Measuring Gender Equality OTHER TITLES IN THE ADePT SERIES Key Labor Market Indicators—Analysis with Household Survey Data: Streamlined Analysis with ADePT Software (2016) by Ina Pietschmann, Steven Kapsos, Evangelia Bourmpoula, Zurab Sajaia, and Michael Lokshin Simulating Distributional Impacts of Macro-dynamics—Theory and Practical Applications: Streamlined Analysis with ADePT Software (2014) by Sergio Olivieri, Sergiy Radyakin, Stainslav Kolenikov, Michael Lokshin, Ambar Narayan, and Carolina Sánchez-Páramo Analyzing Food Security Using Household Survey Data: Streamlined Analysis with ADePT Software (2014) by Ana Moltedo, Nathalie Troubat, Michael Lokshin, and Zurab Sajaia A Unified Approach to Measuring Poverty and Inequality—Theory and Practice: Streamlined Analysis with ADePT Software (2013) by James Foster, Suman Seth, Michael Lokshin, and Zurab Sajaia Health Equity and Financial Protection: Streamlined Analysis with ADePT Software (2011) by Adam Wagstaff, Marcel Bilger, Zurab Sajaia, and Michael Lokshin Assessing Sector Performance and Inequality in Education: Streamlined Analysis with ADePT Software (2011) by Emilio Porta, Gustavo Arcia, Kevin Macdonald, Sergiy Radyakin, and Michael Lokshin For more information about Streamlined Analysis with ADePT software and publications, visit www.worldbankgroup.org/adept. STREAMLINED ANALYSIS WITH ADePT SOFTWARE Measuring Gender Equality Josefina Posadas Pierella Paci Zurab Sajaia Michael Lokshin © 2017 International Bank for Reconstruction and Development / The World Bank 1818 H Street NW, Washington, DC 20433 Telephone: 202-473-1000; Internet: www.worldbank.org Some rights reserved 1 2 3 4 20 19 18 17 This work is a product of the staff of The World Bank with external contributions. The findings, interpretations, and conclusions expressed in this work do not necessarily reflect the views of The World Bank, its Board of Executive Directors, or the govern- ments they represent. The World Bank does not guarantee the accuracy of the data included in this work. The boundaries, colors, denominations, and other information shown on any map in this work do not imply any judgment on the part of The World Bank concerning the legal status of any territory or the endorsement or acceptance of such boundaries. Nothing herein shall constitute or be considered to be a limitation upon or waiver of the privileges and immunities of The World Bank, all of which are specifically reserved. Rights and Permissions This work is available under the Creative Commons Attribution 3.0 IGO license (CC BY 3.0 IGO) http://creativecommons.org/ licenses/by/3.0/igo. Under the Creative Commons Attribution license, you are free to copy, distribute, transmit, and adapt this work, including for commercial purposes, under the following conditions: Attribution—Please cite the work as follows: Posadas, Josefina, Pierella Paci, Zurab Sajaia, and Michael Lokshin. 2017. Measuring Gender Equality: Streamlined Analysis with ADePT Software. Washington, DC: World Bank. doi: 10.1596/978-1-4648-0775-6. License: Creative Commons Attribution CC BY 3.0 IGO Translations—If you create a translation of this work, please add the following disclaimer along with the attribution: This transla- tion was not created by The World Bank and should not be considered an official World Bank translation. The World Bank shall not be liable for any content or error in this translation. Adaptations—If you create an adaptation of this work, please add the following disclaimer along with the attribution: This is an adaptation of an original work by The World Bank. Views and opinions expressed in the adaptation are the sole responsibility of the author or authors of the adaptation and are not endorsed by The World Bank. Third-party content—The World Bank does not necessarily own each component of the content contained within the work. The World Bank therefore does not warrant that the use of any third-party-owned individual component or part contained in the work will not infringe on the rights of those third parties. The risk of claims resulting from such infringement rests solely with you. If you wish to reuse a component of the work, it is your responsibility to determine whether permission is needed for that reuse and to obtain permission from the copyright owner. Examples of components can include, but are not limited to, tables, figures, or images. All queries on rights and licenses should be addressed to World Bank Publications, The World Bank Group, 1818 H Street NW, Washington, DC 20433, USA; e-mail: pubrights@worldbank.org. ISBN (paper): 978-1-4648-0775-6 ISBN (electronic): 978-1-4648-0776-3 DOI: 10.1596/978-1-4648-0775-6 Cover images: Foreground: Man and woman, Benin, © Arne Hoel/World Bank. Further permission required for reuse. Photo ID: AH-BJ-100917-5283. Background: © iStockphoto.com/Olga Altunina. Used with permission. Further permission required for reuse. Cover design: Kim Vilov Library of Congress Cataloging-in-Publication Data has been requested. Contents Foreword .................................................................................................... xi Preface ...................................................................................................... xiii Abbreviations ........................................................................................ xvii PART I: Introducing ADePT Gender Software ......................................... 1 Chapter 1 Applying a Household-Centered Framework to Gender Analysis ........ 3 Framework ...........................................................................................................4 What This Manual Does and How to Use It ...................................................10 Notes..................................................................................................................12 References..........................................................................................................13 Chapter 2 Working with ADePT Software............................................................... 17 Conventions Used in This Chapter .................................................................18 Installing ADePT ..............................................................................................18 Launching ADePT ............................................................................................19 Overview of the Analysis Procedure.................................................................20 Specifying Datasets............................................................................................21 Mapping Variables ............................................................................................23 Selecting Tables and Graphs ............................................................................26 Generating the Report ......................................................................................29 v Contents Examining the Output ......................................................................................30 Working with Variables ....................................................................................30 Setting Parameters ............................................................................................36 Working with Projects ......................................................................................36 Adding Standard Errors or Frequencies to Outputs .........................................37 Applying If-Conditions to Outputs ..................................................................38 Generating Custom Tables ...............................................................................39 Note ...................................................................................................................41 Chapter 3 Data Preparation ...................................................................................... 43 Household Surveys ............................................................................................45 Main Variables ..................................................................................................47 Outcome Variables ............................................................................................55 Parameters .........................................................................................................77 Notes..................................................................................................................79 References..........................................................................................................82 PART II: Producing a Country Gender Diagnostic ................................. 87 Chapter 4 How to Interpret the Results of the Country Gender Diagnostic ....... 89 Demographic and Regional Characteristics .....................................................91 Human Capital ................................................................................................104 Economic Opportunities .................................................................................114 Country Context: Voice, Agency, and Participation ....................................132 Notes................................................................................................................140 References........................................................................................................141 Chapter 5 Technical Notes for the Country Gender Diagnostic .......................... 145 Household Labor Supply Model .....................................................................146 Intrahousehold Allocation Model ..................................................................152 Notes................................................................................................................154 References........................................................................................................155 PART III: Analyzing Gender in Labor Markets ..................................... 159 Chapter 6 How to Interpret the Results of Labor Market Analysis .................... 161 Earnings Inequality..........................................................................................162 Earnings Decompositions ................................................................................175 vi Contents Notes................................................................................................................200 References........................................................................................................201 Chapter 7 Technical Notes on Labor Market Analysis ......................................... 205 Discrimination Theories .................................................................................205 Earnings Equations ..........................................................................................213 Nonrandom Selection into the Labor Force ..................................................222 Notes................................................................................................................228 References........................................................................................................229 PART IV: Conclusions ............................................................................. 235 Chapter 8 Reflections on What ADePT Gender Does and What It Does Not Do ..........................................................................................237 References........................................................................................................239 APPENDIXES ........................................................................................... 241 Appendix A Fields, Variable Definitions, and Variable Requirements ................... 243 Appendix B Demographic and Health Survey Agency Variables ........................... 249 Appendix C Tests of Statistical Significance ............................................................ 253 Means Tests .....................................................................................................253 Statistically Different Distributions ................................................................259 Notes................................................................................................................261 Appendix D Juhn-Murphy-Pierce Decomposition .................................................... 263 Appendix E Mathematical Derivation of the Mills Ratio Variable Included in the Mincer Equation .......................................................................... 265 Notes................................................................................................................269 Index ........................................................................................................ 271 vii Contents Boxes 1.1: A Deeper Look at Agency........................................................................6 3.1: Sampling Design .....................................................................................48 3.2: Mapping Panama’s Education System to the International Standard Classification of Education .....................................................58 3.3: Revised ILO Statistical Standards for Measuring Employment and Work ................................................................................................62 4.1: Other ADePT Modules: Education and Health ..................................107 4.2: Demographic and Health Survey Indicators on Contraception .........135 6.1: Interpreting Different Measures of the Gender Gap in Pay for the Whole Distribution of Earnings ...............................................170 6.2: ADePT Gender Model Specifications for Earnings Equations ...........176 7.1: Interpreting Coefficients in a Log-Linear Regression..........................219 7.2: Technical Terms Related to Nonrandom Selection of Women into the Labor Force...............................................................223 C.1: Test of Variance Equality for Two Populations...................................257 Figures 1.1: WDR Framework for Analyzing Gender Equality ...................................6 4.1: Demographic Typology ..........................................................................94 4.2: ADePT Gender Figure 1b, Panama 2008 ..............................................96 4.3: ADePT Gender Figure 1e, Nepal 2010–11 .........................................102 4.4: Population Pyramid in Countries with Gender Imbalance at Birth..... 103 4.5: Sex Ratio by Age and Urban and Rural Areas for Kenya, 2005, and Russian Federation, 2006 ..............................................................104 4.6: Primary and Secondary Net Enrollment Rates, by Gender and Region, 1999 and 2007 .................................................................111 4.7: U-Shape Relationship between Female Labor Force Participation and GDP, 1990 and 2010...............................................122 4.8: ADePT Gender Figure 5a, Panama, 2008 ...........................................124 4.9: ADePT Gender Figures 5b and 5c, Panama 2008 ...............................127 4.10: Female Labor Force Participation over the Life Cycle ........................130 4.11: Share of Women Who Experience Overlapping Agency Deprivations in Three Domains (Percentage).....................................136 4.12: Mean Age at Marriage for Women and Men and the Difference in Years, Countries Where Women Marry on Average at Age 20 or Earlier, 2002–06 ...............................................138 viii Contents 4.13: Women’s Control over Household Decisions......................................139 5.1: Household Time Allocation Problem ..................................................149 5.2: Budget Constraints in Households with One Female Adult and One Male Adult ...................................................................................151 6.1: Lorenz Curve ........................................................................................167 6.2: ADePT Gender Figure 12b, Panama 2008 ..........................................173 6.3: ADePT Gender Figure 12c, Panama 2008 ..........................................175 6.4: Illustration of Unobserved Skills .........................................................188 6.5: Illustration of Bias from a Lack of Common Support .........................190 6.6: Oaxaca-Blinder Decomposition, ADePT Figure 13a, Nepal 2010–11 .................................................................................... 194 6.7: ADePT Gender Figure 16a, Nepal 2010–11 .......................................197 7.1: Graphic Representation of Employer Discrimination .........................209 7.2: Overcrowding Model Resulting in Discrimination .............................212 7.3: Graphic Visualization of Mincer Equation ..........................................215 7.4: Graphic Representation of Selection Bias ...........................................225 C.1: Rejection and Acceptance Regions .....................................................259 E.1: Graphic Representation of the Reservation Wage ..............................266 Map 4.1: Female Labor Force Participation around the World ..........................121 Tables 1.1: Key Resources for ADePT Gender .......................................................10 3.1: Usual Topic Coverage of Survey Instruments ......................................46 B3.3.1: Comparison of New and Previous Definitions ......................................63 3.2: ILO Conceptual Framework: Informal Employment .............................66 4.1: ADePT Gender Table 1b, Panama 2008 ...............................................98 4.2: Countries by Share of Women in Total Population Living in Poor Households, 1999–2008 ................................................................99 4.3: ADePT Gender Table 1e, Panama 2008 .............................................100 4.4: ADePT Gender Table 1f, Panama 2008..............................................102 4.5: ADePT Gender Table 2a, Nepal 2010 ................................................109 4.6: ADePT Gender Table 3b, Nepal 2010 ................................................112 4.7: Women Receiving Prenatal Care, Skilled Assistance at Birth, and Deliveries in Health Facilities, by Region, 1996 and 2000–08 ..........112 4.8: ADePT Gender Table 4a, Panama 2008 .............................................121 ix Contents 4.9: ADePT Gender Table 4b, Panama 2008 .............................................123 4.10: ADePT Gender Table 5a, Panama 2008 ............................................125 4.11: Distribution of Women and Men by Type of Work............................126 4.12: ADePT Gender Table 6a, Panama 2008 .............................................128 4.13: ADePT Gender Table 7a, Panama 2008 .............................................129 4.14: ADePT Gender Table 8a, Panama 2008 .............................................133 4.15: ADePT Gender Table 10, Nepal 2010–11 ..........................................137 4.16: ADePT Gender Table 11g, Nepal 2010–11 ........................................138 5.1: Daily Hours Spent in Household Work, Paid Work, and Leisure for Men and Women ...............................................................148 6.1: ADePT Gender Table 12c, Panama 2008 ...........................................171 6.2: ADePT Gender Table 12b, Panama 2008 ...........................................172 6.3: ADePT Gender Table 12d, Panama 2008 ...........................................174 6.4: Most Commonly Used Decomposition Methodologies.......................180 6.5: ADePT Gender Table 13a, Nepal 2010–11 ........................................193 6.6: ADePT Gender Table 13b, Nepal 2010–12 ........................................195 6.7: ADePT Gender Table 16a, Panama 2008 ...........................................196 6.8: ADePT Gender Table 15a, Nepal 2010–11 ........................................199 x Foreword Gaps between men and women permeate everyday life—and yet they are often difficult to pin down and quantify. Every year, countries around the world observe equal pay day, a symbolic day on which female earnings “catch up” with male earnings from the past year. However, calculating the gender gap in earnings can seem almost as daunting as efforts to close it, given a myriad of competing measures and computation methods. This book seeks to help its readers navigate the sometimes confusing world of measuring and analyzing gender equality on the basis of household survey data. It is designed as a manual of ADePT Gender, a free software tool developed by the World Bank’s Development Data Group and Gender Cross-Cutting Solution Area, which automates and simplifies the pro- duction of standardized tables and graphs related to the analysis of gaps between males and females. In addition, this book provides the core eco- nomic context needed to interpret—and sometimes challenge—measures of gender equality. Its primary audience is data analysts, who wish to perform hands-on analysis of household survey data to obtain a better understanding of the existing gender gaps within or across countries. However, other data users in government, media, or academia may also find it a useful read. The ADePT Gender software is divided into two core modules. The first module produces a country gender diagnostic for the three core dimensions xi Foreword of gender equality highlighted in the 2012 World Development Report: human capital (or endowments), economic opportunities, and voice and agency. The second module zooms in on gender gaps in labor market outcomes, using more advanced decomposition techniques from the labor economics literature. We hope this book will stimulate the analysis of country gender gaps and contribute to informed decision and policy making. Caren Grown Haishan Fu Senior Director, Gender Director, Development Data Group The World Bank Group The World Bank Group xii Preface Gender equality matters for both development outcomes and policy mak- ing. It is a core development objective in its own right but it is also smart economics as it enhances average productivity and improves prospects for the next generation and for the quality of societal policies and institutions. However, differences between men and women are observed in several dimensions of social and economic life, and throughout the life cycle. The past 50 years have seen marked improvement in the lives of girls and women around the world. Across the globe, more girls and women are educated than ever before, more girls are in school than boys, and women make up nearly half of the global labor force. In some areas, however, progress toward reducing gender gaps has been more limited, especially among disadvan- taged groups such as the poor, women and girls living in remote areas, or those belonging to minority groups. As awareness of the importance of gender equality grows among researchers, development practitioners, and policy makers, so does the demand to better understand the patterns of progress and the nature and sources of persistent gaps. This information is fundamental to ensure that the limited resources are channeled to areas where progress has been harder to achieve and to priority areas of interventions. The increased availability of disaggregated statistics for men and women on many key development indicators is at the same time a reflection of, and fuel for, the growing xiii Preface interest in evidence-based, gender-sensitive policy making. However, access to data in itself is not enough to ensure better understanding of the magni- tude, dynamics, and drivers of gender inequality, especially in the context of limited capacity, and even more limited resources. There is also a clear need for standard approaches, common methodologies, and analytical tools that facilitate the use of these data for systematic and comprehensive diagnostic work. ADePT Gender is just such a tool. Building on the framework pro- posed by the World Development Report (WDR) 2012: Gender Equality and Development, ADePT Gender is designed to guide the broad and diverse gender and development community through the complexity of the diag- nostics of gender inequality and its dynamics. The intuitive software consists of two parts. The first part uses simple statistics and tabulations to profile the extent and dynamics of gender inequalities across three dimensions— namely endowments, economic opportunities, and agency. The second part focuses on gender gaps in economic opportunities by analyzing gender dis- advantages in the labor market and, in particular, wage inequality. Its focus is on being user-friendly and comprehensive, although not exhaustive. This manual provides a guide to working with ADePT Gender with a particular emphasis on helping the wide community of users to interpret the large volume of statistical information generated by the software. Contrary to other ADePT modules, it does not detail the mechanisms behind gender differences in outcomes, as these are extensively covered in the WDR 2012 and in its companion reports. Table 1.1 lists the main references available to the ADePT Gender users on the drivers of gender inequality and offers a short description of how they relate to the WDR framework. ADePT Gender and this manual were made possible by the efforts of the many who have provided invaluable technical support and encourage- ment. This manual and the ADePT Gender module build on an earlier, more limited version of the software developed by Gisela García, Gayatri Koolwal, and Nistha Sinha. Andrea Atencio, Jenifer Golan, Francisco Haimovich, Giulia Mancini, Julieth Santamaria, and Chimedkham Zorigtbaatar went the extra mile in assisting us with data management and analysis of vari- ous countries and surveys in preparation for this version of the software and the manual. Thanks to their tenacious efforts, we have a tool that is flexible and user friendly, while covering a very wide range of indicators and dimensions. Particular thanks goes to our peer reviewers Georgia De Paoli, Elena Ferreras Carreras, and Gayatri Koolwal as well as to Isis Gaddis, xiv Preface Eliana Rubiano Matulevich, and Elena Bardasi, who provided extensive comments at different stages during the development of the software and the preparation of this manual. The insightful feedback we received has raised the quality of both products and their interest and accessibility to a wider audience. Finally, we are grateful for the support we have received over the years from the management of the Gender and Development unit and, in particular, Mayra Buvinic, who supported the idea in its initial stages and Caren Grown, who oversaw the completion of the tool and the manual. In finalizing ADePT Gender and this manual, we benefitted from the feedback on earlier versions received during training events in Armenia, the Lao People’s Democratic Republic, and Timor-Leste, and from its use in selected poverty assessments. We thank Helle Buchave and Nistha Sinha for giving us these opportunities. Various presentations and training ses- sions during PREM Week 2012 and 2013 at the World Bank headquarters in Washington, DC, provided additional useful feedback. The invaluable support of the World Bank ADePT team in the technical aspects of the software development is acknowledged with thanks. xv Abbreviations ADePT Automated DEC Poverty Tables AIDS acquired immune deficiency syndrome BCG Bacillus Calmette–Guérin (vaccination against tuberculosis) CMC century-month code CPR contraceptive prevalence rate DHS Demographic and Health Survey DPT diphtheria, pertussis, and tetanus (vaccination) GER gross enrollment rate HIV human immunodeficiency virus ICLS International Conference of Labour Statisticians IHSN International Household Survey Network ILO International Labour Organization ISCED International Standard Classification of Education ISCO International Standard Classification of Occupations ISIC International Standard Industrial Classification LF labor force MMR measles, mumps, and rubella (vaccination) NER net enrollment rate OECD Organisation for Economic Co-operation and Development PISA Programme for International Student Assessment xvii Abbreviations SPSS Statistical Package for the Social Sciences (software package for statistical analysis) Stata data analysis and statistical software STEP Skills Toward Employment and Productivity TIMSS Trends in International Mathematics and Science Study UN United Nations UNESCO United Nations Educational, Scientific, and Cultural Organization WDR World Development Report WHO World Health Organization xviii PART I Introducing ADePT Gender Software Part I of this book introduces the reader to ADePT software and the ADePT Gender module. ADePT software allows users to analyze microdata from sources such as household surveys to generate print-ready, standard- ized tables and charts. It can also be used to simulate the effect of economic shocks, farm subsidies, cash transfers, and other policy instruments on pov- erty, inequality, and labor. In this case, ADePT can be used to analyze and create standardized tables and charts to construct gender indicators. ADePT software can be customized to the user’s needs. In the ADePT Gender module, the user can disaggregate data across men and women, as well as male versus female heads of household. The software can also disaggregate data across different geographic regions and socioeconomic and demographic backgrounds. The software automates the analysis, helps minimize human errors, and encourages development of new economic analysis methods. ADePT sup- ports datasets in Stata, the Statistical Package for the Social Sciences (SPSS), and tab-delimited text formats. ADePT incorporates Numerics by Stata (installed with ADePT) as its computational engine. For each run, ADePT produces one output file—containing the user’s selection of tables and graphs, an optional original data summary, and errors and notifications—in 1 Introducing ADePT Gender Software: Part I Microsoft Excel format. Optionally, tables of standard errors and frequencies can be added to a report. To learn more about the uses and requirements of ADePT software, the user may consult the ADePT User Guide.1 Chapter 1 introduces the software and the logic of the product. ADePT Gender uses the framework of the World Development Report 2012: Gender Equality and Development (World Bank 2012b) to organize the presentation of tables and graphs. Applying this framework is also recommended when writing the analysis of the results. This framework provides comprehensive coverage of all the dimensions of gender inequalities: endowments, eco- nomic opportunities, and agency. The results can then help identify areas for further analysis and priority actions. For instance, the systematic review of gaps helps answer such questions as: • Should a country focus on addressing missing women or gender gaps in education? • Is domestic violence a problem in the country? • Are gender gaps in employment and wages important? • How much of the gender gap can result from occupational segregation? The next two chapters are intended to familiarize the user with ADePT Gender software. Chapter 2 helps the user understand how to install the program, upload the data, and fill in the variable fields. Chapter 3 helps the user prepare the data and introduces several variable definitions to ensure that the user has no misunderstanding in how the program interprets the data. All these definitions are based in standard practices and conven- tions adopted by international organizations such as the United Nations and the International Labour Organization. This step is critical for the validity of the results. Even when ADePT Gender software carries out some internal checks to establish the validity of the data, it is the user’s responsi- bility to upload data that respond to the definitions of the variables requested by ADePT Gender and are suitable for interpreting the results correctly. Note 1. Michael Lokshin, Sergiy Radyakin, Zurab Sajaia, and William Creitz. 2013. ADePT User Guide. Version 5 (Washington, DC: World Bank). 2 Chapter 1 Applying a Household- Centered Framework to Gender Analysis Gender equality matters for its intrinsic and instrumental value. Gender equality matters intrinsically because the ability to live a life of one’s choosing is a basic human right and should be available to anyone, regardless of one’s sex or gender. Gender equality also matters because it contributes to economic efficiency and the achievement of other desirable development outcomes. Gender equality can contribute to economic development in three ways. First, it can remove barriers that prevent women from accessing edu- cation, economic opportunities, and productive inputs that generate eco- nomic gains. Second, women’s gains promote other desired development outcomes, such as increased economic productivity, as well as child nutri- tion, health, and education, which improve the welfare of future generations. Third, greater equality of female participation in community and political organizations leads to more inclusive and representative institutions, which contributes to development. These messages have been discussed extensively among researchers, development practitioners, and policy makers. Recently, the World Development Report (WDR) 2012: Gender Equality and Development (World Bank 2012b) has resumed the conversation and contributed to establishing guidelines for how to analyze and measure gender equality in a comprehensive manner. This report not only exhaustively examined all 3 Introducing ADePT Gender Software: Part I types of outcomes related to gender equality but also advanced in bringing together findings and methodologies of analyses from economics and other social fields of studies, which are usually disconnected in this complex area of policy work. More important, this report also presented a set of policy actions for tackling gender inequalities. ADePT Gender is designed to help this broad community working on gender equality and development to measure gender equality using the framework proposed in this influential report. With that purpose, this manual is organized in three parts: part I focuses on applying the framework introduced in the WDR 2012: Gender Equality and Development using simple statistics and tabulations across three dimensions, namely, endowments, economic opportunities, and agency; part II covers the output that produces a country gender diagnostic; and part III goes deeper in analyzing labor- market inequalities, particularly wage inequality. This manual and ADePT Gender software refer to gender equality as the equality of outcomes between men and women, even if gender does not refer to men and women. Gender denotes the social, behavioral, and cultural attributes, expectations, and norms associated with being a woman or a man. Gender equality refers to how these aspects determine how women and men relate to each other and to the resulting differences in power between them. This approach is also consistent with that proposed by the WDR and is also applied by ADePT Gender.1 The rest of this chapter discusses how ADePT Gender uses the WDR’s framework. It is not the objective of this chapter to summarize the messages of the WDR. Given the wide impact of the product, there are multiple prod- ucts used for its dissemination that the user can consult and that accom- modate different users’ needs, from quick overviews to extensive discussions.2 The two following chapters provide information for working with the software. Chapter 2 introduces the ADePT software and the gender module, whereas chapter 3 describes the data sources that are suited for use in ADePT Gender. It also defines the concepts needed for proper data manage- ment that must be undertaken before using ADePT Gender. Framework Differences between men and women are observed in several dimensions of social and economic life and throughout the life cycle. The examples are 4 Chapter 1: Applying a Household-Centered Framework to Gender Analysis numerous: in many countries, boys often have better health outcomes than girls and are more likely to achieve higher levels of education than girls. Such inequality persists later on in life: young women are more likely than young men to be unemployed, and a higher proportion of women compared with men do not participate in the labor force. These differences between men and women and between boys and girls are the result of complex inter- actions between households, markets, and formal and informal institutions. Sometimes, the inequalities may be detrimental to men: in various Caribbean countries, poor boys have less schooling than girls, as they drop out to work in agriculture; or in many eastern European countries, adult men have a low life expectancy associated with various life hazards and health problems. One way to depict the complex interactions between households, mar- kets, and institutions and their effects on equality of gender outcomes is to examine how households function. Families decide how much to spend on boys’ and girls’ education and health, how to allocate tasks inside and outside the household, and other matters that influence gender outcomes. This household-centered framework has proved useful for the economic analysis of gender equality and has been the basis of World Bank milestones in promoting gender equality in policy recommendations and allocation of resources. Households make choices on the basis of preferences shaped by social norms, market incentives, and constraints that result from markets and for- mal institutions. The WDR summarizes these interactions in the simple diagram presented in figure 1.1. The interactions of households, markets, and institutions generate growth, which in turn contributes to gender equal- ity as income and economic development alter some of the constraints. At the same time, greater gender equality contributes to economic efficiency and growth. The approach followed by ADePT Gender is to work over this framework and present a country diagnostic on gender equality that looks at gender outcomes from the perspective of households and individuals as household members. This is one way to condense the information concern- ing the interactions between households, markets, and institutions. A country diagnostic on gender equality should be comprehensive, covering both social and economic issues. Following the WDR, ADePT Gender groups outcomes in three dimensions: human endowments (or capital), economic opportunities, and agency (box 1.1).3 All of these aspects are interrelated and matter for individual welfare, gender equality, and economic development. Building on the conceptual and empirical work of others, ADePT Gender was developed to maximize the use of household-level data 5 Introducing ADePT Gender Software: Part I and to create ready-to-print reports that allow users to easily visualize gender inequalities. ADePT Gender is a tool for conducting country diagnostics using micro- data from different types of household surveys. The results allow users to identify broad areas for further analysis or for public action. At the same time, using the WDR helps take advantage of the report’s in-depth analysis Figure 1.1: WDR Framework for Analyzing Gender Equality R EQUALITY po NDE lic GE ies INFORMAL INSTITUTIONS ECONOMIC OPPOR- TUNITIES MARKETS HOUSEHOLDS ENDOW- AGENCY MENTS FORMAL INSTITUTIONS GR OW TH Source: World Bank 2012b. Box 1.1: A Deeper Look at Agency The term agency can be defined as the ability of individuals or groups to give voice to and act on their preferences and to influence outcomes that affect them and others in society. Agency is affected by and also affects individuals’ ownership of and control over endowments and their access to economic opportunities (Kabeer 1999). Within a household or partnership, one’s relative power affects the strength of one’s voice and influence in household decisions, such as how to spend or invest family resources. Similarly, at the community or societal level, the relative power of individuals and groups affects their ability to act on their preferences and to influence (continued) 6 Chapter 1: Applying a Household-Centered Framework to Gender Analysis Box 1.1: A Deeper Look at Agency (continued) outcomes in the economic, social, and political domains. The relative power of differ- ent members of society, which often differs systematically by gender, reflects a complex combination of one’s personal characteristics, prevailing social norms, and the broader legal and institutional environments. The ability to act on one’s preferences, regardless of one’s gender, and to translate those preferences into a desired outcome is a development objective in its own right. Development not only involves raising incomes or reducing poverty but also involves a process of expanding freedoms and choices available to all people (Sen 1999). Agency is a measure of a person’s well-being, reflecting the ability to achieve as well as actual achievements (Sen 1992). Source: Mason and others 2012. and explanation of the mechanisms that generate the observed results. The WDR also concludes with a chapter containing a rich set of policy recom- mendations organized around the framework. It is recommended that the users of ADePT Gender software build on the lessons and policy recommen- dations of this report, as it complements this product. Endowments The term endowment refers to investments in nutrition, health, and educa- tion, beginning with the right to be born. The WDR chooses this term instead of human capital, a concept introduced by Gary Becker (1957), because it aims to capture differences that arise even before birth, and not just investments that take place during the life cycle. Education and health investments have a huge effect on individuals’ ability to function and reach their potential in life and society. Narrowing gender dif- ferences in endowments is important not only because all persons have the same right to health and education, but also because improvements in women’s edu- cation and health will have positive effects on their children, which will in turn result in future economic growth. Evidence shows that a mother’s nutritional status is positively associated with her children’s health and survival, and a mother’s education is positively linked to a range of health benefits for her chil- dren and to their educational attainment. There is no doubt that poor health outcomes in adulthood affect economic outcomes, which are reflected in health- related absences from the labor force, fewer work hours, and lower earnings. 7 Introducing ADePT Gender Software: Part I ADePT Gender allows users to produce a set of tables that compare health and education outcomes for boys and girls and for men and women. The standardized graphs and charts produced include data that are most likely to be available in multitopic household surveys. Chapters 2 and 3 further detail how users of the ADePT software can choose independent variables to produce customized tables and charts. Economic Opportunities The term economic opportunity refers to access to both productive inputs and productive employment, and to the result of economic activities in the form of productivity and earnings. For an economy to function at its maximum potential, women’s skills need to be used productively. Instead, women’s labor tends to be underused or misallocated. For example, women are more likely to occupy low- productivity occupations, do unpaid family work, or hold informal jobs. Many times, women have less access to productive assets, such as land and fertilizers, which in turn affects their productivity in self-employment as entrepreneurs and farmers. In agriculture, farms managed by women typi- cally render lower yields than those operated by men, even if plots belong to the same household. That outcome is partly explained by different access to assets and agricultural inputs. In urban areas, female-managed firms tend to have lower levels of value added per worker. That is also explained by the fact that women often work predominantly in low-productivity sectors and have less access to credit. These differences result from gender differences in endowments and in access to inputs, including time that men and women devote to household duties and market-paid activities, and from market and institutional failures. For example, differences in preferences—which may be specific to the individual but are also shaped by society—result in gaps in employment, occupations, and earnings and persist throughout the labor market. ADePT Gender attempts to capture statistical differences between men and women in such outcomes as employment, unemployment, and occupa- tional segregation. The software also allows users to explore differences in access to productive inputs, including both human capital and physical assets. Since women’s constraints vary along the life cycle, these differences in outcomes are also provided for various age groups, marital status, and fertility outcomes. 8 Chapter 1: Applying a Household-Centered Framework to Gender Analysis Because ADePT Gender aims to understand gender differences in labor- market outcomes, part II of this manual is dedicated to producing measures of inequality in the labor market as well a set of standard decomposition methodologies that allow users to disentangle the contribution of differ- ences in endowments—usually called the composition effect—and differ- ences in labor market payoff of these endowments—usually called the wage structure effect. Agency The term agency refers to one’s ability to make choices and to transform them into desired actions and outcomes. Agency can be exerted at the individual level, within the family, or in society (Kabeer 1999; Sen 1985; World Bank 2012b). In household surveys, agency is typically measured through individuals’ decision-making roles within the household, the community, or both, although their desired goals for different areas of decision making are usu- ally not elicited. Independent of the income level or a country’s economic development, women are less involved than men in decision making in the household, community, and society. In many countries, a large percentage of women do not have access to household resources or do not contribute to decisions about how to use household income. However, evidence shows that when women participate in household spending, greater household expenditures are devoted to items related to nutrition, health, and educa- tion, which later translate into better outcomes for children. Conversely, in many countries, women do not participate in government, political parties, or even civil organizations, which perpetuates inequality. In politics, women’s concerns are not represented if men cannot understand or defend female issues. Women also participate in weaker or even segre- gated networks, which makes it more difficult for them to participate in the institutional machine and to generate change. Most important, women and men internalize social norms and institutional incentives in ways that affect their expectations, aspirations, and choices. This in turn complicates the measurement of agency itself in surveys and the work of the analyst, as it becomes extremely complex—if not impossible—to disentangle “inherent” preferences from constraints that stem from markets and institutions. To complicate matters even further, measuring agency is not simple, and agency is often confounded with other related concepts, such as empowerment. 9 Introducing ADePT Gender Software: Part I Much has been written about measurement challenges and solutions for agency and empowerment (Annan and others 2016; Klugman and others 2014; Narayan 2005; World Bank 2012b). The agreement so far is that most of the available variables are weak proxies for agency. ADePT Gender uses the proxies for agency that are easily accessed by the user. Thus, the results on agency need to be interpreted bearing in mind the assumptions linking the proxy variables with the agency features they try to measure. What This Manual Does and How to Use It This manual provides a guide to working with ADePT Gender software and is intended to help the user interpret the results produced. As opposed to other ADePT modules, it does not detail the mechanisms that lead to gender differences in outcomes, as those are fully developed in the 2012 WDR as well as in its companion reports (see table 1.1). Moreover, the last Table 1.1: Key Resources for ADePT Gender Reference Highlighted contribution • World Development Report 2012: The 2012 WDR takes stock of the evidence and develops Gender Equality and Development a framework for analyzing gender equality, which ADePT (WDR) (World Bank 2012b) Gender uses to organize output. The 2012 WDR provides a wealth of statistics that can be used as comparators/ reference values of ADePT Gender’s output. This report also provides a set of policy recommendations. • The World’s Women 2010 (UN 2010) This excellent report provides reference values for many indicators produced by ADePT Gender. • Global Gender Gap Report: 2014 The 2014 report contains a wealth of indicators that can (World Economic Forum 2014) be used to provide reference values and complementary information from other sources. Regional WDR companion reports • Toward Gender Equality in East The framework of the 2012 WDR is used to more deeply Asia and the Pacific (Mason and analyze countries of the East Asia and Pacific region. It others 2012) provides reference values for many outcomes of ADePT Gender for those countries. • Opening Doors: Gender Equality The framework of the 2012 WDR is used to more deeply and Development in the Middle East analyze countries of the Middle East and North Africa and North Africa (World Bank region. It provides reference values for many outcomes 2013b) of ADePT Gender for those countries. • Work and Family: Latin American Using the household as the center of analysis and based and Caribbean Women in Search of on a rich set of household surveys for Latin America and a New Balance (Chioda 2016) the Caribbean, this report discusses many gender inequalities along the dimensions of the 2012 WDR. It also presents results on wage decompositions using Ñopo’s methodology (see chapter 6). (continued) 10 Chapter 1: Applying a Household-Centered Framework to Gender Analysis Table 1.1: Key Resources for ADePT Gender (continued) Reference Highlighted contribution • Enhancing Women’s Voice, Agency A companion to the 2012 WDR, this report details five and Participation in the Economy: dimensions of agency in Egypt, Arab Rep.; Jordan; Studies in Egypt, Jordan, Morocco, Morocco; Tunisia; and Turkey in relation to economic Tunisia and Turkey (EBRD 2015) participation of women. Thematic companion reports or related references • Getting to Equal: Promoting Gender This companion report to the 2012 WDR analyzes in Equality through Human more detail the gender gaps in endowments. Development (World Bank 2011b) • On Norms and Agency: This companion report to the 2012 WDR covers the Conversations about Gender background qualitative studies on social norms and Equality with Women and Men in agency and further discusses measurement issues for 20 Countries (Munoz Boudet and this dimension of the framework. others 2012) • Voice and Agency: Empowering This report goes deeper into four of the five dimensions Women and Girls for Shared of agency described in the 2012 WDR—women’s control Prosperity (Klugman and others 2014) over assets, control over family formation, freedom from domestic violence, and freedom of physical mobility. • Empowering Women: Legal Rights This report analyzes the links between legal rights and a and Economic Opportunities in set of economic opportunities outcomes for African Africa (World Bank 2011a) countries. Country gender diagnostics using WDR framework • Russian Federation Country Gender Chapter 1 of the country gender assessment takes stock Assessment: Main Report (Munoz of gender inequalities over the three dimensions of Boudet and Posadas 2014) outcomes proposed by the 2012 WDR. Most of the tables and graphs included in this chapter can be produced with ADePT Gender software. • Bosnia and Herzegovina: Gender The country gender assessment takes stock of gender Disparities in Endowments, Access inequalities over the three dimensions of outcomes to Economic Opportunities and proposed by the 2012 WDR. Most of the tables and Agency (Cancho and Elwan 2015) graphs included in this chapter can be produced with ADePT Gender software. • Country Gender Assessment: The report aims to assess gender imbalances in the Economic Participation, Agency areas of economic participation in the labor market, and Access to Justice in Jordan agency, and access to justice in Jordan. Many of the (World Bank 2013a) tables and graphs included in this assessment can be produced with ADePT Gender software. • Papua New Guinea: Country The assessment describes the gender dimensions of Gender Assessment for the Period Papua New Guinea’s development challenges and 2011–2012 (World Bank 2012a) strategies. Chapters 2 and 3 detail access to education and health as well as employment, livelihood, and economic resources, respectively. Many of the tables and figures included in these chapters can be produced using ADePT Gender software. • Vietnam: Country Gender The assessment highlights progress and challenges to Assessment (World Bank 2011c) gender equality in Vietnam over the past several decades. Several of the tables and figures produced in this report, particularly those in chapters 3 and 4, can be produced using ADePT Gender software. Note: Dimensions of agency include (a) access to and control over resources, (b) freedom from the risk of violence, (c) freedom of movement, (d) decision making over family formation, and (e) a voice in society and influencing policy. 11 Introducing ADePT Gender Software: Part I chapter of the WDR contains a rich set of policy recommendations based on the evidence collected from countries all over the world. The WDR companion reports build on the same framework and describe in detail the gender realities of countries in their regions or in a particular topic or dimension of the framework—education, jobs, or agency. ADePT Gender software users have this wealth of existing material at their disposal— table 1.1 lists several of the relevant references and country assessments, including a short description of how those documents relate to the WDR framework. Moreover, since the launch of the WDR and in parallel to the develop- ment of ADePT Gender, a few countries have conducted individual country gender diagnostics that use the WDR framework and present tables and graphs similar to those produced by ADePT Gender software. Those country studies constitute excellent guides for ADePT Gender users, as they illus- trate how to use and interpret ADePT Gender output. The recommendation to the user is to work with the complementary studies when producing the report based on ADePT Gender output to have a rich description of the results and to link them to policy recom- mendations. Part II explains the main indicators produced to cover all the dimensions of gender equality. ADePT Gender software allows the user to produce many of the 52 minimum core gender indicators agreed by the United Nations Interagency and Expert Group on Gender Statistics.4 Notes 1. See box 1 in World Bank (2012b) for more details about the concept of gender in this setup. 2. The various dissemination outputs can be found on the website associ- ated with the report (World Bank 2012b). 3. The definition of agency comes in the following paragraphs and more extensively in box 1.1. 4. For more details on the minimum core gender indicators, see the report of the International Household Survey Network (IHSN 2015) and the paper on mapping gender gaps by Buvinic, Furst-Nichols, and Koolwal (2014). 12 Chapter 1: Applying a Household-Centered Framework to Gender Analysis References Annan, Jeannie, Aletheia Donald, Kathryn Falb, Gayatri Koolwal, and Markus Goldstein. 2016. “Measuring Women’s Agency.” Working Paper, Gender Innovation Lab, World Bank, Washington, DC. Becker, Gary S. 1957. The Economics of Discrimination. Chicago: University of Chicago Press. Buvinic, Mayra, Rebecca Furst-Nichols, and Gayatri Koolwal. 2014. “Mapping Gender Data Gaps.” Data2X, Washington, DC. http://data2x .org/wp-content/uploads/2014/11/Data2X_MappingGenderDataGaps _FullReport.pdf. Cancho, Cesar A., and Nihal Elwan. 2015. Bosnia and Herzegovina: Gender Disparities in Endowments, Access to Economic Opportunities and Agency. Washington, DC: World Bank. http://documents.worldbank.org /curated/en/2015/07/24811714/bosnia-herzegovina-gender-disparities -endowmentsaccess-economic-opportunities-agency. Chioda, Laura. 2016. Work and Family: Latin American and Caribbean Women in Search of a New Balance. Washington, DC: World Bank. http:// siteresources .worldbank.org/LACEXT/Resources/informe_genero _LACDEF.pdf. EBRD (European Bank for Reconstruction and Development). 2015. Enhancing Women’s Voice, Agency and Participation in the Economy: Studies in Egypt, Jordan, Morocco, Tunisia and Turkey. London: EBRD. http:// www.ebrd.com/documents/comms-and-bis/enhancing-womens-voice -agency-and-participation-semed-and-turkey.pdf. IHSN (International Household Survey Network). 2015. “How Well Are Gender Issues Covered in Household Surveys and Censuses? An Analysis Using the IHSN–World Bank Gender Data Navigator.” http://ihsn.org/HOME/sites/default/files/resources/Gender_Issues_July -2015.pdf. Kabeer, Naila. 1999. “Resources, Agency, Achievements: Reflections on the Measurement of Women’s Empowerment.” Development and Change 30 (3): 435–64. Klugman, Jeni, Lucia Hanmer, Sarah Twigg, Tazeen Hasan, Jennifer McCleary- Sills, and Julieth Santamaria. 2014. Voice and Agency: Empowering Women and Girls for Shared Prosperity. Washington, DC: World Bank. https:// openknowledge.worldbank.org/handle/10986/19036. 13 Introducing ADePT Gender Software: Part I Lokshin, Michael, Sergiy Radyakin, Zurab Sajaia, and William Creitz. 2013. ADePT User Guide. Version 5. Washington, DC: World Bank. Mason, Andrew D., Reena Badiani, Trang Van Nguyen, Katherine Patrick, and Ximena Del Carpio. 2012. Toward Gender Equality in East Asia and the Pacific: A Companion to the World Development Report. Washington, DC: World Bank. Munoz Boudet, Ana Maria, Patti Petesch, Carolyn Turk, and Maria Angelica Thumala. 2012. On Norms and Agency: Conversations about Gender Equality with Women and Men in 20 Countries. Washington, DC: World Bank. http:// documents.worldbank.org/curated/en/2012/01/17041656/norms-agency -conversations-gender-equality-women-men-20-countries. Munoz Boudet, Ana Maria, and Josefina Posadas. 2014. Russian Federation Gender Assessment: Main Report. Washington, DC: World Bank. http:// documents.worldbank.org/curated/en/2016/04/19286033/russian -federation-gender-assessment-main-report. Narayan, Deepa. 2005. Measuring Empowerment: Cross Disciplinary Perspectives. Washington, DC: World Bank. Sen, Amartya K. 1985. “Well-Being, Agency and Freedom: The Dewey Lectures 1984.” Journal of Philosophy 82 (4): 169–221. ———. 1992. Inequality Re-Examined. Oxford: Clarendon Press. ———. 1999. Development as Freedom. New York: Knopf. UN (United Nations). 2010. The World’s Women 2010: Trends and Statistics. New York: UN. http://unstats.un.org/unsd/demographic/products /Worldswomen/WW_full%20report_color.pdf. World Bank. 2011a. Empowering Women: Legal Rights and Economic Opportunities in Africa. Washington, DC: World Bank. http://documents .worldbank.org/curated/en/2011/01/16394357/empowering-women -legal-rights-economic-opportunities-africa. ———. 2011b. Getting to Equal: Promoting Gender Equality through Human Development. Washington, DC: World Bank. http://siteresources.worldbank. org/EDUCATION/Resources/278200-1099079877269/547664 -1099080014368/Getting_to_equal.pdf. ———. 2011c. Vietnam: Country Gender Assessment. Washington, DC: World Bank. http://documents.worldbank.org/curated/en/2011/11/15470188 /vietnam-country-gender-assessment. ———. 2012a. Papua New Guinea: Country Gender Assessment for the Period 2011–2012. Washington, DC: World Bank. http://documents.worldbank 14 Chapter 1: Applying a Household-Centered Framework to Gender Analysis .org/curated/en/2012/12/17431121/papua-new-guinea-country-gender -assessment-period-2011-2012. ———. 2012b. World Development Report 2012: Gender Equality and Development. Washington, DC: World Bank. https://openknowledge .worldbank.org/handle/10986/4391. ———. 2013a. Country Gender Assessment: Economic Participation, Agency and Access to Justice in Jordan. Washington, DC: World Bank. http:// documents.worldbank.org/curated/en/2013/07/18423362/country-gender -assessment-economic-participation-agency-access-justice-jordan. ———. 2013b. Opening Doors: Gender Equality and Development in the Middle East and North Africa. Washington, DC: World Bank. http://documents. worldbank.org/curated/en/2013/02/17235637/opening-doors-gender -equality-development-middle-east-north-africa. World Economic Forum. 2014. Global Gender Gap Report: 2014. Geneva: World Economic Forum. http://reports.weforum.org/global-gender-gap -report-2014. 15 Chapter 2 Working with ADePT Software This chapter provides basic information about installing and using ADePT Gender. The instructions here are sufficient for performing a simple analysis. More information is available from the following sources: • Detailed instructions for using ADePT are provided in the ADePT User Guide, which can be downloaded from http://www.worldbank .org/adept Documentation. • Video tutorials are available at http://www.world bank.org/adept Video Tutorials. • ADePT provides online help via the Help Contents command. • For help using any ADePT module, see the appropriate chapters in this book or in another book in the Streamlined Analysis with ADePT Software series. • Module-specific instructions, along with example datasets, projects, and reports, are available at http://www.worldbank.org/adept Modules. • Example datasets and projects are installed with ADePT. They are located in the \example subfolder in the ADePT program folder. Use the examples with the instructions in this chapter to familiarize your- self with ADePT operations. 17 Introducing ADePT Gender Software: Part I Conventions Used in This Chapter • Windows, buttons, tabs, dialogs, and other features you see on-screen are shown in bold. For example, the Save As dialog has a Save button and a Cancel button. • Keystrokes are shown in SMALL CAPS. For example, you may be instructed to press the ENTER key. • Menu commands use a shorthand notation. Project Exit, for exam- ple, means “open the Project menu and click the Exit command.” Installing ADePT System Requirements • A PC running Microsoft Windows XP (SP1 or later), Windows Vista, Windows Server 2003 and later, or Windows 7; ADePT runs in 32- and 64-bit environments. • NET 2.0 or later (included with recent Windows installations), and all updates and patches. • 80-megabyte disk space to install, plus space for temporary dataset copies. • At least 512 megabytes of random-access memory (RAM). • At least 1024 x 768 screen resolution. • At least one printer driver must be installed (even if no printer is connected). • Microsoft Excel for Windows (XP or later), Microsoft Excel Viewer, or a compatible spreadsheet program for viewing reports generated by ADePT. • A Web browser and Internet access are needed to download ADePT. Internet access is needed for program updates and to load Web-based datasets into ADePT. Otherwise, ADePT does not require Internet access to run. 18 Chapter 2: Working with ADePT Software Installation 1. Download the ADePT installer by clicking the ADePT Downloads button at http://www.worldbank.org/adept. Launch the installer and follow the on-screen instructions. ADePT automatically launches after installation. Launching ADePT 2. Click the ADePT icon in the Windows Start menu. 3. In the Select ADePT Module window, double-click the name of the module you want to use (see the up and down arrows in the bar at right in the screenshot below). To open a health module, double- click Health; then click Health Financing or Health Outcomes. 19 Introducing ADePT Gender Software: Part I You now see the ADePT main window. (The example below shows ADePT configured with the gender module. The lower-left and upper-right panels will be different when another module is loaded.) To switch to another module after launching ADePT: Module Select Module.... In the Select ADePT Module window, double-click the name of the module you want to use. Overview of the Analysis Procedure There are four general steps to performing an analysis: 1. Specify one or more datasets that you want to analyze. 2. Map dataset variables to ADePT analysis inputs. 3. Select tables or graphs. 4. Generate the report. 20 Chapter 2: Working with ADePT Software Here is where you perform each step in the ADePT main window: 1. Click Add button to load. 3. Select tables or graphs to be Enter dataset year in Label column. included in report. 2. Map dataset variables to input variables by 4. Click Generate. selecting dataset variables in drop-down lists. The next sections in this chapter provide detailed instructions for the four steps. Specifying Datasets Your first task in performing an analysis is to specify one or more datasets. ADePT can process data in Stata (.dta), SPSS (.sav), and tab-delimited text (.txt) formats.1 21 Introducing ADePT Gender Software: Part I Operations in this section take place in the upper-left corner of the ADePT main window. 1. Click the Add... button. 2. In the Open dataset dialog, locate and click the dataset you want to analyze. Then click the Open button. The dataset is now listed in the Datasets tab. Tip: While learning to use ADePT, you may want to experiment with example data. You can find sample datasets in the ADePT\Example folder. 22 Chapter 2: Working with ADePT Software 3. Specify a label for the dataset: In the Label column, select the default label. Type a label for the dataset. Recommendation: Label the dataset using the year the survey was conducted (for example, 2002). When labels are years, ADePT can calculate differences between surveys. Press ENTER. To remove a dataset: Click the dataset; then click the Remove button. One dataset has been specified in this example. Note: ADePT does not alter original datasets in any way. It always works with copies of datasets. Mapping Variables ADePT needs to know which variables in the dataset(s) correspond to the inputs to its calculations. You must manually map dataset variables to input variables. Operations described in this section take place on the left side of the ADePT main window. These examples show the gender module loaded into ADePT, but the process is similar for the other modules. 23 Introducing ADePT Gender Software: Part I There are two methods for mapping variables: Method 1: In the lower input Variables tab, open the variable’s list; then click the corresponding dataset variable, as shown here for the urban variable. 24 Chapter 2: Working with ADePT Software Method 2: In the upper dataset Variables tab, drag the variable name and drop it in the corresponding field in the lower input Variables tab. Note: You can also type dataset variable names in the input variable fields. The above methods are preferred, however, since typing may introduce spelling errors. A spelling error is indicated by the red excla- mation point next to the input variable field. 