AUS6744 v1 ETHIOPIA POVERTY ASSESSMENT 2014 POVERTY GLOBAL PRACTICE AFRICA REGION Report No. AUS6744 ETHIOPIA POVERTY ASSESSMENT January 2015 Poverty Global Practice Africa Region Document of the World Bank For Official Use Only iii TABLE OF CONTENTS ACKNOWLEDGEMENTS................................................................................................................................. xi ABBREVIATIONS AND ACRONYMS........................................................................................................... xiii EXECUTIVE SUMMARY...................................................................................................................................xv INTRODUCTION............................................................................................................................................xxv 1. PROGRESS IN REDUCING POVERTY AND INCREASING WELLBEING, 1996-2011....................... 1 1.1  Recent progress in poverty reduction...................................................................................................................2 1.2  Sensitivity of poverty estimates............................................................................................................................5 1.3  The incidence of progress and shared prosperity..................................................................................................9 Growth incidence................................................................................................................................................9 Shared prosperity..............................................................................................................................................11 Inequality..........................................................................................................................................................13 Decomposing changes into growth and redistribution......................................................................................14 1.4  Who are the poor and poorest households in 2011?...........................................................................................16 1.5  Outlook: Ending extreme poverty in Ethiopia...................................................................................................21 2. MULTIDIMENSIONAL POVERTY IN ETHIOPIA.................................................................................. 25 2.1 Introduction......................................................................................................................................................25 2.2  Trends in non-monetary dimensions of wellbeing..............................................................................................26 Education.........................................................................................................................................................28 Health...............................................................................................................................................................29 Command over resources and access to information.........................................................................................29 2.3  Overlapping deprivations..................................................................................................................................30 2.4  Perceived improvements in wellbeing................................................................................................................35 2.5  Deprivations that particularly affect girls and women........................................................................................35 2.6 Conclusion........................................................................................................................................................37 3. THE CHANGING NATURE OF VULNERABILITY IN ETHIOPIA....................................................... 39 3.1  Sources of risk in today’s Ethiopia......................................................................................................................39 3.2  Measuring vulnerability in Ethiopia...................................................................................................................42 3.3  Vulnerable places or vulnerable people?.............................................................................................................46 3.4  Summary and conclusion..................................................................................................................................50 iv ETHIOPIA – POVERTY ASSESSMENT 4. DRIVERS OF POVERTY REDUCTION IN ETHIOPIA........................................................................... 51 4.1  Decomposing poverty reduction........................................................................................................................52 4.2  Drivers of poverty reduction..............................................................................................................................56 Has growth contributed to poverty reduction?..................................................................................................56 Understanding the relationship between agricultural growth and poverty reduction..........................................58 Safety nets and investments in public services...................................................................................................63 4.3  Implications for future poverty reduction..........................................................................................................64 5. A FISCAL INCIDENCE ANALYSIS FOR ETHIOPIA................................................................................ 67 5.1  Taxation incidence.............................................................................................................................................68 5.2  Incidence of public expenditure.........................................................................................................................73 Direct transfers made through the PSNP and food aid......................................................................................74 Education.........................................................................................................................................................75 Health...............................................................................................................................................................77 Indirect subsidies..............................................................................................................................................78 Overall incidence of public spending................................................................................................................79 5.3  Overall incidence of taxes and spending and impact on poverty and inequality.................................................80 6. NON-FARM ENTERPRISES AND POVERTY REDUCTION IN ETHIOPIA....................................... 83 6.1 Introduction......................................................................................................................................................83 6.2  Prevalence and nature of NFEs in Ethiopia........................................................................................................84 6.3  The role of NFE in incomes of poor households................................................................................................86 6.4  Constraints to NFE activities.............................................................................................................................89 6.5 Conclusion........................................................................................................................................................93 7. MIGRATION AND POVERTY IN ETHIOPIA.......................................................................................... 95 7.1 Introduction......................................................................................................................................................95 7.2  Migration in Ethiopia........................................................................................................................................96 7.3  Migration and poverty.....................................................................................................................................100 7.4  What constrains migration in Ethiopia?...........................................................................................................104 7.5 Conclusion......................................................................................................................................................106 8. UNDERSTANDING URBAN POVERTY.................................................................................................. 107 8.1  Work and urban poverty..................................................................................................................................108 8.2  Reducing poverty in urban centers through work: a framework.......................................................................113 8.3  Urban poverty among those unable to work....................................................................................................118 8.4  Improving urban safety nets............................................................................................................................120 8.5 Summary.........................................................................................................................................................123 9. GENDER AND AGRICULTURE................................................................................................................ 125 9.1 Introduction....................................................................................................................................................125 9.2  Gender productivity differentials: Ethiopia in a regional comparison...............................................................126 9.3  Zooming in: Refining the decomposition........................................................................................................128 9.4  Explaining gender differences in input-use.......................................................................................................134 9.5 Conclusion......................................................................................................................................................138 Table of Contents v ANNEXES ANNEX 1........................................................................................................................................................... 143 ANNEX 2........................................................................................................................................................... 155 ANNEX 3........................................................................................................................................................... 161 ANNEX 4........................................................................................................................................................... 165 ANNEX 5........................................................................................................................................................... 171 ANNEX 6........................................................................................................................................................... 179 REFERENCES.................................................................................................................................................. 181 LIST OF FIGURES Figure 1.1: Progress in health, education and living standards in Ethiopia from 2000 to 2011......................................2 Figure 1.2: Poverty headcount by region from 1996 to 2011........................................................................................4 Figure 1.3: Incidence of monetary poverty in Ethiopia compared with other African countries....................................5 Figure 1.4: Annual reduction of poverty headcount at US$$1.25 PPP poverty line for selected countries with two poverty measurements in the last decade..............................................................................................5 Figure 1.5: Food share per consumption percentile across time.....................................................................................7 Figure 1.6: Growth Incidence Curves with 95% confidence intervals nation-wide, urban and rural............................10 Figure 1.7: Consumption growth was negative in Addis Ababa from 2005 to 2011....................................................11 Figure 1.8: Average growth for the bottom 10%, bottom 40% and the top 60% from 1995 to 2011.........................11 Figure 1.9: Average growth for the bottom 10%, bottom 40% and top 60% for 1996 to 2011, by rural and urban...12 Figure 1.10: Gini Coefficient in Ethiopia and other African Countries.........................................................................12 Figure 1.11: Gini and Theil index for national, urban and rural Ethiopia, 1996–2011.................................................13 Figure 1.12: Relative and absolute income differences between different income percentiles.........................................15 Figure 1.13: Growth and redistribution decomposition of poverty changes, 1996–2011..............................................16 Figure 1.14: Poverty headcount, depth and severity for children and adults..................................................................17 Figure 1.15: Poverty incidence based on simulations with percentile-specific growth (solid lines) and average growth (dotted line).....................................................................................................................22 Figure 1.16: Poverty statistics in 2030 compared to current values for different simulations.........................................22 Figure 1.17: Simulation of income shares (relative to average income) for bottom 10%, bottom 40%, and top 60% using percentile-specific growth rates..........................................................................................23 Figure 1.18: Alternate simulations................................................................................................................................23 Figure 2.1: Monetary, education and sanitation deprivation in urban and rural areas, 2000–2011..............................32 Figure 2.2: Evolution of overlapping deprivations over time, 2000–2011 (rural Ethiopia)..........................................32 Figure 2.3: Components of the MPI in 2011 and over time, 2000–2011...................................................................34 Figure 2.4: Monetary and education deprivations and wellbeing perception, 2000–2011...........................................35 Figure 2.5: Multiple deprivations affecting women, 2011...........................................................................................37 Figure 3.1: Impact of Drought and food price shocks.................................................................................................41 Figure 3.2: All Ethiopia: Meher crop losses.................................................................................................................41 Figure 4.1: The contribution of rural and urban poverty reduction.............................................................................53 vi ETHIOPIA – POVERTY ASSESSMENT Figure 4.2: The contribution of poverty reduction among different sectors.................................................................53 Figure 4.3: The contribution of poverty reduction among the employed and self-employed.......................................53 Figure 4.4: The contribution of demographics, education, occupational change and urbanization to consumption growth, 1996–2011..................................................................................................................................54 Figure 4.5: The changing relationship between education and consumption, 1996–2011...........................................55 Figure 4.6: The contribution of agricultural growth, services and safety nets to poverty reduction, 1996–2011..........58 Figure 4.7: Services growth is positively correlated with growth in agriculture............................................................58 Figure 4.8: Increased fertilizer use reduced poverty when weather and prices were good.............................................61 Figure 4.9: Proportion of farmers experiencing more than 30% crop loss, 1997–2011...............................................61 Figure 4.10: Travel time to urban centers of 50,000 people or more in 1994 and 2007................................................64 Figure 5.1: Incidence of direct taxes by market income deciles....................................................................................70 Figure 5.2: Concentration curves and incidence of direct taxes...................................................................................71 Figure 5.3: Incidence of Indirect taxes by market income deciles................................................................................72 Figure 5.4: Direct and indirect tax concentration curves in relation to market income Lorenz curve...........................72 Figure 5.5: Concentration of total taxes across socioeconomic groups, cross-country comparison...............................73 Figure 5.6: Ethiopia. Direct transfers by market income deciles..................................................................................75 Figure 5.7: Effectiveness of direct transfers in comparison to direct transfers in other countries..................................76 Figure 5.8: Ethiopia. Incidence and concentration shares of education.......................................................................76 Figure 5.9: Ethiopia: Incidence and concentration shares of health.............................................................................77 Figure 5.10: Health benefit incidence as percent of income by market income decile...................................................78 Figure 5.11: Incidence and concentration curves for indirect subsidies.........................................................................78 Figure 5.12: Transfers and subsidies as a proportion of consumption in rural and urban Ethiopia................................79 Figure 5.13: Proportion of household with electricity (%) by market income category.................................................79 Figure 5.14: Ethiopia. Public expenditure programs (percent of spending included in analysis)....................................79 Figure 5.15: Concentration coefficients of public spending..........................................................................................80 Figure 5.16: Ethiopia. Incidence of taxes and transfers (by market income deciles).......................................................81 Figure 6.1: Age of NFEs.............................................................................................................................................85 Figure 6.2: Households’ reaction to shocks.................................................................................................................88 Figure 6.3: Seasonality of NFE creation......................................................................................................................89 Figure 6.4: Highest months of NFE operation...........................................................................................................90 Figure 6.5: Harvest season and NFE operation, by type NFE sector...........................................................................90 Figure 6.6: NFE activity for Farming and non-Farming households...........................................................................92 Figure 7.1: Migration Flow, 2007...............................................................................................................................97 Figure 7.2: Migrants by duration of stay in current residence.....................................................................................97 Figure 7.3: City size and migration.............................................................................................................................97 Figure 7.4: Migration and employment......................................................................................................................98 Figure 7.5: Proportion of migrants and non-migrants that are male............................................................................98 Figure 7.6: Age distribution of those who migrated in last 5 years..............................................................................99 Figure 7.7: Education levels: Migrants and non-migrants, 2007.................................................................................99 Figure 7.8: Employment status of working migrants and non-migrants, 2007..........................................................100 Figure 7.9: Number of Rooms in the House.............................................................................................................101 Figure 7.10: Access to tap water and electricity among migrants.................................................................................101 Figure 7.11: Migration and poverty............................................................................................................................102 Table of Contents vii Figure 7.12: Distribution of consumption for migrants and non-migrants.................................................................102 Figure 7.13: Subjective measures of wellbeing.............................................................................................................103 Figure 7.14: Housing ownership by duration of migration.........................................................................................103 Figure 7.15: Ownership of radio and television..........................................................................................................104 Figure 8.1: City size, poverty and inequality in Ethiopia...........................................................................................107 Figure 8.2: City size and the nature of jobs...............................................................................................................109 Figure 8.3: City size and unemployment..................................................................................................................109 Figure 8.4: Median wages of employees in Addis Ababa, other big towns and small towns.......................................110 Figure 8.5: Towns and cities with higher rates of employment are less poor..............................................................110 Figure 8.6: Characteristics of the unemployed..........................................................................................................112 Figure 8.7: Unemployment, self-employment and education in Addis Ababa (12 month definition)........................113 Figure 8.8: Rate of finding employment among unemployed youth in Addis Ababa.................................................116 Figure 8.9: Labor markets in large cities: three types.................................................................................................117 Figure 8.10: The urban poverty profile is similar to the rural poverty profile on some dimensions..............................119 Figure 8.11: Being disabled, widowed, and elderly is more associated with poverty in urban areas..............................119 Figure 8.12: The elderly and disabled are less able to cope with shocks in urban areas.................................................120 Figure 8.13: Transfers and subsidies as a proportion of market income in rural and urban Ethiopia...........................120 Figure 8.14: Larger transfers have a larger effect on the poverty rate...........................................................................121 Figure 8.15: Addis Ababa poverty map.......................................................................................................................123 Figure 9.1: Gender Gap in Agricultural productivity, by country..............................................................................125 Figure 9.2: Factors that widen the gender gap in agricultural productivity................................................................127 Figure 9.3: Components of gender differentials in productivity................................................................................132 Figure A1.1: Food share in total consumption across time for different deflators.........................................................143 Figure A1.2:rowth Incidence Curve for 2005 to 2011 for full sample and for partial sample......................................143 Figure A4.1: Scatter of estimated and measured level of poverty by zone.....................................................................168 Figure A5.1: Definitions of income concepts in CEQ methodology............................................................................172 LIST OF TABLES Table 1: Ethiopia then and now: a decade of progress from 2000 to 2011............................................................. xiv Table 2: Poverty, inequality, wellbeing and sector of employment, 2000–2011...................................................... xxi Table 1.1: Poverty headcount ratio for national poverty line (per adult) and the US$1.25 PPP poverty line (per capita).................................................................................................................................................3 Table 1.2: Poverty headcount ratio for national poverty line by region........................................................................4 Table 1.3: Poverty depth and severity from 1996 to 2011 (at national poverty line)....................................................5 Table 1.4: HICES and CPI measures of inflation over 1996 to 2011..........................................................................6 Table 1.5: Test of sensitivity of poverty rates to new survey methodology....................................................................8 Table 1.6: Profile of the poor for 1996, 2000, 2005 and 2011..................................................................................18 Table 1.7: Differences in characteristics between consumption percentiles................................................................19 Table 2.1: Deprivation indicators, definitions and their use for urban and rural overlap analysis...............................27 Table 2.2: Proportions of deprived households, 2000–2011......................................................................................28 Table 2.3: Household’s perception about living standards and price shock, 2011......................................................36 Table 2.4: Deprivation status for school aged children (aged 5–17) by relationship, 2011.........................................37 Table 3.1: Frequency of shocks..................................................................................................................................40 viii ETHIOPIA – POVERTY ASSESSMENT Table 3.2: Measure of vulnerability...........................................................................................................................43 Table 3.3: 2011 vulnerability and poverty national overview.....................................................................................45 Table 3.4: Vulnerability measures over time...............................................................................................................45 Table 3.5: Poverty and vulnerability across the “five Ethiopias” and urban centers, 2011 Table 3.6: The proportion of individuals measured as poor and vulnerable by PSNP status, 2011.............................48 Table 3.7: The number of individuals measured as poor and vulnerable in PSNP woredas, 2011 (million)................48 Table 3.8: Demographic characteristics of vulnerability.............................................................................................49 Table 4.1: Growth, safety nets and infrastructure investments contributed to poverty reduction...............................57 Table 4.2: Agricultural growth and poverty reduction...............................................................................................60 Table 4.3: Annual food inflation in selected countries...............................................................................................61 Table 4.4: Favorable rainfall and improved producer prices contributed to agricultural growth.................................62 Table 4.5: Food gap of poor households, 2005 and 2011..........................................................................................65 Table 5.1: Ethiopia: Tax revenue structure 2011........................................................................................................68 Table 5.2: Average per capita direct taxes in Birr per year and concentration by decile...............................................70 Table 5.3: Ethiopia: General government expenditure 2011......................................................................................73 Table 5.4: Poverty Indicators before and after PSNP and food aid transfers...............................................................75 Table 5.5: Poverty and inequality indicators before and after taxes and spending.......................................................81 Table 5.6: Impoverishment and fiscal policy in Ethiopia...........................................................................................82 Table 6.1: Types of NFEs 1.......................................................................................................................................84 Table 6.2: Proportion of households operating an NFE (%)......................................................................................85 Table 6.3: Prevalence of NFEs by per adult equivalent expenditures..........................................................................86 Table 6.4: Annual agricultural profits per hectare......................................................................................................88 Table 6.5: Source of start-up funds for NFEs............................................................................................................91 Table 6.6: The three main constraints to NFE growth...............................................................................................92 Table 7.1: Migration and agricultural productivity..................................................................................................104 Table 8.1: Mean poverty measures and t-test results, by city size category................................................................108 Table 8.2: National, urban and rural unemployment rates, various definitions........................................................110 Table 8.3: The relationship between poverty, city size and employment...................................................................111 Table 8.4: Poverty and unemployment in Addis Ababa...........................................................................................112 Table 8.5: Two types of unemployed.......................................................................................................................114 Table 8.6: Permanent employment in Ethiopia:......................................................................................................115 Table 8.7: The simulated impact of introducing policies to address urban poverty...................................................118 Table 9.1: Descriptive statistics on the mean and differences, by gender and matching Status.................................130 Table 9.2: Descriptive statistics on the mean and differences for matched farmers...................................................133 Table 9.3: Gender differences by different groups....................................................................................................136 Table A1.1: Difference in means, household characteristics by poverty status and consumption decile (1996–2011) (Total)....................................................................................................................................................144 Table A1.2: Difference in means by percentile of consumption distribution (1996)...................................................147 Table A1.3: Difference in means by percentile of consumption distribution (2000)...................................................148 Table A1.4: Difference in means by percentile of consumption distribution (2005)...................................................150 Table A1.5: Difference in means by percentile of consumption distribution (2011)...................................................152 Table A2.1: Deprivation Indicators............................................................................................................................155 Table A2.2: Deprivation proportions by Venn diagram region in Figure 2.2: urban and rural populations.................158 Table of Contents ix Table A2.3: Deprivation proportions by Venn diagram region in Figure 2.4: rural population...................................158 Table A2.4: Deprivation proportions by Venn diagram region in Figure 2.4: urban population.................................159 Table A2.5: Deprivation proportions by Venn diagram region in Figure 2.5: urban and rural populations.................159 Table A3.1: Full results of regression estimation of log of consumption per adult equivalent (pooled across 2005 and 2011)...............................................................................................................161 Table A4.1: Zonal averages of key variables................................................................................................................167 Table A5.1: Ethiopia. Tax rate schedules on Direct Taxes...........................................................................................173 Table A5.2: Oromia Regional State, land use fee and agricultural income tax rule.....................................................174 Table A5.3: Ethiopia: Goods that are liable to excise tax when either produced locally or imported...........................175 Table A5.4: Current tariff for household electricity consumption (monthly)..............................................................177 LIST OF BOXES Box 1.1: Poverty measures.........................................................................................................................................3 Box 1.2: Inequality measures..................................................................................................................................14 Box 1.3: Poverty, growth, and inequality.................................................................................................................15 Box 2.1: The WIDE-3 qualitative research program................................................................................................26 Box 2.2: Aspirations and educational investments in rural Ethiopia........................................................................31 Box 2.3: The Multidimensional Poverty Index .......................................................................................................33 Box 2.4: Learning how to provide education to out-of-school girls in Addis Ababa.................................................38 Box 3.1: A measure of vulnerability to poverty using cross-sectional data................................................................44 Box 4.1: What does decomposing changes in poverty entail?..................................................................................52 Box 4.2: Agricultural growth in 12 rural communities............................................................................................59 Box 5.1: Terminology.............................................................................................................................................69 Box 8.1: Youth unemployment and job search in Addis Ababa.............................................................................114 Box 9.1: Policy example: Government response and RCBP in Ethiopia................................................................139 xi ACKNOWLEDGEMENTS T he World Bank greatly appreciates the close Hoddinott (IFPRI) and Alemayehu Seyoum Taffesse collaboration with the Government of Ethiopia provided input for the boxes in Chapter 2. (the Ministry of Finance and Economic Chapter 3: “A Vulnerability Assessment for Development and the National Planning Commission, Ethiopia” by Ruth Hill and Catherine Porter (Herriot- in particular) in the preparation of this report. The Watt University), and reviewed by Matthew Hobson core team preparing this report consisted of Ruth (Senior Social Protection Specialist, GSPDR), Hill (Senior Economist, GPVDR) and Eyasu Tsehaye Camilla Holmemo (Senior Economist, GSPDR), Tim (Economist, GPVDR). Many people contributed to Conway, and the PSNP working group. this report through the preparation and review of Chapter 4: “Growth, Safety Nets and Poverty: background papers that form the basis for Chapters Assessing Progress in Ethiopia from 1996 to 2011” 2 to 9 of this report. The list of background papers, by Ruth Hill and Eyasu Tsehaye, and reviewed by authors and reviewers is provided below. Luc Christiaensen (Senior Economist, AFRCE) and The core team received guidance and comments Alemayehu Seyoum Taffesse. on the concept note, drafts of papers, chapters and Chapter 5: “Fiscal Incidence in Ethiopia” by presentations from Ana Revenga (Senior Director, Tassew Woldehanna, Eyasu Tsehaye, Gabriela GPVDR), Pablo Fajnzylber (Practice Manager, Inchauste (Senior Economist, GPVDR), Ruth Hill GPVDR), Lars Moller (Lead Economist and Program and Nora Lustig (University of Tulane), and reviewed Leader, AFCE3), Stefan Dercon (Peer reviewer and by the Commitment to Equity team. Chief Economist, Department for International Chapter 6: “Nonfarm Enterprises in Rural Development, UK (DFID)), Franciso Ferreira (Peer Ethiopia: Improving Livelihoods by Generating reviewer and Chief Economist, AFRCE), Andrew Income and Smoothing Consumption?” by Julia Dabalen (Peer reviewer and Lead Poverty Specialist, Kowalski (LSE), Alina Lipcan (LSE), Katie McIntosh GPVDR), Ambar Narayan (Peer reviewer and Lead (LSE), Remy Smida (LSE), Signe Jung Sørensen (LSE), Economist, GPVDR), Pedro Olinto (Peer reviewer Dean Jolliffe, Gbemisola Oseni (Ecoomist, DECPI), and Senior Economist, GPVDR), Eliana Carranza Ilana Seff (Consultant, DECPI) and Alemayehu (Economist, GPVDR), Alemayehu Seyoum Taffesse Ambel, and reviewed by Kathleen Beegle (Lead (International Food Policy Research Institute Economist, AFRCE) and Bob Rijkers (Economist, (IFPRI)), Tassew Woldehanna (University of Addis DECTI). Ababa), and Tim Conway (DFID). Chapter 7: “Internal Migration in Ethiopia: Chapter 2: “Multidimensional Poverty in Ethiopia, Stylized Facts from Population Census, 2007” by 2000–2011” by Alemayehu Ambel (Economist, Forhad Shilpi (Senior Economist, DECAR) and DECPI), Parendi Mehta (Consultant) and Biratu Jiaxiong Yao (Consultant), and reviewed by Alan de Yigezu (Deputy Director, Ethiopian Central Statistical Brauw (IFPRI). “Migration, Youth and Agricultural Agency), and reviewed by Dean Jolliffe (Senior Productivity in Ethiopia” by Alan de Brauw (IFPRI), Economist, DECPI) and Maria Ana Lugo (Economist, and reviewed by Daniel Ayalew Ali (Economist, GPVDR). In addition Laura Kim (Consultant), John DECAR) and Forhad Shilpi. xii ETHIOPIA – POVERTY ASSESSMENT Chapter 8: “Cities and Poverty in Ethiopia” by GPVDR), Markus Goldstein (Practice Leader, Ruth Hill, Parendi Mehta, Thomas Pave Sohnesen AFRCE), and reviewed by Andrew Goodland (Practice (Consultant), and reviewed by Celine Ferre and Leader, AFCE3) and Gbemisola Oseni. Megha Mukim (Economist, GTCDR). “Work, Funding for the background paper for Chapter 3 Unemployment and Job Search among the Youth in came from the Social Protection Global Practice. The Urban Ethiopia” by Simon Franklin (University of zonal analysis undertaken for Chapter 4 benefited Oxford), and reviewed by Patrick Premand (Senior from funding provided through a Poverty and Social Economist, GPVDR) and Pieter Serneels (University Impact Assessment grant. The background papers of East Anglia). “A Model of Entrepreneurship and behind Chapters 7 and the first three background Employment in Ethiopia: Simulating the Impact of papers listed under Chapter 8 were funded by the an Urban Safety-net” by Markus Poschke (McGill CHYAO trust fund. University), and reviewed by Douglas Gollin Utz Pape contributed to the analysis and writ- (University of Oxford). “Targeting Assessment ing of Chapter 1. Jonathan Karver, Rhadika Goyal, and Ex-Ante Impact Simulations of Addis Ababa Christopher Gaukler and Jill Bernstein provided Safety Net” by Pedro Olinto and Maya Sherpa (ET research assistance for various chapters of the report. Consultant, GPVDR) Martin Buchara, Senait Yifru, and Teshaynesh Michael Chapter 9: “Gender disparities in Agricultural Seltan helped in formatting the report and providing Production” by Arturo Aguilar (Instituto Tecnológico logistical support for travel and meetings undertaken Autónomo de México), Nik Buehren (Economist, in preparation of the report. xiii ABBREVIATIONS AND ACRONYMS AAU Addis Ababa University LEAP Livelihoods, Early Assessment and ADLI Agricultural Development-Led Protection project Industrialization LIAS Livelihoods Impact Analysis and AGP Agricultural Growth Program Seasonality BSG Benishangul-Gumuz MDGs Millennium Development Goals CEQ Commitment to Equity MoARD Ministry of Agriculture and Rural DAs Ethiopia’s Development Agents Development DHS Demographic and Health Survey MoFED Ministry of Finance and Economic EEPCO Ethiopian Electric Power Development Corporation MPI Multi-dimensional poverty index EGTE Ethiopian Grain Trade Enterprise NFEs Non-farm enterprises ERHS Ethiopian Rural Household Survey PASDEP Plan for Accelerated and Sustained ERSS Ethiopian Rural Socioeconomic Development to End Poverty Survey PPP Purchasing Power Parity FDI Foreign Direct Investment PSNP Productive Safety Net Program FTC Farmer Training Centers RCBP Rural Capacity Building Project GDP Gross Domestic Product RIF Recentered Influence Functions GoE Government of Ethiopia SNNPR Southern Nations, Nationalities and GTP Growth and Transformation Plan People’s Region HCES Household Income and UNICEF United Nations International Consumption Expenditure Survey Children’s Emergency Fund HH Household USD United States Dollars HICES Income and Consumption WMS Welfare Monitoring Survey Expenditure Survey HIV/AIDS Human Immunodeficiency Virus/ Acquired Immune Deficiency Syndrome Vice President: Makhtar Diop Senior Director: Ana Revenga Country Director: Guang Zhe Chen Practice Manager: Pablo Fajnzylber Task Team Leader: Ruth Hill xv EXECUTIVE SUMMARY I n 2000 Ethiopia had one of the highest poverty rates Development Goals (MDG), particularly in gender in the world, with 56% of the population living on parity in primary education, child mortality, HIV/ less than US$1.25 PPP a day. Ethiopian households AIDS, and malaria. While in 2000 only one in five experienced a decade of remarkable progress in wellbe- women in rural areas had an antenatal check-up, more ing since then and by the start of this decade less than than one in three women attended an antenatal check- 30% of the population was counted as poor. This Poverty up in 2011. Women are now having fewer births: the Assessment documents the nature of Ethiopia’s success and total fertility rate fell from almost seven children per examines its drivers. Agricultural growth drove reductions women in 1995 to just over four in 2011. At the same in poverty, bolstered by pro-poor spending on basic services time, the prevalence of stunting was reduced from and effective rural safety nets. However, although there is 58% in 2000 to 44% in 2011. The share of popula- some evidence of manufacturing growth starting to reduce tion without education was also reduced considerably poverty in urban centers at the end of the decade, struc- from 70% to less than 50%. Finally, the number of tural change has been remarkably absent from Ethiopia’s households with improved living standards measured story of progress. The Poverty Assessment looks forward by electricity, piped water, and water in residence asking what would be needed to end extreme poverty doubled from 2000 to 2011. in Ethiopia. In addition to the current successful recipe The pace of poverty reduction in Ethiopia has of agricultural growth and pro-poor spending, the role been impressive and particularly so when com- of the non-farm rural sector, migration, urban poverty pared to other African countries. Poverty incidence reduction and agricultural productivity gains for women measured by the population living below the interna- are considered. tional extreme poverty line of US$1.25 PPP fell from 55% in 2000 to 31% in 11 years. This puts Ethiopia 1. Trends in poverty and shared on par with Senegal with a GDP per capita (in PPP prosperity terms) double the size of Ethiopia. Only Uganda has had a higher annual poverty reduction during this Since 2000, Ethiopian households have experienced time. a decade of progress in wellbeing. In 2000 Ethiopia Ethiopia’s record of fast and consistent poverty had one of the highest poverty rates in the world, with reduction from 2000 to 2011 is robust to a number 56% of the population living below US$1.25 PPP a of sensitivity analyses that can be conducted on the day and 44% of its population below the national 2011 poverty estimates. Price deflators allow com- poverty line.1 In 2011 less than 30% of the popula- parisons to be made across time, but during periods tion lived below the national poverty line and 31% of high inflation such as experienced in Ethiopia from lived on less than US$1.25 PPP a day. 2008 to 2011, estimating the right deflator to com- The average household in Ethiopia also has pare living standards across time can be challenging. better health, education and living standards today than in 2000. Life expectancy increased and progress 1 In 1999/2000 less than 10% of countries that conducted household was made towards the attainment of the Millennium surveys recorded a poverty rate higher than Ethiopia. xvi ETHIOPIA – POVERTY ASSESSMENT TABLE 1: Ethiopia then and now: a decade of progress from 2000 to 2011 2000 2011 Percentage of the population: Living below the national poverty line 44 30 Living on less than US$1.25 PPP a day 56 31 Without education 70 50 With electricity 12 23 Piped water 17 34 Percentage of children under 5 years that are stunted 58 44 Percentage of rural women receiving an antenatal checkup 22 37 Life expectancy (years) 52 63 Total fertility rate 6 4 Sources: Ethiopia Demographic and Health Surveys, Household Income and Consumption Expenditure Surveys, World Development Indicators, Carranza and Gallegos (2011), Canning et al. 2014. The official numbers of poverty reduction appropri- distribution in rural areas. Additionally, low levels ately use a relatively high deflator and thus provide of inequality have, by and large, been maintained conservative estimates about the amount of progress throughout this period of economic development. that has been made. In urban areas, all measures of inequality show a Poverty reduction in Ethiopia has been faster substantial increase in inequality from 1996 to 2005 in regions where poverty was highest a decade and and a substantial reduction in urban inequality from a half ago. The proportion of households living in 2005 to 2011. In rural areas, all measures of inequality poverty has fallen in both rural and urban areas, with suggest there has been little change in inequality over stronger reductions in urban poverty since 2005. In time although inequality fell marginally from 1996 to 1996 poverty rates differed greatly between regions. 2005 and increased from 2005 to 2011. Nationally, For example, 56% of the population in Tigray and urban and rural trends offset each other and many SNNP were living in poverty compared to 34% of measures suggest inequality has stayed quite stable the population of Oromia. As a result of particularly from 2005 to 2011. However, measures of inequal- strong agricultural growth and improvements in ity that give more weight to poorer households show basic services, poverty reduction has been faster in that national inequality has steadily increased from those regions in which poverty was higher in 1996. 2000 until 2011. Consequently, the proportion of the population liv- This progress is not without its challenges, ing beneath the national poverty line has converged poverty remains widespread and the very poorest to around one in three in nearly all regions in 2011. have not seen improvements—to the contrary, even Geography still matters; for example those who live a worsening—of consumption since 2005, which in more remote locations are consistently poorer poses a challenge to achieving shared prosperity in than those living in closer proximity to markets and Ethiopia. Prior to 2005 the growth in consumption of services. the bottom 40% was higher than the growth in con- Ethiopia is one of the most equal countries in sumption of the top 60% in Ethiopia, but this trend the world as a result of a very equal consumption was reversed in 2005 to 2011 with lower growth rates Executive Summary xvii observed among the bottom 40 percent. Consumption arise largely because of the divergence between mon- growth benefited many poor households from 2005 etary poverty and the measure of living standards used to 2011, with the highest growth rates experienced by in the MPI. This divergence is due, in part, because the decile below the poverty line. However, the poor- the assets considered in the MPI do not include assets est decile did not experience an increase in consump- important in Ethiopia and the cutoff used in some tion. As a result reductions in poverty rates were not dimensions is too high to reflect recent progress. matched by reductions in poverty depth and severity from 2005 to 2011. The negative growth rate of the 2. Drivers of progress consumption of the bottom decile is robust to the choice of deflator and is a concerning trend. In the last ten years Ethiopia has experienced high There has been considerable progress in reduc- and consistent economic growth driven by high lev- ing the proportion of households experiencing els of public investment and growth in services and multiple deprivations in health, education, and agriculture. Since the early 1990s Ethiopia has pur- living standards at once, particularly in rural areas. sued a “developmental state” model with the objective In many cases, on any three indicators of deprivation of reducing poverty. The approach envisages a strong considered—such as access to sanitation and clean role for the Government of Ethiopia in many aspects water, education, and monetary poverty—the propor- of the economy and high levels of public investment to tion of rural households deprived in all three dimen- encourage growth and improve access to basic services. sions fell from four in 10 to less than one in 10 rural The model has been one of Agricultural Development- households. In the case of education and sanitation, Led Industrialization in which growth in agriculture the proportion of households with improved access is emphasized in order to lead transformation of the has increased, and increases have been largest among economy. Since 2004, Ethiopia’s economy has had disadvantaged groups. strong growth with annual per capita growth rates of However deprivation in some dimensions is 8.3% over the last decade (World Bank 2013). The still quite high, for example Ethiopia still has rela- contribution of agriculture to value added has been tively low rates of educational enrollment, access high throughout this period, however over time the to sanitation, and attended births. Four in five rural importance of agriculture has fallen (from 52% in 2004 households and two out of three urban households still to 40% in 2014) and the importance of the service sec- experience at least one out of three selected depriva- tor has increased (from 37% in 2004 to 46% in 2014). tions. Although much progress has been made, con- Growth was broad-based and has been the main tinued emphasis on investments in education. health. driver of reductions in poverty over the fifteen-year and improving living standards is needed. The need period from 1996 to 2011. Growth has been impor- for continued further progress is reflected in a high tant, but the average growth elasticity is quite low. and slowly moving Multidimensional Poverty Index Each 1% of growth resulted in 0.15% reduction in (MPI). In 2011, 87% of the population was measured poverty, which, although better than the sub-Saharan as MPI poor which means they were deprived in at African average, is lower than the global average. least one third of the weighted MPI indicators. This Growth in agriculture was particularly inclusive put Ethiopia as the second poorest country in the and contributed significantly to poverty reduction. world (OPHDI 2014). While the MPI is useful in Ethiopia has a rural, agricultural-based labor force: drawing attention to the need for further progress in more than four out of every five Ethiopians live in access to basic services in Ethiopia, it not a complete rural areas and are engaged in small-holder agricul- measure of deprivation in Ethiopia today. The higher tural production. Poverty fell fastest when and where rates of poverty and slow progress recorded in the MPI agricultural growth was strongest. For every 1% of xviii ETHIOPIA – POVERTY ASSESSMENT growth in agricultural output, poverty was reduced by and one sixth of poverty reduction respectively. In 0.9% which implies that agricultural growth caused Bangladesh (from 2000 to 2005) and in Cambodia reductions in poverty of 4.0% per year on average post in recent years, growth in light manufacturing accom- 2005 and 1.1% per year between 2000 and 2005. panied agricultural growth and helped spur further There is some evidence that manufacturing poverty reduction. growth and urban employment contributed to However, although the direct impact of non- poverty reduction in more recent years. Although agricultural growth on poverty reduction may nationally growth in manufacturing or services did not have been minimal, a more detailed examination contribute to poverty reduction, in urban Ethiopia, of the role of agricultural growth in reducing pov- manufacturing growth played a significant role in erty shows that increased access to urban centers reducing poverty from 2000 to 2011. For every 1% has been an important part of Ethiopia’s progress. of growth in manufacturing output, urban poverty fell While agricultural growth had a strong impact on by 0.37%. Although manufacturing only employs 3% poverty reduction on average, the positive impact of of the population nationally, the proportion of indi- agricultural growth was only found close to urban viduals employed in manufacturing in urban centers centers of 50,000 people or more. This indicates that is much higher. infrastructure investment and growth in non-agricul- The impact of service sector growth on poverty tural urban demand are essential complements to agri- reduction was small relative to growth in value cultural output growth to achieve poverty reduction. added by the service sector in national accounts. High food prices have ensured high returns to Growth in the service sector has been high in recent investments in agricultural production for many of years, but few poor households are employed in the Ethiopia’s rural households that are connected to service sector, and as a result only a tenth of the poverty markets. Food inflation has been high in recent years reduction in recent years took place among those in and this has shaped the nature of development and the service sector. While a shift to technical and pro- poverty reduction during this period. In 2011 food fessional occupations has helped increase consump- inflation was 39%, three times both the sub-Saharan tion at all consumption levels, this shift has mainly African average of 13%, and the approximate 12% contributed to increases in consumption among the food inflation in China and significantly higher than richest. However there is some evidence that agricul- the 27% food inflation in Vietnam. High prices and tural growth may drive poverty reduction in part by good weather ensured that investments in input-use encouraging rural service sector activity. Service sector brought high returns and gains for poverty reduction growth has been highest when and where agricultural during this period. Increased adoption of modern growth has been highest, and agricultural income is input-use in agriculture, such as fertilizer, has been the source of start-up funds for 64% of non-farm important in reducing poverty but this has only enterprises (often service sector). increased agricultural incomes and reduced poverty Overall, poverty reduction among rural, self- when good prices and good weather has been pres- employed, agricultural households accounts for ent. Over time an increasing proportion of poor the major share of poverty reduction from 1996 households have become self-sufficient in food or net to 2011. Structural change has not contributed producers and as a result high crop prices have helped much to poverty reduction during this time. This poverty reduction. is in contrast to some other economies in the region However high food prices have hurt agricul- and elsewhere. In Uganda and Rwanda agricultural tural households in the poorest decile that produce growth was accompanied by growth in the non-farm very little; high food prices perhaps offer an expla- service sector, which in turn accounted for one third nation for the pattern of broad-based growth with Executive Summary xix losses in the bottom decile observed in Ethiopia Public investment has been a central element from 2005 to 2011. The poorest decile are more of the development strategy of the Government of likely to report producing less than three months of Ethiopia over the last decade and since 2005 redis- consumption than other poor households, and were tribution has been an important contributor to pov- more likely to report suffering from food price shocks erty reduction. This coincides with the introduction than any other group. Broad based growth for the poor of large-scale safety net program in rural areas and the is aided by high food prices, but the high food prices expansion of basic services. Public spending is guided that benefit the majority of the agricultural poor in by the Growth and Transformation Plan (GTP) and is Ethiopia hurt the very poorest decile that continue particularly targeted to agriculture and food-security, to purchase much of their food. This group of house- education, health, roads, and water. Accordingly 70% holds needs compensatory interventions. The majority of total general government expenditure is allocated to (92%) of households own land, and as a result agri- these sectors. Education comprises a quarter of total cultural wage employment is more limited in Ethiopia spending followed by roads, agriculture, and health than in other countries. Those in non-agricultural at 20%, 15%, and 7% respectively. About half of unskilled wage employment are negatively impacted the agricultural budget is allocated to the Productive as wages take four to five months to adjust to food Safety Net Program (PSNP). price increases. As such high food prices do not help The Government of Ethiopia has reduced urban poverty reduction in large urban centers where poverty through the direct transfers provided in the majority of the labor force is in wage employment. the Productive Safety Net Program (PSNP) estab- Indeed, consumption growth was negative for many lished in 2005. The PSNP comprised 1% of GDP households in Addis Ababa from 2005 to 2011. Urban in 2010/11, and it is the largest safety net program households headed by someone with no education in Sub-Saharan Africa. The immediate direct effect of reduced their consumption by 12–14% as a result of transfers provided to rural households in the PSNP food price shocks experienced in the 12 months prior has reduced the national poverty rate by two percent- to the household survey. age points. The PSNP has also had an effect on pov- Consistently good rainfall has benefited agri- erty reduction above and beyond the direct impact cultural production and poverty reduction in recent of transfers on poverty. PSNP transfers have been years, but the dependence of agricultural growth shown to increase agricultural input-use among some on good weather highlights a key vulnerability. beneficiaries thereby supporting agricultural growth. Agricultural output is vulnerable to poor rains given Large-scale public investments in the provision the predominance of rain-fed production and the of basic services such as education and health have dependence of yield-increasing technologies (such as also contributed to poverty reduction both by con- fertilizer) on the weather. Since 2003 the proportion of tributing to growth and by preferentially increasing farmers experiencing crop losses greater than 30% has the welfare of the poor. Access to, and utilization not been more than one standard deviation above the of, education and health services has increased over average. Were a drought similar to 2002 to be experi- the last decade in Ethiopia. From 2006 to 2013 the enced in Ethiopia today, regression estimates suggest number of health posts increased by 159% and the poverty would increase from 30% to 51%. Increasing number of health centers increased by 386%. In the uncertainty around climate change will need to be education sector, from 2005 to 2011, the primary net managed through increased irrigation, development of attendance rate for 7–12 year olds increased from 42 drought-resistant seed varieties and strengthened finan- to 62%. Spending on services that are well accessed cial markets. Further diversification of the Ethiopian by poor households such as primary education and economy out of agriculture is also important. preventative health services is pro-poor. However xx ETHIOPIA – POVERTY ASSESSMENT spending is less progressive on programs where chal- The primacy of access to the labor market as a deter- lenges remain in ensuring utilization by poor house- minant of poverty and vulnerability in urban areas is holds, such as enrollment in secondary and tertiary particularly evident. education or use of curative health services. Individuals everywhere—in every woreda of The Government of Ethiopia has reduced Ethiopia—are vulnerable and as a result safety inequality and poverty through fiscal policy, net programs targeted only to specific rural wore- however because Ethiopia is a poor country this das will necessarily result in many vulnerable reduction in inequality has come about at a cost Ethiopians being left without support. This has to some households who are already poor. Poor implications for how safety nets function in Ethiopia, households pay taxes—both direct and indirect— suggesting that a move from geographically targeted although the amounts paid may be small. For most programs to systems that provide specific support to poor households, the transfers and benefits received individuals at defined points in time may be warranted are higher than the amount paid in taxes. As a result, as Ethiopia develops. fiscal policy brings about poverty reduction. Good fis- Further gains in reducing poverty are also cal policy is designed to meet a number of objectives, needed: in an optimistic growth scenario, extreme not just equity, and is also an important part of the poverty will be substantially reduced to 8%, but social contract. However it is worth noting that one not eradicated, by 2030. In an optimistic growth in 10 households are impoverished (either made poor scenario, all households will experience annual growth or poor households made poorer) when all taxes paid in consumption of 2.5%, which is higher and more and benefits received are taken into account. There equal than the growth Ethiopia experienced in the last are two means by which this negative impact could decade. In a less optimistic scenario annual consump- be reduced: (i) by reducing the incidence of direct tax tion growth rates might be lower, approaching the on the bottom deciles and increasing the progressivity annual consumption growth rate for the last decade of direct taxes, particularly personal income tax and of 1.6%. Or consumption growth rates may vary for agricultural taxes, and (ii) by redirecting spending on poorer and richer households as they did from 2005 to subsidies to spending on direct transfers to the poorest. 2011. Achieving 8% extreme poverty by 2030 requires both high and more equal growth than experienced 3. Ending extreme poverty in Ethiopia in the last ten years. Even very high rates of growth will not result in poverty falling below 12% if the Ending extreme poverty in Ethiopia requires pro- pattern of income losses of the bottom decile from tecting current progress. Many non-poor house- 2005 to 2011 is not reversed. Higher growth rates for holds in Ethiopia today consume only just enough the poorest households are also essential to ensuring to live above the poverty line making reductions shared prosperity. In the last five years incomes of the in poverty vulnerable to shocks: 14% of non-poor poorest 40% have, on average, not grown faster than rural households are estimated to be vulnerable average incomes. to falling into poverty. Weather shocks remain an In addition to continuing the successful mix of important source of risk in rural areas, and food agricultural growth and investments in the provi- price shocks have become increasingly important in sion of basic services and direct transfers to rural urban areas. However, although vulnerability does households, additional drivers of poverty reduction have a geographic footprint in Ethiopia today, it is will be needed, particularly those that encour- not fully determined by location of residence. Factors age the structural transformation of Ethiopia’s such as individual access to assets, or lifecycle events economy. Structural transformation will entail the are often defining features of vulnerable households. transition of labor from agricultural activities into Executive Summary xxi non-agricultural activities and it may also entail seasonality of NFEs, but many do report access to the movement of people from rural to urban areas. market demand as a major constraint. Interventions However, although non-farm enterprise ownership in to increase demand—e.g. continued improvements in rural areas and rural to urban migration are important rural accessibility and agricultural productivity—will realities in Ethiopia today, both have remained quite have the largest impact on increasing the vibrancy of limited. Neither have been significant contributors to this sector and its role in reducing poverty. However, poverty reduction as they have in some other coun- growth in this sector may be more likely in areas that tries in the region (for example the role of non-farm are more densely populated or proximate to such areas. enterprises in Rwanda and Uganda) and elsewhere (for Migration from rural to urban areas is an example the role of rural to urban migration in China). inherent component of the development process, Self-employment in non-farm enterprises but since 1996 rural to urban migration contrib- (NFEs) provides an additional income source for uted very little to poverty reduction in Ethiopia some poor, but the size of the sector is relatively because there was so little of it. About one in 10 small, constrained by limited demand for goods rural workers migrates in Ethiopia, in contrast to one and services in rural areas. In addition to being the in five rural workers in China. Migration has been primary sector of activity for 11–14% of the popula- beneficial for poverty reduction when it occurred. tion, a further 11% of rural households earn about On average, those that migrate experience substantial a quarter of their income from operating non-farm welfare benefits. The evidence is consistent with the enterprises in the service sector. In contrast, 67% notion that rural land policies and cash constraints of rural Rwandan households reported operating a limit the rate of migration. Land policy that has non-farm enterprise (one of the highest rates in the been so good for ensuring an equitable distribution region). While NFEs provide some secondary income of income in rural areas acts as a break on migration in rural areas and a source of income for those unable flows by prohibiting those planning on migrating to secure employment in rural towns, the contribution from liquidating their land. The costs associated with of this sector is small in comparison to other countries. migration and searching for a job in urban areas also Estimates from the 2011 Household Consumption limits the ability of liquidity-constrained poor house- Expenditure Survey suggest it comprises about 10% holds to invest in migration. Policies that make it of household earnings in Ethiopia. In comparison, easier to transfer land and that reduce the costs of job the rural non-farm sector is estimated to account search would likely increase migration. In addition for an average of 34% of rural earnings across Africa policies that protect more vulnerable groups as they (Haggbalde et al. 2010). migrate would increase the poverty reducing effects An initial assessment of constraints to NFEs of migration: young female migrants currently see suggests that limited demand constrains the role much lower welfare gains from migration than their of NFEs in rural income generation and poverty male counterparts. reduction. On the supply side, NFEs appear to Ethiopia is urbanizing and further agglomera- depend on agricultural income for inputs and invest- tion would likely enhance the pace of structural ment capital. On the demand side, they rely heavily transformation. As Ethiopia urbanizes so too does on increased local demand during the harvest period poverty. In 2000, 11% of Ethiopia’s poor lived in cit- to generate household income. As a result they are ies, but this rose to 14% in 2011. In Ethiopia, just most active during harvest and in the months imme- as in other countries, poverty rates fall and inequality diately thereafter and are not an important a source increases as city size increases, however poverty rates of income in the lean season. The need for capital in the two largest cities of Addis Ababa and Dire does not appear to be a major cause for the current Dawa are much higher than this trend would predict. xxii ETHIOPIA – POVERTY ASSESSMENT Improving welfare in large urban centers may in turn “necessity entrepreneurs” to upgrade to wage employ- make further agglomeration more likely by making ment and potentially reduce unemployment. cities more attractive places to live. However, addressing urban poverty will take Addressing poverty in large urban centers will more than encouraging employment. Increased thus become an increasingly important focus of safety nets to support those who do not participate development policy, and increasing the produc- in the urban labor market are needed. The elderly, tivity of urban work will be central to this. The disabled, and female-headed households are much nature of work is much different in larger urban poorer in urban areas. Households with disabled centers than in rural Ethiopia and small towns. Rates members and headed by the elderly are also more of self-employment and work in family enterprises vulnerable to shocks in urban areas than in rural areas. decrease and waged employment increases with city In part this is as a result of informal safety nets being size. In urban centers where waged employment is weaker in urban areas, but also in part as a result of higher, poverty rates are lower. However, as rates of inadequate urban safety nets. Direct transfers are waged employment increase so to do the number of only provided to rural households, with subsidies in people searching for these jobs, resulting in very high electricity, kerosene, and wheat in place to reach the rates of unemployment in the largest urban centers in urban poor. Although urban households do benefit Ethiopia. In Addis Ababa unemployment is strongly more than rural households from subsidies this is not correlated with poverty: nearly half of all households enough to compensate for the lack of direct transfers with an unemployed male in Addis Ababa live in pov- to urban households among the bottom percentiles. erty. Yet those with the lowest levels of education are Poverty, particularly urban poverty, would be reduced more often engaged in informal self-employment, out further if spending on indirect subsidies (on electricity, of necessity, rather than being unemployed looking kerosene and wheat) were converted to direct transfers. for a wage job. These individuals can be thought of An urban safety net can also have productive as choosing self-employment not because it is more benefits. Introducing a safety net in large urban profitable but because the cost of being unemployed centers will have a direct effect on poverty. Evidence while searching for waged employment is too high suggests that transfers can encourage income growth relative to the expected benefit. among recipients by increasing job search, increasing Poverty in large urban centers may be better the productivity of the self-employed and encourag- addressed by encouraging the entry and growth ing some to upgrade from necessity self-employment of larger firms rather than by encouraging self- to employment. employment. Supporting small-scale entrepreneurs Finally, although accelerating poverty reduc- can reduce poverty by increasing the productivity of tion will require looking beyond agriculture for those who currently earn marginal profits from self- sources of pro-poor growth, agricultural growth employment. However, supporting entrepreneurs that will remain an important driver of poverty reduc- have larger firms can also be poverty reducing—and tion in the near future, and ensuring that all often to a greater degree. High productivity entre- individuals in rural areas can participate in this preneurs earn substantial profits, but also employ growth is essential to poverty reduction. Female many workers, and contribute to higher overall wage farm managers in Ethiopia are 23% less productive levels through their demand for labor. As the value of than their male counterparts. They have less time to employment increases so does the value of job-search. spend on farm work and farm less land, more of which This encourages those who are entrepreneurs by neces- is rented. In addition, female managers obtain lower sity to search for and gain employment. Where job output from the productive factors that are employed search is costly, reducing its cost would also encourage compared to men. Differences in productivity arise, in Executive Summary xxiii TABLE 2: Poverty, inequality, wellbeing and sector of employment, 2000–2011 2000 2005 2011 National absolute poverty headcount (National Poverty Line) 44.2% 38.7% 29.6% Urban 36.9% 35.1% 25.7% Rural 45.4% 39.3% 30.4% International extreme poverty headcount (US$1.25 PPP Poverty Line) 55.6% 39.0% 30.7% Population (thousands) 63,493 71,066 84,208 Number of people living beneath the national poverty line (thousands) 28,064 27,523 25,102 Poverty depth (National Poverty Line) 11.9% 8.3% 7.8% Urban 10.1% 7.7% 6.9% Rural 12.2% 8.5% 8.0% Poverty severity (National Poverty Line) 4.5% 2.7% 3.1% Urban 3.9% 2.6% 2.7% Rural 4.6% 2.7% 3.2% Gini coefficient 0.28 0.30 0.30 Urban 0.38 0.44 0.37 Rural 0.26 0.26 0.27 Nutrtitional outcomes among children under 5 years of age* Stunting 58% 51% 44% Wasting 12% 12% 10% Underweight 41% 33% 29% Life expectancy (years) 52 63 Net attendance rate: Primary education (7–12 years of age)* 30.2% 42.3% 62.2% Urban 73.6% 78.8% 84.9% Rural 24.3% 38.8% 58.5% Immunization Rates (BCG, DPT1–3, Polio, Measles)* At least one shot 83.5% 76.0% 85.5% All vaccines 14.3% 20.4% 24.3% Proportion of households reporting shocks Food price n.a. 2.0% 19.0% Drought n.a. 10.0% 5.0% Job loss n.a. 1.0% 0.0% % crop loss (from LEAP) 22.4% 23.5% 13.8% Share of population living in urban areas 13.3% 14.2% 16.8% Proportion of households with at least one member engaged in Agriculture 78.8% 79.7% 78.4% Industry 3.4% 8.7% 8.0% Service 23.0% 20.8% 23.1% Notes: The data source is the HICE and WMS surveys unless otherwise stated. *Denotes that the statistic was calculated using the DHS. Some of the statics are taken from MOFED 2013 using these datasets. Life expectancy data is from the World Development Indicators. International extreme poverty rates estimated using a line of US$1.25 PPP per capita per day are taken from Povcalnet (June 2014). xxiv ETHIOPIA – POVERTY ASSESSMENT part because women are often relegated to, or choose, have been high, underscoring the dependence of low-risk low-skilled activities while men choose high- agricultural growth on increased urban demand for risk, high-value crops and engage in commercializa- agricultural products in a land-locked country such tion. Increasing women’s access to land, extension, as Ethiopia. However, the structural change in value oxen, and labor markets will help address gender- addition that has occurred during the last decade productivity differences, but policies that help change has not been fully matched by structural change in institutions and gender norms that keep female farm- employment and the analytical findings presented here ers in low-return activities are also needed. are consistent with the idea that further agglomeration In summary, the Government of Ethiopia’s focus through urbanization would help increase poverty on agricultural growth and investments in basic ser- reduction. This will require policies that favor the vices for all has ensured improvements in wellbeing entry and growth of firms, in addition to support to for many poor households in Ethiopia. The proportion self-employment in non-agricultural activities. Further of the population living below the national poverty urbanization and growth in non-agricultural sectors line fell from 44% in 2000 to 30% in 2011. Looking would continue to exert upward pressure on food forward, further investment in basic services are prices. This will need to be met by agricultural produc- required to ensure that Ethiopia continues to make tivity growth in order to keep labor costs competitive, additional, needed, progress in education, health and but high prices incentivize the required agricultural living standards. The predominance of agriculture as a investments. Although beneficial for many poor rural source of income for Ethiopia’s poor also suggests that households, high food prices carry costs for the urban agricultural growth will remain an important driver poor. Improving the fiscal position of poor urban of poverty reduction in the future. Poverty reduction households—such as through higher direct transfers from agricultural productivity increases has occurred or raising the minimum income above which personal in places with better market access when cereals prices income tax is levied—would help offset this effect. xxv INTRODUCTION I n 2005 the last Poverty Assessment documented Part I synthesizes progress since 1996, but with a wellbeing in Ethiopia from 1996 to 2000. It focus on progress since 2000. This part of the report showed that little progress had been made in reduc- starts with a focus on monetary poverty in Chapter ing poverty and that many households still experi- 1. It summarizes work undertaken by MOFED enced deprivation on many dimensions of wellbeing. (2013) in measuring poverty and assessing progress Since then life in Ethiopia has been transformed with in poverty reduction from 1996 and extends the work marked progress recorded in a number of surveys by undertaking some sensitivity analysis of the poverty and qualitative studies, particularly since 2003. This estimates, examining additional indicators of distribu- Poverty Assessment documents Ethiopia’s progress in tional change, profiling the bottom decile and simulat- reducing poverty from 1996 to 2011 with a particular ing future poverty trends. Chapter 2 takes as its focus focus on progress since 2000. progress in non-monetary dimensions of wellbeing Ethiopia has a wealth of data and surveys and in particular assesses the degree to which house- that have been used in this work. The core of the holds in Ethiopia experience multiple deprivations of analysis uses the series of Household Income and wellbeing. It draws on work undertaken in Carranza Consumption Expenditure Surveys (HICES) under- and Gallegos (2013) in assessing progress on many taken in 1995/6, 1999/2000, 2004/5 and 2010/11 non-income measures of wellbeing and explores why, (henceforth referred to as 1996, 2000, 2005 and given so much progress, the Multidimensional Poverty 2011). And it is from this series that the official Index ranks Ethiopia as the second poorest country in consumption aggregates and monetary poverty esti- the world. Chapter 3 examines another dimension of mates are derived. However additional nationally wellbeing: that of vulnerability. Wellbeing in Ethiopia representative surveys such as the annual Agricultural has historically been vulnerable to natural events Census Survey, the annual Medium and Large beyond individual control and the chapter examines Scale Manufacturing Census, the Ethiopian Rural the extent to which this is still true in 2011. Socioeconomic Survey of 2012 (representative of The overwhelming conclusion of Part I is that rural Ethiopia), and the Urban Employment and there has been substantial progress in wellbeing Unemployment Survey of 2012 (representative of in Ethiopia over the last decade. In Part II factors urban Ethiopia) are also used. Insights from the that have contributed to this progress are explored. Ethiopian Rural Household Survey panel from 1994 Chapter 4 examines the drivers of poverty reduction to 2009 are also drawn on. Analysis undertaken with through decomposition analysis, but also through the Demographic and Health Surveys collected in regression analysis of a panel constructed for zones 2000, 2005 and 2011 is also referenced. Throughout, in Ethiopia from many different nationally represen- the quantitative work is complemented with rich tative data sources. Agricultural growth emerges as a insights from the WIDE-3 (Wellbeing and Ill-being large contributing factor and the chapter explores the Dynamics in Ethiopia) longitudinal qualitative study nature of agricultural growth that has reduced poverty undertaken in 20 rural communities in Ethiopia from in further detail. Chapter 5 focuses on the role of fis- 1996 to 2013. cal policy in reducing poverty through redistribution. xxvi ETHIOPIA – POVERTY ASSESSMENT It summarizes work undertaken as part of a fiscal inci- activities and the movement of people from rural to dence analysis for Ethiopia using the Commitment urban areas. However, although non-farm enterprise to Equity framework. In documenting the impact of ownership and rural to urban migration feature in fiscal policy on inequality and poverty the chapter also Ethiopia today, they have contributed little to poverty points to a number of ways in which fiscal policy could reduction. Chapter 6 and Chapter7 examine the role be harnessed to reduce poverty further. of non-farm enterprises and migration respectively, In looking back to explain drivers of progress Part and document key constraints to both. Addressing II already points to a number of priorities for ending poverty in large urban centers will be an increasingly extreme poverty in Ethiopia. The importance of agri- important focus of development policy and Chapter cultural growth and good producer prices is empha- 8 considers the nature of urban poverty and work, sized, and the potential for further fiscal redistribution and strategies to further urban poverty reduction. is underscored. However, ending extreme poverty in Finally, although accelerating poverty reduction will Ethiopia will require more than repeating the past and require looking beyond agriculture, ensuring that all in particular it will likely require further structural individuals in rural areas can participate in agricul- transformation than has been observed in the last tural growth is essential to ensuring that the impact decade. Part III of the Poverty Assessment examines of agricultural growth on poverty reduction remains areas that have not been major contributors to national high. In this regard, Chapter 9 examines constraints poverty reduction in the past, but could be in the to the productivity of female farm managers and the future. Structural transformation entails the transi- degree to which policy can help alleviate some of this tion of labor from agricultural to non-agricultural inequality. 