helping households to cope with shocks. Less than two percent of households report the use of safety nets as one of the top four coping mechanisms against a shock. The relative importance of savings and borrowing, compared to SNs, in dealing with shocks is not surprising given the vast microfinance and informal credit sectors that operate in Bangladesh as well as the relatively low coverage and transfer amounts of safety nets. 29. When faced with covariate shocks due to climatic reasons, households appear less able to cope (Figure 8-7). Nearly 60 percent of households that experience climatic shocks were unable to cope. Savings and borrowing remain the predominant coping strategies, though to a lesser extent than for idiosyncratic shocks. Friends are also a much smaller source of help; only five percent of the households Figure 8-6: Coping Mechanisms for Idiosyncratic Figure 8-7: Coping Mechanisms for Covariate Shocks Shocks Could not cope Could not cope NGO/Gvt Asset NGO/Gvt Reduce essential… Health Reduce essential consumption Help from friends Help from friends Economic Deplete assets Deplete assets Loans Loans Savings Savings 0 20 40 60 0 20 40 60 Source: Santos et al. (2011). Source: Santos et al. (2011). 117 turn to relatives. This is consistent with the notion that in the event of climatic or covariate shocks, households are unable to rely on community-based coping instruments. 8.4 Conclusion 30. The Government of Bangladesh spends almost two percent of its GDP on safety net programs. Since 2008, expenditures on social protection have been consistently increasing, largely in response to the global price shocks of 2008 and facilitated by fiscal space, resulting from strong and steady growth. In addition to helping to reduce poverty and inequality, safety nets can also help to mitigate the adverse effects of shocks, improve livelihoods, and enable households to invest in children’s education and health. Well-designed and well-targeted social safety nets can protect the poor, help them to climb out of poverty, and prevent inter-generational transmission of poverty, all while being fiscally cost-effective. 31. Bangladesh implements a wide spectrum of social safety net programs, which include both cash and in-kind transfers. Safety net programs in Bangladesh have traditionally been skewed toward expensive but “leaky” food transfers. Since the 2005-2010 period, however, the government has increasingly shifted towards cash transfer programs. Moreover, regional allocation patterns of SN expenditures have reversed; while it was unrelated to regional poverty rates in 2005, it became highly correlated with poverty rates in 2010. Regions with the highest poverty rates enjoyed the greatest coverage of SN beneficiaries. Thus, the distribution of SN coverage has become skewed toward Khulna, Barisal, and Rajshahi, three regions which also experienced the largest poverty reductions in the 2005- 2010 period. 32. However, these trends have not yielded overall improvement in terms of minimizing targeting errors at the household-level. Despite the progressive coverage of SNs, coverage for the poor is still low. On the other hand, coverage of non-poor beneficiary households has increased by 36 percent. This suggests that, while SN expenditure targeting across divisions improved, within-division targeting of safety net resources has actually worsened over time. Programs are skewed toward food-based emergencies, seasonal shocks, and unconditional cash allowances for special groups. Yet, safety net resources serve a minimal role in assisting beneficiaries in coping with shocks. 33. Benefit adequacy levels have also fallen over the last decade. For poor households, average SN benefit amounts are currently about 11 percent of per-capita monthly consumption, compared to 22 percent in 2005. This reduction implies that the greater amount of resources allocated to safety nets are distributed across a much wider population, consisting of households that are not necessarily the most in need of assistance. Safety net programs in Bangladesh also suffer from leakages, though the extent of such leakages is debatable. Evidence on leakage is somewhat mixed and suggests high variation across programs. However, food transfer programs are more likely to suffer from higher leakage due to the large numbers of intermediaries involved in the system as well as the difficulties associated with the procurement, distribution, and storage of food. 34. Despite the government’s generous and well-intended efforts, the analysis in this chapter suggests that the full potential of SNs to reduce poverty and vulnerability remains largely untapped. In order to be more effective and to achieve this potential, SN programs need to be: (i) well-timed; (ii) well-targeted; and (iii) well-tailored to meet the needs of the poor. These three points imply that current SN programs need to improve their design and strengthen their institutional and administrative capacities. Furthermore, consolidation of the numerous programs is required to adequately address poverty and/or mitigate vulnerability to poverty in a sustainable way. Bangladesh’s safety net system would be efficient if, in the short-term, programs are consolidated and re-structured according to the aforementioned three principles. 35. Through innovation, the performance of Bangladesh’s safety net programs may improve. Some 118 examples include better targeting of the most needy, greater accessibility of payment mechanisms, and linking portions of unconditional cash allowances to promote human development outcomes, such as maternal health and nutrition, an area in which Bangladesh lags behind other nations. 36. Worldwide, ‘conditional’ cash transfer programs have been used to provide incentives for poor children to participate in health and education programs. One such example is Bangladesh’s Female Secondary School Stipend Program, which gives cash incentives to families to keep their daughters in school. The program, which was set up in the early 1990s, was one of the first conditional cash transfer (CCT) programs in the world and has contributed to narrowing gender gaps in secondary education. Moreover, global experience suggests that CCT programs, when properly implemented, are appropriate demand-side interventions that have had a significant, positive impact on education and, in some cases, health outcomes.97 The maternity incentive scheme, operated by the Government, is another CCT program that provides incentives to mothers who receive better antenatal, natal, and postnatal care. Unconditional transfers, such as old age pensions, also have substantial health ramifications; people become increasingly vulnerable as they age. This is also the case for children, who are at risk for a variety of reasons, and for people with disabilities. Historically, Bangladesh’s safety net policy-making has been very dynamic. 37. The largest constraints faced by SN programs are their weak administrative and monitoring capacities. Program implementation can be facilitated and leakages can be reduced through investments in technology for delivery and information systems. The new, digitalized national ID data can provide a stepping-stone toward the use of electronic identity, which, combined with new mobile phone banking technology or smart cards, can sharply reduce administrative costs and the potential for irregularities and errors. The database of information on poor households, currently in the planning process, can also help to improve SN performance. As an example, in India, the Government operates a national health insurance program with technical support provided by the World Bank. Using biometric data embedded in smart cards to identify program enrollees, members can seek care at any participating health service provider. The Government subsidizes insurance premiums for the poor, and others must pay the full amount of the premium. Service providers are required to provide the same pre-defined services to all subscribers and are unaware of the source of the premium, which prevents the problem of “programs for the poor becoming poor programs”. Once Bangladesh has developed an appropriate system to target the poor, the example of India’s health insurance program potentially serves as an ideal model for Bangladesh to consider in the future. 38. Bangladesh would also benefit if the numerous SN programs, implemented by a multitude of agencies, were consolidated under a common strategy and agency. Through program and agency consolidation, the countries of Brazil and Mexico turned their respective cash transfer programs into the best global practices. In these and other Latin American countries, social protection mechanisms became effective, partly through consolidation and rationalization and partly through the development of state of the art implementation and monitoring capacity. As a next stage, the consolidated programs came under the umbrella of an independent social protection agency. While most of these reforms are incremental, some can be instantaneously undertaken. Usually, the reform process is long and requires a committed government that will seize the opportunity to push a welfare reform agenda that better addresses the needs of the most vulnerable members of society. Therefore, wide-ranging consultations and coalition-building will be key starting points to undertaking such reform efforts. 39. Further innovation in safety nets involves linking safety net beneficiaries to more productive employment opportunities, thereby helping them to graduate out of these programs and, ultimately, poverty. Bangladesh’s demographic trends, discussed in Chapter 5, create challenges as well as 97 World Bank 2009. “Conditional Cash Transfers: Reducing Present and Future Poverty” A World Bank Policy Research Report 119 opportunities. Recent analysis shows that the working-age population will rapidly expand over the next decade. Based on UN projections, this expansion will add an estimated 22 million individuals to the working-age population between 2005 and 2015 (World Bank 2008a). Matching this expanding labor supply to productive employment opportunities, either through overseas migration or growing sectors of the economy, will be important. In designing Bangladesh’s overall social protection system, some of the critical elements will include investments in skills development and education as well as enhanced access to non-farm sector jobs for safety net beneficiaries. 40. As a result of the demographic transition, an emerging challenge is to help workers, who have the means, to save for retirement and to insure against risks. Using its large and vibrant network of micro- finance institutions, a unique advantage to the country, Bangladesh can expand its social insurance coverage. The Government can also help to expand this coverage through establishment of the appropriate regulatory framework. Furthermore, the Government can incentivize banks and insurance companies so that they offer pensions or social insurance to formal sector workers who currently have no access to such services; and it can provide fiscal incentives for current workers to take advantage of such services and the “demographic dividend” window, which is expected to close in approximately 20 years. 120 9. Microfinance Expansion and Poverty Reduction 1. The impact of access to microfinance on poverty may occur via two main channels. One possibility is that microfinance can increase levels of household consumption and income. Holding constant all other factors, participation in microfinance programs yields an income effect, which increases total household consumption expenditure. Another possibility is that microfinance can help to smooth consumption and, thus, protect a household from falling into more severe poverty. To identify the effects of microfinance, studies have examined changes in household assets and their related impacts on income levels. Neither poverty reduction nor consumption-smoothing through microcredit is easy since borrowers must first earn sufficient revenue from a microcredit-financed activity to repay a loan before they can consume, save, or make productive use of profits. For borrowers, this process requires time and resources (e.g., entrepreneurial ability), including a favorable local economic environment. Therefore, poverty reduction attainable through enhanced credit is context-specific and may occur only under certain conditions. 2. The role of microfinance in reducing poverty and vulnerability is one of the most researched topics on Bangladesh. Existing literature is as vast as it is contentious. Given the nearly thirty years of microfinance experience as well as the huge microcredit expansion (from barely one-half million members in 1991 to 30 million members in 2010) in Bangladesh, an analysis of its role in poverty reduction for participating households is essential. Both Chapters 6 and 8 highlighted the important role played by savings and loans as coping mechanisms for poor households who are vulnerable to shocks. Using several data sets, including HIES and a three-period panel survey spanning over 20 years, this chapter investigates the link between microcredit expansion and poverty reduction in Bangladesh. The purpose of the analysis is to verify whether or not the growth in income and expenditure has led to poverty reduction similar to that observed at the national-level over the same period. According to the analysis, participation in microcredit programs is associated with higher income, higher consumption, and lower poverty. The findings also suggest that households who continuously participated in the program over a 20-year span experienced better outcomes with respect to several measurable dimensions. 3. This chapter is organized as follows. The first section discusses trends in poverty and the growth of microfinance. The proceeding section describes the dynamics of household participation in microfinance programs as well as borrowing behavior. Section 3 presents welfare outcomes resulting from participation in microfinance programs. The last section concludes. 9.1 Trends in poverty and microfinance growth over the last two decades 4. Over the last 20 years, poverty reduction was substantial in Bangladesh. The country recorded reductions of 25 percentage points for moderate poverty and 23 percentage points for extreme poverty Table 9.1: Moderate and Extreme Poverty Headcounts in Bangladesh, CBN Method Year Moderate poverty rate Extreme poverty rate Rural Urban National Rural Urban National 1991–92 58.7 42.7 (%) 56.6 43.7 (%) 23.6 41.0 1995–96 54.5 27.8 50.1 39.4 13.7 35.1 2000 52.3 35.2 48.9 37.9 20.0 34.3 2005 43.8 28.4 40.0 28.6 14.6 25.1 2010 35.2 21.3 31.5 21.1 7.7 17.6 Source: Background paper by Khandker and Samad (2012b) commissioned for this poverty assessment. Data source: HIES, various rounds. 121 (Table 9.1). Despite significant progress in poverty reduction, the disparity in poverty between urban and rural areas remains a concern (see Chapter 1). Actual food intake, measured as kilocalories per person per-day, as opposed to total consumption expenditures, can be directly used to measure food poverty (known as the direct calorie-intake method of poverty estimation). For this method, two cutoff points for per-capita daily kilocalorie intake are considered: an upper cutoff, which corresponds to 2,122 kilocalories per-capita per-day, referred to as moderate calorie-intake deficiency; and a lower cutoff, which corresponds to 1,805 kilocalorie per-capita per-day, representing severe deficiency. 5. In line with the overall poverty estimates shown in Table 9.1, the proportions of people with moderately and severely calorie-deficient diets has steadily declined since the early 1990s (Table 9.2).98 Contrary to the poverty trends reported in Table 9.1, however, the urban population actually had higher moderate and severe food deficiency rates than the rural population. In 2005, urban areas experienced a moderate poverty rate of about 28 percent but a moderate food deficiency rate of over 43 percent. Overall, while estimates for moderate poverty are about the same using either measure, estimates of severe deficiency (extreme poverty) portray a more encouraging picture when calorie consumption is used as the measure instead of the cost of basic needs. Table 9.2: Population with Moderate and Severe Deficiency in Calorie Intake (%) Moderate deficiency Severe deficiency Year (< Rural 2,122 kilocalories/person/day) Urban National (< Rural 1,805 kilocalories/person/day) Urban National 1991–92 47.6 46.7 47.5 28.3 26.3 28.0 1995–96 47.1 49.7 47.5 24.6 27.3 25.1 2000 42.3 52.5 44.3 18.7 25.0 20.0 2005 39.5 43.2 40.4 17.9 24.4 19.5 Source: Background paper by Khandker and Samad (2012b) commissioned for this poverty assessment. Data source: HIES, various rounds. Trends in microfinance growth indicators 6. Microfinance activities grew explosively, in terms of both scale and scope following the establishment of Grameen Bank and BRAC in the 1970s. Three main factors, particularly during the 1990s, influenced this explosion: the entry of other major NGO Microfinance Institutions (MFIs); the availability of increased donor funds; and the formation of the microfinance apex body, PKSF, As result, new branches were established in rural areas of Bangladesh, disbursements intensified, and service Figure 9-1: Trend of Microfinance Members in Bangladesh 40 Active members (millions) 35 30 25 20 15 10 5 0 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 Source: CDF (1996-2007) InM & CDF (2008-2010) Grameen Bank (2010) 98 The estimates for 2010 are not yet available from the published results. 122 portfolios expanded; these developments placed Bangladesh’s microfinance sector on a phenomenal growth path. 7. Figure 9-1 plots the Table 9.3: Growth of Microfinance Clientele in Bangladesh (%) number of microfinance members in each year during the period Calendar Year Growth of Active Members Outstanding Borrowers from 1996 to 2010. MFI 2003 15.71 17.12 membership grew from about 8 million to over 34 million during 2004 16.49 15.88 this period. Table 9.3 shows the 2005 17.85 21.25 yearly growth rate of MFIs since 2006 12.50 16.14 2003. MFI membership grew by 2007 14.39 16.11 well over 10 percent per year until 2008, after which the growth 2008 14.47 16.35 became slightly negative. This 2009 -0.55 -9.21 table also shows the growth of 2010 -3.04 0.54 MFIs’ outstanding borrowers. Like members, borrowers also grew at a Source: Background paper by Khandker and Samad (2012b) commissioned for this poverty assessment. rapid pace until 2008 before Data source: CDF 1996-2007, InM & CDF 2008-2010, and MIX market 2012. slowing down. This slowdown in membership and borrowers during the later years of the decade may indicate a certain degree of market saturation. 8. With steady membership growth, loan disbursements and savings mobilized by MFIs also increased steadily, as shown in Figure 9-2 and Figure 9-3. While MFI loan disbursements were just over Tk 15 billion in 1997, they rapidly grew, reaching Tk 370 billion by 2010. Similarly, savings mobilization has also grown rapidly. In the mid-1990s, it was around Tk 5 billion, and by 2010, MFIs had mobilized savings of Tk 162 billion. Figure 9-2: Trend in Loans Disbursed by the MFIs Figure 9-3: Trend of Savings Mobilized by the in Bangladesh MFIs in Bangladesh 400 180 160 350 Annual disbursement (billion Tk.) 140 Net savings (billion Tk.) 300 120 250 100 200 80 150 60 100 40 50 20 0 0 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 Source: Background paper by Khandker and Samad (2012b) Source: Background paper by Khandker and Samad (2012b) commissioned for this poverty assessment. commissioned for this poverty assessment. Data source: CDF (1996-2007) InM & CDF (2008-2010) Grameen Data source: CDF (1996-2007) InM & CDF (2008-2010) Grameen Bank (2010). Bank (2010). 123 9.2 Dynamics of microcredit participation 9. Table 9.4 shows microcredit participation status for the three-period panel survey, ranging from 1991/92 to 2011. The original sample included only participants from Grameen Bank (GB), BRAC, and the Bangladesh Rural Development Board (BRDB-12). However, over time, the BRDB-12 substantially lost its membership but then re-emerged under the new name of the Palli Daridra Bimochan Foundation (PDBF), which is an outfit of the Ministry of Local Government and Rural Development (LGRD), after 1998/99. In addition to the three programs identified in 1991/92, a new NGO, named ASA (Association of Social Advancement), rapidly expanded its microcredit service delivery after 1991/92 and is included as a separate program in subsequent survey years. 10. Membership for the four major programs, GB, BRAC, BRDB, and ASA, significantly changed over time. For example, the membership of Grameen Bank increased from 8.7 percent in 1991/92 to 12.1 percent in 1998/99, and later declined to 10.7 percent in 2010/11. In contrast, BRAC’s membership stayed at around 11 percent over 1991-1998 but then dropped to 6.7 percent in 2011. The rates reported in Table 9.4, however, do not equal actual membership rates since counts of members in multiple programs are included. For instance, while membership in multiple programs was non-existent in 1991/92, 8.9 (31.9) percent of households were members of multiple programs in 1998/99 (2010/11). Because of the increase in multiple memberships over time, actual membership in each program is higher in 1998/99 and 2010/11 than indicated by the figures in Table 9.4. Table 9.4: Household Distribution by Microcredit Program Participation: 1991-2011 Survey year GB BRAC BRDB ASA Other (single Multiple Any program Non-participant only only only only program only) programs 1991/92 (N=1,509) 8.7 11.3 6.3 0 0 0 26.3 73.7 1998/99 (N=1,758) 12.1 11.0 4.5 2.4 9.7 8.9 48.6 51.4 2010/11 10.7 6.7 1.3 7.3 10.6 31.9 68.5 31.5 (N=2,322) Note: Figures in parentheses in the column labeled “Any program” are the percentages of borrower households among participants. Findings of this and subsequent tables are based on 1,509 households from 1991/92 which are common to all three surveys. Sample size is higher in 1998/99 and 2011 because of household split-offs. Each household is counted once even if it participates in multiple programs. Participants of more than one program are accounted for in the column “Multiple programs”, not in the columns for individual programs. Source: Background paper by Khandker and Samad (2012b) commissioned for this poverty assessment. Data source: WB-BIDS surveys 1991/92 and 1998/99, and WB-InM survey 2011. 11. Actual program participation rates are presented in Table 9.5. The data show that over a 20-year period, an increased number of eligible non-participant households joined these programs. In 1991/92, non-participants comprised 73.7 percent of the sample. By 1998/99, about one-third of non-participants had joined microcredit programs. After accounting for membership in multiple programs, Grameen Bank membership increased from 8.7 percent in 1991/92 to 27.4 percent in 2010/11, implying a nearly one percentage point gain per year over this 20-year period. 12. Table 9.5 also presents the distribution of borrowers across programs and years (in parentheses). Non-borrower membership increased across all programs over time. For example, while nearly 89 percent of all members (or 23.3 percent out of 26.3 percent) were borrowers in 1991/92, borrowers accounted for only 82 percent of all members by 2010/11 (or 68.5 percent out of 56.2 percent). 124 Table 9.5: Microcredit Program Participation Rate Among Households: 1991-2011 1. Survey year Other programs Non- GB BRAC BRDB ASA Any program (one or multiple) participant 1991/92 8.7 11.2 6.4 0 0 26.3 73.7 (N=1,509) (8.6) (9.0) (5.8) (0) (0) (23.3) 1998/98 15.1 16.2 8.3 4.1 14.9 48.6 51.4 (N=1,758) (13.6) (10.1) (4.4) (3.6) (11.4) (38.0) 2010/11 27.4 20.9 4.7 23.8 32.9 68.5 31.5 (N=2,322) (21.7) (12.3) (1.3) (19.3) (28.2) (56.2) Note: Sample is restricted to 1,509 panel households (from the 1991/92 wave) which are common to all three surveys. Sample size is higher in 1998/99 and 2011 because of household split-offs. Figures in parentheses are percentages of borrowers. Sum of the figures across columns for 1998/99 and 2010/11 exceeds 100% because of household participation in multiple programs. Source: Background paper by Khandker and Samad (2012b) commissioned for this poverty assessment. Data source: WB-BIDS surveys 1991/92 and 1998/99, and WB-InM survey 2011. Dynamics of household-level microcredit portfolio 13. MFI membership is primarily motivated by the ability to access loans that a poor household would otherwise find difficult to obtain from formal financial institutions, or would pay exorbitant interest rates to borrow from informal lenders. Therefore, one way to measure the benefits of microcredit programs is to assess temporal changes in credit access. Table 9.6 presents the distribution of microcredit borrowing from major programs as well as the amount of combined borrowing from all microcredit sources over the 20-year period. Borrowing increased multitudinously: the total amount borrowed was Tk 9,252 in 1991/92 compared to Tk 20,870 in 2011, a change which implies simple annual growth, in real terms, of more than six percent per year over the 20-year period. Over the same period, very high growth occurred in households from “Other” programs, which were relatively new compared to the original programs surveyed. Table 9.6: Household Average Cumulative Borrowing from Microcredit Programs over Time Survey year GB BRAC BRDB ASA Other All Programs 1991/92 16,289.4 5,276.7 6,453.9 0 0 9,252.3 (N=769) (0.73) (0.71) (0.38) (-) (-) (0.67) 1998/99 25,938.4 6,377.1 6,552.4 6,346.8 4,680.2 13,262.1 (N=1,099) (0.84) (0.95) (0.52) (0.99) (0.86) (0.84) 2010/11 11,597.6 13,452.3 2,501.3 7,760.1 10,849.5 17,005.6 (N=1,770) (0.89) (0.38) (0.58) (0.84) (0.79) (0.73) Note: Findings are restricted to microcredit participants. Loans are CPI-adjusted Tk with 1991/92=100. Loans are cumulative for five years preceding the surveys. Figures in parentheses are sample size in column 1 and share of female loans in columns 2-7. Source: Background paper by Khandker and Samad (2012b) commissioned for this poverty assessment. Data source: WB-BIDS surveys 1991/92 and 1998/99, and WB-InM survey 2011. 14. As part of a social agenda, Bangladesh’s microcredit programs target women more frequently than men to facilitate the obtainment of credit and other financial services. Over the years, over two-thirds of loans were received by women. In 2011, women’s shares were highest for Grameen Bank (89 percent) and ASA (84 percent) and lowest for BRAC (38 percent). Interestingly, women’s share of BRAC- financed microloans was much higher in earlier years (95 percent, for example, in 1998/99). Due to the 125 recent intensification of small micro-enterprises (SMEs), and since SME loan recipients are predominantly men, the BRAC portfolio saw a higher share of disbursements to male members. 9.3 Welfare gains from microcredit participation 15. Table 9.7 shows average values of income, expenditure, and poverty for participants and non- participants over the twenty years of the survey. An outcome of particular interest is poverty dynamics over the long period covered by the panel surveys. To compare participants and non-participants, we select a set of four indicators: income, expenditure, moderate poverty, and extreme poverty. Both income and expenditure are in real terms (in 1991/92 Tk). The poverty line is based on the cost-of-basic-needs method, as reported in Chapter 1. Households are divided into two groups according to program participation status: participants and non-participants (ineligible households are excluded).99 16. As Table 9.7 shows, between the 1991/92 and 2010/11 rounds, real per-capita income more than doubled, increasing by 104 percent for program participants and by 125 percent for non-participants. During the same period, the share of nonfarm income consistently increased and grew at a faster rate for participants than for non-participants. Food and non-food expenditure shares of total expenditures remained similar for participants and non-participants during the 20-year span, and both types of households experienced similar growth in the share of non-food consumption. However, non-participants experienced a higher growth in per-capita expenditure (89.6 percent) than participants (74.6 percent). Table 9.7: Income, Expenditure, and Poverty by Microcredit Participation Status Indicators 1991/92 1998/99 2010/11 Non- Non- Non- Participants Participants Participants participants participants participants (N=769) (N=1,014) (N=1,554) (N=483) (N=420) (N=334) Per-capita income 521.8 495.6 502.7 523.1 1,066.0 1,114.3 (Tk/month) t=0.74 t=-0.86 t=-0.36 Share of nonfarm income* 0.627 0.613 0.678 0.641 0.765 0.726 t=0.60 t=2.05 t=2.40 Per-capita expenditure 327.3 318.6 440.0 436.9 571.6 604.0 t=1.04 t=0.17 t=-1.71 Share of food 0.813 0.821 0.751 0.762 0.662 0.650 expenditure* t=-1.23 t=-1.15 t=1.59 Moderate poverty (%) 86.3 87.6 60.6 58.2 32.9 34.6 t=-0.67 t=0.88 t=-0.62 Extreme poverty (%) 75.1 78.5 43.6 46.5 16.2 23.1 t=-1.38 t=-1.05 t=-3.19 Note: Monetary figures are in terms of 1991/92 Tk Figures in parentheses are t-statistics of the differences between participants and non- participant households. The analysis is restricted to 1991/92 eligible households only (those who participated and those who could have but did not participate in microcredit programs in 1991/92) who constitute 84, 82 and 81 percent households of the surveyed sample of 1991/92, 1998/99 and 2010/11 respectively. Source: Background paper by Khandker and Samad (2012b) commissioned for this poverty assessment. Data source: WB-BIDS surveys 1991/92 and 1998/99, and WB-InM survey 2011 99 Eligible households are those who participated in microcredit programs and those that could have but did not participate. Such households constituted 83 percent, 87 percent, and 99 percent of households surveyed in 1991/92, 1998/99, and 2010/11, respectively. 126 17. By 2010/11, the difference in expenditure between participants and non-participants became statistically significant. Unlike the expenditure trend, by 2010/11, the incidence of moderate poverty was lower for participants (32.9 percent) than for non-participants (34.6 percent), although this difference is not statistically significant. Similarly, by 2010/11, the extreme poverty rate for participants (16.2 percent) was lower than that for non-participants (23.1 percent) by a relatively wide and statistically significant margin. While poverty substantially declined for both participants and non-participants over the 20-year period, the reduction was greater for program participants than for non-participants. 18. Do the aforementioned trends indicate that microcredit helped to alleviate poverty? Such a simple comparison fails to reveal information about the counterfactuals, i.e. what would have happened to participants had they not been microcredit program members. We must note that average values mask any underlying differences in unobserved factors among participants, who, as a group, may not have been homogenous. Moreover, the same households do not participate in all three survey years – participants in one survey year are not necessarily participants in all the survey years. 19. Using an alternative definition for program participation, we compare the welfare outcomes of two groups of participants (long-term and short-term participants) against those who never participated, even if they were eligible. The first group (long-term participants) includes households that have been continuous participants of microcredit activities for the last 20 years; that is, in all three panels, 1991/92, 1998/99, and 2011. The second group (short-term participants) includes households that have been irregular participants of microfinance activities. More specifically, these households participated in microcredit for one or two periods of the survey but did not remain in the program for the entire 20-year span covered by the panel. The third group (non-participants) includes households that have never participated in microcredit activities during the time period under consideration. 