72461 New Century, Old Disparities GENDER AND ETHNIC EARNINGS GAPS IN LATIN AMERICA AND THE CARIBBEAN Hugo Ñopo New Century, Old Disparities New Century, Old Disparities GENDER AND ETHNIC EARNINGS GAPS IN LATIN AMERICA AND THE CARIBBEAN Hugo Ñopo a copublication of the inter-american development bank and the world bank © 2012 Inter-American Development Bank 1300 New York Avenue, NW Washington DC 20577 Telephone: 202-623-1000 Internet: www.iadb.org E-mail: res@iadb.org 1234 15 14 13 12 A copublication of the Inter-American Development Bank and The World Bank. The Inter-American Development Bank The World Bank 1300 New York Avenue, NW 1818 H Street, NW Washington, DC 20577 Washington, DC 20433 The findings, interpretations, and conclusions expressed in this work do not necessarily reflect the views of the Inter-American Development Bank or its Board of Governors; The World Bank or its Board of Executive Directors; or the governments they represent. The Inter-American Development Bank and The World Bank do 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 Inter-American Develop- ment Bank or The World Bank concerning the legal status of any territory or the endorsement or acceptance of such boundaries. Rights and Permissions The material in this work is subject to copyright. Because The Inter-American Development Bank and The World Bank encourage dissemination of their knowledge, this work may be reproduced, in whole or in part, for noncommercial purposes as long as full attribution to this work is given. For permission to reproduce any part of this work for commercial purposes, please send a request with complete information to the Copyright Clearance Center Inc., 222 Rosewood Drive, Danvers, MA 01923, USA; telephone: 978-750-8400; fax: 978-750-4470; Internet: www.copyright.com. All other queries on rights and licenses, including subsidiary rights, should be addressed to the Office of the Publisher, The World Bank, 1818 H Street NW, Washington, DC 20433, USA; fax: 202-522- 2422; e-mail: pubrights@worldbank.org. ISBN (paper): 978-0-8213-8686-6 ISBN (electronic): 978-0-8213-9496-0 DOI: 10.1596/978-0-8213-8686-6 Cover: Drew Fasick of the Fasick Group, Inc. Library of Congress Cataloging-in-Publication Data New century, old disparities : gender and ethnic earnings gaps in Latin America and the Caribbean /Hugo Ñopo. p. cm. — (Latin American development forum series) Includes bibliographical references and index. ISBN 978-0-8213-8686-6 — ISBN 978-0-8213-9496-0 (electronic) 1. Sex discrimination against women—Economic aspects—Latin America. 2. Ethnic relations— Economic aspects. 3. Discrimination—Economic aspects—Latin America. I. Ñopo, Hugo. HQ1237.5.L29N39 2012 305.80098—dc23 2012014249 Latin American Development Forum Series This series was created in 2003 to promote debate, disseminate informa- tion and analysis, and convey the excitement and complexity of the most topical issues in economic and social development in Latin America and the Caribbean. It is sponsored by the Inter-American Development Bank, the United Nations Economic Commission for Latin America and the Caribbean, and the World Bank. The manuscripts chosen for publication represent the highest quality in each institution’s research and activity output and have been selected for their relevance to the academic com- munity, policy makers, researchers, and interested readers. Advisory Committee Members Alicia Bárcena Ibarra, Executive Secretary, Economic Commission for Latin America and the Caribbean, United Nations Inés Bustillo, Director, Washington Office, Economic Commission for Latin America and the Caribbean, United Nations Tito Cordella, Deputy Chief Economist, Latin America and the Caribbean Region, World Bank Augusto de la Torre, Chief Economist, Latin America and the Caribbean Region, World Bank Santiago Levy, Vice President for Sectors and Knowledge, Inter-American Development Bank Eduardo Lora, Chief Economist (a.i.) and General Manager, Research Department, Inter-American Development Bank Luis Servén, Senior Adviser, Development Economics Vice Presidency, World Bank Andrés Velasco, Cieplan, Chile v Titles in the Latin American Development Forum Series New Century, Old Disparities: Gender and Ethnic Earnings Gaps in Latin America and the Caribbean (2012) by Hugo Ñopo Does What You Export Matter? In Search of Empirical Guidance for Industrial Policies (2012) by Daniel Lederman and William F. Maloney From Right to Reality: Incentives, Labor Markets, and the Challenge of Achieving Universal Social Protection in Latin America and the Caribbean (2012) by Helena Ribe, David Robalino, and Ian Walker Breeding Latin American Tigers: Operational Principles for Rehabilitating Industrial Policies (2011) by Robert Devlin and Graciela Moguillansky New Policies for Mandatory Defined Contribution Pensions: Indus- trial Organization Models and Investment Products (2010) by Gregorio Impavido, Esperanza Lasagabaster, and Manuel García-Huitrón The Quality of Life in Latin American Cities: Markets and Perception (2010) by Eduardo Lora, Andrew Powell, Bernard M. S. van Praag, and Pablo Sanguinetti, editors Discrimination in Latin America: An Economic Perspective (2010) by Hugo Ñopo, Alberto Chong, and Andrea Moro, editors The Promise of Early Childhood Development in Latin America and the Caribbean (2010) by Emiliana Vegas and Lucrecia Santibáñez Job Creation in Latin America and the Caribbean: Trends and Policy Challenges (2009) by Carmen Pagés, Gaëlle Pierre, and Stefano Scarpetta China’s and India’s Challenge to Latin America: Opportunity or Threat? (2009) by Daniel Lederman, Marcelo Olarreaga, and Guillermo E. Perry, editors Does the Investment Climate Matter? Microeconomic Foundations of Growth in Latin America (2009) by Pablo Fajnzylber, Jose Luis Guasch, and J. Humberto López, editors vii viii titles in the latin american development forum series Measuring Inequality of Opportunities in Latin America and the Carib- bean (2009) by Ricardo de Paes Barros, Francisco H. G. Ferreira, José R. Molinas Vega, and Jaime Saavedra Chanduvi The Impact of Private Sector Participation in Infrastructure: Lights, Shadows, and the Road Ahead (2008) by Luis Andres, Jose Luis Guasch, Thomas Haven, and Vivien Foster Remittances and Development: Lessons from Latin America (2008) by Pablo Fajnzylber and J. Humberto López, editors Fiscal Policy, Stabilization, and Growth: Prudence or Abstinence? (2007) by Guillermo Perry, Luis Servén, and Rodrigo Suescún, editors Raising Student Learning in Latin America: Challenges for the 21st Cen- tury (2007) by Emiliana Vegas and Jenny Petrow Investor Protection and Corporate Governance: Firm-Level Evidence across Latin America (2007) by Alberto Chong and Florencio López-de- Silanes, editors Natural Resources: Neither Curse nor Destiny (2007) by Daniel Lederman and William F. Maloney, editors The State of State Reform in Latin America (2006) by Eduardo Lora, editor Emerging Capital Markets and Globalization: The Latin American Expe- rience (2006) by Augusto de la Torre and Sergio L. Schmukler Beyond Survival: Protecting Households from Health Shocks in Latin America (2006) by Cristian C. Baeza and Truman G. Packard Beyond Reforms: Structural Dynamics and Macroeconomic Vulnerability (2005) by José Antonio Ocampo, editor Privatization in Latin America: Myths and Reality (2005) by Alberto Chong and Florencio López-de-Silanes, editors Keeping the Promise of Social Security in Latin America (2004) by Indermit S. Gill, Truman G. Packard, and Juan Yermo Lessons from NAFTA: For Latin America and the Caribbean (2004) by Daniel Lederman, William F. Maloney, and Luis Servén The Limits of Stabilization: Infrastructure, Public Deficits, and Growth in Latin America (2003) by William Easterly and Luis Servén, editors Globalization and Development: A Latin American and Caribbean Per- spective (2003) by José Antonio Ocampo and Juan Martin, editors Is Geography Destiny? Lessons from Latin America (2003) by John Luke Gallup, Alejandro Gaviria, and Eduardo Lora About the Author Hugo Ñopo, a Peruvian national, is a lead research economist in Education at the Inter-American Development Bank (IDB), based in Bogotá, Colombia. Before joining the IDB, he was an assistant professor at Middlebury College, research affiliate at Group for the Analysis of Development (GRADE), and adviser at the Ministry of Labor and Social Promotion in Peru. His research agenda includes gender and racial inequalities in educational systems, labor markets and the access to public services, impact evaluation of public policies, and trust and reciprocity among economic agents. His research has been published in various specialized academic journals and books. Currently, he is also a research affiliate at the Institute for the Study of Labor (IZA). ix Contents Foreword xxv Acknowledgments xxvii Abbreviations xxviii PART I: OVERVIEW, METHODOLOGY, AND DATA 1 1 Overview 3 Recent Changes on the Situation of Women and Ethnic Minorities 4 Overview of the Book 6 Notes 7 References 7 2 Methodology and Data 9 The Blinder-Oaxaca Decomposition 10 Methodology for This Book: An Extension of the Blinder-Oaxaca Decomposition 10 Data 14 Notes 19 References 19 3 Gender Differences in Education in Latin America and the Caribbean: Girls Outpacing Boys 21 Changes in the Gender Education Gap 23 Decomposing Changes in the Gender Education Gap 24 Gender Differences in Attendance and Attainment among Children of School Age 32 Notes 35 References 35 xi xii contents PART II: GENDER EARNINGS GAPS 37 4 More Schooling, Lower Earnings: Women’s Earnings in Latin America and the Caribbean 39 What Does the Literature Show? 40 How Do Male and Female Workers Differ? 41 Linkages between Unexplained Gender Earnings Gaps and Macroeconomic, Social, and Governance Indicators 57 How Did Differences between Male and Female Workers Change between Circa 1992 and Circa 2007? 64 Notes 80 References 80 5 The Mostly Unexplained Gender Earnings Gap: Peru 1997–2009 83 How Do Male and Female Workers Differ? 84 The Role of Individual Characteristics in Explaining the Gender Earnings Gap 89 Exploring the Unexplained Component of the Gender Earnings Gap 92 Changes in Women’s Participation and Unemployment Rates 97 Notes 100 References 100 6 Is Gender Segregation in the Workplace Responsible for Earnings Gaps? Mexico 1994–2004 101 What Does the Literature Show? 102 Measuring Occupational and Hierarchical Segregation 103 The Role of Individual Characteristics in Explaining the Earnings Gap 105 Notes 112 References 112 7 Low Participation by Women, Heavy Overtime by Men: Chile 1992–2009 115 What Does the Literature Show? 115 How Do Male and Female Workers Differ? 118 The Role of Individual Characteristics in Explaining the Gender Earnings Gap 126 contents xiii Exploring the Unexplained Component of the Gender Earnings Gap 132 Note 135 References 135 8 The Resilient Earnings Gap: Colombia 1994–2006 137 What Does the Literature Show? 138 How Do Male and Female Workers Differ? 139 The Role of Individual Characteristics in Explaining the Gender Earnings Gap 147 Exploring the Unexplained Component of the Gender Earnings Gap 155 Notes 160 References 160 9 Promoting Equality in the Country with the Largest Earnings Gaps in the Region: Brazil 1996–2006 163 What Does the Literature Show? 163 The Role of Individual Characteristics in Explaining the Earnings Gap 165 Exploring the Unexplained Component of the Gender Earnings Gap 169 Notes 172 References 172 10 Gender Earnings Gaps in a Country with a Large Indigenous Population: Ecuador 2003–07 175 What Does the Literature Show? 176 How Do Male and Female Workers Differ? 176 The Role of Individual Characteristics in Explaining the Gender Earnings Gap 177 Exploring the Unexplained Component of the Gender Earnings Gap 179 Note 181 References 181 11 Gender Earnings Gaps in Central American Countries, 1997–2006 183 What Does the Literature Show? 184 How Do Male and Female Workers Differ? 185 xiv contents The Role of Individual Characteristics in Explaining the Earnings Gap 196 Exploring the Unexplained Component of the Gender Earnings Gap 203 Notes 212 References 212 12 The Understudied Caribbean: Barbados (2004) and Jamaica (2003) 215 What Does the Literature Show? 215 History and Development of Barbados and Jamaica 216 Barbados: Men in the Middle, Women at Both Ends 217 The Role of Individual Characteristics in Explaining the Gender Earnings Gap 222 Exploring the Unexplained Component of the Gender Earnings Gap 225 Jamaica: Women in the Middle, Men at Both Ends 231 The Role of Individual Characteristics in Explaining the Gender Earnings Gap 235 Exploring the Unexplained Component of the Gender Earnings Gap 238 Summary 240 Notes 242 References 242 PART III: ETHNIC EARNINGS GAPS 243 13 Overlapping Disadvantages: Ethnicity and Earnings Gaps in Latin America 245 What Does the Literature Show? 246 How Do Ethnic Minorities and Nonminorities in the Work Force Differ? 248 The Role of Individual Characteristics in Explaining the Ethnic Earnings Gap 248 Exploring the Unexplained Component of the Ethnic Earnings Gap 258 Notes 262 References 262 14 Promoting Ethnic Equality: Brazil 1996–2006 265 What Does the Literature Show? 265 contents xv How Do Ethnic Minorities and Nonminorities in the Work Force Differ? 268 The Role of Individual Characteristics in Explaining the Ethnic Earnings Gap 272 Exploring the Unexplained Component of the Ethnic Earnings Gap 273 Notes 277 References 278 15 No Good Jobs and Lower Earnings: Ecuador 2000–07 281 What Does the Literature Show? 281 How Do Ethnic Minorities and Nonminorities in the Work Force Differ? 283 The Role of Individual Characteristics in Explaining the Ethnic Earnings Gap 285 Exploring the Unexplained Component of the Ethnic Earnings Gap 289 Note 291 References 291 16 Ethnic Earnings Gaps for Large Minorities: Guatemala 2000–06 293 How Do Ethnic Minorities and Nonminorities in the Work Force Differ? 294 The Role of Individual Characteristics in Explaining the Ethnic Earnings Gap 296 Exploring the Unexplained Component of the Ethnic Earnings Gap 296 Note 300 References 300 PART IV: POLICY OPTIONS 301 17 Policy Options 303 Investing in Education Early in Life 304 Boosting Productivity and Reducing Labor Market Segregation 305 Fostering a More Equitable Division of Household Responsibilities 305 xvi contents Diminishing Stereotyping 306 References 307 Index 309 Figures 3.1 Average Years of Education of Men and Women, Born 1940–84, and Education Gender Gap in Labor Force in Latin America and the Caribbean 23 3.2 Educational Attainment of Men and Women in Labor Force in Latin America and the Caribbean Born 1940–84 26 3.3 Decomposition of Educational Gender Gap in Latin America and the Caribbean, by Educational Level for Cohorts Born in 1940 and 1984 27 3.4 Decomposition of Difference in Educational Gender Gap between Youngest and Oldest Cohort in Latin America and the Caribbean, by Educational Level 28 3.5 Decomposition of Change in Educational Gender Gap in Latin America and the Caribbean, by Educational Level and Country for Cohorts Born 1940–84 29 3.6 Decomposition of Change in Educational Gender Gap in Latin America and the Caribbean, by Component and Educational Level for Cohorts Born 1940–84 30 3.7 Decomposition of Changes in Educational Gender Gap in Latin America and the Caribbean, by Component and Country 31 3.8 School Attendance Rates in Bolivia, Guatemala, Mexico, and Peru, by Gender, Age, and Per Capita Household Income Quintile, Circa 2003 33 3.9 Average Years of Educational Attainment in Bolivia, Guatemala, Mexico, and Peru, by Gender, Age, and Per Capita Household Income Quintile, Circa 2003 34 4.1 Decomposition of Gender Earnings Gap in Latin America and the Caribbean, by Country, Circa 2007 after Controlling for Demographic and Job Characteristics 51 4.2 Confidence Intervals for Unexplained Gender Earnings Gap in Latin America and the Caribbean, after Controlling for Demographic and Job Characteristics, Circa 2007 52 contents xvii 4.3 Unexplained Gender Earnings Gap in Latin America and the Caribbean, by Percentiles of Earnings Distribution, Circa 2007 56 4.4 Female Employment in Industry versus Unexplained Gender Earnings Gaps, Circa 2007 63 4.5 Female Legislators, Senior Officials, and Managers versus Unexplained Gender Earnings Gaps, Circa 2007 64 4.6 Labor Market Liberalization Index versus Unexplained Gender Earnings Gaps, Circa 2007 65 4.7 Confidence Intervals for Unexplained Gender Earnings Gap in Latin America and the Caribbean after Controlling for Demographic and Job Characteristics, Circa 1992 and 2007 72 4.8 Unexplained Gender Earnings Gap in Latin America and the Caribbean after Controlling for Observable and Job Characteristics, by Percentiles of Earnings Distribution, Circa 1992 and 2007 75 4.9 Confidence Intervals for Unexplained Gender Earnings Gap in Latin America and the Caribbean by Cohort, Circa 1992 and 2007 77 4.10 Confidence Intervals for Original and Unexplained Gender Earnings Gaps in Latin America and the Caribbean by Country, Circa 1992 and 2007 79 5.1 Average Years of Education of Men and Women in Peru’s Labor Force, 1997–2009 87 5.2 Educational Levels of Men and Women in Peru’s Labor Force, 1997–2009 88 5.3 Gender Gap in Hourly Earnings in Peru, 1997–2009 89 5.4 Decomposition of Gender Earnings Gap in Peru after Controlling for Demographic Characteristics, 1997–2009 90 5.5 Decomposition of Gender Earnings Gap in Peru after Controlling for Demographic and Job Characteristics, 1997–2009 91 5.6 Unexplained Gender Earnings Gap in Peru after Controlling for Demographic Characteristics, by Percentiles of Earnings Distribution, 1997–2009 96 5.7 Unexplained Gender Earnings Gap in Peru after Controlling for Demographic and Job Characteristics, by Percentiles of Earnings Distribution, 1997–2009 97 5.8 Confidence Intervals for Unexplained Gender Earnings Gap in Peru after Controlling for Demographic Characteristics, 1997–2009 98 xviii contents 5.9 Confidence Intervals for Unexplained Gender Earnings Gap in Peru after Controlling for Demographic and Job Characteristics, 1997–2009 98 5.10 Labor Force Participation Rates of Men and Women in Peru, 1997–2009 99 5.11 Unemployment Rates of Men and Women in Peru, 1997–2009 100 6.1 Estimated Counterfactual Duncan Indexes of Occupational Segregation in Mexico, 1994–2004 106 6.2 Estimated Counterfactual Duncan Indexes of Hierarchical Segregation in Mexico, 1994–2004 108 6.3 Estimated Counterfactual Gender Earnings Gaps in Mexico, 1994–2004 109 6.4 Estimated Changes in Gender Earnings Gap as a Result of Changes in Occupational and Hierarchical Segregation in Mexico, 1994–2004 111 7.1 Average Years of Education of Men and Women in Chile’s Labor Force, 1992–2009 119 7.2 Percentage of Men and Women in Chile’s Labor Force with University Degrees, 1992–2009 120 7.3 Percentage of Men and Women in Chile’s Labor Force with Less Than Secondary Education, 1992–2009 121 7.4 Labor Force Participation Rates of Men and Women in Chile, 1992–2009 121 7.5 Unemployment Rates of Men and Women in Chile, 1992–2009 122 7.6 Unemployment Rates of Men and Women in Chile, by Educational Level, 1992–2009 123 7.7 Average Years at Same Job by Men and Women in Chile, 2000–09 124 7.8 Average Weekly Hours Worked by Men and Women in Chile, 1992–2009 124 7.9 Average Weekly Hours Worked by Men and Women in Chile by Educational Level, 1992–2009 126 7.10 Gender Gap in Hourly Earnings in Chile, 1992–2009 127 7.11 Decomposition of Gender Earnings Gap in Chile after Controlling for Demographic and Job Characteristics, 1992–2009 130 7.12 Decomposition of Gender Earnings Gap in Chile after Controlling for Age, Marital Status, Education, and Years in Same Occupation, 2000–09 131 contents xix 7.13 Confidence Intervals for Unexplained Gender Earnings Gap in Chile, 1992–2009 133 7.14 Unexplained Gender Earnings Gap in Chile, by Percentiles of Earnings Distribution, 1992–2009 134 8.1 Decomposition of Gender Earnings Gap in Colombia after Controlling for Demographic and Job Characteristics, 1994–2006 158 8.2 Gender Earnings Gaps in Colombia after Controlling for Demographic and Job Characteristics, by Percentiles of Earnings Distribution, 1994–2006 159 9.1 Decomposition of Gender Earnings Gaps in Brazil after Controlling for Demographic and Job Characteristics, 1996–2006 168 9.2 Originial and Unexplained Gender Earnings Gap in Brazil, by Percentiles of Earnings Distribution, 1996–2006 171 10.1 Decomposition of Gender Earnings Gap in Ecuador, 2003–07 178 10.2 Unexplained Gender Earnings Gap in Ecuador after Controlling for Demographic and Job Characteristics, by Percentile of Earnings Distribution, 2003–07 180 11.1 Gender Earnings Gap in Central American Countries, Circa 1997–2006 197 11.2 Decomposition of Gender Earnings Gap in Guatemala after Controlling for Age, Marital Status, and Education, 2000–06 202 11.3 Confidence Intervals for Unexplained Gender Earnings Gap in Costa Rica, El Salvador, Honduras, and Nicaragua after Controlling for Demographic and Job Characteristics, 1995–2007 203 11.4 Unexplained Gender Earnings Gap in Costa Rica after Controlling for Demographic and Job Characteristics, by Percentile of Earnings Distribution, 2006 205 11.5 Unexplained Gender Earnings Gap in El Salvador after Controlling for Demographic and Job Characteristics, by Percentile of Earnings Distribution, 2005 206 11.6 Unexplained Gender Earnings Gap in Guatemala after Controlling for Demographic and Job Characteristics, by Percentile of Earnings Distribution, 2000–06 207 11.7 Unexplained Gender Earnings Gap in Honduras after Controlling for Demographic and Job Characteristics, by Percentile of Earnings Distribution, 2007 208 xx contents 11.8 Unexplained Gender Earnings Gap in Nicaragua after Controlling for Demographic and Job Characteristics, by Percentile of Earnings Distribution, 2005 209 11.9 Confidence Intervals for Unexplained Earnings Gaps in Central America after Controlling for Demographic Characteristics, Circa 2006 210 12.1 Distribution of Weekly Earnings of Men and Women in Barbados, by Earnings Interval, 2004 218 12.2 Women’s Participation in the Labor Force in Barbados, by Earnings Interval, 2004 218 12.3 Proportion of Unmatched Women in Barbados, by Earnings Interval, 2004 225 12.4 Unexplained Gender Earnings Gap in Jamaica, after Controlling for Demographic Characteristics, by Percentile of Earnings Distribution, 2003 239 13.1 Decomposition of Ethnic Earnings Gap in Selected Countries in Latin American after Controlling for Demographic and Job Characteristics, Circa 2005 257 13.2 Confidence Intervals for Unexplained Ethnic Earnings Gap in Latin America after Controlling for Demographic and Job Characteristics, Circa 2005 259 13.3 Unexplained Ethnic Earnings Gap in Latin America after Controlling for Demographic and Job Characteristics, by Percentile of Earnings Distribution, Circa 2005 261 14.1 Decomposition of Ethnic Earnings Gap in Brazil, 1996–2006 273 14.2 Original and Unexplained Ethnic Earnings Gap in Brazil after Matching, by Percentile of Earnings Distribution, 1996–2006 276 15.1 Decomposition of Ethnic Earnings Gap in Ecuador, 2003–07 288 15.2 Unexplained Ethnic Earnings Gap in Ecuador after Controlling for Demographic and Job Characteristics, by Percentile of Earnings Distribution, 2003–07 290 16.1 Real Monthly Earnings of Indigenous and Nonindigenous Workers in Guatemala, 2000–06 294 16.2 Decomposition of Ethnic Earnings Gap in Guatemala after Controlling for Demographic and Job Characteristics, 2000–06 297 contents xxi 16.3 Ethnic Earnings Gap in Guatemala after Controlling for Demographic and Job Characteristics, by Percentile of Earnings Distribution, 2000–06 298 Tables 2.1 Household Survey Data Used, by Country and Chapter 15 2.2 Criteria for Classifying Ethnic Groups as “Minorities,� by Country 18 3.1 Gender Gap in Education in Latin America and the Caribbean for Cohorts Born in 1940 and 1984, by Country 25 4.1 Demographic and Job Characteristics and Relative Hourly Earnings of Men and Women in Latin America and the Caribbean, Circa 2007 42 4.2 Decomposition of Gender Earnings Gap in Latin America and the Caribbean after Controlling for Demographic Characteristics, Circa 2007 45 4.3 Decomposition of Gender Earnings Gap in Latin America and the Caribbean after Controlling for Demographic and Job Characteristics, Circa 2007 47 4.4 Original and Unexplained Components of Gender Earnings Gap in Latin America and the Caribbean by Country, Circa 2007 49 4.5 Correlation between Gender Earnings Gap and Economic Indicators in Latin America and the Caribbean, Circa 2007 59 4.6 Relative Hourly Earnings for Men and Women in Latin America and the Caribbean by Demographic and Job Characteristics, Circa 1992 and 2007 66 4.7 Demographic and Job Characteristics of Men and Women in Latin America and the Caribbean, Circa 1992 and 2007 68 4.8 Decomposition of Gender Earnings Gap in Latin America and the Caribbean after Controlling for Demographic and Job Characteristics, Circa 1992 and 2007 70 4.9 Decomposition of Changes in Unexplained Gender Earnings Gap in Latin America and the Caribbean between Circa 1992 and 2007 74 4.10 Unexplained Gender Earnings Gap in Latin America and the Caribbean by Cohort and Demographic and Job Characteristics, Circa 2007 78 xxii contents 5.1 Demographic and Job Characteristics and Relative Earnings of Men and Women in Peru’s Labor Force, 1997–2009 85 5.2 Demographic and Job Characteristics of Matched and Unmatched Samples of Men and Women in Peru’s Labor Force, 1997–2009 93 6.1 Average Duncan Index of Occupational and Hierarchical Segregation in Mexico, by Demographic and Job Characteristics, 1994–2004 103 7.1 Demographic and Job Characteristics of Matched and Unmatched Samples of Men and Women in Chile’s Labor Force, 1992–2009 128 8.1 Relative Hourly Earnings of Men and Women in Colombia, 1994–2006 140 8.2 Demographic and Job Characteristics of Men and Women in Colombia’s Labor Force, 1994–2006 144 8.3 Decomposition of Gender Earnings Gap in Colombia after Controlling for Demographic Characteristics, 1994–2006 148 8.4 Decomposition of Gender Earnings Gap in Colombia after Controlling for Demographic and Job Characteristics, 1994–2006 152 8.5 Demographic and Job Characteristics of Matched and Unmatched Samples of Men and Women in Colombia’s Labor Force, 2002–06 156 9.1 Demographic and Job Characteristics of Matched and Unmatched Samples of Men and Women in Brazil’s Labor Force, 1996 and 2006 166 9.2 Original and Unexplained Gender Earnings Gap in Brazil, by Demographic and Job Characteristics, 1996 and 2006 170 10.1 Educational Attainment by Men and Women in Ecuador’s Labor Force, 2003 and 2007 176 10.2 Average Hourly Earnings for Indigenous and Nonindigenous Men and Women in Ecuador, 2003–07 177 11.1 Relative Hourly Earnings of Men and Women in Central American Countries, by Demographic and Job Characteristics, Circa 1997–2006 186 contents xxiii 11.2 Demographic and Job Characteristics of Central American Countries, 1997, 2001, and 2006 191 11.3 Decomposition of Gender Earnings Gap in Central America after Controlling for Demographic Characteristics, Circa 1997 198 11.4 Decomposition of Gender Earnings Gaps in Central American Countries after Controlling for Demographic Characteristics, Various Years 200 12.1 Demographic and Job Characteristics and Relative Earnings of Men and Women in Labor Force in Barbados, 2004 219 12.2 Decomposition of Earnings Gap in Barbados after Controlling for Demographic Characteristics, 2004 223 12.3 Decomposition of Gender Earnings Gap in Barbados after Controlling for Demographic and Job Characteristics, 2004 226 12.4 Unexplained Gender Earnings Gap in Barbados after Controlling for Demographic and Job Characteristics, 2004 227 12.5 Demographic and Job Characteristics and Relative Hourly Earnings of Men and Women in Jamaica’s Labor Force, 2003 233 12.6 Decomposition of Gender Earnings Gap in Jamaica after Controlling for Demographic Characteristics, 2003 236 12.7 Decomposition of Gender Earnings Gap in Jamaica after Controlling for Demographic and Job Characteristics, 2003 237 13.1 Demographic and Job Characteristics and Relative Earnings of Nonminority and Minority Workers in Latin America, Circa 2005 249 13.2 Decomposition of Ethnic Earnings Gap in Latin America after Controlling for Demographic Characteristics, Circa 2005 253 13.3 Decomposition of Ethnic Earnings Gap in Latin America after Controlling for Demographic, Job, and Full Set of Characteristics, Circa 2005 255 13.4 Decomposition of Ethnic Earnings Gap by Demographic and Job Characteristics in Selected Countries in Latin America, Circa 2005 256 xxiv contents 14.1 Demographic and Job Characteristics of Matched and Unmatched Samples of Whites and Nonwhites in Brazil, 1996 and 2006 269 14.2 Original and Unexplained Ethnic Earnings Gaps in Brazil, by Demographic and Job Characteristics, 1996 and 2006 275 15.1 Ethnic Minorities in Ecuador, by Gender, 2003–07 284 15.2 Educational Attainment in Ecuador’s Labor Force, by Gender and Minority Status, 2003 and 2007 286 15.3 Occupational Distribution in Ecuador, by Gender and Minority Status, 2003 and 2007 287 16.1 Highest Educational Level Begun or Completed by Indigenous and Nonindigenous Workers in Guatemala, 2000, 2004, 2006 295 16.2 Unexplained Ethnic Earnings Gap in Guatemala after Controlling for Demographic Characteristics, 2000–06 298 Foreword Socioeconomic inequality is a complex issue that has occupied human thought over the ages. Historically, overcoming inequality has been the battle cry of numerous revolutions and uprisings. In modern times, the study of inequality has been elevated to a major academic enterprise. Among social scientists, economists bring a unique perspective to this important topic. Their perspective stems from two interrelated features of economic analysis. The first feature is economic theories of inequality, which derive from a common core of insights about the motives of eco- nomic agents. These theories include explanations of how inequality can arise from individuals’ decisions about investment in human capital and from discrimination against particular demographic groups. The second feature is the body of sophisticated statistical and econometric methodolo- gies for measuring inequality and its components. In the best tradition of economics, empirical methodology is informed by economic models of inequality and discrimination. Latin America and the Caribbean provide a rich environment for studying social inequality, because historical inequalities along gender and ethnic lines persist, despite positive indicators of economic develop- ment. The extent of inequality and its probable causes vary widely across the many countries in the region. Among the many dimensions of socioeconomic status that one can consider—health, education, earnings—Hugo Ñopo’s book places major emphasis on labor market earnings. The book adopts a sophisticated econometric methodology for measuring earnings gaps and applies it con- sistently across and within countries to measure gender and racial or ethnic differences. The analysis includes a dynamic dimension that sheds light on the evolution of earnings gaps over time. The book offers impor- tant insights on economic and political strategies that could be adopted to reduce inequality. As such, it is a must for any academic or policy maker interested in understanding and correcting inequality, with respect to not only Latin America and the Caribbean but also anywhere in the world. Ronald L. Oaxaca University of Arizona and Institute for the Study of Labor (IZA) xxv Acknowledgments The development of the methodology that is the basis of this book benefited from comments and suggestions by Gadi Barlevy, Fabio Caldieraro, Alberto Chong, Juan Jose Diaz, Mauricio Drelichman, Libertad Gonzalez, Luojia Hu, Zsolt Macskasi, Rosa Matzkin, Bruce Meyer, Lyndon Moore, Andrew Morrison, Dale Mortensen, Jaime Saavedra, Jim Sullivan, Chris Taber, and Maximo Torero. Comments from Luz Karime Abadía, Diego �ngel-Urdinola, Martin Benavides, Matias Busso, Anna Crespo, Mario Cuevas, Chico Ferreira, Miguel Jaramillo, Julia Johannsen, Liuba Kogan, Jorge Lavarreda, Osmel Manzano, Natalia Millan, Ana Maria Muñoz, Maria Beatriz Orlando, Claudia Piras, Rocio Ribero, Cynthia Sanborn, Sergei Soares, Ernesto Stein, and Renos Vakis on different chapters were invaluable. The book also benefited from comments at presentations made at the Latin American and Caribbean Economic Association Meetings, the Latin American Meeting of the Econometric Society, the European Associa- tion of Labour Economist Meetings, the Midwest Econometric Group Meet- ings, the Joint Labor/Public Economics Seminar at Northwestern University, the Group for the Analysis of Development (GRADE), the Institute for the Study of Labor (IZA), Centro de Investigación y Docencia Económica (CIDE), Instituto Tecnológico Autónomo de Mexico (ITAM), the Inter-American Development Bank, the World Bank, Middlebury College, the Universidad de los Andes, the Universidad del Rosario, Universidad Javeriana, Universidad del Pacifico, and Fedesarrollo. Many of the chapters’ coauthors also contrib- uted comments and provided research assistance on other chapters. Felipe Balcázar, Deidre Ciliento, Cristina Gomez, Lucas Higuera, John Jessup, Alejandra Jimenez, Melisa Morales, and Georgina Pizzolito pro- vided valuable research assistance at different stages of this project. The able editorial work of Rita Funaro, Barbara Karni, John Smith, and Janice Tuten is also acknowledged. Centro de Estudios Distributivos, Laborales y Sociales (CEDLAS) at the Universidad de La Plata and Sociometro at the Inter-American Development Bank generously provided the harmonized data sources for this book. This book was possible thanks to the generous support I received from my supervisors, Eduardo Lora and Marcelo Cabrol, at different stages of this project. Thank you very much! Olga Aguilar, Irma Ugaz, Miski Ñopo, and Maria Ñopo—the women in my life—provided inspiration, support, guidance, and love. xxvii Abbreviations CASEN Encuesta de Caracterización Socioeconómico Nacional (National Socioeconomic Characterization Survey; Chile) CEO chief executive officer CLFSS Continuous Labor Force Sample Survey (Barbados) ENAHO Encuesta Nacional de Hogares (National Household Survey; Peru) ENCOVI Encuesta Nacional de Condiciones de Vida (National Survey of Living Conditions; Guatemala) ENEI Encuesta Nacional de Empleo e Ingresos (National Survey of Employment and Income; Guatemala) ENEMDU Encuesta de Empleo, Desempleo, y Subempleo (Survey of Employment, Unemployment, and Underemployment; Ecuador) ENEU Encuesta Nacional de Empleo Urbano (National Survey of Urban Employment; Mexico) GDP gross domestic product INEC Instituto Nacional de Estadísticas y Censos de Ecuador (National Institute of Statistics and Census of Ecuador) ISCO International Standard Classification of Occupations ISIC International Standard Industrial Classification PNAD Pesquisa Nacional por Amostra de Domicilios (National Household Sample Survey; Brazil) UNICEF United Nations International Children’s Fund xxviii Part I Overview, Methodology, and Data 1 Overview Despite sustained economic growth at the end of the 20th and the begin- ning of the 21st century, Latin America and the Caribbean still faces high inequality and weak indicators of well-being among certain population groups. Women, people of African ancestry, and indigenous peoples are often at the bottom of the income distribution. Growth in gross domes- tic product, expansion of labor force participation, and (some) increase in formal sector real earnings (ILO 2007) have not been sufficient to remove barriers to access to sustainable income-generating opportunities for these groups (Paes de Barros et al. 2009), among whom unemploy- ment and underemployment rates remain high and the quality of jobs has diminished (Márquez et al. 2007). An increasing share of workers has no access to health or pension benefits, turnover rates have increased, and temporary contracts have become more common (Arias, Yamada, and Tejerina 2004). Within this setup, gender and ethnic earnings gaps persist. Recent decades, however, have seen important changes regarding the situation of women and ethnic minorities in labor markets and, in general, in society (World Bank 2011). This chapter was adapted from the following sources: “New Century, Old Disparities: Gender and Ethnic Wage Gaps in Latin America,� Juan Pablo Atal, Hugo Ñopo, and Natalia Winder, RES Working Paper 4640, Inter-American Development Bank, 2009; “Evolution of Gender Wage Gaps in Latin America at the Turn of the Twentieth Century: An Addendum to ‘New Century, Old Dispari- ties,’� Alejandro Hoyos and Hugo Ñopo, IZA Discussion Paper 5086, Institute for the Study of Labor, 2010; “Pushing for Progress: Women, Work and Gender Roles in Latin America,� Hugo Ñopo, Harvard International Review 33 (2): 315–28, 2011. Juan Pablo Atal is a graduate student in economics at the University of California, Berkeley, and Natalia Winder is a consultant at UNICEF, Division of Policy and Practice, New York. Alejandro Hoyos is a consultant at the Poverty Reduction and Economic Management Network (PREM) at the World Bank. 3 4 new century, old disparities Recent Changes on the Situation of Women and Ethnic Minorities The world, and particularly the region, has experienced important changes in the role of women and men in the past decades. Women’s visibility at home, at school, in the labor market, and in society in general has evolved significantly. Concurrently, men’s roles have evolved as well. In recent decades, women in Latin America and the Caribbean have seen progress in various dimensions of their social, economic, and political situations. For instance, the number of female presidents democratically elected rose from three in the 1980s to seven in the first decade of the 21st century, and women occupy 20 percent of parliamentary seats and make up 22 percent of elected municipal council representatives (ProLead 2012). Women’s school attainment increased more than that of men. Among people born in 1940, men had nearly one more year of schooling than women (5.8 years compared with 5.0 years); among people born in 1980, women had 0.3 year more education than men (9.5 years compared with 9.2). For the region as a whole, the gender gap in schooling reversed from favoring men to favoring women for the cohort born in 1968. The only countries for which the gap has not reversed are Bolivia and Guatemala, both of which have large indigenous populations. The global phenomenon of higher schooling attainment among women began earlier in Latin America and the Caribbean than in the rest of the world. These educational advances were observed particularly in the highest levels of education. In 1992, 16.4 percent of working women in the region and 10.7 percent of working men had some (complete or incomplete) tertiary education; by 2007, these figures had risen to 26.1 percent of women and 17.3 percent of men (Duryea et al. 2007). Women’s labor force participation increased in the region, whereas participation of men remained roughly constant.1 By the beginning of the 1990s, half of women participated in the labor market (worked or looked for work); by 2007, almost two of three women did so in most countries of the region. Most of the increase in women’s labor force participation can be explained by the increase in participation of young married women. Men still dominate labor markets, however: three out of five workers in the region are men, and occupational segregation by gender remains high. The share of female-headed households rose in the past 20 years. By the beginning of the 1990s, women headed 1.2 percent of complete households (households in which both husband and wife are present) and 79.8 percent of single-head households (authors’ calculations based on data in household surveys). These percentages increased to 9.2 percent and 82.3 percent, respectively, by the late 2000s. Women are increasingly heading households even when the father of their children is present. overview 5 Female-headed households are at both extremes of the income distri- bution. Some female household heads correspond to the profile of single young professionals or managers with young children. Others correspond to the profile of a low-educated single mother who holds an informal job in the service or commerce sector, has three or more children, and lives at or below the poverty line. Fertility has declined. In Argentina and Uruguay, fertility rates have decreased since the 1930s. In contrast, Bolivia, Guatemala, Honduras, Nicaragua, and Paraguay still showed high fertility rates in the mid-1990s, although in recent years these rates have also fallen (Ñopo 2011). Statistics about the presence of children reflect these changes. By the begin- ning of the 1990s, one in nine working women lived in a household with at least one child age six or younger; by the end of the first decade of the 2000s, that number had been almost halved. A similar situation exists for men: over the same period, the share of men living in households with children fell from 16 percent to 9 percent. This phenomenon, which has been linked to delays in women’s age at birth of their first child and higher earnings, suggests an alleviation of household responsibilities related to childbearing and child rearing. For countries that began the demographic transition early, however, responsibilities are shifting to the care of the elderly. Marriage, education, and work decisions have changed. The most sig- nificant increase in labor force participation was among women who married men with more education than themselves and, not surprisingly, women with no children or elderly relatives at home. Women who married men less skilled than themselves were more likely to work than women who married equally skilled men. Relative to women in other regions, skilled women in Latin America and the Caribbean are more likely to marry less skilled men. Ganguli, Hausmann, and Viarengo (2010) reveal that skilled women are less likely than unskilled women or skilled men to be married (or cohabiting). Skilled women who are married are less likely to work than skilled women who are not. Occupational segregation is particularly high in Latin America (Blau and Kahn 1992). Hierarchical segregation—the fact that managers tend to be men (white) and subordinates women (minorities)—is commonly accepted as the norm in the region’s labor markets. The persistence of traditional gender roles may be behind this phenomenon. The reduction of gender-based segregation in the workplace represents an area in which policy interventions can improve the efficiency of labor markets. Determining whether addressing occupational rather than hier- archical segregation is more effective is one of the areas of policy design to which this book aims to make a contribution. Latin America and the Caribbean is also a racially and ethnically diverse region, with some 400 ethnic groups (Hopenhayn and Bello 2001). All Latin American countries have indigenous and Afro-descendant populations.2 Recent progress in the region has not benefited indigenous people or people 6 new century, old disparities of African descent (ethnic minorities) as much as whites (ethnic majorities); high inequality remains pervasive (López-Calva and Lustig 2010). Ethnic minorities have fared worse than women. Across the region, they have higher poverty rates and lower income than whites (Psacharopoulos and Patrinos 1994; Gandelman, Ñopo, and Ripani 2008). They face restricted access to public services, lack of political representation, narrower labor market opportunities, and discrimination (Buvinic, Mazza, and Deutsch 2005; IDB 2008; Thorp 1998). They have weaker health indicators.3 Other factors that contribute to this pattern of inequality and poverty include labor force participation in low-productivity and hence poorly remu- nerated activities (Gaviria 2006). For example, throughout the region, indig- enous people are concentrated in informal trade, self-employment, and (for women) domestic service. Indigenous men are concentrated in blue-collar sectors, such as construction and manufacturing, and low-skilled services. This pattern can be traced largely to lower human capital endowments manifested in poorer educational performance and fewer years of job expe- rience of ethnic minorities (Hernández-Zavala et al. 2006; Solano 2002). Furthermore, returns to education have also been shown to vary substan- tially across ethnic groups (Gallardo 2006), which explains a large part of the income differences between ethnic minorities and nonminorities. In this regard, Latin America and the Caribbean have few empirical studies measuring discrimination against indigenous populations and exploring their potential economic costs (Patrinos and Psacharopoulos 1994; Cunningham and Jacobsen 2003; Saavedra et al. 2004; Patrinos and Hall 2006; Inter-American Development Bank 2008). The small number of studies mirrors the limited number of government policies in place to address the inequality between ethnic minorities and ethnic majorities and its impact on the incidence of poverty for the former group. Overview of the Book This book presents a regional overview of gender and ethnic disparities in labor earnings during this last turn of the century. After this introduc- tion, chapter 2 presents the methodology adopted by the book and the data sources employed. Chapter 3 then examines education in the region, highlighting the reversal of the gender schooling gap. Nowadays, girls attend more years of schooling than boys. After these three introductory chapters, the book then turns to the analysis of earnings gaps. Analyses of individual countries and groups of countries appear in chapters 4–16. Chapters 4–12 examine gender earnings gaps. Chapter 4 overviews gender earnings gaps in the region as a whole; chapters 5–12 examine gender earnings gaps in individual countries (Peru, Mexico, Chile, Colombia, Brazil, and Ecuador) and subregions (Central America and overview 7 the Caribbean). Chapters 13–16 examine ethnic earnings gaps, using the harmonized household surveys described previously. Chapter 13 provides an overview of the issue; the three chapters that follow it examine ethnic earnings gaps in Brazil (chapter 14), Ecuador (chapter 15), and Guatemala (chapter 16). Chapter 17 proposes policy options. Notes 1. In some other regions of the world, including the countries of the Organi- sation for Economic Co-operation and Development, labor force participation by men actually dropped. 2. Peru (27 percent of the total), Mexico (26 percent), Guatemala (15 percent), Bolivia (12 percent), and Ecuador (8 percent) account for almost 90 percent of the indigenous and Afro-descendant population in the region. 3. For instance, in Bolivia the provinces with larger proportions of indigenous populations, especially aymará and quechua, have the worst health indicators in the country: child malnutrition levels are above the national average in the provinces of Inquisivi, Tamayo, and Omasuyo de La Paz (Hopenhayn and Bello 2001), all of which have high indigenous density. References Arias, O., G. Yamada, and L. Tejerina. 2004. “Education, Family Background and Racial Earnings Inequality in Brazil.� International Journal of Manpower 25 (3/4): 355–74. Atal, J. P., H. Ñopo, and N. Winder, 2009. “New Century, Old Disparities: Gender and Ethnic Wage Gaps in Latin America.� RES Working Paper 4640, Inter- American Development Bank, Research Department, Washington, DC. Blau, F. D., and L. M. Kahn, 1992. “The Gender Earnings Gap: Learning from International Comparisons.� American Economic Review 82 (2): 533–38. Buvinic, M., J. Mazza, and R. Deutsch, eds. 2005. Social Inclusion and Economic Development in Latin America. Baltimore, MD: John Hopkins University Press. Cunningham, W., and J. P. Jacobsen. 2003. “Earnings Inequality Within and Across Gender, Racial, and Ethnic Groups in Latin America.� Wesleyan Economics Working Paper 2003-001. Wesleyan University, Middletown, CT. Duryea, S., S. Galiani, H. Ñopo, and C. Piras. 2007. “The Educational Gender Gap in Latin America and the Caribbean.� RES Working Paper 4510, Inter- American Development Bank, Research Department, Washington, DC. Gallardo, M. L. 2006. “Ethnicity-Based Wage Differentials in Ecuador’s Labor Mar- ket.� Master’s thesis, Cornell University, Department of Economics, Ithaca, NY. Gandelman, N., H. Ñopo, and L. Ripani. 2008. “Las fuerzas tradicionales de exclusión: Análisis de la bibliografía.� In ¿Los de afuera? Patrones cambiantes de exclusión en América Latina y el Caribe, 17–34. Washington, DC: Inter- American Development Bank. Ganguli, I., R. Hausmann, and M. Viarengo. 2010. “‘Schooling Can’t Buy Me Love’: Marriage, Work, and the Gender Education Gap in Latin America.� 8 new century, old disparities Faculty Research Working Paper RWP10–032, Harvard University, Kennedy School of Government, Cambridge, MA. Gaviria, A. 2006. “Movilidad social en América Latina: Realidades y percepciones.� Universidad de los Andes, Facultad de Economía, Bogota. http://economia .uniandes.edu.co/content/download/9168/44755/file/movilidad%20social.pdf. Hall, G., and H. A. Patrinos, eds. 2006. Indigenous Peoples, Poverty and Human Development in Latin America. London: Palgrave Macmillan. Hernández-Zavala, M., H. A. Patrinos, C. Sakellariou, and J. Shapiro. 2006. “Qual- ity of Schooling and Quality of Schools for Indigenous Students in Guatemala, Mexico, and Peru.� Policy Working Paper 3982, World Bank, Washington, DC. Hopenhayn, M., and A. Bello. 2001. “Discriminación étnico-racial y xenofobia en América Latina y el Caribe.� Serie Políticas Sociales 47, Comisión Económica para América Latina y el Caribe (CEPAL), Santiago. Hoyos, A., and H. Ñopo. 2010. “Evolution of Gender Wage Gaps in Latin America at the Turn of the Twentieth Century: An Addendum to ‘New Century, Old Disparities.’� IZA Discussion Paper 5086, Institute for the Study of Labor, Bonn, Germany. IDB (Inter-American Development Bank). 2008. Economic and Social Progress in Latin America, 2008 Report. Washington, DC: Inter-American Development Bank. ILO (International Labour Organization). 2007. Modelo de tendencias mundiales del empleo. Geneva: International Labour Organization. López-Calva, F., and N. Lustig, eds. 2010. Declining Inequality in Latin America: A Decade of Progress? Washington, DC: Brookings Institution Press. Márquez, G., A. Chong, S. Duryea, J. Mazza, and H. Ñopo, eds. ¿Los de afuera? Patrones cambiantes de exclusión en América Latina y el Caribe. Washington, DC: Inter-American Development Bank. Ñopo, H. 2011. “Pushing for Progress: Women, Work and Gender Roles in Latin America.� Harvard International Review 33 (2): 315–28. Paes de Barros, R., F. H. G. Ferreira, J. R. Molinas Vega, and J. Saavedra Chanduvi. 2009. Measuring Inequality of Opportunities in Latin America and the Carib- bean. Washington, DC: World Bank. ProLead. 2012. “Cuál ha sido el impacto de las leyes de cuotas sobre la representa- ción parlamentaria de las mujeres en América Latina.� Inter-American Develop- ment Bank. http://www.iadb.org/research/geppal/index.cfm. Psacharopoulos, G., and H. Patrinos. 1994. Indigenous People and Poverty in Latin America: An Empirical Analysis. Washington, DC: World Bank. Saavedra, J., Ñopo, H., and M. Torero. 2004. “Ethnicity and Earning in Urban Peru.� IZA Discussion Paper 980, Institute for the Study of Labor, Bonn, Germany. Solano, E. 2002. “La Población Indígena en Costa Rica según el Censo 2000.� Conference paper presented at “Costa Rica a la Luz del Censo 2000,� San José, Costa Rica. Thorp, R. 1998. Progress, Poverty and Exclusion: An Economic History of Latin America in the 20th Century. Baltimore, MD: Johns Hopkins University Press. World Bank. 2011. World Development Report 2012: Gender Equality and Devel- opment. Washington, DC: World Bank. 2 Methodology and Data Individuals’ earnings differ substantially. Within the vast heterogeneity of earnings there are patterns, of course. Some of these patterns correspond to productivity-related characteristics (individuals earn more the higher their educational achievement, the more experience they have, and so forth), but others do not correspond to those types of productivity-related characteristics. On average, men earn more than women and whites earn more than ethnic minorities.1 Gender and ethnicity may be linked indirectly to the extent that on average, men and whites exhibit human capital character- istics that are better rewarded in the labor market than the characteristics of women, people of indigenous descent, and Afro-descendants. What if these differences in human capital characteristics were removed? Would men still earn more than women and whites more than indigenous people and Afro-descendants? The statistical counterfactual question that has been used to address this issue is “what would the average earnings of a working woman (or ethnic minority) be if her labor market characteristics were equal, on average, to those of a working man (white)?� This chapter was adapted from the following sources: “Matching as a Tool to Decompose Wage Gaps,� Hugo Ñopo, Review of Economics and Statistics 90 (2): 290–99, 2008; “New Century, Old Disparities: Gender and Ethnic Wage Gaps in Latin America,� Juan Pablo Atal, Hugo Ñopo, and Natalia Winder, RES Working Paper 4640, Inter-American Development Bank, 2009; “Evolution of Gender Wage Gaps in Latin America at the Turn of the Twentieth Century: An Addendum to ‘New Century, Old Disparities,’� Alejandro Hoyos and Hugo Ñopo, IZA Discus- sion Paper 5086, Institute for the Study of Labor, 2010. Juan Pablo Atal is a graduate student in economics at the University of California, Berkeley, and Natalia Winder is a consultant at UNICEF, Division of Policy and Practice, New York. Alejandro Hoyos is a consultant at the Poverty Reduction and Economic Management Network (PREM) at the World Bank. 9 10 new century, old disparities The Blinder-Oaxaca Decomposition Methodologically, the approach to answer such questions has been the Blinder-Oaxaca decomposition. This partitions the average difference in earnings—the earnings gap—into two components, one attributable to differences in observable characteristics and the other that remains after these observable differences are removed (and hence attributable to dif- ferences in unobservable elements within the labor markets, including discrimination). This decomposition is performed on the estimated differ- ences in (Mincerian) earnings equations (Blinder 1973; Oaxaca 1973). The Blinder-Oaxaca decomposition is the prevailing approach in the empirical work on earnings gaps, but the literature has extensively docu- mented its limitations and drawbacks. Three are particularly worth not- ing. First, the relationship between characteristics and earnings is not necessarily linear, and recent data have been found to violate key implica- tions of the Mincerian model, the key input of the Blinder-Oaxaca decom- positions (Hansen and Wahlberg 1999; Heckman, Lochner, and Todd 2003). Second, Blinder-Oaxaca is informative only about the average earnings gap decomposition, providing no clues about the distribution of the differences in pay (Jenkins 1994; DiNardo, Fortin, and Lemieux 1996; Donald, Green, and Paarsch 2000). Third, Blinder-Oaxaca fails to restrict its comparison to comparable individuals, which is likely to substantially upwardly bias the estimators for unexplained differences in pay (Barsky et al. 2002). Methodology for This Book: An Extension of the Blinder-Oaxaca Decomposition The econometric procedure pursued in this book is the one introduced in Ñopo (2008). Conceived as an extension of the Blinder-Oaxaca decom- position using a nonparametric matching approach, this methodology attempts to explore the extent to which gender and ethnic earnings gaps can be attributed to differences in observable characteristics. This alter- native approach addresses the traditional Blinder-Oaxaca question not only for averages but also, and more interestingly, for the distribution of earnings, emphasizing the role of gender and ethnic differences in the “common support� of the distribution of observable human capital characteristics. The proposed methodology yields a more precise measurement of the explained and unexplained components of the earnings gap. It not only decomposes the earnings gap into “endowment� and “unexplained� blocks, it also allows for the exploration of the distribution of the unex- plained differences in earnings. methodology and data 11 The methodology constrains the comparison of earnings gaps to people with comparable characteristics. In other words, it accounts for the out- comes of minorities and women for whom no whites or men (respectively) with comparable human capital characteristics can be found, an issue often neglected in the earnings gaps literature. Finally, this methodology does not need to assume any sort of functional form for the relationship between characteristics and earnings (such as the Mincerian model). The methodology works by generating synthetic samples of individuals by matching men (whites) and women (ethnic minorities) with the same observable characteristics. The matching characteristics are discrete, so the match is done perfectly, without using propensity scores or any notion of distance between the characteristics. The basic form of the algorithm is shown below for gender earnings gaps (it works in the same way for ethnic earnings gaps): • Step 1: Select one woman (without replacement) from the sample. • Step 2: Select all men who have the same characteristics as the woman selected. • Step 3: Construct a synthetic individual whose earnings are equal to the average of all of individuals selected in step 2 and “match� him to the original woman. • Step 4: Put the observations of both individuals (the synthetic man and the woman) in the new (respective) samples of matched individuals. • Repeat steps 1–4 until it exhausts the original sample of women. Application of this matching algorithm creates three sets of individuals: one of men whose observable characteristics cannot be matched to those of any women in the sample; one of women whose observable characteristics cannot be matched to those of any men in the sample; and one of matched men and women, in which the distribution of observable characteristics for men is identical to that of women. In this last group, observations for men are weighted in such a way that their joint distribution of observable characteristics mimics the distribution of observable characteristics of matched women. Only comparable individuals are compared. It is pos- sible to calculate the earnings distribution of the sample of women if their observable characteristics resemble those of the sample of men. The other two sets—of unmatched men and women—make it possible to determine how much of the calculated gap is accounted for by the out- comes of men and women out of the common support. This issue of lack of comparability between some men and women (uncommon supports) has been largely neglected in the gender earnings gap literature. As Ñopo (2008) shows, failure to recognize the lack of common support in some circumstances may lead to overestimation of the unexplained component of the earnings gaps. 12 new century, old disparities The sets of matched and unmatched individuals are compared. The earnings gap (Δ)—the difference in average earnings of men and women, expressed as a percentage of women’s average earnings—is then decom- posed into four additive elements: Δ = (ΔX + ΔM + ΔF) + Δ0. As in the Blinder-Oaxaca decomposition, one component: ΔX, is attrib- uted to the differences in observable characteristics between men and women. However, as the matching procedure takes into account the fact that not every combination of characteristics of men is found among women (and vice versa), the computation of ΔX is restricted to men and women whose characteristics lie in the common support of both character- istics’ distributions. Further extending the basic Blinder-Oaxaca approach, instead of controlling for differences in average characteristics of men and women, the matching procedure allows controlling for differences in the distributions of those characteristics. The second element, ΔM (ΔW in the ethnic-based decompositions), is the portion of the earnings gap caused by the existence of men with com- binations of characteristics that are not met by any women (for instance, highly educated young workers filling high-profile positions such as chief executive officer [CEO]). The third element, ΔF (ΔNW in the ethnic-based decompositions), is the portion of the gap caused by the existence of women with combina- tions of characteristics that are not met by any men (for instance, old and low-skilled domestic workers). Both ΔM and ΔF exist because the supports of the sets of observable characteristics of men and women do not com- pletely overlap. The element ΔM is referred to as the “CEO effect�; ΔF is referred to as the “maid effect.