1 PROGRESS IN REDUCING POVERTY AND INCREASING WELLBEING, 1996-2011 1 S ince the early 1990s Ethiopia has pursued a of progress in wellbeing. As reported in the last “developmental state” model with the objec- Poverty Assessment (World Bank 2005), households tive of reducing poverty in Ethiopia. The experienced very little consumption growth between strategy has its genesis in the policy of Agricultural 1996 and 2000, and there had been little change in Development-Led Industrialization (ADLI), which the national poverty rate. From 2000 to 2011 the was first articulated in a paper by the then Ministry of wellbeing of Ethiopian households has improved on Planning and Economic Development in 1993. The a number of dimensions and poverty has fallen. In strategy was continued with some modifications in 2000 Ethiopia had one of the highest poverty rates in the Plan for Accelerated and Sustained Development the world, with 56% of the population living below to End Poverty (PASDEP) from 2005 to 2010 and, US$1.25 PPP a day and 44% of its population below since 2010, in the Growth and Transformation Plan the national poverty line.2 In 2011 less than 30% of (GTP), which will end in 2015. The approach envis- the population lives below the national poverty line. ages a strong role for the Government of Ethiopia in However this progress is not without its many aspects of the economy and high levels of public challenges. Ethiopia started from a low base and sector investment to encourage growth and improve attainment remains low on some dimensions. This access to basic services. chapter also documents that in recent years the very In the last ten years Ethiopia has experienced poorest have seen little improvement—even a wors- high and consistent economic growth driven largely ening—of their wellbeing. New challenges such as by growth in services and agriculture. Since 2004, food price shocks (during 2011 food price inflation Ethiopia’s economy has had strong growth with was 39%) have been particularly difficult for house- growth rates between 8–14%. GDP growth outpaced holds who purchase much of the food they consume. population growth (which has averaged about 3% Reversing this trend is essential for reducing extreme during this period) and Ethiopia recorded annual poverty and boosting shared prosperity. per capita growth rates of 8.3% over the last decade This chapter synthesizes and extends existing (World Bank 2013). The contribution of agriculture analysis that documents progress in poverty reduc- to value added has been high throughout this period, tion and wellbeing since 2000. It builds on MOFED however over time the importance of agriculture has (2014) which presents the national poverty estimates fallen and the importance of the service sector has for 2011, documents progress in reducing poverty increased. The contribution of agriculture to value over time and profiles households living beneath the added fell from 52% in 2004 to 40% in 2014 while national absolute poverty line. It summarizes the find- the contribution of the services sector increased from ings of MOFED (2014); conducts sensitivity analysis 37 to 46% during this time. However, although of the 2011 poverty estimates to document that the growth has been high, inflation has also been high progress in poverty reduction is robust; and extends and volatile at the end of this period. This chapter documents that, since 2000, 2 In 1999/2000 less than 10% of countries that conducted household Ethiopian households also experienced a decade surveys recorded a poverty rate higher than Ethiopia. 2 ETHIOPIA – POVERTY ASSESSMENT the distributional analysis conducted in MOFED in 2000. Life expectancy increased and progress was (2014) to further understanding on relative impor- made towards the attainment of the Millennium tance of growth and distribution changes in bringing Development Goals (MDG), particularly in gender par- about poverty reduction in Ethiopia during the last ity in primary education, child mortality, HIV/AIDS, decade. The next chapter focuses on the non-mone- and malaria. While in 2000 only one in five women tary dimensions of wellbeing. This and the following in rural areas had an antenatal check-up, more than chapter also incorporate findings from World Bank one in three women attended an antenatal check-up and other studies that have also documented prog- in 2011 (Figure 1.1). At the same time, the prevalence ress in wellbeing over this period (e.g. Carranza and of stunting was reduced from 58% in 2000 to 44% in Gallagos 2013; Woldehanna et al. 2011; Bevan, Dom 2011. The share of population without education was and Pankhurst 2013 and 2014; and UNICEF 2014). also reduced considerably from 70% to less than 50%. Finally, the number of households with improved living Recent progress in poverty 1.1  standards measured by electricity, piped water and water reduction in residence doubled from 2000 to 2011. This progress is documented further in Chapter 2. The average household in Ethiopia has better Trends in household consumption and mone- health, education and living standards today than tary poverty during this time also point to consistent FIGURE 1.1: Progress in health, education and living standards in Ethiopia from 2000 to 2011 Percentage of Women who had an Antenatal Check up during Prevalence of Stunting and Underweight their Most Recent Pregnancy, by Area of Residence (Children under 5 years of age) 80 77.12 70 70 67.79 69.76 60 50 60 40 50 30 40 36.94 20 30 10 22.17 24.49 20 0 2000 2005 2011 2000 2005 2011 Urban Rural Total Stunting Underweight Share of the Population with no Education, by Gender 100 Share of the Population with Electricity and Water 90 80 76.74 80 70 66.80 70 61.46 60 60 50 52.36 52.11 40 34.38 50 30 23.49 23.02 38.34 20 17.35 13.89 40 12.49 12.47 10 5.07 8.40 30 0 2000 2005 2011 2000 2005 2011 Male Female Total Electricity Piped water Water in residence Source: Ethiopia Demographic and Health Surveys, Carranza and Gallagos (2013). Progress in Reducing Poverty and Increasing Wellbeing, 1996-2011 3 BOX 1.1: Poverty measures In Ethiopia, absolute poverty is measured by comparing a household’s consumption per adult equivalent to the national poverty line defined as 3781 Birr in 2011. The poverty line indicates the minimum money required to afford the food covering the minimum required caloric intake and additional non-food items. The following three poverty measures are commonly used to assess poverty: Incidence of poverty (headcount index): The headcount index for the incidence of poverty is the proportion of individuals in the population living below the poverty line. Depth of poverty (poverty gap): The depth of poverty indicates how far, on average, poor households are from the poverty line. It captures the mean consumption shortfall relative to the poverty line across the whole population. It is obtained by adding up all the shortfalls of the poor (considering the non-poor are having a shortfall of zero) and dividing the total by the population. Thus, the depth of poverty shows the total resources needed per capita to eliminate poverty assuming that all poor individuals would obtain exactly the shortfall between their consumption and the poverty line. Poverty severity (squared poverty gap): The poverty severity takes into account the distance separating the poor from the poverty line (the poverty gap) as well as the inequality among the poor. Conceptually, poverty severity puts a higher weight on households/individuals, who are further below the poverty line. Source: World Bank’s Poverty Handbook. progress. In Ethiopia, poverty is measured by assessing of poverty with 56% of the population in Tigray and whether a household consumes enough to meet their SNNP living in poverty compared to 34% of the pop- basic food needs and other necessary expenditures. ulation of Oromia. Poverty reduction has been faster The national absolute poverty line is set at 3781 Birr in those regions in which poverty was higher and as a per adult equivalent per year in 2011 prices.3 Those result the proportion of the population living beneath falling below this line are considered poor (Box 1.1). the national poverty line has converged to around one The proportion of Ethiopians living beneath this line in 3 in all regions in 2011 (Figure 1.2 and Table 1.2). was reduced from almost one in every two Ethiopians The reason for this convergence is explored in Chapter in 1996 (46%) to less than 30% in 2011 (Table 1.1). 4. Section 1.4 in this chapter and analysis in Chapter The reduction mainly took place between 2000 and 2 point to the fact that poverty is still geographic in 2011. The proportion of households living in poverty nature, but it is geographical characteristics such as has fallen in both rural and urban areas, with stronger remoteness rather than regional location that strongly reductions in urban poverty since 2005. correlates with poverty. Poverty reduction in Ethiopia has been faster in regions where poverty was highest a decade and 3 3781 Birr in 2011 prices is equivalent to 1.24 USD PPP using the 2005 a half ago. In 1996 regions differed strongly in terms International Comparison Project. TABLE 1.1: Poverty headcount ratio for national poverty line (per adult) and the US$1.25 PPP poverty line (per capita) 1996 2000 2005 2011 National Poverty Line 45.5% 44.2% 38.7% 29.6% Urban 33.2% 36.9% 35.1% 25.7% Rural 47.6% 45.4% 39.3% 30.4% US$1.25 PPP Poverty Line 60.5% 55.6% 39.0% 30.7% Source: Own calculations using HICES 1996, HICES 2000, HICES 2005, HCES 2011 and Povcalnet (June 2014). 4 ETHIOPIA – POVERTY ASSESSMENT FIGURE 1.2: Poverty headcount by region this line as show in Table 1.1. The poverty headcount from 1996 to 2011 ratio dropped from 60.5% in 1996 to 30.7% in 2011. 70% The pace of poverty reduction in Ethiopia has 60% been impressive and particularly so when com- pared to other African countries. Poverty incidence 50% measured by the population living below US$$1.25 40% PPP dropped in Ethiopia from 55.6% in 2000 down 30% to 30.7% in 11 years (Figure 1.3). This puts Ethiopia 20% on par with Senegal with a GDP per capita (in PPP 10% terms) double the size of Ethiopia. Only Uganda has 0% a higher annual poverty reduction at almost 10% 1996 2000 2005 2011 compared to Ethiopia with 4% (Figure 1.4). Tigray Afar Amhara Oromia The reduction in the proportion of the Somali Benishangul SNNP Gambela Ethiopian population living in poverty was not matched by reductions in poverty depth and sever- Source: Own calculations using HICES 1996, HICES 2000, HICES 2005 and HCES 2011. ity from 2005 to 2011. From 2000 to 2005, poverty depth decreased from 13% to 8% and poverty severity from 5% to 3% (Table 1.3). In 2005 86 Birr per adult equivalent (in 1996 prices) was the average amount Poverty has also fallen when compared against of money that would have been required to lift poor an international line of extreme poverty. To facilitate households out of poverty. In the years between 2005 international comparisons of poverty rates an extreme and 2011, the shortfall did not change. Given the poverty line of US$1.25 PPP is used. Ethiopia has also substantial reduction in poverty incidence, this indi- exhibited strong poverty reduction in comparison to cates that those who are poor in 2011 are on average TABLE 1.2: Poverty headcount ratio for national poverty line by region National Poverty Line per adult 1996 2000 2005 2011 Tigray 56.0% 61.4% 48.5% 31.8% Afar 33.1% 56.0% 36.6% 36.1% Amhara 54.3% 41.8% 40.1% 30.5% Oromia 34.0% 39.9% 37.0% 28.7% Somali 30.9% 37.9% 41.9% 32.8% Benishangul-Gumuz 46.8% 54.0% 44.5% 28.9% SNNP 55.9% 50.9% 38.2% 29.6% Gambela 34.2% 50.5% 32.0% Harari 22.5% 25.8% 27.0% 11.1% Addis Ababa 30.2% 36.1% 32.5% 28.1% Dire Dawa 29.4% 33.1% 35.1% 28.3% Source: Own calculations using HICES1996, HICES 2000, HICES 2005 and HCES 2011. Progress in Reducing Poverty and Increasing Wellbeing, 1996-2011 5 FIGURE 1.3: Incidence of monetary poverty in Ethiopia compared with other African countries Incidence of Monetary Poverty in Ethiopia and other African Countries (Percentage of the population at US$1.25 PPP poverty line) 100 90 80 70 60 50 40 30 20 10 00 04 11 98 06 97 05 94 03 05 10 04 10 03 11 06 11 05 11 00 07 06 09 04 Ethiopia Ghana Kenya Lesotho Madagascar Malawi Nigeria Rwanda Senegal Tanzania Uganda Zimbabwe Source: World Bank WDI. FIGURE 1.4: Annual reduction of poverty further below the poverty line than those who were headcount at US$1.25 PPP poverty line poor in 2005. Poverty severity measures the gap of for selected countries with two poverty the consumption of the poor to the poverty line by measurements in the last decade putting more emphasis on the poorest. Poverty sever- –15% ity worsened in the same period despite the reduction in poverty. –10% 1.2  Sensitivity of poverty estimates 5% In practice, assessing trends in poverty across time is challenging, particularly during periods of high 0% inflation. One of the challenges in comparing trends in poverty over time is determining how to accurately 5% compare household consumption in one year with Malawi Nigeria Uganda Ethiopia Rwanda Senegal Madagascar another. The bundle of good and services that can be purchased with 3781 Birr (the national poverty Source: World Bank WDI. line) is quite different in 2011 than it was in 2005, TABLE 1.3: Poverty depth and severity from 1996 to 2011 (at national poverty line) Poverty Depth Poverty Severity 1996 2000 2005 2011 1996 2000 2005 2011 Rural 13.5% 12.2% 8.5% 8.0% 5.3% 4.6% 2.7% 3.2% Urban 9.9% 10.1% 7.7% 6.9% 4.2% 3.9% 2.6% 2.7% National 13.0% 11.9% 8.3% 7.8% 5.1% 4.5% 2.7% 3.1% Source: Own calculations using HICES 1996, HICES 2000, HICES 2005 and HCES 2011. 6 ETHIOPIA – POVERTY ASSESSMENT 2000, or 1995. Price deflators allow comparisons per adult equivalent in 2011 prices. A comparison to be made across time, but during periods of high of the two lines allows a survey-based deflator to be inflation, differences in price deflators estimated by constructed. different methods can be quite large. The CPI com- This section presents results from analysis pares prices of the same products of like quality over conducted to assess whether the positive trend a period time; but price collection is biased to urban in poverty reduction in 2005 to 2011 is sensitive markets (even though the price collection exercise to the choice of deflator. The results show that the that contributes to Ethiopia’s CPI is conducted in an official poverty numbers presented in Section 1.1 are impressive number of markets throughout the coun- conservative. The sensitivity of the poverty estimates try) and the basket of goods is not focused on goods to changes in spatial price deflation techniques and consumed by poor households. As a result survey- survey methodology are also discussed. based measures of prices focused on the consumption The proportion of people living beneath the bundle of poor households may suggest a different rate national poverty line would have been six percent- of inflation. The period of high inflation that Ethiopia age points lower had the CPI been used to deflate experienced from 2008 to 2011 results in the poverty prices across time. Had the same method of compar- trend between 2005 and 2011 being quite sensitive to ing poverty across time been used in 2011 as was used the choice of deflator. In addition any changes in the in 2000 and 2005 (converting all prices to 1996 prices methodology used to survey households or quantify using the CPI) the national poverty rate would have poverty can result in changes in estimated poverty fallen to 23.4% instead of the 29.6% rate estimated rates that are artifacts of the method of estimation using the HICES-based deflator. rather than underlying improvements in people’s lives. This is because the HCES-based measure of Conscious of the period of high inflation, food inflation used is lower than the food CPI the official 2011 poverty estimates use a different suggests. This lower HCES-based measure of food method of price deflation to that used in the 2000 inflation could reflect a lower rate of inflation for and 2005 survey. In 2000 and 2005 the poverty rate the goods consumed by the poor during this period. was estimated by converting all food and non-food However, it could also reflect that the quality of the consumption recorded in the 2000 and 2005 surveys food consumed by the poor fell over this period with to 1996 prices based on the CPI and comparing the the smaller increases in prices reflecting a lower qual- resulting consumption aggregate to the national pov- ity bundle of items (Table 1.4). erty line of 1075 Birr per adult equivalent in 1996 In contrast, the HCES data suggests non-food prices. For the 2011 poverty estimates, the cost of the inflation was higher than the non-food CPI would same bundle of goods used to construct the poverty suggest. In order to estimate non-food CPI, the food line in 1996 was re-estimated to generate a poverty share of total consumption for the bottom 25% of line in 2011 prices. The new poverty line is 3781 Birr the distribution was estimated using the 2011 HCES TABLE 1.4: HICES and CPI measures of inflation over 1996 to 2011 Food price index Non-food price index (1996:2010) (1996:2010) Ratio of food to non-food index HICES 1:3.06 1:4.17 1.4 CPI 1:3.70 1:2.78 0.7 Source: Own calculations using HICES 1996 and HCES 2011. Progress in Reducing Poverty and Increasing Wellbeing, 1996-2011 7 data. The resulting proportion (52%) was used to scale FIGURE 1.5: Food share per consumption the food poverty line to provide an absolute poverty percentile across time line. This was a fall in the proportion of food in total 80% consumption compared to that recorded in 1996. Figure 1.5 shows that the share of non-food expendi- 60% ture of the bottom quartile has increased over time, Food share but was quite constant from 2005 to 2011. In 1996 40% when the poverty line was set, the proportion of food in total consumption was 60%. This suggests either 20% higher non-food inflation in Ethiopia over this period than suggested by the non-food CPI (Table 1.4) or an 0% 1 21 41 61 81 increase in the quantity of non-food items consumed Consumption percentile by the poor.4 In the latter case, the recalculation of 1996 2000 2005 2011 the poverty line represents a change in the poverty line, an increase by 504 Birr per adult equivalent (in Source: Own calculations using the HICES1996, HICES 2000, HICES 2011 prices). 2005 and HCES 2011. Increased spending on rent captured in the household survey data explains some of the higher non-food inflation captured in the HCES estimate. and less to increases in real consumption. Reductions The proportion of spending on rent increased from in poverty are thus estimated to be lower. This is quite 22% in 2005 to 25% in 2011. Imputation of rents is remarkable given the fall in poverty using these con- difficult, especially in rural areas where formal rental servative estimates is already sizeable. Although the markets are uncommon. Household conditions have remaining analysis uses the appropriate HCES-based improved and this may be driving some of the increase deflator, results for the CPI-based deflator are also in imputed rent, but it may also be an increase in prices sometimes shown to test the sensitivity of trends to that does not reflect a real increase in the quantity or this assumption. quality of housing consumed. Using a survey-based Comparing consumption expenditures across deflator is thus the appropriate approach. Going for- space can also be challenging in a country as large ward, further work may be warranted on how best as Ethiopia where the cost of living varies from to quantify and include housing in the consumption one region to another. In order to address this chal- aggregate and poverty line for Ethiopia. It may also be lenge the national poverty estimates use a spatial price useful to revisit the food basket used to construct the index in order to measure all consumption expen- poverty line in case the food consumption patterns of diture in a consistent national price. In 1996–2005 poor households have changed significantly over time. price indices were constructed at the regional level, On aggregate the survey-data suggests a higher while in 2011 price indices were constructed at the rate of inflation than that measured using the “reporting level.” There are three reporting levels in CPI for 2005–2011, and as a result official esti- each region—rural, urban and other urban—so this is mates of poverty reduction during this period are a finer level of disaggregation used compared to pre- more conservative than if the CPI had been used. vious years. Had the previous level of disaggregation The CPI annual inflation of 7.9% is lower than the HCES-estimated annual inflation of 8.8%. By using the HCES rate, the official estimates attribute a larger 4 See Annex 1 for an analysis of the food share in total consumption part of nominal consumption increases to inflation across years when using the CPI to deflate consumption. 8 ETHIOPIA – POVERTY ASSESSMENT been used the estimated poverty rate would have been to previous years. The implications of the survey tim- two percentage points higher at 31.8%. This change ing are tested by comparing consumption estimated is driven by higher rural poverty rates. Urban poverty in the two months used in previous HICES rounds rates are lower. with data from the full 2011 sample. The results sug- Ethiopia is exceptional in comparison to many gest that the prior average of a month immediately other countries in the degree of comparability after harvest and a month in the lean season was a across the consumption surveys it has imple- reasonably good average for the whole year, but that mented over time. However, in 2011 there were two the new method marginally reduces the amount of changes to the survey. In previous HICES, data was consumption estimated thereby overestimating the collected at two points in the year—for one month rate of poverty in comparison to prior years. If the old immediately after harvest and for one month in the method had been used poverty may be one percentage lean season—and the consumption aggregate was a point lower (Table 1.5). Annex 1 also shows that the simple average of data collected at these two points change in method does not seem to change the distri- in time. In 2011 the Central Statistical Agency took bution of consumption either: the shape of the growth steps to improve the degree to which seasonality was incidence curve from 2005 to 2011 does not change reflected in the HICES by surveying one twelfth of when only the two months surveyed in both survey sampled households in each month throughout the rounds are used. The impact of the number of visits year. In addition the number of visits within a sur- on recorded consumption is more difficult to ascer- vey month changed in the last round. In the HICES tain. Reported consumption across subsequent visits surveys conducted in 1995, 2000 and 2005 eight in the 2005 HICES is examined to determine whether visits were made to each interviewed household in reported consumption appeared to fall. Evidence was each of the two survey months, while in the 2010/11 found that neither the number of consumption items HICES only two visits were made. The recall period nor the quantities decreased across visits. of the visits remained identical across survey rounds In summary, Ethiopia’s record of fast and (three or four days depending on the visit). However, consistent poverty reduction from 2000 to 2011 if reported consumption fell as the number of visits is robust to a number of sensitivity analyses that to the household increased this would make the con- can be conducted on the 2011 poverty estimates. sumption aggregates higher in the 2010/11 HICES Faster trends would have been observed had the pre- as a result of this methodological change. vious survey methodology been used and had the Poverty estimates do not appear very sensitive to CPI been used to allow comparisons across time, changes in survey methodology that can be tested, if while a slower trend of poverty reduction would have anything the methodology employed in 2011 results been observed had the previous method for deflating in higher estimates of poverty rates in comparison prices across space been implemented. The numbers TABLE 1.5: Test of sensitivity of poverty rates to new survey methodology Poverty rate Average consumption (percent of households living below Sample (Birr per adult equivalent) 3781 Birr per adult equivalent) New 12 month sample 5663 29.6 Old 2 month sample 5869 28.6 Adjusted Wald test of difference F(1, 25432) = 31.43*** F(1, 25432) = 8.69*** Source: Own calculations using HCES 2011. Progress in Reducing Poverty and Increasing Wellbeing, 1996-2011 9 are most sensitive to changes in the deflator used to households experienced broad-based growth of 2.4% assess progress across time. The official numbers of annually with slightly higher growth among the bot- poverty reduction use a relatively high deflator and tom decile. In urban areas, the household consump- thus provide conservative estimates about the amount tion of the bottom decile grew by 2.7% annually of progress that has been made. dropping down to almost zero at the 35th percentile before increasing again. The top 60% in urban areas The incidence of progress and 1.3  had high consumption growth of 4.4% annually shared prosperity (Figure 1.6). The high growth in consumption at all points in the consumption distribution resulted Reducing the number of people living below the in substantial poverty reduction, but as is discussed national poverty line is a significant measure of further below the pattern of consumption growth in progress. However, this is just one measure of how urban areas resulted in increasing urban inequality. Ethiopian households have fared in the last decade and High levels of broad-based consumption a half. Section 1.1 detailed how the depth and severity growth were also realized from 2005 to 2011, but of poverty increased during the period from 2005 to the very poorest households did not participate in 2011. This increase indicates that not all experienced this growth. Consumption in the bottom 15 percen- equal progress during this time. This section takes a tile contracted during this period while consumption closer look at changes in the distribution of consump- growth for the remaining of the distribution averaged tion in Ethiopia from 1996 to 2011 and sheds light 1.2% (Figure 1.6).6 The deteriorating consumption for on the role of growth and redistribution in bringing the poorest is reflected in the constant poverty depth about changes in poverty. between 2005 and 2011 while poverty incidence was reduced. While consumption growth in the bottom Growth incidence four deciles is similar in urban and rural areas, the pattern of growth is quite different in the top half of Although a small reduction in poverty was recorded the distribution. In rural areas the consumption of from 1996 to 2000, household consumption stag- the middle- and high-income population grew 1.4% nated within this same period (Figure 1.6). In rural annually while consumption contracted for the top areas (which dominate the national distribution given half of the urban distribution. The very poorest and Ethiopia has remained 85% rural throughout this those better off in urban areas did not fare well dur- time), the bottom half of the population benefited ing this period, despite large reductions in poverty from growth of 0.81% annually; the income of the as a result of the substantial consumption growth high middle-income population stagnated.5 The small experienced by poor households living just below the reduction in national and rural poverty during this poverty line. The contraction of consumption among period was as a result of the low but broad growth for the better-off urban population resulted in improve- the poor rural population. In urban areas the pattern ments in some measures of inequality. of progress was much different. The bottom 5% and The choice of deflator shifts the growth inci- the top 10% benefited from growth while the middle- dence curves along the growth-axis but does not income population in between had income losses of up to 2% per year (Figure 1.6). Negative growth for 5 Only the top 5% gained income but growth estimation at the extreme the 15% to 85% quantiles created additional urban quantiles is based on a very small sample and, thus, lacks credibility. poverty. 6 Note that the survey in 2011 was carried out in all months while previ- ous surveys were only administered in selected months. However, the From 2000 to 2005 a period of broad-based additional months included in 2011 do not introduce a bias into the growth in household consumption ensued. Rural growth incidence curve (see Annex 1). 10 ETHIOPIA – POVERTY ASSESSMENT FIGURE 1.6: Growth Incidence Curves with 95% confidence intervals nation-wide, urban and rural Growth Incidence, 2005–2011, HICES deflator Growth Incidence, 2005–2011, CPI deflator 5 5 4 4 3 3 Annual growth rate, % Annual growth rate, % 2 2 1 1 0 0 –1 –1 –2 –2 –3 –3 –4 –4 –5 –5 0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100 Expenditure percentile Expenditure percentile National Urban Rural National Urban Rural Growth Incidence, 2000–2005 Growth Incidence, 1996–2000 9 9 7 7 Annual growth rate, % Annual growth rate, % 5 5 3 3 1 1 –1 –1 –3 –3 –5 –5 0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100 Expenditure percentile Expenditure percentile National Urban Rural National Urban Rural Source: Own calculations using HICES 1996, HICES 2000, HICES 2005 and HCES 2011. change the qualitative finding that consumption to be sensitive to changes in the survey methodology. of the poorest deciles did not grow as fast. The However, it is worth noting that measurement error is HCES-based deflator assumes a smaller growth rate higher at the bottom and top of the consumption dis- in contrast to the CPI-based deflator. Accordingly, tribution. This is evident in the higher standard errors only the bottom 5% suffer from income losses based of the growth estimates in Figure 1.6. Despite the on the CPI-based deflator in contrast to the bottom higher measurement error, the consumption growth 15% using the HCES-based deflator. Independent of of the bottom and top deciles was significantly lower the choice of deflator, the shape of the income distri- than the growth in consumption in the middle of the bution stays constant and indicates that the poorest distribution during this period. households in Ethiopia did not fare as well as other The contraction of consumption observed in households from 1996 to 2011. The analysis presented the top half of the urban distribution from 2005 in Annex 1 indicates that this finding does not appear to 2011, reflects a contraction of consumption Progress in Reducing Poverty and Increasing Wellbeing, 1996-2011 11 FIGURE 1.7: Consumption growth was negative in Addis Ababa from 2005 to 2011 Growth Incidence, 2000–2005 Growth Incidence, 2005–2011 10 10 Annual growth rate, % Annual growth rate, % 5 5 0 0 –5 –5 –10 –10 0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100 Expenditure percentile Expenditure percentile Urban Addis Urban Non-Addis Urban Addis Urban Non-Addis Source: Own calculations using HICES 2000, HICES 2005 and HCES 2011. of households in Addis Ababa. Between 2000 and growth for the bottom 10%, the bottom 40% and 2005, the poor in Addis gained from growth as much the top 60%. as the poor in other urban areas and non-poor house- Prior to 2005 the growth in consumption of the holds fared, on average, better (Figure 1.7). From 2005 bottom 40% was higher than the growth in con- to 2011, consumption growth in Addis was worse sumption of the top 60% in Ethiopia, but this trend than in other urban areas. Incomes in Addis shrank was reversed in 2005 to 2011 with lower growth for the poor and for the rich alike. Given average rates observed among the bottom 40%. While consumption levels are higher in Addis Ababa than elsewhere; this explains the contraction in the top of the urban distribution (these are predominantly FIGURE 1.8: Average growth for the bottom 10%, bottom 40% and the top 60% from Addis Ababa residents) observed in Figure 1.6. The 1995 to 2011 particularly bad experience of Addis Ababa during 4 this period may reflect the fact that higher food prices were particularly observed in markets in Addis Ababa 3 and particularly hurt households in Addis Ababa that 2 are predominantly in wage employment and purchase 1 almost all of what is consumed. 0 –1 Shared prosperity –2 –3 Ethiopia’s progress in achieving shared prosper- 1996–2000 2000–2005 2005–2011 ity can be assessed using the growth incidence Bottom 10% Bottom 40% Top 60% analysis performed. Figure 1.8 and Figure 1.9 sum- marize the discussion in the previous paragraphs by Source: Own calculations using HICES 1996, HICES 2000, HICES depicting the average annual rate of consumption 2005 and HCES 2011. 12 ETHIOPIA – POVERTY ASSESSMENT FIGURE 1.9: Average growth for the bottom 10%, bottom 40% and top 60% for 1996 to 2011, by rural and urban 4 3 2 1 0 –1 –2 –3 National Urban Rural National Urban Rural National Urban Rural National Urban Rural 1996–2000 2000–2005 2005–2011 (HICES) 2005–2011 (CPI) Bottom 10% Bottom 40% Top 60% Source: Own calculations using HICES 1996, HICES 2000, HICES 2005 and HCES 2011. growth was generally very low between 1996 and 2000, from growth reversed abruptly with annual growth the bottom 10% were subject to highest consumption rates of –1.9%, 0.3% and 1.1% for the bottom 10%, growth of 1% annually followed by the bottom 40% bottom 40% and top 60% respectively. with 0.7% annually and the top 60% with stagnating The negative growth rate for the poorest is incomes (0.2% annually; Figure 1.10). From 2000 robust to the choice of deflator and is a concern- to 2005, growth was much more pronounced for all ing trend. Using the CPI-based deflator increases income categories: consumption growth increased to the growth rates of all income categories, but shows 3.5%, 2.4% and 1.9% for the bottom 10%, the bot- consumption losses around –0.5% for the poorest tom 40% and the top 60% respectively. In the last six decile. The negative growth for the poorest after 2005 years from 2005 to 2011 however, the pattern of gains is worrying. A more detailed analysis of the profile FIGURE 1.10: Gini Coefficient in Ethiopia and other African Countries 70 60 50 40 30 20 10 0 00 05 11 98 06 97 05 94 03 05 10 04 10 04 10 06 11 05 11 00 05 06 09 95 Ethiopia Ghana Kenya Lesotho Madagascar Malawi Nigeria Rwanda Senegal Tanzania Uganda Zimbabwe Source: World Bank WDI and authors’ calculations.a a Note that the Gini coefficient in WDI is calculated based on a parametric Lorenz curve. Only the Gini coefficients for Ethiopia are based on the survey data directly. Progress in Reducing Poverty and Increasing Wellbeing, 1996-2011 13 FIGURE 1.11: Gini and Theil index for national, urban and rural Ethiopia, 1996–2011 Gini coefficient Theil Index with α = –1 50 0.40 45 0.35 40 0.30 35 30 0.25 25 0.20 20 0.15 15 0.10 10 5 0.05 0 0 1996 2000 2005 2011 1996 2000 2005 2011 1996 2000 2005 2011 1996 2000 2005 2011 1996 2000 2005 2011 1996 2000 2005 2011 National Urban Rural National Urban Rural Within Between Source: Own calculations using HICES 1996, HICES 2000, HICES 2005 and HCES 2011. of the bottom 10% in the next section will help to In urban areas, all measures of inequality show understand the recent precarious decline of consump- a substantial increase in inequality from 1996 to tion for the poorest. 2005 and a substantial reduction in urban inequal- ity from 2005 to 2011. From 1996 to 2005, the urban Inequality top-income households experienced high consump- tion growth as shown by the Growth Incidence Curves The growth incidence analysis also provides some (Figure 1.6) and as a result the increase in inequality indication as to how inequality has changed over is reflected in the Gini and Theil (alpha=–1) measures time and the next paragraphs present information of inequality depicted in Figure 1.11. From 2005 to on summary measures of inequality. Box 1.2 out- 2011, the consumption of urban top-income house- lines the inequality measures used. The Theil index holds deteriorated while the consumption of house- with parameter α=–1 emphasizes inequality for lower holds from the 10th to 40th percentile increased. This incomes (Figure 1.11). decreased the share of income held by the top decile of Ethiopia is one of the most equal countries in households. Accordingly, the Gini coefficient dropped the world as a result of a very equal consumption strongly from 43.4% to 35%. distribution in rural areas. In comparison to other In rural areas, all measures of inequality sug- African countries, Ethiopia has the lowest inequality gest there has been little change in inequality as measured by the Gini coefficient (Figure 1.10). over time although inequality fell marginally Ethiopia’s Gini coefficient has consistently remained from 1996 to 2005 and increased from 2005 to below 30% while other countries have Gini coef- 2011. The Gini coefficient in rural areas decreased ficients around 40%. The Gini for rural Ethiopia is from 26.0 to 25.1 over the course of nine years from particularly low at 27%, and given that the majority 1996 to 2005. The slight reduction is explained by of the population is rural this contributes to a low the higher growth of incomes among the bottom national Gini. Urban Ethiopia has consistently higher 40% relative to the top 60% from 2000 to 2005. In inequality than rural areas, across measures and across the period from 2005 to 2011, inequality measured time, but in comparison to other countries it is still by the Gini coefficient remained at the same level. quite low at 35%. However, inequality measured by the Theil index 14 ETHIOPIA – POVERTY ASSESSMENT BOX 1.2: Inequality measures While poverty measures absolute deprivation with respect to a given threshold, inequality is a relative measure of poverty indicating how little some parts of a population have relative the whole population. In the context of monetary poverty, equality can be defined as an equal distribution of consumption / income across the population. This means that each share of the population owns the same share of consumption / income. The Lorenz Curve compares graphically the cumulative share of the population with their cumulative share of consumption / income. A perfectly equal consumption / income distribution is indicated by a diagonal. The other extreme is complete inequality where one individual owns all the consumption / income. These two (theoretical) extremes define the boundaries for observed inequality. The Gini coefficient is the most commonly used measure for inequality. A Gini coefficient of 0 indicates perfect equality while 1 signifies complete inequality. In relation to the Lorenz Curve, the Gini coefficient measures the area between the Lorenz Curve and the diagonal. The Theil Index measures inequality based on an entropy measure. A parameter α controls emphasis to measure inequality for higher incomes (larger α) or lower incomes (smaller α). The Theil index with parameter α=1 is usually called Theil T while using α=0 is called Theil L or log deviation measure. Relative and absolute income differences can be used to compare inequality dynamics over time. Usually, percentiles are used to compare incomes of different groups. For example, p90/p10 is the ratio (for relative incomes) or difference (for absolute incomes) of the average income in the 90th and 10th percentile. Source: World Bank’s Poverty Handbook. indicated that the poorest increased by about 10%, at Decomposing changes into growth and the same time that the income from the bottom 10% redistribution fell sharply, while the top 60% had higher income gains. The Theil (alpha=–1) measure suggests that Positive average consumption growth has contrib- rural inequality is higher now than it has ever been uted to poverty reduction, especially during 2000 (Figure 1.11). to 2005; and in rural areas since 2005. Poverty Nationally, many measures suggest inequal- reduction can be decomposed into a part that comes ity has stayed quite stable from 2005 to 2011. from an average increase in consumption across Inequality measured by the Gini coefficient remained the population (i.e. the consumption levels of all quite constant between 28 and 29% from 1996 to households increasing) and that which comes from a 2011 (Figure 1.11) and actually fell from 29.3% to change in the shape of the consumption distribution 28.5% between 2005 and 2011. Many measures of (i.e. consumption of the poorest growing faster than relative and absolute income differences also suggest consumption of the richest). Box 1.3 provides more very little change in national inequality (Figure 1.12). details. In the period from 1996 to 2000, the impact However, measures of inequality that give of growth on poverty was minimal (Figure 1.13) given more weight to poorer households show national the low rates of consumption growth during this inequality has steadily increased from 2000 until period. From 2000 to 2005, high average consump- 2011. The Theil (alpha = –1) suggests an increase in tion growth reduced poverty in both urban and rural inequality from 2000 to 2011. The most pronounced areas. In rural areas positive average consumption change in the relative income differences is for the growth resulted in substantial poverty reduction—a relative income of the top 10% in comparison to 20% change—from 2005 to 2011. Given the low the bottom 10%, which increased to above 360% average growth rates in urban areas from 2005, aver- given the contraction in consumption in the bot- age consumption growth in urban areas contributed tom decile. very little to poverty reduction. Progress in Reducing Poverty and Increasing Wellbeing, 1996-2011 15 FIGURE 1.12: Relative and absolute income differences between different income percentiles Relative Percentile Income Differences Relative Income Differences 400% 50% 350% 40% 300% 250% 30% 200% 150% 20% 100% 10% 50% 0% 0% p90/p10 p75/p25 p25/p50 p10/p50 p90/p50 p75/p50 Bottom 10%/ Bottom 40%/ Bottom 90%/ Top 90% Top 60% Top 10% Absolute Income Differences 3000 2500 2000 1500 1000 500 0 Top 90%– Top 60%– Top 10%–Bottom 90% Bottom 10% Bottom 40% 1996 2000 2005 2011 Source: Own calculations using HICES 1996, HICES 2000, HICES 2005 and HCES 2011. Redistribution increased poverty before 2005 role of redistribution in reducing poverty was mini- and then helped to reduce poverty. From 1996 mal. In urban areas the income distribution became to 2005, changes in the distribution of consump- more unequal during this period and this increase in tion were minimal in rural areas and as a result the inequality increased poverty by 6% from 1996 to 2000 BOX 1.3: Poverty, growth, and inequality Poverty, growth, and inequality are closely linked with each other, while at the same time the exact causal relationships are not yet well understood. Three stylized facts, though, help to summarize the current evidence. First, economic growth and changes in inequality are uncorrelated. Second, poverty generally declines as the economy grows. Third, the larger the initial inequality in a given country, the higher the growth rate needed to achieve the same amount of poverty reduction.* Poverty reduction can be formally decomposed into a growth component and a redistribution component. The partial effect of positive growth on poverty reduction is always positive. Thus, growth reduces poverty. However, the redistribution component can increase or decrease poverty reduction. Therefore, poverty can also decrease in a country with positive growth if the redistribution component is disfavoring poverty reduction. Source: Ferreira, 2010. *Note: Adapted from Ferreira 2010. 16 ETHIOPIA – POVERTY ASSESSMENT FIGURE 1.13: Growth and redistribution decomposition of poverty changes, 1996–2011 Poverty Decomposition, 1996–2005 Poverty Decomposition, 2005–2011 –25 –25 –20 –20 Poverty change, % Poverty change, % –15 –15 –10 –10 –5 –5 0 0 5 5 10 10 National Urban Rural National Urban Rural National Urban Rural National Urban Rural 1996–2000 2000–2005 HICES deflator CPI deflator Growth Redistribution Source: Own calculations using HICES 1996, HICES 2000, HICES 2005 and HCES 2011. and also 6% from 2000 to 2005 (Figure 1.13). Urban 10% of the income distribution in 2011 and 2005. growth was not pro-poor. In the most recent period As such no conclusion can be drawn as to whether the from 2005 to 2011, growth became strongly pro-poor same households have seen their livelihoods worsen with redistribution reducing poverty, particularly in or whether some households have become substan- urban areas. Redistribution reduced poverty by 15% tially poorer than any household was in 2005. Better in urban areas and 4% in rural areas. Although aver- understanding is imperative for designing policies to age urban consumption growth was minimal from address the worsening of the consumption distribu- 2005 to 2011, the 10th–40th percentile experienced tion at the bottom. Without such it will be difficult positive income growth and this is reflected in the for Ethiopia to eradicate extreme poverty and maintain contribution of redistribution to poverty reduction. low levels of inequality. In rural areas average growth rates of poor house- holds were also higher than average growth rates of Who are the poor and poorest 1.4  the non-poor (although this was not the case for the households in 2011? very poorest) and this also resulted in redistribution contributing to poverty reduction in rural areas. The lack of nationally representative panel data This analysis suggests that growth was pro- does not allow an analysis of which households have poor from 2005 to 2011, but some households in lost income, but in this section cross-sectional data Ethiopia today are substantially poorer than any is used to profile the characteristics of the poorest household was in 2005. From 2005 to 2011, growth households in Ethiopia. MOFED (2014) provides a was pro-poor based on a positive contribution of redis- comprehensive profile of poor households in Ethiopia tribution to poverty reduction. At the same time, the and this section repeats some of that work by focusing bottom 10% lost relative as well as absolute income. on characterizing the bottom 40% of the consump- Thus, pro-poor growth helped to reduce poverty, tion distribution—those households that are poor in especially for moderately poor households close to the 2011 and vulnerable to being poor. It also extends that poverty line. However, poverty in the bottom 10% was work by examining the characteristics of the bottom exacerbated. Without panel data it is not possible to 10% of the consumption distribution, given it is the say whether it is the same households in the bottom bottom 10% that has worsened in the last six years Progress in Reducing Poverty and Increasing Wellbeing, 1996-2011 17 FIGURE 1.14: Poverty headcount, depth and the same for children as they are for adults. The pov- severity for children and adults erty rate among children dropped from 49% in 1996 Poverty for Children to 32% in 2011, very similar to the magnitude of the 60% drop in the national poverty rate. Woldehanna et al. 50% (2011) report a remarkable increase in asset wealth among households with children in twenty sentinal 40% sites in Ethiopia from 2002 to 2011.In particular, 30% they document larger wealth increases for children in 20% households with uneducated mothers. Poverty depth 10% and severity are similar among children and adults. Thus, children are more often poor—but poverty is 0% 1996 2000 2005 2011 not more extreme among children. Children headcount Adult headcount Children depth As would be expected, individuals in the bot- Adult depth Children severity Adult severity tom 40% of the consumption distribution are very similar to those living beneath the poverty line Source: Own calculations using HICES 2000, HICES 2005 and HCES 2011. and as such are less educated, more remote, more engaged in agriculture, and in households with higher dependency ratios than those in the top (this is robust to choice of deflator). However, without 60%. Table 1.7 details the type of difference found nationally representative panel data it is not possible between those in the bottom 40% (excluding the to say whether the bottom 10% comprises the same poorest decile) and those in the top 60% over 1996 households that saw their consumption worsening to 2011. Full tables on the average characteristics of over the last six years. households in 1996, 2000, 2005 and 2011 are found Ethiopia is predominantly rural and poor in Annex 1. Households in the bottom 40% have Ethiopian households even more so. As a result the household heads that are significantly older and less national poverty profile is driven by the character- educated. They are larger and have larger proportions istics of the rural poor. A profile of the poor shows of unpaid workers, children and dependents. And are poor households being larger than non-poor house- predominantly engaged in agriculture and are more holds (Table 1.6, a fuller list of variables is provided likely to be engaged in agriculture than households in Annex 1). The household heads of poor house- in the top 60%. As a result households in the bot- holds are older, more often male than female, and are tom 40% own more agricultural assets: land as well more often married than non-poor household heads. as livestock, cattle, sheep or goats. Generally, poor households are more often engaged in In many respects households in the bottom agriculture (measured by the sector of the household 10% reflect these patterns, with limited school- head as well as the fraction of adults working in this ing, age, and dependency ratios increasing for sector). Differences in the urban poverty profile are these households as would be expected. Table 1.7 discussed in detail in Chapter 4. compares the bottom 10% to other households in the The positive correlation between dependency bottom 40% (i.e. those in the second, third and fourth ratios and poverty means that children are mar- decile) and shows that those in the bottom 10% have ginally more often poor than adults. In 2011 the even lower levels of education. Likewise those in the poverty rate among children less than 14 years old was bottom 10% are in households of larger size, more 32% compared to the national poverty rate of 30% dependents, and headed by more elderly heads than (Figure 1.14). The dynamics of poverty reduction are other households in the bottom 40%. 18 ETHIOPIA – POVERTY ASSESSMENT TABLE 1.6: Profile of the poor for 1996, 2000, 2005 and 2011 1996 2000 Mean Mean Mean Non- Sign. Mean Non- Sign. Variable Poor Poor Sign. Model Poor Poor Sign. Model Household is urban 0.11 0.18 0.11 0.15 Household size 6.51 5.68 *** *** 6.46 5.50 *** *** Household head age 45.45 44.36 *** 46.31 43.28 *** *** Household head is male 0.84 0.81 ** 0.81 0.80 ** Household head is married 0.85 0.83 ** 0.83 0.83 Household head level of formal education 0.36 0.79 0.38 0.87 Household head is literate 0.22 0.36 Household head’s year of education Household head works in agriculture 0.80 0.73 *** 0.76 0.73 ** Household head works in prof. services 0.00 0.01 0.01 0.02 Household head works in services & trade 0.04 0.07 0.10 0.10 *** Proportion of adults 0.52 0.59 0.54 0.57 in agriculture 0.92 0.85 *** 0.87 0.82 *** in education / health / social services 0.00 0.01 2005 2011 Household is urban 0.13 0.15 *** 0.14 0.18 *** Household size 6.90 5.31 *** *** 6.82 5.49 *** *** Household head age 45.66 42.73 *** * 46.41 43.30 *** ** Household head is male 0.84 0.80 *** 0.83 0.82 * Household head is married 0.85 0.82 *** 0.86 0.83 *** Household head level of formal education 0.56 0.86 0.62 1.06 Household head is literate 0.32 0.38 0.34 0.46 Household head’s year of education 1.36 2.11 1.50 2.62 Household head works in agriculture 0.80 0.76 *** 0.81 0.75 *** Household head works in prof. services 0.01 0.02 0.01 0.03 Household head works in services & trade 0.04 0.06 0.05 0.08 Proportion of adults 0.52 0.58 0.53 0.58 in agriculture 0.87 0.80 *** ** 0.85 0.77 *** ** in education / health / social services 0.03 0.04 0.04 0.05 Source: Own calculations using HICES 2000, HICES 2005 and HCES 2011. Significance values are calculated for each year separately including region fixed effects. Model significance includes all variables and regional fixed effects. *, **, and *** indicate significance level of probit regres- sion at 10%, 5%, and 1% levels correcting for the clustered nature of the errors. Progress in Reducing Poverty and Increasing Wellbeing, 1996-2011 19 TABLE 1.7: Differences in characteristics between consumption percentiles 1996 2000 2005 2011 Bottom Bottom Bottom Bottom 10% vs. Bottom 10% vs. Bottom 10% vs. Bottom 10% vs. Bottom bottom 40%+ vs. bottom 40%+ vs. bottom 40%+ vs. bottom 40%+ vs. Variable 40%+ top 60% 40%+ top 60% 40%+ top 60% 40%+ top 60% Age of household head +++ +++ + +++ ++ +++ Household head is male +++ Household head is married +++ +++ Years of schooling of household head --- --- --- --- Number of household members +++ +++ +++ +++ +++ +++ +++ Highest years of schooling in household -- -- --- Proportion of unpaid workers +++ +++ +++ +++ +++ Proportion of children (<12) ++ +++ +++ +++ +++ Proportion of dependents + +++ ++ +++ +++ + +++ Proportion of children (6–18) in school - -- --- --- Proportion of children (6–12) in school --- -- --- --- Proportion of children (13–18) in school --- Occupation of household head: agriculture +++ - +++ +++ +++ Occupation of household head: manufacturing --- Occupation of household head: construction --- --- Occupation of household head: mining/energy -- ++ - Occupation of household head: social services --- -- --- Occupation of household head: professional --- --- --- --- --- services Occupation of household head: services and trade --- ++ --- --- Household lives in an urban area + --- --- --- --- Floors in households made of hard/solid material Household has a private toilet --- -- --- Household owns livestock +++ Household owns cattle --- -- + - +++ Household owns sheep or goats - + +++ +++ Household owns chickens -- +++ Household owns beehives Household owns land --- + +++ +++ Household located between 1–2km to all weather road Household located more than 2km to all weather ++ +++ road Food gap of at least 9 months +++ + (continued on next page) 20 ETHIOPIA – POVERTY ASSESSMENT TABLE 1.7: Differences in characteristics between consumption percentiles (continued) 1996 2000 2005 2011 Bottom Bottom Bottom Bottom 10% vs. Bottom 10% vs. Bottom 10% vs. Bottom 10% vs. Bottom bottom 40%+ vs. bottom 40%+ vs. bottom 40%+ vs. bottom 40%+ vs. Variable 40%+ top 60% 40%+ top 60% 40%+ top 60% 40%+ top 60% Food gap of 6–8 months + +++ ++ + Food gap of 3–5 months +++ +++ + +++ Food gap < 3 months --- --- --- --- Household shock: drought ++ ++ Household shock to food prices (price rise) ++ Household shock: illness or death of member Non-agricultural household --- --- Months covered by crop production for agr. --- --- --- --- - --- hh: 10+ Months covered by crop production for agr. -- + hh: 7 to 9 Months covered by crop production for agr. +++ +++ ++ +++ hh: 4 to 6 Months covered by crop production for agr. +++ +++ +++ ++ ++ hh: 0 to 3 Source: Own calculations using HICES 2000, HICES 2005 and HCES 2011. Notes: Grey boxes indicate lack of data for estimation. +, ++ and +++ indicate a significant positive difference for the poorer group at a significance level of 10%, 5% and 1%. -, -- and --- denote negative differences accordingly. +Bottom 40% refers to those in the bottom 40% of the consumption distribution, without including the bottom 10%. The food gap refers to the number of months during which the household faced a food shortage during the last 12 months. However this is not always the case and on some and 2011 in Ethiopia and with these gains, the his- key characteristics such as sector of occupation and toric disparity in enrollment rates between those in the remoteness, households in the bottom decile are second to fourth decile and wealthier households were no different from other households in the bottom no longer present. However the difference between the 40%. For example individuals in the bottom 10% bottom decile and those in the 2nd to 4th increased dur- are no more likely engaged in agriculture than others ing this time. Although barriers to school enrollment in the bottom 40% and as such are not likely to own may not be a concern for many in the bottom 40% in more agricultural assets. They are no more remote today’s Ethiopia, they are still significant among the than others in the bottom 40%. lowest decile. Woldehanna et al. (2011) document Children in households in the bottom decile are that parental poverty, a need to work, and illnesses less likely to go to school even though there is no are the main reason for non-attendance. longer a difference between other households in the Broad based growth for the poor is aided by bottom 40% and those in the top 60% of the con- high food prices given that many of these house- sumption distribution. Children of those in the sec- holds are net-sellers, but the poorest decile has a ond to fourth decile were historically less likely to be in significantly higher proportion of marginal agricul- school, however this changed in 2011. Large gains in tural producers (households that produce very little, school enrollment have been achieved between 2005 i.e. not more than three months of consumption Progress in Reducing Poverty and Increasing Wellbeing, 1996-2011 21 needs) than other households in the bottom 40% wages do not adapt. Nationally, few are wage laborers and as a result they are more likely to report wel- (just 8% of household heads) given the widespread fare losses as a result of price shocks. Although the ownership of land (92% of households own land), but HICES have typically not collected information on in urban areas many more are in wage labor. Headey the size of agricultural production and income of a et al. (2012) examines the degree to which unskilled household, from 2000 to 2011 households were asked wages (maids, guards, and casual labor) in 120 urban to state the number of months the household will be centers and rural towns in Ethiopia adjusted to the sustained from crop production or the income from price increases observed in 2008 and 2011. They show crop sales. The number of months covered by crop that in the short run, wages do not adjust, but that in production was categorized whether produced crops the longer run they do. It is quite likely that for many will last for 10 or more months of consumption, for the HCES survey was conducted before wages had 7–9 months of consumption, for 4–6 months of con- fully adjusted to food price increases in 2011. sumption or whether agricultural production covered only three or less months. In 2011, the production of Outlook: Ending extreme poverty 1.5  agricultural households in the bottom 10% signifi- in Ethiopia cantly more often covered not more than three months compared to the second to third deciles. Thus, a large Is Ethiopia on a path to end extreme poverty by number of highly marginal agricultural producers exist 2030? The Government of Ethiopia has set ambitious among the bottom 10%. As such the higher food prices poverty targets in recent years, and it is likely to do so that were present in 2011 specifically hurt the bottom in the second Growth and Transformation Plan, which 10% with their large number of marginal agricultural will be implemented from 2016. This section reports producers. Indeed the poorest 10% of households simulation results to examine what poverty rates may were more likely to report experiencing a food price be in Ethiopia in the next 5, 10, and 15 years if recent shock compared to other households in the bottom patterns of growth continue. Three scenarios are iden- 40%, who were no more likely to report experiencing tified in which the average growth rate is estimated a food price shock than households in the top 60% of based on recent history:7 the consumption distribution. The higher proportion of marginal agricultural  Pessimistic scenario assumes annual average con- producers in the poorest decile offers an explana- sumption growth of 0.8%. This is the consump- tion as to why this decile fared badly during a tion growth recorded from 2005–2011 when period of high food inflation such as was experi- using HICES deflator; enced between 2005 to 2011; providing insight  Intermediate scenario assumes annual average into the pattern of consumption growth observed consumption growth of 1.6%. This is the aver- during this period The high food prices that benefit age annual consumption growth recorded from the majority of the agricultural poor in Ethiopia hurt 2000–2011 when the HICES deflator is used; and the very poorest decile that continue to be marginal  Optimistic scenario assumes annual average agricultural producers and net consumers. This may consumption growth of 2.5%. This is the aver- well explain why consumption growth was negative age annual consumption growth recorded from for the bottom decile during 2005 to 2011 while other 2005–2011 when the CPI deflator is used. poor households experienced positive consumption growth during this period. 7 The label of the scenarios (pessimistic to optimistic) refers to the aver- age assumed growth rate. It does not imply that growth distribution Higher food prices can also impose consider- across the population is ‘better’ in the optimistic scenario than in the able welfare costs for those in waged employment if pessimistic scenario. 22 ETHIOPIA – POVERTY ASSESSMENT FIGURE 1.15: Poverty incidence based on FIGURE 1.16: Poverty statistics in 2030 simulations with percentile-specific growth compared to current values for different (solid lines) and average growth (dotted line) simulations Poverty Incidence Simulation Poverty Statistics in 2030 35% 35% 30% 30% 25% 25% 20% 20% 15% 15% 10% 10% 5% 5% 0% 0% Incidence Depth Severity 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 2025 2026 2027 2028 2029 2030 Current pessimistic Intermediate Optimistic (05–11 CPI) pessimistic per percentile pessimistic average Source: Own calculations using HCES 2011. Intermediate per percentile Intermediate average Optimistic per percentile Optimistic average Assuming the same growth rate for all house- Source: Own calculations using HCES 2011. holds in the population, household consumption is multiplied by 1 plus the growth rate for each year in the simulation. However, as growth incidence curves indicate, the assumption of average growth across the 2030 may be an overly optimistic projection. First, population is usually violated. Therefore, for each this scenario assumes growth rates averaging 2.5%, scenario household consumption is also simulated which is the annual growth rate of the past six years, using percentile-specific growth rates inferred from whereas best estimates indicate growth rates have been the past. This step is repeated for each year of the 0.8%. Secondly, this assumes equal growth rates across simulation. all percentiles and when the more unequal growth In the most optimistic scenario, extreme pov- rates observed in recent years are allowed for, national erty will be substantially reduced to 8%, but not poverty rates would be reduced to between 12% and eradicated, by 2030. Figure 1.15 and Figure 1.16 20%. When percentile-specific growth rates are used, present results from the simulation analysis detailing each scenario has a plateau around 20% where pov- the trend in poverty rates over time under the scenarios erty will not decrease for several years. Unequal dis- considered. Poverty rates in 2030 range between 8 and tribution of growth, with higher growth rates for the 21%. The most optimistic scenario entails reducing third and higher percentiles create an enlarging gap extreme poverty to 8% by 2030 which would be a in incomes between the bottom 20% and the rest of remarkable achievement given 44% of the population the population. Once the top 80% crossed the poverty was in poverty in 2000. line, it takes several years for the bottom 20% to exit Achieving this low level of extreme poverty poverty as well. requires both high and more equal growth than Two further scenarios are tried: region-specific experienced in the last ten years. The scenarios point growth and urban-rural growth rates with migra- to a number of reasons why 8% extreme poverty in tion. Both scenarios use percentile-specific growth Progress in Reducing Poverty and Increasing Wellbeing, 1996-2011 23 FIGURE 1.17: Simulation of income shares (relative to average income) for bottom 10%, bottom 40%, and top 60% using percentile-specific growth rates. Pessimistic (05–11 HICES) Intermediate (00–11 HICES) 10000 12000 8000 10000 8000 6000 6000 4000 4000 2000 2000 0 0 2011 2013 2015 2017 2019 2021 2023 2025 2027 2029 2011 2013 2015 2017 2019 2021 2023 2025 2027 2029 Optimistic (05–11 CPI) 14000 12000 10000 8000 6000 4000 2000 0 2011 2013 2015 2017 2019 2021 2023 2025 2027 2029 Bottom 10% Bottom 40% Top 60% Source: Own calculations using HCES 2011. Note: Dotted lines indicate the unchanged share if average growth is assumed for all percentiles. rates in the pessimistic scenario. Results are shown in FIGURE 1.18: Alternate simulations Figure 1.18. Using region-specific growth rates does Poverty Statistics in 2030 not change the headcount poverty predicted, even 40% though it has an impact on where poor households will be concentrated in 2030. Allowing for specific 30% urban-rural growth rates and migration also predicts a similar headcount poverty rate will be attained in 20% 2030. This is because very little gain from migration has been modeled, and because urban consumption growth was no better than rural consumption growth 10% in the pessimistic scenario. With policies to encour- age urban development and poverty reduction this 0% Incidence Depth Severity may change. Current Urban-rural specific growth If recent trends in the distribution of growth Pessimistic Region specific growth continue, relative income inequality will increase. Assuming average growth across the population Source: Own calculations using HCES 2011. 24 ETHIOPIA – POVERTY ASSESSMENT conserves relative incomes within the population rates from 2005 to 2011. The bottom 40% are slightly (Figure1.17; dotted lines). If growth is distributed better off but will still lose up to three percentage across the population as in past periods, relative points from an initial share of 55% to 52% using the incomes can change over time. In fact, the incomes optimistic growth scenario. This emphasizes the chal- of the bottom 10% will deteriorate from 35% of the lenge Ethiopia faces to bring about structural change average income in 2011 to 21% in 2030 using growth to ensure high growth, and growth for the poorest. 