20. Table 9.8 shows inter-group differences for the outcomes of interest. With respect to per-capita income, differences between participants (either long- or short-term) and non-participants are statistically insignificant. However, differences within participants are statistically significant; long-term participants had significantly higher income than short-term participants. Interestingly, although per-capita Table 9.8: Income, Expenditure and Poverty by Level of Microcredit Participation 2010/11 Outcomes Long-term Short-term Non- t-statistics of the differences in outcomes participants participants participants among participation groups (L) (S) (N) tLN tSN tLS (NHH=694) (NHH=461) (NHH=97) Per-capita income 1,172.3 981.0 1,202.3 0.15 -1.05 1.67 (Tk/month) Per-capita expenditure 596.4 551.4 654.0 2.08 3.18 3.01 (Tk/month) Moderate poverty (%) 29.9 36.3 30.3 0.11 1.25 2.85 Extreme poverty (%) 15.3 18.7 21.3 2.18 0.67 1.87 Note: The analysis is restricted to 1991/92 eligible households only (those who participated and those who could have but did not participate in microcredit programs in 1991/92). Long-term participants are those who are found to have participated in microcredit programs during all three survey years (denoted by L), short-term participants are those who participated in microcredit programs in either one or two of the three survey years (denoted by S), and non-participants are those who did not participate in microcredit programs in any of the survey years (denoted by N). The subscripts of t in the t-statistics columns refer to the two groups that are compared. Source: WB-BIDS surveys 1991/92 and 1998/99, and WB-InM survey 2010/11 127 expenditure of participants is significantly less than non-participants, comparison within participants shows that long-term participants have significantly higher expenditures than short-term participants. A similar pattern holds for both moderate and extreme poverty of participants; long-term participants experience lower rates of poverty than non-participants. 21. This exercise using an alternate definition of participation status highlights two main patterns. First, improvements in poverty outcomes were larger for participating households than non-participating households. Second, among participants, continuous microcredit participation yielded better outcomes compared to irregular participation. We must note that, while this exercise helps to describe trends in the outcomes for different groups of participating households, it does not establish causality between the outcomes and microcredit participation. Identifying a causal relationship requires controlling for unobserved factors that influence microcredit program placement and household participation once a program is in place. 9.4 Conclusion 22. Over the last 15 years, rural Bangladesh has experienced a tremendous expansion of microcredit programs. Membership increased from about 8 million in 1996 to 34.6 million in 2010, implying a growth rate of over 23 percent per year. Membership growth was also accompanied by even greater growth in loan disbursements and savings mobilization. Microcredit programs mobilized Tk 7 billion in savings in 1996 which grew to over Tk 160 billion by 2010, indicating phenomenal growth in the realm of 156 percent per year. This chapter investigates whether the growth and outreach of microcredit programs corresponded with improvements in the welfare of participating households. 23. Findings, using data from a three-period household panel, suggest that total household income and the share of non-farm income steadily grew between 1991 and 2011. However, the share of non-farm income grew at a faster pace for microcredit participants than non-participants. Household consumption also grew for all households but at a faster rate for non-participants relative to participants. While both moderate and extreme poverty rates declined for all households, the reduction was larger for participants than non-participants during the 20-year period. Extreme poverty dropped by 2.9 percentage points per year for program participants compared to 2.8 percentage points per year for non-participants. 24. While the analysis presented in this chapter provides a descriptive overview of a wide range of household-level outcomes, it does not establish a causal pathway between access to microfinance and changes in household-level welfare outcomes. Moreover, this chapter does not explore the impact of access to microfinance on household consumption-smoothing, a relationship that is less contentious than the impact of microfinance on poverty. Murdoch (1998) finds that, among Grameen Bank borrowers, access to financial services can reduce a household’s consumption variability by about 50 percent. In other words, when faced with income shocks, households with access to microfinance experience the shock with half the severity of those without access. More recent research by Collins et al. (2009) also shows that poor households value and use financial services more frequently for the management of household consumption and cash flows. Consumption-smoothing reduces household risk and vulnerability, making access to microfinance important for poor households. 128 Part IV: Revisiting the East-West Divide 129 10. Changing Poverty Patterns Across Regions 1. Natural physical barriers, created by the presence Figure 10-1: Integrated (IR) vs. Less of two major rivers (Padma, or the Ganges, and Jamuna, or Integrated (LIR) Regions in Bangladesh Brahmaputra), slice Bangladesh into two regions (Figure 10.1). The West region, also known as the less integrated regions (LIRs), is defined as the areas that are disconnected Integrated from the country’s two largest cities, Dhaka and Chittagong, on account of the rivers. The East region, or (IR) integrated regions (IRs), consists of territories that lie to the west of the Brahmaputra River (Rajshahi division) and Less Integrated south of the Ganges River (Barisal and Khulna divisions, and a small portion of Dhaka division). (LIR) 2. An important difference between IRs and LIRs is that the former has significantly better access to large and growing markets in major metropolitan areas. Likewise, Source: Background paper prepared by Shilpi (2012). because of significant changes in the flow of the major rivers between the monsoon and dry seasons, unreliable water transportation, and a virtual lack of bridges that cross the rivers,100 year-round commuting across LIRs and IRs is not feasible for work. Similarly large differences exist within the two geographical regions. For example, urban and rural areas within a region observe differential access to infrastructure and services. 3. The last Poverty Assessment Report for Bangladesh (World Bank 2008a) highlighted the need for creating economic opportunities to narrow the development gap between the East and West of Bangladesh. The report showed that, while the East was rapidly improving and benefiting from its geographical proximity to growth poles, the Western region of Bangladesh was lagging behind. In particular, poverty headcount figures for the years 2000 and 2005 revealed that while all of the Eastern- most divisions (Chittagong, Dhaka, and Sylhet) experienced large declines in poverty, their Western counterparts (Barisal, Khulna, Rajshahi) remained practically stagnant with respect to poverty rates (see Figure 10-2). Figure 10-2: Poverty Headcount Across Regions 60.0% 56.7% 53.1% 52.0% 51.2% 50.0% 45.7% 46.7% 45.1% 45.7% 42.4% 39.4% 40.0% 35.6% 34.0% 33.8% 32.0% 32.1% 30.5% 28.1% 30.0% 26.2% 20.0% 10.0% 0.0% Barisal Chittagong Dhaka Khulna Rajshahi Sylhet 2000 2005 2010 Source: HIES 2000, 2005, and 2010. 100 The only bridge crossing the Jamuna River, Jamuna bridge, began to operate in 1999. 130 4. Overall, the 2008 Poverty Assessment depicted a bittersweet chronicle for Bangladesh: while growth was occurring, it was taking place in a highly uneven fashion. The report suggested that a combination of factors contributed to this divergence. Among these factors were the relative lack of remittance income, inadequate public infrastructure, like electricity and access roads to markets, and deficiencies in assets and endowments. 5. The latest round of HIES (2010) describes a changed Bangladesh. Comparison of 2010 poverty headcount figures to 2005 estimates reveals a reversal of poverty patterns (see Figure 10-2). Not only did the Western divisions (Barisal, Khulna, Rajshahi) experience larger reductions in poverty (as measured by the poverty headcount ratio), they also achieved levels of poverty in line with their Eastern counterparts (Chittagong, Dhaka, and Sylhet). Dhaka, the division with the lowest poverty rate in 2005, did not see much change in its poverty headcount. An important exception to this pattern is the Northern part of the country, specifically Rangpur, which has over 42 percent of its population living below the upper poverty line. Even so, the estimates show that while the first half of the decade was characterized by uneven growth and poverty reduction across regions, the second half is characterized by the convergence experienced by the lagging regions. 6. Understanding the evolution of regional poverty patterns as well as patterns of other welfare measures is particularly important to inform policies for the next decade. For instance, looking forward, what types of policies should be considered for more equitable regional growth? To help to reap demographic dividends, should skills policies be regionally focused? Should anti-poverty programs be geographically targeted? To shed light on these questions, this chapter examines trends and patterns of regional convergence for selected outcomes as well as the factors that may have contributed to such patterns. 7. The chapter is organized as follows. Section 1 presents results from a poverty decomposition exercise in order to identify the factors that explain regional poverty reduction patterns. This section is followed by a discussion of the underlying factors, namely labor income and demographic changes. Section 3 assesses whether the convergence in poverty rates across regions has led to more equitable access to certain key basic facilities and services for children. Finally, some key conclusions are laid out in Section 4. 10.1 Decomposing Poverty Figure 10-3: Contributions to Poverty Reduction Reduction Across Regions: The East vs. The West 170 Unexplained 8. How different were the East and West experiences in poverty Other Non-labor reduction? To answer this question, 120 the section begins by decomposing Transfers poverty by region following the 70 Capital Paes de Barros et al. (2006) methodology (for details related to 20 Labor income the underlying methodology, see Appendix 4). This decomposition Occupation share -30 corresponds to Part B of Figure 3-8. Adult population Similar to results at the national- -80 level, discussed in Chapter 3, the National East West Consumption- regional results show that changes income ratio in both labor income and the adult Source: HIES 2000, 2005, and 2010. population were the two most 131 important contributors to poverty reduction in both regions; however, they accounted for a larger share of overall poverty reduction in the West relative to the East (Figure 10-3). This result is consistent with the broader story on the drivers of pro-poor growth between 2005 and 2010. Interestingly, the consumption- to-income ratio increased in the East but decreased in the West, suggesting that in 2010, Eastern households were consuming more whereas Western households were saving more relative to the year 2000. Unlike changes in labor income and the adult population, transfers and the share of occupied adults played more significant roles in poverty reduction in the East relative to the West. The next section explores regional divergence-convergence patterns by analyzing several indicators of well-being, with particular focus on labor income and demographic changes. 10.2 Is there Regional convergence or divergence in Socio-Economic Indicators?101 9. A simple linear probability model is used to investigate whether or not convergence occurs with respect to various welfare measures across four broad geographical regions over the 2000-2010 decade.102 10. Analysis of the log of per-capita real expenditure (LRPCE) reveals two striking patterns (Figure 10-4). Figure 10-4: Log of Per-capita Real Expenditure by Region First, the urban-rural gap in LIRs has narrowed, while that the analogous difference in IRs has remained nearly constant. Second, LRPCE in rural LIRs caught up with 7.20 IRs in the second part of the decade, but the gap between 7.00 urban IR and LIR persists. Both trends are consistent with 6.80 the earlier finding that the growth of agricultural incomes 6.60 outpaced the growth of non-farm incomes. In terms of 6.40 convergence-divergence patterns between IRs and LIRs, 6.20 the results show a statistically significant divergence 6.00 Urban-IR Urban-LIR Rural-IR Rural-LIR between urban and rural areas in 2000, irrespective of 2000 2005 2010 their level of integration. Urban IRs enjoyed the highest level of LRPCE. By 2005, urban IRs jumped ahead in LRPCE relative to rural IRs and urban LIRs, whereas Source: HIES 2000, 2005, and 2010. rural LIRs diverged from their IR counterparts. In other words, between 2000 and 2005, the existing gap between urban IRs and rural LIRs grew larger. By the end of the decade, however, some convergence took place. In particular, the LRPCE improved for urban LIRs, shrinking the gap with their urban IR counterparts, whereas rural LIRs saw no difference from their IR counterparts. That is, even as the statistically significant urban-rural gap within and across IRs and LIRs was persistent, the second half of the decade was characterized by greater improvements among the LIRs (see also Table A10-1). 11. The HIES also provides separate information on earnings from log-daily wages and log-salary. Panels A and B of Figure 10-5 provide estimates for daily earnings from log-daily wages and log-salary (including benefits and in-kind income), respectively. Both measures are adjusted using the relevant spatial price and consumer indices. Overall, we observe an increase in spatially-adjusted wages/salaries over the three survey rounds and across all regions (see also the estimates reported in the bottom panel of 101 This section draws heavily from Shilpi (2012), a background paper commissioned for this poverty assessment report. 102 The empirical specification is given by: , where is an indicator of welfare for the survey year t, UI is an index taking a value of one for urban areas in integrated region, ULI (RLI) is an index for urban (rural) areas in less-integrated region, and is an idiosyncratic error term. Equation (1) is estimated separately for 2000, 2005, and 2010. Convergence between urban and rural areas in integrated region will imply: . 132 Table A10-1). In the case of earnings from log-daily wages, some degree of convergence between the urban-IRs and the other regions is evident, particularly over the second half of the decade, when urban- LIRs and both rural IRs and LIRs experienced the largest increases. With regard to log-salary income, virtually no difference is observed between rural LIRs and IRs in any survey year. More importantly, even as urban areas continue to have significantly higher log-daily wages, the estimated coefficients suggest convergence in this indicator between urban and rural areas and between IRs and LIRs. Figure 10-5: Log of Daily Earnings A. Log(daily wage) B. Log(daily salary) 4.50 5.00 2000 2005 2010 2000 2005 2010 4.50 4.00 4.00 3.50 3.50 Urban-IR Urban-LIR Rural-IR Rural-LIR Urban-IR Urban-LIR Rural-IR Rural-LIR Source: HIES 2000, 2005, and 2010. 12. With respect to the number of hours worked, in Figure 10-6: Hours Worked Per Year and by 2000, the average worker from an urban area worked Region approximately 400 extra hours per year compared to the average rural worker, regardless of the level of integration (Figure 10-6). Moreover, with the exception 3,000 of urban-IRs, the estimates suggest a decline in hours 2,000 worked over the years, particularly in rural areas. Hours worked remained nearly unchanged in urban-IRs. 1,000 Furthermore, workers from LIRs and IRs worked 0 approximately the same number of hours in 2005 and Urban-IR Urban-LIR Rural-IR Rural-LIR 2010. Similar patterns are observed for urban and rural- 2000 2005 2010 LIRs. By the end of the decade, we observe a significant divergence in hours worked between urban-IRs and the Source: HIES 2000, 2005, and 2010. other region types. 13. Overall, the observed patterns suggest that the decade gave way to some degree of regional convergence. Nevertheless, significant regional differences in income measures still persist, particularly between urban and rural areas. The next section revisits this regional pattern, focusing on demographic changes, which was the second largest contributor to poverty reduction over the decade. 10.3 Demographic changes Change in Population by Age Groups 14. Figure 10-7 displays percentage point changes in the shares of children (0-14 age group), adults (15-64 age group), and the elderly (more than 64 years old) across two periods (2000-2005 and 2005- 2010) and regions (East versus West). The figure shows that, while the East experienced a larger drop, relative to the West, in the share of its under-15 population between 2000 and 2005, this pattern reversed between 2005 and 2010 (i.e. the under-15 share experienced a relatively larger decline in the West). At the same time, the increase in the working-age population (15-64 age group) share was larger in the East 133 relative to the West during the first part of the 2000-2010 decade. Yet, similar to the pattern observed for the share of children, the pattern for the working-age population was reversed in the second part of the decade; in other words, the increase in the share of this age group was relatively larger in the West. Over the decade, the share of the elderly population also increased. However, the increase in the share of working-age adults exceeded the increase in the share of the elderly and also coincided with a reduction in the share of children. As a result, these demographic changes worked to lower the overall household dependency ratio. Figure 10-7: Change in Population by Age Groups and Divisions 2000-2005 vs. 2005-2010 3 2 Percentage point 1 0 -1 -2 -3 2000-05 2005-10 2000-05 2005-10 East West [0-14] -2.65 -1.74 -1.42 -2.33 [15-64] 2.55 1.32 2.09 1.77 [64+] 0.11 0.41 -0.68 0.57 Source: own estimates using HIES 2000, 2005, and 2010. 15. Since lower dependency ratios are associated with lower poverty rates, the patterns of changes in the age composition of the population are consistent with the patterns of changes in regional poverty, described in Figure 10-2. In particular, the patterns observed in the East (West) during the first (second) part of the decade (i.e. the relatively larger increase in the number of working-age adults in conjunction with the relatively larger decrease in the share of children) help to explain both the diverging (2000-2005) and the converging (2005-2010) poverty patterns observed over the entire decade. Changing family formations in Bangladesh: The East and the West 16. During the first decade of the millennium, the proportion of elderly living with their sons consistently dropped across all divisions, but similar to trends in other demographic characteristics, regional variations are also very marked. In the first part of the decade, the East witnessed a 2.8 percentage point decline in the proportion of older persons who live with their sons and commensurate increases in those living with their daughter, alone, or with others (Figure 10-8). A more dramatic pattern was observed in the East during the second part of the decade. In particular, while the proportion of elderly who live with their sons decreased by 7.1 percentage points, the proportion living with their daughters and with others increased significantly, 3.65 and 2.8 percentage points, respectively. 17. The West, on the other hand, witnessed increases in the proportion of older persons who live either with their sons or alone (1.45 and 1.33 percentage points, respectively) and proportionate decreases in living with their daughter or with others during the first part of the decade (Figure 10-8). Similar to the East, in the second part of the decade, the West witnessed a large decrease in the number of individuals living with their sons, but unlike the East, the percentage of elderly adults living with others considerably 134 Figure 10-8: Change in the Share of Elderly Living Arrangements 2000-2005 vs. 2005-2010 8 Percentage point 6 4 2 0 -2 -4 -6 -8 2000-05 2005-10 2000-05 2005-10 East West With son (married or unmarried) -2.81 -7.10 1.45 -7.5 With daughter (married or unmarried) 0.41 3.65 -2.44 -0.37 With Others 1.94 2.80 -0.35 6.87 Alone 0.47 0.65 1.33 1 Source: own estimates using HIES 2000, 2005, and 2010. increased. These large increases in the proportion of elderly adults living with their daughters, as observed in the East, or with others, as observed in the West, reveal a breakdown of the traditional patterns of co-residence with sons. Convergence in Non-monetary Measures of Welfare: The Human Opportunity Index103 18. Better poverty outcomes are likely to be associated with better access to basic facilities and services. Thus, LIRs are expected to experience relatively greater improvements in these outcomes compared to IRs over the 2005 to 2010 period. This section investigates whether poverty and other circumstantial indicators impacted children’s opportunities to access basic services and facilities, which are required to ensure improvements in health and education outcomes. These outcomes are particularly important in order to arrest the transmission of intergenerational poverty. Recent research shows that much of the variation in adult income is related to family background during childhood, and childhood development programs are critical so that children do not remain poor as adults.104 The analysis presented below makes use of the Human Opportunity Index (HOI) to explore child-level outcomes across urban and rural areas, IRs and LIRs, and changes over the 2005 to 2010 period.105 103 This section draws heavily from a background paper by Abdallah and Sharif (2012), commissioned for this poverty assessment. 104 Engle, P.L., Black, M.M., Behrman, J.R., Cabral de Mello, M., Gertler, P.J., Kapiriri, L., Martorell, R., Eming Young, M., and the International Child Development Steering Group (2007) Child development in developing countries 3: Strategies to avoid the loss of developmental potential in more than 200 million children in the developing world' Lancet 2007; 369: 229–42. Child Development Series Paper No. 3. January 20, 2007. 105 By construction, the HOI synthesizes two factors into a scalar measure. One is the average coverage rate of a basic good or service, and the other is the relative measure of equality of opportunity. Conceptually, the HOI is computed using the following equation: HOI = C (1 – D), where C is the average coverage rate of the service in question and D is the Dissimilarity Index – a measure of dissimilarity or inequality across the groups considered in the calculation of the coverage rate. D can be interpreted as the percentage of people (of the entire population) whose access to the good or service would have to be redistributed among people belonging to groups with below-average coverage rates such that equality of opportunity can be achieved (Barros et al. 2009). Refer to the Appendix 3 for more details pertaining to the estimation of the HOI. 135 19. The set of services considered in the analysis include: (i) children’s access to professional health care services; (ii) access to education among school children aged six to ten; (iii) access to education among school children aged 11 to 15; (iv) children’s access to electricity; and (v) children’s access to sanitation. The set of circumstances considered are poverty status of the family (whether a family is below the upper poverty line), gender of the household head (whether the household head is female), number of family members, religion of the family (whether or not the family is Muslim), and educational background of the household head (an ordinal variable measuring the educational level of the household head). HOI comparisons are made across poor and non-poor households and across regions (urban versus rural as well as IRs and LIRs). In order to track temporal changes in access to services, the HOI is estimated separately for the years 2005 and 2010. The estimated HOI for each of the services are presented below. Access to Professional Health Care Services 20. Professional health care providers include government, private, and NGO doctors. In 2010, Bangladesh’s national HOI for access to professional health care providers among all children was 41 percent; 39 percent in IRs and 44 percent in LIRs (Figure 10-9.A). Given the low coverage rate, the dissimilarity index is also relatively high: 8.2 percent in rural areas and 9.7 percent in urban areas (Figure 10-9.B). Geographically, LIRs have a higher HOI than IRs. Moreover, the dissimilarity index is higher in IRs than LIRs and is highest in urban IRs. This pattern of the dissimilarity index suggests that IRs and urban areas have greater inequality with respect to access to professional health care services. In 2010, the impact of poverty on the dissimilarity index at the national-level is high at 5.4 percent, contributing 61 percent to the national dissimilarity index, and is highest in urban IRs at 6.8 (contributing 52 percent to the urban IRs dissimilarity index, see Table A10-2).106 In terms of contribution to the dissimilarity index, Figure 10-9: Children’s Access to Health Professionals A. HOI - 2010 B. Dissimilarity Index - 2010 Bangladesh IR LIR Bangladesh IR LIR 51 52 12.9 48 10.9 41 44 42 8.9 9.8 8.6 9.7 8.5 39 39 36 7.9 8.2 Over-all Rural Urban Over-all Rural Urban Dissimilarity Index (%) C. Changes in HOI –over Time D. Changes in Dissimilarity Index over Time HOI 2005 HOI 2010 D-Index 2005 D-Index 2010 51 41 44 15.1 39 39 11.710.9 27 25 26 26 29 8.9 9.3 7.9 8.2 9.7 7.1 5.3 Bangladesh IR LIR Rural Urban Bangladesh IR LIR Rural Urban Source: Background paper prepared by Abdallah and Sharif (2012). Data source: HIES 2005, 2010. 106 The full set of results is presented in the Tables Annex, Tables A.10-1-A10-4 136 the impact of poverty is greatest in rural LIRs (76 percent) and lowest in urban LIRs (13 percent). With regard to temporal changes, the national HOI significantly increased in all regions between 2005 and 2010 (Figure 10-9.C), but the dissimilarity index increased at the national-level, which was primarily driven by the increase experienced in rural areas (Figure 10-9.D). Access to Education of Children aged 6 – 10 years 21. In 2010, the national HOI for access to education for six to ten year-old children was 80 percent. Like access to professional health care providers, the HOI was higher in LIRs relative to IRs and in urban relative to rural areas (Figure 10.10, panel A). The dissimilarity index (Figure 10.10, panel B) explains a substantial portion of the gap in access to education between LIRs and IRs; it is higher in IRs (4.4 percent) than LIRs (3.1 percent). Furthermore, the dissimilarity index was highest in urban IRs (5.4 percent). In 2010, the contribution of poverty to the dissimilarity index is 1.6 (or about 42 percent, see Table A10-2). In rural areas, the impact of poverty has very little variation across regions. In urban areas, the impact of poverty on the dissimilarity index is greater relative to rural areas as well as nationally, except for LIRs, where the impact is smallest. Between 2005 and 2010, significant improvements occurred with respect to both coverage (Figure 10-10.C) and the dissimilarity index (Figure 10-10.D) at the national- and regional-levels. Figure 10-10: Children’s Access to Education (6-10 years old) A. HOI - 2010 B. Dissimilarity Index - 2010 Bangladesh IR LIR Bangladesh IR LIR 88 5.4 84 83 84 4.4 4.3 80 80 82 3.7 3.1 3.4 3.8 3.2 78 77 2.4 Over-all Rural Urban Over-all Rural Urban HOI (%) Dissimilarity Index (%) C. Changes in HOI –over Time D. Changes in Dissimilarity Index over Time HOI 2005 HOI 2010 D-Index 2005 D-Index 2010 84 84 80 78 79 80 6.3 77 5.1 5.8 74 74 4.4 4.3 4.9 4.3 71 3.7 3.1 3.4 Bangladesh IR LIR Rural Urban Bangladesh IR LIR Rural Urban Source: Background paper prepared by Abdallah and Sharif (2012). Data source: HIES 2005, 2010. Access to Electricity Children aged 11 – 15 years 22. The opportunity for access to electricity is defined as whether or not children ages 11 to 15 have access to an electricity connection (i.e. whether the home where the child resides has access to an electricity connection). Access to electricity is an important service since it is a significant determinant of 137 educational attainment. In 2010, the national HOI for access to electricity was 48 percent; 53 percent in IRs and 40 percent in LIRs (Figure 10-11.A). In particular, the electricity HOI was as high as 90 percent in urban IRs and as low as 33 percent in rural LIRs. Moreover, the HOI was significantly higher in urban areas (85 percent) relative to rural areas (37 percent). In general, the dissimilarity index was high: 14 percent nationally and highest at 17 percent in rural LIRs (Figure 10-11.B). The dissimilarity index was lower in IRs than LIRs and in urban relative to rural areas. In 2010, the impact of poverty to the national dissimilarity index is 6.0, contributing about 43 percent to the observed inequality in access to electricity (see Table A10-3). The impact of poverty on the dissimilarity index is largest in rural LIRs at 11.5, contributing about 67 percent, and is smallest in urban IRs at 1.3, contributing about 40 percent. Figure 10-11: Children’s Access to Electricity (11-15 years old) A. HOI - 2010 B. Dissimilarity Index - 2010 Bangladesh IR LIR Bangladesh IR LIR 90 85 72 14.2 13.8 16.0 15.3 15.4 17.3 48 53 10.3 40 37 40 33 5.2 3.3 Over-all Rural Urban Over-all Rural Urban HOI (%) Dissimilarity Index (%) C. Changes in HOI –over Time D. Changes in Dissimilarity Index over Time HOI 2005 HOI 2010 D-Index 2005 D-Index 2010 85 23.5 75 22.8 20.2 16.7 16.0 48 46 53 40 37 14.2 13.8 15.3 36 9.0 25 26 5.2 Bangladesh IR LIR Rural Urban Bangladesh IR LIR Rural Urban Source: Background paper prepared by Abdallah and Sharif (2012). Data source: HIES 2005, 2010. 23. Temporal changes in the HOI for access to electricity reveal that significant improvements took place over the period from 2005 to 2010. The national HOI increased from 36 percent to 48 percent, with the largest improvement occurring in LIRs (Figure 10-11.C). Significant improvements occurred in the dissimilarity index, which decreased nationally and across all regions considered; children from rural areas and LIRs experienced the largest percentage point declines (Figure 10-11.D). Access to Sanitation Children aged 0 – 5 years 24. The opportunity for access to sanitation is defined as whether or not children ages zero to five have access to a sanitary latrine (i.e. whether or not the home where the child resides has access to a sanitary latrine). Overall, the HOI for access to sanitation is lower than for access to electricity. In 2010, the national HOI for access to sanitation was only 39 percent; 39 percent in IRs and 37 percent in LIRs (Figure 10-12.A). Moreover, large differences persist across regions. In particular, the sanitation HOI was as high as 64 percent in urban IRs and as low as 32 percent in rural IRs. Like the HOI for access to electricity, the HOI for sanitation access was significantly higher in urban areas (62 percent) relative to rural areas (33 percent). In general, the dissimilarity index was high: 15.2 percent nationally and highest 138 at 19 percent in rural IRs (Figure 10-12.B). The dissimilarity index was lower in IRs than LIRs (nationally) and in urban areas compared to rural areas. Figure 10-12: Children’s Access to Sanitation (0-5 years old) A. HOI – 2010 B. Dissimilarity Index - 2010 Bangladesh IR LIR Bangladesh IR LIR 64 17.5 19.0 16.8 62 55 15.2 15.0 16.4 13.4 39 39 37 33 32 34 9.4 9.2 Over-all Rural Urban Over-all Rural Urban HOI (%) Dissimilarity Index (%) C. Changes in HOI –over Time D. Changes in Dissimilarity Index over Time HOI 2005 HOI 2010 D-Index 2005 D-Index 2010 31.6 68 62 21.1 23.2 45 39 16.415.0 16.4 17.5 36 39 37 30 33 15.2 22 10.8 9.4 Bangladesh IR LIR Rural Urban Bangladesh IR LIR Rural Urban Source: Background paper prepared by Abdallah and Sharif (2012). Data source: HIES 2005, 2010. 25. Temporal changes in the HOI for access to sanitation reveal lackluster improvements over the period from 2005 to 2010. The national HOI increased from 36 percent to 39 percent, where the largest improvement took place in LIRs (Figure 10-12.C). More significant improvements were achieved in terms of the dissimilarity index, which decreased nationally and across all regions; children from rural areas and LIRs experienced the largest percentage point declines (Figure 10-12.D). 