� These effects reflect the fact that CEOs tend to be men and not women and maids tend to be women and not men. Dávila and Pagán (1999) report that Costa Rican and Salvadoran women are underrepresented in occupational categories such as mana- gerial, services, agricultural labor, and laborer occupational catego- ries and overrepresented in professional, administrative support/cleri- cal, and transportation jobs. Hertz et al. (2008) report that working women are underrepresented in managerial positions and overrepre- sented as service workers, merchants, administrative personnel, and professionals. Marked differences by economic sector are also apparent. Construction and agriculture are sectors dominated by men, whereas community, social, and personal services are dominated by women. These differences may reflect women’s self-selection into segments of the labor market where they enjoy more flexibility to manage their work and household respon- sibilities. Women may choose to permanently or temporarily withdraw from the labor market or work in occupations with flexible or fewer methodology and data 13 working hours (Tenjo, Ribero, and Bernat 2006). As a result, they may accumulate less work experience or invest less in education or on-the-job training (Terrell 1992). The fourth element, Δ0, is the portion of the gap that cannot be explained by the first three elements. It could be attributable to differences in unob- servable characteristics, possibly including discrimination. It is compa- rable to the component of the earnings gap that reflects the differences in rewards to observable characteristics in the traditional Blinder-Oaxaca approach but restricted to the common support of those characteristics. In this way, the methodology yields an alternative estimator for the unexplained earnings gap. This estimator attenuates biases and is more informative about the gap distribution, not only its average. The meth- odology, nonetheless, has some limitations. In addition to the need to define the matching characteristics as categorical variables only, it faces a challenge known as the “curse of dimensionality,� which is behind most nonparametric configurations. This “curse� refers to the fact that the likelihood of finding matches of men and women decreases as the number of control variables (the “dimension�) increases—a problem, given that researchers would like to use the maximum number of observable charac- teristics in order to control the scope of the role of unobservable factors in explaining the earnings gap. The curse of dimensionality forces research- ers to make a trade-off between the number of control characteristics and the size of the nonoverlapping supports. This tradeoff is expressed in the decomposition exercises described in the following chapters as a shrinkage in the size of the common supports as new observable characteristics are added to the matching configuration. Two limitations that the approach introduced by this methodology cannot overcome are selectivity and unobservables. Men and women and whites, indigenous people, and Afro-descendants may differ in their deci- sion-making processes about entering the labor markets. Hence, the way in which they select into the active (and employed) labor force may be different. The observed samples of working women and men and whites and ethnic minorities may not be representative samples of the population as a whole. This limitation can be treated with conventional corrections in the regres- sion-based approach (Heckman 1979), but not in a matching-based one. Another limitation, shared by the regression-based and matching-based approaches is that data on all relevant variables that might affect earnings are not available. Individual abilities and characteristics on which data are not available—including work ethic, commitment, and capacity to work as part of a team—are very relevant for determining earnings. As employ- ers, and labor markets in general, can observe them and reward them appropriately, their effects should be embedded in individuals’ earnings. These features are, however, unobservable. In this sense, the estimators reported in this book for the unexplained differences in pay are just that: gaps that cannot be explained on the basis of observable characteristics.2 14 new century, old disparities Data The methodology was applied to nationally representative household sur- vey data. Table 2.1 indicates the years of each survey analyzed in each chapter of this book. These surveys were processed and harmonized by the Research Department of the Inter-American Development Bank and CEDLAS at the Universidad de La Plata to facilitate cross-country comparisons. Each observation in every household survey has an associated expansion factor that reflects the particularities of the sampling methods involved. The expansion factor can be interpreted as the number of individuals each observation represents; the sum of the expansion factors in any given sur- vey approximates the population size of the country. In this way, pooling the observations in the 18 surveys for the gender studies, each weighted by its expansion factor, creates a sample representative of the working population of Latin America and the Caribbean. The focus of the analyses is on the working population in each country. The variable of interest is labor earnings, measured as hourly earnings. In the pooled data sets, hourly earnings are measured in terms of 2002 dollars using purchasing power parity (PPP) exchange rates and nominal GDP deflators. Every chapter therefore excludes the population below and above certain ages. Also excluded from the data sets are all observa- tions for which hourly income is missing or negative. For the purpose of the decompositions, only observations with values for every one of the characteristics used as control variables are kept. As the gender variable is available in all national data sources, the gen- der earnings gap analysis is performed for the entire sample of countries listed in table 2.1. In contrast, the datasets pooled for the ethnic studies cover only Bolivia, Brazil, Chile, Ecuador, Guatemala, Paraguay, and Peru, which represent about 55 percent of the region’s population. The sample used in the ethnic regional analysis is selected in the same way as the gen- der sample, excluding observations with the same criteria. Cross-country comparisons of ethnic earnings gaps should be inter- preted with caution, because the definition of ethnicity is not the same in all countries. Individuals are classified as either minority or nonminority depending on the ethnic groups each survey considers. Ethnic minorities are defined by individuals’ self-assessment of being part of an indigenous group in Bolivia, Chile, Ecuador, Guatemala, and Peru; by skin color in Brazil; and by mother tongue in Paraguay. The details of this classification are presented in table 2.2. Questions on surveys for educational attainment information are fre- quently expressed in terms of the grade completed in school or university. Calculating years of schooling—obtained by summing the years com- pleted by each respondent—requires taking into account differences in school systems across countries. After years of schooling are calculated, Table 2.1 Household Survey Data Used, by Country and Chapter Gender chapters Ethnic chapters Country Survey Education Regional Subregional Country Regional Country Argentina Encuesta Permanente de Hogares (EPH) 1992 Encuesta Permanente de Hogares-Continua 2006 2006 (EPH-C) Barbados Continuous Labor Force Sample Survey 2004 (CLFSS) Bolivia Encuesta Nacional de Empleo (ENE) 1997 Encuesta Continua de Hogares (ECH)— 2007 2006, 2007 2006 Mejoramiento de las Encuestas y Medición de las Condiciones de Vida (MECOVI) Brazil Pesquisa Nacional por Amostra de Domicilios 2008 1992 1996–2006 2007 1996–2006 (PNAD) 2007, 2008 Chile Encuesta de Caracterización Socioeconómica 2006 1992 1992, 1994, 2006 Nacional (CASEN) 2006 1996, 1998, 2000, 2003, 2006, 2009 Colombia Encuesta Nacional de Hogares — Fuerza de 1992, 2005 1994–2006 Trabajo (ENH—FT) Gran Encuesta Integrada de Hogares (GEIH) 2006 2006 15 (continued next page) 16 Table 2.1 Household Survey Data Used, by Country and Chapter (continued) Gender chapters Ethnic chapters Country Survey Education Regional Subregional Country Regional Country Costa Rica Encuesta de Hogares de Propósitos Múltiples 2007 1992 1995, 2000, (EHPM) 2006, 2007 2006 Dominican Encuesta Nacional de Fuerza de Trabajo 2007 2000 Republic (ENFT) 2003, 2007 Ecuador Encuesta de Condiciones de Vida (ECV) 2006 1995, 2006 Encuesta de Empleo, Desempleo, y Subempleo 2007 2003–07 2007 2003–07 (ENEMDU) El Salvador Encuesta de Hogares de Propósitos Multiples 2007 1991 1995, 2000, (EHPM) 2005, 2007 2005 Guatemala Encuesta Nacional de Condiciones de Vida 2006 2000, 2006 2000, 2006 2006 2000, 2006 (ENCOVI) Encuesto Nacional de Empleo e 2004 2004 Ingresos (ENEI) Honduras Encuesta Permanente de Hogares de Propósitos 2007 1997 1997, 2002, Multiples (EPHPM) 2007 2007 Jamaica Labor Force Survey undertaken by the 2003 Statistical Institute (STATIN) Mexico Encuesta Nacional de Ingresos y Gastos de los 2008 1992, 2008 1994–2004 Hogares (ENIGH) Encuesta Nacional de Empleo Urbano (ENEU) 2004 Table 2.1 Household Survey Data Used, by Country and Chapter (continued) Gender chapters Ethnic chapters Country Survey Education Regional Subregional Country Regional Country Nicaragua Encuesta Nacional de Hogares sobre Medición 2005 1993 1998, 2001, de Nivel de Vida (EMNV) 2005 2005 Panama Encuesta de Hogares, Mano de Obra (EMO) 2006 1991 1997, 2002, Encuesta de Hogares (EH) 2003, 2006 2006 Paraguay Encuesta de Hogares (Mano de Obra) (EH) 2006 Encuesta Permanente de Hogares (EPH) 2007 2006, 2007 Peru Encuesta Nacional de Hogares (ENAHO) 2007 1997 1997–2009 2006 2006, 2007 Uruguay Encuesta Continua de Hogares (ECH) 2007 1992 2005, 2007 Venezuela, Encuesta de Hogares Por Muestreo (EHM) 2006 1992 RB 2004, 2006 Source: The data sources were compiled and harmonized by the Research Department of the Inter-American Development Bank and CEDLAS. 17 18 new century, old disparities Table 2.2 Criteria for Classifying Ethnic Groups as “Minorities,� by Country Percentage of workers Country Criterion 12–65 Bolivia Self-declaration Self-declaration as Quechua, 52.6 Aymara, Guarani, Chiquitano, Mojeño, or other Brazil Skin color Self-declaration of skin color as 48.5 black or brown Chile Self-declaration Self-declaration as Aymara, 6.0 Rapa nui, Quechua, Mapuche, Atacameño, Coya, Kawaskar, Yagan, or Diaguita Ecuador Self-declaration Self-declaration as indigenous, 10.0 black, mulatto, or other Guatemala Self-declaration Self-declaration as K´iche´, 35.1 Q´eqchi´, Kaqchikel, Mam, Q´anjob´al, Achi, Ixil, Itza´, Poqomchi´, Chuj, Awakateko, Poqomam, Ch´orti´, Jakalteko, Sakapulteco, Mopan, Uspanteko, Tz´utujil, Sipakapense, Chalchiteko, Akateko, Xinka, or Garifuna Paraguay Self-declaration Self-declaration as Guarani 33.4 speaking Peru Self-declaration Self-declaration as Quechua, or 31.3 Aymara; from Amazonia; or black, mulatto, Zambo, or other Source: The data sources were compiled and harmonized by the Research Depart- ment of the Inter-American Development Bank and CEDLAS. new variables for educational attainment are created that consider the same education levels across countries. These are seven dummy variables, one for each of the following levels: no education, primary incomplete, primary complete, secondary incomplete, secondary complete, tertiary incomplete, and tertiary complete or more. In general, job characteristics include whether or not the individual works in the formal sector, the firm size, the occupation, and the economic sector of the activity. Formal labor is a dummy variable created from information on social security affiliation; it is equal to 1 if the respon- dent reports paying mandatory social security. Small firm is a dummy methodology and data 19 variable that takes the value 1 for firms with no more than five workers. Occupation is a variable coded to the one-digit level based on categoriza- tions used in each country, which are frequently based on the Interna- tional Standard Classification of Occupations (ISCO) international code. The categories included are professionals and technicians, directors and upper management, administrative personnel, merchants and sellers, ser- vice workers, agricultural workers and similar, nonagricultural blue collar workers, armed forces, and other occupations not classified in the previous categories. Economic sector is a variable coded to the one-digit level based on categorizations used in each country that are frequently based on the International Standard Industrial Classification (ISIC) international code. The categories included are agriculture, hunting, forestry, and fishing; mining and quarrying; manufacturing; electricity, gas, and water supply; construction; wholesale and retail trade and hotels and restaurants; trans- port, storage; financing, insurance, real estate, and business services; and community, social, and personal services. In general, sociodemographic variables will be dummies, which take the value of 1 if the condition is met and 0 otherwise. Notes 1. For simplicity, the term ethnic minorities is used to refer to ethnic and racial groups other than whites. In some countries, these groups represent majorities. 2. See Ñopo (2008) for technical details on the matching procedure, a com- parison between it and the traditional approach based on linear regressions, and proofs of the asymptotic consistency of the estimators derived from this method. The same procedure is used to decompose gender and ethnic earnings gaps. References Atal, J. P., H. Ñopo, and N. Winder, 2009. “New Century, Old Disparities: Gender and Ethnic Wage Gaps in Latin America.� RES Working Paper 4640, Inter- American Development Bank, Research Department, Washington, DC. Barsky, R., J. Bound, K. Charles, and J. P. Lupton. 2002. “Accounting for the Black–White Wealth Gap: A Nonparametric Approach.� Journal of the American Statistical Association 97 (459): 663–73. Blinder, A. 1973. “Wage Discrimination: Reduced Form and Structural Differ- ences.� Journal of Human Resources 7 (4): 436–55. Dávila, A., and J. Pagán. 1999. “Gender Pay and Occupational-Attainment Gaps in Costa Rica and El Salvador: A Relative Comparison of the Late 1980s.� Review of Development Economics 3 (2): 215–30. DiNardo, J., N. Fortin, and T. Lemieux. 1996. “Labor Market Institutions and the Distribution of Wages, 1973–1992: A Semiparametric Approach.� Economet- rica 64 (5): 1001–44. Donald, S., D. Green, and H. Paarsch. 2000. “Differences in Wage Distributions between Canada and the United States: An Application of a Flexible Estimator 20 new century, old disparities of Distribution Functions in the Presence of Covariates.� Review of Economic Studies 67: 609–63. Hansen, J., and R. Wahlberg. 1999. “Endogenous Schooling and the Distribution of the Gender Wage Gap.� IZA Discussion Paper 78, Institute for the Study of Labor, Bonn, Germany. Heckman, J. 1979. “Sample Selection Bias as a Specification Error.� Econometrica 47 (1): 153–61. Heckman, J., L. Lochner, and P. Todd. 2003. “Fifty Years of Mincer Earnings Regressions.� NBER Working Paper 9732, National Bureau of Economic Research, Cambridge, MA. Hertz, T., A. P. de la O Campos, A. Zezza, P. Winters, E. J. Quiñones, and B. Davis. 2008. “Wage Inequality in International Perspective: Effects of Location, Sec- tor, and Gender.� ESA Working Paper 8/08, Food and Agriculture Organiza- tion, Agricultural and Development Economics Division, Rome. ftp://ftp.fao .org/docrep/fao/011/ak230e/ak230e00.pdf. Jenkins, S. P. 1994. “Earnings Discrimination Measurement: A Distributional Approach.� Journal of Econometrics 61 (1): 81–102. Ñopo, H. 2008. “Matching as a Tool to Decompose Wage Gaps.� Review of Eco- nomics and Statistics 90 (2): 290–99. Ñopo, H., and A. Hoyos. 2010. “Evolution of Gender Wage Gaps in Latin America at the Turn of the Twentieth Century: An Addendum to ‘New Century, Old Disparities.’� IZA Discussion Paper 5086, Institute for the Study of Labor, Bonn, Germany. Oaxaca, R. 1973. “Male-Female Wage Differentials in Urban Labor Markets.� International Economic Review 14 (3): 693–70. Tenjo, J., R. Ribero, and L. Bernat. 2006 “Evolución de las diferencias salariales de género en seis países de América Latina.� In Mujeres y trabajo en América Latina, ed. C. Piras, 149–98. Washington, DC: Inter-American Development Bank. Terrell, K. 1992. “Female-Male Earnings Differentials and Occupational Struc- ture.� International Labor Review 131 (4–5): 387–98. 3 Gender Differences in Education in Latin America and the Caribbean: Girls Outpacing Boys Education is fundamental to economic and social development and the end of poverty. Countries with higher average schooling have been more successful in their development paths. As important as the overall level of education is its distribution. A significant dimension of the distribution of education is gender. In most countries, women attain less schooling than men; the gender gap is wider in developing countries than in developed countries. Latin America and the Caribbean is an interesting exception, as girls in the region achieve more schooling than boys. In contrast to Africa, Asia, and the Middle East and North Africa, it has achieved gender parity (or a ratio that favors girls) in education. Furthermore, in most countries of the region, there is a reverse gender gap in education. Women have more average years of schooling than their male counterparts (important excep- tions are the indigenous communities of Bolivia and Guatemala). These surprising outcomes seem to contradict the standard assumption that parents favor investing in boys’ education. This chapter analyzes the evolution of the gender gap in average years of education for cohorts born between 1940 and 1984. A descriptive cross-country analysis of the changes in the distribution of education by This chapter was adapted from “The Educational Gender Gap in Latin Amer- ica and the Caribbean,� Suzanne Duryea, Sebastian Galiani, Hugo Ñopo, and Claudia Piras, RES Working Paper 4510, Inter-American Development Bank, Research Department, 2007. Suzanne Duryea is a lead economist in the Social Sector Unit at the Inter-American Development Bank. Sebastián Galiani is a professor of economics at the University of Maryland. Claudia Piras is a lead social development economist at the Inter- American Development Bank. 21 22 new century, old disparities gender, cohort, and country is produced using household surveys (for a description of the household surveys used in this chapter and the rest of this book, see chapter 2). The chapter attempts to answer the following questions: When did the gender gap in schooling close in Latin America and the Caribbean? Was it a uniform process across the region, or did some countries close the gender gap earlier than others? Is the reversal of the gender gap uni- formly distributed across education levels, or is it explained mostly by changes among the more educated? Are there remaining gender differ- ences in attendance and attainment among 6 to 20-year-olds by income quintile? Strengthening girls’ education opportunities is a strategic priority in many countries, because societies pay a price for gender inequality in terms of slower growth and reduced income (Dollar and Gatti 1999). Studies of rates of return also document the economic benefits of investing in girls’ education (Psacharopoulos and Tzannatos 1992; Psacharopoulos 1994). In addition to generating private returns from labor market par- ticipation, women’s education yields strong social externalities, including the following: • Higher levels of education among women reduce fertility (Schultz 1973; Cochrane 1979), which decreases infant mortality and increases life expectancy (Behrman and Deolalikar 1988). • Mothers’ education has important intergenerational effects on the education, health, and well-being of their children (King et al. 1986; Schultz 1988; Strauss and Thomas 1995; Behrman, Duryea, and Székely 1999). • Adding to a mother’s schooling has a larger beneficial effect on a child’s health, schooling, and adult productivity than adding to a father’s schooling (King and Hill 1993; Schultz 1993). Advances in the education of women represent one of the biggest suc- cess stories in Latin America and the Caribbean. However, little is known about this important and unprecedented accomplishment in the develop- ing world. Most studies that look at educational outcomes have not gone beyond addressing the absence of a gender gap in the region. Knodel and Jones (1996) stress the rapid closure of the gender gap in most of the world, suggesting that the strong emphasis on eliminating gender inequality in schooling is no longer needed, but they do not specifically address the situation in Latin America and the Caribbean. Behrman, Duryea, and Székely (1999) were the first to analyze schooling progress in the region using household surveys. They highlight that for two-thirds of the 18 countries considered, the average years of schooling for women is higher than for men for cohorts born in 1970. gender differences in education 23 One of the few efforts to look at gender differences in education in Latin America and the Caribbean is Parker and Pederzini (2000), who examine the determinants of the level of education of girls and boys in Mexico and the factors that may explain gender differences. Marshall and Calderón (2005) find that enrollment rates of 6 to 11-year-olds were lower among girls than boys in only 4 of 22 countries considered. The picture changes slightly for older age groups, but in the majority of countries, enrollment rates favored girls. Marshall and Calderón also report lower repetition and drop-out among girls, higher promotion rates, and, in most countries, better grade-for-age outcomes. Changes in the Gender Education Gap Figure 3.1 shows the evolution of the average number of years of school- ing completed by women and men and the gap between the two by birth year. The data are computed as three-year moving averages (that is, data reported for the 1940 cohort correspond to people born between 1939 and 1941 and so on). Figure 3.1 Average Years of Education of Men and Women, Born 1940–84, and Education Gender Gap in Labor Force in Latin America and the Caribbean 11 10 average years of schooling 9 8 7 6 5 4 3 2 1 0 –1 –2 40 42 44 46 48 50 52 54 56 58 60 62 64 66 68 70 72 74 76 78 80 82 84 19 19 19 19 19 19 19 19 19 19 19 19 19 19 19 19 19 19 19 19 19 19 19 birth cohort women men gap Source: Based on data gathered from national household surveys, 2001–04. Note: Figures are three-year moving averages. 24 new century, old disparities Among people born between 1939 and 1941, on average, women attained 4.4 years of schooling and men attained 5.1 years. The gender education gap for this birth cohort was thus 0.7 year in favor of men. For people born between 1983 and 1985 (people who were 21–23 at the time of the surveys), the average schooling attainment was 10.1 years for women and 9.6 years for men; the gender education gap was 0.5 year in favor of women. During this period of four decades, women’s schooling attainment increased by 5.7 years while men’s attainment increased by 4.5 years. On average, the gender gap has been declining at a rate of about 0.27 years of schooling per decade. Figure 3.1 suggests that gender parity was achieved beginning with the cohort born around 1965. These average statistics for the region hide intraregional diversity (for graphs for individual countries and descriptions of the data, see Duryea et al. 2007). Table 3.1 reports the birth cohort in which each country achieved gender parity. Six countries (Argentina, Brazil, Colombia, Costa Rica, Panama, and República Bolivariana de Venezuela) achieved parity for cohorts born in the 1950s. The Dominican Republic, Honduras, and Nicaragua achieved parity for cohorts born in the 1960s. Chile, Ecuador, and Paraguay achieved parity for cohorts born in the 1970s (the educa- tional gap in Chile has been close to zero since the mid-1960s). El Salvador, the last country to achieve gender parity, did so for cohorts born in 1984 (but its gap was close to zero for cohorts born in the early 1970s). The gender gap in educational attainment in Uruguay favors women in all years considered, suggesting that it was the first country in which the gap closed (before the period of analysis). In four countries (Bolivia, Guatemala, Mexico, and Peru), the gender educational gap favored men during the whole period. These countries have the largest shares of indigenous people. Data are available for two additional birth cohorts for Mexico (2008) and Peru (2007). These data show that Mexico achieved gender parity for the 1985 birth cohort, with the gap for this year equal to 0.10 year favor- ing women. Peru has an education gender gap that is very close to zero, but it favors men (–0.09) for the last cohort in the survey. A linear extrapolation of the rate at which the gap has been declining in Bolivia and Guatemala suggests that parity will be achieved in Bolivia for the cohort born in 1999. The trends for Guatemala do not allow estima- tion of the year at which parity will materialize. Decomposing Changes in the Gender Education Gap For Latin America as a whole, the gender gap in schooling attainment has been declining at a rate of about 0.27 year of schooling per decade. Since the mid-1960s, the gap has favored women. These changes in the average trend are interesting, but it would be even more interesting to understand the segments of the schooling distribution gender differences in education 25 Table 3.1 Gender Gap in Education in Latin America and the Caribbean for Cohorts Born in 1940 and 1984, by Country (years) Gap for 1940 Birth cohort at Gap for 1984 Country birth cohort which the gap closes birth cohort Argentina –0.89 1951 0.69 Bolivia –2.40 – –0.19 Brazil –0.41 1950 0.82 Chile –0.74 1975 0.18 Colombia –0.28 1958 0.45 Costa Rica –0.57 1956 0.54 Dominican Republic –0.83 1965 0.90 Ecuador –0.69 1971 0.33 Guatemala –0.59 – –0.84 Honduras –0.53 1968 0.72 Mexico –0.83 – –0.13 Nicaragua –0.88 1966 1.18 Panama –1.01 1955 0.72 Peru –1.84 – –0.17 Paraguay –0.83 1975 0.65 El Salvador –1.44 1984 0.11 Uruguay 0.07 ++ 0.91 Venezuela, RB –0.84 1955 1.23 Latin America –0.65 1965 0.46 Source: Based on data from national household surveys, 2001–04. Note: – = gap has not closed for any of the birth cohorts considered, ++ = gap closed for a previous cohort to the 1940 cohort. in which changes were most pronounced. For this purpose, the sample was decomposed into four groups: individuals who acquired no educa- tion or only incomplete primary education, individuals with complete primary or incomplete secondary education, individuals with complete secondary or incomplete university education, and university graduates (figure 3.2) (see Duryea et al. 2007 for figures by country). The proportion of women with no schooling or incomplete primary education fell markedly, decreasing at a faster rate than for men. The 26 Figure 3.2 Educational Attainment of Men and Women in Labor Force in Latin America and the Caribbean Born 1940–84 a. Percentage of population with no education or b. Percentage of population with complete primary school or incomplete primary school incomplete secondary education 70 70 60 60 50 50 percent percent 40 40 30 30 20 20 10 10 0 0 40 44 48 52 56 60 64 68 72 76 80 84 40 44 48 52 56 60 64 68 72 76 80 84 19 19 19 19 19 19 19 19 19 19 19 19 19 19 19 19 19 19 19 19 19 19 19 19 c. Percentage of population with complete secondary or d. Percentage of population with university degree incomplete university education 70 70 60 60 50 50 percent percent 40 40 30 30 20 20 10 10 0 0 40 44 48 52 56 60 64 68 72 76 80 84 40 44 48 52 56 60 64 68 72 76 80 84 19 19 19 19 19 19 19 19 19 19 19 19 19 19 19 19 19 19 19 19 19 19 19 19 birth cohort birth cohort women men women men Source: Based on data from national household surveys, 2001–04. gender differences in education 27 proportion of people with complete secondary or incomplete university increased slightly more rapidly for women than for men. At the upper extreme of the distribution, there are three periods with interesting differ- ences. During the first period (cohorts born 1940–60), university gradu- ation rates increased for women. This period was followed by a period (cohorts born 1960–75) of relative stagnation. A third period, starting with cohorts born around 1975, was marked by a decrease in univer- sity graduation rates for both women and men, although there are good reasons to attribute the decline to the fact that younger people in these cohorts may still be in school. Changes in the gender schooling gap between the oldest and the young- est cohort in the sample are decomposed into changes at each educational level. Results are first reported for the components of the educational gap accounted for by each educational level in each birth cohort (figure 3.3). Each component of the education gap for a cohort corresponds to the dif- ference between women’s average years of schooling at each educational level weighted by women’s participation at that level and men’s average years of schooling at each educational level weighted by men’s participa- tion at that level. The gender schooling gap in the 1940 birth cohort is compared with the gender schooling gap in the 1984 cohort in figure 3.4. Each component of Figure 3.3 Decomposition of Educational Gender Gap in Latin America and the Caribbean, by Educational Level for Cohorts Born in 1940 and 1984 0.8 (average women’s education – average men’s education) gap in years of education 0.6 0.49 0.4 0.2 –0.06 0.23 0 –0.10 –0.06 –0.2 –0.13 –0.17 –0.4 –0.39 –0.6 –0.8 1940 1984 birth cohort no education or incomplete primary complete secondary or incomplete university complete primary or incomplete secondary university degree Source: Based on data from national household surveys, 2001–04. 28 new century, old disparities Figure 3.4 Decomposition of Difference in Educational Gender Gap between Youngest and Oldest Cohort in Latin America and the Caribbean, by Educational Level 1.4 1.2 change in educational gap 1.0 (women–men) 0.8 0.88 0.6 0.4 0.2 0.36 0 –0.05 –0.2 –0.10 –0.4 1984/1940 birth cohort no education or incomplete primary complete secondary or incomplete university complete primary or incomplete secondary university degree Source: Based on data from national household surveys, 2001–04. Note: Figure shows change between cohorts born in 1940 and cohorts born in 1984. the total is the difference between the gender gap at each educational level for the 1984 cohort, calculated as described before, and the gender gap at the same level for the 1940 cohort. The gender education gap for the 1940 cohort is –0.65 (the negative sign indicates that it favors men). Decomposition yields gaps of –0.06, –0.06, –0.13, and –0.39 for each of the four education levels. For the cohort born in 1984, the gender education gap is 0.46, favoring women, decomposed as –0.10, –0.17, 0.23, and 0.49. The change in the education gap between the oldest and youngest cohorts is 0.46 – (–0.65) = 0.11, decomposed as –0.05, –0.10, 0.36, and 0.88. Figure 3.4 indicates that changes among university graduates explain 88 percent of the change in the gender gap. A country-by-country decomposition of the change in the gap reveals some interesting differences across countries (figure 3.5). For most coun- tries, the third and fourth education levels are the most important con- tributors to the change in the schooling gap. For Ecuador, Honduras, and Peru, the second-level component (complete primary and incomplete secondary) is positive; in Mexico and Chile, the gap is positive but small. In the remaining 13 countries in the region, the gap at this level is nega- tive. In Bolivia, Ecuador, El Salvador, Guatemala, Honduras, Mexico, and gender differences in education 29 Figure 3.5 Decomposition of Change in Educational Gender Gap in Latin America and the Caribbean, by Educational Level and Country for Cohorts Born 1940–84 gap in years of education (average women’s 4 education – average men’s education) 3 2 1 0 –1 –2 a ia il le ia a lic dor ala ras xico gua ma eru uay dor ay RB tin liv az Chi mb Ric ub u e a gu a, g en Bo Br o lo sta a m ep Ecu ate ond M icar Pan a P ag r va al Uru uel Ar C Co n R G u H N Pa I S z ica E ne in Ve m Do no education or primary incomplete primary complete or secondary incomplete secondary complete or college incomplete college degree or more Source: Based on data from national household surveys, 2001–04. Peru, the first-level component (no education and incomplete primary) explains why changes in the gap favor women. Figure 3.5 reveals polarization in many countries, particularly Argentina, Nicaragua, and Venezuela. These countries exhibit large changes in the gap that favor women at the higher levels of schooling attainment as well as changes at the lower levels of attainment that favor men. Thus, women are falling behind men at low levels of education even as they are surpass- ing men at higher levels of educational attainment. The change in the gap between cohorts is decomposed into two compo- nents, the probability component and the conditional expectations compo- nent. The probability component accounts for the gender difference in the probability of achieving a given educational level. The conditional expec- tations component accounts for the gender difference in the number of expected years of completed schooling at each level. The probability com- ponent is calculated as the sum of the four differences in the percentages of the female and male population at each educational level multiplied by 30 new century, old disparities the average years of schooling that men reach by level. The conditional expectations component is calculated as the sum of the four differences in average years of schooling between women and men at each educational level weighted by the percentage of women at each level. The results are summarized in figure 3.6, which shows that most of the changes in schooling attainment between cohorts occurred as a result of changes in the probability component rather than the number of com- pleted years of schooling at each level. The figure shows that the prob- ability component accounted for 0.90 and the conditional expectations component for 0.19 year of the gap. Thus, changes in gender differences in the probabilities of achieving higher education levels explain four-fifths of the change in the schooling gender gap. Within the changes in probabili- ties, changes at the completed secondary and completed university levels are most important, although less than a third of the population reaches university. There is thus still much room for improvement in enrollment, attendance, and graduation from the upper levels of education in the region, for women and men alike. Figure 3.7 decomposes the changes into changes in probabilities and changes in expectations. Only the aggregate changes for each are presented. Figure 3.6 Decomposition of Change in Educational Gender Gap in Latin America and the Caribbean, by Component and Educational Level for Cohorts Born 1940–84 1.4 (average women’s education – average men’s education) gap in years of education 1.2 1.0 0.8 0.87 0.6 0.4 0.02 0.00 0.2 0.33 0 0.14 0.03 –0.19 –0.2 –0.11 –0.4 probability conditional expectations component no education or incomplete primary complete secondary or incomplete university complete primary or incomplete secondary university degree Source: Based on data from national household surveys, 2001–04. gender differences in education 31 Figure 3.7 Decomposition of Changes in Educational Gender Gap in Latin America and the Caribbean, by Component and Country gap in years of education (average women’s education – average men’s education) 2.5 0.99 1.80 1.80 2.0 1.16 1.15 1.05 1.67 1.15 0.93 1.5 1.01 0.76 1.06 1.0 0.65 0.47 0.50 0.67 0.61 0.5 0.16 0 .5 2 24 .2 2 26 06 07 06 53 48 10 26 6 2 .5 3 .3 7 .5 6 .3 4 .2 4 –0.5 0 1. 0 0. 0. 0. 0. 0. 0. 0. 0. 0. 0 0 0 0 0 8 .3 –1.0 –0 a a zil hile bia ica ic or ala ras ico a a r m eru uay do uay , RB in vi bl d x gu e nt Boli Bra C lom ta R epu cua tem ndu Me a ra ana P rag lva rug ela rg o s a o c P a a C Co R E u H Ni P S U zu A ica n G EI ne in Ve m Do conditional expectations probability Source: Based on data from national household surveys, 2001–04. Note: Figure shows changes in the educational gender gap between cohorts born in 1940 and cohorts born in 1984. The results show that change in the probability of attaining a given level of education is the more important of the two component in most countries, as is the case with the aggregate data for the region reported earlier. There are two exceptions to this pattern: Bolivia and Ecuador. Guatemala is the only country displaying a negative change in gender differences in the probability of attaining a particular level of educa- tion. It appears to be the only country in the region in which the rate of completion of primary, secondary, and university education grew more rapidly for men than for women. Bolivia experienced the largest changes in the expectations component in the region, followed by Panama and El Salvador. The gender schooling gap changed at a rapid pace during the past four decades. For the oldest cohort in the data (people born in 1940), the gap in attainment was 0.6 year favoring men. For the youngest cohort (people born in 1984), the gap favored women by almost half a year. During this period, the gap in attainment changed by 0.27 year of schooling per decade in favor of women. 32 new century, old disparities One of the plausible implications of these changes has to do with changes in marriage markets. People across the world are delaying mar- riage decisions (Schultz 1973; Cochrane 1979; King et al. 1986; Blau, Kahn, and Waldfogel 2000; Saardchom and Lemaire 2005). It would be useful to understand the extent to which this phenomenon is the result of changes in women’s and men’s schooling and the extent to which other forces are driving these trends. Gender Differences in Attendance and Attainment among Children of School Age Although the main focus of this chapter is to explore gender differences in completed average years of schooling across generations, it is instruc- tive to explore the gender gap in children who are still of school age in the countries in which the gap has not yet been closed: Bolivia, Guatemala, Mexico, and Peru. Of particular interest is the role of household income in schooling decisions, given that household economic constraints rep- resent an important barrier to girls’ schooling. For young children, it is possible to examine how both attendance and attainment vary by house- hold income level, something that cannot be done in the analysis of adult attainment.1 Figure 3.8 presents population-weighted school attendance profiles for 6 to 18-year-olds in Bolivia, Guatemala, Mexico, and Peru by gender and per capita household income quintile. Three income groups are displayed: the bottom 20 percent of the per capita household income distribution, the middle 20 percent, and the top 20 percent. Attendance rates among children 8–11 exceed 95 percent, leaving little room for variation across gender or income group; significant differences in attendance by gender are not evident before age 12. At older ages, there is a slight tendency for boys from the lowest income quintile to have higher attendance rates than girls from the same income group. The opposite pattern is evident at the highest income quintile. No gender differences in attainment are evident for the middle and top income quintiles; there is, however, evidence of a small gender gap in favor of men in the bottom quintile (figure 3.9).2 The most striking differences in school attendance (figure 3.8) and attainment (figure 3.9) occurred across income groups rather than by gender, however. For three out of the four countries that did not close the gender school- ing gap (Bolivia, Guatemala, and Peru), it is possible to explore household ethnicity. In Bolivia and Peru, the indigenous classification is based on “mother tongue�; in Guatemala, it is based on self-identification. Both attendance profiles and schooling attainment vary by gender and ethnicity in these three countries (for graphs for each country, see Duryea gender differences in education 33 Figure 3.8 School Attendance Rates in Bolivia, Guatemala, Mexico, and Peru, by Gender, Age, and Per Capita House- hold Income Quintile, Circa 2003 100 90 percentage attending school 80 70 60 50 40 30 20 10 0 6 7 8 9 10 11 12 13 14 15 16 17 18 age men, bottom quintile women, bottom quintile men, middle quintile women, middle quintile men, top quintile women, top quintile Source: Based on data from national household surveys, circa 2003. et al. 2007). Attendance rates in Peru exceed 90 percent for children ages 6–13 for all groups. There is quite a bit of noise in the data for ages 14–18, with an unclear pattern in attendance rates for indigenous people. By the age of 19 and 20, it becomes clear that indigenous people attend school at much lower rates than their nonindigenous peers. Nonindig- enous women display similar schooling attainment as their male peers. In contrast, indigenous women lag behind their male peers by about two full years of schooling. In Bolivia and Guatemala, school attendance of indigenous people lags that of nonindigenous people both at early ages and in the teen years. At age 6, indigenous children in Bolivia are 12–15 percentage points less likely to attend school than nonindigenous children. Attendance rates for indigenous girls start to lag those of indigenous boys at age 9, with a more rapid decline after age 13. Patterns in Guatemala are not as clear, with noisier data reflecting a much smaller sample. Nonetheless, the data reveal that indigenous girls do not attend school at the same rates as their nonindigenous peers. Patterns of school attainment in Bolivia and Guatemala are similar to patterns in Peru. Nonindigenous boys and girls have similar outcomes, 34 new century, old disparities Figure 3.9 Average Years of Educational Attainment in Bolivia, Guatemala, Mexico, and Peru, by Gender, Age, and Per Capita Household Income Quintile, Circa 2003 120 110 mean years of completed schooling 100 90 80 70 60 50 40 30 20 10 0 6 7 8 9 10 11 12 13 14 15 16 17 18 age men, bottom quintile women, bottom quintile men, middle quintile women, middle quintile men, top quintile women, top quintile Source: Based on data from national household surveys, circa 2003. followed by indigenous boys and then indigenous girls. The differences are greatest after ages 13–15. Although the three countries share some features in the patterns of schooling attainment by gender and ethnicity, there is a striking differ- ence in the levels of attainment. At age 15, indigenous girls have achieved 7.1 years of schooling in Peru, 6.1 years in Bolivia, and 4.6 years in Guatemala. Analysis of attendance and attainment for younger children reveals that the largest gender differences in attendance occur among children in the lowest income quintile. Although higher proportions of boys than girls attend schools, boys nonetheless display lower attainment. This result is consistent with the fact that repetition rates are higher among boys. Educational attainment of nonindigenous boys is similar to that of non- indigenous girls in Bolivia, Guatemala, and Peru. In contrast, attainment of indigenous teenage girls lags behind that of indigenous teenage boys. gender differences in education 35 Notes 1. Monetary labor income generated by children is excluded when computing family income, in order to avoid problems with the causality relationship between income generation and schooling. 2. The number of years completed should not be confused with a measure- ment of number of years spent in the schooling system. The measure used is net of repetition. References Behrman, J., and A. Deolalikar. 1988. “Health and Nutrition.� In Handbook of Development Economics, vol. 1, ed. H. Chenery and T. N. Srinivasan, 633–90. Amsterdam: North-Holland. Behrman, R., S. Duryea, and M. Székely. 1999. “Schooling Investments and Mac- roeconomic Conditions: A Micro-Macro Investigation for Latin America and the Caribbean.� Research Department Working Paper 407, Inter-American Development Bank, Washington, DC. Blau, F. D., L. M. Kahn, and J. 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. Cochrane, S. 1979. Fertility and Education: What Do We Really Know? Baltimore, MD: Johns Hopkins University Press. Dollar, D., and R. Gatti. 1999. “Gender Inequality, Income, and Growth: Are Good Times Good for Women?� Gender and Development Working Paper Series 1, World Bank, Washington, DC. http://www.worldbank.org/gender/prr. Duryea, D., S. Galiani, H. Ñopo, and C. Piras, 2007. “The Educational Gender Gap in Latin America and the Caribbean.� RES Working Paper 4510, Inter- American Development Bank, Research Department, Washington, DC. King, E. M., and M. A. Hill, eds. 1993. Women’s Education in Developing Coun- tries. Washington, DC: World Bank. King, E. M., J. Peterson, S. M. Adioetomo, L. T. Domingo, and S. H. Syed. 1986. Change in the Status of Women across Generations in Asia. Santa Monica, CA: Rand Corporation. Knodel, J., and G. Jones. 1996. “Does Promoting Girls’ Schooling Miss the Mark?� Population and Development Review 22 (4): 683–702. Marshall, J., and V. Calderón. 2005. “Social Exclusion in Education in Latin America and the Caribbean.� Discussion draft, Inter-American Development Bank, Sustainable Development Department, Washington DC. Parker, S., and C. Pederzini. 2000. “Gender Differences in Education in Mexico.� World Bank Departmental Working Paper 21023, Washington, DC. Psacharopoulos, G. 1994. “Returns to Investment in Education: A Global Update.� World Development 22 (9): 1325–43. Psacharopoulos, G., and Z. Tzannatos. 1992. “Latin American Women’s Earnings and Participation in the Labor Force.� Policy Research Working Paper 856, World Bank, Washington, DC. 36 new century, old disparities Saardchom, N., and J. Lemaire. 2005. “Causes of Increasing Age at Marriage: An International Regression Study.� Marriage and Family Review 3: 73–97. Schultz, T. P. 1973. “A Preliminary Survey of Economic Analysis of Fertility.� American Economic Review 63 (2): 77–78. ———. 1988. “Education Investments and Returns.� In Handbook of Devel- opment Economics, vol. 1, ed. H. Chenery and T. N. Srinivasan, 543–630, Amsterdam: North-Holland. ———. 1993. “Economics of Women’s Schooling.� In The Politics of Women’s Education, ed. J. K. Conway and S. C. Bourque, 237–44. Ann Arbor: University of Michigan Press. Strauss, J., and D. Thomas. 1995. “Human Resources: Empirical Modeling of Household and Family Decisions.