25 MULTIDIMENSIONAL POVERTY IN ETHIOPIA 2 2.1 Introduction (WMS) found that more than half of Ethiopians believed their standard of living was worse now than Over the last decade, and particularly since 2005, it was 12 months ago. Is this because some dimen- substantial improvements in education and health sions of wellbeing are not improving or because there investments have been observed as a result of a is a coincidence of deprivations for some people that concerted effort by the Government of Ethiopia to is not changing, or even worsening? Or because gains improve access to health care and educational ser- in welfare over time were reversed in the year prior vices. From 2006 to 2013 the number of health posts to the survey? increased by 159% and the number of health centers This chapter analyzes multidimensional pov- increased by 386% (Federal Ministry of Health 2013). erty in Ethiopia focusing on selected dimensions Immunization coverage increased from 14% in 2000 of education, health, and command over resources to 24% in 2011, modern contraceptive use increased as well as gender equality and access to sources of from 6% to 27%, and the percentage of women age information. Poverty is multidimensional in nature 15–49 years who received antenatal services increased and the dimensions of deprivation considered here from 27% to 34% (EDHS 2011). Infant mortality are those that are reflected in a number of multidi- declined from 97 deaths per 1,000 in 2000 to 59 mensional measures of wellbeing and deprivation, deaths per 1,000 in 2010, and under-five mortality such as the Human Development Index and the MPI. decreased from 166 deaths to 88 deaths per 1,000. In There is however a disagreement on how to measure the education sector, the primary net attendance rate poverty using these deprivation dimensions. The two for 7–12 year olds increased from 42 to 62% from alternative approaches are scalar indices of multidi- 2005 to 2011,. mensional poverty (e.g. Alkire and Santos 2010) and Despite apparent progress on many aspects of the dashboard approach (Ravallion 2011) that con- wellbeing, progress has not been observed to the siders deprivation in each dimension one by one. same degree in the multi-dimensional poverty index Lugo and Ferreira (2012) propose a middle ground (MPI). The MPI, which measures those who are poor to capture the interdependency across dimensions on many dimensions (see Box 2.1), declined by about without aggregating the dimensions into one index 10% compared to the 33% decrease in monetary and this approach is followed here. Levels and poverty recorded during the same period (Carranza trends in non-monetary dimensions of wellbeing and Gallegos 2013). In 2011, 87% of the population are documented and then multidimensional poverty was measured as MPI poor, which means they were in Ethiopia over the last decade is explored using deprived in at least one third of the weighted MPI Venn diagrams. This work draws on a background indicators. This put Ethiopia as the second poorest paper prepared for the Poverty Assessment (Ambel country in the world (OPHDI 2014). et al. 2014). It allows an assessment of progress on Moreover, in 2011 more people reported they each aspect of deprivation and also on the degree to felt worse off than one year previously, than in which individuals experience deprivation in many 2000 or 2005. The 2011 Welfare Monitoring Survey dimensions at once. 26 ETHIOPIA – POVERTY ASSESSMENT BOX 2.1: The WIDE-3 qualitative research program The Wellbeing and Ill-being Dynamics in Ethiopia (WIDE) research program covers 20 communities in Ethiopia selected as exemplars of different types of rural livelihood systems. WIDE is a qualitative research program that began in 1994 when qualitative village studies were undertaken in rural communities selected to be part of the long-run quantitative panel (the Ethiopian Rural Household Survey) undertaken by Addis Ababa University and the University of Oxford. The communities were chosen as exemplars of the main rural agricultural livelihood systems found in the four main regions of Ethiopia at the time. These 15 communities were visited for a second time in 2003 and five new sites were also visited: three new agricultural sites, which had been added to the ERHS panel in 1999 as exemplars of new agricultural livelihood systems, and two pastoralist sites. From 2010 to 2013 the twenty communities have been visited again as part of WIDE-3. These visits were conducted in phases: from 2009–2010 six communities about which the team had most information (three drought-prone and food-insecure and three self-sufficient); from 2011–2012 eight drought-prone and food insecure sites were researched; and from 2012–2013 the remaining six rain-secure higher potential sites were visited. Findings from WIDE 3 are reported in this chapter and in other chapters of the Poverty Assessment. The research focuses on communities and takes a long-term perspective on development, which allows longer-run and inter- dependent changes in the community to be identified. The research looks in particular at how community processes have been affected by government activities and broader modernization processes and changes in the communities’ environments. The case studies documented by the research in the twenty communities, provide additional insights and triangulation of the quantitative findings of nationally representative surveys documented throughout this report. Source: Bevan et al. 2011. The chapter documents considerable progress the weight of any other dimension of living standards, on many aspects of wellbeing and in reducing the and (iii) the cutoff used in some dimensions is too high proportion of households experiencing multiple to reflect recent progress in Ethiopia. This highlights deprivations at once. The proportion of the popula- that while the MPI is useful in drawing attention to tion experiencing multiple deprivations has declined the need for further progress in access to basic services particularly rapidly in rural areas. Experiencing depri- in Ethiopia; it not on its own, a complete measure of vation in many dimensions at once makes it difficult deprivation in Ethiopia today. to escape poverty, and thus this progress is also a positive indication that poor households may be in a Trends in non-monetary 2.2  better position to see improvements in welfare than dimensions of wellbeing in earlier years. However the analysis also documents that there The indicators considered in the analysis reflect are still a large number of households experiencing dimensions considered in most multidimensional one out of any three selected deprivations. Four indices of wellbeing and deprivation. In addition in five rural households and two out of three urban they have relevance to the country’s policies and the households still experience at least one out of three MDGs. A total of 12 indicators are identified covering selected deprivations. This contributes to a high and education, health, command over resources, gender slowly moving MPI. However fundamentally, the equality, access to information, and perceived wellbe- higher rates of poverty and slow progress recorded in ing. Table 2.1 presents the definitions of the indicators the MPI arise because of the divergence of monetary and how households are counted as deprived in each poverty and the measure of living standards used dimension. The HCE and WMS were used rather than in the MPI. The disconnect arises because: (i) the the Demographic and Health Survey (DHS) because choice of assets considered in the MPI does not reflect it allows dimensions of wellbeing to be compared to Ethiopian realities, (ii) electricity access is given twice the monetary poverty data presented in Chapter 1 Multidimensional Poverty in Ethiopia 27 TABLE 2.1: Deprivation indicators, definitions and their use for urban and rural overlap analysis Deprivation Indicator Definition: A household is deprived when… Urban Rural Education of school- …at least one child, age 7–15, in the household is not currently attending school.   aged children Education of female …at least one girl child, age 7–15, in the household is not currently attending   school-aged children school. Health facility quality …the household reported dissatisfaction with at least one health facility visit, or did   not use a health facility due to cost, distance, quality, or other reasons. Health facility access …the household is located more than 5 km away from the nearest health facility  (clinic, health station, hospital, or health post). Institutional birth …at least one child, age 0–4, in the household was not born in a health facility.  Female circumcision …at least one girl child, age 0–14, in the household has been (or will be) circum-   cised. Assets …none of these assets are owned by the household: fridge, phone, radio, TV,   bicycle, jewelry, or vehicle. Source of information …the household does not own a TV, radio, or phone.   Drinking water …a safe drinking water source—piped water, protected water source, or rainwa-  ter—is not used by the household. Sanitation …an improved toilet—private flush toilet or private pit latrine—is not used by the   household. (i.e., A household that uses an improved toilet facility, but it is shared, is deprived.) Living standard per- …the household believes that its overall standard of living is worse (or worst now)   ception compared to 12 months ago. Below poverty line …the household’s real total consumption expenditure per adult is lower than the   poverty line (3781 Birr). Note: The columns Urban and Rural specify which indicators are used in the overlap analysis for urban areas and rural areas. Access to a health facility and access to safe water are present for nearly all urban households, so they are not considered in the overlap analysis. Institutional birth is not considered in overlap analysis for rural households because almost all children in rural areas aged 0–4 years were not born in a health facility. and to data collected on perceived changes in well- the deprivation incidence has changed over time for being. However the trends in wellbeing that were all indicators. In both rural and urban areas there documented in Carranza and Gallegos (2013) using have been significant reductions in the proportions the DHS are reported where relevant. While the HCE of deprived populations in all dimensions and the and WMS surveys conducted in different years are in declines from 2001 to 2011 and 2005 to 2011 were general similar in their coverage and representative- found significant (at the 1% level) for almost all indi- ness, some content differences exist and Tables A1 and cators.8 The result is in line with other recent studies, A2 in Annex 2 provide more details and compare the for example, Carranza and Gallegos (2013) using the indicators used in this study to indicators selected for 2000, 2005 and 2011 DHS, and the WIDE-3 quali- the MDGs and the MPI (See Box 2.3). tative studies on Wellbeing and Ill-being Dynamics There have been significant reductions in many dimensions of deprivation from 2000 to 2011, 8 The sole indicator that captures households’ cultural practices is only particularly in rural areas. Table 2.2 presents how available in 2011 and thus no trends can be confirmed. 28 ETHIOPIA – POVERTY ASSESSMENT TABLE 2.2: Proportions of deprived households, 2000–2011 Urban Rural Absolute Absolute Absolute Absolute Change Change Change Change Deprivation 2005– 2000– 2005– 2000– Indicator 2000 2005 2011 2011 2011 2000 2005 2011 2011 2011 Education of school-aged 0.26 0.26 0.16 –0.10*** –0.10*** 0.83 0.80 0.58 –0.22*** –0.25*** children Education of school-aged 0.22 0.23 0.14 –0.09*** –0.08*** 0.79 0.72 0.46 –0.26*** –0.33*** girls Health facility quality — 0.74 0.67 –0.07*** — — 0.83 0.77 –0.06*** — Health facility access 0.02 0.01 0.04 0.03*** 0.02** 0.62 0.56 0.32 –0.24*** –0.30*** Institutional birth — 0.59 0.52 –0.07*** — — 0.98 0.96 –0.02*** — Female circumcision — — 0.19 — — — — 0.30 — — Assets 0.33 0.21 0.12 –0.08*** –0.21*** 0.86 0.69 0.53 –0.16*** –0.33*** Source of information 0.33 0.25 0.15 –0.10*** –0.18*** 0.86 0.79 0.62 –0.17*** –0.25*** Drinking water 0.08 0.07 0.05 –0.02* –0.03** 0.82 0.77 0.59 –0.18*** –0.23*** Sanitation 0.54 0.51 0.53 0.02 –0.01 0.93 0.83 0.45 –0.37*** –0.48*** Living standard perception 0.33 0.29 0.54 0.26*** 0.22*** 0.38 0.39 0.51 0.11*** 0.12*** Below national poverty line 0.36 0.35 0.26 –0.09*** –0.10*** 0.45 0.39 0.30 –0.09*** –0.15*** Source: Own calculations using HICES 2000, HICES 2005, and HCES 2011. Notes: Deprivation indicators are specified for 2011. Details on these 2011 indicators and notes about the minor differences in definitions for the 2000 and 2005 indicators are included in Appendix A (Table A1 and A2). The two education indicators are defined for those households with at least one school-aged child (aged 7–15) and with at least one school-aged female child, respectively. The institutional birth indicator is defined for those households with at least one child aged 0–4. The female circumcision indicator is defined for those households with at least one female children aged 0–14. The “Change” columns show the coefficient estimate for the difference in proportions from 2000 (or 2005) to 2011. The asterisks indicate the significance level: *** p<0.01, ** p<0.05, * p<0.1. in rural Ethiopia. Their finding confirms that of the considerable progress in education enrollment and Alkire and Roche (2013) results. The salient trends outcomes using the DHS data. The Net Attendance are documented further here. Rate for primary education increased from 30% in 2000 to 62% in 2011. As a result the share of the Education population between 15 and 24 years old able to read at least part of a sentence increased five-fold from 8 The proportion of households with a child between to 36%, the share of the population aged six years the ages of seven and 15 that had a child out of and over with no education declined from 69% to school fell from 83% in rural and areas and 26% in 46 percent, and the average years of schooling of urban areas to 58% in rural areas and 16% in urban this population increased from 4.0 to 4.5 years. The areas. Progress would have been even more dramatic Human Opportunity Index report for sub-Saharan had the age range been restricted to younger children. Africa shows that Ethiopia has increased both the scale The WIDE-3 study found that nearly all 7-year-olds of education enrollment and the degree to which it were enrolled in school in the six study sites visited in is inclusive, favoring disadvantaged groups (Dabalen 2013. Carranza and Gallegos (2013) also document et al. 2014). This corroborates the findings of Khan Multidimensional Poverty in Ethiopia 29 et al. (2014), which show that spending on education The proportion of individuals without access to is higher in more historically disadvantaged areas. improved sanitation fell from 93% in 2000 to 45% The challenge for Ethiopia is increasing atten- in 2011 and the proportion of individuals without dance rates at higher grades and ensuring that access to improved water sources fell from 82% to 59 the quality of education received is adequate. The percent. Not only has the proportion of households WIDE-3 studies document that student attendance with improved sanitation increased, the expansion has is often irregular and interrupted, and that shortages favored underserved groups (Dabalen et al. 2014). of teachers and textbooks are common (Bevan, Dom, Government policies for rural areas seem to and Pankhurst 2013, 2014). Although attendance rates have been particularly successful in ensuring better for secondary and post-secondary schooling are higher access to private toilet facilities and safe drinking than they were in the past, they are still low. Distance water sources. Indeed the WIDE-3 study (see Box 2.1 to secondary school can be a major constraint: for those for more details) found that in all eight of the food that live too far from a secondary school, attendance insecure communities included in the study, provision entails living away from home, which is expensive. of health services, drinking water, and education had expanded considerably since 2003. In the six commu- Health nities with agricultural potential, access to safe water had greatly improved although there were problems of Since 2000, life expectancy has increased by one poorer access for remote residents. Health extension year per year from 52 years to 63 years in 2011. workers had been effective at making people aware of Child mortality and morbidity rates improved: the hygiene and environmental sanitation. share of children under five who were reported to However the substantial progress should not have an episode of acute respiratory infection, fever, or overshadow the substantial challenge that remains: diarrhea fell from 45 to 27% (Carranza and Gallegos almost six out of 10 rural households still do not 2013). Although health outcomes such as these are have access to improved water sources. There were not calculated using the WMS and HCES, health people in seven of the eight food insecure WIDE-3 inputs such as access to health centers, access to clean sites without all-year access to clean water. In the six water and sanitation are recorded. These are used as sites with agricultural potential, access to safe water proxies for improvements in health outcomes in the had improved though there were problems of poorer multidimensional analysis that follows. access for remote residences and slow responses to The proportion of households living further water point failures. Those without access to clean than five km from the nearest health facility almost water were more at risk of infections and there had halved between 2005 and 2011, from 56% to 32 been an outbreak of cholera in a number of the food percent, driven largely in part by the establishment insecure communities. of health posts and a system of health extension workers. There have also been improvements in Command over resources and access to access to quality health facilities in rural areas, but information progress has not been as fast as improvements in access and improvements have been slower in urban In addition to the higher levels of monetary areas. This is probably due to the challenge associated expenditure documented in Chapter 1, Ethiopian with improving health facility quality in this short households today hold more assets than a decade period of time. ago. Although taking a narrowly defined set of There has also been vast improvement regard- assets— fridge, phone, radio, TV, bicycle, jewelry, or ing sanitation and drinking water in rural areas. vehicle—Table 2.2 documents considerable progress 30 ETHIOPIA – POVERTY ASSESSMENT in asset ownership, with deprivation (defined as own- households own fewer assets, and have less access to ing none of these assets) among rural households fall- information and safe drinking water. ing from 86% to 53% in rural areas and from 33 to 12% in urban areas. The proportion of households 2.3  Overlapping deprivations owning livestock has also increased from 74% in 2000 to 83% in 2011, largely representing an increase in In this section we look at the extent to which those ownership of sheep, goats, and poultry. The propor- who are monetarily poor are also deprived in other tion of households owning sheep or goats increased dimensions. Poverty is a multidimensional concept from 39% in 2000 to 51% in 2011 and the proportion and this allows us to take a broader look at who is poor of households owning chicken increased from 47% in Ethiopia today. Additionally this analysis can pro- in 2000 to 55% in 2011. This speaks well for vulner- vide some insight into the likelihood that households ability in Ethiopia as households with more assets are will be able to move out of poverty. In other contexts likely more able to withstand shocks. This is discussed it has been shown that when a household experiences in further detail in Chapter 3. multiple deprivations at once it is more difficult for However, ownership of some types of assets is the household to move out of poverty. still quite low: even in 2011 62% of rural house- The analysis examines sets of three indica- holds did not have access to a TV, phone, or radio. tors reflecting the three dimensions of depriva- This is despite the proportion of households owning tion in health, education and command over a mobile phone increasing by almost fifteen times resources (measured as monetary poverty) used between 2005 and 2011 (Carranza and Gallegos in the UNDP Human Development Index. In the 2013). The lack of access to these “information absence of health outcomes, improved sanitation is assets” limits access to outside information. This first used as a measure of access to health. A Venn in turn limits the horizons and aspirations of rural diagram is presented for sanitation, education, and households, especially those in remote places. The monetary poverty in Figure 2.1. Circle areas in the 2005 Ethiopia Poverty Assessment documented the diagram represent the proportion of the population high degree of remoteness for many households in with the deprivation. Intersection areas represent the Ethiopia. Although there have been improvements proportion of the population with two, or all three, in this regard, this data suggests that for many access deprivations. Changes in deprivation are observed in to outside sources of information remains difficult. two ways: the change in the size of the circles and the This issue is discussed further in Box 2.2. Bernard et change in the overlap area. Improvements in terms al. (2014) show that increased access to information of reduction in a deprivation over time are observed that increases the aspirations window of households when the circle for the deprivation under consider- in a remote location in Ethiopia has a substantial ation is smaller now (2011) than it was before (2000 impact on investments made in children’s education. or 2005). Likewise, improvements in reduction in This suggests that this aspect of deprivation also has multiple deprivations are illustrated as the three circles substantial economic costs. move apart.9 However, given the low base from which Ethiopia started, deprivation on some dimensions 9 For example the top left panel in Figure 2.1 shows that 50% of the is still high, particularly in rural areas. Rural house- population was poor, 83% had a child out of school and 93% did not have improved access to sanitation; 43% of poor households were also holds still have more children out of school; about without sanitation and this area is depicted by the intersecting red and one-third of them still live farther than five kilometers green circles. The poverty rate fell in rural areas from 2000 to 2011 and this is depicted by the red circle decreasing in size. Fewer poor households from a health facility, and the practice of female cir- have children out of school or lack improved sanitation and as a result the cumcision is still more prevalent in rural areas. Rural red, green and blue circles also move apart from 2000 to 2011. Multidimensional Poverty in Ethiopia 31 BOX 2.2: Aspirations and educational investments in rural Ethiopia Aspirations express goals or desired future states. They have been shown to have an important influence on behavior and economic choices such as choice of occupation and educational investments. Aspirations evolve over time in response to life experience and circumstances and are largely formed by observing the outcomes of individuals whose behaviors they can observe and with whom they can identify. Researchers from the International Food Policy Research Institute and the University of Oxford measured aspirations in Doba woreda in 2010–11 (Bernard et al. 2014). Doba is historically a food-insecure woreda and the majority of residents are subsistence farmers growing sorghum and maize. Aspirations were measured in four dimensions: income, wealth, social status and children’s educational attainment. For each of these dimensions, respondents were asked what level on this dimension they would like to achieve. Initial levels of aspirations of income and wealth were found to be quite high, 20 times current levels of income and wealth. Most parents aspired to provide 13 years of education for their children. Similarly high aspirations for education attainment have been found in other sites in Ethiopia. The site selected was very remote—only 13% of surveyed individuals left the woreda more than once a month—and as a result exposure to experience of people outside of the local area was limited. To ascertain whether increasing information would impact aspirations and the sense of control people have about their lives, selected individuals were randomly invited to watch documentaries about people from similar communities who had succeeded in agriculture or small business, without help from government or NGOs. Immediately after the screening of the documentaries, aspirations had increased among those who watched the documentary. Aspirations on the educational attainment of children had increased and six months later aspirations were still higher. Individuals who saw the documentary were also less likely to agree with fatalistic explanations that attribute poverty to luck and fate after six months. These changes in aspirations and attitudes had a significant effect on the investment behavior of households, particularly with regard to education. The number of children enrolled in school increased by 15% among those who had watched the documentary. These households also had more savings and took more credit (Bernard et al. 2014). Using data collected as part of the evaluation of the Agricultural Growth Programme (AGP), analysis has also been undertaken to explore rural household’s attitudes to fate and whether they believe that their efforts can be rewarded (Hoddinott et al. 2014). Specifically, it applies the notion of “locus of control.” Someone with an internal locus of control believes that their actions influence events, while someone with an external locus of control believes that forces beyond their control largely shape life’s events. Those with stronger internal locus of control tend to be better-educated individuals, among those who are married and to be men. There is little difference across age until 60 years after which it declines. Although no experimental evaluation has been conducted to assess the causal relationship between locus of control and investment (as in the case of aspirations), individuals with higher internal locus of control were found to be more likely to ensure that girls are attending school and more willing to buy fertilizer and improved seeds. Source: Bernard et al. 2014, Hoddinott et al. 2014. In 2000 nearly all rural households that Progress in reducing multidimensional depri- experienced deprivation in monetary wellbeing, vation in urban areas is also evident, but prog- education or sanitation experienced it on mul- ress has been slower despite lower initial levels of tiple dimensions, but by 2011 this was no longer deprivation. The proportion of households deprived the case. In 2000 four out of 10 rural households in monetary wellbeing, education, and sanitation is were deprived in all three dimensions considered; much lower in urban areas. Only 9% of households while in 2011 only one in 10 rural households was thus deprived.10 The contrast between rural 10 The proportions of deprivations used to construct all the Venn diagrams Ethiopia in 2000 and 2011 is shown quite dramati- in this section are presented in Appendix II. For example, Table A2.2 provides information used in Figure 2.1. The first three rows of Table cally in the top panel of Figure 2.1. The reductions A2.2 reflect the deprivation incidence for each indicator separately. The in deprivation on all three dimensions also resulted first three rows are similar to the values in Table 2.2 (single deprivation analysis). However, in the Venn diagrams the deprivations rates are in a reduction in the number of households simul- calculated after having dropped those observations with missing data taneously deprived. for any of the three indicators. 32 ETHIOPIA – POVERTY ASSESSMENT FIGURE 2.1: Monetary, education and households were sanitation deprived and in 2011 this sanitation deprivation in urban and rural had fallen only slightly to 47 percent. areas, 2000–2011 When considering a wide variety of indictors, Rural considerable progress has been made in reducing 2000 2011 the proportion of individuals deprived in multiple dimensions on account of improvements in health, water, sanitation, education and poverty. Figure 2.2 depicts the degree to which deprivations overlap with monetary poverty for other indicators. A similar pic- Urban ture of progress emerges when considering access to 2000 2011 healthcare or improved water in place of sanitation, and also when considering other sets such as poverty, information and sanitation. A greater incidence over- lapping deprivations is observed in rural areas when a measure of the quality of the health services received is also incorporated. However, overall, there has been Money poor Children out of school No improved sanitation considerable reduction in the number of individuals experiencing more than one out of any three depriva- Note: Details for these diagrams are in Annex 2. tions. Experiencing deprivation in many dimensions at once makes it difficult to escape poverty, and thus this progress is also a positive indication that poor were deprived in all three dimensions in 2000 and this households may now be in a better position to see fell further to 3% in 2011. Urban households have improvements in welfare. a less substantial reduction in part due to their bet- The analysis also points to substantial depri- ter initial access to education and higher enrolment vations remaining. The proportion of individuals rates but also in part due to slow progress in improv- that are not deprived in any dimension has increased ing sanitation in urban areas. In 2000 51% of urban substantially over time (Figure 2.2), but it consistently FIGURE 2.2: Evolution of overlapping deprivations over time, 2000–2011 (rural Ethiopia) 100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% 2000 2011 2000 2011 2000 2011 2000 2011 2000 2011 Poverty-education- Poverty-education- Poverty-education- Poverty- Poverty-information- sanitation health health quality health-water sanitation None One Two Three Source: Own calculations using HICES 2000 and HCES 2011. Multidimensional Poverty in Ethiopia 33 remains at about one fifth of the rural population in education and health were recorded. However, the and one third of the urban population. This means living standards dimension of the MPI records both that almost four out of five households in rural areas a very high proportion of people deprived in this and two out of three households in urban areas are dimension and very slow progress over time: 84% of deprived for any set of three dimensions considered. people are deprived in this dimension in 2011 and Given this progress why does the MPI remain only 8% improvement was recorded between 2000 high and slow moving? The fundamental reason and 2011. This is despite fast improvement recorded why Ethiopia’s MPI is very high and moving slowly in monetary poverty, used in the overlap analysis to is that it records higher deprivation and slower reflect command over resources. progress on its dimension of living standards than The disconnect between the level and progress is reflected in monetary poverty. The MPI cap- in monetary poverty and the MPI living standards tures three dimensions of deprivation—education, dimension arises because the choice of assets con- health, and living standards—as detailed in Box 2.3. sidered in the MPI are not those best placed to Figure 2.3 depicts the contribution of each dimension reflect asset accumulation among Ethiopian house- to the overall MPI for Ethiopia, and the change in the holds, considerable weight was given to electric- number of people deprived in that dimension over ity access (a dimension on which Ethiopia fares the period 2000 to 2011. Substantial improvement poorly), and the cutoff used in some dimensions BOX 2.3: The Multidimensional Poverty Index Poverty is multidimensional, and the Multidimensional Poverty Index (MPI) tries to capture this by considering overlapping deprivations suffered by people at the same time. The index identifies deprivations across education, health and standard of living. It counts an individual as multi-dimensionally poor if they suffer deprivations in a third of the weighted indicators. The index can be deconstructed by region, ethnicity and other groupings as well as by dimension. Almost 1.5 billion people in the 91 countries covered by the MPI—more than a third of their population—live in multidimensional poverty; that is, with at least 33% of the indicators reflecting acute deprivation in health, education, and standard of living. This exceeds the estimated 1.2 billion people in those countries who live on US$1.25 a day or less The data underlying the index is the data in the Demographic and Health Surveys (DHS). Specifically, the indicators considered and the weights they receive are as follows. Education (each indicator receives a weight of one sixth in the total): • Years of schooling: if no household member has completed at least five years of schooling; and • Child school attendance: if any school-aged child is not attending school in years 1 to 8. Health (each indicator receives a weight of one sixth in the total): • Child mortality: if any child has died in the family; and • Nutrition: if any adult or child for whom there is nutritional information is malnourished. Living standards (each indicator receives a weight of one eighteenth in the total): • Electricity: if the household has no access to electricity; • Drinking water: if the household has no access to clean drinking water or clean water is more than 30 minutes walk from home; • Improved sanitation: if the household does not have an improved toilet or if the toilet is shared; • Flooring: if the household has dirt, sand, or dung floor; • Cooking Fuel: if they cook with wood, charcoal, or dung; • Assets: deprived if the household does not own more than one of: radio, TV, telephone, bike, or motorbike, and do not own a car or tractor. Source: Alkire and Roche 2013 and http://hdr.undp.org/en/content/multidimensional-poverty-index-mpi. 34 ETHIOPIA – POVERTY ASSESSMENT FIGURE 2.3: Components of the MPI in 2011 and over time, 2000–2011 Percentage contribution of deprivations Percentage improvement in proportion of each dimension to overall poverty, 2011 deprived in each dimension over time, 2000–2011 50 46% 40% 35% 34% 40 34% 34% 30% 30% 25% 30 28% 26% 20% 18% 20 15% 10% 8% 10 5% 0 0 Education Health Living standards Education Health Living standards Ethiopia Average Sources: Alkire, S., A. Conconi, and S. Seth (2014) and Alkire, S., J. M. Roche and A. Vaz (2014). is too high to reflect recent progress in Ethiopia. A people from being counted as deprived in that dimen- household is counted as deprived in assets if it does sion to not deprived. For example, in the DHS the not own more than one of a radio, TV, telephone, proportion of households that have no toilet has fallen bike, or motorbike, and does not own a car or tractor. substantially from 82% to 38%, and the proportion Bike ownership is widely prevalent in many countries, of households with a latrine increased from 18% to but is less desirable in some of the more mountain- 54% (Carranza and Gallegos 2012). However these ous highland areas where large proportions of the improvements are not reflected in the MPI sanitation Ethiopia population live. Conversely livestock are the measure as they do not count as improved toilets and most important non-land asset in rural Ethiopia, and thus on this dimension 91% of households in Ethiopia a means by which households store wealth, but this today are still counted as deprived (Alkire, Conconi, asset class is not considered in the MPI. The propor- and Seth 2014).11 tion of households owning one or more of the set of The MPI allows a cross-country comparison assets considered by the MPI increased by 21 percent- on a broad range of dimensions in one index, and age points in rural areas and 23 percentage points in it usefully draws attention to the further need for urban areas. However, the proportion of households progress in Ethiopia, but using the aggregate mea- owning two or more assets in this asset set increased by sure alone as a statement about the level of poverty only 6% over this time (Alkire, Roche, and Vaz 2014). and changes in poverty over time does not reflect Access to electricity is given a relatively high weight in the full reality. The choice of indicators and cutoffs the MPI. No access counts directly as a deprivation in a global index cannot always reflect local reali- and also contributes to poor performance on a second ties. In the case of Ethiopia, the MPI does not fully dimension of wellbeing: the proportion of households reflect living standards or the progress that has been that use clean cooking fuels (defined as electricity, gas, made. Assessing overlapping dimensions to examine or biogas). Ethiopia almost doubled electrification rates from 12% to 23%, but low rates of electrification 11 One non-methodological point is that it is not clear how the DHS are still observed. In sanitation the binary indicator water data is being used. The MPI records the proportion of households without drinking water increasing over time (Alkire, Roche, and Vaz used to measure deprivation does not reflect improve- 2014), yet the DHS data shows clear improvement. The HICES also ments because they have not been enough to move shows improvement in access to clean water as recorded in Table 2.2. Multidimensional Poverty in Ethiopia 35 multidimensional changes in welfare over time proves FIGURE 2.4: Monetary and education to be a useful exercise in the case of Ethiopia. deprivations and wellbeing perception, 2000–2011 Perceived improvements in 2.4  2000 Rural 2011 wellbeing Although progress was observed in education, health, assets, sanitation, water, and monetary poverty, more people in 2011 perceived they were worse off com- pared to a year ago than in 2000. This perception Urban that living standards had worsened was particularly 2000 2011 prevalent in urban areas. Over half of the Ethiopian population perceived that their living standards had worsened over the last year in 2011 (Table 2.2). This represents more than an increase 20 percentage points in urban areas since 2000, and a 12 percentage-point increase in rural area since 2000. This is the only indi- Money poor Children out of school living standard perception cator of wellbeing documented in Table 2.2 that sig- Note: Details for these diagrams are in Annex 2. nificantly worsened in both urban and rural Ethiopia during the last decade. This may be pointing to changing attitudes after the food inflation periods of 2010–11 because the WMS survey asks households about how households that experienced a worsening of wellbeing their level of material comfort has changed from the were more likely to report having experienced a shock previous year. Indeed, the deprivation variable indicates in the last 12 months, particularly a food price shock, that the urban subpopulation has a slightly higher pro- adding credence to this hypothesis (see Table 2.3). portion of negative perceptions than rural households in 2011. In the previous two survey periods, more rural Deprivations that particularly affect 2.5  than urban households perceived their living conditions girls and women had worsened in the last year. Households that reported conditions worsen- This section considers selected indicators that par- ing in the last year were not just those deprived on ticularly affect the wellbeing of girls and women. Of other dimensions of wellbeing. Figure 2.4 indicates the total 12 indicators considered in this study three that in 2011, many households that perceived their are particularly important for the wellbeing of girls and conditions worsening were not living below the pov- women. These are female circumcision, institutional erty line and were not education deprived. Similar birth, and girls’ education. We also consider evidence figures could be shown for the other dimensions of on other indicators of female wellbeing from the DHS. wellbeing. There has been substantial progress in invest- The perception of worsening does not reflect a ments in education for girls aged between seven true worsening from 2005 to 2011. However, it may and 15. In 2000 more than three quarters of rural reflect a worsening from 2010 to 2011. It could be households with school-aged girls had at least one that on some dimensions, wellbeing improved before girl not in school but by 2011 this had fallen to less worsening. Without additional, more frequent surveys than half of all rural households. In urban Ethiopia it is not possible to test this hypothesis. However, progress was also observed, albeit from a much better 36 ETHIOPIA – POVERTY ASSESSMENT TABLE 2.3: Household’s perception about living standards and price shock, 2011 “The overall living standard of the Households reporting household now when compared to 12 Households reporting no-food price months ago” a food price increase increase All households Proportion of households responding: Much worse now 26% 12% 14% Worse now 45% 35% 37% Same 15% 25% 23% A little better now 13% 26% 24% Much better now 1% 2% 2% Total 100% 100% 100% Source: Own calculations using HCES 2011. baseline: In 2000, 22% of households with school- of women and men that believe physical violence aged girls had at least one girl out of school; this fell to is justified remains high. Between 2000 and 2011, 14% in 2011. This progress reflected primary school the share of women who found wife beating accept- net attendance ratios for girls, which rose from 28% able under specific circumstances decreased from in 2000 to 62% in 2011. Remarkably, in the period 85 to 68 percent. The reduction was larger among 2000–2011, the original gap in primary school net younger women (it fell to 64 percent) and among enrollment rates in favor of boys disappeared. men. In 2000, 75% of men justified wife beating and Very few women report giving birth in health in 2011 this was 45% (Carranza and Gallegos 2013). facility although the number of women receiv- The high proportion of women and men who still ing antenatal visits increased. Almost no rural agree with wife beating is concerning. Carranza and women recorded giving birth in a health facility in Gallegos note that the belief that domestic violence is 2011 (4 percent) and one in two urban women were justified is frequently correlated with poorer wellbeing similarly deprived. This represents a considerable outcomes among women and their children. Women health challenge in Ethiopia today. The WIDE-3 who believe that a husband is justified in hitting or studies documented that despite a government beating his wife tend to have a lower sense of entitle- campaign to encourage all babies to be delivered at ment, self-esteem, and status. Such a perception acts health centers launched early in 2013, most births as a barrier to accessing health care for themselves and were still taking place at home with the assistance their children, affects their attitude towards contracep- of traditional birth attendants and in some places tive use, and impacts their general wellbeing. Health Extensions Workers due to practical and The harmful practice of female circumcision cultural preferences (Bevan, Dom, and Pankhurst is still widespread despite its illegality. A 2003 2014). However, the DHS data shows that the UNICEF report ranks Ethiopia among the top proportion of women who had an antenatal visit countries where female genital mutilation or cutting during their most recent pregnancy in the previous (FGM/C) practices are common (UNICEF, 2003). five years, increased from 27 in 2000 to 43% in The report shows that there were 23.8 million girls/ 2011 (Carranza and Gallegos 2013). women who have undergone FGM/C. In 2011, 30% Physical violence against women became less of Ethiopians in rural areas and 19% of Ethiopians socially acceptable during the decade, but the rates in urban areas lived in households in which a girl Multidimensional Poverty in Ethiopia 37 younger than 14 had been or would be circumcised. FIGURE 2.5: Multiple deprivations affecting The WIDE-3 studies documented that the practice women, 2011 was still widespread and that in some sites there was Urban Rural vocal female opposition to the ban (Bevan, Dom, and Pankhurst 2013). Few girls are simultaneously out of school, experiencing poverty, and facing circumcision; but more than three in four rural households with girls and more than two in four urban households with girls are deprived in at least one of these dimen- sions. Figure 2.5 shows that in 2011, women in rural Money poor Girl out of school households had a higher chance of experiencing all Girl has been or will be circumcised three deprivations largely as a result of the higher rates Note: Details for these diagrams are in Annex 2. of education deprivation for girls. In general however, especially in urban areas, the overlap between these different dimensions of wellbeing is low. A number of non-poor households have girls who are out of school puts these children in the better-off category. These and practice female circumcision in both rural and children are rarely employees in poor households urban areas. Few girls are deprived in all three depri- and are most often girls employed by urban families. vations, which is a positive finding. However the flip The disadvantage faced by these children requires side to this is that many girls in Ethiopia today experi- urgent public attention, particularly to encourage ence some form of deprivation, they are either poor, their employers to invest in their employees. Box 2.4 not in school, or underwent (or will undergo) female details a pilot intervention undertaken to address this circumcision. problem. Girls who work as domestic maids are most likely to be deprived in investments in education: 2.6 Conclusion only 20% of school-aged children who are non-rela- tives and employed by the household in which they This chapter has documented the considerable reside are in school. Relatively better-off households, progress that has been made in reducing the pro- especially in urban areas, employ children as maids for portion of individuals deprived in multiple dimen- domestic services including babysitting, cooking, and sions, but also the challenges that still remain. other chores. These unrelated children are less likely Improvements in health, water, sanitation, education, to be in school. Table 2.4 shows enrollment status in and poverty, particularly in rural areas, have reduced 2011 was 20% for these children compared to 65% for deprivation. Further improvements are needed to all children. However, a monetary poverty indicator address continued deprivations faced by households TABLE 2.4: Deprivation status for school aged children (aged 5–17) by relationship, 2011 Child Status Non-relative, employed by household All other children In school 0.20 0.65 Below poverty line* 0.04 0.34 Source: Computed from WMS and HCES 2011. Note: * Household level indicator. 38 ETHIOPIA – POVERTY ASSESSMENT BOX 2.4: Learning how to provide education to out-of-school girls in Addis Ababa Ethiopia has made significant strides in expanding access to education through a range of initiatives, including the alleviation of school fees, social campaigns, and availability of non-formal education. While these initiatives have greatly increased enrollment rates, there is an interest in honing existing policies to extend equity and quality to vulnerable groups. Findings from a recent World Bank research project indicate that out-of-school migrant girls in urban areas are among the disadvantaged, and that it is indeed possible to mobilize them through a community-driven and government-supported approach. From 2013 to 2014, the World Bank implemented “Powering Up: Social Empowerment for Vulnerable Girls” in partnership with the Addis Ababa City Administration Bureau of Women, Children and Youth Affairs and the Population Council. The project trained local female mentors to mobilize out-of-school, 12–18 year-old girls throughout 17 woredas (districts) in five sub-cities of Addis Ababa city into community-donated “safe spaces” (or girls clubs). The mentors made door-to-door visits to inquire about eligible girls and to convince the head of the household for their approval. At times, repeated visits were necessary to obtain consent. Within club settings, girls participated in non-formal education and life skills training. Where possible, they were mainstreamed into the formal education system with the aid of school material provision. A demographic and livelihood profiling of the girls revealed that over 90% were rural-to-urban migrants. Two-thirds of the girls had some years of formal schooling, although none had attained education beyond primary school. Almost half were functionally illiterate. Two out of three stated that they had no friends, revealing a disturbing degree of social isolation. Over a third of girls (34%) lived with their non-relative employers, primarily as domestic workers with an average income of 249 birr a month. One-fifth of these girls also received in-kind payment—mainly food and shelter—equivalent to a value of approximately 325 birr a month. Project results revealed that constraints to formal school enrollment for a large share of the project participants were not as strong as expected. Substantial increases in enrollment can be achieved by strong nudging of household decision-makers, coupled with appropriate coverage of hidden school costs. Further, results also suggest that girls can easily be absorbed into existing alternative basic education centers as they provide a similarly flexible learning environment as the girls clubs. Data was also collected on girls who declined participation in Powering Up. Much like the project participants, approximately one-third of the girls (32 out of 100) lived with their non-relative employers. For girls who lived with relatives, household poverty was cited as the major factor affecting the lack of participation, as coping mechanisms are manifested in high demand for child labor. As such, further pilot projects need to be carried to better understand the opportunity costs to education, and on how to strengthen delivery of existing services to ensure that girls receive an education that is suitable to their lifestyle. For those girls who lived with non-relatives as their employees, household heads stated that the girls’ domestic duties were too great for participation in the clubs, indicating minimal or no value placed in their employees’ education. Stronger government encouragement aimed at the employers of these girls is recommended to adequately address the needs of these girls who remain disenfranchised and invisible. Source: World Bank (2014). and those that particularly affect women and girls. Few (2014) shows that decentralized spending on public girls are simultaneously out of school, experiencing services in Ethiopia is effective in improving outcomes. poverty and facing harmful traditional practices; but Expenditure is broadly equal across woredas with the more than three in four rural households with girls and exception that more goes to historically disadvantaged two in four urban households with girls are deprived areas. The study also identified that spending on public in at least one of these dimensions. services has not yet reached decreasing returns, sug- Continued emphasis on the successful delivery gesting that continued investments in service provision of basic services in rural areas is required as well as will continue to yield beneficial results. The incidence further attention to the needs of urban households of public expenditure on education and health is con- for whom progress has been slower. Khan et al. sidered further in Chapter 5. 39 THE CHANGING NATURE OF VULNERABILITY IN ETHIOPIA 3 Wellbeing in Ethiopia has historically been vul- are vulnerable, and what they are most vulnerable nerable to drought. Almost half of rural households too. Section 3.1 discusses the main sources of vulner- in Ethiopia were affected by drought in a five-year ability reported by households in Ethiopia and their period from 1999 to 2004 (Dercon, Hoddinott, and impact on consumption. In Section 3.2 approaches to Woldehanna 2005), and drought had a significant measuring vulnerability are described and rates of vul- impact on the welfare of these households. The con- nerability are documented. Section 3.3 describes the sumption levels of those reporting a serious drought characteristics of vulnerable households, and Section were found to be 16% lower than those of the families 3.4 draws conclusions. not affected, and the impact of drought was found to have long-term welfare consequences: those who had 3.1  Sources of risk in today’s Ethiopia suffered the most in the 1984–85 famine were still experiencing lower growth rates in consumption in the Unexpected events that cause ill health, a loss of 1990s compared to those who had not faced serious assets, or a loss of income play a large role in deter- problems in the famine (Dercon 2004). mining the fortunes of many people in the devel- How vulnerable are households in Ethiopia oping world. A study exploring welfare dynamics in today? Have improvements in many dimensions of rural Kenya and Madagascar found that every poor wellbeing had an impact on the ability of households household interviewed could ultimately trace its pov- to respond to shocks in the future that might hit them? erty to an asset or health shock (Barrett et al. 2006). Asset ownership in rural areas is substantially higher Dercon et al. (2005) show that just under half of rural today than 10–15 years ago; many more households households in Ethiopia reported to have been affected report themselves being able to raise money to handle by drought in a five-year period from 1999 to 2004. an emergency than in the past and markets have also Additionally, 43% reported to have been affected by improved, limiting the local price impact of local a death in the household and 28% were affected by a supply shocks. In addition in recent years, drought in serious illness. Even in the absence of social safety nets Ethiopia has been increasingly well managed. During or private insurance markets, rural households have the last decade Ethiopia has transitioned from a sys- some informal risk coping mechanisms with which tem of emergency food aid to one in which many they are able to manage some risk. In the absence of vulnerable households are covered under a safety net safety nets households accumulate and liquidate assets program, the Productive Safety Net Program (PSNP). to smooth consumption and provide mutual support The drought of the mid-1980s caused many deaths in the form of gifts and transfers. However, poorer with estimates ranging from half a million to over a households are less able to use their assets to manage million (Dercon and Porter 2010). In comparison the risk and mutual support also has its limits. Yilma et drought in 2002 did not cause many deaths and this al. (2013) find that only 2–5% of households received was similarly the case for 2011. help from friends, relatives, or neighbors in the case of This chapter examines the extent of vulnerabil- economic shocks or health shocks, whereas 22–30% ity in Ethiopia today and the nature of those who reported selling assets, and 18% borrowed to meet the 40 ETHIOPIA – POVERTY ASSESSMENT cost of health shocks. The result is that shocks often better than a 15 year average). Crop conditions were hit poorer households and disadvantaged individuals relatively good throughout Ethiopia in 2010/2011, and harder. For example, fluctuations in adult nutrition better than compared to 2004/5. This may seem sur- were found to be larger among women and individuals prising given this was the onset of the Horn of Africa from poorer households (Dercon and Krishnan 2000). drought, but that drought only affected the pastoral The types of shocks that beset household welfare regions of the country, which are a small proportion in Ethiopia are changing, as are the institutions to of the sample used here. However, drought and crop respond to them. Weather shocks have been less com- damage were still reported by a number of households. monplace in recent years and some climate change A moderate drought causes consumption losses predictions would predict they will become less com- of 8% in drought prone areas, but for PSNP ben- monplace also in the future. Price shocks on the other eficiaries this impact is estimated at 6%, perhaps hand have increased in recent years. At the same time, reflecting the role of the PSNP in helping mitigate market integration is improving. While high food prices the impact of drought. The impact of different types may be beneficial for net sellers they can be problematic of shocks was estimated across households using for net buyers or non-agricultural households for whom objective rainfall measures of data and self-reported wage and non-wage income has not kept up with price food price shocks, job loss shocks and death. The changes. As poverty becomes more urban, sudden food impact was allowed to vary based on the ability of a price increases become more of a challenge (Alem and household to manage the risk. In rural areas house- Soderbom 2012). Health shocks may be better managed holds with more land and access to the PSNP were today with the widespread presence of health extension considered separately from others to reflect the fact workers, and some forms of community based health that higher wealth levels and PSNP transfers may help insurance emerging in some areas, but for individuals them better manage risk. In urban areas households that have migrated to urban centers away from tradi- with more education and male-headed households tional support networks they may pose a much greater were considered separately from other households. challenge than they did in the past. This section exam- In addition drought shocks were allowed to have a ines the types of shocks that households are particularly differential impact depending on whether they were vulnerable to and the characteristics of households that occurring in a drought prone area or in areas that are are more or less resilient to these shocks. not used to having shocks. The results for drought The most prevalent shock in 2011 was an adverse and food shocks—the two most reported shocks—are food price shock, and this was reported more in summarized in Figure 3.1, with full results in Annex urban areas than in rural areas (Table 3.1). In con- 3. While a moderate drought (crop loss of 30%) has trast, 2011 was a better year for crops than 2005 (and little impact on consumption in areas where it is an TABLE 3.1: Frequency of shocks 2005 2011 Proportion of households reporting the following shocks Food price 0.02 0.19 Drought 0.10 0.05 Job loss 0.01 0.00 % crop loss (from LEAP) 23.5 13.8 Sources: Own calculations using HICES 2005 and HCES 2011. The Changing Nature of Vulnerability in Ethiopia 41 infrequent occurrence it reduces consumption by 8% FIGURE 3.1: Impact of Drought and food in drought-prone areas. The number of plots owned price shocks has little impact on a household’s ability to manage 15% Percentage change in consumption risk, but the PSNP did seem to reduce losses some- 10% what, with drought reducing consumption by 6% for 5% PSNP beneficiaries. 0% 0% –2% –4% However, the impact of drought is likely to –5% –6% –8% –8% be higher when crop losses are higher. Using the –10% –10% –13% Ethiopia Rural Household Survey, Porter (2012) found –15% that more extreme shocks impact consumption to a far –20% greater extent than lesser shocks. Rainfall in the bottom –25% Land rich Land poor, no psnp Land poor, psnp Educated Uneducated, female Uneducated, male quintile of the 30-year village distribution caused up to 20% drop in household consumption, whereas for the next quintile of the rainfall distribution (i.e. less than Infreq Frequent Rural Urban average rainfall, but to a lesser extent), the impact was Drought High food prices around 2% (and non-significant). Figure 3.2 depicts Notes: Drought is defined as a crop loss of great than 30%. Points the types of yield shocks that were present in the run indicate the average estimated impact of drought and the length up to the two survey years (2010 and 2004) and com- of the bar depicts the precision of the estimate stretching from one standard deviation below the average to one standard deviation pares them to 2002 which was the most recent very bad above the average. FIGURE 3.2: All Ethiopia: Meher crop losses 0.20 0.15 0.15 0.10 Density Density 0.10 0.05 0.05 0 0 0 20 40 60 0 20 40 60 80 Meher-Value 2010 Meher-Value 2005 0.10 0.08 Density 0.06 0.04 0.02 0 0 20 40 60 80 Meher-Value 2002 Notes: The graphs show the proportion of woredas in each season, in each year, that experienced crop losses of between 0 and 100%, based on LEAP data. For example, in the top left graph, almost 20% of woredas experience no crop loss, and a very few woredas experience crop loss of greater than 20% during the Meher season. This is in contrast to the Meher 2002 graph, directly below, which shows just under 10% that experi- ence no crop loss, but a much higher proportion of woredas that have crop loss greater than 20%. 42 ETHIOPIA – POVERTY ASSESSMENT rainfall year. The values show the proportion of crop (Dercon and Christiaensen 2010). Households in lost, e.g. a value of 60 means that 60% of the crop was Oromia that had access to insurance were 31% more lost in that year due to rainfall deficits. In 2010, few if likely to invest in fertilizer as a result (Berhane et al. any households experienced losses of more than around 2014). Enabling poor households to better deal with 30%, whereas in 2002, a considerable proportion did. shocks—particularly those shocks that are frequent There were some places in 2005 that experienced high and severe—is essential to both improving their wel- crop losses, particularly in 2005. However, overall the fare in the short run and improving their opportunities years of data that are available for the analysis do not for income growth in the long run. capture the very worst that can happen and the impact is likely to be larger in more extreme years. 3.2  Measuring vulnerability in Ethiopia Household welfare is still vulnerable to bad weather: were a drought similar to 2002 to be Vulnerability to poverty is conceptually distinct experienced in Ethiopia today, regression estimates from poverty, since it is what might happen, an suggest poverty would increase substantially. To expectation about the future. Many chronically poor investigate the impact of an extreme weather shock households are poor due to a lack of assets (including on poverty the occurrence of a major drought in large land, able-bodied working-age labor, good health) and parts of the highlands, such as experienced in 2002, opportunities. Such households will have possibly not was simulated. In this event, the regression results sug- exited poverty in a long time. Many household who are gest that poverty would increase from 30% to 51%. above the poverty line however, are at risk of transient Food price shocks are mainly reported by urban poverty. A household may be classified as non-poor since households, and the impact on urban households is its consumption lies above the poverty line in 2011. In also much greater, particularly for households with contrast a household may be classified as vulnerable if little education. In rural areas food price shocks had there is a strong likelihood that it could be poor in the little impact on households, with consumption fall- near future. There is usually quite a degree of overlap ing by two percentage points on average. Food price between poor and vulnerable households. However it is shocks were felt more severely in urban areas, but also possible that a household may be poor today, but not equally. Educated households were not forced to not vulnerable to poverty in the future, if for example reduce consumption as a result of high food prices, but the household had extremely bad luck, relative to what uneducated households reduced their consumption we might otherwise have expected. It is likely that such by 10–13%. The impact of the food price shock was households are experiencing transient poverty and may much higher for these households than the impact of be quite likely to escape poverty in the near future. a moderate drought shock in rural areas. Measuring vulnerability is more complex than Uninsured risk not only has a direct impact of measuring poverty because of the uncertainty that household welfare, it also impacts the decisions has to be incorporated into the measure. Different poor households make about their livelihood. measures have strengths and weaknesses and looking The expectation that something bad may happen at a number of different measures at the same time affects household behavior, causing households who provides a richer picture of vulnerability than one are unprotected to avoid expending effort on risky measure alone. Table 3.2 details the measures used in activities. Fertilizer investment is an example of this this chapter and the sources of data used to estimate in rural Ethiopia—fertilizer returns are high when the them. The first measure is a new estimate generated rainfall is good and negative when the rainfall is too for the Poverty Assessment and Box 3.1 provides more low or too high—and households that are less able to detail on how this measure was estimated. Further manage income risk are less likely to apply fertilizer detail can also be found in Hill and Porter (2014). The Changing Nature of Vulnerability in Ethiopia 43 TABLE 3.2: Measure of vulnerability Indicator Description Data Vulnerability to poverty See Box 3.1 HCES, WMS, LEAP Asset vulnerability The Household Economy Approach is “a livelihoods-based framework for analyz- Livelihood base- ing the way people obtain access to the things they need to survive and prosper” lines and LIAS. (HEA Practitioners guide 2008). This approach underpins the Livelihood Baselines that are used in the Livelihoods Impact Analysis and Seasonality (LIAS) to generate estimates of the numbers of those in need of emergency assistance each year. Food gap This measures is used in a number of programs in Ethiopia to identify households WMS in need—it identifies those that were not able to meet the food needs of their household for all 12 months in the previous year. It is a measure that overcomes the seasonality of some of the other measures of wellbeing (such as amount consumed in the last week or month), but it is still a measure of current wellbeing rather than future wellbeing. Experiencing shock Self-reported experiences of shocks that have negatively affected consumption or WMS assets Not able to raise 200 This question gives an indication of whether the household could access resources WMS Birr to protect itself should a shock materialize. The measures of vulnerability considered are proportion of population that is currently poor. The primarily drawn from data collected in the 2005 proportion of people measured as vulnerable using the and 2011 rounds of the nationally representative food gap measure is higher at 20%. This indicates that Household Income and Consumption Expenditure although on average 13% are expected to be unable and Welfare Monitoring Surveys (HICES/WMS). to meet their food needs, at certain points in the year The advantage of using the HICES/WMS is that it is (during the lean season for example) this proportion nationally representative, and it allows measures of vul- of households is higher. nerability to be estimated at the household level. This Table 3.3 also reports two other indicators of allows the analysis to look at the relative importance vulnerability. The proportion of households who of geographic and household factors in determining experienced decreases in welfare as a result of a self- vulnerability, and to examine how vulnerability var- reported list of nine shocks including illness of a ies across certain groups of households. As detailed household member, death of a household member, in Box 3.1 first measure incorporates data from other drought, livestock loss or death, crop damage, flood- sources as well. The second measure of vulnerability— ing, price shock, job loss, or food shortage is 55%. The asset vulnerability—is from the livelihoods baselines. It proportion of households who say that they would not is nationally representative, but is a measure of vulner- be able to raise 200 Birr, should a sudden need occur is ability of an area rather than an individual. 18%. Many households may experience reductions in As expected, more households are vulnerable wellbeing that do not cause them to fall into poverty, to being poor in the future than are poor today. In which is why more households report experiencing a 2011, vulnerability to poverty was 41%, much higher shock than the number of vulnerable households. In than the national poverty rate of 30% (Table 3.3). addition, raising 200 Birr is not full protection from Asset vulnerability as defined by the livelihood base- larger shocks so again it is not surprising that this lines is 34%, slightly lower than the new measure number is lower than the proportion of households estimated for this report, but also higher than the that are vulnerable. 44 ETHIOPIA – POVERTY ASSESSMENT BOX 3.1: A measure of vulnerability to poverty using cross-sectional data Measuring vulnerability is conceptually and empirically much more complex than measuring poverty because it is about what might happen, an expectation on the future. The perfect dataset would include several years’ worth of observations for each household, and even better, information on what could happen and how probable this is/was in differing states of the world. This box details how a measure of consumption vulnerability has been estimated with the data available in Ethiopia: nationally representative surveys that do not survey the same household twice, and other data sources on the nature of shocks households face. In the future as successive panels of the Ethiopian Rural Socioeconomic Survey have been conducted the nature of data available for this analysis will change. Measures of consumption vulnerability are the most difficult to estimate, but unlike many of the other measures of vulnerability considered in this chapter; they provide an estimate of how likely it is that a household will be poor in the future even if they are not poor today. This is often thought to be an important element of vulnerability to capture. Step 1: Define a wellbeing indicator, and a level of wellbeing, below which a person is considered poor. Consistent with the poverty numbers of the Government of Ethiopia (MOFED 2013), a household is consumption poor or absolute poor in one year if their total expenditure on all items is less than the national absolute poverty line. This is the amount of money needed to purchase food of 2200 kilocalories for every adult-equivalent in the household, and other necessary items for everyone in the household. In this chapter a household is defined as extreme poor if their total expenditure on all items is less than the amount of money needed to purchase food of 2200 kilocalories for every adult-equivalent in the household. This is different from the measure of food poverty used by the Government of Ethiopia, which compares total spending on food (rather than total spending on all items) to the food poverty line. In reality this measure of food poverty results in a similar number of food poor and absolute poor households in Ethiopia. Step 2: Estimate the relationship between shocks and the wellbeing indicator. In the absence of repeated observations for each household, the spatial and historic distribution of shocks is used to estimate the average impact of that shock on household wellbeing. Objective measures of rainfall-induced crop losses are used to measure the impact of drought on wellbeing (as in Thomas et al. 2010 and Anttila Hughes and Hsiang 2013) and household reported shocks are used for other types of shock. Given the impact of a shock on wellbeing is dependent on the ability of a household to manage the shock; the impact of the shock is allowed to vary across different types of individuals. Step 3: Calculate how likely a shock is to happen for each household. This is done in two ways depending on the type of shock. For shocks that do not tend to happen to everyone at once (idiosyncratic shocks), the frequency of similar households that report that shock is used to determine the probability that a given household will experience that shock in the future. For shocks that do happen to many people at once, covariate shocks such as weather-induced yield losses, the probability and severity of shocks is determined by using other sources of data for shocks in that location for the last 18 years. In particular, the Livelihoods, Early Assessment and Protection project (LEAP) system, developed in 2008 by the Government of Ethiopia in collaboration with WFP , is used to calculate the rainfall-induced crop loss in woredas throughout Ethiopia from 1995 to 2012. Step 4: Simulate the likely distribution of wellbeing for each household in one year’s time using the likelihood and estimated cost of shocks. One thousand possible outcomes for each household are simulated in which households experience these shocks according to how likely they are. Each time a household experiences a shock its impact on consumption is calculated using the regression estimates of the impact these shocks have on households. The result is a distribution of likely consumption outcomes for a given household. Step 5: Calculate the probability that the household would have consumption below the national poverty line. A cutoff of probability, above which a household is defined as vulnerable is set. Most studies use 50% probability of poverty to classify vulnerability (i.e. household has more than a 50% chance of being poor), and that is what is used in these estimates also. Across most measures of vulnerability, rural reflecting the fact that rainfall and crop production households appear more vulnerable than urban was better than average in 2010–11. One quarter of households; and relative to poverty rates, rates urban households were estimated to be vulnerable to of vulnerability in rural areas are much higher poverty, but 44% of rural households were estimated The Changing Nature of Vulnerability in Ethiopia 45 TABLE 3.3: 2011 vulnerability and poverty national overview Vulnerable to poverty Other measures of vulnerability Vulnerable to absolute Asset Experienced Cannot Raise Absolute poor poverty vulnerability Food gap shock 200 Birr Total 30% 41% 34% 20% 55% 18% Urban 26% 26% — 8% 50% 21% Rural 30% 44% 34% 22% 56% 18% Source: Own calculations using HICES/WMS 2011 merged with livelihood baseline. to be vulnerable. The rate of urban poverty and vul- Living Standards Measurement Survey becomes avail- nerability is similar but the rate of rural vulnerability able in late 2014. This survey revisited all households is much higher than rural poverty. This reflects the surveyed in 2012 in round 1. Understanding the char- findings in the previous section that the period prior to acteristics of households that were not poor in 2012 the survey in Ethiopia was an unusually difficult time but had become poor by 2014 will help identify the for urban households and an unusually good time for salient characteristics of those vulnerable to poverty. rural households. Many non-poor vulnerable house- Has vulnerability changed in the last six years? holds had fallen into poverty in 2011 and were thus All measures of vulnerability have fallen, but at dif- counted as both poor and vulnerable. Vulnerability to fering speeds. poverty is thus higher in rural areas than the number Vulnerability to poverty has fallen, but not of people currently estimated to be poor. by much, although modeling limitations may Understanding the characteristics of house- contribute to this result. The proportion of house- holds that are vulnerable to poverty but not poor holds calculated as vulnerable to poverty fell from will help in targeting interventions to meet their 43% to 41% (Table 3.4). There are two reasons needs. The regression results in Annex 3 provide some why this may be the case: the favorable crop condi- indication on the characteristics of the vulnerable but tions present in much of the country in 2010/11, not poor households. Further analysis will be possible and modeling limitations. The relatively good crop when the next wave of the nationally representative conditions resulted in lower levels of poverty for TABLE 3.4: Vulnerability measures over time Vulnerable to absolute poverty Food gap Experienced shock Cannot Raise 200 Birr 2011 41% 19% 55% 18% 2005 43% 29% 61% 34% Urban 2011 26% 8% 50% 21% 2005 28% 13% 41% 41% Rural 2011 44% 22% 56% 18% 2005 46% 32% 64% 33% Source: HICES/WMS, LIAS, LEAP. Note that the question about raising cash is 200 Birr in 2011, and 100 Birr in 2005. 46 ETHIOPIA – POVERTY ASSESSMENT many households as higher crop incomes allowed given the widespread predominance of livelihoods them to consume more. However, the measure of that are dependent on rainfed production systems. vulnerability to poverty presented here captures the This characterization of vulnerability has resulted in a risk of being poor in the future, for example if crop widespread understanding of a geographic footprint incomes are low. It may be the case that although of vulnerability. Until recently the Government of households were less poor in 2010/11 they were Ethiopia framed rural policy discussions around “three still quite vulnerable to poverty in the future. In the Ethiopias:” drought-prone highlands, moisture-reli- model used to measure vulnerability, vulnerability able highlands, and pastoral lowlands. This classifica- can only fall between 2005 and 2011 as a result of tion was recently been expanded to a concept of “five changes in household characteristics—such as accu- Ethiopias” according to agricultural productivity and mulation of assets or changing sectors of employ- agroecological conditions (EDRI 2009). The five areas ment over time—because the model assumes that are: drought prone highlands, moisture-reliable cereals the returns to these assets (i.e. the relationship areas, moisture-reliable enset areas, humid moisture- between these characteristics and poverty) and the reliable lowlands, and pastoral areas. distribution of shocks did not change in these five The idea of a geographic footprint of vulner- years. The level of asset accumulation by households ability in Ethiopia has been refined further in the between 2005 and 2011 has been enough to reduce Livelihoods Atlas of Ethiopia. This atlas presents the vulnerability to absolute poverty, but not by much. results of considerable investment in understanding The proportion of rural households experi- the livelihoods and assets of households in different encing a shock fell from 64% in 2005 to 56% of parts of the country and their vulnerability to different households in 2011; however the proportion of types of risks. Communities are grouped in 173 liveli- urban households experiencing a shock increased hood zones. The atlas also allows mapping of many from 41% to 50% during this period. Compared to of the household characteristics that are understood five years ago, the likelihood of reporting a shock has to be deeply related to a vulnerable life, such as the fallen across many regions in the rural areas, with the length of the hunger season. exception of Somali and SNNP regions. The increase The geographical nature of vulnerability in in the proportion of urban households reporting a Ethiopia has influenced the targeting of develop- shock is almost entirely driven by food price shocks. ment interventions. The PSNP is targeted to address A very different picture is observed when look- vulnerability in the most food insecure districts in ing at vulnerability measured by the food gap. This Ethiopia. Interventions in less food insecure woredas perspective suggests that vulnerability has fallen by a have typically not focused on providing a safety net but third during this five-year period: the proportion of instead on how to improve agricultural productivity households unable to meet their food during the last and non-farm income earning enterprises. 