26. To summarize, the estimated HOIs presented in this section provide a mixed set of results. On the one hand, the HOI for access to education was highest relative to all other services considered, and it substantially increased over the 2005-2010 period. Furthermore, its corresponding dissimilarity index was relatively low, suggesting that the extent of inequality in educational opportunity, caused by poverty and other circumstances, is not high. On the other hand, access to professional health care services was generally low for all children, irrespective of location, and the corresponding dissimilarity index was highest for children from urban IRs. 27. The analysis of temporal changes between 2005 and 2010 reveals that LIRs experienced larger gains in the HOIs for access to professional health services, access to sanitation, and access to electricity. A similarly positive finding is that the impact of poverty on the dissimilarity index has significantly declined for nearly all of the outcomes under consideration (see Table A10-2 to Table A10-5). In general, the results reveal that children from less integrated regions had relatively better access to professional health services and education whereas children from integrated regions had better access to electricity and sanitation. Moreover, while access to education was generally high and access to professional health services was generally low, access to electricity and sanitary facilities were dismally low in rural areas relative to urban areas. 139 28. The narrowing of HOI gaps between IRs and LIRs for the various services is consistent with the improvements in consumption and income patterns described in this chapter; patterns which would tend to increase the demand for and access to services, particularly for education and health care. However, we emphasize that the descriptive analysis presented in this chapter cannot rule out the possibility that a deliberate focus on improving services in the economically lagging LIRs could also have contributed toward closing these gaps. 10.4 Conclusions 29. Poverty headcount figures for the first part of the decade reveal that, while poverty decreased in both rural and urban areas, the reduction was highly uneven, particularly favoring the Eastern part of the country. Analogous estimates for the later part of the decade indicate that these East-West poverty differences significantly diminished. 30. The evidence presented shows that the second part of the decade gave way to some degree of convergence in monetary measures of welfare. The patterns described in this chapter, as well as Chapters 3, 4, and 8 of this report, suggest that the convergence observed in the various poverty indicators resulted from the 2008 price shock benefiting agricultural sector employees. Moreover, since the demographic effect was larger in the West relative to the East, the West, lagging relative to the East, may have enjoyed higher returns to demographic changes over the decade. 31. Analysis of temporal changes in the HOIs between 2005 and 2010 suggests some degree of convergence in access to basic facilities and services between IRs and LIRs, with much of the gain experienced by LIRs. For instance, significant improvements occurred in access to professional health services at the national-level, but more pronouncedly in LIRs. Even so, access to professional health care services was generally low for all children, irrespective of location, and the corresponding dissimilarity index was highest for children from urban IRs. The analysis also shows that between 2005 and 2010, LIRs made larger gains in the HOIs for access to sanitation and access to electricity relative to IRs. The HOI for access to education was the highest relative to all other services considered and, similar to the access to professional health services HOI, it significantly increased over the 2005-2010 period. Despite increases in access to electricity and sanitation in LIRs over the 2005-2010 period, special attention needs to be given to rural areas, where access to either service is extremely poor. Children’s human development outcomes are critical for arresting the intergenerational transmission of poverty; thus, poor children must have access to educational opportunities, health, and other basic facilities, which should not be curtailed by their circumstances. 32. Overall, the new poverty estimates suggest that, although the first half of the 2000-2010 decade was characterized by uneven poverty reduction and growth, the second half provided lagging regions with an opportunity to catch up. While this is positive news at the national-level, we emphasize that poverty in rural areas, irrespective of the level of integration, continues to be relatively more pervasive and extreme, and the gap in the speed of poverty reduction between urban and rural areas has, in fact, widened over that last five years. Urban areas remain relatively more unequal, and rural areas remain relatively poorer (see Chapter 1). In other words, even as the dramatic “East/West divide” has been substantially mitigated, the insidious “urban/rural” divide must be underscored. 140 Supplemental Content 141 Appendix 1: Inflation Annex - Temporal and Spatial Price Adjustments 1. Prior to this poverty assessment, the Figure IA-1: Poverty Headcount - Alternative PLs Bangladesh Bureau of Statistics (BBS) organized a committee of experts to produce the official 2010 58.59 58 poverty estimates. Published by BBS (2011), the 56.11 2010 poverty estimates are based on a Cost of 53 Basic Needs (CBN) methodology and are derived 48 48.9 50.47 by adjusting existing poverty lines to reflect Poverty Headcount changes in the cost of meeting basic needs, as 43 43.25 indicated by the HIES 2010 data. More 38 40 specifically, the adjustments to poverty lines for Official PL 2010 were obtained by: (i) updating 2005 food 33 WB-PovcalNet Estimates (CPI) 31.5 poverty lines with food inflation rates calculated 28 WB-PovcalNet Estimates (BNPI) from unit values of HIES 2005 and HIES 2010 2005- Pop. Weighted PL (CPI) 24.59 data; and (ii) re-estimating the non-food poverty 23 2000 2005 2010 line using HIES 2010 data to adjust for the non- Source: HIES 2000, 2005, and 2010. food allowance.107 2. While poverty lines are not the only element necessary to quantify poverty, they are fundamental to the generation of a country’s poverty profile. According to Ravallion (1998), “A poverty line helps focus the attention of governments and civil society on the living conditions of the poor”. Since they are often used to measure the real cost of basic needs over time, poverty lines also provide indirect information about price changes (in addition to measuring and tracking poverty over time). Ideally, poverty lines should reflect the cost of meeting some fixed measure of basic needs (or, even more abstractly, reflect the minimum monetary cost of obtaining some fixed level of utility). In other words, the percentage increase in poverty lines over time can serve as a useful measure of inflation experienced by the poor. 3. In most countries, the consumer price index (CPI) is used to measure inflation, usually at the national-level for the typical consumer. Currently, in Bangladesh, a substantial divergence exists between inflation, as measured by the CPI, and the increasing cost of minimum needs, as measured by changes in national poverty lines over time, referred hitherto as the Basic Need Price Index, or BNPI (see Tables IA- 1.1-IA-1.3). Figure IA-1 presents poverty headcount trends under alternative poverty lines. The alternatives displayed include: the official poverty headcount, which uses the national poverty line (solid purple line); the PovcalNet poverty trend, based on the $1.25 poverty line, that adjusts the 2005 PPP exchange rate using the CPI (dashed orange line); the PovcalNet poverty trend, based on the $1.25 poverty line, that adjusts the 2005 PPP exchange rate using inflation as implied by the BNPI (solid green line); and the poverty headcount implied by the 2005 population-weighted (or national) poverty line projected for 2000 and 2010 using the CPI (dashed blue line). This figure highlights the dramatic difference that exists between inflation measured by the CPI and the BNPI. 4. The large divergence between inflation measured by the CPI and inflation measured by the BNPI is not unique to Bangladesh. Deaton (2008) finds that the Consumer Price Index for Agricultural Labourers (CPIAL, the official national price index for rural India) understates the rate of food price inflation and the nominal poverty line over the five-year period from 1999-2000 to 2004-2005. Using data from India’s household expenditure surveys, he identifies two root causes for his finding: (1) outdated 107 See Chapter 5 and Appendix I in the BBS report (2011) for a detailed description of the methodology used to update poverty lines. 142 CPIAL weights and (2) weights that are too heavy assigned to food in a period when food prices fell relative to non-food prices. 5. While consumption patterns for the poor likely differ from typical, non-poor consumption patterns, the difference in consumption is generally less pronounced in a country with high poverty rates. On the one hand, the CPI in Bangladesh is constructed using a standard Laspeyres-type “based-weighted” price index. Plainly, the CPI informs us about how the cost of the base-year bundle changed over time, but it does not take into account potential substitution effects (i.e. consumer substitution in response to price changes). On the other hand, poverty line (PL)-based inflation allows the underlying basic needs consumer bundle to change over time while the consumer maintains a fixed level of utility. The latter is conceptually closer to the cost of living index (COLI), which reflects changes in all factors beyond the direct consumption of goods and services that affect consumer welfare (Diewert et al. 2009). 6. Even as a clear distinction exists between these two inflation measures, the difference is unable to fully explain the observed divergence. A large divergence between the CPI and the rate of change in the poverty lines is problematic for Bangladesh (a relatively high poverty country) because it suggests that either the CPI (the official inflation measure of Bangladesh), as it is currently estimated, is not a good measure of price changes (at least not for the majority) or that the real value of the poverty line has increased, thus rendering it less effective as a tool for comparing absolute poverty across time. Furthermore, this divergence has significant implications for the real value of all money-metrics (see, for example, the case of consumption expenditure presented in Table IA-1.3). 7. The inflation measure that is suitable for the purposes of the analyses undertaken in a poverty assessment report is not clear, a priori. The aim of this annex is to establish the appropriate measure for Bangladesh. First, we provide a brief recount of how inflation is measured in Bangladesh as well as the construction/re-estimation of poverty lines (PLs) over the last decade. We follow this overview by analyzing the temporal dimension of price changes, as measured by the CPI and the PLs. Then, we analyze the spatial aspect of price changes. To undertake this analysis, we use HIES data from 2000, 2005, and 2010. We conclude the annex by summarizing our main findings and conclusions. I. Measuring Inflation in Bangladesh 8. In Bangladesh, the official measure of inflation is the Consumer Price Index (CPI), which is calculated using Laspeyre’s “based-weighted” price index.108 The weights, which are shares of goods in the initial expenditure basket, are periodically updated using the Household Income and Expenditure Surveys (HIES).109 Currently, these weights correspond to commodity-wise expenditures shares from the 1995/96 HIES.110 The national CPI is a weighted-average of rural and urban CPIs, whose weights are 70.9 percent and 29.1 percent, respectively. Similarly, the general CPI in each region is the weighted-average of the food and non-food CPIs; the weights are the respective expenditure shares of food and non-food items, as estimated from the HIES. At present, the CPI cannot be disaggregated into more specific geographical areas (i.e. divisions or strata). To account for regional differences in expenditures, the urban and rural CPIs differ in the number of goods included in their respective expenditure baskets. The expenditure basket that corresponds to rural (urban) areas is composed of 215 (302) items. The CPI can be used as a measure of national/urban/rural inflation but cannot be used to adjust for spatial price 108 This sections draws heavily from the BBS (2008) report. 109 Before, the weights used in estimating the CPI corresponded to commodity-wise expenditures shares from the 1985/86, 1973/74, and 1969/70 HES (see section 2 of the BBS 2008 report). 110 Recently the CPI weights were revised using the HIES 2005. 143 differences across strata111 (i.e. the CPI is not a spatial price index, as is the case with most CPIs across the world). Year 2000112 2005 2010 Food PL Updated from 1991/92 Re-estimated (CBN) Updated from 2005 Non-food PL Updated from 1991/92 Re-estimated (CBN) Re-estimated (CBN) a. Poverty Lines in Bangladesh 9. As suggested by Ravallion (2001), Bangladesh’s PLs are periodically updated using a price index (as done for both food and non-food PLs in 1995/96 and 2000, and only for the food PL in 2010) or re- estimated using the CBN method (as done in 1991/92 and 2005).113 The expectation is that either of the two methods (price index or CBN) maintains a constant level of wellbeing in real terms and, thus, provides a good measure of poverty and price changes over time. 10. Under the first method, the food basket quantities are fixed but market prices are updated using an appropriate price index. Under the second method, a new food basket is estimated following the CBN method. Using the CBN method, calculation of PLs entails estimation of the average level of per capita expenditure at which individuals can meet basic food and non-food needs. The CBN method is implemented in three steps. In the first step, the cost of a fixed food bundle is computed. In the case of Bangladesh, this bundle consists of eleven food items that include: rice, wheat, pulses, milk, oil, meat, fresh water fish, potato, other vegetables, sugar, and fruits. The bundle provides the minimal nutritional requirements that correspond to 2,122 kcal per day per person. In the second step, two different non-food allowances for non-food consumption are computed: the lower non-food allowance (the median amount spent on non-food items by households whose total consumption is approximately equal to their food- poverty line)114 and the upper non-food allowance (the amount spent on non-food items by households whose food consumption is approximately equal to their food-PL). In the third step, the food and non- food allowances are added together. The sum of the food and upper non-food allowances constitute the upper PL.115,116 11. Both methods have concerns in terms of maintaining a constant measure of wellbeing in real terms. The first method, which maintains the same bundle of good, won’t reflect that as relative prices 111 In 2000/01 (2005 and 2010/11), Bangladesh had a total of 14 (16) strata. 112 The 2005 PLs were also back-casted to 2000. Please see Tables IA-1.4 and IA-1.5. 113 For more details related to the construction and updating of the poverty lines for Bangladesh, please see: http://www- wds.worldbank.org/external/default/WDSContentServer/WDSP/IB/2003/07/26/000094946_03050804024314/ Rendered/PDF/multi0page.PDF, pp. 92-95 for the years 1991/92, 1995/96, and 2000; http://www- wds.worldbank.org/external/default/WDSContentServer/WDSP/IB/2008/12/10/000333037_20081210001004/ Rendered/PDF/443210ESW0P09910Box334107801PUBLIC1.PDF, pg. 111, Box A-1.1 for the year 2005; and the Preliminary Report on Household Income and Expenditure Survey – 2010 BBS, pp. 99-104 for the years 2010/11. 112 The rationale behind this calculation is that the non-food budgets of these households are set to just afford the bare essentials. 113 In 2000/01 (2005 and 2010/11), a total of 14 (16) PLs existed, each corresponding to one of Bangladesh’s 14 (16) strata. 116 The “upper PL” (“lower PL”) is the reference level of per capita consumption expenditures used to estimate the national poverty (extreme poverty) headcount, also known as the moderate poverty (severe poverty) rate. 144 change, people will frequently substitute relatively more expensive good and the relatively cheaper ones. A constant bundle of goods, when relative price are changing, is unlikely to maintain the same utility level. But, a similar critique can be made of the BNPI. When basic needs are re-estimated, the bundle of goods can completely change. If everyone has become better off over time, the average cost per calorie (based on the ‘relatively’ poor) is likely to increase even if prices of goods are unchanged because the bundle of goods being consumed reflects improved wellbeing. In short, the changing bundle may now be reflecting a higher-level of wellbeing. 12. To address these concerns about making appropriate poverty comparisons over time, Lanjouw and Lanjouw (2001) propose using the CBN method. The authors show that re-estimation of PLs based on the CBN method can produce comparable poverty lines in real terms when the following set of assumptions hold: (1) the consumption measures are monotonically increasing in total expenditures (akin to Engel’s Law); (2) relative prices, which determine consumption patterns, are stable across time and the groups under comparison; and (3) no measurement error exists in the expenditure data. The authors demonstrate that the CBN method allows for measuring poverty and price changes over time without having to rely on the existence of a price index. Nevertheless, like the first method, this method also assumes that relative prices and tastes remain unchanged. 13. Bangladesh’s PLs by year and stratum are presented in Tables IA-1.4-1.5. In the next paragraphs, we provide a brief recount of the construction/re-estimation of Bangladesh’s PLs over the last decade. b. Bangladesh Poverty Assessment 2000 14. PLs were first estimated in 1991/92 (the base year) and subsequently updated in 1995/96 and 2000 to account for price changes (see Table IA-1.4). In 2001, different PLs were estimated for Bangladesh’s 14 different geographical areas (nine urban and five rural areas). To update the base-year PLs to 2000, price indices were derived by combining price information obtained from HIES (corresponding to food and non-food items that account for approximately two-thirds of total household expenditure) and the non-food CPI; the price indices are reported in Table IA-1.6.117 c. Bangladesh Poverty Assessment 2005 15 In 2005, PLs were re-based using the HIES 2005 data and following the same CBN method described above. This time, sixteen stratum-specific PLs were estimated for Bangladesh (ten urban and six rural areas). To ensure comparability across years, the estimated PLs were back-casted to 2000 using price indices that combine price information available in the HIES (prices corresponding to food items that account for more than one-half of total household expenditure) and the non-food CPI (see Table IA- 1.5). d. Final Report HIES 2010 16. After examining the data and comparing alternative methods, the technical committee for poverty estimation, which consisted of researchers and experts on poverty measurement in Bangladesh, recommended an approach which uses parts of both methods described above (i.e. updating and re- estimation) to update the 2005 PLs. In particular, while food PLs for 2010/11 were updated from 2005 using stratum-specific, Törnqvist price indices, the upper and lower non-food PLs were re-estimated using the CBN approach (the same procedure used to estimate 2005 non-food PLs) with the HIES 2010/11 data. The primary motivation for this choice is the concern that the non-food CPI may not faithfully reflect inflation faced by the poor. Although the technical committee acknowledges that the CPI 117 For more details on the poverty lines update for the year 2000, please refer to the Technical Appendix of the 2000 PA report (World Bank 2002, pg. 92-96). 145 is widely used to convert nominal values into real values in national accounts, the committee maintains that the CPI is not a good inflation measure for poverty analysis. 17. As in 2005, sixteen stratum-specific PLs were estimated for Bangladesh in 2010 (ten urban and six rural). To update 2005 (base-year) food PLs to 2010, price indices used price information obtained from HIES (which contains prices corresponding to food items that account for more than 60 percent of total household expenditure). The non-food allowance was re-calculated following steps 2 and 3 of the CBN method described above and used the HIES 2010 data. II. Temporal Price Changes: Survey-Based Evidence 18. In this section, we turn to the HIES data to find evidence to support the use of either the CPI or BNPI for our analysis. a. Demographic Change Over Time “An absolute approach in the space of capabilities translates into a relative approach in the space of commodities” (Sen 1983, p. 168) 19. When PLs are estimated following the CBN method, at least two forces are likely to change the real values of PLs: economic growth and demographic changes (Ravallion 1998). In other words, as a country becomes wealthier and as fertility declines, a real increase in the cost of the minimum needs required to fully participate in society is likely. In particular, as Ravallion (1998) explains, when average income increases and fertility falls, consumption shifts away from food (food represents a smaller share of the consumption bundle, so its relative price rises). Then, under the assumption that food and non-food are uncompensated substitutes, a decrease in the relative price of non-food leads to an increase in the PL. In other words, due to the income effect, individuals will adjust their consumption bundles in order to maintain an equivalent level of utility, an adjustment which results in higher, real-valued PLs. 20. This section explores whether or not the effect of Bangladesh’s demographic changes over the 2005-2010 period is significant. If significant, this effect can help to parse out one of the causes of the divergence between the CPI and PLs. To make this assessment, we investigate whether the demographic changes that took place help to predict the observed changes in both poverty headcount rates and per- capita consumption for the period between 2005 and 2010. In particular, we pool the 2005 and 2010 HIES data at the household-level and generate eight age-gender cohort variables, measuring the number of household members falling into each cohort.118 Then, using this pooled sample119, we regress the poverty headcount rate (as implied by the official upper PLs for 2005 and 2010) and per-capita consumption on the eight age-gender cohorts, a time trend (which equals one if the survey year is 2010), and 15 dummies for each stratum (Barisal is the base group). 21. The estimated coefficients from these regressions are presented in Table IA-2.1. First, we obtain separate means for each age-gender cohort for survey years both, 2005 and 2010. Using these year- and cohort-specific means, we compute the average change in household composition by age-gender cohort for the 2005-2010 period by subtracting each 2005 age-gender cohort mean from the 2010 age-gender cohort mean. Then, we multiply the estimated means by the average change in household composition 118 The relevant age-gender-cohorts are: Men (+55), Women (+55), Men (26-55), Women (26-55), Men (18-25), Women (18-25), Boys (10-17), Girls (10-17), Boys (6-9), Girls (6-9), Boys (0-5), and Girls (0-5). 119 We chose to pool the samples to increase efficiency. To deal with the sampling weights, household weights are scaled by a factor of ½. 146 calculated for each age-gender cohort. Summing up these products over each age-gender cohort suggests that approximately one-third of the overall poverty reduction is associated with changes in household composition. In particular, we observe that, while the number of adults and children declined over time, the decline in the number of adults increased the likelihood of being poor, whereas the reduced number of children decreased the likelihood of being poor (not surprisingly, since adults are more likely than children to contribute to household income). Furthermore, the estimates hint at a decline in fertility as the most important demographic change contributing to the reduced likelihood of being poor (the contribution of demographic changes to household composition increases to 40% if considering only the decline in the number of children per household). Overall, the results suggest that households are smaller in size, primarily due to a reduction in fertility, in 2010 relative to 2005. 22. Next, following the same procedure described in the previous paragraph, we predict changes in per capita consumption. The results from this exercise (Table IA-2.2) show that the average change in household composition led to an increase in per capita consumption of approximately Tk 72.1. If we only consider changes in the number of children, per capita consumption increases to Tk 79.4. Re-estimating the poverty headcount rate after reducing the poverty line by Tk 72 (80), we observe that the official poverty headcount rate is reduced by 4.31 (4.76) percentage points, which translates into a 14 (15) percent reduction. These numbers are reported in the bottom panel of Table IA-2.2. For comparison, we also reproduce the official poverty headcount and the poverty headcount estimates under a PL constructed by combining the food component of each year with the non-food component of 2005 and adjusted using a survey-based composite price index. The latter estimates are fairly close to the estimates obtained after adjusting official PLs for changes in household composition. 23. The time trend explains about 93% of the increase in nominal per-capita consumption. As might be expected, this is equivalent to the increase in inflation implied by the BNPI. Because the income effect is expected to be relatively small for the food component, the evidence presented in this section is consistent with the hypothesis that as fertility declines, consumption shifts away from food which causes the non-food share of consumption to be relatively high and, consequently, the real value of the PL to be higher.120 Hence, the evidence suggests that the decline in fertility resulted in a real increase in per-capita consumption. The resulting reallocation of resources from food to non-food consumption partially explains the relatively larger divergence between the non-food CPI and BNPI compared to the divergence between the food CPI and BNPI. 24. These results provide support for use of the BNPI by suggesting a theoretically-based and empirically-backed avenue for the CPI to be biased downwards. To recapitulate, the demographic changes experienced by Bangladesh over the last five years, and the attendant change in relative prices, have downward-biased the CPI relative to the BNPI. This is not surprising since the CPI, unlike the BNPI, was never intended for use as a cost-of-living index. b. Price Changes Over Time 25. The HIES data allow for partial testing of whether or not price changes, as observed in the data, match inflation measured using the CPI. In particular, the 2005 and 2010/11 HIES data contain enough information that allows for the construction of an aggregate price index consisting of food, clothing, footwear, and rent (categories which account for about 70 percent of total consumption). The index of rent prices is based on the hedonic model of housing prices, as used in constructing the measure of total consumption. In this model, the available information on rental prices and housing characteristics is used 120 For a more detailed description of the mechanisms behind this statement, please refer to Ravallion (2001). 147 to predict the cost of housing for the whole sample for each survey year. 121 First, the log of annual rental payments in 2005 is regressed on household characteristics in 2005, and the analogous regression is run for the 2010/11 data. Then, using the estimated parameters from 2010/11 and housing characteristics in 2005, one can estimate the rental value of the 2005 characteristics based on 2010 prices (parameters). Comparison of the rental cost of the 2005 characteristics in 2005 with the cost of these same characteristics in 2010 provides an estimate for rental inflation. The results are presented in Table IA-3. The implied national, rural, and urban inflation rates for housing costs are 46 percent, 43 percent, and 54 percent, respectively; the analogous rates using the CPI are 33 percent, 38 percent, and 20 percent, respectively. Assuming the data collected for the HIES provides a better representation of the overall population than the CPI data, the CPI, then, grossly underestimates the rental inflation rate in urban areas, though it provides a decent approximation of the rental inflation rate in rural areas. 25. The HIES also allows for the construction of a footwear and garment price index. In particular, we calculate the per-unit cost of two footwear items and thirteen ready-made garments, as well as the two categories’ respective budget shares, for each household.122 Then, stratum-level mean budget shares for the base year (2005) and stratum-level prices for each of the surveys (2005 and 2010/11) are computed. This information is used to compute a Laspeyres-type price index for each of the two item groups and for each year. The weighted sum of these indices (their weights correspond to their respective stratum-level expenditure shares) yields a single price index (column 1 of Table IA-3). The national, rural, and urban inflation rates implied by this index are 53 percent, 71 percent, and 47 percent, respectively; the analogous numbers using the CPI are 28 percent, 25 percent, and 34 percent, respectively. Compared to the price index produced using the HIES data, the CPI underestimates footwear and clothing inflation rates for both urban and rural areas. To obtain a single, non-food price index, we compute the weighted sum of the housing price index and the footwear and clothing price index (their weights correspond to their respective CPI weights). The national, rural, and urban inflation rates implied by this non-food price index are 51 percent, 46 percent, and 67 percent, respectively; the analogous numbers using the CPI are 33 percent, 33 percent, and 28 percent, respectively. 27. Next, we construct a food price index using thirteen different food categories, with a modal/representative food item selected from each category. The median price of this item is then measured for each stratum. Each item’s price is assigned a weight based on the national food share of that category. The national, rural, and urban inflation rates implied by this index are 78 percent, 77 percent, and 81 percent, respectively. The analogous rates using the CPI are 54 percent, 54 percent, and 58 percent, respectively. Table IA-3: Temporal Price Changes 2005-2010 Footwear Overall Overall & Housing Non-food Other Food (excluding (including Garment “Other”) “Other”) National 53.43% 45.81% 51.37% 0% 78.46% 71.10% 58.75% Rural 70.90% 42.78% 45.79% 0% 77.39% 69.32% 58.60% Urban 47.20% 54.29% 67.01% 0% 81.46% 76.08% 59.16% 121 In particular, the log of rental payments is regressed on the log of the number of rooms, log of land size, indicators for the existence of a dining area, a separate kitchen, safe water supply, electricity, a phone line, for whether the wall material is brick, as well as stratum-level fixed effects. 122 The selection criteria is to include the item as long as at least ten households in each stratum have valid/non- missing entries for both the value paid for the item and the quantity purchased. 148 28. Combining the three indexes (food, footwear and garment, and housing) into a single index, with each index weighted according to their CPI weights, we find that the implied national, rural, and urban inflation rates are 71 percent, 69 percent, and 76 percent, respectively. The analogous numbers using the CPI are 46 percent, 47 percent, and 44 percent, respectively. 