� In Handbook of Development Economics, vol. 3A, ed. J. Behrman and T. N. Srinivasan, 1885–946. Amsterdam: Elsevier. Part II Gender Earnings Gaps 4 More Schooling, Lower Earnings: Women’s Earnings in Latin America and the Caribbean Gender earnings gaps in Latin America and the Caribbean were smaller than in other regions of the world until the late 1950s. The situation reversed after then (Frankema 2008). Since the mid-1980s, the region has seen a steady increase in women’s labor force participation. By the turn of the 21st century, 58 percent of women actively participated in the labor market.1 Despite this improve- ment, in 2007, the World Economic Forum ranked Latin America and the Caribbean the third most unequal region (among nine) in economic participation of and opportunity for women (Hausmann, Tyson, and Zahidi 2007).2 This chapter presents nonparametric earnings gap decompositions in order to assess the extent to which observed gender earnings gaps corre- spond to gaps in individuals’ demographic and job-related characteristics.3 The analysis focuses on labor income earners ages 18–65 from a pooled data set of 18 countries representative of most of the working population in Latin America and the Caribbean; earnings are measured as hourly earnings in the main job. This chapter was adapted from the following sources: “New Century, Old Dis- parities: Gender and Ethnic Wage Gaps in Latin America,� Juan Pablo Atal, Hugo Ñopo, and Natalia Winder, RES Working Paper 4640, Inter-American Develop- ment Bank, 2009; Evolution of Gender Wage Gaps in Latin America at the Turn of the Twentieth Century: An Addendum to ‘New Century, Old Disparities,’ “Hugo Ñopo and Alejandro Hoyos, IZA Discussion Papers 5086, Institute for the Study of Labor, 2010. Juan Pablo Atal is a graduate student in economics at the University of California, Berkeley, and Natalia Winder is a consultant at UNICEF, Division of Policy and Practice, New York. Alejandro Hoyos is a consultant at the Poverty Reduction and Economic Management Network (PREM) at the World Bank. 39 40 new century, old disparities What Does the Literature Show? The evidence suggests that women’s insertion into the labor market has been facilitated by the region’s economic growth, trade liberalization, rapid urbanization, and changes in fertility patterns (Psacharopoulos and Tzannatos 1992b; Cox and Roberts 1993). The increase in women’s labor participation has been accompanied by a slow but steady rise in relative earnings for nearly two decades, allowing women in most countries to contribute about one-third of household income (Duryea, Edwards, and Ureta 2004). However, in most countries in the region, women are more likely than men to hold low-paid occupations (Márquez and Prada 2007), and gender earnings gaps in the region remain substantial. Several authors have attempted to explain the sources of gender earn- ings differentials in the region, exploring issues such as differences in indi- vidual characteristics and human capital endowments (Atal, Ñopo, and Winder 2009); regulation (Lim 2002); fertility (Madrigal 2004; Urdinola and Wodon 2006; Cruces and Galiani 2007); and occupational segregation (Deutsch et al. 2004; Tenjo, Ribero, and Bernat 2006), among others. The literature has also attempted to relate gender earnings gaps to differences in income-generating opportunities in urban and rural areas; however, no clear link can be found (Hertz et al. 2008). In an analysis of 15 countries in the region for which data were available for the late 1980s, Psacharopoulos and Tzannatos (1992a) show that human capital accounts for one-third of the earnings differential, leaving a large portion of the earnings gap unexplained. By the middle of the current decade, most countries in the region had closed the education attainment gender gap (see chapter 3 of this book; Hausmann, Tyson, and Zahidi 2007). Some empirical research provides insights into the linkages between earnings differentials and differences in the types of jobs men and women hold. A review of 13 countries in the region finds that the gender earnings gap appears to be larger on average in the private sector than in the public sector (Panizza and Qiang 2005). Researchers have also examined occupational segregation—the overrep- resentation or underrepresentation of a group (women, men, youth, ethnic groups) in a specific activity—and its linkage with earnings differentials in the region. Most studies find that, in an effort to manage their housework and childcare responsibilities, women may permanently or temporarily withdraw from the labor market, choose occupations with flexible or fewer working hours (Tenjo, Ribero, and Bernat 2006), or invest less in education or on-the-job training, thereby limiting their work experience (Terrell 1992). As a result, women are concentrated in low-paid jobs and face high steeper barriers when attempting to reach higher-level (better-paid) positions. These factors explain only part of the earnings gap in the region. In Costa Rica, Ecuador, and Uruguay, high and persistent levels of occupa- tional segregation explain only a small portion of earnings differentials more schooling, lower earnings for women 41 (Deutsch et al. 2004). A comparative study of Brazil and Mexico shows that despite higher levels of gender occupational segregation in Mexico, gender earnings gaps are wider in Brazil (Salas and Leite 2007). Women have an important presence in the region’s informal sector. Some authors argue that this factor may provide an additional potential explanation for earnings disparities. Plausible explanations include the small impact of education on earnings in the informal sector and the greater importance of experience, where for the most part, men have an advantage over women (Freije 2009). Furthermore, although there may be no real difference in self-employment rates of men and women, there are considerable gender differences in quality, measured not only in terms of average earnings but also in work conditions and income security (Barrientos 2002). Research has examined the role of regulation, such as maternity laws, gender quotas, and employer child care, as drivers of earnings gaps. Cre- ated to protect and provide flexibility for women in certain occupations, labor legislation in areas such as maternity leave and pregnancy protection increase women’s nonsalary labor costs and may therefore increase earn- ings disparities. The empirical evidence in this regard is not clear (Angel- Urdinola and Wodon 2006). Other policies, such as access to affordable childcare and programs to prevent domestic violence, are correlated with increases in both women’s labor force participation and earnings (Deutsch et al. 2004). Differentials may also correspond to women’s roles in society, which, regardless of their skill levels or potential, leads them to choose low- skilled occupations in low-productive sectors (Contreras and Plaza 2004; Tenjo, Ribero, and Bernat 2006). A review of the literature in Atal, Ñopo, and Winder (2009) provides a list of the studies on gender and ethnic earnings gaps for almost all Latin American countries. Most of the studies in that review use household sur- veys to disentangle the causes or components of the earnings gap. How Do Male and Female Workers Differ? Circa 2007, on average, men earn 10 percent more than women in the region. Men earn more than women at all ages; at every level of education; in all types of employment (self-employed, employers, and employees); and in both large and small firms. Only in rural areas do women earn on average the same as their male counterparts. These earnings disparities are reported in the last two columns of table 4.1, where they are computed as multiples of average women’s earn- ings. These disparities may reflect, to some extent, differences in observ- able individual characteristics. Working women in the region have more years of schooling than men. They are nevertheless underrepresented in managerial positions 42 new century, old disparities Table 4.1 Demographic and Job Characteristics and Relative Hourly Earnings of Men and Women in Latin America and the Caribbean, Circa 2007 Relative earnings Composition (base: average women’s (percent) earnings = 100) Men Women Men Women All 100.0 100.0 110.0 100.0 Personal characteristics Age 37.1 36.6 18–24 79.6 74.9 25–34 106.6 100.9 35–44 122.5 108.7 45–54 127.2 111.3 55–65 113.0 97.8 Education level None or primary incomplete 20.9 15.9 73.1 71.1 Primary complete or secondary incomplete 44.5 37.6 95.3 76.0 Secondary complete or tertiary incomplete 29.1 38.0 141.7 118.1 Tertiary complete 5.5 8.5 202.0 178.9 Presence of children (12 years or younger in household) No 52.6 44.7 117.0 105.0 Yes 47.4 55.3 102.2 95.9 Presence of other household member with labor income No 39.8 23.6 108.8 102.0 Yes 60.2 76.4 110.8 99.4 Urban No 26.6 17.5 91.3 92.5 Yes 73.4 82.5 116.8 101.6 Job characteristics Type of employment Employer 4.9 2.3 195.3 180.1 Self-employed 28.0 26.2 95.9 88.8 Employee 67.1 71.5 109.6 101.5 (continued next page) more schooling, lower earnings for women 43 Table 4.1 (continued) Relative earnings Composition (base: average women’s (percent) earnings = 100) Men Women Men Women Part time No 90.7 75.2 105.0 92.2 Yes 9.3 24.8 158.3 123.6 Formality No 56.4 55.9 95.8 86.8 Yes 43.6 44.1 128.4 116.7 Small firm (five workers or less) No 47.6 45.8 115.9 113.7 Yes 52.4 54.2 85.3 78.1 Occupation Professionals and 9.6 15.1 208.7 182.2 technicians Directors and upper management 3.3 2.7 212.5 176.7 Administrative personnel 5.0 10.5 134.0 107.7 Merchants and sellers 9.2 17.2 106.6 93.3 Service workers 11.8 32.5 93.4 70.9 Agricultural workers and similar 15.6 7.1 63.4 80.4 Nonagricultural blue-collars 32.0 9.4 95.6 70.4 Armed forces 0.8 0.1 105.6 116.2 Occupations not classified above 12.7 5.4 110.5 89.9 Economic sector Agriculture, hunting, forestry and fishing 18.1 3.8 59.1 54.0 Mining and quarrying 1.0 0.1 144.3 175.9 Manufacturing 16.7 15.3 115.5 85.4 Electricity, gas, and water supply 0.9 0.2 153.9 165.6 (continued next page) 44 new century, old disparities Table 4.1 (continued) Relative earnings Composition (base: average women’s (percent) earnings = 100) Men Women Men Women Construction 12.1 0.8 97.3 109.3 Wholesale and retail trade, and hotels and restaurants 21.0 27.9 106.6 88.8 Transport, storage 9.0 1.9 115.7 125.0 Financing, insurance, real estate, and business services 3.1 3.1 150.5 149.1 Community, social, and personal services 18.3 46.9 153.9 110.1 Source: Based on data from national household surveys from circa 2007. and overrepresented in other occupations, such as service workers, mer- chants, administrative personnel, and professionals. Differences by eco- nomic sector are also apparent. Construction and agriculture are sectors dominated by men, whereas community, social, and personal services are dominated by women. Important gender differences are also evident in working hours: almost one-fourth of working women are part-time workers, compared with less than one-tenth of working men.4 This section assesses the role of individual differences in earnings gaps. It first provides decompositions of five sets of observable demographic characteristics as control variables. Each set adds a new characteristic to the previous set, in an order that first considers characteristic that are less likely to be endogenous to a model of earnings determination. The full set of demographic control variables (in the order used in the matching exercise) are age, education, presence of children 12 or younger in the household (dummy), presence of other labor income earner in the household (dummy), and urban area (dummy). Country of residence is an implicit control variable in each specification, as only individuals within the same country are matched. Table 4.2 shows the gender earnings gaps, the four components of its decomposition (for five different sets of controls), and the percentages of men and women belonging to the common support of observable char- acteristics (that is, people who were matched). ΔM (ΔF) is the portion of the earnings gap attributed to the existence of men (women) with com- binations of characteristics that are not met by any women (men). ΔX is the portion of the earnings gap attributed to differences in the observ- able characteristics of men and women. Δ0 is the portion of the earnings Table 4.2 Decomposition of Gender Earnings Gap in Latin America and the Caribbean after Controlling for Demographic Characteristics, Circa 2007 (percent) + Presence of other + Presence of children household member with Age + Education in the household labor income + Urban Δ 10.0 10.0 10.0 10.0 10.0 Δ0 8.9 17.2 17.4 17.9 18.8 ΔM 0.0 0.1 0.2 0.2 –0.3 ΔF 0.0 –0.0 –0.1 –0.4 –0.6 ΔX 1.1 –7.2 –7.5 –7.8 –7.9 Percentage of men in common support 100.0 99.8 99.3 97.7 94.7 Percentage of women in common support 100.0 99.9 99.8 99.1 97.9 Source: Based on data from national household surveys from circa 2007. Note: ΔM (ΔF) is the part of the earnings gap attributed to the existence of men (women) with combinations of characteristics that are not met by any women (men). ΔX is the part of the earnings gap attributed to differences in the observable characteristics of men and women over the “common support.� Δ0 is the part of the earnings gap that cannot be attributed to differences in characteristics of the individuals. It is typically attributed to a combination of both unobservable characteristics and discrimination. The sum of these components equals the total earnings gap (ΔM + ΔF + ΔX + Δ0 = Δ). 45 46 new century, old disparities gap attributed to differences between men and women that cannot be explained by observable characteristics. The sum of ΔX, ΔM, ΔF, and Δ0 is equal to the total earnings gap (Δ). More prime-age workers are men and, on average, male workers are older than female workers (probably because women retire earlier). However, after controlling only for age, most of the gender earnings gap remains unexplained (that is, most of Δ is captured by Δ0): only 1 percent- age point of the 10 percentage points in the gender earnings gap can be explained by the differences in age distributions between men and women in the labor market. After controlling for education, the unexplained component of the gender earnings gap is larger than the original gap: if men and women had the same distribution of age and education in the labor market, the gender gap would increase from 10 percent to 17 percent of average women’s earnings. This increase reflects higher educational achievement among women workers than among men, as shown in table 4.1. The unexplained component of the earnings gap is larger than the original gap after control- ling for each subsequent set of controls, remaining almost constant after the addition of each characteristic. The last two rows of table 4.2 show the percentages of matched men and women for each set of characteristics. These percentages are large even when controlling for the set of five characteristics, suggesting that the inclusion of more matching characteristics does not limit the explana- tory capacity of the exercise. Differences in the “common support� do not play a major role in explaining the earnings gap, as confirmed by the small magnitude of both ΔM and ΔF. Job characteristics can now be added. The new variables considered are type of employment (self-employed, employer, or employee); part-time work (a dummy equal to 1 for people working 35 hours or less a week); formality status (a dummy equal to 1 for people covered by social security obtained from their labor relationship); economic sector (nine categories of the International Standard Industrial Classification [ISIC] revision 2 at the one-digit level); occupation (nine categories of a slight modification of the International Standard Classification of Occupations [ISCO] system at the one-digit level); and small firm (dummy equal to 1 if firm has fewer than six workers).5 Because there was no strong a priori belief regarding which variable is “least endogenous� and some of the variables were strongly correlated, the variables were included in a way that differs from the previous analy- sis. The six job characteristics were added separately to the basic set of five sociodemographic matching variables reported in the last column of table 4.2. Including the variables in this way prevents conclusions from being drawn that are likely to depend on the order in which each variable is included. For ease of comparison, the first column of table 4.3 repro- duces the last column of table 4.2. Table 4.3 Decomposition of Gender Earnings Gap in Latin America and the Caribbean after Controlling for Demographic and Job Characteristics, Circa 2007 (percent) Demographic & Type of & Part & Small set employment time & Formality & Sector & Occupation firm Full set Δ 10.0 10.0 10.0 10.0 10.0 10.0 10.0 10.0 Δ0 18.8 17.2 27.3 18.0 23.6 16.8 18.8 19.5 ΔM –0.3 1.1 –0.3 –0.1 –5.0 –0.8 –0.2 –2.0 ΔF –0.6 –1.2 –2.0 –1.0 –0.3 –1.1 –0.9 –2.9 ΔX –7.9 –7.1 –15.0 –6.8 –8.2 –4.9 –7.8 –4.5 Percentage of men in common support 94.7 87.3 91.3 90.8 64.3 73.0 90.8 27.3 Percentage of women in common support 97.9 95.1 93.5 96.4 88.0 86.8 96.3 44.7 Source: Based on data from national household surveys from circa 2007. Note: ΔM (ΔF) is the part of the earnings gap attributed to the existence of men (women) with combinations of characteristics that are not met by any women (men). ΔX is the part of the earnings gap attributed to differences in the observable characteristics of men and women over the “common support.� Δ0 is the part of the earnings gap that cannot be attributed to differences in characteristics of the individuals. It is typically attributed to a combination of both unobservable characteristics and discrimination. The sum of these components equals the total earnings gap (ΔM + ΔF + ΔX + Δ0 = Δ). 47 48 new century, old disparities As shown in table 4.3, none of the job characteristics is able to offset the increase in the unexplained gender earnings gap after controlling for education. The unexplained component of the gap is considerably larger than the original gap after the addition of every job characteristic inde- pendently (and also when they are added together). The unexplained gap widens substantially after controlling for economic sector, suggesting that gender segregation in economic sectors is not by itself the source of earn- ings differentials. The widening of the gap is driven mainly by the over- representation of men in agriculture, the sector with the lowest average earnings. The unexplained gap also widens substantially after controlling for part-time work, as women are overrepresented in part-time jobs, which have an hourly earnings premium over full-time jobs. The four other job-related characteristics (type of employment, formal- ity, occupation, and small firm size) reduce the unexplained component of earnings gaps after controlling for the five demographic characteristics, but only slightly. These findings challenge the popular belief that occupa- tional segregation contributes to gender earnings gaps, reinforcing previ- ous evidence on this issue (Barrientos 2002). The last column of table 4.3, which shows the decomposition exercise after controlling for the full set of observable characteristics, suggests that the unexplained gender earnings gap in the region reaches 20 percent of average women’s earnings. Indeed, the portion explained by gender dif- ferences in individual characteristics over the common support (Δx) is about –5 percent. Differences in the distribution of characteristics of men and women thus favor women because they share characteristics, such as higher educational levels, that are better rewarded in the labor market. Even though the common support is reduced after controlling for the full set of variables, the portion of the gap attributable to the uncommon support is small (in contrast to the results on ethnic earnings differences, presented in other chapters), indicating that barriers to access are not the most important factor explaining gender earnings gaps. A country-by-country exploration of the gender earnings gap decom- positions, reported in table 4.4, provides evidence of cross-country hetero- geneity behind the averages reported in table 4.2. The table provides mea- sures of the original gap and the unexplained component after controlling for three sets of controls: first, age and education; second, the whole set of demographic matching variables; and third, the whole set of demographic and job-related matching variables. In 7 of the 18 countries examined, the original gender earnings gaps reported in table 4.4 are negative, reflecting higher average earnings for women than men. These results do not stand when comparing men and women with the same observable characteristics. In the first specification, Δ0 is statistically equal to zero in Bolivia and Guatemala and 29.7 percent in Brazil. The influence of controlling by education varies significantly from country to country. Whereas in Peru more schooling, lower earnings for women 49 Table 4.4 Original and Unexplained Components of Gender Earnings Gap in Latin America and the Caribbean by Country, Circa 2007 (percent) Δ0 + Presence of + Part-time, children in the formality, household, occupation, presence of other economic income earner in sector, type of Age and the household, employment, and Country Δ education and urban small firm Argentina 0.5 14.2*** 12.6*** 10.8*** Bolivia –5.5 –1.8 3.0 17.8 Brazil 20.5 29.7*** 31.4*** 26.4*** Chile 10.9 19.3*** 18.6*** 13.1*** Colombia –0.9 7.1*** 6.3*** 7.3*** Costa Rica –5.8 13.7*** 13.6*** 17.9*** Dominican Republic –3.1 16.6*** 17.3*** 23.9*** Ecuador –3.2 16.4*** 13.6*** 5.6 El Salvador 3.3 11.9*** 16.0*** 11.3*** Guatemala –3.3 0.3 –0.7 17.7*** Honduras 5.6 16.3*** 16.3*** 24.2*** Mexico 2.6 7.8*** 10.5*** 15.3*** Nicaragua 1.5 20.3*** 19.3*** 28.4*** Panama –8.6 13.6*** 16.2*** 10.4** Peru 18.3 19.4*** 25.9*** 23.5*** Paraguay 6.2 16.0*** 13.8*** 6.9 Uruguay 5.7 26.3*** 27.5*** 23.4*** Venezuela, RB 0.4 13.9*** 13.8*** 12.3*** Latin America and the Caribbean 10.0 17.2 18.8 19.5 Source: Based on data from national household surveys from circa 2007. Note: **p < 0.05, ***p < 0.01. Δ is the total earnings gap. Δ0 is the part of the gap attributed to differences between men and women that cannot be explained by observable characteristics. 50 new century, old disparities the unexplained component of the gap is almost equal to the original gap, reflecting small educational differences by gender, in Argentina the unex- plained component is almost 30 times the original gap. Gender differences in educational attainment for both countries are large, especially at the extremes of the distributions. At the lower extreme of educational dis- tributions, the proportion of workers without education in Argentina is almost zero for both men and women; in Peru, the situation is unfavor- able for women, as 7 percent of female workers but only 2 percent of male workers have no education. Among people with tertiary education, in Argentina, the educational gaps are wider: 40 percent of women and 25 percent of men have tertiary education. In Peru, 29 percent of women and 24 percent of men have tertiary education. Figure 4.1 presents the four components of the earnings gap by country (sorted by the magnitude of the unexplained component) for the specifica- tion with the full set of control variables. Beyond the heterogeneity in the magnitudes of every component, interesting qualitative patterns arise. The portion of the gap attributable to differences in distributions of observable characteristics over the common support (ΔX) is negative in every country, indicating that in every country in the region, women have combinations of characteristics (especially educational attainment) that are expected to yield higher labor market returns for them than for men. Women’s lower access to well-paid jobs or combinations of observable characteristics explain a substantial part of the earnings gap in Bolivia, Guatemala, Nicaragua, and Paraguay. At the other extreme, women’s confinement to lower-paid segments of the labor market is prevalent in Argentina, Colombia, Costa Rica, Ecuador, El Salvador, Panama, and Peru. In the first group of countries, the evidence suggests that the problem of gender earnings gaps is linked to barriers in access to high-paying occu- pations (the “chief executive officer [CEO] effect�); in the second group of countries, earnings gaps seem to be linked to women’s confinement to low-paying segments of the labor market (the “maid effect�). An advantage of the matching approach over traditional decomposi- tion is that it is informative not only about the average unexplained gap but also about its distribution. Further evidence of the heterogeneity of the decomposition results appears when the unexplained component of the earnings gaps (after controlling for all demographic and job-related characteristics) is reported for different segments of the labor market (figure 4.2). Richer information about the nature of the unexplained gender earnings gaps emerges that can explain the problem and provide policy advice on how to address it. The observations that emerge from the distribution of unexplained gender pay differentials include the following: • The unexplained gender earnings gap increases with age. Although one possible (and optimistic) interpretation of this result is that more schooling, lower earnings for women 51 Figure 4.1 Decomposition of Gender Earnings Gap in Latin America and the Caribbean, by Country, Circa 2007 after Controlling for Demographic and Job Characteristics Nicaragua (Δ = 1.5%) Brazil (Δ = 20.5%) Honduras (Δ = 5.6%) Dominican Republic (Δ = –3.1%) Peru (Δ = 18.3%) Uruguay (Δ = 5.7%) Costa Rica (Δ = –5.8%) Bolivia (Δ = –5.5%) Guatemala (Δ = –3.3%) Mexico (Δ = 2.6%) Chile (Δ = 10.9%) Venezuela, RB (Δ = 0.4%) El Salvador (Δ = 3.3%) Argentina (Δ = 0.5%) Panama (Δ = –8.6%) Colombia (Δ = –0.9%) Paraguay (Δ = 6.2%) Ecuador (Δ = –3.2%) –60 –40 –20 0 20 40 60 percentage of average women’s earnings Δ0 ΔM ΔF ΔX Source: Based on data from national household surveys from circa 2007. Note: ΔM (ΔF) is the part of the earnings gap attributed to the existence of men (women) with combinations of characteristics that are not met by any women (men). ΔX is the part of the earnings gap attributed to differences in the observable characteristics of men and women over the “common support.� Δ0 is the part of the earnings gap that cannot be attributed to differences in characteristics of the individuals. It is typically attributed to a combination of both unobservable characteristics and discrimination. The sum of these components equals the total earnings gap (ΔM + ΔF + ΔX + Δ0 = Δ). earnings gaps are narrowing over time, such an assertion must be made with caution, as this finding could also be driven by unob- servable characteristics correlated with age. For instance, this result may reflect gender differences in labor experience, which could be exacerbated over time as women bear and raise children. Indeed, the unexplained component of the gender gap is slightly larger (although not statistically significant so) among workers with children. • The unexplained gender earnings gap is smaller among people with tertiary education. One possible explanation is that more educated women fill positions in firms in which there is less room for dis- cretionary earnings setting or other discriminatory behavior. This hypothesis is supported by the fact that the unexplained earnings gap 52 new century, old disparities Figure 4.2 Confidence Intervals for Unexplained Gender Earnings Gap in Latin America and the Caribbean, after Controlling for Demographic and Job Characteristics, Circa 2007 percentage of average women’s earnings a. Controlling for age b. Controlling for education 30 40 percentage of average women’s earnings percentage of average women’s earnings 30 25 20 30 10 15 0 ne e e e e e e et et et et et et no pl pl pl pl pl pl m m m m m m 10 co co co co co co in in in y y y ar ar ar y y y ar ar ar im nd rti 24 4 4 4 54 –3 –4 –5 im nd rti te pr co < > te 24 34 44 pr co se se age c. Controlling for presence of children d. Controlling for presence of other under 12 in household household member with labor 22 24 income percentage of average women’s earnings percentage of average women’s earnings 21 22 20 19 20 18 17 17 no yes no yes (continued next page) more schooling, lower earnings for women 53 Figure 4.2 (continued) e. Controlling for urban location f. Controlling for type of employment 24 30 percentage of average women’s earnings percentage of average women’s earnings 25 22 20 20 15 10 18 5 ed ed ee oy oy oy 16 pl pl pl em em em lf- se no yes g. Controlling for part-time employment h. Controlling for formality of employment 24 24 percentage of average women’s earnings percentage of average women’s earnings 22 22 20 20 18 18 16 16 no yes no yes (continued next page) 54 new century, old disparities Figure 4.2 (continued) i. Controlling for occupation j. Controlling for sector percentage of average women’s earnings ag percentage of average women’s earnings 10 40 20 20 0 0 –20 –40 –20 –60 m rec als tra s na icu ice se ve ic ra or rs ra r s m or s t c fo rs ss s ed el an m ure tri ac ing an co su ng tra re stu ply fin spo tau tion rs ial tor s al rv e rv s es is r tu wo r ar l w ker la rce pe nc rt,s ant se ice on se ag in to gr ltu w lle ul l ke no ed ke ti ifi ad di ion ic ty ri n p ec uf in t n s c ci tu ul r s ric es of pr no agr erv d m a s s, el ot ,h de tra k. Controlling for small firm percentage of average women’s earnings 25 20 15 10 no yes Source: Based on data from national household surveys from circa 2007. Note: Figures show results after controlling for demographic and job-related characteristics. Boxes show 90 percent confidence intervals for unexplained earnings; whiskers show 99 percent confidence intervals. is also smaller among formal workers and very high in small firms (where there are fewer highly educated workers). • The unexplained gender earnings gap is larger among informal workers and at small firms. These findings reinforce the idea that better-educated women are able to find niches within the labor mar- ket where there is less room for discriminatory behavior, whereas more schooling, lower earnings for women 55 women with lower education are confined to segments in which there is more room for discretionary earnings setting. • The unexplained gender earnings gap is larger among the self- employed. This finding challenges the view that claims that gen- der earnings gaps reflect discrimination by employers. It does leave room for customer discrimination. Linked to this result, the unex- plained gender earnings gap is also highly dispersed across employ- ers, reflecting possible heterogeneities in entrepreneurial abilities and success. • The unexplained gender earnings gap is negative in the mining sector and the armed forces. These professions and sectors are dominated by men: 0.77 percent of men but just 0.08 percent of women are employed in the armed forces, and 0.95 percent of men and just 0.14 percent of women work in mining. The few women who obtain a job in these environments dominated by men enjoy a considerable premium, however, on average earning more than their male coun- terparts. Presumably, selection plays an important role or the jobs women perform in these sectors differ substantially from the jobs men perform. Figure 4.3 shows the magnitude of unexplained earnings gaps along percentiles of the earnings distribution. The earnings gap between the representative man and woman is calculated at each percentile of the dis- tributions of earnings using the matched samples. Earnings differences are thus the differences that remain unexplained after controlling for observ- able characteristics. The results depicted in figure 4.3 show larger unexplained earnings gaps at the lower end of the earnings distribution, followed by a sharp decrease after the 6th percentile, a somewhat flat or slightly increasing pattern in the middle, and a negative slope in the upper tail of the dis- tribution (after the 80th percentile). The introduction of education as a matching variable increases the unexplained gender earnings gap, but it does not so do homogeneously along the distribution. The introduction of the presence of children and other income earners in the household leaves almost unchanged the magnitude of the unexplained gender earnings gaps for percentiles 40 and above but increases the magnitude by almost 10 percentage points for the lower percentiles (5–15). One job characteristic—part-time work—is particularly important to highlight, because, as in the case of education, its inclusion increases the unexplained gender earnings gap. The increase is not homogenous—in fact, it is negligible until the 25th percentile of the earnings distribution, at which point it starts increasing. The inclusion of the part-time job variable causes an increase of 15 percentage points in the unexplained gap for the top 20 percentiles of the earnings distribution. The introduction of each of 56 new century, old disparities Figure 4.3 Unexplained Gender Earnings Gap in Latin America and the Caribbean, by Percentiles of Earnings Distribution, Circa 2007 70 a. Controlling for age and education percentage of average women’s 60 50 earnings gap 40 30 20 10 0 0 10 20 30 40 50 60 70 80 90 100 percentile age + education 70 b. Controlling by set of observable characteristics percentage of average women’s 60 50 earnings gap 40 30 20 10 0 0 10 20 30 40 50 60 70 80 90 100 percentile age + education + presence of children in household + presence of other household member with labor income + urban + type of employment + part time Source: Based on data from national household surveys from circa 2007. the other labor characteristics reduces the unexplained component of the gap from the level it reaches when adding the part-time job variable. When the complete set of job-related characteristics is included, the unexplained component of the gender gap increases among the lowest- earning individuals (percentiles 1–5), decreases among lower-earning individuals (percentiles 6–35), and increases for workers at the upper end of the distribution (percentiles 65 and above). This finding suggests more schooling, lower earnings for women 57 important differences in the ways gender segmentation occurs in the labor market and the impacts of gender segmentation on labor earnings. Linkages between Unexplained Gender Earnings Gaps and Macroeconomic, Social, and Governance Indicators The gender earnings gaps that remain after controlling for differences in observable characteristics between men and women may reflect mac- roeconomic conditions. Economies may be shaped such that economic sectors that favor men are more developed than others, or the extent to which economies are open for trade with the rest of the world may favor the development of certain occupations that are dominated by men or by women. Along similar lines, it can be argued that the way in which social invest- ments are determined (in health and education, for instance) imposes certain conditions that favor the possibilities for high performance in the labor market differently for men and women. It could also be that the level of interpersonal trust and individuals’ satisfaction with the performance of (political and market) institutions are linked to egalitarian attitudes and actions that operate in the labor market. This section explores the possible linkages between these aggregate conditions and the unexplained gender earnings gaps from a cross-country perspective. It groups the aggregate variables considered for this exercise into four categories: • macroeconomics and fundamentals (growth, gross domestic product [GDP] per capita, foreign investment, expenditure per capita, and so forth) • sociodemographics and social spending (adolescent birth rate, life expectancy at birth, marital status, public spending on education, and so forth) • employment (women’s labor force participation, participation of women in industry, vulnerable employment on women, hiring and firing practices, and so forth) • governance (interpersonal trust, satisfaction with local services, sat- isfaction with the market economy, percentage of female legislators, and so forth). These variables were collected from the following sources: United Nations Children’s Fund (UNICEF); the World Bank’s World Develop- ment Indicators, Millennium Development Goals, Gender Statistics, and Health Nutrition Population Statistics; the Latin American Public Opinion Project’s Americas Barometer; the Fraser Institute’s Economic Freedom of 58 new century, old disparities the World; the World Economic Forum’s Global Competitiveness Report; the Bertelsmann Foundation; and Latinobarómetro. From these data sources, only data that were available for at least 15 countries during the relevant period of analysis (circa 2007) were selected. Table 4.5 shows the variables, the years for which they were available, the number of countries for which data were available, the correlation coefficient between the vari- ables and the unexplained gender earnings gap, and the data source. Only a few variables show a statistically significant correlation with the unexplained gender earnings gaps: employee, industry, female (percent of women’s employment); female legislators, senior officials, and managers (percent of total); and labor market liberalization index. The variables for which there is a significant correlation with the unexplained gender earn- ings gap are plotted in figures 4.4–4.6 Figure 4.4 reports the positive relationship between the earnings gap and the percentage of women employed in industry—a sector clearly dominated by men (there are 12 times more men than women in con- struction and 6 times more men than women in agriculture, for instance). The figure shows that 15 percent of employees in industry in Latin America and the Caribbean are women. Peru is an outlier ( 40 percent of employed women work in industry); for this reason, figure 4.4 includes two fitted lines, one including Peru and one without it. This figure sug- gests that economies with greater participation of women in sectors dominated by men have larger gender earnings disparities. This appar- ently paradoxical result is explored further in chapter 6, on Mexico, where, based on econometrics and a simple theoretical model linking segregation and earnings gaps, the result is substantiated. The finding raises some warnings about the apparent benefits of reducing occupa- tional segregation. The second statistically significant relationship among the variables explored also seems to be paradoxical. Figure 4.5 shows a positive rela- tionship between the percentage of female legislators, senior officials, and managers and the size of the gender earnings gap. Countries in which women’s visibility at top positions is higher tend to have larger unex- plained gender earnings gaps in the aggregate. The same positive correla- tion holds for the subsample of highly educated people, although the cor- relation is no longer statistical significant (this result is not reported but available upon request). In countries in which women hold top positions, their status seems to be coming at the price of lower earnings. Women are thus breaking some “glass doors� (to get into selected high-profile posi- tions) but still facing some “glass ceilings� (in the sense that they are not remunerated accordingly). This result is similar to another finding reported in this book regarding women’ entrance into flexible segments of the labor market at the price of lower earnings. Examining the same variables for European countries (not reported but available upon request to the author of this book) shows no Table 4.5 Correlation between Gender Earnings Gap and Economic Indicators in Latin America and the Caribbean, Circa 2007 Number of Correlation Variable Years countries coefficient Source Macroeconomics and fundamentals Domestic credit provided by banking sector (percentage of GDP) 2003–07 18 0.3 World Development Indicators Exports of goods and services (constant 2000 U.S. dollars) 2003–07 18 0.1 World Development Indicators Foreign direct investment, net inflows (percentage of GDP) 2003–07 18 0.2 World Development Indicators GDP per capita growth (annual percentage) 2003–07 18 0.0 World Development Indicators GDP per capita, (purchasing power parity) (constant 2005 international $) 2003–07 18 –0.2 World Development Indicators Imports of goods and services (constant 2000 U.S. dollars) 2003–07 18 0.1 World Development Indicators Industry, value added (percentage of GDP) 2003–07 18 –0.1 World Development Indicators Sociodemographics and social spending Adolescent birth rate, number of births per 1,000 girls 15–19 years old 2000–08 18 0.1 UNICEF Adolescent fertility rate (births per 1,000 women Health, nutrition, and 15–19) years old 2003–07 18 0.2 population statistics Household final consumption expenditure per capita 59 (constant 2000 U.S. dollars) 2003–07 18 –0.1 World Development Indicators (continued next page) 60 Table 4.5 (continued) Number of Correlation Variable Years countries coefficient Source Fertility rate, total (births per woman) 2003–07 18 0.1 Health, nutrition, and population statistics Life expectancy at birth, female (years) 2003–07 18 –0.2 Health, nutrition, and population statistics Life expectancy at birth, male (years) 2003–07 18 –0.1 Health, nutrition, and population statistics Population growth (annual percentage) 2003–07 18 –0.1 Health, nutrition, and population statistics Public spending on education, total (percentage of government expenditure) 1983–87 18 –0.4 World Development Indicators Survival to age 65, female (percentage of cohort) 2003–07 18 –0.3 Health, nutrition, and population statistics Survival to age 65, male (percentage of cohort) 2003–07 18 –0.2 Health, nutrition, and population statistics Employment Employees, agriculture, female (percentage of female employment) 2003–07 17 0.1 Gender statistics Number of Correlation Variable Years countries coefficient Source Employees, agriculture, male (percentage of male 2003–07 17 0.3 Gender statistics employment) Employees, industry, female (percentage of female 2003–07 17 0.5* Gender statistics employment) Employees, industry, male (percentage of male 2003–07 17 0.1 Gender statistics employment) Employees, services, female (percentage of female 2003–07 17 –0.4 Gender statistics employment) Employees, services, male (percentage of male 2003–07 17 –0.4 Gender statistics employment) Employment to population ratio, 15+, female 2003–07 18 –0.2 Millenium Development Goals (percentage) Employment to population ratio, 15+, male 2003–07 18 0.2 Millenium Development Goals (percentage) Flexibility of earnings determination 2009–10 18 –0.1 Global Competitiveness Report Hiring and firing practices 2009–10 18 0.2 Global Competitiveness Report Labor force participation rate, female (percentage of 2003–07 18 0.2 Gender statistics female population 15-64) Labor force participation rate, male (percentage of 2003–07 18 0.2 Gender statistics male population 15-64) Labor market liberalization index 2007 18 0.4* Economic Freedom of the 61 World (continued next page) 62 Table 4.5 (continued) Number of Correlation Variable Years countries coefficient Source Governance Interpersonal trust 2008–09 18 –0.1 Americas Barometer Satisfaction with local services 2008–09 18 –0.2 Americas Barometer Trust in political parties 2008–09 18 0.2 Americas Barometer Female legislators, senior officials, and managers 2003–07 16 0.4* World Development Indicators (percentage of total) Political transformation (Bertelsmann transformation 2008 18 0.2 Bertelsmann Foundation index) Proportion of seats held by women in national 2003–07 18 0.1 World Development Indicators parliaments (percent) Public institutions index 2009–10 18 0.1 Global Competitiveness Report Satisfaction with democracy 2009 18 0.1 Latinobarómetro Satisfaction with market economy 2009 18 0.1 Latinobarómetro Strength of legal rights index (0 = weak to 10 = strong) 2003–07 17 0.1 World Development Indicators Sources: UNICEF; World Bank’s World Development Indicators, Millennium Development Goals, gender statistics, and health, nutrition, and population statistics; Latin American Public Opinion Project’s Americas Barometer; Fraser Institute’s Economic Freedom of the World; World Economic Forum’s Global Competitiveness Report; Bertelsmann Foundation; Latinobarómetro; and calculations based on Inter–American Development Bank’s harmonized household surveys from circa 2007. Note: * p < 0.10. more schooling, lower earnings for women 63 Figure 4.4 Female Employment in Industry versus Unexplained Gender Earnings Gaps, Circa 2007 30 Nicaragua 2 R = 0.18 Brazil unexplained gender earnings gap (%) 25 Dominican Honduras Republic R 2 = 0.18 Peru Uruguay 20 Costa Rica Guatemala 15 Mexico Chile Venezuela, RB Argentina El Salvador 10 Panama Paraguay Colombia 5 Ecuador 10 20 30 40 50 employees, industry, female (percentage of employment) Δ0 (full set) fitted values (Peru not included) fitted values Sources: Based on data from the World Bank’s gender statistics and data from national household surveys, circa 2007. correlation between women’s participation in top positions and the gender earnings gap. The third positive correlation is between labor market liberalization and the unexplained gender earnings gap (Figure 4.6). Countries in which workers have less job security, allowing more room for earnings negotia- tion, tend to have larger gender earnings disparities. This correlation may be linked to the tendency of women to be less willing to negotiate, in labor markets and out of them (Babcock and Laschever 2003). These findings are merely correlations; there is no attempt to attribute causality. Nonetheless, it is noteworthy that among more than 100 aggre- gate variables explored, only three showed statistically significant correla- tions with the unexplained gender earnings gap (and two of them showed apparently paradoxical results, although one of the apparent paradoxes is disentangled in chapter 6). This finding may suggest that the problem of gender earnings disparities is microeconomic rather than macroeconomic, probably linked more closely to the persistence of cultural biases in favor of men’s role in society and women’s lack of empowerment and less linked to GDP growth or the trade balance. The reasons behind the correlations, 64 new century, old disparities Figure 4.5 Female Legislators, Senior Officials, and Managers versus Unexplained Gender Earnings Gaps, Circa 2007 30 Nicaragua unexplained gender earnings gap (%) Brazil 25 Honduras Peru Dominican Republic Uruguay R 2 = 0.18 20 Costa Rica Bolivia Mexico 15 Chile Venezuela, RB El Salvador Argentina Panama 10 Colombia Ecuador 5 25 30 35 40 45 women legislators, senior officials, and managers (percenatge of total) Δ0 (full set) fitted values Sources: Based on data from the World Bank’s World Development Indicators and data from national household surveys, circa 2007. Note: Legislators, senior officials, and managers corresponds to the ISCO-88, major group 1: legislators and senior officials (government), corporate managers, and general managers (private sector). however, are not entirely known. More research is needed to investigate these linkages. How Did Differences between Male and Female Workers Change between Circa 1992 and Circa 2007? The figures presented up to this point describe gender earnings disparities at a point in time, circa 2007. Do the results for circa 2007 represent a change since circa 1992? The rest of this chapter analyzes the evolution of gender earnings gaps in the same 18 countries between 1992 and 2007. It compares two data points, without making inferences about trajectories of the variables under analysis during the period. Metaphorically, this section compares two more schooling, lower earnings for women 65 Figure 4.6 Labor Market Liberalization Index versus Unexplained Gender Earnings Gaps, Circa 2007 30 Nicaragua unexplained gender earnings gap (%) Brazil 25 Dominican Republic Honduras Peru Uruguay R 2 = 0.16 20 Bolivia Guatemala Costa Rica 15 Mexico Venezuela, RB Chile El Salvador Argentina 10 Panama Paraguay Colombia Ecuador 5 3 4 5 6 7 labor market liberalization index Δ0 (full set) fitted values Sources: Based on data from the World Bank’s World Development Indicators and data from national household surveys, circa 2007. photographs; it does not show the film of what happened between them. The approach is the same as that described in chapter 2. The Evolution of the Earnings Gap at the Turn of the 20th Century Table 4.6 shows relative labor earnings for men and women in circa 1992 and 2007. Earnings are normalized so that average women’s earnings are equal to 100 for both years. Average men’s earnings can be read directly as the gender earnings gap, which declined from 16.3 to 8.9 percent of average women’s earnings between 1992 and 2007. Earning patterns are remarkably similar across years. Working youth show the lowest earnings; as individuals age, earnings rise up to a mature age, at which point they drop slightly. There is also a clear pattern of earn- ings progression along the educational ladder. The presence of children (in this analysis: six years old and younger) in the household is linked to lower labor earnings; the presence of other labor income earners at home seems to be linked to no significant earnings differences. For both women 66 new century, old disparities Table 4.6 Relative Hourly Earnings for Men and Women in Latin America and the Caribbean by Demographic and Job Characteristics, Circa 1992 and 2007 Circa 1992 Circa 2007 (base: average women’s (base: average women’s earnings = 100) earnings = 100) Men Women Men Women All 116.3 100.0 108.8 100.0 Personal characteristics Age 15–24 78.4 72.6 71.1 69.1 25–34 121.0 110.5 106.0 101.0 35–44 139.2 115.9 121.0 109.2 45–54 134.4 105.9 132.5 114.1 55–64 113.4 86.6 119.0 104.7 Education level None 62.0 52.6 55.8 52.3 Incomplete primary 90.7 65.1 74.0 61.2 Primary complete 104.8 80.6 84.1 67.3 Secondary 106.4 83.6 87.9 73.0 incomplete Secondary 148.0 124.2 116.2 90.7 complete Tertiary incomplete 193.8 157.4 156.7 132.2 Tertiary complete 271.6 214.9 242.6 203.6 Presence of children (6 years or younger in the household) No 119.4 102.3 110.9 101.5 Yes 100.2 82.6 87.0 79.2 Presence of other household member with labor income No 124.4 107.8 109.8 103.9 Yes 111.1 98.1 108.3 98.9 Urban No 78.4 66.1 71.7 69.2 Yes 130.4 107.2 117.0 103.8 (continued next page) more schooling, lower earnings for women 67 Table 4.6 (continued) Circa 1992 Circa 2007 (base: average women’s (base: average women’s earnings = 100) earnings = 100) Men Women Men Women Job characteristics Type of employment Employer 197.8 181.9 195.9 187.9 Employee 113.6 103.7 107.4 102.4 Self-employed 104.5 83.1 92.2 81.5 Time worked Part time 148.3 121.1 130.4 114.9 Full time 120.8 102.3 111.3 101.2 Over time 97.0 61.1 93.5 69.7 Source: Based on data from national household surveys from circa 1992 and circa 2007. and men, hourly labor earnings are significantly higher in urban areas, for both employers and part-time workers. Not all observable characteristics used in the analysis for circa 2007 alone can be used here, because some of them are not available for some countries in their surveys circa 1992. This is particularly the case for some variables related to individuals’ jobs. Nonetheless, most of the vari- ables are available and comparable. Table 4.7 shows the distribution of observable individual and job characteristics for men and women for each period. These descriptive statistics show demographic changes among the working population. In both periods, the percentages of men 55–64 years are higher than the percentage of women, although there was an increase for both women and men. Workers are staying in the labor market longer, but gender differences in retirement age remain. The gender gap in educational attainment widened during this 15-year span. In circa 1992, 16 percent of women and just 11 percent of men had (complete or incomplete) tertiary levels of education. By circa 2007, the percentages had increased for both, but the increase was greater for women: 26 percent of women and 17 percent of men had attained at least some tertiary education. Another characteristic that changed during this period is fertility. The percentages of women and men who live with children at home fell by almost half. By circa 2007, only about 7 percent of the working popula- tion had a child six or under at home. 68 new century, old disparities Table 4.7 Demographic and Job Characteristics of Men and Women in Latin America and the Caribbean, Circa 1992 and 2007 (percent) Circa 1992 Circa 2007 Men Women Men Women Personal characteristics Age 15–24 24.1 26.0 20.1 18.7 25–34 29.5 30.4 27.3 28.1 35–44 23.7 24.7 24.4 26.4 45–54 14.5 13.2 18.5 19.0 55–64 8.2 5.8 9.8 7.9 Education None 8.0 7.7 4.1 3.4 Incomplete primary 37.3 31.1 24.7 18.7 Primary complete 14.4 12.1 14.4 12.1 Secondary incomplete 16.6 15.0 20.0 17.3 Secondary complete 13.1 17.8 19.6 22.6 Tertiary incomplete 4.5 6.6 7.1 10.4 Tertiary complete 6.2 9.8 10.2 15.7 Presence of children (6 years or younger in the household) No 84.1 88.6 91.2 93.2 Yes 16.0 11.4 8.9 6.8 Presence of other household member with labor income No 39.4 19.6 34.6 21.3 Yes 60.6 80.4 65.5 78.8 Urban No 27.1 17.6 18.1 11.1 Yes 72.9 82.5 81.9 88.9 Job characteristics Type of employment Employer 6.0 2.2 5.6 2.9 Employee 68.38 71.90 70.65 73.80 (continued next page) more schooling, lower earnings for women 69 Table 4.7 (continued) Circa 1992 Circa 2007 Men Women Men Women Self-employed 25.61 25.94 23.71 23.32 Time worked Part time 11.29 31.41 13.54 32.20 Full time 56.89 48.60 57.78 50.08 Over time 31.83 19.98 28.68 17.71 Source: Based on data from national household surveys from circa 1992 and circa 2007. Another demographic change is marital and cohabitation arrange- ments. The percentage of men who live with another labor income earner at home increased 5 percentage points between circa 1992 and 2007, and the percentage of women dropped 2 percentage points. Both demographic changes are symptomatic of a process of changes in household and gen- der dynamics that societies (and labor markets) in the region have been experiencing. The data also show that the region continued to urbanize. The percent- ages of urban workers increased about 8 percentage points during this 15-year span. During this period, there was also a slight decrease in self- employment and overtime work and a slight increase in part-time work for both women and men.6 Table 4.8 shows the decomposition exercise for the two periods for var- ious sets of observable characteristics: the overall earnings gap dropped from 16.3 percent of average women’s earnings to 8.9 percent during this 15-year span. The components of the gender earnings gap attributable to the segregation of men or women to certain segments of the labor market in which there are no peers of the opposite sex is almost zero: ΔM and ΔF are different from zero with statistical significance (at the 99 percent level) only when all controls are included in period 1 (circa 1992). In some other circumstances, ΔM is statistically significant; in even fewer circumstances, ΔF is statistically different from zero. In addition, the measure of the common supports increases for both men and women in period 2 (circa 2007). Although this change is probably linked to the larger sample sizes in period 2, it may also be indicative of a reduction in gender differences in observable characteristics. The results suggest progress in reducing the access barriers of women and men to all segments of the labor market. More still needs to be done to reduce remaining gender pay differentials, however. Unexplained gender earnings gaps increased between circa 1992 and 2007, particularly after adding education (which increases the unexplained 70 Table 4.8 Decomposition of Gender Earnings Gap in Latin America and the Caribbean after Controlling for Demographic and Job Characteristics, Circa 1992 and 2007 (percent) + Presence of + Presence of other household children in the member with labor + Type of Age + Education household income + Urban employment + Time worked Period Period 1 (circa 1992) Δ 16.3 16.3 16.3 16.3 16.3 16.3 16.3 Δ0 13.4 25.2 25.4 24.9 25.0 24.0 33.7 ΔM 0.0 0.4 0.5 0.8 0.1 2.2 1.3 ΔF 0.0 –0.1 0.1 –0.1 0.1 0.3 –1.4 ΔX 2.9 –9.2 –9.7 –8.4 –8.8 –10.2 –17.2 percentage of men in common support 100.0 99.5 98.2 93.4 89.3 79.6 65.6 percentage of women in common support 100.0 99.9 99.5 98.9 97.4 92.8 80.7 Period 2 (circa 2007) Δ 8.8 8.8 8.8 8.8 8.8 8.8 8.8 Δ0 9.7 22.2 22.2 21.8 22.6 20.8 29.6 ΔM 0.0 0.1 0.1 –0.3 –0.9 –0.3 –2.1 ΔF 0.0 0.1 0.1 0.1 0.2 0.3 0.4 ΔX –0.9 –13.4 –13.4 –12.9 –13.1 –12.0 –19.1 percentage of men in common support 100.0 99.9 99.2 97.4 95.3 89.6 79.4 percentage of women in common support 100.0 100.0 99.7 99.4 98.8 96.4 89.1 Source: Based on data from national household surveys from circa 1992 and 2007. 71 72 new century, old disparities gap to 12 percentage points in both periods) and time worked (which increases the unexplained gap 3–4 percentage points in both periods). The other observable characteristics do not greatly change the unexplained earnings gap. The unexplained gender earnings gaps move in the same direction in the two periods when adding control characteristics, suggest- ing that the role of observable characteristics in explaining gender earn- ings gaps is qualitatively similar during both periods. Figure 4.7 reports confidence intervals for the unexplained gender earnings gaps for various combinations of matching variables during circa 1992 and circa 2007 (the sequence follows the same pattern as in table 4.8). It shows decreasing unexplained earnings gaps for all controls included. In addition, the confidence intervals for circa 1992 do not inter- cept with the corresponding confidence intervals for circa 2007 in any of the pairs of unexplained earnings gaps shown. As a result, the reduction Figure 4.7 Confidence Intervals for Unexplained Gender Earnings Gap in Latin America and the Caribbean after Controlling for Demographic and Job Characteristics, Circa 1992 and 2007 35 percentage of women’s earnings 30 25 20 15 10 age +education +children +other +urban +type of +time labor employment worked income earner circa 1997 circa 2007 circa 1997 circa 2007 Source: Based on data from national household surveys from circa 1992 and 2007. Note: Figures show results after controlling for demographic and job related characteristics. Boxes show 90 percent confidence intervals for unexplained earnings; whiskers show 99 percent confidence intervals. more schooling, lower earnings for women 73 in unexplained earnings gaps is statistically significant and robust to dif- ferent specifications. There are two important increases in the unexplained components of the gender earnings gaps, both statistically significant for both periods of analysis. The first occurs after adding education. No characteristic added after education can offset the fact that the education control results in larger unexplained gender earnings gaps. In fact, the addition of a last characteristic, time worked, increases the unexplained components of the gaps in both periods.7 The declines in the unexplained components of the earnings gaps between circa 1992 and 2007 may reflect the general trend of narrowing gaps for all segments of the labor market. It could also be the result of changes over time in the distribution of individuals’ observable character- istics, which change the composition of the labor market. If it were the case that women moved to segments of the market with less (more) evidence of unexplained earnings gaps during this 15-year span, one would expect a reduction (increase) in earnings gaps like the one shown in figure 4.7. A “matching-after-matching� exercise is conducted to disentangle the effects of general trends versus changes in the composition of the labor market. Using the matching approach, each matched set (in a given year of data) corresponds to a hypothetical world in which men and women have the same distribution of observable characteristics. Performing a matching between women circa 1992 and women circa 2007 would preserve the distribution of men’s characteristics (which, by construction, are the same as those of women for each corresponding year). Three sets of individuals are generated in matching the two sets of data with the methodology described in chapter 2. In this matching after matching exercise, the distributions of observable characteristics in the set of matched individuals will be the same between men and women and the same between circa 1992 and 2007. The increase in the unexplained gen- der earnings gap that remains in the matched set of matched individuals corresponds to a counterfactual situation in which there is no change over time in the distribution of observable characteristics (or no change in the composition of the labor market). The results of this exercise are reported in table 4.9. In all cases, the first stage of matching is performed with all of the observable character- istics shown in figure 4.7. The matching after matching exercise is then performed with each observable characteristic, one at a time. The results show that in the hypothetical situation of no changes over time in the dis- tribution of characteristics, the decline in unexplained gender earnings gaps would have been even greater than what was observed. This narrowing is more pronounced when using age and education independently and even more pronounced when using the whole set of observable characteristics. Figure 4.8 compares unexplained gaps along the earnings percentiles for the two periods. The comparison is made for four sets of matching 74 new century, old disparities Table 4.9 Decomposition of Changes in Unexplained Gender Earnings Gap in Latin America and the Caribbean between Circa 1992 and 2007 (percent) Counterfactual Part of the change change if no change attributed to in observable changes in observable Characteristics characteristics characteristics Age –7.1 3.1 Education –7.3 3.3 Presence of children in the household –4.6 0.5 Presence of other household member with labor income –4.2 0.1 Urban –5.4 1.3 Type of employment –4.2 0.1 Time worked –4.6 0.5 Full set –12.1 7.9 Source: Based on data from national household surveys from circa 1992 and circa 2007. variables (only the results for the full set of variables are reported; for a full set of graphs, see Ñopo and Hoyos 2010). The results indicate that most of the reduction in the average unexplained gender earnings gap in the region occurred at the extremes of the earnings distribution. The unexplained gender earnings gaps at the middle of the distribution (percentiles 35–60) remained almost unchanged. The gaps at the bottom of the distribution narrowed by about 10 percentage points (at the 5th percentile of the distributions of earn- ings, for instance, unexplained gender gaps declined from 38–48 percent to 28–38 percent) The gaps at the top of the distribution narrowed by 3–9 per- centage points (at the 90th percentile of the distribution, for instance, the unexplained gender gaps declined from 10–42 percent to 7–33 percent). The U-shape of the curve of unexplained gender earnings gap with respect to the percentiles of the earnings distributions that was evident in circa 1992 smoothed in circa 2007. Nonetheless, there is still a pattern of larger unexplained earnings gaps at the bottom of the distributions of earnings. The correlation between gender earnings gaps and poverty or low income generation remains prevalent in the region. Having explored changes over time in the patterns of unexplained gender earnings gaps across the earnings distributions, the analysis turns next to more schooling, lower earnings for women 75 Figure 4.8 Unexplained Gender Earnings Gap in Latin America and the Caribbean after Controlling for Observable and Job Characteristics, by Percentiles of Earnings Distribution, Circa 1992 and 2007 50 percentage of average women’s earnings 40 30 20 10 0 0 10 20 30 40 50 60 70 80 90 100 earnings percentile circa 1997 circa 2007 Source: Based on data from national household surveys from circa 1992 and circa 2007. an exploration of unexplained gender earnings gaps for different segments of the labor market (for graphs reporting the results, see Ñopo and Hoyos (2010). Segments of the labor market for which the unexplained gender earnings gaps are larger (or smaller) are similar in both periods. The unex- plained gender earnings gaps decreased for all age groups, especially people 25–44 years old. Regarding education, the earnings gaps increased for workers in the middle of the distribution and decreased for workers at the extremes, especially for workers with no education. The confidence interval for unexplained earnings gaps fell from 40–49 percent to 13–21 percent. The unexplained gaps also narrowed among people who live with children under six, live in rural areas, are self-employed, and work part time. A Cohort Approach to Understanding Unexplained Changes in Gender Pay Differences Until now, results for the evolution of the gap were presented in a format as close as possible to the results for the analysis of circa 2007 alone. How- ever, with data for two points in time, it is possible to analyze in greater depth some of the assertions made earlier. 