12 months has fallen from 29% to 19%. This may Is resilience in Ethiopia primarily driven by reflect better crop conditions present in 2010/11, but geography and in particular, by access to good it may also reflect an improvement in asset accumula- land and reliable rainfall? How do the food price tion among households. shocks in recent years fit with this understanding, do they change the geographic centers of vulnerability or Vulnerable places or vulnerable 3.3  is there a case for abandoning a geographical under- people? standing of vulnerability altogether? The analysis presented in the following tables Unreliable rainfall has historically underpinned suggests that vulnerability does have a geographi- much of the discussion on vulnerability in Ethiopia cal footprint in Ethiopia, but that it is not fully The Changing Nature of Vulnerability in Ethiopia 47 determined by location of residence. There are many area. Estimates of vulnerability to food poverty sug- that are vulnerable in areas that have been considered gest that vulnerability to severe poverty is also high in resilient and there are many that are resilient in vul- moisture reliable lowlands. Lower levels of education nerable areas. The geographic footprint of vulnerabil- and asset ownership result in higher predicted vulner- ity in Ethiopia is assessed by comparing the level of ability in these areas. However, although the moisture vulnerability across the “five Ethiopias,” small towns, reliable lowlands have the highest rate of poverty and and big cities in Table 3.5. The moisture reliable low- vulnerability in Ethiopia, it accounts for only a small lands are the most vulnerable places in Ethiopian in fraction of Ethiopia’s poor and vulnerable households, 2011, followed by the enset-growing lowlands and given that only a small proportion of the population the drought-prone highlands. of Ethiopia lives there (last column of Table 3.5). Vulnerability is highest in the moisture-reliable Across a number of the measures of vulnerabil- lowlands not because residents are more subject to ity, levels of vulnerability in pastoral areas are lower climate shocks that will drive them into poverty, but than may have been initially expected, but further because residents are already poor. It may at first work on the magnitude and nature of vulnerability seem surprising that the moisture reliable lowlands are in pastoral areas is needed. The Livelihoods Atlas the most vulnerable places in Ethiopia, given moisture of Ethiopia attributes lower levels of vulnerability is indeed more reliable in these areas. However, pov- in pastoral areas to the high asset levels recorded in erty, although falling, is still very high in the moisture pastoral households. This also contributes to the low reliable lowlands. More than half (59%) of residents vulnerability to poverty estimates in pastoral areas. were poor in 2005, and although this had reduced In addition, a number of vulnerability measures may substantially to 45% in 2010/11 the level was still 13 have been affected by the fact that survey coverage was percentage points higher than the next geographic quite limited in Afar and Somalia. The 2011 HCE TABLE 3.5: Poverty and vulnerability across the “five Ethiopias” and urban centers, 2011 Household Household Share of Vulnerable has a food experienced Asset national Poor to poverty gap a shock vulnerable Overall rank population Moisture- 45% 75% 31% 87% 26% 1 2% reliable lowlands Enset 29% 47% 36% 75% 57% 2 18% lowlands Drought- 28% 43% 25% 46% 50% 3 33% prone highlands Moisture- 32% 42% 13% 63% 13% 5 42% reliable highlands Pastoral areas 31% 52% 21% 31% 16% 4 2% Town/small 26% 27% 9% 52% 41% 6 7% city Large city 22% 23% 4% 28% 0% 7 12% Note: to calculate the overall rank we rank each geographic area with each measure and take the average of rank overall measures. 48 ETHIOPIA – POVERTY ASSESSMENT TABLE 3.6: The proportion of individuals measured as poor and vulnerable by PSNP status, 2011 Household Vulnerable to Household has a experienced a Absolute poor absolute poverty food gap shock Asset vulnerable PSNP woreda 32% 47% 29% 49% 60% PSNP not in woreda 28% 36% 13% 63% 13% Note: Asset vulnerable is defined at the woreda level here, and does not pick up individual variation within the woreda so this overstates the num- ber of vulnerable in PSNP woredas and understates the number of vulnerable not in PSNP woredas. survey covered all rural and urban areas of the country absolute poverty is 11 percentage points higher in except non-sedentary areas in Afar and Somali (three PSNP woredas than in non-PSNP woredas. However and six zones, respectively). Although vulnerability to there are still a substantial proportion of vulnerable absolute poverty is not particularly high, vulnerability households in non-PSNP woredas. Again taking the to extreme poverty is quite high. This suggests that measure of vulnerability to absolute poverty, 36% of the vulnerability that exists in pastoral areas is to very households living in non-PSNP woredas are vulner- extreme poverty. The results indicate that a separate able to absolute poverty. study on the magnitude and nature of vulnerability Individuals everywhere—in every woreda of among pastoral households is needed. Ethiopia—are vulnerable and as a result safety net Even though vulnerability may have a geo- programs targeted to specific woredas will necessar- graphic footprint in Ethiopia, and even though ily result in many vulnerable Ethiopians being left safety net programs are well targeted to more vul- without safety nets. Table 3.7 estimates the magni- nerable areas of Ethiopia, much vulnerability is tude of this omission, and presents the total number of not geographically determined, but instead deter- vulnerable people in PSNP woredas and the number mined by other factors such as individual access of vulnerable people in non-PSNP woredas. For all to assets, or lifecycle events. The spatial targeting of the measures of vulnerability defined at the house- of PSNP is examined in Table 3.6 by estimating and hold level there are significant numbers of vulnerable comparing the proportion of vulnerable households households in non-PSNP woredas. For example, in PSNP and non-PSNP woredas. For every mea- although 13.9 million individuals who are vulnerable sure of vulnerability, rates of vulnerability are higher to absolute poverty live in PSNP woredas, 15 million in PSNP woredas than in non-PSNP woredas. For individuals that are vulnerable to poverty live outside example the proportion of households vulnerable to of woredas where PSNP programs are run. This means TABLE 3.7: The number of individuals measured as poor and vulnerable in PSNP woredas, 2011 (million) Household Vulnerable to Household has a experienced a Absolute poor absolute poverty food gap shock Asset vulnerable PSNP woreda 9.9 13.9 8.9 15.1 18.5 Non-PSNP woreda 12.0 15.0 5.6 27.0 5.6 Note: Asset vulnerable is defined at the woreda level here, and does not pick up individual variation within the woreda so this overstates the num- ber of vulnerable in PSNP woredas and understates the number of vulnerable not in PSNP woredas. The Changing Nature of Vulnerability in Ethiopia 49 that even if the PSNP were perfectly targeted to all of and more vulnerable than the national average. the vulnerable households in the woredas in which it Households with disabled members are particularly is run, 52% of the vulnerable households in Ethiopia vulnerable to extreme food poverty. A safety net that would be without a safety net. targets these groups in urban areas would be targeting Which people are particularly vulnerable? households that are much more vulnerable than the Vulnerability across the lifecycle is examined by average urban household (see Chapter 8 for further disaggregating households into groups that may be discussion). Households with children under two particularly vulnerable: female-headed households, years of age are equally poor as the average, but have those with very young children, households with chil- higher vulnerability to absolute poverty. Those with dren out of school, unemployed, disabled, and older out of school children have much higher vulnerability household heads. than the average, but poverty is also higher. Similarly The primacy of access to the labor market as a urban households with out of school youth are much determinant of poverty and vulnerability in urban more vulnerable. areas is evident. Vulnerability is considerably higher On average more households are vulnerable in for many of these potentially vulnerable groups in rural areas and the strong patterns of higher vulner- urban areas (see Table 3.8, top panel). Those who ability among the potentially vulnerable lifecycle are unemployed, disabled or elderly are much poorer stages observed in urban areas is not present to TABLE 3.8: Demographic characteristics of vulnerability Vulnerable to Experienced (Percent) Poor poverty shock Raise cash Food gap Urban Overall 26 25 50 79 8 Child under 2 25 28 49 82 9 Out of school child 37 44 47 79 13 Out of school youth 32 33 53 80 8 Unemployed 35 34 59 71 7 Disabled 42 45 63 69 16 Female headed 28 26 50 70 11 Head over 65 35 33 52 73 10 Rural Overall 30 43 56 82 22 Child under 2 30 47 58 84 21 Out of school child 35 51 60 84 22 Out of school youth 37 53 55 84 24 Landless 23 31 55 66 31 Disabled 40 50 70 78 32 Female headed 27 37 59 71 29 Head over 65 30 42 51 78 22 Source: Own calculations using HCE 2011. 50 ETHIOPIA – POVERTY ASSESSMENT the same degree in rural areas. The bottom panel of net program that is targeted to specific woredas will Table 3.8 presents results for rural Ethiopia. The age necessarily result in many vulnerable Ethiopians being of the household head is only associated with increased left without safety nets. Additional interventions will vulnerability in urban areas, not in rural areas. Female- be needed to reduce their vulnerability to shocks. headed households do not appear much poorer or Rural vulnerability is higher than urban vul- vulnerable in rural areas. This is the same finding nerability, and higher than rural poverty measures as in MOFED (2013) and in Chapter 1. Landless for 2011 might suggest in light of the good rains households are also not poorer or vulnerable by any in the run up to the survey. Some projections of measure, but these are also few in number (4.6% of the likely impact of climate change suggest changing the population weighted sample). weather conditions may bring about improvements Vulnerability of households with disabled in yields and wellbeing (for example see Robinson et members (those unable to work due to disability al. 2013) but variability in yields will also increase. or illness) is an issue in both rural and urban areas, This is likely to be particularly high as farmers learn slightly more in urban areas. There is also evidence about new weather patterns and adapt their produc- that those with children under two, and those with tion technologies to the changes they bring. Helping out of school youth and out-of-school children are farmers mitigate the impact of production losses on also more vulnerable than the average rural household. consumption is essential for reducing vulnerability. Further strengthening of rural safety nets through 3.4  Summary and conclusion broadening the geographical reach of the PSNP (particularly to vulnerable areas in the lowlands) and Vulnerability does have a geographic footprint in ensuring the PSNP can scale up effectively at times Ethiopia and the PSNP is targeted to many of the of drought will increase the resilience of rural house- most vulnerable woredas in Ethiopia. The moisture holds. However other mechanisms are also needed to reliable lowlands are the poorest and most vulnerable help provide additional insurance to farmers, as it is places in Ethiopia in 2011, followed by the enset- not realistic to expect a publicly financed safety net to growing lowlands and the drought-prone highlands. fully insure households against weather risk. However, much vulnerability is not geographically Although urban vulnerability is much lower determined, but instead determined by other factors than rural vulnerability, one quarter of urban such as individual access to assets, or lifecycle events. households are vulnerable. The nature of risk faced This causes individuals everywhere—in every woreda by rural households is quite different. Food price of Ethiopia—to be vulnerable. For all of the measures shocks comprise a major risk—and the types of house- of vulnerability defined at the household many vulner- holds that are vulnerable in rural areas are also differ- able households are found in non-PSNP woredas. For ent in urban areas, with labor market access being a example, although 13.9 million individuals who are primary determinant of vulnerability. An urban safety vulnerable to poverty live in PSNP woredas, 15 million net can reduce the vulnerability of urban households, individuals that are vulnerable to poverty live outside but it will need to be a very different type of safety net of woredas where PSNP programs are run. A safety than the rural-based PSNP. 51 DRIVERS OF POVERTY REDUCTION IN ETHIOPIA 4 There are many possible factors that could have of poverty reduction that is not usually possible contributed to Ethiopia’s impressive performance in sub-Saharan Africa. As described in the previous in reducing poverty in recent years. Ethiopia has chapter, the Ethiopian Central Statistical Agency experienced high and consistent economic growth, has collected consumption data four times between recording annual per capita growth rates of 8.3% in 1996 and 2011, and in a comparable manner allow- the last decade, driven largely by growth in services ing changes in poverty to be measured for three time and agriculture (World Bank 2013). Substantial periods for nearly all of Ethiopia’s zones. Multiple improvements in the provision of safety nets and basic surveys and census data are used to construct annual services were also taking place at this time. Ethiopia zonal estimates of poverty, economic output, safety introduced the Productive Safety Net Programme in net beneficiaries and access to public services and 2005, a large rural safety net targeted to those parts of markets. Panel analysis is then used to identify what Ethiopia where reliance on food aid had been highest. has been driving changes in poverty over time. This Expansion of the provision of education and health approach has been used in China (Montalvo and services also increased from a low base during this Ravallion 2009), India (Datt and Ravallion 1996) and time, supported by the Provision of Basic Services Brazil (Ferreira et al. 2011) but not for any African Program. In addition, Ethiopia witnessed tremendous country. Weather shocks are used to further examine investment in infrastructure and market development the causal nature of agricultural growth and poverty during this period. Road networks expanded reducing reduction. remoteness, integrating markets and reducing market- The chapter also examines what type of agri- ing margins (Minten et al. 2012). cultural growth has been most effective at reducing This chapter explores the type of growth and poverty. Agriculture has remained the primary occu- investments in public goods that drove reductions in pation of a large proportion of Ethiopian households poverty and improvements in wellbeing. It exploits during this period (Martins 2014). There has been variation in poverty reduction, sectoral output growth a strong policy focus by the Ethiopian government and provision of public goods across zones and time to on encouraging productivity growth in small-holder examine what has been driving changes in poverty over cereal farming during this period in the Agricultural the period of 1996 to 2011 in Ethiopia. The analysis Development Led Industrialization strategy (ADLI), examines the extent to which growth drove changes in and its later formulation in the PASDEP and GTP. poverty reduction, and what type of growth—output As part of this strategy the government has spent growth in agriculture, manufacturing or services—was considerable resources supporting cereal intensifica- more effective at reducing poverty. The analysis also tion of smallholder farmers, for example through examines whether safety nets and public good provi- investments in agricultural extension services and sion more broadly, had an additional effect on poverty supporting fertilizer distribution. Understanding reduction by increasing redistribution. the effectiveness of this focus and the impact of this Ethiopia is a country rich in data, which strategy on the spatial nature of poverty in Ethiopia allows an approach to understanding the drivers is thus important. The results suggest that the 52 ETHIOPIA – POVERTY ASSESSMENT agricultural growth that has been encouraged by these 4.1  Decomposing poverty reduction investments has paid off, but that access to centers of urban demand, good prices and good weather have Ethiopian households are primarily rural and also been important. self-employed in agricultural production and Before presenting the results of the analysis as a result poverty reduction among rural, self- two decomposition techniques are used to quan- employed and agricultural households has been tify changes that have been important to poverty the major component of poverty reduction from reduction during this period. As Box 4.1 describes, 1996 to 2011. Poverty reduction in rural areas these techniques rely on defining a counterfactual accounted for 2.0, 5.2 and 7.8 percentage points of scenario, which is then used to help identify the poverty reduction during the periods 1996–2000, quantitatively important changes that have occurred 2000–2005 and 2005–2011 (Figure 4.1). The contri- during this period. bution of reductions in poverty among those engaged BOX 4.1: What does decomposing changes in poverty entail? In this chapter the results of two decomposition methods are presented. The first method is the Ravallion and Huppi (1991) inter-sectoral decomposition method that quantifies how much poverty reduction among different groups or movement between different groups accounts for national poverty reduction. The second method uses Recentered Influence Functions (RIF, Firpo et al. 2009) in which traditional Oaxaca-Blinder decompositions are applied to different percentiles of the consumption distribution. This allows an assessment of the amount of poverty reduction that can be accounted for in changes in the characteristics of households and individuals (“endowments”) compared to the changing nature of the Ethiopian economy and poverty. Both decomposition methods rely on defining a counterfactual scenario and estimating what would have happened to poverty had the counterfactual scenario occurred. By defining a counterfactual scenario the changes that have been important to overall poverty reduction can be quantified. The figure below depicts how this can work for two different counterfactual scenarios. In the Ravallion and Huppi method the focus is on a counterfactual of no change in the proportion of population in different sectors; and a counterfactual of no change in poverty among people in a given sector. These counterfactuals are used to examine the amount of poverty reduction that took place within sectors (as if sectors had not changed), and the amount of poverty reduction that took place as a result of people moving between sectors. In the RIF analysis the focus is on a counterfactual of a constant relationship between endowments and poverty in Ethiopia over 1996 to 2011. This counterfactual is used to determine which changes in endowments could have contributed to poverty reduction, and how much poverty reduction could have changed as a result of a changing relationship between poverty and endowments. The latter is sometimes referred to as changes in the returns to endowments, but really it represents how the conditional correlation between a given endowment and consumption has changed. In all decomposition approaches there is an interaction effect which can be interpreted as a measure of the correlation between population shifts and inter-sectoral changes in poverty in the Ravallion and Huppi method, and changes in endowments and returns in the RIF analysis. In the decompositions shown here it is quite small. Using counterfactuals to quantify changes that have been important to poverty reduction Counterfactual: Poverty Change in poverty rates for Poverty if no Change in endowments and Poverty in 1996 people with a given endowment change in “interaction effect” in 2011 endowments Counterfactual: “ Change in poverty rates for Poverty Poverty if only Poverty Change in endowments people with a given endowment in 1996 change in in 2011 and interaction effect” endowments Drivers of Poverty Reduction in Ethiopia 53 FIGURE 4.1: The contribution of rural and FIGURE 4.2: The contribution of poverty urban poverty reduction reduction among different sectors –10 25% –10 Poverty reduction in percentage points Poverty reduction in percentage points –8 20% –8 Urban population share –6 15% –6 –4 10% –4 –2 5% –2 0 0% 0 2 –5% 2 1996–2000 2000–2005 2005–2011 1996–2000 2000–2005 2005–2011 Interaction Population shift Urban Interaction Population shift Other Service Rural Urban population share Construction Manufacturing Agriculture Source: Own calculations using HICES 1996, HICES 2000, HICES Source: Own calculations using HICES 1996, HICES 2000, HICES 2005, and HCES 2011. 2005, and HCES 2011. in agriculture was similar: 1.9, 3.8 and 6.9 percent- FIGURE 4.3: The contribution of poverty age points respectively (Figure 4.2). And among the reduction among the employed and self- self-employed: 1.1, 4.8 and 7.5 percentage points employed respectively (Figure 4.3). –10 100% Poverty reduction in percentage points Self-employed popula6on share Increasingly, reductions in poverty in urban –8 80% areas, among those engaged in the service sector –6 60% and among those who are wage-employed contrib- uted to overall poverty reduction, but structural –4 40% change has not contributed much to poverty reduc- –2 20% tion during this time. Poverty reduction among those 0 0% engaged in the service sector has accounted for about one percentage point of poverty reduction since 2000. 2 –20% 1996–2000 2000–2005 2005–2011 This is about one eighth of total poverty reduction that Interaction Population shift Self-employed has taken place during this time, which suggests the Not Self-employed Self-employed population share contribution of the service sector to growth has been Source: Own calculations using HICES 1996, HICES 2000, HICES much lower than the contribution of the service sector 2005, and HCES 2011. to value addition during this period. Poverty rates fell faster among those that reported employment in the service sector (MOFED 2013) but employment in the poverty reduction. This is in contrast to some other service sector has remained consistently low across this countries with large agricultural sectors that have time period (from 12–14% of the workforce) which experienced fast reductions in poverty. In Rwanda makes it very difficult for service sector growth to have and Cambodia, poverty reduction among agricultural a large direct effect on poverty reduction. Structural households also contributed to poverty reduction, but change—shifts in the share of the population engaged growth in non-farm enterprises and, in Cambodia, in certain sectors, living in urban locations or the urban wage-employment contributed to additional nature of employment—has contributed very little to poverty reduction. 54 ETHIOPIA – POVERTY ASSESSMENT Changes in individual and household charac- (43%) but the contribution of endowments is much teristics, “endowments,” can account for between higher (67%) for a household at the 90th percentile. 46% and 67% of consumption growth during this Of the endowments considered, improvements period. The household surveys that collect data on in education and demographic changes can most consumption expenditure that is used to define pov- account for poverty reduction during this time. erty do not collect much information on household Improvements in primary education were particu- income and productive activities. As a result only lim- larly important among the poorest households, while ited analysis can be done to ascertain how changes in improvements in post-secondary education were par- employment and productive activities contributed to ticularly important among the richest households. poverty reduction in Ethiopia. However, Hassan and Demographic changes include changes in the age dis- Seyoum Taffesse (2014) use data on demographics, tribution of household heads, changes in household education, occupation type, location of residence, size and the dependency ratio of the household. It was and ownership of some productive assets to assess changes in the size and composition of households that the degree to which changes in endowments have contributed the most to the role of demographic change. contributed to poverty reduction in Ethiopia from A shift to technical and professional occupa- 1996 to 2011, or whether poverty reduction has come tions helped increase consumption at all points in about as a result of a changing relationship between the distribution, but particularly among the richest. endowments and poverty. The findings from this study This suggests that some of the growth in services had a (see Figure 4.4) show that changes in endowments of larger impact on wealthier households than on poorer the median household can explain 46% of growth households which may also be one reason why service in consumption for the median household. This is sector growth has been much higher than reductions quite similar for a household at the 10th percentile in poverty among those engaged in the service sector. Controlling for all other factors, urbanization on its own did not contribute to consumption growth for the median household. For the richest FIGURE 4.4: The contribution of households it made them marginally better off and demographics, education, occupational for the poorest households it made them marginally change and urbanization to consumption worse off. growth, 1996–2011 For the majority of households, the changing 0.8 relationship between endowments and consump- Contribution to change in per-adult 0.7 tion was a more important contributor to changes equivalent consumption 0.6 0.5 in consumption from 1996 to 2011 than changes 0.4 in endowments, and this was the case particularly 0.3 for poorer households. For the median household 0.2 changes in the relationship between consumption 0.1 and endowments, holding endowments constant can 0 –0.1 explain 72% of the total change in consumption. This 0 20 40 60 80 100 is somewhat similar for poorer households, but drops Consumption percentile to 55% of the total change. Change in endowments Demographic change As more people have become educated the rela- Increased education Occupational change Urbanization Changing relationship between tionship between education and consumption has consumption and endowments changed dramatically from 1996 to 2011. Although Source: Hassan and Seyoum Taffesse (2014). Figures 4.1 to 4.3 suggest little structural change in Drivers of Poverty Reduction in Ethiopia 55 Ethiopia during this time, Chapters 1 and 2 showed (Figure 4.4). However, with such large changes in the just how much some aspects of life in Ethiopia have proportion of educated individuals there has also been changed from 1996 to 2011. In particular, educational a structural shift in the relationship between educa- attainment has increased substantially and in part tion and poverty. This is depicted in Figure 4.5. The this explains why increased education can account figure shows how much consumption increases with for part of the poverty reduction that has taken place educational attainment (of primary, secondary and FIGURE 4.5: The changing relationship between education and consumption, 1996–2011 A. Primary 0.15 Estimated returns to primary education 0.10 0.05 0 0 20 40 60 80 100 0 20 40 60 80 100 0 20 40 60 80 100 0 20 40 60 80 100 1996 2000 2005 2011 Quantile B. Secondary 0.5 secondary education Estimated returns to 0.4 0.3 0.2 0.1 0 20 40 60 80 100 0 20 40 60 80 100 0 20 40 60 80 100 0 20 40 60 80 100 1996 2000 2005 2011 Quantile C. Post-secondary 1.0 Estimated returns to 0.8 higher education 0.6 0.4 0.2 0 20 40 60 80 100 0 20 40 60 80 100 0 20 40 60 80 100 0 20 40 60 80 100 1996 2000 2005 2011 Quantile Source: Hassan and Seyoum Taffesse (2014). 56 ETHIOPIA – POVERTY ASSESSMENT post-secondary education) for households across the During this period the ratio of GDP growth to poverty consumption distribution. reduction suggests an elasticity of -0.55 which is higher The correlation between educational attain- than the regression based estimate, but still quite low. ment and poverty has fallen over time. As more Christiaensen et al. (2013) find no relationship between people have primary and secondary education the GDP growth and poverty reduction in sub-Saharan additional consumption obtained by primary school Africa so even though the effect of growth on poverty and secondary school graduates lessens, particularly reduction may be small, it is still much higher than the for wealthier households and particularly for primary rest of the region and confirms that Ethiopia has been school graduates. Acquiring some years of primary much more successful than other countries in Africa in education no longer obtains the same increase in converting growth into poverty reduction. consumption in 2011 as it did in 1996. The same Simulations using household survey data find that is true for secondary education, although the gains if all households were to experience equal amounts of in consumption are higher. The correlation between poverty reduction, one percentage point growth in consumption and post-secondary education has been household consumption would result in a fall in pov- more constant across time, although it has also fallen. erty of almost two percentage points (–1.94) given the household consumption distribution in 2011 4.2  Drivers of poverty reduction (MOFED 2013). If the growth poverty elasticity is calculated using household consumption growth rates To assess what has driven these changes, a dataset of rather than GDP rates, a relatively high growth elastic- zone-year observations is used to assess correlates, ity of poverty reduction is found: –1.53 from 2000 to and where possible, determinants of changes in 2011. This is much higher than the regional average of poverty in Ethiopia between 1996 and 2011. Various –0.69 reported in Christiaensen et al (2013) for this sources of nationally representative survey data collected measure, and closer to the global average of –2.02. by the Ethiopian Central Statistical Agency are com- Growth in agriculture, more than growth in bined to create this dataset. Zones are used as the unit of other sectors, has been significantly positively analysis, as it is the lowest level at which data on poverty related with poverty reduction; poverty has fallen and agricultural output can be disaggregated. Fifty zones fastest in those zones in which agricultural growth are followed over a period of 15 years, covering nearly all has been strongest. Columns 2–5 of Table 4.1 present of Ethiopia’s population. The method used and details the results of regression analysis examining the type on how measures of poverty, agricultural, services, and of growth and investments that have contributed to manufacturing output were constructed are provided in poverty reduction. Manufacturing and services output Annex 4. The Annex also details data used to determine growth has not been a significant contributor to pov- changes in infrastructure, educational investments and erty reduction on average during the fifteen years from number of PSNP beneficiaries. 1996–2011, although the coefficients on manufactur- ing and services growth are of the sign expected. The Has growth contributed to poverty reduction? implied elasticities of poverty to growth in agriculture, manufacturing, and services are –0.155, –0.002 and Growth has been a significant driver of reductions –0.027 respectively.12 However, given the imprecision in poverty over the fifteen-year period from 1996 to with which the coefficients on manufacturing and 2011, although each 1% of growth resulted in only 0.15% reduction in poverty. Results are presented in 12 Calculated by multiplying the coefficients in column 1 of Table 4.1 column 1 of Table 4.1. Although growth had an impact with the average share of the sector over the years 1996, 2000 and 2005 the estimated growth elasticity of poverty was quite low. detailed in Table A4.1. Drivers of Poverty Reduction in Ethiopia 57 TABLE 4.1: Growth, safety nets and infrastructure investments contributed to poverty reduction (3) (4) Weighting results by urban population (5) Annualized percentage change in (1) (2) IV headcount poverty rate 1996–2011 1996–2011 1996–2011 2000–2011 1996–2011 Annualized percentage change in…. Output per capita –0.15* (0.09) Agricultural output per capita –0.29** –0.04 0.30 –1.66** (0.14) (0.20) (0.32) (0.70) Manufacturing output per capita –0.03 –0.47 –1.36* 0.20 (0.42) (0.38) (0.73) (0.61) Services output per capita –0.04 0.04 –0.17 0.27 (0.18) (0.24) (0.34) (0.30) Proportion of population in PSNP –0.06** –0.06* –0.09** –0.03 –0.01 (0.03) (0.03) (0.04) (0.05) (0.05) Distance to primary school –0.08 –0.07 0.01 0.37** 0.07 (0.16) (0.16) (0.12) (0.14) (0.24) Distance to public transport 0.18* 0.14 0.22*** –0.44 –0.02 (0.10) (0.11) (0.08) (0.37) (0.17) Constant –0.02 –0.02** –0.01* –0.04*** 0.02 (0.01) (0.01) (0.01) (0.01) (0.09) Observations 147 147 135 91 147 R-squared 0.115 0.129 0.169 0.312 Number of zones 50 50 46 46 50 Source: regression results using data described in Annex 4. Notes: Zonal fixed effects included but not shown. Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1. services sector growth are estimated, a test of equal- magnitude of the relationship between agricultural ity of coefficients across the three sectors (cannot be growth and poverty reduction increases (column 5 of rejected. Table 4.1 and Figure 4.6). This indicates that the rela- Agricultural output growth has had a strong tionship between poverty reduction and agricultural causal impact on poverty reduction: for every 1% of growth does not arise because less poor households are growth in agricultural output, poverty was reduced able to better grow their agricultural incomes (a story by 0.9% which implies that agricultural growth of reverse causality). Instead it indicates that either caused reductions in poverty of 4.0% per year on measurement error might affect agricultural growth average post 2005 and 1.1% per year between 2000 estimates (what is called “attenuation bias”) or agricul- and 2005. Agricultural growth is been instrumented tural growth induced by good weather was particularly with weather shocks in order to assess whether the poverty reducing. This indeed could be the case. relationship between agricultural growth and pov- Although nationally growth in manufacturing erty reduction is causal. When agricultural growth is or services did not contribute to poverty reduction, instrumented with weather shocks the significance and in urban Ethiopia, manufacturing growth played 58 ETHIOPIA – POVERTY ASSESSMENT FIGURE 4.6: The contribution of agricultural FIGURE 4.7: Services growth is positively growth, services and safety nets to poverty correlated with growth in agriculture reduction, 1996–2011 0.15 Poverty reduction (percent) from … 0.10 Growth in agriculture Improved access to services 0.05 0 PSNP –0.05 –0.10 Agricultural growth –0.10 0 0.10 0.20 Growth in services –5 0 5 10 15 20 25 30 35 Source: Own calculations using data described in Annex 4. 1996 to 2000 2000 to 2005 2005 to 2011 Source: Regression results using data described in Annex 4. This finding is corroborated by analysis presented in Chapter 6, which shows that 64% of businesses were a significant role in reducing poverty from 2000 established using funds from agricultural production to 2011. For every 1% of growth in manufacturing and that these businesses are most active in the months output, poverty fell by 0.37%. of harvest and immediately thereafter, suggesting a The insignificance of service sector growth is strong relationship between agricultural production surprising given it contributed to a tenth of poverty and this type of service sector activity. It is quite likely reduction in recent years (Figure 4.3). In all other that any relationship between growth in services and aspects the findings of the zonal regression analysis poverty reduction is being captured in the coefficient have been consistent with the findings of the decom- on agricultural growth. position analysis presented in Section 4.1. First, it This analysis helps explain some of the regional is worth noting that of the three sectors, output esti- convergence in poverty rates reported in Figure 1.2 mates were most imprecise for this sector, relying on and Table 1.2 in Chapter 1. Agricultural growth employment data in the HICES and national estimates was particularly strong in Tigray and Amhara, and of output per worker in this sector. This measurement these regions also benefited from the introduction of error may mask the true relationship between these the PSNP. Although SNNPR did not record strong sectors and poverty reduction. In the measures of agricultural growth through this time, the introduc- service output presented here, not all of service sector tion of the PSNP and strong improvements in access activity is included—for example public employment to basic services and towns helped to reduce poverty. is not included—but the same findings holds when a Oromia experienced both good agricultural growth broader measure of service output is used. and the introduction of the PSNP, but the magnitude Growth in the service sector has been high- of both improvements was smaller than in Tigray and est when agricultural growth has been highest, Amhara and Oromia’s overall poverty reduction was so although it may have contributed to poverty also lower. Although SNNPR did not record strong reduction it has not had an effect independent of agricultural growth through this time, the introduc- growth in agriculture. Figure 4.7 shows the positive tion of the PSNP and strong improvements in access correlation and this correlation is significant at 5%. to basic services and towns helped to reduce poverty. Drivers of Poverty Reduction in Ethiopia 59 The pastoral regions of Afar and Somali did not experi- of improved inputs. The type of agricultural growth ence agricultural growth, and although safety nets were that is most associated with poverty reduction is introduced there this alone was not enough for these quantitatively explored in Figure 4.8 and Table 4.2 regions to realize strong gains in poverty reduction. and indicates similar findings. Recent years have seen high food prices and Understanding the relationship between good rainfall conditions in much of Ethiopia. agricultural growth and poverty reduction Food prices increased in Ethiopia over the period 2000–2011 and particularly in the year prior to the What drove the relationship between agricultural survey during which annual food price inflation was growth and poverty reduction? Box 4.2 presents 39.2%. Table 4.3 compares food prices increases findings on the type of economic growth that mat- to other countries and shows that the food price tered in 12 rural communities of the WIDE-3 study Ethiopia experienced in 2011 was relatively high. In in Ethiopia. The findings highlight the importance general, weather has been good in Ethiopia in recent of good prices, access to markets and increased use years. Figure 4.9 indicates the proportion of farmers BOX 4.2: Agricultural growth in 12 rural communities The WIDE research covers 20 communities in Ethiopia selected as exemplars of different types of rural livelihood systems. Research was conducted in 1995, 2003 and 2010–2013. Findings are reported here for six sites with agricultural potential and six agriculturalist food insecure sites for which research was conducted in 2012–13. In the six communities with agricultural potential, large changes in the local economies since the early 1990s were documented with economic growth in evidence in all. Growth was not driven solely by increasing agricultural incomes but also by increasing involved in trade and other non-farm activities, wage employment in nearby towns, and remittances. Agricultural growth has been driven by improvements in agricultural productivity, increased demand for crops and livestock products, better access to markets, food price increases, and new aspirations. Economic growth had also been experienced in all six agriculturalist but food-insecure communities as a result of improvements in roads, increases in agricultural and no-farm incomes, and the PSNP . Improvements in agricultural incomes were related to agricultural productivity increases, food price inflation, better road access to markets, and diversification into higher-value products, many of which depended on irrigation. Cash-crop production and sale had increased everywhere. Failure to maintain a road had reduced access to markets in one site. Improved seeds, fertilizer, and new planting techniques had contributed to improvements in agricultural productivity. The analysis also highlighted the vulnerability of agricultural growth as a sole driver of improvements in wellbeing. Although economic growth had been experienced, all food insecure communities had suffered at least one severe drought since 2003. Annual rain shortages were experienced although the severity varied by year. Irrigation schemes were of varying importance in the sites but demand for irrigation was high. Finally, the research provides insights on what has been effective in encouraging agricultural growth in the agricultural sites, and what had not been effective: • Minimal agricultural extensions services were available in the mid 1990s but by 2013 the services covered crops, livestock and natural resource management and the government was supporting and monitoring farming activities. • Although limited credit was available in the early 1990s in 2013 most communities had credit for farm and non-farm activities though regional MFIs • Nearly all government investment in rural economic development had gone to adult male farmers, mostly richer ones. This strategy was successful but this group of leading farmers is now sufficiently well-established and aspiring that it does not need nudging anymore. • The engine of growth is the hard work of private individuals trying to change their lives and most cooperatives had failed to work effectively. Source: Bevan, Dom and Pankhurst 2012 and 2013. 60 ETHIOPIA – POVERTY ASSESSMENT TABLE 4.2: Agricultural growth and poverty reduction Annualized percentage change in headcount poverty rate (1) (2) (3) (4) Annualized percentage change in…. Growth in agricultural output per capita interacted with Close to town of 50,000 plus –3.40* (1.81) Far from town of 50,000 plus –0.74 (0.66) Cereal output per capita –0.35** (0.16) Cash crop output per capita 0.45 (0.54) Manufacturing output per capita 0.39 0.02 –0.190 –0.14 (0.70) (0.42) (0.42) (0.41) Services output per capita 0.64 –0.09 –0.16 –0.15 (0.48) (0.18) (0.18) (0.18) Proportion of land planted with improved seed 0.004 –0.007 (0.04) (0.04) Proportion of land applied with fertilizer –0.01 (0.01) Proportion of land applied with fertilizer * bad conditions 0.001 (0.01) Proportion of land applied with fertilizer *good conditions –0.04* (0.02) Weighted crop price index –0.16 –0.14 (0.15) (0.14) Change in predicted rainfall induced crop-loss 0.002*** 0.002*** (0.001) (0.001) Constant –0.039 –0.03** –0.04** –0.03** (0.10) (0.01) (0.01) (0.01) Observations 147 147 143 143 R-squared 0.141 0.141 0.225 0.254 Number of zones 50 50 49 49 Source: regression results using data described in Annex 4. Notes: Zonal fixed effects included but not shown. PSNP, education and infrastructure variables are included but not shown. Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1. experienced a rainfall-induced crop loss of greater than farmers experiencing crop loss was higher than two 30%. This has been calculated using rainfall data and standard deviations above the average was 2003. crop models. On average, 16% of farmers in Ethiopia Good prices and good weather have been essen- experience such crop losses each year. Since 2003 the tial in ensuring that increases in the use of fertil- proportion of farmers experiencing crop losses has izer brought about reductions in poverty. Despite not gone higher than one standard deviation above substantial increases in the use of inputs over this the average. The last year in which the proportion of period, the estimates in column 3 of Table 4.2 indicate Drivers of Poverty Reduction in Ethiopia 61 FIGURE 4.8: Increased fertilizer use reduced TABLE 4.3: Annual food inflation in selected poverty when weather and prices were good countries Poverty reduction from inputs, prices, and 2005–2011 2011 weather (% per year) 2.5 Ethiopia 21.8% 39.2% 2.0 China 8.0% 11.8% 1.5 Vietnam 14.4% 26.5% 1.0 Uganda 14.8% 32.3% 0.5 Zambia 8.0% 4.9% 0 Kenya 17.0% 20.5% –0.5 Rwanda 8.8% 2.0% –1.0 Africa (Total) 11.4% 13.3% Seeds Fertilizer Prices Weather Source: FAO database. Good conditions (good rain, high prices) Poor conditions Source: Regression results using data described in Annex 4. is estimated separately for good and bad conditions. Good conditions are defined as years in which weather FIGURE 4.9: Proportion of farmers was better than average, and when crop prices were experiencing more than 30% crop loss, higher than average (given returns to fertilizer are also 1997–2011 highly price dependent (Spielman et al. 2010). There 0.35 is a significant relationship between the use of fertil- 0.30 izer and poverty reduction when the conditions are 0.25 right and no relationship between fertilizer use and 0.20 poverty in other years. The results suggest that under 0.15 the right conditions, a 10% increase in fertilizer use would reduce poverty by 0.4%. 0.10 An analysis of agricultural growth, for a larger 0.05 number of years, confirms that modern input-use 0 1996 1998 2000 2002 2004 2006 2008 2010 2012 contributed to agricultural growth when weather Source: Rainfall induced crop loss is calculated for each woreda conditions and prices were favorable. Table 4.4 exam- using the LEAP database. These estimates are then weighted using ines the relationship between growth in cereals output the population living in each woreda. Belg and Meher are added in each year, so 1997 represents crop loss from Meher rains of 1996 and weather, prices and the use of improved inputs over harvested in January 1997 and crop loss from Belg rains harvested a longer period of time. Given the focus of this regres- around June of 1997. sion is no longer the relationship between agricultural growth and poverty reduction, years in which poverty data is not available can also be included allowing the that, on average, increased use of in inputs did not panel to be expanded to all years from 1996 to 2011. cause poverty reduction. Returns to use of improved Growth in modern input-use contributed to agricultural inputs is highly weather dependent in Ethiopia. growth when weather conditions and prices were favor- Christiaensen and Dercon (2010) provide estimates able. There was no contribution of growth in improved that show that net-returns are only positive under inputs in other years. The results also highlight the good weather conditions. In column 4 the relationship important role of weather and prices in overall agricul- between growth in fertilizer use and poverty reduction tural output growth. It is also possible that in addition 62 ETHIOPIA – POVERTY ASSESSMENT TABLE 4.4: Favorable rainfall and improved producer prices contributed to agricultural growth Growth in revenue from cereals (1) (2) (3) Change in predicted rainfall induced crop-loss –0.005*** –0.004*** –0.002 (0.001) (0.001) (0.002) Growth in the proportion of land planted with improved seeds –0.026 –0.030 –0.006 (0.030) (0.030) (0.031) Growth in the proportion of land on which fertilizer was applied 0.016 (0.033) Growth in the proportion of land applied with fertilizer * bad conditions –0.026 –0.012 (0.036) (0.037) Growth in the proportion of land applied with fertilizer *good conditions 0.154** 0.159** (0.062) (0.065) Growth in crop prices 0.124*** 0.117** 0.218*** (0.047) (0.047) (0.058) Growth in the area of land cultivated 0.292*** (0.055) Constant 0.064*** 0.059*** 0.050*** (0.019) (0.019) (0.020) Observations 452 452 452 R-squared 0.039 0.054 0.073 Number of zones 38 38 38 Source: regression results using Agricultural Sample Surveys and LEAP data. Notes: Zonal fixed effects included but not shown. Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1. Given the larger number of year-zone observations available for the regressions presented in this table, these regressions focus on the main agricultural zones, namely all zones in Amhara, Oromia, SNNPR, and Tigray. Somali, Benishangul-Gumuz, Harari, Addis Ababa, and Dire Dawa are thus excluded. Agricultural zonal outputs are not available for two years in the middle of the series, and as a result two years of estimates are lost. to ensuring positive returns to fertilizer use, higher and agricultural growth. The relationship between prices may have encouraged farmers to increase the agricultural growth and poverty was compared in area of land farmed, or the number of hours spent on areas that were far (more than 6 hours and 40 min- farming activities. While it is difficult to measure labor utes) from urban centers of 50,000 plus people at the intensification, it is possible to examine whether farmers beginning of the time period in question, to the rela- increased the area of land farmed. Indeed results in col- tionship between agricultural growth and poverty in umn 3 of Table 4.4 show that increased land cultivation areas close to urban centers. Agricultural growth was over time also contributed to rising agricultural revenue. only poverty reducing for those close to urban centers However, regression analysis not shown suggests that (Table 4.2, column 1). expansion of agricultural land is positively correlated This finding suggests an important link with good rainfall but not with price increases. between agricultural growth, access to markets, While agricultural growth had a strong impact and urban demand, which is likely to be fuelled by on poverty reduction on average, the positive non-agricultural growth. Although manufacturing impact of agricultural growth was only found close and services growth did not have a direct effect on aver- to urban centers of 50,000 people or more, indicat- age rates of poverty reduction during this period, the ing the complementary nature of non-agricultural results do point to a potential indirect role of growth Drivers of Poverty Reduction in Ethiopia 63 in these sectors, and to the need for growth in non- reduces poverty by about seven percent. This estima- agricultural sectors. This finding echoes the results of tion strategy controls for initial differences in PSNP the simulation analysis in Diao et al. (2012). Work in a and non-PSNP areas, zone-specific time-trends and quasi-experimental setting in northern Ethiopia shows time-varying differences in growth rates across zones as that remoteness and limited access to markets can have well as proxies for other social spending and infrastruc- a substantial impact on transport costs, reducing net ture investments. The positive impact of the PSNP profits from agricultural sales. Transportation costs found is plausible given the program is well targeted over a 35-kilometer distance, along a route mainly (Chapter 5 and Berhane et al. 2012) and contributed accessible to foot traffic only, led to marketing costs to improved food security for beneficiaries (Berhane increasing from 6% to 23% of the market price. They et al. 2012). However given only one change in pov- also led to a 50% increase in the price of chemical erty is observed after the introduction of the PSNP, fertilizer and a 75% reduction in its use (Stifel et al. it is possible that commensurate changes brought 2012, Minten et al. 2014). There has been a remark- about in PSNP areas at the same time as the PSNP able uniform deployment of extension agents in all was introduced could be an alternative explanation of locations; however the more costly supply of inputs this result. The significance of the effect of the PSNP may result in lower agricultural growth in remote areas is not robust to all specifications. as inputs are less used. The results in Table 4.2 do not There is also some evidence that investments in speak to this as they show that the impact of the same roads may have a direct beneficial effect on poverty amount of agricultural growth on poverty reduction reduction through redistribution in addition to was lower in more remote areas, not that agricultural their role in increasing the poverty reducing impact growth has been lower in more remote locations. This of agricultural poverty gains as identified earlier. does seem to suggest a story of market access for agri- Investment in roads has had an impressive impact cultural output and the ability to access and provide on increasing access to urban markets as evidenced other consumption goods and services. However, it is by Figure 4.10 from Schmidt and Kedir (2009). possible that supply side constraints also play a role in Remoteness is still a defining characteristic of extreme limiting household profits from agricultural revenue poverty in rural Ethiopia. Poverty rates increase by growth or causing only richer household to experience 7% with every 10 kilometers from a market town. As agricultural revenue growth. outlined above, farmers that are more remote are less likely to use agricultural inputs, and are less likely to Safety nets and investments in public services see poverty reduction from the gains in agricultural growth that are made. This makes poverty reduction The introduction of transfers to poor households more challenging in remote locations. Remoteness is in food-insecure rural areas also contributed to something that affects only some individuals within a poverty reduction post 2005. The PSNP has been zone, and a zonal-level analysis will only pick up part shown to increase agricultural input-use among some of the impact of infrastructure on poverty. Further beneficiaries thereby supporting agricultural growth analysis using poverty mapping and smaller geo- (Hoddinott et al. 2012). The results in Table 4.1 graphic units of analysis is really needed to properly show that the implementation of the program from identify the impact of infrastructure and basic services 2005 onwards had an additional annual impact on on poverty reduction. The generally positive impact poverty reduction through redistribution of 0.5 per- of improvements in infrastructure and access to basic cent. The magnitude of the effect is consistent with services such as education complements the evidence the fiscal incidence analysis presented in Chapter 5 for Ethiopia that suggests investing in roads reduces which shows that the direct effect of PSNP payments poverty (Dercon, Gilligan and Hoddinott 2009). 64 ETHIOPIA – POVERTY ASSESSMENT FIGURE 4.10: Travel time to urban centers of 50,000 people or more in 1994 and 2007 1994: Travel time to a city 2007: Travel time to a city of at least 50,000 people of at least 50,000 people Hours Hours 1994 2007 Source: Schmidt and Kedir 2009. Further work will help inform whether continued growth and poverty reduction is conditional on access investment in roads is likely to bring about the same to urban demand. Agricultural households more beneficial effects on poverty reduction. proximate to urban centers can more easily consume goods and services from urban centers and supply Implications for future poverty 4.3  goods and services to these markets. Increased urban reduction demand can also put increasing upward pressure on cereal prices (Minten et al. 2012), which the analysis Explaining past growth performance helps inform presented in this chapter suggests may help poverty what worked and what did not in achieving poverty reduction. This is consistent with the finding of Diao reduction. This section considers the implications of et al. (2012) that simultaneous growth in agricul- these findings for future efforts to reduce poverty in ture and non-agriculture will bring about the fastest Ethiopia. declines in poverty rates. Reducing transportation Agricultural growth is likely to remain impor- costs will also reduce the cost of fertilizer in more tant in reducing poverty. Agricultural output growth remote locations, which may help encourage further was found to explain a large part of Ethiopia’s success agricultural growth. in reducing poverty, and given the large share of house- Adoption of agricultural technologies can holds still engaged in agriculture, this trend is likely to reduce poverty, but their effectiveness is dependent continue. The analysis offers insights on the nature of on good prices and good weather. Increased use of agricultural growth and the interplay between growth improved inputs was beneficial for poverty reduction in agriculture and growth in other sectors. when good weather conditions and favorable crop Agricultural growth will have a larger impact on prices prevailed. The analysis confirms other studies poverty reduction if it is complemented by growth showing that fertilizer, improved seeds and production in urban, or non-agricultural, demand. The results practices have the potential to stimulate agricultural show that the strong relationship between agricultural growth in Ethiopia (Teklu 2006, Dercon and Hill Drivers of Poverty Reduction in Ethiopia 65 2011, Vandercateelen et al. 2013) suggesting their to insure crop income (such as index insurance for increased use may reduce poverty further.13 However, better off farmers and safety nets for poorer farm- the conditional nature of this poverty reduction, is a ers that scale-up when drought occurs) will likely be reminder that: (i) many of the technologies currently important in ensuring Ethiopian farmers can manage on the table offer returns that are highly rainfall depen- climate change well. dent, rendering this a potentially vulnerable source of High crop prices help poverty reduction, but growth, and (ii) improvements in cereal markets and rising food prices will hurt some poor households. increasing urban demand will also be needed to keep Compensatory policies (such as an urban safety crop prices high. net) may be needed to offset this effect. Increases The rainfall dependency of returns to agricul- in producer prices contributed to agricultural growth tural technologies means that increasing uncer- and increased the incentives for technology adoption. tainty around climate change needs to be managed. Higher producer prices will benefit net-producers, In three of the four climate change scenarios con- which comprise a sizeable share of poor households sidered by Robinson et al. (2013) changing weather in Ethiopia. Table 4.5 indicates that households that conditions bring about average improvements in report having a food gap of less than three months cereal yields in Ethiopia. However although climate are a high proportion of poor households in 2011 change may bring about improved yields on average, (42%) and increasing across time (25% in 2005). all scenarios predict an increase in variability of yields However Table 4.5 also indicates that many poor in future years. This increased variability will cause households purchase significant amounts of food and farmers to reduce investments in agriculture unless that the severity of poverty is higher among those farmers are helped to manage this risk (Christiansen reporting a higher food gap. Higher food prices also and Dercon 2011), such as through household irriga- tion where possible, or to insure against these risks. 13 Dercon and Hill (2011) review the agroeconomic literature on the Berhane et al. (2014) show that when farmers are returns to improved seeds and production practices in Ethiopia and sug- gest that increased use of improved maize seeds and production practices provided with access to index insurance that provides can bring about substantial yield gains in Ethiopia. One careful review protection against weather related crop-losses farmers of on-farm trials for wheat suggests that fields with optimal fertilizer ap- plication can produce between 42–109% more than fields without any significantly increase investments in fertilizer and also fertilizer (Teklu et al. 2000). Vandercateelen et al. (2013) show returns improved seeds. Providing the right tools for farmers of 2–17% are available for improved practices in the production of teff. TABLE 4.5: Food gap of poor households, 2005 and 2011 Proportion of poor households that are… Average monthly consumption 2005 2011 in 2011 (Birr per adult) Non-agricultural 11% 12% 2791 Agricultural households with a food 17% 9% 2661*** gap of 9 or more months Agricultural households with a food 25% 16% 2805 gap of 6–9 months Agricultural households with a food 21% 20% 2762*** gap of 3–6 months Net sellers or agricultural households 25% 42% 2816 with a food gap of less than 3 months *** significantly different from agricultural households with a food gap of less than 3 months. 66 ETHIOPIA – POVERTY ASSESSMENT hurt wage employees in the short run if wages do not effect on poverty reduction perhaps because service increase. Few household heads (8%) report being sector growth has been strongly correlated with agri- wage employees, however households that do rely on cultural growth. Growth in agriculture and services has wage labor income are impacted by rising food prices gone hand in hand. This complements the findings in in the short run until wages adjust some 4–5 months Chapter 6 that show that the non-farm sector in rural later (Headey et al. 2012). If higher producer prices Ethiopia is driven by agricultural gains: agricultural are also reflected in higher retail prices, they will hurt profits finance their operation and they tend to oper- food buyers unless there is compensatory intervention ate at times when fellow residents have cash in-hand to improve their wellbeing. Improvements in market from recent harvests. efficiency can help both net producers and consum- The effect of safety nets on poverty reduction, ers by increasing producer’s share of the retail price. even controlling for the sectoral composition of Minten et al. (2012) show that improvements in growth, suggests that they hold potential in help- market efficiency increased farmers’ share of the final ing reach the poorest households that have not teff retail price by 7% from 2001 to 2011. Further been participating in economic growth in recent infrastructure investments and improvements in years. Hoddinott et al. (2013) provides evidence competition in cereal markets will further improve that safety nets have supported agricultural growth market efficiency. Minten et al. (2014) suggests that in Ethiopia. The analysis presented here shows that in particular, investments that allow the last miles of the introduction of the PSNP also reduced poverty access to be improved are needed. through redistribution, in addition to any impact Reducing the gender agricultural productivity through supporting growth. The effect of PSNP gap in Ethiopia is another way in which inclusive coverage on zonal poverty reduction corroborates agricultural growth can be encouraged in Ethiopia. evidence from impact assessments of the PSNP Chapter 9 details the types of interventions that will (Gilligan, Hoddinott, and Seyoum Taffesse 2010; help ensure that female-headed households are able to Berhane et al. 2012) which suggests that the program see increases in agricultural productivity. In particular, has been well targeted to poor households and has the analysis shows that interventions that help female enabled households to acquire and protect assets, farmers access land for cultivation and hire agricultural particularly when safety net payments have been labor will help lower the gender productivity gap. It large and reliable. The evidence provided in the fis- also shows that addressing sources of gender-bias in cal incidence analysis in Chapter 5 suggests that the the types of crops women and men market may be transfers reduce poverty by 6%. Expanding safety important, and would help female farmers realize nets may reduce poverty further. Hill and Porter higher returns on inputs such as fertilizer. (2014) show that although the PSNP is well targeted, Manufacturing growth may play an increasing almost half of the poor households in Ethiopia live role in poverty reduction as Ethiopia urbanizes. In in woredas in which the PSNP is not functioning, urban areas, manufacturing output growth was a more and some very vulnerable areas of the country are not important driver of poverty reduction in recent years covered (such as some lowland areas in Gambela and indicating that growth in this sector may be important Benishangul-Gumez). In addition no urban areas are for poverty reduction. covered by a safety net. Poverty reduction in the service sector has Further analysis on the relative cost of investing contributed to overall poverty reduction, but its in safety nets, roads, education or public invest- contribution has been somewhat lower than its ments to support growth is needed to ascertain large contribution to GDP growth would indicate. which investments would bring about the largest Service sector growth has not had an independent reductions in poverty per Birr invested. 67 A FISCAL INCIDENCE ANALYSIS FOR ETHIOPIA 5 Public investment has been a central element of The analysis shows that income in Ethiopia is the Government of Ethiopia’s development strategy very equally distributed, prior to any redistribution over the last decade. Since the early 1990s Ethiopia by the state through taxes, transfers and subsidies. has pursued a “developmental state” model with the This suggests that other factors in Ethiopia contribute objective of reducing poverty in Ethiopia. In this to keeping the distribution of incomes relatively equal. model, high levels of public sector investment encour- One such factor is the relatively equal distribution of age growth and improve access to basic services. As land in rural Ethiopia as a result of a land policy that Chapter 4 indicated, growth has been the primary allocates land according to need and makes the con- driver of reductions in poverty over the last decade. solidation of land in the hands of a few very difficult. In recent years redistribution has also been an It may be that this policy has other less beneficial important contributor to poverty reduction. This effects on poverty reduction (such as hindering migra- chapter assesses the role of fiscal policy in contrib- tion and structural change—Chapter 7) but it likely uting to that trend. It is an open question as to how also contributes to the equal distribution of pre-fiscal much fiscal policy has contributed to redistribution. redistribution income that this chapter documents. Although Ethiopia has reduced poverty while main- Even though income inequality is low, fiscal pol- taining low levels of inequality, the poorest have not icy still reduces inequality. Fiscal policy has improved fared well in recent years (as documented in Chapter the welfare of those in the bottom decile, and both the 1). Poverty depth did not fall in Ethiopia between 2005 poverty gap and the severity of poverty are also lower and 2011 and the poverty severity index increased. This as a result. Taxes are progressive (the proportion paid chapter assesses the impact of fiscal policy on poverty increases as income increases) and direct transfers are incidence, depth and severity and examines whether made to the poorest households. Subsidies are not there is room for an increased role for fiscal policy in always progressive, and the largest subsidy (on electric- improving the wellbeing of the very poorest. ity) is regressive (comprising a lower share of income for This chapter summarizes findings from the poorer households), but in general spending is progres- first comprehensive analysis of the incidence of fis- sive. In many cases spending is also pro-poor, providing cal policy in Ethiopia. It applies the Commitment more to poorer households in absolute terms. to Equity (CEQ) methodology (Lustig and Higgins However, because Ethiopia is a poor country this 2013) to analyze the distributional impact of fis- reduction in inequality has come about at a cost to cal policy in a holistic and standardized way.14 This many households who are already poor. Poor house- facilitates comparison with other countries in which holds pay taxes—both direct and indirect—and the the CEQ methodology has been applied. The analy- transfers and benefits they receive do not compensate sis assesses the incidence of fiscal policy in 2011, the same year for which poverty estimates were calculated, 14 Led by Nora Lustig since 2008, the CEQ project is an initiative of and includes 83% of tax revenue and 43% of govern- the Center for Inter-American Policy and Research (CIPR) and the Department of Economics at Tulane University, the Center for Global ment spending. Woldehanna et al. (2014) discusses Development and the Inter-American Dialogue. For more details visit the full results. www.commitmentoequity.org 68 ETHIOPIA – POVERTY ASSESSMENT TABLE 5.1: Ethiopia: Tax revenue structure 2011 Million Birr In percent Total Tax Revenue 58,986 100% Direct Taxes 19,554 33% Personal Income Tax 5,733 10% Corporate Income Tax 10,055 17% Ag. Income and Rural Land Use Fee 628 1% Rental Income 377 1% Other Direct Taxes 2,761 5% Indirect Taxes 39,432 67% Domestic indirect taxes 15,706 27% Import duties & taxes 23,726 40% Source: Ministry of Finance and Economic Development (MOFED), Government Finance Account 2011. all households for the taxes they have paid. As a result, equity. In addition paying taxes and receiving benefits although poverty falls as a result of fiscal policy, one in are important aspects of a social contract. By assessing four households are impoverished (either made poor the equity of taxes and spending, the results of this or poor households made poorer15) after direct taxes chapter are one input to public policy making, one are paid and transfers received, and nearly one in 10 that should be weighed with other evidence before households are impoverished when all taxes paid and deciding that a tax or expenditure is desirable. benefits received are taken into account. The analysis presented in this chapter highlights two areas by 5.1  Taxation incidence which this negative impact could be reduced: (i) by reducing the incidence of direct tax on the bottom The structure of Ethiopia’s tax system shares impor- deciles and increasing the progressivity of direct tant features with other underdeveloped economies taxes, particularly personal income tax and agricul- in terms of reliance on indirect taxes and dependency tural taxes, and (ii) by redirecting spending on sub- on international trade (Besley and Persson 2011). sidies to spending on direct transfers to the poorest. Indirect taxes contribute 67% of the total tax collection By considering only the redistributive effects of the general government (Table 5.1). The bulk of indi- of fiscal policy this chapter does not offer a full rect taxes are collected from imports. In 2011 taxes from analysis of whether specific taxes or expenditures imports contributed 40% of the total tax collection. are desirable. When one tax or expenditure is found Analysis of the direct taxes shows that they to be more redistributive to the poor than another, the are progressive and pro-poor. Box 5.1 sets out the temptation is to conclude that the former is preferable. definitions of regressivity and progressivity used in However, redistribution is only one of many criteria this report and Annex 5 details the methodology and that matter when making public policy. Good tax data used in estimations. Note that in the absence of policy will aim to be sufficient, efficient, and simple in income data, the analysis uses the assumption that addition to equitable; and public spending will aim to consumption is equal to disposable income, defined (among other goals) provide the minimal functions of a state (such as security) and invest in necessary public 15 See Higgins and Lustig (2014) for more details on measuring im- goods (such as infrastructure) as well as improving poverishment. A Fiscal Incidence Analysis for Ethiopia 69 BOX 5.1: Terminology It is important to define some basic concepts in incidence analysis as the distributive impact of fiscal policy depends on the extent of progressivity of taxes and transfers. The terms progressive and regressive can be used in two different senses: in absolute and relative terms. Following Lustig and Higgins (2013) the following definitions are used.a Progressive: a subsidy (or tax) is progressive if it is progressive in relative terms, that is, if the proportion of the subsidy (or tax) relative to income decreases (increases) with household income. Pro-poor subsidies and transfers: a subsidy/transfer is pro-poor if it is progressive in absolute terms, that is, if the absolute (i.e., per capita) amount of the subsidy/transfer decreases with household income (and therefore if the share of total spending is higher for lower income deciles). Subsidies/transfers that are not pro-poor: a subsidy is not pro-poor if it is regressive in absolute terms, that is, if the absolute amount of the subsidy increases with household income (and therefore if the share of total spending is higher for higher income deciles). Regressive: a subsidy (or tax) is regressive if it is regressive in relative terms, that is, if the proportion of the subsidy (or tax) relative to income increases (decreases) with household income. Using these definitions, spending can be progressive (i.e., equalizing) but not necessarily pro-poor. a All these definitions apply exactly when the net fiscal system does not cause re-ranking. If there is re-ranking, they are a very good approximation. as income after direct cash transfers, net of taxes and low as a share of GDP, the direct tax that the poorest contributions. Figure 5.1A orders households accord- decile pays as a share of income is higher than that paid ing to their market income, defined as household in all other countries considered. The share of market income from wages, salaries, interest income, private income paid in tax is almost constant from the first to transfers, and pensions, which is constructed based the eighth decile, only increasing for the top deciles. on the prevailing tax legislation. Against this metric, Disaggregating the types of direct taxes paid Figure 5.1 shows that the burden of direct taxes is reveals that although personal income tax and highest for the top decile, while the bottom 50% of the rental income tax are progressive, agricultural income distribution pays less than 2% of their market income tax is regressive which contributes to the income. In fact, the concentration shares shown in relatively high tax burden on the poorest. A good Table 5.2 show that the top 10% of the distribution way to compare the progressivity of taxes is to com- contribute 55% of total direct taxes, while the bottom pare the Lorenz curve with concentration curves for 50% contributes less than 15% of total direct taxes. each of the taxes. A concentration curve is constructed Although the average incidence of direct tax similarly to Lorenz curves but the difference is that collection is relatively low in Ethiopia compared to the vertical axis measures the proportion of the tax some other countries, Ethiopia levies more direct paid by each quantile (with the households ranked taxes on the poorest decile than any other country by income on the horizontal axis). This is done in considered. Typically the collection of direct taxes is low Figure 5.2A for direct taxes. Overall direct taxes are for lower income countries (Besley and Persson 2011), (everywhere) progressive, as the cumulative share of however, for Ethiopia’s level of GDP, direct tax collection tax paid by each quantile of the population is lower is remarkably high (Figure 5.1B). For example, direct than their share in market income.16 In particular, taxes are a higher share of GDP in Ethiopia than in Guatemala even though its GDP per capita is more than 16 This analysis assumes that tax evasion is constant across income levels, which may over-estimate the progressivity of direct taxes if richer house- seven times the GDP per capita of Ethiopia. In addition, holds are more able to evade tax payments and underestimate progressivity although direct tax collection in Ethiopia is relatively if poorer households are more likely to evade payments. 