25 Introducing ADePT Gender Software: Part I To remove a mapping: Select the variable name in the input variable field, then press DELETE. Some modules have multiple input variable tabs. The gender module, for example, organizes variables in four tabs. In some input variable fields, you can specify multiple dataset variables. Household ID, for example, may not be unique within a dataset because the same ID was assigned to a household in another region. In such cases, you can map multiple dataset variables to one input variable. In this example, the id dataset variable has been mapped to the Household ID input variable. The italic variable name indicates that this input variable field accepts multiple dataset variables. The region dataset variable can now be mapped to Household ID using either of the two methods described earlier. ADePT uses this mapping to create its own internal household ID variable to uniquely identify each household. Tip: Open the example project (Project Open Example Project) to see the result of mapping dataset variables to input variables. Selecting Tables and Graphs After mapping variables, you are ready to select the tables or graphs you want ADePT to generate. Operations described in this section take place on the right side of the ADePT main window. 26 Chapter 2: Working with ADePT Software In the upper-right (outputs) panel, select the tables or graphs you want to generate. Note: If a name is gray, it cannot be selected. These tables and graphs cannot be generated because no required variables have been specified. 27 Introducing ADePT Gender Software: Part I To see a description of a table or graph: Click the name. Its description is displayed in the Table description and if-condition tab in the lower left corner of the ADePT window. 28 Chapter 2: Working with ADePT Software Generating the Report Click the Generate button. To stop calculating: Click the Stop button. Examine items in the Messages tab. ADePT lists potential problems in this tab. ADePT can identify three kinds of problems: Notification provides information that may be of interest to you. Notifications do not affect the content of reports generated by ADePT. Warning indicates a suspicious situation in the data. Warnings are issued when ADePT cannot determine whether the situation is impossible. Examples include the violation of parameters, the pres- ence of potential outliers in the data, inconsistent data, and inconsis- tent category definitions. ADePT reports are not affected by warnings. Error prevents a variable from being used in the analysis. For exam- ple, a variable may not exist in a dataset (in this case, ADePT continues its calculations as if the variable was not specified). 29 Introducing ADePT Gender Software: Part I If ADePT can match the problem to a particular variable field, that field is highlighted in the input Variables tab. Correct problems as needed. Then, generate the report again. Note: Notifications, warnings, and errors can negatively affect the results ADePT produces. Carefully review messages, and correct critical problems before drawing conclusions from tables and graphs. Examining the Output When its analysis is complete, ADePT automatically opens the results as a spreadsheet in Excel or Excel Viewer. The results are organized in multiple worksheets: The Contents worksheet lists all the other worksheets, including titles for tables or graphs. The Notifications worksheet lists the errors, warnings, and notifications that ADePT identified during its analysis. This worksheet may be more useful than the Messages tab in the ADePT main window, because the problems are organized by dataset. Table worksheets display tables generated by ADePT. Tip: ADePT formats table data with a reasonable number of decimal places. Click in a cell to see the data with full resolution in the formula bar. Figure worksheets display graphs generated by ADePT. Working with Variables Viewing Basic Information about a Dataset’s Variables 1. In the Datasets tab, click the dataset you want to examine. 2. Click the Variables tab. 30 Chapter 2: Working with ADePT Software To search for a variable: In the Search field, type a few characters in the variable name or variable label. To view statistics for a variable: Double-click the variable name or variable label. This operation opens the MultiDataset Statistics window for that variable. 31 Introducing ADePT Gender Software: Part I Viewing a Dataset’s Data and Variable Details 1. In the Datasets tab, click the dataset you want to examine. 2. Click the Browse... button. This operation opens the ADePT Data Browser. 32 Chapter 2: Working with ADePT Software The Data Browser lists observations in rows and organizes variables in columns. To see underlying data: Click the Hide value labels button. To see value labels: Click the Show value labels button. To view a variable’s statistics: Click in the variable’s column. Click the Show statistics... button. To view detailed information about the dataset’s variables: Click the Variable view tab on the bottom left of the Data Browser. To hide or show variable columns in the Data view tab: In the Variable view tab, click the checkbox next to the variable name. Tip: The ADePT User Guide describes other functions available in the Data Browser. Generating Variables You can create new variables based on the variables present in a dataset. This process might be useful for simulating the effects of changes in param- eters on various economic outcomes. 33 Introducing ADePT Gender Software: Part I 1. In the Datasets tab in the main window, click the dataset you want to modify. 2. Click the Variables | [dataset label] tab. 3. Right-click in the table; then click Add or replace variable... in the pop-up menu. 4. In the Generate/replace variable dialog: a. In the Expression field, define the new variable using the following syntax: = [if <filter_expression>], where is a unique name not already in the dataset(s), calculates new data for the variable, and <filter_expression> (optional) filters observations that are used in the calculation. (See the “Variable Expressions” section below for more information.) b. Optional: Activate the Apply to all datasets option. Note: If you load multiple datasets but do not generate the new variable for all datasets, you will be unable to use the new variable in calculations. However, you may want to generate a different new variable for each dataset in the project. c. Click the Generate button. d. In the Information dialog, click the OK button. The new variable will be listed in the Variables | [dataset name] tab, and in the Data Browser. If the variable was generated for all loaded data- sets, it will appear in the drop-down lists in the input Variables tab. When you save a project, variable expressions are saved with the project, and the variables are regenerated when you open that project. Generating new variables does not change original datasets. Replacing Variables You can replace an existing numeric variable by following the instructions in “Generating numeric dataset variables,” but in the Generate/replace variable dialog (step 4a above), specify an existing variable name instead of a new variable name. 34 Chapter 2: Working with ADePT Software As with generated variables, these expressions are saved with a project, and the variables are regenerated when you open the project. Replacing variables does not change original datasets. Variable Expressions The following operators can be used in expressions: Operator Description + – * / Basic mathematical operators abs sign = == Equality check operators ^ pow sqrt Exponent (for example, x^2 is x squared), power (for example, pow(4,2) is 42 = 16) and square root round truncate Shortening operators min max Range operators ceiling floor Variable expressions can include constants, and strings can be used for variables that are of the string type. Expression examples are as follows: x=1 Sets all variable x observations to 1 x=y+z Sets variable x observations to y observation plus z observation x=y=1 Sets variable x observations to 1 (true) if y is 1, otherwise sets to 0 (false) x = 23 if z == . Sets variable x observations to 23 if z is missing ( . ), otherwise sets to . x = Log(y) if z = 1 Sets variable x observations to log of y observation if z is 1, otherwise sets to . s = “test” Sets all variable x observations to the string “test” Deleting Variables You can remove variables from the working copy of a dataset that ADePT uses for its calculations. This operation does not change the original dataset. Native variables, as well as generated and replaced variables, can be deleted. 1. In the dataset Variables tab, right-click in the row containing the variable you want to delete; then click Drop variable [variable name] in the pop-up menu. 2. In the Confirmation dialog, click the Yes button. 35 Introducing ADePT Gender Software: Part I Setting Parameters Some modules have a Parameters tab next to the input Variables tab. In the Parameters tab, you can set ranges, weights, and other module-specific factors that ADePT will apply during its processing. A Parameters tab may also have input variable fields for mapping dataset variables. The mechanics for setting parameters are straightforward: activate options, set values, and select items in drop-down lists. The analytical reasons for setting parameters can be found elsewhere in this book or in the appropriate book in the Streamlined Analysis with ADePT Software series. Working with Projects After specifying datasets and mapping variables, you can save the configura- tion for future use. A saved project stores links to datasets, variable names, and other information related to analysis inputs. Projects do not retain table and graph selections, corresponding if-conditions, and choices for frequen- cies and standard errors, as they are related to analysis outputs. To save a project: 1. Project Save project or Project Save as... . 2. In the Save as dialog, select a location and name for the project; then click the Save button. 36 Chapter 2: Working with ADePT Software To open a saved project: 1. Project Open project... . 2. In the Open dialog, locate and select the project; then click the Open button. ADePT supports Web-based projects and datasets. To open a Web-based project: 1. Project Open web project... . 2. In the Open web project dialog, enter the project’s URL; then click the OK button. To add a Web-based dataset: 1. In the Datasets tab, SHIFT-click the Add... button. 2. In the Add web dataset dialog, enter the dataset’s URL; then click the OK button. Adding Standard Errors or Frequencies to Outputs To calculate standard errors: Before clicking the Generate button, activate the Standard errors option. Calculating tables with standard errors takes considerably more time than calculating tables without them. A good approach is to obtain the result you want without standard errors and then generate final results with standard errors. To calculate frequencies: Before clicking the Generate button, activate the Frequencies option. 37 Introducing ADePT Gender Software: Part I Tables with frequencies show the unweighted number of observations that were used in the calculation of a particular cell in a table. No significant additional time is needed to calculate frequencies. Results of standard error and frequency calculations associated with a table are provided in separate worksheets, labeled SE and FREQ, within the output report. Applying If-Conditions to Outputs The purpose of if-conditions is to include observations from a particular subgroup of a population in the analysis. The inclusion condition is formu- lated as a Boolean expression—a function of the variables that exist in the dataset. Each particular observation is included in the analysis if it satisfies the inclusion condition (the Boolean expression evaluates to value true). In many cases, the conditions we use are quite simple. Consider the following examples: If-condition Interpretation urban=1 Only those observations that have the value of variable urban equal to 1 will be included in the analysis. region==5 Only observations from the region with code 5 are included in the analysis. age_yrs>=16 Only those individuals who are age 16 or older are included in the analysis. sland!=0 Exclude from analysis those individuals who are not landowners (given that the variable sland denotes the area of the land owned). 1. In the list of tables and graphs, click the table or graph name. 2. Enter the if-condition at the bottom of the Table description and if-condition tab (see list of operators below). 38 Chapter 2: Working with ADePT Software If-condition operators: Operator Description = Equal == Equal >= Greater than or equal <= Less than or equal != Not equal & Logical AND | Logical OR inlist(,n1,n2,n3,...) Include only observations for which has values n1,n2,n3,... inrange(,n1,n2) Include observations for which is between n1 and n2 !missing() Exclude observations with missing values in 3. Click the Set button. A table or graph that has an if-condition is highlighted. Generating Custom Tables You can add a custom table to ADePT’s output. 1. Tools Show custom table tab. 2. In the lower-left panel’s Custom table tab, activate the Define cus- tom table option. 39 Introducing ADePT Gender Software: Part I Design the table by selecting items in the drop-down lists and by activat- ing the options as desired. The Custom table tab in the lower-right corner of the ADePT main window displays a simple preview of your table design. This preview enables you to interactively modify the table to suit your needs. In the outputs panel: 1. Scroll to the bottom of the list. 2. Select Custom table. 40 Chapter 2: Working with ADePT Software The custom table will be included in the report generated by ADePT. Note 1. Stata files have to be Version 9 or later. 41 Chapter 3 Data Preparation This chapter describes the data requirements for producing the complete set of tables and graphs available in ADePT Gender. ADePT Gender requires the user to prepare the data. Data preparation can be done using statistic or econometric packages, such as Stata or the Statistical Package for the Social Sciences (SPSS). ADePT Gender is also capable of creating variables, though its power to do so is limited compared with econometric packages. This chapter does not explain how to create variables; it only describes the data characteristics and the variable definitions to be uploaded in ADePT Gender.1 It also briefly discusses methodological mea- surement issues of some of the key variables and provides key references for further reading. ADePT Gender uses data of individuals or persons, which can be found in many household surveys, such as the Living Standards Measurement Surveys, Demographic and Health Surveys (DHS), Labor Force Surveys, and other multitopic surveys. Without data of individuals to identify their gender, ADePT Gender cannot produce any output. If the household sur- vey collects the data in separate modules—and thus data files—with obser- vations at the individual and household level, all data need to be combined into a single data file with observations at the individual level before uploading them to the ADePT Gender software. However, data files from different years or countries do not need to be merged and can be uploaded to ADePT one at a time. The data fields to be completed in ADePT Gender software are of two types: (a) fields that describe the individual and the household and (b) fields 43 Introducing ADePT Gender Software: Part I that contain outcomes of interest. The first group includes the variables that define the relevant groups of the population the user is interested in analyz- ing, and they are filled in under the Main tab and the Parameters tab. The second group includes outcome variables that are filled in under three tabs: Human Capital, Economic Opportunities, and Agency. As the different fields are completed, ADePT works out which tables and graphs are feasible to produce. The rest of the chapter describes each of the variables to be populated in ADePT Gender’s fields. The book uses different typefaces to indicate a field or a tab name that is fixed in ADePT software and a variable name that is completed by the user. It is fundamental that the user understands the defi- nition that ADePT Gender expects for each variable in order to correctly interpret the tables and graphs. The definition of the variable might require the user to manipulate data depending on the format of the raw survey data before loading it into ADePT Gender software. The complete list of field variables and definitions is also summarized in appendix A. In addition, ADePT Gender provides a short description of the input variable in the status bar (see screenshot 3.1). Notice that ADePT Gender does not perform any data cleaning. However, the software provides information about the data in order to help the user identify potential problems. Thus, the user is responsible for evaluating the quality of the data and therefore the quality of the final output. ADePT Gender performs various internal and background checks on each of the variables filled in, but it does not eliminate observations. The user must perform this task outside ADePT Gender software, if neces- sary. It is recommended that the user eliminate all observations from the sample that have no positive sample weight and the observations with missing values for a few key variables, such as gender and age. It is also recommended that the user eliminates households with no identifiable household head from the sample. As a rule of thumb, the eliminated observations should be no more than 5 percent of the total sample covered by the survey. If the missing information affects a larger percentage of the sample, the user should find additional information to assess the validity of the data. For example, the user can conduct additional statistical analy- sis on the eliminated observations in order to determine the reliability of data and thus of the results. It is good practice to produce a log of these deletions. 44 Chapter 3: Data Preparation Screenshot 3.1: Variable Definition Hint Field Variable A description of the input variable is shown in the status bar. The description appears as the cursor moves onto the field name or when the scroll-down menu is opened. Household Surveys Different types of household surveys can be used in ADePT Gender. Survey instruments are designed with a variety of main objectives. They thus cover issues to varying extents and might be representative of different population groups. For example, labor force surveys are fielded with the main objective of measuring employment and unemployment rates; they have good cover- age of employment variables and are usually representative of the working- age population. Demographic and health surveys aim to measure outcomes indicated by their name. They have poor coverage of labor market perfor- mance and often oversample women of reproductive age. These surveys are collected by national statistics offices, either alone or with the support of 45 Introducing ADePT Gender Software: Part I the Demographic and Health Survey Program.2 An additional resource for understanding gender indicators is the International Household Survey Network (IHSN) study based on an analysis of the IHSN–World Bank Gender Data Navigator database on coverage of gender data and specific types of surveys (IHSN 2015). However, most surveys serve many purposes, such as providing data to complete national accounts, providing weights for consumer price indexes, and so on. Many countries—in particular, those with the support of the World Bank—developed multitopic surveys such as the Living Standards Measurement Surveys, which was originally designed to monitor poverty and inequality but has evolved to collect a wide range of indicators. The number of tables and graphs that ADePT Gender generates depends on the variables populated and thus on the survey instrument. Table 3.1 shows the most common household survey instruments and their usual coverage of gender topics that allow populating ADePT Gender fields. ADePT Gender’s template cannot accommodate microdata surveys in which the sampling unit is not the household. For example, this is the case with Enterprise Surveys, which have firms as the unit of observation.3 However, some of these surveys have employee modules where the unit of observation is the firm employee and that distinguish between male and female employees. Although ADePT Gender is not designed to work with Table 3.1: Usual Topic Coverage of Survey Instruments Demographic and health Labor force Multitopic Household Topic and survey instrument survey survey surveya budget survey Endowments Education *** * ** * Health and nutrition *** * Maternal health *** Economic opportunities Labor market participation * *** ** * Job characteristics *** Access to resources * ** * Voice, agency, and participation Control over resources ** * Freedom of movement ** Decision making over family formation ** * Violence against women ** Voice and participation * Note: The qualification of the coverage is relative; in other words, it was done comparing the average coverage of the other survey instruments and using as a reference the variables needed to populate ADePT Gender fields. *** = excellent coverage; ** = good coverage; * = weak coverage. a. For example, Living Standards Measurement Surveys. 46 Chapter 3: Data Preparation these data, some of the tables and graphs can be produced with them, as long as the variable definitions satisfy the requirements of ADePT Gender fields. Main Variables ADePT Gender uses data files with the observations at the individual level. Some of the variables vary by individual (such as a person’s age) and others vary by household (such as place of residence or the welfare aggregate). Some of the variables require no data preparation and thus can be loaded directly into ADePT Gender (for example, age). Other variables demand very little manipulation and can be created in ADePT Gender (for example, employ- ment). Finally, a few variables are complex, and it is recommended that the user construct them using econometric or statistic packages before loading the data into ADePT Gender software (for example, the welfare aggregate). Next, we discuss each of the variables to be completed in ADePT Gender as well as some concepts behind these definitions. Sample Weights As opposed to census data, in which all the units of the universe are inter- viewed, surveys select a random sample of the population, and thus sample weights are necessary to make the estimates representative of the (country) population. Given that the data are loaded at the individual level—all persons in a household—ADePT Gender requires only household weights, which will automatically be used to produce weighted results representative of the country’s population. The Household weights field should be com- pleted with a continuous variable that indicates the estimated expansion factor of each household in the country population. Household weights are needed because the sampling design does not necessarily select households with equal probability. If all households were selected with equal probability, all of them would have the same weight. However, because of cost and accuracy reasons, the probability of being selected likely differs across households (Deaton 1997). When selection probabilities differ, each household in the survey stands proxy for or repre- sents a certain number of households in the population. Consequently, when the sample is used to calculate estimates of the population, it is necessary to weight the sample data. In other words, the weighted averages used “undo” 47 Introducing ADePT Gender Software: Part I the sample design to obtain estimates that match the population. The rule here is to weight according to the reciprocal of the sampling probabilities, because households with low or high probabilities of selection stand proxy for a large or small number of households in the population. Surveys often include more than one type of weight. They might have individual weights (or person-specific weights) and household weights. If the survey comes from a stratified sample, the data can include strata weights. If the survey is longitudinal (that is, follows the same individual or house- hold over time), the survey may have cross-sectional weights, longitudinal weights, or both. Each of these weights makes the observations representa- tive of different populations. ADePT Gender assumes that the weights used for each observation are representative of the total population, and the expla- nations of how to interpret tables and graphs in chapters 4 and 6 refer to this case.4 Box 3.1 describes the basics of sampling design to understand how probabilities of being selected in the survey might vary across households. Box 3.1: Sampling Design The typical household survey collects data from a national sample of households that are randomly selected from a national list. That list is called the survey frame. The frame is usually the national population census; however, some countries use adminis- trative data. Often, the survey frame does not cover the total population. The term coverage refers to completeness of the survey frame compared with the total popula- tion. Certain groups of the population are likely to be excluded from the sampling frame, such as the homeless, members of the armed forces, seasonal migrants, people in jail, workers who live in factories, and college students. Partial coverage of the survey frame will result in differences between actual and esti- mated statistics. However, this result is typical of household surveys and should not prevent the data from being used. Users simply need to bear in mind the population that the survey will represent, that is, the population that is covered by the frame. In addition to noncoverage, users need to consider nonresponse: those households that either refuse to join the survey (unit nonresponse) or that do not answer specific mod- ules or questions (item nonresponse). If nonresponse is associated with certain observ- able or nonobservable characteristics, users need to take that into account when making inferences from survey estimates. For example, women subject to domestic violence might be more likely not to participate in a survey simply because they do not want to open the door and be seen by strangers. In this case, the survey would underestimate the prevalence of domestic violence because of nonresponse or nonparticipation. (continued) 48 Chapter 3: Data Preparation Box 3.1: Sampling Design (continued) The simplest household survey design would be one in which (a) a reliable, up-to-date list of all households in the population exists; (b) the design assigns an equal probabil- ity to each household selected from the list to participate in the survey; and (c) all households asked to participate actually do so. Under this design, each household has the same weight, since all of them represent the same number of households in the population. However, surveys tend to be more complex, as discussed by Deaton (1997). Most of the time, surveys rely on a two-stage sampling design. However, in any of these cases, we need to know only the weight of each household to generate statis- tics that represent the country. To generate statistics that are valid for population subgroups, the sampling design needs to stratify the population. The most common design has two stages. In the first stage, clusters—groups of households determined by geographic location or another characteristic—are selected. In the second stage, households are selected within each cluster. This type of survey has many advantages with regard to costing and visits. Note that in two-stage sampling, we can still have households that have the same probability of being selected, if clusters are selected with a probability proportional to the number of households in each cluster. Stratification converts a sample from one national population into a sample from many populations. Stratification guarantees enough observations to have estimates by group. For example, suppose we are interested in knowing the percentage of female-headed households in urban and rural areas. Without stratification, we would run two surveys, one for each population—urban and rural. However, with stratification, we might run a single survey in which the national values are obtained by a weighted average of the urban and rural populations, where the weights are the proportion of rural and urban households in the total population. The precision of this combined estimate is assessed (inversely) from its variance over replications of the survey. Because the two compo- nents of the survey are independent, the variance of the overall mean will be the sum of the estimates from each stratum. Hence, the variance depends only on the within- stratum component and not the between-strata component. Although stratification typically enhances the precision of sampling estimates, the clustering of the sample will reduce it. Source: Deaton 1997. Household Information Household A household is a person or group of people who usually live and eat together. The standard definition of household is a group of people who live together, pool their money, and share at least one meal (UN 2008b). 49 Introducing ADePT Gender Software: Part I However, more recent revisions of this definition have removed the condition of sharing a meal (UN 2008b).5 Household members typically share a residential unit and have organized economic production and consumption. In many countries, sharing extends to inheritance and child rearing. Members might not be related. Household is not synonymous with family, which is a social institution that is characterized by common residence, economic cooperation, and reproduction. This definition of household has four main exceptions. First, unrelated individuals may share a dwelling to minimize housing costs. Second, one house might have two families who use different rooms and do not share a budget. Third, servants might live in the same household and even share meals, but they have different budgets. Fourth, renters or pensioners may share meals with the owner of the house, but they also have different incomes. In all four cases, separate households should be considered even if all the people reside in the same dwelling and share meals. The other main concern when identifying household members is to avoid counting one person twice when they reside in more than one dwell- ing. This situation might arise when households split, that is, when house- hold members migrate (especially seasonally), or when members move to study somewhere else. As Beaman and Dillon (2012) show, using different definitions has consequences for measuring household welfare and produc- tion, which are variables that ADePT Gender uses to define groups for comparing gender outcomes. ADePT Gender must specify a household identification variable, or a series of variables, that uniquely identifies the household in the dataset. They are filled in under Household ID. Numeric variables used for Household ID are expected to be integers; this is the only internal check performed by the software. ADePT Gender cannot produce any table if this variable (or set of variables) is not completed. Place of Residence Two fields are used to create groups to examine differences by place of resi- dence. The field urban needs to be completed with a variable that takes a value of 1 when the household resides in an urban area. The field region needs to be completed with a categorical variable that codes the household’s region of residence. This variable can be defined following the political division of the country or any other relevant division the user considers of 50 Chapter 3: Data Preparation interest, such as ethnographic division, level of development, climate, and so on. ADePT Gender checks that the variables have the numerical proper- ties expected. Welfare Aggregate ADePT Gender analyzes the differences in outcomes across population groups that have different standards of living. Two fields are used to describe a household’s welfare. The first requires a continuous variable that measures household well-being. This variable is completed in the Welfare aggregate field and is used to construct quintiles or deciles of the population, accord- ing to the option selected in the Number of quantiles field. The second uses a poverty line to divide the population between poor and nonpoor. The Poverty line field requires a number that is the minimum amount— according to the welfare indicator—that an individual needs to be out of poverty.6 Both variables (the ones that populate the Welfare aggregate and the Poverty line fields) should be expressed in the same unit of measure. For example, if the value of the poverty line variable is per capita, then the welfare aggregate variable should also be per capita; if the value of the poverty line variable is per adult equivalent, then the one corresponding to the welfare aggregate should also be per adult equivalent. Most commonly, there are three direct measures of household well- being: income, consumption, and expenditures. Income refers to the earn- ings of all household members from productive activities, plus the sum of all current transfers. The earnings from productive activities include wages and salaries, sales of home-produced goods (including farming), and rents from land and assets. Consumption is the sum of the value of food and nonfood items consumed by the household. Consumption is usu- ally retrieved from consumption recall modules or consumption diaries. Price indexes are often used to adjust for spatial and temporal differences in cost of living. Expenditure is the sum of all household expenses, both food and nonfood items. It is similar to consumption, but it has some conceptual differences. First, using expenditures excludes the consump- tion of items that were not part of a market transaction (for example, own-produced agricultural goods). This difference can be large in those countries with high levels of home production. Second, expenditures record a transaction and thus are subject to measurement error if the time of transaction differs from the time of consumption. This problem arises 51 Introducing ADePT Gender Software: Part I mostly with durables, which are bought in one period but consumed over several periods. Income and consumption measures differ, since consumption can also come from borrowing (or past savings), and income can be used to accumulate savings. In addition, each of these variables has its own mea- surement problems. For income, it is difficult to accurately measure all possible sources of income, especially in countries with high levels of informality, either in wage employment or in self-employment.7 When direct measurements of welfare are unavailable, the user can construct a welfare index using proxy variables. Welfare indexes can be constructed in different ways. One way is to use principal components analysis to construct a “wealth” index using information on asset and land- ownership as well as household characteristics.8 However, the user must bear in mind that mechanic correlations can be generated between the outcomes of interest and these composite indexes of welfare, if the same variables are used as outcomes and as inputs into the composite index. Other less frequently used measures include direct reports of household well-being.9 Individual Information Gender, Age, and Marital Status The word gender is a social construct that includes male, female, transgender, and third gender, and the word sex refers to biological and anatomical differences between male and female. ADePT Gender software, however, does not include transgender and does not differentiate between the terms sex and gender, even using them interchangeably. The reason for these omissions is simply the limitations in the survey data: the datasets used in ADePT do not code these characteristics, and the objective of the software is to analyze surveys from the highest number of countries possible (screenshot 3.2). The Gender field needs to be completed with an indicator variable that identifies male or female and takes a value of 1 for males. ADePT Gender performs background checks to ensure that each individual’s gender is identified by the variable filled in, and that the variable takes only two values. The Age field expects a continuous integer variable that indicates the age of each household member. This variable is used to construct age groups 52 Chapter 3: Data Preparation Screenshot 3.2: Fields in Main Tab Household characteristics fields Individual characteristics fields relevant for the analysis. By default, ADePT Gender software constructs broad categories that reflect the stages of life that matter for the gender analysis: youth (15–24), adult reproductive age (25–49), adulthood (50–64), and elderly (65 years of age and older). However, ADePT Gender is capable of setting finer or different age groups if desired by the user. Another concern is that in some very poor contexts, respondents might not know their own age or, if there is only one respondent, the age of other household members. To avoid this problem, many surveys also collect infor- mation about date of birth, as it may be easier for some elderly respondents to remember their date of birth rather than their age. Data preparation might require calculating the age using date of birth instead of respondent values. In these cases, it is recommended that users perform basic tabula- tions to see whether any spikes occur in age responses (usually in round numbers like 20, 30, and 40), since these are not part of ADePT Gender internal checks. ADePT Gender verifies that the populated variable is a continuous positive integer between 0 and 99. The Marital status field expects a variable that indicates whether or not household members are in a marital union, and if not, the main cause of its dissolution. Users can define the categories of this variable according to their country context and survey options. For example, polygyny is still 53 Introducing ADePT Gender Software: Part I common. However, if the prevalence of a certain marital status is low or not relevant for the study, users might decide to consider only one large group referred to as “under a marital union.” ADePT Gender tests only for whether the populated variable is categorical.10 Household Head Head of household is a complex concept. The head of household is supposed to be the person that the other members of the household acknowledge as head. The head has primary authority and responsibility for household affairs. However, in cases where such authority and responsibility are not vested in one person, special rules may be needed to identify the head of household.11 In some surveys, the interviewer asks respondents to iden- tify the head of household. In other countries, like the United States, it depends on how the household files taxes. In other surveys, head of house- hold does not denote any particular role and is the code that the interviewer gives the oldest male of working age or the member responding to the questionnaire. As with the definition of household, the definition of its head has consequences for measuring outcomes. For example, Fuwa (2000) exam- ines how poverty incidence for female-headed households varies for differ- ent definitions of female-headed households in Panama. In particular, he compares de jure female-headed households—where women declared themselves as the head—with de facto female-headed households—where the woman is working, and the present male spouse is not; or where the woman is present in the household, and the husband is away for work. Fuwa finds that the poverty incidence varies across these definitions, even when compared with households of similar composition. The Household head field expects a binary variable that takes a value of 1 for the individual that is identified as the head of household. ADePT Gender checks that one of the variable’s two values is 1, that only one household member is coded as head, and that the household head is at least 15 years old. Custom Category ADePT Gender allows the user to add custom variables that can be used to further tabulate the results. A custom variable can provide a household 54 Chapter 3: Data Preparation characteristic (that is, with a common value for all the individuals in the household) or an individual one. Examples of the former could be whether the household head is absent part of the year because of sea- sonal work outside the city,12 whether pensioners are in the household,13 or whether a household member receives cash transfers.14 Examples of individual variables include race or ethnicity.15 Another significant household characteristic to include may be households with individuals who have disabilities, who likely require additional time and resources for their care. If a custom individual field is completed, additional rows will be added in the tables where the outcomes are presented by individual characteristics, such as ADePT tables 1a and 1c. If a custom household field is completed, additional rows will be added in the tables where the outcomes are presented by household characteristics, such as ADePT tables 1b and 1d. Outcome Variables Variables that reflect outcomes are grouped into three categories following the three main dimensions proposed by the World Development Report— human capital (or endowments) (screenshot 3.3), economic opportunities, and agency. We discuss each concept in turn. Human Capital Education ADePT Gender examines gaps in access to education for boys and girls— measured by school enrollment or attendance—and gender gaps in educa- tion attainment for the adult population. Both access to and completed education outcomes require binary variables by level of education—primary, secondary, and postsecondary. For adults (or those who have completed their education), ADePT Gender also offers the option to assess gaps in literacy. Literacy is the “ability to identify, understand, interpret, create, communicate and compute, using printed and written materials associated with varying contexts” (UNESCO 2006).16 Most surveys identify literacy levels through specific questions that assess whether respondents know how to read and write.17 For finer gender gaps in completed education, ADePT Gender also allows the user to input years of education. 55 Introducing ADePT Gender Software: Part I Screenshot 3.3: Fields in the Human Capital Tab Fields to input education Fields to input completed enrollment variables education variables Fields to complete health and nutrition and maternal health variables (fields admit more than one variable) Levels of education are classified by the International Standard Classification of Education (ISCED), adopted by the General Conference of the United Nations Educational, Scientific, and Cultural Organization.18 Primary education (ISCED 1), sometimes called elementary education, refers to programs normally designed to give students a sound basic education in reading, writing, and mathematics, along with an elementary understanding of other subjects, such as history, geography, natural science, social science, art, and music. Religious instruction may also be featured. Secondary educa- tion refers to programs of lower (ISCED 2) and upper (ISCED 3) secondary education. Lower secondary education continues the basic programs of the primary level, but the teaching is typically more subject focused, requiring more specialized teachers for each subject area. In upper secondary education, instruction is often organized along more in-depth subject lines, and teachers typically need an advanced degree or more subject-specific qualification. 56 Chapter 3: Data Preparation One caveat when looking at the variables on current education is that although school enrollment and school attendance are both used to ascer- tain school participation, they are two different issues. Household surveys usually collect data on household members’ school attendance, whereas administrative data—generated from school records—yield information on school enrollment. A child may be enrolled in school but not be attending at the time of the interview. As a result, school enrollment data tend to overstate the effective student population. School attendance at the time of the interview is a more reliable indicator of the proportion of students actu- ally attending school.19 Tertiary education refers to a wide range of programs with more advanced educational content. The first stage of tertiary education (ISCED 5) refers to theoretically based programs intended to qualify a student to enter (a) an advanced research program that is practical, technical, or occupation spe- cific; or (b) a profession with high skill requirements. The second stage of tertiary education (ISCED 6) refers to programs devoted to advanced study and original research. Postsecondary nontertiary education (ISCED 4) refers to programs that, regarding their content, cannot be considered tertiary from an international point of view. These programs might be considered postsecondary in a national context, but they are not significantly more advanced than pro- grams classified as ISCED 3, a level that should be completed before being enrolled in ISCED 4. Some countries do not have ISCED 4 programs. We recommend caution when uploading the human capital variables by checking the country ISCED mapping (see box 3.2 for an example from Panama).20 The ADePT user has the option to pool the levels corre- sponding to ISCED 4, 5, and 6 into one variable and upload it in the Postsecondary field or to reserve this category for tertiary education only, if it makes more sense in the country context. All the fields in ADePT Gender that are related to the current level of school enrollment expect binary variables that take a value of 1 if the indi- vidual is enrolled in that level.21 The same arguments apply to the fields related to completed education, and the variables take a value of 1 to indicate the highest level achieved. For example, if a person has started but not completed college, he or she will have a value of 1 for the variable indicating the second- ary level of education. ADePT Gender performs internal checks of these vari- ables, so that no person is coded with two different levels of education. 57 Introducing ADePT Gender Software: Part I Box 3.2: Mapping Panama’s Education System to the International Standard Classification of Education Panama’s education system is divided into five levels: preschool, basic, presecondary, secondary, and superior. The first four levels correspond to International Standard Classification of Education (ISCED) levels 0, 1, 2, and 3, respectively, and the superior levels exist in several programs that can be classified as ISCED 4, 5, or 6. The secondary level can be accomplished in two ways: the first and most common is called academic secondary education, and the second is called technical and profes- sional education. The two options have the same duration and prerequisites and cor- respond to ISCED 3 programs. However, it is a common mistake to classify technical and professional education as ISCED 4 or 5 since, in the national surveys, it is pre- sented as a separate level, and its name could be easily confused with an ISCED 5 program. Like Panama, several other countries have domestic labels of education that might be misleading, even for national practitioners. For this reason, we recommend checking the country ISCED mapping variable before classifying the specific country levels at the data preparation stage. Source: Panama 1997. Nutrition and Health Data on nutrition and health that are disaggregated by gender are not widely available from household surveys. Thus, ADePT Gender has been designed to let users accommodate the existing data. As opposed to other outcomes for which ADePT Gender requires specific variables, users can fill in the fields with outcomes of their choice and have the option to have outcomes for boys and girls, men and women, and women of reproductive age. The section that follows provides examples of variables that can be used as outcomes. Users can fill in as many variables as they want in the Health and nutri- tion field, as long as they are available for both sexes. These variables can refer to outcomes relevant for children, such as being inoculated with BCG (Bacillus Calmette–Guérin vaccine against tuberculosis), DPT (diphtheria, pertussis, and tetanus), MMR (measles, mumps, and rubella), or all three. Alternatively, the variables can be anthropometric outcomes, such as age for height, weight for height, and age for weight. The field can also be popu- lated with variables that describe adults’ health, such as human immunode- ficiency virus (HIV) and alcohol consumption. 58 Chapter 3: Data Preparation Instead, the Maternal health field expects variables that pertain only to women’s reproductive health. Thus, ADePT Gender will compute the sta- tistics only for women of reproductive age—15 to 49 years old. Examples of outcomes of interest related to maternal health are whether a woman received assistance when delivering a child, prenatal care, and postnatal care. Prenatal care and postnatal care include the services that health care providers give to a pregnant woman before and just after the birth of her infant, respectively. Delivery assistance is usually provided by a skilled birth attendant. Ensuring quality maternity care can save the lives of women and newborns. These services require an accredited health professional—such as a midwife, doctor, or nurse—who is proficient in the skills needed to man- age normal (uncomplicated) pregnancies, childbirth, and the immediate postnatal period. Such professionals are also trained to identify and manage complications in women and newborns and to refer them to specialists. Other important health and maternal health outcomes are more difficult to assess using household survey data. The most important are life expec- tancy at birth, infant mortality (by gender), and maternal mortality. These outcomes are usually constructed using information from vital statistics records. However, Demographic and Health Surveys collect information to calculate a few of them. Maternal mortality is the death of a woman while pregnant or within 42 days of terminating a pregnancy—irrespective of the duration of the pregnancy—from any cause related to or aggravated by the pregnancy or its management, excluding accidental or incidental causes. To facilitate the identification of maternal deaths in circumstances in which cause of death attribution is inadequate, a new category has been introduced: pregnancy-related death.22 For the variables populated in Health and nutrition and in Maternal health, ADePT Gender checks only nonmissing values for the relevant populations; that is, it checks the variables that are available for men and women for the former field and that are available only for women ages 15–49 in the latter. Economic Opportunities Labor Status The analysis of economic opportunity is limited to the working-age popula- tion. The minimum age range for defining the working-age population needs 59 Introducing ADePT Gender Software: Part I to be specified by the user according to national circumstances, such as the minimum age for admission to employment and the extent of child labor. These circumstances vary so greatly among countries that it is impossible to specify any universally applicable minimum age limit at the international level. By default, ADePT Gender assumes that the working-age population is between 15 and 64 years of age. The user can define the limits through the Parameters tab (see below). Tables for outcomes on economic opportu- nities will be computed only for the working-age population. According to the International Labour Organization (ILO) 1982 guide- lines, which most national statistical agencies continue to use as a refer- ence to date, individuals are employed if they worked for at least one hour during the previous seven days (a) for a wage, (b) in a household enterprise or on a farm, or (c) as an unpaid apprentice or trainee. The definition also includes all persons who had a job or enterprise during that period but were temporarily absent from it—persons who during the reference period were sick, on vacation, on maternity leave, on strike, or temporarily laid off (Hussmanns 2007; ILO 1982). However, the International Conference on Labour Statisticians’ 2013 change in the definition of employment will also affect employment modules going forward by classifying employment as work for pay. That change may affect comparability of estimates over time. Measuring employment in developing countries, in particular for women, is especially difficult when many income-generating activities are performed outside of markets. Respondents may not think of such activities as “work” or “employment.” To help respondents understand the concept of work, employment modules usually include a series of questions that ask about different kinds of work (work on own farm, nonfarm house- hold enterprise, and wage employment), as well as reasons for temporary absences from a job. It is recommended that such questions come with examples to help respondents understand the range of activities to which the interviewer is referring (Anderson Scheffner 2000). In addition, the user should be aware of who the respondent of the labor module is, since it has been found that responses depend on how and to whom questions are asked (Bardasi and others 2011). A person of working age is unemployed if during the period of refer- ence—usually one week—he or she was not employed, was available for work, and was actively seeking work,23 which is understood as taking con- crete actions in a specified recent period to seek employment. The period of reference for seeking employment does not have to coincide with the 60 Chapter 3: Data Preparation reference period of employment. The ILO abstains from making recommen- dations about the length of the reference period, or whether the reference periods used for employment and unemployment have to coincide. Some countries might prefer to have longer reference periods (one month or four weeks) based on the idea that lags in the process of obtaining a job can exceed the reference period for employment (Hussmanns 2007). The other condition that requires clarification is being available for work. Availability for work means that, if given a work opportunity, a per- son is able and ready to take the job and start working. The idea behind this condition is to eliminate people like students, who might be looking for a job at a later date—for example, after graduation. The condition also excludes those who cannot accept work because of certain impediments, such as family responsibilities, illness, or other commitments. The ILO makes no recommendations regarding the time reference for this condi- tion either, and it does not have to coincide with any of the previous refer- ence periods. A related concept is underemployment, which refers to a less extreme situation of partial lack of work. Underemployment captures the idea that a person is willing to work more hours, referred to as time-related underem- ployment. Two concepts are measured by underemployment: the number of persons in time-related underemployment and the intensity of underem- ployment. A person is considered underemployed if he or she (a) is willing to work additional hours, (b) is available to work more hours, and (c) is not working more than a certain number of hours (or days) during the reference period—in order to avoid extra hours or any other unusual circumstances. The intensity of underemployment refers to the gap in hours (or days) of work and the threshold fixed as normal hours of work.24 If a person of working age is neither employed nor unemployed, he or she is said to be out of the labor force. Thus, every person of working age in the sample has to be employed, unemployed, or out of the labor force. In some cases, the labor status variable is already coded by the national statistics office. However, there are other cases in which the user needs to code these variables before loading the data into ADePT Gender software. ADePT Gender requires the user to complete the Employed and Unemployed fields and assumes that all persons of working age that are neither employed nor unemployed are out of the labor force. ADePT Gender does not explic- itly include a space for underemployed. Users interested in this type of employment should work with the ADePT Labor module. 61 Introducing ADePT Gender Software: Part I Even if the definitions above can be found in most household surveys, the user should be aware that the ILO changed the official definition of employment in 2013 (ILO 2013). The transition toward this new definition will take time, as it first has to be tested in several countries before it is applied worldwide. Box 3.3 describes it in more detail. It should be noted that even though the definition of employment has changed, the ADePT Gender soft- ware can still be used to assess labor status in line with the new definitions, as long as the employment variable is defined according to the new standards. Box 3.3: Revised ILO Statistical Standards for Measuring Employment and Work In October 2013, the 19th International Conference of Labour Statisticians (ICLS)—which makes recom- mendations on selected topics of labor statistics to the governing body of the International Labour Organization (ILO)—adopted the resolution concerning statistics of work, employment, and labor underuse (hereafter referred to as the new ICLS standards). The resolution revises previous international statistical standards and guidelines for labor market statistics. An important element of the new ICLS standards is to narrow the definition of employment to work performed for pay or profit. Unlike in the previous definition, the production of goods for one’s own consumption (par- ticularly subsistence agriculture) is now excluded from the employment category (see table B3.3.1 for a schematic overview). In addition, the 2013 resolution introduces a new category of work, which recognizes all productive activities—paid and unpaid—and proposes several measures of labor underuse. Once fully implemented, the new ICLS standards are expected to have significant implications for the mea- surement of employment and labor force participation at the aggregate level, but especially for the large share of the population in developing countries who are engaged in subsistence activities, especially women and the rural poor. • Under the revised definition of employment, farmers who produce only for subsistence purposes are no longer counted as employed and are considered to be out of the labor force. Given the large number of workers—especially women—involved in these activities in low-income countries, the revised standards are likely to result in significantly lower estimates of employment and labor force participation. • Services produced for own final use (such as childcare and eldercare, food preparation, and other house- hold chores), which are often performed by women and were not captured by the previous employment definition, are now recognized as work under the category own-use production. Full and separate mea- surement of women’s participation in these unpaid activities allows a more comprehensive assessment of their overall workload. It is important to note that the new definition of employment is narrower and the new definition of work broader than that of the 2008 System of National Accounts (SNA) frontier, as the latter includes own- produced goods but not own-produced services (see table B3.3.1). Hence, attention must be taken when comparing employment statistics to National Accounts production estimates. (continued) 62 Chapter 3: Data Preparation Box 3.3: Revised ILO Statistical Standards for Measuring Employment and Work (continued) Table B3.3.1: Comparison of New and Previous Definitions Productive activities Nonproductive activitiesa Nonmarket units Households (producing Market units (government, nonprofit) for own final use) Goods Services Goods Services Goods Services Previous activity scope for “employment” New activity scope for “employment” New concept of “work” = productive activities Activities in the SNA production boundary Activities in the SNA general production boundary Source: Diez de Medina and Benes 2014. Note: SNA = System of National Accounts. a. For example, sleeping, learning, own creation, begging, stealing. Work Characteristics The measures of labor force participation are as important as the character- istics of the work. Women might be as likely as men to engage in economic activities, especially in very poor countries, but their economic opportuni- ties are very different. The disparities in economic opportunity can be ana- lyzed only by looking at the type of work they do in more depth. The Work category field can be completed by a categorical variable that can take several values, depending on the country context. At a minimum, it should differentiate between wage work and self-employment, but it can also accommodate other work classifications, such as unpaid or family work. Self-employed refers to working for oneself. Self-employed people work for themselves instead of an employer, drawing income from a trade or busi- ness that they operate. To be self-employed is not necessarily the same as being a business owner; many self-employed people conduct the day-to-day opera- tions of the business as managers, as line workers, or as both. A business owner may or may not work in the business and is not required do so. Self-employed persons without workers are known as own-account workers. Self-employed persons who employ family members are known as household enterprises. Wage workers or paid employees are those who hold jobs in which the basic remuneration is not directly dependent on the revenue of the employer. 63 Introducing ADePT Gender Software: Part I Paid employees include regular employees, workers in short-term employment, casual workers, outworkers, seasonal workers, and other categories of workers who hold paid jobs. Paid employees are not own-account workers, contributing family workers, members of worker cooperatives, or workers unclassifiable by status (ILO 2012). A finer category, called the International Classification of Status in Employment, uses a scale according to the type of contract the person has with other people or organizations when performing a particular job. The type of contract is determined by the type of economic risks and authority that are involved when doing the job. The International Classification of Status in Employment consists of the following groups: • Employees, who receive basic remuneration not directly dependent on the revenue of the employer—among whom countries may need and be able to distinguish “employees with stable contracts” (including “regular employees”); • Employers, who are self-employed—that is, whose remuneration depends directly on the (expectation of) profits derived from the goods and services produced—and who engage one or more person to work for them as “employees” on a continuous basis; • Own-account workers, who are self-employed and do not engage “employees” on a continuous basis; • Members of producer cooperatives, who are self-employed in a coopera- tive producing goods and services, where the members participate equally in making major decisions concerning the cooperative; • Contributing family workers, who hold jobs in an establishment oper- ated by a relative, whose degree of involvement in its operation is too limited to be considered a partner; and • Workers not classifiable by status, for whom insufficient relevant infor- mation is available, or who cannot be included in any of the preced- ing categories. The ILO monitors employment vulnerability and the corresponding gender gap, in which vulnerable employment comprises own-account workers and contributing family workers. ADePT Gender allows the user interested in exploiting this differentiation to input the categorical vari- able that best reflects the country context and needs. However, users should be aware that the differentiation between these two categories is 64 Chapter 3: Data Preparation subject to measurement errors, since the distinction between own-account and family workers can be unclear. Informality The term informality means different things to different people, but its connotation as a type of work is almost universally negative. It refers to unprotected workers, inadequate regulation, low productivity, unfair competition, evasion of the law, underpayment or nonpayment of taxes, and working “underground” or in the shadows (Perry and others 2007). Informality can refer to the worker as well as the firm, if both workers and firms do not integrate (or at least not fully) the state’s role as regulator and public service provider in their economic activities. There are three main types of informality: (a) firms that can be formal or informal; (b) firms that can be partly formal and partly informal; and (c) workers, who can be formal and informal in the way in which they operate in the labor market. These margins are not exhaustive and are related to each other, but they cover much of the discussion (Perry and others 2007). Thus, the final definition of informality depends on two aspects or margins. The intrafirm margin refers to firms that are partially informal across several dimensions—because (a) they underreport sales, (b) they are partially compliant with all registration requirements (firms might be registered at the municipal level but not at the national level), (c) they underreport wages (wages are partly paid on the books and partly off the books), or (d) they are partially compliant with labor regulations (firms can have just a part of their workforce compliant with labor regulations). The intersectoral margin of firms is the threshold that divides formal firms from informal ones and relates to whether workers are covered by labor legislations. A formal firm can be regarded as one that complies with all the labor, business, and tax regulations. However, such legislation often does not reach or cover microfirms, which are then by definition consid- ered informal. Many times, people consider informal workers as those who are not covered by labor legislation, including those in microfirms. The ILO defines informal employment as (a) own-account workers and employers working in their own informal sector; (b) enterprises with at least some market production that are unregistered or small in the number of persons employed (for example, fewer than five employees); (c) all con- tributing family workers; (d) employees who hold informal jobs, that is, 65 Introducing ADePT Gender Software: Part I employees not covered by legal protection or social security or unentitled to other employment benefits, such as paid annual or sick leave; (e) members of informal producer cooperatives (not established as legal entities); and (f) own-account workers who produce goods exclusively for final use by their household (if considered employed). Table 3.2 provides a schematic frame- work of the definition. However, in practice, data limitations shape the definition of infor- mality. Not all labor force surveys collect information about the status of the firm or employer, and employees do not necessarily know whether the firm they work for complies with all the regulations. Thus, many times, countries and policy makers define formality according to whether a firm or employer has any of the following nonexclusive criteria: (a) offers per- manent as opposed to temporary contracts, (b) offers written as opposed to oral contracts, (c) provides pension coverage, or (d) provides health coverage. Other benefits of formal jobs—such as vacations or maternity or sick leave—are seldom incorporated in the analysis. The user thus needs to assess the quality of the variable that can be constructed using the available information. ADePT Gender expects the user to fill in the Formal status field with a binary variable (or expression) that takes a value of 1 if the worker is formal. The only background check performed by the software consists of assessing whether one of the two values of the variable is equal to 1. Table 3.2: ILO Conceptual Framework: Informal Employment Job by status in employment Own-account Contributing Members of producer workers Employers family workers Employees cooperatives Production by type Informal Formal Informal Formal Informal Informal Formal Informal Formal Formal sector 1 2 enterprises Informal sector 3 4 5 6 7 8 enterprisesa Householdsb 9 10 Source: Hussmanns 2004. Note: Cells shaded in dark gray refer to jobs that by definition do not exist in the type of production unit in question. Cells shaded in light gray refer to formal jobs. Unshaded cells represent the various types of informal jobs. Informal employment: cells 1–6 and 8–10. Employment in the informal sector: cells 3–8. Informal employment outside the formal sector: cells 1, 2, 9, and 10. ILO = International Labour Organization. a. As defined by the 15th International Conference of Labour Statisticians 1993 (excluding households that employ paid domes- tic workers). b. Households that produce goods exclusively for their own use and households that employ paid domestic workers. 