76 new century, old disparities In particular, the unexplained gender earnings gap was shown to increase with age. It was argued that this result implied either a narrowing of the gender earnings gap over time or a correlation between this gap and unob- servable characteristics, such as labor experience or the bearing and raising of children. To determine which explanation is more accurate, the rest of this chapter is dedicated to detecting changes in earnings gaps over time through a cohort analysis. The analysis examines gender earnings gaps among individuals who were age 15–29, 30–44, and 45–59 in 1992.8 Figure 4.9 shows the results for the analysis after controlling for the full set of observable characteristics. The results show that the unexplained gender earnings gaps for the two older cohorts decreased as individuals aged. For the youngest cohort, the gap increased. For this cohort, the secular trend of reduction of gender earnings disparities was outweighed by the increase in gender earnings gaps workers faced as they entered adulthood. The increases in unexplained gender earnings gaps shown for the three cohorts are disaggregated for different segments of the market in table 4.10. The analysis corresponds to a pseudo-panel analysis, in the sense that the same individuals are not followed in both periods; instead, the same segments of the labor market are compared in periods 1 and 2. The results suggest differences across the life cycle. In the youngest cohort, the largest increases in the unexplained earnings gaps occurred among workers who completed primary and secondary education; for the other two cohorts, the largest increases occurred among the least edu- cated workers. For the oldest cohort, the unexplained earnings gaps fell among the least educated individuals, which may suggest that the earnings penalty faced by women with little education declines with maturity (and perhaps experience). Among workers with children in the household, the largest increases in unexplained gender gaps occurred in the two youngest cohorts and the oldest cohort. For workers with children at home, the unexplained gender earnings gaps narrowed over time for all three cohorts. Regarding the presence of other income generators at home, the data show no differences for the two oldest cohorts. The narrowing in gender earnings gaps was similar for workers with and without other labor income earners at home. For the youngest cohort, however, the largest increase in the gap occurred among workers who lived with another income genera- tor. For workers with no other labor income earner at home, the unex- plained gender earnings gap narrowed for all cohorts: women who had no other option than generating income to maintain their households were successful at reducing their gender earnings disparities. The reductions in unexplained earnings gaps also occurred among all cohorts in rural areas. It changed substantially among employers as well, increasing for the two youngest cohorts and falling for the oldest one. more schooling, lower earnings for women 77 Figure 4.9 Confidence Intervals for Unexplained Gender Earnings Gap in Latin America and the Caribbean by Cohort, Circa 1992 and 2007 45 percentage of average women’s earnings 40 35 30 25 20 15 to 29 in 1992 30 to 44 in 1992 45 to 59 in 1992 cohort circa 1997 circa 2007 circa 1997 circa 2007 Source: Based on data from national household surveys from circa 1992 and 2007. Note: Figures show results after controlling for demographic and job-related characteristics. Boxes show 90 percent confidence intervals for unexplained earnings; whiskers show 99 percent confidence intervals. To what extent do the reported changes correspond to changes in the earnings gap within segments of the labor market, and to what extent do they correspond to changes in the composition of those segments? The same “matching after matching� exercised shown in table 4.9 was conducted within the cohorts in this pseudo-panel to answer this question (for more detailed results, see Ñopo and Hoyos 2010). The evidence points to the same results, which attribute a small role to the composition of the labor market. Most of the changes during this period can be attributed to changes within the segments of the labor market. Table 4.10 identifies the segments of the labor market within which most of the reductions in gender earnings gaps occurred. The cross-country heterogeneity in unexplained gender gaps shown ear- lier can be seen in terms of the evolution of these differences. Figure 4.10 shows confidence intervals for the original earnings gap and the unex- plained component of the gender earnings gaps by country, after control- ling for the full set of observable characteristics. The original earnings 78 new century, old disparities Table 4.10 Unexplained Gender Earnings Gap in Latin America and the Caribbean by Cohort and Demographic and Job Characteristics, Circa 2007 (percent) Age 15–29 30–44 45–59 Characteristics in 1992 in 1992 in 1992 Overall 7.6 –3.8 –12.4 Education None 8.9 –51.6 –7.2 Primary incomplete –4.1 –17.4 –43.1 Primary complete 19.5 0.4 19.8 Secondary incomplete 12.8 12.6 –15.9 Secondary complete 22.5 5.8 –1.1 Tertiary incomplete 11.6 -0.6 18.1 Tertiary complete –1.4 –2.4 –4.3 Presence of children (6 years or younger in household) No 9.01 –4.4 –12.6 Yes –10.3 –14.1 –3.4 Presence of other household member with labor income No –3.2 –3.9 –10.3 Yes 9.1 –3.8 –13.2 Urban No –1.8 –16.7 –23.5 Yes 8.1 –3.5 –12.1 Type of employment Employer 21.5 5.8 –46.0 Employee 7.5 –1.8 –12.7 Self-employed 5.4 –11.1 –6.0 Time worked Part time 5.9 –7.9 –8.9 Full time 8.2 –2.9 –15.5 Over time 4.1 –2.0 –15.1 Source: Based on data from national household surveys from circa 2007. more schooling, lower earnings for women 79 Figure 4.10 Confidence Intervals for Original and Unexplained Gender Earnings Gaps in Latin America and the Caribbean by Country, Circa 1992 and 2007 50 a. Original gender earnings gap percentage of women’s earnings 40 30 20 10 0 –10 il c r r s y y ina livia raz hile bia Rica ubli ado ado o u ala ura xic gua ma ua Per ua , RB nt C lom a m a a g g rge Bo B o s e t Rep Ecu alv ate ond M icar an ara ru ela A C C o S u H P P U zu ca n El G N ne ini Ve o m D b. Unexplained gender earnings gap after controlling for full 50 set of demographic and job characteristics percentage of women’s earnings 40 30 20 10 0 –10 ina via zil ile bia ica lic or or ala as ico ua a ay ru ay B nt oli Bra Ch om a R pub uad lvad em dur ex rag nam agu Pe ugu la, R rge B o l s t Re Ec Sa a t o n M ica P a a r U r ue A C C o u H P z n El G N ne ica Ve in o m D circa 1997 circa 1997 circa 2007 circa 2007 Source: Based on data from national household surveys from circa 1992 and 2007. Note: Boxes show 90 percent confidence intervals for unexplained earnings; whiskers show 99 percent confidence intervals. gap peaks in Chile in period 1 (circa 1992) and Bolivia in period 2 (circa 2007). However, these measures of earnings gaps incorporate differences in observable characteristics. Regarding unexplained gender earnings gaps, the most salient result, consistent with the results reported for circa 2007 alone, is that Brazil shows the largest gap across both periods, although the 80 new century, old disparities gap decreased. Brazil, El Salvador, and Guatemala show the largest drops in unexplained gender earnings gaps. In contrast to the regional trend of declining unexplained earnings gaps, these gaps increased in Nicaragua and República Bolivariana de Venezuela between circa 1992 and 2007. To a lesser extent (and one that is not statistically significant), the gaps also widened in Argentina and Mexico. It is precisely this cross-country heterogeneity that motivates the studies included in this book. Chapters 5 (on Peru) and 6 (on Mexico) examine countries that have not achieved gender educational parity. The next six chapters examine countries and regions that have achieved parity: Chile (chapter 7), Colombia (chapter 8), Brazil (chapter 9), Ecuador (chapter 10), Central America (chapter 11) and the Caribbean (chapter 12). Notes 1. Among working women, almost 10 percent worked in the agriculture sec- tor, 14 percent in industry, and 76 percent in the service sectors circa 2006. The percentage of women in services is significantly higher than in other regions of the world. Women’s unemployment rate in Latin America and the Caribbean was about 10 percent (ILO 2007). 2. This ranking is based on an index that includes earnings disparities and other variables. The index also includes differences in labor participation and access to certain type of occupations (legislators, senior officials, and managers, and professional and technical workers). For more details, see Hausmann, Tyson, and Zahidi (2007). 3. For a description of the methodology used in this chapter, see chapter 2. 4. Part-time workers are people who work 35 hours or less a week at their main occupation. 5. In the Dominican Republic, workers are considered formal if they report having a contract. Firm size is not used as a control variable in Brazil, because it was not possible to construct the “small firm� variable there. 6. Time worked is divided in three categories: part time (less than 35 hours a week), full time (35–48 hours a week), and overtime (more than 48 hours a week). 7. The cross-country heterogeneity reported for circa 2007 alone is also evi- dent in these data. For an analysis of the unexplained component of the gender earnings gaps by country, see Ñopo and Hoyos (2010). 8. The Dominican Republic and Guatemala were dropped from this part of the analysis, because data for the 15-year span were not available. References Angel-Urdinola, D., and Q. Wodon. 2006. “The Gender Wage Gap and Poverty in Colombia.� Labour 20 (4): 721–39. Atal, J. P., H. Ñopo, and N. Winder, 2009. “New Century, Old Disparities: Gen- der and Ethnic Earnings Gaps in Latin America.� RES Working Paper 4640, Research Department, Inter-American Development Bank, Washington, DC. more schooling, lower earnings for women 81 Babcock, L. and S. Laschever. 2003. Women Don’t Ask: Negotiation and the Gen- der Divide. Princeton, NJ: Princeton University Press. Barrientos, A. 2002. “Women, Informal Employment and Social Protection in Latin America.� IDPM Discussion Paper 30557, Institute for Development Policy and Management, University of Manchester, United Kingdom. Contreras, D., and G. Plaza. 2004. “Participación femenina en el mercado labo- ral chileno¿cuánto importan los factores culturales?� Universidad de Chile, Santiago. http://.estudiosdeltrabajo.cl2Fwp-content2Fuploads2F20092F122F- factores-culturales-en-participacion-laboral-femenina-d-contreras-g-plaza .pdf&ei=SSA8T5GMOcHd0QH3iKiqCw&usg=AFQjCNFQqH1vr79I69Uyrd KzBPNwhn5PXg&sig2=Tl5wueYQWxtana-oGtfNRA. Cox, A., and J. Roberts. 1993. “Macroeconomic Influences on Female Labor Force Participation: The Latin American Evidence.� Estudios de Economía 20: 87–106. Cruces, G., and S. Galiani. 2007. “Fertility and Female Labor Supply in Latin America: New Causal Evidence.� Labour Economics 14 (3): 565–73. Deutsch, R., A. Morrison, H. Ñopo, and C. Piras. 2004. “Working within Confines: Occupational Segregation by Gender in Costa Rica, Ecuador, and Uruguay.� In Women at Work: Challenges for Latin America, ed. C. Piras, 187–226. Washington, DC: Inter-American Development Bank. Duryea, S., A. C. Edwards, and M. Ureta. 2004. “Women in the LAC Labor Market: The Remarkable 1990s.� In Women at Work: Challenges for Latin America, ed. C. Piras, 27–60. Washington, DC: Inter-American Devel- opment Bank. Frankema, E. 2008. “Wage Gaps in Twentieth Century Latin America: Persistent Inequality or Distributional Change.� Utrecht University, the Netherlands. Freije, S. 2009. “Informal Employment in Latin America and the Caribbean: Causes, Consequences, and Policy Recommendations.� Inter-American Devel- opment Bank, Washington, DC. Hausmann, R., L. D. Tyson, and S. Zahidi, eds. 2007. The Global Gender Gap 2007. Davos, Switzerland: World Economic Forum. Hertz, T., P. Winters, A. Paula de la O, E. J. Quiñones, B. Davis, and A. Zezza. 2008. “Wage Inequality in International Perspective: Effects of Location, Sector, and Gender.� ESA Working Paper 8/08, Food and Agriculture Organization of the United Nations, Agricultural and Development Economics Division (ESA), Rome. ftp://ftp.fao.org/docrep/fao/011/ak230e/ak230e00.pdf. ILO (International Labour Organization). 2007. Modelo de tendencias mundiales del empleo. Geneva: International Labour Organization. Lim, L. L. 2002. “Female Labour Force Participation.� International Labour Orga- nization, Gender Promotion Programme (GENPROM), Geneva. http://www .un.org/esa/population/publications/completingfertility/RevisedLIMpaper.PDF. Madrigal Correa, A. L. 2004. “La evolución de la participación laboral de la mujer en Mexico: el efecto del tamaño de la familia 1970–2000.� Master’s thesis, Centro de Investigaciónes y Docencia Económicas, Mexico City. Márquez, G., and M. F. Prada. 2007. “Bad Jobs, Low Productivity, and Exclusion.� Inter-American Development Bank, Research Department, Washington, DC. Ñopo, H. 2008. “Matching as a Tool to Decompose Wage Gaps.� Review of Eco- nomics and Statistics 90 (2): 290–99. 82 new century, old disparities Ñopo, H., and A. Hoyos. 2010. “Evolution of Gender Wage Gaps in Latin America at the Turn of the Twentieth Century: An Addendum to ‘New Century, Old Disparities.’� IZA Discussion Paper 5086, Institute for the Study of Labor, Bonn, Germany. Panizza, U., and C. Z.-W. Qiang. 2005. “Public-Private Wage Differential and Gender Gap in Latin America: Spoiled Bureaucrats and Exploited Women?� Journal of Socio–Economics 34 (6): 810–33. Psacharopoulos, G., and Z. Tzannatos, eds. 1992a. Case Studies on Women’s Employment and Pay in Latin America. Washington, DC: World Bank. ———. 1992b. “Latin American Women’s Earnings and Participation in the Labor Force.� World Bank Policy Research Working Paper 856, Washington, DC. Salas, C., and M. Leite. 2007. “Segregación sectorial por género: una comparación Brasil-México.� Cadernos PROLAM/USP. Ano 7 - vol. Tenjo, J., R. Ribero, and L. Bernat. 2006 “Evolución de las diferencias salariales de género en seis países de América Latina.� In Mujeres y trabajo en América Latina, ed. C. Piras, 149–98. Washington, DC: Inter-American Development Bank. Terrell, K. 1992. “Female-Male Earnings Differentials and Occupational Struc- ture.� International Labor Review 131 (4–5): 387–98. Urdinola, A. D. F., and Q. Wodon. 2006. “The Gender Wage Gap and Poverty in Colombia.� Labour 20 (4): 721–39. 5 The Mostly Unexplained Gender Earnings Gap: Peru 1997–2009 Gender disparities in the Peruvian labor market are pronounced. There are substantial gaps in participation and employment rates, occupations, and hourly and monthly earnings. Peru has high occupational segregation (Blau and Ferber 1992), and a sizable share of jobs tend to fail at least one of the formality conditions (formal contract or access to insurance). Formality affects men and women differently: 55 percent of men and 65 percent of women have jobs in the informal sector. Gender gaps are also associated with differences in observable characteristics of the work- ing population, such as age, schooling, marital status, and household responsibilities. Peru experienced labor market reforms during the early 1990s.1 These reforms included dramatic reductions in firing costs, linked to reduc- tions in formality, and a subsequent increase in turnover rates, as a result of shorter durations of both employment and unemployment (Saavedra 2000; Saavedra and Torero 2000). These reforms may have influenced women’s participation in labor markets, but the theoretical literature has no clear predictions as to how these kind of changes in employment dynamics affect earnings differentials. In addition to gender differences in labor market outcomes, there are also gender disparities in individual characteristics. Men in Peru tend to have more years of education than women and longer tenure in higher- paying occupations. This chapter was adapted from “The Gender Wage Gap in Peru 1986–2000: Evidence from a Matching Comparisons Approach,� Hugo Ñopo, Economica, La Plata, vol. L, 1–2, 2004. 83 84 new century, old disparities The extent to which these differences in observable characteristics account for gaps in labor market outcomes is a longstanding question. This chapter analyzes both the evolution of the gender earnings gap between 1997 and 2009 and the role of individual characteristics in explaining earnings gaps during this period. The results suggest a steady reduction in gender differences in participation and employment rates, accompanied by cyclical evolution of the gender gap in hourly earnings. The analysis in this chapter is based on 1997–2009 data from the Encuesta Nacional de Hogares (ENAHO), Peru’s national household survey, conducted by the National Institute of Statistics and Informatics (INEI). As the main objective of this chapter is to estimate and explain gen- der earnings gaps, only the working population ages 16–75 is examined. How Do Male and Female Workers Differ? It can be argued that the gender earnings gap simply reflects gender differ- ences in some observable characteristics of the individuals that are deter- minants of earnings. To some extent, this is a valid argument, as there are gender differences in age, education, occupational experience, and occu- pations, among other characteristics rewarded in labor markets. However, these differences only partially explain the earnings gap. The purpose of this chapter is to measure the extent to which differences in characteristics explain differences in pay in Peru.2 On average during 1997–2009, working men in Peru were 0.65 years older than working women. This result contrasts with figures for the Peru- vian population as a whole, in which the average age is slightly higher for women than for men. The difference in the average age among workers may reflect women’s earlier entrance into or earlier retirement from the labor market. Either circumstance is expected to have a negative impact on earnings. Early entry into the labor market may imply fewer years of schooling; early retirement implies shorter tenure. There are also significant differences between men and women in edu- cational attainment (table 5.1). Although the proportion of working men and women that completed high school or have some years of university is fairly similar, there are important differences in all other educational levels. Women with university degrees represent a larger proportion of the labor force than men with the same educational level, even though, on average, working women attain fewer years of education than working men in most years of the sample. Women at the other extreme of the educational ladder also participate more in the Peruvian labor market than men at that educational level. As a result, working women are concentrated at the extremes of the educational distribution. high unexplained gaps: peru, 1997–2009 85 Table 5.1 Demographic and Job Characteristics and Relative Earnings of Men and Women in Peru’s Labor Force, 1997–2009 Relative earnings (average Composition women’s earnings for (percent) each year = 100) Women Men Women Men Personal characteristics Age 16–24 20.8 20.3 76.2 85.9 25–34 27.5 26.5 109.2 113.6 35–44 23.5 22.3 117.2 139.0 45–54 15.6 16.4 107.1 150.1 55–65 12.5 14.5 78.3 127.6 Education None 9.1 3.7 35.0 40.9 Primary incomplete 9.5 9.5 51.9 57.7 Primary complete 13.5 16.3 63.6 72.3 Secondary incomplete 12.8 16.7 74.5 91.3 Secondary complete 26.3 29.0 88.9 110.1 Tertiary incomplete 8.2 8.2 128.5 158.3 Tertiary complete 20.6 16.6 193.2 257.6 Urban No 22.8 32.2 46.7 62.6 Yes 77.2 67.8 115.7 149.7 Job characteristics Part-time work No 66.0 77.6 82.4 103.8 Yes 34.0 22.4 134.2 183.5 Small firm No 31.3 37.6 148.3 160.0 Yes 68.7 62.4 78.0 98.6 Occupation Professionals and technicians 16.9 14.6 211.9 265.9 (continued next page) 86 new century, old disparities Table 5.1 (continued) Relative earnings (average Composition women’s earnings for (percent) each year = 100) Women Men Women Men Directors and upper management 0.5 0.8 315.3 552.8 Administrative personnel 6.7 3.9 145.3 177.8 Merchants and sellers 27.9 10.3 80.5 111.9 Service workers 22.0 9.9 72.4 90.3 Agricultural workers and similar 13.9 30.3 42.8 60.1 Nonagricultural blue- collars 12.2 30.2 70.8 108.6 Armed forces 0.0 0.2 107.6 90.3 Economic Sector Agriculture, hunting, forestry, and fishing 14.1 31.0 44.4 62.3 Mining and quarrying 1.2 1.8 102.8 175.3 Manufacturing 11.0 11.7 87.0 139.7 Electricity, gas, and water supply 0.1 1.2 201.1 128.9 Construction 1.0 7.3 154.9 137.2 Wholesale and retail trade and hotels and restaurants 36.6 16.7 86.1 125.0 Transport, storage 2.7 10.2 141.1 124.7 Financing, insurance, real estate, and business services 4.2 5.4 215.8 236.8 Community, social, and personal services 29.1 14.7 126.4 169.6 Source: Based on 1997–2009 data from ENAHO. high unexplained gaps: peru, 1997–2009 87 Figure 5.1 shows average years of education of the labor force. The edu- cation gap favors working men for most of the period. However, the gap has been almost zero since 2006, and the figure for 2009 shows a gap in favor of women. These figures lie in contrast with figures for the Peruvian population as a whole, where a gender education gap remains. The finding may reveal that women in the labor force have more human capital than the average Peruvian woman, reflecting selection into the labor market. Figure 5.2 reveals the evolution of the gender composition of the labor force by educational level. It shows that the gap between men and women at each educational level decreased throughout the period. Women’s participation was greatest at the extremes of the educational ladder. There are gender differences in human capital accumulation, probably the observable characteristic most rewarded in the labor market. However, this difference narrowed over the period. This relationship partially explains the gender earnings gap and its evolution. Figure 5.3 shows average hourly earnings gaps as multiples of women’s average hourly earnings. It shows that the gender earnings gap fluctuated around an average value of 21 percent (that is, men earned an average of 21 percent more per hour than women). However, there are significant fluctuations around this average measure, and there are two years in the sample (2007 and 2009) when men reportedly earned less than women (the earnings gap was negative). Figure 5.1 Average Years of Education of Men and Women in Peru’s Labor Force, 1997–2009 12 average years of education 11 10 9 8 7 6 97 98 99 00 01 02 03 04 05 06 07 08 09 19 19 19 20 20 20 20 20 20 20 20 20 20 women men Source: Based on 1997–2009 data from ENAHO. 88 new century, old disparities Figure 5.2 Educational Levels of Men and Women in Peru’s Labor Force, 1997–2009 a. No or incomplete primary education 40 35 30 25 percent 20 15 10 5 0 97 98 99 00 01 02 03 04 05 06 07 08 09 19 19 19 20 20 20 20 20 20 20 20 20 20 b. Complete primary or incomplete secondary education 40 35 percent 30 25 20 15 10 5 0 97 98 99 00 01 02 03 04 05 06 07 08 09 19 19 19 20 20 20 20 20 20 20 20 20 20 40 c. Tertiary education 35 30 25 percent 20 15 10 5 0 97 98 99 00 01 02 03 04 05 06 07 08 09 19 19 19 20 20 20 20 20 20 20 20 20 20 women men Source: Based on 1997–2009 data from ENAHO. The measures of the gap (multiples of average hourly earnings for women) are crude data, as they consider all men and women regardless of differences in observable characteristics or whether it is possible to compare them. If variation in these gender differences in average hourly earnings according to individual characteristics is explored, the results displayed in table 5.1 are obtained. The gender earnings gap tends to increase at about age 30, reaching a peak at age 45–54. It increases monotonically with educational attainment. high unexplained gaps: peru, 1997–2009 89 Figure 5.3 Gender Gap in Hourly Earnings in Peru, 1997–2009 50 percentage of average women’s 40 30 earnings 20 10 0 –10 –20 97 98 99 00 01 02 03 04 05 06 07 08 09 19 19 19 20 20 20 20 20 20 20 20 20 20 Source: Based on 1997–2009 data from ENAHO. The largest gap occurs among people with university degrees. The gap is larger in urban than in rural areas and for people who work part time. It is largest in the best rewarded occupations, directors and upper-level managers. The Role of Individual Characteristics in Explaining the Gender Earnings Gap Figure 5.4 presents the earnings gap in relative terms (as a multiple of women’s earnings) and decomposes it into the four components intro- duced in chapter 2. The height of each bar is proportional to the earnings gap in each year. The height of each component is proportional to the value of the component; a component with a negative value is illustrated below the zero line. The first set of decompositions was calculated using a combination of explanatory variables, such as age, education, marital status, and residence in an urban area. The results show that most of the earnings gap remains unexplained after including these controls, as the unexplained gender earnings gap (Δ0)—the portion of the gap attributed to differences between men and women that cannot be explained by observable characteristics—is large in all years. ΔX is the portion of the gap due to differences in characteristics between men and women in the “common support.� It is negative except in 1998, when it is positive and particularly large. When negative, this 90 new century, old disparities Figure 5.4 Decomposition of Gender Earnings Gap in Peru after Controlling for Demographic Characteristics, 1997–2009 60 percentage of average women’s earnings 50 40 30 20 10 0 –10 –20 –30 97 98 99 00 01 02 03 04 05 06 07 08 09 19 19 19 20 20 20 20 20 20 20 20 20 20 ΔX ΔF ΔM ΔO Source: Based on 1997–2009 data from ENAHO. Note: ΔM (ΔF) is the part of the earnings gap attributed to the existence of men (women) with combinations of characteristics that are not met by any women (men). ΔX is the part of the earnings gap attributed to differences in the observable characteristics of men and women over the “common support.� Δ0 is the part of the earnings gap that cannot be attributed to differences in characteristics of the individuals. It is typically attributed to a combination of both unobservable characteristics and discrimination. The sum of these components equals the total earnings gap (ΔM + ΔF + ΔX + Δ0 = Δ). component indicates that matched women exhibit a distribution of char- acteristics that is better rewarded by the labor market than the distribu- tion of characteristics exhibited by men. This is the case for education, for example. Within the working population, a larger percentage of women than men hold university degrees. In 1998, when the ΔX component is positive, two events come into play. Both the gap in average years of edu- cation between men and women and the share of working women with no education are largest in 1998. The other components—the portions of the earnings gap attributable to the nonoverlapping supports of women (ΔF) and men (ΔM)—are fairly close to zero in all years analyzed. ΔF, however, is positive in most years, indicating that unmatched women earn less than matched ones. ΔM is negative, implying that unmatched men earn less than matched men. high unexplained gaps: peru, 1997–2009 91 In general, the average components of the earnings gap for the whole period point to an insignificant role of ΔM, ΔF, and ΔX in explaining the earnings gap. For the whole period, the average value of these components is zero. In contrast, Δ0 has an average value (21.3) that is almost equal to the entire gender earnings gap (21.5). The demographic characteristics considered as controls thus cannot account for gender differences in pay. The decompositions in figure 5.5 use different combinations of age, education, economic sector, occupation, and firm size (a dichotomous variable equal to one for firms with five workers or less) as controls. After controlling for these job characteristics, the average unexplained gender earnings gap is about 23.1 percent—slightly higher than the average Figure 5.5 Decomposition of Gender Earnings Gap in Peru after Controlling for Demographic and Job Characteristics, 1997–2009 60 percentage of average women’s earnings 50 40 30 20 10 0 −10 −20 −30 97 98 99 00 01 02 03 04 05 06 07 08 09 19 19 19 20 20 20 20 20 20 20 20 20 20 ΔX ΔF ΔM Δ0 Source: Based on 1997–2009 data from ENAHO. Note: ΔM (ΔF) is the part of the earnings gap attributed to the existence of men (women) with combinations of characteristics that are not met by any women (men). ΔX is the part of the earnings gap attributed to differences in the observable characteristics of men and women over the “common support.� Δ0 is the part of the earnings gap that cannot be attributed to differences in characteristics of the individuals. It is typically attributed to a combination of both unobservable characteristics and discrimination. The sum of these components equals the total earnings gap (ΔM + ΔF + ΔX + Δ0 = Δ). 92 new century, old disparities total gap when these variables are not considered. The results show that most of the earnings gap remains unexplained even after including the complete set of controls. The role of the other three components play in the decomposition is also similar to the roles explained earlier. In most years, ΔX is negative and ΔF close to zero. However, there is a change in ΔM, which becomes larger and is positive in all years, meaning that when controlling for job characteristics, men matched with women tend to earn lower earnings than unmatched men. The components Δ0 and ΔM explain more than 80 percent of the earn- ings gap during all years when using the full set of observable characteris- tics. These components may be regarded as noisy discrimination measures or unexplained differences. The first of them is determined in the labor market, while the second is outside of it (in the acquisition of particular characteristics). Whereas discrimination measures are linked to differences in pay, unexplained differences are presumably linked to differences in access to particular combinations of characteristics that are rewarded in the labor market. Table 5.2 shows descriptive statistics for women in and out of the common support. Just 1 percent of working women exhibit combinations of age, education, location (urban or rural), and marital status that can- not be matched by any men in the sample; 0.2 percent of working men report combinations of these characteristics that cannot be matched by any women in the sample. The percentage of unmatched individuals grows when more characteristics are included: 3 percent of working women and 11 percent of working men exhibit combinations of age, education, economic sector, occupation, and firm size that cannot be matched by any individual of the opposite sex in the sample. Unmatched men and women are older than matched men and women when controlling by both sets of characteristics explored. Unmatched women are concentrated in the lowest educational levels, whereas unmatched men are frequently found among workers with some high school or university education. Most unmatched women are service workers, whereas most unmatched men are agricultural or blue collar workers. Most of the matched working population is concentrated in wholesale and retail trade; the hotel and restaurant sector; and com- munity, social, and personal services. This pattern may reflect women’s concentration in services. Exploring the Unexplained Component of the Gender Earnings Gap This section analyzes the distribution of unexplained gender differ- ences in earnings obtained from the matching process by comparing the distribution of earnings for women with the counterfactual distribution Table 5.2 Demographic and Job Characteristics of Matched and Unmatched Samples of Men and Women in Peru’s Labor Force, 1997–2009 (percent) Age, education, marital status, and Age, education, firm size, occupation, and urban area economic sector Matched Matched women and Unmatched Unmatched women and Unmatched Unmatched Characteristics men men women men men women Average hourly earnings 2.6 2.4 4.8 5.5 (constant 1994 Peruvian soles) Average age 37.1 42.8 59.6 36.90 43.5 43.6 Average years of schooling 9.6 5.4 9.8 9.58 8.8 10.3 Education level No education or primary 18.2 59.0 21.5 18.20 29.6 12.9 incomplete Primary incomplete or 26.4 12.8 34.2 26.50 19.2 33.9 secondary complete Secondary complete or 34.7 16.9 28.4 34.70 29.7 34.4 tertiary incomplete Tertiary complete 20.7 11.2 15.9 20.60 21.5 18.8 (continued next page) 93 Table 5.2 (continued) Age, education, marital status, and Age, education, firm size, occupation, and 94 urban area economic sector Matched Matched women and Unmatched Unmatched women and Unmatched Unmatched Characteristics men men women men men women Marital status Single 30.4 9.1 32.0 Married 51.0 2.8 29.5 Divorced 12.3 39.7 26.8 Widower 6.3 48.5 11.7 Living in urban area 77.4 60.5 24.7 Working in small firm 68.9 60.6 42.2 Occupation Professionals and technicians 16.9 14.9 16.9 Directors and upper management 0.4 4.1 4.2 Administrative personnel 6.2 22.6 5.8 Merchants and sellers 28.4 8.9 4.2 Service workers 21.5 35.7 11.7 Agricultural workers and similar 14.3 0.5 2.4 Nonagricultural blue-collars 12.2 13.2 53.6 Table 5.2 (continued) Age, education, marital status, and Age, education, firm size, occupation, and urban area economic sector Matched Matched women and Unmatched Unmatched women and Unmatched Unmatched Characteristics men men women men men women Armed forces 0.0 0.1 1.2 Economic sector Agriculture, hunting, forestry, and fishing 14.4 3.1 7.2 Mining and quarrying 0.9 10.2 5.7 Manufacturing 10.8 15.4 8.8 Electricity, gas, and water supply 0.1 1.5 9.4 Construction 0.9 4.7 16.1 Wholesale and retail trade and hotels and restaurants 37.0 25.0 15.7 Transport, storage 2.5 9.8 17.7 Financing, insurance, real estate, and business services 4.0 10.6 9.3 Community, social, and personal services 29.3 19.8 10.2 95 Source: Based on 1997–2009 data from ENAHO. Note: Blank cells indicate that the variable is not being controlled for. 96 new century, old disparities of earnings for men when they are resampled to mimic the distribution of women’s characteristics. Figure 5.6 shows the relative earnings gap as a percentage of women’s earnings for each percentile of the earnings distribution. The gap exhibits a slight U-shape when controlling for both demographic and job characteristics. When controlling only for demo- graphic characteristics, an increase in the gap is observed after including the control dummy for urban area, indicating that the unexplained earn- ings gap is greater in urban areas across the distribution of earnings. The gap reaches a maximum for people in the 20th percentile of the earnings distribution, after which it monotonically decreases until the 80th per- centile before reaching another peak at the 95th percentile. Men in the 20th percentile earn on average 60 percent more than women; men in the top percentiles earn on average 25 percent more than women. When introducing job market controls (see figure 5.7), the gap again shows a slight U-shape, but this time the gender differences at the lowest percen- tiles of the earnings distribution are larger. Figure 5.6 Unexplained Gender Earnings Gap in Peru after Controlling for Demographic Characteristics, by Percentiles of Earnings Distribution, 1997–2009 60 percentage of average women’s earnings 40 20 0 −20 −40 0 10 20 30 40 50 60 70 80 90 100 earnings percentile year, age, and education set 1 (year, age, and education) + marital status set 1 (year, age, and education) + urban full demographic set (year, age, education, marital status, and urban location) Source: Based on 1997–2009 data from ENAHO. high unexplained gaps: peru, 1997–2009 97 Figure 5.7 Unexplained Gender Earnings Gap in Peru after Controlling for Demographic and Job Characteristics, by Percentiles of Earnings Distribution, 1997–2009 percentage of average women’s earning 80 60 40 20 0 0 10 20 30 40 50 60 70 80 90 100 earnings percentile year, age, and education sector set 1 (year, age, and education) + firm size set 1 (year, age, and education) set 1 + full labor set (firm size, occupation) + occupation Source: Based on 1997–2009 data from ENAHO. The distribution of unexplained gender earnings differences can also be analyzed by computing confidence intervals (figures 5.8 and 5.9). The extremes of the boxes correspond to a 90 percent confidence interval for the average unexplained differences in pay; the extremes of the whiskers correspond to a 99 percent confidence interval. The figures show no evi- dence of a monotonic decrease in earnings differences when controlling for either demographic or job characteristics. The unexplained hourly gender earnings gap reached its lowest levels in 1999, 2006, and 2007; it attained peaks in 2000 and 2005, evolving in a way that seems correlated with the cycle of the Peruvian economy. Changes in Women’s Participation and Unemployment Rates The measure of gender differences shown in the previous section was earnings. This section examines changes in women’s participation and unemployment rates. 98 new century, old disparities Figure 5.8 Confidence Intervals for Unexplained Gender Earnings Gap in Peru after Controlling for Demographic Characteristics, 1997–2009 70 percentage of average women’s earnings 60 50 40 30 20 10 0 −10 −20 −30 97 98 99 00 01 02 03 04 05 06 07 08 09 19 19 19 20 20 20 20 20 20 20 20 20 20 Source: Based on 1997–2009 data from ENAHO. Note: Boxes show 90 percent confidence intervals for unexplained earnings; whiskers show 99 percent confidence intervals. Figure 5.9 Confidence Intervals for Unexplained Gender Earnings Gap in Peru after Controlling for Demographic and Job Characteristics, 1997–2009 percentage of average women’s earnings 80 60 40 20 0 −20 −40 97 98 99 00 01 02 03 04 05 06 07 08 09 19 19 19 20 20 20 20 20 20 20 20 20 20 Source: Based on 1997–2009 data from ENAHO. Note: Boxes show 90 percent confidence intervals for unexplained earnings; whiskers show 99 percent confidence intervals. high unexplained gaps: peru, 1997–2009 99 The gender gap in participation decreased over the period, as a result of both a slight decrease in men’s participation and a larger increase in women’s participation (figure 5.10). In 1997, 64 percent of women were participating in the labor market; by 2009, this proportion reached 71 percent. The proportion for men was 85 percent in 1997 and 2009, with slight changes during the period. Gender differences in unemployment rates decreased between 1997 and 2009 (figure 5.11). The unemployment rate among men fell from 3.5 percent to 3.0 percent; the unemployment rate among women rose and fell over the period, declining from 3.5 percent to 2.5 percent over the period as a whole. There are also differences in the number of hours worked. On average, over the whole period, men worked 45 hours a week and women worked 40 hours, an 11 percent difference. These differences decreased between 1986 and 2000. Whereas men worked 15 percent more hours than women in 1997, they worked 11 percent more hours than women during 2005. The difference decreased to almost zero in 2006 and 2007 before increas- ing again in 2008 and 2009. Both participation and unemployment rates show the significant pres- ence of women in the Peruvian labor market. In fact, women’s participa- tion force was the second highest in the region in the early 1990s and the Figure 5.10 Labor Force Participation Rates of Men and Women in Peru, 1997–2009 90 85 80 participation rate (%) 75 70 65 60 55 50 45 40 97 98 99 00 01 02 03 04 05 06 07 08 09 19 19 19 20 20 20 20 20 20 20 20 20 20 women men Source: Based on 1997–2009 data from ENAHO. 100 new century, old disparities Figure 5.11 Unemployment Rates of Men and Women in Peru, 1997–2009 4.5 4 unemployment rate (%) 3.5 3 2.5 2 1.5 1 97 98 99 00 01 02 03 04 05 06 07 08 09 19 19 19 20 20 20 20 20 20 20 20 20 20 women men Source: Based on 1997–2009 data from ENAHO. highest in the mid-2000s (Elías and Ñopo 2010). In contrast, Mexico, analyzed in the next chapter, experienced the lowest women’s participa- tion rates in the region. Notes 1. The two waves of reform occurred in 1991 and 1995. 2. For a description of the methodology used in this chapter, see chapter 2. References Blau, F., and M. Ferber. 1992. The Economics of Women, Men, and Work, 2nd ed. Englewood Cliffs, NJ: Prentice-Hall. Elías, J., and H. Ñopo. 2010. “The Increase in Female Labor Force Participation in Latin America 1990–2004: Decomposing the Changes.� Inter-American Devel- opment Bank, Research Department, Washington, DC. Saavedra, J. 2000. “La flexibilización del mercado laboral.� In La reforma incom- pleta: rescatando los noventa, ed. R. Abusada, 379–428. Lima: Universidad del Pacífico. Saavedra, J., and M. Torero. 2000. “Labor Market Reforms and Their Impact on Formal Labor Demand and Job Market Turnover: The Case of Peru.� Research Network Working Paper R-394, Inter-American Development Bank, Washington, DC. http://pws.iadb.org/res/laresnetwork/files/pr111 finaldraft.pdf. 6 Is Gender Segregation in the Workplace Responsible for Earnings Gaps? Mexico 1994–2004 Low women’s labor force participation rates by women make Mexico an interesting country to analyze. Mexico had the lowest women’s participa- tion rate in Latin America by the early 1990s, at 37 percent. Since then, it has experienced important changes in the labor market from a gender perspective. The increase in the labor market participation rate of women has been the largest in a region where the women’s participation rate has increased substantially. Nonetheless, women’s labor participation in Mexico is still below the Latin American average and gender segregation in the workplace is still pervasive (Elías and Ñopo 2010). This chapter links the gender pay differential and labor market segre- gation. It explores the linkages between gender differences in observable human capital characteristics, (occupational and hierarchical) segrega- tion, and earnings. The data are drawn from the National Survey of Urban Employment (Encuesta Nacional de Empleo Urbano [ENEU]), Mexico’s national urban employment survey.1 These quarterly data cover the period from the third quarter of 1994 to the fourth quarter of 2004. This chapter was adapted from the following sources: “Gender Segregation in the Workplace and Wage Gaps: Evidence from Urban Mexico 1994–2004,� Sebastián Calónico and Hugo Ñopo, Research Department Working Paper 636, Inter-American Development Bank, Washington, DC; and Sebastián Calónico and Hugo Ñopo, “Gender Segregation in the Workplace and Wage Gaps: Evidence from Urban Mexico 1994–2004,� in Occupational and Residential Segregation (Research on Economic Inequality, Volume 17), ed. Yves Flückiger, Sean F. Reardon, and Jacques Silber, (Emerald Group Publishing Limited), 245–70. Sebastián Calónico is a graduate student in economics at the University of Michigan, Ann Arbor. 101 102 new century, old disparities What Does the Literature Show? Gender pay differentials in Mexico have been documented from various perspectives (Sánchez 1998; Pagan and Sánchez 2000; López Acevedo 2003; Chinhui and Airola 2005). Brown, Pagan, and Rodriguez-Oreggia (1999) study the effect of occupational attainment on the increase in gender earnings differentials between 1987 and 1993. Using data from the ENEU, they find that the decline in gender differences in occupational attainment somewhat attenuated the increase in the gender earnings dif- ferential. They also find important roles for labor supply decisions (hours of work per week) and changes in the regional structure of earnings. This finding contrasts with the results of Parker (1999), who exam- ines the gender earnings gap in rural areas of Mexico between 1986 and 1992 by looking at skill levels within groups of occupations. She finds that earnings differentials among labor income earners were low and remained roughly constant throughout the period, although they varied widely across occupations. She finds the largest earnings gaps in mana- gerial positions (in both the private and public sectors) and the smallest among public service workers and administrative positions. Rendón (2003) analyzes gender differences in employment, segrega- tion, and earnings. She documents that, in spite of the large increase in women’s labor force participation in recent decades, there is still a large concentration of women in certain activities. She documents an increase in segregation by productive sectors from 1990 to 2000. However, she sug- gests that there are reasons to believe that such segregation should decline in the future, because women tend first to enter activities more populated by other women before entering activities that are more gender neutral. She also provides estimates for the high degree of hierarchical (vertical) segre- gation (the holding of higher-ranking positions by men). When analyzing the evolution of the gender earnings gap, she argues that the observed reduction can be explained by an increase in women’s working hours. Rendón and Maldonado (2004) study the relationship between domestic work, occupational segregation, and the gender earnings gap in Mexico. Their motivation is the large increase observed in women’s labor force participation, which reflects both cultural factors and changes in the country’s occupational and productive structure (namely, the increase in the relative importance of professionals, office workers, and salespeople). However, this increase in participation did not imply that conditions faced by men and women equalized. Occupational segregation and earnings gaps are still notable, partly because of the number of hours worked, and they vary substantially across sectors and occupations. Colmenares (2006) analyzes occupational segregation by gender and its relation with earnings difference in the industry sector. She finds vari- ability across regions in gender occupational segregation. gaps and workplace segregation: mexico, 1994–2004 103 Measuring Occupational and Hierarchical Segregation In this chapter, occupational and hierarchical (vertical) segregation by gender are measured using the Duncan index (Duncan and Duncan 1955). The occupational index shows the percentage of men (women) that would need to switch from jobs that are dominated by men (women) to jobs that are dominated by women (men) in order to achieve a labor force with no segregation. The hierarchical index shows the percentage of women that would need to be promoted to better labor positions in order to eliminate segregation. The index ranges from zero to one, with a higher index representing greater segregation. The occupational index is computed using disaggre- gated information on seven occupations at the one-digit level (professionals and technicians, managers, administrative personnel, salespeople, work- ers in the service sector, workers in agricultural activities, and workers in industrial activities).2 The Duncan index of hierarchical segregation uses hierarchical cat- egories instead of occupations. The ENEU survey includes five hierarchi- cal categories (managers, independent workers, piece-rate or commission workers, fixed-salary workers, and members of a cooperative). Table 6.1 reports average measures of occupational and hierarchical segregation for various segments of the market for the period under analysis. Table 6.1 Average Duncan Index of Occupational and Hierarchical Segregation in Mexico, by Demographic and Job Characteristics, 1994–2004 Occupational Hierarchical Characteristics segregation segregation Years of schooling 0 0.40 0.11 1–6 0.39 0.10 7–12 0.33 0.10 13 or more 0.24 0.14 Age 15–24 0.31 0.09 25–49 0.34 0.10 50–64 0.33 0.09 (continued next page) 104 new century, old disparities Table 6.1 (continued) Occupational Hierarchical Characteristics segregation segregation Marital status Single (never married) 0.29 0.11 Married 0.34 0.09 Separated 0.30 0.12 Firm size (number of workers) 1–5 0.49 0.19 5–50 0.30 0.09 50+ 0.28 0.01 Management Private 0.34 0.10 Public 0.33 0.01 Economic sector Agriculture 0.31 0.25 Extraction and electricity 0.57 0.02 Manufacturing 0.05 0.07 Construction 0.81 0.29 Commerce 0.18 0.11 Communications and transports 0.70 0.50 Services 0.31 0.19 Public administration and defense 0.40 0.00 All 0.33 0.09 In 1994: III quarter 0.35 0.11 In 2004: IV quarter 0.33 0.08 Source: Based on data from 1994 to 2004 ENEU. Occupational segregation by gender in Mexico, as in most labor mar- kets, is less pronounced among people with more years of schooling. Inter- estingly, however, hierarchical segregation is more pronounced among people with more years of schooling. Although younger workers display lower levels of occupational segregation, hierarchical segregation appears to remain constant over the life cycle. Occupational segregation is lower among single (including both never-married and separated) individuals gaps and workplace segregation: mexico, 1994–2004 105 than among married people; the opposite is true for hierarchical segre- gation. Both types of segregation are significantly more pronounced in smaller firms. Although the Mexican public sector exhibits almost no hierarchical segregation, it displays levels of occupational segregation similar to those in the private sector. The ENEU records eight firm activities (agricul- ture, extraction and electricity, manufacturing, construction, commerce, communications and transport, services, and public administration and defense). The rankings of sectors according to occupational and hierarchi- cal segregation show some differences across these sectors. The greatest occupational segregation by gender is found in construction firms, fol- lowed by communications and transport; the lowest is found in manufac- turing. The greatest hierarchical segregation by gender is among people who work in communications and transport; the lowest is in public admin- istration and defense. Overall, occupational segregation is substantially greater than hierar- chical segregation. Both have been decreasing, albeit slightly. During the 10-year span analyzed, occupational segregation dropped 2 percentage points (from 0.35 to 0.33), and hierarchical segregation dropped 3 per- centage points (from 0.11 to 0.08). The Role of Individual Characteristics in Explaining the Earnings Gap To some extent, gender differences in individual characteristics that are important for the labor market can explain gender differences in occupa- tions and hierarchies. It could be the case that gender disparities in edu- cation, for example, which are still prevalent in Mexican labor markets, somehow determine occupational and hierarchical sorting by gender. The extent to which this argument is valid is evaluated here by analyz- ing three counterfactual situations, in which, first, there are no gender differences in age, schooling, or marital status; second, there are no gender differences in hierarchies; and third, there are no gender differences in age, schooling, marital status, or hierarchies. The evolution of occupational segregation in each of these situations is explored by comparing the origi- nal Duncan index with the index that would prevail in each hypothetical counterfactual situation (figure 6.1). The counterfactual situations are generated with the same matching approach used to decompose earnings gaps, illustrating the versatility of the matching approach.3 The results suggest that eliminating all gender differences in age, schooling, and marital status in the labor market would have reduced occupational segregation by 2–3 percentage points for the period 1994– 2004. Eliminating gender differences in hierarchies would have reduced occupational segregation by about 1 percentage point. Eliminating both 106 new century, old disparities Figure 6.1 Estimated Counterfactual Duncan Indexes of Occupational Segregation in Mexico, 1994–2004 a. Matching by age, education, and marital status 0.40 0.35 index 0.30 0.25 0.20 94-i 94-iv 95-iii 96-ii 97-i 97-iv 98-iii 99-ii 00-i 00-iv 01-iii 02-ii 03-i 03-iv 04-iii quarter original matching on age, education, and marital status b. Matching by hierarchies 0.40 0.35 index 0.30 0.25 0.20 94-i 94-iv 95-iii 96-ii 97-i 97-iv 98-iii 99-ii 00-i 00-iv 01-iii 02-ii 03-i 03-iv 04-iii quarter original matching on hierarchies c. Matching by age, education, marital status, and hierarchies 0.40 0.35 index 0.30 0.25 0.20 94-i 94-iv 95-iii 96-ii 97-i 97-iv 98-iii 99-ii 00-i 00-iv 01-iii 02-ii 03-i 03-iv 04-iii quarter original matching on age, education, marital status, and hierarchies Source: Based on data from 1994–2004 ENEU. sets of differences would have reduced the Duncan index by about 4 per- centage points. The differences between the counterfactual and the actual indexes are roughly constant over the period. The same exercise is conducted to analyze the evolution of hierarchical segregation. In this case, the first counterfactual situation is one in which gaps and workplace segregation: mexico, 1994–2004 107 there are no gender differences in age, schooling, or marital status; the sec- ond is one in which gender differences in occupations are eliminated; and the third is one in which there are no gender differences in age, schooling, marital status, or occupations (figure 6.2). The results suggest that the impact of the observable characteristics on the reduction in hierarchical segregation is greater than the impact of occupational segregation (especially when taking account of the fact that the original levels of hierarchical segregation are lower than the levels of occupational segregation). The role of occupations decreases in importance during the later portion of the period under analysis. The hypothetical situation in which working men and women have the same age, schooling, and marital status leads to a hierarchical segregation that would have been lower by 1 percentage point than the one actually observed between 1994 and 2004. Eliminating occupational segregation would have reduced hierarchical segregation by as much as 6–7 percentage points in the mid- 1990s and about 3 percent in 2004. The combined effect of eliminating occupational segregation and gender differences in individual characteris- tics (age, schooling, and marital status) would have reduced hierarchical segregation by 7–8 percentage points in the mid-1990s and by 4 percent- age points in 2004. These results indicate that individual characteristics play a (somewhat small) role in determining gender segregation in the Mexican labor mar- ket. Occupational and hierarchical segregation are linked, in the sense that a reduction in one leads to a reduction in the other. The counterfactual analysis seeks to answer the questions “by how much would the gender earnings gap change if (occupational or hierarchi- cal) segregation were reduced to zero?