70 ETHIOPIA – POVERTY ASSESSMENT FIGURE 5.1: Incidence of direct taxes by market income deciles A. Ethiopia B. Selected Developing Countries 6% 25% 16% Tax as share of market income 14% Share of GDP (right axis) Share of market income 5% 20% 12% 4% 15% 10% 3% 8% 10% 6% 2% 4% 1% 5% 2% 0% 0% 0% Poorest decile 2 3 4 5 6 7 8 9 Richest decile Armenia (2011) Brazil (2009) El Salvador (2011) Ethiopia (2011) Guatemala (2010) Mexico (2010) Peru (2009) South Africa (2010) Uruguay (2009) Poorest decile Richest decile Share of GDP (right axis) Source: Own estimates based on HCES 2011. Source Armenia: Younger et al. 2014; Brazil: Higgins and Pereira 2014; Mexico: Scott 2014; Peru: Jaramillo 2014, South Africa: Inchauste et al. 2014; Uruguay: Bucheli et al. 2014 and Lustig (2014) based on Beneke et al. 2014; and Cabrera et al. 2014. For Ethiopia, own estimates based on HCES 2011. the PIT and rental taxes are quite progressive as their presents the incidence of the three main types of concentration curves are everywhere below the Lorenz direct taxes paid relative to the market income of curve for market income. In contrast, the agricultural each decile. As shown, the agricultural land tax makes income tax and land use fee and other direct taxes are up a larger share of the market income of the poor- regressive as the share paid by the poorest quantiles is est deciles compared to the higher income deciles. higher than their share in market income. Figure 5.2B Agricultural households are likely to be poorer than TABLE 5.2: Average per capita direct taxes in Birr per year and concentration by decile Share of total taxes paid by Cumulative share of total Market income decile Average per capita (in Birr) decile (%) taxes paid by decile (%) 1 151 1.7 1.7 2 226 2.5 4.2 3 266 2.9 7.1 4 308 3.4 10.5 5 370 4.1 14.6 6 471 5.2 19.8 7 504 5.6 25.3 8 676 7.5 32.8 9 1156 12.7 45.5 10 4942 54.5 100.0 Source: Own estimates based on HCES 2011. A Fiscal Incidence Analysis for Ethiopia 71 FIGURE 5.2: Concentration curves and incidence of direct taxes A. Concentration Curves for Direct Taxes B. Incidence of direct taxes (percent of market income decile) 1.0 0.2% 6% Cumulative proportion of tax 0.8 5% % of market income 4% 0.6 0.1% 3% 0.4 2% 0.2 1% 0 0% 0% Poorest 2 3 4 5 6 7 8 9 Richest 0 0.2 0.4 0.6 0.8 1.0 Cumulative proportion of the population Population shares Personal income tax Ag. income tax Rental income tax Lorenz curve for market income Ag. income and land use fee Personal income tax (right axis) Direct taxes Rental income tax Other direct tax Source: Own estimates based on HCES 2011. non-agricultural households, and this may be one incidence of indirect taxes is assessed with respect reason why the agricultural income tax may appear to disposable income (which is defined as the sum regressive when considered on its own. In addition, of market income plus direct transfers, net of direct agricultural income tax rules are set by regional and taxes) because households make their consump- local governments and are mainly levied according to tion decisions taking into account government cash land holding size, which does not necessarily deter- transfers as part of their income, and therefore con- mine income earned. In only a few places are assets sume (and are taxed) more than what their market such as cattle size also considered. For the most part income would allow them in the absence of these per hectare tax rates do not increase with land hold- transfers. Although VAT, customs duties and excise ing size (for example, in Oromia they tend to slightly taxes apply to everyone at the same rate on the pur- fall with land-holding size as detailed in Annex 5). chase of goods or services, regardless of the level of However, personal income tax is the largest income of the household, indirect taxes are progressive direct tax levied on individuals, and although it (Figure 5.3A).17 The progressivity has been achieved is progressive it is striking to note that the aver- because higher tax rates are applied to those goods age tax rate is constant across the first five deciles consumed more by richer households (see Annex 5 at 1.1% of market income. Currently any personal for a discussion of the tax system). For example, the income above 150 Birr per month (or 1800 Birr per richest decide income group spend ten times more year) is taxed. This is much less than the poverty line than the poorest decile on alcohol and beverages as a of 3781 Birr per adult equivalent and increasing this share of total spending and these products have among minimum cut-off would reduce the direct tax burden the highest taxes rates of excise tax. on the bottom deciles. The loss in tax revenue that Comparatively, Ethiopia’s indirect taxes relative this would comprise could be compensated by higher to GDP are average, but indirect taxes are a lower personal income tax rates on higher deciles. Indirect taxes are slightly progressive in 17 This analysis assumes that effective tax rates are equal across house- Ethiopia, taxing a higher share of the pre-tax holds, which may underestimate the progressivity of indirect taxes (if income of the richest deciles (Figure 5.3). The richer urban households are more likely to purchase in formal markets). 72 ETHIOPIA – POVERTY ASSESSMENT FIGURE 5.3: Incidence of indirect taxes by market income deciles A. Ethiopia B. Selected Developing Countries 100% 20% 16% Share of disposable income 14% 80% 15% 12% Share of GDP 60% 10% 10% 8% 40% 6% 5% 4% 20% 2% 0% 0% 0% 0 1 2 3 4 5 6 7 8 9 10 Bolivia Brazil Ethiopia Indonesia Mexico South Africa Disposable Income Import duty Indirect Taxes Excise tax VAT Population shares Poorest decile Richest decile Share of GDP Source: Own estimates based on HCES 2011. Source: Bolivia: Paz et al. 2014; Brazil: Higgins and Pereira 2014; Indo- nesia: Jellema et al. 2014; Mexico: Scott 2014; South Africa: Inchauste et al. 2014. For Ethiopia, own estimates based on HCES 2011. share of market income than in all other countries VAT, and excise taxes is a slightly progressive tax sys- considered, and in most cases also more progressive. tem. This compares well to middle income countries For instance, while indirect taxes amount to 3% of dis- considered and is similar to Peru’s tax system where posable income of the poorest decile in Ethiopia, they indirect tax systems are also progressive. amount to 18% of the disposable income of the poor- However, although indirect taxes are more est decile in Bolivia, and 11% in Brazil (Figure 5.3B). progressive in Ethiopia compared to other coun- The combined incidence of personal income taxes, tries, they are still less progressive than direct taxes. The concentration curves of both direct and indirect taxes are further away from the 45-degree line than the Lorenz curve of market income, which FIGURE 5.4: Direct and indirect tax indicates that they are both progressive and decrease concentration curves in relation to market inequality (Figure 5.4).18 However, the concentra- income Lorenz curve tion curve for direct tax is much to the right of the 1.0 curve for indirect tax documenting that direct taxes Cumulative proportion of tax 0.8 are indeed much more progressive than indirect taxes in Ethiopia. 0.6 In aggregate, taxes are low and progressive 0.4 compared to other countries, but because Ethiopia is a poor country, the share of the tax bill paid by 0.2 households living under US$1.25 PPP a day is very 0 high, highlighting the fundamental challenge of 0 0.2 0.4 0.6 0.8 1.0 Cumulative proportion of the population Population share Direct taxes 18 Note that taxes cannot reduce poverty as they can inequality because Market income Indirect taxes they reduce incomes. The best case from a distributional perspective would be that no poor people pay taxes and the FGT remains unchanged Source: Own estimates based on HCES 2011. after the tax. A Fiscal Incidence Analysis for Ethiopia 73 pro-poor revenue generation in a low income coun- FIGURE 5.5: Concentration of total taxes try. Together, Figures 5.1 and 5.3 indicate that taxes across socioeconomic groups, cross-country are relatively low and progressive in Ethiopia. However, comparison even though this is the case, the share of the total tax 100% burden paid by households living on less than US$1.25 90% 80% a day is much higher in Ethiopia than in other countries 70% as Figure 5.5 indicates. This highlights the challenge fac- 60% ing Ethiopia: even with low and progressive taxes, taxes 50% make many poor households poorer and some non-poor 40% 30% households poor. To the extent possible taxes should be 20% made more progressive to limit the impoverishing effect 10% of taxes. It is perhaps unlikely that Ethiopia can reduce 0% Armenia Bolivia Brazil El Salvador Ethiopia Guatemala South Africa its reliance on indirect taxes or make them more pro- gressive given how well it compares to middle income countries on these fronts, but consideration should be 50.00 <= y 10.00 <= y < 50.00 given to the extent that direct taxes can be made more 4.00 <= y < 10.00 2.50 <= y < 4.00 progressive. For example, the minimum income above 1.25 < = y < 2.50 y < 1.25 which personal income tax is levied could be raised, and Note: “y” is market income. Source: Armenia: Younger et al. 2014; Brazil: Higgins and Pereira, 2014; South Africa: Inchauste et al. agricultural income taxes can be made more progressive 2014; Lustig (2014) based on Beneke et al, 2014 and Cabrera et al., by encouraging a higher per hectare tax rate for house- 2014. For Ethiopia, own estimates based on HCES 2011. holds with larger land holdings. 5.2  Incidence of public expenditure plan. The pro-poor sectors of the GTP are agriculture and food-security, education, health, roads and water, Public spending is guided by the Growth and and accordingly 70% of total general government Transformation Plan (GTP) and is particularly expenditure is allocated to these sectors. Table 5.3 targeted to the pro-poor sectors identified in this indicates how government spending is allocated. TABLE 5.3: Ethiopia: General government expenditure 2011 Million Birr In% Total General Gov. Expenditure 93,831 100% General Services 15,655 17% Economic Development 38,422 41% o/w Agriculture 14,183 15% Road 18,318 20% Social Development 32,936 35% o/w Education 23,345 25% Health 6,307 7% Urban Dev’t and Housing 2,762 3% Labor and Social Welfare 179 0% Source: Ministry of Finance and Economic Development (MOFED), Government Finance Account 2011. 74 ETHIOPIA – POVERTY ASSESSMENT Education spending comprises the highest share of The following subsections present findings on the total spending (25%), followed by roads and agricul- progressivity of each type of spending and conclude ture at 20% and 15% respectively. About half of the with a discussion on the overall progressivity of gov- agricultural budget is allocated to the ongoing food ernment spending. security and Productive Safety Net Program (PSNP). Health spending accounts for 7% of the general gov- Direct transfers made through the PSNP ernment budget. and food aid This incidence analysis covers 43% of all gov- ernment spending, mostly covering social spend- Direct transfers made in the PSNP and food aid ing. It assesses the incidence of spending on education programs are progressive and pro-poor with more and health, and half of the agricultural budget (that than 58% of the benefits going to households below spent on the PSNP). Spending on general services the national poverty line. Direct transfers are pro- and roads were not included given the difficulty of gressive in relative terms, as measured by the benefits attributing benefits to specific households. Non- received by the poorest deciles relative to their market PSNP agricultural spending and spending on urban income (Figure 5.6A), as well as in absolute terms, as development and housing were not included in the measured by the share of benefits received by each analysis at this stage, given data challenges, but can decile (Figure 5.6B). In fact 66% of all direct transfers be considered in future work. were concentrated in the bottom 40% of the market The government also subsidizes items off- income distribution. budget through the operation of public enterprises The finding that PSNP transfers are more and funds and the analysis also includes some off- progressive than emergency food aid reflects the budget spending. In 2011—the year for which this findings of the broader literature on food aid tar- incidence analysis was conducted—the government geting in Ethiopia and the results of PSNP external subsidized electricity, kerosene and wheat through the evaluations. Food aid is targeted to communities operations of Ethiopian Electric Power Corporation particularly affected by disasters, and while there is (EEPCO), the Oil stabilization Fund and Ethiopian often targeting of poor households within these com- Grain Trade Enterprise (EGTE). These are included munities this is done in an ad hoc fashion in order to in the fiscal incidence analysis. They are off-budget ensure aid is provided in a timely manner. As a result operations that are not included in general govern- targeting errors in the selection of individuals at the ment finance. Electricity subsidies to households are local level can be quite high. The PSNP has clear tar- the main indirect subsidy with an estimated benefit of geting rules and identification of beneficiaries and as Birr 1.5 billion (equivalent to 2.6% of general govern- a result targeting errors have been found to be much ment budget) to households in 2011. Kerosene was lower (Gilligan et al. 2010). subsidized in 2011 through the Oil Stabilization Fund. Transfers made in the PSNP and food aid have The government also subsidizes wheat to reduce the a sizeable direct effect on poverty, reducing it by effect of food inflation on the urban poor. In 2011, two percentage points. Given both PSNP and emer- the government had a program of import and distri- gency food aid are progressive and pro-poor, they both bution of wheat in Addis Ababa at a subsidized price , made substantial contributions to poverty reduction which was later expanded to other regional towns. The (Table 5.4). The direct effect of these transfers reduced transfer was not targeted and the sales were rationed poverty rates from 33% to 31% (estimated by com- to all households of the city through local adminis- paring consumption with and without the size of the trative units (kebeles). The estimated subsidy was Birr transfer provided). The transfers reduce the poverty 150 per quintal of wheat. gap by 1.4 percentage points (14.3 percent) and reduce A Fiscal Incidence Analysis for Ethiopia 75 FIGURE 5.6: Ethiopia. Direct transfers by market income deciles A. Direct Transfers as Share of Income B. Concentration Curves for Direct Transfers (by market income decile) 18% 1.0 16% Cumulative proportion 14% 0.8 Benefit as share of market tincome 12% of transfer 0.6 10% 8% 0.4 6% 4% 0.2 2% 0% 0 Poorest decile 2 3 4 5 6 7 8 9 Richest decile 0 0.2 0.4 0.6 0.8 1.0 Cumulative proportion of the population PSNP Food aid Population shares Food aid Market Income Lorenz PSNP Source: Own estimates based on HCES 2011. the squared poverty gap by 0.9 percentage points transfer programs, targeting them to more households. (21.5 percent). However, they do make up about 20% of market In comparison to other countries, PSNP transfers income of the poorest decile, which is comparable to are effective at reducing poverty but could become what direct transfers do in Mexico (31%) and more more so, and could also become more generous. The than what direct transfers achieve in Indonesia and Peru. effectiveness of PSNP transfers is calculated as the per- centage point reduction in poverty headcount as a ratio Education of the share of transfers to GDP. On this comparison, the PSNP compares well to other countries but there Overall, spending on education is progressive in is room for improvement, with transfers in Indonesia, relative terms but only primary education spend- South Africa, and a number of Latin American countries ing is pro-poor. Table 5.3 documented the large share proving to be more effective (Figure 5.7). In terms of of public spending going to education. Half of this their generosity, direct transfers from PSNP and food spending was to tertiary education, of which a con- aid make up a smaller share of market income of the siderable amount was spent on building universities. poorest deciles when compared to countries such as Once investments in university buildings in 2011 South Africa, Argentina, Uruguay, or Armenia sug- were distributed across 10 years, spending on primary gesting that there is room to increase the size of direct education comprises the largest share of education TABLE 5.4: Poverty indicators before and after PSNP and food aid transfers Before transfers After transfers Head count ratio (US$1.25 PPP) 32.9% 30.9% Poverty Gap (US$1.25 PPP) 9.5% 8.2% Squared Poverty Gap (US$1.25 PPP) 4.1% 3.2% Source: Own estimates based on HCES 2011. 76 ETHIOPIA – POVERTY ASSESSMENT FIGURE 5.7: Effectiveness of direct transfers in decile the value of primary education benefits received comparison to direct transfers in other countries is 5.6% of market income compared to 0.5% for the –14% 4.0 richest decile. The absolute amount of primary educa- 3.5 tion benefits received by poor households is also larger –12% 3.0 than those received by rich households (not shown), 2.5 –10% and as a result education spending is pro-poor in addi- 2.0 –8% tion to being progressive. Secondary education spend- 1.5 1.0 ing is also progressive in relative terms, comprising a –6% 0.5 larger share of market income for poor households 0% 0.0 than for rich households, but it is not pro-poor; richer South Africa (2010) Argentina (2009) Brazil (2009) Uruguay (2009) Armenia (2011) El Salvador (2011) Bolivia (2009) Ethiopia (2011) Mexico (2010) Guatemala (2010) Peru (2009) Indonesia (2012) households receive a larger share of the secondary education spending (Figure 5.8B). Tertiary education is neither progressive nor pro-poor. It is regressive in that the direct benefits go more to richer households. Change in poverty rate Effectiveness indicator (right axis) Students in the richest decile receive 40% of spending Source: Argentina: Lustig and Pessino 2014; Armenia: Younger et on tertiary education, while the poorest decile only a., 2014; Bolivia: Paz et al. 2014; Brazil: Higgins and Pereira 2014; receives 2.5% of spending. However, tertiary spending Indonesia: Jellema et al. 2014; Mexico: Scott 2014; Peru: Jaramillo 2014; South Africa: Inchauste et al. 2014; Uruguay: Bucheli et al. has beneficial impacts on national growth rates and 2014; and Lustig (2014) based on Beneke et al, 2014 and Ca- service delivery (for example through more educa- brera et al. 2014. For Ethiopia, own estimates based on HCES 2011. Note: Poverty line of US$1.25 PPP is used for Ethiopia. For all the tion to primary school teachers) and should not be other countries the poverty line is US$2.5 PPP. reduced; rather a focus on increased access for poorer families is needed. Low enrollment rates in secondary and tertiary spending. Figure 5.8 shows that spending on primary education limit the progressivity of spending on education as a proportion of market income is very non-primary education. Primary education is avail- high for poorer households: for those in the poorest able in almost all villages in Ethiopia, resulting in high FIGURE 5.8: Ethiopia. Incidence and concentration shares of education A. Incidence (share of benefits by market income decile) B. Concentration Curves Cumulative proportion of benefits 6% 1 Share of market income 5% 0.8 4% 0.6 3% 0.4 2% 0.2 1% 0% 0 Poorest decline 2 3 4 5 6 7 8 9 Richest decline 0 0.2 0.4 0.6 0.8 1 Cumulative proportion of the population Primary Secondary Tertiary Population shares Market Income Education total Primary Secondary Tertiary Source: Own estimates based on HCES 2011 and WMS 2011. A Fiscal Incidence Analysis for Ethiopia 77 FIGURE 5.9: Ethiopia: Incidence and concentration shares of health A. Health Benefit Incidence as Share of Income (by market income decile) B. Health Benefit Concentration Curves Benefit as share of market income Cumulative proportion of benefits 5% 1 4% 0.8 3% 0.6 2% 0.4 1% 0.2 0% 0 Poorest decile 2 3 4 5 6 7 8 9 Richest decile 0 0.2 0.4 0.6 0.8 1 Cumulative proportion of the population Population shares Market Income Lorenz Health Preventive Curative Source: Own estimates based on HCES 2011 and WMS 2011. enrollment, which reached 96% in 2013, but second- received by households in the poorest deciles are higher ary schools are found only in limited, mostly urban, as a share of their market income than benefits received areas. Although secondary education (like primary by higher income deciles (Figure 5.9A). However, education) is free, those living far from secondary health spending is not pro-poor (Figure 5.9B). About schools have to pay for travel and sometimes boarding 9% of health spending is concentrated in the poorest costs (if the distance makes daily commuting prohibi- decile, while 14% is concentrated in the richest decile. tive) for children to attend. These costs are prohibi- Nevertheless, this difference in the concentration of tive for the poorest families and as a result secondary spending is not as large as in other countries such as Peru. enrollment rates are much lower in poorer deciles than Health extension agents are present in all in richer deciles. Since many poor parents may not kebeles and ensure that a basic range of health be able to afford to send their children to secondary services are readily available to all households. education in nearby towns, spending on secondary This ensures that preventative health care spend- education is not as pro-poor as that of primary educa- ing—which is about 27% of overall health spend- tion. A quarter of total secondary education spending ing—is progressive in relative terms but curative benefits the richest decile, compared to only 5% that health care is less progressive. Although preventive benefits the poorest decile. Completion of second- health care services are provided for free, marginal ary school is a prerequisite for tertiary enrollment, so user fees are usually charged for curative public inequalities in secondary school enrollment are also health services, which are much lower than the cost reflected in tertiary enrollment, despite stipends for of service. To protect the poor against the financial attendance available to all households. burden of user fees, there are fee waiver and exemp- tion systems at the public health center and hospital. Health However, poorer households are less likely to avail themselves of curative health services and as a result Spending on health is progressive in relative terms, public spending on preventive health care is more and even though it is not pro-poor its progressivity progressive in relative terms than spending on cura- compares well to other countries. Health benefits tive health (Figure 5.10). 78 ETHIOPIA – POVERTY ASSESSMENT FIGURE 5.10: Health benefit incidence as subsidies are progressive, benefiting the poor in relative percent of income by market income decile terms more than the rich. In contrast, electricity com- 5% 1.6% prises a smaller share of spending among poorer house- holds than among richer households and as a result 4% 1.2% electricity subsidies are highly regressive (Figure 5.11A). 3% The richest 30% of the population received 65% of 0.8% 2% electricity subsidies while the poorest 30 percent—those 0.4% living below the national poverty line—obtained only 1% 10% of the subsidy for electricity. Among these three 0% 0% subsidies, electricity is the largest subsidy. 1 2 3 4 5 6 7 8 9 10 Subsidies are more progressive among the Curative Preventive (right axis) urban population they are designed to benefit, Source: Own estimates based on HCES 2011 and WMS 2011. but they are less progressive than direct transfers. Indirect subsidies are designed to benefit the urban poor who are particularly reliant on purchases of these Indirect subsidies goods and who do not benefit from direct transfer pro- grams that are present in rural areas. Figure 5.12 shows Indirect subsidies are present for electricity, kerosene that urban households do benefit more than rural and wheat, and although they are progressive for households from subsidies, and that subsidies are, on wheat and kerosene in relative terms they are highly aggregate, progressive for urban households. However regressive for electricity. Poorer households consume the figure also shows that subsidies in urban areas are less electricity, kerosene, and wheat than richer house- not as progressive as direct transfers, and that the size holds and as a result none of these subsidies are pro-poor of subsidies relative to direct transfers is low. Poverty, (Figure 5.11B). However, wheat and kerosene comprise particularly urban poverty, would be reduced further a larger share of spending among poorer households were spending on indirect subsidies (on electricity, than among richer households and as a result these two kerosene and wheat) converted to direct transfers. FIGURE 5.11: Incidence and concentration curves for indirect subsidies A. Ethiopia. Incidence of Indirect Subsidies (share of benefits by market income decile) B. Concentration Curves 0.7% Cumulative proportion of subsidy 1 0.6% Share of market income 0.8 0.5% 0.4% 0.6 0.3% 0.4 0.2% 0.2 0.1% 0% 0 Poorest 2 3 4 5 6 7 8 9 Richest 0 0.2 0.4 0.6 0.8 1 Cumulative proportion of the population Electricity Kerosene Wheat Population share Market Income Lorenz Electricity Kerosene Wheat Source: Own estimates based on HCES 2011. A Fiscal Incidence Analysis for Ethiopia 79 FIGURE 5.12: Transfers and subsidies as a proportion of consumption in rural and urban Ethiopia Direct transfers Subsidies 0.2 0.2 Subsidies as a proportion Transfer as a proportion 0.15 0.15 of consumption of consumption 0.1 0.1 0.05 0.05 0 0 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10 Decile of consumption Decile of consumption Urban Rural Urban Rural Source: Own estimates based on HCES 2011. Another objective of electricity subsidies is to Overall incidence of public spending encourage the use of electricity among Ethiopian households. In 2011, just 17% of households in Overall, the progressive nature of taxes is comple- Ethiopia spent anything on electricity, and the house- mented by progressive social spending, however holds that had access were the wealthier households less than half of total spending is pro-poor. Of the (Figure 5.13). Subsidizing access to electricity by sub- total social spending included in the study, 81% of the sidizing the cost of connections may be a better way spending is progressive, of which 44% is pro-poor and to encourage electrification in Ethiopia than subsidiz- 37% is not pro-poor. Nineteen percent of spending is ing the use of electricity which benefits those already regressive (Figure 5.14). with connections. Bernard and Torero (2011) show Spending on direct transfers in the PSNP is that subsidizing the cost of a connection increases particularly pro-poor, while spending on subsidies electrification in rural Ethiopia. FIGURE 5.14: Ethiopia. Public Expenditure FIGURE 5.13: Proportion of households with programs (percent of spending included in electricity (%) by market income category analysis) 0.7 0.25 0.6 0.20 0.5 0.4 0.15 0.3 0.10 0.2 0.1 0.05 0 0 Less than Between Between Between More than PSNP Food aid Primary education Secondary education Health Kerosine subsidy Wheat subsidy Tertiary education Electricity subsidy USD 1.25 USD 1.25 USD 2.5 PPP USD 4 PPP USD 10 PPP PPP PPP and USD and USD and USD 4 PPP 10 PPP 2.5 PPP Progressive and pro-poor Progressive, but not pro-poor Regressive Source: Own estimates based on HCES 2011. Source: Own estimates based on HCES 2011 and WMS 2011. 80 ETHIOPIA – POVERTY ASSESSMENT is never pro-poor and sometimes regressive. Overall Incidence of taxes and 5.3  Disaggregating spending, it is clear that spending on spending and impact on poverty the PSNP and food aid and primary education is not and inequality only progressive, but also pro-poor (Figure 5.15). Of these three programs, spending on the PSNP is the This section documents the overall impact of fis- most progressive. Figure 5.15 shows that while wheat cal policy on poverty and inequality. A household’s and kerosene subsidies, health, education, secondary income prior to the payment of taxes and receipt of education are progressive, they are not pro-poor. The benefits (market income), is first compared to a house- electricity subsidy and tertiary education are regres- hold’s income after all direct taxes have been paid and all sive as their concentration coefficients are greater than direct transfers have been received (disposable income). the Gini coefficient for market income. Once indirect txes and subsidies are included, this is Moving resources from off-budget subsidies called post-fiscal income. Post-fiscal income excludes into direct transfer programs targeted to the in-kind benefits to households for health and educa- populations the subsidies are designed to ben- tion. Last, market income is compared to final income efit (such as the urban poor) would improve the which takes into account all taxes paid—both direct progressivity of public spending. If all financing and indirect—and all benefits received—both direct of subsidies were used to provide transfers to poor transfers and in-kind benefits received through subsi- households at the same level of effectiveness, this dies or in-kind receipt of health and education services. would result in a further 2% reduction of the pro- When focusing on disposable income of house- portion of the population living below the national holds, the results show that the top 30% are net pay- poverty line. It would also result in a 5% reduction ers to the government and the bottom 40% are net in the poverty gap and a 12.5% reduction in the recipients (Figure 5.16). As a result Table 5.5 shows severity of poverty. that direct taxes and transfers reduce poverty by one FIGURE 5.15: Concentration coefficients of public spending Total Social Spending Gini of market income Wheat subsidy Kerosine subsidy Electricity subsidy Health Tertiary education Secondary education Primary education Education Food Aid PSNP –0.5 –0.4 –0.3 –0.2 –0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 Concentration coefficient Progressive and pro-poor Progressive but not pro-poor Regressive and not pro-poor Source: Own estimates based on HCES 2011 and WMS 2011. A Fiscal Incidence Analysis for Ethiopia 81 TABLE 5.5: Poverty and inequality indicators before and after taxes and spending Market Income Disposable Income Post-fiscal Income Final Income National Poverty Line Incidence 31.2% 30.2% 32.4% Gap 9.0% 7.9% 8.7% Severity 4.3% 3.1% 3.4% US $1.25 a day Incidence 31.9% 30.9% 33.2% Gap 9.2% 8.2% 8.9% Severity 3.9% 3.2% 3.5% Gini coefficient 0.322 0.305 0.302 0.302 Source: Own estimates using HCES 2011 and WMS 2011. percentage point (measured using the national poverty FIGURE 5.16: Ethiopia. Incidence of taxes lines, as well as the extreme international poverty line and transfers (by market income deciles) calculated in PPP terms) and reduce inequality by 30% two percentage points (as measured by the Gini coef- Share of market income ficient).19 The poverty gap and the severity of poverty 20% also decline with transfers, so that the overall effect is 10% a decline in these indicators. However, when focusing on post-fiscal income, the results show that all but the 0% bottom 20% are net payers to the government. The poverty headcount rate for post-fiscal income increases –10% Poorest 2 3 4 5 6 7 8 9 Richest to 32.4 percent, which signals the fact that total gov- ernment transfers and subsidies do not make up for In-kind education In-kind health Indirect subsidies the impact of indirect taxes around the poverty line Indirect taxes Direct taxes Direct transfers (Table 5.5). However, although the headcount ratio Source: Own estimates based on HCES 2011 and WMS 2011. goes up, the poverty gap and poverty gap squared pov- erty are lower for post-fiscal income indicating that on average, the poorest of the poor receive net transfers. poverty rates after in-kind transfers because house- Once in-kind transfers are included, the net holds may not be aware of the actual amount spent on impact of all fiscal policy is progressive with all their behalf and may not value this spending as much but the top 20% receiving more benefits relative as they would value a direct cash transfer. However to their market income than the taxes they pay Figure 5.16 shows that the effect of taxes is compen- (Figure 5.16) and as a result fiscal policy reduces sated for by the services households receive in the form inequality in Ethiopia. The overall decline in inequal- of education and health services. ity is 2.3%. This is not surprising given the progres- sive nature of taxes, and the overall progressivity of social spending. PSNP transfers and primary educa- 19 Note that typically Ethiopia measures welfare using a household con- sumption aggregate, which this analysis sets equal to disposable income. tion spending drive this result. Following standard Using the National Moderate Poverty line, poverty headcount is 30%, conventions, this analysis refrains from calculating coinciding with the official headcount rate for 2010/11. 82 ETHIOPIA – POVERTY ASSESSMENT TABLE 5.6: Impoverishment and fiscal policy in Ethiopia National US$1.25 PPP Impoverishment Headcount Index (% of population impoverished) Market income to disposable income 25.0 25.6 Market income to final income 9.1 9.3 Percentage of population that was non-poor and became poor (%) Market income to disposable income 0.9 0.9 Market income to final income 1.1 0.9 Source: Own estimates using HCES 2011 and WMS 2011. Although on average poverty does not increase received into account. When all taxes paid and benefits with fiscal policy, fiscal policy impoverishes 25% received are considered, 9% are still impoverished by of households, when considering disposable fiscal policy; that is by moving from market income income, and 9% of households, when considering to final income. In both cases, 1% of the population final income. Standard incidence measure can fail was non-poor and became poor. to capture the extent to which the poor are further Reducing the burden of taxation and improving impoverished by tax and benefit systems. Therefore the the progressivity of spending by redirecting fund- standard incidence results are substantiated by impov- ing for subsidies to expanded rural and urban direct erishment index analysis, proposed by Higgins and transfer programs can reduce the impoverishment Lustig (2013). The impoverishment headcount index of poor households. The minimum income at which measures the percentage of the population impover- income tax is levied can be increased and agricultural ished by the tax and transfer system. Households who tax rates can also be made more progressive in order are impoverished are those who are either non-poor to reduce the burden on households living beneath the before taxes and transfers and made poor by the fis- national poverty line. Spending on subsidies can be cal system, or they are poor before taxes and transfers made more progressive if it were provided via transfers and made even poorer by the tax and transfer system. to the population it is targeted to benefit—the urban Table 5.6 summarizes the impoverishment indices poor—as in the rural safety net program. Chapter 8 at various poverty lines: from market to disposable discusses how a transfer program of 0.2% targeted to income to post-fiscal income and to final income. poorest households in Addis Ababa could halve the One quarter of the population were made poorer as a current poverty rate in Addis Ababa. This is the same result of direct taxes, even when taking direct transfers as the cost of electricity subsidies to the richest 40%. 83 NON-FARM ENTERPRISES AND POVERTY REDUCTION IN ETHIOPIA 6 6.1 Introduction the proportion of households engaged in non-farm activities is very low and there is no clear evidence of In addition to being the primary sector of activ- recent growth in this sector since 2008 when 25% of ity for 11–14% of the population, a further 11% households reported owning a non-farm enterprise. of rural households earn about a quarter of their The rural non-farm sector is estimated to account for income from operating non-farm enterprises between 35–50% of household earnings in the devel- (NFEs) in the service sector. This income is earned oping world and an average of 34% of rural earnings largely during harvest months and months immedi- across Africa (Haggbalde et al. 2010). The numbers ately following harvest. The income earned from these reported in this chapter from 2011–12 suggests it com- activities improves the wellbeing of households and its prises about 10% of household earnings in Ethiopia. role in reducing poverty can be missed in the standard Additionally NFEs in Ethiopia are largely complemen- decomposition and growth regression techniques used tary to agriculture and driven by growth in this sector. in Chapter 4 of this report. As a result they are not able to provide any activity and Ascertaining the impact of these types of service income for households during the lean season and they sector activities on poverty reduction is difficult do not allow households to manage any agricultural but this chapter provides more information on the losses they might experience. The close dependence of amount of income these activities generate and for NFE activity on agricultural income means that this is which types of households. These individuals often not a driver of poverty reduction on its own. have a primary categorization in agriculture and the An initial assessment of constraints to NFEs non-farm income they earn is highly correlated with suggests that interventions to increase demand will agricultural income, causing growth analyses to attri- have the largest impact on increasing the vibrancy bute this impact to agricultural growth. Simply ascer- of this sector and its role in reducing poverty. On taining whether households with NFEs are poorer or the supply side, NFEs appear to depend on agri- richer than other households also does not address this cultural income for inputs and investment capital. question. If households with NFEs are richer it could On the demand side, they rely heavily on increased be that operating NFEs is a means by which some local demand during the harvest period to generate poor, uneducated households grow their incomes and household income. The need for capital does not escape poverty. On the other hand, it could simply be appear to be a major cause for the current seasonal- the case that it these households are already better off ity of NFEs, but many do report access to market and are able to invest in high-return NFE activities, demand as a major constraint. Increasing demand and are thus more likely to operate them. will require further investments in infrastructure, While NFEs provide some secondary income increased employment the manufacturing sector on in rural areas and a source of income for those non-seasonal service sector activities, and increased unable to secure employment in rural towns, the agricultural revenues. contribution of this sector is small in comparison The chapter uses detailed data on the liveli- to other countries. In comparison to other countries, hoods of households in rural and small-town 84 ETHIOPIA – POVERTY ASSESSMENT TABLE 6.1: Types of NFEs 1 Overall Small Town Rural Difference (1) (2) (3) (2)–(3) (1) Non-agricultural services from home/ household-owned 0.283 0.429 0.279 0.150*** shop (e.g. mechanic, tailor, barber) (.029) (.048) (.029) (2) Processing/sale of agricultural by-products (e.g. flour, 0.285 0.256 0.286 –0.030 excluding livestock by-products and fish) (.030) (.044) (.031) (3) Trading business on a street/market 0.242 0.223 0.242 –0.019 (.035) (.042) (.036) (4) Sale of products/services offered on a street/in a market 0.098 0.052 0.099 –0.047** (e.g. firewood, mats, bricks) (.016) (.016) (.017) (5) Professional office, professional services from home (e.g. 0.012 0.006 0.012 –0.006 doctor, translator, midwife) (.004) (.003) (.005) (6) Transportation or moving services (e.g. driving a house- 0.013 0.017 0.013 0.004 hold-owned taxi or pick-up truck) (.005) (.012) (.005) (7) Bar/restaurant ownership 0.006 0.045 0.005 0.040*** (.002) (.015) (.002) (8) Other non-agricultural business from home/on a street 0.148 0.129 0.148 0.019 (.023) (.029) (.024) N 1,113 286 827 Source: ERSS 2011–12. Note: Proportions do not add up to 1.0 because some households qualified NFEs with several responses. Standard errors corrected for clustering and stratification in parentheses. * p < 0.10, ** p < 0.05, *** p < 0.01. Ethiopia collected in the 2011–2012 Ethiopian of the definition of NFE ownership is similar to that Rural Socioeconomic Survey (ERSS). The full analy- of Rijkers and Söderbom (2013), and consistent with sis and results can be found in Kowalski et al. (2014). the broader literature, allowing for comparability of The ERSS sample is representative of rural Ethiopia results. The exhaustive list of NFE types found in the and towns less than 10,000 people. The data includes survey is displayed in Table 6.1. A household was con- both those counting the service sector as a primary sidered to operate a NFE in the survey if it reported to occupation—the service sector in rural Ethiopia and have operated one or more of these types of enterprises small towns comprises 67% self-employed activities— in the twelve months prior to the survey, including and those who engage in service sector activities but those ventures that had been shut down permanently as a secondary activity. or temporarily during that time. In Ethiopia, NFE activity is primarily concen- Prevalence and nature of NFEs in 6.2  trated in processing and sale of agricultural prod- Ethiopia ucts, trade of other products or offering a service from home or a shop. The most prevalent NFE type NFE ownership is defined as the operation of a is the processing and sale of agricultural byprod- nonfarm enterprise involved in the provision of ucts (Table 1), which is strongly tied to agricultural non-agricultural services such as carpentry, the pro- activities. Twenty-eight percent of NFE-operating cessing and sale of agricultural by-products such households operate this type. A further 28.3% of as flour, trade, professional services, transportation NFE-owning households offer a service from home or services, and food services. This operationalization a household-owned shop and 24.2% trade in a market NON-FARM ENTERPRISES AND POVERTY REDUCTION IN ETHIOPIA 85 TABLE 6.2: Proportion of households operating an NFE (%) ERSS (2011–2012) RICS (2008) Woldehanna and Oskam (2001) Tigray 19 22 28 Amhara 16 20 Oromia 16 23 SNNPR 25 37 Other regions 33 Source: Kowalski et al. 2014. or on a street. Stark differences in the prevalence of FIGURE 6.1: Age of NFEs non-agricultural NFEs that are operated from home or Density a shop exist between rural areas and small towns with over 40% of households in small town areas reporting 0.2 to operate a NFE of this kind. One in five households in rural Ethiopia own an NFE. NFEs dominated economic activities in small 0.1 towns with 55% of small town households operat- ing at least one NFE. On the basis of the ERSS, it can 0.0 be estimated that there exist approximately 2.9 million 0 10 20 30 40 Age NFEs in Ethiopia with 20.2% of all households in Source: ERSS 2011–12. rural and small town areas owning at least one NFE. There does not appear to have been much growth in NFE ownership in recent years. The proportion of households owning an NFE is slightly of a few older NFEs (Figure 6.1). Only 17% of NFEs lower than the NFE participation rate of 25% esti- that were reported to be in operation at the time of mated by Loening et al. (2008) for the four largest the survey were 10 years or older. A third of all NFEs regions of Oromiya, Tigray, SNNP, and Amhara.20 It were started in the year leading up to the survey. In also varies from Woldehanna and Oskam (2001) who the absence of clear evidence of high growth in the estimate that 28% of households were engaged in any proportion of households operating an NFE, this nonfarm activity, with 7% engaged in nonfarm self- suggests significant churn in the operation of NFEs.21 employment according to Tigray data. Although the NFEs in small towns are somewhat older with entry samples of these surveys are different, a comparison rates almost half those of rural areas (17.5%), a mean does not suggest noticeable growth in this sector. The age of 8.3 years and a median age of 4.1 years as well rates of NFE ownership by region are presented in as a higher percentage of NFEs being at least 10 years Table 6.2. old (27.6%). NFEs are found to be mostly young with a mean age of approximately six years and a median 20 In a World Bank report (2009) using the same data (RICS-AgSS) age of two years. This is similar to the mean age of NFE participation rates are placed within a broader range of 10–35%. NFEs estimated by Loening et al. (2008). The differ- 21 It will be possible to conduct further analysis on the exit rates of NFEs, ascertain the reason for closure and identify if households with closed ence in median and mean arises because of the high NFEs open new ones in a different or similar sector when the second number of very young enterprises and the presence wave of the ERSS data becomes available. 86 ETHIOPIA – POVERTY ASSESSMENT TABLE 6.3: Prevalence of NFEs by per adult equivalent expenditures 1st Quintile 5th Quintile (poorest) 2nd Quintile 3rd Quintile 4th Quintile (richest) NFE ownership 0.168 0.151 0.198 0.263 0.213 (.028) (.021) (.028) (.026) (.027) N 407 849 773 998 882 Source: ERSS 2011–12. Notes: Standard errors corrected for clustering and stratification in parentheses. The differences in NFE prevalence between the consumption quin- tiles are statistically significant at the 5% level. The role of NFE in incomes of poor 6.3  This gap narrows but becomes statistically significant households for households in rural areas. In addition house- holds that operate an NFE are less likely to own There is some indication that it is the less educated livestock, suggesting these are alternate sources of households in small towns that operate NFEs, as livelihood for households. Households without a opposed to more educated households in rural NFE own more livestock than NFE households in areas. The average years of schooling of a house- both subpopulations with a comparably sized gap. hold head is 2.4 years for NFE-owning households, Households who are unable to support themselves whereas this drops to 1.7 years for households that solely off their land may be more likely to diversify do not own a NFE. The results are mainly driven by into NFE operation. the rural subpopulation and show an opposite pat- There is no significant difference in the rates of tern in small towns. In rural areas, NFE household NFE ownership between male and female-headed heads have an average of 0.5 more years of education households in rural areas, nor in small towns once than households without a NFE suggesting that bet- other characteristics of households such as educa- ter educated households may be better equipped to tion and consumption per capita have been taken choose to engage in NFE activities. Conversely, in into account. In small towns female-headed house- small towns they have on average 3.3 fewer years of holds represent a greater proportion of households education than households without a NFE, point- with NFEs at 38.3% than without NFEs at 29.3%. ing to higher education potentially providing better This suggests a slightly more conservative role of access to public sector and wage jobs. Households in Ethiopian women22 in the NFE sector than noted small towns do have higher access to wage jobs with elsewhere (Rijkers and Costa 2012; Loening et al. over 15% of those seven years and older working in 2008) but this dissipates in regression analysis. wage jobs, compared to less than 3% of those in rural One out of every six households (16%) in the areas; most of the wage jobs are with the government bottom 40% operate an NFE, but rates of NFE or private enterprises. ownership are higher among non-poor households: NFE participation is more prevalent among one out of every four households (24%) in the top households with lower landholdings per head, 40% own an NFE. Table 6.3 displays the prevalence which may indicate some households are pushed by necessity into NFE operation. In small towns, where households generally have very little land, 22 It should be noted that most of the female-headed households are widowed and this could potentially limit their access to land. This could those that do not operate a NFE own on average increase their likelihood to operate an NFE. However, we only find an more than double the land assets of NFE-owners. increased likelihood of women in NFEs in the small towns. NON-FARM ENTERPRISES AND POVERTY REDUCTION IN ETHIOPIA 87 of NFEs by household adult equivalent annual con- expenditures than those households not engaged in sumption expenditure quintiles.23 Prevalence is higher NFE activity. among higher quintiles with 26.3% and 21.3% of Over half (54%) of NFE operating households households in the fourth and fifth quintiles, respec- report that NFEs generate approximately a quarter tively, operating a NFE. NFE households do not of their income: these are households for whom appear to be significantly less likely to be food insecure. service sector activities contribute significantly to This is in contrast to the finding in Beegle and Oseni household welfare but who do not report their pri- (2008) that finds that households with non-farm mary sector of occupation as services. Households enterprises are more likely to be food secure in rural in small towns, however, report more often to be Ethiopia even after controlling for the distance to the generating a share of around half or three quarters of nearest major agricultural market and to the nearest total household income through the operation of a all-weather road. nonfarm business, and 21.9% indicate that it generates Households with NFEs in rural areas have the household’s entire income, compared to only 4.5% higher levels of consumption on average, but it is in rural areas. Calculating average annual incomes per not clear whether those households that are bet- NFE, we find a median annual NFE income of 700 ter off are better able to engage in NFE activities, Birr. Median annual NFE income in small towns is or whether NFE activities help some households 1600 Birr, relative to a much lower value of 650 Birr in become less poor. Analysis using panel data in Bezu rural areas, indicating that most small town NFEs are et al. (2012) also find that higher consumption growth generating more income than their rural counterparts. is positively correlated with a higher initial share of However there are some NFEs that earn much higher non-farm enterprise income. Households with NFEs levels of income and this is indicated in the fact that in rural areas consume an average of 280 Birr more the mean income in rural areas is much higher than the per annual adult equivalent than those without NFEs, mean income in small towns. Using the source of con- which rely primarily on agriculture. The difference in sumption data in the 2011 Household Consumption consumption is significant when household size and Expenditure survey suggests that nationally, 10% of education and age of the household head have been consumption is funded through non-agricultural controlled for. It could be that operating NFEs is a household enterprises. means by which some poor, uneducated households In combination with the prevalence rates grow their incomes and escape poverty. On the other reported in Table 3 this suggests that service sec- hand, it could simply be the case that it is those tor activities not reported in official surveys, pro- households already better off, that are able to invest in vide a non-negligible source of income for about high-return NFE activities, and are thus more likely 9% of households living in poverty in Ethiopia. to operate them. Households for whom NFE activities comprise more There is no difference in the consumption level than half of their income will report this as a primary of households with and without an NFE in small sector of occupation in national surveys. Some NFEs towns. In small towns, households that operate NFEs, on average, consume approximately 250 Birr less per annual adult equivalent than households that do 23 Annual consumption expenditures include measures of food consump- not. However, these differences are not significantly tion, non-food expenditures, and educational expenditures, indexed for regional spatial price. Costing of consumption of own produced food is different. The results for small towns are similar to done by taking the median local price per gram. To ensure stable prices, those from Rijker and Söderbom’s (2012) study of a local price is only defined when a minimum of 10 unit prices are avail- able. Hence, if an enumeration area has 10 price observations for a given Amhara in which households that run a NFE are not good, the median of these price observations is taken. If less than 10 were found to have considerably higher per adult annual available, the median price for the kebele is used, etc. 88 ETHIOPIA – POVERTY ASSESSMENT TABLE 6.4: Annual agricultural profits per hectare NFE No NFE Difference (1) (2) (1)-(2) Agricultural profits (mean) 3,905.866 4419.889 –514.023 (1,002.634) (609.558) Agricultural profits (median) 3,412.611 3,995.523 N 646 1,947 Source: ERSS 2011–12. Notes: Standard errors corrected for clustering and stratification in parentheses. Standard errors are not reported for medians as we were unable to bootstrap in order to obtain them. This is due to the fact that there is little literature at the intersection of variance estimation in the presence of complex sample design and bootstrapping. This analysis attempts to use replicate weights, but median estimation using them was not possible. * p < 0.10, ** p < 0.05, *** p < 0.01. earn negligible service sector activities and thus con- The data offers no support to the hypothesised tribute little to household welfare. role of NFE ownership as an insurance mecha- NFEs generate, on average, one sixth of the nism. In a number of settings NFE income allows returns generated by a hectare of land used for households to become more resilient in the face of agricultural production. Table 4 reports a measure agricultural shock such as weather. However, in the of mean agricultural profits per hectare per year, which case of NFEs considered here, NFE and non-NFE suggests that households that own a NFE do not report households report statistically similar incidence of significantly different agricultural profits per hectare decreases in income, assets, food production, food than households that do not own a NFE. The median stocks, and food purchases (see Figure 6.2) which NFE increases household income by 20% and gener- suggests NFE income does not mitigate the negative ates the income equal to about 0.16 to 0.18 hectares effects of a shock. Regression analysis shows that being of land. This finding, as well as the overall picture of exposed to a shock is associated with a 31 percentage NFEs painted in this section, present further evidence point increase in the probability of being food insecure to support the claim that NFEs represent only limited but there is no significant indication that NFE owner- income-generating opportunities for households. ship is associated with a lower likelihood of reporting being food insecure conditional on receiving a shock. When interpreted in light of the strong links FIGURE 6.2: Households’ reaction to shocks between NFEs and agricultural production as well as the local nature of NFE markets discussed below, Decrease in income the result that NFEs do not significantly reduce Decrease in assets household vulnerability to aggregate weather shocks is somewhat unsurprising. Dependency on Decrease in food production seasonal local markets, which are highly susceptible Decrease in food stocks to weather shocks, renders NFE households likewise exposed to risk. Decrease in food purchases NFEs have little impact on other households 0% 25% 50% 75% 100% as few employ workers outside of the household non-NFE households that owns them. The average number of workers per NFE households NFE is 1.5, a figure that is consistent with findings Source: ERSS 2011–12. of Loening et al. (2008) and Söderbom and Rijkers NON-FARM ENTERPRISES AND POVERTY REDUCTION IN ETHIOPIA 89 (2012). Most workers on any enterprise are household typically lasts between September and February workers with an average of 1.2 household workers per (Taffesse et al. 2011). If NFE activity began or peaked NFE. The number of hired workers per NFE is sub- during the lean season, and thus was counter-cyclical stantially lower with an average of 0.3 for all NFEs. with agriculture, this would provide some prima facie Approximately 71.4% of all enterprises employ only evidence that NFEs aid households to smooth con- one person, a further 17.0% employ two workers sumption throughout the year. However, the opposite and only the remaining 11.6% employ three or more is observed: NFE activities strongly correspond to workers. Less than one in 10 NFE has a formal license. the timing of the Meher season. NFEs tend to begin operation largely coinciding with the timing of the 6.4  Constraints to NFE activities Meher season as shown in Figure 6.3, and these are also the highest months of NFE activity (Figure 6.4 The majority of NFE-operating households reported and Figure 6.5). Nonfarm enterprises tend to be most that the activities of their NFEs were seasonal. A third active during the months of November, December, of small town NFEs reporting to be seasonal compared and January, with 42.7%, 44.5%, and 32.2% of NFEs to 54% of rural NFEs. NFE’s of a larger income size listing these as one of their three most important and of longer duration in the market are less seasonal. months of activity. NFE seasonality does not appear to be significantly The pro-cyclical nature of the activity suggests associated with household annual per adult equivalent that supply side considerations (for example the expenditures, nor rural location, once NFE age and need for inputs from agricultural production) or income are controlled for. These results vary and have demand concerns (for example demand from an significantly different policy implications from those agriculturally financed consumer base) are impor- of Loening et al. (2008), who show that rural NFEs tant determinants of NFE activity. The association are highly seasonal but countercyclical with agriculture. between NFE start-ups and the main agricultural The seasonality of NFE activities coincides with period suggests that business activity was taken up in the agricultural season; NFE activities are pro- anticipation of or in response to highly active agricul- cyclical not counter-cyclical with agriculture. The tural activities and heightened local demand. The vari- main harvest period in Ethiopia, or the Meher season, ability in timing of NFE start-ups is less pronounced FIGURE 6.3: Seasonality of NFE creation 25% 20% % of NFE creation 15% 10% 5% 0% September October November December January February March April May June July August Pagune Overall Small Town Rural Source: ERSS 2011–12. 90 ETHIOPIA – POVERTY ASSESSMENT FIGURE 6.4: Highest months of NFE operation 60% % of households reporting highest months of NFE 50% 40% 30% 20% 10% 0% September October November December January February March April May June July August Pagune Overall Small Town Rural Source: ERSS 2011–12. FIGURE 6.5: Harvest season and NFE operation, by type NFE sector 80% 60% % of households 40% 20% 0% September October November December January February March April May June July August Pagune Harvest Season Retail NFEs Farm-related NFEs Transport, communication, storage NFEs Utility services Source: ERSS 2011–12. in small towns than in rural areas, a finding that is On the supply-side, most households rely on expected given that small town NFEs tend to be agricultural income to fund the creation of NFEs. less seasonal and less strongly linked in agricultural Overall, agricultural income is reported to be either production. In rural areas, most NFEs list the top the primary or secondary source of start-up capital for three months of activity as November, December 64% of NFEs (see Table 6.5). Loening et al. (2008), and January and in small towns NFEs list December, found that agricultural income represented 60% of January, and February as the most important months start-up capital for NFEs. NFE households report for activity. There thus appears to be a small, one- the next important source of start-up capital to be month lag in peak NFE activity between small town nonfarm self-employment income, noted as a primary and rural sub-populations. This lag may indicate a or secondary source of funds by 18% of households. rural supply-chain trend. This result can be explained by the fact that some NON-FARM ENTERPRISES AND POVERTY REDUCTION IN ETHIOPIA 91 TABLE 6.5: Source of start-up funds for NFEs Overall Small Town Rural Difference (1) (2) (3) (2)–(3) Agricultural income 0.642 0.137 0.657 0.520*** (.030) (.020) (.031) NFE self-employment 0.175 0.369 0.169 0.200*** (.024) (.049) (.025) Family/friends 0.116 0.312 0.111 0.201*** (.018) (.040) (.018) Money Lender 0.076 0.095 0.076 0.019 (.017) (.027) (.018) Microfinance Institution 0.029 0.045 0.028 0.017 (.009) (.013) (.009) Wage employment 0.016 0.088 0.014 0.074*** (.004) (.020) (.004) Remittances 0.003 0.005 0.003 0.002 (.002) (.003) (.002) Sale of assets 0.009 0.011 0.009 0.002 (.004) (.006) (.004) Bank loan 0.006 0.014 0.006 0.008 (.003) (.009) (.003) Other 0.055 0.101 0.054 0.047 (.011) (.026) (.011) N 1,315 345 970 Source: ERSS 2011–12. Standard errors corrected for clustering and stratification in parentheses. * p < 0.10, ** p < 0.05, *** p < 0.01. Note: Columns do not sum to one as numbers account for the proportion of NFEs reporting each source as either a primary or secondary source of start-up capital. households operate multiple NFEs and may thus use demonstrates the temporal nature of NFEs for both the income from one NFE to start another. farming and non-farming households. Although there Rural NFEs tend to rely more heavily on agri- is a statistically significant difference in the propor- cultural income for start-up capital than small town tion of households reporting September, October, and NFEs, with 65.7% of rural households citing agri- November as a high month for NFE activity, there is cultural income as a main source of funds for NFEs, no significant difference in the overall trend through- as opposed to only 13.7% of small town households. out the year for farming and non-farming households. This result can be explained by the greater prevalence Despite the fact that non-farm households cannot rely of nonfarm activities in towns, and the stronger direct on agricultural income to fund the operation of their links with agriculture in rural areas. NFEs, they still exhibit increased NFE activity from However, further exploration suggests that November to February. agricultural income’s contribution to starting an The customer base of most NFEs appears to NFE only partially explains the cyclical relationship primarily comprise the local market, local consum- between NFE and agriculture activity. Figure 6.6 ers or passers-by, and traders, indicative of the local 92 ETHIOPIA – POVERTY ASSESSMENT FIGURE 6.6: NFE activity for Farming and non-Farming households 60% highest months of NFE activty % of households reporting 50% 40% 30% 20% 10% 0% September October November December January February March April May June July August Pagune Farming households Non-farming households Source: ERSS 2011–12. nature of the markets they serve. Locals and passers- often insufficient to generate sizeable NFE income by constitute a somewhat higher share of the customer throughout the year. Table 6.6 lists the three most base in small towns than in rural areas, with 41.6% commonly cited constraints to NFE growth, all of and 29.5% of NFEs reporting this as one of their two which are related to markets. While 37.3% of NFEs main customer bases, respectively. Additionally, sell- identified access to markets as one of three main ing to traders appears to be more common for rural obstacles, another 21.0% and 16.9% viewed low NFEs, as 16.8% of rural households reported traders demand and difficulty to obtain market information, as a main customer base, relative to 10.4% of small respectively, as key constraints. Therefore, the top three town NFEs. constraints identified, out of more than 30 catego- NFEs perceive low demand and a lack of access ries, were all related to markets. A lower proportion to better markets as major operational barriers. This of NFE households in small towns reporting access provides further evidence that local market demand, to markets as a constraint to growth (23.0%) than in which is mainly driven by seasonal agriculture, is rural areas (37.7%). TABLE 6.6: The three main constraints to NFE growth Overall Small Town Rural (1) (2) (3) Access to markets 0.373 0.230** 0.377** (.036) (.060) Low demand for goods/services 0.210 0.274 0.209 (.023) (.055) Difficult to obtain information about the market 0.169 0.148 0.169 (.026) (.033) N 1,382 362 1,020 Source: ERSS 2011–12. Notes: Standard errors corrected for clustering and stratification in parentheses * p < 0.10, ** p < 0.05, *** p < 0.01. NON-FARM ENTERPRISES AND POVERTY REDUCTION IN ETHIOPIA 93 6.5 Conclusion local demand during the harvest period to generate household income. The need for capital does not NFE activity is seasonal and pro-cyclical with agricul- appear to be a major cause for the current seasonality ture. On the supply side, NFEs appear to depend on of NFEs, but many do report access to market demand agricultural income for inputs and investment capital. as a major constraint. On the demand side, they rely heavily on increased 95 MIGRATION AND POVERTY IN ETHIOPIA 7 7.1 Introduction characteristics of migrants. It shows that one of the main reasons migration has not contributed to Migration from rural to urban areas is an inherent poverty reduction is that there is so little of it. The component of the development process. As such, it evidence presented in this section is however consis- has long been a focus of the development literature tent with “pull” forces driving migration in Ethiopia, (Lewis 1954; Fei and Ranis 1964). Whereas initial suggesting it should have positive effects when it theories effectively suggested that migration took place takes place. primarily so that migrants could obtain higher returns Second, the chapter examines the poverty to their human capital (“pull” migration), more recent and welfare effects of migration. It examines theory suggests that migration may also arise as part whether the limited effect of migration on poverty of a household strategy to overcome other constraints is because migration has been welfare reducing (“push” migration, Lucas and Stark 1985). Such con- when it occurs. Instead of “pull” migration, which straints can include credit constraints, liquidity con- is usually beneficial for both the origins and desti- straints, or a lack of insurance against risk. In either case, nations, “push” migration due to adverse income consistent with higher returns to human capital outside and other shocks in the origins can lead to growth of agriculture, migration should be poverty reducing. of slums in urban areas (World Bank 2009; Fay and As an example of its potential poverty reducing effects, Opal 2000; Barrios, Bertinelli, and Strobl 2010; Beegle, de Weerdt, and Dercon (2011) report that in Gollin, Jedwab, and Vollrath 2012). However, the Tanzania, the poverty rate among those who moved evidence indicates that although migrants may out of Kagera region dropped by 23 percentage points suffer welfare losses in the initial year after transi- compared with a 12 percentage point drop among those tion, there do appear to be substantial benefits to who moved within the region and a four percentage migration. point drop among those who did not move. Finally the chapter reviews current evidence However, since 1996, migration and the struc- on constraints to migration in Ethiopia. The very tural change that it brings contributed very little focus of government policy that has been so beneficial to poverty reduction in Ethiopia (see Chapter 4). for rural poverty reduction in Ethiopia may act as an This chapter examines why the role of migration in implicit barrier to migration by improving produc- poverty reduction in Ethiopia has been so limited. tivity and safety nets in rural areas preferentially over Have migration rates been too low to have an impact? urban locales. However, this would not explain why Has migration had positive or negative effects when in the presence of welfare gains few still choose to it has taken place—on those migrating or on families migrate. There is some evidence that credit constraints being left behind? If beneficial effects are found, what may limit the ability of poor households to invest in constrains migration and limits the beneficial role it migration. Limited land markets in rural areas also could potentially play? act as a break on migration flows. First the chapter documents the speed Various data sources are used in the evi- and nature of migration in Ethiopia and the dence presented. The Household Consumption 96 ETHIOPIA – POVERTY ASSESSMENT Expenditure Survey and the Urban Employment and small relative to other countries. For example, using Unemployment Surveys do not collect information on a similar definition, about 30% of India’s population the migration status of respondents. This considerably can be classified as migrants, and India is a country limits analysis of the employment and welfare status known for low population mobility. In Vietnam in of migrants and their sending households. The 2013 1992, 22% of the population were migrants and in Labor Force Survey did collect some information on Uganda in 2001, 25% of people aged 25 to 49 were this and this is likely to provide some additional key not living in the district of their birth (World Bank evidence on migration but the data is not yet available. 2009, World Bank 2012). A relatively large proportion This chapter thus draws on two background papers of the Ethiopian population continues to live in rural commissioned for the Poverty Assessment which draw areas (Taylor and Martin 2001) for its per capita GDP. on two alternative sources of information: (i) the pub- There is little data on temporary migration within licly available sample data of the long-form of the 2007 Ethiopia. The ERSS 2012 data suggests temporary Ethiopian Population and Housing Census which migration flows are not large. identified which respondents are migrants (Shilpi and Migration between rural areas accounts for Yao 2014), and (ii) a unique panel dataset from 18 nearly half of all who migrate for work. It is useful to villages in which migrants were tracked over time (de distinguish between those who migrate for work and Brauw 2014).24 This second data set is the Ethiopian those who migrate for other reasons such as marriage. Rural Household Survey (ERHS) from 2004 and Although the census data did not collect information 2009, and the migrant tracking survey conducted on the reasons for migration, employment status is after the 2009 round of data collection in which all used to define those who migrated for work. A migrant migrants from these villages between 2004 and 2009 is defined as “working” if he or she was employed in were tracked. The data was collected by Addis Ababa productive activities during the last 12 months even University, the International Food Policy Research if partially. Nearly all migrants in Ethiopia (86%) are Institute and the University of Oxford and uniquely working migrants. As shown in Figure 7.1, nearly half provides information on sending households before of all migrants are rural-to-rural migrants. Migrants and after they sent a migrant, and on migrants them- from rural to urban areas comprise about 25% of all selves. Together the nationally representative 2007 migrants in Ethiopia. This suggests about one in ten census and the in-depth data of the ERHS allow a rural residents migrate, in contrast to one in five rural fairly comprehensive assessment of migration and workers in China (World Bank 2009). As expected, poverty in Ethiopia at the end of the last decade. urban to rural migration is very small whereas there Additional insights are provided from literature on is considerable migration between urban areas. The poverty and wellbeing in Ethiopia. ERHS migrant panel was able to identify those who migrated for work and finds very similar patters: 50% 7.2  Migration in Ethiopia of migrants from these rural villages moved to other rural areas, 31% moved to urban areas and 18% The migration rate in Ethiopia is low compared moved outside of Ethiopia (a category of migrant that with most developing countries. Counting a migrant is not found in the census data). as any individual who resides in a different woreda or city to the one of their birth (a definition of migration 24 The publicly available data consists of 2% of the entire sample survey that is considered most expansive), the Intercensal data and thus has about 1.3 million individual records from 289345 Population Survey conducted by CSA in 2012 reveals households. Such a large dataset is useful in portraying the pattern of migration with statistical precision. Most household surveys have a very that migrants comprise 13.7% and 16.2% of the male small sample of migrants and are thus ill-suited for the analysis of migra- and female population, respectively. This is quite tion at finer geographical details. MIGRATION AND POVERTY IN ETHIOPIA 97 FIGURE 7.1: Migration Flow, 2007 FIGURE 7.2: Migrants by duration of stay in 60% current residence 50% 50% 45% 40% 40% 35% 30% 30% 25% 20% 20% 15% 10% 10% 5% 0% 0% Rural to Rural Rural to Urban Urban to Rural Urban to Urban Short-term Medium Term Long Term All Migrants Working Migrants All Migrants Urban Migrants Source: 2007 census. Source: 2007 census. Despite the low overall migration rate, migra- smaller: medium-term migrants are 15% and 18% tion rates have been increasing. Migrants in the for all and urban migrants respectively. census are categorized as short-term migrants who In Addis Ababa migrants represent nearly half are in their current residence for less than five years, of the population, but recent migration to urban medium-term migrants who came to their current centers has favored smaller cities. Addis Ababa residence between five and nine years ago, and long- accounts for 4% of Ethiopia’s population but 10% of term migrants who migrated 10 or more years ago. all migrants (all those not residing in the woreda or city Short-term migrants account for 40% of all and 43% of their birth) and 22% of urban migrants. However, of all migrants to urban areas (Figure 7.2). Migration smaller towns have had higher rates of recent migra- rates in the five years prior to that had been much tion than large cities. Figure 7.3 shows that migration FIGURE 7.3: City size and migration Those who migrated in the last 12 months Those who migrated 5+ years ago 0.25 0.5 Migration within the past year Migration over 5 years ago 0.20 0.4 Addis Ababa 0.15 0.3 0.10 0.2 0.05 Addis Ababa 0.1 0.00 0.0 10 11 12 13 14 15 10 11 12 13 14 15 In(city size) In(city size) Fitted values of quadratic prediction do not include Addis Ababa Source: 2007 census. 98 ETHIOPIA – POVERTY ASSESSMENT FIGURE 7.4: Migration and employment Employment among existing residents and migration rates Unemployment among existing residents and migration rates for last 7 days (excl. recent migrants) Proportion of labor force unemployed 0.6 0.30 Proportion of labor force employed 0.25 (excl. recent migrants) Addis Ababa 0.4 Addis Ababa 0.20 0.15 0.2 0.10 0.0 0.05 0.00 0.05 0.10 0.15 0.20 0.25 0.00 0.05 0.10 0.15 0.20 0.25 Migration within past year Migration within past year Fitted values of linear prediction do not include Addis Ababa Source: 2007 census. in the year prior to the census favored smaller cities. suggests that males form the majority of international The proportion of the city comprised of those that migrants, although younger migrants to the Middle migrated in the 12 months prior to the census falls East are more likely to be girls. with city size. In contrast the proportion of people in Migration is nearly always in the form of a the city that migrated over five years ago is higher in child leaving a household and migrating, very few larger cities. This could reflect the fact that small towns whole families migrate, and as a result migrants are just stopping places, with migrants quickly moving are quite young at the time of migration. The on to larger cities, or that there was comparatively less ERHS migrant tracking survey found that only 5% migration to smaller towns before 2006. of migrants had migrated with their entire family, and Migration to urban centers is strongly cor- related with the labor market in the destination. Rates of migration to cities have been higher in cities FIGURE 7.5: Proportion of migrants and with higher rates of employment, and lower rates of non-migrants that are male unemployment (Figure 7.4). This is consistent with 58% pull factors being a strong determinant of migration. 56% Most migrants are women, but men account 54% for a larger share of working migrants. The share 52% of males in total migrants is less than 50% in all 50% cases with the exception of urban to rural migrants 48% (Figure 7.5), but men form a larger share of work- 46% 44% ing migrants that move to urban areas. Women may 42% migrate not only for work but also for marriage and Rural to Rural to Urban to Urban to Rural Urban other family obligations. The ERHS data, which is Rural Urban Rural Urban better able to distinguish those who migrated for work Migrants Non-Migrants and other reasons show that the 62% of migrants All Work who migrate for work are male. The ERHS data also Source: 2007 census. MIGRATION AND POVERTY IN ETHIOPIA 99 in all cases this was migration to other rural areas. In FIGURE 7.6: Age distribution of those who most cases—80% of the time—migration is of a child migrated in last 5 years within the family, and the average age of a migrant is 45% 23–26 years at time of migration. Taking all migrants 40% together, there is no difference in the age distribution, 35% 30% but many of the migrants in this group migrated many 25% years prior. The age distribution of those who migrated 20% between 2003 and 2007 is presented in Figure 7.6 15% and shows that migrants are much younger than the 10% non-migrant population: 81% of all recent migrants 5% are less than 30 years old compared to 56% of the 0% 15–20 21–30 31–40 41–50 51+ non-migrant population in this age category. Non-migrants Those who migrated in last 5 years Migrants in Ethiopia, as in other developing Source: 2007 census. countries, tend to be more educated than non- migrants, and this is suggestive of pull factors encouraging migration. About 73% of non-migrants FIGURE 7.7: Education levels: Migrants and have education up to 5th grade or below whereas non-migrants, 2007 the share is 43% for migrants (Figure 7.7). This is 80% unlikely to be an age effect, as the age distribution of 70% all migrants is similar to that of non-migrants. About 60% 44% of migrants have education level between 6th to 50% 12th grade, and another 13% have higher secondary 40% (more than 12th grade) education. These compare far 30% better than non-migrants among whom 25% have 20% education between 6–12 grade and only 2% have 10% higher than 12th grade of education. However, in 0% comparison to non-migrants at destination, migrants <=Grade 5 Grade 6–12 Higher Education to Addis Ababa appear to be a little less educated: for Migrant Non-migrant example 50% of migrants in Addis Ababa completed Source: 2007 census. grades 6–12, compared to 55% of non-migrants. Migrants from rural to urban areas come from wealthier families, which would also be consistent characteristics of the household. The relationship with rural-urban migration driven by pull factors. between agricultural productivity and migration is However this is not true for all migrants, with migrants particularly strong in the lower half of the land distri- to other rural areas and international destinations bution. For households in the lower half of the land coming from poorer households within the ERHS distribution, a 10% increase in agricultural revenue migrant tracking survey. results in a 0.45% increase in the likelihood of send- Households with higher levels of agricultural ing a migrant. A 10% increase in agricultural revenue production are more likely to send a migrant even has no positive impact on the likelihood of sending a once controlling for other factors. There is a small, migrant for those in the top half of the distribution. positive correlation between migration and the value These results suggest that good harvests allow poorer of agricultural production by the household, and this households to overcome credit constraints, which correlation is robust to controlling for many other may otherwise constrain migration (see Section 7.4). 