29. Survey-based evidence, using the HIES, suggests that the CPI underestimates inflation as faced by the poor in Bangladesh. At the national-level, the CPI-implied inflation rate is 25 percentage points (or 35 percent) lower than the implied survey-based inflation rate; while the PL-implied inflation rate is 15 percentage points higher than the survey-based rate. If we take into account the seven percent increase in consumption, attributable to changes in household composition, for each of the indices, the gap between the implied survey-based and BNPI-based inflation rates decreases to only seven percentage points (71 percent and 78 percent, respectively). Based on this evidence, we conclude that the BNPI (or PL-implied inflation rate), relative to the CPI, is a more accurate measure of inflation for Bangladesh. III. Spatial Price Changes 30. Using unit prices from the HIES survey, we construct spatial price indices for clothing and footwear, housing, and food items123 to contrast with the observed spatial dispersion (as implied by changes in PLs over the 2005-2010 period). We don’t suggest that the HIES nonfood price index is the correct measure, but we do believe it contains some signal on the spatial distribution of nonfood prices. And this distribution can help us assess the nonfood poverty lines. The constructed indices are presented in Table IA-4.1 and Figure IA-2, with rural Dhaka serving as the reference group. The survey-based nonfood price indices follow a similar pattern (and exhibit significant variation), but they do not fully coincide with those implied by the stratum-level, PL-based indices. In particular, when considering just the footwear and clothing indices, we observe that the spatial price dispersion implied by the survey- based index is lower relative to the PL-based index. On the other hand, as we take into account the dispersion due to differences in housing prices (i.e. non-food prices), the implied survey-based inflation rates are higher than the PL-based rates (Figure IA-3). 31. Ravallion (1998) suggests the use of the methodology proposed by Lanjouw and Lanjouw (2001) Figure IA-2: Spatial Price Indices Source: HIES 2010. 123 The items included in the construction of these indices are the same as those included in the construction of the temporal price indices. 149 in the presence of measurement error in non-food consumption and/or price changes. Lanjouw and Lanjouw (2001) demonstrate that, under certain assumptions (in particular, in the absence of mismeasurement in the data), poverty comparisons can be made regardless of the comprehensiveness of the data. In other words, measuring poverty using total expenditures should yield the same estimates as using a less comprehensive component of expenditures, such as the food component. 32. Table IA-4.2 explores the Lanjouw and Lanjouw (2001) methodology at the national- and stratum-levels in Bangladesh. At the national-level, poverty estimates using total expenditures are similar to estimates using food expenditures. However, at the stratum-level, the differences are larger and systematic. In particular, according to official numbers and relative to the food expenditure approach, poverty is underestimated in Barisal Rural, Khulna Rural, and Rajshahi Rural and Municipal (all of which are regions in the West). Poverty is overestimated for Barisal Municipal, Chittagong SMA, Dhaka Municipal, Khulna SMA, Rajshahi SMA, and Sylhet Municipal (all of which are non-rural areas). These systematic differences and the evidence presented earlier in the draft do not provide solid evidence in support of the BNPI (nor do they provide evidence for use of CPI). 33. With the exception of Sylhet, the magnitude of the gaps between the spatial price differences (implied by the 2010 PLs and the survey-based price index) for footwear and ready-made garments (Figure IA-3) are associated with the differences in poverty headcounts (implied by food PLs versus overall PLs) (Table IA-4.2). In other words, whenever the food-based poverty headcount is lower than the upper PL poverty headcount, the spatial price differences implied by the PLs exceeds the price dispersion implied by the footwear and ready-made garment-based index. Figure IA-3: Non-food PL vs. Survey-Based Non-food Indices Non-Food Indices 2.600 2.100 1.600 1.100 0.600 Barisal Barisal Chittagong Chittagong Chittagong Dhaka Dhaka Dhaka Khulna Khulna Khulna Rajshahi Rajshahi Rajshahi Sylhet Sylhet (Rural) (Muni.) (Rural) (Muni.) (SMA) (Rural) (Muni.) (SMA) (Rural) (Muni.) (SMA) (Rural) (Muni.) (SMA) (Rural) (Muni.) HIES-Non-Food PL-Non-Food (Base: Rural Dhaka) HIES-Footwear & Garment (Base: Rural Dhaka) Source: HIES 2010. Conclusion 34. The purpose of this annex is to determine the best approach for measuring price changes in Bangladesh. First, we provide a brief recount of how inflation is measured and of how poverty lines were constructed/re-estimated in Bangladesh. Then, we analyze both temporal and spatial aspects of inflation, as measured by the CPI and the PLs (or BNPI). 35. We find some evidence that the poverty line in 2010, relative to 2005, seems to be somewhat greater in real terms. A plausible explanation for the increase in real value of the official PLs is provided by our analysis of the relationship between the incidence of poverty and both per-capita consumption and changes in household demographics. The analysis shows that decreases in the number of household members (in particular, young children) are associated with significant increases in per-capita 150 consumption. This result suggests that in the 2005-2010 period, as incomes rose and the number of household members declined, the relative cost of food increased, causing the value of PLs to increase in real terms. 36. HIES-based temporal price indices suggest that the CPI may underestimate inflation as faced by the poor in Bangladesh (essentially an implicit assumption of the expert committee on poverty estimates). Furthermore, the indexes constructed using HIES data suggest that inflation, as experienced by the poor, is closer to what is implied by changes in the PLs. A further piece of evidence in favor of using PLs rather than the CPI to measure inflation faced by consumers is provided by a poverty headcount robustness check. In particular, following the methodology proposed by Lanjouw and Lanjouw (2001), we corroborate that the poverty headcount at the national-level is robust to consumption aggregates (i.e. both the food PL and the upper PL virtually imply the same poverty rate). 37. Evidence for spatial price indices constructed using the HIES is mixed. On the one hand, whenever the food-based poverty headcount is lower than the upper PL-based poverty headcount, the spatial price differences implied by the PLs exceeds the price dispersion implied by the footwear and ready-made garment-based index. When we take into account price dispersion due to differences in housing prices, we obtain the opposite result. Again following the methodology suggested by Lanjouw and Lanjouw (2001), we are unable to corroborate that poverty headcounts at the stratum-level are robust to consumption aggregates (i.e. the food PL and the upper PL imply different poverty rates). In particular, we find that the gaps between the spatial prices (implied by the 2010 PLs and the survey-based footwear and clothing price index) are associated with the differences in poverty headcounts implied by the food PLs versus overall PLs. Unlike the BNPI (which, by definition, is closer to a cost of living index), the survey-based price index intends to capture price differences across regions, regardless of whether or not utility levels are held constant. Ultimately, since both of the approaches for measuring spatial price differences has potential concerns, we argue that there is much value in using the spatial price index as implied by the poverty lines (and agreed to by the Government of Bangladesh). 38. Hence, depending on whether we seek to measure absolute poverty, inequality, inflation faced by consumers, or aggregate-level inflation, we propose: (1) the use of the national PL-based price index (or BNPI) to make temporal comparisons; (2) the use of the national PL as the baseline for adjusting consumption when making cross-sectional poverty comparisons across regions; (3) the use of the survey- based spatial price indices when making cross-sectional inequality comparisons across regions; and until a better measure of aggregate-level inflation is available, (4) the use of the CPI for adjusting nominal macroeconomic aggregates (in part, to reduce mis-communications when making comparisons from this report on macroeconomic indicators with existing published statistics). 39. Given the findings from this report as well as concerns raised by other researchers124, we also recommend that the construction of the CPI basket and, in particular, the non-food component should be revised such that a more reliable/appropriate measure of consumer prices can be formulated for Bangladesh. Ideally, this price measure should be periodically validated as new rounds of HIES data are made available. 124 See, for example, the article published by the Policy Research Institute http://www.pri- bd.org/index.php?option=com_content&view=article&id=208:inflation-and-statistical- paradox&catid=47:bangladesh-economy&Itemid=59 151 Table IA-1.1: Alternative Measures of National Inflation National Index (Base 2000) 2000 2005 2010 General-CPI (Base Year: 2000/01=100) 100 126 184 General-UPL (Pop. Weighted) 100 126 234 General-UPL (Rural Dhaka) 100 129 230 Food-CPI (Base Year: 2000/01=100) 100 127 196 Food-UPL (Pop. Weighted) 100 125 224 Food-UPL (Rural Dhaka) 100 129 219 Non-Food-CPI (Base Year: 2000/01=100) 100 126 167 Non-Food-UPL (Pop. Weighted) 100 127 252 Non-Food-UPL (Rural Dhaka) 100 130 252 CPI = Consumer Price Index. U(L)PL = Upper (Lower) Poverty Line. Table IA-1.2: Alternative Measures of Inflation by Area Urban Rural Index (Base 2000) 2000 2005 2010 2000 2005 2010 General-CPI (Base Year: 2000/01=100) 100 125 180 100 124 182 General-UPL (Pop. Weighted) 100 118 221 100 127 232 General-UPL (Rural Dhaka) 100 129 230 100 129 230 Food-CPI (Base Year: 2000/01=100) 100 127 201 100 125 192 Food-UPL (Pop. Weighted) 100 118 213 100 127 224 Food-UPL (Rural Dhaka) 100 129 219 100 129 219 Non-Food-CPI (Base Year: 2000/01=100) 100 124 159 100 123 164 Non-Food-UPL (Pop. Weighted) 100 118 235 100 126 247 Non-Food-UPL (Rural Dhaka) 100 130 252 100 130 252 CPI = Consumer Price Index. U(L)PL = Upper (Lower) Poverty Line. Table IA-1.3: Nominal and Per-capita Consumption Using Alternative Measures of Inflation %00- %05- %00- Consumption 2000 2005 2010 05 10 10 Nominal per-capita 877 1231 2447 40 99 179 Real per-capita (CPI) 877 990 1350 13 36 54 Real per-capita (UPL-Rural Dhaka) 877 951 1064 8 12 21 Real per-capita (UPL-Stratum) 877 991 1067 13 8 22 Real per-capita (UPL-Pop. weighted) 877 978 1047 12 7 19 CPI = Consumer Price Index. UPL = Upper (Lower) Poverty Line. 152 Table IA-1.4: Poverty Lines Region LPL UPL FOOD NON-FOOD LPL UPL FOOD NON-FOOD 1995/96 2000 Rural Barisal Pathuakali 413 467 360 107 494 558 428 128 Rural Noakhali Chittagong 438 541 395 146 522 645 470 174 Other Urban Chittagong 517 609 408 201 619 730 490 241 SMA Chittagong 523 722 448 274 627 867 537 329 Rural Dhaka 425 512 379 133 492 593 440 154 Other Urban Dhaka 399 482 328 154 480 580 393 185 SMA Dhaka 480 660 389 271 574 791 467 325 Rural Khulna Jessore Kushtia 420 497 363 134 499 592 432 160 Urban Khulna 482 635 425 210 552 727 485 239 Rural Rajshahi Pabna 459 540 394 146 535 630 461 171 Urban Rajshahi 446 582 390 192 496 647 433 213 Rural Sylhet Comnilla 432 558 430 128 499 644 494 148 Rural Faridpur Tangail Jamalpur 432 472 373 99 484 529 418 111 Rural BograRa Dajpur 426 487 365 122 468 535 402 134 2005 2010 Barisal (Rural) 753 926 583 343 1284 1485 982 503 Barisal (Muni.) 800 951 599 352 1419 1963 1100 863 Chittagong (Rural) 753 891 568 323 1404 1687 1023 664 Chittagong (Muni.) 749 963 561 402 1495 1825 1064 762 Chittagong (SMA) 766 1171 578 593 1479 1876 1047 829 Dhaka (Rural) 728 842 565 277 1276 1497 958 538 Dhaka (Muni.) 749 890 579 311 1314 1793 1018 775 Dhaka (SMA) 806 1018 601 417 1406 2038 1089 948 Khulna (Rural) 652 743 510 233 1192 1435 884 551 Khulna (Muni.) 670 825 517 308 1262 1680 932 748 Khulna (SMA) 706 938 552 386 1348 1639 970 669 Rajshahi (Rural) 656 766 509 257 1236 1487 957 529 Rajshahi (Muni.) 696 857 530 327 1312 1585 987 598 Rajshahi (SMA) 722 856 523 333 1223 1556 931 625 Sylhet (Rural) 697 822 549 273 1240 1311 953 358 Sylhet (Muni.) 806 1020 549 471 1286 1558 992 566 153 Table IA-1.5: Harmonized Poverty Lines 2000, 2005, and 2010 2000 2005 2010 Region LPL UPL LPL UPL LPL UPL Barisal (Rural) 580 714 753 926 1284 1485 Barisal (Muni.) 643 764 800 951 1419 1963 Chittagong (Rural) 619 733 753 891 1404 1687 Chittagong (Muni.) 643 827 749 963 1495 1825 Chittagong (SMA) 639 978 766 1171 1479 1876 Dhaka (Rural) 563 651 728 842 1276 1497 Dhaka (Muni.) 625 742 749 890 1314 1793 Dhaka (SMA) 678 855 806 1018 1406 2038 Khulna (Rural) 511 582 652 743 1192 1435 Khulna (Muni.) 561 690 670 825 1262 1680 Khulna (SMA) 582 773 706 938 1348 1639 Rajshahi (Rural) 511 598 656 766 1236 1487 Rajshahi (Muni.) 575 707 696 857 1312 1585 Rajshahi (SMA) 576 682 722 856 1223 1556 Sylhet (Rural) 560 661 697 822 1240 1311 Sylhet (Muni.) 666 843 806 1020 1286 1558 154 Table IA-1.6: Price Indices Region Food Covered Non- Composite Food Covered Non- Composite HIES budget Food Price HIES budget Food Price Index sh. CPI Index Index sh. CPI Index 1991/92 - 1995/96 1995/96 - 2000 SMA Dhaka 1.20 59% 1.20 1.20 1.10 53% 1.16 1.13 Other Urban Dhaka 1.20 68% 1.20 1.20 1.03 60% 1.16 1.08 Rural Dhaka 1.12 74% 1.26 1.16 1.07 72% 1.20 1.11 Rural Faridpur Tangail 1.08 79% 1.26 1.12 1.07 74% 1.20 1.12 SMA Chittagong Jamalpur 1.20 62% 1.20 1.20 1.09 59% 1.16 1.12 Other Urban 1.20 67% 1.20 1.20 1.09 60% 1.16 1.12 Rural Sylhet Comnilla Chittagong 1.12 77% 1.26 1.15 1.11 71% 1.20 1.15 Rural Noakhali 1.17 73% 1.26 1.19 1.06 67% 1.20 1.11 Urban Khulna Chittagong 1.12 67% 1.20 1.14 1.06 62% 1.16 1.10 Rural Barishal 1.17 77% 1.26 1.19 1.05 70% 1.20 1.10 Rural Khulna Jessore Pathuakali 1.16 73% 1.26 1.19 0.98 69% 1.20 1.05 Urban KushtiaRajshahi 1.07 67% 1.20 1.11 1.08 61% 1.16 1.12 Rural Rajshahi Pabna 1.13 73% 1.26 1.17 1.04 71% 1.20 1.10 Rural BograRa Dajpur 1.04 75% 1.25 1.10 1.01 70% 1.20 1.09 2000 – 2005 2005 - 2010 Barisal (Rural) 1.31 48% 1.28 1.30 1.69 66% 1.33 1.55 Barisal (Muni.) 1.28 51% 1.24 1.24 1.84 56% 1.29 1.59 Chittagong (Rural) 1.17 52% 1.28 1.22 1.80 61% 1.33 1.60 Chittagong (Muni.) 1.13 50% 1.24 1.16 1.89 58% 1.29 1.60 Chittagong (SMA) 1.19 41% 1.24 1.20 1.81 56% 1.29 1.53 Dhaka (Rural) 1.31 52% 1.28 1.29 1.70 64% 1.33 1.55 Dhaka (Muni.) 1.20 54% 1.24 1.20 1.76 57% 1.29 1.54 Dhaka (SMA) 1.18 50% 1.24 1.19 1.81 53% 1.29 1.55 Khulna (Rural) 1.28 54% 1.28 1.28 1.73 62% 1.33 1.57 Khulna (Muni.) 1.19 52% 1.24 1.20 1.80 55% 1.29 1.56 Khulna (SMA) 1.22 49% 1.24 1.21 1.76 59% 1.29 1.55 Rajshahi (Rural) 1.29 52% 1.28 1.28 1.88 64% 1.33 1.66 Rajshahi (Muni.) 1.22 51% 1.24 1.21 1.86 62% 1.29 1.60 Rajshahi (SMA) 1.30 49% 1.24 1.25 1.78 60% 1.29 1.54 Sylhet (Rural) 1.23 54% 1.28 1.24 1.74 73% 1.33 1.59 Sylhet (Muni.) 1.22 45% 1.24 1.21 1.81 64% 1.29 1.56 155 Table IA-2.1: Predicted Change in Poverty Due to Changes in Household (HH) Composition HH-Comp. Total Trend Children (All) -0.0849 39.7% (32.99%) 64.75% Change in poverty Return- Change in Poverty Coefficient HH-Comp. Headcount Constant 0.4376 Time Trend -0.0549 1.0000 -0.0549 Men (+55) 0.0006 -0.0154 -0.0382 Women (+55) -0.0002 0.0055 -0.0371 Men (26-55) 0.0034 -0.0468 -0.0735 Women (26-55) 0.0012 -0.0155 -0.0793 Men (18-25) 0.0005 -0.0473 -0.0116 Women (18-25) 0.0001 -0.0051 -0.0204 Boys (10-17) -0.0026 -0.0817 0.0318 Girls (10-17) -0.0027 -0.0735 0.0370 Boys (6-9) -0.0029 -0.0223 0.1279 Girls (6-9) -0.0020 -0.0184 0.1095 Boys (0-5) -0.0103 -0.0685 0.1497 Girls (0-5) -0.0132 -0.0892 0.1485 Change in HH Comp. Abs. change Out of total Resulting change Out of total change in poverty change Total change -0.478 100% -2.80% -100% Change in male members -0.282 59% -1.11% -40% Change in female members -0.196 41% -1.69% -60% Change in male adults -0.110 23% 0.46% 16% Change in female adults -0.015 3% 0.11% 4% Change in male children -0.173 36% -1.57% -56% Change in female children -0.181 38% -1.80% -64% 156 Table IA-2.2: Predicted Change in Per-Capita Consumption Due to Changes in Household (HH) Composition HH-Comp. Total Trend Children (All) Change in consumption 1217 6.53% (5.93%) 92.51% Return- Change in Per-capita Coefficient HH-Comp. Consumption Constant 1165.83 1165.83 Time Trend 1125.60 1.0000 1125.60 Men (+55) -2.51 -0.0154 162.59 Women (+55) 0.62 0.0055 111.74 Men (26-55) -5.03 -0.0468 107.58 Women (26-55) -3.33 -0.0155 215.33 Men (18-25) 2.99 -0.0473 -63.25 Women (18-25) -0.03 -0.0051 6.13 Boys (10-17) 11.97 -0.0817 -146.56 Girls (10-17) 5.86 -0.0735 -79.67 Boys (6-9) 6.28 -0.0223 -281.20 Girls (6-9) 5.20 -0.0184 -282.42 Boys (0-5) 21.28 -0.0685 -310.81 Girls (0-5) 28.79 -0.0892 -322.60 Change in HH Comp. Abs. change Out of total Resulting change Out of total change in poverty change Total change -0.478 100% 72.10 100% Change in male members -0.282 59% 34.99 49% Change in female members -0.196 41% 37.11 51% Change in male adults -0.110 23% -4.55 -6% Change in female adults -0.015 3% -2.75 -4% Change in male children -0.173 36% 39.54 55% Change in female children -0.181 38% 39.85 55% Estimated poverty headcount Year Actual 2005-NFPL-PC1 Consumption Consumption + 79.4 Tk2 72.1 Tk3 2000 48.85 48.88 48.85 48.85 2005 39.99 40.00 40.00 40.00 2010 31.51 25.24 26.74 27.19 1 This constructed poverty line combines the food component of each year with the non-food component of 2005, adjusted using the survey-based price index. 2 This is the estimated average increase in consumption due to changes in the number of children (less than 18 years of age). 3 This is the estimated average increase in consumption due to changes in household demographics (includes children and adults). 157 Table IA-4.1: Spatial Price Indices HIES Spatial Price Indices – 2010 (Base: Rural Dhaka) Others set to mimic Bangladesh PL-Based Spatial Others set to 1 footwear and Price Indices – 2010 (Base: (i.e. no change) housing Rural Dhaka) Housing Housing Footwear (Base: Overall (3 Overall (3 PL-Non- Stratum & Garment (Base: Rural, 3 Food Non-Food Overall groups) Overall groups) Food Food Overall Rural) groups) Barisal (Rural) 0.939 1.000 1.000 1.044 0.981 1.021 1.021 1.016 1.016 1.025 0.935 0.992 Barisal (Muni.) 1.072 2.181 1.580 1.146 1.422 1.214 1.148 1.456 1.390 1.147 1.603 1.311 Chittagong (Rural) 1.082 1.000 1.000 1.056 1.026 1.035 1.035 1.042 1.042 1.068 1.233 1.127 Chittagong (Muni.) 1.256 2.181 2.465 1.120 2.087 1.208 1.240 1.467 1.498 1.110 1.415 1.219 Chittagong (SMA) 1.265 2.181 2.465 1.185 2.090 1.245 1.276 1.504 1.536 1.093 1.539 1.253 Dhaka (Rural) 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 Dhaka (Muni.) 1.154 2.181 2.465 1.072 2.056 1.177 1.208 1.426 1.458 1.062 1.440 1.198 Dhaka (SMA) 1.297 2.181 2.465 1.184 2.100 1.246 1.277 1.508 1.540 1.137 1.761 1.361 Khulna (Rural) 1.050 1.000 1.000 0.952 1.016 0.976 0.976 0.980 0.980 0.922 1.023 0.959 Khulna (Muni.) 1.231 2.181 1.580 0.987 1.471 1.135 1.068 1.391 1.325 0.973 1.389 1.122 Khulna (SMA) 1.138 2.181 1.580 1.022 1.442 1.149 1.083 1.397 1.331 1.012 1.242 1.095 Rajshahi (Rural) 0.959 1.000 1.000 0.985 0.987 0.990 0.990 0.986 0.986 0.999 0.983 0.993 Rajshahi (Muni.) 1.100 2.181 1.580 1.027 1.430 1.150 1.084 1.394 1.328 1.030 1.110 1.059 Rajshahi (SMA) 1.168 2.181 1.580 0.998 1.451 1.137 1.071 1.388 1.322 0.971 1.161 1.039 Sylhet (Rural) 1.073 1.000 1.000 1.038 1.023 1.024 1.024 1.031 1.031 0.995 0.665 0.876 Sylhet (Muni.) 1.388 2.181 1.580 1.100 1.520 1.204 1.138 1.475 1.409 1.035 1.051 1.041 158 Table IA-4.2: Food Poor versus Overall Poor by Region Diff. Food Level Poor (percentage Poor points) National 0.32 0.31 0.85 National Rural 0.35 0.35 -0.17 Urban 0.21 0.18 3.70 Barisal 0.39 0.40 -0.16 Chittagong 0.26 0.25 0.92 Dhaka 0.31 0.28 2.60 Division Khulna 0.32 0.32 0.45 Rajshahi 0.36 0.37 -1.32 Sylhet 0.28 0.27 1.51 Barisal (Rural) 0.39 0.40 -0.84 Barisal (Muni.) 0.40 0.37 3.24 Chittagong (Rural) 0.31 0.30 0.68 Chittagong (Muni.) 0.22 0.22 0.27 Chittagong (SMA) 0.07 0.04 2.34 Dhaka (Rural) 0.39 0.38 0.98 Dhaka (Muni.) 0.30 0.25 4.72 Dhaka (SMA) 0.15 0.10 5.10 Stratum Khulna (Rural) 0.31 0.32 -1.41 Khulna (Muni.) 0.32 0.28 4.18 Khulna (SMA) 0.40 0.30 9.58 Rajshahi (Rural) 0.37 0.38 -1.45 Rajshahi (Muni.) 0.31 0.32 -1.47 Rajshahi (SMA) 0.32 0.28 3.52 Sylhet (Rural) 0.30 0.30 0.98 Sylhet (Muni.) 0.15 0.11 4.45 159 Appendix 2: Dietary Diversity Score 1. The dietary diversity score (DDS) estimates a household’s economic ability to consume a set of nutritionally diverse food items. Customarily, food consumption is recorded over a period of 24 hours, and the food tally is used to calculate the household dietary diversity score (FAO 2007). However, as HIES collects detailed consumption data over a period of fourteen days,125 the information required to measure the diversity score, as suggested by FAO (2007) and FANTA (2006), can be replicated up to fourteen times for each household. For the purposes of our analysis, food items consumed on each of the fourteen days for which data is available were categorized into one of twelve food groups (see table below).126 If a household consumed an item from a particular food group, the household was assigned a value of “1” for that food group and “0” otherwise. Hence, for each household, a set of twelve parameters indicates whether or not a certain food group was consumed by any member of the household on each day during the 14-day period. For the analysis presented in this section, the last food group, “other” (see table below), is dropped. Summing over the eleven indicators for each household and for each of the fourteen days yields the household-day HDDS. Household-day is the primary unit of analysis. Classification of Food Group Food Group Name Examples Group Number CEREALS bread, noodles, biscuits, cookies or any other foods made from 1 millet, sorghum, maize, rice, wheat, and other local foods WHITE TUBERS AND ROOTS white potatoes, white yams, cassava, or foods made from roots 2 VITAMIN A RICH pumpkin, carrots, squash, or sweet potatoes that are orange VEGETABLES AND TUBERS inside and other locally available vitamin-A rich vegetables DARK GREEN LEAFY dark green/leafy vegetables, including wild ones and locally VEGETABLES available vitamin-A rich leaves, such as cassava leaves, etc. OTHER VEGETABLES other vegetables (e.g. tomato, onion, eggplant) , including wild 3 vegetables VITAMIN A RICH FRUITS ripe mangoes, cantaloupe, dried apricots, dried peaches, and 4 other locally available vitamin A-rich fruits OTHER FRUITS other fruits, including wild fruits ORGAN MEAT (IRON-RICH) liver, kidney, heart, or other organ meats or blood-based foods 5 FLESH MEATS beef, pork, lamb, goat, rabbit, wild game, chicken, duck, or other birds EGGS 6 FISH fresh or dried fish or shellfish 7 LEGUMES, NUTS, AND beans, peas, lentils, nuts, seeds, or foods made from these 8 SEEDS MILK AND MILK PRODUCTS milk, cheese, yogurt, or other milk products 9 OILS AND FATS oil, fats, or butter added to food or used for cooking 10 SWEETS sugar, honey, sweetened soda, or sugary foods such as 11 chocolates, sweets, or candies SPICES, CONDIMENTS, Spices (black pepper, salt), condiments (soy sauce, hot sauce), 12 BEVERAGES coffee, tea, alcoholic beverages OR local examples Source: FAO (2007). 125 In particular, the HIES revisits the same household seven times over the given period on alternate days to record all food items consumed over the previous two days. 126 The entire per-capita consumption distribution is estimated for each food item for each day of positive consumption. If a household’s food consumption of a particular food item belongs to the bottom one percentile of the respective food item distribution (i.e. when the food item was consumed in trace amount), the household is assigned a zero for that particular food group. 160 Appendix 3: Human Opportunity Index 1. Estimation of the Human Opportunity Index (HOI) involves three steps. The first step is to identify the set of goods and services for which an HOI will be estimated. The goods and services under consideration for our analysis are outlined in the main text. Second, for a given good or service, identification of a set of indicators that may influence access to the good or service is required; these indicators are individual-specific and include factors over which households have no control. These indicators are also referred to as circumstances, which Barros et al. (2003) define as: “personal, family or community characteristics that a child has no control over, and that, for ethical reasons, society wants to be completely unrelated to a child’s access to basic opportunities”. In this analysis, we consider the following circumstances: poverty status of the household; gender of the household head; number of family members; religion of the household; and educational background of the household head. 2. The third step involves estimation of the Coverage rate and the Dissimilarity index. This estimation requires six steps: 1. Estimate a separable logistic model. The dependent variable is a binary variable that takes a value of one if the household accesses the service and zero otherwise. 2. Using the logistic model and the coefficient estimates obtained in step 1, estimate the predicted probability, ̂ . 3. Estimate the Coverage rate, C, computed as ∑ ̂ , where represents an 127 appropriate sampling weight. 4. Compute the Dissimilarity Index, D, by ̂ ∑ ̂ . 5. Compute the Human Opportunity Index as HOI = C (1 – D). 3. Since poverty is included as an additional circumstance using Shapley Decomposition the circumstance impact of poverty is estimated by tracking the change in the Dissimilarity Index.128 More specifically, Suppose D(A, B) is defined as the Dissimilarity Index with circumstances A and B. Then, the circumstance impact of poverty can be computed as D(A, B, Poverty) – D(A, B). We note that this difference is path-dependent; that is, sensitive to a pre-existing set of circumstances prior to the introduction of poverty (here, A and B). Hence, all potential cases, and their relative weights, need to be considered. Then, the circumstance impact of poverty can be calculated as: ( ) ∑ [ ( { }) ( )] where A is circumstance poverty, N is the set of all circumstances containing n elements, S is a subset of N containing s elements that excludes circumstance poverty, D(S) is the dissimilarity index estimated with S circumstances, and D(S ⋃ {A}) is the dissimilarity index computed with the circumstances contained in S and A (poverty). The contribution of poverty to the dissimilarity index can be calculated as ( ) . 127 In this paper, we use the HIES survey design weights or population weights. 128 See Hoyos and Narayan (2012) for more details. 161 Appendix 4: Decomposition Method Chapter 1 The Datt and Ravallion Decomposition Datt and Ravallion (1992) observe that poverty measures (Pt) may be fully characterized by the poverty line (z), the mean of the distribution of economic welfare (μ), and relative inequality, as represented by the Lorenz curve (L), such that: ( ) Then, the overall change in poverty from base period 0 to end period 1 can be written as follows: [ ] [ ( ) ( )] [ ( ) ( )] where is known as the growth component and is the redistribution component. Chapter 3 The Consumption Model 1. We begin with the household consumption identity. Per-capita consumption in household h is defined by: [ ] (1) where n is the number of people in household h, is the consumption-to-income ratio, and y represents household income. The consumption-to-income ratio depends on the average propensity to consume in household h. Empirically, may also capture measurement error or underreporting of household income. If we further disaggregate income into its sources, we can rewrite (1) as: [ ] (2) where , , and are household salaried labor, daily labor129, and self-employed (non-farm) labor income, respectively; is the farm household net revenue function; and is household non-labor income. We slightly modify the Bourguignon and Ferreira (2005) approach and model the household income generating function as: [∑ ( ) ∑ ( ) ∑ ( ) ( ) ] (3) 129 Daily workers in Bangladesh are agricultural or non-farm workers who are hired on a daily basis rather than continually into one or more jobs. They are classified separately as they may potentially belong to multiple sectors, seamlessly transitioning between agricultural and non-farm work. Note that they tend to be less educated and generally disadvantaged in terms of asset ownership compared to other types of workers. 162 where , and are indicator variables equal to one if individual i in household h is a salaried, daily, or self-employed worker, respectively; , and are the corresponding earnings of individual i in household h and depend on individual and household endowments ( ) and the returns to these endowments ( ); is household net revenue from farm activities, which depends on household endowments ( ) and the returns to these endowments; and is household non-labor income. 2. The allocation of individuals across occupations is represented through a multinomial logit model (McFadden 1974a, 1974b), specified as follows: ( ) (4) where is a vector of characteristics specific to individual i and household h; are coefficient vectors, for the following activities j={salaried, daily worker, self-employed, not employed}; and are random variables identically and independently distributed across individuals and activities according to the law of extreme values. Within a discrete utility-maximizing framework, is interpreted as the utility associated with activity s, with as the unobserved utility determinants of activity s and the utility of inactivity arbitrarily set to 0. Similarly, following Bourguignon, Ferreira, and Leite (2008), we estimate a multinomial logit model for an individual’s education level and sector of employment. This estimation allows for a representation of the occupational, sectoral, and educational composition of the work force. 3. Once the composition of the workforce is determined, we then model the heterogeneity in individual earnings for each occupation type j using a log-linear Mincer model: ( ) (5) where is a vector of individual characteristics, is a vector of coefficients, and is a random variable identically and independently distributed across individuals, according to the standard normal law. Farm net revenue is modeled as: (6) where ( ) includes endowments and household characteristics. As before, are vectors of coefficients, and are random variables following a standard normal distribution. 1. Decomposing changes in poverty a la Bourguignon, Ferreira, and Lustig (2005) 5. Given the model presented in equations (1) – (6), two important steps remain in order to obtain decomposition results. The first step consists of defining an estimation strategy with the purpose of obtaining a set of reduced-form parameters. The second entails decomposition based on the construction of approximated counterfactual distributions. a. Estimation strategy 6. The reduced-form models established earlier require the estimation of different sets of parameters, ranging from the occupational choice model, the educational and economic sector conditional distributions, and (random) estimates of the residual terms. This subsection presents the estimation strategy that has been applied. 163 i. Occupational choice model: non-farm workers and farm workers 7. Based on the assumption that the random terms are identically and independently distributed according to the law of the extreme values, a multinomial logit estimator is used for the occupational choice model for the following options: not employed (base category), salaried, daily workers, and self- employed for all individuals in the non-farm sector.130 Separate models are estimated for household heads, spouses, and other members.131 The vector of characteristics is given by a set of individual and household characteristics, such as age, or a range of ages, education level, dependency rates, region, and area, among others. 