66 Chapter 3: Data Preparation Sector of Employment and Occupation Knowing the sector of the economy and the occupations in which men and women are employed is as important as knowing if they are employed, unemployed, or out of the labor force. Sector refers to the type of economic activity in which workers are involved. It refers to the main activity of the firm, enterprise, or establish- ment as opposed to the main activity of the worker, which is the occupation. Sectoral classification is usually collected in the data using detailed industry classifications. The ILO recommendation is to follow the International Standard Industrial Classification (ISIC) for all eco- nomic activities.25 The scope of ISIC covers all (or almost all) productive activities—that is, economic activities within the production boundary of the System of National Accounts (EC and others 2008). These eco- nomic activities are subdivided into a hierarchical, four-level structure of mutually exclusive categories, which facilitates data collection, presenta- tion, and analysis at detailed levels of the economy in an internationally comparable, standardized way. The categories at the highest level are called sections, which are alphabetically coded categories intended to facilitate economic analysis. The sections subdivide the entire spectrum of productive activities into broad groupings, such as agriculture, forestry, and fishing (section A); manufacturing (section C); and information and communication (section J). The classification is then organized into successively more detailed categories, which are numerically coded: two- digit divisions, three-digit groups, and, at the greatest level of detail, four- digit classes.26 ADePT Gender works with three fields that capture different levels of detail, with different purposes. The Broad sector field should be completed with a categorical variable that differentiates agriculture, industry, and services.27 The Sector field should be completed with a categorical variable that offers more detail on the activity—for example, the one-digit ISIC. Finally, the Detailed sector field should be completed with a detailed sector classification, such as the four-digit ISIC, and is used to compute industry segregation indexes. The Agriculture field expects a bivariate variable to denote employment in the agriculture sector. This variable can include farmwork but also off-farm work in agriculture—for example, everything captured under ISIC section A according to the fourth ISIC revision (UN 2008a). Industry is commonly defined as ISIC sections B to F; thus, 67 Introducing ADePT Gender Software: Part I it includes mining, manufacturing, utilities, and construction. The services category includes all other ISIC sections (for example, ISIC sections G–U). Similar to the industry classification, there is an occupation classification. The most widely used classification is the International Standard Classification of Occupations (ISCO) produced by the ILO.28 ISCO is a tool for organizing jobs into a clearly defined set of groups according to the job’s tasks and duties. Its main aim is threefold: (a) to provide a basis for international reporting, comparison, and exchange of statistical and admin- istrative data about occupations; (b) to provide a model for the development of national and regional classifications of occupations; and (c) to establish a system that can be used directly in countries that have not developed their own national classifications. As opposed to sector, ADePT Gender has only one field to complete with occupation. The field Occupation requires a categorical variable like the one-digit ISCO. ADePT Gender expects categorical variables for the Broad sector, Sector, Detailed sector, and Occupation fields. ADePT Gender performs no background checks except for the variable input into the Broad sector field, for which ADePT Gender ensures that it only takes three values. Work Intensity Work intensity is measured in two ways: (a) by differentiating full-time and part-time jobs and (b) by looking directly at the number of hours worked. Both variables have complexities that are worth discussing. Many household surveys include a question about the full-time or part-time status of respondents’ employment. The ILO Part-Time Work Convention (1994, no. 175) defines a part-time worker as an employed person whose normal hours of work are fewer than those of comparable full- time workers (ILO 2004). Part-time work can be identified through a ques- tion or through the number of hours worked. Asking a direct question has advantages and disadvantages. On the one hand, it reflects the job require- ments with regard to time, which might be different from the number of hours worked. On the other hand, a self-assessment is subject to biases, since workers might be unaware of how many hours constitute a full-time job in their country. For example, the threshold between part-time and full-time work is 30 hours per week in Canada, Finland, and New Zealand; 35 hours in Australia, Japan, and the United States; and 37 hours in Norway (Messenger 2004). 68 Chapter 3: Data Preparation Another measure of work intensity is the hours of work per week (or equivalent period of reference) that are spent on activities that contrib- ute to the production of goods and services. The normal hours of work are those that workers are expected to spend on work activities during a short reference period, such as one day or one week, as stipulated in laws, regula- tions, collective agreements, arbitral awards, or establishments’ rules or customs. The usual hours of work are the average hours worked during a reference period, like the past calendar year. The hours actually worked are the hours spent on work activities during a specified reference period, such as the past week. These are presented as a list of elements of a day of work or work components and include (a) productive time (hours actually worked during normal periods of work and any additional time worked that is generally paid at a higher rate than normal rates, such as overtime); (b) time spent on ancillary activities (time spent at the workplace on such tasks as repairs and maintenance, preparation and cleaning of tools, and preparation of receipts, time sheets, and reports); (c) unproductive time spent in the course of the production process (time spent at the place of work waiting or standing by for such reasons as lack of workload, break- down of machinery, accidents, or time spent at the place of business during which no work is done but for which payment is made under a guaranteed employment contract); and (d) resting time (time corresponding to short rest periods at the workplace, including tea and coffee breaks). The defini- tion explicitly excludes paid time not worked, such as paid annual leave, paid public holidays, paid sick leave, meal breaks, and time spent on travel from home to work and vice versa. Surveys typically ask either the usual number or the actual number of work hours. Actual hours are preferred if the user is interested in exploring how the number of work hours varies over the year, for example, to study seasonality. Usual hours, in contrast, are less influenced by short-term fluc- tuations. If the interviewer asks for the number of hours of work for each separate job, the users should sum the hours of work from all jobs. Another variable that can be used to indicate work intensity is the number of weeks worked during a period of reference (such as the past calendar year or past 12 months). As with the number of hours, the phras- ing of the question can refer to the normal, the usual, or the actual weeks worked in a reference period. This question can be important if women are more likely than men to have temporary jobs or to hold jobs that have long out-of-work periods (such as teachers and farmers). 69 Introducing ADePT Gender Software: Part I ADePT Gender analyzes the full-time or part-time labor status and the number of hours worked, which are to be completed in the Full time and Hours fields, respectively. The Full time field can be filled in with the minimum number of hours required to be considered a full-time worker. A related issue that ADePT Gender does not explicitly include—though it can be analyzed through the use of custom variables or tables—is scheduling. This aspect can be critical for gender analysis, since women’s household responsibilities might constrain their access to jobs that require night shifts or unusual work hours. Scheduling work hours relates to the periods of the day, week, or month when work is done, such as in the morning, afternoon, or evening; from Monday to Friday; on weekends; as overtime; and so on. Scheduling work hours can be combined with the hours of work and information on their fixed or variable nature to derive a vast number of different working time arrangements. These relate to schedules that are different from regular full- time working schedules—that is, where workers are required by their employer or choose to work (a) less or more than full time, (b) only part of the year, (c) only part of the week, (d) at night, and (e) on weekends, in addition to starting or ending at different times and having variable daily or weekly schedules as part of flexible schedules or as part of “annualized” work- ing schemes, which fix working time over a long period of one year, allowing weekly schedules to vary. Labor Income Several concepts related to labor income need to be differentiated (Ehrenberg and Smith 2009). For paid employees, differences are made among wages, earnings, and income.29 The concept of wage rates relates to the basic price of a unit of labor, before adding any bonuses for overtime, shift work, or family allowance and before deducting contributions for social security schemes or advanced tax payments. Wage rates can be expressed in units of time, such as an hour, a week, or a month, or as piece rates. It is the smallest of all pay concepts and applies only to workers in paid employment. The concept of earnings typically relates to the pay that employers provide directly to their employees regularly during a specified reference period. It includes basic pay for time worked or work done, as well as for time not worked, such as vacation, holiday, and sick leave. It also includes 70 Chapter 3: Data Preparation other payments granted by the employer for various reasons, such as over- time work, hazard payments, regular bonuses, and fringe benefits, such as family allowances. However, it excludes all irregular bonuses, even if they are provided by the employer. Like time rates, earnings are recorded gross of social security contributions or tax deductions. The concept of income related to paid employment is the most comprehen- sive measure of workers’ remuneration. In addition to earnings, it includes all irregular bonuses and payments and all social security benefits received from the employer directly or from a social security scheme, if they are related to employment. These include family and education allowances, as well as sick and maternity benefits. They also include benefits received by those who are no longer employed, such as unemployment benefits, pen- sions, and invalidity benefits. All of these social benefits will be part of income from paid employment only insofar as workers received them because they participated in work activities. In countries with general social security systems, whereby family and other allowances are provided inde- pendently of work activities, these allowances are not part of income related to paid employment. To avoid double counting, these benefits are recorded net of contributions the worker makes to social security schemes. The Earnings field should be completed with the total earnings as per the above definition and summing the earnings from multiple jobs. ADePT Gender expects a continuous variable and will check only for usual nonre- sponse values, such as negative values or 9999. The most important require- ment is to have consistency in the reference period of the variables used for the Earnings and Hours fields. Access to Resources Productive assets determine the scale of production, investment, and growth. Farmers depend on land, labor, water, seeds, fertilizer, pesticides, machinery, and other inputs to produce crops. Entrepreneurs require labor and— depending on the business’s size and sector of operation—capital for investment. Access to credit is crucial for farmers and entrepreneurs. Despite women’s diverse and fundamental roles in agricultural and nonagricultural activities, women tend to have more limited access to productive assets, as well as inputs and services. Women also face additional constraints on their use of time, which are often tied to local norms and beliefs. ADePT Gender gives the user the flexibility to decide which outcomes to measure (screenshot 3.4). 71 Introducing ADePT Gender Software: Part I Screenshot 3.4: Economic Opportunities Tab Fields to input work characteristics variables Fields to input labor status variables Fields to input resource variables (admits more than one variable) The Resources field can be completed by one or many categorical vari- ables. The variables are expected to have a missing value if the household does not have access to or ownership of the resource, and positive integers are used to indicate the three types of ownership, access, or use of the resource: (a) only man, (b) only woman, or (c) both man and woman. It is up to the user to decide whether the variable should be restricted to all men and all women in the household or to the husband and wife. The following section offers examples and clarifies some measurement issues that are com- mon to all candidate variables. Household versus individual: Many household surveys collect informa- tion on access to assets at the household level. This approach implicitly assumes that the resources are equally shared or that consensus exists within the household on the allocation and use of resources. However, this approach does not do justice to the reality of different spheres of deci- sion making within households and intrahousehold inequalities in access to resources. Even when surveys ask questions about individual versus joint asset ownership, caution is warranted if this information is gath- ered from a single respondent in the household, usually a self-identified 72 Chapter 3: Data Preparation “most knowledgeable” household member. In particular, different percep- tions about ownership rights within the household may exist. Women may provide different responses when interviewed together with their partners, or household members may hide resources from each other. A recent survey experiment in Uganda under the UN Evidence and Data for Gender Equality initiative—which seeks to improve measure- ment of asset ownership from a gender perspective—sheds light on these issues by comparing different interviewing modalities. The results, though preliminary, suggest that reported ownership rates for a broad range of physical and financial assets increase significantly if all adult household members are interviewed individually (and alone) compared with inter- viewing a single household member (Kilic and Moylan 2016). Access to versus use of resources: Not using a resource does not necessar- ily mean that someone did not have access to it; it simply means that it was not optimal for the person to use it. This difference matters when we analyze financial services, which include managing financial transactions (including borrowing, saving, making payments, and having insurance). Having access to finance is not the same as using finance; it is having the possibility of using a financial service (Cull and Scott 2009). Having access to finance therefore implies the absence of price and nonprice obstacles regarding financial services. It also implies that measuring access to finance requires a combination of information on current use of formal and informal financial services, savings, payments, insurance, unmet demand, and details on all types of barriers faced by men and women, households, and firms (Demirgüç-Kunt, Beck, and Honohan 2008). However, it is more likely to have indicators that measure use of financial services—for example, having bank accounts or loans from either formal or informal institutions.30 Ownership or management of productive resources: Ownership and manage- ment of productive resources are very different concepts. Land rights are those property rights that pertain to real estate—that is, land. Because land is a limited resource and property rights include the right to exclude others, land rights are a form of monopoly. Landownership matters because it can be used as collateral to buy assets or to get a loan. It might also be indicative of intrahousehold bargaining power (Quisumbing and Maluccio 2003). However, from the viewpoint of inputs necessary for an income-generating activity, what matters is whether women can make decisions about how land is used; about the use of inputs such as fertilizer, pesticides, poultry, 73 Introducing ADePT Gender Software: Part I and so on (Croppenstedt, Goldstein, and Rosas 2013); and about the use of the output generated by the land. Agency The World Development Report on gender equality and development (World Bank 2012) proposes five outcomes or expressions of agency, all of which are closely associated with a woman’s ability (or inability) to make choices. These outcomes are related and often overlap; as a result, a woman’s ability to choose and act at any point in time partly reflects foundations laid earlier in her life, often starting in childhood.31 Household surveys can cover many of the indicators of each of these expressions, though users should exercise caution when using and interpreting these types of variables (see chapter 4 of this book). Marriage and Fertility One of the expressions of agency is a woman’s ability to decide when and whom to marry, when and how many children to have, and when to leave a marriage. ADePT Gender translates this concept into specific indicators. The Age at first marriage field should be completed with a continuous integer variable. Surveys like the DHS that contain fertility histories usually include this type of question. Ideally, it should be retrieved from a direct question, such as “When did you start living with your first husband/partner? In what month and year was that?” Surveys often instead ask, “In what month and year did you start living with your husband/partner?” The latter question captures only the length of the current marriage and does not nec- essarily indicate a woman’s ability to decide the timing of marriage, because previous marriages are taken into consideration. The DHS includes all the relevant information. Another indicator is average age at first birth. Again, the user should complete the field using a direct question, as opposed to computing the vari- able using the age of the oldest child. If the woman had a miscarriage or any of her children had died, the latter “estimated” variable would bias the indicator upward. ADePT Gender expects a continuous integer variable in the Age at first birth field. A woman’s control over when to have children and how many to have can be measured in several ways. The total fertility rate as well 74 Chapter 3: Data Preparation as the age-specific fertility rate will serve this purpose. To compute these two rates, the user needs to know the mother’s date of birth and the birth dates of all the children who were born alive. ADePT Gender computes these rates using the DHS’s method. Thus, ADePT Gender requires the date of the interview (Interview date field), the mother’s date of birth (Mother’s birth date field), and the birth dates of all the chil- dren (Children’s birth dates field) in century-month code format (CMC). For example, consider a woman interviewed in December 2001 who was born in May 1970. Her CMC date of interview would be 12 × (2001 − 1900) + 12 = 1224. Her CMC date of birth would be 12 × (1970 − 1900) + 5 = 845. WDR’s Five Dimensions of Agency Given the scarce availability on agency outcomes, ADePT Gender lets the user select variables. The only requirement is that the mean of the variable (or set of variables) to be completed in the field Agency has an economic meaning (screenshot 3.5). Hence, they have to be either bivariate variables Screenshot 3.5: Fields in the Agency Tab Four (out of five) dimensions of agency identified in the World Development Report and that can be described or proxied with household surveys 75 Introducing ADePT Gender Software: Part I with values of 0 or 1 or continuous variables. Examples of variables from the DHS for expressions of agency are as follows: • Control over resources: “Who usually decides how the money you earn will be used: mainly you, mainly your husband/partner, or you and your husband/partner jointly?” “Who usually decides how your husband’s/partner’s earnings will be used: you, your husband/partner, or you and your husband/partner jointly?” “Who usually makes deci- sions about your health care: you, your husband/partner, you and your husband/partner jointly, or someone else?” “Who usually makes decisions about major household purchases?” “Who usually makes decisions about making purchases for daily household needs?” Of course, this expression of agency is closely related to the Resources field in the Economic opportunities tab. • Ability to move freely: “Who usually makes decisions about visits to your family or relatives?” “In the past 12 months, how many times have you been away from home for one or more nights?” • Decision making over family formation: “Would you say that using con- traception is mainly your decision, mainly your husband’s/partner’s decision, or do you decide together?” • Freedom from the risk of violence: The DHS has a module on domestic violence that focuses mainly on physical abuse. However, a compre- hensive assessment of domestic violence will include physical, sexual, psychological, and economic violence (UN 2014). • Ability to have a voice in society and influence policy: Most of the time, these variables are collected at a different level. For example, the indicators on political participation come from administrative data on parliamentary representation. The information is collected via community questionnaires. However, very few of these sources have sex-disaggregated information. Although scarce, in some cases, household surveys include questions that can be used to con- struct indicators, such as whether they are members of cooperatives and other community associations. One example is the Indonesia Family Life Survey, which includes a module on community par- ticipation and such questions as “Have you participated in a com- munity association meeting in the past 12 months?” “How many community association meetings have you participated in during the past 12 months?” 76 Chapter 3: Data Preparation When choosing variables related to voice and agency, the user needs to bear in mind some characteristics that affect their interpretation. Variables can be (a) direct versus indirect measures of voice and agency, (b) intrinsic versus extrinsic, or (c) universal versus context specific.32 Direct versus indirect: Direct measures of voice and agency aim to capture increases in an individual’s ability to achieve goals and control his or her life. Indirect, or proxy, measures of voice and agency traditionally focus on the possession of resources necessary for agency or the determinants of being empowered—such as education or asset ownership—rather than on empow- erment itself. For example, decision-making variables are more direct mea- sures of agency, whereas indicators of asset ownership are more indirect measures. Although the general recommendation is to use direct measures as much as possible, it would be useful to explore further how direct mea- sures correlate with indirect measures. Intrinsic versus extrinsic: The intrinsic variables capture the value that people give to voice and agency, whereas the extrinsic variables relate to what people can do with those skills and abilities independently of the value they assign to them. Universal versus context specific: Voice and agency are context-specific concepts, since they are highly determined by socioeconomic, cultural, and political conditions. Currently, some initiatives feature indicators that can be compared across contexts and time, such as the Assets Project and the Women’s Empowerment in Agriculture Index (see chapter 8). Parameters Several age brackets relevant for the analysis vary from country to country. ADePT Gender allows the user to define age brackets according to the specific circumstance or the interest of the analyst, as explained in the fol- lowing section (screenshot 3.6). General Age Brackets Several tables in ADePT Gender present statistics for subgroups of the popu- lation defined by age. By default, ADePT Gender divides the population into five main groups: (a) children ages 0–14; (b) youth ages 15–24; (c) the adult population with women still in their fertility years, ages 24–49; (d) the adult 77 Introducing ADePT Gender Software: Part I Screenshot 3.6: Fields to Be Completed in the Parameters Tab population with women past their fertility years, ages 50–64; and (e) the elderly population, age 65 and older. These age brackets are detailed enough to earmark the main aspects of a woman’s life cycle. However, some users might be interested in analyzing certain groups in more detail, for example, teenagers ages 13–19 or elderly people between 65 and 70 years old. This analysis can be done using customized tables. Youth According to the United Nations World Population Statistics, the fastest- growing population is that of 15- to 24-year-olds who are also considered youths. An increase in this age group could result in high unemployment rates, if countries are unable to create jobs to meet the fast-growing labor force. High youth unemployment can discourage job searching among those ages 15–24. In addition, high youth unemployment can put downward pres- sure on employment and earnings in many countries (World Bank 2012). ADePT Gender generates some statistics that refer explicitly to this demo- graphic group. Education As explained earlier, years of education and the entrance age for primary education can vary from country to country. The number of years of educa- tion at each level can be tailored using the Primary school and Secondary school fields. 78 Chapter 3: Data Preparation Working-Age Population The minimum age for legal work, as well as the retirement age, varies from country to country. The user has the option to define the relevant age brackets for the working-age population using the Working age field. The general recommendation is to use the minimum age of legal work as the lower limit and the age of retirement as the upper limit. If these bounds do not coincide with the available information from the survey, the user should correct them to reflect the age of the people to whom the labor module was administered. Notes 1. The use of ADePT Gender software is explained in chapter 2. A detailed explanation of the software’s capabilities is provided by Lokshin and others (2013). 2. For more details, see the DHS Program website, http://www.measuredhs .com/Who-We-Are/About-Us.cfm. 3. Enterprise Surveys can be downloaded at http://www.enterprisesurveys.org. 4. For further details on how to compute and understand basic concepts related to sampling weights, the user can consult Deaton (1997), par- ticularly, chapter 1. 5. The latest revision of UN recommendations for conducting surveys and censuses eliminated the condition of sharing one meal a day, and instead it focuses on one-person households and multiperson households. A household may be either (a) a one-person household, that is, a person who makes provision for his or her own food or other essentials for living without combining with any other person to form part of a multiperson household; or (b) a multiperson household, that is, a group of two or more people living together who make common provision for food or other essentials for living. The people in the group may pool their resources and have a common budget; they may be related or unrelated persons or a combination of both related and unrelated. This arrangement exemplifies the “housekeeping” concept. Some countries use a concept different from the housekeeping con- cept, namely, the “household-dwelling” concept, which regards all persons living in a housing unit as belonging to the same household. According to this concept, there is one household per occupied 79 Introducing ADePT Gender Software: Part I housing unit. Therefore, the number of occupied housing units and the number of households occupying them are equal, and the loca- tions of the housing units and households are identical. Countries should specify in their census reports whether they used the “house- keeping” or the “household-dwelling” concept of a private household (UN 2008b, paragraph 2.108). 6. Notice that ADePT Gender admits only a single variable, as opposed to ADePT Poverty, where the Poverty line field can be filled with more than one variable,. 7. For a discussion of the advantages and disadvantages of each of these measures, see Deaton and Zaidi (2002). 8. See the Demographic and Health Surveys Program’s “Wealth Index” web page for a description of the construction of the wealth index, http://www.dhsprogram.com/topics/wealth-index/Wealth-Index -Construction.cfm. 9. Examples of direct variables could be those that evaluate household well-being, such as the Cantril ladder used by Gallup’s World Poll (Cantril 1965; Gallup 2009). 10. Categorical means that each category represented by the variable takes an integer or alphanumeric combination. 11. One reason why the term head of household is a complex concept is that the title does not necessarily correlate with household decision making. For example, if a woman’s husband migrates for work, she may be con- sidered the de facto head of household, although other male family members—such as an older son, father-in-law, or another adult male family member—may be making household financial decisions. Data analysts and policy makers should therefore consider the reasons why certain women become heads of households and consider this variation when looking at this group. 12. For an application of such a variable, see the Country Gender Assessment of Lesotho. 13. This variable might be relevant for eastern European countries; see Sattar (2012) for examples. 14. For example, in Mexico, whether or not the household is a beneficiary of PROGRESA (Programa de Educación, Salud, y Alimenación), a program to alleviate poverty. 15. Ethnicity is defined as large groups classified by cultural factors, such as nationality, culture, ancestry, language, and beliefs. Ethnicity is a variable 80 Chapter 3: Data Preparation that in certain contexts could be defined at the household level or at the individual level, if there are interracial marriages and within-household differences in ethnic affiliation. 16. For a detailed discussion on the definition of literacy, see UNESCO (2006), chapter 6. 17. Literacy level can be determined by different methods: (a) by respon- dents reporting their literacy level (self-declaration), (b) by another individual—typically, the head of the household—reporting on the literacy level of household members (third-party assessment), and (c) by the number of years of schooling completed determining the “literate” from the “‘nonliterate” (educational attainment proxy), which varies from country to country, ranging from one year up to eight years but typically four or five years of schooling. However, all of these methods are subject to measurement error and difficulty in the cross-country comparison. To respond to this problem, surveys and censuses started to use direct measures of literacy in the 1990s. These methods range from the interviewer giving a card to the respondent to evaluate his or her own literacy level to more complex skills measurement surveys, such as PISA (Programme for International Student Assessment), TIMSS (Trends in International Mathematics and Science Study), and STEP (Skills Toward Employment and Productivity). For more details on this discussion, see UNESCO (2006, 2015). 18. The current version of the ISCED was adopted in 1998 and is called ISCED 1997. For more details consult UNESCO (1997). 19. A more detailed discussion of the differences between enrollment and attendance can be found in the ADePT education manual (Porta and others 2011). 20. The ISCED 1997 mappings are available at http://www.uis.unesco.org /Education/ISCEDMappings/Pages/default.aspx. 21. Or expressions that ADePT interprets as binary variables. 22. For more details, see WHO (2010). 23. The three conditions must hold to be considered unemployed. 24. The underemployment rate is the number of underemployed persons as a percentage of the number of employed persons (Hussmanns 2007). The rate of volume of time-related underemployment is the ratio between the volume of time-related underemployment and the poten- tial time for work of employed persons, calculated as the sum of the 81 Introducing ADePT Gender Software: Part I total hours actually worked by employed persons and the volume of time-related underemployment (Hussmanns 2007). 25. The ISIC has four revisions: ISIC Revision 2 (1968), Revision 3 (1990), Revision 3.1 (2002), and Revision 4 (2008). The details of each classification and the correspondence among revisions can be found at http://unstats.un.org/unsd/cr/registry/isic-4.asp. 26. For more details, see UN (2008a). 27. If users consider it appropriate for the country context, they could use the alternative classification of economic sectors: (a) the primary sector com- prising agriculture, mining, and forestry; (b) the secondary sector compris- ing manufacturing and construction; (c) the tertiary sector comprising services, such as retail sales, entertainment, and financial services; and (d) the quaternary sector comprising intellectual activities like education. 28. ISCO was first adopted in 1957 and is also known as ISCO-58. The clas- sification was revised many times and is superseded by ISCO-68 (1966), ISCO-88 (1987), and ISCO-08 (2007). More details can be found at http://www.ilo.org/public/english/bureau/stat/isco/isco08/index.htm. 29. Definitions are taken from the ILO “Income Statistics” web page, http://www.ilo.org/global/statistics-and-databases/statistics-overview -and-topics/income/lang--en/index.htm. All definitions refer to a gross concept of remuneration. 30. Good questions have a definite reference period (such as, within the past year or past six months). 31. See chapter 4 of World Bank (2012) for a proper definition of each dimension of agency and more details on these concepts. 32. This discussion is based on Alkire and others (2012). References Alkire, Sabina, Ruth Meinzen-Dick, Amber Peterman, Agnes Quisumbing, Greg Seymour, and Ana Vaz. 2012. “The Women’s Empowerment in Agriculture Index.” OPHI Working Paper 58, Oxford Poverty and Human Development Initiative, Oxford University. Anderson Scheffner, Julie. 2000. “Module for Chapter 9: Employment.” In Designing Household Survey Questionnaires for Developing Countries: Lessons from Ten Years of LSMS Experience, edited by Margaret Grosh and Paul Glewwe, 147–250. Washington, DC: World Bank. 82 Chapter 3: Data Preparation Bardasi, Elena, Kathleen Beegle, Andrew Dillon, and Pieter Serneels. 2011. “Do Labor Statistics Depend on How and to Whom the Questions Are Asked? Results from a Survey Experiment in Tanzania.” World Bank Economic Review 25 (3): 418–47. Beaman, Lori, and Andrew Dillon. 2012. “Do Household Definitions Matter in Survey Design? Results from a Randomized Survey Experiment in Mali.” Journal of Development Economics 98 (1): 124–35. Cantril, Hadley. 1965. The Pattern of Human Concerns. New Brunswick, NJ: Rutgers University Press. Croppenstedt, Andre, Markus Goldstein, and Nina Rosas. 2013. “Gender and Agriculture: Inefficiencies, Segregation, and Low Productivity Traps.” Policy Research Working Paper 6370, World Bank, Washington, DC. https://openknowledge.worldbank.org/handle/10986/13171. Cull, Robert, and Kinnon Scott. 2009. “Measuring Household Usage of Financial Services: Does It Matter How or to Whom You Ask?” Policy Research Working Paper 5048, World Bank, Washington, DC. Deaton, Angus. 1997. The Analysis of Household Surveys: A Microeconometric Approach to Development Policy. Baltimore: Johns Hopkins University Press. Deaton, Angus, and Salman Zaidi. 2002. “Guidelines for Constructing Consumption Aggregates for Welfare Analysis.” Living Standards Measurement Study Working Paper 135, World Bank, Washington, DC. Demirgüç-Kunt, Ash, Thorsten Beck, and Patrick Honohan. 2008. Finance for All: Policies and Pitfalls in Expanding Access. Washington, DC: World Bank. Diez de Medina, Rafael, and Elisa Benes. 2014. “Redrawing the Boundaries: From Employment and Unemployment towards Work and Labour Underutilization.” International Labour Organization, Geneva. EC, IMF, OECD, UN (European Commission, International Monetary Fund, Organisation for Economic Co-operation and Development, United Nations), and World Bank. 2008. System of National Accounts: 2008. New York: EC, IMF, OECD, UN, and World Bank. http://unstats .un.org/unsd/nationalaccount/sna2008.asp. Ehrenberg, Ronald G., and Robert S. Smith. 2009. Modern Labor Economics. 10th ed. Boston: Pearson/Addison-Wesley. Fuwa, Nobuhiko. 2000. “The Poverty and Heterogeneity among Female- Headed Households Revisited: The Case of Panama.” World Development 28 (8): 1515–42. 83 Introducing ADePT Gender Software: Part I Gallup. 2009. “Understanding How Gallup Uses the Cantril Scale.” http:// www.gallup.com/poll/122453/Understanding-Gallup-Uses-Cantril -Scale.aspx?version=print. Hussmanns, Ralf. 2004. “Measuring the Informal Economy: From Employment in the Informal Sector to Informal Employment.” Working Paper 53, International Labour Organization, Geneva. ———. 2007. “Measurement of Employment, Unemployment and Underemployment: Current International Standards and Issues in Their Application.” International Labour Organization, Geneva. http://www .ilo.org/global/statistics-and-databases/WCMS_088394/lang--en /index.htm. IHSN (International Household Survey Network). 2015. “How Well Are Gender Issues Covered in Household Surveys and Censuses? An Analysis Using the IHSN–World Bank Gender Data Navigator.” http://ihsn.org/HOME/sites/default/files/resources/Gender_Issues_July -2015.pdf. ILO (International Labour Organization). 1982. “Resolution Concerning Statistics of the Economically Active Population, Employment, Unemployment and Underemployment, Adopted by the 13th International Conference of Labour Statisticians.” http://www.ilo.org /public/english/bureau/stat/download/res/ecacpop.pdf. ———. 1993. “Report of the Conference,” Fifteenth International Conference of Labour Statisticians. Geneva, January 19–28. http://www .ilo.org/public/libdoc/ilo/1993/93B09_65_engl.pdf. ———. 2004. “80 Years of ILO Statistical Standard Setting,” Seventeenth International Conference of Labour Statisticians, Geneva, November 24–December 3, 2003. http://www.ilo.org/public/english/bureau/stat /download/articles/2004-1.pdf. ———. 2012. Measuring Informality: A Statistical Manual on the Informal Sector and Informal Employment. Geneva: ILO. ———. 2013. “Resolution Concerning Statistics of Work, Employment and Labour Underutilization.” Presented at the Nineteenth International Conference of Labour Statisticians, Geneva, October 2–11. http://www .ilo.org/wcmsp5/groups/public/---dgreports/---stat/documents /normativeinstrument/wcms_230304.pdf. Kilic, Talip, and Heather Moylan. 2016. “Methodological Experiment on Measuring Asset Ownership from a Gender Perspective (MEXA).” 84 Chapter 3: Data Preparation Technical report, World Bank, Washington, DC. http://siteresources .worldbank.org/INTLSMS/Resources/3358986-1423600559701/MEXA _Technical_Report.pdf. Lokshin, Michael, Sergiy Radyakin, Zurab Sajaia, and William Creitz. 2013. ADePT User Guide. Version 5. Washington, DC: World Bank. Messenger, Jon C., ed. 2004. Working Time and Workers’ Preferences in Industrialized Countries: Finding the Balance. London and New York: Routledge. Panama, Government of. 1997. “ISCED Mapping.” Ministry of Education, Panama City. http://www.uis.unesco.org/Education/ISCEDMappings /Pages/default.aspx. Perry, Guillermo, William F. Maloney, Omar S. Arias, Pablo Fajnzylber, Andrew D. Mason, and Jaime Saavedra-Chanduvi. 2007. Informality: Exit and Exclusion. Washington, DC: World Bank. Porta, Emilio, Gustavo Arcia, Kevin Macdonald, Sergiy Radyakin, and Michael Lokshin. 2011. Assessing Sector Performance and Inequality in Education. Washington, DC: World Bank. Quisumbing, Agnes, and John A. Maluccio. 2003. “Resources at Marriage and Intrahousehold Allocation: Evidence from Bangladesh, Ethiopia, Indonesia, and South Africa.” Oxford Bulletin of Economics and Statistics 65 (3): 283–327. Sattar, Sarosh. 2012. Opportunities for Men and Women in Emerging Europe and Central Asia. Washington, DC: World Bank. UN (United Nations). 2008a. “International Standard Industrial Classification of All Economic Activities (ISIC), Rev 4.” Statistical Papers Series M, no. 4, New York. ———. 2008b. Principles and Recommendations for Population and Housing Censuses Revision 2. New York: UN. http://unstats.un.org/unsd/publication /seriesM/seriesm_67Rev2e.pdf. ———. 2014. Guidelines for Producing Statistics on Violence against Women: Statistical Surveys. New York: UN. UNESCO (United Nations Educational, Scientific, and Cultural Organization). 1997. “International Standard Classification of Education: ISCED 1997.” Paris. http://www.unesco.org/education/information /nfsunesco/doc/isced_1997.htm. ———. 2006. Education for All: Literacy for Life. Paris: UNESCO. http:// unesdoc.unesco.org/images/0014/001416/141639e.pdf. 85 Introducing ADePT Gender Software: Part I ———. 2015. Education for All 2000–2015: Achievements and Challenges. Paris: UNESCO. http://unesdoc.unesco.org/images/0023/002322/232205e.pdf. WHO (World Health Organization). 2010. World Health Statistics, 2010. Geneva: WHO. World Bank. 2012. World Development Report 2012: Gender Equality and Development. Washington, DC: World Bank. https://openknowledge .worldbank.org/handle/10986/4391. 86 PART II Producing a Country Gender Diagnostic Part II of this manual guides the user through the ADePT output that feeds into a country gender diagnostic. This is the set of tables and graphs that describe the three dimensions of gender inequality highlighted by the 2012 World Development Report (WDR) framework: human endowments, economic opportunities, and agency. Chapter 4 provides guidance on how to produce and interpret the tables and graphs. Using the household surveys from Nepal and Panama, the chapter discusses selected tables and graphs to show the user how to inter- pret them. Chapter 5 provides additional information about theoretical models and statistical techniques that aid the interpretation of results. This organization aims to help different audiences to quickly access nec- essary inputs and outputs. Users familiar with quantitative analysis and gender issues can consult chapter 4, whereas users less familiar with quanti- tative analysis or the economic angle of gender issues should consult chapters 4 and 5. 87 Chapter 4 How to Interpret the Results of the Country Gender Diagnostic This chapter discusses how to interpret the results of the tables and graphs produced by ADePT Gender, which aspires to be as comprehensive as pos- sible, while using a minimum set of key variables. At the same time, it leaves the door open to further customization to fully exploit the richness of the data on hand. ADePT Gender can be used for diagnostics, for tracking prog- ress over time, and for evaluating country performance with respect to peer countries. In addition, many of the statistics here are part of the 52 mini- mum core gender indicators identified by the United Nations (UN) Inter- Agency and Expert Group on Gender Statistics.1 It also helps to highlight needs for further analysis in areas that have persistent gender inequality or knowledge gaps. The set of tables and graphs also allows users to identify gaps in gender-relevant data. The country gender diagnostic aims to answer such questions as: Are female infants as likely as male infants to be born in a country? Do female and male infants receive the same attention and care? Are girls as likely as boys to attend school? Are women as engaged in paid work as men? And if so, do they have access to the same jobs and payment for their work and education? How much can differentiated access to productive assets such as land and credit be associated with the gaps in economic opportunities for men and women? Do women and men have equal decision-making power in 89 Producing a Country Gender Diagnostic: Part II the household and in society? These are among the questions that this chapter will help the user answer using simple statistical tools. The output produced is by nature descriptive. However, in combination with the broader economic literature, it can help the user to infer which mechanisms are most likely to produce those outcomes. If needed, the user can pursue further analysis to establish causal links and measure the effects from different drivers of change. These needs may arise (a) because the user is interested in disentangling different mechanisms that could be working simultaneously, (b) because it is an area of high priority, or (c) because a unique opportunity exists to measure a policy’s effect (via a quasi-natural experiment, for example). The diagnostic uses simple basic tabulations to describe the differences between men and women in the main outcomes and by relevant population subgroups. Progress in some domains is tempered by the sobering realities that women face in others (World Bank 2012). This chapter is divided into four parts. The first discusses how to place gender in the country’s demographic context. Tables and graphs present the distribution of women across the country’s main demographic and regional groups. The following three sections describe gender differences in outcomes in the three main domains identified by the WDR 2012: Gender Equality and Development: human capital (or endowments), economic opportunities, and agency (see chapter 1). Each section is organized in the same way. The content of the output produced by ADePT Gender is first presented. Next, basic concepts are introduced that help illustrate the content of the tables and graphs. The last section describes how to interpret the output. When interpreting the results, the user should bear in mind that differ- ent values of the statistics for men and women do not necessarily reflect the existence of gender gaps. These differences might not be statistically significant; they might result from lack of precision2 of the estimates to be confident—from a statistical point of view—of the existence of such difference. The user interested in assessing whether the difference is statisti- cally significant needs to conduct additional tests. These tests—discussed more extensively in appendix A—can be carried out by simple math calcu- lations based on ADePT Gender outputs (means and standard deviations) or by using standard statistical software packages. Two datasets are used to provide the user with practical examples of output and as reference to interpret the results: (a) the Nepal Living Standards Survey for 1996, 2003, and 2010 and (b) the Panama Living 90 Chapter 4: How to Interpret the Results of the Country Gender Diagnostic Standard Measurement Study for 2008.3 All results are based on the defini- tions discussed in chapter 3. Applying the same definitions to all datasets is key to ensure comparability across countries. Only selected tables and graphs are described in detail in the chapter as an illustration. Demographic and Regional Characteristics Content This section guides the reader through seven tables (tables 1a–1g) and five graphs (figures 1a–1e) that show the distribution of women across differ- ent groups of the population, defined by either individual or household characteristics. Generally, the graphs show the share of women within each group, whereas tables show the distribution of men and women according to individual and household characteristics. ADePT Gender tables 1a and 1b show the distribution of men and women according to such characteristics, respectively. Tables 1c and 1d show the distribution of the population by the household head’s gender and individual characteristics and by the household head’s gender and select household characteristics, respectively. Figures 1a–1d show the share of women (or women heads of household) for each of the groups defined in tables 1a–1d. Finally, detailed information on household composition and age composition is provided in tables 1e and 1f and figure 1e. Some examples from specific country datasets are also presented below. Concepts When preparing gender diagnostics or examining gender disadvantages within the context of poverty assessments, labor market studies, or other analytical country work, it is important to ascertain whether women are disproportionately represented among certain demographic, regional, or income groups. This is because gender inequalities tend to be amplified for severely disadvantaged populations, even when women have the same poten- tial and right as men to a full, meaningful life. Across and within countries, gender gaps widen at lower incomes. And in the poorest economies, gender gaps are even larger. In general, the benefits of economic growth have not accrued equally to all men and women of a society. Household poverty can mute the effect of national development, and gender differences—even if 91 Producing a Country Gender Diagnostic: Part II shared by all—often compound or exacerbate other types of exclusion, such as geography and ethnicity. Understanding the degree of overlap between gender disparities and other vulnerabilities is fundamental for correctly interpreting gender differences in outcomes. For example, in some countries, gender disparities remain significant only for those who are poor. In both India and Pakistan, although boys and girls from the top income quintile participate in school at similar rates, the bottom income quintile has a gen- der gap of almost five years (World Bank 2012). In the former Yugoslav Republic of Macedonia, Roma women are considerably less likely than non- Roma women to work (Angel-Urdinola and Macias 2008; Gamberoni and Posadas 2012). Often, certain population groups are more vulnerable to external shocks— economic, political, and institutional—which can erase gender equality gains or cause reversals, with welfare losses for boys and men. For example, economic shocks in many poor Latin American countries result in boys leaving school at young ages. Such adverse circumstances early in life can have long-lasting effects, even irreversible ones. Moreover, when it comes to gender equality, it seems that countries or population groups often fall in either virtuous or vicious circles, or “sticky gets stickier,” as the World Bank (2012) puts it. Progress in one dimension of gender can multiply the effects of other dimensions. Under this hypothesis, it is useful to understand which groups of the population may fall into virtuous or vicious circles. Missing Women Basic demographic analysis helps explore the high rates of missing women and adult male mortality. The term missing women highlights a country’s excess female mortality that results from different behaviors, preferences, and expo- sure to health risks: (a) prenatal sex preferences that translate into “stopping behavior”4 and sex-selective abortions, (b) excessive female mortality in early childhood, (c) maternal mortality, and (d) higher human immunodeficiency virus (HIV) risks. Age-specific mortality rates differ between males and females because of various biological and behavioral factors. Although female life expectancy exceeds male life expectancy in high-income countries, in poor-income countries, women, especially young women and girls, die at higher rates than men. To a large degree, the reason is the lack of access to health services, in particular, those related to reproductive health. However, sex-selective abortions are also an important factor in some countries. 92 Chapter 4: How to Interpret the Results of the Country Gender Diagnostic A combination of three factors led to falling numbers of female births (World Bank 2012). First, fertility started to drop as female education and returns to the labor market increased. Second, new technologies allow pre- natal sex screening to become widespread and available to all sectors of society. Third, the preference for sons remains unchanged, and families that want to have only two children strongly prefer to have at least one son. Population Pyramids and Demographic Transitions Population pyramids typically consist of two back-to-back bar graphs, with the population plotted on the x-axis and age on the y-axis. One graph shows the number of males, and the other shows females in a particular population in five-year age groups (also called cohorts). Males are conventionally shown on the left and females on the right, and they may be measured by raw num- ber or as a percentage of the total population. Demographic structures are classified by four types: (a) demographic explosion, (b) demographic window of opportunity, (c) demographic implosion, and (d) demographic hourglass.5 See figure 4.1 for a plot of each type. It is important to understand the population’s demographic composition for policy making and other issues. Gender inequality affects demographic processes, which in turn further affect economic outcomes (Buvinic, Das Gupta, and Casabonne 2009). For example, if a country is transitioning from a demographic explosion to a demographic window of opportunity, it should ensure that the present cohort of girls will be ready to participate in the labor market when they become adults. Population pyramids allow missing women to be identified when the imbalance is acute. For instance, an asymmetric population pyramid may suggest differences in sex ratios at birth, mortality differences by gender, or gender-differentiated migration patterns. However, as described above, males and females of different ages face different survival probabilities, so that the quantification of missing women requires more complex computations and the use of model life tables from countries expected to have little or no sex discrimination as benchmark (Klasen and Wink 2003; World Bank 2012). Household Composition The term household composition refers to the description of household mem- bers by gender and age. Before analyzing the results, it is important to 93 Producing a Country Gender Diagnostic: Part II Figure 4.1: Demographic Typology a. Demographic explosion, b. Demographic window of opportunity, Niger 2005 Republic of Korea 2005 80+ 80+ 70–74 70–74 60–64 60–64 50–54 50–54 Age Age 40–44 40–44 30–34 30–34 20–24 20–24 10–14 10–14 0–4 0–4 0.10 0.05 0 0.05 0.10 0.10 0.05 0 0.05 0.10 Proportion of total population Proportion of total population c. Demographic implosion, d. Demographic hourglass, Japan 2025 Germany 1950 80+ 80+ 70–74 70–74 60–64 60–64 50–54 50–54 Age Age 40–44 40–44 30–34 30–34 20–24 20–24 10–14 10–14 0–4 0–4 0.10 0.05 0 0.05 0.10 0.10 0.05 0 0.05 0.10 Proportion of total population Proportion of total population Male Female Source: Buvinic, Das Gupta, and Casabonne 2009. understand whether household composition differs across vulnerable groups to assess household differences in access to resources and other constraints. Two types of indicators are usually used that present the same type of infor- mation but in different contexts—dependency ratio and a direct description of household members’ age and gender. 94 Chapter 4: How to Interpret the Results of the Country Gender Diagnostic The dependency ratio is an indicator intended to capture the house- hold’s potential—from a demographic perspective—to earn income. Table 1e shows the dependency ratio, defined as ⎛ number of HH members ages 0 to 14 + ⎞ ⎜ ⎝ number of HH members 65 and older ⎟ ⎠ Dependency ratio = × 100. number of HH members ages 15 to 64 As the ratio increases, the burden on the household’s productive members to support the economically dependent—children and elderly—also increases. This effect results in direct impacts on household expenditures, in particular, on expenses related to health and education. The dependency ratio is typically higher for female-headed households, because women often assume headship when the working-age male is absent. Thus, at the aggre- gate level, the dependency ratio needs to be analyzed jointly with the per- centage of the population in female-headed households and with their heads’ characteristics. Two other indicators are used in the literature but are not included in ADePT Gender. They are the child dependency ratio and elderly depen- dency ratio: number of HH members ages 0 to 14 Child dependency ratio = × 100, number of HH members ages 15 to 64 number of HH members 65 and older Elderly dependency ratio = × 100. number of HH members ages 15 to 64 Interpreting the Results ADePT Gender figures 1a–1d are all interpreted in the same way. They show the percentage of women or female heads for different groups of the population defined using individual or household characteristics. For example, ADePT Gender figure 1a shows the percentage of women in each of five age groups—children, youth, reproductive-age adults, mature adults, and elderly. Figure 4.2 reproduces ADePT Gender figure 1b and shows that in Panama in 2008, more women than men lived in female-headed 95 Producing a Country Gender Diagnostic: Part II Figure 4.2: ADePT Gender Figure 1b, Panama 2008 head’s sex residence Household Area of Urban Rural Male Female Poverty status Poor Nonpoor Residence and Urban, poor poverty Urban, nonpoor Rural, poor Rural, nonpoor Quintiles of welfare Lowest quintile aggregate 2 3 4 Highest quintile Western Region Eastern Metropolitan Central 0 20 40 60 80 100 Percent Male Female Source: Based on ADePT Gender using the Panama 2008. households (60 percent versus 40 percent, respectively). In male-headed households, the gender balance is more even (54 percent versus 46 percent, respectively).6 ADePT Gender tables 1a–1d also show the distribution of certain groups of the population among categories defined by a certain characteristic. 96 Chapter 4: How to Interpret the Results of the Country Gender Diagnostic The population group being analyzed is indicated in the column head, and the categories among which the group is distributed are indicated in the rows in successive blocks. Take, for example, ADePT table 1a. It examines three population groups—total country population, men, and women— and three individual characteristics—age, marital status, and completed education. The top left block indicates the distribution of the total country population by age categories, whose sum is the total population, that is, 100. ADePT table 1d examines other population groups—male- and female- headed households, alone (for the total country population), and combined with two characteristics typically found to exacerbate the gender imbalances—rural and poor populations. As before, these groups are com- bined with other characteristics—household area of residence, household poverty status, the combination of residence and poverty status, household quintile, and region of residence—indicated in the table rows. ADePT figures 1a and 1b and ADePT tables 1a and 1b analyze men and women by select individual (figure 1a and table 1a) and household (figure 1b and table 1b) characteristics. ADePT Gender figures 1c and 1d and ADePT Gender tables 1c and 1d analyze female- and male-headed house- holds by select individual characteristics of the head (figure 1c and table 1c) and the household (figure 1d and table 1d). The characteristics used to define the groups of ADePT figures 1a–1d are aligned with the characteris- tics used in ADePT tables 1a–1d. Women are often, though not always, overrepresented in vulnerable groups, such as female-headed, poor, and rural households. In the case of Panama, a larger fraction of women live in female-headed households than in male-headed households (31 percent versus 21 percent, ADePT table 1b). Whether this difference is mechanical (that is, resulting from the fact that no or fewer male adults are present in female-headed households) depends on the definition of household head (see chapter 3) and on the country context. In Panama, more men than women live in rural households (37 percent versus 34 percent, table 4.1 that corresponds with ADePT table 1b), though this difference is not statistically significant.7 The per- centage of women living in poor households is low compared with other countries. Table 4.2 lists 65 countries grouped by region and by the three ranges of the percentage of poor that are women. No difference exists in the poverty status of the household. It should be noted that because the welfare measures are at the household level, such analysis ignores any differences in resource allocations among household members. 97 Producing a Country Gender Diagnostic: Part II Table 4.1: ADePT Gender Table 1b, Panama 2008 Table 1b: Distribution of Males and Females in the Population across Selected Household Characteristics Gender All Male Female Total 100.0 100.0 100.0 Area of residence Urban 64.3 63.0 65.5 Rural 35.7 37.0 34.5 Gender of household head Male 73.7 79.0 68.3 Female 26.3 21.0 31.7 Poverty status Poor 32.7 33.2 32.2 Nonpoor 67.3 66.8 67.8 Residence and poverty status Urban, poor 11.4 11.2 11.5 Urban, nonpoor 52.9 51.8 54.0 Rural, poor 21.4 22.0 20.7 Rural, nonpoor 14.4 15.0 13.8 Quintile of welfare aggregate Lowest quintile 20.0 20.6 19.3 2 20.0 19.8 20.2 3 20.0 19.9 20.1 4 20.0 19.6 20.4 Highest quintile 20.0 20.1 19.9 Region Western 19.7 19.7 19.6 Eastern 2.7 2.8 2.6 Metropolitan 58.0 57.2 58.8 Central 19.6 20.3 18.9 Source: Based on ADePT Gender using Panama 2008. Note: The complete ADePT output for Panama 2008 and Nepal 2011 can be found on the ADePT web- site, under the tab for “Gender,” at http://go.worldbank.org/0GA4FDMQY0. Hence, any disproportionate share of women living in poor households reflects solely demographic patterns. If gender gaps in the allocation of resources among household members were considered, gender gaps in poverty rates would be considerably more pronounced in some country contexts. Moreover, in general, there are overlapping vulnerabilities. For example, women are more likely to live in poor households and in female-headed households, and female-headed households may also be more likely to be poor.8 Or women may be more likely to live in female-headed households, and female-headed households may be more likely to be headed by an elderly woman. For example, in Panama, 25 percent of women and 17 per- cent of men live in female-headed households, and 70 percent of 98 Chapter 4: How to Interpret the Results of the Country Gender Diagnostic Table 4.2: Countries by Share of Women in Total Population Living in Poor Households, 1999–2008 Below 50 percent 50–54 percent 55–61 percent Latin America Latin America More More and the and the developed developed Africa Asia Caribbean Africa Asia Caribbean regions Asia regions Benin China Panama Burkina Faso Bhutan Belize Belgium Cyprus Austria Mali Philippines Paraguay Cameroon Bolivia Denmark Armenia Bulgaria Cabo Verde Brazil Finland Czech Congo, Rep. Chile France Republic Congo, Dem. Colombia Germany Estonia Rep. Costa Rica Greece Iceland Guinea Dominican Hungary Italy Kenya Republic Ireland Latvia Niger Ecuador Luxembourg Lithuania El Salvador Malta Norway Guatemala Netherlands Slovak Haiti Poland Republic Honduras Portugal Slovenia Jamaica Romania United Mexico Serbia States Nicaragua Spain Peru Sweden Uruguay United Kingdom Venezuela, RB Source: UN 2010. Note: Year is the latest available in the period of reference. female-headed households are headed by a woman with no education, as opposed to 43 percent of male-headed households (ADePT tables 1b and 1c for Nepal, not shown9). Hence, women might be at a disadvantage for being in a female-headed household and in a household whose head has no education. Users should interpret results with caution when overlapping vulnerabilities exist, since the confounding effects prevent drawing conclu- sions about which factors are more oppressive. Also, overlapping vulnerabilities might mutually reinforce each other. In general, women tend to have less access to resources, tend to be unprotected by legislation, or are subject to social norms that affect their agency, which in turn affects economic opportunities. For example, in Panama, female- headed households are more likely to be headed by widows. If widowed or divorced women have less access to resources because of inheritance or divorce laws, then gender disparities in outcomes might be a consequence of women’s unequal access to resources upon marriage dissolution and may not be due to the fact that women are household heads (Buvinic and Gupta 1997; World Bank 2012). 