100 ETHIOPIA – POVERTY ASSESSMENT Although the age and relative wealth of migrants FIGURE 7.8: Employment Status of working suggests that pull factors characterize migration, migrants and non-migrants, 2007 push factors also play a role. Households with more 70% male adults are more likely to have a migrant. In 60% particular households with more adult males for a given 50% land size are more likely to send a child to migrate. 40% Regression analysis showed that the relative abundance 30% of labor within the household is an important predic- 20% tor of migration in rural Ethiopia. A household with an additional adult male is eight percentage points 10% more likely to send a migrant, even when holding 0% Government Self employed Unpaid family Other many other characteristics constant. This suggests worker that although it may be relatively well-educated adults Migrant Non-migrant from wealthier households that migrate, there is some Source: 2007 census. degree of necessity that encourages migration to occur. 7.3  Migration and poverty diversification regardless of whether it is within rural The characteristics of migrants suggest that pull areas or from rural to urban areas. Such employment factors may be a strong driver of migration in diversification is usually associated with higher house- Ethiopia, with migrants leaving their place of birth hold incomes and lower poverty incidence. to find a better job. What evidence is there that they Although migrants as a whole are more likely to attain the improvements in living standards that they be employed than non-migrants this is not the case expected when they left home? This section examines for recent migrants. Census data shows that employ- the welfare of migrants. ment rates among those who migrated in the last 12 Migrants are more likely to be employed and months are lower (6.7%) than employment rates less likely to be self-employed than non-migrants among those already resident in the city (8.9%). On in both rural and urban destinations. Migrants are average the proportion of recent migrants (those who less likely to be employed as unpaid family work- moved to the city in the last year) who are employed ers and self-employed compared with non-migrants is higher than the proportion of recent migrants who (Figure 7.8). For instance, more than 60% of non- are unemployed, but employment rates are lower. migrants are self-employed compared with 50% of Migrants live in smaller houses after migration, migrants; 27% of non-migrants are unpaid family but this may indicate smaller household sizes rather workers compared to 16% of migrants.25 This is than higher levels of poverty, as access to electricity consistent with evidence from other countries. The and tap water is higher among recent migrants than World Development Report 2009 documented that non-migrants. Migrants who migrated within five years in 24 out of 35 countries considered, migrants were of the survey to both urban and rural areas were more equally or more likely to be employed than locally native people of working age. Migrants are however 25 This analysis defines four broader employment categories: self-employ- ment, unpaid family worker, government employee and “other” which more likely to be government employees or employees accounts mainly for employees in private organizations/households. The of private firms/households. This is true regardless of analysis is carried out only for those individuals who reported their em- ployment status, which caused a drop in sample size (about 128 thousand whether a migrant migrates to a rural or an urban area. migrants and 492 thousand non-migrants). All the comparisons are also This suggests that migration facilitates employment for working population. MIGRATION AND POVERTY IN ETHIOPIA 101 FIGURE 7.9: Number of Rooms in the House Urban Rural 70% 70% 60% 60% 50% 50% 40% 40% 30% 30% 20% 20% 10% 10% 0% 0% Non-migrants ST Migrants MT Migrants LT Migrants Non-migrants ST Migrants MT Migrants LT Migrants 1 Room 2 Rooms >=3 Rooms Source: 2007 census. Note: ST Migrants are migrants who migrated in the last 5 years, MT Migrants are migrants who migrated between 5 and 10 years ago, and LT Migrants refer to those who migrated more than 10 years ago. likely to live in one-room houses than non-migrant the propensity of migrants to live in one-room houses households: 48% of recent migrants live in one room declines sharply such that migrants who migrated over houses compared to 32% of non-migrants in urban 10 years ago are less likely to live in one-room houses areas and 50% of recent migrants compared to 52% than non-migrants. Although recent migrants live in of non-migrants in rural areas (Figure 7.9). With an smaller places they are more likely to have electricity increase in the duration of stay in the current residence, or light than non-migrants (Figure 7.10). This may FIGURE 7.10: Access to tap water and electricity among migrants Urban Rural 100% 100% 80% 80% 60% 60% 40% 40% 20% 20% 0% 0% Non-migrants ST Migrants MT Migrants LT Migrants Non-migrants ST Migrants MT Migrants LT Migrants Tap Water Electricity for Lights Source: 2007 census. Note: ST Migrants are migrants who migrated in the last 5 years, MT Migrants are migrants who migrated between 5 and 10 years ago, and LT Migrants refer to those who migrated more than 10 years ago. 102 ETHIOPIA – POVERTY ASSESSMENT FIGURE 7.11: Migration and poverty Those who migrated in the last 12 months Those who migrated 5+ years ago 0.4 0.4 Headcount poverty rate Headcount poverty rate 0.3 Addis Ababa 0.3 Addis Ababa 0.2 0.2 0.1 0.1 0.0 0.0 0.00 0.05 0.10 0.15 0.20 0.25 0.0 0.1 0.2 0.3 0.4 0.5 Migration within past year Migration over 5 years ago Fitted values of linear prediction do not include Addis Ababa Source: 2007 census. indicate that they are living in smaller households rather find large gains in consumption expenditure per than living in poorer living conditions upon arrival to capita, around 110 percent. The difference is not only their new place of residence. at the average, but also across the whole distribution At the city level there is little relationship (Figure 7.12) and remains after controlling for dif- between recent rates of migration and poverty, ferences in characteristics across migrants and non- but there is a strong negative relationship between migrants. Migrants eat more meat (41% compared to the proportion of medium/long term migrants 18%) and animal products (68% compared to 48%). and poverty. City level poverty rates were estimated The difference is larger for urban migrants. Migrants for the 95 largest cities in Ethiopia for this Poverty Assessment (see Chapter 8 for more details) and Figure 7.11 graphs city rates of migration against city poverty rates. Even though migrants are more FIGURE 7.12: Distribution of consumption for likely to move to cities with vibrant labor markets, migrants and non-migrants migration has a weakly negative association with the 0.6 proportion of the people in the city living in poverty. However, the better employment outcomes experi- 0.4 enced by longer-term migrants results in a strong Density negative relationship between the proportion of individuals who migrated to the city over five years 0.2 ago and the head count poverty rate. Large increases in consumption are observed 0.0 for migrants in comparison to individuals in their 1 3 5 7 villages of origin who did not migrate. de Brauw, Logarithm, Per Capita Consumption (1944 birr), 2009 Mueller, and Woldehanna (2013) use a number of Individuals who stay within community techniques to measure the impacts of migration on Individuals who move out of community the welfare of migrants versus non-migrants. They Source: de Brauw et al. (2013). MIGRATION AND POVERTY IN ETHIOPIA 103 to rural areas experience 68% of the consumption FIGURE 7.13: Subjective measures of gain of urban migrants. wellbeing However, there is no evidence that migrants 4.5 consider themselves better off subjectively than 4.0 non-migrants. Seventeen percent of migrants are 3.5 happy, compared to 16% of non-migrant household 3.0 heads. In fact, by a few measures, migrants may con- 2.5 sider themselves worse off (Figure 7.13). Controlling for individual and households characteristics in an 2.0 empirical model, however, renders any differences 1.5 insignificant. 1.0 In most The conditions I am I’ve gotten If I could live Asset ownership is lower among migrants as ways, my life of my life are satisfied the important my life over, a result of much lower rates of house ownership. is ideal excellent with life things I want I would out of life change almost Nearly all non-migrants (90%) own the house they nothing live in; while only 64% of migrants own the house that Household Heads Migrants they occupy (Figure 7.14). As time passes, migrants Source: de Brauw et al. (2013). start catching up with natives in terms of house owner- ship—house ownership is 56% among migrants who migrated less than five years ago and 70% among those There are large disparities between different who migrated more than 10 years ago—but never catch types of migrants, with female migrants and less up fully. Ownership of other assets such as a radio and educated migrants experiencing much lower wel- TVs are higher among migrants (54% own a radio and fare gains from migration. Female migrants expe- 16% own a TV) than non-migrants (35% own a radio rience about half (56%) of the consumption gain and 4% own a TV) and does not vary with length of experienced by male migrants. This is in part because migration. However, this difference is driven by differ- employment outcomes of female migrants are not as ences in migrants in rural areas, not migrants in urban good as employment outcomes of the average migrant. areas who have similar levels of asset ownership to households native to the city (Figure 7.15). Given that migrants in urban areas are less likely to own houses FIGURE 7.14: Housing ownership by duration this indicates lower levels of asset ownership among of migration migrants in urban areas compared to non-migrants. 100% This is not necessarily the case in rural areas. The fact that rural migrants are seemingly better off than their 80% neighbors in terms of TV and radio ownership could 60% indicate that they are indeed better off or it may reflect the fact that rural migrants may be in rural destina- 40% tions with better access to electricity and transmission. 20% Regression analysis that compares migrants and non- migrants within the same district shows that migrants 0% are indeed more likely to own a TV and radio (and Non-migrants Migrants ST Migrants MT Migrants LT Migrants more likely to rent and to have access to electricity and Source: 2007 census. water). This suggests that differences are not driven by Note: ST Migrants are migrants who migrated in the last 5 years, MT Migrants are migrants who migrated between 5 and 10 years ago, selection of specific rural destinations. and LT Migrants refer to those who migrated more than 10 years ago. 104 ETHIOPIA – POVERTY ASSESSMENT FIGURE 7.15: Ownership of radio and as households without a migrant, post-migration. television Table 7.1 shows that, if anything, income increases 100% after sending out a migrant between 2004 and 2009, but this increase is not robust to controlling for other 80% household characteristics. This finding is important, 60% given that one would have expected a negative effect of migration on agricultural productivity (which may 40% be offset by increased remittance income). It appears 20% that households that migrants leave are able to shift resources on the intensive margin in order to maintain 0% at least the same level of productivity. Alternatively, Ethiopia Urban Ethiopia Urban migrants may not have been productive agricultural Radio TV workers prior to leaving. Non-migrant Migrant Source: 2007 census. What constrains migration in 7.4  Ethiopia? Female migrants are 4% less likely to gain employ- Given the clear welfare benefits to internal labor ment and seven percentage points more likely to be migration and the limited negative effect on the an unpaid family worker than the average migrant. sending household, why are migration rates not Migration is beneficial for migrants, and higher in Ethiopia? The low migration rates for reduces their poverty, and the evidence also suggests employment suggest that constraints of some type little loss for sending households. Sending house- hinder migration. Depending upon whether returns holds may experience income gains from migration to migration are defined as per capita consumption if they receive remittances from the migrant, but they or consumption per adult equivalent, the returns to also lose the income or contribution to agricultural migration appear to be 83–113% (de Brauw, Mueller production that the migrating member could have and Woldehanna, 2013). earned by staying at home. The ERHS is used to In general, migration can be limited by policy, assess how the agricultural productivity of a household credit and information constraints. Policy barriers changes after migration. Migration was found to have that limit migration can be both explicit (for example, no negative impact on agricultural productivity in that in China the hukou system explicitly limited move- households sending migrants were just as productive ment from rural to urban areas, e.g. Fan 2008), and TABLE 7.1: Migration and agricultural productivity Migrant Households Non-migrant Households Average value of all production, 2004/5 (Birr) 1705 1607 (2714) (2055) Average value of all production, 2009 (Birr) 2589 2138 (4456) (2964) Source: ERHS 2004/5 and 2009. Notes: Standard deviations are in parentheses. The value at the 99th percentile has been set as the maximum value and all values above that have been set to that value to minimize the influence of outliers. All results are reported in 2004 birr, and number of observations are reported for 2004/5 (seven additional observations dropped from the 2009 sample. MIGRATION AND POVERTY IN ETHIOPIA 105 implicit (for example, governments may implement serious obstacles to migration, urbanization, and struc- policies that foster agricultural production to the detri- tural transformation in the medium to longer term. ment of worker movement). Alternatively, constraints In Ethiopia, de Brauw and Mueller (2012) show that affecting households may limit migration. Two poten- land tenure appears to be a constraint to migrating, tially important constraints relate to information and although the magnitude of such effect was found to capital. Potential migrants may lack information about be quite small and other constraints to migration are the types of employment available in urban areas, also likely important. While it may not be the case that particularly if migrant networks do not reach them. changing policies will result in higher rates of poverty Uncertainty about potential returns to labor could reduction it is important to note that these policies lead to perceptions of migration as too risky. In sub- may be limiting structural change and the develop- Saharan Africa, such uncertainty may be exacerbated, ment and poverty reduction that it could bring. as most urban opportunities are in the informal sector The evidence is consistent with credit con- (Fox and Gaal, 2008). Capital, in the form of credit straints also limiting migration at the household or liquidity constraints, may be a further constraint. level. The costs of migration can be large, comprising Migration implies the movement from one place to not just the costs of travel, but also the costs of sup- another, which implies both costs of transportation porting the migrant in the destination location until and a place to live when away from the source house- they are able to access employment. As described in hold. Without a source of capital for these start-up Chapter 8 the majority of young migrants to Addis costs and in the absence of capital, potential migrants Ababa report being supported by their families as might not be able to move. they search for work (Franklin 2014). If households Policies focused on equitable land distribu- in general face credit constraints against investing in tion and rural development have aided broad- migration, households must have enough income or based growth in rural Ethiopia but have hindered savings to support the initial migration. Agriculture migration and the structural change and develop- is the primary source of rural household income. ment it brings. Ethiopia has pursued a rural-focused As a result if access to credit constrains migration, development strategy since the 1990s, encouraging migration will be more likely to occur from more productivity improvements in agriculture through households that are wealthier and more agriculturally investments in extension and modernization and pro- productive. Alternatively, if less productive house- viding effective safety nets in food insecure areas (see holds were sending out migrants, one would infer Chapter 8 for more details on differences in transfers that credit constraints are not an issue. Analysis of received between rural and urban Ethiopia). This the ERHS data shows that wealthier households and policy focus preferentially favors rural areas, and may households that are more agriculturally productive are be acting as a check on migration trends in Ethiopia. more likely to have migrants suggesting that credit Recent land registration and certification programs constraints are important. A 10 % increase in agricul- have improved land user rights and land security tural income increases the probability of migration by for the farmers, but land transfers outside of family 0.45 percent. This analysis also finds that households members are prohibited. This has allowed Ethiopia to that report being able to access funds in time of need maintain a very equitable land distribution in rural are also more likely to send migrants, which would areas, but it also means that a household that would also be consistent with credit constraints negatively benefit from selling their land and migrating to an affecting migration. urban area is not able to do so. The experience of Information constraints may also be important, other countries—particularly China with similar land but very little is known about the role of informa- sales restrictions—shows that land restrictions become tion constraints on migration patterns in Ethiopia. 106 ETHIOPIA – POVERTY ASSESSMENT 7.5 Conclusion There is a fundamental trade-off facing policy- makers in Ethiopia today: current policies focused Migration in Ethiopia is increasing but rates of on equitable land distribution and rural develop- rural-urban migration remain low in light of the ment may continue to aid broad-based growth in welfare improvements experienced by migrants. rural Ethiopia but they will also limit migration and When migration does occur it is more educated the structural change and development it brings. individuals that migrate from rural households Experience from China shows that policies that restrict that are more agriculturally productive than their migration will only become more binding when eco- neighbors. Migrants experience improvements in nomic growth and employment transformation (from welfare and sending households experience little agriculture to non-agriculture) accelerates (Au and loss in production, suggesting that migration can Henderson 2006a and 2006b; Deininger, Jin, and Xia reduce poverty and encourage development in 2012) Similar findings have been documented in the Ethiopia. Continued improvements in agricultural case of Sri Lanka (Emran and Shilpi forthcoming). productivity are likely to spur migration, but the The removal of these restrictions could help stimulate evidence presented in this chapter suggests that “pull” migration, which in turn will facilitate “good” addressing liquidity constraints of households may urbanization and help to reap benefits of agglomera- also be needed. tion economies in the medium to long term. 107 UNDERSTANDING URBAN POVERTY 8 Although rates of urbanization in Ethiopia are Poverty Assessment. A simple scatterplot confirms quite low compared to other countries (Schmidt that the negative relationship between poverty and and Kedir 2009) urbanization is taking place, city size that underpins the oft-stated metropolitan and as Ethiopia urbanizes, poverty becomes more bias also holds true for Ethiopia cities excluding urban. In 2000, 11% of Ethiopia’s poor lived in cities, Addis Ababa (Figure 8.1, left panel). Very small urban but this rose to 14% in 2011. As a result the number centers—rural towns—are poorer than larger urban of urban poor stayed almost constant between 2005 centers as shown in Table 8.1.26 The depth and sever- and 2011 at 3.2 million even though urban poverty ity of poverty tend to fall with city size, but inequal- rates fell by almost ten percentage points (from 35% ity is marginally higher in larger cities (Figure 8.1, to 26%). right panel). In Ethiopia, just as in other countries, pov- Poverty rates in Addis Ababa and Dire Dawa, erty rates fall and inequality increases as city size however, are much higher than this trend would increases. A number of cross-country studies have predict. Addis Ababa is a city on a different scale to shown that smaller towns have deeper, more wide- other cities in Ethiopia. The intercensal population spread poverty and higher infant mortality rates (Ferré survey estimated the population of Addis Ababa city et al. 2010, Brockerhoff and Brennan 1998). The first small areas estimates of poverty for the 95 largest cit- 26 The findings of Table 8.1 are robust to choosing different population ies and towns in Ethiopia were constructed for this cut-offs to define the two groups and when Addis Ababa is excluded. FIGURE 8.1: City size, poverty and inequality in Ethiopia City Size and Head Count Poverty (FTGO) City Size and Inequality 0.4 0.4 0.3 Gini coefficient Addis Ababa FGTO 0.2 0.35 0.1 Addis Ababa 0.0 0.3 10 11 12 13 14 15 10 11 12 13 14 15 In(city size) In(city size) Fitted values of quadratic prediction do not include Addis Ababa Fitted values of linear prediction do not include Addis Ababa Tigray Amhara Oromia Somali Benishangui Gumuz SNNP Gambella Harari Dire Dawa Addis Ababa Fitted values Source: Own calculations using 2007 census and city poverty rates estimated using the 2007 census and HCES 2011. 108 ETHIOPIA – POVERTY ASSESSMENT TABLE 8.1: Mean poverty measures and t-test results, by city size category Urban Urban Urban Urban centers centers centers centers All 95 smaller than larger than t-test of smaller than larger than t-test of cities 95,000 95,000 difference 30,000 30,000 difference Headcount poverty rate 0.24 0.25 0.21 0.041** 0.26 0.22 0.037** No. of urban centers 95 82 13 53 42 Source: Own calculations using city poverty rates estimated using the 2007 census and HCES 2011. Note: The t-test columns report the p-value from testing the equality of (FGT0) means between the two groups of cities. *** p<0.01, ** p<0.05, * p<0.1. administration as almost three million people, com- important to ensuring welfare improvements continue prising 21% of the urban population of Ethiopia in both rural and urban Ethiopia. (CSA 2012). The second largest city, Dire Dawa, has This chapter seeks to inform the design of poli- a population of about a quarter million. Although cies to address urban poverty by characterizing the Addis Ababa is much larger than other cities or town nature of urban poverty and presenting results of in Ethiopia, the headcount poverty rate in Addis simulations of possible policy interventions. The Ababa is quite high, and higher than it should be if nature of work and poverty in urban Ethiopia is dis- the relationship between poverty and city size found cussed in Section 8.1 and a framework for thinking in other Ethiopian cities was extrapolated. Dire Dawa about policies to reduce poverty through work is pre- also records much higher poverty rates than its size sented in Section 8.2. Unemployment is a feature of would predict. poverty in the urban landscape—particularly in large As more of the poor live in large urban centers, cities—and Box 8.1 provides a special focus on youth expanding development programming to address unemployment and job search in Addis Ababa. Section key challenges to urban poverty reduction is imper- 8.3 examines poverty rates among those unable to work ative. Until now, Ethiopia’s development strategy has and points to weaker informal and formal safety nets been rural-focused. This strategy has been successful in among many urban poor. Section 8.4 thus examines ensuring agricultural growth and rural safety nets have what an urban safety net policy would consist of, a safety made significant contributions to reducing extreme net that improves the welfare of those unable to work poverty (see Chapters 4 and 5). However, for Ethiopia and that increases the productivity of those who are able to eliminate extreme poverty in the future this strategy and willing to participate in the urban labor market. needs to be complemented with specific programming designed to address urban poverty. 8.1  Work and urban poverty Interventions targeted at addressing urban poverty are also important to mitigate unintended This section considers the nature of work in urban impacts of high food prices, which are beneficial to centers in Ethiopia, particularly in large urban rural poverty alleviation. In addition, development centers, and examines the relationship between policy that favors rural areas and rural poverty reduc- work and poverty in these cities. It highlights that tion—namely high food prices—can have negative poverty is particularly a concern for the unemployed welfare consequences on urban areas (Chapter 4). A and those who are engaged in marginally productive policy framework that allows these beneficial effects self-employment activities out of necessity. on rural poverty reduction to be in place, while nega- The nature of work is much different in Addis tive impacts on urban households are mitigated is Ababa and other big towns than in smaller urban UNDERSTANDING URBAN POVERTY 109 FIGURE 8.2: City size and the nature of jobs A. City size and wage-employment rates B. Job types among youth by city size 70 0.3 Proportion of active labor force 60 Proportion of labor force Addis Ababa 50 that is employed 0.2 aged 15–29 40 30 0.1 20 10 0.0 0 10 11 12 13 14 15 Government or NGO Private (temporary) Private (permanent) Paid domestic work Self- employed Unpaid work for family Agriculture In(city size) Fitted values of linear prediction do not include Addis Ababa Addis Ababa Other big towns Small towns Source: 2007 Census and CSA Urban Employment and Unemployment Survey, 2012. centers: self-employment and work in family enter- also has a higher prevalence of permanent jobs than prises decreases, and waged employment increases, elsewhere (Figure 8.2, panel B). Median wages in with city size. Panel A of Figure 8.2 shows a strong Addis Ababa are also, on average, higher for all educa- positive relationship between city size and the share of tion levels (Figure 8.4) and the premium tends to be the population in wage employment. Panel B details higher at higher levels of education. In Addis Ababa how, for youth, the prevalence of self-employment, individuals are usually more satisfied with the work family work and agricultural activities falls with city they have, possibly because they are more likely to be size and private sector waged employment increases permanent, paid more, and have work for more hours. with city size. In small towns up to 83% of working In urban centers where waged employment— youths are self-employed or in unpaid family work both private and public—is higher, poverty rates but this share falls to 24% in Addis Ababa. Unemployment rates are also higher in the largest urban centers in Ethiopia (Figure 8.3). FIGURE 8.3: City size and unemployment Unemployment in Ethiopia is an urban phenom- 0.30 enon. Fewer than 5% of all rural households have Dire Dawa Proportion of labor force that was unemployed for last 7 days an unemployed adult member (Table 8.2). However, 0.25 in urban Ethiopia 14.8% of all households report an Addis Ababa 0.20 adult member—male or female—as unemployed. In particular, there are high unemployment rates in Addis 0.15 Ababa. More than one in four households in Addis 0.10 Ababa report an unemployed adult (28.7%) com- pared to one in 10 households in other urban areas 0.05 (10.8%), Table 8.2. 10 11 12 13 14 15 The higher prevalence of good jobs in large In(city size) cities encourages more people to search for them. Fitted values of quadratic prediction do not include Addis Ababa Addis Ababa has high rates of unemployment, but it Source: 2007 census. 110 ETHIOPIA – POVERTY ASSESSMENT TABLE 8.2: National, urban and rural unemployment rates, various definitions Population living in a household with an unemployed adult Individuals unemployed Unemployed in the Predominantly unemployed over Unemployed in the last 7 days (WMS) the last 12 months (HCES) last 7 days (UEUS) Male Female Male Female Male Female Adult adult adult Adult adult adult Adult adult adult National 6.4% 3.8% 3.4% 1.7% 0.9% 1.0% Rural 4.8% 3.0% 2.5% 0.4% 0.2% 0.2% Urban 14.8% 8.4% 8.1% 8.7% 4.4% 5.2% 18.4% 11.5% 25.9% Urban, not Addis 10.8% 6.6% 5.0% 5.6% 2.9% 3.1% 15.4% 10.3% 23.9% Addis Ababa 28.7% 14.8% 18.7% 19.9% 9.7% 12.8% 21.7% 15.5% 32.0% Sources: Own calculations using the Welfare Monitoring Survey 2011 (WMS), Household Consumption Expenditure Survey 2011 (HCES) and Franklin (2014) using the Urban Employment and Unemployment Survey 2012 (UEUS). Note: the definition of available for work is that used in Franklin (2014) which is quite broad and contributes to high female unemployment rates. are lower. Poverty rates are lower in cities in Ethiopia poverty rates, but manufacturing output per se is not, that have a higher share of the labor force in wage suggesting that it is employment, not output per se, employment rather than self-employment (see that matters for poverty. Employment in both the Figure 8.5). This relationship is part of what under- public and private sector is significantly negatively pins the relationship between city size and poverty correlated with poverty, although employment in in Ethiopia: once the share of labor force in waged the public sector is slightly more strongly negatively employment is considered, there is no longer a sig- correlated. This is consistent with evidence from nificant association between city size and poverty in Chapter 4 that indicates that manufacturing growth Ethiopia (Table 8.3). Higher rates of both private and and the waged employment it brings helped reduce public sector employment are associated with lower FIGURE 8.5: Towns and cities with higher FIGURE 8.4: Median wages of employees rates of employment are less poor in Addis Ababa, other big towns and small towns 0.4 3000 Headcount poverty rate 0.3 Addis Ababa 2500 Wage (Birr) 2000 0.2 1500 1000 0.1 500 0 0.0 No Education Primary Secondary Incomplete Secondary Higher Incomplete Diploma Degree 0.0 0.2 0.4 0.6 Proportion of labor force that is employed Fitted values of linear prediction do not include Addis Ababa Small town Addis Ababa Other big towns Source: Own calculations using 2007 census and city poverty rates Source: CSA Urban Employment and Unemployment Survey, 2012. estimated using the 2007 census and HCES 2011. UNDERSTANDING URBAN POVERTY 111 urban poverty from 2000 to 2011. In addition, it poverty gap is 13.5% compared to 7.6% on average highlights the potential relationship between public and poverty severity is 5.2% compared to an average employment and poverty reduction. However, both of 2.8%. Although 28.7% of all households in Addis Figure 8.5 and Table 8.3 show that there is substan- Ababa report an adult member in unemployment, tial variance in poverty rates for cities with the same this increases to 40.1% and 41.9% of households level of employment. This indicates the importance of below the poverty line and in the bottom 10% of the other factors such as the quality of work, profitability income distribution. In rural areas unemployment is of self-employment, and provision of basic services. much less common and when it is reported it is less In many developing economies being unem- strongly correlated with being poor. ployed and searching for waged employment is not Young people are particularly affected by high strongly correlated with poverty, as only better- unemployment and young women are more likely off middle class families can afford this type of to be unemployed than men, even though they are search. However in Addis Ababa unemployment, less likely to be engaged in the labor market due particularly male unemployment, is strongly cor- to family responsibilities. Rates of unemployment related with poverty: nearly half of all households are as high as 21% among urban men actively seeking with an unemployed male in Addis Ababa live in work between the age of 15 and 24. This falls to 6% poverty (Table 8.4). The probability of a household for men between the age of 31 and 50 (Figure 8.6). with a male unemployed member living in poverty is Unemployment among young women is 22% com- 48.0% compared with an average poverty rate in Addis pared to 14.5% among young men. The main reason Ababa of 29.0%. The poverty gap and poverty sever- that women give for not engaging in the labor market ity is also particularly high for these households. The is responsibility of home activity (34.5%). It is quite TABLE 8.3: The relationship between poverty, city size and employment Dependent variable is the proportion of households living below the poverty line (1) (2) (3) (4) (5) Log of city size –0.02164* –0.00924 –0.01711 0.00917 0.01070 [0.01240] [0.01247] [0.01355] [0.01599] [0.01506] Proportion of labor force employed –0.24114*** –0.28381*** [0.06825] [0.07397] Log of manufacturing output per capita –0.00186 [0.00308] Proportion of labor force employed in private sector –0.24180* [0.14514] Proportion of labor force employed in public sector –0.31003*** [0.10646] Proportion of labor force unemployed for last 7 days –0.46846** –0.46918*** [0.17798] [0.17598] Constant 0.46767*** 0.41670*** 0.42807*** 0.31042** 0.29423** [0.12966] [0.12502] [0.13705] [0.15107] [0.14009] Observations 94 94 94 94 94 R-squared 0.02826 0.11599 0.03182 0.18551 0.18453 Source: Regression results using data from the 2007 census and city poverty rates estimated using the 2007 census and HCES 2011. Notes: Robust standard errors in brackets*** p<0.01, ** p<0.05, * p<0.1. 112 ETHIOPIA – POVERTY ASSESSMENT FIGURE 8.6: Characteristics of the unemployed Unemployed (15–29 year olds) Unemployment (%) 30% 35 25% 30 25 20% 20 15% 15 10% 10 5% 5 0% 0 15–24 25–30 31–40 41–50 51–65 None Primary Secondary incomplete Secondary Some Higer Diploma Degree Age Category Education Level Female Male Source: Urban Employment and Unemployment Survey 2012. possible that if women were to face good job opportu- Those with the lowest levels of education are nities these women would also enter the labor market. more often engaged in informal self-employment, Pregnancy and delivery is the second most common out of necessity, rather than being unemployed reason for not seeking a job (19.9%). looking for a wage job. Although the unemployed Unemployment exhibits an inverted-U shape are poor, many are often well-educated. At low levels in education, with employment rates being highest of education the unemployed are extremely poor and among those with only secondary education, with less successful at gaining wage employment, making rates at their lowest among the highest and lowest self-employment a better alternative for many with education levels (Figure 8.6). There is a considerable low levels of education. In Addis Ababa the poverty need for creation of jobs for individuals who have just rate among the unemployed who have not completed completed high school. In Addis Ababa, unemploy- primary education is 44% compared to 28% among ment rates are as high as 36% among male youth who the self-employed and 32% among the unemployed have just graduated high school. Box 8.1 details the who have completed primary education. As a result, problem of youth unemployment in Addis Ababa. those with no education or primary education are TABLE 8.4: Poverty and unemployment in Addis Ababa Percentage of … bottom Total Addis Poor in 10% in Poverty population Addis Addis Poverty rate Poverty gap severity Household has Unemployed member 28.7% 40.1% 41.9% 40.5% 10.7% 4.0% Unemployed member, male 14.8% 24.4% 28.1% 48.0% 13.5% 5.2% Unemployed member, female 18.7% 24.0% 22.2% 37.3% 9.5% 3.5% Source: HCES 2011. UNDERSTANDING URBAN POVERTY 113 much more likely to become self-employed than those FIGURE 8.7: Unemployment, with higher levels of education. Figure 8.7 shows that self-employment and education in Addis for Addis Ababa, self-employment is more prevalent Ababa (12 month definition) until about seven years of education after which point unemployment becomes more prevalent. 0.10 Job status is changing on a weekly basis for 0.08 many in urban Ethiopia and as a result many with Density 0.60 some work are in need of more work and more permanent work. Many working in wage employ- 0.40 ment are on jobs that do not last long, just as many 0.20 unemployed individuals undertake temporary work to 0.00 get by while they look for a permanent job (Box 8.1). 0 5 10 15 20 Years of education Reducing poverty in urban centers 8.2  Unemployed Self-employed through work: a framework Source: CSA Household Consumption Expenditure Survey 2011. kernel = epanechnikov, bandwidth = 2.0000 On the basis of the description of work and poverty in Section 8.2, three types of workers can be char- acterized in large urban centers: the necessity self- entrepreneurs that employ others comprise 2%. employed, those in wage work or searching for wage A macro-labor model has been developed with these work, and opportunity entrepreneurs. Labor market three types and has been parameterized to predict imperfections cause high rates of unemployment to be rates of self-employment, wage-employment and observed. Those with little education choose neces- unemployment that are found in Addis Ababa. This sity self-employment rather than searching for wage parameterization also captures the fact that large employment because the high probability of being firms are few but account for a relatively large share unemployed makes job search costly and the wages of employment. earned do not compensate them for the cost of look- Within this type of labor market the avail- ing. The productivity and income of these individuals ability and quality of work for poor households is lower than it would be if they were employed, but can be improved by encouraging the necessity the cost of being unemployed and searching is not self-employed to upgrade to employment, reduc- worth the gain. Those with moderate levels of educa- ing unemployment rates, increasing wages for tion look for and gain employment. They may be in those with lower levels of ability or helping the unemployment for some time before they secure a job, self-employed become more profitable. The exact but the returns to being employed are worth the search. nature of policy interventions requires a clearer under- Those of the highest ability are also self-employed standing of what drives the labor market to have the entrepreneurs, but operate businesses that are of a scale characteristics it has in Addis Ababa. In particular, such that they employ others. Those entrepreneurs are the sources of labor market inefficiency that causes termed opportunity entrepreneurs. This model is set high positive rates of unemployment to be sustained. out in Poschke (2014) and is summarized in Figure 8.9. Costly search processes are detailed in Box 8.1, but In Addis Ababa those in self-employment com- this may not be the only cause of high unemployment prise 21% of the work force, those in employment rates. Other reasons could be some type of stickiness or searching for employment comprise 77% of the in wages, or queuing for specific jobs that carry a labor force (of which 29% are unemployed) and lifetime earnings premium (or job security), such as 114 ETHIOPIA – POVERTY ASSESSMENT BOX 8.1: Youth unemployment and job search in Addis Ababa High levels of youth unemployment have long been a feature of the Ethiopian economy and their persistence in the face of economic growth and increased educational attainment is a cause of concern. In 2012 unemployment rates among males age 15–24 in Addis Ababa are 21%, and 36% among those who have just graduated from high school. The striking finding for Ethiopia is just how poor the unemployed are as they search for work. What are their aspirations for employment opportunities? What do they do and how do they survive while they are without work? How do the youth find jobs? This box draws on a background paper prepared for the Poverty Assessment that uses two datasets, the 2012 Ethiopian Central Statistics Agency’s Urban Employment and Unemployment Survey (a large survey taken to be representative of all urban areas in Ethiopia) and a panel dataset collected by Oxford University studying the lives of a small sample of unemployed youth from areas in and around Addis Ababa conducted in 2013 (Franklin 2014). Survey data on unemployed youth in Addis Ababa identify two different types of unemployed: lower educated youth who are native to Addis Ababa and higher educated recent migrants. One type of unemployed youth is native to Addis and living with their parents. They have some secondary education (60%) but no tertiary education and are often not actively looking for jobs as they have become discouraged by a long period of unemployment. They are likely to have done some temporary work in the past and have a relatively low reservation wage. In contrast the second type of unemployed have just finished school or university and moved to Addis either for education or upon graduating. They live on their own or with relatives and they are actively engaged in formal job search. They have not been unemployed for as long, they are less likely to have work experience, and they are looking for higher paid jobs. In the Oxford survey, the first type of unemployed youth was primarily sampled in slum areas in non-central sub-cities of Addis Ababa and the second type was primarily sampled around the job vacancy boards, which are described further below. Descriptive statistics of these two samples of unemployed youth are presented in Table 8.5. TABLE 8.5: Two types of unemployed Type 1: lower educated, native Type 2: higher educated, recent Addis Ababa unemployed migrants to Addis Ababa (sampled (sampled in slum areas) around vacancy boards) Age (years) 23.3 23.7 Female (%) 32.7 13.1 Has a degree (%) 1.0 43.8 Finished Grade 10 (%) 59.9 94.7 Moved to Addis last year (%) 62.4 18.6 Lives in parents’ house (%) 55.9 20.1 Discouraged (%) 25.2 1.9 Has work experience (%) 63.4 39.7 Reservation wage (Birr) 1135 1376 Source: Oxford Survey of Unemployed Youth from 2013. Employment aspirations Unemployed youth aspire to escape poverty through a permanent job often in an administrative position, however finding permanent employment is difficult, particularly for those without higher levels of education. More than 50% of unemployed youth in Addis Ababa were looking only for permanent jobs. In comparison 35% said they were looking for any type of work. Although many unemployed seek permanent employment, few have permanent jobs. On average, 27.7% of males have permanent jobs and 17.3% of females. However the proportion is much lower among 15–24 year olds with 13.1% of males in this age group that are in the labor market having a permanent job and 12.8% of females (Table 8.6). Those with higher levels of education are more likely to have permanent jobs. (continued on next page) UNDERSTANDING URBAN POVERTY 115 BOX 8.1: Youth unemployment and job search in Addis Ababa (continued) TABLE 8.6: Permanent employment in Ethiopia: Proportion of labor force with a permanent job in different age groups (%) Age group Male Female 15–24 13.1 12.8 25–30 28.9 21.0 31–40 31.0 19.8 41–50 40.8 20.2 50–65 33.8 10.6 Source: CSA Urban Employment and Unemployment Survey 2012. Few youth actively seek self-employment, but some end up in self-employment when job search does not provide employment. Almost no unemployed youth said they were planning on making a living in self-employed activities (just 2%). However, four times this number end up in self-employment just four months later. The UEUS finds that self-employment is more prevalent among old cohorts. Increasingly, unemployed youth are looking for and attaining private sector jobs. Serneels (2007) describes the Ethiopian labor market in the mid-1990s as characterized by the phenomenon of queuing, where reasonably well educated wait for good jobs usually in the public sector. The presence of relatively well-educated aspiring to high-paying government or administrative-type jobs is still very much in evidence in 2012/3, but jobs in the private sector are increasingly desirable. In 1994 there were twice as many jobs in government as there were wage-paying jobs in the private sector (Serneels, 2007). In 2012, the private sector provided more jobs to youth (45.1%) than did the government (33.2%). However, government jobs are still more likely to be permanent jobs (82.1%) than private sector jobs (23.7%). Public sector jobs are still highly sought after, with 34% of unemployed youth in Addis Ababa stating a preference to work in the public sector compared to 65% in 1993. However the private sector is increasingly desirable: 55% of unemployed youth in Addis Ababa stated a preference to work in the private sector, compared to 16% in 1993. Unemployed life The unemployed youth of Addis Ababa rely heavily on money from their parents, particularly those who have just graduated or moved to Addis Ababa. Half of those without work reported that money from immediate family (excluding friends, spouse, and partner) was their main source of income. The reliance on family was higher for those who had recently migrated to Addis, or those who had just graduated from school or university. Those with degrees, and recent migrants, are getting three times the financial support from their parents than someone who was born in Addis, or had no education. Families provide savings to youth on graduating or on moving to the capital in order to support themselves while they look for work. Unemployment spells, however last longer than the number of weeks of search that family support can sustain and many youth engage in short spells of temporary employment to sustain job search, very often manual labor in the construction sector. Recent migrants are far less likely to have savings, formal or informal than those who are unemployed and native to Addis. On average, they have only enough money to survive a few weeks on their own savings, at their regular rates of expenditure. The unemployed are no longer just wealthy elite that can wait for many months for a permanent job (as found in the mid-1990s by Serneels 2007). Many unemployed engage in temporary employment to earn money to continue searching for work. More than half of those who were unemployed and remained after four months, had engaged in temporary work during this time. Half of casual/daily jobs for men are in construction. Only one fifth (22%) of unemployed youth did not work at all during four months. Unemployed individuals of all education levels are just as likely to take temporary work, but recent migrants and those not living with their parents are much more likely to take temporary work. Well-educated individuals were no less likely to have taken work over the 16 weeks during which they were tracked. As a result many well educated are engaged in temporary jobs (such as those in the construction sector) for which they are over qualified. Recent migrants seem the least able to avoid taking work while searching, they are about 10 percentage points less likely on average to have done no work over four months, relative to 35% of those who have been in Addis for longer than a year. Individuals that did not take any work were more likely to be relying on family money and more likely to be living at home. (continued on next page) 116 ETHIOPIA – POVERTY ASSESSMENT BOX 8.1: Youth unemployment and job search in Addis Ababa (continued) Unemployment contains considerable boredom on a daily basis for unemployed youth. The unemployed spend on average two thirds (16 hours) of their time in their own homes or yards. Of those waking hours, remarkably, respondents report spending at least three hours per day on average “doing nothing,” even after having been asked about 20 different activity categories, and asking about any other time spent that had not been accounted for. This time spent doing nothing does not include all other leisure activities reported (on average 3.4 hours a day), nor does it include time spent socializing with friends (1.7 hours per day on average). This sort of time use behavior fits with the picture of time spent among young unemployed men in the anthropological work of Mains (2012), who discussing the considerable boredom of waiting for their lives to progress, and having little to do in the meantime. Job search Visiting vacancy boards is the most common form of job-search method and the one that yields the highest probability of finding a permanent job. Job seekers usually try a range of different routes into work, including asking their social networks and going door to door to ask businesses for vacancies. However, vacancy boards and newspapers, particularly vacancy boards, are the most common forms of job search. They are used by 44% of the urban unemployed, compared to 22% who ask friends or relatives for a job (UEUS 2012). They are particularly used for finding permanent jobs: although 38% of jobs found by unemployed youth in Addis Ababa were found at job boards, 69% of permanent jobs were found at job boards (compared to 63% and 31% found through networks respectively). Searching at vacancy boards can be time consuming and expensive involving many visits to the central vacancy boards, each of which costs more than the median daily expenditure of unemployed youth. Those with lower levels of education that do not visit vacancy boards state it is because they will not find work there (60%) reflecting the fact it is more often skilled jobs that are posted on the boards. The majority of those with higher levels of education that do visit more but the costs of transport were prohibitively high (82%). Of the sample of unemployed youth sampled at vacancy boards, 83% had stopped visiting these boards after four months because it was too expensive to travel to the board. The average cost of a trip to the town center to look for work is estimated to be 15 Birr, which is higher than the average median expenditure of 14 Birr per day among the two samples of unemployed youth. One in four educated, active job seekers secured permanent employment in four months of search, but rates of success are much lower for those who are less educated and less actively looking for work. After 16 weeks, 21% of the type 2 unemployed had found permanent jobs compared to only 6% of the type 1 unemployed (Figure 8.8). This means that type 2 unemployed will stay in unemployment and poverty for a longer period of time. Among those well-educated, actively seeking work, one third (32.8%) had been unemployed for 6–12 months and almost one fifth (18.9%) for longer than this. Among those native to Addis, 35.3% had been unemployed for longer than one year. Rates of discouragement are much higher among the type 1 unemployed. FIGURE 8.8: Rate of finding employment among unemployed youth in Addis Ababa Type 2: Those sampled at job boards Type 1: Those sampled in slum areas 1.0 1.0 0.8 0.8 0.6 0.6 0.4 0.4 0.2 0.2 0.0 0.0 0 1 2 3 4 5 6 7 8 9 10 11 16 0 1 2 3 4 5 6 7 8 9 10 11 16 Discouraged Searching Temp Job (no search) Temp Job (search) Perm Job Source: Oxford Survey of Unemployed Youth from 2013. Franklin 2014. UNDERSTANDING URBAN POVERTY 117 FIGURE 8.9: Labor markets in large cities: three types Type Low levels of education Secondary and tertiary education High levels of entrepreneurial ability Strategy These individuals determine they are better These individuals realize that they could These individuals realize they can earn not searching for work as it is costly and earn more in the long run if they were higher returns if they choose to be the wages would not compensate the cost. employed rather than self-employed, even self-employed rather than employed. They They choose self-employment. though they know this will require spells of are successful entrepreneurs and employ unemployment while they look for work. others. Outcome Petty trading, shoe-shining (“necessity Employed in blue or white collar jobs OR Owners of medium-sized enterprises self-employ-ment”) unemployed looking for work (“opportunity entrepreneurs”) Individual ability is shown on the x axis, and on the y axis the value of Entrepreneurship different activities. The blue line indicates the value of employment for individuals of differing levels of ability. The red line indicates the value Value Employment of entrepreneurship and the black line indicates the value of searching for employment (i.e. unemploy-ment) for different levels of Unemployment ability. The current value of unemployment is zero at any point in time but it secures employment in the future, so the total value is positive Ability and increasing with ability. Source: Adapted from Poshcke (2014). some public sector jobs. Although the exact reason is in scale and easing these costs can improve the wel- unknown, this framework allows a discussion of the fare of the very poorest by increasing demand for likely qualitative impacts of different types of interven- labor which increases wages and encourages some tions, by allowing for unemployment in a market in in necessity self-employment to engage in more which workers are choosing between self-employment profitable wage employment. The World Bank’s and wage employment. In particular simulations high- Doing Business project contains detailed measures of light how supporting large-scale entrepreneurs can be the compliance cost for a “typical” firm on entry. In very beneficial for poverty reduction, perhaps more so Ethiopia, this cost is about the size of GDP per capita, than supporting necessity entrepreneurship. that is, a starting entrepreneur could employ one and Supporting entrepreneurs, both large and a half employees for a year at the same cost. Given the small, can be poverty reducing. Supporting small- very small average size of firms in Ethiopia, this cost scale entrepreneurs can reduce poverty by increasing is high, not only in global comparison (the OECD the productivity of those who currently earn marginal average is 3.6% of GDP), but also in the African profits from self-employment. However, supporting context. Compared to other measures of the business entrepreneurs that have large firms can also be poverty environment also collected by the Doing Business reducing—if not more so. High productivity entre- project, entry regulation is the aspect of the business preneurs earn substantial profits, but also employ environment where the burden on Ethiopian firms is many workers, and contribute to higher overall wage largest relative to other countries. Model simulations levels through their demand for labor. As the value of suggest that a subsidy of six times the entry cost results employment increases so does the value of job search. in a benefit that is slightly below one times the aver- This encourages necessity entrepreneurs to search for age period profit for opportunity entrepreneurs. This and gain employment. allows these firms to hire more workers and as indicated An important concern for potential large-scale by Table 8.7 would increase the proportion of workers entrepreneurs is the cost of entering or increasing in employment and increase wages. Unemployment 118 ETHIOPIA – POVERTY ASSESSMENT TABLE 8.7: The simulated impact of introducing policies to address urban poverty Proportion of workers Number of employed by Direct effect on Unemployment necessity opportunity Average Wage poverty rate rate entrepreneurs* entrepreneurs wage inequality Reducing hiring costs 0 – – + + – Reducing entry costs for 0 +/– – + + + large firms Safety net for all poor 36% reduction + – + – + *Whenever the number of necessity self-employed falls, the income of those in necessity self-employment that switch to job search and wage employment increases. **Measured as a ratio of the wages of the 90th percentile to the wages of the 10th percentile. rates may fall and some in necessity self-employment Urban poverty among those 8.3  would move into more profitable wage employment. unable to work A policy that reduces hiring costs can raise wages, reduce unemployment and encourage a num- Addressing urban poverty also requires improving ber in necessity self-employment to upgrade to new the wellbeing of those not in the labor market. On jobs. One possible source of labor market inefficiency some dimensions, poor households in urban areas is the cost of hiring new employees. If hiring were not have similar characteristics to those in rural areas. costly, firms would post so many vacancies that work- They are less likely to be educated and household ers would find jobs instantaneously. If matching were size is larger (Figure 8.10). However, households infinitely efficient, all profitable matches would be with members who cannot engage in labor markets made instantaneously. In practice, of course, neither is are more likely to be poorer in urban Ethiopia than the case, and matching efficiency and hiring costs are in rural Ethiopia. determined by technology and regulation. Consider a Households with elderly members, widows, policy that reduces hiring costs, for example by cover- and with elderly or female heads are much more ing the costs of on-the-job training for new recruits. likely to be poor if they are located in urban areas This change induces firms to post more vacancies and compared to rural areas (Figure 8.11). In urban reduces the unemployment rate. It also reduces neces- areas households with female widows have poverty sity self-employment as job search becomes less costly rates 10 percentage points above the urban poverty and more choose to upgrade to wage employment rate while in rural areas households with female where they are more productive. Lower hiring costs widows are no more likely to be poor than other also imply higher aggregate output, higher profits for households. In fact in rural areas households with the opportunity entrepreneurs that hire workers, and, an elderly household member or an elderly head are as a consequence, higher wages for workers (Table 8.7). less likely to be poor than other rural households. Reducing the costs of job search (perhaps by encourag- In urban areas households with an elderly member ing the use of technology to search for jobs rather than or an elderly head are 12 and 13 percentage points vacancy boards, see Box 8.1) would encourage workers more likely to be poor. A similar pattern is observed to apply for more jobs, which increases the probability for female-headed households who are less likely to of finding a job and reduces unemployment and neces- be poor in rural areas and more likely to be poor in sity entrepreneurship. urban areas. UNDERSTANDING URBAN POVERTY 119 FIGURE 8.10: The urban poverty profile is Although households with a disabled house- similar to the rural poverty profile on some hold member are poorer in rural areas, the increase dimensions in poverty associated with disability is double in Household Size urban areas. In urban areas household with disabled 8 member have poverty rates 19 percentage points 6 above the urban average, compared to households 4 with disabled members in rural areas who have 2 poverty rates 10 percentage points above the rural 0 average. Rural Urban Households with disabled members and headed by the elderly are also more vulnerable Number of Adults in Household 5 to shocks in urban areas than in rural areas but 4 this is not the case for female headed households 3 or widows. Although imperfect, one measure of 2 vulnerability is a household’s response to the ques- 1 tion of whether they could access 200 Birr at a time 0 of emergent need. Urban households with disabled Rural Urban members are seven percentage points less likely to Proportion of Household Heads with no Formal Schooling be able to access 200 Birr when needed than rural 0.8 households with disabled members. Households 0.6 with elderly heads in urban areas are five percentage 0.4 points less likely to be able to access 200 Birr at a 0.2 time of need than households with elderly heads in 0.0 rural areas (Figure 8.12). Rural Urban There is currently no safety net for poor and vul- Non-poor Poor nerable households, such as the elderly and disabled, Source: Own calculations using HCES 2011. in urban areas. Many urban households in Ethiopia FIGURE 8.11: Being disabled, widowed, and elderly is more associated with poverty in urban areas Difference in the headcount poverty rate based on the following characteristics... 0.20 0.15 0.10 0.05 0.00 –0.05 –0.10 Primary school Secondary school Disabled household Widowed household Elderly household Elderly head Female head child not in school child not in school member member member Urban Rural Source: Own calculations using HCES 2011 and WMS 2011. 120 ETHIOPIA – POVERTY ASSESSMENT FIGURE 8.12: The elderly and disabled are 8.4  Improving urban safety nets less able to cope with shocks in urban areas Difference in probability of accessing 200 Birr in time of need Strengthening urban safety nets can further pov- based on the following characteristics… 0.00 erty reduction in Ethiopia. Chapters 1 to 4 of the –0.02 Poverty Assessment documented that the high food –0.04 prices that help reduce rural poverty hurt the urban –0.06 poor and an urban safety net is a policy tool that –0.08 allows this imbalance to be addressed. Section 8.3 –0.10 also documented that informal safety nets are weaker –0.12 in urban areas and that fiscal transfers do not fill this –0.14 gap. However, although the existing fiscal system –0.16 does not provide well-targeted transfers to the urban Disabled Widowed Elderly Elderly Female household household household head head poor, Ethiopia has a proven track record of provid- member member member ing well-targeted productive transfers in rural areas Urban Rural and this experience could be harnessed to address Source: Own calculations using HCES 2011 and WMS 2011. urban poverty. The cost of an urban safety net could be quite low and have a substantial impact on urban poverty receive no direct support for the government as food aid rates. Spending on subsidies currently designed to and the PSNP are targeted only to rural households and alleviate the cost of living for the urban poor is about those in small towns in rural areas. Urban households 0.55% of GDP (Chapter 5), but a transfer program of do benefit more than rural households from indirect 0.2% of GDP would reach 25% of the Addis Ababa subsidies in fuel and food, but this is not large enough population if transfers were generously sized (1500 to compensate for the lack of direct transfers among Birr per annum in 2011 prices). A transfer program of the bottom percentiles (Figure 8.13). This topic is dis- this size would halve the poverty rate in Addis Ababa cussed further in Chapter 5. (see Figure 8.14). FIGURE 8.13: Transfers and subsidies as a proportion of market income in rural and urban Ethiopia Direct transfers Subsidies 0.2 0.2 Subsidies as a proportion Transfer as a proportion 0.15 0.15 of consumption of consumption 0.1 0.1 0.0 0.0 0.0 0.0 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10 Decile of consumption Decile of consumption Urban Rural Source: Own estimates using HCES 2011. UNDERSTANDING URBAN POVERTY 121 For a given program budget, higher transfers FIGURE 8.14: Larger transfers have a larger have the largest impact on the poverty rate even effect on the poverty rate though they reach fewer households (Figure 8.14). 30 The direct impact of safety nets on poverty is simu- Initial poverty rate (28.1) lated by examining how many poor households 26 Poverty headcount would be lifted out of poverty as a direct result of transfers received under reasonable targeting modali- 22 ties (see Olinto and Sherpa 2014 for more details). 18 Three different cash transfer amounts were considered (all 2011 prices): a low transfer of 500 Birr per eli- 14 gible individual, a medium transfer of 1,000 Birr per 0.05 0.1 0.15 0.2 eligible individual and a high transfer of 1,500 Birr Cost as a % of GDP per eligible individual. The PSNP currently provides Transfer Amounts 650 Birr per household member. For a program that Low (500 Birr) Medium(1000 Birr) High (1500 Birr) costs 0.2% of GDP, transfers of 1500 Birr would Source: Olinto and Sherpa (2014). halve the poverty rate in Addis from 28% to 14% and transfers of 500 Birr would reduce poverty to 20%. A program that targets many households with If labor-intensive public works are time-intensive a small amount of money includes both poor and they would preclude the self-employed from par- non-poor households, and does not provide poor ticipating. It might be easier to design livelihood and households that are targeted with enough money to employment generation schemes to fit around exist- exit poverty. When considering those that remain ing self-employment activities. The Productive Safety in poverty after receiving transfer income, medium- Net Program (PSNP) employs labor-intensive public sized transfers have the largest impact on those that works during the slack season in rural areas. There remain poor for a given budget. This is because those is unlikely to be the same clear seasonal variation in that are still poor have received a meaningful amount labor demands in large urban centers as found in rural to make them less poor. Ethiopia, so this will require some thought. A labor-intensive public works scheme would Unconditional transfers will also be needed benefit fewer poor people than a livelihood and for some households. Almost one quarter of poor employment generation scheme as many poor individuals live in a household with an elderly or already work, but in low-productivity self-employ- disabled member. Not all of these households have ment. Although many poor households in urban an unemployed able-bodied adult and as such some areas have an unemployed household member (as form of unconditional transfers or transfers condi- high as 38% in Addis Ababa and 18% in all urban tional on non-labor activities, will be needed for these areas) there are also a significant number of poor households. households in which all adult members work in self- In an urban safety net program, cash transfers employment (26% in Addis Ababa, and 51% in all can increase the productivity of beneficiaries if they urban areas). This suggests that a program in which allow improved job-search outcomes among the both the unemployed and the self-employed can par- unemployed, and if they encourage self-employed ticipate will be relevant for more of the urban poor to upgrade to employment or allow the self- than a program which would preclude self-employed employed to increase their productivity. One of key members from participating on account of the time features of the PSNP has been its focus on increasing commitment, particularly for smaller urban centers. the productivity of beneficiaries. The approach will 122 ETHIOPIA – POVERTY ASSESSMENT necessarily be different in urban areas among non- increase in self-employment as a result of transfers agricultural households and in the presence of a larger (Haushofer and Shapiro 2013). waged labor market. There are three ways in which  Cash transfers can increase wage employment productivity can be increased by transfers: by providing liquidity to search for jobs. A ran- domized control trial conducted in Addis Ababa  Cash transfers can also provide the necessary found that providing active job seekers living in support to the necessity self-employed for non-central locations with small amounts of cash them to upgrade to employment. Providing (330 Birr in a program that ran for 8–11 weeks) to transfers to the necessity self-employed reduces look for jobs increased their probability of find- the cost of being unemployed while searching ing a job by seven percentage points from 19% to for a job. Some workers who, before benefits, 26% over a four months period (Franklin 2014). preferred entrepreneurship are now able to The impact was particularly strong among cash search for a job and as a result upgrade from constrained respondents. Respondents reduced self-employment to employment and become work at temporary jobs when subsidies were avail- more productive. There may be an improve- able. The transfer was unconditional, but individ- ment in the quality of matching and therefore uals had to arrive at job notice boards in the center overall productivity as job seekers have money of Addis Ababa (the main source of information to search longer for better jobs (Acemoglu and for jobs) in order to receive the transfer, resulting Shimer 2000). This may result in an increase in in a de facto job search condition for receiving the unemployment as more people search for work, money. It remains to be seen whether such sup- but it will also increase the number of people port would have the same effect if introduced on in employed positions (Table 8.7). The size of a large scale and it may be that reducing the cost the upgrading effect would depend on the con- of search by increasing the availability of informa- ditionality imposed on receiving the transfer. tion or opportunities for matching between firms If the training is complementary to staying in and workers (such as through job fairs) would be self-employment then the upgrading effect will more cost-effective. be smaller. Fewer self-employed individuals will transition from self-employment to unemploy- Transfers can also be conditioned on activi- ment as for some self-employed individuals it will ties that increase skills, job experience, and job be more beneficial to stay in self-employment search. The most common interventions to increase and receive the transfer rather than to engage employment are training, job search assistance, in job search and receive the transfer. However public works, and wage subsidies. Lessons from the some necessity self-employed will still upgrade. World Development Report on Jobs (2013) on the  While unconditional cash transfers are clas- relative effectiveness of these programs in increasing sically recognized as a social safety net, employment suggest that: (i) training can encourage increasing evidence shows a positive effect on employment if well-implemented and combining non-agricultural self-employment income. both classroom sessions with on-the-job training, Small amounts of cash, training, and supervision and (ii) job search assistance programs are successful doubled earnings and increase microenterprise in increasing employment and wages at low costs if ownership and profitability for unemployed youth job vacancies are available. Although public works are and the ultra-poor in Uganda (Blattman et al. an effective way to provide a safety net to the urban 2011; Blattman et al. 2014.) and in Kenya those poor and can be an effective self-targeting tool, the receiving unconditional transfers recorded a 38% impact of public works on employability was found UNDERSTANDING URBAN POVERTY 123 to be low to insignificant, and wage subsidies for firms 2014). Eligibility defined through a PMT system had limited effects as standalone programs. Trainings works quite well. Simulations show that nearly all to increase the profitability of entrepreneurs have also beneficiaries in a program of 500,000 would be in the been employed in a number of settings with mixed bottom 50% and three-quarters would be below the success. poverty line if were PMT targeting used. Combining Targeting a safety net within urban areas will PMT with self-selection into the program through likely rely less on geographical targeting than imposing some form of conditionality or with refine- the PSNP has. Ethiopia’s two largest cities of Addis ments of targeting by kebele officials may further Ababa and Dire Dawa are much poorer than one improve targeting. would expect and an urban safety net program would be well placed in these two cities. Other factors, such 8.5 Summary as regional equity will also need to be factored into the decision. Targeting within Addis Ababa will not Addressing poverty in Ethiopia’s large urban centers be able to rely on geographical targeting as, although will become an increasingly important component small collections of poorer houses can visibly be seen of development policy in Ethiopia. This chapter has throughout the city, these are not concentrated in shown it will require a different approach than the type specific kebeles, but instead are spread out through- of policy interventions that have been used in rural out the city. Very few kebeles have poverty rates areas given that the nature of work and social support higher than 50% or lower than 10% (Figure 8.15). systems are different. Introducing a safety net in large A proxy means test (PMT) model was constructed urban centers in Ethiopia will have a sizeable direct to assess whether poverty status could be accurately effect on poverty reduction and can be designed to have predicted in Addis Ababa using a few easily observed additional productive effects that encourage growth. characteristics of a household (Olinto and Sherpa While direct transfers can play an important role in FIGURE 8.15: Addis Ababa poverty map Source: Sohnesen 2014. 124 ETHIOPIA – POVERTY ASSESSMENT reducing poverty in large cities in Ethiopia policies reduce labor market inefficiencies will also contribute that encourage the entry and growth of large firms and a lot to poverty reduction in large urban centers. 125 GENDER AND AGRICULTURE 9 9.1 Introduction FIGURE 9.1: Gender gap in agricultural productivity, by country Agriculture is an important driver of economic Ethiopia 23%*** growth and poverty reduction in Ethiopia, but Malawi 25%*** female farmers benefit less from this because they Niger 19% are less productive than their male counterparts. Chapter 4 emphasized the importance of the agricul- Nigeria (North) 4% tural sector for poverty reduction over the past few Nigeria (South) 23%* years in Ethiopia. If the gender gap in agricultural Tanzania 6%*** productivity was narrowed, economic growth from Uganda 13%*** agriculture would increase further and everyone would 0 5 10 15 20 25 30 benefit. This chapter will look within the household in order to explore the gender gap in agricultural Source: Levelling the Field (2014). Notes: The symbols */**/*** denote statistical significance at the 10%, productivity. 5% and 1% levels respectively. The gender productivity gap in Ethiopian agriculture is one of the highest in sub-Saharan Africa, with female farm managers (largely com- in Ethiopia will help the design of agricultural prised of female heads of households) being 23% programs that are aimed at addressing this gender less productive than their male counterparts.27 gap and to ensure higher agricultural income for A recent report from the World Bank and the ONE female farmers. Identifying the main sources of these Campaign, titled Levelling the Field (2014), profiles inequalities will also help policy-makers to more effec- six countries that comprise more than 40% of Sub- tively target the most vulnerable and disadvantaged Saharan Africa’s population and presents a synthesis of farmers. Therefore, this analysis is of great value from existing evidence attesting to the breadth and depth of an economic perspective, since the alleviation of the the gender gap in African agriculture. The report draws gender gap can translate into further agricultural and upon nationally representative data from the Living countrywide economic growth (FAO 2011). The next Standards Measurement Study-Integrated Surveys on section will provide the big picture of the gender gap Agriculture (LSMS-ISA). The comparison of average in agricultural productivity in Ethiopia and compare male and female productivity across Ethiopia, Malawi, Niger, Nigeria, Tanzania, and Uganda shows that the gaps range from 13% in Uganda to 25% in Malawi as 27 Previous work carried out by Tiruneh et al. (2001) found a gender gap shown in Figure 9.1.28 This glaring gender gap pres- of 26% in a geographically limited sample of farmers. 28 The comparison by gender is made on the plot or land manager level ents an important barrier for the agricultural sector whereby a manager is defined as the person who is in charge of decision to reach its full potential.29 making for the respective piece of agricultural land. 29 Mekonnen et al. (2013), for example, find that average farmers in Developing a better understanding of the spe- Ethiopia produce less than 60% of the most efficient farmers and that the cific reasons for gender differences in productivity gender of the household head is an important determinant of inefficiency. 