8. Next, we estimate the residual terms, which are unobserved, of the occupational choice model. They must be drawn from extreme value distributions in a way that is consistent with observed occupational choices. Train and Wilson (2008) define the distribution functions of the extreme value errors, conditional on the chosen alternative. In particular, assume that alternative zero is chosen ( j = 0), and denote ̂ for j = 0,…,J. Define ̂ and ∑ ( ̂ ) where is the logit choice probability. Then, the CDF for the chosen alternative, ,is: ( ) ( ( )) . Calculating the inverse of this distribution yields: ̂ ( ) ( ( )) (a) where µ is drawn from a uniform distribution between 0 and 1. Error terms for other alternatives ( with ) must be calculated conditioning on the error terms of the chosen alternative ( ̂ ). The distribution for these errors is: ( ( )) ( alternative 0 is chosen ) for (̂ ) ( ( (̂ ̂ ))) 10. The inverse of the above distribution is: ̂ ( ( ( ̂ ) )) where (̂ ) ( ( (̂ ̂ ))) (b) where µ is drawn from a uniform distribution between 0 and 1. We repeat (a) and (b) when alternatives other than zero are chosen. 11. In the case of farm workers, we estimate a model for secondary occupational choice in order to capture the probability of diversification into other non-farm activities. We assume that the residuals are independently and identically distributed according to a logistic function. A logit model serves as the estimator for the secondary occupational choice of household heads self-employed in agriculture. The 130 We do not model the occupational decision in primary occupation if the household head is self-employed in agriculture. 131 Since grouping spouses separately for each model yields an insufficient number of observations, we group them into the same category as other members. 164 vector of characteristics includes individual and household variables such as age, gender, education level, region, and areas, among others. Random terms are drawn conditional on the choice that was made at the initial point. ii. Earning equations: non-farm and farm workers 12. Turning to the labor market determination of earnings, we separate the sample into two different groups based on the type of activities performed: non-farm and farm workers. Individual earnings equations for the first group are estimated separately for household heads, spouses, and other members if they perform as daily workers, self-employed, or salaried. The specification includes individual characteristics, such as age, gender, and education level, among others, as well as characteristics of other members of the household. For instance, in the case of spouses and other members, characteristics of the household head, such as education level and employment status, are also included. The second step requires the estimation of the residual terms, as random numbers following a normal distribution, and their variances. 13. As mentioned before, farm net revenues are modeled at the household-level, and parameters are estimated using ordinary least squares. The vector of characteristics includes endowments, such as land and irrigation, and individual and household characteristics of the household head, for instance, educational level, gender, civil status, and number of members involved in the farm activity. Random estimates for the residual terms are drawn from a standard normal distribution. Earnings from the secondary occupation are estimated only for farm workers and as a function of individual characteristics (age, gender, education level, and economic sector of secondary job) as well as a random term following a standard normal distribution. iii. Other characteristics: educational structure and economic sectors for main occupation 14. In the absence of panel data, we are unable to observe the same individuals in both years. Hence, for the estimation of endowments and demographic effects, we must simulate their distributions in year s using the population of year t. We estimate conditional distributions for education levels and economic sectors by occupation categories based on an individual’s age group, gender, region, and area. Following Bourguignon, Ferreira, and Leite (2008), the estimation of both distributions uses a multinomial model. These models are estimated separately for the group of household heads, spouses, and other members within the working-age population. iv. Non-labor income and consumption-income ratio 15. We non-parametrically estimate the conditional distributions of total non-labor income as well as its different components (e.g. remittances, public transfers, and other private transfers) on gender, education level, and area for quantiles of non-labor income. Specifically, we create cells for household heads with the same education level, gender, and region (urban-rural). Within each cell, we create quantiles of non-labor income. A similar approach is employed for estimation of the conditional distribution of the consumption-income ratio. b. Decomposition approach 16. After estimation of each reduced-form model for two years, t and s (early 2000 and late 2010), we decompose distributional changes by formulating the appropriate counterfactual distributions for income and consumption. i. Changes in distribution due to changes in returns to endowments 165 17. We first estimate the following components of household income at time t and s as: ( ) ∑ ( ) ∑ ( ) ( ) ( ), (7) which, for simplicity, we express as: ( ) ( ) (8) We can then estimate the counterfactual household income distribution by computing the earnings of every household at time t with the estimated returns to individual and household characteristics ( ) computed for period s.132 We have: ( ) (̂ ̂ ̂ ̂ ) (9) where the notation “^” stands for ordinary least squares estimates. This simulation yields the earnings for each household in the sample if the returns to each observed characteristic had been what would be observed at time s rather than the actual returns observed at time t. Returns to unobserved characteristics that may be captured by the residual term, ̂ , are assumed to be unchanged. The difference between this simulated distribution of household incomes, { } , and the actual distribution is equivalent to the price effect in the Oaxaca-Blinder calculation. ii. Distributional changes due to changes in unobservable factors 18. To simulate the effect of changes in unobservable factors between s and t, we rescale the estimated residuals of the earning and net revenue equations for non-farm and farm workers of time t by the ratio of standard deviations at time s and t. This counterfactual is defined as: ̂ ( ) (̂ ̂ ̂ ̂ ( ⁄ )) . (10) ̂ iii. Distributional changes due to changes in occupation, education structure, and economic sectors 19. Whenever the occupational, educational, or sectoral coefficients from the multinomial logit model of year t are replaced by those for year s, individuals may be reallocated across occupations, education levels, or economic sectors. Estimated error terms for each reduced-form equation (for occupation, education, and economic sector) remain constant in each decomposition exercise. The labor income imputed for individuals who change occupation status (to a remunerated one), education level, or sector is a linear projection with a relevant vector of parameters and residuals drawn from a standard normal distribution. The difference between the distributions of this set of simulated incomes, { } and the actual set of incomes for period t is comparable to the endowment effect in the Oaxaca-Blinder decomposition. iv. Distributional changes due to changes in demographics 20. The next decomposition consists of altering the joint distribution of exogenous household characteristics, such as age, gender, region, and area for each individual in the household. These variables are not dependent on other exogenous variables in the model; the simulation is performed by recalibrating the population using weights corresponding to the joint distribution of these attributes in the target year. v. Distributional changes due to changes in non-labor income & consumption-income ratio 132 The notation refers to estimating earnings in period t using the returns to characteristics, , estimated at time s. 166 21. The conditional distributions estimated in the previous step are used for the rank-preserving transformation of the observed distribution of non-labor income in each year. In particular, we create cells for household heads with the same level of education, gender, and region (urban-rural). Within each cell, we create quantiles of non-labor income. We estimate the counterfactual distribution of non-labor income in year t by assigning the mean value of non-labor income of quantile q in cell c in year s to the same quantile and cell in year t. In other words, we rank the two distributions according to per-capita household, non-labor income, and if a household with income at time t had a rank of q, we replace with the non-labor income of the household with the same rank at time s. We apply the same decomposition methodology for the consumption-income ratio. 167 Appendix 5: Oaxaca – Blinder decompositions 1. Blinder (1973) and Oaxaca (1973) proposed a methodology to decompose wage gaps into explained and unexplained components. The method is based on the separate estimation of linear earnings equations for the two groups under comparison: for i = {1,2}. (1) Subscript i denotes two states, in this case, public and private sector workers, but it could just as easily represent male and female workers, or period one and two, when decomposing changes over time. Abstracting from the subscript, the conditional mean outcome can be written as ( ) . Therefore, β is a measure of the effect of X on the conditional mean outcome. Furthermore, the law of iterated expectations implies that the unconditional mean outcome is ( ) [ ( )] ( ) . (2) 2. This result implies that β also measures the effect of changing the mean value of X on the unconditional mean value of y. Thus, the average wage gap can be expressed as: [ ( ) ( )] [ ( ) ( ) ]. (3) 3. We construct a counterfactual outcome, namely, the average outcome for state 2, which is valued on the basis of the parameters for state 1, as ( ) . When comparing public to private sector wages, this counterfactual situation can be interpreted as follows: an average private sector worker earns the same reward for his characteristics as an average public sector worker. 4. If we add and subtract this term from the average wage gap, equation (3) can be expressed as: [ ( ) ( )] [ ( )( )]. (4) 5. The first component of the right-hand side, [ ( ) ( )] , represents differences in average characteristics between states 1 and 2; in this case, differences in the average characteristics of male and female workers. The second component, [ ( )( )], represents differences in the average rewards to individuals’ characteristics. However, as Jun, Murphy, and Pierce (1993) point out, these differences may also be attributed to unobserved heterogeneity. Ñopo matching decompositions 6. Several authors have pointed out limitations in the Oaxaca-Blinder approach. For instance, Dolton and Makepeace (1987) and Munroe (1988) explain that the wage gap decomposition is only informative with regard to average unexplained pay differences but not its distribution. Others have noted the method’s reliance on the linearity assumption. For instance, DiNardo, Fortin, and Lemieux (1996) non-parametrically estimate earnings equations to explore the distribution of unexplained differences. 7. In the context of Bangladesh, the most problematic assumption of this approach is the failure to recognize the potential for deep segmentation between the two states under consideration. Although construction of counterfactuals for female workers’ earnings if rewarded at male prices is possible, this construction may be meaningless if, in the actual distribution, female workers do not have the same characteristics as male workers. Similarly, when markets are deeply segmented, a significant number of 168 males, who exhibit a specific set of characteristics, may have no female counterparts. For instance, finding female construction workers or male nurses may be impossible. As described in Ñopo (2008), failure to recognize this problem of segmentation implies an overestimation of the unexplained component of the wage gap. In addition, moving away from linear earnings equations allows for analysis of the distribution of the differences in earnings. Following Ñopo (2008), we model as the conditional cumulative distribution function of individual characteristics, X, conditional on being in state i, where denotes the corresponding probability measure. Let ( ) represent the expected value of earnings, conditional on characteristics and state i: [ ] ( ). Then, the expected values of earnings conditional on individual characteristics and state are: [ ] ∫ ( ) ( ) [ ] ∫ ( ) ( ) where denotes the support of the distribution of characteristics for state i. The wage gap can be defined as: [ ] [ ] ∫ ( ) ( ) ∫ ( ) ( ). (5) 9. Since the support of the distribution of characteristics for state 1 is different from the support of the distribution of characteristics for state 2, each integral is split over its respective domain into two parts, within their intersection and out of the common support, as follows: [∫̅̅̅̅ ( ) ( ) ∫ ( ) ( )] [∫ ( ) ( ) ∫ ̅̅̅̅ ( ) ( )], where the integrals over ̅̅̅ and ̅̅̅ represent the distributions from common support. This expression can then be manipulated to rewrite the wage gap as: 11. In this framework, is analogous to the Oaxaca-Blinder component, which is the portion of the wage gap explained by differences in the distributions of characteristics of individuals in states 1 and 2 over common support. Similarly, is analogous to the unexplained part in Oaxaca-Blinder decomposition; namely, the share of the wage gap that cannot be attributed to differences in characteristics of the individuals and is typically attributed to a combination of unobservable characteristics and discrimination. The new components are and . The former component is the part of the wage gap explained by differences between two groups of individuals in state 1: those with characteristics that can be matched to individuals in state 2, and those with characteristics that cannot be matched to individuals in state 2. Similarly, is the part of the wage gap that can be explained by differences between two groups of individuals in state 2: those possessing characteristics that can be matched to individuals in state 1, and those with characteristics that cannot be matched to individuals in state 1. 169 Appendix 6: The Investment Model 1. Following previous literature, we assume that a firm’s decision to invest is affected by three factors: (i) its organizational structure, (ii) its productivity level, and (iii) external influence. The organizational structure of the firm is proxied by whether or not the owner of the firm is also a manager. Firm productivity is proxied using the following measures: the number of research staff members in the firm; the firm’s use of modern communication modes, such as email and internet; whether or not the firm exports; and whether or not the firm provides in-house training. Factors that might hinder productivity, such as electricity and water shortages, are also included in the regression. External factors affecting investment decisions are proxied by variables such as transportation problems, corruption, unpredictability of law, and problems with the political scenario (as perceived by the firm manager). The model we estimate for capital-deepening is as follows: ( ) , (1) where investment per worker ( ) is measured as a firm’s previous year fixed investment, (which includes investment in property and machinery), divided by the workers employed by that firm, . Capital-deepening is a function of X, which is a vector of controls that measure the firm’s perception of the business environment. As mentioned before, only about one-half of firms in the sample invested in fixed assets. Thus, we employ a Heckman Selection model to account for selection bias. In the first step, we run a probit regression in which the dependent variable (D) equals 1 if the firm invested in fixed assets and 0 otherwise: ( ) ( ) , (2) where Φ(∙) is a cumulative normal distribution and Z is a vector of controls. From the probit, we calculate the inverse Mills’ ratio (the ratio of the PDF to CDF for all i) and then use the ratio as an independent variable in equation (1). Thus, equation (1) becomes ( ) , (3) where INVi is the inverse Mills’ ratio for firm i. We then estimate equation (3) for all firms that invested in fixed assets using the following maximum likelihood estimator (for firm j): ( ) ( ) ( ) ( √ ) ( ) ( ) √ (4) { ( ) where τ is the set of coefficients in equation (3) and x is the vector of all variables used in equation (3). Correlation between the error terms in equations (2) and (3) is represented by ρ, and σ is the standard deviation of the error term in equation (3). Table D4-2 in the Data Annex further describes the variables used in the regression. To estimate the coefficients in equations (2) and (3), we employ maximum likelihood estimation.133 133 Specifically, we use Full Information Maximum Likelihood (FIML). We also use a two-step Limited Information Maximum Likelihood (LIML) to test the validity of the results. We find that the coefficients obtained from using FIML are not very different from using LIML. Therefore, we only present results of the FIML coefficients. 170 Appendix 7: Estimation of wage price elasticity – The Model 1. We seek to exploit micro-data and the exogenous variation induced by the recent hike in international food prices. The idea is to regress district-level changes in rural wages (calculated from household surveys) on changes in an agricultural commodity price index, . Even if all price changes are the same for each district, the changes in will vary across districts because of differences in the shares of total revenue from agricultural production. This also means that, when simulating the welfare impacts of a change in a single price, say, for example, rice, the impact on rural wages in each district will vary according to the district’s share of agricultural revenue from rice production. On similar ground, the variation in wages across districts will depend on agriculture’s labor share, which will vary based on the district’s manufacture and service labor as well. We control for these variations by appropriately weighting wages and prices as follows: Regression: ̃ ̃ , (1) where ̃ ̃ [ ], ̃ ∑ ( ), lagged share of district revenue from crop c to major crops planted ( ), ̃ , and district fixed effects. 2. Among others, one factor that might influence a household’s true exposure to higher food prices is the incidence of tenancy. Given a high incidence of tenancy, the first component of equation (1) may not be a true representation of the net benefit ratio and, thus, must be adjusted. Based on information available in HIES 2010, tenants’ shares for all major crops (and rice, separately) are estimated as: where τ is the adjustment coefficient for tenant i’s take-home production; Loi is land owned by tenant i; Lri is land rented by i; and s is the average share that tenant i pays as rent for land leases. Given this formulation, τ < 1 for tenants since 0 < s < 1, while τ = 1 for owner cultivators since s = 1 and Lri = 0. 3. Similarly, we adjust the landlord’s value of total production by incorporating earnings from lands that have been leased out. Since one-to-one correspondence does not exist between tenant and landlord, we assume that the landlord’s average earning per plot of leased-out land in a particular district is (1 – s), where s is the average share of all tenants in total production. Based on this assumption, we calculate τ for each landlord as: ( ) where, Lcj is land cultivated; Lrj is land leased out; and given 0 < sd < 1, τ > 1 for landlords. Given these specifications, the tenancy-adjusted production, τy, is:  y for tenants   y  =y for owner cultivators   >y for the landlords. Revision of Exposure Index: A GE approach 4. The above partial equilibrium model (equation 1) can be defended in the short-run, when an abrupt change in food prices may not yet have permeated into other sectors. Neither is the change 171 expected to affect labor mobility in the short-run. However, the model is difficult to defend for medium- or long-run analysis. The question arises: what happens if prices in the service sector are endogenously determined as food prices rise? Similarly, how does the wage price elasticity vary across the labor market? Or, how does labor mobility across sectors influence the elasticity? The above partial equilibrium model fails to address these questions. In order to touch upon these issues, we adopt a general equilibrium framework incorporating three production sectors in each district: agriculture (A), manufactures (M), and services (S). Output from the agricultural and manufacturing sectors can be traded across district borders, but service sector output is not tradable across district borders. Labor is perfectly mobile across sectors but immobile across districts. Capital (land) is sector-specific and assumed to be immobile. For our medium-term analysis, we revise our previous exposure index: (3) ( ) ( ) where is the share of expenditures on non-tradable services and η is the elasticity of prices in the service sector (S) with respect to agriculture (A). We assume η = ε. 5. In the medium-term, price adjustments in the labor market require revision of the wage-price equation stated in (1). The revised regression is as follows: Regression: ̃ ̃ where ̃ [ ] ̃ ̃ ∑ ( ) lagged share of district revenue from crop to major crops planted ( ) ̃ where and ( ) district fixed effects. (4) 6. The parameters λj represent labor shares in sector j (j = A, M, and S). Other specifications remain similar to the previous model. Notice that the previous model (1) is a special case of this model (4) when assuming that prices in other sectors (e.g., service sector) are not affected; in other words, λS,d,t–5 = 0. Given that the value of ∑ , when we consider all major crops for our analysis, the wage-price ̃ elasticity, ε, for district d becomes ̃ , where w is the rural wage relative to a fixed price in the manufacturing sector, and PA is the agricultural sector price index. 7. The value of γ lies within 0 and 1 and depends on the severity of labor market friction. If the labor market is perfectly mobile, then γ = 0. Alternatively, if the labor market is perfectly immobile (i.e. no mobility across sectors), then γ = 1. The underlying assumption is that under perfect labor mobility, laborers from a given sector will move into another sector with higher wages until wages across sectors equalize. On the other hand, wages proportionally adjust when labor supply is perfectly inelastic. 172 Tables Annex Chapter 2 Table A2-1: Gross and Net Enrollment - 2005 and 2010 Primary Secondary Higher secondary Both Male Female Both Male Female Both Male Female Quintile Gross enrollment 1 77 71 83 30 25 35 7 8 6 2 91 91 91 41 36 48 12 11 15 3 98 98 98 60 54 67 17 23 10 4 101 101 102 78 74 83 25 25 26 5 99 102 96 98 99 96 57 57 57 Poor 83 80 87 36 31 41 14 16 11 Non-poor 100 100 99 78 74 82 31 30 32 Rural 91 90 93 59 55 63 19 20 19 Urban 93 92 93 71 68 73 45 48 43 2005 Quintile Net enrollment 1 60 56 63 23 16 30 3 2 3 2 64 63 66 36 33 40 3 2 4 3 71 70 72 45 38 53 5 6 3 4 75 75 77 55 49 62 8 7 10 5 80 81 79 69 66 71 20 19 20 Poor 61 60 62 27 21 32 3 3 3 Non-poor 77 77 78 56 52 61 12 12 12 Rural 68 67 69 43 38 47 6 6 6 Urban 75 74 77 54 50 56 18 19 17 Quintile Gross enrollment 1 93 85 102 37 31 43 11 12 10 2 103 96 109 53 47 59 23 20 26 3 104 104 105 66 62 71 39 35 46 4 106 106 105 75 70 81 48 47 50 5 105 104 106 82 81 84 81 83 80 Poor 93 86 101 41 35 47 15 15 14 Non-poor 107 105 109 73 70 77 54 52 58 Rural 102 98 106 62 59 66 39 39 40 Urban 100 95 105 66 63 70 64 57 42 2010 Quintile Net enrollment 1 72 66 78 31 25 37 5 3 7 2 78 73 82 44 39 49 13 12 13 3 78 77 78 51 45 56 17 15 21 4 79 79 79 58 52 65 19 17 23 5 81 80 82 66 64 67 36 40 31 Poor 72 67 77 34 28 41 7 7 8 Non-poor 81 79 83 57 54 62 25 23 28 Rural 77 74 81 49 45 54 18 17 20 Urban 78 75 82 54 50 59 28 25 32 Source: HIES 2005 and 2010. 173 Chapter 3 Table A3-1: Percentage Change in Average Per-capita Income (PCI) by Source (constant terms 2005 in Rural Dhaka prices) Deciles Total Labor Non-labor Intl. Remit. Deciles Total Labor Non-labor Intl. Remit. of PCI of PCI 2000 vs. 2005 2005 vs. 2010 Overall 19 22 9 34 Overall 54 66 19 106 1 7 10 -4 22 1 29 35 6 76 2 6 9 -9 -50 2 35 45 -15 234 3 7 13 -16 17 3 37 47 -17 79 4 10 15 -7 -6 4 40 55 -24 118 5 13 17 1 -25 5 43 55 -10 151 6 16 20 -1 9 6 44 58 -11 172 7 17 25 -5 24 7 45 58 -3 117 8 19 25 3 13 8 48 60 12 91 9 21 20 23 80 9 50 66 8 50 10 27 31 18 35 10 74 85 49 126 Note: Income is adjusted for spatial cost of living differences using the poverty line and for inflation using the CPI. Source: World Bank staff estimates based on HIES 2000, 2005, 2010 Table A3-2: Percentage Change in Per-capita Income (PCI) Shares by Source (constant 2005 terms in rural Dhaka prices) Deciles Labor Non-labor Intl. Remit. Deciles Labor Non-labor Intl. Remit. of PCI of PCI 2000 vs. 2005 2005 vs. 2010 Overall 3 -9 28 Overall 7 -22 65 1 4 -17 2 1 7 -35 105 2 5 -23 -24 2 7 -43 294 3 6 -24 60 3 9 -46 87 4 5 -19 4 4 10 -47 178 5 4 -14 -12 5 9 -38 140 6 4 -14 15 6 9 -38 175 7 8 -22 32 7 8 -31 105 8 5 -13 14 8 8 -23 51 9 -1 2 48 9 11 -29 24 10 2 -5 17 10 4 -12 53 Note: Income is adjusted for spatial cost of living differences using the poverty line and for inflation using the CPI. Source: World Bank staff estimates based on HIES 2000, 2005, 2010. 174 Table A3-3: Characteristics of the Population and Labor Force 2000 2005 2010 Population and demographics Total (millions) 71.0 81.9 89.8 Men (percent of total) 50.4 49.9 48.4 Women (percent of total) 49.6 50.1 51.6 Rural (percent of total) 78.2 73.3 72.0 Urban (percent of total) 21.8 26.7 28.0 Average household size 5.2 4.8 4.5 Number of adults per household 3.13 3.07 2.94 Share of adults per household 63.4 66.3 68.4 Occupied adults (as a share of number of adults) 48.4 49.5 50.1 Labor force participation (percent of working age population) All 49.4 47.6 49.2 Men 83.3 83.1 84.9 Women 14.9 12.3 15.6 Employment (percent of working age population) All 46.6 47.1 47.8 Men 81.6 82.3 82.9 Women 11.1 12.1 14.8 Unemployment (percent of labor force) All 5.6 1.1 1.4 Men 2.0 1.0 2.0 Women 25.9 1.7 0.8 Education levels (percent of working age population) Illiterate & Incomplete primary 57.2 49.8 47.1 Complete primary & Lower secondary 33.8 40.3 43.4 Higher secondary & Tertiary 8.9 9.9 9.4 Labor relation (percent of employed population) Daily workers 33.5 32.3 32.4 Self-employed 46.3 45.2 42.3 Salaried 20.2 22.5 25.4 Economic Sector (percent of employed population) Agriculture 49.2 44.3 41.8 Manufacturing 22.9 23.8 24.4 Services 27.9 32.0 33.7 Area (percent of employed population) Rural 78.6 73.4 71.6 Urban 21.4 26.6 28.4 Note: Working-age population includes individuals between 15 and 64 years old. Source: Own estimations based on HIES 2000, 2005, and 2010. 175 Table A3-4: Consumption-to-income ratio 2000 2005 2010 Average 1.26 1.36 1.22 Deciles (per-capita income) 1 3.69 4.22 3.46 2 1.39 1.56 1.53 3 1.14 1.32 1.29 4 1.07 1.16 1.13 5 1.05 1.08 1.03 6 1.01 1.02 0.92 7 0.97 0.95 0.83 8 0.95 0.85 0.76 9 0.88 0.81 0.67 10 0.76 0.65 0.50 Spearman rank 0.79 0.63 0.56 Correlation 0.62 0.53 0.18 Coefficient of Variation Consumption 0.87 0.80 0.78 Income 1.06 1.19 2.99 Note: Household consumption as a share of total household income. Source: Own estimates based on HIES 2000, 2005, and 2010. 176 Table A3-5: Simulating the Characteristics of Household Heads 2000 2005 2010 Simulated Simulated Simulated Actual Actual Actual 2005 2010 2000 2010 2000 2005 Education Structure Illiterate & Incomplete primary 63.2 61.7 61.9 57.7 59.0 57.6 57.0 57.8 56.9 Primary & Low Secondary 29.0 29.6 29.6 32.6 32.4 32.6 33.1 33.2 33.2 High Secondary & Tertiary 7.7 8.7 8.5 9.7 8.7 9.8 9.9 9.0 9.9 P-value of Pearson chi-square 0.918 0.944 0.934 0.999 0.956 1.000 Occupation Non-employed 9.6 8.7 10.1 7.0 7.5 7.5 8.0 7.7 6.9 Daily workers 45.5 42.3 40.7 45.2 49.0 44.1 43.4 49.9 45.4 Self-employed - Non-Agriculture 24.7 25.3 25.1 23.5 22.3 23.3 24.1 23.9 25.3 Salaried 20.2 23.7 24.2 24.3 21.2 25.2 24.5 18.6 22.5 P-value of Pearson chi-square 0.812 0.716 0.848 0.992 0.495 0.926 Economic Sectors Daily workers -Agriculture 58.8 57.4 56.4 50.3 51.5 50.4 51.8 51.8 52.1 -Manufacturing 12.3 13.5 12.8 10.3 9.8 10.5 12.6 13.1 12.9 -Industry 7.9 8.2 8.5 11.2 11.4 11.6 11.9 12.3 11.3 -Services 21.0 21.0 22.4 28.2 27.3 27.5 23.7 22.8 23.7 P-value of Pearson chi-square 0.984 0.969 0.994 0.998 0.995 0.998 Self-employed -Manufacturing 34.9 35.5 35.3 31.4 30.4 30.9 17.6 17.0 16.9 -Industry 6.2 6.9 7.8 5.5 5.8 6.7 1.7 1.9 2.2 -Services 58.9 57.6 57.0 63.1 63.8 62.4 80.7 81.1 81.0 P-value of Pearson chi-square 0.938 0.785 0.971 0.866 0.985 0.933 Salaried -Agriculture 8.6 6.7 7.3 5.7 6.7 4.9 5.4 6.9 5.8 -Manufacturing 27.7 27.5 26.8 26.6 27.8 28.2 33.4 34.3 32.1 -Industry 3.7 4.0 3.8 5.2 5.2 5.0 3.9 4.0 4.3 -Services 60.0 61.8 62.2 62.6 60.3 62.0 57.3 54.9 57.2 P-value of Pearson chi-square 0.919 0.958 0.956 0.975 0.919 0.987 Source: Own estimates based on HIES 2000, 2005, and 2010. 177 Table A3-6: Simulating the Characteristics of Other Household Members 2000 2005 2010 Simulated Simulated Simulated Actual Actual Actual 2005 2010 2000 2010 2000 2005 Education Structure Illiterate & Incomplete primary 54.4 54.1 54.5 45.9 45.7 46.2 42.1 42.3 41.9 Primary & Low Secondary 36.2 36.2 36.0 44.0 44.5 43.6 48.6 48.2 48.8 High Secondary & Tertiary 9.5 9.7 9.6 10.1 9.8 10.2 9.3 9.4 9.3 P-value of Pearson chi-square 0.996 0.998 0.994 0.996 0.996 0.999 Occupation Non-employed 79.8 79.2 83.0 78.3 79.3 81.3 78.0 80.1 79.1 Daily workers 8.6 8.4 4.7 8.4 8.4 4.7 8.2 5.6 5.7 Self-employed - Non-Agriculture 4.4 4.8 4.8 5.2 5.0 5.8 4.1 4.4 4.7 Salaried 7.2 7.7 7.5 8.2 7.3 8.1 9.7 9.9 10.6 P-value of Pearson chi-square 0.996 0.586 0.991 0.618 0.809 0.806 Economic Sectors Daily workers -Agriculture 54.8 55.3 53.4 47.5 49.5 47.6 42.7 44.1 45.6 -Manufacturing 19.0 17.3 18.3 20.4 18.6 20.5 21.5 20.4 18.4 -Industry 8.4 9.6 8.9 11.4 11.0 11.3 15.8 16.4 15.8 -Services 17.8 17.9 19.4 20.8 20.8 20.6 20.0 19.2 20.2 P-value of Pearson chi-square 0.952 0.968 0.969 1.000 0.985 0.885 Self-employed -Manufacturing 41.1 42.0 45.7 36.5 34.0 38.2 26.8 22.1 26.1 -Industry 7.1 5.9 6.9 6.4 5.2 6.6 3.1 2.6 2.4 -Services 51.8 52.1 47.4 57.1 60.8 55.2 70.1 75.4 71.5 P-value of Pearson chi-square 0.890 0.630 0.720 0.927 0.515 0.897 Salaried -Agriculture 4.4 4.2 3.4 4.7 6.1 4.4 3.0 3.7 2.8 -Manufacturing 42.6 39.9 42.4 41.7 43.6 44.3 47.3 47.5 44.8 -Industry 2.5 2.5 2.0 2.9 2.9 2.6 2.6 2.7 2.9 -Services 50.5 53.3 52.2 50.8 47.4 48.9 47.2 46.0 49.6 P-value of Pearson chi-square 0.953 0.942 0.858 0.966 0.973 0.962 Source: Own estimates based on HIES 2000, 2005, and 2010. 178 Table A3-7: Non-Farm Earnings – Household Heads (Individuals between 15 and 64 years old) 2000 2005 2010 Daily Self- Self- Self- Salaried Daily workers Salaried Daily workers Salaried workers employed (1) employed (1) employed (1) Primary & Lower 0.132*** 0.370*** 0.330*** 0.100*** 0.350*** 0.308*** 0.0530*** 0.312*** 0.357*** secondary (0.0392) (0.0484) (0.0499) (0.0241) (0.0413) (0.0465) (0.0150) (0.0387) (0.0387) Higher secondary 0.420** 0.994*** 0.796*** 0.440*** 0.941*** 0.782*** 0.248*** 0.632*** 0.826*** & Terciary (0.164) (0.0810) (0.0551) (0.131) (0.0659) (0.0509) (0.0654) (0.0625) (0.0420) Age 0.0510*** 0.0591*** 0.0529*** 0.0328*** 0.0571*** 0.0482*** 0.0224*** 0.0411*** 0.0916*** (0.00989) (0.0186) (0.0161) (0.00616) (0.0152) (0.0136) (0.00426) (0.0149) (0.0117) Age squared -0.000666*** -0.000664*** -0.000570*** -0.000415*** -0.000663*** -0.000475*** -0.000267*** -0.000541*** -0.00107*** (0.000119) (0.000219) (0.000191) (7.39e-05) (0.000180) (0.000162) (5.09e-05) (0.000172) (0.000138) Female -0.894*** -1.477*** -0.956*** -0.665*** -1.226*** -0.594*** -0.688*** -1.080*** -0.709*** (0.0621) (0.166) (0.0849) (0.0417) (0.122) (0.0734) (0.0266) (0.138) (0.0609) Urban 0.0863* 0.234*** 0.108** -0.0210 0.210*** 0.0224 -0.0217 0.264*** 0.117*** (0.0447) (0.0495) (0.0421) (0.0243) (0.0415) (0.0367) (0.0164) (0.0396) (0.0315) Barisal 0.0601 -0.193** -0.0421 -0.0124 0.0254 -0.236*** 0.0120 -0.0592 -0.0989 (0.0641) (0.0966) (0.0880) (0.0424) (0.0854) (0.0776) (0.0293) (0.0857) (0.0671) Chittagong -0.0891* -0.0956 -0.0181 0.212*** 0.0676 0.202*** 0.0488** -0.152** -0.0456 (0.0461) (0.0622) (0.0521) (0.0286) (0.0658) (0.0453) (0.0194) (0.0597) (0.0405) Khulna 0.103** -0.108 -0.0193 0.0383 0.0171 0.00712 -0.216*** -0.212*** -0.184*** (0.0478) (0.0747) (0.0677) (0.0300) (0.0698) (0.0626) (0.0194) (0.0621) (0.0506) Rajsahi -0.182*** -0.242*** -0.195*** -0.0138 -0.136*** -0.00594 -0.179*** -0.314*** -0.112** (0.0364) (0.0589) (0.0610) (0.0247) (0.0503) (0.0542) (0.0160) (0.0466) (0.0454) Sylhet -0.0841 -0.183 -0.370*** 0.137*** 0.306*** -0.129 -0.0994*** 0.302*** -0.0124 (0.0604) (0.131) (0.111) (0.0441) (0.0781) (0.0801) (0.0298) (0.0846) (0.0812) Manufacturing 0.432*** 0.446*** 0.396*** 0.231*** 0.0674*** 0.107 (0.0456) (0.0780) (0.0316) (0.0792) (0.0196) (0.0699) Industry 0.178*** 0.00876 0.433*** 0.421*** 0.0872 0.217** 0.272*** -0.0833 0.250** (0.0547) (0.0962) (0.121) (0.0307) (0.0879) (0.102) (0.0201) (0.146) (0.0969) Services 0.429*** -0.0220 0.403*** 0.391*** -0.0116 0.233*** 0.111*** -0.0844* 0.0882 (0.0379) (0.0472) (0.0740) (0.0221) (0.0420) (0.0752) (0.0156) (0.0475) (0.0670) Public job 0.178*** 0.224*** 0.320*** (0.0452) (0.0405) (0.0393) Constant 6.395*** 6.636*** 6.256*** 6.675*** 6.657*** 6.406*** 7.190*** 7.164*** 5.698*** (0.198) (0.386) (0.334) (0.125) (0.315) (0.283) (0.0872) (0.317) (0.248) Observations 2092 1269 1134 2829 1531 1532 3390 2007 1820 R-squared 0.216 0.240 0.401 0.245 0.260 0.321 0.289 0.183 0.373 Adj R-squared 0.211 0.232 0.393 0.241 0.253 0.314 0.286 0.177 0.368 Note: (1) Self-employed in Non-agriculture. Dhaka is the base region. Illiterate and Incomplete primary education is the base for education levels. Agriculture is base sector for Daily and Salaried while Manufacturing is base sector for Self-employed in Non-Agriculture. Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1. Source: Own estimates based on HIES 2000, 2005, and 2010. 179 Table A3-8: Non-Farm Earnings – Other Members (Individuals between 15 and 64 years old) 2000 2005 2010 Self- Self- Self- Daily workers Salaried Daily workers Salaried Daily workers Salaried employed (1) employed (1) employed (1) Primary & Lower 0.133** 0.449*** 0.459*** 0.0293 0.234*** 0.439*** 0.0492** 0.0246 0.292*** secondary (0.0554) (0.0831) (0.0674) (0.0363) (0.0651) (0.0486) (0.0211) (0.0676) (0.0372) Higher secondary & 0.114 0.975*** 0.837*** 0.210 0.715*** 0.814*** 0.466*** 0.331*** 0.954*** Terciary (0.270) (0.124) (0.0854) (0.203) (0.0966) (0.0612) (0.0992) (0.107) (0.0478) Age 0.0514*** 0.0693*** 0.0478*** 0.0425*** 0.0574*** 0.0958*** 0.0384*** 0.124*** 0.0693*** (0.0131) (0.0222) (0.0178) (0.00949) (0.0147) (0.0117) (0.00593) (0.0167) (0.00975) Age squared -0.000633*** -0.000774** -0.000444 -0.000521*** -0.000548** -0.00114*** -0.000470*** -0.00170*** -0.000846*** (0.000203) (0.000350) (0.000282) (0.000149) (0.000223) (0.000183) (9.17e-05) (0.000246) (0.000153) Female -1.078*** -1.244*** -0.682*** -0.957*** -0.831*** -0.406*** -0.691*** -1.132*** -0.462*** (0.0698) (0.121) (0.0656) (0.0502) (0.0783) (0.0418) (0.0301) (0.0841) (0.0331) Urban -0.0135 0.0209 -0.147** -0.000481 0.0855 -0.149*** -0.104*** 0.287*** -0.0407 (0.0737) (0.0803) (0.0619) (0.0475) (0.0607) (0.0405) (0.0276) (0.0615) (0.0316) Barisal 0.216* -0.0186 -0.228* -0.167** -0.0541 -0.257*** 0.0885* 0.156 -0.0396 (0.111) (0.151) (0.124) (0.0842) (0.131) (0.0809) (0.0502) (0.127) (0.0752) Chittagong -0.200*** -0.123 -0.178** 0.0361 0.100 0.170*** 0.101*** -0.149* -0.0281 (0.0756) (0.101) (0.0704) (0.0538) (0.0808) (0.0477) (0.0321) (0.0883) (0.0388) Khulna -0.107 -0.0433 -0.215* -0.142** -0.0310 -0.269*** -0.204*** -0.297*** -0.237*** (0.0855) (0.121) (0.117) (0.0574) (0.0996) (0.0726) (0.0338) (0.0907) (0.0562) Rajsahi -0.227*** -0.396*** -0.295*** -0.119** -0.162** -0.294*** -0.103*** -0.00253 -0.0741 (0.0665) (0.103) (0.0910) (0.0466) (0.0760) (0.0624) (0.0281) (0.0756) (0.0481) Sylhet -0.231** -0.481*** -0.636*** 0.132* 0.328*** -0.119 0.0448 0.125 -0.0857 (0.103) (0.166) (0.125) (0.0718) (0.110) (0.0795) (0.0389) (0.112) (0.0741) Manufacturing 0.300*** 0.463*** 0.176*** 0.251*** -0.169*** 0.149* (0.0674) (0.145) (0.0460) (0.0934) (0.0274) (0.0904) Industry 0.448*** -0.0366 0.440** 0.326*** 0.0989 0.305** 0.174*** 0.700*** 0.0853 (0.0897) (0.148) (0.221) (0.0566) (0.117) (0.138) (0.0296) (0.171) (0.126) Services 0.469*** 0.0881 0.297** 0.345*** 0.0454 0.182* 0.0930*** 0.195*** -0.0847 (0.0692) (0.0777) (0.144) (0.0448) (0.0591) (0.0931) (0.0276) (0.0707) (0.0905) Public job 0.555*** 0.371*** 0.437*** (0.0921) (0.0606) (0.0522) Constant 6.270*** 6.379*** 6.198*** 6.538*** 6.503*** 5.676*** 6.974*** 5.456*** 6.227*** (0.192) (0.334) (0.285) (0.140) (0.231) (0.194) (0.0892) (0.273) (0.163) Observations 1153 673 1104 1552 1021 1470 1857 973 1984 R-squared 0.307 0.317 0.285 0.288 0.239 0.359 0.383 0.343 0.348 Adj R-squared 0.299 0.304 0.275 0.281 0.229 0.352 0.379 0.334 0.343 Note: (1) Self-employed in Non-agriculture. Dhaka is the base region. Illiterate and Incomplete primary education is the base for education levels. Agriculture is base sector for Daily and Salaried while Industry is base sector for Self-employed in Non-Agriculture. Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1. Source: Own estimates based on HIES 2000, 2005, and 2010. 180 Table A3-9: Farm Earnings Individuals between 15 and 64 years old 2000 2005 2010 0.132** 0.0732 0.0331 Primary & Lower secondary (0.0633) (0.0463) (0.0489) Higher secondary & Tertiary 0.295** 0.479*** -0.00290 (0.147) (0.103) (0.119) Age 0.142*** 0.0976*** 0.0795*** (0.0214) (0.0153) (0.0172) Age squared -0.00175*** -0.00110*** -0.000843*** (0.000245) (0.000172) (0.000190) Female -2.243*** -1.661*** -0.926*** (0.123) (0.0662) (0.0605) Urban 0.197 -0.00530 -0.161** (0.123) (0.0781) (0.0816) Barisal -0.285** -0.177* 0.270*** (0.123) (0.0907) (0.0920) Chittagong -0.240** 0.0481 0.164** (0.0977) (0.0665) (0.0680) Khulna 0.405*** 0.345*** 0.204*** (0.0955) (0.0730) (0.0751) Rajshahi 0.0125 0.297*** 0.303*** (0.0763) (0.0557) (0.0629) Sylhet 0.0104 0.0697 0.253** (0.126) (0.106) (0.108) Low land -0.107 0.609*** 0.743*** (0.0986) (0.0669) (0.0666) High land 0.457*** 1.095*** 1.333*** (0.111) (0.0772) (0.0786) Irrigation 0.458*** 0.276*** 0.408*** (0.0941) (0.0732) (0.0707) Number of members 0.525*** 0.496*** 0.429*** (0.0507) (0.0458) (0.0525) Constant 4.007*** 4.077*** 4.780*** (0.462) (0.344) (0.381) Observations 1678 2612 2956 R-squared 0.340 0.400 0.323 Adj. R-squared 0.334 0.396 0.320 Note: Sample: Self-employed in Agriculture. Dhaka is the base region. Illiterate and Incomplete primary education is the base for education levels. No-land is the base category. Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1. Source: Own estimates based on HIES 2000, 2005, and 2010. 181 Table A3-10: Marginal Contributions to the Change in Poverty Head Count Ratio 2000-2005 2005-2010 2000-2010 Percentage Share of Percentage Share of Percentage Share of point total point total point total change change change change change change A. LABOR INCOME I. Non-Farm Labor Income Returns to characteristics -3.54 39.9 -0.66 7.8 -3.52 20.3 Occupational-choice -0.09 1.0 -0.65 7.6 -1.61 9.3 Economic Sector -0.44 4.9 0.07 -0.9 -0.48 2.8 Education -0.58 6.5 0.05 -0.6 -0.55 3.1 Unobservable factors 1.30 -14.7 0.56 -6.6 1.59 -9.2 II. Farm Labor Income Returns to characteristics -1.64 18.5 -6.27 73.9 -6.98 40.2 Occupational-choice 0.35 -4.0 0.17 -2.1 0.56 -3.2 Education -0.20 2.2 0.32 -3.8 0.13 -0.8 Unobservable factors 0.24 -2.8 0.12 -1.4 0.26 -1.5 B. NON-LABOR INCOME Capital 0.55 -6.2 0.74 -8.7 1.31 -7.5 Domestic Transfers 0.36 -4.0 1.02 -12.0 1.10 -6.3 International Transfers -0.51 5.8 -1.33 15.7 -1.94 11.2 Other transfers 0.30 -3.4 0.28 -3.3 0.58 -3.3 C. OTHER Age-gender-regional structure -2.35 26.6 -2.17 25.6 -4.41 25.4 Consumption-to-income ratio -2.89 32.6 5.06 -59.7 0.93 -5.4 Unexplained 0.26 -2.9 -5.81 68.4 -4.34 25.0 TOTAL -8.86 100.0 -8.49 100.0 -17.34 100.0 Source: Own estimates based on HIES 2000, 2005, and 2010. 182 Table A3-11: Contributions to the Change in Poverty Head Count Ratio - Parameter Effects 2000-2005 2005-2010 2000-2010 Non-Farm Households Farm Households Non-Farm Households Farm Households Non-Farm Households Farm Households Percentage Share of Percentage Share of Percentag Share of Percentage Share of Percentage Share of Percentag Share of point total point total e point total point total point total e point total change change change change change change change change change change change change Returns to characteristics -3.54 39.9 -1.64 18.5 -0.66 7.8 -6.27 73.9 -3.52 20.3 -6.98 40.2 Education 1.07 -12.0 0.19 -2.1 0.87 -10.2 0.70 -8.3 1.77 -10.2 0.78 -4.5 Experience 3.68 -41.5 6.01 -67.8 2.62 -30.9 2.08 -24.5 3.94 -22.7 4.19 -24.2 Female -0.50 5.7 -0.23 2.6 0.12 -1.4 0.01 -0.1 -0.44 2.6 -0.20 1.1 Urban premium 0.57 -6.4 0.02 -0.2 -0.59 7.0 0.51 -6.0 0.11 -0.6 0.45 -2.6 Dhaka premium -4.76 53.7 -1.98 22.3 3.43 -40.4 -0.20 2.3 -0.87 5.0 -1.82 10.5 Manufacturing and service sector premium 2.09 -23.6 0.14 -1.6 4.96 -58.5 0.09 -1.1 5.85 -33.7 0.21 -1.2 Returns to Land -6.21 70.1 -1.43 16.9 -7.25 41.8 Irrigation 0.18 -2.0 0.15 -1.8 0.30 -1.8 Other members 0.04 -0.5 1.27 -14.9 1.48 -8.5 Constant -4.92 55.5 -0.10 1.1 -11.73 138.2 -7.31 86.2 -14.87 85.7 -7.74 44.7 Source: Own estimates based on HIES 2000, 2005, and 2010. 183 Table A3-12: Cumulative Contributions to the Change in Poverty Head Count Ratio 2000-2005 2005-2010 2000-2010 Percentage Share Percentage Share Percentage Share point of total point of total point of total change change change change change change Demographics -2.35 26.6 -2.17 25.6 -4.41 25.4 Education -0.81 9.2 -0.05 0.6 -0.88 5.1 Occupation -0.57 6.4 -0.52 6.2 -1.38 8.0 Sector -0.55 6.2 0.07 -0.8 -0.51 2.9 Returns non-farm -3.49 39.3 -0.50 5.9 -2.93 16.9 Returns farm -1.90 21.5 -7.77 91.5 -8.18 47.1 Residuals non-farm 1.04 -11.8 0.55 -6.4 1.39 -8.0 Residuals farm 0.38 -4.3 -0.07 0.8 -0.06 0.4 International Remittances -0.71 8.0 -1.31 15.5 -1.93 11.2 Others 0.10 -1.1 3.29 -38.80 1.55 -9.0 Total -8.86 100.0 -8.49 100.0 -17.34 100.0 Source: Own estimates based on HIES 2000, 2005, and 2010. 184 Table A3-13: Multinomial Logit on Occupational Choice – Household Heads Individuals between 15 and 64 years old 2000 2005 Self- Self- Daily Daily employed Salaried employed Salaried workers workers (1) (1) Primary & Lower -1.204*** -0.0182 0.614*** -1.587*** -0.253 0.574*** secondary (0.160) (0.161) (0.166) (0.155) (0.158) (0.159) Higher secondary -2.408*** 0.301 2.002*** -3.931*** -0.426* 1.426*** & Tertiary (0.368) (0.294) (0.287) (0.349) (0.236) (0.227) Age 0.0408 0.126** 0.114** 0.180*** 0.294*** 0.290*** (0.0510) (0.0544) (0.0556) (0.0487) (0.0513) (0.0510) Age squared -0.00141** -0.00228*** -0.00208*** -0.00271*** -0.00395*** -0.00391*** (0.000572) (0.000611) (0.000627) (0.000551) (0.000583) (0.000581) Urban -0.354** 0.421*** 0.795*** -0.668*** -0.0584 0.223 (0.160) (0.161) (0.162) (0.145) (0.147) (0.146) Barisal 0.0892 -0.0402 -0.313 1.081*** 0.346 0.219 (0.263) (0.275) (0.288) (0.358) (0.363) (0.362) Chittagong -0.490*** -0.310* -0.216 0.368** -0.890*** -0.285 (0.179) (0.184) (0.186) (0.177) (0.186) (0.177) Khulna 0.432* 0.273 0.0118 0.298 -0.847*** -0.931*** (0.231) (0.240) (0.248) (0.195) (0.205) (0.206) Rajshahi 1.735*** 1.400*** 0.834*** 2.163*** 1.266*** 0.654*** (0.227) (0.234) (0.242) (0.238) (0.241) (0.245) Sylhet 0.541* -0.0682 0.126 0.457 0.228 -0.00582 (0.281) (0.319) (0.323) (0.294) (0.297) (0.300) Attends school -3.549*** -3.393*** -2.742*** -2.485** -4.637*** -2.212*** (0.753) (0.866) (0.724) (1.113) (1.768) (0.691) Remittances -0.646*** -1.014*** -0.715*** -0.595*** -0.606*** -0.398** (0.172) (0.178) (0.181) (0.181) (0.185) (0.185) Female -2.213*** -2.993*** -1.178*** -2.395*** -2.786*** -1.785*** (0.409) (0.489) (0.444) (0.458) (0.497) (0.471) Remitt x Female -0.328 -0.985* -1.450*** -1.530*** -1.463*** -1.233*** (0.321) (0.595) (0.431) (0.338) (0.501) (0.351) Married 0.558 0.683* 0.593 0.424 0.387 0.0429 (0.367) (0.384) (0.389) (0.419) (0.432) (0.424) Married x Female -2.269*** -1.432** -1.876*** -1.610*** -1.481** -1.189** (0.496) (0.651) (0.569) (0.545) (0.655) (0.550) Other member -0.521*** -0.385*** -0.513*** -0.509*** -0.426*** -0.527*** employed (0.0700) (0.0717) (0.0762) (0.0748) (0.0766) (0.0775) Number of children 0.246*** 0.267*** 0.213*** 0.0556 0.0609 -0.0238 (0.0484) (0.0503) (0.0521) (0.0508) (0.0525) (0.0530) Constant 3.198*** 0.0640 -0.443 0.630 -2.512** -2.824*** (1.030) (1.094) (1.113) (0.923) (0.977) (0.966) Observations 4974 4974 4974 6353 6353 6353 Pseudo R2 0.233 0.233 0.233 0.253 0.253 0.253 Notes: (1) Self-employed in Non-agriculture. Dhaka is the base Non-Employed is the base category region. Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1. Source: Own estimates based on HIES 2000, 2005, and 2010. 185 Table A3-13. Cont. 2010 Self- Daily workers Salaried employed (1) Primary & Lower -1.082*** 0.122 0.682*** secondary (0.145) (0.147) (0.147) Higher secondary & -3.461*** -0.599*** 1.227*** Tertiary (0.267) (0.203) (0.194) Age 0.109** 0.281*** 0.238*** (0.0423) (0.0449) (0.0440) Age squared -0.00211*** -0.00382*** -0.00360*** (0.000473) (0.000504) (0.000496) Urban -1.150*** -0.411*** -0.00785 (0.134) (0.135) (0.133) Barisal 0.101 -0.330 -0.310 (0.255) (0.262) (0.258) Chittagong 0.228 -0.525*** -0.264 (0.161) (0.168) (0.161) Khulna 1.127*** 0.352 0.128 (0.217) (0.224) (0.223) Rajshahi 1.641*** 1.141*** 0.390** (0.190) (0.194) (0.197) Sylhet 0.564** 0.467* -0.0835 (0.262) (0.270) (0.276) Attends school -0.263 -14.93 -1.508 (1.439) (565.3) (1.281) Remittances -0.129 -0.00592 -0.178 (0.214) (0.215) (0.218) Female -1.903*** -2.802*** -1.019*** (0.376) (0.444) (0.379) Remitt x Female -1.534*** -1.583*** -2.358*** (0.338) (0.506) (0.380) Married 1.295*** 1.471*** 1.235*** (0.355) (0.375) (0.355) Married x Female -2.703*** -1.877*** -1.897*** (0.467) (0.590) (0.464) Other member -0.395*** -0.492*** -0.363*** employed (0.0730) (0.0760) (0.0736) Number of children 0.114 0.0960 -0.115 (0.0697) (0.0713) (0.0722) Constant 1.710* -3.336*** -2.206** (0.915) (0.976) (0.943) 7862 7862 7862 Observations 0.259 0.259 0.259 Pseudo R2 Notes: (1) Self-employed in Non-agriculture. Dhaka is the base Non-Employed is the base category region. Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1. Source: Own estimates based on HIES 2000, 2005, and 2010 . 186 Table A3-14: Multinomial Logit on Occupational Choice – Other members Individuals between 15 and 64 years old 2000 2005 Self- Self- Daily Daily employed Salaried employed Salaried workers workers (1) (1) Primary & Lower -0.964*** -0.0334 0.259*** -0.872*** 0.0448 0.514*** secondary (0.0977) (0.117) (0.0965) (0.0790) (0.0942) (0.0823) Higher secondary & -2.894*** 0.0292 1.227*** -2.649*** 0.154 1.638*** Tertiary (0.358) (0.197) (0.155) (0.327) (0.154) (0.122) Age 0.180*** 0.231*** 0.190*** 0.112*** 0.165*** 0.128*** (0.0239) (0.0311) (0.0255) (0.0211) (0.0241) (0.0211) Age squared -0.00286*** -0.00345*** -0.00308*** -0.00193*** -0.00239*** -0.00212*** (0.000345) (0.000454) (0.000381) (0.000305) (0.000347) (0.000310) Urban -0.172 0.623*** 0.781*** -0.143 0.564*** 0.731*** (0.108) (0.112) (0.0884) (0.0902) (0.0860) (0.0710) Barisal 0.219 -0.117 -0.284 -0.0570 -0.655*** -0.405*** (0.179) (0.206) (0.174) (0.164) (0.179) (0.139) Chittagong 0.107 -0.0791 0.0725 -0.0329 -0.702*** -0.295*** (0.119) (0.136) (0.103) (0.105) (0.112) (0.0853) Khulna 0.504*** 0.155 -0.540*** 0.582*** -0.462*** -0.635*** (0.141) (0.166) (0.155) (0.117) (0.137) (0.119) Rajshahi 1.508*** 0.433*** -0.141 1.024*** 0.0629 -0.594*** (0.109) (0.142) (0.121) (0.0956) (0.105) (0.1000) Sylhet 0.414** 0.0886 0.193 0.358** -0.0339 -0.104 (0.171) (0.222) (0.168) (0.145) (0.156) (0.132) Attends school -3.253*** -3.346*** -3.866*** -4.475*** -3.393*** -4.542*** (0.281) (0.287) (0.228) (0.375) (0.213) (0.220) Remittances -0.319** -0.440*** -0.367*** -0.361*** -0.230** -0.0920 (0.127) (0.139) (0.125) (0.100) (0.108) (0.100) Female -3.038*** -3.288*** -1.753*** -3.101*** -3.039*** -1.577*** (0.161) (0.269) (0.136) (0.146) (0.185) (0.110) Remitt x Female 0.143 -0.0907 0.0784 -0.112 0.0267 -0.602*** (0.201) (0.329) (0.189) (0.178) (0.209) (0.162) Married 0.822*** 1.240*** 0.932*** 0.254** 0.457*** 0.282** (0.148) (0.159) (0.149) (0.115) (0.121) (0.113) Married x Female -2.151*** -2.283*** -2.376*** -1.415*** -1.425*** -1.622*** (0.202) (0.309) (0.189) (0.174) (0.211) (0.145) Employed HH head -0.413*** -0.412*** -0.331*** 0.199 -0.169 0.0100 (0.121) (0.135) (0.112) (0.132) (0.124) (0.107) HH Head with primary & -0.952*** -0.244** -0.379*** -0.783*** 0.0843 -0.198*** Lower secondary (0.104) (0.115) (0.0948) (0.0863) (0.0877) (0.0751) HH Head with Higher -1.844*** -0.492** -0.465*** -2.225*** -0.0434 -0.155 secondary & Tertiary (0.314) (0.204) (0.156) (0.363) (0.154) (0.120) Constant -1.512*** -3.638*** -2.838*** -0.866*** -2.648*** -2.277*** (0.364) (0.475) (0.387) (0.332) (0.379) (0.329) Observations 14058 14058 14058 18883 18883 18883 Pseudo R2 0.362 0.362 0.362 0.350 0.350 0.350 Notes: (1) Self-employed in Non-agriculture. Dhaka is the base Non-Employed is the base category region. Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1. Source: Own estimates based on HIES 2000, 2005, and 2010 . 187 Table A3-14. Cont. Individuals between 15 and 64 years old 2010 Self- Daily employed Salaried workers (1) Primary & Lower -0.802*** 0.321*** 0.240*** secondary (0.0739) (0.0986) (0.0704) Higher secondary & -2.253*** 0.114 1.460*** Tertiary (0.264) (0.171) (0.111) Age 0.154*** 0.336*** 0.183*** (0.0207) (0.0281) (0.0194) Age squared -0.00243*** -0.00459*** -0.00303*** (0.000296) (0.000407) (0.000285) Urban -0.134 0.358*** 0.966*** (0.0826) (0.0907) (0.0617) Barisal 0.0287 -0.406** -0.677*** (0.157) (0.183) (0.136) Chittagong 0.189* -0.704*** -0.429*** (0.100) (0.124) (0.0766) Khulna 0.902*** 0.219* -0.506*** (0.109) (0.128) (0.0996) Rajshahi 0.976*** 0.0781 -0.748*** (0.0901) (0.108) (0.0847) Sylhet 0.791*** 0.0883 -0.629*** (0.130) (0.163) (0.132) Attends school -3.964*** -3.359*** -3.996*** (0.251) (0.239) (0.151) Remittances -0.516*** -0.295** -0.394*** (0.110) (0.123) (0.106) Female -3.197*** -3.580*** -1.563*** (0.146) (0.256) (0.0973) Remitt x Female 0.0908 0.140 -0.352** (0.202) (0.251) (0.166) Married 0.817*** 0.794*** 0.639*** (0.116) (0.128) (0.114) Married x Female -1.856*** -1.477*** -2.115*** (0.176) (0.276) (0.138) Employed HH head 0.170 -0.0691 0.0817 (0.123) (0.135) (0.0982) HH Head with primary & 1.414*** 0.171 0.213** Lower secondary (0.240) (0.160) (0.106) HH Head with Higher 0.630*** 0.0284 -0.0465 secondary & Tertiary (0.244) (0.156) (0.100) Constant -3.080*** -5.940*** -2.808*** (0.414) (0.482) (0.329) Observations 21715 21715 21715 Pseudo R2 0.359 0.359 0.359 Notes: (1) Self-employed in Non-agriculture. Dhaka is the base Non-Employed is the base category region. Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 . Source: Own estimates based on HIES 2000, 2005, and 2010 . 188 Chapter 4 Table A4-1: Summary Statistics for Employment-Related Variables in Bangladesh Employment Share Hours Worked Education in Years 2003 2005 2010 Pct 2003 2005 2010 Pct 2003 2005 2010 Pct Activity (% of Working Population) Salaried Employment 8.02 10.23 10.52 0.55 47.43 52.86 52.71 -0.06 7.49 7.73 7.26 -1.26 Self-Employment 23.98 23.28 22.85 -0.37 42.29 51.09 51.24 0.06 3.71 4.39 3.98 -1.92 Day-Labor 11.31 10.53 11.11 1.08 42.55 51.01 54.14 1.20 2.30 2.45 2.56 0.87 Sector Public Sector 6.16 4.61 3.30 -6.47 47.25 48.45 47.22 -0.51 7.62 11.15 10.60 -1.02 Private Sector 93.84 95.39 96.70 0.27 40.70 40.99 41.59 0.29 3.52 3.60 3.65 0.28 Industry Agriculture 51.34 47.58 46.59 -0.42 36.15 37.76 41.37 1.84 3.12 3.33 3.81 2.75 Manufacturing 13.98 15.08 18.19 3.81 44.40 53.25 50.57 -1.03 4.04 4.56 4.27 -1.29 Services 34.68 37.34 35.23 -1.16 45.89 53.23 50.22 -1.15 5.42 6.09 5.40 -2.38 Note: Pct indicates the variable’s annualized percentage change. Source: LFS 2003, 2005, and 2010 data. 189 Table A4-2: Differences Between Eastern and Western Bangladesh East West 2003 2005 2010 Pct 2003 2005 2010 Pct % of Total Population Residing 54.46 59.71 57.49 -0.76 45.54 40.29 42.51 1.08 54.75 56.88 58.82 0.67 58.08 59.59 62.55 0.97 Working Age % Urban % 23.95 28.23 28.70 0.33 22.72 16.80 15.00 -2.24 Employment % 53.92 56.48 55.14 -0.48 54.33 58.01 59.80 0.61 Proportion Working in Agriculture 49.19 44.39 42.00 -1.10 53.74 51.99 51.94 -0.02 Manufacturing 15.55 17.09 20.90 4.11 12.24 12.31 15.01 4.05 Services 35.26 38.52 37.10 -0.75 34.02 35.70 33.05 -1.53 Proportion Active as Salaried 15.80 19.90 23.18 3.10 12.37 13.34 11.53 -2.88 Self-Employed 41.95 37.70 35.63 -1.12 43.01 41.00 42.66 0.80 Day-Laborer 18.28 15.36 17.23 2.33 21.97 20.92 20.85 -0.07 Non-Salaried 19.00 21.83 20.09 -1.65 18.51 21.43 22.94 1.37 Note: Pct indicates the variable’s annualized growth between 2005 and 2010. Western Bangladesh is comprised of Barisal, Khulna, Rajshahi, and Rangpur Divisions, while eastern Bangladesh is comprised of Dhaka, Chittagong, and Sylhet Divisions. Source: LFS 2003, 2005, and 2010. 190 Table A4-3: Labor Force Participation of Women 2003 2005 2010 Coeff. Coeff. Coeff. (t-stat) (t-stat) (t-stat) Between 5 to 10 Years of Schooling -0.1269** -0.1745** 0.1231** (-7.37) (-9.69) (8.32) Above 10 Years of Schooling 0.0913** 0.0147 0.0975** (3.85) (0.58) (4.48) Experience -0.0337** 0.0410** 0.0660** (-16.23) (17.88) (31.44) Experience^2 0.0002** -0.0010** -0.0020** (5.39) (-19.53) (-41.09) Barisal Dummy -0.0336 0.1417** -0.0814** (-1.31) -5.62 (-3.96) Chittagong Dummy 0.0550** -0.0513** -0.1668** (2.93) (-2.50) (-9.22) Khulna Dummy -0.0706** -0.1258** -0.003 (-3.46) (-5.93) (-0.16) Rajshahi Dummy 0.0318* -0.1170** 0.1178** (1.84) (-6.51) (6.92) Sylhet Dummy 0.1190** 0.1282** -0.1143** (4.17) (4.57) (-5.73) Rural Dummy -0.0446** 0.0887** 0.1049** (-3.38) (5.45) (7.00) Total Child Below Age of 6 -0.0822** -0.0127 -0.0523** (-11.67) (-1.47) (-7.09) Training Dummy 0.6569** 0.5675** 0.4262** (16.72) (13.01) (11.57) Married Dummy -0.4537** -0.3565** -0.2777** (-24.35) (-18.09) (-14.81) Constant 0.3987** -0.4715** -0.3812** (15.13) (-14.88) (-14.65) Note: The dependent variable is binary and equals 1 if a working-age woman is in the labor force and 0 otherwise. Source: LFS 2003, 2005, 2010. 191 Table A4-4: Heckman Mincer Equation for Women 2000 2005 2010 Coeff. Coeff. Coeff. (t-stat) (t-stat) (t-stat) Wage Equation Between 5 to 10 Years of Schooling -0.230 -0.128 0.148 (-1.57) (-1.09) (1.50) Above 10 Years of Schooling 0.908** 0.760** 0.632** -4.91 -5.74 -4.93 Experience 0.103** 0.051** 0.148** (6.17) (4.03) (11.62) Experience^2 -0.002** -0.001** -0.002** (-5.31) (-2.93) (-9.47) Barisal Dummy -0.423 -0.500** -0.09 (-1.34) (-3.24) (-0.60) Chittagong Dummy -0.779** (0.00) 0.236** (-3.17) (-0.02) -2.17 Khulna Dummy (0.20) (0.18) (0.10) (-0.72) (-1.36) (-0.78) Rajshahi Dummy 0.21 -0.291** (0.09) -1.26 (-2.68) (-0.82) Sylhet Dummy 0.08 (0.08) (0.05) (0.26) (-0.55) (-0.33) Rural Dummy -0.567** -0.276** -0.222** (-4.15) (-2.93) (-2.32) Agriculture Dummy 2.116** 0.17 2.116** (11.69) (1.02) (13.18) Manufacturing Dummy 1.954** 0.420** 1.288** (11.84) (3.07) (11.73) Salary Employed Dummy 3.155** 1.216** 2.555** (17.92) (8.81) (16.82) Day-Laborer Dummy 2.038** 0.659** 1.476** (11.44) (6.62) (15.34) Public Sector Dummy 1.080** 0.664** 0.757** (3.69) (6.73) (4.91) Constant -1.670** 3.560** 0.077 (-3.73) -11.51 -0.25 Mills Ratio 2.406 1.182 1.608 (standard error) (0.22) (0.06) (0.07) 192 Table A4-4. Cont. Selection Equation Between 5 to 10 Years of Schooling -0.133** -0.156** -0.138** (-3.07) (-3.63) (-3.81) Above 10 Years of Schooling 0.179** 0.172** 0.097** (3.24) (3.53) -2.36 Experience 0.023** 0.010** 0.015** (4.60) (1.97) (3.36) Experience^2 -0.001** 0.00 -0.000** (-4.56) (-1.54) (-3.60) Barisal Dummy -0.148** -0.333** -0.199** (-2.03) (-5.00) (-3.57) Chittagong Dummy 0.718** -0.018 0.143** (16.42) (-0.40) -3.61 Khulna Dummy -0.030 -0.183** (0.06) (-0.49) (-3.52) (-1.45) Rajshahi Dummy 0.284** 0.004 -0.05 (6.32) (0.09) (-1.34) Sylhet Dummy 0.02 (0.07) 0.184** (0.27) (-1.01) (3.63) Rural Dummy -0.260** -0.351** -0.227** (-7.05) (-9.76) (-7.15) Total Child Below Age of 6 -0.054** -0.201** -0.213** (-2.72) (-9.56) (-10.55) Total Family Income -0.239** -0.706** -0.784** (-3.90) (-8.74) (-9.96) Total Family Income^2 0.008* 0.033** 0.035** (1.84) (5.93) -6.8 Constant 0.22 2.745** 3.229** (0.90) (8.78) (10.42) Note; The reference group is: Dhaka, services sector, less than five years of education, and self-employed. Source: HIES 2000, 2005, and 2010. 193 Table A4-5: Summary Statistics of Firms Variable All Firms Firms That Invested Mean Std. Err. Mean Std. Err. Ln((investment+1)/employee) 2.305 0.154 8.282 0.067 Proportion of Land Owned by Firm 0.478 0.012 0.502 0.018 Using email to communicate (1 – yes, 0 – no) 0.397 0.013 0.576 0.018 Internet (1 – yes, 0 – no) 0.018 0.003 0.038 0.007 Proportion of workers using Computer 0.012 0.002 0.020 0.003 Security a threat (1 – 4 scale) 0.834 0.028 0.975 0.042 Number of Research Staff 0.974 0.067 1.489 0.106 Hours of Power outages in a day 1.117 0.010 1.108 0.015 Hours of water shortage in a day 0.206 0.022 0.255 0.036 Number of Generators used 0.552 0.028 0.837 0.050 Owner also a Top Manager (1 – yes, 0 – no) 0.554 0.013 0.675 0.017 Unpredictability of Law (1 - 4 Scale) 2.107 0.026 2.013 0.035 Bribe/extortion paid as a proportion of sales (100%) 0.142 0.012 0.184 0.017 Percentage of sales lost due to spoilage on the road (100%) 0.146 0.022 0.097 0.021 Problems with transportation 1.051 0.025 1.200 0.037 Politics a Problem (1 – 4 scale) 2.982 0.024 2.995 0.033 In-house training for workers 0.109 0.008 0.167 0.014 Education level of workers 3.849 0.084 4.626 0.117 Export Dummy 0.189 0.010 0.289 0.016 Observations 1504 759 Weighted Observations 26,839 12,891 Note: Weighted averages. For variables on a 1 to 4 scale, 1 indicates no problem, while 4 indicates highest severity of problem. Source: Enterprise Survey 2006. 194 Table A4-6: Results of the Probit Regression Probit 1 Probit 2 Proportion of total land owned by company -0.002 -0.002** (-1.553) (-2.336) Research Staff 0.024** 0.026** (2.295) (2.212) Using email to Communicate 0.432** 0.467** (10.529) (21.673) Percentage of Staff using Computer 0.014** 0.013** (4.966) (10.830) Internet facilities 2.430** 2.451** (4.538) (4.748) Education level of Workers 0.013 0.017 (0.483) (0.672) In-house training for Workers 0.194** 0.209** (2.967) (2.490) Top Manager also an Owner 0.363** 0.298** (2.558) (3.955) Number of Generators Used 0.211** 0.219** (10.381) (13.738) Export Dummy 0.025 0.056 (0.248) (0.741) Hours of power outage in a day faced by firm -0.075 -0.105** (-1.154) (-2.722) Proportion of sales lost in transit -0.112** -0.107** (-10.25) (-7.687) Problem with Transportation 0.164** 0.163** (9.864) (9.274) Unpredictability of Law -0.036** -0.028 (-2.061) (-1.090) Politics A Problem -0.098** -0.089** (-3.281) (-3.079) Regional Dummies No Yes Constant -0.262** -0.238** (-4.731) (-5.007) Note: The Dependent Variable equals 1 if the firm invested and 0 otherwise. Standard errors were clustered by divisions where the firm is located. T-statistics are in parenthesis. * indicate significance at 10 percent, ** indicate significance at 5 percent. 195 Table A4-7: Heckman Second Stage Regression Results (MLE) 1 2 Proportion of Land Owned by the Firm 0.007** 0.008** (22.671) (17.600) Using email to communicate 0.024 (0.070) (0.174) (-0.506) Internet 0.962** 0.973** (21.146) (23.769) Proportion of workers using Computer 0.017** (13.309) Security a threat -0.297** -0.317** (-14.595) (-95.621) Hours of Power outages in a day 0.665** 0.693** (14.365) (11.209) Hours of water shortage in a day -0.157** -0.163** (-3.394) (-6.432) Number of Generators used -0.050** -0.057** (-6.483) (-5.261) Unpredictability of Law (1 - 4 Scale) -0.072** -0.038* (-2.816) (-1.905) Ln(Bribe/extortion paid as a proportion of -0.074** -0.086** sales) (-2.855) (-2.244) Problems with transportation -0.189** -0.210** (-3.479) (-3.364) In-house training for workers 0.433** 0.474** (13.413) (12.367) Education level of workers 0.073** 0.097** (8.002) (12.994) Export Dummy -0.348** -0.348** (-13.443) (-11.459) Regional Dummies No No Other Regional Variables Used as Controls No Yes Constant 5.929** 5.671** (33.876) (15.162) Rho (correlation between error terms of -0.743** -0.997** first and second stage regressions) (-14.557) (-15.250) Ln(Sigma) 0.581** 0.655** -36.722 -40.609 Note: The Dependent Variable is the natural log of investment to workers in a firm. Standard errors were clustered by divisions where the firm is located. T-statistics are in parenthesis. * indicate significance at 10 percent, ** indicate significance at 5 percent. The regressions also controlled for variables ‘politics a problem’ (on a 1 to 4 scale), ‘percentage of sales lost due to spoilage on the road,’ and ‘number of research staff’. 196 Table A4-8: Income Differences between East and West East West 2000 2005 2010 Pct 2000 2005 2010 Pct Average Labor 3202.21 3575.15 4236.76 3.45 2880.80 3092.77 4028.368 5.43 Income % Labor Income 83.21 84.08 90.38 1.46 88.28 87.26 95.00 1.71 out of Total Income Poverty Rate 48.1% 42.5% 32.4% -5.3 53.0% 44.3% 33.9% -5.2 (Upper Line) Income Source Agriculture 30.50 24.35 36.61 8.50 50.84 46.38 58.01 4.58 Manufacturing 29.69 28.22 23.33 -3.73 19.09 20.26 13.29 -8.09 Services 39.81 47.43 40.05 -3.33 30.07 33.37 28.70 -2.97 Income Source Salaried 30.28 35.88 30.74 -3.04 17.73 17.71 16.07 -1.92 Self 50.12 47.54 52.25 1.91 58.82 57.32 60.65 1.14 Day 19.60 16.58 17.01 0.51 23.45 24.97 23.28 -1.39 Note: Pct is the annualized percentage change of variable. Western Bangladesh is comprised of Barisal, Khulna, Rajshahi, and Rangpur Divisions, while eastern Bangladesh is comprised of Dhaka, Chittagong, and Sylhet divisions. Source: HIES 2000, 2005, and 2010. 197 Table A4-9: Male Wage Regression 2000 2005 2010 Coeff. Coeff. Coeff. (t-stat) (t-stat) (t-stat) Between 5 to 10 Years of Schooling 0.211** 0.213** 0.188** (3.84) (4.27) (4.09) Above 10 Years of Schooling 0.187** 0.227** 0.179** (2.73) (3.65) (3.08) Experience 0.194** 0.165** 0.181** (30.99) (28.91) (30.59) Experience^2 -0.004** -0.003** -0.003** (-26.98) (-24.05) (-25.85) Barisal Dummy -0.201** -0.112 -0.068 (-2.35) (-1.40) (-0.88) Chittagong Dummy -0.318** -0.358** -0.307** (-4.77) (-5.66) (-5.10) Khulna Dummy 0.132* -0.07 -0.013 (1.75) (-1.04) (-0.23) Rajshahi Dummy 0.076 0.223** 0.079 (1.37) (4.37) (1.62) Sylhet Dummy -0.228** 0.328** 0.174** (-2.42) (3.88) (2.32) Rural Dummy (0.03) (0.02) 0.03 (-0.50) (-0.38) (0.58) Agriculture Dummy 1.722** 1.559** 1.485** (25.67) (28.16) (30.69) Manufacturing Dummy 1.839** 1.574** 1.125** (32.81) (32.61) (27.96) Salary Employed Dummy 1.919** 2.013** 1.643** (31.29) (34.24) (29.75) Day Laborer Dummy 0.892** 1.099** 0.867** (21.62) (28.04) (22.68) Public Sector Dummy 0.603** 0.475** 0.479** (8.21) (7.59) (8.39) Constant 3.298** 3.540** 3.779** (31.82) (33.34) (34.39) Note: The reference group is: Dhaka, services sector, less than five years of education, and self-employed. Source: HIES 2000, 2005, and 2010. 198 Table A4-10: Returns to Education 2000 2005 2010 Additional year (percent) Male Female Male Female Male Female Average 0.0175** 0.0415** 0.0195** 0.0889** 0.0110** 0.0474** of which in Primary 0.3724** -0.4104** 0.2361** -0.2284 0.2845** 0.0911 Incomplete secondary 0.0338 0.042 0.1643** 0.0289 0.0403 0.1853 Junior secondary (HSC) -0.0527 0.0789 -0.0169 0.5305** -0.0176 -0.0937 Senior secondary (HSC) 0.3836** 1.1648** 0.3978** 1.3109** 0.0688 0.4251 More than 12 Years of Education 0.5598** 1.0591** 0.5239** 1.8890** 0.4363** 1.5902** Daily wage worker 0.0263** 0.0675** 0.0188** 0.0154 0.0108** -0.0063 Non-farm self-employed 0.0656** 0.0978** 0.0543** 0.0447** 0.0411** 0.0093 Salaried workers 0.0624** 0.1129** 0.0630** 0.1027** 0.0491** 0.0804** Public sector 0.0530** 0.2056** 0.0584** 0.0790** 0.0699** 0.1173** Private Sector workers 0.0056 -0.0299** 0.0087 -0.0017 0.0021 -0.0025 Urban workers 0.0193** -0.0421** 0.0295** 0.0379** 0.0248** 0.0339** Rural workers 0.0142** 0.0162* 0.0101* 0.0213** 0.0019 0.0216** Note: Mincer regressions of log-wages on years of education, experience, experience squared, regional dummies, the number of children under age 6, a dummy for marriage, and a constant. Female wage equations were corrected for self-selection using a Heckman selection model in the case of all workers. Source: HIES 2000, 2005, and 2010. 