99 Producing a Country Gender Diagnostic: Part II Thus, if the data allow, it is useful to complement the output of ADePT Gender with a description of the factors that are pushing women to fall into each vulnerable group—for example, what causes them to become heads of households. Women might become heads because of marriage dissolution or because husbands had to migrate for a job or are unemployed. The conse- quences in each case are very different for household well-being. Female- headed households with migrant husbands might receive remittances—and thus are less likely to be poor—whereas female-headed households with unemployed husbands depend on female earnings only—and thus are more likely to be poor. In other words, the percentage of female-headed house- holds might be endogenous to other factors, either economic or social. If there is a large percentage of female-headed households, users should con- sider additional custom variables, such as whether the husband is present or his labor status, to obtain cross-tabulations and further examine the factors behind the phenomenon. Household Composition Table 4.3 (that corresponds to ADePT table 1e) shows a higher dependency ratio in female-headed households than in male-headed households—78 for male-headed households and 89 for female-headed households. Tabulation Table 4.3: ADePT Gender Table 1e, Panama 2008 Table 1e: Dependency Ratio by Selected Household Characteristics and by Male- and Female-Headed Households: Total, Rural, and Poor Households Total Rural Poor Male Female Male Female Male Female Total 78.0 88.9 98.1 107.8 120.5 126.2 Area of residence Urban 64.8 83.0 113.7 131.2 Rural 98.1 107.8 98.1 107.8 123.4 121.3 Poverty status Poor 120.5 126.2 123.4 121.3 120.5 126.2 Nonpoor 56.3 72.0 60.8 82.1 Residence and poverty status Urban, poor 113.7 131.2 — — 113.7 131.2 Urban, nonpoor 54.9 70.6 — — — — Rural, poor 123.4 121.3 123.4 121.3 123.4 121.3 Rural, nonpoor 60.8 82.1 60.8 82.1 — — (continued) 100 Chapter 4: How to Interpret the Results of the Country Gender Diagnostic Table 4.3: ADePT Gender Table 1e, Panama 2008 (continued) Total Rural Poor Male Female Male Female Male Female Quintile of welfare aggregate Lowest quintile 132.9 129.2 134.3 128.3 132.9 129.2 2 88.8 112.7 86.7 104.9 98.7 122.7 3 66.6 82.9 67.6 89.4 — — 4 58.8 67.1 54.2 65.3 — — Highest quintile 38.5 54.2 38.3 53.0 — — Region Western 98.7 108.7 114.0 124.0 138.8 132.4 Eastern 111.4 125.8 112.0 131.4 121.9 166.0 Metropolitan 68.1 84.5 89.1 104.1 117.9 132.2 Central 79.5 81.7 84.1 91.1 103.4 107.3 Source: Based on ADePT Gender using Panama 2008. Note: Individual (population) weights. This table is representative of the national population. — = not applicable. of the household head’s marital status (not shown, ADePT table 1c) further suggests that most female-headed households are a consequence of marriage dissolution. ADePT Gender output shows the household composition in more detail by calculating the average number of household members by gender and age in ADePT table 1f. As with ADePT tables 1c–1e, the averages are presented for the total population and separately for rural and poor households, each time distinguishing between male- and female-headed households. For example, in Panama, on average households have 3.7 members, one adult male, one adult female, and most likely three children, with relatively equal probability of having boys and girls (table 4.4). Very few households have elderly men and women. Reference values of dependency ratios can be found in UNFPA (2011). Population Pyramids Additional details on the demographic composition of the population are important for understanding other issues. ADePT figure 1e illustrates the population pyramid for Nepal, which shows a decrease in the number of young men ages 20 to 45, which might reflect international migration (see figure 4.3). Figure 4.4 shows population pyramids for Azerbaijan and Armenia, which have among the highest levels of missing women in the 101 Producing a Country Gender Diagnostic: Part II Table 4.4: ADePT Gender Table 1f, Panama 2008 Total Rural Poor Total Male Female Male Female Male Female Total 3.7 3.8 3.5 4.1 3.7 5.4 5.2 1 0.8 1.0 0.5 0.9 0.5 1.0 0.6 2 0.8 0.8 1.0 0.7 0.9 0.9 1.0 3 0.3 0.3 0.3 0.3 0.3 0.5 0.5 4 0.3 0.3 0.3 0.3 0.3 0.4 0.5 5 0.4 0.4 0.3 0.5 0.4 0.7 0.7 6 0.3 0.3 0.3 0.4 0.4 0.7 0.7 7 0.2 0.2 0.2 0.3 0.2 0.5 0.4 8 0.2 0.2 0.2 0.3 0.2 0.4 0.4 9 0.1 0.2 0.1 0.2 0.1 0.2 0.1 10 0.1 0.1 0.3 0.1 0.3 0.1 0.3 Source: Based on ADePT Gender using Panama 2008. Note: Individual (population) weights. This table is representative of the national population. Adult men and adult women include head of household. If the number of adult men in a male-headed house- hold is 1, this means he is the head. Figure 4.3: ADePT Gender Figure 1e, Nepal 2010–11 95–99 90–94 85–89 80–84 75–79 70–74 65–69 60–64 Age group 55–59 50–54 45–49 40–44 35–39 30–34 25–29 20–24 15–19 10–14 5–9 0–4 2,500 2,000 1,500 1,000 500 0 500 1,000 1,500 2,000 2,500 Population Male Female Source: Based on ADePT Gender using Nepal 2012b. 102 Chapter 4: How to Interpret the Results of the Country Gender Diagnostic Figure 4.4: Population Pyramid in Countries with Gender Imbalance at Birth a. Azerbaijan (resident population b. Armenia (resident in 2009) population in 2013) 70 and over 80+ 60–69 75–79 70–74 55–59 65–69 50–54 60–64 45–49 55–59 Age category Age category 40–44 50–54 45–49 35–39 40–44 30–34 35–39 25–29 30–34 20–24 25–29 15–19 20–24 15–19 10–14 10–14 5–9 5–9 0–4 0–4 0 0 0 0 0 0 10 00 15 00 20 00 0 15 00 10 00 50 0 50 0 00 00 00 00 00 00 00 00 00 ,0 0 0 0 0 ,0 0 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 60 40 20 20 40 60 20 Male Female Sources: South Caucasus Country Gender Assessment 2015; World Bank 2014. Note: The Azerbaijan pyramid and the Armenia pyramid are in absolute numbers of persons. world, after China and India. The gender asymmetry in the base of the pyramid hints at this problem. In some countries, it is worthwhile to recalculate the population pyra- mids using other background characteristics. For example, if important internal waves of migration occur from rural to urban areas, where women are left behind, the pyramids should also be generated for urban and rural areas. Figure 4.5 shows the case of Kenya and the Russian Federation, where the same information that is shown in the population pyramid is plotted in a bar graph. In panel a, we see that more men than women are in urban areas, while the opposite is true in rural areas. However, the gender imbal- ance is not symmetric, suggesting that aside from an internal migration pat- tern of men from rural to urban areas, there might be migration waves of women outside the country or adult female mortality. In the Russian Federation, the pattern points to the adult mortality rate, which seems to be more acute in urban areas than in rural areas. 103 Producing a Country Gender Diagnostic: Part II Figure 4.5: Sex Ratio by Age and Urban and Rural Areas for Kenya, 2005, and Russian Federation, 2006 a. Kenya Sex ratio (men per 100 women) 250 200 150 100 50 0 4 9 + 4 9 4 9 4 9 4 9 4 9 4 9 4 0– 5– 75 –1 –1 –2 –2 –3 –3 –4 –4 –5 –5 –6 –6 –7 10 15 20 25 30 35 40 45 50 55 60 65 70 Age groups b. Russian Federation Sex ratio (men per 100 women) 120 100 80 60 40 20 0 4 9 0+ 15 4 20 9 25 4 30 9 35 4 40 9 45 4 50 9 55 4 60 9 65 4 70 9 75 4 80 9 85 4 90 9 95 4 10 9 0– 5– –1 –1 –2 –2 –3 –3 –4 –4 –5 –5 –6 –6 –7 –7 –8 –8 –9 –9 10 Age groups Urban Rural Source: UN 2010. Human Capital Content This section navigates the user through five ADePT tables (tables 2a–3b) that profile gender disparities in human capital: education, nutri- tion, and health. ADePT Gender tables 2a–2c show education outcomes, 104 Chapter 4: How to Interpret the Results of the Country Gender Diagnostic whereas ADePT tables 3a and 3b show health and nutrition outcomes. ADePT tables 2a–2c and 3a compare boys and girls, or men and women, whereas ADePT table 3b analyzes reproductive health outcomes for women of reproductive age. All outcomes are presented for the total country population and for select groups on the basis of household char- acteristics: gender of the household head, area of residence, poverty sta- tus, residence and poverty status combined, well-being quintile, and region of residence. Concepts We interpret human capital as the investment an individual receives or makes over the course of life in health, nutrition, and education. Investments in human capital have an intrinsic value, and a society that cares about equal opportunities for boys and girls and men and women should exhibit gender parity in human capital outcomes. This view is reflected in the Millennium Development Goals, which include gender par- ity targets. Most countries also have mandatory schooling laws that are gender blind, implying that countries value boys’ and girls’ education equally. For girls, better outcomes in health and education reinforce each other. Studies show that girls who are in school are more likely to be healthy—they are more likely to have a higher body mass index—and have a higher probability of finding wage work when compared with girls who are not in school. In contrast, education does not significantly change the out- comes for boys (World Bank 2012). In addition, human capital is instrumental, as it determines both current and future productivity. Poor health outcomes cause health-related absences from the labor force and lower numbers of work hours and thus lower earn- ings (Case and Deaton 2003). Therefore, gender gaps in human capital accumulation generate and perpetuate wedges in labor market outcomes as reflected, for example, in the Human Opportunity Index (Molinas and oth- ers 2010). Children who are unhealthy have an increased probability of not growing into healthy adults; consequently, they cannot make meaningful economic contributions to the country. Low birth weights and childhood exposure to diseases have been linked to lower cognitive development, lower school attainment, less learning in adolescence, poorer maternal health, and even higher crime rates (Almond, Currie, and Hermann 2012; Case, Ferting, and Paxon 2005; Currie and others 2010; Currie and Tekin 2012). 105 Producing a Country Gender Diagnostic: Part II Better-educated children will be more productive adults, which increases countries’ economic opportunities and reduces poverty and inequality. Finally, women’s human capital also matters for household allocation of resources and affects the transmission of inequalities to future genera- tions (Strauss and Thomas 1995; Thomas 1990). Healthier and better- educated women are more likely to make better choices for themselves and also for their offspring (Currie and Moretti 2003; Oreopoulos, Page, and Stevens 2006). ADePT Gender lets the user choose health and nutrition indicators. In addition, ADePT Gender examines gender disparities in net and gross enrollment rates and in education attainment. The net enrollment rate (NER) is an indicator of access to education with respect to gender and other dimensions. Net enrollment rates are calculated as population enrolled of gender g of the official age group for the level of education Net enrollment rate eg = × 100. population of gender g of the official group for the level of education The NER is not calculated for all education levels. Although theoreti- cally possible and computationally feasible in ADePT Gender, tertiary education is left out since wide variations occur in the duration of programs at this level, which introduces difficulties in determining the denominator in the definition above. The gross enrollment rate (GER) is also an indicator of access to education and shows the general level of participation in a given level of education. Unlike the NER, it does not count only children of official school age in the numerator. Hence, the GER is a complementary indicator, which denotes— in conjunction with the NER—the extent of overage and underage enroll- ment. Gross enrollment rates can exceed 100 percent and are calculated as population enrolled of gender g for the level e of education Gross enrollment rate eg = × 100. population of gender g of the official age group for the level e of education Education attainment is measured as the percentage of the population— men or women—that has completed a certain level of education. Therefore, education attainment is calculated for only the adult working-age population. 106 Chapter 4: How to Interpret the Results of the Country Gender Diagnostic Gender gaps in other education indicators can be examined using ADePT Education (described in box 4.1). The past decades have seen much progress toward gender parity in edu- cation. Three main factors have been identified as increasing female enroll- ment in school: (a) higher returns to schooling in the labor market, (b) the removal of institutional constraints, and (c) higher household income (World Bank 2012). First, when returns to education in the labor market increase for women, parents’ investments in girls’ schooling also increase. Box 4.1: Other ADePT Modules: Education and Health ADePT Education The ADePT Education module of the ADePT software—with its accompanying manual—produces output on education indicators and education inequality. It ana- lyzes inequality in school participation, progression, and attainment for boys and girls. The output of ADePT Gender follows the same reasoning and structure, though it does not go into the same level of detail as ADePT Education. For example, ADePT Gender produces average gross and net enrollment rates by gender, but ADePT Education also produces such indicators as proportion out of school, gross and net intake rate, grade 1 students older than official grade 1 age, typology of those out of school, and so on. It also gives further disaggregation by gender and background char- acteristics (age, place of residence, gender, and quintile). Aside from these core statistics, the software produces three additional sets of tables and graphs. The first group analyzes education inequality by computing standard inequality measures for years of schooling and earnings and for several slices of the population (gender crossed with age, place of residence, quintile). The two inequality measures are the concentration and the Theil indexes. Another group of tables exam- ines household spending on education using household consumption data and goes into as much detail as possible with regard to type of expenditure. The third group of tables and graphs concentrates on youth and covers several employment outcomes and their interaction with education. ADePT Health ADePT Health is divided into two parts: health outcomes and health financing. The part on health outcomes can provide additional details, as this module covers more indicators than ADePT Gender and some of the tables are disaggregated by gender. A few of the tables also focus on maternal health. However, overall, this module is less suited to gender analysis than ADePT Education. Users interested in going into more depth in these areas can easily switch between ADePT modules by clicking on the Module tab in the top bar. Sources: Porta and others 2011; Wagstaff and others 2011. 107 Producing a Country Gender Diagnostic: Part II Higher returns to education for women can result from better labor market opportunities, for instance, from the introduction of new technologies, outsourcing of production, and so forth (Jensen 2010; Oster and Millet 2010). Second, the removal of economic barriers to enrollment can be direct, reducing the cost of schooling (fees, uniforms, and books), or indi- rect, reducing the opportunity cost (wages that children could earn outside school or the value of household chores). Households weigh the price of schooling and the opportunity cost of children. The balance might turn out to be different for boys and girls. Higher children’s employment opportunities are associated with low school enrollment for boys (Edmonds, Pavcnik, and Topalova 2009). This is because boys are on average more likely than girls to engage in agricultural or other productive work. It has been estimated that the opportunity cost of education is about 10 times higher for boys than for girls (World Bank 2012). On the other hand, girls tend to have more responsibilities for house- hold chores. Girls collect water, take care of younger siblings, and substitute for their mother in domestic activities. If domestic work is costly, house- holds will be more likely to take girls out of school. Higher household income reduces the need to rely on children’s work. However, when households face economic shocks—because of a national economic crisis, a drought, or the unemployment of a breadwinner—a com- mon coping mechanism is to reduce investments in education. Again, given that the opportunity cost and the returns to schooling are different for boys and girls, the shock might affect boys and girls differently. In middle-income countries, girls are more likely to drop out of school when household income experiences a shock. However, in high-income countries, boys with higher labor market opportunities are more likely to leave school (World Bank 2012). By the same token, safety nets that help households weather such shocks help keep children in school. The same forces that affect the investment in children’s education affect investments in health. It is worth noting that typically hardly any differ- ences exist between boys’ and girls’ vaccination rates or nutrition outcomes. In addition, little or no association exists between excess mortality of girls and gender differences in vaccinations or access to nutrition or health care. The main factors explaining excess female mortality for infants and young girls is related to sanitation (World Bank 2012). Besides the social prefer- ences for sons (discussed earlier) and households’ unequal investment in boys and girls, there is a concern that health service providers discriminate 108 Chapter 4: How to Interpret the Results of the Country Gender Diagnostic in treatment, though hard evidence of such discrimination is difficult to come by (World Bank 2012). Interpreting the Results ADePT tables 2a–2c, 3a, and 3b show the mean value of the indicator speci- fied in the column head for boys and girls (tables 2a, 2b, and 3a) or men and women (tables 2c and 3a), or reproductive-age women (3b) for the total of the country’s population and for select groups defined by household characteristics—indicated in the table rows. For example, table 4.5 reproduces ADePT table 2a using data from the 2010 Nepal Living Standards Survey. It shows that, overall, girls are more Table 4.5: ADePT Gender Table 2a, Nepal 2010 Table 2a: Male and Female Net Enrollment Rates by Selected Household Characteristics: Primary and Secondary Education Primary Secondary Male Female Male Female Total 81.2 84.2 4.1 4.1 Gender of household head Male 79.5 83.1 4.1 3.3 Female 86.1 87.4 4.3 6.2 Area of residence Urban 82.2 84.7 9.0 9.2 Rural 81.0 84.1 3.0 3.0 Poverty status Poor 78.6 81.6 0.5 0.8 Nonpoor 82.8 86.0 5.5 5.6 Residence and poverty status Urban, poor 75.3 81.3 1.7 1.3 Urban, nonpoor 83.3 85.4 9.9 10.5 Rural, poor 78.8 81.6 0.4 0.7 Rural, nonpoor 82.6 86.2 4.1 4.1 Quintile of welfare aggregate Lowest quintile 79.1 80.5 0.0 0.8 2 79.3 84.3 1.3 0.4 3 80.8 82.8 3.1 2.1 4 82.7 86.2 5.0 7.3 Highest quintile 87.2 92.0 10.9 11.1 Region Eastern 83.2 87.0 4.2 4.2 Central 76.1 77.4 5.9 4.2 Western 82.1 87.1 4.6 4.8 Midwestern 86.0 86.2 0.7 3.9 Far western 83.9 92.3 1.9 2.6 Source: Based on ADePT Gender using Nepal 2012b. 109 Producing a Country Gender Diagnostic: Part II likely than boys to enroll in primary school (84 percent versus 81 percent, respectively), but they are equally likely to enroll in secondary school (4.1 percent for both). The primary school advantage of girls persists when gender is crossed with other population groups (poverty, area of residence, and so forth). Figure 4.6 provides some reference values for primary and secondary enrollment rates by region of the world. Table 4.6 further shows the results on maternal health for Nepal in 2010. Although 80 percent of pregnant women had at least one prenatal care consultation during preg- nancy, only 30 percent were assisted during the delivery. As table 4.7 shows, this percentage is extremely low when compared with the global average for developing countries but also with the 50 percent average rate for the Southern Asia region. In general, gender imbalances in school enrollment vary with level of education. Despite the narrowing trend in gender education gaps, in most countries, boys are more likely than girls to be enrolled in primary educa- tion, and young women are more likely than boys to be enrolled in tertiary education—especially in countries with high rates of tertiary enrollment (World Bank 2012). However, important heterogeneity emerges both within and across countries, reflecting factors such as the gender of the household head, the level of household income, the labor market opportu- nities, and social norms, which can vary by region and between rural and urban areas. In particular, higher income tends to increase the education enrollment and attainment of both boys and girls, and it often narrows gender education gaps or even reverses them in favor of girls. In addition, women tend to be more aware than men about the importance of investing in human capital, especially for female household members. Thus, in households where women have more control over resources, we observe larger investments in health and education resulting in higher gender parity in education and health. However, female-headed households typically have lower incomes, and this may have adverse effects on gender equality within the household as resources may be concentrated on boys, and girls might drop out of school to substitute for working mothers in household responsibilities. Higher opportunity costs may also reduce investments in education in rural areas. Boys might be needed on the farm, whereas girls might be needed to care for younger siblings. Distance to school can increase the cost of education (in money, time, or both) and might contribute to favor- ing boys’ enrollment over girls’. Also, lack of school infrastructure—such 110 Chapter 4: How to Interpret the Results of the Country Gender Diagnostic Figure 4.6: Primary and Secondary Net Enrollment Rates, by Gender and Region, 1999 and 2007 a. Primary World Less-developed regions More-developed regions Africa Eastern Africa Middle Africa Northern Africa Southern Africa Western Africa Asia Eastern Asia South-Central Asia South-Eastern Asia Western Asia Europe Eastern Europe Northern Europe Southern Europe Western Europe Latin America and the Caribbean Caribbean Central America South America North America Oceania 0 10 20 30 40 50 60 70 80 90 100 Percent b. Secondary World Less-developed regions More-developed regions Africa Eastern Africa Middle Africa Northern Africa Southern Africa Western Africa Asia Eastern Asia South-Central Asia South-Eastern Asia Western Asia Europe Eastern Europe Northern Europe Southern Europe Western Europe Latin America and the Caribbean Caribbean Central America South America North America Oceania 0 10 20 30 40 50 60 70 80 90 100 Percent 1999 Women 2007 Women 1999 Men 2007 Men Source: UN 2010, figures 3.8 and 3.13. Note: Year is the latest available in the period of reference. 111 Producing a Country Gender Diagnostic: Part II Table 4.6: ADePT Gender Table 3b, Nepal 2010 Table 3b: Maternal Health Outcomes for Women Ages 15–49 and Select Household Characteristics Delivery Prenatal Postnatal assistance control control Total 0.3 0.8 0.2 Gender of household head Male 0.3 0.8 0.2 Female 0.4 0.8 0.2 Area of residence Urban 0.7 0.9 0.3 Rural 0.3 0.8 0.2 Poverty status Poor 0.2 0.6 0.1 Nonpoor 0.4 0.9 0.2 Residence and poverty status Urban, poor 0.3 0.8 0.1 Urban, nonpoor 0.8 0.9 0.3 Rural, poor 0.2 0.6 0.1 Rural, nonpoor 0.4 0.9 0.2 Quintile of welfare aggregate Lowest quintile 0.2 0.6 0.1 2 0.2 0.8 0.2 3 0.3 0.8 0.2 4 0.5 0.9 0.2 Highest quintile 0.7 1.0 0.4 Region Eastern 0.4 0.8 0.2 Central 0.3 0.8 0.2 Western 0.4 0.9 0.2 Midwestern 0.2 0.7 0.2 Far western 0.3 0.7 0.2 Source: Based on ADePT Gender using Nepal 2012b. Table 4.7: Women Receiving Prenatal Care, Skilled Assistance at Birth, and Deliveries in Health Facilities, by Region, 1996 and 2000–08 Percentage pregnant Percentage Percentage women receiving prenatal deliveries with deliveries in care (at least 1 visit) skilled assistance health facilities 1996 2000–08 1996 2000–08 1996 2000–08 Africa Northern Africa 65 80 66 82 57 78 Southern Africa 86 92 67 78 64 72 Eastern, Middle and 66 79 42 53 37 48 Western Africa Asia Eastern Asia 93 94 95 98 89 94 South-Eastern Asia 77 77 64 62 52 48 (continued) 112 Chapter 4: How to Interpret the Results of the Country Gender Diagnostic Table 4.7: Women Receiving Prenatal Care, Skilled Assistance at Birth, and Deliveries in Health Facilities, by Region, 1996 and 2000–08 (continued) Percentage pregnant Percentage Percentage women receiving prenatal deliveries with deliveries in care (at least 1 visit) skilled assistance health facilities 1996 2000–08 1996 2000–08 1996 2000–08 Southern Asia 49 68 39 52 28 46 Central Asia 90 94 93 96 92 91 Western Asia 82 91 82 89 79 86 Latin America and the Caribbean Caribbean 95 96 88 92 86 79 Central America 75 90 70 82 62 76 South America 79 91 80 86 76 85 Oceania 84 – 81 81 87 – Eastern Europe 97 97 99 100 98 99 Source: UN 2010, table 2.4. Note: Year is the latest available in the period of reference. as separate bathrooms for boys and girls—as well as sexual harassment might push girls and young women out of school. In general, the absolute and relative costs of education for girls and boys are the most relevant fac- tor for determining gender educational level and parity. Policies to elimi- nate primary-level school fees and provide universal primary education were often associated with improving gender parity at the primary level (World Bank 2012). When comparing and interpreting ADePT tables 2a and 2b, the user should remember that the net and gross enrollment rates provide different information. Both rates show education coverage for a certain level of edu- cation; however, the NER is constrained by age, whereas the GER is not. Thus, a positive difference between the GER and NER indicates incidence of underage and overage enrollment because of early or late entrants and grade repetition, respectively. If the GER exceeds 100 percent, additional information about the extent of repetition and late entrants is required to better interpret this indicator. The complement of the NER is usually thought to provide a measure of the proportion of children not enrolled at the specified level of education, but some of these children and youths could be enrolled at other levels. Some possible explanations for observed gender differences between GER and NER include (a) girls can enroll later than boys or be the first to drop out in difficult times (World Bank 2012), and (b) if overage children are more likely to drop out of school, then boys might be more likely to be 113 Producing a Country Gender Diagnostic: Part II overage than girls, if they have to combine school with work. Boys and young men often exhibit higher repetition and dropout rates than their female counterparts at the primary level. However, whether the pressure to drop out is higher for boys or for girls depends on the country (UNESCO 2012, 43, 56). The same explanations applied to schooling are relevant to gender gaps in boys’ and girls’ health and nutrition. Because of country efforts to vacci- nate all children, vaccination rates tend to indicate full coverage of boys and girls. However, gender gaps might still be observed in some countries or for other outcomes, such as stunting, underweight, wasting, or vitamin deficien- cies (such as vitamin A and salt iodization). Studies in South Asian countries using siblings who still live in the same household have found a higher per- centage of boys who are fully immunized compared with girls (Singh 2012). However, differences in human capital investments are not only a matter of boys and girls but also of the adult population. Given that adult education is rare, changes over time in adult education attainment are the result of education improvements in younger cohorts. However, other factors might be at play for adult health outcomes. Men and women might have different risks for certain diseases and factors, such as HIV or maternal health. As before, income, access to infrastructure, and access to information are key factors associated with preventive health care. For maternal health, access to prenatal care, skilled assistance at birth, and postnatal care are important factors that determine maternal mortality rates and children’s health. Health outcomes are often strongly associated with the availability of care, infrastructure, and similar factors. ADePT Education and ADePT Health are great resources that go beyond ADePT Gender (see box 4.1). Economic Opportunities Content This section guides the user through 27 tables (ADePT tables 4–7) and seven figures that describe gender disparities in economic opportunities. ADePT tables 4 and ADePT figure 4 show the gender gap at the extensive margin—that is, in engagement in economic activities. ADePT tables 5 and 6 and ADePT figure 5 show gender gaps at the intensive margin, including type of employment and hours of work. ADePT tables 7a–7d 114 Chapter 4: How to Interpret the Results of the Country Gender Diagnostic show labor market wages. ADePT table 8 profiles gender differences in access to resources, and ADePT table 9 tabulates the main labor market outcomes for different groups of the population that could be linked to channels affecting economic opportunity: (a) number of children to cap- ture family burden, (b) levels of education to capture investments in human capital, and (c) age to capture life-cycle channels, such as fertility and firm- specific and on-the-job human capital investments. Most of the tables are presented for the total working-age population, disaggregating for men and women. When describing employment, tables and figures represent most of the times men and women in wage employment. As before, further analysis is done for three vulnerable groups: rural and poor households, separately for male- and female-headed households. Concepts The indicators of economic opportunities profiled in the ADePT output are grouped into four categories. The first two refer to the extensive and intensive margin of participation in economic activities. The extensive margin simply measures participation, not taking into account the depth of the engagement. The intensive margin looks at depth by examining the type of employment and the number of hours worked. The third group is a measure of labor productivity—wages—and the fourth focuses on access to resources that might influence the observed labor market outcomes. This comprehensive set of out- comes is important given that the main message from the WDR 2012: Gender Equality and Development (World Bank 2012) is that gender gaps in productiv- ity do not reflect that women are less productive farmers, entrepreneurs, or workers than men. Rather, gender differences in productivity and earnings are a result of differences in the type of economic activities that employ women (occupations and sectors) and women’s limited access to resources. Labor Market Participation The working-age population can be employed, unemployed, or neither. A person is in the labor force if he or she is either employed or unemployed. A person not in the labor force is said to be out of the labor force or inactive. The size of the labor force (LF) is given by LF = E + U, 115 Producing a Country Gender Diagnostic: Part II where E is the number of employed persons, and U is the number of unem- ployed persons. Note that, as described in chapter 3, characterizing a person as employed does not take into account how many hours he or she works. Thus, this indicator does not say anything about the “intensity” of work. The labor force participation rate is defined as E +U LFP = × 100, WAP where LFP is the labor force participation rate, and WAP is the total working-age population of the country. If we want to construct the labor force participation of men or women, we need to adjust the formula in the following way: Eg + U g LFPg = × 100, WAPg where g is the gender (male or female), LFPg is the labor force participation rate of gender g, Eg is the number of employed persons of gender g, Ug is the number of unemployed persons of gender g, and WAPg is the number of working-age persons of gender g. The employment rate is defined as Eg Employment g = × 100. WAPg And the unemployment rate is Ug Unemployment g = × 100. LFg The above rates can be adjusted to represent a population group. In each case, it needs to make the subgroup the reference group. Take, for example, the labor force participation rate E gx + U gx LFPgx = × 100, WAPgx 116 Chapter 4: How to Interpret the Results of the Country Gender Diagnostic where x represents persons belonging to a certain group. The group can be defined either by individual or household characteristics. A group of particular interest is the working poor, defined as employed persons who live in a household whose members are estimated to be below the nationally defined poverty line. The number of working poor can be calculated using the equation Working poor = poverty rate × labor force15, where labor force15 is the labor force ages 15 years and older. The key assump- tion behind using the labor force instead of employment numbers is that all, or nearly all, of the poor in the labor force are employed. This assumption is made because in countries with no social safety nets, poor individuals must work to maintain a subsistence level. Note that 15 years and older is typically used to define a country’s standard working-age population. Some countries, however, apply other age limits. The nationally defined working-age popula- tion is what should be used here. The working poverty rate is the proportion of working poor in total employment: EP Working poverty rate = × 100, E where Ep is the number of employed persons living in a household with income below the poverty line, and E is the total number of employed per- sons. The version of the working poverty rate by gender is E Pg Working poverty rate g = × 100, Eg where we condition the expression on the gender of the employed population. Another group of interest could be youths. Low labor force participation of young people should not be a concern if they are studying. However, a common problem is idleness. A young person is said to be idle if he or she is not in school or training, not employed, and not looking for job (World Bank 2014).10 In other words, the concept of idleness serves to sort the percentage of out-of-labor force youths who are studying from those who are not. 117 Producing a Country Gender Diagnostic: Part II Job/Employment Characteristics The intensive margin is used to express the intensity of work. To measure the intensive margin, ADePT Gender provides two types of indicators: (a) the type of job, and (b) the number of hours worked as a measure of job engagement. Systematic differences exist between men’s jobs and women’s jobs, whether across sectors, industries, occupations, job types, or firm types. The term employment segregation or occupational segregation refers to the unequal concentration of men and women in different jobs or occupations. Women are often more likely than men to work in agriculture and in many service sectors. Women are also overrepresented among unpaid workers and in the informal sectors. In many countries, better educated women seek out work in the public sector, which is viewed as socially acceptable. Women are more likely to be teachers, nurses, clerical workers, and sales and service employees. Men are more likely to work in construction and transport sectors. The clustering of women in particular groups is in part the reflection of the fact that some jobs are by nature less demanding with regard to hours and responsibilities and thus are friendlier to women who have significant family responsibilities. However, persisting social norms are another impor- tant factor. The evidence suggests that various factors are at play that result in gender segregation and gender gaps in hours worked. One is economic structure. As countries develop, their economic structure changes and with it the types of jobs that are available (Gaddis and Klasen 2014; Mammen and Paxson 2000). Rural farm employment is replaced by city jobs, for instance, in factories or the service sector. This change of economic struc- ture means more salaried and fewer informal jobs. These same patterns that are seen across countries are observed within a country by comparing low- and high-income households. Beyond the economic structure, other factors influence occupational segregation and the number of hours worked. The WDR (World Bank 2012) argues that the four main factors are (a) gender differences in time– use patterns, (b) access to productive inputs, (c) the effects of markets and institutional failures, and (d) the constraints imposed by the choice in edu- cational field (Flabbi 2011). Women are more likely to respond to market signals—higher wages—when other market and institutional failures are absent. For example, female labor force participation tends to grow faster 118 Chapter 4: How to Interpret the Results of the Country Gender Diagnostic when childcare services are in place or when labor participation (or the occupation) is seen as socially acceptable. More details on each of these factors can be found in World Bank (2012). Earnings Earnings are the main outcome by which to compare men’s and women’s performance in the labor market, either when running their own business or when working for others. A large part of the gender wage gap and the aver- age productivity difference between male- and female-led firms is explained by occupational and industry segregation, as the jobs that are also more likely to be performed by women tend to be paid less. Furthermore, in a fairly large number of countries and occupations, segregation is universal, which prevents the analysis from finding a comparator group to even calcu- late a gender wage or productivity gap (Ñopo, Daza, and Ramos 2011). Moreover, the difference in earnings can be masked by benefits; thus, as explained in chapter 3, earnings should include all types of remunerations. However, it is not always possible to have reliable data in earnings, and the analysis should use wages.11 To control for differences in productivity and compensation on the job that are related to the number of hours worked, ADePT Gender compares the earnings per hour of work for men and women. Interpreting the Results Labor Market Participation ADePT tables 4a–4d show three main indicators for economic participation— labor force participation rate, employment rate, and unemployment rate— for men and women by select individual and household characteristics. Each number represents the rate corresponding to the label indicated in the column head (for example, labor force participation rate) and for a popula- tion group that is determined by the combination of the column subhead (for example, men) and the row (for example, ages 15–24). The tables show the same breakdown used in previous tables, according to individual characteris- tics (table 4a), household characteristics (table 4b), or the combination of the two (tables 4c and 4d). ADePT figure 2 shows the percentage of women and men in each of three mutually exclusive labor statuses: employment, 119 Producing a Country Gender Diagnostic: Part II unemployment, and out-of-the labor force. Knowing the percentage of women in the employed or the unemployed population is highly important for policy makers. If a country is planning to introduce an unemployment benefit or raise labor taxes, for example, this information would indicate which group of the population would be most affected by the reform. Globally, labor force participation is typically higher for men than for women. For example, in Panama in 2008, about 50 percent of women and 82 percent of men participated in the labor market (see table 4.8 showing ADePT table 4a). Factors affecting female and male participation relate to household-level decisions of labor supply and the division of tasks (see chap- ter 5 for a brief presentation of the household labor supply model). Map 4.1 provides a world map showing the wide variation of female labor force par- ticipation across countries Moreover, this indicator is highly correlated to development (measured by gross domestic product per capita) exhibiting a U-shaped function (see figure 4.7).12 The gender gap in unemployment rate can favor men or women, since opposing forces exist. On the one hand, women often face greater diffi- culties in finding a job than men. Women might have smaller networks and thus less access to information about job vacancies; they might be more selective in the jobs they apply for, so as to balance work and fam- ily; and they might face discrimination in the hiring process, resulting in higher unemployment rates and longer unemployment spells. On the other hand, women might become discouraged after a period of unem- ployment, abandon the job search sooner than men, and exit the labor force. If the latter effect is stronger than the former, women will have both lower unemployment rates and lower labor force participation rates than men. As before, these factors could result in various outcomes for different sectors of the population. For example, better-educated women might have greater incentives to work, since the cost of staying at home (measured by the lost wage) is higher than for less educated women. However, if better- educated women marry better-educated men (assortative matching), their nonlabor income tends to be higher, and thus they also have more incen- tives to stay at home.13 If the latter effect dominates, we would observe in ADePT table 4a that labor force participation decreases with the level of education. In the case of Panama, we observe that the first effect dominates the second, since female labor force participation increases with the level of education (table 4.9). 120 Chapter 4: How to Interpret the Results of the Country Gender Diagnostic Table 4.8: ADePT Gender Table 4a, Panama 2008 Table 4a: Male and Female Labor Force Participation, Employment, and Unemployment Rates by Selected Individual Characteristics Labor force Share of population participation rate Employment rate Unemployment rate out of the labor force Male Female Male Female Male Female Male Female Total 82.2 49.7 80.3 47.7 1.9 2.0 17.8 50.3 Age 15–24 59.3 31.4 56.1 28.5 3.2 2.9 40.7 68.6 25–34 93.1 58.2 91.5 55.7 1.6 2.5 6.9 41.8 35–44 96.2 63.6 95.1 61.2 1.1 2.3 3.8 36.4 45–54 92.4 59.3 91.2 58.3 1.2 1.1 7.6 40.7 55–64 76.9 35.8 75.4 35.8 1.5 0.0 23.1 64.2 Marital status Union 95.1 46.1 93.5 44.4 1.5 1.6 4.9 53.9 Married 90.8 49.8 89.8 48.7 1.0 1.2 9.2 50.2 Married separated 89.7 73.5 86.9 71.1 2.8 2.5 10.3 26.5 Union separated 89.5 66.9 88.6 63.6 0.9 3.3 10.5 33.1 Divorced 81.6 71.8 77.3 70.0 4.3 1.8 18.4 28.2 Widowed 83.7 44.8 83.7 43.4 0.0 1.4 16.3 55.2 Single 66.1 45.7 63.4 42.7 2.7 3.0 33.9 54.3 Education No education 85.2 39.2 84.2 38.9 1.0 0.3 14.8 60.8 Primary 78.5 36.9 77.0 35.4 1.6 1.5 21.5 63.1 Secondary 87.3 55.6 84.1 52.0 3.2 3.6 12.7 44.4 Post-secondary 85.9 72.7 84.0 70.0 1.9 2.7 14.1 27.3 Source: Based on ADePT Gender using Panama 2008. Note: The working-age population is ages 15–64. Primary and secondary refer to completed levels. Map 4.1: Female Labor Force Participation around the World 70+ 60 – <70 50 – <60 40 – <50 0 – <40 No data IBRD 38771 August 2011 Source: World Bank 2012. 121 Producing a Country Gender Diagnostic: Part II Figure 4.7: U-Shape Relationship between Female Labor Force Participation and GDP, 1990 and 2010 100 Russian Federation Soviet Union 2010 Female labor force participation 1990 80 2010 60 1990 40 20 0 6 7 8 9 10 11 12 Natural logarithm of GDP per capita, PPP adjusted Source: Gamberoni, Munoz Boudet, and Posadas 2014. Note: GDP = gross domestic product; PPP = purchasing power parity. Alternatively, women in poor households may be more likely to work outside the home to supplement household income or to do unpaid work (working in the household or on the farm). If this hypothesis were true, ADePT table 4b should show that female labor force participation rates decrease when we move from the bottom to the top quintile of the welfare aggregate. Our example of Panama does not reflect this case. Female labor force participation for the bottom quintile is 36 percent, but it rises to 65 percent at the top of the welfare distribution. Obviously, these factors do not operate in isolation. Both education and household income level matter, as do many other factors. ADePT tables 4a–4e show only partial effects, which ought to be complemented with further regression analysis and in-depth research to understand the contribution of each factor. Part III of this manual describes how to further analyze gender gaps in labor market outcomes to estimate the contribution of different variables to gender gaps in the labor market. In addition, the 122 Chapter 4: How to Interpret the Results of the Country Gender Diagnostic Table 4.9: ADePT Gender Table 4b, Panama 2008 Table 4b: Male and Female Labor Force Participation, Employment, and Unemployment Rates by Selected Household Characteristics Labor force Share of population participation rate Employment rate Unemployment rate out of the labor force Male Female Male Female Male Female Male Female Total 82.2 49.7 80.3 47.7 1.9 2.0 17.8 50.3 Gender of household head Male 83.8 45.6 82.1 43.7 1.7 1.9 16.2 54.4 Female 75.8 58.5 73.2 56.2 2.6 2.3 24.2 41.5 Area of residence Urban 81.0 54.4 78.7 51.9 2.3 2.5 19.0 45.6 Rural 84.6 39.2 83.5 38.2 1.1 1.0 15.4 60.8 Poverty status Poor 82.7 36.8 81.2 35.1 1.5 1.7 17.3 63.2 Nonpoor 82.0 54.4 80.0 52.2 2.0 2.2 18.0 45.6 Residence and poverty status Urban, poor 77.0 37.7 74.4 34.0 2.6 3.7 23.0 62.3 Urban, nonpoor 81.6 57.3 79.4 54.9 2.2 2.3 18.4 42.7 Rural, poor 85.5 36.3 84.6 35.8 0.9 0.5 14.5 63.7 Rural, nonpoor 83.5 42.6 82.3 41.1 1.2 1.5 16.5 57.4 Quintile of welfare aggregate Lowest quintile 83.7 36.3 82.3 35.0 1.3 1.2 16.3 63.7 2 81.0 39.7 79.1 38.0 1.8 1.7 19.0 60.3 3 80.6 48.0 78.2 45.7 2.4 2.3 19.4 52.0 4 82.7 53.0 80.5 50.4 2.2 2.6 17.3 47.0 Highest quintile 83.1 64.7 81.5 62.6 1.5 2.1 16.9 35.3 Region Western 82.1 44.3 80.6 43.3 1.4 1.0 17.9 55.7 Eastern 85.6 58.4 85.4 58.1 0.2 0.2 14.4 41.6 Metropolitan 81.4 53.0 79.2 50.3 2.2 2.7 18.6 47.0 Central 84.4 43.2 83.0 42.1 1.4 1.1 15.6 56.8 Source: Based on ADePT Gender using Panama 2008. Note: The working-age population is ages 15–64. ADePT Labor and ADePT ILO modules provide useful resources to further understand the functioning of the labor market and some of the gender gaps. The main feature of ADePT Labor is that it combines labor markets and poverty angles by looking at the labor income of the household. In addition, ADePT ILO produces International Labour Organization labor market indi- cators that can be computed using household surveys. Job Characteristics ADePT Gender figure 5 (figure 4.8) shows the percentage of women in dif- ferent types of jobs defined by the type of work (wage work, self-employment, 123 Producing a Country Gender Diagnostic: Part II Figure 4.8: ADePT Gender Figure 5a, Panama, 2008 Percentage of Women in Each Employment Type Part-time Full-time Informal Formal Agriculture Industry Services Wage worker Self-employed Private/family employment Public employment 0 10 20 30 40 50 60 Source: Based on ADePT Gender using Panama 2008. and so on), the sector, the type of firm ownership (public or private), the formality status, the full-time status, the occupation, and the industry. Generally, women are more likely to be part-time workers and to have ser- vice or public sector jobs. Women are often clustered in certain industries— such as health, education, and hotels and restaurants—and in certain occupations related to administrative tasks. For example, in Panama in 2008, 51 percent of public sector workers were female, and only 35 percent of private sector jobs were held by women. ADePT tables 5a (table 4.10) and 5b show the distribution of employ- ment by job characteristics for men and women separately. All the numbers in the two tables are read in blocks of rows. The first block shows the distri- bution of the population that is employed, unemployed, and out of the labor force. The following blocks show the distribution of the employed popula- tion by specific job characteristics as indicated by the block subtitle 124 Chapter 4: How to Interpret the Results of the Country Gender Diagnostic Table 4.10: ADePT Gender Table 5a, Panama 2008 Table 5a: Male and Female Employment by Selected Employment Characteristics: Total, Rural, and Poor Households Total Rural Poor Male Female Male Female Male Female Total 100.0 100.0 Employed 80.3 47.7 83.5 38.2 81.2 35.1 Unemployed 1.9 2.0 1.1 1.0 1.5 1.7 Out of labor force 17.8 50.3 15.4 60.8 17.3 63.2 Public sector employment Private/family employment 88.8 80.0 94.0 88.6 96.4 94.4 Public employment 11.2 20.0 6.0 11.4 3.6 5.6 Full-time status Part-time 17.4 28.0 25.4 47.7 27.1 52.2 Full-time 82.6 72.0 74.6 52.3 72.9 47.8 Formal status Informal 52.2 50.0 72.7 78.2 73.1 84.2 Formal 47.8 50.0 27.3 21.8 26.9 15.8 Broad sector Agriculture 22.8 7.2 55.5 25.8 54.3 27.5 Industry 24.5 8.6 18.2 14.1 19.3 14.2 Services 52.7 84.2 26.3 60.2 26.4 58.4 Work category Wage work 79.3 81.5 75.7 63.4 77.7 64.2 Self-employed 20.7 18.5 24.3 36.6 22.3 35.8 Source: Based on ADePT Gender using Panama 2008. Note: The working-age population is age 15–64. Private employment includes family employment. (for example, part time or full time). The last two columns present results for two vulnerable populations—rural areas and the poor. For example, in Panama, about half of employed men and women are informal workers. However, this percentage rises considerably if the analysis is restricted to men and women living in poor households, where 73 percent of employed poor men and 84 percent of employed poor women are informal workers. Table 4.11 provides some reference values of the distribution for women and men by type of work. This information is complemented with four additional graphs that show women’s participation along the welfare distribution in agricultural and non- agricultural employment (ADePT figures 5b and 5c) and in wage work versus self-employment (ADePT figures 4c and 4d). Nonagricultural jobs are typi- cally more productive, and hence better remunerated, than agricultural jobs. Moreover, individuals from the upper end of the welfare distribution are more likely to have benefited from secondary or tertiary education that allows 125 Producing a Country Gender Diagnostic: Part II Table 4.11: Distribution of Women and Men by Type of Work Women Men Wage and Own- Contributing Wage and Own- Contributing salaried account family salaried account family workers Employers workers workers workers Employers workers workers (%) (%) (%) (%) (%) (%) (%) (%) Africa Northern Africa (3) 46 2 19 34 58 8 22 11 Southern Africa (3) 76 3 17 4 82 7 9 2 Eastern and Western 20 1 47 32 24 1 56 18 Africa (6) Asia Eastern Asia (3) 86 2 7 5 80 7 13 <1 South-Eastern 52 2 23 23 52 4 34 9 Asia (6) Southern Asia (5) 30 1 22 46 44 3 40 12 Western Asia (6) 80 1 6 12 79 5 13 2 CIS in Asia (4) 45 1 39 15 50 3 39 7 Latin America and the Caribbean Caribbean (5) 80 2 16 2 67 3 27 1 Central America (6) 64 3 25 7 64 6 24 6 South America (9) 62 3 28 6 62 6 28 3 More developed regions Eastern Europe (8) 84 2 10 4 78 4 16 1 Northern Europe (5) 93 2 4 1 84 5 10 <1 Southern Europe (9) 81 3 10 6 74 6 17 2 Western Europe (4) 89 3 6 3 84 7 8 1 Other more 88 2 7 2 83 5 11 1 developed regions (4) Source: UN 2010. Note: Parenthetical numerals indicate the number of countries. CIS = Commonwealth of Independent States. them to access employment opportunities outside agriculture. Thus, we would expect that the share of agricultural employment declines as we move along the welfare distribution. Figure 4.9 illustrates this pattern for Panama. Regarding the measures of work intensity, ADePT tables 5c and 5d show the average number of hours worked (in the specified reference period; see chapter 3) for different groups of the population indicated in the column heads and for different types of employment indicated in the rows. In gen- eral, it is observed that men work on average more hours than women. For example, in Panama, the average man works 189 hours per month (45 hours per week), compared to 161 hours for the average woman (38 hours per week).14 It is often argued that women choose informal jobs because they are more flexible than formal ones. In Panama, women in informal jobs work 126 Chapter 4: How to Interpret the Results of the Country Gender Diagnostic Figure 4.9: ADePT Gender Figures 5b and 5c, Panama 2008 a. Share of Working Women in and out of Agriculture Females 100 90 Share of women, % 80 70 60 50 40 30 20 10 0 1 2 3 4 5 6 7 8 9 10 Deciles b. Percentage of the Working-Age Population in Agricultural and Nonagricultural Jobs, by Decile Male 100 90 80 Employment, % 70 60 50 40 30 20 10 0 1 2 3 4 5 6 7 8 9 10 Deciles Female 100 90 80 Employment, % 70 60 50 40 30 20 10 0 1 2 3 4 5 6 7 8 9 10 Deciles Nonagriculture Agriculture Source: Based on ADePT Gender using Panama 2008. Note: The working-age population is ages 15–64. 127 Producing a Country Gender Diagnostic: Part II fewer hours per week than the average woman. Also, women in part-time jobs typically work much less than the threshold of 35 hours per week. In Panama, they work fewer than 20 hours per week. It is also found that many women prefer public sector jobs because of their stability, relatively higher pay, and lower time demands. However, in Panama, men and women employed in the public sector report working slightly more hours than men and women employed in the private sector (see table 4.12, which corres- ponds to ADePT table 6a). Earnings ADePT tables 5e–5h show monthly earnings by gender and type of job. The user needs to ensure that the variables populated in the Hours and Earnings fields have the same periodicity (see chapter 3 for more details). ADePT tables 5e and 5f show mean earnings, while ADePT tables 5g and 5h show median earnings. Tables 5e and 5g present the statistics for different types of jobs, while tables 5f and 5h show them for industry and occupation. All statistics are interpreted as in ADePT tables 5a and 5b. They show the Table 4.12: ADePT Gender Table 6a, Panama 2008 Table 6a: Male and Female Mean Monthly Hours Worked by Selected Employment Characteristics: Total, Rural, and Poor Households Total Rural Poor Male Female Male Female Male Female Total 188.9 160.9 179.2 139.0 176.2 132.1 Public sector employment Private/family employment 188.4 160.0 177.9 136.4 175.8 131.1 Public employment 192.7 164.0 197.1 157.2 186.9 147.8 Full-time status Part-time 83.2 71.6 84.5 66.3 83.1 67.0 Full-time 211.0 195.4 211.3 204.8 210.9 202.9 Formal status Informal 176.9 141.8 168.0 126.9 165.3 121.1 Formal 200.4 178.7 204.2 177.7 201.5 185.5 Broad sector Agriculture 168.2 97.8 166.4 95.2 162.3 91.8 Industry 190.5 155.3 192.7 128.6 192.4 145.3 Services 195.7 166.2 193.0 158.5 189.4 146.5 Work category Wage work 195.1 170.8 198.1 161.3 196.4 155.7 Self-employed 179.8 134.9 178.8 142.8 179.9 131.5 Source: Based on ADePT Gender using Panama 2008. Note: The working-age population is ages 15–64. 128 Chapter 4: How to Interpret the Results of the Country Gender Diagnostic statistic for a group of the population that is defined by the combination of the column’s heads and subheads and the row titles. For example, the top left cell of ADePT table 7a (see table 4.13) contains the mean earnings for all employed men. More vulnerable jobs are usually associated with larger gender earn- ings gaps. If women lack access to better paid formal jobs, they will also have problems accessing well-paid informal jobs. Instead, some men who choose informal jobs most likely had the option of taking formal jobs but preferred to remain in the informal sector. In addition, the interpretation of ADePT tables 5e–5h needs to be linked to the previous findings. For example, a larger percentage of women in vulnerable jobs could face restricted access to good job opportunities,15 and thus higher gender earnings gaps should be expected. Missing data on wages are particularly common for women. Further understanding of these issues can be gained from regression analysis (see chapters 5 and 6) and by looking at access to productive resources and repeating the analysis for population sub- groups, as discussed next. Table 4.13: ADePT Gender Table 7a, Panama 2008 Table 7a: Male and Female Mean Monthly Earnings by Selected Employment Characteristics Total Rural Poor Male Female Male Female Male Female Total 529.3 454.8 311.4 210.1 233.3 153.1 Public sector employment Private/family employment 482.9 387.0 287.3 151.7 223.5 135.7 Public employment 859.7 694.1 622.2 527.1 448.8 359.7 Full-time status Part-time 259.0 179.5 147.1 86.1 115.5 73.5 Full-time 579.4 545.8 355.7 291.0 267.8 215.5 Formal status Informal 349.0 211.7 229.1 106.3 162.9 97.2 Formal 690.1 652.8 472.0 453.3 374.6 351.4 Broad sector Agriculture 218.8 109.9 192.3 93.0 142.7 54.8 Industry 479.1 350.0 383.6 63.1 300.7 84.6 Services 652.0 476.0 452.2 260.7 322.4 183.8 Work category Wage work 630.0 519.3 449.9 302.1 348.0 217.6 Self-employed 414.1 212.7 318.5 83.3 205.8 68.1 Source: Based on ADePT Gender using Panama 2008. Note: The working-age population is ages 15–64. 129 Producing a Country Gender Diagnostic: Part II Disaggregating Economic Outcomes–By Life-Cycle Stage and Education Economic outcomes, particularly for women, are strongly influenced by education and family responsibilities. ADePT tables 9b–9d show economic outcomes by broad age group (suffix A, for age)—which is meant to reflect (a woman’s) life-cycle stage—the number children in the household (suffix C for children), and the level of education (suffix E for education). ADePT table 9b shows labor force participation and employment rates (extensive margin), ADePT table 9c presents numbers of hours (intensive margin), and ADePT table 9d earnings. The interpretation of the tables is otherwise the same as in the corresponding ADePT tables 5a––5c. Figure 4.10 shows labor force participation rates for men and women of different ages in selected countries. For example, in Japan, women sub- stantially decrease their engagement in the labor market when they have young children, but they return later on. In Chile, female labor force par- ticipation is generally much lower, but a much smaller dip occurs for women in their 30s, suggesting that some women work and others stay out of the labor market all their life. In Tunisia, there is evidence of cohort Figure 4.10: Female Labor Force Participation over the Life Cycle (a) Japan 100 Labor force participation rate (%) 90 80 70 60 50 40 30 20 10 0 + 9 29 39 19 49 59 4 24 44 54 4 4 75 –6 –6 –3 –7 – – – – – – – – 65 25 35 15 45 55 60 20 40 50 30 70 Age Women 1990 Men 1990 Women 2008 Men 2008 (continued) 130 Chapter 4: How to Interpret the Results of the Country Gender Diagnostic Figure 4.10: Female Labor Force Participation over the Life Cycle (continued) (b) Chile 100 Labor force participation rate (%) 90 80 70 60 50 40 30 20 10 0 9 9 9 9 9 9 4 4 4 + 4 4 –1 –2 –6 –3 –4 –5 –4 –2 –3 –5 –6 70 15 25 65 35 45 55 40 20 30 50 60 Age Women 1990 Men 1990 Women 2007 Men 2007 (c) Tunisia 100 Labor force participation rate (%) 90 80 70 60 50 40 30 20 10 0 9 9 + 9 9 9 4 4 4 4 4 –1 –2 –3 –5 –4 65 –3 –4 –6 –5 –2 15 25 35 55 45 30 40 60 50 20 Age Women 1989 Men 1989 Women 2005 Men 2005 Source: UN 2010. Note: Year is the latest available in the period of reference. 131 Producing a Country Gender Diagnostic: Part II effects—comparatively few women work, but younger cohorts are more likely to participate in the labor market than older cohorts. Access to Resources Finally, ADePT table 8a explores whether men and women have differential rates of access to household assets and whether they use them differently (World Bank 2012). Table 4.14 illustrates the logic of table 8a, using the example of credit use, an important financial asset. The first column of ADePT table 8a shows the percentage of households owning these assets for the population subgroups defined by the household characteristics indicated in the rows. For example, in Panama, 10 percent of rural households and 19 percent of urban households reported having a loan. In the columns farther to the left of ADePT table 8a, we see the within-household gender differences in access, defined here by who has the primary responsibility for the loan. In particular, the second to fourth col- umns show the percentage of households in which (a) only men, (b) only women, or (c) both men and women have the primary responsibility for the loan (out of all households with loans). For instance, in 69 percent of the households with loans in rural Panama, the primary responsibility for the loan rests with male household members, whereas in only 23 percent it rests only with females, and in 7 percent the responsibility was joint. Country Context: Voice, Agency, and Participation Content This section describes nine tables (ADePT tables 10–11h) that profile voice and agency outcomes. The tables are grouped according to the five dimensions of voice and agency presented in the World Development Report (World Bank 2012). ADePT tables 10, 11a, and 11b focus on the dimension decision making over family formation, ADePT table 11 on free- dom of movement, and so on. ADePT tables 10, 11a, and 11b require spe- cific variables, whereas ADePT tables 11d–11h leave flexibility to explore other dimensions for which data are available in the survey. As before, all tables show outcomes for all of the population (tables 11a, 11c, 11e, and 11g) and for selected subgroups (tables 11b, 11d, 11f, and 11h). Examples in this section are based on the Living Standards Management Survey for 132 Chapter 4: How to Interpret the Results of the Country Gender Diagnostic Table 4.14: ADePT Gender Table 8a, Panama 2008 Table 8a: Household Access to Economic Resources and Intrahousehold Access by Selected Household Characteristics Primary credit borrowers HH uses credit Primary credit Only borrowers Only men women Both No Yes Total 100.0 55.0 34.3 10.7 83.7 16.3 Household head’s sex Male 100.0 66.8 23.2 10.0 83.8 16.2 Female 100.0 20.6 66.6 12.8 83.5 16.5 Area of residence Urban 100.0 51.2 37.2 11.7 80.6 19.4 Rural 100.0 69.4 23.4 7.2 90.0 10.0 Poverty status Poor 100.0 70.2 25.5 4.2 93.8 6.2 Nonpoor 100.0 52.5 35.7 11.8 80.7 19.3 Residence and poverty Urban, poor 100.0 71.0 29.0 0.0 91.1 8.9 Urban, nonpoor 100.0 49.2 38.0 12.8 79.2 20.8 Rural, poor 100.0 69.4 21.7 8.9 95.1 4.9 Rural, nonpoor 100.0 69.4 24.1 6.5 85.7 14.3 Quintiles of welfare aggregate Lowest quintile 100.0 75.8 21.4 2.8 96.5 3.5 2 100.0 68.7 23.5 7.8 88.9 11.1 3 100.0 59.4 30.9 9.7 83.7 16.3 4 100.0 54.4 34.6 10.9 79.9 20.1 Highest quintile 100.0 41.7 44.1 14.2 78.1 21.9 Region Región Occidental 100.0 54.1 33.4 12.4 84.7 15.3 Región Oriental 100.0 84.0 13.1 2.9 91.3 8.7 Región Metropolitana 100.0 53.2 36.4 10.4 82.3 17.7 Región Central 100.0 59.8 29.3 10.8 86.2 13.8 Source: Based on ADePT Gender using Panama 2008. Note: HH = household. The working-age population is ages 15–64. Panama (2008) and the Living Standards Survey for Nepal (2012b), but Demographic and Health Surveys (DHS) can also be used, as they typically include a wealth of information in this area (see chapter 3 and appendix C). Concepts Women’s agency matters in its own right, but evidence is emerging that it also matters for economic development (Narayan 2005; World Bank 2012). This section focuses on selected proxies of agency for which data 133 Producing a Country Gender Diagnostic: Part II are commonly available. Given that this area of research is new—at least among economists—most of the findings show associations, but they have not been proved to be causal. Agency outcomes are boosted by economic growth and affected by markets and formal and informal institutions (World Bank 2012). For example, social norms shape women’s agency. Available indicators of decision making over family formation tend to be less controversial than for other dimensions but fail to provide a com- plete description of this dimension. ADePT Gender measures the age at first marriage and the age at first birth and gives the option of adding variables chosen by the user, as surveys vary considerably in the available information on this topic. For example, variables that could be informa- tive in some cases and are usually included in DHS are the percentage distribution by months since the preceding birth, the median number of months since the preceding birth, and knowledge/use of contraception methods. Box 4.2 lists the relevant indicators that can be constructed using DHS data. There is less agreement on how to measure the other dimension of agency, and access to data on the relevant indicators is more limited, at least at the household or individual level. Chapter 6 discusses the progress made to collect better data and analyze more directly and accurately all measures of agency. Given the scarcity of indicators and the variability in availability across surveys, ADePT Gender does not request any specific variable. The rest of the report further discusses the other dimensions of agency using the DHS as a reference. The main indicators used to capture outcomes related to control over resources are related to decision making over income, property rights, or management of individual or household assets. The DHS has a set of ques- tions on decision making regarding income and expenditures, and a few questions on asset ownership (see appendix B). In general, women with more education and from wealthy households are more likely to have more decision-making power and more access to and ownership of resources. Travel records are usually used as a proxy of the ability to move freely (see appendix B). Better infrastructure, good roads, and safe public transpor- tation promote this dimension of agency (World Bank 2012). In addition, social norms are of great importance for women to have control of their mobility. Freedom from violence is particularly relevant for women in poor households and women without bargaining power. 134 Chapter 4: How to Interpret the Results of the Country Gender Diagnostic Box 4.2: Demographic and Health Survey Indicators on Contraception The diagnostic on contraception covers all women between ages 15 and 49 for the lifetime of the woman interviewed (or since she was age 15). Knowledge of contraception: Percentage of women who know of any, a modern, or a traditional contraceptive method. The numerator for any of these indicators is the number of women who say they know of a (specific) contraceptive method (or meth- ods). Any method refers to the number of women who say they know of at least one method. The denominator is all women ages 15–49. Use of contraception: Percentage of women who have ever used any, a modern, or a traditional contraceptive method. The numerator for any of these indicators is the number of women who say they used the (specific) contraceptive method (or methods). Any method refers to the number of women who say they used at least one of the methods. The denominator is all women ages 15–49. Current use of contraception: Percentage of women who currently use any, a modern, or a traditional contraceptive method. The indicators are calculated as those above, and the period of reference is defined by the respondent. One indicator is more widespread—the contraceptive prevalence rate (CPR), defined as Number of currently married women who use any method of contraception CPR = × 100. Number of currently married women Need for family planning: Percentage of currently married women with (a) an unmet need for family planning, (b) a met need for family planning, or (c) total demand for family planning. All of the above indicators can be computed in ADePT Gender using the open fields, since they are simply averages of dummy variables. Source: Rutstein and Rojas 2006. The last dimension of agency—ability to have voice in society—is easily measured at macro levels and often comes from administrative data. However, many household surveys and surveys conducted to evaluate programs ask respondents about their involvement in community net- works and public life. Understanding women’s participation in public life is important, since women’s voice has been found to be positively associ- ated with women’s ability to transform societies (World Bank 2012). 135 Producing a Country Gender Diagnostic: Part II Women can influence the social, economic, and political environment by participating in informal associations and through collective actions. However, their collective success starts with their ability to have indi- vidual agency. More important, women’s agency matters, since it shapes their children’s future. Children who experience or witness violence are more likely to per- petrate crimes and are less likely to be productive workers when adults (Currie and Tekin 2012; World Bank 2012). In addition, social norms are formed at home, and children’s perceptions—those of both boys and girls— of what is acceptable (and what is not) are often shaped at home and early in life (Farré and Vella 2007; Fernández, Fogli, and Olivetti 2004). Interpreting the Results ADePT table 10 shows the median age at first marriage and age at first birth for select household characteristics. Across countries, we find that different dimensions of agency are highly correlated to each other (Klugman and others 2014). Figure 4.11 presents results for three of the four dimen- sions that can be analyzed with simple cross tabulations of microdata. Figure 4.11: Share of Women Who Experience Overlapping Agency Deprivations in Three Domains (Percentage) Lack of control over household resources 42 Absence of deprivations 10 21 8 11 13 Condones wife beating Child marriage 12 43 10 15 51 Source: Klugman and others 2014, based on Demographic and Health Surveys for 54 countries using the latest data available, 2001–12. 