126 ETHIOPIA – POVERTY ASSESSMENT the findings with those obtained from other African In Ethiopia female farmers own less land, rent countries. Then, the analysis of the gender gap will less land and have fewer hours to allocate to agri- zoom in and distinguish between different groups of cultural production than male-headed households, farmers, which will provide additional, targeted policy all of which contributes significantly to lower levels implications. of productivity. On average, women plot managers spend 8.2 hours less per week on agricultural activi- Gender productivity differentials: 9.2  ties, hold 41% fewer hectares of land and have a 7.4 Ethiopia in a regional comparison percentage point lower likelihood of working on rented fields compared to men. These three factors In Ethiopia, almost half (43 percent) of the gender are the main driving forces behind the endowment gap in agricultural productivity is as a result of differ- effect estimated in Aguilar et al. (2014). ences between the amount of productive inputs used In Ethiopia, differences in returns to inputs are by men and women, a difference in endowments. The primarily related to the benefits female managers size of this endowment effect is relatively small com- reap from fertilizer, extension services, land certi- pared to other countries. In Malawi, for example, this fication, land under agricultural production, and effect was found to explain 82% of the gender produc- oxen availability. These differences in returns may be tivity gap (Kilic et al. 2013).30 In Ethiopia, the remain- explained by several aspects: complementarities with ing 57% of the gender gap is frequently attributed to other productive factors (e.g. women may need one differences in the returns women receive from the use of type of productive factor to get the most out of another the same quantity of the same inputs, stemming from factor), differences in input quality (e.g. women access structural disadvantages (Aguilar et al. 2014).31 These lower quality fertilizers or land that is less fertile), and could include, for example, the unequal treatment of gender discrimination, as well as other unobservable men and women within formal and informal societal determinants. institutions, markets, or social programs. If such struc- The regional comparison shows that returns tural issues are a key constraint, then policies would to labor and fertilizer are important drivers of the need to address broader issues of disadvantage (includ- gender gap. These factors were already identified as ing factors such as discrimination) that hinder women’s key factors contributing to the gender productivity gap productivity in the agricultural sector. However, it is in terms of the differing amounts used by men and important to note that the differences in returns can women. However, they also appear to be key factors also capture differences in other unobservable or omit- in terms of the returns men and women are able to ted explanatory variables and other factors including get from their use. Therefore, a focus on these factors errors related to measurement. may offer a particularly promising avenue for policies Although the factors accounting for the aimed at narrowing the gender gap. gender gap vary by country, Levelling the Field Land characteristics are additional factors that (2014) reveals several key determinants of particu- contribute to the gender gap through unequal lar relevance across countries. Figure 9.2 indicates returns. It is not only simple access to land that is that most factors considered in standard models of agricultural production, with the notable exception of credit, are also decisive in explaining the gender gap. 30 For Nigeria, Oseni et al. (2014) report that the structural effect is larger than the endowment effect in the North but that the opposite is Levels of labor and non-labor (seeds, chemicals, and true in the South. fertilizer) productive factors appear to have the great- 31 These estimates are based on data from the 2011–2012 Ethiopia Rural Socioeconomic Survey and the Oaxaca-Blinder decomposition methodol- est influence across the studied sample of countries, ogy which is widely used in labor economics and equally offers significant including Ethiopia. value for understanding gender productivity gaps in agriculture. GENDER AND AGRICULTURE 127 FIGURE 9.2: Factors that widen the gender gap in agricultural productivity Nigeria Nigeria Ethiopia Malawi Niger (North) (South) Tanzania Uganda Land Size Land Size Other Land Characteristics* Household Size Labor Household Male Farm Labor† Household Female Farm Labor† Hired Farm Labor† Time Spent on Farm Activities Ratio of Children to Adults within Household Improved/ Purchased Seeds Non-Labor Inputs Pesticide/ Herbicide Use† Fertilizer Use (Organic or Inorganic)† Irrigation Farm Tools & Equipment Agricultural Extension Information Export/Cash Crops Access to Markets Credit or Agricultural Capital Distance to Market or Road Non-Farm Income/ Activity Age Age & Human Capital Years of Schooling Wealth/ Consumption Wealth Levels of factor found to widen the gender gap Factor included in country analysis but not found to widen the gender gap Returns to factor found to widen gender gap Factor not included in country analysis Levels and returns to factor found to widen gender gap Source: Levelling the Field (2014). Notes: * Number of plots managed and plot-level slope, elevation, soil quality, ownership and documentation. † Includes both use and intensity of factor use (quantity/value per hectare or acre). in both access to and security of control over land. important; land tenure security is also vital due to its Deeply embedded norms and customary institu- impact on incentives to make productivity-enhanc- tions govern women’s access to land in much of rural ing investments in land. Women are disadvantaged 128 ETHIOPIA – POVERTY ASSESSMENT sub-Saharan Africa, and women are often disadvan- for example, shows that more than one household taged under both statutory and customary land tenure member manages a large proportion of plots in those systems. Insecure tenure is shown to reduce investment countries. For instance, in Uganda 1,711 out of in land, leading to lower agricultural productivity. 2,224 of the plots managed by more than one house- The empirical literature has established strong links hold member are under male-headed households. between the security of land tenure and the level of In Ethiopia, by comparison, plots are usually only investment in that land. For example, research in managed by the household head: the large majority Ethiopia found that the threat of expropriation tends of male plot managers (1,268 out of 1,277) in the sur- to reduce investment in soil conservation measures, veyed sample reside in male-headed households, while whereas land certification (which increases security of most female plot managers (231 out of 241) reside in tenure) boosts investment and rental market activity female-headed households. Therefore, the comparison (Deininger et al. 2011). Similarly, soil quality is a of productivity in Ethiopia is almost equivalent to a major determinant of crop productivity in Africa, and comparison of the productivity of female- and male- it is often claimed that land managed by women may, headed households. This finding may explain why the on average, have poorer soil quality than that managed productivity gap is so high in Ethiopia, something that by men. However, the high cost and the logistics of is explored further in this chapter. large-scale soil testing has limited the availability of Marital status appears to be an important quality data at the farm level. determinant throughout the decomposition Informal social networks also play a critical role analyses. In the decomposition estimation, most in the exchange of agricultural information and the of the gender difference results from the disadvan- adoption of agricultural technologies among farm- tage of non-married females with respect to males. ers. Detailed data on the role of social networks is This group of women farmers exhibits agricultural beyond the scope of the LSMS-ISA surveys, but existing productivity that is 30.2% lower than for their male literature has suggested that women’s social networks counterparts, with most of the difference explained tend to differ from men’s. Moreover, women and men by the structural effect (80 percent). Within the appear to use their social networks differently, which sample of non-married females, divorced women are has implications for their agricultural productivity. For the most disadvantaged. example, Kondylis and Mueller (2012) find that female extension agents in a pilot study in Mozambique were Zooming in: Refining the 9.3  successful in terms of both teaching farmers modern decomposition agricultural techniques and to induce a process of peer-to-peer learning in the targeted communities. The gender productivity differences discussed in In addition, evidence for the advantage of function- the previous section assume that female farmers can ing network structures is presented in a recent Oxfam be compared to male farmers as long as farmer and (2013) publication where collective action groups household characteristics are taken into account. (co-operatives) were found to provide substantial eco- This analysis followed the classical approach used in nomic benefits to female farmers particularly in terms the labor economics literature that describes gender of revenues and prices when marketing their produce. wage differentials and decomposes the gender pro- A distinct characteristic that separates Ethiopia ductivity differential into two parts: (i) the part of from the other countries in the regional compari- the differential explained by different levels of pro- son is the fact that women are rarely reported as ductive inputs, and (ii) the part explained by unequal managing a plot unless they are also the household returns. Recently, Ñopo (2008) proposed an alterna- head. Survey results from Tanzania and Uganda, tive method to relax the assumption that all farmers GENDER AND AGRICULTURE 129 are comparable.32 The data reveals that, for example, differential in productivity. This suggests that the seg- females tend to produce in gardens close to their regation of males and females into specific agricultural homes, focus on staple food crops and perform specific products, based on their characteristics, is an impor- tasks in the production chain (e.g. weeding), while tant component in explaining gender differentials. In males dominate cash crop production and marketing. particular, the group of males that are the most pro- These differences are clearly important determinants ductive (on average) cannot be compared (matched) to of agricultural productivity. Hence, using the alter- any females based on their characteristics. In contrast, native method, farmers are classified in two groups: the analysis also identifies a group of female plot (i) the matched: those that have an individual in the managers comprised of the most disadvantaged in other gender group with similar characteristics, and terms of productivity, and that cannot be compared (ii) the unmatched: those that do not have a similar to any male manager based on their characteristics. counterpart in the other gender group. The aim of Table 9.1 shows the average difference in a select group applying this methodology is to investigate whether of characteristics between the matched and unmatched the contribution of unequal returns to the gender gap groups on a gender basis. From this information we (which amount to a staggering 57% in Ethiopia) is can conclude that female matched managers are 9.8% at least partially driven by those farmers that do not more productive than their unmatched counterparts. have adequate counterfactuals in the opposite gen- This difference accounts for three percentage points der group. To the best of our knowledge this is the of the overall 21% gender productivity differential. In first time that this alternative methodology has been contrast, unmatched male managers are 1.75% more employed in the agricultural context. productive than the matched male group, explaining Applying the alternative method, the two tradi- 1.1 percentage points of the overall gender differential. tional terms that describe the part of the gender gap Female matched and unmatched managers explained by unequal levels of productive inputs or differ in terms of intercropping, availability of by differences in returns to these inputs are derived agricultural tools, access to female labor, income using only the matched group of individuals only. levels, and household size. Meanwhile, the male To match plot managers, a set of preexisting condi- groups differ in terms of age, years of schooling, tions were employed. The selected variables include: disabilities, use of agricultural inputs, house- age of the manager, types of crops produced, agro-eco- hold size, and of the proportion of output that logical regions, and household demographic character- is consumed. Overall, 77 female managers (32% istics (number of household members and dependency of the female managers) form the unmatched group. ratio).33 The notion of matched and unmatched male Table 9.1 shows that the unmatched managers (com- and female farmers, however, additionally allows the pared to the matched group) practice intercropping identification of and comparison of different sub- to a lower extent, use less female household labor groups to each other. This analysis uses three groups: and agricultural tools, and have a lower value of matched men and women, unmatched men, and household weekly consumption. In regards to the unmatched women. The analysis shows that they are male managers, 743 (59% of male managers) were quite different from each other, not only in charac- left unmatched. The unmatched male group is advan- teristics, but also in terms of agricultural outcomes. taged in most dimensions with respect to the matched Female managers that can be matched to male male group: they are, on average, 3.7 years younger, mangers are more productive than female managers that cannot be matched to male managers, whereas 32 The methodology proposed by Ñopo (2008) is detailed in Annex 9. the opposite result is found for male managers. 33 Crops were classified in eight different categories: cereals, pulses, oil Together, this explains 20% of the overall gender seeds, spices, root crops, fruits, vegetables, and cash crops. 130 ETHIOPIA – POVERTY ASSESSMENT TABLE 9.1: Descriptive statistics on the mean and differences, by gender and matching status Female sample Male sample Variable Unmatched Matched Difference Unmatched Matched Difference Outcome Variable Log (Self-Reported Productivity) 8.197 8.295 0.0980 8.481 8.464 –0.0175 Holder Characteristics Age (years) 49.99 47.57 –2.415 43.37 47.03 3.662*** Marital Status: Married † 0.219 0.274 0.0552 0.950 0.949 –0.00114 Years of Schooling 0.503 0.516 0.0133 1.985 1.424 –0.560*** Holder Disability † 0.0755 0.133 0.0576 0.022 0.077 0.0542*** Hours per Week for Agriculture Activities 15.49 14.11 –1.384 23.49 21.77 –1.720 Access to Extension Program † 0.245 0.356 0.111 0.354 0.375 0.0215 Access to Credit Services † 0.187 0.197 0.00956 0.249 0.309 0.0601* Holder Land Tenancy Total Land Managed (Hectares) 0.990 1.199 0.209 1.421 1.509 0.0878 Number of Fields Managed by Holder 11.18 12.32 1.141 13.43 13.04 –0.389 Total Number of Crops Produced 6.164 7.100 0.936 7.702 7.248 –0.454* Fields for which HH has a Certificate 0.681 0.528 –0.153* 0.512 0.543 0.0305 Holder’s Plot Occupation: Rented (% of 0.0143 0.0376 0.0233* 0.112 0.0933 –0.0190 parcels) Holder’s Plot Characteristics Intercropping (% of fields) 0.129 0.243 0.114*** 0.251 0.245 –0.00610 Slope 12.33 11.83 –0.502 12.91 13.82 0.916 Distance to Household 1.114 0.634 –0.480 1.241 1.768 0.526 Holder’s Agricultural Non-Labor Input-use (for Season) Fields that Use (% of Total) Irrigation 0.0199 0.0133 –0.00662 0.0272 0.0331 0.00593 Organic Fertilizer 0.351 0.343 –0.00794 0.289 0.269 –0.0200 Pesticide, Herbicide, or Fungicide 0.0796 0.108 0.0286 0.0792 0.132 0.0530*** Improved Seeds 0.0633 0.0431 –0.0202 0.0476 0.0461 –0.00143 Chemical Fertilizer Used per Hectare (KG/ 37.38 41.24 3.859 35.81 40.25 4.434 HA) Oxen per Hectare 0.797 0.989 0.192 0.989 1.365 0.376*** Agricultural Implement Access Index –0.389 0.116 0.505** 0.270 0.457 0.187* Holder’s Agricultural Labor Input-use (for Season) Household Male Labor Use (Hours/HA) 824.0 1096.1 272.1 420.9 349.6 –71.26 Household Female Labor Use (Hours/HA) 214.2 979.6 765.4*** 1227 1255 27.89 Household Child Labor Use (Hours/HA) 17.55 33.69 16.14 6.330 18.15 11.82 Total Hired Labor Use (Days/HA) 37.08 13.25 –23.83 23.14 11.67 –11.47 Total Exchange Labor Use (Days/HA) 37.65 41.64 3.981 24.30 27.73 3.434 (continued on next page) GENDER AND AGRICULTURE 131 TABLE 9.1: Descriptive Statistics on the Mean and Differences, by Gender and… (continued) Female sample Male sample Variable Unmatched Matched Difference Unmatched Matched Difference Household Characteristics Weekly Value of HH Food Consumption 165.3 206.5 41.25** 231.5 236.6 5.175 (Birr) Distance to Closest Market (KM) 60.13 52.46 –7.665 60.21 60.22 0.00609 Household Size 3.037 4.267 1.230*** 5.789 5.287 –0.503*** Dependency Ratio 0.438 0.598 0.160 0.677 0.661 –0.0162 More than Half of the Household 0.0661 0.0455 –0.0206 0.0373 0.005 –0.0324*** Production Sold † Non-agricultural Labor Income † 0.196 0.214 0.0173 0.167 0.159 –0.00814 Household Agro-Ecological Zone Classification Tropic-Warm/Semiarid † 0.0299 –6.94e–18 –0.0299 0.0231 0.000 –0.0231*** Tropic-Cool/Semiarid † 0.190 0.279 0.0891 0.226 0.345 0.120*** Tropic-Cool/Subhumid † 0.422 0.461 0.0397 0.518 0.495 –0.0230 Tropic-Cool/Humid † 0.349 0.254 –0.0952 0.213 0.158 –0.0541** Shocks Crop Damage † 0.377 0.403 0.0263 0.447 0.424 –0.0228 Total of Observations 77 161 743 511 32.4% 67.6% 59.3% 40.7% Source: ERSS 2011–12. Notes: The symbols */**/*** denote statistical significance at the 10%, 5% and 1% levels respectively. The symbol † denotes a dummy variable. achieved 0.6 more years of schooling, have a lower By construction, matched farmers are similar rate of disability, produce a higher diversity of crop in terms of the characteristics on which they were groups, and sell a higher proportion of their produc- matched but differ on other dimensions, including tion. The only contrasting characteristic is that they educational attainment, use of inputs and access to use fewer agricultural inputs (pesticides, herbicides markets.34 Significantly, relative to matched females, and fungicides, oxen per hectare, and agricultural matched males: (i) are more educated, (ii) spend more tools). These differences suggest a number of avenues hours per week on agricultural activities, (iii) are more through which relatively disadvantaged farmers in both groups, male and female, can be identified and targeted for policy intervention. 34 The variables used for matching male and female farmers contain: (i) crop categories (cereals, pulses, oil seeds, root crops, cash crops, spices, When female and male managers that can be vegetables, and fruits), (ii) categories for age of manager (less 35 years, matched are compared there is almost no difference between 35 and 49, and 50 or more), (iii) categories for household size (1–2, 3–4, 5–6, 7 or more members), categories for the dependency in returns to productive factors. Matched males are ratio (dependency ratio equal to 0, more than 0 and less than 1, and 1 16.9% more productive than females, but most of this or more—the dependency ratio that is defined as the number of children below age 10 over the number of individuals above 10 in the household, difference (97%) can be explained by the disparity in and (iv) agro-ecological zone categories (Tropic-Warm/Semiarid, Tropic- the levels or endowments of productive resources using Cool/Semiarid, Tropic-Cool/Subhumid, Tropic-Cool/Humid). A female manager will be matched to a corresponding male only if the two farmers the Ñopo (2008) methodology. match on all categories simultaneously. 132 ETHIOPIA – POVERTY ASSESSMENT likely to have access to credit, (iv) manage larger pieces FIGURE 9.3: Components of gender of agricultural land, (v) use rented field to a higher differentials in productivity extent, (vi) manage plots with higher slopes, (vii) use 21.1%** agricultural inputs, such as oxen, tools, and irrigation, 3.10%*** Segregation to a larger extent (except for organic fertilizer, which is 16.88% 8.11% 1.10%*** more frequently used by females), (viii) live in larger and wealthier households, (ix) are located closer to markets, and (x) sell their production to a lower extent. 16.40%*** These differences illustrate that despite defining a com- 13% mon support for the decomposition analysis, there is 0.50%* still variation in terms of farmer characteristics in the Traditional Alternative matched sample of male and female farmers that can methodology methodology be attributed to that part of the gender gap caused by Unmatched females Unmatched males differences in the levels of productive factors. Table 9.2 Differences in levels Differences in the returns presents the comparison of matched female and male Source: ERSS 2011–12. Notes: The symbols */**/*** denote statistical significance at the 10%, farmers using a wide range of variables. 5% and 1% levels respectively. A comparison between the traditional method and the alternative Ñopo (2008) methodology sug- gests that the managers’ classification into matched and unmatched matters and that the approach sig- calculated from comparing matched female and male nificantly influences the interpretation of results. farmers only.38 To make this comparison, Figure 9.3 shows three sets In conclusion, the application of a more refined of results. In the first bar, the traditional method from decomposition technique shows that the gender section 9.2 is employed.35 In this case, 13 percentage productivity difference is importantly explained by points (62 percent) of the overall 21.1% female-male the presence of managers that do not have a compa- productivity differential is explained by the difference rable individual within the opposite gender group. in the level of inputs.36 The remaining 8.1 percent- More precisely, this finding implies that for a subgroup age points (38 percent) are explained by differences of managers, it is not possible to find adequate coun- in returns.37 The second bar shows the result when terfactual farm managers from the other gender group, employing the alternative methodology proposed by Ñopo (2008). The main change that results from 35 In order to allow for comparability, the estimate of the decomposition using this approach is that a non-negligible part of is based on only those variables that were selected to create female- the gender differential is explained by the differ- male matches as explanatory variables. Therefore, the estimates for the contribution of the two components diverge slightly from the results ence between the matched and unmatched groups. cited before. Specifically, 4.2 percentage points (20 percent) of 36 The estimates of the gender differences in agricultural productivity presented in Figure 9.3 are based on a slightly more restricted sample the overall differential are explained by the difference of respondents. Therefore, the gap differs from the 23.4% differential between the matched and non-matched farmers in reported in Aguilar et al. (2014). 37 As already mention in section 9.2, it is, however, important to be cau- both the group of male and female farmers (1.1 and tious with the interpretation of the results presented here, particularly 3.1 percentage points, respectively). Also, varying when trying to make causal interpretations. The concern related to an omitted variable bias mentioned before is present both in the traditional levels of input-use between the two groups explain and alternative method suggested here. 16.4 percentage points of the gender gap, which 38 An estimation of the differences in levels and returns was also done using only the matched sample and the traditional methodology. The amounts to the largest part of the remaining differ- results find that differences in levels and in returns account for 50% each ence. Importantly, this difference in levels of inputs is of the gender difference. The results can be made available upon request. GENDER AND AGRICULTURE 133 TABLE 9.2: Descriptive statistics on the mean and differences for matched farmers Matched sample Variable Total Male Female Difference Outcome Variable Log (Self-Reported Productivity) 8.422 8.464 8.295 –0.169 Holder Characteristics Age (years) 47.17 47.03 47.57 0.539 Marital Status: Married † 0.782 0.949 0.274 –0.675*** Years of Schooling 1.200 1.424 0.516 –0.908*** Holder Disability † 0.0906 0.0767 0.133 0.0564 Hours per Week for Agriculture Activities 19.88 21.77 14.11 –7.663*** Access to Extension Program † 0.370 0.375 0.356 –0.0193 Access to Credit Services † 0.281 0.309 0.197 –0.112** Holder Land Tenancy Total Land Managed (Hectares) 1.432 1.509 1.199 –0.310** Number of Fields Managed by Holder 12.86 13.04 12.32 –0.714 Total Number of Crops Produced 7.212 7.248 7.100 –0.148 Fields for which HH has a Certificate 0.539 0.543 0.528 –0.0152 Holder’s Plot Occupation: Rented (% of parcels) 0.0795 0.0933 0.0376 –0.0557*** Holder’s Plot Characteristics Intercropping (% of fields) 0.244 0.245 0.243 –0.00184 Slope 13.33 13.82 11.83 –1.993** Distance to Household 1.488 1.768 0.634 –1.133 Holder’s Agricultural Non-Labor Input-use (for Season) Fields that Use (% of Total) Irrigation 0.0282 0.0331 0.0133 –0.0198** Organic Fertilizer 0.287 0.269 0.343 0.0744** Pesticide, Herbicide, or Fungicide 0.126 0.132 0.108 –0.0241 Improved Seeds 0.0454 0.0461 0.0431 –0.00301 Chemical Fertilizer Used per Hectare (KG/HA) 40.49 40.25 41.24 0.995 Oxen per Hectare 1.272 1.365 0.989 –0.376** Agricultural Implement Access Index 0.373 0.457 0.116 –0.341** Holder’s Agricultural Labor Input-use (for Season) Household Male Labor Use (Hours/HA) 534.1 349.6 1096.1 746.5*** Household Female Labor Use (Hours/HA) 1186.8 1254.9 979.6 –275.3 Household Child Labor Use (Hours/HA) 21.99 18.15 33.69 15.54 Total Hired Labor Use (Days/HA) 12.06 11.67 13.25 1.580 Total Exchange Labor Use (Days/HA) 31.17 27.73 41.64 13.90 (continued on next page) 134 ETHIOPIA – POVERTY ASSESSMENT TABLE 9.2: Descriptive statistics on the mean and differences for matched farmers (continued) Matched sample Variable Total Male Female Difference Household Characteristics Weekly Value of HH Food Consumption (Birr) 229.2 236.6 206.5 –30.11* Distance to Closest Market (KM) 58.30 60.22 52.46 –7.755* Household Size 5.035 5.287 4.267 –1.019*** Dependency Ratio 0.645 0.661 0.598 –0.0630 More than Half of the Household Production Sold † 0.0149 0.00488 0.0455 0.0406* Non-agricultural Labor Income † 0.173 0.159 0.214 0.0547 Household Agro-Ecological Zone Classification Tropic-Warm/Semiarid † 0 0 0 0 Tropic-Cool/Semiarid † 0.329 0.345 0.279 –0.0661 Tropic-Cool/Subhumid † 0.487 0.495 0.461 –0.0336 Tropic-Cool/Humid † 0.182 0.158 0.254 0.0950** Shocks Crop Damage † 0.419 0.424 0.403 –0.0213 Total of Observations 672 511 161 76.0% 24.0% 24.2% Source: ERSS 2011–12. Notes: The symbols */**/*** denote statistical significance at the 10%, 5% and 1% levels respectively. The symbol † denotes a dummy variable. since their unique characteristics cannot be matched to are discussed in more detail in the next section. For any individual in that group. Adding these individu- the least productive female farmers (the unmatched als to the classical decomposition analyses violates the women), inputs play a role, but crop choice and common support assumption. other factors matter. Therefore, policies targeting the In conclusion, the application of a more refined institutions and gender norms that trap some women decomposition technique delivers three distinct farmers at the bottom of the productivity distribution groups: (1) unmatched women (who are the least may be of particular relevance. These policies will not productive); (2) unmatched men (who are the most only have the greatest potential for poverty reduction productive); and (3) matched male and female but could also benefit female farmers who have already farmers. Focusing on these groups separately provides reached higher productivity levels. us with critical insights into what drives the gender gap in agricultural productivity. More precisely, this Explaining gender differences in 9.4  novel approach suggests that when matched men and input-use women are compared, the vast majority of the gender gap in productivity can be explained by differences in Differences in endowments matter as almost half levels of productive factors. Therefore, for this group, of the gender gap in agricultural productivity interventions that increase women’s access to inputs (43% considering all farmers in the compari- such as land, labor, and technology are critical. The son) in Ethiopia is explained by differences in man- factors that may be constraining access to these inputs agers’ characteristics, land attributes and access to GENDER AND AGRICULTURE 135 productive resources. On average, female managers variety of underlying causes for these findings, formal farm smaller plots, spend less time on agricultural and informal institutions that govern how women are activities, are less likely to use rented fields, use less non- treated according to their marital status are likely to be labor inputs, and tend to inhabit smaller households of significant importance. However, it is worth not- with lower average income. The purpose of this section ing that the evidence also indicates that even married is to further analyze the gender difference in levels of use females are also restricted in terms of endowment levels. of some of these variables. A more detailed understand- Gender differences in productivity and hours ing of disparities in these variables is valuable from a spent on agricultural activities are largest for the policy perspective since it allows us to characterize the oldest farmers, while disparities in land tenancy are most disadvantaged female farm managers. This section greater for the youngest farmers. Table 9.3 suggests specifically focuses on those variables that were found that for older age cohorts the gender gap in productivity to be the most decisive in explaining the gender gap is particularly severe, ranging from a negligible 2.8% in terms of their level of use by male and female farm- difference in productivity among the youngest group to ers: managers’ time allocated to agricultural activities, a 38% difference for the oldest cohort. In contrast, the land size, and the proportion of fields that are rented. youngest group of females are most disadvantaged in On average, female managers spend 8.39 fewer their access to land: the youngest group of females holds hours in agricultural activities, manage 42.9% less 72% less land and rents 13.4 percentage points less of land, and a lower proportion of such managed land their land with respect to same-aged males, while the is rented (7.7 percentage points less). Table 9.3 corresponding difference for the oldest group indicates allows the identification of differences in the size of a 39% difference in land farmed and a 4.6 percentage the gender gap for a range of subgroups of female point lower proportion of land rented.39 farm managers. These subgroups were formed using Gender differences in productivity and use of personal, household, and community characteristics. productive factors (except for land managed) are Table 9.3 shows the average gender differences in the largest for female farmers in small households. agricultural productivity for each subgroup, as well as Females in the smallest sized households (one to three gender differences in the level of the selected endow- members) are 33.4% less productive than males in ments used (hours of agricultural activity, land man- the same group. The gender gap for those in the next aged and proportion of land rented). The differences household size group (four to six members) is a non- are useful from a descriptive perspective, though no significant 22.7 percent. Finally, for the largest sized causal relationship can be inferred. households (seven members or more) the gender gap Marital status is one of the key determinants of in productivity is only 15.1 percent. Gender differ- the gender gap and the analysis shows that widowed ences in hours spent on agricultural activities and the females are the most disadvantaged group in terms rented proportion of land follow a similar pattern of time available to spend on agricultural activities. ranging from a 9.9 hour and 10.7 percentage point Widowed females are not only 29% less productive differential for the smallest households to a 6 hour than the average male (which echoes the evidence pre- and 1.4 percentage point differential for the largest sented in Section 9.2) but they also spent 11 hours less households.40 on agricultural activities, manage 29% less land, 7.5 percentage points less of which was rented. Divorced 39 A separate analysis (not shown in the table) finds that the differences females were 24% less productive than the average for the oldest group, except for land holding, are significantly reduced man and are the most disadvantaged group in terms after taking into account marital status. 40 A similar systematic pattern is not found in the differences of land of land tenancy, since they manage 80% less land than managed. Yet, the gender difference for the largest and smallest household the average male. Although it is likely that there are a size is 44 and 38% in detriment of females. 136 ETHIOPIA – POVERTY ASSESSMENT TABLE 9.3: Gender differences by different groups Females Agricultural Hours on Area of land Land (%) productivity agriculture farmed rented (1) Overall differenceᵃ –0.2340*** –8.3995*** –0.4293*** –0.0767*** (0.0871) (1.7524) (0.1190) (0.0111) (2) Manager Marital Statusb Married 31.4% –0.0032 –4.1852* –0.5874*** –0.0798*** (0.1520) (2.4064) (0.1620) (0.0145) Divorced 14.4% –0.2425 –4.1860 –0.8061*** –0.0760*** (0.1825) (4.3465) (0.2764) (0.0223) Widowed 54.2% –0.2895** –11.0067*** –0.2871* –0.0756*** (0.1249) (2.1569) (0.1602) (0.0122) (3) Manager Age Aged less than 35 16.1% –0.0281 –8.3600** –0.7217*** –0.1340*** (0.2106) (3.6420) (0.2493) (0.0236) Aged between 35 and 49 41.5% –0.0869 –6.9289*** –0.5188*** –0.0592*** (0.1405) (2.3503) (0.1355) (0.0186) Aged 50 or more 42.4% –0.3839** –9.1338*** –0.3922** –0.0464*** (0.1497) (2.6181) (0.1698) (0.0114) (4) Household (HH) Size HH with 1 to 3 members 43.2% –0.3342** –9.8929*** –0.3812** –0.1072*** (0.1372) (2.6258) (0.1838) (0.0240) HH with 4 to 6 members 47.5% –0.2271 –7.3952*** –0.1093 –0.0792*** (0.1410) (2.1201) (0.1285) (0.0153) HH with 7 or more members 9.3% 0.1512 –6.0365 –0.4409** 0.0148 (0.3008) (5.4255) (0.1980) (0.0447) (5) Household (HH) Compositionb HH w/ no males † 20.8% –0.0695 –12.7897*** –1.2766*** –0.0930*** (0.1836) (2.1093) (0.2110) (0.0130) HH w/ oldest male aged 12 or 23.7% –0.4144*** –4.2705 –0.6376*** –0.0822*** less (0.1418) (2.9375) (0.2334) (0.0162) HH w/ oldest male aged from 38.1% –0.3390** –9.4566*** 0.0186 –0.0630*** 13 to 24 (0.1413) (2.3071) (0.1254) (0.0149) HH w/ oldest male aged 25 or 17.4% 0.1389 –6.2195 –0.2669 –0.0825*** more (0.1888) (4.5112) (0.1750) (0.0143) (6) Main Crop Category (by land farmed) Cereals 75.3% –0.1630 –8.8182*** –0.5620*** –0.0849*** (0.1089) (1.9255) (0.1390) (0.0136) Pulses 7.2% 0.1887 –12.7985*** –0.2635 –0.0637** (0.3470) (3.5653) (0.3479) (0.0313) Oil seeds 6.0% –0.1166 –10.2889 –0.1338 –0.1125*** (0.2217) (7.2798) (0.2343) (0.0397) Root crops 1.7% –0.9783 –15.1157*** –0.1277 –0.1213 (0.6212) (3.4099) (0.5953) (0.1089) (continued on next page) GENDER AND AGRICULTURE 137 TABLE 9.3: Gender differences by different groups (continued) Females Agricultural Hours on Area of land Land (%) productivity agriculture farmed rented Cash crops main crop 7.2% –0.8608*** 8.3754 –0.0210 –0.0286** (0.3073) (7.1200) (0.2832) (0.0126) Spices, vegetables or fruits 2.6% –1.3230** –28.4292*** –1.7821*** –0.0333 (0.6291) (8.1975) (0.3763) (0.0232) (7) Administrative Regions Tigray 9.8% –0.0667 –6.0840 –0.4511* –0.0889** (0.1338) (3.9267) (0.2386) (0.0386) Amhara 27.2% –0.3906** –12.6692*** –0.9837*** –0.1355*** (0.1770) (2.6532) (0.1949) (0.0200) Oromiya 18.3% –0.0108 –8.1829** –0.0503 –0.0423*** (0.1825) (3.9650) (0.1702) (0.0138) SNNP 29.8% –0.2552* –3.3150 –0.0258 –0.0436** (0.1516) (3.2092) (0.1681) (0.0186) Other Regions 14.9% –0.2685 –3.3453 –0.6550*** –0.0524 (0.3395) (2.6219) (0.1483) (0.0497) (8) Enumeration Area (EA) Populationc Less than 3.5k 27.2% –0.4136** –10.9936*** –0.2053 –0.0780*** (0.1841) (3.9050) (0.1930) (0.0193) Between 3.5 and 6.5k 38.3% –0.2749* –8.5941*** –0.5150** –0.0803*** (0.1525) (2.7188) (0.2124) (0.0205) 6.5k or more 40.4% –0.0117 –6.2384** –0.4860*** –0.0729*** (0.1207) (2.6740) (0.1596) (0.0173) (9) Distance to Woreda Town (WT)c Less than 10 kilometers 32.8% –0.3748* –7.3419** –0.4452* –0.0745*** (0.1970) (3.2068) (0.2345) (0.0193) Between 10 and 20 kilometers 26.8% –0.1210 –10.9091*** –0.3992** –0.0413** (0.1261) (3.1203) (0.1832) (0.0170) 20 kilometers or more 40.4% –0.1530 –7.5747*** –0.4227** –0.1041*** (0.1485) (2.6264) (0.1846) (0.0193) Observations 15.8% 1,481 Source: ERSS 2011–12. Notes: Agricultural productivity is measured as value of logged birrs per hectare. Hours on agriculture are weekly hours spend on agricultural activities. Area of land farmed is the logged hectares the manager farmed. Land rented is the proportion of rented fields. Symbols */**/*** denote statistical significance at the 10%, 5% and 1% levels respectively. Clustered Standard Errors, at Enumeration Area level, are presented in paren- theses. a The reported overall agricultural productivity gender difference corresponds to the number estimated in Aguilar et al. (2014). The rest of the differentials are estimated using more restricted sample defined by information availability. Using this group an agricultural productivity gap of –0.2100*** would result. b Female disadvantage of groups (2) and (4) are with respect to the average male manager. c According to the LSMS-ISA ERSS documentation, Enumeration Area should not be interpreted in a sociological sense but rather as the primary geographical classification. Women benefit from greater land access in in the household have the greatest disadvantage in direct relationship to the age of the oldest man access to land (127% lower than the average male in their household. Females with no males present manager) and in the proportion of land rented 138 ETHIOPIA – POVERTY ASSESSMENT (9.3 percentage points less than the average male). suggests that closing this gap requires both types of Interestingly, these differences remain after taking policies: (i) changing gender norms and institutions into account marital status. Again, this set of results in order to economically empower female farmers; suggest that gender norms play an important role in and (ii) ensuring that differences in endowments determining the extent to which women have access between male and female farmers are addressed. to productive resources. This conclusion draws on findings from two recent The largest gender productivity differentials by World Bank publication, Aguilar et al. (2014) and crop groups are found for two categories: (i) cash Levelling the Field (2014), as well as on novel analyti- crops; and (ii) spices, vegetables or fruits. Table 9.3 cal work that builds on decomposition methods used indicates that for these groups, women are 86% and to determine the extent to which the differences in 132% less productive, respectively. Interestingly, the productivity are explained by: (i) gender disparity in former group does not show big disparities in the the levels of productive inputs (endowment effect); endowment levels, which suggests that structural and/or (ii) gender inequality in the returns to those factors are the most important cause of the gender inputs (structural effect). productivity gap for cash crops. The only endowment The results suggest that endowments par- difference that is significant (though small) for cash ticularly matter in determining the gender gap in crops is the proportion of land rented (a 2.8 percent- agricultural productivity among male and female age point gap). In contrast, women mainly growing farmers who share similar characteristics. For this “spices, vegetables or fruits” experience significant group (as well as for the others) interventions that disadvantages with respect to males in the same group increase access to labor, land, and other inputs, as for two endowments: hours spent on agriculture (28.4 well as how effectively they are used, are critical. For fewer hours for females) and land managed (178% less some of these interventions, evidence exists on what for females). Considering these results, it is no sur- approaches may work. For example, Rwanda’s expe- prise that many interventions targeting the economic rience with joint land titling has been documented empowerment of women are aimed at transitioning to increase women’s control over land. However, women into higher-value and more commercially ori- other interventions are promising, but have yet to ented production. It is important, however, that this be proven. For example, to address women’s labor approach takes into account that female farmers who shortages, possible interventions include financing already transitioned at least partially face disadvantages mechanisms such as vouchers to hire labor or coop- in terms of access to productive factors that constrain erative labor pools. Given the importance of the them from reaping equal benefits. gender gap for agricultural productivity in Ethiopia, Finally, in geographical terms, females in less piloting and learning which works best will have populated locations are more disadvantaged. In significant payoffs. particular, those female managers living in less inhab- In contrast to the farmers who share charac- ited enumeration areas are 41% less productive than teristics, the disadvantage captured by the struc- males in the same areas. This difference is reflected in tural effect appears to be especially relevant when 10 fewer hours spent on agricultural activities. considering male and female farmers who are observationally too different to include in the pro- 9.5 Conclusion ductivity decomposition: within this group, women are found at the bottom and men at the top of the This evidence presented in this chapter builds on productivity distribution on average. This result sug- a variety of decomposition techniques to assess gests that for this group of female farmers (the low level the gender gap in agricultural productivity and producers) disadvantages caused by gender norms and GENDER AND AGRICULTURE 139 institutions matter relatively more. Relative to address- provision of extension services that are more gender ing the gaps in endowments, there are fewer proven sensitive, is detailed in Box 9.1. policy interventions for addressing gender norms and Finally, focusing on the most important endow- institutions. Pilot interventions, for example, to help ments, the analysis highlights a number of socio-eco- female farmers move into higher value/cash crops and nomic and community characteristics, in particular marketing, can help provide valuable lessons that can the marital status of farmers, which correlate with be taken to scale. differences in the level of endowments between men One example of a project in Ethiopia that tried and women in order to help identify and target the to address both dimensions, namely through the most disadvantaged groups of female farmers. BOX 9.1: Policy example: Government response and RCBP in Ethiopia The Rural Capacity Building Project (RCBP), implemented between 2006 and 2012, was designed to strengthen agricultural services and systems and make them more responsive to clients’ needs. The project encompassed five major components: (1) Assisting Agricultural TVET colleges in training Ethiopia’s Development Agents (DAs); (2) Improving and scaling up the effectiveness of this agricultural extension system’s capacity to respond to farmers’ demands to enhance women’s participation; (3) Strengthening agricultural research, through institutional strengthening of the National Agricultural Research System (NARS); (4) Improving Ministry of Agriculture and Rural Development (MoARD) capacity; and (5) Assisting with analytical work. The RCBP was implemented in 10 regions, 127 woredas, 635 kebeles and 2,500 Farmer Training Centers (FTC) in the country, beginning in 2007. From the beginning, it was decided to make an impact evaluation an integral part of the RCBP in order to rigorously assess its effectiveness. The evaluation carried out by Buehren et al. (2014) primarily relied on two rounds of survey data collected from farming households in both RCBP project and non-project woredas. The first round of data collection was carried out soon after the launch of project implementation, and efforts were made to revisit all baseline respondents after project completion in 2012. The resulting panel dataset comprises 1,485 households and nearly 300 DAs spread across four regions: Amhara, Oromia, Southern Nations, Nationalities and People’s Region (SNNPR), and Benishangul-Gumuz (BSG). The impact analysis builds on an estimation of difference-in differences using matching techniques and focuses on intention- to treat estimates (comparing farmers in project woredas with their counterfactuals in non-project woredas). The results suggest that the RCBP has had a significant impact on economic activity: RCBP households utilize more farm labor relative to non-RCBP households, with an additional one-half of a person contributing to income in RCBP households. The increase in the number of people who contribute to household income is not only statistically significant but it is also estimated to be a sizeable 23–27% increase over the baseline value. Over the evaluation period, the amount of farm labor declines overall. However, this decline is significantly lower in RCBP areas. On average, RCBP households use 10–12% more labor, in terms of the number of people within a household that work on farming, compared to households in the non-RCBP counterfactual group. In addition, there is a positive impact of 17–24% on the size of land under agricultural production. It should also be noted that the total size of land under agricultural production fell between the baseline and follow-up surveys for all households included in the studied sample. However, the decline is significantly lower among RCBP households, indicating that RCBP households have a larger area of land under farm use by the end of the project. The authors define high value crops as those with a higher value than traditional staple crops and that are used primarily to sell in the market, as well as an extended set of marketable crops termed somewhat high-value crops.a The analysis indicates that households located in RCBP woredas are more likely to grow high value crops, in response to the program. The estimated increase is in the range between 9–11 percentage points for high value crop production. For the sale of these crops, an increase in RCBP areas and a decrease in non-RCBP areas are observed. Considering the net difference, this yields a statistically significant and large increase in the sale of high value crops among RCBP households relative to control areas by 8–12 percentage points. A similar pattern is documented for the somewhat high value crops and incidence of sale of high and somewhat high value crops. Repeating the estimations and disaggregating by the gender of the household head shows that, while the positive impact of the program on the number of individuals who contribute to income is lower in female-headed households, this difference is not statistically significant. Additionally, there is no significant statistical difference in the impact of the number of people who work on the farm or for income from agriculture. The same is true for consumption. Finally, in the case of farm size and growing somewhat high value or high value crops, there is no statistically significant difference between the impact of the program on male- and (continued on next page) 140 ETHIOPIA – POVERTY ASSESSMENT BOX 9.1: Policy Example: Government response and RCBP in Ethiopia (continued) female-headed households. A positive program impact only for male-headed households can merely be demonstrated in the case of livestock. Hence, the evidence indicates that for most of the outcomes of interest, the program seems to have benefitted men and women equally. These results powerfully suggest that government intervention does not necessarily lead to a reinforcement of the mechanisms that underlie the prevailing gender gap in agricultural productivity. Instead, the authors argue and the results of the impact evaluation suggest that adapting the extension system supported by the RCBP , which traditionally serves male farmers, to the needs of women farmers, can contribute considerably to closing the gender gap and unleashing the full potential of farming households in Ethiopia. a High value crops include coffee, mango, avocado, banana, guava, casmir, sesame, peanuts, clove, ginger, tobacco, khat, dinbilal, water melon, eucalyptus, pineapple, orange, papaya, garlic, lemon, sunflower, cumin, cauliflower, rapeseed, cucumber, apple, and spices. Some- what high value crops additionally include teff, lentil, onion, pepper, sugar cane, and Ethiopian hops, which are often consumed rather than sold in the market. ANNEXES 143 ANNEX 1 Sensitivity analysis of poverty estimates: curves between 2005 and 2011 using (i) the restricted further results sample in 2011 and (ii) the full sample shows that the different timing of the surveys did not introduce bias The food share of total consumption varies consider- at this level of aggregation (Figure A1.2). ably across years, consumption and deflator choice (Figure A1.1). The poorest have the highest food share FIGURE A1.2: Growth incidence curve for 2005 dropping down to around 20% for the richest. When to 2011 for full sample and for partial sample the CPI deflator is used the food share increases from Growth incidence, 2005–2011 10 1996 to 2000 for the bottom 60% of the population but then drops from 2000 to 2005 and further in Annual growth rate, % 5 2011. When the HICES-based deflator is used, the food share in 2011 is similar for many poor households 0 than the food share in 2005. While the HICES surveys of 1996, 2000 and –5 2005 were implemented in February (Yekatit) and July (Hamele), the HCES 2011 was implemented –10 in all months. To assess the implication of the differ- 0 10 20 30 40 50 60 70 80 90 100 ent timing of data collection, the HCES 2011 round Expenditure percentile can artificially be restricted to the same months of Full sample Same months sample February and July. Comparing the growth incidence Source: own calculations using HICES 2005 and HCES 2011. FIGURE A1.1: Food share in total consumption across time for different deflators HICES-based deflator CPI deflator 80% 80% 60% 60% Food Share Food Share 40% 40% 20% 20% 0% 0% 1 21 41 61 81 1 21 41 61 81 Consumption percentile Consumption percentile 1996 2000 2005 2011 Source: own calculations using the HICES1996, HICES 2000, HICES 2005 and HCES 2011. 144 ETHIOPIA – POVERTY ASSESSMENT Poverty profile: further results TABLE A1.1: Difference in means, household characteristics by poverty status and consumption decile (1996–2011) (Total) 1996 2000 2005 2011 Mean, Mean, Mean, Mean, non- Mean, non- Mean, non- Mean, non- Mean, Variable poor poor sig. poor poor sig. poor poor sig. poor poor sig. Household Adult Equivalence 4.66 5.33 * 4.45 5.30 * 4.30 5.65 * 4.49 5.63 * Household size from roster 5.68 6.51 * 5.50 6.46 * 5.31 6.90 * 5.49 6.82 * Household head age 44.36 45.45 * 43.28 46.31 * 42.73 45.66 * 43.30 46.41 * Household head gender (male) 0.81 0.84 * 0.80 0.81 0.80 0.84 * 0.82 0.83 Household head marital status 0.83 0.85 * 0.83 0.83 0.82 0.85 * 0.83 0.86 * (married) HH head education level: No formal 0.65 0.78 * 0.64 0.78 * 0.64 0.72 * 0.59 0.70 * schooling completed HH head education level: Grade 1–3 0.15 0.13 0.14 0.12 0.12 0.11 0.13 0.12 HH head education level: Grade 4–6 0.09 0.05 * 0.10 0.06 * 0.11 0.10 0.12 0.11 HH head education level: Grade 7–8 0.04 0.02 * 0.04 0.02 * 0.05 0.04 * 0.06 0.04 * HH head education level: Grade 9–11 0.02 0.01 * 0.02 0.01 * 0.02 0.01 * 0.04 0.02 * (or incomplete certificate) HH head education level: Grade 12 or 0.04 0.01 * 0.04 0.01 * 0.04 0.01 * 0.02 0.01 * completed certificate HH head education level: Degree/ 0.01 0.00 * 0.01 0.00 * 0.02 0.00 * 0.04 0.01 * Diploma program Household head years of completed 2.11 1.36 2.62 1.50 schooling Household head work status (past 12 0.96 0.97 0.97 0.96 0.97 0.96 0.98 0.97 months) Household head is self-employed 0.72 0.78 * 0.76 0.82 * 0.78 0.82 * 0.82 0.86 * Household head is an unpaid family 0.00 0.00 0.00 0.00 0.01 0.01 0.01 0.01 worker Household head is an employer 0.06 0.04 * 0.08 0.06 * 0.06 0.05 * 0.02 0.01 * Household head is an employee 0.10 0.08 0.07 0.04 * 0.07 0.04 * 0.08 0.05 * Total number of members currently 2.43 2.63 * 2.29 2.52 * 2.20 2.61 * 2.24 2.56 * employed in HH Household employment as share of 0.90 0.91 0.90 0.88 * 0.90 0.88 * 0.88 0.86 * working age population Household male employment as share 0.42 0.43 0.41 0.40 0.40 0.41 0.40 0.40 of male working age population Household female employment as 0.48 0.48 0.49 0.49 0.50 0.47 * 0.48 0.45 * share of working age women (continued on next page) Annex 1 145 TABLE A1.1: Difference in means, household characteristics by poverty status and consumption decile (1996–2011) (Total) (continued) 1996 2000 2005 2011 Mean, Mean, Mean, Mean, non- Mean, non- Mean, non- Mean, non- Mean, Variable poor poor sig. poor poor sig. poor poor sig. poor poor sig. Total HH members who are self 0.80 0.86 * 0.95 1.10 * 1.01 1.10 * 1.13 1.16 employed Total HH members who are employers 0.07 0.04 * 0.52 0.55 0.06 0.05 0.02 0.01 * Total HH members who are employees 0.26 0.16 * 0.23 0.13 * 0.21 0.15 * 0.23 0.18 * Total number of HH members involved 1.11 1.50 * 0.96 1.29 * 1.03 1.64 * 0.99 1.47 * in unpaid Total number of HH members involved 0.79 0.85 0.61 0.73 * 0.44 0.52 * 0.32 0.38 * in domestic work HH sector of occupation: agriculture 0.78 0.85 * 0.78 0.81 * 0.79 0.85 * 0.78 0.85 * (hhead) HH sector of occupation: 0.03 0.02 0.01 0.01 * 0.03 0.03 0.02 0.02 manufacturing (hhead) HH sector of occupation: construction 0.01 0.00 0.00 0.00 0.01 0.01 0.02 0.01 (hhead) HH sector of occupation: mining/ 0.00 0.00 0.00 0.01 0.00 0.00 0.01 0.00 energy (hhead) HH sector of occupation: education, 0.00 0.00 0.02 0.00 * 0.04 0.03 * 0.05 0.03 * health, social services (hhead) HH sector of occupation: professional 0.01 0.00 * 0.02 0.01 * 0.03 0.02 * 0.03 0.01 * services (pub or private) (hhead) HH sector of occupation: services & 0.08 0.05 * 0.11 0.12 0.07 0.04 * 0.09 0.06 * trade (hhead) Number of adults in households 3.17 3.34 * 2.98 3.35 * 2.87 3.50 * 2.98 3.55 * Total number of non-working age 2.88 3.57 * 2.86 3.53 * 2.77 3.83 * 2.86 3.72 * dependents in HH Total number of working age adults 2.80 2.95 * 2.63 2.93 * 2.54 3.08 * 2.63 3.10 * in HH Total number of employed dependents 0.48 0.61 * 0.39 0.56 * 0.46 0.71 * 0.41 0.57 * in HH Total number of employed working 2.43 2.63 * 2.29 2.52 * 2.20 2.61 * 2.24 2.56 * age adults in HH HH sector of occupation: agriculture 2.09 2.48 * 1.91 2.20 * 1.79 2.31 * 1.75 2.18 * (total in HH) HH sector of occupation: 0.06 0.06 0.03 0.01 * 0.09 0.10 0.06 0.07 manufacturing (total in HH) HH sector of occupation: construction 0.02 0.02 0.01 0.01 0.02 0.03 * 0.03 0.04 (total in HH) HH sector of occupation: mining/ 0.01 0.02 0.01 0.02 0.01 0.01 0.01 0.01 energy (total in HH) (continued on next page) 146 ETHIOPIA – POVERTY ASSESSMENT TABLE A1.1: Difference in means, household characteristics by poverty status and consumption decile (1996–2011) (Total) (continued) 1996 2000 2005 2011 Mean, Mean, Mean, Mean, non- Mean, non- Mean, non- Mean, non- Mean, Variable poor poor sig. poor poor sig. poor poor sig. poor poor sig. HH sector of occupation: education, 0.00 0.00 0.03 0.00 * 0.08 0.07 0.10 0.09 health, social services (total in HH) HH sector of occupation: professional 0.02 0.00 * 0.04 0.01 * 0.05 0.03 * 0.05 0.02 * services (total in HH) HH sector of occupation: services & 0.22 0.16 * 0.33 0.38 0.21 0.19 * 0.26 0.21 * trade (total in HH) HH sector of occupation: other or not 0.12 0.07 * 0.01 0.01 0.00 0.00 0.00 0.00 defined (total in HH) HH has at least one member 0.76 0.83 * 0.77 0.81 * 0.77 0.84 * 0.76 0.84 * employed in agriculture HH has at least one member 0.04 0.03 0.02 0.01 * 0.06 0.07 0.05 0.05 employed in manufacturing HH has at least one member 0.02 0.01 0.01 0.01 0.01 0.02 0.02 0.02 employed in construction HH has at least one member 0.01 0.01 0.01 0.01 * 0.01 0.01 0.01 0.01 employed in mining/energy HH has at least one member 0.00 0.00 0.02 0.00 * 0.06 0.05 0.07 0.06 * employed in education/social services HH has at least one member 0.01 0.00 * 0.02 0.01 * 0.03 0.02 * 0.04 0.02 * employed in professional sector HH has at least one member 0.13 0.10 * 0.19 0.22 0.13 0.12 * 0.17 0.13 * employed in services/transport Fraction of HH members over 6 with 0.29 0.19 * 0.30 0.22 * 0.35 0.32 * 0.48 0.44 * formal education (grade 1–3) Fraction of HH members 12+ with 0.31 0.21 0.32 0.24 0.37 0.35 0.48 0.44 formal education (grade 1–3) Maximum years of schooling in HH 3.81 3.67 5.16 4.74 * Source: own calculations using HICES 1996, 2000, 2005 and 2011. Note: All standard errors are clustered by enumeration area. * represents a significant difference at the 5% level. Annex 1 147 TABLE A1.2: Difference in means by percentile of consumption distribution (1996) Bottom 10% vs. Bottom 40%* vs. top Bottom 10% vs. top bottom 40%* 60% 60% Mean Mean bottom bottom Mean top Variable 10% 40% 60% Diff. t-stat Diff. t-stat Diff. t-stat Age of household head 45.7 45.6 44.4 0.2 0.249 1.2 3.133*** 1.4 2.319** Household head is male 0.832 0.833 0.819 –0.001 –0.042 0.013 1.269 0.013 0.713 Household head is married 0.861 0.845 0.829 0.016 0.973 0.016 1.522 0.032 1.939* Number of household 6.666 6.507 5.734 0.159 1.211 0.773 8.833*** 0.932 7.239*** members Proportion of unpaid 0.219 0.226 0.188 –0.007 –0.597 0.038 5.103*** 0.031 2.366** workers Proportion of children (<12) 0.461 0.444 0.395 0.017 2.003** 0.049 8.532*** 0.066 7.509*** Proportion of dependents 0.547 0.534 0.490 0.014 1.721* 0.044 8.282*** 0.058 6.848*** Caloric intake (def. varies 1245.8 1580.8 2260.4 –335.0 –11.426*** –679.6 –24.444*** –1014.5 –28.268*** by year)** Occupation of household 0.789 0.798 0.741 –0.009 –0.475 0.058 3.548*** 0.048 2.079** head: agriculture Occupation of household 0.020 0.020 0.025 0.000 –0.041 –0.005 –1.150 –0.005 –0.790 head: manufacturing Occupation of household 0.003 0.005 0.008 –0.002 –1.177 –0.003 –1.471 –0.005 –2.173** head: construction Occupation of household 0.002 0.002 0.004 0.000 0.426 –0.002 –2.451** –0.002 –1.471 head: mining/energy Occupation of household 0.001 0.001 0.009 0.000 –0.410 –0.008 –6.476*** –0.008 –6.337*** head: prof. services Occupation of household 0.053 0.041 0.068 0.013 1.552 –0.027 –3.924*** –0.014 –1.447 head: services & trade Household lives in an urban 0.127 0.099 0.171 0.028 1.942* –0.071 –4.855*** –0.044 –2.089** area Source: CSA Household Income and Consumption Expenditure Surveys 1996, 2000, 2005 and 2011. Notes: *Bottom 40% refers to those in the bottom 40% of the consumption distribution, without including the bottom 10%. ** Caloric intake is measured differently across time, as such these measures are not comparable. Significance levels are defined as follows: * 10%, ** 5%, *** 1%. 148 ETHIOPIA – POVERTY ASSESSMENT TABLE A1.3: Difference in means by percentile of consumption distribution (2000) Bottom 10% vs. Bottom 40%* vs. top Bottom 10% vs. top bottom 40%* 60% 60% Mean Mean bottom bottom Mean top Variable 10% 40% 60% Diff. t-stat Diff. t-stat Diff. t-stat Age of household head 46.9 46.2 43.4 0.7 0.992 2.8 6.386*** 3.5 5.491*** Household head is male 0.805 0.814 0.802 –0.010 –0.536 0.012 1.136 0.002 0.144 Household head is 0.825 0.832 0.833 –0.006 –0.377 –0.001 –0.148 –0.008 –0.469 married Number of household 6.946 6.365 5.528 0.581 4.978*** 0.836 11.015*** 1.418 12.236*** members Proportion of unpaid 0.186 0.196 0.170 –0.010 –0.943 0.026 3.590*** 0.015 1.348 workers Proportion of children 0.446 0.432 0.407 0.014 1.518 0.025 4.004*** 0.039 4.679*** (<12) Proportion of dependents 0.547 0.532 0.498 0.016 2.050** 0.033 6.193*** 0.049 6.682*** Caloric intake (def. varies 1445.0 2063.8 3070.9 –618.8 –22.039*** –1007.1 –26.332*** –1625.9 –38.491*** by year)** Proportion of children 0.293 0.328 0.359 –0.035 –1.934* –0.031 –2.358** –0.066 –3.795*** (6–18) in school Proportion of children 0.255 0.282 0.354 –0.027 –1.331 –0.072 –4.824*** –0.100 –4.954*** (6–12) in school Proportion of children 0.379 0.418 0.413 –0.039 –1.331 0.005 0.235 –0.035 –1.218 (13–18) in school Occupation of household 0.729 0.768 0.727 –0.039 –1.941* 0.041 3.062*** 0.002 0.090 head: agriculture Occupation of household 0.005 0.005 0.013 0.000 –0.031 –0.008 –4.213*** –0.008 –3.246*** head: manufacturing Occupation of household 0.001 0.004 0.002 –0.003 –2.812*** 0.002 1.565 –0.001 –2.490** head: construction Occupation of household 0.013 0.004 0.005 0.009 2.119** –0.001 –0.653 0.008 1.983** head: mining/energy Occupation of household 0.001 0.001 0.013 0.000 –0.501 –0.012 –5.876*** –0.012 –5.827*** head: social services Occupation of household 0.007 0.005 0.018 0.002 0.309 –0.012 –6.178*** –0.011 –1.953* head: prof. services Occupation of household 0.127 0.097 0.103 0.031 2.406** –0.006 –0.772 0.025 1.843* head: services & trade Household lives in an 0.128 0.109 0.150 0.019 1.481 –0.041 –4.771*** –0.021 –1.345 urban area Household has a private 0.091 0.093 0.140 –0.003 –0.169 –0.047 –4.689*** –0.050 –2.793*** toilet Household owns cattle 0.662 0.775 0.773 –0.114 –4.818*** 0.002 0.156 –0.112 –4.568*** (continued on next page) Annex 1 149 TABLE A1.3: Difference in means by percentile of consumption distribution (2000) (continued) Bottom 10% vs. Bottom 40%* vs. top Bottom 10% vs. top bottom 40%* 60% 60% Mean Mean bottom bottom Mean top Variable 10% 40% 60% Diff. t-stat Diff. t-stat Diff. t-stat Household owns sheep or 0.419 0.465 0.432 –0.046 –1.767* 0.033 1.757* –0.014 –0.531 goats Household owns chickens 0.470 0.538 0.528 –0.068 –2.385** 0.010 0.537 –0.058 –2.151** Household owns land 0.914 0.943 0.934 –0.029 –2.815*** 0.009 1.837* –0.020 –1.713* Household located 1–2km 0.127 0.122 0.144 0.005 0.177 –0.022 –1.576 –0.017 –0.619 to all weather road Household located >2km 0.618 0.630 0.578 –0.013 –0.370 0.052 2.447** 0.040 1.005 to all weather road Months covered by crop 0.124 0.193 0.307 –0.068 –3.128*** –0.114 –5.695*** –0.182 –6.897*** production for agr.hh: 10+ Months covered by crop 0.148 0.205 0.224 –0.057 –2.382** –0.019 –1.106 –0.077 –3.318*** production for agr.hh: 7 to 9 Months covered by crop 0.329 0.365 0.313 –0.036 –1.241 0.052 2.692*** 0.017 0.556 production for agr.hh: 4 to 6 Months covered by crop 0.398 0.237 0.156 0.162 5.200*** 0.081 4.041*** 0.242 7.317*** production for agr.hh: 0 to 3 Source: CSA Household Income and Consumption Expenditure Surveys 1996, 2000, 2005 and 2011. Notes: ** Caloric intake is measured differently across time, as such these measures are not comparable. Significance levels are defined as follows: * 10%, ** 5%, *** 1%. 150 ETHIOPIA – POVERTY ASSESSMENT TABLE A1.4: Difference in means by percentile of consumption distribution (2005) Mean Mean Bottom 10% vs. Bottom 40%* vs. top Bottom 10% vs. top bottom bottom Mean top bottom 40%* 60% 60% Variable 10% 40% 60% Diff. t-stat Diff. t-stat Diff. t-stat Age of household head 46.4 45.4 42.7 1.0 1.850* 2.7 7.422*** 3.7 6.932*** Household head is male 0.843 0.835 0.797 0.008 0.510 0.039 4.349*** 0.047 3.150*** Household head is 0.846 0.854 0.818 –0.008 –0.534 0.036 4.264*** 0.028 1.943* married Years of schooling of 1.097 1.441 2.124 –0.344 –2.903*** –0.683 –9.110*** –1.027 –8.327*** household head Number of household 7.537 6.665 5.293 0.872 6.984*** 1.372 20.293*** 2.244 17.411*** members Highest years of schooling 3.766 3.623 3.816 0.143 0.923 –0.193 –1.999** –0.050 –0.310 in household Proportion of unpaid 0.251 0.223 0.183 0.028 2.785*** 0.040 6.373*** 0.068 6.558*** workers Proportion of children 0.447 0.451 0.399 –0.004 –0.486 0.052 10.020*** 0.048 5.972*** (<12) Proportion of dependents 0.548 0.547 0.493 0.001 0.133 0.055 11.418*** 0.056 7.795*** Proportion of children 0.371 0.388 0.415 –0.017 –1.093 –0.027 –2.733*** –0.044 –2.743*** (6–18) in school Proportion of children 0.283 0.322 0.377 –0.039 –2.209** –0.055 –4.629*** –0.094 –5.278*** (6–12) in school Proportion of children 0.530 0.528 0.524 0.002 0.087 0.004 0.236 0.006 0.241 (13–18) in school Occupation of household 0.794 0.801 0.753 –0.008 –0.515 0.048 4.894*** 0.041 2.717*** head: agriculture Occupation of household 0.033 0.028 0.029 0.005 0.921 –0.001 –0.446 0.004 0.712 head: manufacturing Occupation of household 0.010 0.007 0.009 0.003 1.071 –0.002 –1.577 0.000 0.220 head: construction Occupation of household 0.002 0.003 0.003 –0.001 –0.649 –0.001 –1.118 –0.001 –1.635 head: mining/energy Occupation of household 0.021 0.027 0.035 –0.006 –1.454 –0.008 –2.116** –0.014 –3.475*** head: social services Occupation of household 0.014 0.013 0.025 0.001 0.135 –0.012 –4.934*** –0.011 –2.532** head: prof. services Occupation of household 0.041 0.041 0.063 0.001 0.096 –0.023 –5.460*** –0.022 –3.767*** head: services & trade Household lives in an 0.130 0.126 0.152 0.004 0.402 –0.026 –3.894*** –0.022 –1.918* urban area Household has a private 0.157 0.200 0.234 –0.043 –2.411** –0.034 –2.908*** –0.077 –4.114*** toilet Household owns cattle 0.628 0.683 0.660 –0.055 –2.361** 0.023 1.882* –0.032 –1.382 (continued on next page) Annex 1 151 TABLE A1.4: Difference in means by percentile of consumption distribution (2005) (continued) Mean Mean Bottom 10% vs. Bottom 40%* vs. top Bottom 10% vs. top bottom bottom Mean top bottom 40%* 60% 60% Variable 10% 40% 60% Diff. t-stat Diff. t-stat Diff. t-stat Household owns sheep or 0.551 0.546 0.482 0.004 0.179 0.065 4.641*** 0.069 2.758*** goats Household owns chickens 0.623 0.615 0.562 0.008 0.359 0.053 3.992*** 0.061 2.729*** Household owns land 0.924 0.927 0.903 –0.003 –0.425 0.024 4.774*** 0.021 2.596*** Household located 1–2km 0.129 0.118 0.107 0.011 0.673 0.011 1.311 0.022 1.295 to all weather road Household located >2km 0.625 0.632 0.609 –0.007 –0.281 0.023 1.415 0.015 0.512 to all weather road Household with a food 0.027 0.027 0.012 0.000 –0.027 0.015 2.903*** 0.015 1.925* gap of at least 9 months Household with a food 0.068 0.046 0.027 0.022 1.727* 0.019 3.177*** 0.041 3.206*** gap of 6–8 months Household with a food 0.256 0.176 0.123 0.080 3.472*** 0.053 5.033*** 0.134 5.861*** gap of 3–5 months Household with a food 0.649 0.751 0.838 –0.102 –3.857*** –0.088 –6.668*** –0.190 –7.303*** gap of less than 3 months Household shock: drought 0.150 0.106 0.083 0.044 2.324** 0.022 2.304** 0.067 3.223*** Household shock to prices 0.017 0.022 0.022 –0.005 –0.754 0.000 0.023 –0.005 –0.782 (rise or fall) Household shock: illness 0.264 0.271 0.270 –0.007 –0.303 0.002 0.115 –0.005 –0.240 or death of member Non-agricultural 0.102 0.096 0.129 0.007 0.719 –0.033 –5.533*** –0.027 –2.780*** household Months covered by crop 0.204 0.304 0.394 –0.101 –4.048*** –0.090 –5.413*** –0.191 –7.486*** production for agr. hh: 10+ Months covered by crop 0.229 0.259 0.257 –0.029 –1.302 0.001 0.080 –0.028 –1.269 production for agr. hh: 7 to 9 Months covered by crop 0.348 0.262 0.227 0.086 3.334*** 0.035 2.425** 0.121 4.706*** production for agr.hh: 4 to 6 Months covered by crop 0.219 0.175 0.121 0.044 1.568 0.054 4.225*** 0.098 3.241*** production for agr. hh: 0 to 3 Source: CSA Household Income and Consumption Expenditure Surveys 1996, 2000, 2005 and 2011. Notes: *Bottom 40% refers to those in the bottom 40% of the consumption distribution, without including the bottom 10%. Household shock to prices refers to any positive or negative shock to prices for (any and all) consumption goods, while the food price shock refers specifically to a rise in food prices. The food gap refers to the number of months during which the household faced a food shortage during the last 12 months. Significance levels are defined as follows: * 10%, ** 5%, *** 1%. 152 ETHIOPIA – POVERTY ASSESSMENT TABLE A1.5: Difference in means by percentile of consumption distribution (2011) Mean Mean Bottom 10% vs. Bottom 40%* vs. top Bottom 10% vs. top bottom bottom Mean top bottom 40%* 60% 60% Variable 10% 40% 60% Diff. t-stat Diff. t-stat Diff. t-stat Age of household head 47.1 45.7 43.0 1.4 2.402** 2.8 8.534*** 4.1 7.177*** Household head is male 0.843 0.828 0.815 0.015 1.044 0.013 1.463 0.028 2.099** Household head is mar- 0.864 0.854 0.821 0.010 0.855 0.033 4.243*** 0.043 3.555*** ried Years of schooling of 1.241 1.647 2.778 –0.406 –3.823*** –1.131 –13.311*** –1.537 –13.227*** household head Number of household 7.265 6.479 5.361 0.786 8.050*** 1.118 17.068*** 1.904 20.136*** members Highest years of schooling 4.482 4.782 5.252 –0.300 –2.045** –0.470 –4.656*** –0.770 –4.799*** in household Proportion of unpaid 0.215 0.201 0.167 0.014 1.168 0.035 5.599*** 0.048 3.922*** workers Proportion of children 0.448 0.438 0.390 0.010 1.110 0.048 8.853*** 0.058 6.126*** (<12) Proportion of dependents 0.547 0.533 0.486 0.014 1.744* 0.046 9.563*** 0.061 7.290*** Proportion of children 0.524 0.592 0.587 –0.067 –3.850*** 0.005 0.435 –0.062 –3.241*** (6–18) in school Proportion of children 0.475 0.556 0.565 –0.081 –3.667*** –0.009 –0.638 –0.091 –3.839*** (6–12) in school Proportion of children 0.610 0.676 0.664 –0.066 –2.793*** 0.012 0.819 –0.054 –2.252** (13–18) in school Occupation of household 0.824 0.804 0.742 0.020 1.473 0.062 6.440*** 0.082 5.435*** head: agriculture Occupation of household 0.021 0.021 0.022 0.000 –0.059 0.000 –0.172 –0.001 –0.164 head: manufacturing Occupation of household 0.008 0.016 0.015 –0.008 –3.270*** 0.000 0.183 –0.007 –3.581*** head: construction Occupation of household 0.003 0.002 0.005 0.000 0.081 –0.003 –1.857* –0.002 –1.244 head: mining/energy Occupation of household 0.026 0.027 0.042 –0.001 –0.237 –0.016 –5.076*** –0.017 –3.874*** head: social services Occupation of household 0.007 0.013 0.031 –0.006 –3.125*** –0.018 –7.667*** –0.024 –10.558*** head: prof. services Occupation of household 0.050 0.050 0.081 0.001 0.121 –0.032 –6.840*** –0.031 –4.026*** head: services & trade Household lives in an 0.146 0.138 0.183 0.008 0.766 –0.045 –6.032*** –0.037 –2.879*** urban area Floors in households of 0.000 0.000 0.001 0.000 –0.218 0.000 –0.766 –0.001 –0.816 hard/solid material Household has a private 0.552 0.529 0.532 0.023 0.916 –0.002 –0.151 0.021 0.789 toilet (continued on next page) Annex 1 153 TABLE A1.5: Difference in means by percentile of consumption distribution (2011) (continued) Mean Mean Bottom 10% vs. Bottom 40%* vs. top Bottom 10% vs. top bottom bottom Mean top bottom 40%* 60% 60% Variable 10% 40% 60% Diff. t-stat Diff. t-stat Diff. t-stat Household owns livestock 0.840 0.862 0.806 –0.022 –1.569 0.056 7.141*** 0.034 2.294** Household owns cattle 0.643 0.683 0.650 –0.040 –1.790* 0.033 2.623*** –0.007 –0.315 Household owns sheep or 0.562 0.544 0.485 0.018 0.741 0.059 3.996*** 0.077 3.163*** goats Household owns chickens 0.540 0.562 0.541 –0.022 –0.997 0.021 1.436 –0.001 –0.029 Household owns beehives 0.152 0.146 0.135 0.006 0.323 0.011 0.975 0.017 0.859 Household owns land 0.935 0.935 0.897 0.000 0.024 0.038 8.612*** 0.038 5.217*** Household between 0.143 0.137 0.140 0.006 0.342 –0.003 –0.276 0.003 0.151 1–2km to all weather road Household more than 0.678 0.653 0.575 0.025 1.012 0.077 4.779*** 0.102 3.711*** 2km to all weather road Household with a food 0.014 0.008 0.004 0.006 1.070 0.003 1.771* 0.009 1.583 gap of at least 9 months Household with a food 0.044 0.023 0.016 0.021 1.971** 0.007 1.842* 0.028 2.577** gap of 6–8 months Household with a food 0.142 0.110 0.085 0.032 1.767* 0.025 2.866*** 0.057 3.209*** gap of 3–5 months Household with a food 0.801 0.859 0.895 –0.058 –2.752*** –0.036 –3.638*** –0.094 –4.242*** gap of less than 3 months Household shock: drought 0.061 0.044 0.044 0.017 1.236 0.001 0.121 0.018 1.176 Household shock to prices 0.229 0.178 0.189 0.051 2.173** –0.011 –0.853 0.040 1.626 (rise or fall) Household shock: illness 0.091 0.096 0.089 –0.004 –0.309 0.007 0.777 0.003 0.195 or death of member Non-agricultural house- 0.119 0.122 0.174 –0.003 –0.283 –0.052 –7.507*** –0.055 –4.829*** hold Months covered by crop 0.464 0.515 0.593 –0.051 –1.750* –0.078 –4.548*** –0.129 –4.168*** production for agr. hh: 10+ Months covered by crop 0.228 0.214 0.190 0.014 0.576 0.024 1.888* 0.038 1.553 production for agr. hh: 7 to 9 Months covered by crop 0.188 0.182 0.147 0.006 0.301 0.035 2.830*** 0.041 2.186** production for agr. hh: 4 to 6 Months covered by crop 0.119 0.088 0.070 0.031 1.994** 0.018 2.145** 0.049 3.084*** production for agr. hh: 0 to 3 Source: CSA Household Income and Consumption Expenditure Surveys 1996, 2000, 2005 and 2011. Notes: *Bottom 40% refers to those in the bottom 40% of the consumption distribution, without including the bottom 10%. Household shock to prices refers to any positive or negative shock to prices for (any and all) consumption goods, while the food price shock refers specifically to a rise in food prices. The food gap refers to the number of months during which the household faced a food shortage during the last 12 months. Significance levels are defined as follows: * 10%, ** 5%, *** 1%. 155 ANNEX 2 TABLE A2.1: Deprivation Indicators Urban/ Atkinson & Lugo OPHI MPI MDG indicators Ethiopia WMS-HCES Rural Indicator (2010) (2013) (2008) 2000, 2005, 2011 2000 2005 2011 Indicator Education of school deprived: any school- net enrollment at least one child (age    U, R School-aged household has aged child is ratio in primary 7–15) in the household Children at least one child not attend- education; propor- is not currently attending 5–16 years old ing school in tion of pupils start- school who is not in years 1 to 8 ing grade 1 who 2000, 2005: currently school reach last grade of registered in school primary school Education “ “ “ at least one girl child    U, R of Female (age 7–15) in the house- School-aged hold is not currently Children attending school 2000, 2005: currently registered in school Health household was dissatis-   U, R Facility fied with at least one Quality health facility visit, or did not use a health facility due to cost, distance, quality, or other reasons Health household is located    R Facility more than 5 km away Access from the nearest health facility (clinic, health station, hospital, health post) 2000: health posts did not exist. Institutional antenatal care cov- at least one child (age   U Birth erage; proportion 0–4) in the household of births attended was not born in a health by skilled health facility personnel Female at least one girl (age  U, R Circumcision 0–14) in the household underwent/will undergo female circumcision (continued on next page) 156 ETHIOPIA – POVERTY ASSESSMENT TABLE A2.1: Deprivation Indicators (continued) Urban/ Atkinson & Lugo OPHI MPI MDG indicators Ethiopia WMS-HCES Rural Indicator (2010) (2013) (2008) 2000, 2005, 2011 2000 2005 2011 Indicator Assets asset deprived: household mobile-cellular/ household does not own    U, R household does does not own fixed telephone a motorcycle, car, or not own a car, and a car or truck, subscriptions per bajaj, and does not own owns fewer than and does not 100 inhabitants a fridge, phone, radio, one small asset- own more TV, bicycle, or jewelry -TV, radio, phone, than one of 2005: motorcycle, bajaj bicycle, refrigera- the following not in list of assets tor, motorcycle assets: radio, 2000: phone, jewelry television, not in list of assets telephone, bi- cycle, scooter, or refrigerator Source of mobile-cellular/ household does not own    U, R Information fixed telephone a TV, radio, or phone subscriptions per 2000: phone is not 100 inhabitants specified in list of assets Drinking water deprived: household proportion of household does not    R Water household does does not have population using use a safe drinking not have access access to safe an improved drink- water source defined as to piped or other drinking water ing water source piped water, a protected protected source defined as source, or rainwater of drinking water piped water, public tap, borehole or pump, protected well, protected spring or rainwater, and it is within a distance of 30 minutes’ walk roundtrip Sanitation household’s proportion of household does not use    U, R sanitation population using an improved toilet facil- facility is not an improved sani- ity defined as a private improved tation facility flush toilet or private pit (according to latrine MDG guide- lines), or it is improved but shared with other house- holds. (continued on next page) Annex 2 157 TABLE A2.1: Deprivation Indicators Urban/ Atkinson & Lugo OPHI MPI MDG indicators Ethiopia WMS-HCES Rural Indicator (2010) (2013) (2008) 2000, 2005, 2011 2000 2005 2011 Indicator Living household believes that    U, R Standards its overall living stan- Perception dard is worse/worst now compared to 12 months ago 2005: is much worse/ worse now 2000: has decreased Below proportion of household lives below    U, R Poverty Line population below the poverty line of 3781 US$1 (PPP) per day Birr per adult equivalent (or below country- (using real total con- level poverty line) sumption expenditure per adult) 2000, 2005: below the poverty line of 1075 Birr (in 1996 prices) 158 ETHIOPIA – POVERTY ASSESSMENT TABLE A2.2: Deprivation proportions by Venn diagram region in Figure 2.2: urban and rural populations Urban Rural 2000 2011 Change 2000 2011 Change Money poor 0.41 0.31 –0.09*** 0.50 0.34 –0.16*** Education deprived 0.26 0.16 –0.10*** 0.83 0.58 –0.25*** Sanitation deprived 0.51 0.47 –0.04 0.93 0.42 –0.50*** Not deprived 0.28 0.33 0.05*** 0.01 0.18 0.16*** Only money poor 0.11 0.12 0.01 0.01 0.08 0.07*** Only education deprived 0.07 0.05 –0.01 0.03 0.21 0.18*** Only sanitation deprived 0.19 0.25 0.07*** 0.08 0.12 0.03*** Money poor, education deprived 0.04 0.03 –0.01* 0.02 0.11 0.09*** Education, sanitation deprived 0.06 0.05 –0.02 0.37 0.16 –0.21*** Sanitation deprived, money poor 0.17 0.14 –0.03* 0.06 0.05 –0.02* All three deprivations 0.09 0.03 –0.06*** 0.41 0.10 –0.31*** Source: own calculations using HICES 2000 and HCES 2011. Note: The “Change” column the difference in proportions from 2000 to 2011. The asterisks indicate the significance level: *** p<0.01, ** p<0.05, * p<0.1. TABLE A2.3: Deprivation proportions by Venn diagram region in Figure 2.4: rural population 2000 2011 Change Money poor 0.50 0.34 –0.16*** Worsened living standards perception 0.39 0.51 0.12*** Education deprived 0.83 0.58 –0.25*** Not deprived 0.06 0.15 0.09*** Only money poor 0.04 0.06 0.02** Only worsened living standards perception 0.03 0.14 0.11*** Only education deprived 0.26 0.18 –0.08*** Money poor, perception deprived 0.03 0.07 0.04*** Perception, education deprived 0.14 0.18 0.04** Education deprived, money poor 0.24 0.10 –0.14*** All three deprivations 0.19 0.11 –0.07*** Source: own calculations using HICES 2000 and HCES 2011. Note: The “Change” column shows the difference in proportions across years. The asterisks indicate the significance level: *** p<0.01, ** p<0.05, * p<0.1. Annex 2 159 TABLE A2.4: Deprivation proportions by Venn diagram region in Figure 2.4: urban population 2000 2011 Change money poor 0.41 0.32 –0.09*** worsened living standards perception 0.33 0.55 0.22*** education deprived 0.26 0.16 –0.10*** not deprived 0.33 0.28 –0.04** only money poor 0.18 0.10 –0.08*** only worsened living standards perception 0.14 0.30 0.16*** only education deprived 0.09 0.04 –0.04*** money poor, perception deprived 0.10 0.16 0.06*** perception, education deprived 0.04 0.06 0.01 education deprived, money poor 0.08 0.02 –0.06*** all three deprivations 0.05 0.04 –0.01 Source: own calculations using HICES 2000 and HCES 2011. Note: The “Change” column shows the difference in proportions across years. The asterisks indicate the significance level: *** p<0.01, ** p<0.05, * p<0.1. TABLE A2.5: Deprivation proportions by Venn diagram region in Figure 2.5: urban and rural populations Urban Rural Difference 2011 2011 (Rural-Urban) money poor 0.32 0.36 0.04* female circumcision deprived 0.24 0.36 0.12*** girls’ education deprived 0.14 0.46 0.32*** not deprived 0.47 0.22 –0.25*** only money poor 0.19 0.11 –0.09*** only female circumcision deprived 0.12 0.14 0.02* only girls’ education deprived 0.07 0.20 0.12*** money poor, female circumcision deprived 0.08 0.07 0.00 female circumcision, girls’ education deprived 0.02 0.09 0.06*** girls’ education deprived, money poor 0.03 0.12 0.09*** all three deprivations 0.02 0.06 0.04*** Source: own calculations using HCES 2011. Note: The “difference” column shows the difference in proportions. The asterisks indicate the significance level: *** p<0.01, ** p<0.05, * p<0.1. 