� and “by how much would the gen- der earnings gap change if gender differences in observable characteristics were reduced to zero?� To answer these questions, the analysis matched men and women based on first, age, schooling, and marital status; second, hierarchies; and third, occupations. The gender earnings gap shows a decreasing trend during most of the period under analysis, interrupted by only two years of increase (Figure 6.3). By the mid-1990s, on average, men earned about 18 percent more than women per hour worked. This gap declined to almost 12 percent by 2004. The role of age, schooling, and marital status in explaining gender dif- ferences in earnings changed as well. During the mid-1990s, these charac- teristics explained almost half of the earnings gap. After 2002, they seem to play almost no role in determining gender differences in pay. During the late 1990s, a hypothetical world in which there was no hierarchical segregation but everything else remained the same would have shown gender earnings gaps similar to those in a hypothetical world in which there were no gender differences in age, schooling, or marital status in the labor market. Later, the hypothetical gender earnings gap without hierarchical segregation becomes somewhat smaller than the hypothetical 108 new century, old disparities Figure 6.2 Estimated Counterfactual Duncan Indexes of Hierarchical Segregation in Mexico, 1994–2004 a. Matching by age, education, and marital status 0.20 0.15 index 0.10 0.05 0.00 I -II -I -IV I -II -I -IV I -II -I -IV I -II -II -II -II -II 96 99 02 95 98 01 04 94 97 00 03 96 99 02 quarter original matching on age, schooling, and marital status b. Matching by occupation 0.20 0.15 index 0.10 0.05 0.00 I -II -I -IV I -II -I -IV I -II -I -IV I -II -II -II -II -II 96 99 02 95 98 01 04 94 97 00 03 96 99 02 quarter original matching on occupations c. Matching by age, education, marital status, and occupations 0.20 0.15 index 0.10 0.05 0.00 I II -I -IV I -II -I IV I -II -I -IV I -II -II -II -II -II - 96 99 02 95 98 01 04 - 94 97 00 03 96 99 02 quarter original matching on age, schooling, marital status, and occupations Source: Based on data from 1994–2004 ENEU. gaps and workplace segregation: mexico, 1994–2004 109 Figure 6.3 Estimated Counterfactual Gender Earnings Gaps in Mexico, 1994–2004 percentage of average women’s earnings a. Matching by age, education, and marital status 30 20 10 0 I -II -I -IV I -II -I -IV I -II -I -IV I -II -II -II -II -II 96 99 02 95 98 01 04 94 97 00 03 96 99 02 quarter original matching on age, schooling, and marital status percentage of average women’s earnings 30 b. Matching by hierarchies 20 10 0 I -II -I -IV I -II -I -IV I -II -I -IV I -II -II -II -II -II 96 99 02 95 98 01 04 94 97 00 03 96 99 02 quarter original matching on hierarchies percentage of average women’s earnings 30 c. Matching by occupations 20 10 0 I -II -I -IV I -II I -IV I -II I -IV I -II -II -II - -II - -II 96 99 02 95 98 01 04 94 97 00 03 96 99 02 quarter original matching on occupations Source: Based on data from 1994–2004 ENEU. 110 new century, old disparities gender earnings gap without age, gender, and marital status differences. For 2000–04, the average gender earnings gap was about 14 percent; in the hypothetical world with no hierarchical segregation, that gap would have reached only 10 percent. The hypothetical world in which there is no occupational segregation shows results that are somewhat surprising, as the earnings gap exceeds the one actually observed. Moreover, the difference between the hypo- thetical and the actual gap increases over time, mainly during the 1990s, so that by 2004 the earnings gap would have been 3 percentage points larger than the gap at the beginning of the period. Why is it that a reduction in hierarchical segregation would lead to a reduction of earnings gaps but a reduction in occupational segregation would not? What forces were behind this development, and how did they evolve during the period of analysis? To approach these questions, one would like to know how the earnings gap changes when occupational segregation varies. Analyzing this question requires defining occupations dominated by men and women and studying the earnings structure in each type of occupations. Mathematically, the element of interest is ∂G/∂D, the rate at which the earnings gap (G) varies for infinitesimal changes in occupational segrega- tion (D). This element has two components (WMM – WMF) and (WFM – WFF). (For a demonstration of this result and the theoretical framework behind it, see Calónico and Ñopo 2008.) These components can be inter- preted as two different gaps. The first is the earnings gap for men—the dif- ference between the average earnings for men in occupations dominated by men and the average earnings for men in occupations dominated by women. The second is the gap for women: the difference between the aver- age earnings for women in occupations dominated by men and the average earnings for women in occupations dominated by women. Male and female dominance were defined on the basis of the gender composition in each occupation over the period under analysis. Three out of seven occupations at the one-digit level (managers, workers in agri- cultural activities, and workers in industrial activities) were considered dominated by men. The other four (professionals and technicians, admin- istrative personnel, salespersons, and workers in the service sector) were considered dominated by women. The upper panel of Figure 6.4 shows the estimation of ∂G/∂D and its components. Both components, and hence ∂G/∂D, are negative for the whole period under analysis—that is, average earnings of men and women in occupations dominated by women were higher than average earnings of men and women in occupations dominated by men. Hence, it is not surprising to observe that a reduction in gender occupational segregation would lead to an increase in gender earnings gaps in Mexico. The difference between the actual earnings gap and the hypothetical earnings gap without occupational segregation increased during the period gaps and workplace segregation: mexico, 1994–2004 111 Figure 6.4 Estimated Changes in Gender Earnings Gap as a Result of Changes in Occupational and Hierarchical Segregation in Mexico, 1994–2004 a. Estimated changes in earnings gap as a result of changes in occupational segregation percentage of average women’s 0 –10 –20 earnings –30 –40 –50 –60 4 5 5 6 6 7 7 8 8 9 9 0 0 1 1 2 2 3 3 4 4 -9 I-9 -9 I-9 -9 I-9 -9 I-9 -9 I-9 -9 I-0 -0 I-0 -0 I-0 -0 I-0 -0 I-0 -0 III III III III III III III III III III III quarter b. Estimated changes in earnings gap as a result of changes hierarchical segregation percentage of average women’s 160 120 earnings 80 40 0 4 5 5 6 6 7 7 8 8 9 9 0 0 1 1 2 2 3 3 4 4 -9 I-9 -9 I-9 -9 I-9 -9 I-9 -9 I-9 -9 I-0 -0 I-0 -0 I-0 -0 I-0 -0 I-0 -0 III III III III III III III III III III III quarter men component (MC) women component (FC) total (MC+FC) Source: Based on data from 1994–2004 ENEU. of analysis. This result is observed in the estimation of ∂G/∂D, which was more negative in the later years of the analysis. A more than proportional decrease in earnings in occupations dominated by men guides this increas- ing difference. The analogous exercise was performed with hierarchies instead of occupations (panel b in Figure 6.4). For this purpose, three of the five hierarchical categories (managers, piece-rate or by commission workers, and members of cooperatives) were considered dominated by men, and two categories (independent workers and fixed-salaried workers) were considered dominated by women. In contrast to the results for occupa- tions, these results indicate that reductions in hierarchical segregation are expected to be linked to reductions in the earnings gap. This finding is in line with the results reported earlier in this chapter. 112 new century, old disparities Hence, although hierarchical segregation by gender has been substan- tially lower than occupational segregation during the last two decades, the two types of segregation have had highly dissimilar impacts on earn- ings. Eliminating hierarchical segregation would reduce the observed gender earnings gap by about 5 percentage points, whereas eliminating occupational segregation would increase it by about 6 percentage points. A reduction in gender differences in age, schooling, and marital status would have a greater impact on the reduction of observed occupational segregation than on hierarchical segregation. A reduction in occupational segregation would have a significant impact on the reduction of hierar- chical segregation, although the reverse may not necessarily be true. The results also suggest that gender equalization of human capital charac- teristics would help reduce not only gender earnings gaps but also both hierarchical and occupational segregation. The next four chapters analyze what happened in countries in which this equalization occurred. As in all Latin American countries, women’s labor participation in these countries increased in the past two decades. However, the ranking of these countries varied. In Chile (chapter 7), women’s labor participation has been below the Latin American aver- age since the early 1990s. In Colombia (chapter 8), women’s labor force participation has been about average for the region since the early 1990s. Brazil (chapter 9) and Ecuador (chapter 10) are among the top five Latin American countries based on women’s labor participation in both the early 1990s and the 2000s. Notes 1. At the end of 2001, the Encuesta Nacional de Empleo (ENE) replaced the ENEU, extending coverage to the entire country. The analysis here is restricted to the urban subsample of the ENE, however, which is comparable to the ENEU sample. At the beginning of the period under consideration, the cities included in the sample represented about 40 percent of Mexico’s working population. During the 10-year span examined in this chapter, coverage increased to 60 percent (48 cities). 2. Computation at the two-digit level, which includes 18 occupations, yields results that are qualitatively similar (albeit somewhat stronger). These computations are available from the authors upon request. For a discussion of the influence of the number of categories on the computation of the Duncan index, see Anker (1998). 3. For a description of the matching methodology, see chapter 2. References Anker, R. 1998. Gender and Jobs: Sex Segregation of Occupations in the World. Geneva: International Labor Organization. Brown, C., J. Pagan, and E. Rodriguez-Oreggia. 1999. “Occupational Attainment and Gender Earnings Differentials in Mexico.� Industrial and Labor Relations Review 53 (1): 123–35. gaps and workplace segregation: mexico, 1994–2004 113 Calónico, S., and H. Ñopo. 2008. “Gender Segregation in the Workplace and Wage Gaps: Evidence from Urban Mexico, 1994–2004.� Research Department Working Paper 636, Inter-American Development Bank, Washington, DC. Calónico, S., and H. Ñopo. 2009. “Gender Segregation in the Workplace and Wage Gaps: Evidence from Urban Mexico 1994–2004.� In Occupational and Resi- dential Segregation (Research on Economic Inequality, Volume 17), ed. Yves Flückiger, Sean F. Reardon, and Jacques Silber, 245–70. Bingley, U.K.: Emerald Group Publishing Limited. Chinhui, J., and J. Airola. 2005. “Wage Inequality in Post-Reform Mexico.� Work- ing Paper 2005–01, Department of Economics, University of Houston, Hous- ton, TX. Colmenares, G. 2006. “Segregación en el empleo por sexo: salario y ocupación en los modelos de industrialización de las regiones centro-occidente y fronteriza.� Frontera Norte 18 (35): 87–110. Duncan, O. D., and B. Duncan. 1955. “A Methodological Analysis of Segregation Indexes.� American Sociological Review 20 (2): 210–17. Elias, J., and H. Ñopo. 2010. “The Increase in Women Labor Force Participa- tion in Latin America 1990–2004: Decomposing the Changes.� Report, Inter- American Development Bank, Washington, DC. López Acevedo, G. 2003. “Wages and Productivity in Mexican Manufacturing.� Policy Research Working Paper 2964, World Bank, Washington, DC. Pagan, J., and S. Sánchez. 2000. “Gender Differences in Labor Market Decisions: Evidence from Rural Mexico.� Economic Development and Cultural Change 48 (3): 619–37. Parker, S. 1999. “Niveles salariales de hombres y mujeres: diferencias por ocupa- ción en las áreas urbanas de México.� In México diverso y desigual: enfoques sociodemográficos, coord. B. F. Campos, 373–90. Serie Investigación Demo- gráfica en México 4, Colegio de México/Sociedad Mexicana de Demografía, Mexico City. Rendón, T. 2003. “Empleo, segregación y salarios por género.� In La situación del trabajo en México, ed. E. de la Garza and C. Salas, 129–50. Mexico City: Plaza y Valdés. Rendón, T., and V. M. Maldonado. 2004. Vínculos entre trabajo doméstico, segre- gación ocupacional y diferencias de ingreso por sexo, en el México actual. Instituto de Estudios del Trabajo, Mexico City. Sánchez, S. 1998. “Gender Earnings Differentials in the Microenterprise Sector: Evidence from Rural and Urban Mexico.� Report, World Bank, Latin America and the Caribbean Region, Finance, Private Sector, and Infrastructure Sector Unit, Washington, DC. 7 Low Participation by Women, Heavy Overtime by Men: Chile 1992–2009 Despite major advances in the education of women in Chile’s labor force relative to men, gender differences in earnings remain. This chapter explores the relatively low remuneration of women’s human capital. All of the statistics and estimations presented in this chapter are based on the CASEN, the official household survey of Chile conducted by the Ministry of Social Development, (earlier named, Ministry of Planning and Cooperation [MIDEPLAN]) since 1987. The survey covers Chile’s entire population, in both urban and rural areas. The period under analysis runs from 1992 to 2009. As the main objec- tive of this chapter is to estimate and explain gender earnings gaps in Chile, the population under consideration is all employed men and women 16–75 years old. Selection issues are ignored; earnings are measured as hourly earnings. What Does the Literature Show? The literature on the Chilean gender earnings gap is not new. The first studies on the topic include work by Paredes (1982) and Paredes and Riveros (1994). Performing Blinder-Oaxaca decompositions for the period 1958–90 in the metropolitan area of Santiago, they provide evidence of This chapter was adapted from the following sources: “The Gender Wage Gap in Chile 1992–2003 from a Matching Comparisons Perspective,� Hugo Ñopo, IZA Discussion Paper 2698, Institute for the Study of Labor, 2007; “The Gender Wage Gap in Chile 1992–2003 from a Matching Comparisons Perspective,� Hugo Ñopo, RES Working Paper 4463, Inter-American Development Bank, 2007. 115 116 new century, old disparities unexplained gender earnings differences, which they find correlated with the business cycle. Along similar lines, both methodologically and with respect to the data set utilized, Contreras and Puentes (2000) study the evolution of the gender gap for the period 1958–96 in Greater Santiago, reaching similar conclusions. Their evidence suggests that unexplained differences in earn- ings decreased from the 1960s to the 1980s, before this trend was reversed in the 1990s. Additionally, they find that these unexplained gender differ- ences in pay are largely a result of the underpayment of women rather than the overpayment of men. Montenegro and Paredes (1999) introduce a quantile regressions approach to the analysis, complementing the Blinder-Oaxaca decompo- sitions with a deeper exploration of the distribution of unexplained pay differences. Using the same data set as the previous studies for the period 1960–98, they find systematic gender differences in returns to education and experience along the conditional earnings distribution. Returns to education are higher for women in lower quantiles and lower for women in upper quantiles. The authors do not find a systematic pattern in the level of unexplained differences in pay over time except for the last decade, when, despite a tighter labor market, they observe an increase in the gender earn- ings differential. They show that the gender earnings gap is much larger in the upper quantiles and report that although the gender earnings gap was falling in Chile, the unexplained component of it was increasing. This result is consistent with the findings of García, Hernández, and López (2001). Montenegro (2001) analyzes gender differentials in returns to educa- tion, returns to experience, and earnings. Using quantile regressions with Blinder-Oaxaca decompositions and micro data from Chile’s National Socioeconomic Characterization Survey (Encuesta de Caracterización Socioeconómico Nacional [CASEN]), which are nationally and region- ally representative for the period 1990–98, he finds systematic gender differences in returns to education and experience along the conditional earnings distribution. The results show that returns to education are sig- nificantly different for men and women by quintiles, although returns to education at the median produce very similar results for men and women, implying that an ordinary least squares mean estimate will not detect the richness of these gender differences. The results for returns to years of experience show that in the lower quantiles, men and women have similar rates of return, whereas in the upper quintiles men tend to have higher rates of return. Montenegro also finds evidence that the unexplained earn- ings differential is larger in the upper quintiles of the conditional earnings distribution. In particular, he shows that the unexplained earnings gap steadily increases from 10 percent to 40 percent when moving from the lower to the upper part of the conditional earnings distribution. Bravo, Sanhueza, and Urzúa (2008) use the Chilean Social Protec- tion Survey 2002 (SPS02), which includes information on variables that gaps and uneven participation: chile, 1992–2009 117 determine social security participation in Chile. They focus their attention on individuals age 28–40 at the time of the survey, who have most likely completed their last level of schooling and studied under the same educa- tion system (which changed radically in 1980 in Chile). Their study uses traditional linear and nonlinear models of earnings differentials with selec- tion correction (Heckman 1979). The authors find that gender earnings gaps are about 23 percent of women’s earnings and grow to 29 percent after correcting for selection. They also find that the gender gap is larger (36–38 percent of women’s earnings) among university graduates, regard- less of their experience. All other labor market characteristics explored show no significant gender difference among university graduates. The largest gender gaps in weekly hours worked, unemployment, and experi- ence are found among less educated groups. Using the same database, Perticará (2007) estimates gender earnings differentials with a sensitivity analysis that explores the earnings gap obtained from ordinary least squares estimations for different levels of actual experience. The information in the data set allows the construction of a variable for actual experience that takes into account the fact that pat- terns of experience differ for men and women, because women are out of the labor market for longer periods than men. As a result, although aver- age years of work after school (potential experience) are similar for both men and women, average actual experience is 16.7 years for men and 9.3 years for women, and women’s experience is more volatile. Gender differ- ences in experience are smaller among the more educated. Perticará finds that the inclusion of variables measuring actual experience reduces the gender earnings gap about 50 percent, but when controlling for selection bias, the unexplained component of the gender earnings gap increases. Perticará and Bueno (2008) explore the gender earnings gap and its relation to years of actual experience. Based on a detailed sensitivity analy- sis with ordinary least squares estimations, instrumental variables, and selection correction, they find that gender earnings gaps are negative for all variables analyzed and that gaps are larger after controlling for actual experience only. Recent actual experience yields higher labor market returns, which may help explain the increase in real earnings from 2002 to 2006. The different estimation approaches presented reveal the impor- tance of correcting for endogeneity and selection problems. When not correcting for the endogeneity of the variables for educational attainment and work experience, the effect of education on earnings differentials is overestimated and the effect of experience underestimated. Perticará and Bueno calculate earnings differentials across occupations after correct- ing for labor market selection. They observe larger gender earnings gaps among blue-collar workers and salespeople and smaller gaps among pro- fessionals and administrative personnel. Using the CASEN survey, García (2000) studies the labor market par- ticipation of women and the gender earnings gap for the period 1990–98. 118 new century, old disparities She observes that the participation of women in the labor market increased across income quintiles: 24 percent of the women in the bottom income quintile and 42 percent of women in the top quintile work. The gender earnings gap also varies across income quintiles, from 43 percent in the bottom quintile to 59 percent in the top quintile. García finds similar results when analyzing the gender earnings gap for different sectors and types of job. As the difference in earnings for men and women remained stable over the period, she concludes that there is evidence that underpay- ment of women is a persistent phenomenon in Chile. Perticará and Astudillo (2008) use quantile regression techniques and the decomposition technique suggested by Mata and Machado (2005) to evaluate the unexplained component of the gender earnings gaps along the conditional earnings distribution after controlling for actual experience. They find that the portion of the gender earnings gap explained by char- acteristics is small and not statistically significant until the 50th percentile, where it becomes positive and thus favors women, growing monotonically until it reaches 7 percent in the 90th percentile. At the top of the distribu- tion, women compensate for “discrimination� with attributes that are better rewarded in the labor markets. How Do Male and Female Workers Differ? When trying to explain differences in earnings between men and women, one can attribute them to some observable individual characteris- tics that determine earnings. Doing so would be a valid argument in cases where differences in age, education, occupational experience, and occupa- tion, among other characteristics, exist. The purpose of this chapter is to measure the extent to which these differences in characteristics explain differences in pay between men and women in Chile.1 Exploring some descriptive statistics by gender elucidates this notion. This section explores the main characteristics of working men and women, including education, labor market participation, unemployment rates, average working hours, and hourly earnings. Differences in Education Female workers in Chile have higher levels of education than men (figure 7.1). On average, women have one more year of schooling than men in Chile. In 1992, on average, women had 10.2 years of education and men had 9.1 years. In 2009, the average was 12.0 years of schooling for women and 11.2 years for men. The observed increase in average years of schooling during this period was slightly greater for men than for women: between 1992 and 2009, average years of education increased 18.5 percent for women and 22.8 percent for men. As a result, the schooling gender gaps and uneven participation: chile, 1992–2009 119 gap narrowed slightly during the 1990s and 2000s, although it still favors women. The educational gender gap in Chile is evident in both rural and urban areas. Over the 1992–2009 period, the average number of years of school- ing for all workers was 7.5 years in rural areas and 11.1 years in urban areas. The average increased in both areas. Large differences are observed in years of schooling of working men and women in rural areas, where women have on average 1.6 years of education more than men. In urban areas, this difference is 0.5 year. The share of less educated workers decreased sharply between 1992 and 2009, and the share of workers with university education rose. In 2009, 11.2 percent of working people had no education or had not com- pleted primary school. This proportion was 25.2 percent in 1992. Over the same period, the share of working people who completed university rose from 8.0 percent to 14.6 percent (figure 7.2). Both men and women saw important improvements over the period, during which the difference in favor of female workers remained almost constant. The percentage of the working population with university education is lower in rural areas (2.2 percent) than in urban areas (10.8 percent). This difference widened during the period of analysis. Gender differences persisted over the whole period, and the difference in the percentage of Figure 7.1 Average Years of Education of Men and Women in Chile’s Labor Force, 1992–2009 14 12 average years of education 10 8 6 4 2 0 1992 1994 1996 1998 2000 2003 2006 2009 women men Source: Based on data from 1992–2009 CASEN. 120 new century, old disparities Figure 7.2 Percentage of Men and Women in Chile’s Labor Force with University Degrees, 1992–2009 16 14 12 10 percent 8 6 4 2 0 1992 1994 1996 1998 2000 2003 2006 2009 women men Source: Based on data from 1992–2009 CASEN. men and women with at least a university degree grew. The gap widened more quickly in rural areas, although it is larger in urban areas. Figure 7.3 presents the percentage of employed men and women with less than high school education. It shows evidence of both a general improvement in education and a gender gap in favor of women. About 80 percent of employed men and women in rural areas have not completed high school; in urban areas, this percentage falls to 42 percent. Differences in Labor Force Participation, Unemployment, and Hours Worked The Chilean labor market has several particularities. Two striking stylized facts are low female labor force participation and the high number of hours of work, especially among men. Figure 7.4 shows the evolution of participation rates for men and women. The evidence indicates that the gender gap in participation nar- rowed during the last decade, as a result of both a decrease in male partici- pation and an increase in female participation. In 1992, only 25 percent of women participated in the labor market; by 2009, this proportion reached 33 percent. Accompanying this increase in participation, Chile experienced an increase in unemployment, particularly in 1998, when the unemployment gaps and uneven participation: chile, 1992–2009 121 Figure 7.3 Percentage of Men and Women in Chile’s Labor Force with Less Than Secondary Education, 1992–2009 70 60 50 40 percent 30 20 10 0 1992 1994 1996 1998 2000 2003 2006 2009 women men Source: Based on data from 1992–2009 CASEN. Figure 7.4 Labor Force Participation Rates of Men and Women in Chile, 1992–2009 60 50 40 percent 30 20 10 0 1992 1994 1996 1998 2000 2003 2006 2009 women men Source: Based on data from 1992–2009 CASEN. 122 new century, old disparities rate jumped from 2.3 percent to 4.0 percent, affecting men and women equally (figure 7.5). Overall, gender differences in unemployment did not change much from the beginning to the end of the period, although they increased between 1996 and 2000 and decreased between 2000 and 2006, when unemployment and the difference between men and women increased again. Gender differences in unemployment rates are evident by level of educa- tional attainment as well (figure 7.6). Less educated people in Chile, espe- cially women, display higher unemployment rates. Among less educated people, the increase in unemployment in 1998 (with respect to 1996) was similar for men and women. For university graduates, however, the change in unemployment disproportionately affected women, unemployment of whom more than doubled between 1996 and 1998. The evolution of the gender composition of the labor force by occupa- tions shows a slight reduction in the gap among merchants and workers in the service and agricultural sectors. Another stylized fact to highlight is the apparent lack of a gap among managers. Women’s participation in the labor force is concentrated in the service sector (about 45 percent of working women are employed as service workers, merchants, or salespeo- ple). In contrast, about 60 percent of men are blue-collar or agricultural workers. An important variable to take into account when analyzing earn- ings gaps is occupational experience. Traditionally, studies have used a Figure 7.5 Unemployment Rates of Men and Women in Chile, 1992–2009 6 5 4 percent 3 2 1 0 1992 1994 1996 1998 2000 2003 2006 2009 women men Source: Based on data from 1992–2009 CASEN. gaps and uneven participation: chile, 1992–2009 123 Figure 7.6 Unemployment Rates of Men and Women in Chile, by Educational Level, 1992–2009 a. Women 8 7 6 5 percent 4 3 2 1 0 1992 1994 1996 1998 2000 2003 2006 2009 no education or incomplete primary complete secondary or incomplete university complete primary or incomplete secondary tertiary education 8 b. Men 7 6 5 percent 4 3 2 1 0 1992 1994 1996 1998 2000 2003 2006 2009 no education or incomplete primary complete secondary or incomplete university complete primary or incomplete secondary tertiary education Source: Based on data from 1992–2009 CASEN. proxy—potential experience—computed as a linear combination of age and schooling. The evidence suggests that this approach tends to produce biased estimates of the gender gap (see Weichselbaumer and Winter-Ebmer 2003). The CASEN data provide a rare opportunity to use occupational experience, at least for the last four years under analysis (figure 7.7). Aver- age years at the same occupation remained fairly constant over 2000–09, but gender differences grew in the last year and favored men during the whole period. Figure 7.8 presents the average number of hours of work per week by gender. Working hours increased from 49.5 hours per week in 1992 to 51.6 in 1998, decreasing after that to 43.6 in 2009. The peak—observed in 1998, when men worked an average of 53.4 hours a week—can be 124 new century, old disparities Figure 7.7 Average Years at Same Job by Men and Women in Chile, 2000–09 9 8 7 number of years 6 5 4 3 2 1 0 2000 2003 2006 2009 women men Source: Based on data from 2000–09 CASEN. Figure 7.8 Average Weekly Hours Worked by Men and Women in Chile, 1992–2009 60 50 number of hours 40 30 20 10 0 1992 1994 1996 1998 2000 2003 2006 2009 women men Source: Based on data from 1992–2009 CASEN. gaps and uneven participation: chile, 1992–2009 125 linked to the recession of 1998 (and the corresponding increase in unem- ployment). The gender gap in hours of work was about 3.5 hours at the beginning of the 1990s; it increased until 2006, reaching 5.9 hours, before decreasing to 4.4 hours in 2009. However, there was an overall increase in the gap during the period of analysis. Working hours is one of the few individual characteristics for which the gender gap widened during the decade. Workers with less education used to devote more hours to the labor market than skilled workers (figure 7.9). However, this gap, which was wide in the early 1990s, narrowed until 2003, as hours worked fell for all levels of skill but especially for less skilled workers. In 2009, a typical highly educated Chilean worked one hour less than in 1990, but a typical unskilled worker worked nearly five hours less than in the early 1990s. The differences in the hours of work by workers of different education levels started to narrow in 1998 and had almost disappeared by 2003. At the beginning of the decade, the difference in the number of hours worked by educational level was larger for women than for men; by 2009, such differences had become almost negligible for both men and women. How- ever, by 2009, the average hours of work of employed men and women seem to be independent of educational level. Differences in Earnings Working women in Chile have more schooling than men, in both rural and urban areas. As education is an important determinant of earnings, it would be expected that women would have higher earnings than men. In fact, the statistics show the opposite result. During the 1990s, the Chilean economy performed better than that of all other countries in the region. Average annual gross domestic product growth was 6.3 percent, and the rate of inflation was the lowest in four decades. As a result, earnings increased considerably since 1996, even in 1998, when the economy suffered a slowdown. Between 1990 and 2009, average real hourly earnings (deflated by the consumer price index) increased 51 percent (54 percent for men, 51 percent for women). Figure 7.10 shows the average hourly earnings gap as a multiple of average hourly earnings of women. The gap between men and women reached the widest level of the decade in 2000, when men earned on aver- age 35 percent more than women. The gender earnings gap ranged from 25 percent at the beginning of the decade to 35 percent in 2000, when it started decreasing, reaching 15 percent of women’s earnings in 2006. The period ended with an increase in the gap, which rose to 30 percent in 2009. The gender gap in hourly earnings is substantially higher in urban than in rural areas. 126 new century, old disparities Figure 7.9 Average Weekly Hours Worked by Men and Women in Chile by Educational Level, 1992–2009 60 a. Women 50 number of hours 40 30 20 10 0 1992 1994 1996 1998 2000 2003 2006 2009 no education or incomplete primary complete secondary or incomplete university complete primary or incomplete secondary teritary education b. Men 60 50 number of hours 40 30 20 10 0 1992 1994 1996 1998 2000 2003 2006 2009 no education or incomplete primary complete secondary or incomplete university complete primary or incomplete secondary teritary education Source: Based on data from 1992–2004 CASEN. The Role of Individual Characteristics in Explaining the Gender Earnings Gap The gender earnings gap reported in Figure 7.10 does not take account of the fact that men and women differ in observable characteristics that the labor market rewards. It is important to measure the extent to which gender differences in observable human capital characteristics explain the gender earnings gap and the extent to which gender differences remain unexplained. Doing so involves decomposing the gender earnings gap. gaps and uneven participation: chile, 1992–2009 127 Figure 7.10 Gender Gap in Hourly Earnings in Chile, 1992–2009 40 percentage of average women’s earnings 35 30 25 20 15 10 5 0 1992 1994 1996 1998 2000 2003 2006 2009 Source: Based on data from 1992–2009 CASEN. In this section, men and women are matched on the basis of five com- binations of observable characteristics. The first set considers age, marital status, and years of schooling. The second set adds a variable that captures whether the worker works full time or part time. The third set replaces full- time and part-time status with occupational category (which aggregates occupations at the one-digit level). The fourth set simultaneously considers all the variables considered in the three previous sets. The fifth set adds years of occupational experience to the set of variables in the fourth set. Table 7.1 reports the average statistics for men and women in and out of the “common support� for each set of matching characteristics. In general, unmatched men and women are older than their matched coun- terparts. In contrast, average years of education are higher in matched groups than in unmatched ones (except in the set of controls that includes occupational experience). Most of the men and women in the common support are (formally or informally) married. This is also the case for male workers outside the common support. In contrast, most unmatched women are separated or widows. Most of the matched men and women (about 30 percent) are service workers. A smaller percentage of men and women in the common support work as directors or managers relative to men and women outside the common support. On average, matched men and women work more hours than unmatched workers, and the difference is larger for women. When average years at the 128 Table 7.1 Demographic and Job Characteristics of Matched and Unmatched Samples of Men and Women in Chile’s Labor Force, 1992–2009 (percent) Age, education, marital status, Age, education, marital status, time Age, education,marital status, and and occupation worked, and occupation years at the same job Matched Unmatched Unmatched Matched Unmatched Unmatched Matched Unmatched Unmatched Characteristics sample women men sample women men sample women men Average age 37.7 48.9 48.0 37.5 47.3 49.3 36.8 46.0 46.5 Average years of schooling 11.3 10.8 10.8 11.4 10.2 10.3 11.7 11.8 10.5 Marital status Single 35.0 20.6 17.9 35.4 19.7 20.1 34.0 31.0 15.7 Married 51.1 2.9 62.8 51.8 12.3 60.7 57.8 28.0 74.0 Divorced 11.2 26.5 11.5 10.5 33.0 11.3 7.5 28.7 7.8 Widower 2.8 50.0 7.8 2.3 34.9 7.9 0.8 12.3 2.5 Education level No education or primary incomplete 16.6 17.5 20.4 16.4 22.6 23.1 12.5 15.4 23.5 Primary incomplete or secondary incomplete 21.2 31.8 21.0 21.1 30.1 23.5 20.4 23.0 30.7 Secondary incomplete or tertiary incomplete 50.7 36.3 41.3 51.0 36.2 37.4 55.9 43.0 32.6 Tertiary complete 11.5 14.3 17.2 11.6 11.1 16.0 11.3 18.6 13.2 Table 7.1 (continued) Age, education, marital status, Age, education, marital status, time Age, education,marital status, and and occupation worked, and occupation years at the same job Matched Unmatched Unmatched Matched Unmatched Unmatched Matched Unmatched Unmatched Characteristics sample women men sample women men sample women men Occupation Professionals and technicians 22.2 13.8 8.4 22.4 12.7 10.0 Directors and upper management 5.0 12.9 6.9 4.9 10.4 9.0 Administrative personnel 14.9 20.6 6.8 15.0 15.3 6.3 Merchants and sellers 12.6 18.7 4.7 12.5 18.0 5.5 Service workers 30.7 28.0 2.9 30.3 37.3 4.1 Agricultural workers and similar 5.1 1.1 11.7 5.2 1.3 12.5 Nonagricultural blue-collars 9.5 3.3 21.2 9.7 4.2 28.1 Armed forces 0.0 0.4 33.5 0.0 0.2 21.9 Average hours of work 45.3 32.3 42.8 Average Years at the same job 4.8 11.2 13.7 Source: Based on data from 1992–2009 CASEN. Note: Blank cells indicate that a variable is not being controlled for. 129 130 new century, old disparities same job are used as a control for the matching, men and women in the common support remain at the same job for 4.8 years on average, whereas women out of the common support stay 11.2 years and men 13.7 years. Figures 7.11 and 7.12 report the earnings gap decompositions in which some empirical regularities arise. First, the differences in observable characteristics (ΔX), and to some extent the component of the earnings gap that reflects the fact that women achieve certain combinations of characteristics that men do not (ΔF ), are negative. This result reflects the fact that human capital characteristics, espe- cially education, are better rewarded for women than for men in Chile. Second, the component of the gap that reflects the fact that men achieve certain combinations of characteristics that women do not (ΔM) is generally Figure 7.11 Decomposition of Gender Earnings Gap in Chile after Controlling for Demographic and Job Characteristics, 1992–2009 0.6 0.5 percentage of average women’s earnings 0.4 0.3 0.2 0.1 0 –0.1 –0.2 –0.3 1992 1994 1996 1998 2000 2003 2006 2009 ΔX ΔF ΔM Δ0 Source: Based on data from 1992–2009 CASEN. Note: ΔM (ΔF) is the part of the earnings gap attributed to the existence of men (women) with combinations of characteristics that are not met by any women (men). ΔX is the part of the earnings gap attributed to differences in the observable characteristics of men and women over the “common support.� Δ0 is the part of the earnings gap that cannot be attributed to differences in characteristics of the individuals. It is typically attributed to a combination of both unobservable characteristics and discrimination. The sum of these components equals the total earnings gap (ΔM + ΔF + ΔX + Δ0 = Δ). gaps and uneven participation: chile, 1992–2009 131 Figure 7.12 Decomposition of Gender Earnings Gap in Chile after Controlling for Age, Marital Status, Education, and Years in Same Occupation, 2000–09 0.5 percentage of average women’s earnings 0.4 0.3 0.2 0.1 0 –0.1 –0.2 2000 2003 2006 2009 ΔX ΔF ΔM Δ0 Source: Based on data from 2000–09 CASEN. Note: ΔM (ΔF) is the part of the earnings gap attributed to the existence of men (women) with combinations of characteristics that are not met by any women (men). ΔX is the part of the earnings gap attributed to differences in the observable characteristics of men and women over the “common support.� Δ0 is the part of the earnings gap that cannot be attributed to differences in characteristics of the individuals. It is typically attributed to a combination of both unobservable characteristics and discrimination. The sum of these components equals the total earnings gap (ΔM + ΔF + ΔX + Δ0 = Δ). positive, suggesting the existence of a sort of glass ceiling effect—that is, there are men with combinations of observable characteristics for which there are no comparable women, and these men have earnings that are on average higher than those in the rest of the economy. Third, the component that remains unexplained by observable charac- teristics (Δ0)—the component that cannot be attributed to differences in observable characteristics between men and women—is slightly larger than the original measure of gender earnings gaps (measured before matching). This is equivalent to saying that the measure of the gender earnings gap that remains unexplained after a Blinder-Oaxaca decomposition is larger than the original measure of the earnings gap, as reported in the literature on gender gaps in Chile summarized at the beginning of this chapter. 132 new century, old disparities Exploring the Unexplained Component of the Gender Earnings Gap Figure 7.13 presents confidence intervals for Δ0, the component that mea- sures the extent to which the gender earnings gap cannot be explained by observable individual characteristics. The extremes of the boxes represent confidence intervals at the 90 percent level; the extremes of the whiskers represent confidence intervals at the 99 percent level. The confidence inter- vals obtained from the last set of matching characteristics are larger than all others, because of the smaller number of matched men and women that corresponds to this large number of matching variables. This combination of individual characteristics best controls for gender differences (the unex- plained component is the smallest of all combinations). However, it is so restrictive that it imposes a cost in terms of standard errors. The next set of figures report on the distribution of the unexplained component. Figure 7.14 shows the distribution for the whole period, 1992–2009, using four sets of matching characteristics of the unexplained differences in earnings by percentiles of the earnings distribution. The results suggest that the unexplained component is larger among people in the highest percentiles of the earnings distribution. At the bottom of the earnings distribution, men earn an unexplained premium of 10–20 percent over comparable women; at the top of the distribution, this premium increases to 40–80 percent, depending on the set of matching characteristics. This result differs from all previous results for the rest of Latin America and the Caribbean, where earnings gaps are larger for the poorest seg- ments of the income distribution. In Chile alone, for all controls used, the gender earnings gap and its unexplained component is largest among the richest income quintile. This component increases more for the rich with each characteristic added to the controls, especially when time worked is added. Analysis of the distribution of the unexplained component of the gen- der earnings gap by other observable characteristics shows that the largest and most disperse measures are among people with more education. The largest and most dispersed gap is among managers and, to a lesser extent, professionals. There is some evidence of larger and more dispersed gaps among older individuals for almost all combinations of control characteristics, except for the one that includes on-the-job experience, for which larger and more dispersed gaps are found among middle-age individuals. By marital sta- tus, the largest gaps are found among married people. When experience (measured as years working at the same job) is introduced, however, all groups seem to have similar unexplained gaps, although the gaps are more dispersed among people who are separated. The unexplained earnings gap gaps and uneven participation: chile, 1992–2009 133 Figure 7.13 Confidence Intervals for Unexplained Gender Earnings Gap in Chile, 1992–2009 a. Unexplained earnings gap after controlling for age, marital status, education, time worked, and occupation, 1992–2009 60 percentage of average women’s earnings 50 40 30 20 10 0 1992 1994 1996 1998 2000 2003 2006 2009 30 b. Unexplained earnings gap after controlling for age, marital status, education, and years at the same job, 2000–09 percentage of average women’s earnings 25 20 15 10 5 0 2000 2003 2006 2009 Source: Based on data from 1992–2009 CASEN. Note: Boxes show 90 percent confidence intervals for unexplained earnings; whiskers show 99 percent confidence intervals. is substantially larger and more dispersed among people who work less than 20 hours per week than among the rest of the labor market. The evidence of unexplained gaps by geographic location is mixed. When experience is not taken into account, the unexplained gap is larger 134 new century, old disparities Figure 7.14 Unexplained Gender Earnings Gap in Chile, by Percentiles of Earnings Distribution, 1992–2009 80 percentage of average women’s earnings 60 40 20 0 0 10 20 30 40 50 60 70 80 90 100 earnings percentile year, age, marital status, and set 1 (year, age, marital status, and education education) + time worked set 1 (year, age, marital status, and full set (year, age, marital status, education, education) + occupation time worked, and occupation) Source: Based on data from 1992–2009 CASEN. in Santiago than in the rest of the country. But when experience is consid- ered as one of the matching variables, the unexplained gap in the provinces is larger (and more disperse) than the gap in Santiago. These results indicate that the earnings gender gap is proportionately larger among highly paid people, people with university education, direc- tors, older workers, married workers, and part-time workers. There is no clear evidence that the unexplained gender earnings gap is higher in Santiago than in the rest of the country. The results suggest the existence of a glass ceiling effect in Chile. There are particular combinations of experience, age, marital status, and educa- tion for which it is not possible to make gender comparisons. Married, older men (in their 50s and 60s) with more than 10 years of occupational experience are more likely to have no female counterparts actively work- ing in the Chilean labor market. These men are more likely to work in managerial occupations and earn hourly earnings that are substantially higher than the national average. This segment of the labor force may account for 5–8 percentage points of the gender earnings gap in Chile. gaps and uneven participation: chile, 1992–2009 135 Occupational experience seems to play an important role in explaining gender earnings gaps in Chile. Unfortunately, this variable is not available for all the years under consideration. For the years for which data are available, there are important differences in favor of men: men average eight years of occupational experience, whereas women average just six. These differences account for a large part of the earnings gap. If public policies in Chile led to increased occupational experience for women, there are good reasons to think that the gender earnings gap would narrow. The next chapter examines the gender earnings gap in a country that has passed laws in this direction. Legislation in Colombia helps prevent women from dropping out of the labor market when they give birth and begin to raise their children. Such legislation may provide disincentives to hire them but can encourage women to stay longer in their jobs. Note 1. For a description of the methodology used in this chapter, see chapter 2. References Bravo, D., C. Sanhueza, and S. Urzúa. 2008. “An Experimental Study of Labor Market Discrimination: Gender, Social Class, and Neighborhood in Chile.� Research Network Working Paper R–541, Inter-American Development Bank, Washington, DC. Contreras, D., and Puentes, E. 2000. “Is Gender Earnings Discrimination Decreas- ing in Chile? Thirty Years of ‘Robust’ Evidence.� Department of Economics, University of Chile, Santiago. García Durán, B. 2000. “Diagnóstico de la disciminación salarial de la mujer en el mercado laboral chileno.� Ph.D. diss., Universidad de Chile, Santiago. García, J., P. J. Hernández, and A. López. 2001. “How Wide Is the Gap? An Inves- tigation of Gender Wage Differences Using Quantile Regression.� Empirical Economics 26 (1): 149–67. Heckman, J. 1979. “Sample Selection Bias as a Specification Error.� Econometrica 47(1): 153–61. Mata, J., and J. A. F. Machado. 2005. “Counterfactual Decomposition of Changes in Wage Distributions Using Quantile Regression.� Journal of Applied Econo- metrics 20 (4): 445–65. Montenegro, C. 2001. “Wage Distribution in Chile: Does Gender Matter? A Quan- tile Regression Approach.� Policy Research Report on Gender and Development Working Paper 20, Development Research Group/Poverty Reduction and Eco- nomic Management Network, World Bank, Washington, DC. Montenegro, C., and R. Paredes. 1999. “Gender Wage Gap and Discrimination: A Long Term View Using Quantile Regression.� World Bank, Washington, DC, and University of Chile, Santiago. http://www.ricardoparedes.cl/paperweb/ Montenegro_Paredes.PDF. 136 new century, old disparities Paredes, R. 1982. “Diferencias de ingreso entre hombres y mujeres en el gran San- tiago 1969 y 1981.� Estudios de Economía 18: 99–121. Paredes, R., and L. Riveros. 1994. “Gender Wage Gaps in Chile. A Long Term View: 1958–1990.� Estudios de Economía 21(Número especial): 211–30. Perticará, M. 2007. “Brechas salariales por género en Chile: un análisis de sensibili- dad.� ILADES–Georgetown University Working Paper inv195, Latin American Institute of Doctrine and Social Studies (ILADES), Universidad Alberto Hurtado, Santiago, and Georgetown University, Washington, DC. Perticará, M., and A. Astudillo. 2008. “¿Qué tan alta puede ser la brecha de salarios en CHILE? Investigando diferencias salariales entre hombres y mujeres a partir de regresiones de cuantiles.� ILADES–Georgetown University Work- ing Paper inv211, Latin American Institute of Doctrine and Social Studies (ILADES), Universidad Alberto Hurtado, Santiago, and Georgetown Univer- sity, Washington, DC. Perticará, M., and I. Bueno. 2008. “Entendiendo las brechas salariales por gen- ero en Chile.� ILADES–Georgetown University Working Papers inv212, Latin American Institute of Doctrine and Social Studies (ILADES), Universidad Alberto Hurtado, Santiago, and Georgetown University, Washington, DC. Puentes, E., and D. Contreras. 2000, “Is the Gender Wage Discrimination Decreas- ing in Chile? Thirty Years of ‘Robust’ Evidence.� Latin American and Carib- bean Economic Association, Chile. Weichselbaumer, D., and R. Winter-Ebmer. 2003. “A Meta-Analysis of the Inter- national Gender Wage Gap.� IZA Discussion Paper 906, Institute for the Study of Labor, Bonn, Germany. 8 The Resilient Earnings Gap: Colombia 1994–2006 Colombia’s labor market experienced important changes during the last three decades. Despite strong improvement in the labor market characteristics of women and the legal framework to promote equal- ity, however, the gender earnings gap changed little between 1994 and 2006. Moreover, the unemployment rate among women over the past two decades has consistently been about 5 percentage points higher than the rate for men (Sabogal 2009). This chapter estimates the gender earnings gap between 1994 and 2006—a period that includes booms and recessions—and decomposes it using the methodology described in chapter 2. The data are drawn from quarterly household surveys conducted by the Colombian National Statistical Agency. Up to 2000, every other year the survey included an extensive labor market module in its second quarter release. This module included information on labor earnings, social security coverage, and firm size, among other areas. Since 2000, the extensive labor module has been included annually. This chapter analyzes all data from 1994 to 2006. As a result, 10 shifts of the survey for the period under analysis are included. The evolution of hourly earnings for women and men during this period can be divided into three qualitatively different periods. During the first period, 1994–98, earnings grew, but with marked fluctuations. This period is characterized by a slowdown in overall economic activity. During the sec- ond period, 2000–01, earnings fell more than 10 percent, and the economy suffered a steep economic decline. During the third period, 2002–06, real This chapter was adapted from “The Persistent Gender Earnings Gap in Colombia, 1994–2006,� Alejandro Hoyos, Hugo Ñopo, and Ximena Peña, IZA Discussion Paper 5073, Institute for the Study of Labor, 2010. Alejandro Hoyos is a consultant at the Poverty Reduction and Economic Man- agement Network (PREM) at the World Bank. Ximena Peña is assistant professor of economics at the Universidad de los Andes, Bogota. 