136 Chapter 4: How to Interpret the Results of the Country Gender Diagnostic Within a country, women in general—but particularly women with less education or in poor households—often marry and have children at a young age. For example, in the case of Nepal in 2010–11 (table 4.15, cor- responding to ADePT table 10), women in the bottom quintile were more likely to be married younger than women in the top quintile (Nepal 2012b). To give an idea of variation across countries, figure 4.12 reproduces a graph from the United Nations (2010) showing the mean age of marriage in countries where women marry at a very young age. The same report offers additional statistics, such as the proportion of girls ages 15–20 in marriages or consensual unions, which can be used to compare the coun- try’s situation with the rest of the world. Table 4.15: ADePT Gender Table 10, Nepal 2010–11 Table 10: Mean age at first marriage and first birth among females by selected household characteristics Age at first marriage Age at first birth Total 17.3 19.3 Household head’s sex Male 17.4 19.4 Female 17.2 19.2 Area of residence Urban 18.5 19.7 Rural 17.1 19.2 Poverty status Poor 16.6 19.3 Nonpoor 17.7 19.4 Residence and poverty Urban, poor 16.4 18.0 Urban, nonpoor 18.7 19.8 Rural, poor 16.6 19.3 Rural, nonpoor 17.3 19.2 Quintiles of welfare aggregate Lowest quintile 16.5 19.2 2 16.8 19.3 3 17.2 19.2 4 17.5 19.1 Highest quintile 18.5 19.8 Regions Eastern 17.9 19.9 Central 17.3 19.3 Western 17.5 19.3 Midwestern 16.9 18.8 Far western 16.8 18.9 Source: Based on ADePT Gender using the Nepal 2012b. Note: Working women ages 15–64. 137 Producing a Country Gender Diagnostic: Part II Figure 4.12: Mean Age at Marriage for Women and Men and the Difference in Years, Countries Where Women Marry on Average at Age 20 or Earlier, 2002–06 40 Singulate mean 35 age marriage 30 25 20 15 so na a a r er al i ad i ca ne ni al aw ig Fa ep a M a Ch as uy ui al N nz N a G ag M G in Ta ad rk Bu M Men Women Source: UN 2010. Note: Year is the latest available in the period of reference. Table 4.16: ADePT Gender Table 11g, Nepal 2010–11 Table 11g: Mean Outcomes Related to Control over Resources by Selected Individual Characteristics Number of Health of Major Use of School children children Food purchase Assets Loans loans Migration Remittances Age 15–24 0.7 0.8 0.8 0.7 0.7 0.6 0.7 0.7 0.6 0.8 25–34 0.8 0.8 0.8 0.8 0.8 0.7 0.7 0.7 0.7 0.9 35–44 0.8 0.8 0.8 0.8 0.8 0.7 0.7 0.7 0.7 0.9 45–54 0.7 0.7 0.8 0.8 0.7 0.7 0.7 0.7 0.7 0.8 55–64 0.7 0.5 0.7 0.8 0.7 0.7 0.7 0.7 0.7 0.7 65+ 0.6 0.7 0.7 0.7 0.7 0.7 0.7 0.7 0.6 0.7 Marital status Never married 0.6 — 0.9 1.0 0.9 0.9 0.7 0.5 1.0 1.0 Married 0.7 0.8 0.8 0.8 0.7 0.7 0.7 0.7 0.7 0.8 Widowed 0.8 0.7 0.9 0.9 0.9 0.9 0.9 0.9 0.8 0.8 Divorced — — 1.0 1.0 1.0 1.0 1.0 1.0 — 1.0 Education No education 0.7 0.7 0.8 0.8 0.7 0.7 0.7 0.7 0.7 0.8 Primary 0.8 0.8 0.9 0.8 0.8 0.7 0.7 0.7 0.7 0.8 Secondary 0.8 0.8 0.9 0.8 0.8 0.6 0.7 0.8 0.9 0.9 Postsecondary 0.8 0.9 0.8 0.9 0.9 0.8 0.7 0.7 0.7 0.8 Employment status Employed 0.7 0.8 0.8 0.8 0.8 0.7 0.7 0.7 0.7 0.8 Unemployed 0.8 0.9 0.9 1.0 0.9 1.0 0.8 0.7 1.0 1.0 Out of labor force 0.7 0.8 0.8 0.8 0.8 0.7 0.7 0.7 0.7 0.8 Source: Based on ADePT Gender using Nepal 2012a. Note: — = No observations to produce the estimate. 138 Chapter 4: How to Interpret the Results of the Country Gender Diagnostic ADePT tables 11a–11h present average values of the variables that are proxies for each of the four dimensions of agency: (a) control over family formation, (b) freedom of movement, (c) freedom from domestic violence, and (d) control over resources. As with all the tables in ADePT Gender diagnostics, there is one table in which the information is also disaggre- gated for selected individual characteristics (ADePT tables 11a and 11g) and household characteristics (ADePT tables 10, 11b, and 11h). Table 4.16 reproduces the results from ADePT Gender for Nepal 2010–11. It shows the average percentage of women who participate in decision making ranging from small decisions—such as household purchases of food—to important decisions—such as getting a loan and how to spend it. One feature worth highlighting is that in Nepal, women’s involvement in deci- sion making is fairly constant by wealth of household. This result is some- what atypical, as in most countries, women in poor households participate less in any decision. Figure 4.13 reproduces results from the 2012 WDR. Figure 4.13: Women’s Control over Household Decisions Large Visits to How own earnings purchases family/relatives are used 100 Share of women with some control 90 over decisions, % 80 70 60 50 40 Q1 Q2 Q3 Q4 Q5 Q1 Q2 Q3 Q4 Q5 Q1 Q2 Q3 Q4 Q5 Quintiles East Asia and Pacific Europe and Central Asia Latin America and the Caribbean Middle East and North Africa South Asia Sub-Saharan Africa Source: World Bank 2012. 139 Producing a Country Gender Diagnostic: Part II Notes 1. For more details on the core gender indicators, the reader can consult IHSN (2015). 2. Estimates might not be precise enough when the data are too noisy (that is, significant dispersion in the values of the observations) or because not enough observations exist for the type of statistic the ana- lyst is trying to estimate. 3. If another survey is used to illustrate particular results, it will be clarified in the appropriate table or graph. 4. Stopping behavior is the term used to describe a situation where parents who gave birth to a boy—most of the time the first birth—would stop having additional children. This behavior results in low fertility rates and missing women at birth. This was the case in the Republic of Korea in the 1980s and is currently prevalent in Armenia. 5. For a visualization of the typology, see Buvinic, Das Gupta, and Casabonne (2009, 350). 6. The interpretation of the results relies on not having biases coming from missing data. If data are missing, and as briefly discussed in chapter 3, the user needs to evaluate whether the missing observations could alter the interpretation of the results. 7. See appendix C for a discussion of the steps needed to evaluate whether a difference in indicators is statistically significant. 8. This, however, depends on the country. In many parts of Sub-Saharan Africa, female-headed households have lower poverty rates than male- headed households. Such findings may be due to the use of non- comparable or inconsistent measures of living standards or benchmarks for judging deprivation, most likely for how adjustments are made for household size and composition, as well as economies of scale in con- sumption (Milazzo and van de Walle 2015). For example, in some countries, only women with adequate economic positions can afford not to remarry; thus, they head households after the marital dissolution. 9. The complete ADePT output for Panama 2008 and Nepal 2011 can be found on the ADePT website, under the tab for “Gender,” at http:// go.worldbank.org/0GA4FDMQY0. 10. This concept is also known as NEET (not in education, employment, or training). 140 Chapter 4: How to Interpret the Results of the Country Gender Diagnostic 11. One of the reasons why people find it difficult to report earnings is that they do not know their exact job benefits or how to value them. 12. This does not imply that individual countries follow a U-shaped trajec- tory over time (Gaddis and Klasen 2014). 13. This is called an income effect. 14. The common practice for moving from hours worked per month to hours worked per week is to divide by 4.2, since many months exceed four weeks. 15. Good jobs could be defined as being formal, having higher wages, hav- ing higher career progress opportunities, stability, and so forth. References Almond, Douglas, Janet Currie, and Mariesa Hermann. 2012. “From Infant to Mother: Early Disease Environment and Future Maternal Health.” Labour Economics 19 (4): 475–83. Angel-Urdinola, Diego F., and Victor Macias. 2008. “FYR Macedonia Labour Market Profile 2004–2007: A Policy Note.” World Bank, Washington, DC. Buvinic, Mayra, and Geeta Rao Gupta. 1997. “Female-Headed Households and Female-Maintained Families: Are They Worth Targeting to Reduce Poverty in Developing Countries?” Economic Development and Cultural Change 45 (2): 259–80. Buvinic, Mayra, Monica Das Gupta, and Ursula Casabonne. 2009. “Gender, Poverty and Demography: An Overview.” World Bank Economic Review 23 (3): 347–69. Case, Anne, and Angus Deaton. 2003. “Consumption, Health, Gender, and Poverty.” Policy Research Working Paper 3020, World Bank, Washington, DC. Case, Anne, Angela Ferting, and Christina Paxon. 2005. “The Lasting Impact of Childhood Health and Circumstance.” Journal of Health Economics 24 (2): 365–89. Currie, Janet, and Enrico Moretti. 2003. “Mother’s Education and the Intergenerational Transmission of Human Capital: Evidence from College Openings.” Quarterly Journal of Economics 118 (4): 1–22. 141 Producing a Country Gender Diagnostic: Part II Currie, Janet, Mark Stabile, Phongsack Manivong, and Leslie L. Roos. 2010. “Child Health and Young Adult Outcomes.” Journal of Human Resources 45 (3): 517–48. Currie, Janet, and Erdal Tekin. 2012. “Understanding the Cycle: Childhood Maltreatment and Future Crime.” Journal of Human Resources 47 (2): 509–49. Edmonds, Erik, Nina Pavcnik, and Petia Topalova. 2009. “Child Labor and Schooling in a Globalizing World: Evidence from Urban India.” Journal of the European Economics Association Papers and Proceedings 7 (2–3): 498–507. Farré, Lídia, and Francis Vella. 2007. “The Intergeneration Transmission of Gender Role Attitudes and Its Implications for Female Labor Force Participation.” IZA Discussion Paper 2802, Institute for the Study of Labor, Bonn. Fernández, Raquel, Alessandra Fogli, and Claudia Olivetti. 2004. “Mothers and Sons: Preference Formation and Female Labor Force Dynamics.” Quarterly Journal of Economics 119 (4): 1249–99. Flabbi, Luca. 2011. “Gender Differentials in Education, Career Choices and Labor Market Outcomes on a Sample of OECD Countries.” Background paper for World Development Report 2012. Gaddis, Isis, and Stephan Klasen. 2014. “Economic Development, Structural Change, and Women’s Labor Force Participation.” Journal of Population Economics 27 (3): 639–81. Gamberoni, Elisa, Ana Maria Munoz Boudet, and Josefina Posadas. 2014. Russian Federation Gender Assessment. Report ACS7863. Washington, DC: World Bank. Gamberoni, Elisa, and Josefina Posadas. 2012. “Gender Gaps in Labor Market Outcomes: Participation, Unemployment, and Wage Gaps in FYR Macedonia.” Mimeo, World Bank, Washington, DC. IHSN (International Household Survey Network). 2015. “How Well Are Gender Issues Covered in Household Surveys and Censuses? An Analysis Using the IHSN–World Bank Gender Data Navigator.” http://ihsn.org /HOME/sites/default/files/resources/Gender_Issues_July-2015.pdf. Jensen, Robert. 2010. “The (Perceived) Returns to Education and the Demand for Schooling.” Quarterly Journal of Economics 125 (2): 515–48. Klasen, Stephan, and Claudia Wink. 2003. “‘Missing Women’: Revisiting the Debate.” Feminist Economics 9 (2–3): 263–99. 142 Chapter 4: How to Interpret the Results of the Country Gender Diagnostic Klugman, Jeni, Lucia Hanmer, Sarah Twigg, Tazeen Hasan, Jennifer McCleary- Sills, and Julieth Santamaria. 2014. Voice and Agency: Empowering Women and Girls for Shared Prosperity. Washington, DC: World Bank. Mammen, Kristin, and Christina Paxson. 2000. “Women’s Work and Economic Development.” Journal of Economic Perspective 14 (4): 141–64. Milazzo, Annamaria, and Dominique van de Walle. 2015. “Women Left Behind? Poverty and Headship in Africa.” Policy Research Working Paper 7331, World Bank, Washington, DC. Molinas, José, Ricardo Paes de Barros, Jaime Saavedra, and Marcelo Giugale. 2010. “Do Children Have a Chance? The 2010 Human Opportunity Index Report for Latin America and the Caribbean.” World Bank, Washington DC. Narayan, Deepa. 2005. Measuring Empowerment: Cross Disciplinary Perspectives. Washington, DC: World Bank. Nepal, Government of. 2012a. “Nepal Demographic and Health Survey 2011.” Population Division, Ministry of Health and Population, Kathmandu. ———. 2012b. “Nepal: Living Standards Survey 2010–2011, Third Round.” Central Bureau of Statistics, National Planning Commission Secretariat, Kathmandu. http://microdata.worldbank.org/index.php /catalog/1000. Ñopo, Hugo R., Nancy Daza, and Johanna Ramos. 2011. “Gender Earnings Gaps in the World.” IZA Discussion Paper 5736, Institute for the Study of Labor, Bonn. Oreopoulos, Philip, Marianne E. Page, and Ann Huff Stevens. 2006. “The Intergenerational Effects of Compulsory Schooling.” Journal of Labor Economics 24 (4): 729–60. Oster, Emily, and M. Bryce Millett. 2010. “Do Call Centers Promote School Enrollment? Evidence from India.” NBER Working Paper 15922, National Bureau of Economic Research, Cambridge, MA. Panama, Government of. 2008. “Encuesta de Niveles de Vida 2008” (“Living Standards Survey 2008”). Ministerio de Economia y Finanzas, Panama City. http://microdata.worldbank.org/index.php/catalog/70. Porta, Emilio, Gustavo Arcia, Kevin Macdonald, Sergiy Radyakin, and Michael Lokshin. 2011. Assessing Sector Performance and Inequality in Education. Washington, DC: World Bank. Rutstein, Shea Oscar, and Guillermo Rojas. 2006. “Guide to DHS Statistics.” Demographic and Health Surveys, ORC Macro, Calverton, MD. 143 Producing a Country Gender Diagnostic: Part II Singh, Ashish. 2012. “Gender Based Within-Household Inequality in Immunization Status of Children: Some Evidence from South Asian Countries.” Working paper, Indira Gandhi Institute of Development Research, Mumbai. Strauss, John, and Duncan Thomas. 1995. “Human Resources: Empirical Modeling of Household and Family Decisions.” In Handbook of Development Economics, vol. 3, edited by Hollis Chenery and T. N. Srinivasan, 1883–2023. Amsterdam and New York: North-Holland. Thomas, Duncan. 1990. “Intra-Household Resource Allocation: An Inferential Approach.” Journal of Human Resources 25 (4): 635–64. UN (United Nations). 2010. The World’s Women 2010: Trends and Statistics. New York: UN. http://unstats.un.org/unsd/demographic/products /Worldswomen/WW_full%20report_color.pdf. UNESCO (United Nations Educational, Scientific, and Cultural Organization). 2012. World Atlas of Gender Equality in Education. Paris: UNESCO. UNFPA (United Nations Population Fund). 2011. “Population Dynamics in the Least Developed Countries: Challenges and Opportunities for Development and Poverty Reduction.” New York. https://www.unfpa .org/sites/default/files/pub-pdf/CP51265.pdf. Wagstaff, Adam, Marcel Bilger, Zurab Sajaia, and Michael Lokshin. 2011. Health Equity and Financial Protection: Streamlined Analysis with ADePT Software. Washington, DC: World Bank. https://openknowledge .worldbank.org/handle/10986/2306. World Bank. 2012. World Development Report 2012: Gender Equality and Development. Washington, DC: World Bank. https://openknowledge .worldbank.org/handle/10986/4391. ———. 2014. “South Caucasus Country Gender Assessment 2015.” World Bank, Washington, DC. 144 Chapter 5 Technical Notes for the Country Gender Diagnostic This chapter provides guidance and a brief overview of topics that serve as background when interpreting the tables and figures presented in chapter 4. The framework of the World Development Report 2012: Gender Equality and Development (World Bank 2012)—discussed in chapter 1—places house- holds at the center, since they make choices on the basis of preferences, incentives (which come from the markets), and constraints (which arise from both formal and informal institutions). This chapter provides an over- view of two household models, which will help the user better frame the results from the tables and figures of the country gender profile. This manual relies on the concepts and models of classical economics, par- ticularly, the household labor supply model and the intrahousehold decision model. However, other schools of thought also support the approach taken by ADePT Gender. The most salient is the feminist school of thought that has run parallel to the classical economics for a long time. This approach has been less quantitative but equally rigorous and highly influential at the intersec- tion of gender and development. Moreover, in recent decades, advances in economics and other social sciences have narrowed the distance between these two schools of thought by developing (quantitative) theoretical models and trying to quantify the concepts of these other social sciences. It is beyond the objective of this manual to summarize the wealth of knowledge produced over so many years by the social sciences, by the feminist and economics schools of thought and their interlinks. 145 Producing a Country Gender Diagnostic: Part II Users interested in knowing more can consult England (2003), which discusses how the two schools of thought have become closer with the development of bargaining and endogenous taste models in economics and by care models in feminist economics. Moreover, if England were to write her paper today, more recent models offered by behavioral econom- ics would help further narrow the distance between these schools of thought. Users interested in learning more about neoclassical economics can consult Blau, Ferber, and Winkler (2006). The first section of this chapter describes in a very simple manner the household labor supply model. This neoclassical economics model, which has been at the core of family economics for a long time, relies on strong assumptions but produces results that replicate many outcomes observed in reality. The second section describes the intrahousehold allocation models, also called bargaining models. These models are richer in the sense that they relax some of the strong assumptions of the household labor supply model and assume that individuals do not live in isolation but in a family that is subject to constraints and has its own rules. Thus, it sometimes makes sense to focus on the behavior of the household. At the same time, the reader should bear in mind that, in economics, a theory is not to be judged by its resemblance to reality but rather by the extent to which it enables us to grasp the salient features of reality. Household Labor Supply Model As stated previously, households are at the center of the framework used to design ADePT Gender tables and graphs. Families make decisions on how many children to have, when to have them, and how much to invest in their sons’ and daughters’ health and education. These decisions are shaped by household members’ preferences, incentives coming from the market, and constraints from formal and informal institutions. The household labor sup- ply model was developed several decades ago and is still relevant for under- standing the mechanisms through which market incentives and institutional constraints are transmitted to households.1 Since the seminal work of Gary Becker (1965, 1974) and Jacob Mincer (1962) on family decisions on labor allocation between the household and the market, economists and social scientists have developed and used many models. Despite its limitations, the unitary model is still useful for predicting 146 Chapter 5: Technical Notes for the Country Gender Diagnostic certain patterns of a family’s division of labor. For example, household labor supply models have contributed to explaining observed facts, such as wom- en’s increasing labor force participation, declining fertility rates, and increasing investments in children’s education, especially for girls. The household labor supply model is based on three main assumptions. First, household well-being depends on goods that are produced at home using market goods and household members’ time. For example, food needs to be bought at the market and requires some preparation, furniture needs to be bought and later arranged at home and maintained, and children need care. As a result, time can be allocated among three competing tasks: (a) market work outside the home, (b) production of household goods (using market goods as inputs), and (c) leisure. Second, most households have more than one adult; thus, decisions about the time allocation for these three competing activities are made jointly.2 In some cases, we can further assume that the productivity of household members varies with the task. For example, women might be more productive in producing house- hold goods, and men might be more productive in market work. Third, the models can be extended to incorporate a time dimension. For example, the time spent at home can be substituted by time in the market (substitution of time between activities in and out of the household), time spent working for pay today can be substituted by leisure tomorrow (substitution of time between activities over the life cycle). Labor Supply Model with Household Production This section shows a simple model in which people can use their time for two tasks: (a) working for pay outside the home and (b) working in house- hold production of goods. Time spent in household production includes time spent on chores or leisure.3 In reality, people spend time on several activities. Table 5.1 shows the average number of hours that men and women spent on different activities, based on survey data from the Lao People’s Democratic Republic (Lao PDR 2008). Although no differences occur in the number of hours spent sleeping (or these differences are not significant; see appendix C), significant gender differences exist in time spent in household production and leisure. Women spend almost one hour more per day than men in childcare and eldercare, and about half an hour less per day on leisure activities. They also spend slightly less time than men in market activities, but this is a highly country specific result.4 147 Producing a Country Gender Diagnostic: Part II Table 5.1: Daily Hours Spent in Household Work, Paid Work, and Leisure for Men and Women Hours spent Hours spent on Hours spent on Hours spent on sleeping childcare/eldercare leisure market activities Male Female Male Female Male Female Male Female Total 8.8 8.7 0.2 0.9 4.4 3.8 1.5 1.3 Gender of household head Male 8.8 8.7 0.2 0.9 4.4 3.8 1.5 1.2 Female 8.9 8.7 0.1 0.6 4.7 4.4 2.4 2.5 Area of residence Urban 8.7 8.4 0.2 0.8 4.7 4.2 3.2 2.8 Rural 8.9 8.8 0.2 0.9 4.3 3.7 0.9 0.7 Poverty status Poor 9.0 8.8 0.3 1.1 4.5 3.5 0.7 0.7 Nonpoor 8.8 8.7 0.2 0.8 4.4 3.9 1.7 1.5 Residence and poverty status Urban, poor 8.9 8.3 0.3 0.9 4.6 4.0 2.8 3.3 Urban, nonpoor 8.6 8.4 0.2 0.8 4.7 4.2 3.2 2.7 Rural, poor 9.0 8.8 0.3 1.2 4.4 3.4 0.3 0.2 Rural, nonpoor 8.9 8.8 0.2 0.9 4.3 3.8 1.0 0.9 Quintile of welfare aggregate Lowest quintile 9.0 8.8 0.3 1.1 4.4 3.5 0.7 0.7 2 8.8 8.7 0.2 1.0 4.4 3.9 1.0 0.7 3 8.8 8.7 0.2 0.9 4.4 3.8 1.4 1.2 4 8.8 8.7 0.2 0.8 4.5 3.9 1.6 1.4 Highest quintile 8.7 8.5 0.2 0.7 4.4 4.1 2.6 2.3 Region Vientiane 8.7 8.3 0.1 0.7 5.1 4.4 3.5 3.6 North 8.7 8.5 0.3 0.9 4.4 3.6 0.9 0.7 Central 8.9 8.8 0.2 1.0 4.2 3.7 1.7 1.3 South 9.0 9.0 0.1 0.9 4.4 4.1 0.9 0.8 Source: Authors’ calculations based on Lao PDR 2008 using ADePT Gender, adapted table 3a. Note: Includes all men and women, ages 15–64. Gender differences in the use of time are particularly high for vulnerable groups. Whereas women at the bottom of the welfare distribution spend more than one hour per day in childcare and eldercare, women at the top spend half an hour less than women at the bottom in these activities. As a result, women at the top of the welfare distribution can have more leisure time, which closes the gender gap in leisure time. The time allocation is also affected by social norms and women’s opportunity cost. Women in female- headed households spend less time on childcare and eldercare than women in male-headed households. The same result is observed for the Vientiane region. This difference may be attributed to the fact that women in female- headed households and in the Vientiane region are more likely to work 148 Chapter 5: Technical Notes for the Country Gender Diagnostic (not shown), or these women have more voice and agency because of differ- ent social pressures or household dynamics. These facts can be partially explained by the household labor supply model. The model assumes that the household behaves as a single unit; thus, for ease of exposition, it can be regarded as if it has only one decision maker. Consider a household with a single female adult and two children below the working age. The adult is the sole decision maker and takes into account the well-being of all family members, giving the same importance to each of them. She has 24 hours per day to devote to work outside the home, household work, and leisure. Allowing for 8 hours of sleeping, she needs to decide how to allocate the remaining 16 hours between paid work and household production to maximize the household well-being subject to the time constraint. The problem the decision maker faces is graphically represented in figure 5.1. The space depicted in figure 5.1, panel a, is money and hours, over which a budget constraint and a utility (or well-being) function can be defined. The y-axis indicates money that can be used to purchase goods that derive well-being, and the x-axis represents time that can be used for household work, which generates well-being through consumption of goods produced at home, or for work for pay, which serves to buy goods in the market— which in turn might or might not require additional time at home. The x-axis Figure 5.1: Household Time Allocation Problem a. Budget constraint and b. Optimal point indifference curves Money Money (value of Nonmarket C goods and income services) + w x 16 D Utility level Z C Utility level Z Utility level Y B Utility level Y A Utility level X Nonmarket B income A 0 16 Hours in HH work 0 H1 H0 16 Hours in HH work 16 0 Time at paid work 16 0 Time at paid work Note: HH = household. 149 Producing a Country Gender Diagnostic: Part II thus indicates hours spent in household and market work, as hours of house- hold work increase moving right, the hours of paid work decrease (this is indicated with two arrows associated with the axis name). The maximum possible value of hours is 16. The line that connects points A, B, and C represents the possible combinations of time and money for the household.5 Without any work, the household has some money from nonlabor activities that can be spent on goods that generate well-being (the vertical segment from A to B). As the adult reallocates time away from leisure toward paid work, disposable money increases by an amount equal to the wage earned in that time. If all available hours are spent on paid work, the decision maker would have earned the hourly wage times 16 hours of work. Point C indi- cates the total earnings plus the nonlabor income. The well-being (or util- ity) is represented by indifference curves (that is, curves that present combinations of leisure and work that produce the same level of well-being), and the farther right the indifference curve is in the diagram, the higher the well-being of the household. The optimization problem is solved at point D in figure 5.1, panel b. At point D, the household achieves the maximum level of utility that is feasible with the available time and wage. A point like A will not maximize the util- ity of the household, since higher indifference curves can be achieved by substituting hours in household work for paid work (from A to B). This substitution improves the household well-being, since more money is gained (from B to D) than what was necessary to maintain the same level of well- being (from B to C). Next, assume the household has more than one decision maker and thus more than one person generating income. The allocation of time is clearly affected by the power of each person in the couple, differences in utility functions between husband and wife, and social norms or customs. However, part of the decision responds to economic incentives, for example, how much men and women are paid for one hour of work in the market. Couples often find it beneficial to specialize, to some extent, in both household and market work. Often, one person will assume the main responsibility of working outside the home, while the other person focuses on household production. Two economic justifications make this outcome optimal. First, it makes more sense for the person who can earn more money outside the home to spend more hours in the labor market (see figure 5.2, panel a). Second, labor specialization might lead to economies of scale. It might be more productive for one person to work only in the labor market 150 Chapter 5: Technical Notes for the Country Gender Diagnostic Figure 5.2: Budget Constraints in Households with One Female Adult and One Male Adult Money Money Nonmarket Nonmarket C income income + wM × 16 + w1 × 16 Nonmarket D Nonmarket D income income + wW × 16 + w2 × 16 E C Nonmarket Nonmarket B B income income A A 0 16 Hours in HH work 0 8 16 Hours in HH work 16 0 Time at paid work 16 8 0 Time at paid work Note: HH = household. and the other only at home, than if each of them split their time between the two activities.6 Figure 5.2, panel b, shows this as a kink at eight hours of work, based on the assumption that the wage jumps when a person, either male or female, works eight hours. Empirically, it has been found that as wives’ wages increase relative to their husbands’ wages, it is more likely that women spend more time outside the household working, and men increase the time they spend on house- hold work (Hersch and Stratton 1994). Clearly, noneconomic factors influence not only women’s participation in paid work but also the labor division of household chores. This is partly due to the positive association between female bargaining power in the household and female economic empowerment (of which wages may be a determinant). In addition, higher female wages give rise to substitutions within the household. As women earn money outside the household, they can outsource the production of certain household goods—for example, women can buy ready-to-wear clothes instead of fabric to make clothes at home. At the same time, changes in the relative price of household production goods will generate substitutions between hours in household work and hours in paid work. The changes in relative prices are often driven by new technologies. The inven- tion of the washing machine was found to have liberated time for women, which was later used for income-generating activities (Greenwood, Seshardri, and Yorukoglu 2005). Other technological developments, 151 Producing a Country Gender Diagnostic: Part II such as electricity, were found to have similar effects (Dinkelman 2011; van de Walle and others 2013). A study on increased access to water found that women’s freed-up time did not result in greater off-farm work, but that children’s health and education improved—so the extra time was most likely spent caring for children (Koolwal and van de Walle 2013). Although rudimentary, the description of the household labor supply model is intended to remind ADePT Gender users about the trade-offs among the competing activities behind the data. Although the summary tables cannot show all the factors behind the results, users need to keep them in mind when interpreting the tables and graphs, and when they propose additional analysis beyond ADePT Gender. Intrahousehold Allocation Model The household labor supply model assumes that the household behaves as a single unit, with one well-being function. However, it is now widely accepted that households do not behave as a single entity but rather as a group of persons with individual preferences. The unitary model for house- hold decisions has been rejected by the empirical evidence from a substantial body of research (for example, Attanasio and Lechene 2002; Duflo 2003; Lundberg, Pollak, and Wales 1997; Schultz 1990; Thomas 1990; Ward-Batts 2008; and many others). This led to the development of the so-called col- lective models, in which each household member maximizes his or her own utility or well-being. These models assume that household members care about one another, but because they have different preferences, they might disagree. Within this group, cooperative models can be differentiated from noncooperative models, with the main distinction being whether an optimum outcome7 is achieved (Lundberg and Pollak 1993, 2003; Manser and Brown 1980; McElroy and Horney 1981). Under the assumptions of collective models, outcomes usually differ from the unitary model solutions and individual optimal choices. The differences arise from changes in the household budget constraint and in relative prices that influence the “power” of each member’s household decision making, which is influenced by the endowments (both in the form of human capital and productive resources), the social norms, and the legal environment. Finally, the outcomes that are affected by household decisions are of a dif- ferent kind, as described in chapter 1. 152 Chapter 5: Technical Notes for the Country Gender Diagnostic The predictions of all these models have been subject to empirical tests. For the unitary model, most of the studies tested the income-pooling hypothesis, which implies that the effect of a transfer of income on house- hold behavior is independent of the recipient’s identity (husband or wife). The first generation of these studies investigated whether a correlation existed between the origin of household income (typically, what fraction comes from the wife) and the way it is spent. Using data from Brazil, Thomas (1990) finds that the relative share of nonlabor income from the wife has a very significant effect on the health status of the children. However, this finding was challenged, arguing that this phenomenon can be driven by unobserved differences, preferences, and income composition. For example, women who have higher nonlabor income because they saved in the past or they received larger inheritances may also be more likely to spend more on their children’s health and education. The unitary model and the income-pooling hypothesis were also rejected using tests based on exogenous variation in income. Lundberg, Pollak, and Wales (1997) studied the impact of the April 1977 reform of the United Kingdom’s child public support system. Before that time, families with children received a child allowance paid together with their wage. That relationship effectively meant that the child benefits were paid to the higher earner, primarily the father. After April 1977, the old scheme was dropped in favor of a nontaxable child benefit, which is paid directly to the mother. This reallocation of income within the household can be reason- ably treated as exogenous to the affected households. Moreover, the child benefit was a sizable transfer (equal to 8 percent of male earnings for a two-child household). The authors focus attention on the ratio of expen- ditures on children’s clothing and women’s clothing, both relative to men’s clothing. Their findings are unequivocal—both ratios rose signifi- cantly after the reform. Another strong rejection is provided by Duflo (2003), who analyzed a reform of the South African social pension program for the elderly that extended benefits to a large, previously not covered, black population. Because of the eligibility criteria, the coverage was not universal—in some households, only one of the grandparents receives the benefit. Duflo uses a difference-in-differences approach8 based on the demographics of the household to control for selection in eligibility. She shows that the recipi- ent’s gender is of considerable importance for the effect transfers have on children’s health—a payment to the grandfather has no significant effect, 153 Producing a Country Gender Diagnostic: Part II whereas the same amount paid to the grandmother results in a huge improvement in the health status of girls in the family. As for the empirical evidence that rejects the income-pooling predic- tion mounted, collective models were developed. One of these models’ predictions is that outcomes depend on the “power” or “distribution fac- tors” of spouses, since they influence the household decisions (Bourguignon and Chiappori 1992; Chiappori 1988). These models predict that a trans- fer to women might have a different effect than a transfer to men, and that the differential effect depends on the “power” the recipient has in house- hold decisions (Attanasio and Lechene 2002). Later, Bourguignon, Browning, and Chiappori (2009) translated this argument into a testable prediction. If the collective setting is correct, all distribution factors should operate in a similar way and thus their respective effects on various aspects of household behavior, or equivalently into what the experts in this sub- field call the property of the “z-conditional demands.” Empirically, this means that once the effect of distribution factors is controlled for—which is what z-conditional demands do—they should have no effect on house- hold decision making. Thus, the key questions become what influences the “distribution fac- tors” or “power” within the household, and how they can be measured. It is with these two questions in mind, along with an awareness of power’s role in outcomes, that tables on voice and agency should be interpreted and included in an analysis. Notes 1. This section draws on the presentation of the household labor supply model in Blau, Ferber, and Winkler (2006); Blundell and Macurdy (1999); and Ehrenberg and Smith (2009). 2. Or decisions are made by a head of household who takes into account the well-being of each household member, giving all of them equal importance. 3. Leisure is another consumption good. 4. Other countries’ time patterns can be found in World Bank (2012), chapter 5, 217–22. 5. This is equivalent to a budget constraint in the simple consumption maximization problem. 154 Chapter 5: Technical Notes for the Country Gender Diagnostic 6. For example, because they become faster at doing the job, because they save in transportation costs to the workplace, and so on. 7. This optimum outcome is called a Pareto outcome, based on the exten- sive research of Vilfredo Pareto. 8. A difference-in-differences approach is an econometric method used to estimate causal relationships. References Attanasio, Orazio, and Valérie Lechene. 2002. “Tests of Income Pooling in Household Decisions.” Review of Economic Dynamics 5 (4): 720–48. Becker, Gary. 1965. “A Theory of the Allocation of Time.” Economic Journal 75 (2999): 493–517. ———. 1974. “A Theory of Marriage.” In Economics of the Family: Marriage, Children, and Human Capital, edited by Theodore W. Schultz, 299–351. Chicago: University of Chicago Press. Blau, Francine D., Marianne A. Ferber, and Anne E. Winkler. 2006. The Economics of Women, Men, and Work. 5th ed. Upper Saddle River, NJ: Pearson/Prentice Hall. Blundell, Richard, and Thomas Macurdy. 1999. “Labor Supply: A Review of Alternative Approaches.” In Handbook of Labor Economics, vol. 3, edited by Orley Ashenfelter and David Card, 1559–1695. Amsterdam: Elsevier. Bourguignon, François, Martin Browning, and Pierre-André Chiappori. 2009. “Efficient Intra-Household Allocations and Distribution Factors: Implications and Identification.” Review of Economic Studies 76 (2): 503–28. Bourguignon, François, and Pierre-André Chiappori. 1992. “Collective Models of Household Behavior: An Introduction.” European Economic Review 36 (2–3): 355–64. Chiappori, Pierre-André. 1988. “Rational Household Labor Supply.” Econometrica 56 (1): 63–90. Dinkelman, Taryn. 2011. “The Effects of Rural Electrification on Employment: New Evidence from South Africa.” American Economic Review 101 (7): 3078–108. Duflo, Esther. 2003. “Grandmothers and Granddaughters: Old-Age Pensions and Intrahousehold Allocation in South Africa.” World Bank Economic Review 17 (1): 1–25. 155 Producing a Country Gender Diagnostic: Part II Ehrenberg, Ronald G., and Robert S. Smith. 2009. Modern Labor Economics. 10th ed. Boston: Pearson/Addison-Wesley. England, Paula. 2003. “Separative and Soluble Selves: Dichotomous Thinking in Economics.” In Feminist Economics Today: Beyond Economic Man, edited by Marianne A. Ferber and Julie A. Nelson, 33–59. Chicago: University of Chicago Press. Greenwood, Jeremy, Ananth Seshardri, and Mehmet Yorukoglu. 2005. “Engines of Liberation.” Review of Economic Studies 72 (1): 109–33. Hersch, Joni, and Leslie S. Stratton. 1994. “Housework, Wages, and the Division of Household Time for Employed Spouses.” American Economic Review 84 (2): 120–25. Koolwal, Gayatri, and Dominique van de Walle. 2013. “Access to Water, Women’s Work, and Child Outcomes.” Economic Development and Cultural Change 61 (2): 369–405. Lao PDR, Government of. 2008. “Household Survey 2008.” Department of Statistics, Ministry of Planning and Investment, Vientiane. Lundberg, Shelly J., and Robert A. Pollak. 1993. “Separate Spheres Bargaining and the Marriage Market.” Journal of Political Economy 101 (6): 988–1010. ———. 2003. “Efficiency in Marriage.” Review of Economics of the Household 1 (3): 153–67. Lundberg, Shelly J., Robert A. Pollak, and Terence J. Wales, 1997. “Do Husbands and Wives Pool Their Resources? Evidence from the United Kingdom Child Benefit.” Journal of Human Resources 32 (3): 463–80. Manser, Marilyn, and Murray Brown. 1980. “Marriage and Household Decision-Making: A Bargaining Analysis.” International Economic Review 21 (1): 31–44. McElroy, Marjorie B., and Mary Jean Horney. 1981. “Nash-Bargained Household Decisions: Toward a Generalization of the Theory of Demand.” International Economic Review 22 (2): 333–49. Mincer, Jacob. 1962. ‘‘Labor Force Participation of Married Women: A Study of Labor Supply.’’ In Aspects of Labor Economics, edited by H. Gregg Lewis, 63–105. Princeton, NJ: Princeton University Press. Schultz, T. Paul. 1990. “Testing the Neoclassical Model of Family Labor Supply and Fertility.” Journal of Human Resources 25 (4): 599–634. Thomas, Duncan. 1990. “Intra-Household Resource Allocation: An Inferential Approach.” Journal of Human Resources 25 (4): 635–64. 156 Chapter 5: Technical Notes for the Country Gender Diagnostic van de Walle, Dominique, Martin Ravallion, Vibhuti Mendiratta, and Gayatri Koolwal. 2013. “Long-Term Impacts of Household Electrification in Rural India.” Policy Research Working Paper 6527, World Bank, Washington, DC. Ward-Batts, Jennifer. 2008. “Out of the Wallet and into the Purse: Using Micro Data to Test Income Pooling.” Journal of Human Resources 43 (2): 325–51. World Bank. 2012. World Development Report 2012: Gender Equality and Development. Washington, DC: World Bank. https://openknowledge .worldbank.org/handle/10986/4391. 157 PART III Analyzing Gender in Labor Markets Part III of the ADePT Gender manual goes into more depth on gaps in economic opportunities, particularly gaps in employment and earnings, and occupational segregation. In this section, ADePT Gender aims to work with tools that have long been used in country analyses of labor market inequali- ties and also introduces a few of the more recent methods for studying the intersection of gender and labor economics. Chapter 6 provides guidance on how to produce and interpret the tables and graphs to further analyze gender gaps in the labor market. Using the household survey from Nepal in 2010–11 and Panama in 2008, the chapter discusses select tables and graphs to show the user how to interpret them. Mirroring the organization of part II, chapter 7 provides technical notes for understanding the theories behind the empirical analysis. This organization aims to help different audiences quickly access the necessary inputs and interpret outputs. The user familiar with quantitative analysis and gender issues may need to consult only chapter 6, whereas the user less familiar with quantitative analysis or the economic angle of gender issues can also consult chapter 7. 159 Chapter 6 How to Interpret the Results of Labor Market Analysis This chapter discusses how to interpret the results from the tables and graphs of ADePT Gender of the labor market analysis. The description of the tables and graphs is directed toward practitioners with a basic knowledge of econometrics and statistics in the context of labor markets. To comple- ment the guidance of this chapter, chapter 7 includes a brief discussion of key topics of econometrics and labor markets for those who need to acquire or review them, including references for those with an appetite for more material. The chapter relies on techniques that have been used extensively by researchers and policy makers to explain gender gaps in employment and earnings. The first section covers the output that provides measures of earnings inequality: overall, within gender, and between men and women. These measures of inequality offer context for the analysis of gender. We know that gender imbalances are present everywhere, but are they more or less important than other inequalities that exist for vulnerable and minority groups? How large is the gap in pay between men and women? Does the pay gap depend on the type of employment? Are these results consistent with the evolution of other outcomes as those described in part II? These are some of the questions that these groups of tables and graphs aim to shed light on. The second section tries to help the user understand the different factors that influence the gender gap in pay. Following the wealth of literature in this area, ADePT Gender focuses on reduced-form approaches that provide 161 Analyzing Gender in Labor Markets: Part III an estimate of the potential (net) effect of a bundle of factors. That is done using decomposition techniques that separate “composition” and “wage structure” effects—that is, factors coming from observable differences in labor market skills of men and women and all the other residual factors. The main question that decomposition methodologies seek to answer is the extent of the gender gap in pay there would be if women received the same pay for their skills as men. Or conversely, how large the gender gap would be if women had the same labor market skills as men. Thus, decomposition methodologies help the user understand how much of the gender gap in pay can be “solved” or narrowed by helping women access education and employment in certain sectors and occupations to match men. Instead, the wage structure effect is a blurred measure of any other issue that translates into differences in wages between men and women. These differences could range from unobserved skills, such as ability to work in teams, to more complex issues, such as discrimination, which would require a change in society’s mentality. All of the analysis is conducted using logs of wages per hour as the main variable to be explained. This chapter and the next might loosely refer to wages without clarifying that the variable being analyzed is the natural loga- rithm of the hourly wage. It is standard to use logs, because they are linked to the formal model of investment in human capital, and because they pro- vide a parsimonious specification that fits data from different countries extremely well. Also, this chapter uses wages and earnings interchangeably, though it was noted in chapter 3 that they correspond to different compo- nents of labor income. Earnings Inequality Content This section describes four tables (ADePT tables 12a–12d) and three graphs (ADePT figures 12a–12c) with measures of earnings inequality.1 All the measures are presented separately for wage workers and self-employed workers. To understand gender inequality, it is necessary to first understand the overall wage structure prevalent in the labor market.2 Gender differences in employment and pay can be affected by how dispersed or compressed the distribution of earnings is and the supply and demand of low- and 162 Chapter 6: How to Interpret the Results of Labor Market Analysis high-skilled workers in the labor market, which in turns affects the price of skills. For example, the gender wage gap is more likely to be larger in econo- mies with high returns to education and a disproportionately large group of working women with little education. The overall earnings inequality measures distributional aspects of earnings, computed using all employed men and women. The term within-gender earn- ings inequality refers to differences in earnings, but for men and women sepa- rately. ADePT tables 12a–12d and ADePT figure 12b show several measures of earnings inequality—both overall and within gender. Notice that in this module of ADePT, inequality measures are computed using earnings as opposed to household or per capita income. Many of these measures can also be computed for income or other welfare aggregates in the ADePT Poverty and inequality module, and a few of them for household labor income in the ADePT Labor module. In this ADePT Gender module, all measures are computed separately for wage workers and the self-employed. As discussed in chapter 3, earnings for the two groups are calculated using different survey questions—even modules3—leaving room for different types of measurement errors that might invalidate the comparison. In addition, the sample of wage workers and self- employed can differ in the observable and unobservable characteristics; thus, gender biases can show up if men and women are sorted differently into these two types of occupations. ADePT table 12a reports the three most used inequality measures in labor economics: (a) the log wage differential between the 90th and 10th percentiles, (b) the log wage differential between the 90th and 50th percentiles, and (c) the log wage differential between the 50th and 10th percentiles. Looking into more detail at the whole distribution of earn- ings, ADePT table 12b reports the share of total earnings captured by each welfare decile (see table 6.2). ADePT table 12c goes into further details on inequality measures that labor economics borrows from the poverty and inequality field. It reports inequality measures from the Theil index, the Atkinson index, and the generalized entropy inequality index. These indica- tors are included for sensibility analysis. Accompanying these tables, ADePT figure 12a plots the earnings distribution for men and women, wage earners, and the self-employed separately. The term gender gap in earnings—or simply gender gap in pay—refers to the differences in earnings between men and women. ADePT table 12d shows the “raw gap” in earnings computed at different points of the earnings distribution, whereas ADePT figures 12b and 12c compare the earnings 163 Analyzing Gender in Labor Markets: Part III distributions of men and women. The term raw gap is usually used to denote the differences in pay by simply comparing men’s earnings and women’s earnings (average or other statistics), before conditioning in other variables that could serve to explain it. Concepts Overall and Within-Gender Earnings Inequality The analysis of the wage structure of labor markets focuses on understand- ing the wages associated with different jobs. Understanding the wage struc- ture of the labor market is important for several reasons. From a gender perspective, it is useful to understand how large gender inequality is com- pared with other inequalities in the labor market. For example, is the gen- der gap in pay as large as the difference in pay between low-skilled and high-skilled workers? Is the gender gap in pay as important as the racial or ethnic gap in pay? Moreover, understanding the overall earnings inequality is crucial for poverty and income inequality, as labor income is the poor’s main source of income. Labor economists are usually interested in understanding the difference in pay between low-skilled and high-skilled workers. With that purpose in mind, it is assumed that workers with the lowest earnings (bottom of the distribution density function of earnings) are low skilled and workers with the highest earnings (top of the distribution density function of earnings) are high skilled. This is probably the most common measure used in studies of labor markets and of gender wage gaps (Autor, Katz, and Kearny 2008; Blau and Kahn 1997; Katz and Autor 1999; Katz and Murphy 1992; Mulligan and Rubinstein 2008). Others, however, prefer to work directly with the ratio of earnings between different education levels, such as the ratio of wages between college and high school workers. If the user is also interested in comparing countries with very different education attain- ments, the ratio of wages in the 90th to 10th percentiles of the cumulative density distribution curve might be better suited to show the relative wage differences. The ratio of wages in the 90th and the 10th percentiles of the earnings distribution curve, mathematically, is 90p/10p = logw90p − logw10p, 164 Chapter 6: How to Interpret the Results of Labor Market Analysis where logw90p is the natural logarithm of the wage that leaves 90 percent of employed workers below that cutoff after sorting all workers according to hourly rate of pay. Similarly, logw10p is the wage that leaves 10 percent of employed workers below that level of earnings. Other ratios can be com- puted in the same vein. For example, the ratio 90p/50p and 50p/10p inform whether the distribution of earnings is symmetric around the median. As expected, they are mathematically defined as 90p/50p = logw90p − logw50p, 50p/10p = logw50p − logw10p. The main advantage of these measures—based on ratios of logs of wages taken at different points in the distribution curve—is that computing and interpreting them is straightforward. However, they use information only from a few points (or percentiles) of the distribution curve and miss its other characteristics. To complete the description of the distribution curve of earnings, ADePT table 12c shows additional inequality measures borrowed from the field of poverty and inequality. Working with these measures is important for sensibility analysis, as they capture different features of the distribution’s functions. Moreover, users might be interested in comparing labor income and total income inequality, in which case it is necessary to have the same measures. Furthermore, these measures are associated with the theory of welfare, which has a long tradition in economics. These mea- sures are explained in more detail by Foster and others (2013), and this section presents only a quick description of each of them. In all cases, ADePT Gender focuses on inequality of outcomes. The user interested in inequality of opportunities can use other available tools, such as the World Bank’s Human Opportunity Index. The coefficient of variation is a measure of the distribution curve’s disper- sion that is normalized by its mean to allow—among other reasons— comparisons over time or across countries. The normalization removes the scale of the variable. Mathematically, i =1 ( w i − μw ) /N ⎤ ⎡∑ N 2 CV = ⎣ ⎦, μw where wi represents the earnings of individual i, mw is the sample mean of wages, and N is the total number of employed persons in the sample. 165 Analyzing Gender in Labor Markets: Part III The Gini coefficient is one of the most popular inequality indicators. Its objective is to summarize in one number the information coming from the Lorenz curve (see figure 6.1). This curve shows the percentage of total income that belongs to the poorest p percent of the population. If the income were perfectly even in distribution, the Lorenz curve would be equivalent to a 45-degree line—or the line of equality. The farther away the Lorenz curve is from the line of equality, the more pervasive the inequal- ity. In the case of earnings (in contrast with income), the adjusted Lorenz curve shows the percentage of total earnings that belongs to the poorest p percent of the population. Comparing the Lorenz curves generated with total income with those coming from earnings allows us to examine the contribution of nonlabor income to inequality. The Gini coefficient measures the area between the Lorenz curve and the line of equality. It can be expressed as a function of the Lorenz curve,4 A Gini = , A + B′ Gini = ( 21 − B) , 1 2 Gini = 1 − 2. B, 1 ∫ Gini = 1 − 2 L ( p ). dp, 0 where p stands for the lowest p percent of the working population, and L(p) refers to the share of total earnings received by the lowest p percent of the working population—that is, the Lorenz curve. The Gini coefficient takes values between 0 and 1: the larger the Gini, the more severe the inequality. There are two extreme cases. When earnings are evenly distrib- uted among the entire working population, the Lorenz curve coincides with the line of equality, the area A equals 0; hence, the Gini coefficient is 0. When only one individual has access to all earnings of the group, the Lorenz curve is 0 for all p <100, the area B is 0, and the Gini coefficient is 1, its maximum value. 166 Chapter 6: How to Interpret the Results of Labor Market Analysis Figure 6.1: Lorenz Curve Percentile of earnings A Lorenz curve L(p) B Percentile of population Behind each of the following inequality measures are implicit value judg- ments. The Atkinson index has the advantage of making these judgments explicit. Assume a social welfare function, W(w) = W(w1, w2, w3, w4, … wN), where this function depends on the earnings of each person in the population—that is, its independent variable is N × 1 vector of earnings for a given population of size N, wi represents the income of individual i, and mw is the mean wage under this distribution of earnings. Next, consider a certain level of wage w* such that W(w1, w2, w3, …, wN) = W(w*, w*, w*, …, w*). Under this earnings distribution, every individual would earn the same, w*, and the society will achieve an equivalent level of social welfare to that under the current distribution of earnings. Then, the Atkinson index is μw − w ∗ A(w ) = . μw It measures the fraction of the average earnings that a social planner would be willing to give up to get an even distribution of earnings. 167 Analyzing Gender in Labor Markets: Part III In practice, the specific value of A(w) depends on the specification of W(w). Atkinson (1970) proposes the following functional form: ywi(1−ε ) 1 ∑ N W (w ) = . N i =1 (1 − ε ) The great advantage of this functional form is that it is possible to sum- marize the value judgment in the parameter e, known as inequality aversion. Note that when e equals 0, W(w) corresponds to mw, the average income of the population. This is the case in which the social planner cares about only the “size of the pie.” Mathematically, for e = 0 ∑ wi N W (w ) = . i =1 N Alternatively, when the inequality aversion e approaches infinity (e → É), it turns into the Rawlsian welfare function W(w) = min{w1, w2, w3, w4, …, wN} = wpoorest. In this case, the lower the ratio wpoorest/mw, the higher the earnings inequality. In practice, researchers use values of e that fall between these two extremes, such as e = 0.5, e = 1, and e = 2. In a similar vein, the generalized entropy index of inequality and Theil index have their origins in the theory of information (Cowell 2000). The formula is ⎡ c ⎤ ∑ ⎛ wi ⎞ 1 N E(c ) = .⎢ ⎜ ⎟ − 1⎥ . N .c( c − 1) ⎣ i =1 ⎝ μ w ⎠ ⎦ One if the main advantages of this index is that it allows us to focus on the inequality in different components of the distribution by choosing the value of c. The larger the value of c, the larger the weight given to changes in the right tail of the distribution. It is possible to show that when c → 1, the entropy index converges to the Theil index.5 Adjusting these indexes to labor market analysis is not fully possible, since differences in earnings are necessary to maintain the incentives for individuals to invest in education. However, the Atkinson index can be used as a signal of the reallocation of skills and jobs that should happen to 168 Chapter 6: How to Interpret the Results of Labor Market Analysis achieve greater equality in the labor market. This way of thinking about the differences gets closer to the concept of equality of opportunity. Even if ADePT Gender users’ ultimate goal is to understand the underly- ing factors that contribute to the gender gap in pay, analyzing the within- gender inequality is important, since it is a way to gather information about the economy’s wage structure. All the measures described above are thus computed for men and women separately. Gender Earnings Inequality ADePT Gender computes the differences in wages between men and women at different points along the earnings distribution curve. The mean gender pay gap is computed as the differences between the log wage of men and women. This gender pay gap is usually referred to as the raw gender pay gap, because it does not take into account any factor that could explain the differences in wages between men and women. The gender pay gap is usually presented as the percentage difference. Mathematically, GWG = ln( w M ) − ln( w F ), where GWG is the raw average gender wage gap; ln( w M ) is the natural loga- rithm of the mean male (hourly) wage; and ln( w F ) is the natural logarithm of the mean female (hourly) wage. This is an approximation of the following: wM − wF GWG = − . wF Analysts use the average wage of women, men, or total employed as a reference point, and the calculation has subtle differences that have implica- tions for the interpretation. The different formulas are presented in box 6.1. The gender gap in pay can be estimated at other points of the earnings distribution in the following way: GWGp = ln(wMp) − ln(wFp), where GWGp is the raw gender wage gap measured at the pth percentile; ln(wMp) is the natural logarithm of the male (hourly) wage evaluated at the pth percentile of the male earnings distribution curve; and ln(wFp) is the 169 Analyzing Gender in Labor Markets: Part III Box 6.1: Interpreting Different Measures of the Gender Gap in Pay for the Whole Distribution of Earnings The following are some of the most widely used metrics of the gender gap in pay: wM − wF G1a = , wM wM − wF G1b = , wF wM − wF G1c = , w G2 = ln( w M ) − ln( w F ), G3 = ln( w M ) − ln( w F ), where the upper bar indicates the (arithmetic) mean. Even when all these metrics are used interchangeably with the same objective—to measure the gender gap in pay— they may yield different values. The first two metrics indicate the difference between men’s and women’s average wages as a percentage of the “reference group,” which can be either men (G1a ) or women (G1b ). The third metric uses the average of all employed workers. Since most of the time the analyst works with logs of wages, the gender gap in pay can be approximated by G2 , which is a good approximation of G1b , if the difference between w M and w F is “small.” The reason is that the logarithm function passes through the point (0, 1), with a local slope of 1. Thus, for x = 1 wM or a small deviation, it holds that y = ln(x) ≈ x − 1. If we say that x = , then wF ⎛ wM ⎞ wM wM − wF G2 = ln ⎜ ≈ − 1= = G1b . ⎝ wF ⎟ ⎠ wF wF In general, when w M > w F , GWG1b ≈ GWG2 ≥ GWG1a. All metrics can be interpreted as the percentage difference between the average wages for men and women, regardless of which gender is used as the reference. Another measure is the difference between the means of the log wages as opposed to the log of mean wages. On the basis of the properties of natural logarithms, this measure is equivalent to the difference of the logs of the geometric—as opposed to the arithmetic—means of the wages: n s ⎡⎛ n ⎞ 1 n ⎤ ⎡⎛ s ⎞ 1 s ⎤ G3 = 1 n ∑ ln( w Mi ) − 1 s ∑ ln( w Fj ) = ln ⎢⎜ ⎢⎝ ∏ w Mi ⎟ ⎠ ⎥ − ln ⎢ ⎥ ⎢⎜ ∏ w Fj ⎟ ⎥. ⎥ i =1 j =1 ⎣ i =1 ⎦ ⎣⎝ j =1 ⎠ ⎦ To some extent, it is possible to establish a relationship between the values of G2 and G3 . Notice that the arithmetic mean is always greater than the geometric mean, except when all wages are the same. The difference between the two increases with the level of wage inequality. 170 Chapter 6: How to Interpret the Results of Labor Market Analysis natural logarithm of the female (hourly) wage evaluated at pth percentile of the female earnings distribution curve.6 Interpreting the Results This section presents concrete examples using data from the 2008 Panama liv- ing standards household survey to show how to interpret the tables and graphs produced by ADePT Gender. It first discusses how to interpret most of the output related to the overall and the within-gender earnings inequality. Later, we address measures of gender gap in pay, including data limitations, such as the issue that women are more likely to have missing data on earnings. Overall and Within-Gender Inequality ADePT tables 12a–12c present different measures of overall and within-gender earnings inequality. ADePT table 12a presents three different ratios of earnings at different percentiles: 90p/10p, 90p/50p, and 50p/10p (see table 6.1). The three measures can be interpreted in the same way as the log point difference of two wages—for example, w90p and w10p—for all wage workers. This difference Table 6.1: ADePT Gender Table 12c, Panama 2008 Table 12c: Male and Female Earnings Shares and Ratios: Wage Workers and Self-Employed Wage worker Self-employed Total Male Female Total Male Female Decile 1 0.9 1.2 0.5 3.1 3.5 1.9 2 2.5 3.2 1.3 3.5 3.8 2.3 3 3.8 4.7 2.1 5.2 5.1 5.7 4 5.6 6.2 4.5 5.4 5.5 5.2 5 6.4 6.9 5.4 7.1 7.5 5.7 6 8.9 9.5 7.7 8.0 8.8 5.3 7 9.4 10.2 8.1 9.5 9.6 9.3 8 11.8 10.7 13.6 14.2 14.3 14.0 9 16.3 14.7 19.2 18.7 18.8 18.5 10 34.4 32.6 37.6 25.1 23.3 32.0 Earnings ratio p90/p10 6.9 5.4 9.4 62.4 46.7 57.6 p90/p50 2.6 2.5 2.7 5.8 4.7 6.9 p50/10 2.7 2.1 3.4 10.8 10.0 8.4 Source: Based on ADePT Gender using Panama 2008. Note: p = percentile. 171 Analyzing Gender in Labor Markets: Part III can also be interpreted as the growth rate—for example, wage workers in the 90th percentile earn almost twice as much as wage workers in the bottom per- centile of the earnings distribution. Note that when comparing the measures of all workers and those by gender, the dispersion of women’s earnings is larger than that of men’s, probably resulting from more women earning very little rela- tive to men. This factor can be further corroborated with more specific analyses, which are described in the coming sections of this chapter. The measures of inequality borrowed from the poverty field are shown in ADePT table 12b (table 6.2). Within-gender inequality is always higher for women than for men, regardless of the chosen measure.7 Notice also that inequality is higher among the self-employed than among the wage workers. Measures featuring the whole distribution of earnings can be found in ADePT table 12c and figure 12b.8 As can be seen at first glance, the distri- bution of earnings for the self-employed is more dispersed than the one for wage workers, and even more so for women than for men (see figure 6.2). These results are consistent with those from ADePT table 12c that indicate that the dispersion is 10 times larger among the self-employed than among wage workers; the 90p/10p ratio is 2 times larger for the self-employed than for wage workers (table 6.1). Table 6.2: ADePT Gender Table 12b, Panama 2008 Table 12b: Male and Female Earnings Inequality Indexes (Detailed): Wage Workers and Self-Employed Work category Total Male Female Wage worker Gini coefficient 0.434 0.423 0.448 Coefficient of variation 1.208 1.235 1.142 A(½) 0.165 0.158 0.174 A(−1) 0.570 0.517 0.619 A(0) 0.298 0.277 0.323 GE(0) 0.353 0.324 0.391 GE(1) 0.388 0.382 0.391 GE(2) 0.781 0.798 0.723 Self-employed Gini coefficient 0.664 0.632 0.723 Coefficient of variation 2.438 2.256 3.016 A(½) 0.382 0.346 0.453 A(−1) 0.924 0.907 0.931 A(0) 0.656 0.611 0.716 GE(0) 1.069 0.944 1.259 GE(1) 0.907 0.805 1.172 GE(2) 2.458 2.096 3.795 Source: Based on ADePT Gender using Panama 2008. Note: A = Atkinson index; GE = General Entropy index. 