199 Table A4-11: Wage Regression for Men and Women using 2010 data Men Women West East West East Education in Years 0.0154** 0.0059 0.0281 0.0413** (2.78) (0.88) (1.58) (2.88) Experience 0.0873** 0.1172** -0.1916** 0.1622** (7.40) (8.48) (-7.96) (7.27) Experience^2 -0.0016** -0.0024** 0.0043** -0.0030** (-7.23) (-9.28) (8.34) (-6.42) Rural Dummy 0.2529** 0.01 0.4238** -0.8394** (4.46) (0.08) (3.17) (-6.36) Total Child Below Age 6 -0.0029 -0.1001** 0.4079** -0.1231 (-0.08) (-2.44) (3.67) (-1.35) Married Dummy 1.0318** 1.3463** 3.2627** -0.0006 (8.12) (9.76) (14.77) (-0.00) Constant 5.2950** 5.1927** 9.6501** 4.1531** (42.13) (39.79) (29.50) (10.87) Note: The regression controlled for regions. Western Bangladesh is comprised of Barisal, Khulna, Rajshahi, and Rangpur Divisions, while eastern Bangladesh is comprised of Dhaka, Chittagong, and Sylhet Divisions. Men Wage Regression was estimated using OLS, while Heckman selection model was used to estimate Women Wage Equation. The selection equation controlled for education, experience, experience^2, rural dummy, total children below age 6, marriage dummy, acres of land owned by family, total family income, total family income squared, and constant. Source: using Data from LFS 2010. 200 Table A4-12: Decomposing the Gender Wage Gap Gap due to Differences in Gap due to The Part of the Gap Income Gap Characteristics Differences in Explained by (difference between Males Characteristics Differences in between Unexplained who were between Females Characteristics of male and Part matched and who were matched Males and Females Year females) of the Gap unmatched and unmatched who were Matched 2000 59.0 43.2 8.5 22.0 26.3 2005 16.8 95.3 5.7 1.9 -2.9 2010 17.5 73.5 5.6 12.1 8.8 Note: The method outlined in Ñopo (2008) was used to decompose the gender wage gap. Males and females were matched by labor force participation, age, education, region and area of residence, sector, and employment type before decomposing their income. Source: HIES 2000, 2005, and 2010. Table A4-13: Decomposing the Public/Private Wage Gap Gap due to Gap due to Differences in The Part of the Differences in Characteristics Gap Explained by Income Gap Characteristics between Private Differences in (difference between Public Sector Characteristics of between public Unexplained Sector Employees Employees who Public and Private and private Part who were matched were matched Sector Employees Year sector) of the Gap and unmatched and unmatched who were Matched 2000 19.6 42.7 0.5 40.3 16.5 2005 13.9 53.1 -0.3 36.1 11.1 2010 13.9 50.0 0.1 39.4 10.5 Note: The method outlined in Ñopo (2008) was used to decompose the public-private sector wage gap. Workers were matched by labor force participation, age, gender, and region of residence before decomposing their income. Source: HIES 2000, 2005, and 2010. 201 Chapter 7 Table A7-1: Rural Nominal and Real Wages by Gender and Season (HIES) Panel A: Nominal Wages Peak Season Lean season Year Male Female Male Female 2010 194.33 141.62 154.92 113.08 2005 89.09 56.57 70.39 47.43 2000 70.29 48.20 55.95 39.96 1995 53.38 35.15 42.09 29.54 Panel B: Real Wages (CPI) Peak Season Lean season Year Male Female Male Female 2010 193.55 141.03 154.16 112.57 2005 133.66 84.87 105.52 71.13 2000 134.01 92.25 106.60 76.31 1995 128.57 84.72 101.49 71.20 Panel C: Real Wages (BNPI) Peak Season Lean season Year Male Female Male Female 2010 193.55 141.03 154.16 112.57 2005 169.20 107.43 133.58 90.04 2000 164.66 113.35 130.98 93.77 Panel D: Real Wages (Rice Price) Peak Season Lean season Year Male Female Male Female 2010 202.79 147.89 161.74 118.17 2005 170.83 108.50 134.87 90.91 2000 182.29 125.01 145.13 103.60 1995 137.95 91.11 108.72 76.41 Note: Real wages are in 2010 prices, weighted and spatially deflated. Nominal wages are from HIES community surveys. Since BNPI does not exist for 1995, real wage calculations do not include the year 1995. Source: Background paper prepared by Zhang et al. (2012) for this poverty assessment report. 202 Table A7-2: Urban Nominal and Real Wages by Gender (HIES) Panel A: Nominal Wage Year Male Female National 2010 270.11 232.32 260.26 2005 161.65 149.75 157.57 2000 112.85 71.51 107.27 Panel B: Real Wage (CPI) Year Male Female National 2010 253.43 223.21 245.11 2005 230.33 216.51 225.24 2000 203.23 129.26 194.07 Panel C: Real Wage (BNPI) Year Male Female National 2010 253.43 223.21 245.11 2005 291.57 274.08 285.12 2000 249.72 158.84 238.46 Note: Real wages are in 2010 prices, weighted and spatially deflated. Nominal wages are obtained from HIES household surveys. Urban daily wage is calculated as (annualized daily wage + annual salary)/number of days worked in year. Wages are from wage earners 15 years and older. Source: Background paper prepared by Zhang et al. (2012) for this poverty assessment report. 203 Chapter 8 Table A8-1: Probit Regressions Reporting Marginal Effects of the Probability of Receiving a Transfer as a Function of Household Characteristics SN (2005) Any SN Head = male -0.22a -0.56 Head: age 0.03a 0.04a Head: age2 0b 0a Chronically ill 0.14a HH size -0.07a -0.01 Dependency 0.13 0.52a Education (omitted: none) Primary -0.05 0.06 Secondary -0.16a -0.07 Higher -0.27 -0.27b Main work is in agriculture 0.13a Not working /daily laborer 0.32a 0.11b Log (real per capita cons.) -0.64a 0.13a Urban -0.36a -0.36a Spouse = no education -0.41a Electricity -0.22a 0.19a Tube well 0.09* Microcredit 0.34a Remittances -0.09 -0.03 Amount of land 0 Division (omitted: Dhaka) Barisal -0.33b 0.33a Chittagong -0.19 0.03 Khulna -0.49a 0.42a Rajshahi -0.57a -0.04 Sylhet 0.1 -0.11 Constant 3.24a 0.67 Num. of obs. 8921 7932 Wald 2 418 490 Prob > χ2 0 0 Pseudo R2 13.8 12.6 204 Chapter 10 Table A10-1: Regional Differences in Welfare and Income Measures: Convergence or Divergence? LRPCE Hours Worked 2000 2005 2010 2000 2005 2010 Urban-Integrated 0.317 0.281 0.313 421.024 670.513 747.114 (7.09)*** (5.65)*** (7.03)*** (8.30)*** (13.51)*** (14.77)*** Urban-Less Integrated 0.182 -0.037 0.094 425.954 510.035 448.896 (2.64)*** (0.93) (2.75)*** (6.88)*** (9.04)*** (9.15)*** Rural-Less Integrated -0.043 -0.139 -0.022 93.326 -36.339 -69.679 (1.41) (5.96)*** (0.92) (1.56) (0.63) (1.45) intercept 6.505 6.694 6.920 2,457.401 2,191.041 2,098.282 (287.36)*** (386.30)*** (391.50)*** (56.73)*** (52.70)*** (56.20)*** No of observations 7440 10080 12240 11748 12951 16847 Log(daily wage) Log (salary) Urban-Integrated 0.283 0.205 0.070 0.274 0.360 0.233 (6.18)*** (4.27)*** (1.49) (3.75)*** (4.48)*** (2.97)*** Urban-Less Integrated 0.080 -0.008 -0.041 0.324 0.228 0.217 (1.16) (0.17) (1.01) (3.50)*** (2.63)*** (3.29)*** Rural-Less Integrated -0.164 -0.212 -0.222 0.019 -0.012 0.022 (4.14)*** (6.55)*** (6.19)*** (0.21) (0.14) (0.30) intercept 4.081 4.166 4.384 4.122 4.229 4.332 (143.96)*** (169.25)*** (142.60)*** (64.31)*** (60.61)*** (83.12)*** No of observations 3846 4844 5649 2387 3050 3910 Note: Robust t-statistics in parentheses. * significant at 10%; ** significant at 5%; *** significant at 1% Source: Background paper prepared by Shilpi (2012). Data source: HIES 2000, 2005, and 2010. 205 Table A10-2: Coverage Rate, Damage Index, HOI, and Circumstance Impact of (Upper) Rural and Urban Poverty (2010) Health- Coverage Rate (%) HOI (%) Damage Index (%) Circumstance Impact Poverty Contribution Professional Overall Rural Urban Overall Rural Urban Overall Rural Urban Overall Rural Urban Overall Rural Urban Bangladesh 45 42 56 41 39 51 8.9 8.2 9.7 5.4 5.4 4.3 61% 65% 44% IR 43 40 56 39 36 48 10.9 9.8 12.9 5.6 4.1 6.8 51% 42% 52% LIR 48 46 57 44 42 52 7.9 8.6 8.5 5.1 6.6 1.1 65% 76% 13% Enrollment of Coverage Rate (%) HOI (%) Damage Index (%) Circumstance Impact Poverty Contribution 6 - 10 year olds Overall Rural Urban Overall Rural Urban Overall Rural Urban Overall Rural Urban Overall Rural Urban Bangladesh 83 82 87 80 80 84 3.7 3.4 4.3 1.6 1.4 1.7 42% 41% 40% IR 82 80 86 78 77 82 4.4 3.8 5.4 1.8 1.5 2.1 41% 39% 39% LIR 87 86 90 84 83 88 3.1 3.2 2.4 1.4 1.4 1.0 45% 45% 39% Enrollment of Coverage Rate (%) HOI (%) Damage Index (%) Circumstance Impact Poverty Contribution 11 - 15 year olds Overall Rural Urban Overall Rural Urban Overall Rural Urban Overall Rural Urban Overall Rural Urban Bangladesh 83 83 83 79 79 80 4.2 4.7 3.8 2.7 2.4 3.3 65% 51% 89% IR 79 78 82 75 73 79 5.8 6.9 3.8 3.6 3.4 3.4 62% 50% 89% LIR 88 89 85 86 87 82 2.3 2.3 4.4 1.7 1.4 3.1 74% 59% 71% Source: HIES 2005 and 2010. 206 Table A10-3: Coverage Rate, Damage Index, HOI, and Circumstance Impact of (Upper) Rural and Urban Poverty (2010) Electricity Coverage Rate (%) HOI (%) Damage Index (%) Circumstance Impact Poverty Contribution Facilities 6 - 10 Overall Rural Urban Overall Rural Urban Overall Rural Urban Overall Rural Urban Overall Rural Urban Bangladesh 49 39 87 41 32 81 16.8 18.1 6.3 8.5 9.6 3.0 51% 53% 48% IR 55 43 92 47 36 88 14.2 15.5 4.2 6.5 7.7 1.2 46% 50% 28% LIR 39 32 72 31 25 63 20.4 22.2 12.8 11.0 13.3 7.2 54% 60% 56% Electricity Coverage Rate (%) HOI (%) Damage Index (%) Circumstance Impact Poverty Contribution Facilities 11 - 15 Overall Rural Urban Overall Rural Urban Overall Rural Urban Overall Rural Urban Overall Rural Urban Bangladesh 56 44 90 48 37 85 14.2 15.3 5.2 6.0 7.6 2.5 43% 50% 48% IR 61 47 93 53 40 90 13.8 15.4 3.3 5.0 5.8 1.3 36% 37% 40% LIR 47 40 81 40 33 72 16.0 17.3 10.3 8.3 11.5 4.3 52% 67% 42% Sanitation Coverage Rate (%) HOI (%) Damage Index (%) Circumstance Impact Poverty Contribution Facilities 0 - 5 Overall Rural Urban Overall Rural Urban Overall Rural Urban Overall Rural Urban Overall Rural Urban Bangladesh 46 40 68 39 33 62 15.2 17.5 9.4 5.8 6.6 1.1 38% 38% 12% IR 46 39 71 39 32 64 15.0 19.0 9.2 4.6 5.5 0.1 31% 29% 1% LIR 45 41 64 37 34 55 16.4 16.8 13.4 7.2 8.2 4.2 44% 49% 32% Source: HIES 2005 and 2010. 207 Table A10-4: Coverage Rate, Damage Index, HOI, and Circumstance Impact of (Upper) Rural and Urban Poverty (2005) Health Coverage Rate (%) HOI (%) Damage Index (%) Poverty Impact Poverty Contribution Professional Overall Rural Urban Overall Rural Urban Overall Rural Urban Overall Rural Urban Overall Rural Urban Bangladesh 29 27 34 27 26 29 7.1 5.3 15.1 2.2 1.9 3.2 32% 35% 21% IR 29 27 34 25 24 29 11.7 11.1 15.7 1.7 1.4 1.9 14% 13% 12% LIR 28 28 33 26 26 28 9.3 9.4 16.1 2.2 1.2 5.9 24% 13% 36% Enrollment of Coverage Rate (%) HOI (%) Damage Index (%) Poverty Impact Poverty Contribution 6 - 10 year olds Overall Rural Urban Overall Rural Urban Overall Rural Urban Overall Rural Urban Overall Rural Urban Bangladesh 78 78 81 74 74 77 5.1 4.9 5.8 2.6 2.5 3.0 50% 51% 51% IR 76 75 80 71 70 74 6.3 6.0 6.9 3.5 3.4 3.8 55% 56% 55% LIR 82 82 85 79 78 81 4.3 4.2 4.6 2.1 2.1 1.7 48% 51% 38% Enrollment of Coverage Rate (%) HOI (%) Damage Index (%) Poverty Impact Poverty Contribution 11 - 15 year olds Overall Rural Urban Overall Rural Urban Overall Rural Urban Overall Rural Urban Overall Rural Urban Bangladesh 69 69 71 64 63 66 8.2 8.6 7.7 3.5 3.4 3.8 42% 39% 49% IR 68 67 71 62 60 66 8.8 9.4 7.6 4.2 4.1 4.0 48% 44% 53% LIR 71 72 71 66 66 64 8.2 8.1 9.7 2.8 2.8 3.0 35% 34% 31% Source: HIES 2005 and 2010. 208 Table A10-5: Coverage Rate, Damage Index, HOI, and Circumstance Impact of (Upper) Rural and Urban Poverty (2005) Electricity Coverage Rate (%) HOI (%) Damage Index (%) Poverty Impact Poverty Contribution Facilities 6 - 10 Overall Rural Urban Overall Rural Urban Overall Rural Urban Overall Rural Urban Overall Rural Urban Bangladesh 36 27 77 27 19 69 26.1 30.5 10.7 13.8 17.7 6.0 53% 58% 56% IR 42 30 85 32 21 79 23.0 29.4 6.6 11.6 16.8 3.4 51% 57% 52% LIR 27 22 58 19 15 47 28.6 30.0 19.1 14.3 16.6 9.4 50% 55% 49% Electricity Coverage Rate (%) HOI (%) Damage Index (%) Poverty Impact Poverty Contribution Facilities 11 - 15 Overall Rural Urban Overall Rural Urban Overall Rural Urban Overall Rural Urban Overall Rural Urban Bangladesh 45 34 82 36 26 75 20.2 23.5 9.0 9.7 12.9 5.3 48% 55% 59% IR 55 41 90 46 33 86 16.7 21.1 4.6 6.9 10.0 2.7 41% 47% 59% LIR 32 27 62 25 20 52 22.8 25.4 16.1 11.4 14.1 8.1 50% 56% 50% Sanitation Coverage Rate (%) HOI (%) Damage Index (%) Poverty Impact Poverty Contribution Facilities 0 - 5 Overall Rural Urban Overall Rural Urban Overall Rural Urban Overall Rural Urban Overall Rural Urban Bangladesh 46 39 76 36 30 68 21.1 23.2 10.8 8.6 9.3 4.4 41% 40% 40% IR 54 45 84 45 37 77 16.4 18.1 8.2 5.6 5.8 2.6 34% 32% 31% LIR 33 29 56 22 19 45 31.6 34.9 18.8 13.7 15.9 7.1 43% 45% 38% Source: HIES 2005 and 2010. 209 Data Annex Chapter 3: Measuring Household Income in Bangladesh – HIES 2010 1. This annex provides a summary of how the household income aggregate and its subcomponents (i.e. labor and non-labor income) were constructed based on the data collected in the Household Income and Expenditure Survey (HIES). Additionally, we overview the adjustments used for spatial and temporal price variations as well as the definition for individual labor market status. We also include a section describing methodological differences between the Bangladesh Bureau of Statistics (BBS) and World Bank (WB) total household income aggregates. A. TOTAL HOUSEHOLD INCOME 2. When adding together the two main sources of family income (labor income and non-labor income, we obtain total income in household h. The first component, labor income, is defined at the individual-level while the second, non-labor income, is defined at the household-level. Formally, [∑ ∑ ] (1) where = labor income of individual i of household h in economic activity j; j = economic activity: self-employed or employer in agriculture or non-agriculture, daily-laborer or salaried in agriculture or non-agriculture; i = individual five years or older; I = indicator function equal to 1 if the individual is engaged in a given activity; n = total number of members of household h; and = non-labor income. 1. Labor income 3. Total family labor income is the sum of income earned in different economic activities by each household member. For each individual, we observe whether or not (s)he participates in a particular economic activity as well as whether or not (s)he receives labor income for that activity. 4. The HIES asks every individual aged older than five years if she/he has performed different labor activities in the year prior to when the survey was collected.134 Each labor activity is divided into agricultural and non-agricultural activities, and each of these economic activities comprises four employment statuses: day-laborer, self-employed, employer, and employee. Labor income is defined as the amount earned or received (in-cash or in-kind) from each activity for a particular employment status over the period of reference (previous year). 5. Following the labor status structure of HIES, we are able to identify individual labor income associated with each economic activity. First, we focus on the self-employed and employers135 and then on day-laborers and employees136 for agriculture and non-agricultural activities. 134 HIES, Section 4: Economic Activities Self and Wage Employment, part A. 135 Section 5 and 7 of HIES 2010. 136 Section 4 part B of HIES 2010. 210 a. Self-employed & employers 6. For the self-employed and for employers, labor income is defined as net revenue of farm business, profit from family business, and incomes from other agriculture business. In the case of revenue and profit, total household income is divided across family members according to the number of hours of work allocated to this activity by each member.137 When no information is available for hours of work, total household income is equally divided among members who work in the family business. We note that all income variables are expressed in Taka (Tk) per month. i. Net-Revenue (Farm activities): 7. Net-Revenue (NR) is calculated as the difference between total Gross Revenue (G) and total Expenditures (E) for each crop and non-crop product.138 We minimally alter the original data and only make adjustments when necessary, for instance, in the presence of missing values in unit prices or outliers; otherwise, we retain the information as reported in the HIES 2010. The methodological steps are outlined below. a. Gross revenue (G): we calculate total production139 revenue by crop and non-crop products: livestock & poultry; livestock products; fish-farming; and forestry. In the presence of missing values for gross revenue but complete information for total production (quantity), we impute the regional median unit price by crop/non-crop product if the number of observations exceeds five. Otherwise, we consider the country median unit price by crop/non-crop product. We also correct for outliers if the unit price is greater or less than two times the median price.140 Note that we only impute prices, not quantities. b. Expenses in inputs (E): we apply the same methodology for inputs as for gross revenue. We impute the regional median by input when the number of observations is greater than five, or in case the number of observations is too low, we consider the country median price unit by input. Additionally, we correct for outliers if the unit price is greater than two times the median price or less than one-half of the median price.141 Note that we only impute prices, not quantities. c. Net revenue (NR): since each individual could earn gross revenue from different activities, and since expenses are not captured by activity, we allocate expenses based on the gross revenue share of crop and non-crop activities for that individual. For instance, suppose an individual has revenues from different products, such as crop, livestock, and fishing. We first calculate total gross revenue from all three activities as well as each activity’s share of the total. Second, we multiply these shares by total expenses to spread the expenses across these activities.142 Finally, the difference between the gross revenue and the expense for each activity equals its net revenue. 137 Note the variable “hours of work” has been adjusted for outliers when necessary. 138 Non-crop products are: livestock and poultry; livestock products (i.e. milk, eggs, meat, etc.); fish-farming and fish capture; and farm forestry. 139 For crops, total production revenue includes: the quantity consumed, sold, stocked, and used by the household for different purposes, such as given to landlords, wages, etc., over the past 12 months. For non-crop products, it includes how much the household consumed and sold over the year. 140 We implement this adjustment in order to be consistent with the previous Poverty Assessment. 141 Idem. 142 We assume that activities which have higher gross revenues also have higher expenditures, given the need for relatively greater amounts of inputs. 211 (2) where G = total gross revenue (all activities); E = total expenditures; Ei = expenditure in activity i; Gi = gross revenue in activity i; and i = crop and non-crop activities. ii. Profit (Non-Agriculture business): 8. We identify three main difficulties associated with the calculation of individual labor income from family business. The first problem is that the Industrial Codes (IDs) are non-exact-matched between two of the HIES modules/questionnaires: Economic Activity143 and Non-Agricultural Enterprises144. To proceed, we assume that IDs from the first questionnaire (i.e. Economic Activity) are correct. Then, from the Economic Activity module, we only keep individuals who report being occupied in a non-agriculture activity and are self-employed or employers. We make few corrections to the Industrial and Occupation codes in order to maintain consistency. Finally, we identify several cases of non-exact-matched observations based on the number of household members involved in each activity and each member’s number of activities. For instance, for members with only one activity, more than 5% of observations do not match between datasets. We assume that the IDs in the Economic Activity dataset are correct. We treat each case separately in order to improve the match between datasets and to reduce the loss of observations. 9. The second problem is that some individuals, before previous corrections, answered the second questionnaire (i.e. Non-Agricultural Enterprise) but did not report themselves as self-employed in a non- agriculture activity in the first questionnaire (i.e. Economic Activity). For these cases, we impute answers into the Economic Activity questionnaire in order to maintain consistency between the two modules. Note that we do not impute hours of work for this activity. 10. The last difficulty arises in the division of total profit between members of households that worked in the family business. To do so, we first calculate total family hours of work per month for each activity. Secondly, we estimate the share of hours per month that each member spent in the family business. Thirdly, we spread the profit among members using the shares calculated in step two. In case of missing hours of work, we divide the profit equally among family members. iii. Other agriculture business 11. We consider sales of agricultural assets and rent from said assets.145 When the price unit is not available, we impute the price for each asset based on the median regional price as long as the number of observations is greater than five. If the number of observations is less than five, we impute the price using the country median price. We correct for outliers if the unit price is greater than two times the median price or less than one-half of the median price. These values are expressed in Tk per month.146 b. Daily laborer & Salaried 143 Section 4 part A, HIES 2010. 144 Section 5, HIES 2010. 145 Section 7, Part E: Agriculture Enterprises, HIES 2010. 146 We implement this adjustment in order to be consistent with the previous Poverty Assessment. 212 12. For the sake of consistency, we apply the same adjustment to the Industrial Codes (ID) as before. We also check for and correct certain inconsistencies in the dataset, such as outliers for hours of work or individuals who report being employed on a daily basis but appear as salaried workers and vice versa, among other minor corrections. Lastly, we express labor income, in-cash or in-kind, in Tk per month. 2. Imputed rent from land & dwelling 13. We consider two sources of imputed rent: land and dwelling rental values. In the first case, the values come from expenses on agricultural inputs in rent (agricultural land), and we impute rents only for individuals who own land. As explained before, we imputed regional rents by area median price, regional median price or the country median price when the number of observations is small. In the second case, we consider the information in the expenditure module, and for owners who do not report, we assign a value based on the median rent for dwellings by district, area, and region. However, this variable was not included as part of the total household income analysis. 3. Non-labor income 14. The second component of total household income is total family non-labor income. This is the sum of various components at the household-level. We use the Other Income module147, which includes rents from other properties, transfers (i.e. international and domestic remittances), social income (such as insurances, charity in-cash and in-kind, etc.), and other non-labor income (i.e. gratuity, retirement, other cash or in-kind). All of these components are defined at the household-level and expressed in Tk per month. Figure D3-2 presents the estimated distribution of remittances receipts by deciles of per-capita income. B. SPATIAL AND TEMPORAL PRICE ADJUSTMENTS 15. Household income estimates must be deflated to capture differences in the needs and prices faced by households. To control for spatial price differences, we use a ratio of the poverty line of Rural Dhaka to the regional poverty line. Then, we multiply income variables by this ratio and express the variables in Rural Dhaka prices.148 16. Temporal adjustment is also necessary in order to make consistent comparisons over time. We take the variation in the nominal upper poverty line across years and re-express all income variables in terms of 2005 Tk. This adjustment yields comparable real income measures, expressed in 2005 Rural Dhaka prices. C. LABOR MARKET STATUS 17. The HIES has three main sections suitable for labor market analysis. A first section is on individual employment status over the previous week; a second section is on the economic activity or activities in which the individual was involved over the last year; and a third section in which the income earned during the past year is recorded. One of many issues highlighted by the previous Poverty Assessment is the inconsistency of responses between modules. For instance, using responses from the 147 Section 8: Other assets and income, HIES 2010. 148 This is a simplistic way to implement this adjustment. The usual approach to controlling for spatial price differences is to use a price index formula that approximates the true cost-of–living index. A common choice is Laspeyre’s index, which calculates the relative cost in each region/area of buying the base region’s basket of goods. In other words, it calculates the price of a fixed bundle of goods rather than the price of a fixed utility level. 213 first module to define individual labor status could create inconsistencies when examining labor income. The main reason is that the variables in the first two modules consider different time horizons; therefore, the possibility remains that individuals that are classified as inactive or unemployed may have positive labor incomes. 18. In order to maintain consistency and comparability over time, we follow the same labor status definitions used in the previous Poverty Assessment. In general, employment is based on the income space: “…a person is classified as employed if any wage or self-employed income is recorded…”149 19. Economic sectors of the various activities undertaken by an individual are ordered according to the number of work hours allocated to each sector. The maximum number of work hours defines the main activity and, in decreasing order, the secondary and tertiary occupations, respectively. 20. The HIES 2010 also presents the same problem with “Other business” or ID sector code 74. We apply the same rules as in the previous Poverty Assessment and spread it among other economic sectors within services. D. THE BBS AND THE WORLD BANK PER-CAPITA INCOME 21. Two main differences exist between the BBS and WB household per-capita income aggregates. First, the BBS income aggregate includes imputed rent and incomes from social safety net programs 150 while the WB income aggregate does not. The WB total household income measure does not include transfers from social safety net programs in order to maintain comparability over the decade. The safety net questionnaire was first included in 2005. However, significant changes occurred between the initial (2005) and final (2010) questionnaires. For instance, one major change, among others, was the observational unit of reference from the household to the individual.151 Second, the WB aggregate incorporates several adjustments as described above (i.e. price and quantity adjustments in the farm and non-farm sectors and for the labor component, and only value adjustments for the non-labor component) in order to maintain consistency between Poverty Assessments over time. 22. Compared to the distribution of the BBS income aggregate, the distribution of the WB household per-capita income aggregate is skewed to the right and has a higher mean and standard deviation, as shown in Figure D3-1: 149 Paci, P. and Sasin, M. (2008). 150 Based on the STATA do-file from BBS. 151 See Methodological Box in Chapter of Safety Nets of Poverty Assessment 2012. 214 Figure D3-1: Density function of Per-capita Income (Bangladesh Bureau of Statistics (BBS) and World Bank (WB) estimations) .0004 .0003 .0002 .0001 0 0 2000 4000 6000 8000 10000 Per capita Income Per capita Income - BBS Per capita Income - WB Source: Own estimation based on HIES 2010 Figure D3-2: Estimated Remittances Receipts by Deciles of Per-capita Income 50.00 45.00 40.00 35.00 30.00 25.00 20.00 15.00 10.00 5.00 0.00 1 2 3 4 5 6 7 8 9 10 Decile of per-capita income 2000 2005 2010 Source: Own estimation based on HIES 2000, 2005, and 2010. 215 Chapter 4: Consistency between HIES and LFS 1. This report uses both the LFS (2003, 2005, and 2010) and HIES (2000, 2005, and 2010). The LFS is used to demonstrate labor force trends and employment changes, and HIES is used to show income/consumption trends among different segments of the population. In order to establish the validity of results obtained from LFS and HIES, the data from the two surveys must be comparable and consistent. 2. Both surveys are nationally representative, and their questionnaires are comparable over time. However, one important difference between the two surveys is the way in which labor force participation is measured. Typically, the labor force is composed of all people who are currently employed as well as those who are unemployed. Differences in definitions of employed and unemployed imply that the labor force will also be measured differently. 3. In terms of employment, the LFS includes non-salaried individuals as employed and, therefore, part of the labor force, but the HIES does not ask if an individual works without pay. More specifically, the LFS has the following categories of economic activity: salaried employment, employer, self- employed, day-laborer, and non-salaried employee. In contrast, the HIES classifies the employed as being day laborers, self-employed, employer, and employee. To the extent that it does not capture unpaid family workers, the HIES may underestimate labor force participation, particularly for women. 4. In addition, the two surveys differ in how they measure unemployment. In order to classify them as unemployed, the LFS asks individuals who are not currently employed if they are currently looking for work. This question is comparable over time for the 2003 and 2005 surveys. However, for the 2010 LFS, the question asks whether or not a person has been looking for work in the past four weeks. In contrast, while the HIES has no comparability issues over time, it has two sections suitable for labor market analysis. The first refers to occupation status, while the second reports economic activity and source of income. The problem between these two modules is their period of reference; while the first section asks about employment in the past seven days, the income section refers to employment over the last 12 months. Therefore, we observe cases in which a person is currently unemployed or inactive but has positive annual labor income. In order to resolve this issue and maintain comparability and consistency with the previous Poverty Assessment, labor force participation is based on the income section, so a person is classified as employed and active if any income is recorded. In this sense, we may overestimate the activity rate of males and underestimate the activity rate of females in the HIES compared to the LFS. 5. Given the aforementioned definitions for employment status, female and male labor force participation in each of the two surveys is documented in Table D4-1. 6. In the LFS, the exclusion of non-salaried employment significantly reduces female labor force participation to 12.7 percent for 2003, 11 percent for 2005, and 15.2 percent for 2010. On the other hand, as expected, the HIES overestimates the proportion of males in the labor force. Table D4-1: Comparing Male and Female Labor Force Participation Between HIES and LFS Year HIES LFS Male Female Male Female 2000 91.32% 15.51% 2003 85.74% 26.53% 2005 91.21% 13.08% 88.47% 29.94% 2010 94.05% 16.80% 81.69% 35.98% 216 7. One predictor of consistency is to compare individual income distributions of the two datasets. However, income data from LFS are not a good indicator for individual and family income, as the incomes are highly inconsistent with corresponding education levels and occupation types recorded in the LFS. Therefore, as a check, we estimate wage equations for men and women from the HIES for the years 2005 and 2010, and using the estimated coefficients, we impute male and female wages into the LFS data. Since HIES was not surveyed in 2003, we used the 2005 HIES coefficients to impute LFS wages in 2003. Figure D4-1 shows that the distributions of the actual HIES incomes and imputed LFS incomes are very similar in structure for all years. The imputed female income distributions are not perfect matches with actual income values. However, for the majority of women, imputed income is predicted well; only the incomes of women in the extreme tails of the distributions are not well-predicted. Figure D4-1: Comparison of Imputed Wages from LFS with Actual Wages from HIES Year Males Females 2003 .4 .6 .3 .4 .2 .2 .1 0 0 0 5 10 15 0 2 4 6 8 10 x x kdensity lnwg_lfs kdensity lnwg_hies kdensity lnwg_lfs kdensity lnwg_hies 2005 .4 .6 .3 .4 .2 .2 .1 0 0 0 5 10 15 0 2 4 6 8 10 x x kdensity lnwg_lfs kdensity lnwg_hies kdensity lnwg_lfs kdensity lnwg_hies 2010 .8 .6 .6 .4 .4 .2 .2 0 0 0 5 10 15 0 5 10 15 x x kdensity lnwg_lfs kdensity lnwg_hies kdensity lnwg_lfs kdensity lnwg_hies 217 Chapter 4: Enterprise Survey 1. The World Bank’s Enterprise Survey collects information from the manufacturing and service sectors from every region of the world. According to the Sampling Methodology, the Enterprise Survey aims to: (i) provide a picture of the investment climate of a country, that can be used to compare firms’ perceptions of the investment climate across countries; (ii) understand what is hindering the growth of a private sector and job creation in a country; (iii) attempt to construct a panel of firms for each country; and (iv) encourage policies/dialogues for reform. 