137 138 new century, old disparities earnings and gross domestic product grew. The analysis of the evolution of earnings gaps is based on pooled datasets that include all available surveys in each period. The analysis is restricted to the 10 largest metropolitan areas in Colombia (Barranquilla, Bucaramanga, Bogotá, Manizales, Medellín, Cali, Pasto, Villavicencio, Pereira, and Cúcuta). Within these cities, the sample includes only people 18–65 years old who reported positive earn- ings and who worked 10–84 hours a week. The sample excludes people on whom information regarding their observable characteristics was missing. For each year and gender, the top 1 percent of the earnings distributions was dropped, because these data likely represented measurement error outliers for the variable of interest. What Does the Literature Show? Several studies measure the average gender earnings gap in Colombia (Tenjo 1993; Tenjo, Ribero, and Bernat 2006). Their findings identify a substantial gender earnings gap, which is explained largely by differences in the rewards to labor market characteristics rather than gender differ- ences in characteristics. Abadía (2005) tries to determine whether statistical discrimination can explain gender pay disparities in Colombia.1 She argues that if firms do not apply statistical discrimination, the gender earnings gap will not change with experience, whereas if they do, the gap will depend less on easily observable characteristics (such as gender and education). Based on such intuition, she finds evidence of statistical discrimination against women in the private sector but not in the public sector. In response to the possible discrimination against women in the labor market, Colombia has issued labor regulations favoring women. Angel– Urdinola and Wodon (2006) document the long–term increase in the gender earnings gap in the years following the issuing of Law 50 of 1990, which ensures that pregnant women cannot be fired and gives them 12 weeks of paid leave following childbirth. The law raised the cost to firms of employ- ing women. Sabogal (2009) finds that the gender earnings gap is procyclical for workers 25–55 years old. Three mechanisms contribute to the procycli- cality of the gap: the additional worker effect, which leads to an increase in the labor supply from a nonworking household member when another member becomes unemployed; changes in the composition of the formal and informal worker forces; and changes in the sectoral composition of the labor force. Although gender earnings gaps in Colombia appeared to start diminish- ing in 2000, no major advances have been made in reducing discrimination (Bernat 2007). Studies that go beyond the analysis confined to averages the resilient gaps: colombia, 1994–2006 139 and analyze gender earnings gaps along the distributions of earnings or its conditional distribution yield other interesting finding on the persistence of differences between men and women in the Colombian labor market. Bernat (2007) explores the distribution of the gender gap, using discrimi- nation curves for 2000, 2003, and 2006. She reports that the percentage of women discriminated against actually increased throughout the period of analysis. Her results also suggest the existence of a glass ceiling (barriers that prevent women from reaching the top of the earnings distribution) for professional women. Fernández (2006) also finds evidence of glass ceilings for women in Colombia’s urban labor market, where the gap favoring men reaches 25 percent of hourly earnings at the top of the income distribution. Using the urban subsamples of the 1997 and 2003 Living Standards Measure- ment Survey, she reports no statistically significant gender differences in earnings along the distribution of income. The behavior along the whole distribution portrays interesting varia- tions in addition to the glass ceiling effect. At the bottom 3 percent of the distribution, the earnings gap favors men; in percentiles 4–85, the earnings gap favors women. Using quantile regression analysis, Fernández reports that the gap largely reflects differences in rewards rather than observable characteristics. Badel and Peña (2009) use the Colombian household sur- veys to measure the gender earnings gap for people 25–55 years old in the country’s seven main cities in 1986, 1996, and 2006. They use quantile regression techniques to examine the degree to which differences in the distribution of observable characteristics explain the gender earnings gap. The gender earnings gap for their sample is always positive, significant, and U–shaped with respect to earnings: women’s earnings fall farther below men’s at the extremes of the distribution and are closer to men’s earnings around the middle of the distribution. The authors account for selection, as self–selection of women into work is important. They find that more able women self–select into work. If all women worked, the observed gender earnings gap would be 50 percent higher than it is. How Do Male and Female Workers Differ? Table 8.1 reports normalized relative earnings by different sets of observ- able individual characteristics (the data are normalized to make average women’s earnings equal to 100 at each period). The gender earnings gap was higher during the earlier period (reaching almost 18 percent of aver- age women’s earnings) and similar in the intermediate and later period (at almost 14 percent). Some gender differences are worth noting over the life cycle. Men reach their earnings peak at 45–54 during all three periods under analysis. Table 8.1 Relative Hourly Earnings of Men and Women in Colombia, 1994–2006 140 1994–98 2000–01 2002–06 (Base: average women’s earnings (Base: average women’s (Base: average women’s for each year = 100) earnings for each year = 100) earnings for each year = 100) Characteristics Women Men Women Men Women Men All 100.0 118.3 100.0 113.8 100.0 113.5 Age 18–24 76.2*** 79.4 74.1*** 76.3 72.4 72.9 † 25–34 102.2 114.0 102.3 109.6 101.3 109.2 35–44† 113.1 134.1 108.8 127.1 107.0 126.4 † 45–54 106.3 143.6 112.7 135.7 114.7 131.8 † 55–65 89.9 130.7 91.5 124.6 95.0 129.6 Education None or primary incomplete† 50.4 68.4 48.6 61.8 45.5 56.9 Primary complete or secondary incomplete† 66.4 85.7 62.6 78.7 58.4 71.4 Secondary complete or tertiary incomplete† 109.6 124.2 103.1 116.3 92.8 107.0 † Tertiary complete 223.5 291.8 244.1 286.3 229.4 277.3 Presence of children (6 years or younger in household) No† 100.9 120.4 102.6 115.6 101.4 115.5 † Yes 97.0 113.3 90.5 108.7 94.3 107.2 Table 8.1 (continued) 1994–98 2000–01 2002–06 (Base: average women’s earnings (Base: average women’s (Base: average women’s for each year = 100) earnings for each year = 100) earnings for each year = 100) Characteristics Women Men Women Men Women Men Marital status Cohabiting† 81.2 94.7 79.2 90.9 81.5 88.7 Married† 124.6 144.6 126.4 142.2 129.7 147.9 Widowed, divorced or separated† 88.4 113.3 90.8 106.7 91.1 107.9 † Single (never married) 95.0 101.4 97.6 101.9 95.0 98.6 Presence of other household member with labor income No 94.7 113.1 101.0 113.8 103.5 113.2 Yes 101.4 121.2 99.6 113.7 98.8 113.7 Type of employment Employer 157.6*** 192.1 185.6 186.3 177.4*** 202.6 † Self-employed 82.3 105.6 71.8 88.2 74.9 89.8 † Private employee 98.8 106.4 105.4 109.0 107.5 105.9 Public employee 178.8** 183.9 218.0 216.4 233.2 230.4 Domestic servants† 40.2 54.3 46.9 67.2 45.0 64.6 (continued next page) 141 Table 8.1 (continued) 1994–98 2000–01 2002–06 142 (Base: average women’s earnings (Base: average women’s (Base: average women’s for each year = 100) earnings for each year = 100) earnings for each year = 100) Characteristics Women Men Women Men Women Men Time worked Part time† 129.5 165.8 112.7 155.9 104.6 147.7 † Full time 105.0 126.4 113.3 130.0 115.8 131.4 † Over time 67.1 96.3 67.8 87.1 70.2 89.1 Formality No† 73.4 94.6 67.3 81.3 63.3 77.0 † Yes 119.2 139.2 129.3 145.6 130.1 142.6 Small firm No 121.2*** 131.3 134.1*** 139.9 135.7* 136.9 † Yes 75.4 102.9 69.1 90.0 67.3 88.8 Economic sector Primary 113.5*** 140.8 93.2*** 115.1 123.5 128.2 † Secondary 88.3 103.5 91.2 101.4 90.4 98.6 † Tertiary 103.5 125.8 102.4 119.3 102.3 120.5 Occupation White collar† 128.4 162.4 134.6 156.8 137.5 159.7 † Blue collar 66.3 89.2 65.4 84.6 62.7 80.5 Source: Based on data from 1994–2006 household surveys conducted by the Colombian National Statistical Agency. Note: *p < 0.1, **p < 0.05, ***p < 0.01. † indicates that the earnings differences between men and women are statistically different at the 99 percent level in all three periods. the resilient gaps: colombia, 1994–2006 143 In contrast, women’s earnings profile over the life cycle changed slightly during the period. In the earliest period, their earnings peaked at 35–44. For the two later periods, the peaks were achieved at 45–54. This change may reflect a secular trend in which women are remaining longer in the labor market, maintaining their productive cycles and avoiding early retirement. Regarding education, men earn more than women in all education cat- egories in all three periods. Individuals living in households with children younger than six tend to earn less than individuals living in households with no children. This difference remained constant over the period of analysis. The presence of other income earners in the household does not seem to play an important role in earnings differentials. Married people earn more than unmarried people. Never–married people earn almost the same as people who live together, but people living together outside of a formal marriage persistently earn the lowest earn- ing. Gender earnings gaps are more pronounced among people cohabit- ing than among people who never married. The largest earnings gaps are among people who are widowed, divorced, or separated. Not surprisingly, in the private sector, employers earn much more than employees, who earn more than the self-employed, who earn more than domestic servants. An unexpected result is that public employees are at the top of average earnings by type of employment. Part-time workers (people who work less than 35 hours a week) earn much more per hour than people working full time, who in turn earn more per hour than people working overtime (more than 48 hours a week). Informal work- ers earn less than their formal counterparts, and people working at small firms (five workers or less) earn less than people working at larger firms. Services (business and social) and construction are among the highest-paid economic sectors, especially for women. At the other extreme, household and personal services was the lowest-paid sector during the whole period of analysis, for both men and women. White-collar workers earn more than blue-collar workers. Table 8.2 describes the differences in observable characteristics between men and women for the three periods under study. Working men are slightly older than working women. However, both women and men are staying longer in the labor market, creating an older labor force, especially for women. The percentage of workers with secondary and tertiary educa- tion increased in each subperiod. Although the majority of working men live in households with children six or under, the prevalence of children decreased over the period of study, and gender differences narrowed. In line with the findings by Amador and Bernal (2009), important changes took place in patterns of family forma- tion and dissolution in Colombia, similar to changes that have occurred in the rest of the region. The percentages of cohabiting people increased for men and women (although cohabitation is more common among men). Table 8.2 Demographic and Job Characteristics of Men and Women in Colombia’s Labor Force, 1994–2006 (percent) 144 1994–98 2000–01 2002–06 Characteristics Women Men Women Men Women Men Real hourly earnings (1998 Colombian pesos) 2,225 2,632 1,948 2,217 1,998 2,269 Personal characteristics Age 18–24 19.8 18.6 18.4 17.5 17.2 16.6 25–34 36.0 33.8 32.2 32.2 30.5 30.6 35–44 27.4 25.5 29.3 27.2 29.0 26.7 45–54 12.4 14.6 14.8 15.7 17.7 18.0 55–65 4.4 7.5 5.2 7.4 5.7 8.2 Education None or primary incomplete 11.6 12.2 10.4 11.4 8.8 9.5 Primary complete or secondary incomplete 38.5 45.9 37.9 41.0 32.6 35.8 Secondary complete or tertiary incomplete 37.7 30.9 38.9 36.3 42.0 40.6 Table 8.2 (continued) 1994–98 2000–01 2002–06 Characteristics Women Men Women Men Women Men Tertiary complete 12.2 11.1 12.7 11.2 16.5 14.1 Presence of children 22.7 29.6 21.3 26.6 19.2 23.5 (6 years or younger in the household) Marital Status Cohabiting 15.2 24.5 18.9 29.3 19.8 29.8 Married 28.5 41.6 25.9 36.7 24.6 34.9 Widowed, divorced, 20.6 4.6 23.4 6.7 22.3 7.0 or separated Single (never married) 35.6 29.3 31.8 27.4 33.3 28.3 Presence of other 79.5 64.5 73.3 62.1 73.5 63.8 household member with labor income Job characteristics Type of employment Employer 3.3 7.1 2.6 5.6 2.8 5.9 Self-employed 22.0 26.4 27.3 32.6 26.9 30.9 (continued next page) 145 146 Table 8.2 (continued) 1994–98 2000–01 2002–06 Characteristics Women Men Women Men Women Men Private employee 54.8 58.4 49.7 54.6 50.8 57.0 Public employee 10.5 7.9 8.0 6.8 6.1 5.7 Domestic servants 9.4 0.2 12.4 0.4 13.4 0.5 Time worked Part time 16.5 6.9 21.6 10.5 20.6 8.9 Full time 59.6 57.2 49.5 45.2 49.9 45.4 Overtime 23.9 35.9 29.0 44.2 29.6 45.7 Small firm (five workers 46.3 45.8 52.5 52.3 52.2 48.7 or less) Formality 58.1 53.1 52.8 50.4 54.9 55.7 Economic sector Primary 0.8 1.8 0.9 2.0 0.8 2.1 Secondary 23.5 35.0 20.9 30.5 20.4 32.5 Tertiary 75.7 63.2 78.3 67.5 78.8 65.4 Occupation White collar 54.2 39.8 50.0 40.4 49.8 41.7 Blue collar 45.8 60.2 50.0 59.6 50.2 58.3 Source: Based on data from 1994–2006 household surveys conducted by the Colombian National Statistical Agency. the resilient gaps: colombia, 1994–2006 147 About two out of three men and less than half of working women in Colombia are married (either formally or informally). Working women are more likely than working men to live in households in which other members earn labor income. The presence of other income earners at home did not change for men between 1994 and 2006 but changed slightly for women. The percentage of women who share the breadwinning responsibilities in their household dropped almost 5 percentage points dur- ing the period of analysis, reflecting the increase in the number of house- holds headed by women that Colombia and the region have experienced in recent years. Very few workers are employers, and about two–thirds of employers are men. The share of self-employment increased at the expense of the share of employees. The percentage of men working overtime increased, such that in the last period, almost half of men reported working more than 48 hours a week. For women, the data show both an increase in over- time and an increase in part-time work (at the expense of full–time work). About half of workers (both men and women) are formal employees at small firms. The transportation sector increased its share of employment among men. The prevalence of white-collar workers decreased slightly among women and increased among men. The Role of Individual Characteristics in Explaining the Gender Earnings Gap Table 8.3 decomposes the earnings gap for the three subperiods.2 Each column adds a demographic variable to the set in the previous one. The full set of demographic control variables, in the order included in the matching exercise, is as follows: year, urban area, age, education, presence of children in the household, marital status, and presence of other labour income earner in the household. The first panel in Table 8.3 shows that during 1994–98, men earned 18.3 percent more than women (as a percentage of average women’s earn- ings). Year, city, and age group account for just 0.1 percentage points of average women’s earnings; the rest of the gap remains unexplained. Adding education increases the unexplained earnings gap (Δ0)—the part of the gap attributed to differences between men and women that cannot be explained by observable characteristics that––slightly, reflecting women’s higher education attainment. The component that reflects the fact that men achieve certain combinations of characteristics that women do not (ΔM) reaches almost 2 percent; this percentage remains after the addition of other demographic characteristics. The addition of the other demographic characteristics slightly reduces the unexplained component of the earnings gap and ΔM but increases ΔF, the component that captures the lack of matchable women. 148 Table 8.3 Decomposition of Gender Earnings Gap in Colombia after Controlling for Demographic Characteristics, 1994–2006 (percent) + Presence of + Presence of other Year, metropolitan children in the + Marital household member area, and age + Education household status with labor income 1994–98 Δ 18.3 18.3 18.3 18.3 18.3 Δ0 18.2 19.9 19.4 17.5 18.5 ΔM 0.0 1.8 2.5 1.6 0.7 ΔF 0.0 –0.1 –0.4 0.8 1.7 ΔX 0.1 –3.3 –3.3 –1.6 –2.6 Percentage of men in common support 99.8 97.4 93.0 76.5 57.6 Percentage of women in common support 100.0 99.3 97.5 78.4 68.3 Table 8.3 (continued) + Presence of + Presence of other Year, metropolitan children in the + Marital household member area, and age + Education household status with labor income 2000–01 Δ 13.8 13.8 13.8 13.8 13.8 Δ0 13.0 15.4 15.8 13.8 14.5 ΔM 0.0 2.4 3.8 5.1 4.9 ΔF 0.0 –0.7 –1.2 –2.3 –2.0 ΔX 0.7 –3.4 –4.7 –2.7 –3.6 Percentage of 99.9 96.9 92.3 73.6 55.9 men in common support Percentage of 100.0 98.7 96.1 74.1 61.2 women in common support (continued next page) 149 150 Table 8.3 (continued) + Presence of + Presence of other Year, metropolitan children in the + Marital household member area, and age + Education household status with labor income 2002–06 Δ 13.5 13.5 13.5 13.5 13.5 Δ0 13.9 17.5 17.4 16.1 15.1 ΔM 0.0 1.1 1.5 1.3 1.4 ΔF 0.0 –0.3 –0.7 –1.3 –1.9 ΔX –0.5 –4.7 –4.7 –2.6 –1.0 Percentage of 99.9 97.6 93.6 75.9 57.9 men in common support Percentage of 100.0 98.7 96.3 74.5 61.6 women in common support Source: Based on data from 1994–2006 household surveys conducted by the Colombian National Statistical Agency. Note: ΔM (ΔF) is the part of the earnings gap attributed to the existence of men (women) with combinations of characteristics that are not met by any women (men). ΔX is the part of the earnings gap attributed to differences in the observable characteristics of men and women over the “common support.� Δ0 is the part of the earnings gap that cannot be attributed to differences in characteristics of the individuals. It is typically attributed to a combination of both unobservable characteristics and discrimination. The sum of these components equals the total earnings gap (ΔM + ΔF + ΔX + Δ0 = Δ). the resilient gaps: colombia, 1994–2006 151 The overall gender earnings gap is larger during the first period than in the other two. The pattern remains after controlling for observable individual demographic characteristics. In fact, most of the gap remains unexplained after matching by the whole set of demographics. The second most important element, but an order of magnitude smaller, is the one that exists because women fail to achieve certain combinations of character- istics that men do. These characteristics—that men have, but women do not—are in well-paid segments of the labor market. Investigating the effect of job–related individual characteristics on top of the demographics reported in table 8.3 is not a simple task because of the “curse of dimensionality.�3 In order to leave space for the inclu- sion of job-related characteristics, the analysis ignores some demographic characteristics. Table 8.4 uses the set of demographic matching variables that includes year, city, age, and education and adds the job–related characteristics one by one (as opposed to cumulatively, as done in table 8.3). The new variables considered are small firm (dummy equal to 1 if firm has no more than five workers); economic sector (primary, secondary, or ter- tiary); occupation category; type of employment (self-employed, employer, or employee); formality status (a dummy variable taking the value 1 for people covered by social security obtained from their labor relationship and 0 otherwise); and time commitment (part, full, or overtime). The last column includes all six job-related characteristics on top of the basic set of demographics. The patterns are similar to the patterns shown in table 8.3. Most of the gender earnings gap is left unexplained by these observable characteristics. Furthermore, the unexplained gender gap after controlling for observable characteristics is frequently larger than the observed one. The one-by-one inclusion of job-related characteristics increases the magnitude of the com- ponent of the earnings gap attributable to the existence of men with char- acteristics that are not achieved by women (ΔM). In the most dramatic case (the one obtained after adding type of employment to the demographic characteristics), this lack of “common support� in favor of men explains more than a third of the earnings gap in all three subperiods. For the two later periods, the role of type of employment accounts for about half of the observed gender earnings gap, partly because of the overrepresentation of women as domestic servants. The component attributable to the existence of men with characteristics that are not achieved by women (ΔM) plays a prominent role in explaining the gender gap when controlling for demographic characteristics alone. The component that reflects the existence of women with characteristics that are not achieved by men (ΔF) is just as important when also control- ling for job-related characteristics. This finding implies greater gender segmentation in job-related characteristics, particularly regarding job type and hours worked. Table 8.4 Decomposition of Gender Earnings Gap in Colombia after Controlling for Demographic and Job 152 Characteristics, 1994–2006 (percent) Demographic & Small & Type of & Time set firm & Sector & Occupation employment & Formality worked Full set 1994–98 Δ 18.3 18.3 18.3 18.3 18.3 18.3 18.3 18.3 Δ0 19.9 20.3 19.3 23.4 16.0 20.3 24.0 19.9 ΔM 1.8 3.0 2.9 1.8 7.4 3.1 4.0 1.0 ΔF –0.1 –0.7 –0.3 –0.1 2.3 –0.5 –2.6 2.3 ΔX –3.3 –4.3 –3.6 –6.9 –7.4 –4.6 –7.1 –4.9 Percentage of men in common support 97.4 92.4 90.3 93.2 83.7 93.0 88.9 29.6 Percentage of women in common support 99.3 97.3 96.9 97.3 84.6 97.1 91.1 38.6 (continued next page) Table 8.4 (continued) Demographic & Small & Type of & Time set firm & Sector & Occupation employment & Formality worked Full set 2000–01 Δ 13.8 13.8 13.8 13.8 13.8 13.8 13.8 13.8 Δ0 15.4 15.4 14.6 17.7 12.7 16.6 20.4 20.1 ΔM 2.4 4.1 3.5 2.8 8.8 3.7 5.8 –5.9 ΔF –0.7 –1.9 –1.5 –0.7 0.2 –1.4 –3.5 5.3 ΔX –3.4 –3.8 –2.8 –6.0 –8.0 –5.1 –8.9 –5.7 Percentage of men in common support 96.9 91.8 88.9 91.9 83.0 92.1 86.8 23.9 Percentage of women in common support 98.7 96.1 95.1 95.8 80.0 96.0 88.1 28.7 (continued next page) 153 154 Table 8.4 (continued) Demographic & Small & Type of & Time set firm & Sector & Occupation employment & Formality worked Full set 2002–06 Δ 13.5 13.5 13.5 13.5 13.5 13.5 13.5 13.5 Δ0 17.5 18.0 16.7 19.9 14.5 17.3 21.2 17.9 ΔM 1.1 1.6 1.6 0.6 6.8 1.3 2.8 –7.2 ΔF –0.3 –1.2 –0.7 –0.4 1.8 –0.5 –1.9 10.5 ΔX –4.7 –4.8 –4.0 –6.6 –9.6 –4.5 –8.5 –7.7 Percentage of men in common support 97.6 92.5 89.4 92.6 83.6 93.3 88.5 25.8 Percentage of women in common support 98.7 95.8 95.0 95.8 78.9 95.2 86.4 28.7 Source: Based on data from 1994–2006 household surveys conducted by the Colombian National Statistical Agency. Note: ΔM (ΔF) is the part of the earnings gap attributed to the existence of men (women) with combinations of characteristics that are not met by any women (men). ΔX is the part of the earnings gap attributed to differences in the observable characteristics of men and women over the “common support.� Δ0 is the part of the earnings gap that cannot be attributed to differences in characteristics of the individuals. It is typically attributed to a combination of both unobservable characteristics and discrimination. The sum of these components equals the total earnings gap (ΔM + ΔF + ΔX + Δ0 = Δ). the resilient gaps: colombia, 1994–2006 155 The addition of all job-related characteristics to the basic set of demo- graphics yields a negative ΔM component and a positive ΔF, implying that the presence of barriers to women’s access to certain job profiles works in opposite directions at both extremes of the earnings distribution. The combinations of human capital characteristics that women fail to achieve (that is, the characteristics of the men that are not part of the common support) are not linked to higher earnings than those of matched men. Women who combine human capital characteristics for which there are no comparable men earn less than women without such combinations of characteristics. Who are the women and men in and out of the common support of observable characteristics? The results in table 8.5 indicate that men out of the common support are married, older, and less educated than other men; are self-employed or employers in the secondary sector or blue-collar workers in small and less formal firms; and work more overtime than other men. There is no clear pattern indicating that out-of-support men share human capital characteristics that are better rewarded than those of other men. Women out of the common support are older, less educated, more likely to be separated, and more likely to be domestic servants or self-employed than women in the common support. They tend to work at both extremes (part time and overtime), in smaller firms with less formality, as blue- collar workers in the tertiary sector. Unmatched women thus seem to have combinations of human capital characteristics that are less rewarded than those of women in the common support. Exploring the Unexplained Component of the Gender Earnings Gap Figure 8.1 shows the decomposition of the gender earnings gap after matching on the set of demographic variables, year by year. It illustrates the narrowing of the unexplained gender earnings gap during the period of analysis. However, the reduction is not statistically significant (Hoyos, Ñopo, and Peña 2010). The unexplained gap can be analyzed along different segments of the labor market. Most cities show similar unexplained gender earnings gaps. The only statistically significant differences in unexplained gaps are between Medellín on the one hand and Bucaramanga and Pereira on the other (the gap is smaller in Medellín). Younger people show smaller earnings gaps than people in middle age. The unexplained gaps are highly dispersed among people 55–65 years old. The unexplained gap along education categories is very similar to that of the whole distribution: it is larger among people in the low (incomplete 156 new century, old disparities Table 8.5 Demographic and Job Characteristics of Matched and Unmatched Samples of Men and Women in Colombia’s Labor Force, 2002–06 (percent) Matched women and Unmatched Unmatched men women men Real hourly earnings (1998 Colombian pesos) 1,979 2,337 Personal characteristics Age 18 to 24 20.6 17.0 16.3 25 to 34 39.1 29.2 29.2 35 to 44 27.8 29.0 26.1 45 to 54 10.8 18.0 18.4 55 to 65 1.7 6.9 10.0 Education None or primary incomplete 3.1 13.0 13.4 Primary complete or secondary incomplete 31.3 37.0 41.0 Secondary complete or tertiary incomplete 50.1 35.8 33.3 Tertiary complete 15.5 14.3 12.4 Presence of children in the household 20.9 20.3 25.5 Marital status Cohabiting 18.0 18.6 29.0 Married 26.6 25.6 38.3 Widowed, divorced, or separated 18.5 23.7 6.6 Single (never married) 37.0 32.1 26.2 Presence of other household member with labor income 76.2 74.5 63.2 Job characteristics Type of employment Employer 0.5 4.0 8.3 Self-employed 19.1 28.7 32.9 (continued next page) the resilient gaps: colombia, 1994–2006 157 Table 8.5 (continued) Matched women and Unmatched Unmatched men women men Private employee 72.5 42.1 51.9 Public employee 7.3 7.8 6.6 Domestic servants 0.6 17.4 0.4 Time worked Part time 6.6 25.6 9.5 Full time 71.3 43.8 42.2 Overtime 22.1 30.6 48.3 Small firm 26.3 61.8 55.6 Formality 73.3 47.2 48.0 Economic sector Primary 0.1 1.1 2.6 Secondary 26.0 19.2 33.6 Tertiary 73.9 79.7 63.8 Occupation White collar 60.6 46.7 37.1 Blue collar 39.4 53.3 62.9 Source: Based on data from 2002–06 household surveys conducted by the Colom- bian National Statistical Agency. secondary) and high (complete tertiary) education groups and smaller for people with intermediate education (complete secondary or incomplete tertiary). The unexplained gaps are also smaller among widows, public employees, full-time workers, workers in construction and transporta- tion, white-collar workers, workers at larger firms, and formal sector workers. The unexplained gender earnings gaps are smaller than average for middle-income earners, larger than average at both extremes of the earn- ings distribution, and slightly larger than average at the bottom of the earnings distribution (figure 8.2). What generates the observed U–shape in both the observed and unexplained gender earnings gaps? The mini- mum earnings may be behind the lower levels of the unexplained gender earnings gap in the middle of the distribution. Because people at the middle of the distribution are close to the minimum earnings (in the sample, 52 percent of men and 58 percent of women earn earnings less than or equal to the minimum wage), the minimum earnings may exert a gender–equalizing effect on intermediate–paying jobs. The “bite� of the 158 new century, old disparities Figure 8.1 Decomposition of Gender Earnings Gap in Colombia after Controlling for Demographic and Job Characteristics, 1994–2006 1994 (Δ = 17.2%) 1996 (Δ = 18.4%) 1998 (Δ = 19.8%) 2000 (Δ = 13.3%) 2001 (Δ = 14.1%) 2002 (Δ = 14.8%) 2003 (Δ = 15.5%) 2004 (Δ = 14.4%) 2005 (Δ = 13.4%) 2006 (Δ = 10.6%) –10 0 10 20 30 percentage of average women’s earnings Δ0 ΔM ΔF ΔX Source: Based on data from 1994–2006 household surveys conducted by the Colombian National Statistical Agency. Note: ΔM (ΔF) is the part of the earnings gap attributed to the existence of men (women) with combinations of characteristics that are not met by any women (men). ΔX is the part of the earnings gap attributed to differences in the observable characteristics of men and women over the “common support.� Δ0 is the part of the earnings gap that cannot be attributed to differences in characteristics of the individuals. It is typically attributed to a combination of both unobservable characteristics and discrimination. The sum of these factors equals the total earnings gap (ΔM + ΔF + ΔX + Δ0 = Δ). minimum earnings varies along the income distribution. It barely affects the earnings of people earning less than the minimum, usually informal workers; is very binding at and around the level of the minimum wage; and loses importance as one moves up the income distribution curve toward high earners (Cunningham 2007). After controlling for the demographic set of observable characteristics, the unexplained gap is slightly larger than when matching on a smaller set of characteristics, especially at the upper end of the earnings distribution. After controlling for the full set of demographic and job characteristics, the situation is similar: the unexplained gaps above the median of the the resilient gaps: colombia, 1994–2006 159 Figure 8.2 Gender Earnings Gaps in Colombia after Controlling for Demographic and Job Characteristics, by Percentiles of Earnings Distribution, 1994–2006 70 percentage of average women’s earnings 60 50 40 30 20 10 0 0 10 20 30 40 50 60 70 80 90 100 earnings percentile original gap after controlling for demographic characteristics after controlling for demographic and job characteristics Source: Based on data from household surveys for 1994–2006 conducted by the Colombian National Statistical Agency. earnings distributions are larger than they are for the smaller set of char- acteristics. The novelty arises below the median of the earnings distribu- tions. There, the unexplained gaps are substantially smaller than the gaps observed with the other sets of matching characteristics. Thus, observable characteristics do a better job of explaining gender earnings differentials at the lower end of the earnings distribution. The results presented in this chapter suggest that the gender earnings gap in Colombia is unexplained largely by differences in observable char- acteristics, both demographic and job related. The gap that remains unex- plained after accounting for these differences displays a U-shape with respect to earnings: it is smaller for middle-income individuals and larger at both extremes of the earnings distribution. This shape may reflect the gender-equalizing effect of the minimum wage. The largest unexplained earnings gaps are found among less educated people and people who work part time, in the primary sector and enter- tainment or household services, at small firms or in the informal sec- tor, and as domestic or blue-collar workers. Among people with these 160 new century, old disparities characteristics, two distinct profiles are evident. One consists of low- productivity individuals, the other comprises women who, in need of flexibility to participate in the labor market, have to work under arrange- ments of precarious attachment to the market. Some women seem to be confined to combinations of human capital characteristics that are less well rewarded than those of the rest of the labor force. Policy implications regarding the potential effectiveness—or ineffectiveness—of different measures to narrow the gender earnings gaps can be derived from these results. First, the gender earnings gap may reflect discrimination. Some observers argue that discrimination will decrease over time on its own, as society becomes accustomed to women in the working force. The high participation rates of women in Colombia and the fact that the gender earnings gap has changed little in the last decade suggest that this channel may not be effective. Like Colombia, Brazil has implemented policies to reduce inequali- ties. It has addressed both gender inequality and inequalities faced by its large Afro–descendant population. Chapter 9 examines the evolution of the gender earnings gaps in Brazil. Notes 1. Statistical discrimination is a theory of why women or minorities are paid lower earnings. It occurs when rational agents use aggregate group characteristics to evaluate individual characteristics, which leads agents belonging to different groups to be treated differently. If, for example, firms believe that women of child- bearing age are more likely to have babies, and therefore have breaks during their careers, than older women, they would pay such women less, to account for the higher probability of losing them. 2. For a description of the methodology used in this chapter, see chapter 2. 3. The curse of dimensionality refers to the fact that the likelihood of finding female–male matches decreases as the number of control variables (the “dimen- sion�) increases. This is a problem because researchers would like to use the maxi- mum number of observable characteristics in order to control the scope of the role of unobservable factors in explaining the earnings gap. References Abadía, L. K. 2005. “Discriminación salarial por sexo en Colombia: un análisis desde la discriminación estadística.� Documentos de Economía 17, Pontificia Universidad Javeriana, Bogotá. Amador, D., and R. Bernal. 2009. “Marriage vs. Cohabitation: The Effects on Children’s Well–Being.� Universidad de los Andes, Bogotá. Angel–Urdinola, D., and Q. Wodon. 2006. “The Gender Wage Gap and Poverty in Colombia.� Labour 20 (4): 721–39. the resilient gaps: colombia, 1994–2006 161 Badel, A., and X. Peña. 2009 “Decomposing the Gender Gap with Sample Selec- tion Adjustment: Evidence from Colombia.� Universidad de Los Andes, Bogotá. http://ximena.pena.googlepages.com/gender.pdf. Bernat, L. F. 2007. “¿Quiénes son las mujeres discriminadas? enfoque distributivo de las diferencias salariales por género� Borradores de Economía y Finanzas 13, Universidad Icesi, Cali, Colombia. Cunningham, W. 2007. Minimum Wages and Social Policy: Lessons from Develop- ing Countries. Washington, DC: World Bank. Fernández, P. 2006. “Determinantes del diferencial salarial por género en Colombia, 1997–2003.� Desarrollo y Sociedad 58 (2): 165–208. Hoyos, A., H. Ñopo, and Ximena Peña. 2010. “The Persistent Gender Earnings Gap in Colombia, 1994–2006.� IZA Discussion Paper 5073, Institute for the Study of Labor, Bonn, Germany. Sabogal, A. 2009. “Brecha salarial por género y ciclo económico en Colombia.� Universidad de los Andes, Bogotá. Tenjo, J. 1993. “1976–1989: cambios en los diferenciales salariales entre hombres y mujeres.� Planeación y Desarrollo 24: 103–16. Tenjo, J., R. Ribero, and L. Bernat. 2006. “Evolución de las diferencias salariales de género en seis países de América Latina.� In Mujeres y trabajo en América Latina, ed. C. Piras, 149–98. Washington, DC: Inter–American Development Bank. 9 Promoting Equality in the Country with the Largest Earnings Gaps in the Region: Brazil 1996–2006 Promoting gender and racial equality has been one of Brazil’s major challenges in recent years. Some observers believe that this challenge has begun to be met; others believe that the work of implementing effective policies has just started. Disentangling group inequalities in Brazil will help researchers inform public policies. This chapter analyzes the composition and evolution of gender earnings differentials over a decade (1996–2006), using the National Household Sample Survey (PNAD) conducted by the Brazilian Institute of Geography and Statistics and the matching comparison methodology described in chapter 2. The analysis is restricted to workers 15–65 years recording nonzero earnings. The variable of analysis is hourly earnings at the pri- mary occupation. What Does the Literature Show? Camargo and Serrano (1983) investigate gender pay differentials, specifying earnings equations using not only personal characteristics, such as level of This chapter was adapted from “Gender and Racial Wage Gaps in Brazil 1996–2006: Evidence Using a Matching Comparisons Approach,� Luana Marquez Garcia, Hugo Ñopo, and Paola Salardi, RES Working Paper 4626, Inter-American Development Bank, 2009. Luana Marques Garcia is a young professional at the Inter-American Devel- opment Bank. Paola Salardi is a research fellow in the Economics Group at the University of Sussex, in Brighton, United Kingdom. 163 164 new century, old disparities education, but also aspects of sectoral features, such as concentration, capital intensity, and size. Their findings suggest that the structure of economic sec- tors plays a negligible role in the determination of women’s earnings. One of the first studies to explore gender pay gaps using the Blinder- Oaxaca decomposition is Birdsall and Fox (1985). Extracting a 1 percent sample from the 1970 Brazilian census focused on a specific occupational category (school teachers), they find that the explained component of the gap is greater than the unexplained component. As 74 percent of the earnings gap can be explained, the authors claim that job discrimination (a proxy measured by the unexplained component) does not represent the main source of gender earnings differentials for school teachers. Stelcner et al. (1992) examine gender differentials in earnings using the 1980 census by correcting the earnings equation estimations for selection bias. They find that unexplained components are larger than the total earnings differential and that a negative explained component reflects women’s better endowments (such as education). Exploring differences in the formal and informal labor market, Tiefen- thaler (1992) finds that gender earnings differentials tend to be larger in the formal sector. The unexplained component dominates in the formal sector, whereas the explained component dominates in the informal sector. This finding is supported by evidence that better educated women tend to work in formal occupations.1 Barros, Ramos, and Santos (1995) investigate the role played by educa- tion and occupational structure in the evolution of gender differentials. In addition to confirming previous results on the effect of education on gen- der pay gaps, they provide evidence for the “glass ceiling� phenomenon, which prevents women from reaching managerial positions. Ometto, Hoffmann, and Alves (1999) use the Blinder-Oaxaca decom- position technique as revised by Brown, Moon, and Zoloth (1980), which isolates the extent of gender pay gaps caused by interoccupation and intraoccupation differentials. They find that gender earnings gaps in Pernambuco are mainly the result of intraoccupational differentials. In contrast, in wealthier São Paulo, both kinds of differentials play a role. Leme and Wajnman (2000) confirm findings of previous studies that education cannot explain gender pay gaps in Brazil. Returns to education favor women; gender earnings gaps thus reflect the unex- plained component, not endowment differences. They find that returns to education are more favorable to women born after the 1950s, a find- ing compatible with improvements in women’s educational attainment over time. Arabsheibani, Carneiro, and Henley (2003) show that gender differ- entials in earnings decreased markedly over time, mainly because of the decline in the explained component. Women’s endowments, particularly educational achievement, have had an important effect. gaps vis-à-vis equality: brazil, 1996–2006 165 Loureiro, Carneiro, and Sachshida (2004) find larger earnings gaps in urban areas than in rural areas. When the Blinder-Oaxaca decomposition is used, unexplained components generally dominate gender differentials. These findings do not hold, however, once the sample is restricted to a more homogenous occupational group, such as school teachers (Birdsall and Fox 1985). Although gender earnings gaps have shrunk over time, the unexplained component has tended to increase (Arabsheibani, Carneiro, and Henley 2003). The Role of Individual Characteristics in Explaining the Earnings Gap Table 9.1 presents the average characteristics of men and women who were either matched or not matched based on their individual characteristics.2 The matching was done based on six combinations of human capital and labor market characteristics. The first set includes only the number of years of schooling. The second set adds age and education, the third adds region,3 the fourth adds occupation, the fifth adds sector, and the sixth adds a vari- able that identifies whether the individual works in the formal sector. The sequence in which extra variables were added to the set of controlling char- acteristics was chosen so that it leaves to the last sets variables that may end up being endogenous in a model of earnings determination à la Mincer (a pricing equation or hedonic earnings function revealing how the labor mar- ket rewards productive attributes such as schooling and work experience). There are significant differences in characteristics of men and women that are and are not matched. The age patterns are similar, although unmatched individuals are likely to be older. Unmatched women are on average better educated than unmatched men over time. In 1996, 9.2 percent of unmatched women completed more than 15 years of education, compared with 6.2 percent of unmatched men; in 2006 these percent- ages increased to 16.6 percent for unmatched women and 7.6 percent for unmatched men. Unmatched men are more likely to be nonwhite and to live in rural areas. The regional distribution of matched and unmatched individuals does not differ, with the South-East and the North-East showing the high- est densities. Labor characteristics reveal interesting differences by gender: in 1996, 14.0 percent of unmatched women worked as professionals and 77.3 percent worked at the intermediate level. In contrast, only 5.2 percent of unmatched men were professionals, and 67.5 percent were blue-collar workers. Over time, the number of unmatched individuals working as professionals increased, to 22.7 percent for women and 17.5 percent for men. In addition, unmatched men were more likely to be employed in the informal sector and concentrated in economic activities such as agriculture 166 Table 9.1 Demographic and Job Characteristics of Matched and Unmatched Samples of Men and Women in Brazil’s Labor Force, 1996 and 2006 (percent) 1996 2006 Unmatched Unmatched Matched Unmatched Matched Matched Characteristics women men women and men women men women and men Personal characteristics Age 15–24 28.2 26.6 27.2 19.7 22.4 25.9 25–34 27.5 27.3 31.0 26.8 26.7 29.2 35–44 24.9 23.4 24.9 26.4 24.9 24.8 45–54 13.6 15.0 12.5 19.1 17.7 14.7 55–65 5.8 7.8 4.4 8.0 8.3 5.3 Years of education Less than 4 28.4 33.7 27.9 19.5 25.7 19.4 4–10 59.2 58.9 59.8 58.6 64.8 60.4 11–15 3.3 1.3 1.2 5.4 1.9 2.5 More than 15 9.2 6.2 11.1 16.6 7.6 17.7 Ethnicity (white) 54.3 52.9 55.7 51.3 44.8 49.3 Urban 92.1 84.6 84.9 93.4 85.7 87.6 Table 9.1 (continued) 1996 2006 Unmatched Unmatched Matched Unmatched Matched Matched Characteristics women men women and men women men women and men Regions North 10.9 9.0 4.1 14.9 15.3 10.3 North-East 22.4 23.0 32.1 21.3 25.2 31.5 South-East 29.9 33.7 40.1 25.9 28.2 33.9 South 21.4 19.7 16.8 21.5 16.5 15.4 Central-West 15.4 14.5 7.0 16.3 14.8 8.9 Job characteristics Formal job 45.3 44.9 50.9 42.6 43.9 52.7 Occupation Professional 14.1 5.2 14.4 22.7 17.5 23.7 Intermediate 77.3 27.3 49.6 69.2 19.0 51.5 Blue collar 8.6 67.5 36.0 8.1 63.6 25.0 Agriculture 0.7 15.6 13.0 1.0 13.7 10.2 Construction 0.6 19.8 0.3 1.2 21.8 0.2 Social services 71.0 20.2 46.5 55.4 13.0 45.4 Source: Based on data from 1996 and 2006 PNAD. 167 168 new century, old disparities and construction. Among unmatched women, 71.0 percent were employed in social services. Figure 9.1 reports the decomposition of gender earnings gaps using the full set of characteristics.4 The total gap shrinks by 13 percent, from 52 percent in 1996 to 39 percent in 2006. The dominance of the unexplained component is striking: the main portion of the gender earnings gap (Δ) is unexplained even when the full set of characteristics is included as con- trols. In fact, the part of the gap that cannot be attributed to differences in characteristics of the individuals (Δ0) is much higher than the total earnings gap. The explained component (ΔX)—attributed to differences in observable characteristics—is always negative for gender earnings dif- ferentials. This negative sign is explained by women’s better endowments, particularly in terms of educational achievement. Although the total gender earnings gap decreased over time, the change resulted mainly from the decrease in explained differences rather than Figure 9.1 Decomposition of Gender Earnings Gaps in Brazil after Controlling for Demographic and Job Characteristics, 1996–2006 100 percentage of average women’s earnings 80 60 40 20 10 −20 −40 1996 1997 1998 1999 2001 2002 2003 2004 2005 2006 ΔX ΔF ΔM Δ0 Source: Based on data from 1996–2006 PNAD. Note: ΔM (ΔF) is the part of the earnings gap attributed to the existence of men (women) with combinations of characteristics that are not met by any women (men). ΔX is the part of the earnings gap attributed to differences in the observable characteristics of men and women over the “common support.� Δ0 is the part of the earnings gap that cannot be attributed to differences in characteristics of the individuals. It is typically attributed to a combination of both unobservable characteristics and discrimination. The sum of these components equals the total earnings gap (ΔM + ΔF + ΔX + Δ0 = Δ). gaps vis-à-vis equality: brazil, 1996–2006 169 a drop in the unexplained component. The portion of the earnings gap attributable to unmatched individuals is negligible. In particular, the small size of ΔM—the part of the earnings gap caused by the existence of men with combinations of characteristics that are not met by any woman— highlights the limited extent of men’s advantage. Exploring the Unexplained Component of the Gender Earnings Gap Table 9.2 reports gender earnings gaps by characteristic, considering only the first year (1996) and last year (2006) of the period under study.5 Earnings gaps increased with age, becoming larger at higher levels of education and for top job positions. The gap for the youngest cohort was much smaller than for other age cohorts. This finding may be explained by the fact that many young people are still in school. In the construction sector, women tend to earn higher earnings than men. The unexplained component is greater than the total earnings gap for most subgroups considered, as it is for the whole sample. For higher levels of education and job position, Δ0 is smaller than the total dif- ferential. In these cases, the number of people out of support tends to be greater, and the earnings gap is explained largely by differences in characteristics in and out of support. Gender earnings gaps are larger among whites than nonwhites, and they are larger in urban regions than in national averages. Geographically, the gaps are higher in the South and South-East. The analysis is enriched by considering unexplained earnings differentials in individual income. Earnings are rescaled such that average women’s earnings are normalized to 100 in each year. This change neu- tralizes nominal changes in earnings, so that real changes in the gaps are evident. At each percentile of the earnings distribution, the earnings of the representative men and women in each distribution are compared and the earnings gap computed. Figure 9.2 reports the entire distribution for both total and unexplained gender earnings gaps, after controlling for the richer set of observable characteristics. The gender earnings gap, particularly the unexplained gap, displays a U-shape along the earnings distribution. The unexplained gap tends to be larger at the bottom of the distribution: low- earning women suffer larger differentials. Observable individual characteristics cannot completely account for gender earnings gaps in Brazil. Unexplained gender earnings gaps increase with workers’ age and education; they are larger among professionals and among people living in the South-East. The unexplained gender earnings gap is highest among the poor, lowest among middle-income earners, and higher among those with high income. 170 new century, old disparities Table 9.2 Original and Unexplained Gender Earnings Gap in Brazil, by Demographic and Job Characteristics, 1996 and 2006 (percent) 1996 2006 Δ Δ0 Δ Δ0 Personal characteristics Age 15–24 15.3 22.3 11.0 15.6 25–34 44.4 66.2 30.4 45.1 35–44 72.0 81.5 50.4 66.6 45–54 96.6 88.4 66.9 82.5 55–65 70.8 48.2 68.2 69.0 Years of education Less than 4 27.4 23.2 22.0 18.9 4–10 56.0 44.8 39.1 28.8 11–15 141.7 129.9 118.3 68.3 More than 15 277.0 149.4 207.9 140.3 White 71.2 72.4 57.6 63.5 Urban 63.0 63.5 47.3 52.4 Regions North 37.3 50.3 28.7 43.9 North-East 31.7 44.6 20.6 35.8 South 63.9 70.3 53.2 60.1 South-East 64.9 67.9 54.7 53.8 Central-West 49.0 64.2 41.7 69.3 Job characteristics Formal job 41.6 62.3 29.7 54.0 Type of occupation Professionals 202.8 97.6 120.0 109.5 Intermediate 133.3 55.9 32.4 27.7 Blue collar 40.1 43.6 31.6 33.6 Agriculture 24.5 18.3 24.0 21.4 Construction –47.0 31.1 –113.0 –145.9 (continued next page) gaps vis-à-vis equality: brazil, 1996–2006 171 Table 9.2 (continued) 1996 2006 Δ Δ0 Δ Δ0 Social services 95.9 65.6 95.4 58.4 Total 52.2 60.0 39.1 49.8 Source: Based on data from 1996–2006 PNAD. Note: Δ is the total earnings gap. Δ0 is the part of the earnings gap that cannot be attrib- uted to differences in characteristics of the individuals. It is typically attributed to a combi- nation of both unobservable characteristics and discrimination. Figure 9.2 Original and Unexplained Gender Earnings Gap in Brazil, by Percentiles of Earnings Distribution, 1996–2006 a. Original gender earnings gap percentage of average women’s earnings 120 100 80 60 40 20 0 0 10 20 30 40 50 60 70 80 90 100 percentile of earnings distribution b. Unexplained component of gender earnings gap after controlling for demographic and job characteristics percentage of average women’s earnings 120 100 80 60 40 20 0 0 10 20 30 40 50 60 70 80 90 100 percentile of earnings distribution Source: Based on data from 1996–2006 PNAD. 172 new century, old disparities Brazil has a large Afro-descendant population, which faces inequalities that may be comparable to the inequalities faced by women. This issue is addressed in chapter 14. Notes 1. Kassouf (1997, 1998) and Silva and Kassouf (2000) correct the earnings equation estimation for participation in the formal and informal labor market sectors. 2. For a description of the methodology used in this chapter, see chapter 2. 3. The regions are North (Rondônia, Acre, Amazonas, Roraima, Parà, Amapà, Tocantins); North-East (Maranhão, Piauì, Cearà, Rio Grande do Norte, Paraiba, Pernambuco, Alagoas, Sergipe, Bahia); South-East (Minas Gerais, Espìrito Santo, Rio de Janeiro, São Paulo); South (Paraná, Santa Catarina, Rio Grande do Sul); and Central-West (Mato Grasso do Sul, Mato Grosso, Goiás, Distrito Federal). 4. For graphs reporting different sets of controls, see Garcia, Ñopo, and Salardi (2009). The results are qualitatively similar to those reported here. 5. Only results for the first and last year are reported, because the trend over the decade is fairly stable and smoothly decreasing. For all subsamples of popula- tion, both explained and unexplained earnings gaps decrease over time. References Arabsheibani, G. R., F. G. Carneiro, and A. Henley. 2003. “Gender Wage Differ- entials in Brazil: Trends over a Turbulent Era.� Policy Research Working Paper 3148, World Bank, Washington, DC. Barros, R., L. Ramos, and E. Santos. 1995. “Gender Differences in Brazilian Labor Markets.� In Investment in Women’s Human Capital, ed. T. P. Schultz, 345–79. Chicago: University of Chicago Press. Birdsall, N., and M. L. Fox. 1985. “Why Males Earn More: Location and Train- ing of Brazilian Schoolteachers.� Economic Development and Cultural Change 33 (3): 533–56. Brown, R. S., M. Moon, and B. S. Zoloth. 1980. “Incorporating Occupational Attainment in Studies of Male-Female Earnings Differentials.� Journal of Human Resources 15 (1) 3–2. Camargo, J. M., and F. Serrano. 1983. “Os dois mercados: homens e mulheres na indústria Brasileira.� Revista Brasileira de Economía 34: 435–48. García, L. M., H. Ñopo, and P. Salardi. 2009. “Gender and Racial Wage Gaps in Brazil 1996–2006: Evidence Using a Matching Comparisons Approach.� RES Working Paper 4626, Inter-American Development Bank, Research Depart- ment, Washington, DC. Kassouf, A. L. 1997. “Retornos à escolaridade e ao treinamento nos setores urbano e rural do Brasil.� Revista de Economia e Sociologia Rural 35 (2): 59–76. ––––. 1998. “Wage Gender Discrimination and Segmentation in the Brazilian Labour Market.� Brazilian Journal of Applied Economics 2 (2): 243–69. Leme, M. C., and S. Wajnman. 2000. “Tendencias de coorte nos diferenciais de rendimentospor sexo.� In Desigualdade e pobreza no Brasil, org. R. Henriques, 251–70. Instituto de Pesquisa Econômica Aplicada, Rio de Janeiro. gaps vis-à-vis equality: brazil, 1996–2006 173 Loureiro, P. R. A., F. G. Carneiro, and A. Sachsida. 2004. “Race and Gender Discrimination in the Labor Market: An Urban and Rural Sector Analysis for Brazil.� Journal of Economic Studies 31 (2): 129–43. Ometto, A. M. H., R. Hoffmann and M. C. Alves. 1999. “Participação da mulher no mercado de trabalho: discriminação em Pernambuco e São Paulo.� Revista Brasileira de Economia 53 (3): 287–322. Silva, N. D. V., and A. L. Kassouf. 2000. “Mercados de trabalho formal e informal: uma analise da discriminação e da segmentação.� Nova Economía (1): 41–47. Stelcner, M., J. B. Smith, J. A. Breslaw, and G. Monette. 1992. “Labor Force Behavior and Earnings of Brazilian Women and Men, 1980.� Case Studies of Women’s Employment and Pay in Latin America, Vol. 2 of ed. G. Psacharopoulos and T. Zatiris, 39–88. Washington, DC: World Bank. Tiefenthaler, J. 1992. “Female Labor Force Participation and Wage Determina- tion in Brazil 1989.� Case Studies of Women’s Employment and Pay in Latin America, Vol. 2 of ed. G. Psacharopoulos and T. Zatiris, 89–118. Washington, DC: World Bank. 10 Gender Earnings Gaps in a Country with a Large Indigenous Population: Ecuador 2003–07 Ecuador has made important advances in reducing gender disparities and addressing gender-related development issues. The country’s gender disparities in education and labor force participation have continued to close. Women’s labor force participation has steadily increased since the 1980s, and women have made significant advances in professional, managerial, and technical fields (Correia and Van Bronkhorst 2000.) In rural areas, women continue to play an important role in subsistence farming and commercial agriculture. However, gender disparities in edu- cational and employment opportunities are still significant, particularly among indigenous people. This chapter analyzes the gender earnings gap in Ecuador, using data from the Survey on Employment, Unemployment, and Underemployment (Encuesta de Empleo, Desempleo, y Subempleo [ENEMDU]), conducted annually by the Instituto Nacional de Estadísticas y Censos de Ecuador (National Institute of Statistics and Census of Ecuador, INEC). The sample studied includes 15 to 65-year-old employers, employees, and the self- employed reporting positive earnings (measured as hourly earnings) who lived in the coastal, highland, and Amazon regions of Ecuador. (Chapter 15 examines the indigenous earnings gap in Ecuador.) This chapter was adapted from “Ethnic and Gender Wage Gaps in Ecuador,� Lourdes Gallardo and Hugo Ñopo, RES Working Paper 4625, Inter-American Development Bank, 2009. Lourdes Gallardo is an investment officer at the Inter-American Development Bank. 175 176 new century, old disparities What Does the Literature Show? Correia and Van Bronkhorst (2000) document that Ecuador’s dispari- ties in educational and labor force participation have continued to close. García-Aracil and Winter (2006) document that endowments account for slightly less than half of the total earnings differentials between men and women in Ecuador. This means that more than half of the earnings dis- parity is unexplained by observable human capital characteristics. García- Aracil and Winter conclude that equalizing educational opportunities for girls would only marginally reduce gender earnings differentials. However, in the case of indigenous women, equalizing educational opportunities would be important in reducing the earning differential with other groups (other studies, focused on ethnic minorities, are addressed in chapter 15). How Do Male and Female Workers Differ? Table 10.1 reports educational completion rates for men and women. On average, women’s educational attainment slightly surpasses that of men. In addition, larger percentages of women have both higher education and no education. Gender differences did not change much during the period of analysis. Table 10.2 presents average hourly earnings for indigenous and non- indigenous men and women between 2003 and 2007. It shows that the gender earnings gap for 2007 (7.4 percent) is much smaller than the indig- enous earnings gap (44.9 percent). Table 10.1 Educational Attainment by Men and Women in Ecuador’s Labor Force, 2003 and 2007 (percent) Level of 2003 2007 education Men Women Men Women None 5.3 7.8 4.1 6.1 Pre-school 0.3 0.3 0.3 0.4 Basic 52.2 48.3 53.3 49.9 Bachilleratoa 28.7 28.8 27.5 27.6 Tertiary 13.5 14.7 14.8 16.0 Total 100 100 100 100 Source: Based on data from 2003–07 ENEMDU. a. Equivalent to last three years of high school. gaps and indigenous populations: ecuador, 2003–07 177 Table 10.2 Average Hourly Earnings for Indigenous and Nonindigenous Men and Women in Ecuador, 2003–07 (current U.S. dollars) Gender 2003 2004 2005 2006 2007 Women 1.0 1.1 1.1 1.3 1.4 Men 1.1 1.2 1.2 1.4 1.5 Gender earnings gap (percent) 7.1 11.2 7.8 9.2 7.4 Ethnicity 2003 2004 2005 2006 2007 Ethnic minorities 0.8 0.8 0.9 0.9 1.0 Nonminorities 1.1 1.2 1.3 1.4 1.5 Ethnic earnings gap (percent) 44.9 48.7 45.4 48.2 44.9 Source: Based on data from 2003–07 ENEMDU. The Role of Individual Characteristics in Explaining the Gender Earnings Gap Men and women in the sample were matched on four combinations of human capital characteristics.1 The first combination includes area (rural or urban), education, ethnicity, and age. The second adds a dummy vari- able that identifies whether the respondent is the head of household. The third adds occupation (coded at the one-digit classification). The fourth adds a variable that reports whether the respondent’s income is comple- mented by remittances from abroad. Figure 10.1 presents the results of the decomposition. Gender earn- ings differentials range from 7.1 percent in 2003 to 11.2 percent in 2007. The contribution of the endowment of productive characteristics to the total earnings gap, ΔX, is negative, indicating that despite having a bet- ter endowment of human capital characteristics, women earn less than men. The component of the earnings gap attributed to the existence of men with observable characteristics that were not met by any woman (ΔM) was small over the whole period but slightly higher in 2007 than in 2003. This result may suggest the existence of a glass ceiling effect, as there are men with combinations of observable characteristics for whom there are no comparable women and these men earn earnings that are, on average, higher than the earnings of the rest of the population. 