172 Chapter 6: How to Interpret the Results of Labor Market Analysis Figure 6.2: ADePT Gender Figure 12b, Panama 2008 Figure 12b: Male and Female Earnings Distributions: Wage Earners and Self-Employed a. Male b. Female 0.8 0.35 0.7 0.30 0.6 0.25 Density 0.5 Density 0.20 0.4 0.15 0.3 0.2 0.10 0.1 0.05 0 0 –6 –5 –4 –3 –2 –1 0 1 2 3 4 –6 –5 –4 –3 –2 –1 0 1 2 3 4 Log(hourly wages) Log(hourly wages) Wage earners Self-employed Median wage earners Source: Based on ADePT Gender using Panama 2008. Complementing these two graphs, ADePT table 12c (table 6.1) shows how much workers in each population decile—determined by sorting house- holds according to the welfare aggregate measure selected by the user— contribute to the total earnings of a certain employed population. Note that by construction, the sum of each column equals 100, reflecting the total earnings of the group indicated in the column (that is, female wage earners). For example, in Panama in 2008, employed men in the bottom 10th wealth decile got only 5 percent of their total earnings from wage employment. The contribution of workers to earnings increases as the decile becomes wealthier. Their contribution after the eighth decile is more than 10 percent, larger than the percentage of the population. Gender Gap in Pay This section discusses how to interpret the tables and graphs that compare women’s and men’s earnings. ADePT table 12d presents the most common measures of the gender gap in pay. They include the average (or mean) gender gap in pay and the percentile that the median female earning repre- sents in the male earnings cumulative distribution curve. It is useful to look 173 Analyzing Gender in Labor Markets: Part III at the gender gap in pay evaluated at different points of the earnings distri- bution, but these are provided later when presenting the Juhn-Murphy- Pierce decomposition. Each of the measures is computed separately for wage workers and the self-employed. Table 6.3 shows the measures for the example illustrated throughout the book. In Panama in 2008, employed women in wage jobs earned 7 percent less than employed men. For an idea of the magnitude of the gap in the local con- text, the next measure evaluates the percentile in the male distribution at which the median wage of women lands. The median wage of women is equivalent to moving 12 percentiles down from the midpoint of the male distribution. The same concept is plotted with more detail in ADePT figure 12c (figure 6.3), which captures where women stand in the male earnings distribution. In the graph, the x-axis represents the percentile of earnings for men, and the y-axis represents the percentile of earnings for men or women. Hence, the 45-degree line represents male percentiles—or line of gender equality in pay—and the curve represents female percentiles in the men’s earnings distri- bution. When the women’s curve is above the 45-degree line, the gender wage gap favors men; when the women’s curve is below the 45-degree line, women earn more than men. If the curve crosses the 45-degree line from left to right, low-skilled women earn less than low-skilled men, and high-skilled women earn more than high-skilled men. The distance of the curve to the 45-degree line offers an idea of the size of the gap—the farther away the curve is from the diagonal line, the larger the size of the gap. For example, if we take the wage of the 20th percentile of men’s earnings distribution, the curve indicates that a woman with that same wage is in the 31st percentile of the women’s earnings distribution curve. In general, the farther the curve is from the 45-degree line, the larger the gender wage gap. Table 6.3: ADePT Gender Table 12d, Panama 2008 Table 12d: Male-Female Gap in Earnings: Wage Workers and Self-Employed Mean gender Percentile of earnings gap women’s median in (in %) men’s distribution Work category Wage worker 6.6 38.3 Source: Based on ADePT Gender using Panama 2008. Note: The mean gender earnings gap is the difference of the mean log female earnings and the mean log male earnings. 174 Chapter 6: How to Interpret the Results of Labor Market Analysis Figure 6.3: ADePT Gender Figure 12c, Panama 2008 Figure 12c: Male-Female Earnings Concentration Curves: Wage Earners and Self-Employed a. Wage earners b. Self-employed 100 100 Cumulative % of earnings by females Cumulative % of earnings by females 90 90 80 80 70 70 60 60 50 50 40 40 30 30 20 20 10 10 0 0 0 20 40 60 80 100 0 20 40 60 80 100 Men’s earnings percentile Men’s earnings percentile Wage earners Self-employed Line of equality Line of equality Source: Based on ADePT Gender using Panama 2008. To shed additional light on how to measure the distribution of earnings, ADePT table 16a also shows the gender gap in pay computed at different percentiles of the earnings distribution of men and women. For example, in ADePT table 16a for Panama 2008, women in the 10th percentile of women’s wage distribution earn 28 percent less than men in the 10th percentile of the men’s distribution. Earnings Decompositions Contents ADePT Gender produces 16 tables (ADePT tables 13a to 17d) and 8 graphs (ADePT figures 13a to 17b) with different earnings decomposi- tions by gender. All the decompositions are done using two specifications: the human capital model and the full model (see box 6.2 for explanations of these models). ADePT Gender generates estimates for three types of 175 Analyzing Gender in Labor Markets: Part III Box 6.2: ADePT Gender Model Specifications for Earnings Equations Human capital specification—This specification includes education and experience variables and controls for place of residence. Education—Education can be included in the regression model in three ways: (a) using number of years of completed education, (b) using a set of dummy variables for levels of completed education, or (c) using a combination of the two previous specifications, including years of completed education and dummies for the levels of education. The choice of education variables clearly depends on the available information, as well as on the country context. Generally, researchers use both years of education and dummy variables to indicate completion of education levels, since level of completion can have an additional payoff.a The ideal combination of the two sets of variables differs from country to country, depending on the prevalent distribution of education. One possible specification could be to add secondary, vocational, and college. However, for certain African countries, it might be reasonable to use secondary and postsecondary or to use only vocational and college for a middle-income country. Experience—ADePT Gender uses age as a proxy for potential experience. Potential experience is defined as age – years of education – age of entrance to school. However, given that users might decide to compare countries that have different ages of school entrance, ADePT Gender uses age. Age and potential experience can be thought of as a similar variable—given that the education variables are also included in the regres- sion estimation—but they differ on its interpretation. Place of residence—ADePT Gender controls for place of residence with urban dummies and region of residence dummies. Full model specification—This specification includes all the variables from the human capital specification plus industry and occupation dummies. Occupation—Occupation is usually included as a set of dummy variables. The general recommendation is to include them as one-digit International Standard Classification of Occupations (ISCO) classifications (see chapter 3 for more details). However, on occasion, some analyses use two- and three-digit dummies when they have a large number of observations. For ADePT Gender, the omitted category is the lowest number in the occupation classification, which, if the user follows the ISCO classifi- cation, is agriculture. Users following other classifications should avoid having an “other” category for omitted occupations, which could make interpretation difficult. Industry—Similar to occupation, the most common approach for industry is to include dummies at the one-digit International Standard Industrial Classification (ISIC) level. The same rules for the omitted category apply. Other specifications—Other authors have used other specifications, depending on the richness of their data. Variables can be classified in two types: (a) those that directly (continued) 176 Chapter 6: How to Interpret the Results of Labor Market Analysis Box 6.2: ADePT Gender Model Specifications for Earnings Equations (continued) affect productivity (as explained by theories) and (b) those that are good proxies for a household’s responsibilities. The first group (those that affect productivity) includes controls for union affiliation and size of firm (that is, number of employees). Regarding household duties, the typical controls are marital status and number and age of children. ADePT Gender does not include any of these variables because they are not widely avail- able or—as in the case of household responsibilities—must be interpreted with caution, as they might be endogenous to the model. The user always has the option to add them. Source: Based on Blau and Kahn 1997. Notes: a. The additional payoff of having completed a certain level of education can be explained by different factors. First, the government or large enterprises might have policies about hiring employees with a certain level of education. The most common example for middle- and high-income countries is the secondary degree. Companies like Walmart hire employees with a secondary degree. Second, signaling theories applied to labor markets can be tested using completed levels of education (Bedard 2001). decompositions methodologies that are the most commonly used and with the easiest interpretation. The two parametric decompositions are the Oaxaca-Blinder, which decomposes earnings at the mean of the earnings distribution curve, and the Juhn-Murphy-Pierce, which decomposes earn- ings at different percentiles of the earnings distribution curve. Finally, ADePT Gender also presents the results of one of the nonparametric decompositions—the Ñopo decomposition—that have been applied to several developing countries (World Bank 2012). Decompositions must be used with caution, and users should avoid drawing overly ambitious conclusions from them. It is important to understand the models’ underlying assumptions as well as to remember that decompositions are partial equilibrium approaches and do not mea- sure causal relations. Decompositions do not deepen our understanding of the mechanisms underlying the relationships between factors and out- comes; rather, they signal potential mechanisms at play to engage in further analysis. For example, if the decomposition indicates that differ- ences in occupational affiliation account for a large fraction of the gender wage gap, one should explore in more detail why and how men and women choose their fields of study and occupations and whether discrimi- nation exists in accessing certain occupations or jobs. Decompositions are an excellent tool for tracking progress and assessing the magnitude of the problem. 177 Analyzing Gender in Labor Markets: Part III All decomposition methodologies use the Mincer earnings equation as the starting point. That means that this reduced-form analysis relies on the assumptions of the human capital model described in chapter 5 and other econometric assumptions of the earnings equation. Users who are unfamiliar with the Mincer earnings equation can find an introduction to it in chapter 7. Box 6.2 describes the variables used for the earnings equa- tions by ADePT Gender. In addition, for both the Oaxaca-Blinder and the Juhn-Murphy-Pierce decompositions, the results are also presented cor- recting for nonrandom selection of women into the labor force. Nonrandom selection of women is a very important problem in measuring the gender pay gap. Chapter 7 also includes a description of the nature of the problem and econometricians’ proposed solutions. These econometric models were not developed in isolation. They were the empirical response to the economic theories of discrimination that were proposed in the 1960s and 1970s. Although still included in all undergradu- ate textbooks, these theories have become increasingly obsolete, as more sophisticated theoretical and empirical models have emerged. Nevertheless, a brief summary of these concepts and theories can be found in chapter 7 to give users who are not that familiar with economics further context and understanding on why and how the Oaxaca-Blinder decomposition originated. Concepts Decomposition methods are applied to several diverse topics. In particu- lar, decompositions are widely used in the area of gender gaps in pay. Decompositions can be grouped in several ways according to different characteristics: (a) parametric versus nonparametric, (b) mean versus other moments of the earnings distribution (for example, variance or quantiles), (c) uncorrected versus corrected for nonrandom selection of women in the labor force, and (d) over groups (for example, men and women) versus over time. This section reviews the methods used by ADePT Gender and provides some discussion of the advantages and dis- advantages of these methods compared with others. The user interested in knowing more or in working with more sophisticated methods can consult Fortin, Lemieux, and Firpo (2011). Assume that there are only two types of workers—men and women. Male workers’ average wage is w M , and female workers’ average wage is w F . 178 Chapter 6: How to Interpret the Results of Labor Market Analysis The simplest way to measure gender inequality is to compute the difference in wages, usually called the raw gender wage gap: w = wM − wF . (6.1) This definition of inequality is informative but subject to flaws, as many factors beyond discrimination can influence the raw gender gap in wages, some of which result from choices and others from constraints. Perhaps the most critical is the difference in labor market skills. Those with profes- sional degrees earn more; if one of the two groups has more members with professional degrees, the raw gender wage gap would capture a “pure inequality” or discrimination effect as well as the returns to having a profes- sional degree. If more women have professional degrees than men, the raw gender gap would underestimate the pure gender inequality or discrimina- tion effect. Instead, if more men than women have professional degrees, then the raw gender gap would overestimate a discrimination effect. Therefore, a good measure of gender inequality in pay should take into account differences in characteristics that affect labor productivity. The adjustment is conducted by constructing counterfactual earnings of how much members of a group would earn if they shared one another’s characteristics. The seminal papers by Oaxaca (1973) and Blinder (1973) developed decomposition methods in labor economics. The linear regression is the basic building block of the Oaxaca-Blinder decomposition. This methodol- ogy explains differences in mean outcomes. As a result of the dramatic increase in wage inequality that has taken place in the United States since the late 1970s, decomposition methods have been further improved with the objective of analyzing distributional parameters other than the mean, such as the variance or the quantiles of the outcome distribution. Tools from the program evaluation literature have also been applied to extend the Oaxaca- Blinder setup. Contributions in this domain have been to clarify the assump- tions underlying the Oaxaca-Blinder framework, to introduce alternative estimators for elements of the decomposition, and to formally derive formal statistical properties of decomposition terms. Table 6.4 summarizes the main decomposition methodologies and their advantages and disadvantages. Oaxaca and Blinder The papers by Oaxaca (1973) and Blinder (1973) are among the most heav- ily cited in labor economics as well as in other fields of economics. The basic 179 Analyzing Gender in Labor Markets: Part III Table 6.4: Most Commonly Used Decomposition Methodologies Decomposition In ADePT methodology Gender Description Advantages Disadvantages Oaxaca (1973) and Yes Decomposes the gender Allows estimating the May hide different results for Blinder (1973) gap in pay in a effect of each variable; other groups that are not composition and wage simple to compute and represented by the mean; structure effect measured interpret attention needs to be paid to at the mean interpretation depending on how the baseline category is defined Juhn, Murphy, and Yes Decomposes the gender Provides an idea of the Does not allow estimating the Pierce (1991, 1993) gap in pay in a variation of the effect of each variable composition and wage composition and wage structure effect measured structure effect at different at the different percentiles points of the distribution or for the variance DiNardo, Fortin, and No Decomposes the gender Allows the analyst to Is complex to estimate and Lemieux (1996) gap in pay in a series of examine the whole requires a large number of marginal distributions distribution of wages observations for the semiparametric estimation, and results are path dependent;a works better with categorical variables, ideally binary variables; does not allow for selection correction Ñopo (2008) Yes Estimates the effect from Estimates the results after Requires a large number of the lack of overlapping controlling for lack of observations for the support (for both men overlapping support; semiparametric estimation, and and women) and a relatively easy to interpret results are path dependent; composition and wage Does not allow for selection structure effect correction Fortin, Lemieux, and No Estimates the Allows the analyst to Is complex to estimate and Firpo (2011) composition and wage examine the whole requires a large number of structure effect for any distribution of wages and observations for the statistics, including the to estimate the semiparametric estimation; percentiles contribution of each does not allow for selection covariate correction Note: a. Path dependent means that the results vary with the order in which the marginal distributions are estimated. assumption of the Oaxaca-Blinder decomposition is that earnings (measured with the logarithm of the hourly rate of pay) can be estimated by a linear regression that yields M M M M ln w M = α + ρ SM + β 1 XM + β 2 XM 2 , F F F F (6.2) ln w F = α + ρ SF + β 1 XF + β 2 XF 2 , where ln w M is the average of the log of hourly rate of men, SM is average number of years of schooling of men, XM is the average years of experience, M and the coefficients estimated in the linear regression are α for 180 Chapter 6: How to Interpret the Results of Labor Market Analysis M M the constant, and ρ and β 1 for each of the corresponding variables. The separate regression for women is indicated with the subscript F or superscript μ F. The estimated average gender wage gap, Δ O , can be written as μ Δ O = lnw M − lnw F , μ ( Δ O = (α M − α F ) + SF ρ − ρ M F )+ρ M ( SM − SF ) . (6.3) μ μ Δ S (Unexplained) Δ X (Explained) For expositional purposes, the contribution of experience and its squared value—X variable in equation (6.2)—are dropped; however, it can be easily added by repeating the terms for education but applying them to experience. The components of equation (6.3) are as follows: U • Δ O , the overall or raw gender wage gap; μ • Δ X , the composition effect, also called the explained component or part; and μ • Δ S , the wage structure, which includes the average differences between men and women, also called the unexplained component. The results from the decomposition are usually presented following these μ μ μ three components: Δ O = Δ S + Δ X , or in full detail—usually called detailed decomposition—where the contribution of each covariate or variable is shown. The intuition behind the Oaxaca-Blinder decomposition can be explained using a graph with a simplified version of the Mincer equation. Assume that earnings depend solely on the number of years of education: workers with more years of education earn more. Women are paid less than men when neither of them have any education, and women also have lower payoffs for each additional year of education than men. Assume men have on average more years of education than women, ( SF < SM ) . The raw gen- der wage gap is ( w M − w F ) . However, it has two components: one related to the differential treatment of women with respect to men (unexplained or wage structure component) and the other related to the gender difference in years of schooling (explained or composition component). If women ∗ were paid like men, they would earn w F ; under discrimination toward women, this is usually larger than w F and smaller than w M . Therefore, (wF − w F ) is the wage structure effect, and ( w M − w F ) is the composition ∗ ∗ 181 Analyzing Gender in Labor Markets: Part III effect due to differences in labor market skills. Moreover, the wage struc- ture effect can be divided into its two components: one coming from differ- ( ) ences in the returns to schooling SF ρ M − ρ F and the other coming from “everything else”—that is, the constant (aM − aF). A very useful graphical representation of the Oaxaca and Blinder decomposition can be found in Borjas (2005).9 The accuracy of the discrimination measurement from the Oaxaca- Blinder decomposition depends on the validity of the estimation of the wage equations. If variables that are omitted bias the estimation of the coefficients of the included variables, and for which differences exist between men and women, the estimated wage structure (unexplained) effect will be biased. For example, consider the case in which the analyst leaves the experience variable out of the wage equation. Assume that men and women have the same number of years of education, but women accumulate fewer years of experience because of employment interruptions for child rearing. A wage gap exists between men and women, and it would be incorrect to assume that the entire wage gap is due to discrimination. In reality, it is partly explained by differences in other unmeasured skills, such as experience.10 Thus, one of the disadvantages of measuring discrimination—usually identified with the unexplained component—using decomposition tech- niques is that it can always be argued that part of the discrimination effect is actually capturing the gender difference in some unobserved (that is, excluded in the regression) labor market skills. Even when working with rich datasets, there will always be some other unobserved characteristic— such as ability, effort, motivation, or responsibility—that can be used to argue that differences between men’s skills and women’s skills exist. As a result, modern labor economics more often refers to this factor as the wage structure component as opposed to a discrimination effect. In response to this criticism, it can be argued that any other gender dif- ference in unobserved variables is due to the influence of formal and infor- mal institutions on men and women. If women exert less effort at a job than men, it could be due to the fact that they are rewarded less or have too many other responsibilities in the household. In other words, some other factors are responsible for the difference. Very likely, the reality lies somewhere in the middle of these two points of view. In summary, the Oaxaca-Blinder decomposition can be a powerful tool for understanding the factors behind gender inequality. Analysts must be careful about how they interpret the unexplained coefficient—depending 182 Chapter 6: How to Interpret the Results of Labor Market Analysis on the set of included variables in the Mincer wage equation—and they must be familiar with the country’s cultural and social norms so as to under- stand the possible gender differences in the omitted unobserved variables. It is critical to correctly interpret the unexplained component. Switching Reference Groups and Pooled Decomposition So far, the discussion has focused on using men as a reference group by add- M ing and subtracting ρ SF to obtain the rearrangement in equation (6.3). This approach amounts to interpreting the “explained” component as the value of the difference in skills (or endowments as more generally discussed in the introduction) between men and women, as valuated by the men’s earnings equation; and the “unexplained” component as the difference between how the men’s and women’s earnings equations value the average skills of women. Two points are worth noting. First, women can also be used as a reference group—which would yield equation (6.4) with some adjustment in the interpretation of the results—to mirror what has been discussed so far using men as the reference point. Oaxaca (1973) refers to this as the “index number problem”—the decomposition is sensitive to the group chosen as a reference. Δ O = (α M − α F ) + SM ρ − ρ μ ( M F )+ρ F ( SM − SF ) . (6.4) μ μ Δ S (Unexplained ) Δ X (Explained) Second, some may implicitly assume that discrimination goes only one way (targeting women if men are the reference, and vice versa). A more general version of the decomposition assumes that underevaluation of one group is likely to go hand in hand with overevaluation of the other. To address this assumption, and to separate the effects of positive and negative discrimination in the unexplained component, a “nondiscrimi- ∗ ∗ natory” vector of parameters (composed of α and ρ ) could be intro- duced as the reference point. This vector of parameters will lie in the middle of the male and female reference point. ( ) ( ) ( Δ O = α M − α ∗ + SM ρ − ρ + (α ∗ − α F ) + SM ρ − ρ + ρ ∗ ( SM − SF ) . μ M ∗ ∗ (6.5) F ) μ μ Δ S (Unexplained) Δ X (Explained) 183 Analyzing Gender in Labor Markets: Part III Different methodologies have been proposed to estimate such nondis- criminatory reference points. Possible solutions include using an average of the group-specific coefficients (simple or weighted by population size) or the coefficients from a pooled regression over both groups—Oaxaca-Ransom decomposition, also called a pooled decomposition (Oaxaca and Ransom 1994). This approach is a way to overcome the “index number problem” and to focus on positive and negative discrimination. Detailed Decomposition Because the linear regression is the basic building block of the Oaxaca- Blinder decomposition, each of its components is in fact a sum over the effects of individual variables. If instead of limiting the discussion to educa- tion, we include all variables in the Mincer earnings equation, an equivalent full version of equation (6.5) is Δ O = (α M − α F ) + SF ρ − ρ μ ( M F ) + X (β F M 1 F ) ( M F − β 1 + XF β 2 − β 2 + 2 ) (6.6) (SM − SF ) + β 1 (XM − XF ) + β 2 (XM ) M M M ρ 2 − XF 2 . The main advantage of the detailed decomposition is that it allows us to further identify each factor’s contribution to the gender gap. However, users must keep two caveats in mind. First, it can be shown that the unexplained component of the decomposition is sensitive to arbitrary scaling decisions if the variables do not have natural zero points (that is, are not so-called ratio variables, for which the level “zero” is a meaningful value and indicates absence of the property being measured). In this case, simply rescaling the variable affects the relative magnitude of the part of the unexplained com- ponent captured in the constants and the part due to the coefficients of the variable in question.11 Second, the unexplained portion of the detailed decomposition is also sensitive to the choice of the excluded category when the regressions include dummy variables to capture the effect of categorical predictors. This arbitrary choice is again reflected in a trade-off between the compo- nents, as changes in the base category alter the (slope) coefficients, which determine the wage structure effect. The solution to the problem— proposed by Gardeazabal and Ugidos (2004) and Yun (2005)—is to restrict the coefficients of the single categories to summing to zero, that is, 184 Chapter 6: How to Interpret the Results of Labor Market Analysis to express coefficients as deviations from the overall mean. This approach is equal to averaging the results obtained from a series of decompositions in which categories are excluded one by one (Yun 2005). However, this solution would cover the differences in means across countries if working with cross-country data. Threefold Decomposition Equation (6.5) is not the only possible decomposition that separates the role of observable characteristics from that of returns to labor market skills. An alternative decomposition, yielding three terms instead of two, can be written as ( Δ O = (α M − α F ) + SF ρ − ρ μ M F )+ ρ F (SM − SF ) + (SM − SF ) ( ρ M −ρ . F ) (6.7) μ μ μ Δ C (Coefficients) Δ E (Endowments) Δ I (Interaction) The effect of different predictors, the “endowments” component E, can be viewed as the expected change in the mean raw gender wage gap if they had men’s characteristics; the effect of a different coefficient, C, can be seen as the expected change of the gender wage gap if women had unchanged characteristics but were paid as men. The last addendum is an interaction term that captures the differences in mean characteristics, and coefficients exist simultaneously. On the one hand, the interpretation of this version of the decomposition might be, in a way, more convenient, given that in both the endowments and coefficients components, the reference group is the same (in this case, women). On the other hand to equate the right-hand side and left-hand side, the interaction term must be added, and its interpre- tation in economic terms is not straightforward, as noted in Blinder (1973). Oaxaca-Blinder with Selection Correction It is possible to incorporate a selection correction in the Oaxaca-Blinder decomposition by modifying the underlying regressions accordingly. The selec- tion correction is needed when the analyst has reasons to believe that women who work are different from women who do not work, with regard to their unobservable skills. As a result, the wages of working women are not good predictors of the potential wages of nonworking women, if they were employed. In other words, working women are not a random selection of all women of 185 Analyzing Gender in Labor Markets: Part III working age. The correction was introduced by Heckman (1974) and later applied to the Oaxaca-Blinder decomposition. Simply, the parameters of the earnings equations can be estimated through a Heckit12 procedure for both groups (or more commonly, for only women, since men’s labor force participa- tion is usually high, random, or both). The basis of the decomposition will then be the Heckman-corrected equivalent of equation (6.3), which includes a regressor, the Mills ratio:13 μ Δ O = ln w M − ln w F =⎡ ⎤ M M M M 2 ⎣α + ρ SM + β 1 XM + β 2 XM + λ Mγ M ⎦ (6.8) −⎡ ⎤ F F F F 2 ⎣α + ρ SF + β 1 XF + β 2 XF + λ Fγ F ⎦ . As Neuman and Oaxaca (2004) note, it is not immediately obvious how the added Mills ratio should be regarded in the overall decomposition scheme. They discuss several alternative ways of allocating it to the decom- position terms: as a component of the explained part, as a component of the unexplained part, or as a stand-alone term. However, a common way to work around this issue is simply to net out the “selection” term from the left-hand side, thus leaving the familiar decomposition terms on the right-hand side: μ ΔO − ⎡ ⎣ λ Mγ M − λ Fγ F ⎤ M F M ( F ) ⎦ = (α − α ) + SM ρ − ρ + ρ ( SM − SF ) , F (6.9) where S is the adjusted difference in outcomes. The resulting decomposition is not conceptually different from the nonselection-corrected version seen above; however, one needs to use some care in interpreting it. What we are left with on the right-hand side are the “true” (that is, random sample) popu- lation parameters that describe the effect of regressors on the underlying, uncensored dependent variable. Accordingly, the left-hand side is no longer equal to the observed difference overall gender wage gap, but to the selection-corrected difference—the difference that would be observed if our selected sample were indeed a random draw from the population. The inter- pretation of the explained and unexplained terms is the usual (and comes with the usual caveats), but the quantity being decomposed now incorporates the effect of selection (that is, it is bigger if there is a larger positive selection into the labor force for women than for men, as it is usually the case). Several decomposition methodologies seek to shed light on aspects that are left out of the Oaxaca-Blinder decomposition framework. Below 186 Chapter 6: How to Interpret the Results of Labor Market Analysis are some of these extensions, which are part of the ADePT Gender out- put. We first discuss issues they are designed to tackle and then outline the expression of the decomposition, highlighting the differences with respect to Oaxaca-Blinder. Juhn, Murphy, and Pierce The value that the Juhn-Murphy-Pierce methodology adds to the Oaxaca- Blinder decomposition discussed above is that it allows the analyst to sepa- rate the effect of skills (or endowments) from the effect of changes in returns to observable characteristics and the effect of unobservable characteristics. The methodology takes the name of the authors who introduced it in 1991 when analyzing reduction in the wage gap between blacks and whites in the United States (Juhn, Murphy, and Pierce 1991) and changes in wage inequality over time in the United States (Juhn, Murphy, and Pierce 1993). From a methodological point of view, it represents a step up toward meth- odologies that move away from the mean by looking at the overall distribu- tion of a variable. From a conceptual point of view, it allows one to measure the price of skills that the analyst cannot observe. Moreover, this methodol- ogy has also been applied to perform double decompositions—or the change over time of the wage decomposition across groups, such as men and women or blacks and whites—providing an additional advancement with respect to the seminal Oaxaca-Blinder work. For simplicity, this subsection works with a reduced version of the earn- ings equation introduced above that includes only the constant and the education variable. The notation is compacted too, where Xgi indicates both the constant and the only regressor (denotes a matrix), and β g indicates both β 0 and ρ1 (denotes a vector): g g ln w gi = β 0 g + ρ1 g Sgi + ε gi = β g X gi + ε gi , (6.10) where g = M, F, that is male and female. The error term in the equation has no role in the Oaxaca-Blinder framework, since wage regressions are estimated by least squares and then averaged over the sample, which implies that residu- als equal zero (it is an algebraic property of ordinary least squares). But the error term can be further exploited, as it has information about the unobserved skills. Take a group of workers who have the same education and experience. If observable characteristics were their only determinant of productivity, they 187 Analyzing Gender in Labor Markets: Part III Figure 6.4: Illustration of Unobserved Skills Men’s earnings Dollars High function unobserved skills WM WF * Women’s earnings aM function Low WF unobserved skills aF S Education would all earn the same wage per hour. However, that is not the reality. It can be said that those workers with higher wages—all other things constant— have better unobserved labor market skills. The unobserved skills can then be thought of as the position in the distribution of the residuals of the earnings equation, or the percentile in the distribution. Figure 6.4 illustrates the logic: take level of education S —the expected return for that level of education is reflected by the line that shows the mean fitted equation for men and women. Men with high unobserved skills will earn more that the value of the mean. Similar to the Oaxaca-Blinder decomposition, the next step is to work out the resulting summary statistics of the regressions to express the raw wage gap in terms of separate components. Appendix D includes the trans- formation step by step. The Juhn-Murphy-Pierce decomposition accom- plishes this by summarizing observed and counterfactual wages at each percentile and then arranging the gender differences among the terms specified above in the following way: w Mp − w Fp = ⎡ ⎣( w Mp − w Fp ) − ( w Mp − w Fp ) ⎤ ⎦+⎡ ⎣w Mp − w Fp ⎤ X X obs obs X X ⎦ (6.11) ⎣( w Mp − w Fp ) − ( w Mp − w Fp ) ⎤ +⎡ unobs unobs obs obs ⎦, 188 Chapter 6: How to Interpret the Results of Labor Market Analysis μ μ μ μ Δ O = Δ Xobs + Δ Xunobs + Δ S , (6.12) μ where Δ Xp = ⎣ ⎡w Mp X − w Fp X ⎤ ⎦ is the part of the raw wage gap at percentile p due to gender differences in observed characteristics, in this case education (the ⎣( w Mp − w Fp ) − ( w Mp − w Fp ) ⎦ is the part of the μ “endowments” effect); Δ Rp = ⎡ obs obs X X ⎤ raw wage gap at percentile p due to gender differences in the returns to educa- ⎡( w Mp ) − (wMp )⎤ μ tion (the “prices” effect); and Δ Rp = ⎣ unobs − w Fp unobs obs − w Fp obs ⎦ is the part of the raw wage gap at percentile p due to unobserved characteristics, as well as their returns (the “unobservables” effect). Thus, the innovation of the Juhn- Murphy-Pierce decomposition with respect to the basic Oaxaca-Blinder set- ting is the ability to measure gender gaps at different positions in the distribution of residuals—or percentiles—and the explicit separation of the effect of unobservable characteristics and their “prices” from the effect of whatever we can observe—in this example, education and its returns. As described in the next section, ADePT Gender computes the decom- position at the 5th, 10th, 25th, 50th, 75th, 90th, and 95th percentiles. Ñopo Another more recent trend in the literature on decompositions uses meth- odologies from program evaluation to generate the counterfactuals. This approach has generated some controversy among academics who believe the spirit of program evaluation cannot be applied to analyses of gender or race. Some purists argue that the core principle of program evaluation relies on the fact that a program can be delivered to a treatment and a control group. However, when analyzing gender differences, we cannot deliver the program “male” to a woman; in other words, no interven- tion can convert a woman into a man from a labor market perspective. Thus, differences that cannot be addressed will always remain between the two groups. In this strand of literature, Ñopo (2008) identified that a potential bias can arise when using the Oaxaca-Blinder decomposition if no common sup- port exists in the distribution of the observable individual characteristics— the support of a random variable being the set of values that has a positive probability of being observed. In simpler terms, estimates of differences between the returns to schooling might be biased if there are no examples of men having levels of education as low as women or no women having 189 Analyzing Gender in Labor Markets: Part III levels of education as high as men. The reason is that the comparison of the returns to education of men and women will be based on an implicit assumption: the linear estimators of returns to education are also valid for the out-of-support individual characteristics for which they were estimated. The empirical evidence suggests that such an assumption of a common support tends to overestimate the component of the gap attributable to differences in the prices of the skills. Figure 6.5 illustrates the possible bias arising from the lack of common support in the Oaxaca-Blinder linear decomposition. The light gray points on the x-axis indicate the levels of education observed for women and are used to estimate the returns for women; the dark gray points indicate the levels of education observed for men and are used to estimate men’s returns. Because the observed samples do not have women with high levels of education, the predicted return for high levels of education for women is underestimated, Figure 6.5: Illustration of Bias from a Lack of Common Support Men’s earnings dollars function if estimated with common support W M Men’s earnings function W * F Women’s earnings function if estimated a with common M support W F Women’s earnings a function F S S education F M Note: The dashed lines indicate the out-of-support prediction based on a linear regression model. Although in this graph the out-of-support predictions increase the role of returns (that is, out-of- support regression from men is steeper than the support regression and vice versa for women), the results could also go in the opposite direction. 190 Chapter 6: How to Interpret the Results of Labor Market Analysis indicated by the dashed light gray line, whereas the opposite is true for men, indicated by the dashed dark gray line. Ñopo (2008) developed a matching technique from the impact evalu- ation literature to solve the bias explained above. This technique refines the Oaxaca-Blinder decomposition terms by (a) changing the meaning of the “unexplained” component, which is estimated only using the sample of men and women with characteristics in the common support; and (b) separating the “explained” component into two parts: one due to a differ- ent distribution of characteristics within the common support, and the other due to the fact that some men have no comparable female counterparts and vice versa, that is, the effect of being outside the common support. The methodology does not require one to estimate an earnings equation; hence, there is no need to estimate validity assumptions for the out-of- support levels of skills. The methodology consists of finding an opposite- sex match for each man and woman in the sample. A match is simply a person (or group of persons) with the same observable characteristics. Thus, the sample can be partitioned into four groups: (a) men with a match, (b) women with a match, (c) men without a match, and (d) women without a match. Then, using the wages of each of these groups and using the wages computed for the match, the following decomposition is proposed: ΔO = (ΔM + ΔF + ΔX) + ΔU, (6.13) where the overall gender wage gap is now decomposed as follows: • The difference in wages between men with a match and men without a match, Δ M: This component accounts for the part of the gap that would disappear if there were no males with characteristics that remain unmatched by the females in the sample—or if these unmatched males were paid on average the same as the matched ones (basically, characteristics would need to have no effect on wages outside the support). • The difference in wages between women with a match and women without a match, Δ F: This component is the part of the gap that would disap- pear if there were no females with characteristics that remain unmatched by the males in the sample, or if unmatched females were paid, on average, the same as matched females. 191 Analyzing Gender in Labor Markets: Part III • The two components similar to the characteristics, ΔX, and wage structure effects of the Oaxaca-Blinder decomposition, computed for the men and women in the common support, ΔU: The interpretation of these two components is essentially the same as for the Oaxaca-Blinder decom- position, but one needs to keep in mind that they are estimated on a sample that is different from the one Oaxaca-Blinder would use—the sample of men and women that are matched on the basis of observed characteristics. In other words, given that no earnings equation was estimated, the skills and wage structure effects are estimated compar- ing the counterfactual distributions of each sex over the common support of characteristics. ΔM, ΔF, and ΔX can all be ascribed to differences in observed character- istics between men and women, so in a sense they all amount to an “explained” component of the wage gap. The term ΔU is the “unexplained” component, the term attributable to differences in unexplained characteris- tics of matched individuals, as well as to the existence of discrimination. Interpreting the Results Oaxaca-Blinder Table 6.5 and figure 6.6 show the results of Oaxaca-Blinder decompositions for Nepal using the 2010–11 household survey. These results do not correct for selection. The decomposition figures summarize the main messages of the decomposition results. For example, ADePT figure 13a (figure 6.6) shows that of the 0.47 log point difference between male and female wages, 0.19 log points are explained by observable characteristics using the human capital model (education, age, and residence) and women as the reference group. If men were used as the reference group, the explained component results in 0.13 log points out of the 0.47. The user may find it useful to describe the results in percentages as opposed to log points. The tables provide a more detailed description of the decomposition results. The first column of ADePT table 13a (table 6.5) shows the estimates of the decomposition for the human capital model using females as the refer- ence group for the counterfactual, and the second column uses men. The first row indicates the raw gender wage gap, that is, the simple difference between the average of the log wage of men and women. The average log 192 Chapter 6: How to Interpret the Results of Labor Market Analysis Table 6.5: ADePT Gender Table 13a, Nepal 2010–11 Table 13a: Oaxaca-Blinder Decomposition (Human Capital Model): Wage Earners Female Male Average Gap 0.47 0.47 0.47 Explained Total 0.19 0.13 0.16 Age 0.07 0.05 0.06 Education 0.15 0.10 0.13 Region and urban residency −0.03 −0.02 −0.02 Unexplained Total 0.28 0.34 0.31 Age −0.14 −0.12 −0.13 Education −0.30 −0.25 −0.28 Region and urban residency −0.14 −0.16 −0.15 Constant 0.86 0.86 0.86 In percentages Gap 100.00 100.00 100.00 Explained Total 39.71 27.91 33.81 Age 14.46 9.82 12.14 Education 32.31 21.67 26.99 Region and urban residency −7.06 −3.58 −5.32 Unexplained Total 60.29 72.09 66.19 Age −29.80 −25.15 −27.48 Education −64.22 −53.58 −58.90 Region and urban residency −29.76 −33.25 −31.51 Constant 184.08 184.08 184.08 Source: ADePT Gender using Nepal 2012. earnings of men are higher than those of women by 0.47 log points. The explained component or composition effect is 0.13 log points and the unex- plained component or wage structure effect is 0.34 log points, using men as the reference. Each of the components is further decomposed according to the detailed version of the decomposition by highlighting the contribution of each variable or group of variables to explain the gender gap in pay. The explained component of the decomposition is interpreted as the additional amount of money women would earn—on average (and in log points)—if they were to have, as a group, the same average characteristics that prevail among men. This reason is mostly due to education, implying that if women were as educated as men, the gender gap in pay would reduce in 0.1 log points—or by 20 percent (the bottom panel of the table presents the num- bers in percentages). 193 Analyzing Gender in Labor Markets: Part III Figure 6.6: Oaxaca-Blinder Decomposition, ADePT Figure 13a, Nepal 2010–11 Figure 13a: Oaxaca-Blinder Decomposition (Human Capital Model): Wage Earners 50 40 28 30 34 Percent 20 10 19 13 0 Female Male Unexplained Explained Source: ADePT Gender using Nepal 2012. Turning to the unexplained component, the difference between men’s and women’s average log wages is positive and large. That factor indicates that both discrimination and unobservables may play a very important role in the gender gap in pay. Women’s estimated returns to experience (proxied by age), education, and place of residence are higher than those of men, but the effect of the constant (which captures the effect of “group membership” and can be seen in this context as a catchall term for unobservables) offsets that effect. The constant captures the pay gap for the base category, which in this case corresponds to the youngest, without any education, and living in rural areas. Moreover, part of the wage structure or discrimination component can reveal differences in unobserved skills. Men and women have very different employment histories. It should be noted that the magnitude of the unex- plained component could also reflect the fact that the parsimonious human capital model may not do a good job of explaining more of the gender gap in pay; important unobserved productivity characteristics could play a rele- vant role. 194 Chapter 6: How to Interpret the Results of Labor Market Analysis Table 6.6: ADePT Gender Table 13b, Nepal 2010–12 Table 13b: Oaxaca-Blinder Decomposition (Human Capital Model with Selection Correction): Wage Earners Female Male Average Gap −0.23 −0.23 −0.23 Explained Total 0.11 0.11 0.11 Age 0.06 0.07 0.06 Education 0.10 0.06 0.08 Region and urban residency −0.04 −0.02 −0.03 Unexplained Total −0.34 −0.34 −0.34 Age 0.13 0.12 0.13 Education −0.06 −0.02 −0.04 Region and urban residency −0.21 −0.23 −0.22 Constant −0.21 −0.21 −0.21 In percentages Gap 100.00 100.00 100.00 Explained Total −49.44 −49.06 −49.25 Age −25.54 −29.54 −27.54 Education −41.49 −26.88 −34.19 Region and urban residency 17.60 7.35 12.48 Unexplained Total 149.44 149.06 149.25 Age −57.62 −53.63 −55.62 Education 24.99 10.38 17.68 Region and urban residency 89.53 99.77 94.65 Constant 92.54 92.54 92.54 Source: ADePT Gender using Nepal 2012. Note: Unweighted results. Weights not supported for the two-step model. ADePT Gender also produces richer versions of this model. ADePT Gender table 13c includes a richer specification that includes occupation and industry controls (full model), and ADePT tables 13b and 13d correct for nonrandom selection of women in the labor market. For the full model, the results are interpreted in the same way as for the simple Oaxaca-Blinder decomposition. The occupation and industry variables capture not only the differences in the payoffs of the different jobs (due to supply and demand equilibrium effects) but also the fact that women self-select–by choice or given constraints—into different jobs. For the selection-corrected decompo- sition, the effect of the additional variable—the Mills ratio—can combine in different ways into a pure selection effect, or in the wage structure. The user with large and significant selection effects interested in further explor- ing this aspect of the results can consult Neuman and Oaxaca (2004). 195 Analyzing Gender in Labor Markets: Part III Juhn-Murphy-Pierce ADePT table 16a (table 6.7) and ADePT figure 16a (figure 6.7) report results from the Juhn-Murphy-Pierce decomposition performed on 2008 data from Panama. The first column reports the difference between the pth percentiles of the male and female wage distribution, as it arises in the sample (the observed wages). The first column shows the raw gender wage gap at each percentile—that is, the simple differences in the aver- age log wage between men and women. As expected, the raw gap evalu- ated at the median is close to that of the mean calculated for the Oaxaca-Blinder decomposition (0.42 versus 0.47 log points, respec- tively). The gap at the median tends to be lower than the gap at the mean, as some outlier value always raises the mean average. The gap is larger at the bottom of the wage distribution curve, which suggests the existence of “sticky floor” effects—it then narrows at higher percentiles. The term sticky floor refers to a situation when women cannot start climbing the job ladder: they are stuck in low-remunerated jobs com- pared with similar men. In the case of Panama, no “glass ceiling” effect is observed. The term glass ceiling refers to the situation when women have the same high skills as men but cannot achieve high-level posi- tions, such as top executives or government officials. The next columns of table 6.7 allow the user to pin down which factors are more relevant for determining these trends. As with the Oaxaca-Blinder presentation of the results, ADePT figure 16a allows the user to quickly visualize the major patterns by looking at how the different components change along the wage distribution, whereas the table Table 6.7: ADePT Gender Table 16a, Panama 2008 Female Male Gap Endowments Price Unobservables Endowments Price Unobservables Percentile p5 0.319 −0.105 0.345 0.078 −0.100 0.338 0.081 p10 0.277 −0.107 0.323 0.061 −0.094 0.317 0.055 p25 0.105 −0.238 0.336 0.007 −0.189 0.298 −0.004 p50 0.048 −0.230 0.297 −0.020 −0.197 0.251 −0.006 p75 −0.106 −0.325 0.262 −0.043 −0.289 0.204 −0.021 p90 −0.034 −0.214 0.223 −0.043 −0.191 0.186 −0.029 p95 0.078 −0.113 0.201 −0.010 −0.103 0.175 0.006 Source: ADePT Gender using Panama 2008. Note: p = percentile. 196 Chapter 6: How to Interpret the Results of Labor Market Analysis presents the exact magnitude of each component and how the results vary if the group of reference is changed (men versus women). First, the “endow- ments” component shows the difference between the quintiles of the male distribution of wages and the counterfactual distribution, the one that would arise if characteristics were distributed as in the female group (see appendix D for mathematical derivation). It can be seen that the differences are consis- tently positive—women’s observable characteristics are such that if returns to them and the distribution of unobservables were identical to those prevailing among men, the gender gap would be smaller. The size of this effect varies along the wage distribution, with very large effects at the very bottom (0.281 log points at the 5th percentile). The second column shows the part of the gap that is explained by differences in the returns to labor market skills—the result of women being paid less than men for their observable characteristics. This com- ponent of the gap that is attributable to different labor market prices of observ- able characteristics is sizable and positive across quintiles. The fact that the Figure 6.7: ADePT Gender Figure 16a, Nepal 2010–11 Figure 16a: Decomposition of Earnings by Percentile (Human Capital Model): Wage Earners 120 100 32 80 60 14 Percent 4 42 40 43 27 21 38 19 20 9 28 17 19 17 17 11 12 0 –4 –1 –9 –16 –20 p5 p10 p25 p50 p75 p90 p95 Unobservables Price Endowments Source: ADePT Gender using Nepal 2012. Note: p = percentile. 197 Analyzing Gender in Labor Markets: Part III price effect is smaller at higher quintiles indicates that women are more likely to be remunerated for their observable characteristics than men are. The third column displays what changes when we finally compare the male earnings distribution to the female earnings distribution—a change that is attributable to unobserved characteristics and their prices. This component also decreases from the bottom to the top of the wage distribution, and it even becomes negative for the top half of the distribution. A negative coefficient means that the gender gap in pay would be even larger if women were paid the same as men. This result might seem odd at first, but it can be explained by a nonrandom selection effect. The problem of selection of women does not apply only to participation in the labor market; it can also apply to the occu- pation or sector in which they work, or how likely they are to reach top posi- tions compared with men. Thus, one possible explanation for this result is that high-skilled women that work are very likely to reach top positions, whereas top positions for men are occupied by high-skilled and low-skilled men. Thus, on average, women at the top would have better unobservable characteristics than men at the top, and that is why they are paid relatively better than men. A story like this would undoubtedly require additional research. The analyst who encounters this type of result should look further into who these women are, how large the group is, and what types of jobs they do. Ñopo Finally, this section switches to the most recently developed methodology that ADePT Gender uses. As stated above, this is a nonparametric method that builds on the program evaluation literature. Its main contribution is that it takes into account the fact that not all working women have the same characteristics as working men, and it measures how much of the gap can be attributed to that fact (called the lack of common support). Table 6.8 corresponds to ADePT table 15a and shows the output of the Ñopo decomposition for Nepal in 2010–11. Each row shows the control that was added in the regression. In each successive row, a variable (or set of variables) is added. This stepwise practice responds to the fact that the methodology is path dependent, as it works with marginal distributions. Thus, by comparing rows, it can be observed how much more is explained by the additional variable (or set of variables). The contribution of each addi- tional variable to explain the gap depends on what was explained before its addition and the correlation between the variables in each consecutive step. 198 Table 6.8: ADePT Gender Table 15a, Nepal 2010–11 Male Female % % % % D DO DM DF DX male female D DO DM DF DX male female Age 0.473 0.432 0.041 1.000 1.000 −0.473 −0.462 −0.011 1.000 1.000 Age, urban 0.473 0.456 0.017 1.000 1.000 −0.473 −0.519 0.046 1.000 1.000 Age, urban, region 0.473 0.472 0.001 0.000 0.997 1.000 −0.473 −0.506 −0.001 0.034 1.000 0.997 Age, urban, region, education 0.473 0.344 0.080 0.005 0.044 0.820 0.974 −0.473 −0.285 −0.005 −0.080 −0.103 0.974 0.820 Age, urban, region, education, sector 0.473 0.321 0.027 0.047 0.078 0.488 0.802 −0.473 −0.168 −0.047 −0.027 −0.231 0.802 0.488 Source: ADePT Gender using Nepal 2012. Note: D = overall or raw gender wage gap; DF = difference in wages between women with a match and women without a match; that is, difference from being out of support from women; DM = difference from being out of support for men; DO = unexplained component; DX = explained component. 199 Analyzing Gender in Labor Markets: Part III The change in the coefficient previously included when more variables are added responds to the correlation among the previously and newly included variables. The first column of data shows the raw gender gap in pay, which should be the same value presented in the first row of the Oaxaca-Blinder table. The second column shows the part that is unexplained (D0) and the fifth column the part that is explained by differences in characteristics (DX). The two columns in the middle indicate the part of the gap that is explained by lack of common support: by having men who do not have women’s charac- teristics (DM) or women who do not have men’s characteristics (DF). For example, when looking at the fourth row (“Age, urban, region, education”), it can be seen that 17 percent of the gender wage gap (the result of dividing 0.08 by 0.473) is due to men’s average educational attainment that is higher than women’s. Only an additional 10 percent (the result of dividing 0.05 by 0.473) comes from having women with levels of education for which there are no comparable men. Overall, then, the part of the raw gender gap that is explained by lack of common support (0.080 + 0.005 = 0.085 log points) is very small. If these effects are large in magnitude, the user should examine them further by identifying those persons and characteristics that are pres- ent in only one group of men (or women). A natural next step, not reported in this version of ADePT Gender but that is straightforward to compute, is to further decompose the common sup- port using Oaxaca-Blinder. The user can do this by recalculating the Oaxaca- Blinder tables combined with an if-condition that excludes those observations for which there are men but no similar women, and vice versa.14 Notes 1. Chapter 3 of this book includes the definition of earnings and their relationships to such associated concepts as wages and income. 2. The term wage structure refers to the relationship of wages and different jobs, either within a firm or more generally in the labor market. 3. Although earnings are computed using wages, and hours of work are usually collected as part of the labor module of a labor force survey for wage workers, earnings for the self-employed are computed using vari- ables on revenues and costs (in some cases, benefits) from household enterprise modules in multitopic surveys. 200 Chapter 6: How to Interpret the Results of Labor Market Analysis 4. Notice that in the substitutions below, area A equals 1/2. 5. Another advantage of this family of indicators is that it is possible to decompose the measure of inequality into two components: inequality within groups and inequality between groups. 6. In this case, given that the log function is a monotonic function, the percentiles can be calculated over the distribution of wages or over the distribution of log wages. 7. The only exception is the entropy measure for c = 2. 8. Density distribution functions of the log of a variable have the property of being approximate to normal. 9. See figure 10.6 in chapter 10. 10. Biases could be even more complicated. Assume, for example, that the returns to education for men and women are the same, but the returns to education also increase with the number of years of experience. If men have more experience than women, and experience is not included in the wage equation, the estimated education coefficient will pick up the returns to education and part of the experience return. The unaware analyst may misinterpret the results and conclude that a gender gap exists in the returns to education. 11. For more details, consult Jann (2003). 12. The term Heckit is the informal way of referring to the Heckman two- step selection model, as the term is the combination of Heckman and probit (which is the model used in the first step of the two-step estima- tion procedure). 13. The term Mills ratio is the coefficient g that enters in the equation to describe the correction for nonrandom selection. Chapter 7 succinctly shows the derivation of the coefficient. 14. The output of the Ñopo decomposition provided in the econometric software Stata provides this level of detail. References Atkinson, Anthony B. 1970. “On the Measurement of Inequality.” Journal of Economic Theory 2 (3): 244–63. Autor, David H., Laurence F. Katz, and Melissa S. Kearny. 2008. “Trends in U.S. Wage Inequality: Revising the Revisionists.” Review of Economics and Statistics 90 (2): 300–323. 201 Analyzing Gender in Labor Markets: Part III Bedard, Kelly. 2001. “Human Capital versus Signaling Models: University Access and High School Dropouts.” Journal of Political Economy 109 (4): 749–75. Blau, Francine D., and Lawrence M. Kahn. 1997. “Swimming Upstream: Trends in the Gender Wage Differential in the 1980s.” Journal of Labor Economics 15 (1): 1–42. Blinder, Alan S. 1973. “Wage Discrimination: Reduced Form and Structural Estimates. Journal of Human Resources 8 (4): 436–55. Borjas, George. 2005. Labor Economics. 3rd ed. New York: McGraw-Hill. Cowell, Frank A. 2000. “Measurement of Inequality.” In Handbook of Income Distribution, vol. 1, edited by Anthony B. Atkinson and François Bourguignon, 87–166. Amsterdam and New York: Elsevier. DiNardo, John, Nicole M. Fortin, and Thomas Lemieux. 1996. “Labor Market Institutions and the Distribution of Wages, 1973–1992: A Semiparametric Approach.” Econometrica 64 (5): 1001–44. Fortin, Nicole, Thomas Lemieux, and Sergio Firpo. 2011. “Decomposition Methods in Economics.” Handbook of Labor Economics, vol. 4, edited by Orley Ashenfelter and David Card, 1–102. Amsterdam: Elsevier. Foster, James, Suman Seth, Michael Lokshin, and Zurab Sajaia. 2013. A Unified Approach to Measuring Poverty and Inequality: Theory and Practice. Washington, DC: World Bank. Gardeazabal, Javier, and Arantza Ugidos. 2004. “More on Identification in Detailed Wage Decompositions.” Review of Economics and Statistics 86 (4): 1034–36. Heckman, James. 1974. “Shadow Prices, Market Wages, and Labor Supply.” Econometrica 42 (4), 679–94. Jann, Ben. 2003. “The Blinder-Oaxaca Decomposition for Linear Regression Models.” Stata Journal 8 (4): 453–79. Juhn, Chinhui, Kevin M. Murphy, and Brooks Pierce. 1991. “Accounting for the Slowdown in Black–White Wage Convergence.” In Workers and Their Wages: Changing Patterns in the United States, edited by Marvin H. Kosters, 107–43. Washington, DC: AEI Press. ———. 1993. “Wage Inequality and the Rise in Returns to Skill.” Journal of Political Economy 101 (3): 410–42. Katz, Lawrence F., and David H. Autor. 1999. “Changes in the Wage Structure and Earnings Inequality.” In Handbook of Labor Economics, vol. 3, edited by Orley Ashenfelter and David Card, 1463–555. Amsterdam: Elsevier. 202 Chapter 6: How to Interpret the Results of Labor Market Analysis Katz, Lawrence F., and Kevin M. Murphy. 1992. “Changes in Relative Wages, 1962–1987: Supply and Demand Factors.” Quarterly Journal of Economics 107 (1): 35–78. Mulligan, Casey B., and Yona Rubinstein. 2008. “Selection, Investment, and Women’s Relative Wages over Time.” Quarterly Journal of Economics 123 (3): 1061–110. Nepal, Government of. 2012. “Nepal: Living Standards Survey 2010–2011, Third Round.” Central Bureau of Statistics, National Planning Commission Secretariat, Kathmandu. http://microdata.worldbank.org/index.php /catalog/1000. Neuman, Shoshana, and Ronald L. Oaxaca. 2004. “Wage Differentials in the 1990s in Israel: Endowments, Discrimination and Selectivity.” CEPR Discussion Paper 4709, Centre for Economic Policy Research, London. Ñopo, Hugo R. 2008. “An Extension of the Blinder-Oaxaca Decomposition to a Continuum of Comparison Groups.” Economics Letters 100 (2): 292–96. Oaxaca, Ronald L. 1973. “Male-Female Wage Differentials in Urban Labor Markets.” International Economic Review 14 (3): 693–709. Oaxaca, Ronald L., and Michael R. Ransom. 1994. “On Discrimination and the Decomposition of Wage Differentials.” Journal of Econometrics 61 (1): 5–21. Panama, Government of. 2008. “Encuesta de Niveles de Vida 2008” (“Living Standards Survey 2008”). Ministerio de Economia y Finanzas, Panama City. http://microdata.worldbank.org/index.php/catalog/70. World Bank. 2012. World Development Report 2012: Gender Equality and Development. Washington, DC: World Bank. https://openknowledge .worldbank.org/handle/10986/4391. Yun, Myeong-Su. 2005. “A Simple Solution to the Identification Problem in Detailed Wage Decompositions.” Economic Inquiry 43 (4): 766–72. 203 Chapter 7 Technical Notes on Labor Market Analysis This chapter introduces technical and complementary discussions on topics that contribute to the understanding of the tables and graphs covered in chapter 6. The first part discusses the main theories of discrimination that inspired the use of decomposition methods to measure it. Discrimination Theories The bulk of the literature on gender inequality focuses on defining and measuring discrimination. Most of chapter 6 focuses on decomposition methods that were originally developed to measure discrimination. Even if nowadays there is a better sense of what the unexplained component in the decomposition analysis captures—and of what it measures beyond discrimination1—it is worthwhile spending some time reviewing the theo- retical models that motivated this strand of the literature. Thus, this section provides an overview of the concepts and models on discrimination. The review is not intended to be comprehensive but instead aims to provide a minimum theoretical framework for interpreting the results produced by ADePT Gender. Examples of more comprehensive reviews made by econo- mists are Altonji and Blank (1999); Bertrand (2011); and Blau, Ferber, and Winkler (2006). This section also briefly discusses how these formative theories for economists can be bridges to the theories developed by other social sciences that also focus on women’s issues. 