2. The surveys are conducted with the intent to focus on how constraints in a country’s investment climate affect the productivity level of firms and the potential for firms to create jobs. To achieve this goal, the survey creates a representative sample of a country’s non-agriculture private sector, which can be used to generate sample means at the industry-level. However, summary statistics for firms are not representative at the sub-national-level. The industries surveyed include manufacturing, construction, services, transport, storage, and communication. Furthermore, the following types of firms are not surveyed: (i) public sector firms, (ii) firms in the non-formal sector of the economy, and (iii) firms with fewer than five employees. Thus, the Enterprise Survey looks at the investment climate of the private sector, or, alternatively, the non-agriculture formal economy of a country. 3. In case of non-response due to a firm’s reluctance to participate in the survey, the Enterprise Survey substitutes such firms with those that are willing to take part in the survey. However, they do take note of the non-response rate and indicate the causes of the substitutions (not willing to respond, firm out of business). The survey results show that almost one-half of firms did not invest in any fixed capital in 2006. One possibility is that firms did not need to invest in any capital. Nonetheless, the investment question was very broad (asking about any kind of investment in building construction and repair, land, machinery, computers), so a firm that does not at all invest in a given year seems almost implausible. Thus, other reasons may underlie the lack of investment by private sector firms in Bangladesh. 4. The Enterprise Survey surveyed a total of 1,504 firms in Bangladesh. From this survey, we collect a number of variables, whose summaries and short descriptions are tabulated in Table D4-2. In Table A4-5, we present their respective summary statistics – disaggregated by all firms, and firms that invested in fixed capital in 2006. All values are weighted averages. A total of 1,504 firms are in the sample, but the weighted sample is about 26,700. Table A4-5 shows that firms that invested in 2006 were more likely to use email and internet to communicate, have a larger pool of research staff, provide in- house training, and have a slightly more-educated workforce than the average firm that did not invest. Investing firms’ perceived evaluation of external problems is similar to that of the average, showing that firms that invest are also susceptible to facing the same problems as firms that do not invest. 5. In Heckman Selection models, at least one identification parameter is required (i.e. at least one variable in the first-stage regression must not be present in the second-stage regression). In our case, this must be a variable that can affect the probability that a firm invests but, conditional on investing, cannot affect the firm’s level of investment. We believe that a firm’s ownership structure can affect the probability of investment but not its level. As explained in the literature review section, firm ownership structure can affect firm profitability and efficiency. In this assessment, we assume that the identifying variable is a dummy which indicates if the manager of the firm is also an owner. 6. We also include the percentage of sales paid as bribes/extortion in the second-stage of the regression but not in the first-stage. Payment of bribes/extortion reduces the firm’s pool of capital to invest but does not necessarily reduce the firm’s propensity to invest. 218 Table D4-2: Description of Variables used in the Regressions Variable Used Description Land Owned Proportion of land that the firm owns. Between 0 and 100 The proportion of workers that use a computer in their work. Proportion of workers using Computer Between 0 and 100 Security a threat Firm’s perception that security is a threat in the operation of their business. On a scale of 1 through 4, with 1 being the least threatening and 4 the most. Research Staff The number of research staff that the firm has Hours of Power outages in a day The average number of hours a firm normally does not have power in a day Hours of water shortage in a day The average number of hours a firm faces water shortage in a day Number of Generators used The number of generators that a firm uses The proportion of sales that the firm has to pay as bribes and Bribe/extortion paid as a proportion of sales extortion. Between 0 and 100 Percentage of sales lost due to spoilage on the road The proportion of sales that is lost due to spoilage due to problems with road transportation Problems with transportation The rating of whether a firm faces transportation problems. The scale is between 1 to 4, with 1 indicating no problem and 4 indicating the most problems. Politics a Problem The rating of whether political situation is a problem. The scale is between 1 to 4, with 1 indicating no problem and 4 indicating the most problems. A dummy variable whether the firm provides in-house training In-house training for workers to its workers Education level of workers The average education level of factory workers in the firm. Variables of Divisions where the firm is located Proportion employed in agriculture The proportion of workers in the division employed in the agriculture sector Proportion employed in services The proportion of workers in the division employed in the service sector Education Level of Division Average education level in years in division Proportion of workers in Div. Day laborers Proportion of workers working as day-laborers in the division Ln(working-age population) The natural log of the working-age population in division Population Density in Division Population divided by area of division 219 Chapter 9: WB-BIDS and WB-InM and Longitudinal Household Survey 1. In 1991/92, the World Bank in conjunction with the Bangladesh Institute of Development Studies (BIDS) carried out the first survey to study the role of microfinance in economic and social upliftment among the poor. The survey randomly drew 1,798 households from 87 villages in 29 thanas across rural Bangladesh.152 Out of the twenty-nine thanas, twenty-four were program thanas (with eight from each of the three microfinance programs: Grameen Bank, BRAC, and BRDB RD-12 project), and five were non- program thanas.153 For each program thana, three villages were randomly selected from a list of program villages in which a program had been in operation for at least three years. For each non-program thana, three villages were also randomly selected using the village census of the Government of Bangladesh. Villages with an unusually low or high number of households (fewer than 51 or higher than 600) were excluded from the survey design. In total, 87 villages were selected, and from these villages, a total of 1,798 households were selected based on landholding. The household survey was conducted three times during 1991/92, based on the three cropping seasons: round one during Aman rice (November-February), round two during Boro rice (March-June), and round three during Aus rice (July-October). However, due to attrition, only 1,769 households were available by the third round. A more detailed description of this survey can be found in Khandker (1998). 2. Surveyed households from the 87 villages were revisited in 1998/99, again with the help of BIDS. Unlike the 1991/92 survey, which was conducted three times, the 1998/99 round surveyed households just once. However, among the 1,769 households surveyed in the 1991/92 round, 131 could not be re- traced in 1998/99, leaving 1,638 households available for the re-survey. The attrition rate is, therefore, 7.4 percent. The re-survey also included new households from old villages and newly included villages. Three new, non-target households were randomly selected from each of the existing 87 villages. Also, three new thanas were randomly selected from the southern and south-eastern regions, which were excluded in the first round survey because of the 1991/92 cyclone. Three villages were randomly drawn from each new thana, adding nine more villages. In these new villages, twenty households were drawn from both target and non-target households. Altogether, 2,599 households were surveyed in 1998/99, out of which 2,226 were from old villages and 373 were from new villages. Among the 2,226 households in old villages, 279 are newly sampled households and 1,947 are households previously surveyed in 1991/92. The number of panel households surveyed in 1998/99 (1,947 households) is greater than the number surveyed in 1991/92 (1,638 households) because some old households split after the first survey to form multiple new households. These split households are logically merged with the original households from which they separated. a. In conjunction with the Institute of Microfinance (InM), the households were again surveyed in 2011. The re-survey tried to re-visit all households surveyed in 1998/99 (2,599). However, due to attrition, 2,342 households were identified and 257 households failed to be interviewed. The attrition rate during the 2011 round survey is about 10 percent. However, due to household split-off, the survey interviewed 3,082 households, with 740 households split-off, in 2011. This survey round began in March 2012 and was completed in September 2012. 152 A thana is an administrative unit that is smaller than a district and consists of a number of villages. 153 The twenty-nine thanas were selected from Bangladesh’s 391 rural thanas. The country has a total of 460 thanas. 220 References Abdallah, Wahid and Iffath Sharif. 2012. Using Human Opportunity Index to estimate equality of opportunity in basic services and assessing impact of poverty using HIES 2010. Washington, DC: World Bank Working Paper Series (Forthcoming). Ahmed, Akhter U., 2012. “Operational performance of the employment generation program for the poorest.” Bangladesh Policy Research and Strategy Support Program. Mimeo. IFPRI, Dhaka. Ahmed, Akhter U., Agnes R. Quisumbing, John F. Hoddinott, Mahbuba Nasreen, and Elizabeth Bryan. 2007. “Relative Efficacy of Food and Cash Transfers in Improving Food Security and Livelihoods of the Ultra Poor in Bangladesh.” Washington, DC: WFP/IFP. Ahmed, Akhter U., Shahidur Rashid; Manohar Sharma, and Sajjad Zohir. 2004. “Food Aid Distribution in Bangladesh: Leakage and Operational Performance.” Food Consumption and Nutrition Division Discussion Paper 173, IFPRI, Washington, DC. Ahmed, Akhter, Shahidur Rashid, Manohar Sharma, and Sajjad Zohir. 2003. “A Study on Food-Aid Leakage in Bangladesh.” WFP/IFPRI Brief, Washington, DC. Ahmed, Sadiq. 2000. “Bangladesh Since Independence: Development, Performance, Constraints and Challenges.” Internal Discussion Paper, World Bank, Washington, DC. Aker, Jenny C. 2008. “Does Digital Divide or Provide? The Impact of Cell Phones on Grain Markets in Niger.” BREAD Working Paper 177. Amin, Sajeda. 1998. “Family Structure and Change in Rural Bangladesh.” Population Studies 52 (2): 201–213. Andaleeb, Syed Saad. 1996. “Explaining the Commitment of Family Planning Fieldworkers in Bangladesh.” International Family Planning Perspectives 22 (1):10–15. Bandyopadhyay, Sushenjit, and Emmanuel Skoufias. 2012. “Rainfall Variability, Occupational Choice, and Welfare in Rural Bangladesh.” Policy Research Working Paper No. 6134. World Bank, Washington, DC. Bangladesh Bureau of Statistics. 2012. Bangladesh Household Income and Expenditure Survey: Key Findings and Results. Washington, DC: World Bank, and Dhaka, Bangladesh: Bangladesh Bureau of Statistics. Bangladesh Bureau of Statistics. 2011. Preliminary Report on Household Income and Expenditure Survey – 2010. Dhaka: Momin Offset Press. Bangladesh Bureau of Statistics. 2008. Inflation in Bangladesh: Does the Changing Consumption Pattern Affect Its Measurement? Policy Analysis Unit: Bangladesh Bank. Banks, J. and Johnson, P. (1994) Equivalence scale relativities revisited. Economic Journal, 104, pp. 883- 890. Banks, Ken. 2011. “From Smart Phones to Smart Farming: Indigenous Knowledge Sharing in Tanzania.” National Geographic News Watch, November, 30. Barro, Ricardo, 1991 “Economic Growth of a Cross-Section of Countries.” Quarterly Journal of Economics 106 (2): 407–423. Basu, Ananya, Meera Mahadevan, and Nobuo Yoshida. 2010. “Understanding Seasonal Extreme Poverty in the Northwestern Region of Bangladesh: Preliminary Analysis of the Monga Survey Data, 2008-09.” Volume I, Washington, DC: World Bank. Bates, Lisa M., Joanna Maselko, and Sidney Ruth Schuler. 2007. “Women’s Education and the Timing of Marriage and Childbearing in the Next Generation: Evidence from Rural Bangladesh” Studies in Family Planning 38 (2): 101–112. Bhaskar, V., and Mushtaq Khan, 1995. “Privatization and Employment: A Study of the Jute Industry in Bangladesh.” American Economic Review 85 (1): 267–273. 221 Bhuiya, Abbas, and K. Streatfield. 1991. “Mothers’ Education and Survival of Female Children in a Rural Area of Bangladesh.” Population Studies 45: 253-64. Bhuiyan N I., Paul, D. N. R. and Jabber, M. A. 2002. Feeding the Extra Millions by 2025: Challenges for Rice Research and Extension in Bangladesh. A key note paper presented in the National Workshop on Rice Research and Extension-2002. BRRI, Gazipur, January 29-31. Blinder, Alan S. 1973. “Wage Discrimination: Reduced Form and Structural Estimates.” The Journal of Human Resources 8 (4): 436–455. Bourguignon, François, and Francisco H. G. Ferreira. 2005. “Decomposing Changes in the Distribution of Household Incomes: Methodological Aspects.” In The Microeconomics of Income Distribution Dynamics in East Asia and Latin America, ed. François Bourguignon, Francisco H. G. Ferreira and Nora Lustig, 17-46. Washington, DC: World Bank. Bourguignon, François, Francisco H. G. Ferreira, and Phillippe George Leite. 2008. “Beyond Oaxaca- Blinder: Accounting for Differences in Households Income Distributions.” Journal of Economic Inequality 6 (2): 117–148. Brown, Gordon, 2012. Out of Wedlock, Into School: Combating Child Marriage through Education. London: The Office of Gordon and Sarah Brown. Buhmann,B., Rainwater, L., Schmaus, G., and T. Smeeding. “Equivalence Scales, Well-Being, Inequality, and Poverty: Sensitivity Estimates across Ten Countries using the Luxembourg Income Study (LIS) Database.” Review of Income and Wealth 34(June 1988):115-142. Burket, Mary. K., Mohammad Alauddin, Abdul Malek, and Mizanur Rahman. 2006. Raising the Age of Marriage for Young Girls in Bangladesh. Dhaka, Bangladesh: Pathfinder International/NSDP. Cain, Mead, Syeda Rokeya Khanam, Shamsun Nahar. 1979. “Class, Patriarchy, and Women’s Work in Bangladesh.” Population and Development Review 5 (3): 405-438. Chaudhury, Nazmul, and Jeffrey S. Hammer. 2004. “Ghost Doctors: Absenteeism in Rural Bangladeshi Health Facilities.” The World Bank Economic Review 18 (3). Chen, Lincoln, C., Emdadul Huq, and Stan D’Souza. 1981. “Sex Bias in the Family Allocation of Food and Health Care in Rural Bangladesh.” Population and Development Review 7(1): 55–70. Chou, Shin-Yi, Jin-Tan Liu, Michael Grossman, and Ted Joyce. 2010. “Parental Education and Child Health: Evidence from a Natural Experiment in Taiwan.” American Economic Journal: Applied Economics 2(1): 33–61. Currie, Janet, and Enrico Moretti. 2003. “Mother’s Education and the Intergenerational Transmission of Human Capital: Evidence From College Openings.” Quarterly Journal of Economics 118(4): 1495–1532. Daryl Collins, Jonathan Morduch, Stuart Rutherford, and Orlanda Ruthven, 2009. Portfolios of the Poor, Princeton: Princeton University Press. Das, Jishnu, Jeffrey Hammer, and Carolina Sánchez-Paramo. 2012. “The impact of recall periods on reported morbidity and health seeking behavior.” Journal of Development Economics 98 (1): 76– 88. Datt, Gaurav, and Martin Ravallion. 1992. “Growth and Redistribution Components of Changes in Poverty Measures: A Decomposition with Applications to Brazil and India in the 1980s.” Journal of Development Economics 38: 275–296. Davis, B., P. Winters, G. Carletto, K. Covarrubias, E. J. Quiñones, A. Zezza, K. Stamoulis, C. Azzarri, and S. DiGiuseppe. 2010. “A Cross-Country Comparison of Rural Income Generating Activities.” World Development 38 (1): 48–63. Davis, Peter. 2008. “The Dynamics of Poverty in Rural Bangladesh.” Washington, DC: International Food Policy Research Institute. Deaton, Angus. 2008. “Price Trends in India and Their Implications for Measuring Poverty.” Economic and Political Weekly 7 (September): 3729–48. Deaton, Angus 1989. “Rice Prices and Income Distribution in Thailand: A Nonparametric Analysis.” Economic Journal 99: 1–37. 222 Demeke, Mulat, Guendalina Pangrazio, and Materne Maetz. 2009. "Country responses to the food security crisis: Nature and preliminary implications of the policies pursued." Devereux, Stephen, Bapu Vatila, and Samuel Swan. 2008. Seasons of Hunger: Fighting Cycles of Quiet Starvation among the World’s Rural Poor. London : Pluto Press. Diewert, W. Erwin, John S. Greenless, and Charles R. Hulten, eds. 2009. Price Index Concepts and Measurement. Chicago: University of Chicago Press. DiNardo, John, Nicole Fortin, and Thomas Lemieux. 1996. “Labor Market Institutions and the Distribution of Wages, 1973-1992: A Semi-Parametric Approach.” Econometrica 64 (5): 1001– 1044. Dorosh, Paul A., and Shahidur Rashid. 2012. “Bangladesh Rice Trade and Price Stabilization: Implications of the 2007/08 Experience for Public Stocks.” IFPRI Discussion Paper 01209, IFPRI, Washington, DC. Faveau, Vincent, Kathryn Stewart, S. A. Khan, and Jyotsnamoy Chakraborty. 1991. “Effect on mortality of a community-based maternity-care programme in rural Bangladesh.” The Lancet 338 (8776): 1183–1186. Field, Erica, and Attila Ambrus. 2008. “Early Marriage, Age of Menarche, and Female Schooling Attainment in Bangladesh.” Journal of Political Economy 116 (5): 881–930. Hashemi, S. M., Schuler, S. R., & Riley, A. P. 1996. “Rural credit programs and women's empowerment in Bangladesh.” World development 24(4): 635-653. Haughton, Jonathan, and Shahidur R. Khandker. 2009. Handbook on Poverty and Inequality. Washington, DC: World Bank. Hossan, Chowdhury Golam, Md. Atiqur Rahman Sarker, and Rumana Afroze. 2012. “Recent Unrest in the RMG Sector of Bangladesh: Is this an Outcome of Poor Labour Practices?” International Journal of Business and Management 7 (3). Hoyos, Alejandro, and Ambar Narayan. 2012. “Inequality of Opportunities among Children: How much does gender matter?” Background paper for World Development Report 2012: Gender Equality and Development, World Bank, Washington, DC. IDS (Institute of Development Studies). 2009. Seasonality Revisited International Conference. IDS Bulletin: 17 (3)-Seasonality. IHS Global Insight. 2010. “Indian Wage Inflation to Eclipse China in 2010 As GM India Ends Strike on Pay Issues.” Country & Industry Forecasting: HIS Global Insight, November 20. Inchauste, Gabriela. 2012. “Jobs: Linking growth and poverty reduction”. Mimeo. World Bank, Washinton DC. Inchauste, Gabriela, and Sergio Olivieri. 2012. “Understanding Poverty Reduction in Bangladesh: A Microdecomposition Approach.” Background Paper prepared for the 2012 Poverty Assessment, World Bank, Washinton DC. International Institute for Population Sciences (IIPS) and Macro International. 2007. National Family Health Survey (NFHS-3), 2005–06: India: Volume I. Mumbai: IIPS. Ivanic, Maros, and Will Martin. 2008. “Implications of Higher Global Food Prices for Poverty in Low- income Countries.” Agricultural Economics 39 (Suppl.), 405–416. Jacoby, Hanan G., and Basab Dasgupta. 2012. “Household Exposure to Food price Increase in Rural Bangladesh.” Washington, DC: World Bank. Jenkins, S. P. and Cowell, F. A. (1994) Parametric equivalence scales and scale relativities, Economic Journal, 104, pp. 891-900. Kabeer, Naila, and Simeen Mahmud. 2004. “Rags, Riches and Women Workers: Export-oriented Garment Manufacturing in Bangladesh.” In Chains of Fortune: Linking Women Producers and Workers with Global Markets, ed. Marilyn Carr, London: Commonwealth Secretariat. Khan, M. R. A, and F. Wadud. 2012. “Food Security and Case of Bangladesh” Department of Agricultural Marketing. [http://www.foodsecurityportal.org/bangladesh/food-secur7.6ity-and- case-bangladesh-report-prepared-department-agricultural-marketing] 223 Khandker, Shahidur. R. 1998. “Fighting poverty with microcredit: Experience in Bangladesh”. Washington, DC: World Bank; New York: Oxford University Press. Khandker, Shahidur R. 1987. “Labor Market Participation of Married Women in Bangladesh.” The Review of Economics and Statistics 69 (3): 536–541. Khandker, Shahidur R., and Hussain A. Samad. 2012a. “Is Monga Persistent in Northwest Bangladesh? Evidence Using Recent HIES Data.” Developing Research Group Working Paper (Forthcoming), World Bank, Washington, DC. Khandker, Shahidur R., and Hussain A. Samad. 2012b. “Microfinance Growth and Poverty Reduction in Bangladesh: What Does the Longitudinal Data Say?” Developing Research Group Working Paper (Forthcoming), World Bank, Washington, DC. Khandker, Shahidur R., Hussain A. Samad, and Zahed H. Khan. 1998. “Income and Employment Effects of Micro-credit Programmes: Village-level Evidence from Bangladesh.” Journal of Development Studies 35 (2): 96–124. Khandker, Shahidur R., M. Abdul Khaleque, and Hussain A. Samad. 2011. “Can Social Safety Nets Alleviate Seasonal Deprivation? Evidence from Northwest Bangladesh.” Policy Research Working Paper 5865, World Bank, Washington, DC. Khandker, Shahidur R., and Wahiduddin Mahmud. 2012. Seasonal Hunger and Public Policies: Evidence from Northwest Bangladesh, Washington, D.C.: World Bank Press. Khundker, Nasreen. 2002. “Globalization, Competitiveness and Job Quality in the Garment Industry in Bangladesh”. In Bangladesh: Economic and Social Challenges of Globalization, ed. Muhammed Muqtada, Andrea M. Singh and Mohammed Ali Rashid. University Press Limited. Lanjouw, Jean Olson, and Peter Lanjouw. 2001. “How to Compare Apples and Oranges: Poverty Measurement Based on Different Definitions of Consumption.” Review of Income and Wealth 47 (1): 25–42. Lanjouw, Peter, and Martin Ravallion. “Poverty and Household Size.” The Economic Journal 105(November 1995):1415-1434. Larson, Ann, and S. N. Mitra. 1992. “Family planning in Bangladesh: An unlikely success story.” International Family Planning Perspectives 18 (4): 123–129. Lesthaeghe, Ron J. 2010. “The Unfolding Story of the Second Demographic Transition.” Population and Development Review 36: 211–251. Lewis, William Arthur. 1954. “Economic Development with Unlimited Supplies of Labor.” The Manchester School 22 (2): 139–191. Mahadevan, M., S. Takamatsu, and Nobuo Yoshida. 2012. Understanding Seasonal Extreme Poverty in the Northwestern Region of Bangladesh Volume 2: The extent of seasonal deprivation and informal/formal coping mechanisms. Washington, DC: World Bank. Mankiw, Gregory N., David Romer, and David N. Weil. 1992. “A Contribution to the Empirics of Economic Growth.” The Quarterly Journal of Economics 107 (2): 407–423. Matin Khan A. 2012. “The Demographic Dividend in Bangladesh : An Illustrative Study.” Paper presented at the 18th Biennial Conference of the Bangladesh Economic Association, Dhaka, July 12-14. McFadden, Daniel 1974a. “Conditional logit analysis of qualitative choice behavior.” In Frontiers in Econometrics, ed. Paul Zarembka. New York: Academic Press. McFadden, Daniel 1974b. “The Measurement of Urban Travel Demand.” Journal of Public Economics 34: 63-82. Menon, P., M. T. Ruel, C. U. Loechl, M. Arimond, J. Habicht, G. Pelto, and L. Michaud. 2007. “Micronutrient Sprinkles Reduce Anemia among 9- to 24-Mo-Old Children When Delivered through an Integrated Health and Nutrition Program in Rural Haiti.” Journal of Nutrition 137 (4): 1023–1030. 224 Minten Bart, Thomas, Reardon, and Kevin Chen, and. 2012. “The Quiet Revolution of ‘Traditional’ Agricultural Value Chains in Asia: Evidence from Staple Food Value to Four Mega-cities.” Unpublished draft, International Food Policy Research Institute, Washington, DC. Mitra, K. Amal, Mohammad M Rahman, and George J. Fuchs. 2000. “Risk factors and gender differentials for death among children hospitalized with diarrhea in Bangladesh.” Journal of Health, Population, Nutrition 18 (3): 151–6. Mulat Demeke, Guendalina Pangrazio and Materne Maetz. 2009. “Country responses to the food security crisis: Nature and preliminary implications of the policies pursued.” Murdoch, Jonathan. 1998. “Does Microfinance Really Help the Poor? New Evidence from Flagship Programs in Bangladesh.” Princeton University Working Paper No. 198, Princeton University. Nargis, Nigar, and Mahabub Hossain. 2006. “Income Dynamics and Pathways Out of Rural Poverty in Bangladesh, 1988-2004.” Agricultural Economics 35: 425–435. National Institute of Population Research and Training (NIPORT), Mitra and Associates, and Macro International. 2011. Bangladesh Demographic and Health Survey 2011. Dhaka and Calverton: National Institute of Population Research and Training, Mitra and Associates, and Macro International. National Institute of Population Research and Training (NIPORT), Mitra and Associates, and Macro International. 2009. Bangladesh Demographic and Health Survey 2007. Dhaka and Calverton: National Institute of Population Research and Training, Mitra and Associates, and Macro International. National Institute of Population Research and Training (NIPORT), Mitra and Associates, and Macro. 2005. Bangladesh Demographic and Health Survey 2004. Dhaka and Calverton: National Institute of Population Research and Training, Mitra and Associates, and Macro International. Newman, John L., Jao Pedro Azevedo, Jaime Saavedra, and Ezequiel Molina. 2008. “The Real Bottom Line: Benchmarking Performance in Poverty Reduction in Latin America and the Caribbean.” Paper presented at LACEA conference, Washington, DC, July 22. Ñopo, Hugo, 2008 “Matching as a Tool to Decompose Wage Gaps.” The Review of Economics and Statistics 90 (2): 290–299. Oaxaca, Ronald. 1973. “Male-Female Wage Differentials in Urban Labor Markets.” International Economic Review 14 (3): 693–709. Paci Pierella, and Marcin Sasin. 2008. “Making Work Pay in Bangladesh. Employment, Growth and Poverty Reduction.” Washington, DC: World Bank. Paes de Barros, Ricardo, Francisco H. G. Ferreira, Jose R. Molinas Vega, and Jaaime Saavedra Chanduvi. 2009. Measuring Inequality of Opportunities in Latin America and Caribbean. Washington, DC: World Bank. Paes de Barros, Ricardo, Mirela de Carvalho, Samuel Franco, and Rosane Mendoça. 2006. “Uma Análise das Principais Causas da Queda Recente na Desigualdade de Renda Brasileira.” Revista Econômica 8 (1): 117-147. Rabbani, Atonu. 2012. “Status of Food Security beyond Poverty Measures: Evidence from HIES 2005 and 2010.” Dhaka, Bangladesh: World Bank. Rah, J. H., Akhter, N., Semba, R. D., de Pee, S., Bloem, M. W., Campbell, A. A., et al. 2010. “Low dietary diversity is a predictor of child stunting in rural Bangladesh.” European Journal of Clinical Nutrition, 64: 1393–1398. Rahman, Jyoti, and Asif Yusuf. “Economic Growth in Bangladesh: Experience and Policy Priorities.” Unpublished manuscript, last modified February 5, 2013. Portable Document Format (PDF). Ravallion, Martin. 2001. “Poverty Lines: Economic Foundations of Current Practices.” World Bank. Unpublished manuscript. Ravallion, Martin. 1998. “Poverty Lines in Theory and Practice.” LSMS Working Paper 133, World Bank, Washington, DC. 225 Ravallion, Martin, 1990. “Reaching The Poor Through Rural Public Employment; A Survey Of Theory And Evidence.” Discussion Papers 94, World Bank, Washington, DC. Ravallion, Martin, and Shaohua Chen. 2003. ‘Measuring Pro-Poor Growth.” Economics Letters 78 (1): 93–99. Ruthbah, Ummul Hasanath. 2012. “EGPP: Determinants of Participation and Implementation: Evidence from Household Surveys”. Mimeo, World Bank, Dhaka. Santos, Indhira, Iffath Sharif, Hossain Zillur Rahman, and Hassan Zaman. 2011. “How Do the Poor Cope with Shocks in Bangladesh? Evidence from Survey Data.” Policy Research Working Paper 5810, World Bank, Washington, D.C. Sasin, Marcin 2009. “Making work pay: Growth, Employment and the Labor Market in Bangladesh.” In Breaking Down Poverty in Bangladesh, ed Amabar Narayan, and Hassan Zaman, 109-133. Dhaka: The University Press. Sen, Amartya. 1983. “Poor, Relatively Speaking.” Oxford Economic Papers 35 (2): 153–69. Sharif, Iffath. 2009. “Building a Targeting System in Bangladesh based on Proxy-Means Testing.” Social Protection Discussion Paper 0914, World Bank, Washington, DC. Shilpi, Forhad 2012. “Trends and Sources of Spatial Differences in Welfare in Bangladesh.” World Bank Working Paper Series (Forthcoming), World Bank, Washington, DC. Sincavage, Jessica R., Carl Haub, and O. P. Sharma. 2010. “Labour Costs in India’s Organized Manufacturing Sector”, Monthly Labor Review 133 (5): 3-23. Stanton, Bonita F., and John D. Clemens. 1988. “The Influence of Gender on Determinants of Urban Childhood Mortality in Bangladesh.” International Journal of Epidemiology 17 (1): 129–35. Steyn, N. P., J. H. Nel, G. Nantel, G. Kennedy, and D. Labadarios. 2006. “Food Variety and Dietary Diversity Scores in Children: Are They Good Indicators of Dietary Adequacy?” Public Health Nutrition 9 (5): 644-650. UN (United Nations). 2011. UN World Population Prospects 2011: The 2011 Revision Population Database. UNICEF (United Nations Children’s Fund). 2011. The State of the World’s Children 2011: Adolescence – An Age of Opportunity. NY:UNICEF. United Nations Children’s Fund (UNICEF). 2001. Early Marriage Child Spouse. Innocenti Digest 7, UNICEF Innocenti Research Centre, Florence. US Census Bureau. 2012. Bangladesh: Population Pyramids, 2000, 2010. Retrieved on November 3, 2012, from http://www.census.gov/population/international/data/idb/region.php. US Department of Agriculture. 2012. Bangladesh Grain and Feed Annual, by Sayed Sarwer Hussain. GAIN Report BG2001. Washington, DC. Van de Kaa, Dirk J. 1987. “Europe’s Second Demographic Transition.” Population Bulletin 42 (1). Population Reference Bureau, Washington, DC. Vleminckx, K., and T. Smeeding. Child Well-being, Child Poverty and Child Policy in Modern Nations. Bristol, UK: The Policy Press, 2003. Waterlow, John C., and Beat Schürch. 1994. Causes and Mechanisms of Linear Growth Retardation: Proceedings of an I.D.E.C.G. Workshop Held in London, January 15-18, 1993. Basingstoke: MacMillan Press. World Bank. 2012a. Towards Accelerated, Inclusive and Sustainable Growth: Opportunities and Challenges. 2 vols. Washington, DC: World Bank. World Bank. 2012b. Bangladesh: Economic Growth in Bangladesh: Achievements, Prospects and Challenges. World Bank Report 67991. Washington, DC: World Bank. World Bank. 2012c. Food Price Watch, Full Report. Permanent URL: http://go.worldbank.org/JA4CUPKLJ0. World Bank. 2012d. Information and Communications for Development 2012: Maximizing Mobile. Washington, DC: World Bank. 226 World Bank. 2008a. Poverty Assessment for Bangladesh, Creating Opportunities and Bridging the East- West Divide. Dhaka: World Bank. World Bank. 2008b. Whispers to Voices. Gender and Social Transformation in Bangladesh. Bangladesh Development Series Paper 22. Washington, DC: World Bank. World Bank. 2005a. Bangladesh - Attaining the Millennium Development Goals in Bangladesh. Washington, DC: World Bank. World Bank. 2005b. Maintaining Momentum Toward the MDGs: An Impact Evaluation of Interventions to Improve Maternal and Child Health and Nutrition Outcomes in Bangladesh. Washington, DC: World Bank. World Bank. 2002. Poverty Assessment for Bangladesh, Building on Progress. Washington, DC: World Bank. World Food Program. 2010. Agricultural Production & National Food Balance: Bangladesh. Zhang, Xiaobo, Jin Yang, and Shenglin Wang. 2011. “China Has Reached the Lewis Turning Point”, China Economic Review 22 (4): 542–554. Zhang, Xiaobo, Shahidur Rashid, Kaikaus Ahmad, Valerie Muller, Hak Lim Lee, Solomon Lemma, Saika Belal, and Akhter Ahmed. 2012. “Rising Wages and Poverty Reduction in Bangladesh,” IFPRI (Forthcoming), Washington, DC. 227