178 new century, old disparities Figure 10.1 Decomposition of Gender Earnings Gap in Ecuador, 2003–07 a. Controlling for area, education, ethnicity, and age 30 25 percentage of average women’s 20 15 10 earnings 5 0 −5 −10 −15 −20 2003 2004 2005 2006 2007 b. Controlling for area, education, ethnicity, age, and head of household 30 25 percentage of average women’s 20 15 10 earnings 5 0 −5 −10 −15 −20 2003 2004 2005 2006 2007 c. Controlling for area, education, ethnicity, age, head of household, and occupation 30 percentage of average women’s 25 20 15 earnings 10 5 0 −5 −10 −15 −20 2003 2004 2005 2006 2007 Δ0 ΔF ΔM ΔX (continued next page) gaps and indigenous populations: ecuador, 2003–07 179 Figure 10.1 (continued) d. Controlling for area, education, ethnicity, age, head of household, occupation, and remittances percentage of average women’s 30 25 20 15 earnings 10 5 0 −5 −10 −15 −20 2003 2004 2005 2006 2007 Δ0 ΔF ΔM ΔX Source: Based on data from 2003–07 ENEMDU. Note: ΔM (ΔF) is the part of the earnings gap attributed to the existence of men (women) with combinations of characteristics that are not met by any women (men). ΔX is the part of the earnings gap attributed to differences in the observable characteristics of men and women over the “common support.� Δ0 is the part of the earnings gap that cannot be attributed to differences in characteristics of the individuals. It is typically attributed to a combination of both unobservable characteristics and discrimination. The sum of these components equals the total earnings gap (ΔM + ΔF + ΔX + Δ0 = Δ). In 2006, the component of the earnings gap attributed to the existence of women with observable characteristics that were not met by any men (ΔF) accounted for a larger proportion of the earnings differential than ΔM. This finding suggests the existence of a large “maid effect�—that is, the presence of many indigenous women in the segments of the labor markets that work as maids. This contrasts with the “chief executive officer (CEO) effect,� which refers to the fact that men and not woman tend to be CEOs. A large maid effect indicates that on average, women’s earnings are lower than the earnings of the rest of the population. Exploring the Unexplained Component of the Gender Earnings Gap All combinations of human capital characteristics used in the matching exercise show that the unexplained component of the gap accounts for 180 new century, old disparities most of the earning differential between men and women. Figure 10.2 shows the distribution of the unexplained component for different per- centiles of the earnings distribution for women and men. The unexplained component is larger at the lower end of the income distribution. Introduc- ing the head of household control into the matching reduces the unex- plained component by more than half. This effect is particularly strong between the 80th and 90th percentile of the income distribution, where being the head of household somewhat eliminates the unexplained compo- nent. At the low end of the income distribution, the occupational variable has a significant effect on reducing the unexplained component. Occupa- tional sorting thus plays an important role in determining gender earnings gaps among lower-income workers, whereas heading a household matters more for higher-income workers. Different policy approaches are needed to combat gender disparities in labor markets for different segments of the earnings distribution. As in other countries, observable differences between men and women do not explain gender earnings gaps in Ecuador, suggesting that gen- der inequalities in labor markets there cannot be reduced through poli- cies that improve human capital endowments for women. Instead, action must be oriented toward changing practices that may discriminate against women. Figure 10.2 Unexplained Gender Earnings Gap in Ecuador after Controlling for Demographic and Job Characteristics, by Percentile of Earnings Distribution, 2003–07 60 percentage of average 50 women’s earnings 40 30 20 10 0 0 10 20 30 40 50 60 70 80 90 100 percentile of earnings distribution area, education, ethnicity, and age (set 1) set 1, household set 1, household and occupation set 1, household, occupation, and remittances Source: Based on data from 2003–07 ENEMDU. gaps and indigenous populations: ecuador, 2003–07 181 Ecuador has a large indegenous population, which faces high inequalities in the labor markets. This issue is addressed in chapter 15. Note 1. For a description of the methodology used in this chapter, see chapter 2. References Correia, M., and B. Van Bronkhorst. 2000. Ecuador: Gender Review: Issues and Recommendations. Washington, DC: World Bank. Gallardo, L., and H. Ñopo. 2009. “Ethnic and Gender Wage Gaps in Ecua- dor.� RES Working Paper 4625, Inter-American Development Bank, Research Department, Washington, DC. García-Aracil, A., and C. Winter. 2006. “Gender and Ethnicity Differentials in School Attainment and Labor Market Earnings in Ecuador.� World Develop- ment 34 (2): 289–307. 11 Gender Earnings Gaps in Central American Countries, 1997–2006 Central America has a relatively young labor force (29 percent under 25), in which women are underrepresented (38 percent of the labor force). The average unemployment rate in Central American countries was 4.3 percent in 2008, 4.8 percent for women and 4.1 percent for men. Almost two-fifths of the economically active labor force lives in rural areas, where the unemployment rate was 3.1 percent (the rate in urban areas was 5.1 percent). Educational achievement is low, with 39 percent of the labor force not having completed primary education and 58 percent having no more than a primary education. This chapter presents a general picture of Central America, using a pooled database for four countries: Costa Rica, El Salvador, Honduras, and Nicara- gua. The pooled dataset includes data for three points in time: the mid-1990s, the early 2000s, and the mid-2000s. The countries are then analyzed individ- ually using the same surveys and years as in the pooled dataset (Enamorado, This chapter was adapted from the following sources: “Gender Wage Gaps in Central American Countries: Evidence from a Non-Parametric Approach,� Ted Enamorado, Ana Carolina Izaguirre, and Hugo Ñopo, RES Working Paper 4639, Inter-American Development Bank, 2009; “Gender and Ethnic Wage Gaps in Guatemala from a Matching Comparisons Perspective,� Hugo Ñopo and Alberto Gonzales, RES Working Paper 4587, Inter-American Development Bank, 2008; and Hugo Ñopo and Alberto Gonzales, “Brechas salariales por género y etnicidad,� in Más crecimiento, más equidad, ed. Ernesto Stein, Osmel Manzano, Hector Morena, and Fernando Straface, Banco Interamericano de Desarrollo, 265–98, 2009. Ted Enamorado is a PhD student in the department of political science at Vanderbilt University, in Nashville, TN. Ana Carolina Izaguirre is a researcher at the Inter-American Development Bank. Alberto Gonzales is a PhD student in the department of economics at the University of Virginia in Charlottesville. 183 184 new century, old disparities Izaguirre, and Ñopo 2009). An analysis for Guatemala is also included (see Ñopo and Gonzales 2008).1 Earnings are measured as hourly earnings. What Does the Literature Show? Psacharopoulos and Tzannatos (1993) were among the first to address gender disparities in Central America. Using historical census data and household surveys in a set of Latin American countries including Costa Rica and Honduras, they find that gender differences in human capital characteristics cannot account for the observed earnings differentials between men and women. They also find that women in the public sector are paid more than their counterparts in the private sector and that pay is more unequal in the public sector than in the private sector. These differences reflect the fact that women in the public sector tend to be more educated than both women and men in the private sector. Panizza and Qiang (2005) show similar results for Costa Rica and El Salvador, where they find a premium of more than 10 percent associ- ated with working in the public sector. Although this premium is often larger for women than men, it still does not compensate for the wide overall gender earnings gap. Dávila and Pagán (1999) analyze the sources of intercountry differences between Costa Rica and El Salvador in the gender earnings gap during the late 1980s from an occupational segregation approach. They report that women in both countries are underrepresented in occupational categories such as managerial, service, agricultural labor, and laborer occupational categories and overrepresented in professional, administrative support and clerical, and transportation jobs. They also find that differences in weekly hours worked and occupational attainment explain the differ- ences in the gender earnings gap. Using data for urban Costa Rica in 1989, 1993, and 1997, Deutsch et al. (2005) find that occupational segregation did not decrease dur- ing this period. Human capital endowments reduced the gender gap in earnings, but a larger problem involved returns to that human capital. Occupational segregation is much more severe among the less educated than the more educated. Furthermore, in all years studied, differences in earnings that cannot be explained by differences in human capital charac- teristics account for the largest portion of the earnings gap. Corley, Perardel, and Popova (2005) show trends in low- and high- skilled occupational earnings across countries. They find that between 1990 and 2000, Nicaragua enjoyed particularly strong earnings growth in both high-skilled and low-skilled occupations. In El Salvador, the gen- der earnings gap in the manufacturing sector increased from 5 percent in 1996 to almost 16 percent in 2003. The opposite occurred in Costa Rica, where the gap narrowed from 28 percent in 1996 to 18 percent in 2006. heterogeneity within central america, 1997–2006 185 Pisani and Pagán (2004) conduct a similar exercise, focusing on high and low educational attainment groups. They find that workers in Nicaragua with higher levels of education were most likely to be employed in the much higher-paying formal sector; people with little education were most likely to be found in the low-paying informal sector. They also find that women earn less than men in both educational groups. How Do Male and Female Workers Differ? Table 11.1 presents statistics for each period in the pooled database for Costa Rica, El Salvador, Honduras, and Nicaragua. It shows relative hourly earnings by various sets of observable individual characteristics, normaliz- ing them by the average women’s earnings. Table 11.2 presents descriptive statistics of the distribution of these characteristics in the samples. In circa 1997, men earn 8.9 percent more than women. This relation is reversed in circa 2001 and 2006: men earn 1.3 percent less than women in circa 2001 and 2.6 percent less in circa 2006. In circa 1997, men earn more than women at every age interval. In circa 2001 and 2006, for the popula- tion 15–34, women earn slightly more than men. In circa 1997, men earn more than women at every level of education. However, in circa 2001 and 2006, women at the bottom of the education distribution (no education or incomplete primary) earn more than men with the same educational level. The original gender earnings gap differs by economic sector, type of employment, firm size, and other characteristics. However, these differ- ences are just simple mean comparisons; they do not take into account gender differences in observable characteristics, which matter in the deter- mination of earnings. Women in the labor force are more educated than men. The propor- tion of women with tertiary education increased by 3.3 percentage points between circa 1997 and 2006, whereas the proportion for men increased by just 1.1 percentage points. The prevalence of self-employed people is greater for women in all three years. Women are more likely than men to work part time. There are also significant differences in economic sector by gender: women are concentrated in wholesale and retail trade and the hotel and restaurants sectors, whereas men are concentrated in agricul- ture, hunting, forestry, and fishing. Women represent just 30–40 percent of the paid work force in Costa Rica, El Salvador, Honduras, and Nicaragua (for tables on each country, see Enamorado, Izaguirre, and Ñopo 2009). However, participation by women increased over the period examined, especially in Costa Rica and Honduras. The countries in the pooled sample show patterns of gender schooling gaps similar to the patterns in the rest of the region, with a marked reversal in recent decades (see chapter 3). On average, women have about one more 186 Table 11.1 Relative Hourly Earnings of Men and Women in Central American Countries, by Demographic and Job Characteristics, Circa 1997–2006 Circa 1997 Circa 2001 Circa 2006 Base: average women’s Base: average women’s Base: average women’s earnings in each year and earnings in each year and earnings in each year and country = 100 country = 100 country = 100 Women Men Women Men Women Men All 100 108.9 100.0 98.7 100.0 97.4 Personal characteristics Age 15–24 71.4 75.0 71.3 68.6 72.7 67.8 25–34 104.6 111.6 102.2 98.8 102.5 99.3 35–44 117.5 132.2 113.5 115.1 108.4 113.0 45–54 108.9 128.8 113.4 122.4 114.8 115.0 55–64 86.7 112.0 94.0 100.4 93.8 103.2 Education None 52.5 59.4 59.9 53.2 59.6 51.8 Primary incomplete 65.6 79.1 68.5 69.4 73.4 71.0 Primary complete 75.9 96.4 73.4 85.1 70.7 81.7 Secondary incomplete 85.0 105.3 79.7 91.7 76.4 86.9 Secondary complete 117.2 145.1 117.2 126.8 104.7 117.9 Table 11.1 (continued) Circa 1997 Circa 2001 Circa 2006 Base: average women’s Base: average women’s Base: average women’s earnings in each year and earnings in each year and earnings in each year and country = 100 country = 100 country = 100 Women Men Women Men Women Men Tertiary incomplete 197.7 207.9 167.4 186.6 152.8 170.9 Tertiary complete 247.7 280.7 232.5 274.4 215.5 244.1 Presence of children (12 years or younger) in the household No 111.0 117.6 107.6 107.1 106.5 103.3 Yes 88.4 100.6 90.4 89.3 90.3 89.4 Presence of other household member with labor income No 97.5 111.2 98.6 98.5 104.3 98.4 Yes 100.9 107.3 100.5 98.9 98.4 96.8 Dependency More independents than dependents in the household 105.9 111.5 104.0 102.8 103.2 99.9 187 (continued next page) 188 Table 11.1 (continued) Circa 1997 Circa 2001 Circa 2006 Base: average women’s Base: average women’s Base: average women’s earnings in each year and earnings in each year and earnings in each year and country = 100 country = 100 country = 100 Women Men Women Men Women Men Same independents as dependents in the household 100.2 116.2 101.5 100.0 101.1 100.8 More dependents than independents in the household 84.7 95.0 84.9 82.6 85.1 81.2 Urban No 82.0 86.6 84.8 74.2 82.6 74.6 Yes 108.0 125.9 105.8 116.1 106.7 112.6 Labor characteristics Type of employment Employer 143.0 161.0 172.8 144.8 138.2 152.4 Self-employed 80.1 99.7 81.6 83.0 82.4 84.6 Employee 107.5 106.4 106.3 99.0 106.8 96.8 Table 11.1 (continued) Circa 1997 Circa 2001 Circa 2006 Base: average women’s Base: average women’s Base: average women’s earnings in each year and earnings in each year and earnings in each year and country = 100 country = 100 country = 100 Women Men Women Men Women Men Time worked Part time 120.5 136.0 115.1 121.1 114.5 113.3 Full time 115.6 114.2 109.9 100.7 107.3 100.9 Overtime 65.0 93.6 72.6 87.8 72.6 85.7 One job No 112.5 123.6 116.1 104.6 110.4 96.8 Yes 99.3 107.6 98.9 98.1 99.2 97.5 Small firm (five workers or less) No 132.4 125.1 130.6 119.0 129.8 114.8 Yes 73.7 92.3 77.0 80.4 74.2 75.4 Economic sector Agriculture, hunting, forestry, and fishing 54.9 67.2 60.1 55.0 58.1 52.9 189 (continued next page) Table 11.1 (continued) 190 Circa 1997 Circa 2001 Circa 2006 Base: average women’s Base: average women’s Base: average women’s earnings in each year and earnings in each year and earnings in each year and country = 100 country = 100 country = 100 Women Men Women Men Women Men Elementary manufacturing 77.2 100.0 74.0 92.5 74.0 94.4 Other manufacturing 116.3 120.2 96.8 102.7 105.3 102.0 Construction 132.0 102.4 134.0 93.8 114.0 89.6 Wholesale and retail trade and hotels and restaurants 87.0 119.3 91.5 105.3 89.6 103.3 Electricity, gas, water supply, transport, and communications 182.3 137.2 152.2 131.3 153.8 121.3 Financing, insurance, real estate, and business services 185.9 168.1 154.7 148.4 150.5 136.4 Public administration and defense 170.3 156.4 176.5 153.1 165.0 152.2 Education, health, and personal services 151.0 151.3 136.8 150.3 142.7 150.3 Source: Based on 1995–2007 national household surveys of Costa Rica, El Salvador, Honduras, and Nicaragua. Table 11.2 Demographic and Job Characteristics of Central American Countries, 1997, 2001, and 2006 (percent) Circa 1997 Circa 2001 Circa 2006 Women Men Women Men Women Men Real Earnings Personal characteristics Age 15–24 22.7 26.9 21.0 25.5 19.0 24.9 25–34 30.7 28.8 29.2 27.4 28.6 27.5 35–44 26.5 22.4 26.6 23.2 27.0 22.4 45–54 14.0 14.5 16.4 15.7 18.0 16.3 55–64 6.0 7.5 6.9 8.3 7.4 9.0 Education None 10.4 11.9 9.3 11.2 7.4 8.8 Primary incomplete 26.3 29.4 24.8 28.0 21.5 24.2 Primary complete 18.6 21.7 18.4 22.8 18.0 23.4 Secondary incomplete 12.6 14.2 13.0 14.0 15.5 17.4 Secondary complete 17.7 12.7 17.6 13.0 17.5 13.8 Tertiary incomplete 8.0 5.7 9.6 6.4 10.5 6.8 Tertiary complete 6.5 4.4 7.2 4.7 9.7 5.5 191 (continued next page) 192 Table 11.2 (continued) Circa 1997 Circa 2001 Circa 2006 Women Men Women Men Women Men Presence of children (12 years or younger) in the household No 51.3 48.5 59.8 52.9 59.8 57.6 Yes 48.8 51.5 44.1 47.2 40.2 42.4 Presence of other household member with labor income No 26.2 39.7 25.3 38.2 27.1 38.5 Yes 73.8 60.3 74.8 61.8 72.9 61.5 Dependency More independents than dependents in the household 59.9 62.6 64.7 65.5 68.4 70.0 Same independents as dependents in the household 16.7 16.6 16.6 16.6 15.7 15.9 More dependents than independents in the household 23.3 20.8 18.7 17.9 15.9 14.0 Table 11.2 (continued) Circa 1997 Circa 2001 Circa 2006 Women Men Women Men Women Men Urban No 30.7 43.3 27.6 41.4 27.7 40.0 Yes 69.3 56.7 72.4 58.6 72.3 60.0 Job characteristics Type of employment Employer 2.5 7.3 3.0 8.0 3.0 6.0 Employee 67.0 70.7 63.5 67.5 65.3 72.0 Self-employed 30.5 22.0 33.5 24.5 31.7 22.0 Time worked Part time 23.7 11.7 25.0 12.7 26.1 13.2 Full time 43.2 50.0 44.5 51.6 47.4 53.2 Overtime 33.0 38.3 30.2 35.6 26.5 33.7 One job No 5.6 8.0 6.3 9.6 6.8 8.5 Yes 94.4 92.0 93.7 90.3 93.2 91.5 (continued next page) 193 194 Table 11.2 (continued) Circa 1997 Circa 2001 Circa 2006 Women Men Women Men Women Men Small firm (five workers or less) No 44.9 50.5 43.0 47.4 44.4 50.3 Yes 55.2 49.5 57.0 52.6 46.2 44.4 Not reported – – – – 9.5 5.3 Economic sector Agriculture, hunting, forestry, and fishing 4.6 28.2 2.8 27.8 3.3 24.6 Elementary manufacturing 17.3 8.3 16.5 7.5 15.6 7.7 Other manufacturing 2.7 7.3 2.6 7.4 2.5 7.2 Construction 0.3 10.2 0.5 10.5 0.4 11.7 Wholesale and retail trade and hotels and restaurants 31.2 19.4 31.7 19.4 32.7 20.4 Table 11.2 (continued) Circa 1997 Circa 2001 Circa 2006 Women Men Women Men Women Men Electricity, gas, water supply, transport, and communications 1.45 8.37 1.76 8.85 1.98 9.26 Financing, insurance, real estate, and business services 3.13 3.52 4.79 4.85 4.99 5.89 Public administration and defense 4.95 5.82 4.67 5.00 4.59 5.05 Education, health, and personal services 19.13 8.10 20.91 7.74 19.95 6.85 Domestic servants 15.33 0.82 13.74 0.98 13.99 1.17 Source: Based on 1995–2007 national household surveys of Costa Rica, El Salvador, Honduras, and Nicaragua. 195 196 new century, old disparities year of schooling than their male counterparts. In Costa Rica, about half of workers report being a head of household. This percentage is slightly smaller in El Salvador, Honduras, and Nicaragua. Marital arrangements are similar across countries and stable over time. Except in El Salvador in 1995, about 1 in 4 workers is single and about 5–6 in 10 workers are in a (formal or informal) marital union.2 Age groups display similar patterns across countries, with almost 40 percent of the sample in each country between the ages of 25 and 40. Descriptive statistics for Guatemala for 2000, 2004, and 2006 show that the gender composition of the labor market was stable over the period of analysis (see Ñopo and Gonzales 2008).3 About 70 percent of workers in Guatemala are men, and this share did not change significantly during the period of analysis. Participation by gender is more balanced in urban (60 percent men) than in rural (80 percent men) areas. Real monthly earn- ings (expressed in 2006 quetzals) declined slightly for men and remained constant for women during 2000–06. As a result, the gender earnings gap narrowed, from 28 percent to 18 percent, during this period. Average urban earnings are almost twice average earnings in rural areas, but the decline in men’s average earnings was more pronounced in urban areas. There are no significant differences in gender gaps between urban and rural areas, except in 2000. Monthly earnings differ widely by educational attainment. The ratio between average earnings of people with university degrees and people with less than secondary education is five to one, although this gap has been closing since 2000. Income disparities between the least educated and most educated are in line with the findings of Auguste, Artana, and Cuevas (2007), who find that the returns to education in Guatemala are among the highest in Latin America. Among employed people in Guatemala, women have about one year more education than men. This result is in apparent contradiction with the findings reported in chapter 3, which indicate that Guatemalan men from recent cohorts are more educated than women. The results pre- sented in this chapter refer only to the working population. The difference between the two results may reflect the nonrandom selection of men and women into the labor market. Given their more limited opportunities to participate in labor markets, women may be acquiring more education to compete with men for jobs. The Role of Individual Characteristics in Explaining the Earnings Gap Figure 11.1 shows the evolution of the original gender earnings gap by country. Except for Costa Rica, the earnings gap decreased between circa heterogeneity within central america, 1997–2006 197 Figure 11.1 Gender Earnings Gap in Central American Countries, Circa 1997–2006 20 percentage of average women’s earnings 15 10 5 0 −5 −10 −15 (circa 1997) (circa 2001) (circa 2006) Costa Rica El Salvador Honduras Nicaragua Source: Based on 1995–2007 national household surveys of Costa Rica, El Salvador, Honduras, and Nicaragua. 1997 and circa 2006. The widest gaps appear in circa 1997 in El Salvador and Honduras, circa 2001 in Guatemala, and circa 2006 in Nicaragua. In circa 2006, the original gender earnings gap is not statistically different from zero in Costa Rica and El Salvador. Table 11.3 decomposes the gender earnings gap using the matching methodology described in chapter 2. Six observable demographic charac- teristics are considered as controls. In circa 1997, men earn 8.9 percent more than women. After control- ling for age, most of the gender earnings gap remains unexplained. Adding education to the controls, the unexplained earnings gap (Δ0)—the part of the earnings gap that cannot be attributed to differences in characteristics of individuals—is considerably larger than the total earnings gap (Δ). The component that captures differences in observable characteristics (Δx) is negative, reflecting the fact that women have more education than men. After adding new characteristics to the set of controls, the unexplained component of the earnings gap remains constant. Matching by demographic characteristics, the unexplained earnings gap is 18.3 percent (that is, if men and women had the same distribution of observable demographic characteristics, men would earn 18.3 percent more than women). The total earnings gap is smaller than the unexplained earn- ings gap because women have characteristics that are better remunerated 198 Table 11.3 Decomposition of Gender Earnings Gap in Central America after Controlling for Demographic Characteristics, Circa 1997 (percent) Circa 1997—Guatemala not included + Presence of + Presence of other household children in the member with Age + Education household labor income + Dependency + Urban Δ 8.9 8.9 8.9 8.9 8.9 8.9 Δ0 11.7 18.9 18.2 16.8 16.6 18.3 ΔM 0.0 0.9 1.1 1.7 1.7 –1.7 ΔF 0.0 –0.3 –0.5 –1.5 –2.5 –1.9 ΔX –2.8 –10.7 –9.9 –8.1 –6.9 –5.9 Percentage of men in the 100.0 98.5 96.0 89.2 72.9 59.5 common support Percentage of women in 100.0 99.5 98.7 95.8 83.6 74.0 the common support Source: Based on circa 1997 national household surveys of Costa Rica, El Salvador, Honduras, and Nicaragua. Note: ΔM (ΔF) is the part of the earnings gap attributed to the existence of men (women) with combinations of characteristics that are not met by any women (men). ΔX is the part of the earnings gap attributed to differences in the observable characteristics of men and women over the “common support.� Δ0 is the part of the earnings gap that cannot be attributed to differences in characteristics of the individuals. It is typically attributed to a combination of both unobservable characteristics and discrimination. The sum of these components equals the total earnings gap (ΔM + ΔF + ΔX + Δ0 = Δ). heterogeneity within central america, 1997–2006 199 in the labor market (ΔX = –5.9 percent) and because of differences in the “common support� of characteristics. Unmatched men earn lower earn- ings than matched men (ΔM = –1.6 percent), and unmatched women earn higher earnings than matched women (ΔF = –1.9 percent). This pattern is the same in all periods, except for the reversal of the gap in favor of women reported earlier. Table 11.4 compares results for each country after matching on two sets of individual characteristics. The first set considers only area and education; the second adds age, head of household, marital status, and occupation. Costa Rica stands out as a country with a negative gender earnings gap, although the gap is relatively small (and likely not statisti- cally different from zero). Nicaragua has a small positive gender earnings gap. Honduras shows a slightly larger gender earnings gap, and El Salvador is the country in the sample with the largest gap. The set of countries can thus be grouped into countries with small gender earnings gaps (Costa Rica and Nicaragua) and countries with larger gender earnings gaps (El Salvador and Honduras). In all four countries, the unexplained component of the gap exceeds the original measure of the gender earnings gap. This result is a consequence of the fact that women have more years of education than men. The extent to which Δ0 exceeds Δ varies across countries and time. For the two countries with large earnings gaps (El Salvador and Honduras), the portion of the gap that cannot be explained by gender differences in observed char- acteristics tends to be closer to the total earnings gap in the mid-1990s than in later years, especially when controlling for the broader set of individual characteristics. For countries with smaller gaps (Costa Rica and Nicaragua), the unexplained components are larger than the original earnings gaps. Regarding the out-of-common-support components, in most cases ΔM is positive and ΔF is negative. In the two countries with larger earnings gaps, ΔM dominates ΔF; in the two countries with small earnings gaps, the opposite is true. The rural earnings gap has a larger unexplained component than the national gap in three of the four countries (the exception being El Salvador) (for tables reporting these results, see Enamorado, Izagu- irre, and Ñopo 2009). The national findings on out-of-common-support components prevail in rural areas, in both the high and low earnings gap countries. For the urban earnings gap decomposition, the situation changes slightly. In Costa Rica and Nicaragua (countries with low earnings gaps), the unexplained component of the gap is larger than the original gap. In Honduras and El Salvador (countries with high earn- ings gaps), the situation resembles a traditional gender earnings gap decomposition: the unexplained component is no longer larger than the original gap. Regarding the out-of-common-support components for the low earnings gap countries, in Nicaragua, the pattern observed at the national and rural 200 Table 11.4 Decomposition of Gender Earnings Gaps in Central American Countries after Controlling for Demographic Characteristics, Various Years (percent) Costa Rica El Salvador Honduras Nicaragua Urban, Urban, Urban, Urban, education, education, education, education, age, age, age, age, head of head of head of head of household, household, household, household, marital marital marital marital Area and status, and Area and status, and Area and status, and Area and status, and Period education occupation education occupation education occupation education occupation Circa 1997 Δ –1.9 –1.9 24.7 24.7 11.4 11.4 5.1 5.1 Δ0 14.6 11.8 30.1 22.9 26.0 10.1 22.3 30.1 ΔM 0.0 22.9 0.0 21.4 0.1 10.7 0.0 15.0 ΔF 0.0 –28.7 –0.1 –12.1 0.0 –4.9 0.0 –24.6 ΔX –16.5 –7.8 –5.3 –7.5 –14.8 –4.5 –17.2 –15.4 Circa 2001 Δ –3.5 –3.5 12.9 12.9 0.0 0.0 –4.6 –4.6 Δ0 15.7 7.8 16.7 11.0 16.4 8.9 12.9 18.6 ΔM 0.0 15.2 0.1 15.9 0.0 8.3 0.0 9.9 ΔF 0.0 –19.2 –0.1 –3.9 0.0 –8.2 –0.1 –17.5 ΔX –19.2 –7.3 –3.9 –10.0 –16.5 –9.0 –17.4 –15.5 Table 11.4 (continued) Costa Rica El Salvador Honduras Nicaragua Urban, Urban, Urban, Urban, education, education, education, education, age, age, age, age, head of head of head of head of household, household, household, household, marital marital marital marital Area and status, and Area and status, and Area and status, and Area and status, and Period education occupation education occupation education occupation education occupation Circa 2006 Δ –2.9 –2.9 14.3 14.3 2.6 2.6 2.6 2.6 Δ0 17.2 12.2 20.6 20.5 14.2 12.3 20.3 16.4 ΔM 0.0 7.8 0.1 –9.3 0.1 7.5 0.1 11.6 ΔF 0.0 –7.2 –0.2 4.8 0.0 –7.3 0.0 –14.8 ΔX –20.2 –15.7 –6.1 –1.6 –11.6 –9.9 –17.8 –10.5 Source: Based on 1995–2007 national household surveys of Costa Rica, El Salvador, Honduras, and Nicaragua. Note: ΔM (ΔF) is the part of the earnings gap attributed to the existence of men (women) with combinations of characteristics that are not met by any women (men). ΔX is the part of the earnings gap attributed to differences in the observable characteristics of men and women over the “common support.� Δ0 is the part of the earnings gap that cannot be attributed to differences in characteristics of the individuals. It is typically attributed to a combination of both unobservable characteristics and discrimination. The sum of these components equals the total earnings gap (ΔM + ΔF + ΔX + Δ0 = Δ). 201 202 new century, old disparities levels remains when controls are added for the urban sample. In contrast, in Costa Rica, the relationship between ΔF and ΔM changes, with ΔM now dominating ΔF. In El Salvador and Honduras, the results for the national and rural samples (that is, ΔM dominating ΔF) reverses in the mid-2000s. The earnings gaps in Guatemala were decomposed for the entire work- ing population and for urban and rural working populations. Only the decompositions for the entire population that control for age, marital status, and education are shown, because they are closer to the controls used in the other countries (figure 11.2).4 About half of the earnings gaps are explained by differences in the distribution of characteristics, both where these distributions are comparable for men and women (ΔX) and where they are not (ΔF and ΔM). The components that control for the lack of common support between men and women are very small and not statistically significant in most combinations. Only in the last set of controls do ΔM and ΔF play impor- tant roles. This result is very similar to the results for Chile (chapter 7) and Peru (chapter 5). Age, marital status, and education provide enough Figure 11.2 Decomposition of Gender Earnings Gap in Guatemala after Controlling for Age, Marital Status, and Education, 2000–06 30 percent of average women’s 25 20 earnings 15 10 5 0 −5 2000 2004 2006 ΔX ΔF ΔM Δ0 Source: Based on data from the 2000 and 2006 ENCOVI and 2004 ENEI. Note: ΔM (ΔF) is the part of the earnings gap attributed to the existence of men (women) with combinations of characteristics that are not met by any women (men). ΔX is the part of the earnings gap attributed to differences in the observable characteristics of men and women over the “common support.� Δ0 is the part of the earnings gap that cannot be attributed to differences in characteristics of the individuals. It is typically attributed to a combination of both unobservable characteristics and discrimination. The sum of these components equals the total earnings gap (ΔM + ΔF + ΔX + Δ0 = Δ). heterogeneity within central america, 1997–2006 203 information to assess the unexplained gender earnings gap. Of these three variables, it is education that drives gender earnings gaps. The decomposition of the national earnings gap is largely similar to the decomposition in urban areas. In contrast, in rural areas, the decom- position is slightly different. The unexplained component accounts for about 80 percent of the earnings gap and the component attributable to unpaired women is negative. Apparently, segmentation (or segregation) operates negatively on women’s earnings in urban areas and positively in rural areas. Exploring the Unexplained Component of the Gender Earnings Gap The decompositions described in table 11.3 and figure 11.2 describe the mean gaps, without reference to either their distribution or variability. Figure 11.3 Confidence Intervals for Unexplained Gender Earnings Gap in Costa Rica, El Salvador, Honduras, and Nicaragua after Controlling for Demographic and Job Characteristics, 1995–2007 25 a. Costa Rica, 1995–2006 percentage of average women’s 20 15 earnings 10 5 0 1995 2000 2006 35 b. El Salvador, 1995–2006 percentage of average women’s 30 25 earnings 20 15 10 5 0 1995 2000 2005 (continued next page) 204 new century, old disparities Figure 11.3 (continued) 25 percentage of average women’s c. Honduras, 1997–2007 20 15 earnings 10 5 0 –5 –10 1997 2002 2007 60 d. Nicaragua, 1998–2005 percentage of average women’s 50 40 30 earnings 20 10 0 –10 –20 –30 1995 2000 2005 Source: Based on 1995–2007 national household surveys of Costa Rica, El Salvador, Honduras, and Nicaragua. Note: Figures show results after controlling for demographic and job-related characteristics. Boxes show 90 percent confidence intervals for unexplained earnings; whiskers show 95 percent confidence intervals. Figure 11.3 presents confidence intervals for the unexplained compo- nent of the gender earnings gap that remains after controlling for the full set of individual characteristics (area, education, age, household head, marital status, and occupation) for Costa Rica, El Salvador, Hon- duras, and Nicaragua. The extremes of the boxes represent 90 percent confidence intervals for the mean unexplained gender earnings gaps; the whiskers represent 95 percent confidence intervals (for figures at the urban and rural levels, see Enamorado, Izaguirre, and Ñopo 2009). Although the hypothesis that the gender earnings gaps remained con- stant over time cannot be statistically ruled out, the figures show a nar- rowing in the gaps between the mid-1990s and 2000, after which the gaps widen. The following subsections present the results for the empirical distribu- tions of the unexplained earnings gap for each country, using the latest heterogeneity within central america, 1997–2006 205 survey data available and three different sets of individual characteristics: first, area; second, area, education, and age; and third, area, education, age, household head, marital status, and occupation. Costa Rica The unexplained part of the gender earnings gap in Costa Rica is larger at the lowest percentiles; gaps are close to zero after the 57th percentile (figure 11.4). After controlling by more characteristics, the gaps remain about 20 percent. At the upper extreme of the earnings distribution, after controlling by the full set of characteristics, the earnings gaps narrow, approaching zero. El Salvador Much of the unexplained gaps in El Salvador appears at the bottom of the earnings distribution (figure 11.5). Qualitatively, the plots for the Figure 11.4 Unexplained Gender Earnings Gap in Costa Rica after Controlling for Demographic and Job Characteristics, by Percentile of Earnings Distribution, 2006 120 percentage of average women’s earnings 100 80 60 40 20 0 –20 0 10 20 30 40 50 60 70 80 90 100 percentile of earnings distribution area area, education, and age area, education, age, head of household, marital status, and occupation Source: Based on data from Costa Rica’s 2006 national household surveys. 206 new century, old disparities Figure 11.5 Unexplained Gender Earnings Gap in El Salvador after Controlling for Demographic and Job Characteristics, by Percentile of Earnings Distribution, 2005 180 percentage of average women’s earnings 160 140 120 100 80 60 40 20 0 0 10 20 30 40 50 60 70 80 90 100 percentile of earnings distribution area area, education, and age area, education, age, head of household, marital status, and occupation Source: Based on data from 2005 El Salvador’s national household surveys. three sets of controls are similar. Between the 1st and 10th percentiles, the gaps are large but decrease rapidly, moving from 160 percent to 80 percent in these first 10 percentiles. Between the 11th and 55th percentiles, there is still a decrease of the gender gap along the percentiles, but the rate of decrease is slower, falling from 80 percent to 30 percent. In this interval, the use of extra controls (head of household, marital status, and occupation) reduces unexplained gap. Around the 65th percentile, there is a peak in unexplained earnings differences. Thereafter the gap declines, ending up with values close to zero at the top of the earnings distribution. Guatemala The unexplained component of the earnings gap is larger among low-income workers than among high-income workers in Guatemala (figure 11.6). The gap decreases rapidly, becoming negative after the 70th percentile of the earnings distribution, a sign of significant inequality heterogeneity within central america, 1997–2006 207 Figure 11.6 Unexplained Gender Earnings Gap in Guatemala after Controlling for Demographic and Job Characteristics, by Percentile of Earnings Distribution, 2000–06 110 percentage of average women’s earnings 90 70 50 30 10 –10 0 10 20 30 40 50 60 70 80 90 100 percentile of earnings distribution Source: Based on data from Guatemala’s 2000–06 national household surveys. within social classes. This distribution is similar to that found in the other Central American countries. Honduras As in El Salvador, larger unexplained differences in earnings are found at the lower percentiles of the earnings distribution in Honduras (figure 11.7). At the lowest percentile of the earnings distribution the unexplained gender earnings gap is 60–100 percent, declining to 20–30 percent around the 40th percentile. For higher percentiles of the earnings distribution, the unex- plained gender gap also decreases but at a slower rate. As in El Salvador, at the upper part of the earnings distributions (85th percentile and above), the unexplained gender earnings gap is almost zero for all three sets of control- ling characteristics. Nicaragua The unexplained gender gaps in Nicaragua behave slightly differently from the other countries (figure 11.8). At the lowest percentiles of the earnings distributions, the gap is negative when the smaller sets of controls 208 new century, old disparities Figure 11.7 Unexplained Gender Earnings Gap in Honduras after Controlling for Demographic and Job Characteristics, by Percentile of Earnings Distribution, 2007 120 percentage of average women’s earnings 100 80 60 40 20 0 –20 0 10 20 30 40 50 60 70 80 90 100 percentile of earnings distribution area area, education, and age area, education, age, head of household, marital status, and occupation Source: Based on data from Honduras’ 2007 national household survey. are used; it is positive only for the set that controls for area, education, age, head of household, marital status, and occupation. The unexplained gap increases with earnings up to the 15th percentile. After that point, the gap decreases but at a slower rate than in Honduras and Guatemala, so that in statistical terms the unexplained gap can be assumed to be constant between the 30th and 95th percentiles. Figures 11.4–11.8 show more similarities than differences in the distri- bution of unexplained gender differences in pay in the five countries. All five countries show larger gaps at the bottom of the earnings distribution and almost zero gaps at the top. For this reason, in the remainder of the analysis, only results for the pooled database are shown. The pool selected corresponds to data for Costa Rica, El Salvador, Honduras, and Nicaragua in the latest time period for which data were available (circa 2006). To what extent do unexplained gender earnings gaps (after control- ling for the fullest set of observable characteristics) differ across different segments of labor markets? Figure 11.9 shows confidence intervals for the unexplained component of the gender earnings gap by area, age, years of heterogeneity within central america, 1997–2006 209 Figure 11.8 Unexplained Gender Earnings Gap in Nicaragua after Controlling for Demographic and Job Characteristics, by Percentile of Earnings Distribution, 2005 120 percentage of average women’s earnings 100 80 60 40 20 0 –20 –40 –60 0 10 20 30 40 50 60 70 80 90 100 percentile of earnings distribution area area, education, and age area, education, age, head of household, marital status, and occupation Source: Based on 2005 data from Nicaragua’s national household survey. education, marital status, head of household, and occupation. As before, the extremes of the boxes represent 90 percent confidence intervals for the mean unexplained gender earnings gaps, and the whiskers represent a 95 percent confidence interval. The results illustrate that gender earnings gaps do not statistically differ in rural and urban areas (panel a). They decrease with age, becom- ing statistically indistinguishable from zero among the oldest cohort (people passed the traditional retirement age) (panel b). In contrast with other countries in Latin America, the unexplained gender earnings gap seems to be larger among people with 6–11 years of completed school- ing (panel c). The unexplained gaps are smaller among widowed people, among whom the gap is negative at the 95 percent confidence level (panel d). Although the average unexplained gaps do not statistically dif- fer between people who are heads of household and people who are not, the dispersion is greater among household heads (panel e). Unexplained gender earnings differences are large and dispersed among agricultural workers and negative among professionals (panel f). A similar analysis for Guatemala shows that the unexplained gender earnings gaps are larger among young people, people with higher education, people who 210 new century, old disparities Figure 11.9 Confidence Intervals for Unexplained Earnings Gaps in Central America after Controlling for Demographic Characteristics, Circa 2006 a. By area 20 percentage of women’s earnings 18 16 14 12 10 08 06 04 02 0 urban rural area 40 b. By age percentage of women’s earnings 30 20 10 0 –10 –20 –30 0–14 15–24 25–40 41–64 65 or older age c. By years of education percentage of women’s earnings 25 20 15 10 5 0 low medium high education (continued next page) heterogeneity within central america, 1997–2006 211 Figure 11.9 (continued) 40 d. By marital status percentage of women’s earnings 30 20 10 0 –10 –20 –30 –40 –50 –60 –70 single formal or divorced or widow informal partners separated years 30 e. By head of household status percentage of women’s earnings 25 20 15 10 5 0 not household head household head head of the household 6 f. By occupation percentage of women’s earnings 5 4 3 2 1 0 –1 –2 ke al rs rs s s es rs ts or r al or ke he w ltu rs ke an tiv on ct or u ot or ra ch re ic si lw w st gr di es er i e ra na in m of ic tu m pr no rv d ul ad an se ric rs ag lle se occupation Source: Based on data from circa 2006 national household surveys of Costa Rica, El Salvador, Honduras, and Nicaragua. Note: Figures show results after controlling for demographic and job-related characteristics. Boxes show 90 percent confidence intervals for unexplained earnings; whiskers show 99 percent confidence intervals. 212 new century, old disparities are separated, migrants, and people living in the capital (Ñopo and Gonzales 2008). This chapter portrays the evolution of gender earnings gaps in Central American countries during the past decade. Some trends suggest improve- ments in gender equity in labor markets: participation by women increased (particularly in Costa Rica and Honduras), and women acquired more years of schooling than men during the period under study. However, sub- stantial gender earnings gaps persist. The results show a pattern in which the unexplained part of the gender earnings gaps is larger among poorer people than it is at the top of the income distribution. This pattern can be very harmful in countries with high incidences of poverty. Notes 1. The population examined is working people between the ages of 15 and 65, except in Guatemala, where the working population is age 18–65. 2. The category of “informal union� was not included in the Salvadoran sur- vey until 2000. 3. Data for 2000 and 2006 come from the National Survey of Living Condi- tions (Encuesta Nacional de Condiciones de Vida [ENCOVI]); data for 2004 come from the National Survey of Employment and Income (Encuesta Nacional de Empleo e Ingresos [ENEI]). 4. For urban and rural decompositions and for results using the other sets of controls refer to Ñopo and Gonzales (2008). References Auguste, S., D. Artana, and M. Cuevas. 2007. “Tearing Down the Walls: Growth and Inclusion in Guatemala.� Inter-American Development Bank, Country Department Central America, Mexico, Panama, and Dominican Republic (CID), Washington, DC. Corley, M., Y. Perardel, and K. Popova. 2005. Wage Inequality by Gender and Occupation: A Cross-Country Analysis. Geneva: International Labour Organization. Dávila, A., and J. Pagán. 1999. “Gender Pay and Occupational-Attainment Gaps in Costa Rica and El Salvador: A Relative Comparison of the Late 1980s.� Review of Development Economics 3 (2): 215–30. Deutsch, R., A. Morrison, C. Piras, and H. Ñopo. 2005. “Working within Con- fines: Occupational Segregation by Sex for Three Latin American Countries.� Icfai University Journal of Applied Economics 4 (3): 50–59. Enamorado, T. A., C. Izaguirre, and H. Ñopo. 2009. “Gender Wage Gaps in Cen- tral American Countries: Evidence from a Non-Parametric Approach.� RES Working Paper 4639, Inter-American Development Bank, Research Depart- ment, Washington, DC. heterogeneity within central america, 1997–2006 213 Ñopo, H., and A. Gonzales. 2008. “Gender and Ethnic Wage Gaps in Guatemala from a Matching Comparisons Perspectives.� RES Working Paper 4588, Inter- American Development Bank, Research Department, Washington, DC. ———. 2009. “Brechas salariales por género y etnicidad.� In Más crecimiento, más equidad, ed. Ernesto Stein, Osmel Manzano, Hector Morena, and Fernando Straface, 265–98. Banco Interamericano de Desarrollo. Panizza, U., and C. Z.-W. Qiang. 2005. “Public-Private Wage Differential and Gender Gap in Latin America: Spoiled Bureaucrats and Exploited Women?� Journal of Socioeconomics 34 (6): 810–83. Pisani, M. J., and J. A. Pagán. 2004. “Sectoral Selection and Informality: A Nicaraguan Case Study.� Review of Development Economics 8 (4): 541–56. Psacharopoulos, G., and Z. Tzannatos. 1993. “Economic and Demographic Effects on Working Women in Latin America.� Journal of Population Economics 6 (4): 293–315. 