205 Analyzing Gender in Labor Markets: Part III Labor market discrimination exists when two equally qualified workers are treated differently solely on the basis of their gender (or any another characteristic that does not affect their productivity). Under this definition, discrimination will result in lower wages for women than for men who are equally productive. However, discrimination can take other forms that result in differences in productivity between men and women. For example, women may end up being less productive if they are discriminated against by receiving inferior on-the-job training. Moreover, discrimination may discourage women from investing in human capital and thus can result in lower productivity. Some authors refer to these effects as indirect or feed- back effects. ADePT Gender produces several decomposition methodologies that should be used with caution. Decomposition methodologies are measure- ment tools and do not allow the user to identify any of the mechanisms that result in the different forms of discrimination. Next, this section describes the most prominent models used to formalize some of these mechanisms that were developed by economists a few decades ago, when they started measuring the gender wage gap and its sources. Three main theories in labor economics frame gender discrimination in employment and wages. The first is the theory of taste discrimination developed by Gary Becker (1957), with three types of economic agents exerting dis- crimination: employers, employees, and consumers. Each of these versions results in different labor market equilibrium outcomes. The second theory is based on statistical differences in men’s and women’s characteristics. The last model—overcrowding—examines the case where too many women chose one specific occupation. Taste Discrimination Theory Becker’s theory of discrimination is grounded in prejudice and translates this concept to basic labor economics models. Discrimination is introduced into the models by monetizing the cost or disutility of an economic agent who is prejudiced against a certain group. The monetary cost or disutility is usually denominated discrimination effect d, which is introduced in the models as a positive number (Borjas 2005).2 Taste discrimination models have thus far three different versions, depending on the identity of the agent who holds the prejudice: the employer, the employee, or the customer. Each of the versions of the theory is based on a small number of assumptions, 206 Chapter 7: Technical Notes on Labor Market Analysis but with powerful results that lead to very different outcomes that explain the observed facts, as described next. Employer Discrimination Assume that there are two types of workers: men and women. Employers must decide how many of each type to hire: Em and Ef. Both types are equally productive and can be perceived as perfect substitutes for the firm. Thus, the firm output will depend on the total number of workers, regardless of the workers’ sex. For the firm, hiring an additional worker increases the product in MPE (that is, the marginal product of labor for the firm). Without discrimination, wages for members of each group should be equal, as they are equally productive. The firm hires workers until the point at which the additional cost of a worker is higher than the additional revenue, usually called the value of the marginal product of labor, VMPE. The employer’s decision rule regarding hiring becomes • If wM < wF, then hire only men; • If wM > wF, then hire only women; and • If wM = wF, then hire either men or women. In equilibrium and without discrimination, all men and women offering labor should be hired, and all would receive the same wage because it is assumed that they are equally productive. Assume now that all employers are prejudiced against women. That prejudice is modeled by increasing the cost the employers face when hiring women to wF(1 + d). As explained above, d is the monetization of the cost or disutility that generates to the employer for hiring female workers. The employer’s decision rule now becomes • If wM < wF(1+d), then hire only men; • If wM > wF(1+d), then hire only women; and • If wM = wF(1+d), then hire either men or women. Then, women will earn less than men (in an amount equal to d), employ- ers have lower profits (they are reduced from an amount represented by the triangle ACwM1 to an amount represented by the triangle ABwM0), and overall employment is lower from E1 to E0, as shown in figure 7.1, panel a. 207 Analyzing Gender in Labor Markets: Part III If some employers discriminate (for example, male managers) but others do not (for example, female managers), then the result is a fully segregated workforce. Male owners will hire only men, and female owners will hire only women. But female firms will have higher profits (as explained above); thus, we may think they would kick some male employers out of the market. If very few women are business owners because of any other constraints, the result will be more male employers, and some women will need to work for them. In this latter case, the average raw gap in pay will depend on how many women work for male employers. This situation can be visualized in a graph that shows the relative wage of men and women and the labor supply of women. Figure 7.1, panel b, shows two cases: (a) one in which all women are employed in female- managed firms represented by the supply SF0 and (b) one in which some female workers need to work for male managers; thus, a positive gender gap in pay occurs. These two cases can also be represented by looking at the firms. In the former case, male-managed firms will hire EM0 and pay wages wM0 and female- managed firms will hire EF0 and pay the same wages (see figure 7.1, panels c and d). However, if some women need to work in male firms, wages will go up for men to wM1 = wM0 + d (see figure 7.1, panel c). In sum, as long as prejudice exists among employers—and thus segre- gated employment—firms will have different profits. Firms that are preju- diced have lower profits, as they have higher costs since they hire the relatively more expensive type of labor. The relationship between profits and the coefficient of discrimination is decreasing: the greater the discrimi- nation, the smaller the profits. This behavior, together with the minority group’s labor supply, determines the equilibrium wages in the labor market. If women’s wages are too high relative to men’s wages, no firm will hire them. If women’s wages are too low relative to men’s wages, then all firms— even those with high discrimination effects—will hire them. Thus, women’s wages must be low enough to induce some firms—those with the lower discrimination effect—to hire women but high enough to still have some firms—those with the higher discrimination effect—hire men. In equilib- rium, the female-to-male wage ratio will be less than 1. This model generates two results that can be contrasted with real data: (a) women are paid less than men, and (b) firms have segregated workforces—that is, they hire only men or only women. This fact is consistent with occupational segregation. Most important, this theory implies that discrimination should not persist over time, since competi- tion will push discriminatory firms out of the market. 208 Chapter 7: Technical Notes on Labor Market Analysis Figure 7.1: Graphic Representation of Employer Discrimination a. All employers discriminate b. Relative labor demand for women Wages Relative wages A WF /WM SF0 SF0 SF1 B B WM1 = WF + d WM0 C 0.75 DF = VMPE DF EM1 EM0 Employment EF0 EF1 Female employment c. Only male-managed firms d. Only female-managed firms Wages Wages A A WM1 = WF + d WM1 = WF + d WM0 WM0 D DF EM1 EM0 Employment EF1 EF0 Employment Employee Discrimination Different equilibrium outcomes arise if employees are prejudiced against women, even if the employers are gender neutral. In other words, it can be assumed that male workers have a disutility from working with female workers. As before, this cost or disutility can be monetized by introducing a loss to their wages that becomes wM(1−d). For simplicity, assume that women do not care about the gender of their coworkers, so the value of their wage remains wF. Men and women are equally productive and thus perfect substi- tutes in the production function of the employer. Because workers with 209 Analyzing Gender in Labor Markets: Part III prejudice will not be willing to work in firms that hire men and women, the best strategy for employers is to hire only men or only women. If the firm would instead choose to have both genders in its workforce, it will have to pay more to men (who are prejudiced against women) to induce them to work, which will incur higher costs. In equilibrium, and because of the perfect segregation of men and women, all firms have the same profits. As in the previous model, here firms have segregated workforces, but there is no gap in wages. Thus, with equally productive men and women, employee discrimination is not enough to explain the observed gender wage gap. Customer Discrimination Another source of discrimination might be customers, who make purchasing decisions on the basis of a seller’s gender and not solely on the product’s price. Assume now that all products are homogeneous and produced in a competitive market at cost (equal to its price) p. Male customers prefer to buy from male sellers as opposed to female sellers. When male customers buy from female sellers, they experience a disutility that can be monetized via the price, increasing it to p(1+d) if a male customer buys from a female seller.3 Under this scenario, the best strategy for profit-maximizing employers is to hire both men and women and to place men in those occupations or positions that require interaction with customers who discriminate against women. This strategy does not result in any profit loss or workforce segrega- tion, as long as firms manage to sort workers into occupations according to whether they face discrimination. If this is the case, women’s wages won’t be affected. However, if employers are unable to place women in those occupations where they do not need to interact with discriminatory clients, their wages can be reduced to compensate for the revenue loss. This model predicts that women are segregated into occupations that do not require client contact, and that firms that have more client interaction would hire fewer women than those that do not. Other Forms of Discrimination It is important to note that other forms of interaction occur where women may be subject to discrimination. For example, it is common for men to ben- efit more than women from mentor–protégé relationships. Women are often 210 Chapter 7: Technical Notes on Labor Market Analysis excluded from this form of career development, both when the mentor is a man and when the mentor is a woman. Despite the qualitative evidence in this area, there are still very few quantitative studies that could measure these effects. Rothstein (2000) has measured the consequences of having male bosses for male and female employees. Cardoso and Winter-Ebmer (2010) find that women benefit more from higher wages in female-led firms than in male-led firms. More recently, Flabbi and others (2014) and Gagliarducci and Paserman (2014) discuss the effects of managers’ gender on gender gaps in pay and other labor market outcomes. Women are often excluded from informal networks that tend to arise at the workplace, both those that involve supervisors and those that involve peers. Job-related meetings were or are held at men’s clubs or in sports bars that males know their female peers are likely to avoid. The lack of inclusion in informal networks may result in differences in productivity as well as differences in career and promotion opportunities. Some of these examples are also discussed by Babcock and Laschever (2003) when presenting evi- dence of why women do not negotiate salaries and promotions the same way men do. Statistical Discrimination Model The models described above show how prejudice can give rise to gender wage gaps and employment segregation—either by occupation or by firm. However, similar results can arise even in the absence of prejudice, when belonging to a particular group suggests information about a worker’s productivity. Statistical discrimination arises when an employer cannot observe all the characteristics of the work in an individual but gathers information about them using statistics from the group to which the individual belongs. Suppose an employer needs to decide whether to hire a man or a woman. Each of the candidates has shown the same observable characteristics: they have the same qualifications, they performed identically in the interview, and they express the same interest and attitude toward the job. For the employer, it is important to hire someone who is reliable and who will be available to work longer hours or on weekends, if needed. During the inter- view, both candidates state they are able and willing to work extra hours if needed. Thus, on paper and with all the observable individual information, the man and the woman appear equal. 211 Analyzing Gender in Labor Markets: Part III However, the employer knows that women, on average, tend to work fewer hours than men, because they are more prone to having family respon- sibilities. Therefore, the employer infers that the woman has a higher prob- ability of being unable to work longer hours. As a result, the employer uses information derived from the group and hires the man. Although the employer does not hold any prejudice against women in general, and both candidates might be equally productive and will exert the same effort on the job, the woman ends up being discriminated against because of a character- istic of the group to which she belongs. Finally, notice that the models of statistical discrimination, as well as the models of taste for discrimination, may have indirect or feedback effects, as they might influence women’s behavior toward productive investments. Other Theories of Discrimination Bergmann (1971) developed another interesting model that results in dis- crimination.4 Her overcrowding model describes the consequences of discrimination in a partial equilibrium approach. If for whatever form of discrimination, women are circumscribed to female occupations for which there is less demand or more supply than in male occupations, women will end up with lower wages, regardless of their productivity and their sector of employment. This results from the fact that no movement of labor occurs between male and female sectors. Figure 7.2 shows the theory in a simple supply-and-demand graph. Figure 7.2: Overcrowding Model Resulting in Discrimination a. Women b. Men SM1 Wages Wages SF0 SM0 SF1 WM1 WF0 WM0 WF1 DF DF EF1 EF0 EF1 Employment EM1 EM0 Employment 212 Chapter 7: Technical Notes on Labor Market Analysis Models of glass ceilings—and before that, of internal labor markets—are also used to explain women’s lower wages. In these models, women are not promoted in the same way as men or are placed into units that do not benefit from career development. One example is the model developed by Doeringer and Piore (1985) that divides jobs into primary and secondary types, and where employers generate separate labor markets within the firm. Each of the units or type of job benefits from different career perspec- tives within the firm, regardless of the skills needed or the productivity of the tasks. The models described above for taste discrimination or statistical discrimination are also consistent with these models coming from the personnel economics strand of literature. For more details, see Bertrand (2011) or Blau, Ferber, and Winkler (2006). Earnings Equations The earnings equation is the workhorse of modern labor economics. Despite being developed more than 50 years ago, it is still a relevant instru- ment for research and policy making. The main reason is that because of its parsimonious specification, it fits several data sources and country contexts extremely well, and it allows measuring productivity for different groups of workers. It has been widely used mainly to understand returns to education but also to investigate other problems, such as gaps in pay by gender, ethnicity, language, labor turnover, and occupational choice, to name a few. It has been applied to both developed and developing countries, from wage earners to farmworkers. For example, Psacharopoulos compiles and updates Mincerian returns to education with an emphasis on developing countries in a series of papers (1972, 1973, 1985, 1989, 1994; and Patrinos and Psacharopoulos 2011). and 2011 with Patrinos). Banerjee and Duflo (2005) and Montenegro and Patrinos (2014) follow up using the same standard Mincer specification to ensure comparability. The Mincer earn- ings equation is also used extensively by cross-country studies analyzing the relationship between human capital and education and economic growth, which include developing countries in their samples (for example, Hanushek and Kimko 2000; Krueger and Lindahl 2001). The earnings equation is mostly attributed to Jacob Mincer (Mincer 1958; Mincer and Polachek 1974), but it is also considered part of the human capital theory developed by Gary Becker (1964).5 The earnings equation is 213 Analyzing Gender in Labor Markets: Part III tremendously suitable for studying the gender gap in pay, as it serves to measure differences in wages that could be attributed to education, experi- ence, and more. It has become the cornerstone of measuring gender gaps in pay, and it opens the way to several decomposition techniques described in chapter 6, and that is why it is important to understand them. The human capital theory is built on a simple model that explains labor productivity. It was initially based on a few key assumptions that led to straightforward testing and interpretation. In a nutshell, the idea is that productivity differences can be influenced (only) by differences in indi- vidual productivity, which in turn are determined by investments in educa- tion or training by individuals throughout their lives. This idea was supported by a theoretic model of investment in human capital (Becker 1964; Ben-Porath 1967; Mincer 1958; Mincer and Polachek 1974) and by an econometric model based on a log-linear function of wages on education and experience—the Mincer earnings equation. Although superseded in some aspects by modern approaches that address causality and heterogene- ity concerns, the earnings equation is still the cornerstone of the analysis of many labor issues, including gender gaps in pay (Lemieux 2006). Initially, the earnings equation was associated with a model for com- pensating wage differentials (Mincer 1958). To induce workers to under- take the additional schooling necessary for certain occupations, they must be compensated by sufficiently large earnings over the course of their working lives. Two conditions must be satisfied in equilibrium: (a) the present value of future earnings in occupations that require education minus the cost of education should be equal to the present value of future earnings of occupations that do not require occupation, and (b) labor markets are in equilibrium. Thus, the equilibrium determines the rate of return on education investments. Later, Becker (1964) formulated this model by differentiating between general human capital that can be applied to any job and specific human capital that enhances the productiv- ity of only one particular job. Then, Ben-Porath (1967) introduced the dynamic human capital model that incorporates the possibility of continu- ing training throughout the working life. Next, the earnings equation began to be interpreted as a hedonic price function that reflects the equilibrium of supply and demand for workers at each level of schooling and experience, or what Heckman, Lochner, and Todd (2006) call the accounting-identity model based on Mincer (1962) and the dynamic human capital theory of Ben-Porath (1967). 214 Chapter 7: Technical Notes on Labor Market Analysis Both formal schooling (general human capital) and on-the-job training (specific human capital) assumed to be valuable only to a particular firm determine the dynamics of life-cycle earnings. These models are usually used to explain gender gaps in education and in earnings. Because women (or their parents) anticipate labor force inter- ruptions over the working life because of childbearing, or simply because a society’s values may dictate that women should work at home, gaps in education disfavoring women appear. However, gender differences in pay can also result from investing in skills that are associated with low-paid occupations. Figure 7.3 plots different patterns of earnings associated with the working life (that is, evolution of earnings with time or age) and how earnings equations can explain gender gaps in pay. The line marked by the segment o‫މ‬J in figure 7.3 reflects the maximum earnings capacity associ- ated with the maximum level of education. As expected, earnings rise continuously with each level of (potential) experience. Workers with lower Figure 7.3: Graphic Visualization of Mincer Equation J Log wage K E G F D o’ C A B o e1 H e2 Time Source: Goldin and Polachek 1987. 215 Analyzing Gender in Labor Markets: Part III levels of education (as is the case of women in many countries) have lower levels of earnings, indicated by the line oE. Moreover, women tend to be intermittent workers, as they usually leave the labor force during the years they rear their children. Intermittent workers usually have a different age-earnings profile. The slope (that is, returns to experience) is smaller, indicated by the segment oA on the line oF. When interrupted, earnings do not increase, as workers do not accumulate experience or continue to invest in skills to avoid depreciation. This outcome is indicated with the flat seg- ment AC. Finally, most intermittent workers take a wage cut in real terms when they return to employment. Thus, the cost of the intermittency can be separated into two components: the depreciation of skills (BC) and the lost wages due to lack of accumulation of experience (CD). DG represents additional earnings lost because of the lack of on-the-job training associated with forecasted interruptions (both from employers and employees). Now moving on to econometrics, the baseline formula of the earnings equation is gi + e gi , ln w gi = b 0g + r1g Sgi + b 2g Exp gi + b 3g Exp2 (7.1) where i indicates individuals and g is gender: g = male (M), female (F), wgi is the hourly rate of pay of individual i of gender g, Sgi is years of school- ing, Expgi is the potential experience of individual i in years as a proxy of experience, and egi is an independently and identically distributed (normal) error term. The error term is assumed not to be correlated with the educa- tion decision. Potential experience is defined as age – years of education – age of entrance to elementary school.6 The key parameters of interest are r M and r F, which are the rates of return to education for men and women, and b 2M and b 2F , which are the returns to experience, also for men and women. It is worth noting a few assumptions embedded in the specification of the earn- ings equation. First, we discuss those related to the connection between the theoretic model and the econometric model. Then, we tackle the functional form of the equation and how it fits the data. Regarding the link between the theoretic model and the empirics, it has been noted that the earnings equation fits the accounting-identity model better than the compensating wage differentials model. As explained by Heckman, Lochner, and Todd (2006), in the compensating wage differentials model, individuals invest in education on the basis of its internal rate of return. Instead, r g in equation (7.1) captures the average 216 Chapter 7: Technical Notes on Labor Market Analysis rate of schooling across all schooling investments and labor market experience levels. Generally, people refer to r as the rate of return and not as the internal rate of return or marginal return of education as it comes out of the theory. Although the theory refers to the stream of earnings over the working life cycle of an individual—male or female—equation (7.1) is usually esti- mated over a cross-section of observations at a point in time, as opposed to the longitudinal analysis of a birth cohort. Using cross-section data to approximate the working life cycle implicitly assumes that the prices—or rates of return for skills—are in a steady state; that is, constant over time. It is assumed that no productivity growth occurs, and if it does, the supply and demand of skills are perfectly adjusted to maintain the equilibrium prices. However, there are reasons to believe that this is not the case. First, the rate of return of education (or more correctly, the average marginal return of education) has changed over time for different cohorts, as supply and demand do not fully adjust at the same time, thus moving equilibrium prices over time. Second, it has been found that cohorts of new entrants face relatively lower earnings (possibly because of more on-the-job train- ing needed at the start of their careers) compared with older cohorts with the same level of education. This effect results in a steeper age-wage pro- file.7 In the same way, notice that the theory model refers to the ex ante stream of earnings, whereas the estimates are done with the ex post stream of earnings. All of these concerns are relevant regardless of the worker’s gender. However, if they have different magnitudes for men and women, they become a concern for the use of the decomposition results described in chapter 6. Take, for example, the last caveat of ex ante versus ex post stream of earnings. If both men and women correctly predict their future labor force participation and stream of earnings, working with the ex post earnings will not necessarily produce any additional bias in the differences in rates of return of education between men and women. However, if women over- estimate their future stream of earnings—or underestimate labor market discrimination in the form of lower wages—they can invest more than the optimal amount in education and have lower rates of return to education than men. Alternatively, the equilibrium prices of the returns to skills can be thought to be within occupation, and it can be easily conceived that some occupations have no productivity growth and some occupations have changes in equilibrium prices. If skills’ price movements are different for 217 Analyzing Gender in Labor Markets: Part III men and women, then the assumptions again can have implications for estimating the gender gap in pay and its determinants. Functional Form of the Earnings Equation Moving to the functional form of the earnings equation, the log-linear regression model in education is a key empirical implication of the multi- plicative effect of education on identical individuals’ earnings.8 Though most of the testing has been done for developed countries, in general, the log-linear function has been found to fit the data well in developed and developing countries. Most of the studies use a linear form on years of education, though there are theoretical and practical reasons to consider alternative specifications. For example, signaling models of education assume that education does not affect labor productivity but instead regard it as a device used by workers to signal employers and thus sort themselves into high- and low-ability types.9 This theory in practice is also called “credential” or “sheepskin” effects. Moving into the returns to experience, the earnings equation (7.1) shows linear effects on the quadratic polynomial in experience. If the relation between log earnings and experience is approximately parallel for different levels of schooling, the relation between log earnings, schooling, and experi- ence becomes the sum of the effects. The relationship between age and log earnings does not have the same shape across different levels of education— the age-earnings profiles get steeper as education increases. This outcome implies that log earnings are not an additively separable function of age and education. Nevertheless, Mincer noted that the relationship between log earnings and potential experience remained approximately constant in shape across levels of education (experience-earnings profiles were parallel). This effect allows for an additive specification of the earnings equation when using potential experience instead of age, which is more parsimonious. The quadratic term captures the concave relationship between experience and log earnings. The picture is similar for developing countries, with the age-earnings profiles somewhat flatter in many cases. Box 7.1 explains how to interpret the coefficients in the Mincer earnings equation, and box 6.2 in chapter 6 shows the specifications chosen by ADePT Gender to be used in the decomposition analysis. The specifications of the regression models used as a basis for the decomposition analysis 218 Chapter 7: Technical Notes on Labor Market Analysis follows the models used by Blau and Kahn (1996, 1997, 2004) and many others in the literature (Weichselbaumer and Winter-Ebmer 2005). ADePT Gender does not include marital status or number of children as control variables. These variables could affect productivity, but they can also affect discrimination, choice of tasks within certain occupations, and so on. Thus, their inclusion has raised some controversy. Empirical evidence shows that marital status and children raise men’s wages but lower women’s wages (Blau and Kahn 1997; Korenman and Neumark 1991; Neumark and Korenman 1994; Waldfogel 1997, 1998a, 1998b). Box 7.1: Interpreting Coefficients in a Log-Linear Regression The standard interpretation of a regression coefficient b is that a one-unit change of the regressor in question results in b units of change in the average of the dependent variable, while holding all other regressors constant. The dependent variable will often enter the regression equation in log form: ln y = a + b x + e . In this case, coefficients will represent marginal effects on the average of the log of the dependent variable. However, we are usually interested in quantifying the effects of regressors on the original dependent variable in levels. To do so, we should interpret the coefficients as percentage changes of the original dependent variable. More specifically, a one-unit change in x will result in (exp(b) − 1) percentage points of change in the geometric mean of y (since taking the arithmetic average of the loga- rithm of a variable amounts to taking the geometric average of the original variable). This expression is actually very close to b itself when the latter is “small” (between −0.1 and 0.1 as a rule of thumb), so that directly interpreting the regression coeffi- cients as percentage changes of the dependent variable measured in levels is a good approximation of the true marginal effects. In the case of the regressor of interest, x, which is a binary 0–1 dummy variable, the same expression and approximation apply. However, when the dependent variable is in levels, one must be careful to interpret the discrete change of the dummy in the right way: (exp(b) − 1) represents the percentage points of change in the geometric mean of y when x switches from 0 to 1, which is the effect of switching from the group indicated by a 0 (the excluded category) to the group indicated by a 1. 219 Analyzing Gender in Labor Markets: Part III Econometric Concerns Model specification and goodness of fit. A key assumption underlying the econometric model of the earnings equation is that schooling accurately represents the opportunity set by a typical individual. The main problem stems from the failure to observe ability—which is a factor in the education decision—or other skills (see more below). In this case, the estimated earn- ings equation will produce a biased estimation of the returns to education. This problem is a clear example of omitted variable bias. The literature has overcome this problem in two ways: (a) by using natural experiments (Card 1999) and (b) by using quantile regressions (Fortin, Lemieux, and Firpo 2011). The other key assumption underlying the econometric model of equation (7.1) is that there are no heterogeneous effects. It was explained above that the return to experience b might be different for different levels of education. And then without saturating the regression, effects working at the (sub)group level can be counterbalanced and can be missed in the average estimate.10 Adding variables that capture a richer set of skills to the Mincer equa- tion can be important for improving the measure of productivity by gender, reducing potential biases arising from potentially different correlations between ability and education by gender and increasing the percentage of the gender gap in pay that is explained by observed characteristics. The term skills refers to the capacity to perform different tasks. The Skills Toward Employment and Productivity (STEP) surveys conducted by the World Bank classify skills into three groups: (a) cognitive skills, (b) socioemotional skills, and (c) job-relevant skills. Box 7.2 presents the definitions of each type of skill, as well as the main variables that the STEP initiative collects around the world. These measures of skills have quickly gained acceptance. Economists have applied more sophisticated quantitative methods to theories that have a long tradition among psychologists and sociologists. However, achieve- ment tests, grades, and credited education do not explain everything about life earnings,11 implying something important was still missing in the Mincer equation. Moreover, quantifiable and reliable cross-country comparable measures are now being collected, including the Organisation for Economic Co-operation and Development’s (OECD) Programme for the International Assessment of Adult Competencies and the World Bank’s STEP program. However, these studies still suffer from identification problems, as recently 220 Chapter 7: Technical Notes on Labor Market Analysis described by Heckman and Kautz (2012). Fortin (2008) adds measures of several noncognitive factors to the basic Mincerian specification before decomposing the gender wage gap for young workers in the United States. He finds that these noncognitive factors—in particular, a composite measure of the extent to which workers find money and career success valuable in life—account for a small but nontrivial part of the gender gap in pay. Meanwhile, Mueller and Plug (2006) use the “Big Five,” an established tax- onomy of personality traits.12 Besides adding variables that are associated with productivity such as skills, authors have added other variables that are indirectly related to skills. On the one hand, analysts add variables related to competing respon- sibilities in the household. In particular, they add dummies for marriage and number and age of children. One of the main researchers of marriage penalty is Waldfogel, who has shown in various papers (Berger and Waldfogel 2004; Blau, Kahn, and Waldfogel 2000; Han and others 2008; Joshi, Paci, and Waldfogel 1999; Sigle-Rushton and Waldfogel 2007; Waldfogel 1997, 1998a, 1998b, 1999) that marriage constitutes a penalty for women, whereas it increases men’s wages. However, many have argued that there is no reason to predict that married women with children should be expected to earn less than men per hour. If that is the case, it is because women do not have certain skills or are being discriminated against, and thus the variables should be left out the regression. Authors sometimes include other variables that aim to capture other wage-enhancing theories. For example, workers earn more when they are in a better job match, and job matches are improved by searching for the best opportunity. It has been proved that changing jobs helps a worker find the right match and increases wages (Light and Ureta 1990, 1992, 1995; Posadas 2009; Royalty 1998). Thus, variables that capture previous job-switching behavior can help explain current wages. Heterocedasticity. The Mincer earnings equations are estimated using a linear regression model corrected by heterocedasticity. The linear regression model gives the best linear unbiased estimates, under certain assumptions. The first is that the error terms are homocedastic; that is, the standard devia- tion of the error term is not correlated with any of the included explanatory variables. The transformation to log of earnings in practice ensures that the errors are approximately homocedastic, but it is still best to obtain homoce- dastic consistent standard errors (Cameron and Trivedi 2005). ADePT Gender corrects the estimate for heterocedasticity using robust standard 221 Analyzing Gender in Labor Markets: Part III error correction. Neglecting the correction for heterocedasticity when the errors are heterocedastic—as is often the case for cross-sectional data—can lead to deflation or inflation of the true standard errors, with implications for the statistical significance of gender differences. If errors are indeed homocedastic, White’s estimated standard errors are less efficient than stan- dard errors estimated with ordinary least squares. One way to deal with heterocedasticity is to estimate quantile regressions while simultaneously being informative about distributional issues. Omitted variable bias. Schooling is not randomly assigned; rather, it is an outcome that depends on choices made by individuals and their parents. The human capital theory treats schooling as an investment by individuals in themselves, and r is interpreted as a measure of the return to education. Thus, education becomes an endogenous variable. Under these circum- stances, unless we can argue that schooling depends on a set of variables that is independent of the error term, the estimates of r should not be taken as causal—that is, interpreted as the effects of an additional year of education on earnings. However, even if the econometrician cannot make the latest assumption but can assume that education depends in the same way on the same set of variables for men and women, then it is like assuming that the bias is equal for both men and women. In addition, although r would not be causal, the gender difference in the returns can be consistently estimated. As long as the bias is the same for men and women, omitted variables should not bias the decomposition results that follow, which require only caution in their interpretation. Partial equilibrium. The earnings equations are derived from a partial equilibrium approach, and thus the interpretation of the results is limited. For example, under partial equilibrium, it is not possible to capture an increase in the returns to education that incentivize women to acquire more educa- tion and enter the labor force, causing in turn a change in the price. This chain of effects is more likely to be measured using structural models, which are based on several assumptions and are computationally more demanding. Nonrandom Selection into the Labor Force Estimates of the Mincer earnings equation can be biased if they are based on a nonrandom sample of the working-age population. The key technical terms used in this subsection are introduced in box 7.2. The understanding 222 Chapter 7: Technical Notes on Labor Market Analysis Box 7.2: Technical Terms Related to Nonrandom Selection of Women into the Labor Force Selection bias exists in many problems in economics. It is present in most cases where the analyst can observe only the outcome of a previous maximization choice. For example, when looking at the returns of occupations, it might be tempting to say that certain occupations have larger returns than others. However, the counterfactual—that is, an individual’s earnings had he or she chosen a different occupation—cannot be observed. More generally, this is a problem of nonrandom sample selection, where respondents fail to provide answers to certain questions, which leads to missing data (either for dependent or independent variables). Since the analyst cannot observe the complete distribution, the estimates will be biased unless the missing data are random—that is, not correlated with the included variables. Incidental truncation occurs when the analyst does not observe the complete distribu- tion of the outcome variable y because of the outcome of another variable. The leading example is estimating the so-called wage offer function in labor economics. When estimating a Mincer equation, the analyst observes the wage offer only for those indi- viduals who work. But for those currently out of the labor force, we do not observe the wage offer. Because working may be systematically correlated with unobservables that affect the wage offer and the working decision, the estimates of the Mincer equation will be biased. of the correction of selection is important for the interpretation of the results discussed in chapter 6, so the analyst can assess whether to work with the simple Oaxaca-Blinder decomposition with selection correction of women into the labor force or with more sophisticated decompositions that do not control for it (Juhn-Murphy-Pierce or Ñopo). This section also helps explain the interpretation of the Mincer equation using the Heckit method- ology that corrects the problem. But before delving into the details of the econometric formulation, a simple example illustrating the origin and mag- nitude of the nature of the problem is presented. Suppose you are interested in estimating the returns to education for women in a country where female labor force participation is low, around 50 percent. If women are randomly selected from the working-age popula- tion, the estimates will be unbiased.13 Now, let’s assume the opposite: women are not randomly selected and that less educated women are less likely to work than more educated women. One reason is that women with less education are more likely to earn lower wages, which in turn might not 223 Analyzing Gender in Labor Markets: Part III be enough to buy household goods, such as food and childcare. To further simplify the case, assume that the women have two levels of education: poorly educated and highly educated. In our simple example, only a fraction (less than 100 percent) of poorly educated women end up working, whereas all highly educated women work. Let’s assume wages are observed for the subsample of women who work— the nature of the phenomenon we are observing forces us to limit our analy- sis to a selected sample. However, this sample is not random, in the sense that it might be very different from the overall population of women. Thus, educated women will be overrepresented in the selected sample, with respect to the general female population. On its own, however, this fact does not bias the estimation of the earnings equation (Achen 1986). If the only variable that influences selection (education itself) is exogenous, then we have what is called deterministic selection,14 and it is solved by including education variables in the regression. Now, let’s make the assumptions more realistic. Assume that participa- tion does not depend on the level of education (or skills) but on an unob- servable variable called effort or motivation. High-effort women are more likely to be more productive, to receive higher (potential) wages, and thus to work. Very likely, high-effort women will also be better educated as part of their type. Thus, the selected sample, made up mostly of educated women, will still include some uneducated women; however, they will not be representative of uneducated women in general, because their high level of motivation will “compensate” for their lack of education. The uneducated women in our sample will be (selectively) the ones with higher potential wages. Education and unmeasured motivation will be correlated in the selected sample whether or not they are in the overall population, which will lead us to confound their effects on wages. In prac- tice, it will underestimate the effect of education on wages, because in the selected sample, women with little education are unusually motivated and still earn high wages. In fact, if women were to enter the labor force randomly, their level of education would pay off more. This type of pattern is usually denominated positive selection (Blundell and others 2007; Olivetti and Petrongolo 2008). Finally, returns to educa- tion are such that highly educated women’s wages are on average higher than those of poorly educated women. Figure 7.4 illustrates the selection problem graphically for the two instances. The x-axis represents the two levels of education, poorly educated 224 Chapter 7: Technical Notes on Labor Market Analysis and highly educated, and the y-axis reflects wages. If we could observe the wages for all women, conditional on their education level, the sample would be the one depicted by the dark gray circles. The implied effect of education on earnings as estimated by linear regression would be the slope of the dark gray line. However, we observe the wages for only those women who choose to participate in the labor market—the light gray dots. We see selection at work for the poorly educated group—only some women here choose to work. Panel a represents deterministic selection; not all women with little educa- tion enter the sample, but their participation is random conditional on educa- tion. The fact that they are missing does not affect the estimated slope of the regression line. In panel b, we see endogenous selection. What changes is that the women we do observe are not representative of the whole distribution of wages conditional on having a low level of education (that is, are not ran- domly selected, even after controlling for their level of education). The rea- son is that an unobserved variable—motivation—mediates participation, so that the sample of women for which the wage is observable (the orange dots) gives us a biased estimate of the education coefficient—the slope of the orange line is flatter. We are underestimating the effect of education on wages, because we are confounding it with the effect of motivation on wages via selection. Figure 7.4: Graphic Representation of Selection Bias a. Deterministic selection b. Endogenous selection Actual Wage Wage Actual = Selection Selected L H Education L H Education All sample is observed/selected Not observed but part of the sample Observed and thus selected Note: L = poorly educated; H = highly educated. 225 Analyzing Gender in Labor Markets: Part III It is clear from this simple illustration that the selection bias is exacer- bated (a) the lower the participation of low-skill or low-effort females in the labor market and (b) the stronger the correlation between the (endogenous) participation decision and the variable of interest—wages. No concerns about selection bias exist for men, because usually all of them work. However, in countries with low male labor force participation (postconflict or high levels of disabilities), the user should explore whether the probabil- ity of being disabled is correlated with education and thus potentially affects the estimated returns to education for men as well. Two commonly used methods address selection bias in econometrics. The first is the two-step Heckman correction, also known as Heckit. The second is simply the (one-step) maximum likelihood that might yield more efficient estimates. ADePT Gender shows that the Heckit estimates as the interpretation of the coefficient is more intuitive—the user can simply think that another variable is added to the earnings equation to capture the decision to work or not to work, and correct the bias. This variable is the Mills inverse ratio, and its omission can be thought as caus- ing the bias. Heckman Two-Step Selection Correction Model The Heckit model is computed in two steps. First, a participation equa- tion is estimated, which estimates the probability that a woman is work- ing on the basis of variables that are related to her productivity in the labor market and to other factors that influence her decision to work but that do not affect wages. These are usually called exclusion variables, as they are needed to identify the model. In the original work of Heckman, the probability of working is estimated using a probit model like the following: Pr[Working ] = j 1 F + j2 F SF + j 3 F XF + j 4 XF + Z + e , F 2 (7.2) where all the variables of the Mincer equation (7.1) are included, e is a white noise error, and the vector Z is a set of exclusion variables, such as number of children, dependent elderly, and marital status, that are determi- nants of labor force participation but not of wages. In a way, it assumes that women with care responsibilities are less likely to work than women without them. Some academics have noticed that these variables are not actually 226 Chapter 7: Technical Notes on Labor Market Analysis exogenous to the potential wage (Kunze 2008). It can be hypothesized that women with lower potential wages are more likely to become housewives and have more children than women who pursue a career. In other words, both the marital status and the number of children are endogenous variables to the labor market participation decision and to wages. Thus, some authors have argued in favor of adding nonearned income (such as a husband’s labor income) as the exclusion variable in vector Z. Women whose husbands earn higher wages have less need to work than women who live in households that struggle to make ends meet. However, some authors have also argued that this variable can be endogenous to women’s wages if there is assortative matching—that is, women with higher potential earnings marrying high-earning husbands (Burdett and Coles 1999; Chiappori, Iyigun, and Weiss 2009). Despite these criticisms, these variables are still the most common choice, even in recent academic publications. For example, Christofides, Polycarpou, and Vrachimis (2013) use as exogenous variables the number of children under 16, childcare provisions, and income from property rents and financial assets. Mulligan and Rubinstein (2008) use the number of children ages 0–6 interacted with marital status, and Beblo and others (2003) use marital status, number of children, and nonearned income. In the second stage, the earnings equation is estimated adding the Mills inverse ratio, l, as an explanatory variable to correct for selection issues.15 The earnings equation for the human capital model is ln w F = a 0 F + r F SF + b 1F XF + b 2F XF 2 + gl ( Zjf ) +¨i . (7.3) The regression (7.3) is the basis of the Oaxaca-Blinder decomposition with selection correction. Most of the time, the bias correction is introduced for women only, since it is assumed that all men work. The coefficient accompanying the inverse Mills ratio, g, indicates the type of selection effect—positive or negative. When g is positive, skilled women are more likely to work than unskilled women. This situation is typical of most devel- oped countries.16 When g is negative, skilled women are less likely to work than unskilled women. Although higher-skilled women would earn more and thus have a larger opportunity cost for staying out of the labor force, they are usually married to husbands who also have high earnings and thus can afford to stay out of the labor market. This was the case of the United States in the 1940s, 1950s, and 1960s, when the rates of participation for 227 Analyzing Gender in Labor Markets: Part III black women were high and rates for white women were lower (Cunningham and Zalokar 1992; Mulligan and Rubinstein 2008). Notes 1. That is, the unexplained component in the decomposition captures the effects of unobservable skills and the underlying sorting of women because of preferences or any other decision. 2. If a positive preference for a certain group is present—for example, female employers prefer to work with female workers—then the coeffi- cient d will be positive. 3. Assume that women face no disutility by buying from men or women. 4. For more details on these theories, see Borjas (2005). 5. Although Becker and Mincer did not coauthor any scientific papers, they mutually benefited from their interactions, mostly in the years they led the Labor Lab at Columbia University (Grossbard 2006). 6. The most common age of entrance to elementary school is six. 7. See, for example, Beaudry and Green (2000) for Canada. 8. For more details, see Lemieux (2006, 130–32). 9. See Cahuc and Zylberberg (2004) for a brief explanation of the Spencer model applied to education, and Bedard (2001). 10. See Angrist and Pischke (2008) for a description of fully saturated models. 11. For example, adolescent achievement test scores explain only about 15 percent of the variance in later-life earnings (Heckman and Kautz 2012). 12. The “Big Five” personality traits are (a) extraversion, (b) agreeableness, (c) conscientiousness, (d) neuroticism, and (e) openness to experience (Mueller and Plug 2006). 13. If female labor force participation is very low, the precision of the estimates could be affected in small household surveys. 14. Notice, however, that if the true value of b varies across observations, then what we estimate on the selected sample is the parameter for the specific selected population. 15. See appendix E for the steps that explain how the inverse Mills ratio is computed. 228 Chapter 7: Technical Notes on Labor Market Analysis 16. See Blundell and others (2007) for the United Kingdom and Olivetti and Petrongolo (2008) for a sample of 26 countries in the Organisation for Economic Co-operation and Development. References Achen, Christopher H. 1986. Statistical Analysis of Quasi-Experiments. Berkeley: University of California Press. Altonji, Joseph G., and Rebecca M. Blank. 1999. “Race and Gender in the Labor Market.” In Handbook of Labor Economics, vol. 3, edited by Orley Ashenfelter and David Card, 3143–259. Amsterdam: Elsevier. Angrist, Joshua D., and Jörn-Steffen Pischke. 2008. Mostly Harmless Econometrics: An Empiricist’s Companion. Princeton, NJ: Princeton University Press. Babcock, Linda, and Sara Laschever. 2003. Women Don’t Ask. Princeton, NJ: Princeton University Press. Banerjee, Abhijit V., and Esther Duflo. 2005. “Growth Theory through the Lens of Development Economics.” In Handbook of Economic Growth, vol. 1, edited by Philippe Aghion and Steven N. Durlauf, 473–552. Amsterdam: Elsevier. Beaudry, Paul, and David Green. 2000. “Cohort Patterns in Earnings and the Skill-Biased Technical Change Hypothesis.” Canadian Journal of Economics 33 (4): 907–36. Beblo, Miriam, Denis Beninger, Anja Heinze, and François Laisney. 2003. “Measuring Selectivity-Corrected Gender Wage Gaps in the EU.” Discussion Paper 03-74, Centre for European Economic Research, Mannheim, Germany. Becker, Gary. 1957. The Economics of Discrimination. Chicago: Chicago University Press. ———. 1964. “Investment in Human Capital: A Theoretical Analysis.” Journal of Political Economy 70 (5, part 2): 9–49. Bedard, Kelly. 2001. “Human Capital versus Signaling Models: University Access and High School Dropouts.” Journal of Political Economy 109 (4): 749–75. Ben-Porath, Yoram. 1967. “The Production of Human Capital and the Life Cycle of Earnings.” Journal of Political Economy 75 (4): 352–65. 229 Analyzing Gender in Labor Markets: Part III Berger, Lawrence, and Jane Waldfogel. 2004. “Maternity Leave and the Employment of New Mothers in the United States.” Journal of Population Economics 17 (2): 331–49. Bergmann, Barbara R. 1971. “The Effect on White Incomes of Discrimination in Employment.” Journal of Political Economy 79 (2): 294–313. Bertrand, Marianne. 2011. “New Perspectives on Gender.” In Handbook of Labor Economics, vol. 4, edited by Orley Ashenfelter and David Card, 1543–90. Amsterdam: Elsevier. Blau, Francine D., Marianne A. Ferber, and Anne E. Winkler. 2006. The Economics of Women, Men, and Work. 5th ed. Upper Saddle River, NJ: Pearson/Prentice Hall. Blau, Francine D., and Lawrence M. Kahn. 1996. “International Differences in Male Wage Inequality: Institutions versus Market Forces.” Journal of Political Economy 104 (4): 791–837. ———. 1997. “Swimming Upstream: Trends in the Gender Wage Differential in the 1980s.” Journal of Labor Economics 15 (1): 1–42. ———. 2004. “The US Gender Pay Gap in the 1990s: Slowing Convergence.” NBER Working Paper 10853, National Bureau of Economic Research, Cambridge, MA. Blau, Francine D., Lawrence M. Kahn, and Jane Waldfogel. 2000. “Understanding Young Women’s Marriage Decisions: The Role of Labor and Marriage Market Conditions.” Industrial and Labor Relations Review 53 (4): 624–47. Blundell, Richard, Amanda Gosling, Hidehiko Ichimura, and Costas Meghir. 2007. “Changes in the Distribution of Male and Female Wages Accounting for Employment Composition Using Bounds.” Econometrica 75 (2): 323–63. Borjas, George J. 2005. Labor Economics. New York: McGraw-Hill. Burdett, Kenneth, and Melvyn G. Coles. 1999. “Long-Term Partnership Formation: Marriage and Employment.” Economic Journal 109 (456): 307–34. Cahuc, Pierre, and André Zylberberg. 2004. Labor Economics. Cambridge, MA: MIT University Press. Cameron, A. Colin, and Pravin K. Trivedi. 2005. Microeconomics: Methods and Applications. lst edition. Cambridge, UK: Cambridge University Press. Card, David. 1999. “The Causal Effect of Education on Earnings.” In Handbook of Labor Economics, vol. 3, ed. Orley Ashenfelter and David Card, 1802–59. Amsterdam: Elsevier. 230 Chapter 7: Technical Notes on Labor Market Analysis Cardoso, Ana Rute, and Rudolf Winter-Ebmer. 2010. “Female-Led Firms and Gender Wage Policies.” Industrial and Labor Relations Review 64 (1): 143–63. Chiappori, Pierre-André, Murat Iyigun, and Yoram Weiss. 2009. “Investment in Schooling and the Marriage Market.” American Economic Review 99 (5): 1689–713. Christofides, Louis N., Alexandros Polycarpou, and Konstantinos Vrachimis. 2013. “Gender Wage Gaps, ‘Sticky Floors’ and ‘Glass Ceilings’ in Europe.” Labour Economics 21 (C): 86–102. Cunningham, James S., and Nadja Zalokar. 1992. “Economic Progress of Black Women, 1940–1980: Occupational Distribution and Relative Wages.” Industrial and Labor Relations Review 45 (3): 540–55. Doeringer, Peter B., and Michael J. Piore. 1985. Internal Labor Markets and Manpower Analysis. Armonk, NY: M. E. Sharpe. Flabbi, Luca, Mario Macis, Andrea Moro, and Fabiano Schivardi. 2014. “Do Female Executives Make a Difference? The Impact of Female Leadership on Gender Gaps and Firm Performance.” CEPR Discussion Paper 10228, Centre for Economic Policy Research, London. Fortin, Nicole M. 2008. “The Gender Wage Gap among Young Adults in the United States: The Importance of Money versus People.” Journal of Human Resources 43 (4): 884–918. Fortin, Nicole, Thomas Lemieux, and Sergio Firpo. 2011. “Decomposition Methods in Economics.” In Handbook of Labor Economics, vol. 4, edited by Orley Ashenfelter and David Card, 1–102. Amsterdam: Elsevier. Gagliarducci, Stefano, and M. Daniele Paserman. 2014. “Gender Interactions in Firm Hierarchies: Evidence from Linked Employer-Employee Data.” Unpublished manuscript. Institute for the Study of Labor, Bonn. Goldin, Claudia, and Solomon Polachek. 1987. “Residual Differences by Sex: Perspectives on the Gender Gap in Earnings.” American Economic Review 77 (2): 143–51. Grossbard, Shoshana. 2006. “Jacob Mincer 1922–2006.” Review of Economics of the Household 4 (4): 441–42. Han, Wen-Jui, Christopher Ruhm, Jane Waldfogel, and Elizabeth Washbrook. 2008. “The Timing of Mothers’ Employment after Childbirth.” Monthly Labor Review 131 (6): 15–27. Hanushek, Eric A., and Dennis D Kimko. 2000. “Schooling, Labor-Force Quality, and the Growth of Nations.” American Economic Review 9 (5): 1184–208. 231 Analyzing Gender in Labor Markets: Part III Heckman, James J., and Tim Kautz. 2012. “Hard Evidence on Soft Skills.” Labour Economics 19 (4): 451–64. Heckman, James J., Lance J. Lochner, and Petra E. Todd. 2006. “Earnings Functions, Rates of Return and Treatment Effects: The Mincer Equation and Beyond.” In Handbook of the Economics of Education, vol. 1, edited by Eric A. Hanushek and Finis Welch, 307–458. London and Amsterdam: Elsevier, North-Holland. Joshi, Heather, Pierella Paci, and Jane Waldfogel. 1999. “The Wages of Motherhood: Better or Worse?” Cambridge Journal of Economics 23 (5): 543–64. Korenman, Sanders, and David Neumark. 1991. “Does Marriage Really Make Men More Productive?” Journal of Human Resources 26 (2): 282–307. Krueger, Alan B., and Mikael Lindahl. 2001. “Education for Growth: Why and for Whom?” Journal of Economic Literature 39 (4): 1101–36. Kunze, Astrid. 2008. “Gender Wage Gap Studies: Consistency and Decomposition.” Empirical Economics 35 (1): 63–76. Lemieux, Thomas. 2006. “The ‘Mincer Equation’ Thirty Years after Schooling, Experience, and Earnings.” In Jacob Mincer: A Pioneer of Modern Labor Economics, edited by Shoshana Grossbard, 127–45. New York: Springer Science. Light, Audrey, and Manuelita Ureta. 1990. “Gender Differences in Wages and Job Turnover among Continuously Employed Workers.” American Economic Review 80 (2): 293–97. ———. 1992. “Panel Estimates of Male and Female Job Turnover Behavior: Can Female Nonquitters Be Identified?” Journal of Labor Economics 10 (2): 156–81. ———. 1995. “Early-Career Work Experience and Gender Wage Differentials.” Journal of Labor Economics 13 (1): 121–54. Mincer, Jacob. 1958. “Investment in Human Capital and Personal Income Distribution.” Journal of Political Economy 66 (4): 281–302. ———. 1962. ‘‘Labor Force Participation of Married Women: A Study of Labor Supply.’’ In Aspects of Labor Economics, edited by H. Gregg Lewis, 63–105. Princeton, NJ: Princeton University Press. Mincer, Jacob, and Solomon Polachek. 1974. “Family Investments in Human Capital: Earnings of Women.” Journal of Political Economy 82 (Supplement): S76–S108. 232 Chapter 7: Technical Notes on Labor Market Analysis Montenegro, Claudio E., and Harry Anthony Patrinos. 2014. “Comparable Estimates of Returns to Schooling around the World.” Policy Research Working Paper 7020, World Bank, Washington, DC. Mueller, Gerrit, and Erik J. S. Plug. 2006. “Estimating the Effect of Personality on Male and Female Earnings.” Industrial and Labor Relations Review 60 (1): 3–22. Mulligan, Casey B., and Yona Rubinstein. 2008. “Selection, Investment, and Women’s Relative Wages over Time.” Quarterly Journal of Economics 123 (3): 1061–110. Neumark, David, and Sanders Korenman. 1994. “Sources of Bias in Women’s Wage Equations: Results Using Sibling Data.” Journal of Human Resources 29 (2): 379–405. Olivetti, Claudia, and Barbara Petrongolo. 2008. “Unequal Pay or Unequal Employment? A Cross-Country Analysis of Gender Gaps.” Journal of Labor Economics 26 (4): 621–54. Patrinos, Harry A., and George Psacharopoulos. 2011. “Education: Past, Present and Future Global Challenges.” Policy Research Working Paper 5616, World Bank, Washington, DC. Posadas, Josefina. 2009. “An Investigation of Gender Differentials in the Labor Market.” PhD dissertation, Boston University. Psacharopoulos, George. 1972. “Rates of Return on Investment in Education around the World.” Comparative Education 16 (1): 54–67. ———. 1973. Returns to Education: An International Comparison. Amsterdam: Elsevier; San Francisco: Jossey-Bass. ———. 1985. “Returns to Education: A Further International Update and Implications.” Journal of Human Resources 20 (4): 583–604. ———. 1989. “Time Trends of the Returns to Education: Cross-National Evidence.” Economics of Education Review 8 (3): 225–31. ———. 1994. “Returns to Education: A Global Update.” World Development 22 (9): 1325–43. Psacharopoulos, George, and Harry A. Patrinos. 2004. “Returns to Investment in Education: A Further Update.” Education Economics 12 (2): 111–34. Rothstein, Donna S. 2000. “Early Career Supervisor Gender and the Labor Market Outcomes of Young Workers.” In Gender and Family Issues in the Workplace, edited by Francine D. Blau and Ronald G. Ehrenberg, 210–55. New York: Russell Sage Foundation. 233 Analyzing Gender in Labor Markets: Part III Royalty, Anne Beeson. 1998. “Job-to-Job and Job-to-Nonemployment Turnover by Gender and Education Level.” Journal of Labor Economics 16 (2): 392–443. Sigle-Rushton, Wendy, and Jane Waldfogel. 2007. “Motherhood and Women’s Earnings in Anglo-American, Continental European, and Nordic Countries.” Feminist Economics 13 (2): 55–91. Reprinted in Feminist Economics, edited by Lourdes Benaria, Ann Marie May, and Diane Strassman. Cheltenham, UK: Edward Elgar, 2009. Waldfogel, Jane. 1997. “The Effect of Children on Women’s Wages.” American Sociological Review 62 (2): 209–17. ———. 1998a. “The Family Gap for Young Women in the United States and Britain: Can Maternity Leave Make a Difference?” Journal of Labor Economics 16 (3): 505–45. ———. 1998b. “Understanding the ‘Family Gap’ in Pay for Women with Children.” Journal of Economic Perspectives 12 (1): 137–56. ———. 1999. “Family Leave Coverage in the 1990s.” Monthly Labor Review 122 (10): 13–21. Weichselbaumer, Doris, and Rudolf Winter-Ebmer. 2005. “A Meta-Analysis of the International Gender Wage Gap.” Journal of Economic Surveys 19 (3): 479–511. 234 PART IV Conclusions 235 Chapter 8 Reflections on What ADePT Gender Does and What It Does Not Do Despite considerable advances in recent decades, gender inequality remains pervasive worldwide and in many dimensions of life. The nature and extent of gender-related differences vary considerably across countries and regions, but the evidence is striking. In no region of the world are women equal to men in legal, social, and economic rights, nor do they achieve equal outcomes. According to the World Development Report 2012: Gender Equality and Development (World Bank 2012), gender gaps are significant and wide- spread in endowments, economic opportunities, and agency, and they are particularly large for the poor, for ethnic minorities, and for other disadvan- taged groups. Women and girls bear the largest and most direct costs of these in equalities, but the costs cut more broadly across society, as gender inequality impedes a country’s ability to grow, to reduce poverty, and to build effective institutions. For that reason, gender equality is an important development issue. Similarly, strengthening the capacity to quantify the size of the prevailing gender differentials, to understand their nature and causes, and to monitor progress in reducing existing gaps is a priority for policy mak- ers and development practitioners more broadly. ADePT Gender is an important tool for building this capacity. It is designed to facilitate the standardized analysis of gender inequalities—and their determinants—by allowing users to easily derive a detailed profile of 237 Conclusions: Part IV existing gender gaps in outcomes in the three dimensions identified by the World Development Report—endowments, economic opportunities, and agency. If several years of the same survey are available, users can also assess progress in reducing inequalities by comparing results over time. Moreover, the easiness with which profiles can be derived facilitates cross-countries comparison. Above all, the software’s strength lies in its ability to produce very quickly—with relatively limited resources—a large volume of quantitative information. Users can also focus on specific groups of interest—such as the young, the poor, and so on—by disaggregating the analysis results on the basis of existing variables. Additional pluses are: (a) the information is systematically organized in standardized tables and preformatted graphs, so that they can represent the skeleton of a country gender profile; and (b) the use of standardized commands and elaborate error messages significantly reduces the likelihood of errors. In addition to profiling gender gaps across a number of indicators, users can also carry out simple analyses of the factors that lead to gender inequality and that perpetuate it over time. The speed at which ADePT carries out the analysis and its high degree of accu- racy also foster the opportunity to use international comparisons to assess the effects of different policy and social environments and to enrich the policy dialogue. The software can also be used for simple simulations of the potential effect on gender equality of both gender-sensitive and gender- blind policy reforms. Particular attention is given to the analysis of gender inequality in the labor market, as differential access to economic opportunities is a major source of disadvantages in other areas for women and girls. In this area, ADePT Gender can do a simple analysis of the gender gaps in earnings and perform a number of decompositions. The technical chapters in this manual provide interested readers with useful theoretical background to the more sophisticated analysis carried out by ADePT. Gender equality is an important development issue that has attracted a wealth of research and that continues to receive a great deal of attention in the academic community and in the policy debate. It covers a broad range of issues and has used a variety of methodologies and approaches, with different degrees of complexity. ADePT is a software package designed simply to generate standardized tables and charts summarizing the results of simple diagnostics based on household data (Lokshin and others 2013). It allows users to profile gender differences in a variety of core indicators of 238 Chapter 8: Reflections on What ADePT Gender Does and What It Does Not Do welfare and empowerment and to identify some of the major driving forces. However, ADePT software is not intended to be used for sophisticated econometrics. In addition, although it is important to obtaining a compre- hensive profile of gender inequalities, a number of key issues cannot be addressed using only standard household data. As ADePT is designed for use with household data, these issues are better analyzed with other instruments. Thus, the gender profile that emerges is comprehensive but not exhaustive, and the analysis allowed in the program has only a limited degree of sophis- tication. The user interested in broader and deeper analysis of gender equal- ity is well advised to look at the wide range of alternative tools available. References Lokshin, Michael, Sergiy Radyakin, Zurab Sajaia, and William Creitz. 