161 ANNEX 3 TABLE A3.1: Full results of regression estimation of log of consumption per adult equivalent (pooled across 2005 and 2011) Coefficient (Standard error) P-value Death of household member For rural household with 2 or more plots –0.015 (0.02) 0.478 For rural household with none or 1 plot 0.016 (0.03) 0.577 For urban educated household –0.095 (0.03) 0.003 For urban uneducated household with female head –0.054 (0.04) 0.223 For urban uneducated household with male head –0.077 (0.04) 0.079 Loss of job For rural household –0.110 (0.07) 0.141 For urban educated household 0.014 (0.04) 0.744 For urban uneducated household with female head 0.177 (0.12) 0.15 For urban uneducated household with male head 0.231 (0.18) 0.206 High food prices For rural household –0.027 (0.02) 0.143 For urban educated household 0.003 (0.02) 0.91 For urban uneducated household with female head –0.119 (0.04) 0.002 For urban uneducated household with male head –0.141 (0.05) 0.002 Rainfall induced crop-loss For rural household in non-drought prone area –0.001 (0.00) 0.539 For rural household in drought prone area with 2 or more plots –0.003 (0.00) 0.005 For rural household in drought prone area with PSNP –0.002 (0.00) 0.146 For rural household in drought prone area with no PSNP –0.003 (0.00) 0.005 Drought prone 0.126 (0.04) 0.004 PSNP beneficiary –0.057 (0.02) 0.011 Own 2 or more plots 0.052 (0.01) 0 Can raise cash in time of need 0.128 (0.01) 0 Uneducated household with female head –0.048 (0.04) 0.199 Uneducated household with male head –0.028 (0.04) 0.43 Age of household head –0.002 (0.00) 0 Female household head –0.044 (0.01) 0.003 (continued on next page) 162 ETHIOPIA – POVERTY ASSESSMENT TABLE A3.1: Full results of regression estimation of log of consumption per adult equivalent (pooled across 2005 and 2011) (continued) Coefficient (Standard error) P-value Head has primary education 0.007 (0.04) 0.846 Head has secondary education 0.152 (0.04) 0 Head has tertiary education 0.020 (0.04) 0.616 Proportion of household members in agriculture 0.062 (0.02) 0.002 Proportion of household members in manufacturing 0.082 (0.03) 0.018 Proportion of household members in construction 0.153 (0.04) 0 Proportion of household members in mineral extraction 0.311 (0.10) 0.001 Proportion of household members in education 0.200 (0.03) 0 Proportion of household members in professional services 0.344 (0.04) 0 Proportion of household members in services 0.316 (0.02) 0 Urban unemployed 0.077 (0.04) 0.046 Rural 0.442 (0.05) 0 Rural unemployed 0.447 (0.15) 0.003 Log of household size –0.419 (0.03) 0 Proportion of household members female between 16 and 64 0.214 (0.05) 0 Proportion of household members female 15 and under 0.134 (0.05) 0.009 Proportion of household members female 65 and over 0.167 (0.09) 0.075 Proportion of household members male 15 and under 0.085 (0.04) 0.038 Proportion of household members male 65 and over 0.170 (0.09) 0.046 Dependency ratio 0.024 (0.03) 0.364 Highest education grade in household 0.000 (0.00) 0.969 Highest education grade in household, male 0.006 (0.00) 0.006 Highest education grade in household, female 0.011 (0.00) 0 Log of distance to market 0.009 (0.01) 0.463 Good market access –0.011 (0.02) 0.487 Distance to town of 50,000 plus 0.000 (0.00) 0.628 Frequency of crop loss greater than 50 percent –0.008 (0.01) 0.356 Urban land –0.056 (0.02) 0.003 Rural no land –0.066 (0.02) 0.008 Urban improved toilet 0.119 (0.02) 0 Rural no improved toilet –0.063 (0.02) 0.001 Urban good roof 0.118 (0.02) 0 Rural no good roof –0.107 (0.01) 0 Urban own toilet 0.051 (0.01) 0 Rural shared toilet 0.020 (0.02) 0.3 (continued on next page) Annex 3 163 TABLE A3.1: Full results of regression estimation of log of consumption per adult equivalent (pooled across 2005 and 2011) (continued) Coefficient (Standard error) P-value Urban electricity –0.119 (0.02) 0 Rural no electricity –0.091 (0.03) 0.008 Urban renter 0.113 (0.02) 0 Square of log of household size –0.009 (0.01) 0.357 Square of proportion of household members female between 16 –0.116 (0.05) 0.01 and 64 Square of proportion of household members female 15 and –0.027 (0.08) 0.724 under Square of proportion of household members female 65 and over 0.001 (0.10) 0.989 Square of proportion of household members male 15 and under –0.059 (0.04) 0.142 Square of proportion of household members male 65 and over –0.112 (0.11) 0.286 Square of log of distance to market –0.004 (0.00) 0.056 Square of log of distance to town of 50,000 plus 0.000 (0.00) 0.11 2011 –0.027 (0.02) 0.099 Constant 7.616 (0.08) 0 Source: Regression results using HICES 2005, HICES 2011, WMS 2005, WMS 2011 and LEAP. 165 ANNEX 4 This annex describes the method employed and the ∆ ln pzt = b0 + bY a s zt a -1 ∆ lnY zta + data used to estimate the relationship between sec- bY m s zt m -1 ∆ lnY ztm + bY r s zt r -1 ∆ lnY ztr toral growth and public goods provision and poverty reduction. The analysis starts by abstracting from the +b N ∆ ln N zt + b E ∆ ln E zt + sectoral pattern of output growth and examining b D ∆ ln Dzt + uz + e zt whether changes in poverty rates have been driven (2) by aggregate output growth in the zone. In addition the analysis examines whether public good provi- Where Y zti , i = a , m , r is the output of agriculture sion—specifically the introduction of safety nets, (a), manufacturing (m) and services (r) a respectively i ∆ ln pzt = ∆bln +pztb = sb0-+ ∆b ln Y aa a s zt -+ ∆ lnY zta + investments in primary education and roads—has had and s zt -1 is the share of output of 0sector Y ai zt at the 1 Ybegin- 1 an additional effect on poverty reduction (in addition ning of the period. In bYlater m m s zt -1 ∆bln Y smm specifications Y m zt -1 + ∆b ln Y s rm,+ Y r zt -1 ∆b and ln Y s rr ∆ lnY ztr Y r zt -1 to any effect that has resulted from their impact on are proxied with growth +in bN the share ∆ ln+N bzt of + ∆the blnEN population ∆ ln + Eb + ∆ ln E zt + N zt zt E growth) via redistribution. Specifically the following living in urban areas in the zone. Interacting the rate b D ∆ ln D b + ∆u ln + Dzt + uz + e zt e zt regression is estimated: of growth of sector i with the share of ztDsector z i in total output allows growth in a given sector to influence ∆ ln pzt = b0 + bY ∆ lnY zt + bN ∆ ln N zt + (1) poverty according to the size of the sector. The com- bined expression, bY i s zt i , provides a measure of the bE ∆ ln E zt + bD ∆ ln Dzt + uz + e zt -1 elasticity of poverty to growth in that sector. Where ∆ ln pzt = is b the 0 bY ∆ lnYrate +poverty zt +bin ∆ ln N zt +at time t, N the zone This specification controls for a number of other pzt∆=lnbp += bYb∆ +lnbYYzt ∆+ islnzonal bY bE∆ ∆ +ln ln output, bN E∆ ln + +N bDzt∆ + islnthe Dztproportion + uz + e zt of people in factors that might confound the relationship between 0 zt 0 N zt N ztzt ∆ ln pzt = b the+ b zone ∆ ln Y covered + bby ∆ ln N N the safety + net program at time t, sectoral composition and poverty rates. The regres- ∆ lnbE ∆ lnbE + ∆+ln0 D b ∆ + Y lnu D + + zt uz + e zt e zt zt E ∆ ln Ezt pzt bE ∆= ln b D zt E0zt+ D zt + bY is b ∆∆ ln z zt increasedlnYD ++ bNu∆ zt access +ln toN e primary zt + schools in the zone sion is estimated in differences to control for any D zt z zt bE ∆ lnat time E zt + bD ∆ ln Dzt is t and +a + umeasuree of infrastructure invest- initial zonal characteristics that affect the relation- z zt ments reducing remoteness in the zone. ship between the output of one sector and poverty.41 Next, the relationship between the nature of Zone-specific time trends are included in the model, sectoral output growth and poverty reduction is uz , through the inclusion of zone-specific fixed effects, examined by decomposing zonal output growth which allows each zone to have a zonal specific trend into that coming from agricultural growth and that in poverty reduction over the period. The inclusion coming from manufacturing and services. Following of measures of public goods provision also allows us Ravallion and Datt (1996) and the subsequent litera- ture on the relationship between the composition of 41 Annualized growth rates are calculated for each variable by dividing each growth rate by the number of years over which the growth occurred growth and poverty reduction the following regres- (4 years for differences from 1996 to 2000, five years for differences from sion is estimated: 2000 to 2005, and 6 years for differences from 2005 to 2011). 166 ETHIOPIA – POVERTY ASSESSMENT to control for a number of time-variant characteristics zones in the Gambela region were not included in the that may be important in determining the relation- analysis as poverty data is not available for 1996 or ship between the pattern of growth and poverty. The 2005 for this region. The following paragraphs detail inclusion of N, D and E also controls for a number the construction of the zonal panel. Table A4.1 pres- of time-variant characteristics that may be important ents summary statistics. in determining poverty and which may affect the estimation of bY i . Poverty estimates To address concerns about reverse causality, growth in agriculture is instrumented with weather. The description of the sampling for the HICES indi- Even with a fully specified model, the estimation cates that enumeration areas are stratified by zone, strategy outlined is subject to a concern that reverse with a similar number of EAs selected by zone in each causation may be driving the results. In some papers year. Zonal level poverty estimates are reported for on the relationship between sectoral growth and pov- 1996, 2000, and 2005 in MOFED (2013). However, erty this goes unaddressed, and in other papers it is these zonal estimates are not often cited, and similar addressed by instrumenting growth rates with growth estimates were not presented for 2011. Although the rates of neighbors (Ligon and Sadoulet 2008, Loayza sample is stratified by zone, the sampling strategy used and Raddatz 2010) or lagged growth (Loayza and in the HICES is not designed to sample enough house- Raddatz 2010). Henderson et al. (2011) has explored holds to generate precise zonal level poverty estimates. the use of rainfall as a measure of exogenous variation Poverty mapping can be used to generate small- in agricultural growth and the same approach is taken area estimates of poverty (Elbers et al. 2003), and for here using WRSI data available at the zonal level in the type of analysis conducted in this paper, poverty Ethiopia from 1996–2011. Weather shocks (calculated maps estimated at multiple points in time could pro- as the sum of annual estimates of crop loss for the zone vide the required data. Although no official poverty through a crop WRSI model) are used as an estimate map estimates exist for Ethiopia, the 2007 census has of exogenous variation in agricultural growth. Ethiopia been used with the HICES 2011 data to generate is characterized by both significant weather risk and zonal and woreda level poverty estimates (Sohnesen significant heterogeneity in weather risk across space 2014). The poverty mapping report presents the cor- and time. It is likely that agricultural output is the relation between poverty-map estimates at the zonal main mechanism by which local weather shocks affect level with estimates calculated directly from the data. local livelihoods, and that increased market integration As indicated in Figure A2.1, which is taken from this throughout this period limits the impact of small local report, although survey based estimates of poverty weather shocks on prices and growth in other sectors. rates are perhaps noisier than those estimated using This is something that is tested empirically and found poverty mapping techniques they compare well. Until to hold true (see Hill and Tsehaye 2014 for full details). poverty-mapping estimates are available across time Fifty zones are followed over a period of 15 years, in Ethiopia, Figure A2.1 suggests that the zonal esti- covering nearly all of Ethiopia’s population. Zonal mates can be used with some confidence for regression boundaries from 1996 are used and all aggregates are analysis. If measurement error in poverty estimates can calculated using these zonal boundaries. Three pastoral be considered white noise, it will not affect coefficient zones in the Somali region were excluded from this estimates given poverty is the dependent variable. analysis because no poverty data is available for them In addition to using the HICES to estimate (three Somali zones are included). Afar’s five zones poverty for each zone, it is also used to estimate the were excluded from the analysis because of missing number of people in each zone by aggregating the agricultural data in some years. In addition, the three weights at the zonal level, and the number of people Annex 4 167 TABLE A4.1: Zonal averages of key variables Data source 1996 2000 2005 2011 Poverty Poverty headcount rate (%) HICES 47.7 47.4 40.3 28.1 (18.9) (15.3) (11.6) (13.7) Poverty gap (%) HICES 13.89 13.64 8.65 7.44 (8.43) (7.69) (4.01) (4.58) Poverty severity (%) HICES 5.57 5.42 2.79 2.92 (4.37) (4.18) (1.74) (2.13) Output by sector Agricultural output per capita (Birr p.c.) AgSS, HICES 162.9 155.0 190.5 275.8 (93.3) (98.4) (134.7) (191.6) Manufacturing output per capita (Birr p.c.) LMSMS, HICES 52.2 78.3 99.3 159.9 (173.9) (232.7) (296.8) (444.0) Trade services output per capita (Birr p.c.) DTSS, HICES 126.1 198.2 165.1 216.6 (128.1) (275.6) (139.7) (115.7) Cereal output per capita (Birr p.c.) AgSS, HICES 136.6 124.6 139.4 193.1 (88.2) (87.4) (115.7) (152.2) Cash crop output per capita (Birr p.c.) AgSS, HICES 15.3 18.1 38.7 62.6 (18.5) (23.6) (62.1) (83.5) Proportion of output coming from: Agriculture 0.60 0.49 0.52 0.55 (0.30) (0.27) (0.26) (0.26) Manufacturing 0.07 0.09 0.09 0.09 (0.14) (0.16) (0.17) (0.19) Services 0.34 0.42 0.39 0.36 (0.24) (0.24) (0.23) (0.20) Safety nets, basic services, and infrastructure Proportion of households in the PSNP (%) PSNP data 0 0 0 8.3 (11.4) Distance to the nearest primary school (km) WMS 4.77 4.11 (1.55) 4.14 2.74 (2.27) (2.36) (0.86) Distance to bus or taxi service (km) WMS 20.9 20.5 17.5 13.6 (17.8) (11.9) (10.7) (8.8) Distance to town of 50,000 or more (minutes) Schmidt and 566 (397) 486.9 (335.7) 408.0 317.4 Kedir (2009) (279.3) (217.5) Agricultural variables Predicted crop loss due to rainfall (%) LEAP 11.4 22.4 26.6 15.7 (13.5) (18.8) (23.1) (16.2) Land planted to improved seeds (%) AgSS 0.5 1.4 2.3 4.1 (0.8) (1.5) (2.2) (4.6) Land on which fertilizer is applied (%) AgSS 15.3 9.6 16.7 27.6 (21.2) (10.6) (16.5) (22.3) Weighted index of crop prices (Birr per kg) AgSS 1.12 0.86 (0.21) 1.03 1.26 (0.25) (0.36) (0.37) Note: Standard deviation in brackets. All Birr values are in 1996 prices. p.c. stands for per capita. 168 ETHIOPIA – POVERTY ASSESSMENT FIGURE A4.1: Scatter of estimated and estimates of land cultivated to scale the area cultivated measured level of poverty by zone in 2000. A survey of producer price data is collected 0.8 to complement the annual agricultural sample survey. 0.7 Producer prices are collected throughout the year. 0.6 Data from January of each year are used, as this is the 0.5 main harvest month. Producer price data is combined Measured 0.4 with production data to estimate the value of agricul- 0.3 tural output in each zone. From this the growth rate 0.2 of agricultural output per capita is derived. 0.1 The AgSS data was also used to provide estimates 0 for the proportion of land planted to fertilizer and 0 0.2 0.4 0.6 0.8 the proportion of land planted to improved seeds. Estimated The price data from the AgSS was used to construct Tigray Afar Amhara Oromia Somali Benishangui Gumuz SNNP a weighted crop price index in which all crop prices Source: Sohnesen 2014.Bubbles indicate the number of HICES obser- were weighted by the share of land planted to that vations in each zone. crop in the zone. Changes in the price index reflect both changes in prices for a given crop and also shifts into higher or lower valued crops. in the zone employed in different aspects of services to predict zonal service sector output. Manufacturing output Agricultural output A census of large and medium sized manufacturing establishments is conducted every year. An establish- Annual zonal estimates of agricultural production ment is considered eligible for this survey if it has more are estimated using the Central Statistical Agency than 10 employees and uses electricity. The survey Agricultural Sample Survey (AgSS). The AgSS collects collects information on output, assets, operating costs data on average landholding, area cultivated, total and employment. The town of each establishment is production, yield, and use of fertilizer and improved recorded and in some cases the zone. By matching seeds during the main Meher season. This data is towns to zones, zonal manufacturing output can be available for 1996–2011 with the exception of the estimated. year during which the agricultural census took place These estimates do not include manufacturing (2002) and the year following the census in which the output of smaller firms. Nationally, this is a small full AgSS could not be conducted. The Meher season proportion of manufacturing output. Soderbom is responsible for about 80% of crop production in (2012) compares micro-manufacturing firms in Ethiopia, but for some zones the smaller Belg season Ethiopia with larger firms included in the annual cen- is an extremely important part of agricultural produc- sus and shows that the value added of larger manufac- tion. A Belg crop survey is also undertaken each year, turing firms is eight times that of firms with less than but the production estimates are not representative at 10 employees. Focusing only on the larger firms for the zonal level. Zonal Belg output is estimated using an estimate of manufacturing output thus captures a zonal estimates of land cultivated to each crop, and large share of the manufacturing output in Ethiopia. regional estimates of average yield for each crop each However, it may be the case that the smaller manu- year. For years prior to 2000 no zonal level land esti- facturing firms matter more for poverty reduction. mates are available and so we use trends in national The regression estimation strategy used here allows Annex 4 169 for the share of manufacturing output produced by measured using the Schmidt and Kedir (2009) esti- small firms to vary across zones and to change with mates of time to travel (using type of road and distance time, but it relies on the growth rate manufacturing to generate the estimates) to a town of 50,000 people output of small firms to be constant within a given in 1994 and in 2007. The distance at each square kilo- zone across the full period 1996–2011. meter in the zone is averaged across the zone to prove a zonal average estimate. Finally administrative data Services on the number of PSNP beneficiaries per zone per year is used to estimate the proportion of households The most systematic survey of service industries is a in the zone benefiting from the PSNP. survey of trade and distributive services that was con- ducted in 1995, 2002 and 2009. This survey allows Weather shocks regional estimates of productivity of service enterprises to be generated, but it is a sample survey and does not The Livelihoods, Early Assessment and Protection allow for an estimate of the zonal service output. It project (LEAP) system, developed in 2008 by the also does not include information on personal services Government of Ethiopia in collaboration with WFP, such as hotel, restaurant, and domestic help. In order uses crop-modeling approaches to estimate rainfall- to generate a zonal estimate of service output data on induced crop loss in woredas throughout Ethiopia. the number of individuals engaged in trade and dis- Water-balance crop models and yield reduction coef- tributive services in the zone from the HICE surveys ficients are defined for the crops grown in the zone. is estimated and multiplied with national estimates of Evapotranspiration coefficients for the zone are used value added per worker to generate a measure of zonal with data on decadal rainfall in a given year to gen- output per worker from these surveys. The value of erate an estimate of the proportion of crop that was hotel and restaurants are however not captured in this lost in a given year as a result of insufficient rainfall. measure of services output per capita. These models essentially provide a weighted average of rainfall in which rainfall at times of the year in which Public goods provision: data on safety the development of the crop is particularly moisture nets and access to basic services dependent is given a higher weight. The weights are provided by agronomic crop models. Crop loss esti- The average distance to a primary school recorded mates are generated for each 50 km by 50 km square. at four points in time in the Welfare Monitoring This is aggregated to generate a zonal estimate of crop- Surveys (WMS) was used as a proxy for education and loss. The LEAP database contains crop loss estimates health investments. Secondly investments in roads are from 1996 to 2012 for both Belg and Meher seasons. 171 ANNEX 5 The incidence of direct and indirect taxes and subsidies personal income tax, agricultural income tax, can be calculated using either income or consumption rental income tax, and rural land use fees. This (e.g., Abramovsky, Attanasio, and Phillips, 2011). In measure of market income is meant to capture the this incidence analysis, consumption was used as the value of Wages and salaries, income from capital, basic welfare indicator because the household survey private transfers; before government taxes, social available for our analysis collects consumption expen- security contributions and transfers and contribu- diture but no income data. Moreover, in the Ethiopian tory pensions. case (as in many low income countries), consumption  Moving down Figure A5.1 from disposable is more accurately measured relative to income, and income to post-fiscal income, indirect taxes were forms the basis for poverty measures. subtracted and subsidies added. In the Ethiopian The Commitment to Equity (CEQ) method of case, the indirect taxes are VAT, sales tax, import benefit incidence analysis developed by Lustig and duties, excise tax, stump duties, and SUR tax Higgins (2012) was used. This method is applied on while the indirect subsidies are wheat, kerosene several countries in Latin America, Asia, and Africa. It and electricity subsidies. analyzes the distributional impact of fiscal policy in a  Post fiscal income plus in-kind transfers (educa- holistic and standardized way, facilitating comparison tion and health) minus co-payment and user fees with other countries in which the CEQ methodology gives final income. has been applied. The CEQ methodology defines dif- ferent income concepts of pre and post transfers and The following subsections detail the data sources taxes so that the distributional impacts of transfers and used and the assumptions made in order to estimate taxes can be easily identified. These different income taxes paid and transfers and subsidies received. concepts are summarized in Figure A5.1. These five income concepts are defined for Data sources Ethiopia as follows. The 2010/11 Household Consumption Expenditure  Disposable income was set equal to household Survey (HCES) and Welfare Monitoring Survey consumption expenditure. (WMS) data sources collected by the Central  Moving up Figure A5.1 from disposable income, Statistical Agency (CSA) of Ethiopia were used. net market income was derived by subtracting Household survey data are combined with data direct transfers received by the household from from National Income Accounts and public finance disposable income. Direct transfers include trans- accounts from the Ministry of Finance and Economic fers from the Productive Safety Net Program and Development. These accounts are used in order to food aid. obtain the public revenue and expenditures corre-  To get market income, direct taxes and contribu- sponding to the 2010/11 Ethiopian fiscal year. This tory pensions were added to net market income. information is complemented with data from the In the case of Ethiopia, direct taxes include Productive Safety Net and Household Asset Building 172 ETHIOPIA – POVERTY ASSESSMENT FIGURE A5.1: Definitions of income concepts in CEQ methodology Market Income = Im Transfers Wages and salaries, income from capital, private transfers; before government taxes, Taxes social security contributions and transfers. _ Personal income taxes and employee contributions to social security Net Market Income = In + Direct transfers Disposable Income = Id + Indirect subsidies _ Indirect taxes Post-fiscal Income = Ipf + _ In-kind transfers (free or subsidized government Co-payments, user fees services in education and health) Final Income = If Source: Lustig and Higgins (2013). Programs Annual Work Plan for 2010/11; Ministry the informal sector, we applied the business tax sched- of Trade, the Ethiopian Electric Power Corporation ule as detailed in Table A5.1B to determine personal Report on Accountability Issues (2013); and Ministry income tax. Tax evasions (calculated as the difference of Health. Finally, the 2005 Social Accounting Matrix between total actual tax collected and tax estimated (SAM) produced by the Ethiopian Development based on income) is assumed to be borne by all self- Research Institute (EDRI) were used to estimate the employed and employees of the informal sector in indirect effect of indirect taxes as described below. proportion to income. Agricultural income taxes and rural land used Direct Taxes fees are, for the most part, calculated on the basis of land holding size. The tax schedule for this tax and To estimate household level personal income tax, the fee is set by regional and local governments and as income tax schedule in Table A5.1A was applied on such varies from locale to locale. However, many of the disposable income of urban individuals who were the main tax schedules were examined and found to employed by formal private or public organizations. levy similar per hectare tax rates regardless of land For self-employed individuals and those employed in size. An example for Oromia region is included in Annex 5 173 TABLE A5.1: Ethiopia. Tax rate schedules on Direct Taxes Tax rate schedule for income from employment/ A.  Tax rule on taxable business income/Net Profit B.  personal income Tax Rate Deduction Taxable Business Income/ Tax Rate Deduction Income bracket per month (%) (Birr) Net Profit/bracket per year (%) (Birr) 0–150 Exempted 0–1800 Exempted 151–650 10 15.0 1,801–7800 10 180 651–1400 15 47.5 7801–16800 15 570 1,401–2350 20 117.5 16801–28200 20 1,410 2,351–3550 25 235.0 28201–42,600 25 2,520 3,551–5000 30 412.5 42601–60,000 30 4,950 Over 5000 35 662.0 Over 60, 000 35 7,950 Source: Ministry of Finance and Economic Development. Table A5.2. This example suggests that if anything the electricity, water, and transport.42 Although these per hectare tax rate falls with land holding size. The exemptions are meant to benefit low-income house- estimated agricultural tax and land use fees assumes holds, the leakage is substantial as richer households that the rates are always constant per hectare. The size consume these commodities as well. of land holding was collected in the WMS 2011 but Import duties account for 20% of indirect tax rev- standardized units were often not recorded making enue. Import duty rates vary depending on the type of it impossible to use. For this reason the ERSS was commodity. Exemptions from import duties or other used to define the association between land size and taxes levied on imports are granted for raw materials consumption in each region, which was then used to that are necessary for the production of export goods impute a land size for each household in the HCE. A and selected investment items. There are six bands on region’s total tax revenue was divided by total agricul- import duty with a maximum rate of 35%. The simple tural land holdings in the region in order to generate average tariff rate is 16.7%. In addition to import duty, an average tax rate per hectare. This rate was used a 10% surcharge on imported consumer imports was with the imputed land size in order to estimate the introduced in 2007 and implemented to date. Import amount of agricultural tax paid by each household. This method implicitly assumes that the average tax rate per hectare is constant across farm size. 42 VAT exempted goods and services: sale or transfer of a used dwelling, or the lease of a dwelling, financial services, the supply or import of national or foreign currency and of securities, import of gold to be transferred to Indirect taxes the National Bank of Ethiopia, services of religious organizations, medi- cines and medical services, educational services and child care services for children at pre-school institutions, goods and services for humanitarian VAT which is levied at the rate of 15% is a major aid and rehabilitation after natural disasters, industrial accidents, and component of indirect taxes and contributes to about a catastrophes, electricity, kerosene, and water; goods imported by the government, organizations, institutions or projects exempted from duties third of the total tax collection. There are VAT exemp- and other import taxes to the extent provided by law or by agreement, tions on various goods and services, most of which are postal service, transport, permits and license fees, goods or services by a workshop employing disabled individuals if more than 60 % of the aimed at favoring the low-income groups. Exemptions employees are disabled, books and other printed materials, unprocessed include unprocessed food items, medicine, kerosene, food items, palm oils used for food, bread, ‘injera,’ or milk 174 ETHIOPIA – POVERTY ASSESSMENT TABLE A5.2: Oromia Regional State, land use fee and agricultural income tax rule Rural Land Use Average Tax Rate Land size (hectare) Payment (Birr) Income Tax (Birr) Total (Birr) (Birr per hectare) < 0.5 15 Exempted 15 40.0 0.5–1 20 20 40 53.3 1–2 30 35 65 43.3 2–3 45 55 100 40.0 3–4 65 70 135 38.6 4–5 90 100 190 42.2 > 5.0 120 140 260 34.7 Source: Oromia Regional State, Proclamation to Amend Rural Land Use Payment and Agricultural Income Tax (No. 131/2007). The tax rate is own calculation. duty and surcharge on imports contribute to 30% of informal sector transaction and tax evasion in the the indirect tax revenue. formal sector. As a result, the incidence analysis has Excise taxes contribute 12% of indirect taxes and used the effective tax rate of the SAM. The effective they are levied on goods that are deemed to be ‘luxury’ tax rate of each commodity directly taken from the or harmful to health. The tax is levied on items such SAM represents the first round effect/burden of indi- as beverages and tobacco, electronics, textile, garments rect taxes on consumers. and motor vehicles imported or produced locally. The second round effects of indirect taxes are the The rates range from 10% (levied on items such as price burden on consumers resulting from indirect textile products) to 100% (levied on items such as taxes paid for inputs used in the production process. perfumes, alcohol, tobacco and high-power personal The input-output table is used to calculate the effect vehicles) (see A5.3 for details). of taxes on intermediate inputs on prices of final goods Indirect taxes are estimated by price multiplier anal- and services. The overall effect is the sum of the direct ysis using the Social Accounting Matrix (SAM) devel- and indirect effect of indirect taxes. The overall effect oped in 2006 by the Ethiopian Development Research of indirect taxes on prices of commodities from the Institute (EDRI). The SAM has 93 commodity input-output table is simulated using SIMSIP simula- accounts and distinguishes between purchased and tor, which gives the burden of indirect taxes for each own consumed commodities (77 are purchased and product (as a percent of the value of supply) in the 16 are own consumed commodities). The indirect tax commodity account. Using item level consumption account corresponding to each good and service in in the HCES survey data, the estimate of the price the SAM represents the actual indirect tax collected. burden on each household is based on the propor- This means that the ratio of the indirect tax to the tional increase in the price of each good and services total supply value of each commodity represents the and household’s expenditure on corresponding goods effective tax rate of each product. For own consumed and services, which is assumed to be borne entirely commodities, there is no indirect tax in the SAM as the by the consumers. actually collected tax from such commodities is zero. One concern is informality and the potential eva- The statutory tax rate is significantly higher than sion of consumption taxes. It is impossible to know the actual rate as the legal tax rates are not univer- from the survey whether a household has made a sally applied in all transactions. There is a significant purchase from a shop that pays VAT or not. Further, Annex 5 175 TABLE A5.3: Ethiopia: Goods that are liable to excise tax when either produced locally or imported Ser. Excise Tax Rate No. Type of Product (in %) 1 Any type of sugar (in solid form) excluding molasses 33 2 Drinks 2.1 All types of soft drinks (except fruit juices) 40 2.2 Powder soft drinks 40 2.3 Water bottled or canned in a factory 30 2.4 Alcoholic Drinks 2.4.1 All types of beer & stout 50 2.4.2 All types of wine 50 2.4.3 Whisky 50 2.4.4 Other alcoholic drinks 100 3 All types of pure Alcohol 75 4 Tobacco & Tobacco Products 4.1 Tobacco Leaf 20 4.2 Cigarettes, cigar, cigarillos, pipe tobacco, snuff and other tobacco products 75 5 Salt 30 6 Fuel-Super Benzene, Regular Benzene, Petrol, Gasoline and other Motor Spirits 30 7 Perfumes and Toilet Waters 100 8 Textile and textile products 8.1 Textile fabrics, knitted or woven of natural silk, rayon, nylon, wool or other similar materials 10 8.2 Textile of any type partly or wholly made from cotton, which is grey, white, dyed or printed, in pieces 10 of any length or width (except Mosquito net and Abudgedi) and including, blankets, bed-sheets, counterpanes, towels, table cloths and similar articles 8.3 Garments 10 9 Personal adornment made of gold, silver, or other materials 20 10 Dish washing machines of a kind for domestic use 80 11 Washing machines of a kind for domestic purposes 30 12 Video decks 40 13 Television and Video Cameras 40 14 Television broadcast receivers whether or not combined with gramophone, radio, or sound receivers and reproducers 10 15 Motor passenger cars, station wagons, utility cars, and Land Rovers, Jeeps pickups, similar vehicles (including motorized caravans), whether assembled, together with their appropriate initial equipment 15.1 Up to 1,300 c.c. 30 15.2 From 1,301 c.c. up to 1800 c.c. 60 15.3 Above 1,80 c.c. 100 16 Carpets 30 17 Asbestos and Asbestos Products 20 18 Clocks and watches 20 19 Dolls and toys 20 Source: Ethiopia Revenue and Customs Authority. 176 ETHIOPIA – POVERTY ASSESSMENT in a standard competitive model, prices at shops tax on different income groups does not change which- that do not pay VAT would be the same as those at ever method is used. VAT-paying shops, with the benefits of non-payment going to the firm owner rather than to the govern- Direct transfers ment. Households suffer the incidence of the tax regardless of the tax status of the seller, though not The 2011 HCES identifies households who received all the benefits go to the fiscal authorities. In essence, payment from the Productive Safety Net Program the assumption here is that all households buy the (PSNP) and households that receive food aid. Both same share of tax-paying goods so that the effects of PSNP payments and food aid payments were based tax avoidance or evasion on market prices are spread on household size and so the beneficiary status of across the population in proportion to each house- the households and the household size was used in hold’s expenditures. conjunction with government PSNP and food aid A sensitivity analysis is also made using an alter- expenditures to impute the value of transfers received native way of estimating the impact of indirect taxes. by each household. We assume that Food Aid and In the benchmark estimate both the first and second PSNP transfers were distributed to all beneficiaries round effects of all types of indirect taxes including equally. VAT are included. This approach considers VAT to be similar to sales tax in which additional taxes are Subsidies paid in each chain of the transaction. In the alterna- tive way, only the first effect of VAT on prices is esti- Item-level HCES data was used to estimate the mated because in principle, producers and retailers are amount of household’s consumption of wheat, kero- entitled for a refund of the VAT payments for input sene, and electricity. The subsidy per kg, liter and purchases making intermediate inputs tax-free. The kilowatt-hours, for each good respectively was then only exceptions to this are items that are VAT exempt, applied to estimate the total value of the subsidy which would have some indirect impact of VAT on received by the household. The wheat subsidy was intermediate goods. This is because if a good is VAT present in Addis Ababa City Administration only exempt, producers are not entitled for VAT refund for and so was only applied to households living in Addis the inputs used in producing the item. As a result, in Ababa. The wheat subsidy was 150 Birr per quintal. the sensitivity analysis, the first round effect of VAT is The electricity subsidy is provided in Table A5.4 and estimated for items on which VAT is levied, and then depends on the amount of electricity consumed. The only the 2nd round effects is included for goods and tariff rate is progressive, but the rates in all ranges are services that are VAT exempt. below the unsubsidized tariff. The government regu- Since the sensitivity analysis excludes the second lates petroleum prices and kerosene was subsidized at round effects of VAT on most items, the estimate Birr 2.17 per liter. of indirect tax burden using this method is slightly smaller than the estimate in the benchmark estimate. Education As a result, the associated income measures of post fis- cal income and final income become slightly higher in The numbers of students enrolled in primary, second- the sensitivity analysis. Apart from the slight change ary and tertiary education recorded in the Welfare in level, the pattern of incidence of indirect taxes on Monitoring Survey are used to determine the total the different income groups based on this method is number of individuals enrolled in primary, second- similar to the pattern in the benchmark estimate. Thus ary and tertiary education in each region. Unit costs the overall storyline of the relative burden of indirect of primary, secondary and tertiary education were Annex 5 177 TABLE A5.4: Current tariff for household electricity consumption (monthly) Monthly consumption of electricity Tariff without subsidy From (kwh) To (kwh) Tariff (Birr/kwh) (Birr/kwh) subsidy (Birr) per kwh 0 50 0.273 0.967 0.694 51 100 0.356 0.967 0.611 101 200 0.499 0.967 0.468 201 300 0.550 0.967 0.417 301 400 0.567 0.967 0.401 401 500 0.588 0.967 0.379 501 1E+07 0.694 0.967 0.273 Source: EEPCO, 2003. obtained by dividing the total regional public spend- Health ing obtained from the MOFED 2013 Government Finance Report by total regional enrollment. The in- For health, total public health spending from MOFED kind transfer of education spending at household level (2013) Government Finance Report was distributed is determined by multiplying the number children to all individuals that received public health service as enrolled in primary, secondary and tertiary education recorded in the Welfare Monitoring Survey (WMS). in 2010/11 by the unit costs. Public education spend- For curative health services, in-kind health benefits ing includes salary, wages, and operational costs as are estimated in proportion to households’ expendi- well as the administration and capital expenditure for ture on public health fees. For households that are primary and secondary education. For tertiary educa- exempted from user fees, average benefit is assumed. tion, a significant proportion of capital expenditure is The WMS is used to identify households that received excluded because there were very large expenditures in health service for free. For preventive health services, expansion of higher education infrastructure, which the benefits are distributed to all households equally. will serve another generation in the future. Only 10% Based on budget on different health programs, the of the capital expenditure is considered in the analysis proportion of preventive and curative health services to account for the benefits the current students are is estimated to be 27% and 73% respectively of total receiving. government health budget. 179 ANNEX 6 F The Oaxaca-Blinder decomposition (introduced by ∆O = ( b M - b F ) X i , which is usually referred to as Blinder (1973) and Oaxaca (1973) and hereon OB), the “unexplained” part and corresponds to the differ- on which the estimates presented in section 9.2 are ence in the returns to inputs. based, estimates the following specifications: To solve this problem, Ñopo (2008) suggests the use of matching methods to divide the male and Yi M = X 'iM b M (1) female samples in two parts: (i) the matched part of the sample, which corresponds to the observations Yi F = X 'iF b F (2) that can be matched based on their input levels (X) with at least one observation of the other group, Then, it decomposes the overall difference in Y in and (ii) the unmatched part of the sample, which the following terms: corresponds to observations that cannot be matched. This method is employed in section 9.4. As a result, M F M F F Y -Y = X i b M - X i b F ± X i b M = (3) the unmatched individuals are those that have specific M F F levels of inputs or characteristics (X) that are only ( X i - X i )b M + ( b M - b F ) X i found in one group. This would be the group out of In this methodology, OLS estimations of equa- the common support. tions (1) and (2) are done using only male and female Once these groups are identified, the overall dif- observations, respectively. Then, in equation (3) the ference in Y is decomposed in the following terms, M F M F F Y -Y = b ±( X i b )= X i b - X i term M counterfactual F M is added and subtracted. which will be non-parametrically estimated following M This term represents the value of the dependent vari- F F Ñopo’s methodology: ( X i - X i )b M + ( b M - b F ) X i able (Y) that females would obtain if they have the The components of the decomposition are the same returns to inputs (X) as males. The common following: support assumption is relevant at this point since the counterfactual term uses the results of an estima- 1. Endowment effect: tion done with male observations and employs it ∆X = E m M (Y | M )- Em F (Y | M ) with the female observations. If the distribution of inputs (X) of either group has regions in which the This component represents the difference in the aver- other group has no observations, the counterfactual age level of the dependent variable (Y) that results from implies an extrapolation in the returns to inputs that differences in the distribution of inputs. The sub-index may not be valid. m indicates that the estimation is only done using the In the OB literature, equation (3) is interpreted matched observations (i.e. the sample in the common as the addition of two terms: the endowment effect, ∆support). X = Em M (Y The|M )- Em term F ( Y | M ) is the counterfactual M F ∆X = ( X i - X i )b M , which corresponds to the average income that females would receive if they part of the difference explained by the average differ- were “paid as males.” The values for this component ence in the levels of inputs, and the structural effect, are calculated using matching methods. In summary, 180 ETHIOPIA – POVERTY ASSESSMENT each female in the common support is matched to 3. Unmatched male effect. one or more males. A counterfactual Y for the female ∆M = ( E uM ( Y | M ) - E m M ( Y | M ))p M ( u ) is formed by averaging the observed levels of Y of the matched males. As a result, this counterfactual keeps This component represents the difference of the the distribution of inputs of the female group. This dependent variable (Y) between unmatched and F component is analogous to the X i b M component matched males, weighted by the probability of being in the OB decomposition. unmatched =(E ∆M male M u (Y | M )- conditional M m ( Y male onEbeing | M )) (p M ( u )). This term quantifies the part of the overall gender dif- 2. Structural effect: ferential that is explained by the advantage (or disad- vantage if negative) that unmatched males have with ∆O = E m F (Y | M )- Em F (Y | F ) respect to matched males. This component represents the difference in the aver- age level of the dependent variable (Y) that results 4. Unmatched female effect. from differences in the returns to inputs (the unex- ∆F = ( E m F ( Y | F ) - E uF ( Y | F ))p F ( u ) plained part). In practice, to estimate the structural effect term, similar steps to estimating an average This component represents the difference of the treatment effect on the treated (ATT) using matching dependent variable (Y) between matched and are followed. Adding ∆X + ∆O results in the overall unmatched females, weighted by the probability of male-female difference for the observations in the being unmatched female conditional on being female common support. ∆F = ( E m F ( Y | F ) - E uF ( Y | F (p F ( u )). This term quantifies the part of the overall )) gender differential that is explained by the advantage (or disadvantage if negative) that matched females have with respect to unmatched females. 181 REFERENCES Abramovsky, Laura, Orazio Attanasio, and David Alkire, S., J. M. Roche and A. Vaz. 2014. Phillips. 2011. A tax micro-simulator for Mexico “Multidimensional Pover ty Dynamics: (MEXTAX) and its application to the 2010 tax Methodology and Results for 34 countries.” reforms. Oxford Poverty and Human Development Acemoglu, D. and R. Shimer. 2000. “Productivity Initiative, Oxford University. ophi.qeh.ox.ac.uk gains from unemployment insurance.” European Anttila-Hughes, Jesse K. and Solomon M. Hsiang. Economic Review 44(7), 1195–1224. 2013. Destruction, Disinvestment, and Death: Aguilar, A., E. Carranza, M. Goldstein, T. Kilic, and Economic and Human Losses Following G. Oseni. 2014. Decomposition of gender dif- Environmental Disaster. Mimeo. University of ferentials in agricultural productivity in Ethiopia. San Francisco. Policy Research Working Paper Series 6764, The Atkinson, A. B. 2003. “Multidimensional depriva- World Bank. tion: Contrasting social welfare and counting Alem, Yonas and Måns Söderbom. 2012. “Household- approaches.” Journal of Economic Inequality, l Level Consumption in Urban Ethiopia: The (1): 51–65. Effects of a Large Food Price Shock.” World Atkinson, A. B. and M. A. Lugo. 2010. Growth, pov- Development, Elsevier, vol. 40(1), pages 146–162. erty and distribution in Tanzania. London School Alkire, Roche & Seth. 2013.: Multidimensional of Economics, London: International Growth Poverty Index 2013.: Oxford Poverty & Human Centre, Working Paper 10/0831. Development Initiative, University of Oxford. Au, Chun-Chung and J. Vernon Henderson. 2006. Alkire, S. and J. Foster. 2007. “Counting and “How migration restrictions limit agglom- Multidimensional Poverty Measurements.” OPHI eration and productivity in China.” Journal of Working Paper No. 7, University of Oxford. Development Economics, Elsevier, vol. 80(2), pages Alkire, S. and J.M. Roche. 2011. “How MPI Went 350–388, August. Down: Dynamics and Comparisons.” Oxford Au, Chun-Chung and J. Vernon Henderson. 2006. Poverty & Human Development Initiative, “Are Chinese Cities Too Small?” Review of University of Oxford. Economic Studies, Oxford University Press, vol. Alkire, S. and M. E. Santos. 2013. “Measuring Acute 73(3), pages 549–576. Poverty in the Developing World: Robustness and Barrett, Christopher, Paswel Phiri Marenya, John Scope of the Multidimensional Poverty Index.” The Mcpeak, Bart Minten, Festus Murithi, Willis Oxford Poverty & Human Development Initiative Oluoch-Kosura, Frank Place, Jean Claude Working Paper No. 59, University of Oxford. Randrianarisoa, Jhon Rasambainarivo and Justine Alkire, S., A. Conconi, and S. Seth. 2014. Wangila. 2006. “Welfare dynamics in rural Kenya “Multidimensional Poverty Index 2014: Brief and Madagascar.” Journal of Development Studies Methodological Note and Results.” Oxford 42(2): 248–277. Poverty and Human Development Initiative, Barrios, S., L. Bertinelli, and E. Strobl 2010. “Trends Oxford University. ophi.qeh.ox.ac.uk. in rainfall and economic growth in Africa: A 182 ETHIOPIA – POVERTY ASSESSMENT neglected cause of the African growth tragedy.” Bezu, Sosina, Christopher B Barrett, and Stein T. The Review of Economics and Statistics. Holden. 2012. “Does the Nonfarm Economy Beegle, K. and G. Oseni. 2008. “Nonfarm Enterprises, Offer Pathways for Upward Mobility? Evidence Food Security, and Economic Shocks in Ethiopia.” from a Panel Data Study in Ethiopia.” World Background paper for the Ethiopia Rural ICA. Development , Elsevier, vol. 40(8), pages Beegle, Kathleen, Joachim De Weerdt, and Stefan 1634–1646. Dercon. 2011. “Migration and Economic Blinder, A. S. 1973. “Wage Discrimination: Reduced Mobility in Tanzania: Evidence from a Tracking Form and Structural Estimates.” Journal of Human Survey.” The Review of Economics and Statistics. Resources, University of Wisconsin Press, vol. 8(4), Beneke, Margarita, Nora Lustig, and José Andrés pages 436–455. Oliva. 2014. El impacto de los impuestos y el Brockerhoff, Martin, and Ellen Brennan. 1998. gasto social en la desigualdad y la pobreza en El “The Poverty of Cities in Developing Regions.” Salvador. CEQ Working Paper No. 26, Center Population and Development Review 24, No. 1. for Inter-American Policy and Research and (March). Department of Economics, Tulane University and Bucheli, Marisa, Nora Lustig, Máximo Rossi, and Inter-American Dialogue. Forthcoming. Florencia Amábile. 2014. “Social Spending, Berhane, G. R.V. Hill, S. Dercon, and A. Seyoum Taxes and Income Redistribution in Uruguay.” Taffesse. 2014. The impact of insurance on agri- In Lustig, Nora, Carola Pessino and John Scott. cultural investment: evidence from Ethiopia. 2014. Editors. The Redistributive Impact of Taxes Mimeo, IFPRI. and Social Spending in Latin America. Special Issue. Bernard, Tanguy, Stefan Dercon, Kate Orkin, and Public Finance Review, May, Volume 42, Issue 3. Alemayehu Seyoum Taffesse. 2014. “The Future in Buehren, N., M. Goldstein, T.A. Ketema, E. Molina, Mind: Aspirations and Forward Looking Behavior and A.Y Teklu. 2014. “The Impact of Strengthening in Rural Ethiopia.” Centre for Study of African Agricultural Extension Services: Evidence from Economies Working Paper 16, 1–48. Ethiopia.” Unpublished Manuscript. Besley, T. and T. Persson. 2011. “Public Finance and Cabrera, Maynor, Nora Lustig, and Hilcías Morán. Development.” Draft chapter for the Handbook 2014. Fiscal Policy, Inequality and the Ethnic of Public Economics. Divide in Guatemala. CEQ Working Paper Bevan, Philippa, Catherine Dom, and Alula Pankhurst. No. 20, Center for Inter-American Policy and 2011. Long Term Perspectives on Development Research and Department of Economics, Tulane Impacts in Rural Ethiopia: WIDE 3 Stage 1 University and Inter-American Dialogue, April Mixed Communities: Short summary, March 2014. Forthcoming. 2011. Mokoro. Canning, David, Sangeeta Raja, and Abdo Yazbeck. Bevan, Philippa, Catherine Dom, and Alula Pankhurst. 2014. People powered growth for Africa’s Young 2013. Long Term Perspectives on Development Lions. World Bank Report, Africa Region. Impacts in Rural Ethiopia: WIDE 3 Stage 2 Eight Carranza, E. and J. Gallegos. 2013. Ethiopia Trends Food Insecure Communities: Short summary, of Wellbeing 2000–2011. Poverty Reduction and March 2013. Mokoro. Economic Management, Africa Region, World Bevan, Philippa, Catherine Dom and Alula Pankhurst. Bank. 2014. Long Term Perspectives on Development Central Statistical Agency. 2013. Welfare Monitoring Impacts in Rural Ethiopia: WIDE 3 Stage 3 Six Survey-2011. Addis Ababa. Communities with Agricultural Potential: Short Central Statistical Agency [Ethiopia] and ICF Summary, May 2014. Mokoro. International. 2012. Ethiopia Demographic and REFERENCES 183 Health Survey 2011. Addis Ababa, Ethiopia and Discussion. Paper prepared as part of a study on Calverton, Maryland, USA: Central Statistical Agriculture and Growth in Ethiopia, March 2009. Agency and ICF International. Dercon, S. and R. V. Hill. 2009. Growth from Central Statistics Agency. 2012. Inter-censal Agriculture in Ethiopia: Identifying Key Population Survey, 2012. Constraints. Paper prepared as part of a study on Christiaensen, L., L. Demery, J. Kuhl. 2011. “The Agriculture and Growth in Ethiopia, March 2009. (evolving) role of agriculture in poverty reduc- Dercon, S., Daniel O. Gilligan, John Hoddinott, tion—An empirical perspective.” Journal of and Tassew Woldehanna. 2009. “The Impact of Development Economics 96: 239–254. Agricultural Extension and Roads on Poverty Christiaensen, Luc, Punam Chuhan-Pole, and Aly and Consumption Growth in Fifteen Ethiopian Sanoh. 2013. “Africa’s Growth, Poverty and Villages.” American Journal of Agricultural Inequality Nexus – Fostering Shared Prosperity.” Economics 91(4), pages 1007–1021. World Bank, 2013, mimeo. Dercon, Stefan and Luc Christiaensen. 2011. Dabalen, Andrew, Ambar Narayan, Jaime Saavedra- “Consumption risk, technology adoption and Chanduvi, and Alejandro Hoyos Suarez, with poverty traps: Evidence from Ethiopia.” Journal Ana Abras and Sailesh Tiwari. 2014. Do African of Development Economics 96(2), pages 159–173 Children Have an Equal Chance? A Human Dercon, Stefan and Catherine Porter. 2012. “Live Opportunity Report for Sub-Saharan Africa. aid revisited: long-term impacts of the 1984 World Bank Report. Ethiopian famine on children.” Journal of the De Brauw, A., V. Mueller, and T. Woldehanna. European Economic Association. 2013. “Motives to Remit: Evidence from Dercon, Stefan and Pramila Krishnan. 2000. “In Tracked Internal Migrants in Ethiopia.” World Sickness and in Health: Risk Sharing within Development. Households in Rural Ethiopia.” Journal of Political De Brauw, Alan. 2014. Migration, Youth and Economy, University of Chicago Press, vol. 108(4), Agricultural Productivity in Ethiopia. Background pages 688–727, August. Paper prepared for the Ethiopia Poverty Dercon, Stefan, John Hoddinott, and Tassew Assessment. Woldehanna. 2005. “Shocks and consumption Deininger, K., D. A. Ayalew, and T. Alemu. 2011. in 15 Ethiopian villages, 1999–2004.” Journal of “Impacts of Land Certification on Tenure Security, African Economies Investment, and Land Market Participation: Development Planning and Research Directorate, Evidence from Ethiopia.” Land Economics , Ministry of Finance and Economic Development University of Wisconsin Press, vol. 87(2), pages (MOFED). 2013. Poverty and growth in Ethiopia 312–334. (1995/96–2010/11): The poverty reduction Deininger, Klaus, Jin, Songqing, and Xia, Fang. 2012. legacy of Meles Zenawi and future challenges. Moving off the farm: Land institutions to facilitate MOFED, Addis Ababa. structural transformation and agricultural produc- Diao, Xinshen, Alemayehu Seyoum Taffesse, Paul A. tivity growth in China. Policy Research Working Dorosh, James Thurlow, Alejandro Nin-Pratt, and Paper Series 5949, The World Bank. Bingxin Yu. 2012. “Ethiopia.” in Xinshen Diao, Dercon, Stefan. 2004. “Growth and Shocks: Evidence James Thurlow, Samuel Benin, and Shenggen Fan from Rural Ethiopia.” Journal of Development (eds.), Strategies and priorities for African agricul- Economics 74 (2): 309–29. ture: Economy-wide perspectives from country stud- Dercon, S. and A. Zeitlin. 2009. Rethinking ies, Chapter 5, pp. 107–140, International Food Agriculture and Growth in Ethiopia: A Conceptual Policy Research Institute (IFPRI). 184 ETHIOPIA – POVERTY ASSESSMENT Dorosh and Thurlow. 2013. Beyond Agriculture: Ferreria, F., P.G. Leite, and M. Ravallion. 2010. Decomposing Non-Agricultural Growth Poverty “Poverty reduction without economic growth? Linkages. Explaining Brazil’s poverty dynamics.” 1985– Elbers, C., J.O. Lanjouw, and P. Lanjouw. 2003. 2004. Journal of Development Economics 93(1): “Micro-level Estimation of Poverty and Inequality,” 20–36 Econometrica, 71(1):355–364. Firpo, S., N. M. Fortin, and T. Lemieux. 2009. Emran, M.S. and F. Shilpi. “The General Equilibrium “Unconditional Quantile Regressions.” Effect of Land Market Restriction on Labor Econometrica, Vol. 77, No. 3, 953–973 Market: Evidence from Wages in Sri Lanka.” Foster, J., J. Greer, and E. Thorbecke. 1984. “A class American Journal of Agricultural Economics, of decomposable poverty measures.” Econometrica forthcoming. Vol. 52: 761–766. Ethiopian Development Research Institute (EDRI). Fox, L. and M. Gaal. 2008. Working Out of Poverty. 2009. Ethiopian Social Accounting Matrix. The World Bank, Washington DC. Fafchamps, M. and A. Quisumbing. 2005. “Assets at Franklin, S. 2014. Work, Unemployment and Job marriage in rural Ethiopia.” Journal of Development Search among the Youth in Urban Ethiopia. Economics, Elsevier, vol. 77(1), pages 1–25, June. Background Paper prepared for the Ethiopia Fan, C. Cindy. 2008. China on the Move: Migration, Poverty Assessment. the State, and the Household. London and New Fullerton, D. and G. Metcalf. 2002. “Tax Incidence.” York, NY: Routledge. In Handbook of Public Economics 4, Auerbach, FAO. 2011. “The State of Food and Agriculture A. and Feldstein, M. (eds.), Elsevier Science, 2010–2011: Women in Agriculture: Closing the Amsterdam, 1787–1872. Gender Gap for Development.” Rome. Gilligan, D., J. Hoddinott, N. Kumar, A. Taffesse. Fay, M., and C. Opal. 2000. Urbanization without 2010. Targeting food security interventions growth: a not-so-uncommon phenomenon. when “everyone is poor.” The case of Ethiopia’s World Bank Policy Research Working Paper 2412 Productive Safety Net Programme. International Federal Ministry of Health, 2013. Annual Performance Food Policy Research Institute, Washington DC. Report 2012/13. Addis Ababa, Ethiopia. Gilligan, Daniel, John Hoddinot, and Alemayehu Fei, J.C.H., and G. Ranis. 1964. Development of the Seyoum Taffesse. 2009. “The Impact of Ethiopia’s Labor Surplus Economy: Theory and Policies. R. D. Productive Safety Net Programme and its Irwin: Homewood, Ill. Linkages.” Journal of Development Studies 45(10), Ferré, C., F.H.G. Ferreira, and Lanjouw, P. 2010. pages 1684–1706. Is There a Metropolitan Bias? The Inverse Gollin, D., R. Jedwab, and D. Vollrath. 2013. Relationship between Poverty and City Size in Urbanization with and without Industrialization. Selected Developing Countries. Washington, mimeo. DC: The World Bank, Policy Research Working Gollin, Douglas, David Lagakos, and Michael Waugh. Paper 5508. 2014. “The Agricultural Productivity Gap.” Ferreira, F.H.G. 2010: Distributions in Motion: Quarterly Journal of Economics 129(2). Economic Growth, Inequality, and Poverty Haggblade S., P. Hazell, and T. Reardon. 2010. “The Dynamics. World Bank. Rural Nonfarm Economy: Prospects for Growth Fe r r e i r a , F. H . G . a n d M . A . L u g o. 2 0 1 2 . and Poverty Reduction.” World Development, Vol. Multidimensional Poverty Analysis: Looking 38, No. 10, pp. 1429–1441. for a Middle Ground. Policy Research Working Hassan, Ibrahim Worku and Alemayehu Seyoum Paper 5964. Taffesse. 2014. Decomposing the Growth of REFERENCES 185 Household Consumption Expenditure to Identify Jellema, Jon, Matthew Wai-Poi, and Rythia Afkar. the Contribution of Wellbeing Covariates in 2014. “The Distributional Impact of Fiscal Policy Ethiopia. ESSP Working Paper, forthcoming. in Indonesia.” World Bank (forthcoming). Headey, Derek, Fantu Bachewe Nisrane, Ibrahim Jolliffe, Dean, Gbemisola Oseni, Ilana Seff. Alemayehu Worku, Mekdim Dereje, and Alemayehu Seyoum Ambel, Julia Kowalski, Alina Lipcan, Katie McIntosh, Taffesse. 2012. Urban Wage Behavior and Food Remy Smida, and Signe Jung Sørensen. 2014. Price Inflation: The Case of Ethiopia. ESSP Nonfarm Enterprises in Rural Ethiopia: Improving Working Paper 41. Livelihoods by Generating Income and Smoothing Higgins, Sean and Claudiney Pereira. 2014. “The Consumption? Background Paper prepared for Effects of Brazil’s Taxation and Social Spending the Ethiopia Poverty Assessment, World Bank. on the Distribution of Household Income.” In Kakwani, N. 1993. Statistical inference in the mea- Lustig, Nora, Carola Pessino and John Scott. surement of poverty. Review of Economics and 2014. Editors. The Redistributive Impact of Taxes Statistics 75 (4): 632–639. and Social Spending in Latin America. Special Issue. Kalwij, Adriaan and Arjan Verschoor. 2005. A Public Finance Review, May, Volume 42, Issue 3. Decomposition of Poverty Trends across Regions: Higgins, Sean and Nora Lustig, 2014. Measuring The Role of Variation in the Income and Fiscal Impoverishment. Dissertation Essay, Inequality Elasticities of Poverty, Working Department of Economics, Tulane University, Paper Series  RP2005/36, World Institute May. f o r D e v e l o p m e n t E c o n o m i c Re s e a r c h Hill, R.V. and C. Porter. 2014. A vulnerability (UNU-WIDER). Assessment for Ethiopia. Background Paper pre- Khan, Qaiser M., Jean-Paul Faguet, Christopher pared for the Productive Safety Net Programme Gaukler, and Wendmsyamregne Mekasha. 2014. and the Ethiopia Poverty Assessment. Improving Basic Services for the Bottom Forty Hoddinott, John, Alemayehu Seyoum Taffesse, and Percent: Lessons from Ethiopia. World Bank, Yisehac Yohannes. 2014. Locus of control and Washington, DC. economic behavior in rural Ethiopia. IFPRI Kilic, T., A. Palacios-Lopez, and M. Goldstein. 2013. Mimeo. Caught in a productivity trap: a distributional Hoddinott, John, Guush Berhane, Daniel O. Gilligan, perspective on gender differences in Malawian Neha Kumar, and Alemayehu Seyoum Taffesse. agriculture. Policy Research Working Paper Series 2012. The Impact of Ethiopia’s Productive 6381, The World Bank. Safety Net Programme and Related Transfers Kondylis, F. and V. Mueller. 2012. Seeing is Believing? on Agricultural Productivity. Journal of African Evidence from a Demonstration Plot Experiment Economies 21(5): 761–786. in Mozambique. MSSP working papers 1, Inchauste, Gabriela, Nora Lustig, Mashekwa Maboshe, International Food Policy Research Institute Ingrid Woolard. 2014. “The Distributional (IFPRI). Impact of Fiscal Policy in South Africa” World Kowalski, Julia, Alina Lipcan, Katie McIntosh, Bank (forthcoming). Remy Smida, Signe Jung Sørensen, Dean Jolliffe, Jaramillo, Miguel. 2014. “The Incidence of Social Gbemisola Oseni, Ilana Seff and Alemayehu Spending and Taxes in Peru.” In Lustig, Nora, Ambel. 2014. Nonfarm Enterprises in Rural Carola Pessino and John Scott. 2014. Editors. The Ethiopia: Improving Livelihoods by Generating Redistributive Impact of Taxes and Social Spending Income and Smoothing Consumption? in Latin America. Special Issue. Public Finance Background paper prepared for the Ethiopia Review, May, Volume 42, Issue 3. Poverty Assessment. 186 ETHIOPIA – POVERTY ASSESSMENT Lanjouw, P. and R. Murgai. 2009. “Poverty Decline, Argentina, Bolivia, Brazil, Mexico, Peru and Agricultural Wages, and Non-Farm Employment Uruguay. Introduction to Special Issue,” Public in Rural India 1983–2004.” Agricultural Economics. Finance Review, May 2014, Volume 42, Issue 3, 40(2): 243–263 published online November 20, 2013 Lewis, W. Arthur. 1954. “Economic Development with Lustig, Nora, et.al. 2013. Commitment to Equity: The Unlimited Supplies of Labour,” The Manchester Impact of Taxes and Social Spending in Inequality School 22(2): 139–191. and Poverty in Argentina, Bolivia, Brazil, Mexico, Ligon, E., and E. Sadoulet. 2007. Estimating the Peru, and Uruguay: an Overview. CEQ working Effects of Aggregate Agricultural Growth on the paper #13. Distribution of Expenditures. Background paper Lustig, Nora. 2014. Taxes, Transfers, Inequality and for the World Development Report 2008. the Poor in the Developing World. Round 1. Loayza, N., and C. Raddatz. 2010. “The composition CEQ Working Paper No. 23, Center for Inter- of growth matters for poverty alleviation.” Journal American Policy and Research and Department of of Development Economics 93 (1): 137–151. Economics, Tulane University and Inter-American Loening, J. L., B. Rijkers, and M. Söderbom. 2008. Dialogue. Forthcoming. Non-farm microenterprises and the investment Lustig, Nora. and Sean Higgins. 2013. Commitment climate: Evidence from rural Ethiopia. World to Equity Assessment (CEQ):  Estimating the Bank Policy Research working paper 4577 Incidence of Social Spending, Subsidies and Lucas, Robert E.B., and Oded Stark (1985). “The Taxes. Handbook. CEQ Working Paper No. 1, New Economics of Labor Migration,” American Center for Inter-American Policy and Research Economic Review Papers and Proceedings 75(2): and Department of Economics, Tulane University 173–178. and Inter-American Dialogue. Lustig and Higgins. 2013. Measuring Impoverishment: Martins, P. 2014. Structural Change in Ethiopia: An An Overlooked Dimension of Fiscal Incidence. Employment Perspective. World Bank Policy Tulane University Economics Working Paper Research Working Paper No. 6749 1315, New Orleans, Louisiana, April 2013 and Mekonnen, D. K., J. H. Dorfman, and E. G. Fonsah. CEQ Working Paper No. 14, New Orleans, 2013. Productivity and Efficiency of Small Scale Louisiana, April Agriculture in Ethiopia. 2013 Annual Meeting, Lustig, Nora and Carola Pessino. 2014. “Social February 2–5, 2013, Orlando, Florida 143038, Spending and Income Redistribution in Argentina Southern Agricultural Economics Association. in the 2000s: the Rising Role of Noncontributory Ministry of Finance and Economic Development. Pensions.” Public Finance Review, May 2014, MOFED. 2006. Ethiopia: Building on Progress: A Volume 42, Issue 3. Plan for Accelerated and Sustained Development Lustig, Nora and Sean Higgins. 2012. Fiscal to End Poverty (PASDEP) 2005/06–2009/10, Incidence, Fiscal Mobility and the Poor: a Addis Ababa, Ethiopia New Approach. CEQ Working Paper No. 4, Ministry of Finance and Economic Development September. Available at: http://stonecenter.tulane. (MOFED). 2008. Dynamics of Growth and Poverty edu/uploads/Lustig.Fiscal_Incidence,_Fiscal_ in Ethiopia 1995/96–2004/05. Addis Ababa, Mobility_and_the_Poor-_A_New_Approach._ Ethiopia. April2012–1344004195.pdf Ministry of Finance and Economic Development Lustig, Nora, Carola Pessino, and John Scott. 2014. (MOFED). 2010. Ethiopia Growth and “The Impact of Taxes and Social Spending Transformational Plan 2010/11–2014/15. Volume on Inequality and Poverty in Latin America: I, Main Text. Addis Ababa, Ethiopia. REFERENCES 187 Ministry of Finance and Economic Development Bolivia.” In Lustig, Nora, Carola Pessino and John (MOFED). 2012. Ethiopia’s Progress Towards Scott. 2014. Editors. The Redistributive Impact Eradicating Poverty: An Interim Report on Poverty of Taxes and Social Spending in Latin America. Analysis Study (2010/11). Addis Ababa, Ethiopia. Special Issue. Public Finance Review, May, Volume Ministry of Finance and Economic Development 42, Issue 3. (MOFED). 2013. Development and Poverty Plavgo, Ilze, Martha Kibur, Mahider Bitew, Tesfayi in Ethiopia 1995/96–2010/11. Addis Ababa, Gebreselassie, Matusda Yumi, and Roger Pearson. Ethiopia 2013. Multidimensional child Deprivation Trend Minten, B., D. Stifel, and S. Tamru. 2012. Structural Analysis in Ethiopia, Further analysis of the 2000, Transformation in Ethiopia: Evidence from Cereal 2005, and 2011 Demographic and Health Surveys. Markets. ESSP 2 Working Paper 39. DHS Further analysis Reports No. 83. Calverton Minten, B., B. Koro and D. Stifel. 2014. The Last MD: ICF International. Mile(s) in Modern Input Distribution: Evidence Porter, Catherine. 2012. “Shocks, Consumption and from Northwestern Ethiopia. ESSP 2 Working Income Diversification in Rural Ethiopia.” Journal Paper 51. of Development Studies, Taylor & Francis Journals, Montalvo, Jose G. and Martin Ravallion. 2010. vol. 48(9), pages 1209–1222, September. “The pattern of growth and poverty reduction Poschke, Markus. 2014. A model of Entrepreneurship in China.” Journal of Comparative Economics, and Employment in Ethiopia: Simulating the Elsevier, vol. 38(1), pages 2–16, March. impact of an urban safety-net. Background Paper Ñopo, H. 2008. “An extension of the Blinder-Oaxaca prepared for the Ethiopia Poverty Assessment. decomposition to a continuum of comparison PovcalNet. World Bank. http://iresearch.worldbank. groups.” Economics Letters, Elsevier, vol. 100(2), org/PovcalNet/index.htm. Downloaded on June pages 292–296, August. 4, 2014. Oaxaca, R. 1973. “Male-Female Wage Differentials Ravallion, M. 1992. Poverty comparisons—A guide in Urban Labor Markets.” International Economic to concepts and method s . Living Standards Review, vol. 14(3), pages 693–709, October. Measurement Survey Discussion Paper 88. Olinto, Pedro and Maya Sherpa. 2014. Targeting Washington, D.C.: World Bank. Assessment and Ex-Ante Impact Simulations of Ravallion, Martin and Monika Huppi. 1991. Addis Ababa Safety Net. Background Paper pre- “Measuring Changes in Poverty: A Methodological pared for the Ethiopia Poverty Assessment. Case Study of Indonesia during an Adjustment Oseni, G., P. Corral, M. Goldstein, and P. Winters. Period.” World Bank Economic Review, World 2014. Explaining gender differentials in agri- Bank Group, vol. 5(1), pages 57–82, January. cultural production in Nigeria. Policy Research Ravallion, Martin, and Gaurav Datt. 1996. “How Working Paper Series 6809, The World Bank. important to India’s poor is the sectoral com- Oxfam. 2013. “Women’s Collective Action Unlocking position of economic growth?” The World Bank the potential of agricultural markets,” Oxford: Economic Review 10 (1), 1–2 Oxfam International. Rijkers, B. and R. Costa. 2012. Gender and Rural Non- Oxford Poverty and Human Development Initiative. Farm Entrepreneurship. World Development. 2014. Global Multidimensional Poverty Index Rijkers, B. and M. Söderbom. 2013. The Effects of Risk Databank. OPHI, University of Oxford. and Shocks on Non-Farm Enterprise Development Paz Arauco, Verónica, George Gray Molina, Wilson in Rural Ethiopia. World Development. Jiménez Pozo, and Ernesto Yáñez Aguilar. 2014. Robinson, S., K. Strzepek and R. Cervigni. 2013. The “Explaining Low Redistributive Impact in Cost of Adapting to Climate Change in Ethiopia: 188 ETHIOPIA – POVERTY ASSESSMENT Sector-Wise and Macro-Economic Estimates. Teklu E., T. Mamo, S. Kidanu, and M. Abebe. 2000. ESSP2 Discussion Paper 53. Response of Some Durum Wheat Landraces to Schmidt, E. and M. Kedir. 2009. Urbanization and Nitrogen Application on Ethiopian Vertisols. Spatial Connectivity in Ethiopia: Urban Growth Paper presented on the 11th Regional Wheat Analysis Using GIS. ESSP2 Discussion Paper 3. Workshop for Eastern, Central and Southern Scott, John. 2014. “Redistributive Impact and Africa Sept. 18–24, 2000. Addis Ababa, Efficiency of Mexico’s Fiscal System.” In Lustig, Ethiopia. Nora, Carola Pessino and John Scott. 2014. Thomas, Timothy, Luc Christiaensen, Quy Toan Do Editors. The Redistributive Impact of Taxes and and Le Dang Trung. 2010. Natural Disasters Social Spending in Latin America. Special Issue. and Household Welfare Evidence from Vietnam. Public Finance Review, May, Volume 42, Issue 3. World Bank Policy Research Working Paper Series Serneels, Pieter. 2007. “The nature of unemployment 5491. among young men in urban Ethiopia.” Review of Tiruneh, A., T. Tesfaye, W. Mwangi, and H. Verkuijl. Development Economics. 2001. Gender Differentials in Agricultural Shilpi, Forhad and Jiaxiong Yao. 2014. Internal Production and Decision-Making Among Migration in Ethiopia: Stylized Facts from Smallholders in Ada, Lume, and Gimbichu Population Census, 2007. Background Paper Woredas of the Central Highlands of Ethiopia. prepared for the Ethiopia Poverty Assessment. EARO, CIMMYT, funded by the European Söderbom, M. 2012. “Firm Size and Structural Union. Change: A Case Study of Ethiopia.” Journal of United Nations Children’s Fund. 2013. Female Genital African Economies, Vol. 21, AERC Supplement Mutilation/Cutting: A statistical overview and 2, pp. ii126–ii151 exploration of the dynamics of change. UNICEF, Sohnesen, Thomas P. 2013. 2011 Ethiopia Poverty New York. Map Technical Report. Mimeo, World Bank, Vandercasteelen, Joachim, Mekdim Dereje, Bart November, 2013 Minten, and Alemayehu Seyoum Taffesse. 2013. Spielman, David J., Derek Byerlee, Dawit Alemu, and Scaling-up adoption of improved technologies: Dawit Kelemework. 2010. “Policies to promote The impact of the promotion of row planting on cereal intensification in Ethiopia: The search for farmers’ teff yields in Ethiopia. LICOS Discussion appropriate public and private roles.” Food Policy, Paper Series Discussion Paper 344/2013 Elsevier, vol. 35(3), pages 185–194, June. Woldehanna, T. and A. Oskam. 2001. Income diver- Stifel, D. B. Minten, and B. Koro. 2012. Economic sification and entry barriers: evidence from the Benefits and Returns to Rural Feeder Roads: Tigray region of northern Ethiopia. Journal of Evidence from a Quasi-Experimental Setting in Food Policy. Ethiopia. ESSP 2 Working Paper 40. Woldehanna, Tassew, Retta Gudisa, Yisak Tafere and Taffesse, A. S., P. Dorosh, S. and Asrat. 2011. Crop Alula Pankhurst. 2011. Understanding Changes Production in Ethiopia: Regional Patterns and in the Lives of Poor Children: Initial Findings Trends. Ethiopia Strategy Support Program II from Ethiopia. Young Lives Round 3 Survey Working Paper No. 0016. Report. Taylor, J. E. and P.L. Martin. 2001. Human capi- World Bank & International Food Policy Research tal: Migration and rural population change. Institute. 2010. “Gender and governance in in G. Rausser and B. Gardner, eds., Handbook rural services: insights from India, Ghana, and of Agricultural Economics. Elsevier Science Ethiopia. Washington.” Washington DC: The Publishers, New York. World Bank. REFERENCES 189 World Bank & The ONE Campaign. 2014. “Levelling World Bank. 2009. World Development Report 2009. the Field. Improving Opportunities for Female World Bank. 2012. World Development Report 2012. Farmers in Africa.” Washington DC: The World Younger, Stephen D. and Artsvi Khachatryan. 2014. Bank. Fiscal Incidence in Armenia. Background Paper World Bank. 2013. Ethiopia Economic Update II: for World Bank Armenia Public Expenditure Laying the foundation for achieving middle- Review. (forthcoming) income status. Yilma, Zelalem, Anagaw Mebratie, Robert Sparrow, World Bank. 2014. Final Report of “Powering Up: Degnet Abebaw, Marleen Dekker, Getnet Alemu Social Empowerment for Vulnerable Girls.” A and Arjun S. Bedi. 2013. Coping with Shocks in project of the Italian Trust Fund of Children and rural Ethiopia. International Institute of Social Youth in Africa (CHYAO). Studies Working Paper 560, Rotterdam. P.O.Box 5515 Addis Ababa Bole Road/ Africa Avenue www.worldbank.org/ethiopia