2013. ADePT User Guide. Version 5. Washington, DC: World Bank. World Bank. 2012. World Development Report 2012: Gender Equality and Development. Washington, DC: World Bank. https://openknowledge .worldbank.org/handle/10986/4391. 239 APPENDIXES 241 Appendix A Fields, Variable Definitions, and Variable Requirements Field Variable definition Variable requirement Internal check Main Household ID Household identifier. One variable or set of If household ID is one variables, numeric or variable or a set of numeric alphanumeric (numeric variables, ADePT Gender variables have to be checks whether it or they are integers) integers Household Survey sample weights. Household surveys assign a Continuous variable weights specific household weight to each household. The weight is used to give each sample household a level of representation in the total household population. Household weights adjust for the differences in the probability of selecting a household in the household population. Weights need to be applied when tabulations have to produce a proper representation. In a database of individuals, the household weights should not vary among individuals within the household. Urban Rural or urban household residence (urban = 1). Binary variable Binary variable and one of the values is 1 Region Variable indicating the geographical region. Categorical variable Integer variable Poverty line (a) The threshold level of per capita consumption, Constant value variable Single value, no larger than expenditure, or household income, above which a mean/median/p60 value of person is no longer considered poor or (b) an indicator the welfare aggregate of poverty status of household (poor = 1) (binary variable). It should be measured for the same frequency (annual, monthly, and so on) and currency (local currency or U.S. dollars, real or nominal) as the welfare aggregate. (continued) 243 Appendixes Field Variable definition Variable requirement Internal check Welfare The variable that captures well-being; typically, total per Continuous variable Continuous variable aggregate capita consumption, but it can also be total per capita household expenditure or total per capita household income. Alternatively, the measures can be corrected for adult equivalent. If the welfare aggregate is a categorical variable for quintiles or deciles, the value 1 corresponds to the bottom quintile or decile of consumption, expenditure, or income. The quintile (decile) is a categorical variable that divides the population, sorted according to the welfare aggregate, in 5 (10) equal parts. Gender Sex of individual (male = 1). Binary variable Binary variable with one of the values equal to 1 Age Age of individual (in years). Integer variable Integer variable between 0 and 99 Household Head of household (head = 1). Binary variable or Binary variable with one of head expression the values equal to 1, only one household member per household is head, and head is at least age 15 Marital status Marital status. Categorical variable Human capital Current school year Primary Attended primary school in current year Binary variable or Binary variable with one of (attended = 1). expression the values equal to 1 Secondary Attended secondary school in current year Binary variable or Binary variable with one of (attended = 1). expression the values equal to 1 Postsecondary Attended postsecondary school in current year Binary variable or Binary variable with one of (attended = 1). expression the values equal to 1 Grade Grade of attendance within specified education level. Categorical variable Completed education Primary Completed primary school (completed = 1). Binary variable or Binary variable with one of expression the values equal to 1 Secondary Completed secondary school (completed = 1). Binary variable or Binary variable with one of expression the values equal to 1 Postsecondary Completed postsecondary school (completed = 1). Binary variable or Binary variable with one of expression the values equal to 1 Education Grade completed at the highest level. Categorical variable Categorical variable (years) Literate Reads and writes (yes = 1). Binary variable or Binary variable with one of expression the values equal to 1 (continued) 244 Appendix A: Fields, Variable Definitions, and Variable Requirements Field Variable definition Variable requirement Internal check Health and nutrition Health and Variable or set of variables that represents health or Binary variable(s) or Available for men and nutrition nutrition outcomes with nonmissing values for men continuous variable(s) women in the same age and women, boys and girls. range Examples of nutrition indicators are prevalence of underweight for children under five years of age, such as weight for age (underweight), height for age (stunting), weight for height (wasting), the Z-scores coming from the latter; or consumption below the minimum level of dietary energy consumption. Maternal Variable or set of variables that represents maternal Binary variable(s) or Missing values for men and health health outcomes. These variables are only for women continuous variable(s) women outside the of reproductive age (15–49). Examples are prenatal reproductive age (15–49) and postnatal care variables: attended prenatal control visits at health facility, birth attended at a health facility, postnatal control visits at health facility, unmet need for family planning, and so on. Economic opportunities Economic status and work characteristics Employed Employment status. An individual (of legal Binary variable or Binary variable with one of employment age) is considered employed if during expression the values equal to 1 the survey reference period (usually the past week or past month) the individual (a) worked for a wage or salary, (b) worked for profit or family gain, (c) was either employed by a third party or was self-employed for at least one hour, or (d) had a wage job, self- employment, or enterprise but was temporarily absent from work (because of vacation, maternity leave, sick leave, or other type of leave) (employed = 1). Unemployed Unemployment status. Working-age individuals are Binary variable or Binary variable with one of considered unemployed if, during the survey expression the values equal to 1 reference period (usually the past week), they were (a) without work in either paid employment or self- employment, (b) available for work, and (c) actively looking for work (unemployed = 1). Work category Wage worker, self-employed, and so on. The work Categorical variable Categorical variable (that can category variable indicates the type of worker’s take a maximum of XX employment. For example, wage and salaried values) workers, self-employed workers, unpaid family workers. Sometimes, wage workers are divided into private sector and public sector. Agriculture Works in agriculture (yes = 1). Binary variable or Binary variable with one of expression the values equal to 1; with nonmissing values for employed individuals Broad sector Categorical variable indicating the sector of Categorical variable Categorical variable that can employment: agriculture, manufacturing, or services. take a maximum of three values; with nonmissing values for employed individuals (continued) 245 Appendixes Field Variable definition Variable requirement Internal check Sector Categorical variable indicating a finer sector of Categorical variable Categorical variable that can employment classification, such as one-digit ISIC. take a maximum of 10 values; with nonmissing values for employed individuals Detailed sector Categorical variable indicating an even finer sector Categorical variable Categorical variable; with of employment classification, such as four-digit ISIC. nonmissing values for employed individuals Occupation Job occupation (such as professional) for example, Categorical variable Categorical variable; with one-digit ISCO. nonmissing values for employed individuals Formal status Employed in the formal sector (yes = 1). The Binary variable or Binary variable with one of definition of formal work varies. The term informality expression the values equal to 1; with means different things to different people. It refers nonmissing values for here to owners (self-employed) and workers (informal employed individuals salaried). Informal workers are usually those who do not have social security or medical benefits and are therefore unprotected. Formal salaried workers are defined as those who enjoy labor protections. Full time Works full time (yes = 1). Whether a worker is Binary variable or Binary variable with one of employed full time or part time depends on the expression the values equal to 1; with standard hours worked in the week of reference (or the nonmissing values for usual week in the past calendar year), as defined by the employed individuals country’s labor code. Most countries define full time as working at least 35 hours per week, but in some countries, this number increases to 40 or even 48 hours per week. If a full-time variable is not specified but the variable number of hours worked is, ADePT Gender assumes 40 hours is the full-time threshold. Earnings Monthly, weekly, daily, or hourly earnings (consistent Continuous variable with hours of work if specified). Hours Hours worked per day, week, or month (consistent Continuous variable, with earnings). nonnegative Access to productive resources Resources For example, savings account, land. Categorical Categorical variable variable that indicates ownership/control/ management/use arrangements, with mutually exclusive categories: (a) only man, (b) only woman, or (c) both man and woman. Voice, agency, and participation Marriage and fertility Age at first Woman’s age at first marriage (in years). Continuous integer marriage variable Age at first Woman’s age at first birth (in years). Continuous integer birth variable (continued) 246 Appendix A: Fields, Variable Definitions, and Variable Requirements Field Variable definition Variable requirement Internal check Birth date Date of mother’s birth, in CMC format. Continuous variable Interview date Date of the interview, in CMC format. Continuous variable Children’s Date of birth of each child, in CMC format. Continuous variable birth date Agency Agency For example, decision making (female respondent), Dummy variable (0 or 1 outcomes involvement in decisions about food expenditure value) or continuous (wife only, husband only, or both). variable Note: CMC = century-month code; ISIC = International Standard Industrial Classification; ISCO = International Standard Classification of Occupations; p = percentile. 247 Appendix B Demographic and Health Survey Agency Variables This appendix describes five tables and two graphs that explore voice and agency outcomes. DHS module Women Men V VI V VI Expression/question Control over resources Who usually makes decisions about major household purchases? x x x x Who usually decides how your (husband’s/partner’s) earnings will be used: x x x you, your (husband/partner), or you and your (husband/partner) jointly? Who usually decides how the money you earn will be used: you, your x x x x (husband/partner), or you and your (husband/partner) jointly? Would you say that the money that you earn is more than what your x x (husband/partner) earns, less than what he earns, or about the same? Who usually makes decisions about health care for yourself: you, your x x x (husband/partner), you and your (husband/partner) jointly, or someone else? Do you own this or any other house, either alone or jointly with someone x x else? Do you own any land, either alone or jointly with someone else? x x Household owns TV or radio: x x x x Do you listen to the radio at least once a week, less than once a week, or not at all? Do you watch television at least once a week, less than once a week, or not at all? (continued) 249 Appendixes DHS module Women Men V VI V VI Does this household own any livestock, herds, other farm animals, or x x x x poultry? Does any member of this household have a bank account? x x x x Ability to move freely In the past 12 months, how many times have you been away from home for x x x x one or more nights? Who usually makes decisions about visits to your family or relatives? x x x Decision making over family formation Age at first marriage? x x x x How old were you when you first started living with your (husband/partner)? x x x x How old were you when you had sexual intercourse for the very first time? x x x x Can you say no to your (husband/partner) if you do not want to have sexual x x intercourse? If a wife knows her husband has a disease that she can get during sexual x x x x intercourse, is she justified in asking that they use a condom when they have sex? Could you ask your (husband/partner) to use a condom if you wanted him to? x x Would you say that using contraception is mainly your decision, mainly your x x (husband’s/partner’s) decision, or did you decide together? Have you ever heard of (contraception method)? x x x x Which method are you using? x x x x Reason for not using x x Do you know of a place where a person can get female/male condoms? x x If you wanted to, could you yourself get a condom? x x x x If you could go back to the time when you did not have any children and x x x x could choose exactly the number of children to have in your whole life, how many would that be? If you could choose exactly the number of children to have in your whole life, how many would that be? How many of these children would you like to be boys, how many would you x x x x like to be girls, and for how many would the sex not matter? Have you ever used anything or tried in any way to delay or avoid getting x x pregnant? Do you agree with the following statement? “Contraception is women’s x x business and a man should not have to worry about it.” Do you agree with the following statement? “Women who use contraception x x may become promiscuous.” (continued) 250 Appendix B: Demographic and Health Survey Agency Variables DHS module Women Men V VI V VI Freedom from the risk of violence In your opinion, is a husband justified in hitting or beating his wife in the x x x x following situations: if she goes out without telling him, if she neglects the children, if she argues with him, if she refuses to have sex with him, if she burns the food? Knowledge Have you ever heard of a sexually transmitted infection? x x x x Have you ever heard of an illness called AIDS? x x x x Knowledge of transmission methods: Can people reduce their chance of x x x x getting the AIDS virus by having just one uninfected sex partner who has no other sex partners? Can people get the AIDS virus from mosquito bites? Can people reduce their chances of getting the AIDS virus by using a condom every time they have sex? Can people get the AIDS virus by sharing food with a person who has AIDS? Can people get the AIDS virus because of witchcraft or other supernatural means? Is it possible for a healthy-looking person to have the AIDS virus? Do you know of a place where people can go to get tested for the AIDS virus? x x x x Have you ever been tested to see whether you have the AIDS virus? x x x x When you are sick and want to get medical advice or treatment, is it a x problem to get permission to go to the doctor? Do you read a newspaper or magazine at least once a week, less than once a x x x x week, or not at all? Note: AIDS = acquired immune deficiency syndrome; DHS = Demographic and Health Surveys. 251 Appendix C Tests of Statistical Significance Users might be interested in knowing whether the differences between men’s and women’s indicators are statistically significant. Whether differ- ences are significant depends on the distribution of values of the indicators or variables and the number of observations, not simply on the absolute value of the difference itself. Differences that are relatively small in mag- nitude might be statistically significant, whereas differences that are large in magnitude might not be. This appendix explains the basic tests for establishing whether observed differences for men and women are statisti- cally significant. We first discuss tests to compare means and then consider tests to compare distributions. All the inputs for these calculations are produced by ADePT, as the formulas require means, standard deviations, and frequencies. Means Tests The student’s t-test is a widely used statistical method to compare group means. It was developed by the statistician William Gossett, who called himself Student. To perform a t-test, the mean of the variable to be compared should be interpretable and random—that is, a variable whose values change randomly, not a constant. 253 Appendixes The t-test is based on the following assumptions: (a) samples are randomly drawn (b) from normally distributed populations with (c) unknown population variances. Assumption (a) guarantees the absence of selection bias, such that the two groups to compare have a systematic difference due to a nonrandom sampling to select individuals with properties that the researcher prefers or that follow a pattern known but not observable to the analyst. If this were the case, the comparison of means between the nonrandom samples with another one is neither reliable nor generalized. Assumption (b) is the key assumption underlying the t-test. If this assumption is not fulfilled, the sample mean is not an unbiased estimator of central tendency, and the t-test will not be valid. The violation of normality is more problematic in the one-tailed test than in the two-tailed test, since in the former this violation could more easily influence statistical inferences. Nevertheless, thanks to the central limit theorem, the normality assumption is not that problematic. The theorem states that the distribution of a sample mean is approximately normal when its sample size is sufficiently large (n1 + n2 ≥ 30).1 Hence, the t-test can be safely used if the sample size is moderate, except when there are severe outliers. Three types of t-tests can be performed: one-sample t-tests, paired t-tests, and independent sample t-tests. Independent sample t-tests are the ones relevant for the users of ADePT Gender, since the objective is to assess whether the observed mean differences between men and women are statistically significant. Instead, the one-sample t-test2 is used to assess whether the population mean is different from a hypothesized value—usually zero—and the paired t-test3 is used to compare two means on the basis of samples that are matched in some way; in other words, to examine whether the mean of the differences between the pairs is discernible from zero. Hence, the underlying methods of these two types of t-tests are identical and are not accurate for studying gender differences in most cases. Independent Sample t-Test When assessing gender differences, the researcher usually wants to know whether the observed differences in the mean of the variables of interest of the two samples are statistically significant—that is, whether the mean difference of the two groups is discernible from zero in a statistical way, which means that the two-sample means are sufficiently different from each other 254 Appendix C: Tests of Statistical Significance to be declared different, and the difference observed is not because of chance or a data peculiarity. To prove this, an independent sample t-test must be performed, since the two groups of interest (women and men) contain indi- viduals who are not paired or matched in any way and who were selected for the same population and thus exposed to identical conditions. In statistics, the hypothesis to reject (or not) in order to answer the research question is called a null hypothesis (H0). It is contrasted with an alternate hypothesis (Ha). The null hypothesis in an independent two-sample t-test is H0: μF = μM. In other words, the population means of the male and female samples are the same, which is exactly the same to postulate H0: μF − μM = 0. In other words, the mean difference between the male and female samples is zero. The researcher may or may not have expectations about the direction of the findings based on previous theoretical or empirical work. Depending on the design of the study, the researcher should decide the type of alternate hypothesis to formulate: a one-sided hypothesis or a two-sided hypothesis. The former corresponds to a two-tailed test and the latter to a one-tailed test. The two-tailed t-test should be performed if the researcher is interested only in testing whether the population means are equal and does not have an a priori expectation about the direction in which the alternate hypoth- esis should move regarding the null hypothesis. In this case, the alternate hypothesis is Ha: μF ≠ μM. In other words, the population means of the male and female samples are different. Reorganizing the terms, Ha: μF − μM ≠ 0. The mean difference between the male and female samples is different from zero. 255 Appendixes The one-tailed t-test is appropriate if the researcher is interested in whether one mean is larger than the other. The alternate hypothesis could be Ha: μF > μM. In other words, the population mean of the female group is larger than the population mean of the male group. Again, this is exactly the same as Ha: μF − μM > 0. The mean difference between the female and male samples is greater than zero. The null hypothesis is the same for both the two-tailed and one-tailed t-tests. The t-statistic is used to evaluate whether a statistic (for example, a mean) is significantly different from a certain value. Algebraically, it is defined as ˆ − μ0 μ ˆ = tμ , σ where μ ˆ is an estimator of the parameter μ, μ is the value that the param- 0 eter takes under the null hypothesis, and s is the standard error of the estimator μˆ . The t-test statistic applied to compare the means of two popu- lations is μF − μM − 0 ˆ = tμ , σ ˆ where σˆ is the standard deviation of the sampling distributions. Under the normality assumption, the null hypothesis, and equal population vari- ances, the t-statistic follows a student’s t-distribution with n−2 degrees of freedom.4 The t-statistic is calculated as follows: 1. Calculate the mean value for each distribution, that is, men and women. 2. Estimate the variance to obtain the standard deviation. This can be done in two ways. If the variances of the populations from which the samples are drawn are assumed to be equal, then compute the test as 256 Appendix C: Tests of Statistical Significance μF − μ M − 0 ˆ = tμ ~ t ( nF + n M − 2 ) 1 1 σ ˆ pool + nF n M where ( n − 1)σ F + ( nM − 1)σ M 2 2 σ ˆ pool 2 = F . nF + nM − 2 Alternatively, the variances can be assumed to come from populations with different variances. Some authors recommend always using the t-statistic that assumes the population variances are unequal, since in most cases both versions of this test lead to the same result. Some studies argue that the tests used to determine variance equality are unreliable. This decision is up to the researcher. Under this assumption, the t-test is computed as μF − μ M − 0 ˆ = ap tμ ~ t ( estimated degrees of freedom ) . σ 2 ˆF σ ˆM2 + nF n M The user can actually test which assumption is correct, as described in box C.1. Box C.1: Test of Variance Equality for Two Populations The folded form F-test is commonly used to test variance equality. It is also a hypoth- esis test, in which the null hypothesis is that the two populations have the same vari- ( ance H 0 : σ F2 = σM 2 ). The test statistic for the F-test is an F-statistic that is the ratio ( )2 between the larger sample variance σ l and the smaller one σ s . It has an ( ) 2 F-distribution with nl−1 and ns−1 degrees of freedom: 2 σl 2 ~ F ( nl − 1, nS − 1) . σs If the null hypothesis of equal variances is not rejected—the value of the F-statistic is inside the confidence interval with a given level of significance—the pooled variance can be used to obtain the denominator of the t-statistic by taking its square root. The pooled variance is a weighted average of the two sample variances. 257 Appendixes Several methods are used to approximate the degrees of freedom, but the most common is the Satterthwaite method. It is important to keep in mind that the approximation is a real number, not necessarily an integer. Once the test statistic is calculated, it should be determined whether this value is contained in the interval in which the true value of the statistic is within the probability chosen by the researcher—(1−a)100%. If it is, the null hypothesis is not rejected; remember that to calculate the t-statistic, the null hypothesis was assumed to be true. Otherwise, it is rejected, and the differences observed in the mean of men and women samples are statisti- cally significant with a confidence level of (1−a)100%. The confidence level (1−a)100% is the percentage of all possible samples that can be expected to include the true population parameter; a is the level of significance, which is the probability that the true value of the statistic is not contained in the confidence interval, and therefore the null hypothesis is rejected although it should not be—this mistake is called a type 1 error. Usually, a is set at 1 percent, 5 percent, or 10 percent. The confidence interval for the t-statistic calculated with a given level of significance α and for a two-tailed test is Pr ⎡ ⎣ −tα /2,df ≤ t μ ⎦ = 1− α, ˆ ≤ tα /2, df ⎤ where t α /2 and −tα /2 are the critical values of t μ ˆ (limits of the confident interval) that could be drawn from the statistic t table for a α / 2 level of significance and the corresponding degrees of freedom. The region between the critical values in the distribution is called the acceptance region, whereas the rest is called the rejection or critical region (see figure C.1). The confidence interval for the t-statistic calculated with a given level of significance α and for a one-tailed test is Pr ⎡ ⎦ = 1− α ˆ < tα , df ⎤ ⎣t μ or Pr ⎡ ⎦ = 1− α, ˆ > − tα , df ⎤ ⎣t μ depending on the alternative hypothesis formulated. The equation in panel a of figure C.1 corresponds to the example used in the one-tailed t-test section 258 Appendix C: Tests of Statistical Significance Figure C.1: Rejection and Acceptance Regions a. Two-tailed t-test b. One-tailed t-test Ha: μF – μM ≠ 0; α = 5%; df → ∞ Ha: μF – μM > 0; α = 5%; df → ∞ f(t) f(t) t t 0 0 (Ha: μF − μM > 0), since the rejection area is always associated with the alternative hypothesis (see figure C.1). Again, if the t-statistic calculated is not contained in the confidence interval, there is evidence to reject the null hypothesis. Therefore, the mean differences between men and women in the variable of interest are statisti- cally different from zero and thus statistically significant. If a one-tailed test was also evaluated, the differences observed are different from zero, and there is evidence to conclude that one mean is greater than the other—for the example, this would be the mean of the female population. It is worth mentioning that the same inference can be made using the p-value of the statistic, which is the exact level of significance—that is, the lowest level of significance at which a null hypothesis can be rejected. For example, if the researcher knows that the exact probability of the t-statistic is outside the confidence interval, he or she can decide whether to reject the null hypothesis at this level instead of deciding a priori the a . Evidently, if p < a , then the null hypothesis is rejected with a 1−p confidence level and therefore with a 1−a confidence level too. Statistically Different Distributions When a user is interested in knowing whether the differences observed between women and men in a certain categorical variable are statistically different from zero, a chi-square goodness-of-fit test should be performed to compare the two distributions. Given that the comparison is between distri- butions as opposed to a single measure—such as the mean—all the catego- ries into which the data have been divided are used. 259 Appendixes Follow these steps to perform a chi-square goodness-of-fit test: 1. Form a hypothesis about the way in which the distribution of the variable of interest behaves. For example, the observed distribution of women’s educational attainment behaves as the observed distribution of men’s education attainment, as in ADePT table 2a (described in chapter 4). The null and alternative hypotheses of a chi-square test is H0: Oi = Ei. The observed number of cases in each category is equal to the expected number of cases in each category, Ha: Oi ≠ Ei, where Oi is the frequency of occurrence for each category i in which the women’s sample has been grouped, and Ei is the expected frequency of occurrence for each category i that corresponds to the men’s sample, follow- ing the example taken from ADePT table 1a. If the observed number and expected number of cases are different for the two populations, the null hypothesis should be rejected. 2. Calculate the test statistic. Suppose that the data have been grouped in m categories, i = 1, 2, 3, ..., m. Our example of ADePT table 1a has four categories (m = 4): (a) no education, (b) primary complete, (c) secondary complete, and (d) postsecondary complete. The statistic is (Oi − E i )2 m χ2 = ∑ i Ei and has a chi-square distribution with m−1 degrees of freedom if the data are obtained from a random sample, and the expected frequency of each category is at least five. Large values of the statistic lead to a rejection of the null hypothesis, since they indicate that the differ- ence between the observed and expected values is large as well. 260 Appendix C: Tests of Statistical Significance 3. Reject (or accept) the null hypothesis by comparing the value obtained in the second with the critical value of the chi-squared dis- tribution with m−1 degrees of freedom and the chosen level of significance a (concepts explained in the previous section on means tests). If the value of the statistic is greater than the critical value, the null hypothesis should be rejected; otherwise, it should be accepted with a confidence level of (1−α)100%. The chi-square goodness-of-fit test is always a right-tail test. Notes 1. ni is the number of observations in the group i. 2. The one-sample t-test is adequate for those problems that look for evidence to conclude that the mean of the population from which the sample is taken is different from a specified value of interest that could be a standard drawn from the literature. 3. The paired t-test is appropriate for those studies that assess the mean differences between two paired groups: before and after data on a single group of individuals, two variables on the same individual, or a group matched one-to-one to a second group. 4. Numbers of values that are not estimated are free to vary. 261 Appendix D Juhn-Murphy-Pierce Decomposition This appendix describes the mathematical steps for progressing from equa- tion (6.10) to equation (6.11) in chapter 6 for the Juhn-Murphy-Pierce decomposition. The earnings equation (6.10) can be rewritten to extract −1 this information from the error term. It can be rewritten as ε gi = Fg ( pgi | X gi ), where pgi is the percentile of individual i of gender g in the distribution of residuals computed using all workers of gender g; Fg (·) is the cumulative distribution function of residuals for gender g, conditional on observed char- −1 () acteristics; and Fg ⋅ is its inverse (or quantile function). This yields equation (D.1): ln Wgi = β g X gi + Fg−1 ( pgi | X gi ) . (D.1) Next, define a benchmark coefficient, b * (including both constant and education coefficients), and a benchmark distribution of residuals, F*−1(·). In practice, these benchmarks can be chosen in several ways. They can coincide with one of the gender-specific coefficients and distributions, either the male or female one; or, in the same spirit as the pooled Oaxaca- Blinder decomposition, with some “summary” of coefficients and residual distributions—either from a pooled model using all workers (men and women) or from a simple average of gender-specific coefficients and distributions. 263 Appendixes ADePT Gender defines the benchmark as the coefficients and residual distribution from the pooled regression as in equation (6.5) in chapter 6. The intuition behind the Juhn-Murphy-Pierce decomposition is that by using equation (D.1), one can create different counterfactuals based on the returns to observed and unobserved skills. In particular, one can simulate three types of counterfactuals, each with an expression for males and one for females: Outcomes if only observed characteristics vary between groups: YFX = β * X Fi + F *−1 ( pFi | X Fi )wM X = β * X Mi + F *−1 ( pMi | X Mi ). (D.2) Outcomes if only observed characteristics and returns to observed char- acteristics vary between groups: obs wM = β M X Mi + F *−1 ( pMi |X Mi ) , (D.3) obs wF = β F X Fi + F *−1 ( pFi |X Fi ) . Outcomes if observed characteristics, returns to observed characteristics, and unobservable characteristics and their returns (that is, the distribution of residuals) all vary between groups: unobs wM = β M X Mi + F −1 ( pMi |X Mi ) , (D.4) w unobs F = β F X Fi + F −1 ( pFi |X Fi ) . unobs Naturally, wg coincides with the original gender-specific wage equation. From this last step, the derivation of equation (6.11) is straightforward. 264 Appendix E Mathematical Derivation of the Mills Ratio Variable Included in the Mincer Equation This appendix shows the mathematical derivation of the reservation wage that is used for the conceptual development of the selection procedure developed by Heckman and discussed in chapter 7. The contribution of Gronau’s model and its use by Heckman to move to the econometric Heckit model can be explained using the household time allocation model intro- duced in chapter 5. Figure E.1 plots the maximization problem that results in the hours of work supplied. The figure plots three different wages, which are indicated by the slope of the different dashed lines. At the lowest wage (the slope of the small dashed line), the woman prefers not to work, as she can obtain a level of utility Y that is higher than any other utility obtained by working a positive number of hours. At the highest wage (the slope of the dotted line), the woman prefers to work a positive number of hours and to reach utility level Z. The reservation wage is that which makes a woman indifferent to working or not working. In the figure, this is indicated by the slope of the large dashed line, for which any small increase in the wage (the slope of the budget restriction pivoting upward) will induce her to work, and any smaller wage will make her prefer not to work. 265 Appendixes Figure E.1: Graphic Representation of the Reservation Wage Money Nonmarket income + w ×16 Utility level Z Nonmarket income Utility level Y Utility level X 0 16 Hours in household work 16 0 Time at paid work Mathematically, the market wage is determined by an equation for a subsample of the working-age population: ln wi = Xib + ¨i, and the reservation wage can be assumed to depend on productivity vari- ables and variables that affect the utility of staying at home, R ln w Fi = XFi β R + ZFiψ R + ξFi , R where w F is the reservation wage of woman i, Xi is a vector of variables capturing productivity (or the potential wage), and Zi is a vector of variables capturing the utility of staying home. Here, for simplicity in the notation, the vector XF β R = α 0R + ρ R SF + β1 R X F + β1 R 2 XF of equation (6.1). The subin- dex F is also dropped for simplicity. A woman would work if and only if the utility of (at least) one hour of work is greater than the utility of not working at all. The wage that makes a woman indifferent to working or staying home is called the reservation wage. Women work if their potential wage, wi, is higher than the reservation wage Pr [ working i ] > 0 ⇔ ln wi − ln wiR > 0, 266 Appendix E: Mathematical Derivation of the Mills Ratio after replacement by respective equations, this is equivalent to Pr [ working i ] = Pr ⎣ ⎡ y* ⎦ ≡ ( Xi β + ∈i ) − ( Xi β + Ziψ + ξi ) i = 1⎤ R R ≡ Xi β − X i β R − Z iψ R + ∈i − ξ i ≡ Xiϕ + Ziα + ε i > 0, but it can see only employed or selected workers, where y* = 1 indicates that the worker is employed and otherwise is 0. The regression conditional expectation function for y over the selected sample is i = 1) = Xβ + E ( ∈ E ( wi | X, y* i |X, y i = 1) . * It is clear that for b to be consistent, we need the usual ordinary least squares’ assumption that ∈i be mean independent of X; ∈i also needs to be mean independent of the selection rule. The Heckman selection model tackles this problem by obtaining a consistent estimate of the term E (∈i| X, s = 1) and including it in the regression. The model is defined by an outcome equation (which contains the parameters of interest to the researcher) and a selection equation: ln wi = Xi β + ∈i , i = 1] = Xiϕ + Ziδ + ε i . Pr[y* Here, the notation of equations (7.2) and (7.3) is used. The model is estimated under the following assumptions: * • yi and Zi are observed in the full sample. 1 • yi is observed only over the selected sample, or when y* i =1 . • Both the outcome and the selection equations are “good” models, in that the regressors are exogenous: E (∈i |X, Z ) = E ( ε i|X, Z ) = 0. • The error term of the selection equation is distributed as a standard normal, e ∼ N (0,1). • E ( ε i|ε i ) = γ ε (u is not mean independent from e, and the relation- ship between the two random errors is linear; this in turn implies ei = g ei + x, where x is a random disturbance). 267 Appendixes In this setup, the conditional expectation function becomes i = 1) = Xβ + E ( ε i|X, ε > − Zδ ) , E ( w|X, y* = Xβ + E (γ ε + ξ|X, ε > −Zδ ) , = Xβ + γ E ( ε|ε > −Zδ ) . Having assumed a functional form for ε allows one to use the properties of the truncated normal distribution and obtain φ ( −Zδ ) φ ( Zδ ) i = 1) = Xβ + γ E ( w | X, y* = Xβ + γ = Xβ + γ λ ( Zδ ) , 1 − Φ ( −Zδ ) Φ ( Zδ ) where l(.) is the inverse Mills ratio, a nonlinear monotone decreasing func- tion of the probability that an observation is selected. It provides a func- tional form for the term that was omitted from the naive regression of y on X in the selected sample. It might be useful to mention that the case in which g = 0 (there is no correlation between the error terms of the outcome and selection equa- tions) amounts to a scenario where the selection is random, or the selection process is ignorable; whereas in the case of g ≠ 0, if X and Z are completely independently distributed, omitting the last term will not result in the endogeneity of X, and will not impede identification of b (exogenous or deterministic selection).2 Heckman proposed a way to estimate the missing term of the regression using a two-step method. First, obtain consistent estimates of d by running a probit model of s on Z, Pr ( yi* = 1 | Z ) = Φ ( Zδ ) ; then, plug δ into l and obtain consistent estimates of b and γ by running ordinary least squares on the selected sample. It must be noted that the usual ordinary least squares variance–covariance matrix is no longer adequate in this case, given that (a) the error term of the outcome equation is intrinsically heterocedastic because of selection, and (b) one of the regressors, λ ( Z δ ), is an estimate itself. Heckman provides a consistent estimator of the covariance matrix in this case, but asymptotic efficiency is lost. 268 Appendix E: Mathematical Derivation of the Mills Ratio Notes 1. Xi can be observed either in the full sample or for only the selected observations; the crucial part is that the outcome equation can be esti- mated only over the restricted sample, for which y* i = 1. 2. In practice, it is common for X and Z to be partly or even completely overlapping. The higher the correlation among X and Z, the greater the multicollinearity between X and the regressor containing Z. At the estimation stage, this will cause problems with identifying the coeffi- cients. In the extreme case in which X and Z are indeed identical, identification relies solely on the nonlinearity of the Mills ratio. 269 Index Boxes, figures, maps, notes, and tables are indicated by b, f, m, n, and t following the page number. A household-centered framework, 3–15, abortion, 92 6–7b, 6f access to resources, 132, 133t if-conditions, applying to output, 38–39 access vs. use of resources, 73 installing, 18–19 household vs. individual, 72–73 labor market analysis, 159–234. See also outcome variables, 71–74 labor market analysis ownership or management of resources, launching, 19–20 73–74 output, 30, 37–39 accounting-identity model, 214 overview, xi–xii, xiv ADePT Gender software parameter setting, 36 advantages of, 237–39 projects, 36–37 analysis procedure, 20–21 report generation, 29–30 country gender diagnostic, 87–157. resources, xiv, 10–11t See also country gender diagnostic standard errors, adding to output, 37–38 Data Browser, 33 tables data preparation, 43–86. See also data custom table generation, 39–41 preparation selecting for generation, 26–28 dataset specification, 21–23 variables, 30–35 earnings equation, model specifications definitions, 243–47 for, 176–77b deleting, 35 frequencies, adding to output, 37–38 expressions, 35 graphs, selecting for generation, 26–28 generating, 33–34 271 Index ADePT Gender software (cont.) BCG (Bacillus Calmette–Guérin) mapping, 23–26 vaccinations, 58 replacing, 34–35 Beaman, Lori, 50 requirements, 243–47 Beblo, Miriam, 227 working with, xiv, 17–41 Becker, Gary, 7, 146, 206, 213, 214, 228n5 age Ben-Porath, Yoram, 214 elderly, age bracket for, 53 Bergmann, Barbara R., 212 at first birth, 74–75, 134, 136–37, 137t Bertrand, Marianne, 205, 213 at first marriage, 74, 134, 136–37, bias, 201n10, 222 137t, 138f Blank, Rebecca M., 205 variable, 52–53, 77–78 Blau, Francine D., 146, 205, 213, 219 youth, age bracket for, 53, 78 Blinder, Alan S., 179, 185. See also Oaxaca- agency Blinder methodology control over resources, 76, 134, 139 Borjas, George, 182 decision making over family formation, Bourguignon, François, 154 76, 132, 134 Brazil, intrahousehold allocation model in, development and, 3 153 direct vs. indirect, 77 broad sector, 67, 68 freedom from risk of violence, 76, 134 Browning, Martin, 154 freedom of movement, 76, 132, 134 Buchave, Helle, xv in household-centered framework, Buvinic, Mayra, xv 6–7b, 9–10 intrinsic vs. extrinsic, 77 C outcome variables, 44, 74–77 Canada, part-time employment policy influence, 76 threshold in, 68 universal vs. context specific, 77 Cardoso, Ana Rute, 211 voice in society, 76, 135 casual workers, 64 WDR’s five dimensions of, 75–77 Chiappori, Pierre-André, 154 agreeableness, 228n12 child dependency ratio, 95 agriculture, 67 child labor, 60 Altonji, Joseph G., 205 child nutrition, 3 Armenia Chile, life-cycle disaggregation in, fertility rate in, 140n4 130, 131f population pyramid for, 101, 103f chi-square goodness-of-fit test, 260–61 Assets Project, 77 Christofides, Louis N., 227 Atencio, Andrea, xiv Cluster sampling, 49b Atkinson, Anthony B., 168 coefficient of variation, 165 Atkinson index, 163, 167, 168–69 cognitive skills, 220 Australia, part-time employment collective action, 136 threshold in, 68 collective models for household availability for work, 61 decisions, 152–53 Azerbaijan, population pyramid for, compensating wage differentials 101, 103f model, 216 conscientiousness, 228n12 B consumption, 51–52 Babcock, Linda, 211 context specific agency, 77 Banerjee, Abhijit V., 213 contraception, 135b Bardasi, Elena, xv contributing family workers, defined, 64 272 Index cooperative models for household “credential” effects, 218 decisions, 152–53 credit, access to, 71, 73 country gender diagnostic, 87–157 crime rates, 105 country context, 132–39 customer discrimination, 210 concepts, 133–36 content, 132–33 D interpreting results, 136–39, 136f, Data Browser, 33 137–38t, 138–39f data preparation, 43–86 demographic and regional dataset specification, 21–23 characteristics, 91–104 household surveys, 45–47, 46t concepts, 91–95 main variables, 47–55 content, 91 age, 52–53, 77–78 demographic transitions, 93 custom category, 54–55 household composition, 93–95, definitions, 243–47 100–101, 100–101t elderly, age bracket for, 53 interpreting results, 95–103, 96f, gender, 52 98–99t head of household, 54, 80n11 missing women, 92–93 household, defined, 49–50 population pyramids, 93, 94f, household information, 49–52 101–3, 102–4f individual information, 52–54 economic opportunities, 114–32 marital status, 53–54 access to resources, 132, 133t place of residence, 50–51 concepts, 115–19 sample weights, 47–48 content, 114–15 welfare aggregate, 51–52 disaggregating by life-cycle stage and youth, age bracket for, 53, 78 education, 130–32, 130–31f outcome variables, 55–77 earnings, 119, 128–29, 129t access to resources, 71–74 interpreting results, 119–32 agency, 44, 74–77 job/employment characteristics, definitions, 243–47 118–19, 123–28, 124f, 125–26t, economic opportunities, 44, 59–74 127f, 128t education, 55–57, 58b, 78 labor market participation, 115–17, health and nutrition, 58–59 119–23, 121m, 121t, 122f, 123t human capital, 44, 55–59 education, 107b, 130–32, 130–31f informality, 65–66, 66t goals for, 5 labor income, 70–71 health, 107b labor status, 59–62, 62–63b human capital, 104–14 marriage and fertility, 74–75 concepts, 105–9 maternal health, 59 content, 104–5 occupation, 67–68 interpreting results, 109–14, 109t, sector of employment, 67–68 111f, 112–13t work characteristics, 63–65 interpreting results, 89–144 working age, 79 technical notes, 145–57 work intensity, 68–70 household labor supply model, 146–47 parameters, 77–79 intrahousehold allocation model, sampling design, 48–49b 152–54 Deaton, Angus, 49b labor supply model with household decision making over family formation, production, 147–52, 148t, 149f, 151f 76, 132, 134 273 Index decomposition methodologies, 162 E Demographic and Health Survey (DHS) earnings, 119, 128–29, 129t agency variables, 249–51 outcome variables, 70–71 contraception indicators, 135b earnings decompositions, 175–200 household survey data, 43, 46 concepts, 178–92, 180t, 188f, 190f maternal health data, 59 contents, 175–78, 176–77b outcome variables, 59 detailed decomposition, 184–85 demographic and regional characteristics, interpreting results, 192–200 91–104 Juhn-Murphy-Pierce methodology, 174, concepts, 91–95 177–78, 187–89, 188f, 196–98, 196t, content, 91 197f, 263–64 demographic transitions, 93 Ñopo methodology, 177, 189–92, 190f, household composition, 93–95, 198–200, 199t 100–101, 100–101t Oaxaca-Blinder methodology, 177–78, interpreting results, 95–103, 96f, 179–83, 185–87, 192–95, 193t, 98–99t 194f, 195t missing women, 92–93 pooled decomposition, 183–84 population pyramids, 93, 94f, 101–3, reference group switching, 183–84 102–4f threefold decomposition, 185 demographic transitions, 93 earnings equations, 213–22, 215f, 219b De Paoli, Georgia, xiv econometric concerns, 220–22 dependency ratio, 94–95, 100–101, goodness of fit, 220–21 100–101t heterocedasticity, 221–22 detailed decomposition, 181 model specification, 220–21 detailed decomposition of earnings, 184–85 omitted variable bias, 222 detailed sector, 67, 68 partial equilibrium, 222 deterministic selection, 224 earnings inequality, 162–75 DHS. See Demographic and concepts, 164–71, 167f, 170b Health Survey content, 162–64 difference-in-differences approach, 153, gender earnings inequality, 169–71, 170b, 155n8 173–75, 174t, 175f Dillon, Andrew, 50 interpreting results, 171–75 direct vs. indirect agency, 77 overall and within-gender, 163, 164–69, disaggregation by life-cycle stage and 167f, 171–73, 171–72t, 173f education, 130–32, 130–31f economic opportunities, 114–32 discrimination theories, 205–13 access to resources, 132, 133t customer discrimination, 210 concepts, 115–19 employee discrimination, 209–10 content, 114–15 employer discrimination, 207–8, 209f disaggregating by life-cycle stage and overcrowding model resulting in education, 130–32, 130–31f discrimination, 212–13, 212f earnings, 119, 128–29, 129t statistical discrimination model, 211–12 in household-centered framework, 8–9 taste discrimination, 206–7 interpreting results, 119–32 Doeringer, Peter B., 213 job/employment characteristics, 118–19, DPT (diptheria, pertussis, and tetanus) 123–28, 124f, 125–26t, 127f, 128t vaccinations, 58 labor market participation, 115–17, Duflo, Esther, 153, 213 119–23, 121m, 121t, 122f, 123t dynamic human capital theory, 214 outcome variables, 44, 59–74 274 Index education extraversion, 228n12 country gender diagnostic, 107b, 130–32, extrinsic vs. intrinsic agency, 77 130–31f disaggregation by, 130–32, 130–31f F in earnings equations, 216, 218 family planning, 135b earnings inequality and, 164, 176b, family vs. household, 50 189–90 Ferber, Marianne A., 146, 205, 213 endowments and, 7–8 Ferreras Carreras, Elena, xiv fertility rate and, 92 fertility, 74–75, 92, 140n4 gender equality’s impact on, 3, 5 Finland, part-time employment gross enrollment rate (GER), 106 threshold in, 68 human capital linked to, 105–6 Firpo, Sergio, 178 net enrollment rate (NER), 106, 109–10, first birth, age at, 74–75, 134, 136–37, 137t 109t, 111f first marriage, age at, 74, 134, 136–37, outcome variables, 55–57, 58b, 78 137t, 138f returns to, 108, 217 Flabbi, Luca, 211 elderly former Yugoslav Republic of Macedonia, age bracket for, 53 Roma women’s employment in, 92 dependency ratio, 95 Fortin, Nicole, 178, 221 employee discrimination, 209–10 Foster, James, 165 employees, defined, 64 freedom from risk of violence, 76, 134 employer discrimination, 207–8, 209f freedom of movement, 76, 132, 134 employers, defined, 64 frequencies, adding to output, 37–38 employment full time employment, 70 calculation of, 116 Fuwa, Nobuhiko, 54 characteristics, 118–19, 123–28, 124f, 125–26t, 127f, 128t G defined, 60, 62b, 63t Gaddis, Isis, xiv segregation, 118 Gagliarducci, Stefano, 211 vulnerability, 64 García, Gisela, xiv empowerment, 9–10. See also agency Gardeazabal, Javier, 184 endowments gender variable, 52. See also ADePT Gender concepts, 105–9 software; country gender diagnostic; content, 104–5 labor market analysis in household-centered framework, 7–8 generalized entropy index of inequality, 168 human capital theory, 213–14 GER (gross enrollment rate), 106, 113 interpreting results, 109–14, 109t, 111f, Germany, demographic transition in, 93, 94f 112–13t Gini coefficient, 166 outcome variables, 44, 55–59 glass ceiling, 196, 213 England, Paula, 146 Golan, Jenifer, xiv Enterprise Surveys, 46 goodness of fit, 220–21 errors graphs, selecting for generation, 26–28 in report generation, 29–30 gross enrollment rate (GER), 106, 113 standard errors added to output, 37–38 Grown, Caren, xv ethnicity, 80–81n15 expenditure variable, 51–52 H expressions for variables, 35 Haimovich, Francisco, xiv external shocks, 92 head of household 275 Index head of household (cont.) content, 104–5 decision making and, 154n2 education, 55–57, 58b poverty and, 140n8 health and nutrition, 58–59 variables, 54, 80n11 interpreting results, 109–14, 109t, 111f, health 112–13t country gender diagnostic, 107b maternal health, 59, 105, 112t, 114 endowments and, 7–8 outcome variables, 44, 55–59 gender equality’s impact on, 3, 5 theory, 213–14 outcome variables, 58–59 Human Opportunity Index, 105, 165 Heckit procedure, 186, 201n12, 223, 226–28 I Heckman, James, 186, 214, 216, 221 if-conditions, applying to output, 38–39 Heckman two-step selection correction IHSN (International Household Survey model, 226–28 Network), 46 heterocedasticity, 221–22 IHSN–World Bank Gender Data Navigator high-skilled workers, 164, 174 database, 46 HIV, 92, 114 ILO. See International Labour hours of work per week Organization country gender diagnostic, 126, 128t immunizations. See vaccinations in statistical discrimination model, incidental truncation, 223b 211–12 income. See also earnings variable, 69 labor income, 70–71 household-centered framework, 3–15, 6f in welfare aggregate variable, 51–52 agency in, 6–7b, 9–10 income effect, 141n13 economic opportunities in, 8–9 independent sample t-test, 254–59, 261n2 endowments in, 7–8 index number problem, 184 “household-dwelling” concept, 79–80n5 indirect vs. direct agency, 77 household ID variable, 50 individual information variables, 52–54 household labor supply model, 145, industry variable, 67–68, 176b 146–47 inequality aversion, 168 households infant mortality, 59, 108 collective models for household informal employment decisions, 152–53 country gender diagnostic, 128, 128t composition of, 93–95, 100–101, outcome variables, 65–66, 66t 100–101t International Classification of Status in cooperative models for household Employment, 64 decisions, 152–53 International Conference on Labour defined, 49–50, 79–80n5 Statisticians, 60, 62b, 63t head of household. See head of household International Household Survey Network information variables, 49–52 (IHSN), 46 time allocation in, 147–50, 148t, 149f International Labour Organization (ILO) unitary model for household decisions, definitions, 2 152–53 on employment sectors, 67 household weights, 47 on informal employment, 65–66, 66t “housekeeping” concept, 79–80n5 on occupation classifications, 68 human capital, 104–14. See also Part-Time Work Convention, 68 endowments standards for measuring employment, concepts, 105–9 60–62, 62b, 63t 276 Index International Standard Classification of overcrowding model resulting in Education (ISCED), 56, 58b, 81n18 discrimination, 212–13, 212f International Standard Classification of statistical discrimination model, Occupations (ISCO), 68, 82n28 211–12 International Standard Industrial taste discrimination, 206–7 Classification (ISIC), 67, 82n25 earnings decompositions, 175–200 intersectoral margin, 65 concepts, 178–92, 180t, 188f, 190f intrafirm margin, 65 contents, 175–78, 176–77b intrahousehold allocation model, 152–54 detailed decomposition, 184–85 intrahousehold decision model, 145 interpreting results, 192–200 intrinsic vs. extrinsic agency, 77 Juhn-Murphy-Pierce methodology, 174, 177–78, 187–89, 188f, 196–98, J 196t, 197f, 263–64 Japan Ñopo methodology, 177, 189–92, 190f, demographic transition in, 93, 94f 198–200, 199t life-cycle disaggregation in, 130, 130f Oaxaca-Blinder methodology, 177–78, part-time employment threshold in, 68 179–83, 185–87, 192–95, 193t, job characteristics 194f, 195t country gender diagnostic, 118–19, pooled decomposition, 183–84 123–28, 124f, 125–26t, 127f, 128t reference group switching, 183–84 outcome variables, 63–65, 67–68 threefold decomposition, 185 job-relevant skills, 220 earnings equations, 213–22, 215f, 219b Juhn-Murphy-Pierce methodology, 174, econometric concerns, 220–22 177–78, 187–89, 188f, 196–98, 196t, goodness of fit, 220–21 197f, 263–64 heterocedasticity, 221–22 model specification, 220–21 K omitted variable bias, 222 Kahn, Lawrence M., 219 partial equilibrium, 222 Kautz, Tim, 221 earnings inequality, 162–75 Kenya, population pyramid for, 103, 104f concepts, 164–71, 167f, 170b Koolwal, Gayatry, xiv content, 162–64 Korea, Republic of gender earnings inequality, 169–71, demographic transition in, 93, 94f 170b, 173–75, 174t, 175f fertility rate in, 140n4 interpreting results, 171–75 overall and within-gender, 163–69, L 167f, 171–73, 171–72t, 173f labor force participation interpreting results, 161–200 country gender diagnostic, 115–17, earnings decompositions, 175–200 119–23, 121m, 121t, 122f, 123t earnings inequality, 162–75 gender differences, 5 nonrandom selection into labor force, Labor Force Surveys, 43 222–28, 223b, 225f labor income, 70–71. See also earnings Heckman two-step selection correction labor market analysis, 159–234 model, 226–28 discrimination theories, 205–13 technical notes, 205–34 customer discrimination, 210 discrimination theories, 205–13 employee discrimination, 209–10 earnings equations, 213–22, 215f, 219b employer discrimination, nonrandom selection into labor force, 207–8, 209f 222–28, 223b, 225f 277 Index labor specialization, 150–51 welfare aggregate, 51–52 labor status variables, 59–62, 62–63b youth, age bracket for, 53, 78 labor supply model with household Mancini, Giulia, xiv production, 147–52, 148t, mapping variables, 23–26 149f, 151f marital status land rights, 73 in earnings equations, 220, 221 Lao People’s Democratic Republic, labor main variables, 53–54 supply model with household outcome variables, 74–75 production in, 147–49, 148t market work outside home, 147 Laschever, Sara, 211 maternal health, 59, 105, 112t, 114 leisure, 147 maternal mortality, 59, 92 Lemieux, Thomas, 178 Matulevich, Eliana Rubiano, xv life-cycle stage, disaggregation by, 130–32, means tests, 253–59, 259f 130–31f members of producer cooperatives, life expectancy, 59 defined, 64 line of equality, 166 Mexico, PROGRESA program in, 80n14 literacy, 55, 81n17 Mills ratio variable, 186, 195, 201n13, 227, Living Standards Measurement Surveys, 265–68, 266f 43, 46 Mincer, Jacob, 146, 213, 218, 228n5 Lochner, Lance J., 214, 216 Mincer equation, 178, 183–84, 213–14, Lorenz curve, 166, 167f 215f, 221, 265–68, 266f low birth weights, 105 missing women, 92–93 lower secondary education, 56 MMR (measles, mumps, and rubella) low-skilled workers, 164, 174, 226 vaccinations, 58 Lundberg, Shelly J., 153 Montenegro, Claudio E., 213 movement, freedom of, 76, 132, 134 M Mueller, Gerrit, 221 Macedonia, former Yugoslav Republic of, Mulligan, Casey B., 227 Roma women’s employment in, 92 main variables, 47–55 N age, 52–53, 77–78 neoclassical economics model, 146 custom category, 54–55 Nepal definitions, 243–47 agency measures in, 133, 136–37, elderly, age bracket for, 53 137–38t, 139 gender, 52 earnings inequality in, 192–93, 193t, head of household, 54, 80n11 194f, 195t, 198, 199t, 200 household information, 49–52 education in, 109–10, 109t household, defined, 49–50 Living Standards Survey, 90 place of residence, 50–51 maternal health in, 112t, 114 welfare aggregate, 51–52 population pyramid for, 101, 102f individual information, 52–54 net enrollment rate (NER), 106, 109–10, age, 52–53 109t, 111f, 113 gender, 52 Neuman, Shoshana, 186, 195 head of household, 54, 80n11 neuroticism, 228n12 marital status, 53–54 New Zealand, part-time employment marital status, 53–54 threshold in, 68 place of residence, 50–51 Niger, demographic transition in, sample weights, 47–48 93, 94f 278 Index noncoverage, 48b fertility, 74–75 nonrandom sampling, 223b health and nutrition, 58–59 nonrandom selection into labor force, human capital, 44, 55–59 222–28, 223b, 225f education, 55–57, 58b nonresponse, 48b health and nutrition, 58–59 Ñopo, Hugo R., 189, 191 maternal health, 59 Ñopo methodology, 177, 189–92, 190f, informal employment, 65–66, 66t 198–200, 199t labor income, 70–71 normal hours of work variable, 69 labor status, 59–62, 62–63b Norway, part-time employment threshold marital status, 74–75 in, 68 maternal health, 59, 105, 112t, 114 number of weeks worked, 69 occupation, 67–68 Numerics by Stata, 1 sector of employment, 67–68 nutrition voice, 249–51 country gender diagnostic, 108 work characteristics, 63–65 outcome variables, 58–59 working age, 79 work intensity, 68–70 O out of labor force, 61 Oaxaca, Ronald L., 179, 183, 186, 195 output. See also outcome variables Oaxaca-Blinder methodology, 177–78, examination of, 30 179–83, 185–87, 192–95, 193t, 194f, frequencies, adding to output, 37–38 195t if-conditions, applying to output, 38–39 Oaxaca-Ransom methodology, 184 standard errors, adding to output, occupation 37–38 classification of, 68 outworkers, 64 earnings equation, model specifications overcrowding model resulting in for, 176b discrimination, 212–13, 212f gender differences, 8 overlapping vulnerabilities, 98 outcome variables, 63–65, 67–68 own-account workers, defined, 64, 65 occupational segregation, 118 omitted variable bias, 222 P openness to experience, 228n12 paid employees, defined, 63–64 Organisation for Economic Co-operation Panama and Development (OECD), 220 access to resources in, 132, 133t outcome variables, 55–77 agency measures in, 132–33 access to resources, 71–74 dependency ratio in, 100–101, agency, 44, 74–77, 249–51 100–101t definitions, 243–47 earnings in, 128–29, 129t economic opportunities, 44, 59–74 earnings inequality in, 171–75, 171–72t, access to resources, 71–74 173f, 174t, 175f, 196–97, 196t, 197f informal employment, 65–66, 66t education classification in, 58 labor income, 70–71 employment characteristics in, 124–25, labor status, 59–62, 62–63b 124f, 125t occupation, 67–68 head of household and poverty in, 54, sector of employment, 67–68 95–96, 96f, 98–99, 98t work characteristics, 63–65 hours of work per week, 126, 128, 128t work intensity, 68–70 informal employment in, 126, education, 55–57, 58b, 78 128, 128t 279 Index Panama (cont.) R labor force participation rate in, 120, Rawlsian welfare function, 168 121t, 122, 123t reference group switching, 183–84 Living Standard Measurement Study, region (variable), 50. See also demographic 90–91 and regional characteristics part-time employment in, 128, 128t report generation, 29–30 rural households in, 97, 98t Republic of Korea. See Korea, Republic of parameter setting, 36 returns to education, 108, 201n10 Pareto outcome, 155n7 Rothstein, Donna S., 211 partial equilibrium, 222 Rubinstein, Yona, 227 part-time employment, 68, 128, 128t rural households Part-Time Work Convention (ILO), 68 education investment in, 110 Paserman, M. Daniele, 211 employment characteristics in, 118 Patrinos, Harry Anthony, 213 gender differences, 97, 98t, 103 Piore, Michael J., 213 Russian Federation, population pyramid for, PISA (Programme for International 103, 104f Student Assessment), 81n17 place of residence, 50–51, 176b S Plug, Erik J. S., 221 sample weights, 47–48 policy influence, 76 sampling design, 48–49b Pollak, Robert A., 153 sanitation, 108 Polycarpou, Alexandros, 227 Santamaria, Julieth, xiv polygyny, 53–54 scheduling of work hours, 70 pooled decomposition of earnings, seasonal workers, 64 183–84 secondary education population pyramids, 93, 94f, 101–3, country gender diagnostic, 109t, 102–4f 110, 111f positive selection, 224 outcome variables, 56, 58b, 78 postnatal care, 59 sector of employment, 67–68 postsecondary nontertiary education, 57 selection bias, 223b, 224–25, 225f poverty self-employment head of household and, 54, 95–96, 96f, country gender diagnostic, 125, 128t 98–99, 98t, 140n8 defined, 63 working poverty rate, 117 earnings inequality in, 172, 172t, poverty line, 51, 80n6 174–75, 175f prenatal care, 59, 92, 112–13t, 114 severely disadvantaged populations, 91 primary education sex-selective abortions, 92 country gender diagnostic, 109–10, 109t, “sheepskin” effects, 218 110, 111f short-term employment, 64 outcome variables, 56, 58b, 78 Sinha, Nistha, xiv, xv production of household goods, 147 Skills Toward Employment and Programme for International Assessment of Productivity (STEP), 81n17, 220 Adult Competencies, 220 social norms, 134, 136 Programme for International Student socioemotional skills, 220 Assessment (PISA), 81n17 South Africa, intrahousehold allocation PROGRESA program (Mexico), 80n14 model in, 153 property rights, 73 standard errors, adding to output, 37–38 Psacharopoulos, George, 213 Stata format, 1, 21, 43 280 Index statistical discrimination model, 211–12 Science Study (TIMSS), 81n17 Statistical Package for the Social Sciences Tunisia, life-cycle disaggregation in, 130, (SPSS), 1, 21, 43 131f, 132 statistical significance tests, 253–61, 257b chi-square goodness-of-fit test, 260–61 U independent sample t-test, 254–59, Ugidos, Arantza, 184 261n2 underemployment, 61, 81n24 means tests, 253–59, 259f unemployment statistically different distributions, calculation of, 116 259–61 defined, 60–61 STEP (Skills Toward Employment and gender differences, 5, 8, 120 Productivity), 81n17, 220 surveys to measure, 44 “sticky floor” effects, 196 youth, 78 stopping behavior, 92, 140n4 unitary model for household decisions, stratification sampling, 49b 152–53 stunting, 114 United Kingdom, intrahousehold survey frame, 48b allocation model in, 153 System of National Accounts, 62b, 63t, 67 United Nations, 2, 89, 137 United Nations Educational, Scientific, T and Cultural Organization tables (UNESCO), 56 custom table generation, 39–41 United Nations World Population selecting for generation, 26–28 Statistics, 78 taste discrimination, 206–7 United States technical notes labor force participation in, country gender diagnostic, 145–57 227–28 household labor supply model, part-time employment 146–47 threshold in, 68 intrahousehold allocation model, wage inequality in, 187 152–54 universal vs. context specific labor supply model with household agency, 77 production, 147–52, 148t, upper secondary education, 56 149f, 151f urban (variable), 50 labor market analysis, 205–34 usual hours of work, 69 discrimination theories, 205–13 earnings equations, 213–22, 215f, 219b V nonrandom selection into labor force, vaccinations, 58, 108, 114 222–28, 223b, 225f variables, 30–35. See also main variables; tertiary education outcome variables country gender diagnostic, 109t, 110 definitions, 243–47 outcome variables, 57 deleting, 35 Theil index, 163, 168 expressions, 35 Thomas, Duncan, 153 generating, 33–34 threefold decomposition of earnings, 185 mapping, 23–26 time allocation in households, 147–50, replacing, 34–35 148t, 149f requirements, 243–47 Todd, Petra E., 214, 216 violence, freedom from, 76, 134 Trends in International Mathematics and virtuous or vicious cycles, 92 281 Index vitamin deficiencies, 114 work intensity, 68–70 voice, 3, 6–7b, 76, 135. See also agency World Bank Vrachimis, Konstantinos, 227 Development Data Group, xi vulnerable groups, 97–100, 98t Gender Cross Cutting Solution Area, xi W gender database, 46 wage rates, 70 Human Opportunity Index, 165 wage workers, 63–64. See also earnings on virtuous or vicious cycles, 92 Waldfogel, Jane, 221 World Development Report 2012: Gender Wales, Terence J., 153 Equality and Development warnings in report generation, 29 (World Bank) wasting, 114 on agency, 74, 75–77 wealth indexes, 52 household-centered gender equality welfare aggregate, 51–52 framework, xiv, 2, 3–4, 6, 10, welfare indexes, 52 90, 237 Winkler, Anne E., 146, 205, 213 Winter-Ebmer, Rudolf, 211 Y Women’s Empowerment in Agriculture youth, age bracket for, 53, 78 Index, 77 Yun, Myeong-Su, 184 work characteristics variables, 63–65 working age, 79 Z working poor, 117 z-conditional demands, 154 working poverty rate, 117 Zorigtbaatar, Chimedkham, xiv 282 ECO-AUDIT Environmental Benefits Statement The World Bank Group is committed to reducing its environmental footprint. In support of this commitment, we leverage electronic publishing options and print-on-demand technology, which is located in regional hubs worldwide. Together, these initiatives enable print runs to be lowered and shipping distances decreased, resulting in reduced paper consumption, chemical use, greenhouse gas emissions, and waste. We follow the recommended standards for paper use set by the Green Press Initiative. The majority of our books are printed on Forest Stewardship Council (FSC)–certified paper, with nearly all containing 50–100 percent recycled content. The recycled fiber in our book paper is either unbleached or bleached using totally chlorine-free (TCF), processed chlorine-free (PCF), or enhanced elemental chlorine-free (EECF) processes. More information about the Bank’s environmental philosophy can be found at http://www.worldbank.org/corporateresponsibility.