12 The Understudied Caribbean: Barbados (2004) and Jamaica (2003) The Caribbean is an understudied region in economic terms. On labor markets issues, the body of empirical research is small. This chapter attempts to fill this void by examining gender earnings gaps in Barbados and Jamaica, two large economies by Caribbean standards, with diverse labor market, social, and economic issues. The chapter focuses on these two countries for a number of reasons. First, both countries have reliable data for representative samples of workers at the national level. Second, the countries have many similarities and differences in terms of social, economic, and labor market issues. Examining gender earnings gaps for the two countries will illuminate peculiarities within the national labor markets, facilitating conjectures on whether the presence of gender earn- ings gaps is an endemic feature of Caribbean labor markets, as in the rest of Latin America and the world. What Does the Literature Show? Only a small number of studies examine gender gaps in the Caribbean.1 A few studies investigate gender issues in labor markets in Barbados, Jamaica, and Trinidad and Tobago. This chapter was adapted from “Gender Earnings Gaps in the Caribbean: Evidence from Barbados and Jamaica,� Alejandro Hoyos, Annelle Bellony, and Hugo Ñopo, IDB Working Paper IDB-WP-210, Inter-American Development Bank, 2010. Alejandro Hoyos is a consultant at the Poverty Reduction and Economic Man- agement Network (PREM) at the World Bank. Annelle Bellony is a senior associate in the Education Division at the Inter-American Development Bank. 215 216 new century, old disparities The evidence from the literature on the gender earnings gap generally indicates that women in Caribbean countries earn less on average than men. Scott (1992) finds that women in Jamaica earn on average 58 percent of men’s earnings. Hotchkiss and Moore (1996) report that average earnings for women in Jamaica are 80 percent of men’s earnings. The two studies are based on different data sources for the same period (the late 1980s), revealing the heterogeneity of results of studies of this kind. Whereas Scott uses labor force survey data, Hotchkiss and Moore use a special dataset compiled for a one-time tax project. Notwithstanding the discrepancy in the magnitude of the gender earnings gap, both studies find that the bulk of the gender earn- ings differential is unexplained by differences in individual characteristics. Using the 1994 Continuous Household Sample Survey (CHSS) for Barbados, Coppin (1996) finds a women’s/men’s earnings ratio of 0.87.2 Olsen and Coppin (2001) use the 1993 Continuous Sample Survey of the Population (CSSP) to estimate the gender earnings gap for Trinidad and Tobago. Their findings suggest that differences in human capital and other measured factors valued by the labor market do not do a good job of explaining earnings differentials. Terrell (1992) cites an unpublished study by Brendan (1991) that estimates the women’s and men’s earnings ratio for Haiti, derived from a 1987 survey of large-scale enterprises in Port-au-Prince, at 0.87. Furthermore, Sookram and Watson (2008) find evidence that workers in the informal sector suffer an earnings penalty, particularly women. History and Development of Barbados and Jamaica British colonization, from 1625–1966, dominates the history of Barbados, in the eastern part of the Caribbean archipelago. An estimated 90 percent of its 270,000 people are of African descent. Like many other Eastern Caribbean countries, Barbados has a history of dependence on one crop as the main export commodity, in its case, sugarcane. The economy has evolved over time to focus primarily on ser- vices, particularly tourism and finance. Jamaica, located in the western Caribbean, gained independence from the United Kingdom in 1962. It was a Spanish colony until 1655, when the British took control. Once the British settled in Jamaica, sugar pro- duction became the mainstay of the economy. First, African slaves and, later, Chinese and Indian indentured servants worked the land. Their descendants remain on the island, contributing to the ethnic diversity of the Jamaican people. The population of Jamaica is slightly less than 3 million. Tourism forms the mainstay of the economy, followed by bauxite and manufacturing. The confluence of diverse ethnic groups resulted in the creation of the Jamaican Creole language, which is widely spoken. Use of Creole has understudied gaps: barbados, 2004 and jamaica, 2003 217 contributed to low educational outcomes, especially among men (Ministry of Education, Youth, and Culture 2001). In many respects, the historical and economic pasts of Barbados and Jamaica have followed the same trajectory. However, in terms of progress on social indicators, the two countries display some noteworthy differences. Barbados has consistently ranked in the top 40 countries on the United Nations Human Development Index. In contrast, Jamaica ranked 100th on this index in 2007 (UNDP 2009). Both countries are home to two of the three campuses of the University of the West Indies (UWI), but the effect of the campuses is markedly different. Barbados has capitalized on the presence of the university: the government provides free tuition to qualified candidates as an investment in the future economic and social development of the country. Exposure to tertiary education, although low by international standards, is high for the Caribbean. In Jamaica, tertiary educational outcomes are much weaker, especially among men. The inci- dence of poverty is also much higher than in Barbados. Barbados: Men in the Middle, Women at Both Ends The data used in the analysis for Barbados are derived from the Continu- ous Labor Force Sample Survey (CLFSS) for 2004. The Barbados Statisti- cal Service conducts the CLFSS quarterly. The data were purged to include only people between the ages of 15 and 64. Data on labor earnings are coded in intervals. Coding the data in inter- vals imposes some challenges on the computation of gender earnings gaps, as the computation of average earnings requires assuming particular values for earnings within the given intervals. For simplicity, the lowest extreme of each earnings interval is assumed to be the representative value. Figure 12.1 shows the distribution of men and women along the earnings intervals. The distribution for women is skewed to the left of the distribution for men. However, at the high end of the earnings distribu- tions, there are almost no gender differences. Women’s labor force participation also varies with earnings. At the lower-middle portion of the earnings distribution, women make up less than 40 percent of the labor force. In contrast, at the two lowest extreme income brackets and the upper-middle part of the distribution, women account for more than 60 percent of the labor force (figure 12.2). An additional challenge that the dataset imposes is that not only earn- ings but also hours worked per week are coded in intervals. Fortunately, almost three out of four male workers and four out of five female workers in Barbados work 40–44 hours a week. There are some gender differences in the percentages of overtime workers: 20 percent of men and 10 percent of women fall within this 218 new century, old disparities Figure 12.1 Distribution of Weekly Earnings of Men and Women in Barbados, by Earnings Interval, 2004 a. Men b. Women 30 30 20 20 percent percent 10 10 0 0 $3 0–$ 0 $4 0–$ 9 $5 –$ 9 $6 0–$ 9 $7 0–$ 9 $8 0–$ 9 – 9 ,0 0– 99 ov –$1 99 $1 00 00 $3 0–$ 0 $4 0–$ 9 $5 0–$ 9 $6 0–$ 9 $7 –$ 9 $8 0–$ 9 – 9 ,0 0– 99 ov –$1 99 $1 00 00 0 20 0 29 00 39 0 49 0 59 0 69 00 79 0 20 0 29 0 39 0 49 00 59 0 69 00 79 $1 90 $8 9 er ,3 ,3 $1 $90 $8 9 er ,3 ,3 $2 er $ 00 $ $2 er $ 00 $ d d un un $ Source: Based on data from 2004 CLFSS. Figure 12.2 Women’s Participation in the Labor Force in Barbados, by Earnings Interval, 2004 70 60 percent 50 40 30 00 0 9 9 9 9 9 9 9 9 0 20 29 39 49 59 69 79 89 99 30 ,3 r$ –$ –$ –$ –$ –$ –$ –$ –$ 1, $1 –$ 00 00 00 00 00 00 00 00 de er 00 $2 $3 $4 $5 $6 $7 $8 $9 un ov ,0 $1 Source: Based on data from 2004 CLFSS. category. This difference complicates the calculation of hourly earnings. The approach adopted here is to adjust the distributions so that weekly hours worked are the same for men and women. Table 12.1 provides the descriptive statistics used in the analysis. It shows that the average gender earnings gap in Barbados reaches 18.9 percent of average women’s earnings. Regarding age, the data indicate a slight predominance of men at both extremes of the age distribution, with a predominance of women among understudied gaps: barbados, 2004 and jamaica, 2003 219 Table 12.1 Demographic and Job Characteristics and Relative Earnings of Men and Women in Labor Force in Barbados, 2004 Earnings index (Base: average Composition (%) women’s earnings = 100) Men Women Men Women All 100 100 118.9 100.0 Personal characteristics Age 15–24 13.9 10.8 75.4 66.7 25–34 23.7 24.8 107.2 98.0 35–44 27.7 31.0 127.5 107.4 45–54 23.6 24.0 134.9 108.1 55–64 11.1 9.3 142.0 99.4 Education None 1.3 0.9 97.6 80.0 Primary 16.1 12.4 89.0 57.7 Secondary 60.0 58.9 107.3 79.0 Tertiary 22.6 27.8 171.2 165.7 Presence of children (12 years or younger) in household No 75.5 70.0 115.2 98.4 Yes 24.5 30.0 130.6 103.9 Presence of other household member with labor income No 28.7 23.0 118.3 107.7 Yes 71.3 77.0 119.2 97.5 Stratum (based on socioeconomic development) 1 (urban) 32.8 32.4 111.9 93.5 2 (mixed) 28.3 29.5 124.8 101.2 3 (mixed) 22.2 23.7 125.5 111.3 4 (rural) 16.7 14.5 113.9 93.6 Job characteristics Type of employment Employer 1.0 0.4 151.9 117.8 Self-employed 17.1 7.9 135.5 104.9 (continued next page) 220 new century, old disparities Table 12.1 (continued) Earnings index (Base: average Composition (%) women’s earnings = 100) Men Women Men Women Public employee 21.9 25.6 138.5 134.4 Private employee 60.0 66.1 107.7 85.9 Occupation Legislators and senior officials 6.5 6.8 189.7 169.9 Professionals 8.2 14.3 203.6 185.2 Technicians and associate professionals 10.1 6.4 157.7 122.7 Clerks 4.9 19.8 121.0 111.6 Service, shop, and market sales workers 12.2 26.8 103.0 66.0 Skilled agricultural and fishery workers 4.4 1.0 83.5 52.8 Craft and related trades workers 23.8 3.0 108.2 67.1 Plant and machine operators and assemblers 8.9 2.8 103.2 58.6 Elementary occupations 21.0 19.1 77.1 52.5 Economic sector Agriculture and mining 4.9 3.5 98.7 61.2 Manufacturing 6.2 6.4 105.7 69.1 Electricity, gas, and water 2.1 1.9 127.1 88.5 Construction 17.6 1.0 104.3 110.9 Wholesale and retail trade and hotels and restaurants 12.1 17.8 107.1 73.1 Transport, storage, and communication 14.2 14.8 116.3 91.2 Finance, insurance, real estate, and business services 22.8 30.2 126.1 101.4 (continued next page) understudied gaps: barbados, 2004 and jamaica, 2003 221 Table 12.1 (continued) Earnings index (Base: average Composition (%) women’s earnings = 100) Men Women Men Women Community, social, and personal services 20.0 24.5 139.8 135.4 Experience Less than 1 year 8.6 11.1 87.2 69.6 1–5 years 33.7 38.6 104.0 91.8 6–10 years 20.1 20.6 114.8 98.6 11–15 years 11.7 9.8 128.6 106.5 16–20 years 7.8 5.9 132.7 117.2 20 or more years 18.1 14.0 154.8 139.3 Source: Based on data from 2004 CLFSS. middle-age (25–54) workers. The data also show that earnings evolve with age in a monotonic way for men whereas women’s earnings increase monotonically up to age 54, after which they decline slightly. Women’s educational achievement surpasses that of men: 27.8 percent of women and 22.6 percent of men completed university. However, at every level of education, men earn more than women. Average earnings for women with no, primary, or secondary education are statistically similar; earnings for women increase markedly only for women with university education. The incidence of children and other labor income earners in the house- hold is higher among women than among men. The earnings premium linked to children living in the household is larger for men than for women, however. The earnings premium linked to the presence of other labor income earners at home is nonexistent for men and negative for women—that is, women who are the sole income earners in their house- holds tend to have higher earnings than women who live with another earner. In the sample design, the 11 parishes in Barbados were grouped into four strata based on socioeconomic development and geographical prox- imity.3 For this chapter, the four strata were reclassified as urban, mixed, and rural. Stratum 1 contains the capital city (Bridgetown), which is clas- sified as urban. Strata 2 and 3 contain parishes that are both suburban and rural (defined as areas with low population density); they are classified as mixed. Stratum 4, which includes the parishes farthest from Bridgetown, 222 new century, old disparities is classified as rural. Earnings are higher in the two mixed strata than in the other two strata for both men and women. This finding reflects the socioeconomic make-up of these regions. The majority of workers in Barbados (82 percent of men and 92 percent of women) are employees. As in most labor markets, most employers are men. Self-employment is also a category dominated by men in Barbados, in sharp contrast with the rest of the developing world, where it is domi- nated by women. The highest-earning men are employers; the highest- earning women work in the public sector. The highest-paid occupational group consists of professionals (8 percent of men and 14 percent of women). The sectors of finance, insurance, real estate, and business services and community, social, and personal services have large shares of women workers (55 percent), with large gender gaps in the business sectors and almost no gaps among social workers. The Role of Individual Characteristics in Explaining the Gender Earnings Gap To what extent do the observed differences in earnings correspond to dif- ferences in observable characteristics that labor markets reward? What would the distribution of men’s earnings look like if their distribution of observable characteristics were exactly the same as the distribution for women? What would the gender earnings gap be in this case? Counterfactual situations are created using the matching technique described in chapter 2. Table 12.2 decomposes the earnings gap for various combinations of observable demographic characteristics. The combina- tions of characteristics are constructed so that each combination builds on the previous one by adding one characteristic. The comparison of the decomposition exercises is analyzed next. First, the Barbados labor market tends to have a larger proportion of prime- age women than men. In a hypothetical world in which men and women have the same age distribution, the gender earnings gap would reach 20.4 percent of average women’s earnings (up from the 18.9 percent observed). A more pronounced result in the same direction is found when consid- ering education as a second matching characteristic. The counterfactual gender earnings gap that would be observed in a world in which men and women have the same distribution of age and education in the labor mar- ket exceeds that observed in the real world by almost 7 percentage points, reaching 25.7 percent of average women’s earnings. Inclusion of the presence of children and other labor income earners in the household does not change the measure of unexplained gender differences in earnings much, but the components attributable to the existence of uncommon supports become pronounced, reaching about Table 12.2 Decomposition of Earnings Gap in Barbados after Controlling for Demographic Characteristics, 2004 (percent) + Presence of other + Presence of children household member with Age + Education in the household labor income + Stratum Δ 18.9 18.9 18.9 18.9 18.9 Δ0 20.4 25.7 25.9 25.0 20.4 ΔM –2.6 –3.6 –10.8 –11.4 –10.4 ΔF 2.7 2.2 9.8 10.8 11.0 ΔX –1.7 –5.4 –5.9 –5.5 –2.1 Percentage of women in common support 96.3 92.6 90.4 86.9 73.7 Percentage of men in common support 97.6 93.0 88.8 83.5 67.7 Source: Based on data from 2004 CLFSS. Note: ΔM (ΔF) is the part of the earnings gap attributed to the existence of men (women) with combinations of characteristics that are not met by any women (men). ΔX is the part of the earnings gap attributed to differences in the observable characteristics of men and women over the “common support.� Δ0 is the part of the earnings gap that cannot be attributed to differences in characteristics of the individuals. It is typi- cally attributed to a combination of both unobservable characteristics and discrimination. The sum of these components equals the total earnings gap (ΔM + ΔF + ΔX + Δ0 = Δ). 223 224 new century, old disparities 10 percent (positive for women and negative for men). Socioeconomic stratum reduces the measure of unexplained earnings gap, maintaining at the same 10 percent level the components attributed to the existence of uncommon supports. The likelihood of finding matches falls as the number of matching char- acteristics increases (as shown in the last two rows of table 12.2). Linked to this result is the fact that the measures of the gender earnings gap can be attributed to the existence of men and women with unmatchable charac- teristics, whose number grows as the number of matching characteristics increases. In contrast to what is typically observed in this decomposition in other countries in Latin American and the Caribbean, the ΔM component (the part of the earnings gap attributed to the existence of men with combina- tions of characteristics that are not found in any women) is negative and the ΔF component (the part of the earnings gap attributed to the existence of women with combinations of characteristics that are not met by any men) is positive. Men whose characteristics cannot be compared with those of women in the labor market tend to have lower earnings than men whose characteristics are matchable. This pattern is shown in figure 12.3, which reports the percentages of unmatched women in each earnings bracket. The two extremes of the earnings distribution have the largest percentages of unmatched women. This result may suggest some segmentation in the labor market, in which there are low-earning men at the bottom extreme of the earnings distribu- tion and high-earning women at the other extreme. Table 12.3 adds job characteristics to the demographic characteris- tics used in table 12.2. The variables are added separately, in order to facilitate exploration of the effects of each variable and avoid the “curse of dimensionality.�4 The results show that sector is the job characteristic that best explains the gender earnings gap. Inclusion of this variable as a matching charac- teristic reduces the unexplained component of the gap from 20.4 percent to 14.1 percent of average women’s earnings. Thus, elimination of gender segregation by sector would reduce more than 6 percentage points of the gender earnings gap. Another variable that helps explain gender earnings gaps in Barbados is experience. Elimination of gender differences in experience would reduce the gender earnings gap by about 2.5 percentage points. Reduction of occupational segregation by gender would not reduce the gender earnings gap. On the contrary, elimination of gender occupational segregation is linked to an increase of more than 6 percentage points in the gender earnings gap. Type of employment does not change the decomposition of the earnings gap. However, there are differences in unexplained earnings gaps across types of employment (as shown in the next section). understudied gaps: barbados, 2004 and jamaica, 2003 225 Figure 12.3 Proportion of Unmatched Women in Barbados, by Earnings Interval, 2004 50 percent 40 30 20 10 0 9 9 9 9 9 9 9 9 0 00 20 29 39 49 59 69 79 89 99 30 ,3 r$ –$ –$ –$ –$ –$ –$ –$ –$ 1, $1 –$ 00 00 00 00 00 00 00 00 de er 00 $2 $3 $4 $5 $6 $7 $8 $9 un ov ,0 $1 Source: Based on data from 2004 CLFSS. The last column of table 12.3 includes the full set of matching variables (the five demographic characteristics and the four job characteristics). As shown in the last two rows of that column, only about 3 percent of women and men can be compared when using this set of nine matching characteristics. Exploring the Unexplained Component of the Gender Earnings Gap Table 12.4 shows the magnitude of the unexplained earnings gap for dif- ferent segments of the labor market (and using different sets of matching characteristics). As before, the matching variables are added sequentially but with replacement as one moves to the right of the table. Regarding age, the evidence seems to be mixed. When using only demo- graphic characteristics, the unexplained gender earnings gap increases with age. When using the full set of matching characteristics, however, the situation is almost reversed. The results show more consistency regarding education. For all sets of matching characteristics shown in the table, the unexplained gaps are smaller (and in some cases even negative) among university graduates. With regard to the effect of children in the household, for almost all sets of matching characteristics, the unexplained earnings gaps seem to be larger among workers with no children at home. When considering experience as a matching variable, however, the result is reversed. After accounting for experience (in the last two columns of the table), workers 226 Table 12.3 Decomposition of Gender Earnings Gap in Barbados after Controlling for Demographic and Job Characteristics, 2004 (percent) Demographic & & & & set Type of employment Occupation Sector Experience Full set Δ 18.9 18.9 18.9 18.9 18.9 18.9 Δ0 20.4 20.4 26.7 14.1 17.8 15.3 ΔM –10.4 –2.5 –38.4 –41.1 –27.4 –68.0 ΔF 11.0 1.8 31.7 45.4 29.1 72.1 ΔX –2.1 –0.8 –1.1 0.5 –0.6 –0.5 Percentage of women in common support 73.7 56.8 30.6 35.2 44.6 3.3 Percentage of men in common support 67.7 50.9 24.3 30.5 41.4 2.7 Source: Based on data from 2004 CLFSS. Note: ΔM (ΔF) is the part of the earnings gap attributed to the existence of men (women) with combinations of characteristics that are not met by any women (men). ΔX is the part of the earnings gap attributed to differences in the observable characteristics of men and women over the “common support.� Δ0 is the part of the earnings gap that cannot be attributed to differences in characteristics of the individuals. It is typically attributed to a combination of both unobservable characteristics and discrimination. The sum of these components equals the total earnings gap (ΔM + ΔF + ΔX + Δ0 = Δ). Table 12.4 Unexplained Gender Earnings Gap in Barbados after Controlling for Demographic and Job Characteristics, 2004 (percent) Demographic & Type of & & & set employment Occupation Sector Experience Full set All 20.4 20.4 26.7 14.1 17.8 15.3 Personal characteristics Age 15–24 13.1 14.2 42.1 22.2 12.8 20.4 25–34 8.0 9.7 20.8 6.1 11.8 11.9 35–44 16.3 13.0 19.9 11.0 12.3 25.8 45–54 22.3 23.2 29.1 17.0 32.0 12.1 55–64 38.9 28.1 23.9 32.5 22.9 10.2 Education None 40.5 –41.2 55.4 134.1 –41.2 55.4 Primary 47.0 32.8 32.6 39.3 44.5 16.3 Secondary 35.7 31.9 37.7 21.8 34.5 19.4 Tertiary –0.3 –0.9 9.5 1.0 –1.7 7.5 Presence of children (12 years or younger) in Household No 21.8 21.6 27.3 14.4 16.5 15.3 227 Yes 14.6 13.6 20.6 11.2 19.8 23.9 (continued next page) 228 Table 12.4 (continued) Demographic & Type of & & & set employment Occupation Sector Experience Full set Presence of other household member with labor income No 12.0 11.1 19.5 5.9 15.9 26.3 Yes 22.8 22.2 27.6 15.8 17.7 15.2 Stratum (based on socioeconomic development) 1 (urban) 24.1 17.2 24.3 18.2 21.7 16.2 2 (mixed) 17.1 21.8 21.7 10.8 11.8 13.5 3 (mixed) 9.5 9.2 23.1 7.3 14.3 –3.2 4 (rural) 26.0 28.2 38.9 11.1 25.9 21.2 Job characteristics Type of Employment Employer 0.0 0.0 Self-employed 23.0 17.3 Public employee 5.9 1.6 Private employee 16.1 24.4 Occupation Legislators and senior officials 2.9 0.1 Professionals 5.9 2.7 Demographic & Type of & & & set employment Occupation Sector Experience Full set Technicians and associate 16.6 4.8 professionals Clerks 13.1 15.5 Service, shop, and market 43.0 27.5 sales workers Skilled agricultural and 8.4 8.4 fishery workers Craft and related trades 52.1 23.1 workers Plant and machine operators 76.9 197.7 and assemblers Elementary occupations 39.0 24.8 Economic sector Agriculture and mining 26.4 56.8 Manufacturing 37.7 41.3 Electricity, gas, and water 26.5 26.5 Construction –14.6 –54.9 Wholesale and retail trade and 19.0 22.9 229 hotels and restaurants (continued next page) 230 Table 12.4 (continued) Demographic & Type of & & & set employment Occupation Sector Experience Full set Transport, storage, and 20.5 20.9 communication Finance, insurance, real 11.8 21.3 estate, and business services Community, social, and 5.2 1.5 personal services Experience Less than 1 year 12.2 24.0 1–5 years 15.1 16.5 6–10 years 15.7 11.2 11–15 years 16.5 28.5 16–20 years 11.9 36.5 20 or more years 19.5 2.3 Source: Based on data from the 2004 CLFSS. Note: Blank cells appear when the related variable(s) is(are) not used as controls. understudied gaps: barbados, 2004 and jamaica, 2003 231 living with children at home show larger unexplained earnings gaps than other workers. The data also suggest that when no other labor income earner lives at home, earnings differences between men and women are smaller. This finding holds true for all sets of matching characteristics except the one that uses the full set of nine variables. The third stratum shows the smallest unexplained gender earnings gaps. Although type of employment does not explain much of the gender earnings gap in Barbados in the aggregate (see table 12.3), some differences in earnings gaps within types deserve highlighting. Unexplained earnings gaps are larger among the self-employed and private sector employees. They are larger among clerks, craft workers and workers in related trades, plant and machine operators and assemblers, and workers in elementary occupations. Among high-skilled occupations (professionals and senior officials), unexplained earnings gaps are smaller and in some cases close to zero. This finding is consistent with the finding that unexplained gaps are smallest among university graduates. The economic sectors with the largest unexplained gender earnings gaps are manufacturing, agriculture, and mining. The gender earnings gap among community, social, and personal service workers is almost zero. The construction sector, which is dominated by men in most economies, deserves special mention. In Barbados, 18 percent of men and just 1 per- cent of women work in construction. The few women who participate in construction, however, have higher earnings than their male peers. One possible explanation for this phenomenon is that the few women who dare work in segments of the labor market dominated by men represent a selected subsample with unobservable traits (such as work ethic, commit- ment, and motivation) that are rewarded in the market. As a result, these women work as managers. The differences in unexplained earnings gaps across the experience lad- der are mixed. Controlling for the set of demographic characteristics plus experience yields larger unexplained earnings gaps among the most expe- rienced workers. However, when using the full set of control variables, the unexplained gaps among the most experienced workers are the smallest. The interplay of experience with the other demographic and job charac- teristics should be taken into account when trying to use this variable as an explanatory source for gender earnings gaps. Jamaica: Women in the Middle, Men at Both Ends The data employed in the estimation for Jamaica are from the 2003 Labor Force Survey undertaken by the Statistical Institute. These quarterly sur- veys sample about 1 percent of the population. The sample enumerates households spread across Jamaica’s 14 parishes, drawing a representative 232 new century, old disparities mix of urban and rural dwellers. The original sample for 2003 contained 22,692 observations; following data cleaning and deletion of observations with missing values, 4,974 observations remained in the final sample; earnings are measured as hourly earnings. Table 12.5 shows descriptive statistics for Jamaica. Having normalized average women’s earnings to 100, average men’s earnings can be directly read as the measure of gender earnings gaps. Men’s earnings below 100 indicate a negative earnings gap. On average, women earn more than men in Jamaica. However, the earnings difference is very small (0.8 percent of average women’s earn- ings), and a significance test would fail to reject the null hypothesis of gender equality in earnings. Regarding age, the recurrent pattern for most countries in the region of higher prevalence of men at both extremes of the age distribution is evident in Jamaica. The pattern of earnings progres- sion along the life cycle is also similar to the pattern observed in other countries. In terms of education, 12.3 percent of working women and just 4.5 percent of working men completed university. Earnings for each level of schooling below university show little variation. It is only university graduates, especially men, whose earnings are significantly higher. The presence of children is much more prevalent among working women than among working men. Whereas for working men there are no earnings differences between men who live with children at home and men who do not, for working women the presence of children is linked to lower earnings. Working women are also more likely than working men to live in urban areas. Unlike elsewhere in Latin American and the Caribbean, self- employment in Jamaica has higher participation of men than women, and dependent relationships, in both the private and public sectors, are more prevalent among women. As in most countries, the data show no gender earnings differences in earnings in public sector employment; surprisingly, no gender earnings differences are evident in self-employment either. The segments of the labor markets showing earnings disparities in favor of men are private employment and, to a greater extent, employers. Occupational segregation is also prevalent in Jamaica. Women tend to be overrepresented among professionals, elementary occupations, ser- vices, and store and market sales workers. Men tend to be overrepresented among skilled agricultural and fishery workers, craft workers and workers in related trades, and plant and machine operators and assemblers. Women tend to work in wholesale and retail trade; hotels and restaurants; and community, social, and personal services. Men are engaged in agriculture, mining, and construction. The highest-paying occupations for both men and women are in the professional sector. The highest-paying activities for men are electricity, gas, and water; for women, the highest-paying activities are finance, insurance, real estate, and business services. There are some gender differences in job tenure. Two-thirds of men have been at their job for five years or more; the corresponding figure for understudied gaps: barbados, 2004 and jamaica, 2003 233 Table 12.5 Demographic and Job Characteristics and Relative Hourly Earnings of Men and Women in Jamaica’s Labor Force, 2003 Earnings Index (Base: average Composition women’s earnings = (%) 100) Men Women Men Women All 100 100 99.2 100.0 Personal characteristics Age 15–24 16.9 13.2 84.8 99.3 25–34 28.9 30.4 98.5 107.1 35–44 26.8 30.7 109.7 94.6 45–54 17.7 17.3 102.2 104.6 55–64 9.7 8.4 91.9 85.6 Education None 0.2 0.2 64.9 62.0 Primary 26.3 19.7 79.8 73.0 Secondary 69.0 67.8 97.5 87.1 Tertiary 4.5 12.3 240.0 214.8 Presence of children (12 years or younger) in household No 70.2 61.6 98.6 106.2 Yes 29.8 38.4 100.6 90.0 Presence of other household member with labor income No 49.8 42.0 98.8 107.5 Yes 50.2 58.0 99.6 94.6 Urban No 64.1 53.6 84.6 81.1 Yes 35.9 46.4 125.2 121.8 Job characteristics Type of employment Employer 2.4 1.6 171.8 140.7 Self-employed 42.5 28.7 71.3 70.1 Public employee 9.2 17.1 160.7 156.6 Private employee 45.9 52.5 108.8 96.6 (continued next page) 234 new century, old disparities Table 12.5 (continued) Earnings index (Base: average Composition women’s earnings = (%) 100) Men Women Men Women Occupation Armed forces 0.3 0.0 161.3 — Legislators and senior 2.6 7.6 166.5 121.1 officials Professionals 3.9 10.3 212.6 208.5 Technicians and associate 3.6 5.8 175.8 124.3 professionals Clerks 2.5 12.1 133.5 122.2 Service, shop, and 11.1 24.4 108.9 86.3 market sales workers Skilled agricultural and 30.8 8.1 58.1 47.2 fishery workers Craft and related trades 22.2 4.0 116.8 70.5 workers Plant and machine 8.5 1.7 113.7 78.6 operators and assemblers Elementary occupations 14.4 26.0 74.5 70.2 Economic sector Agriculture and mining 34.2 9.7 63.4 50.9 Manufacturing 7.0 4.9 109.2 90.7 Electricity, gas, and water 0.4 0.3 165.6 113.9 Construction 15.8 0.7 116.9 92.7 Wholesale and retail trade and 13.8 32.3 94.3 81.9 hotels and restaurants Transport, storage, and 8.0 2.7 119.9 157.1 communication Finance, insurance, real 4.4 5.7 133.1 177.2 estate, and business services Community, social, and 16.3 43.8 136.1 111.8 personal services Experience Less than 3 months 1.9 2.3 84.3 80.2 (continued next page) understudied gaps: barbados, 2004 and jamaica, 2003 235 Table 12.5 (continued) Earnings index (Base: average Composition women’s earnings = (%) 100) Men Women Men Women 3–6 months 2.0 4.4 114.3 76.1 6–9 months 2.3 3.0 82.2 83.0 9–12 months 2.5 3.2 83.0 68.1 1–2 years 6.0 7.7 99.7 91.6 2–5 years 18.5 22.4 97.7 98.1 5 years or more 66.9 57.0 100.7 107.2 Small firm (five workers or less) No 41.8 43.4 129.4 134.7 Yes 58.2 56.6 77.5 73.3 Time worked Part time 6.0 12.6 99.0 85.7 Full time 72.7 74.6 102.5 105.1 Overtime 21.3 12.8 87.9 84.1 Source: Based on data from the 2003 Labor Force Survey. women is 57 percent. Gender differences in regular time worked per week are also substantial. Women dominate part-time work, and men dominate overtime work. The Role of Individual Characteristics in Explaining the Gender Earnings Gap Tables 12.6 and 12.7 show the earnings gap decompositions exercise. Each column shows one decomposition, based on a set of matching variables. The first table uses only demographic characteristics, adding them sequentially without replacement as one moves to the right. The second table adds job characteristics to the set of demographic ones; in order to avoid the curse of dimensionality, it does so with replacement. The last column of table 12.7 uses the full set of demographic and job characteristics. The first thing to note is that the –0.8 percent earnings gap in Jamaica for the overall economy masks the fact that women have more schooling than men and are not compensated for it appropriately. When comparing 236 Table 12.6 Decomposition of Gender Earnings Gap in Jamaica after Controlling for Demographic Characteristics, 2003 (percent) + Presence of other + Presence of children household member with Age + Education in the household labor income + Stratum Δ –0.8 –0.8 –0.8 –0.8 –0.8 Δ0 0.2 12.2 11.0 9.7 12.0 ΔM 0.0 –0.1 0.1 0.1 2.2 ΔF 0.0 –1.8 –2.8 –4.4 –7.8 ΔX –1.0 –11.1 –9.1 –6.2 –7.2 Percentage of women 100.0 99.5 98.0 94.8 88.7 in common support Percentage of men in 99.9 98.3 96.9 94.3 88.7 common support Source: Based on data from the 2003 Labor Force Survey. Note: ΔM (ΔF) is the part of the earnings gap attributed to the existence of men (women) with combinations of characteristics that are not met by any women (men). ΔX is the part of the earnings gap attributed to differences in the observable characteristics of men and women over the “common support.� Δ0 is the part of the earnings gap that cannot be attributed to differences in characteristics of the individuals. It is typically attributed to a combination of both unobservable characteristics and discrimination. The sum of these components equals the total earnings gap (ΔM + ΔF + ΔX + Δ0 = Δ). Table 12.7 Decomposition of Gender Earnings Gap in Jamaica after Controlling for Demographic and Job Characteristics, 2003 (percent) & & Demographic & Type of & & & Firm Time set employment Occupation Sector Experience size worked Full set Δ –0.8 –0.8 –0.8 –0.8 –0.8 –0.8 –0.8 –0.8 Δ0 12.0 15.3 16.5 14.6 11.2 7.9 12.4 16.8 ΔM 2.2 5.7 6.6 5.9 1.1 3.9 0.9 18.8 ΔF –7.8 –15.3 –15.5 –14.0 –6.1 –8.6 –6.4 –24.8 ΔX –7.2 –6.5 –8.4 –7.3 –7.1 –4.0 –7.7 –11.6 Percentage of women 88.7 69.4 37.4 41.3 70.7 79.7 72.4 5.8 in common support Percentage of men in 88.7 72.6 46.3 52.0 64.9 82.5 74.0 5.5 common support Source: Based on data from 2003 Labor Force Survey. Note: ΔM (ΔF) is the part of the earnings gap attributed to the existence of men (women) with combinations of characteristics that are not met by any women (men). ΔX is the part of the earnings gap attributed to differences in theobservable characteristics of men and women over the “common support.� Δ0 is the part of the earnings gap that cannot be attributed to differences in characteristics of the individuals. It is typically attributed to a combination of both unobservable characteristics and discrimination. The sum of these components equals the total earnings gap (ΔM + ΔF + ΔX + Δ0 = Δ). 237 238 new century, old disparities men and women with the same age and education, the unexplained dif- ferences in earnings reach 12.2 percent of average women’s earnings in favor of men. To a lesser extent, the inclusion of the presence of another labor income earner in the household as a matching variable reduces the explained gender differences in earnings. The addition of other demo- graphic controls does not alter much the unexplained gaps. The measure of the common supports is nearly 90 percent of men and women. The addition of job characteristics changes the panorama a bit. Two job characteristics that do not greatly change the measure of unexplained earnings gap are tenure and time worked. One variable, firm size, markedly reduces this measure. A hypothetical world in which all gender differences in firm size of workers were eliminated would reduce the unexplained differences in earnings by 4 percentage points. For type of employment, occupation, and economic sector, reduction of gender differences would increase unexplained gender earnings gaps. Elimination of gender occupational segregation in Jamaica would increase the gender earnings gap by 4.5 percentage points. Jamaica is thus another country in the region in which a reduction of gender occu- pational segregation seems to be the wrong target for reducing gender earnings gaps. The matching exercise including occupation as a match- ing variable leads to the smallest measures of the common supports among all job characteristics—that is, gender occupational segrega- tion is a prevalent feature in Jamaican labor markets. However, reduc- ing this segregation may have detrimental effects on gender earnings disparities. The decomposition exercises for all countries (in this and previous chap- ters) except Barbados exhibit components caused by the lack of common support that are positive for men and negative for women. Jamaica reveals the same behavior as the rest of the region, with unmatched men and women earnings more than the national average. The matching decomposition exercise after the inclusion of all demo- graphic and job characteristics is shown in the last column of table 12.7. The results are qualitatively similar to but stronger than the results shown after the inclusion of job variables. The measures of the common supports become smaller: only about 5 percent of men and women are fully compa- rable under these sets of demographic and job characteristics. Exploring the Unexplained Component of the Gender Earnings Gap The exploration of unexplained differences in earnings along the distri- bution of income reveals a pattern similar to that found in other Latin American and Caribbean countries. The unexplained gap is larger among understudied gaps: barbados, 2004 and jamaica, 2003 239 lower-income workers, suggesting that the problem of earnings gaps is linked to the problem of low income generation and hence poverty. For some intermediary percentiles (10th–20th), the earnings gap attains a minimum and increases thereafter. The original gap is larger than the gap obtained after controlling for observable characteristics from the first to the seventh percentiles. After these percentiles, the situation is as similar to that in other Latin American and Caribbean countries: the controlled earnings gaps is larger than the original one, as women have completed more schooling (figure 12.4). To conclude the analysis for Jamaica, a description of unexplained gender differences in earnings for different segments of the labor mar- kets is presented (for a complete set of graphs reporting these results, see Bellony, Hoyos, and Ñopo 2010). Some patterns found in Jamaica are similar to those found in other countries. Unexplained gender differ- ences in earnings increase with age (although most of the differences are not statistically significant) and show an inverted U-shape with respect to education (where the largest unexplained gaps are found among high school graduates). Workers with young children at home experience larger unexplained gender differences in earnings. The presence of other income earners at home is also linked with larger gender disparities, but the result is not statistically significant. In some other aspects, Jamaica shows peculiarities with respect to the distribution of unexplained earnings gender differences along segments Figure 12.4 Unexplained Gender Earnings Gap in Jamaica, after Controlling for Demographic Characteristics, by Percentile of Earnings Distribution, 2003 100 percentage of average women’s 50 earnings 0 −50 0 10 20 30 40 50 60 70 80 90 100 earnings percentile original gap age and education full set of demographic characteristics Source: Based on data from 2003 Labor Force Survey. 240 new century, old disparities of the labor market. Earnings gaps are similar in urban and rural areas and across types of employment, occupations, economic sectors, firm size, and time worked. Regarding type of employment, however, there is huge heterogeneity within the “employer� category. The only seg- ment of the market for which there seems to be statistically significant differences in earnings is private sector employment. Four occupational categories show statistically positive unexplained gender differences in earnings, and five others are not distinguishable from zero. Among legis- lators, technicians, and machine operators, gender disparities in earnings are very heterogeneous. Something similar happens across economic sec- tors, where the only categories with statistically significant earning gaps are agriculture and social services and, to a lesser extent, trade, hotels, and restaurants. The dispersion of unexplained gender differences in earnings is greater among part-time and overtime workers than among people who work full time. On average, differences in earnings are smaller for part-time and overtime workers than for full-time workers; when dispersion is considered, however, the differences are not statistically significant. Unex- plained gender gaps are larger in small firms than in larger firms, as in most countries in the region, but these differences are not statistically significant in Jamaica. The last point to highlight is job tenure, which seems to have no link to gender differences in earnings. The data show, however, some gender gaps among people with 3–6 months and 9–12 months of job tenure. For workers at the top of the distribution of job tenure (five years or more), the unexplained gender earnings gap is positive and statistically significant. Women are less able to accumulate enough occupational experience, and when they do accumulate that experience, they earn substantially less than their male counterparts. Summary This chapter explores gender earnings gaps in two Caribbean econo- mies, Barbados and Jamaica, emphasizing the similarities and differences between the two countries as well as between them and the rest of Latin American and the Caribbean. In both countries, as in most of the region, women’s educational achievement is greater than that of men. Jamaica shows lower educational achievement and larger gender disparities than Barbados. Nonetheless, men’s earnings surpass those of their female peers. Comparison of earnings for men and women with the same age and educa- tion reveals that men earn 25 percent more than women in Barbados and 12 percent more in Jamaica. The unexplained gender earnings gaps after adding extra control variables are larger in Barbados than in Jamaica. understudied gaps: barbados, 2004 and jamaica, 2003 241 Both countries confirm a finding that is recurrent in the analysis of gender earnings gaps conducted with this matching approach and that challenges some popular beliefs about gender occupational segregation—namely, the notion that elimination of gender occupational segregation would increase rather than reduce gender earnings gaps. Occupational segregation seems to be one of the wrong culprits the literature has emphasized. Both Barbados and Jamaica show the smallest unexplained earnings gaps among the high skilled and the largest gaps among the low skilled. Regarding segregation by economic sectors, the evidence for Barbados and Jamaica is also in line with what has been found in other countries in the region, and it is mixed. The results indicate that elimination of gen- der sector segregation would reduce the observed gender earnings gap in Barbados but increase it in Jamaica. Occupational experience in Barbados and job tenure in Jamaica help explain gender earnings gaps. Elimination of gender disparities in these variables is linked to a reduction of 1–2 percentage points in unexplained earnings gaps. The data coding of earnings in intervals poses some challenges to the analysis of Barbados. Thanks to the nonparametric nature of the matching approach used, however, most of the analysis can be performed as it is when earnings are coded as a continuous variable. One of the few results that cannot be replicated is the exploration of unexplained earnings gaps along percentiles of the earnings distribution. This result is available only for Jamaica, where gender earnings gaps are larger among low-income work- ers, as in most of Latin America and the Caribbean. This finding suggests linkages between gender earnings disparities and low income generation (or poverty). Reducing these inequities would also help reduce poverty. Another issue that calls for further exploration is ethnicity. Some coun- tries in the region have large indigenous and Afro-descendant populations that are worse off in many measures of well-being. Chapter 13 presents an overview of the issue, based on data on Bolivia, Brazil, Chile, Ecuador, Guatemala, Paraguay, and Peru. Chapters 14–16 explore ethnic earnings disparities in Brazil, Ecuador, and Guatemala. It is suggestive of a problem that among 18 countries with data on gen- der earnings differences, only 7 have data on ethnic earnings differences. The paucity of data may reflect the invisibility of these populations or the lack of interest in their situation on the part of policy makers. However, the analysis is relevant, as ethnic “minorities� make up an important fraction of the Latin American population and participation of ethnic minorities in the labor market is considerable in most of them. Ethnic minorities in both Bolivia and Brazil make up about half the labor force; they represent more than 30 percent of the working population in Guatemala, Paraguay, and Peru. Ten percent of Ecuador’s work force and 5 percent of Chile’s are ethnic minorities. 242 new century, old disparities Notes 1. Of the 21 studies in the edited volume of Psacharopoulos and Tzannatos (1992), just one examines a Caribbean country (Jamaica). 2. The CHSS was later changed to the Continuous Labor Force Sample Survey (CLFSS) 3. Stratum 1: St. Michael; Stratum 2: Christ Church, St. Phillip; Stratum 3: St. George, St. James, St. Thomas; Stratum 4: St. John, St. Joseph, St. Andrews, St. Peter, St. Lucy. 4. The curse of dimensionality refers to the fact that the likelihood of finding female-male matches decreases as the number of control variables (the “dimen- sion�) increases. This is a problem because researchers would like to use the maxi- mum number of observable characteristics in order to control the scope of the role of unobservable factors in explaining the earnings gap. References Bellony, A., A. Hoyos, and H. Ñopo. 2010. “Gender Earnings Gaps in the Caribbean: Evidence from Barbados and Jamaica.� RES Working Paper 4683, Inter-American Development Bank, Research Department, Washington, DC. Brendan, D. 1991. “Male and Female Wage Differentials in Haiti.� Graduate School of Public and International Affairs, University of Pittsburgh, Pittsburgh. Coppin A. 1996. “An Analysis of Earnings in Barbados by Age and Sex.� Eco- nomic Review (Central Bank of Barbados) 23 (3): 14–21. Hotchkiss, J., and R. Moore. 1996. “Gender Compensation Differentials in Jamaica.� Economic Development and Cultural Change 44 (3): 657–76. Hoyos, A., A. Bellony, and H. Ñopo. 2010. “Gender Earnings Gaps in the Carib- bean: Evidence from Barbados and Jamaica.� IDB Working Paper IDB-WP- 210, Inter-American Development Bank, Washington, DC. Ministry of Education, Youth, and Culture. 2001. “Language Education Policy.� Ministry of Education, Youth, and Culture of Jamaica, Kingston. http://www .moec.gov.jm/policies/languagepolicy.pdf. Olsen, R. N., and A. Coppin. 2001. “The Determinants of Gender Differentials in Income in Trinidad and Tobago.� Journal of Development Studies 37 (5): 31–56. Psacharopoulos, G., and Z. Tzannatos, eds. 1992. Vol. 2 of Women’s Employment and Pay in Latin America. Washington, DC: World Bank. Scott, K. 1992. “Female Labor Force Participation and Earnings: The Case of Jamaica.� In Case Studies on Women’s Employment and Pay in Latin America, ed. G. Psacharopoulos and Z. Tzannatos, 323–38. Washington, DC: World Bank. Sookram, S., and P. Watson. 2008. “The Informal Sector and Gender in the Carib- bean: The Case of Trinidad, and Tobago.� Journal of Eastern Caribbean Studies 33 (4): 42–66. Terrell, K. 1992. “Female-Male Earnings Differentials and Occupational Struc- ture.� International Labor Review 131 (4/5): 387–98. UNDP (United Nations Development Programme). 2009. Human Development Report 2009. New York: UNDP. Part III Ethnic Earnings Gaps 13 Overlapping Disadvantages: Ethnicity and Earnings Gaps in Latin America Gender earnings gaps have been the subject of much analytical work; the study of ethnic earnings gaps has been somewhat constrained, partly because of limited data, especially in household surveys and national cen- suses. Only nine countries in Latin America include an “ethnic� question in their national censuses and seven include it in their national household surveys. These questions usually refer to mother tongue or self-identification with an ethnic group (table 2.2 in chapter 2 describes the survey questions used in each country to identify individuals from ethnic minorities). Another important constraint is the number of people belonging to ethnic “minori- ties� (often majorities) who are not officially registered or lack an identity document. Invisibility in national statistics and systems for delivering public services is a sign of the inferior situation in which ethnic minorities often live. Despite these constraints, studies of ethnic earnings gaps have been made. Because of the importance of the interplay between ethnic and gen- der earnings gaps, the analysis in this chapter and the following ones frequently refer to comparisons with information presented in chapters 4–12, on gender differentials. The studies on ethnic earnings gaps try to This chapter was adapted from the following sources: “New Century, Old Dis- parities: Gender and Ethnic Wage Gaps in Latin America,� Juan Pablo Atal, Hugo Ñopo, and Natalia Winder, RES Working Paper 4640, Inter-American Develop- ment Bank, 2009; Evolution of Gender Wage Gaps in Latin America at the Turn of the Twentieth Century: An Addendum to ‘New Century, Old Disparities,’� Hugo Ñopo and Alejandro Hoyos, IZA Discussion Papers 5086, Institute for the Study of Labor, 2010. Juan Pablo Atal is a graduate student in economics at the University of Califor- nia, Berkeley, and Natalia Winder is a consultant at UNICEF, Division of Policy and Practice, New York. Alejandro Hoyos is a consultant at the Poverty Reduction and Economic Management Network (PREM) at the World Bank. 245 246 new century, old disparities use the same formats, measures, and methodologies used to analyze gen- der differentials. However, data on ethnic gaps are available for only 7 of the 18 countries examined elsewhere in this book: Bolivia, Brazil, Chile, Ecuador, Guatemala, Paraguay, and Peru; as usual, earnings are computed as hourly earnings in the main job. What Does the Literature Show? Some 28–34 million indigenous people live in Latin America, represent- ing roughly 10 percent of the population (Hall and Patrinos 2006). In all countries, these groups are disproportionately represented among the poor and extreme poor, a situation that has not changed significantly over time. Moreover, since the 1990s, despite decreasing poverty rates in most countries in the region, poverty among indigenous groups either increased or declined at a significantly slower pace than in the rest of the popula- tion (Psacharopoulos and Patrinos 1994; Jiménez, Casazola, and Yáñez Aguilar 2006). On average, 63–69 percent of the indigenous population in the region is economically active. Indigenous people are overrepresented among the self-employed and in the agricultural sector. Despite higher levels of labor force participation over time, in most countries their earnings are sig- nificantly lower than those of their nonindigenous peers. This gap nar- rowed in the past decade, but it remains high in some countries, including Bolivia, Brazil, Chile, and Guatemala (ILO 2007). Attempts to explain ethnic earnings gaps have analyzed differences in human capital, especially education but also age, migrant status, and the interplay of ethnicity and gender. Despite improvements in educational attainment, indigenous groups earn significantly less than their nonindig- enous counterparts (Psacharopoulos 1992). Although low education indica- tors may explain much of the persistent ethnic earnings differential in some countries, productive characteristics explain only half the earnings gap in other countries (Patrinos 2000). Rangel (1998) explores indicators such as quality of education, measured in terms of certification of teachers, teacher/ pupil ratio, and materials, as potential drivers of ethnic earnings differen- tials in the region. Hall and Patrinos (2006) consider differences in returns by levels of education. None of these studies fully explain pay differentials. Rangel (1998) shows that indigenous groups tend to be concentrated in low-paid sectors and low-skilled and low-paid jobs. One possible explana- tion for this concentration could be the impact of social networks, which may have a significant influence on the economic sector, type, and even quality of jobs obtained by indigenous workers, especially migrants. This factor is subject to significant heterogeneity across countries and across eth- nic groups within countries (Hall and Patrinos 2006; Fazio 2007). overlapping disadvantages: ethnicity and earnings gaps 247 The literature also examines the impact of proficiency in the domi- nant language (Chiswick, Patrinos, and Hurst 2000) and regional dif- ferences (Contreras and Galván 2003). Important issues, such as the significant share of rural income represented by unsalaried labor and the socioeconomic dynamics of indigenous people in urban zones remain unexplored. Analysis of many topics has been constrained to country case studies, limiting the conclusions to a specific labor market and earnings structure. Most authors agree, however, that human capital endowments are a criti- cal contributor to earnings differences. Significant narrowing of earnings gaps could be achieved if interventions focus on improving human capi- tal accumulation by indigenous peoples while exploring complementary policies to increase their return on investments in human capital (Hall and Patrinos 2006). The interplay of ethnicity and gender is of crucial importance: one of the most recurrent stylized facts is that indigenous women appear to fare worst in labor markets. Statistics in this area are unreliable, how- ever, and large discrepancies exist across sources. Indigenous women represent 20–35 percent of the population in Bolivia and Guatemala and 0.2–5.0 percent in Brazil, Ecuador, and Panama. They represent about 25–50 percent of the economically active population in some countries, not including people involved in unpaid work (Calla 2007). Despite increases in female labor force participation and earnings, indigenous women persistently remain at the bottom of the earnings distribution, showing the highest levels of poverty and exclusion (Piras 2004). In Bolivia, for example, being indigenous and female is the most unfavor- able condition when entering the labor market and securing earnings (Contreras and Galván 2003). Latin America also has a large population of African descent: 150 mil- lion people. Most of these people live in Brazil (50 percent of the regional total), Colombia (20 percent), and Républica Bolivariana de Venezuela (10 percent) (Hopenhayn and Bello 2001). Brazil’s Afro-descendent pop- ulation is the largest in the region. It suffers more from unemployment, low earnings, and glass ceilings than the rest of the population. Occupational differences by race are evident. In 1988 in Rio de Janeiro, 81 percent of Afro-descendent men (and about 60 percent of whites) worked in manual occupations. Among women who worked, the share of domestic workers was 40 percent among Afro-descendents and 15 per- cent among whites (Rangel 1998, using data from the 1988 Pesquisa Nacional por Amostra de domicílios [PNAD]). Gender earnings gaps are also important among the Afro-descendent population (Hopenhayn and Bello 2001). Despite their achievements in education and occupational attainment, Afro-Brazilian women continue to earn significantly less than men (Lovell 2000). 248 new century, old disparities How Do Ethnic Minorities and Nonminorities in the Work Force Differ? Wide earnings disparities are evident between minorities and nonminori- ties in the seven countries for which data were available (table 13.1). Minorities have significantly lower educational attainment than non- minorities. As in the gender case, disparities are evident in type of employ- ment and occupation. However, ethnic differences in economic sectors are substantially smaller than along the gender divide. Also in contrast to the gender case, there are important ethnic differences in firm size: less than half of nonminorities and almost three-quarters of minorities are employed in firms with five or fewer workers. The Role of Individual Characteristics in Explaining the Ethnic Earnings Gap How much of the earnings gap is explained by the striking differences in observable characteristics of minorities and nonminorities just shown? To answer this question, the analysis decomposes ethnic earnings gaps fol- lowing the strategy developed for gender1. In order to make the ethnic earnings gap decompositions comparable to those reported along the gender dimension, it is necessary to decompose the gender earnings gap using only the seven countries used in the ethnic analysis (Atal, Ñopo, and Winder 2009). This subsample of countries dis- plays wider gender earnings gaps than the region as a whole (15.7 percent compared with the 10.0 percent reported in table 4.3 in chapter 4). The wider gaps reflect the fact that gender earnings gaps are large in Brazil, Paraguay, and Peru. Controlling for ethnicity alone provides little explanation for gender gaps. The results in table 13.2 are qualitatively similar to those reported in table 4.3, with a jump in the unexplained component of the gap after add- ing education as a matching variable. The set of matching variables and the sequence in which these variables are added follows the same pattern as in the gender decompositions. The total ethnic earnings gap (37.8 percent) is considerably larger than the gender earnings gap (15.7 percent for this set of countries). The unex- plained components of the earnings gap after controlling for gender and age are also larger. However, unlike in the gender analysis, once educa- tion is added to the matching variables, the unexplained component of the ethnic gap decreases significantly. The fact that ethnic minorities have considerably lower educational attainment than nonminorities explains the large drop in the unexplained component (from 40 percent of average minorities’ earnings to 28 percent) after education is added. A considerable Table 13.1 Demographic and Job Characteristics and Relative Earnings of Nonminority and Minority Workers in Latin America, Circa 2005 Composition Earnings index (Base: average minority (percentage) earnings = 100) Nonminorities Minorities Nonminorities Minorities All 137.8 100.0 Personal characteristics Age 37.0 36.4 18 to 24 98.4 77.9 25 to 34 133.6 98.2 35 to 44 149.5 109.5 45 to 54 159.8 113.5 55 to 65 151.2 100.1 Education None or primary incomplete 14.9 24.8 108.7 74.7 Primary complete or secondary incomplete 38.7 43.0 113.4 90.8 Secondary complete or tertiary incomplete 38.4 27.6 155.7 127.1 Tertiary complete 8.0 4.6 223.7 160.2 (continued next page) 249 250 Table 13.1 (continued) Composition Earnings index (Base: average minority (percentage) earnings = 100) Nonminorities Minorities Nonminorities Minorities Presence of children (12 years or younger) in the household No 50.7 45.5 144.7 104.4 Yes 49.3 54.5 130.7 96.3 Presence of other household member with labor income No 29.2 34.0 140.5 96.3 Yes 70.8 66.0 136.7 102.0 Urban No 15.0 20.2 92.5 68.0 Yes 85.1 79.8 145.7 108.1 Job characteristics Type of employment Employer 4.5 2.5 264.3 215.4 Self-employed 24.1 28.2 135.0 95.1 Employee 71.5 69.3 130.8 97.8 Table 13.1 (continued) Composition Earnings index (Base: average minority (percentage) earnings = 100) Nonminorities Minorities Nonminorities Minorities Part time No 86.8 85.2 133.0 94.3 Yes 13.2 14.8 169.2 132.7 Formality No 47.8 56.6 113.5 83.9 Yes 52.2 43.4 160.0 121.0 Small firm (five workers or less) No 50.8 30.0 152.1 113.8 Yes 49.2 70.1 123.0 87.6 Occupation Professionals and technicians 13.6 8.5 237.0 180.3 Directors and upper management 4.8 2.3 271.7 211.0 Administrative personnel 9.6 6.5 136.5 114.0 Merchants and sellers 12.4 11.4 117.5 102.2 251 (continued next page) 252 Table 13.1 (continued) Composition Earnings index (Base: average minority (percentage) earnings = 100) Nonminorities Minorities Nonminorities Minorities Service workers 19.0 24.3 95.0 79.9 Agricultural workers and similar 12.0 16.7 85.3 57.7 Nonagricultural blue-collars workers 27.6 29.0 126.1 102.1 Armed forces 0.0 0.0 409.1 260.1 Occupations not classified above 1.1 1.4 170.3 161.4 Economic sector Agriculture, hunting, forestry, and fishing 12.2 16.9 87.6 58.3 Mining and quarrying 0.8 0.7 195.6 144.8 Manufacturing 16.8 14.5 136.9 103.9 Electricity, gas, and water supply 0.6 0.5 178.4 151.3 Construction 7.3 9.6 124.2 94.5 Wholesale and retail trade and hotels and restaurants 24.0 21.9 132.3 102.7 Transport and storage 6.6 5.4 158.2 129.3 Financing, insurance, real estate, and business services 3.7 1.7 196.8 143.4 Community, social, and personal services 28.0 28.8 153.2 112.3 Source: Based on data from national household surveys from circa 2005. Table 13.2 Decomposition of Ethnic Earnings Gap in Latin America after Controlling for Demographic Characteristics, Circa 2005 (percent) + Presence of + Presence of other children in the household member with Gender + Age + Education household labor income + Urban Δ 37.8 37.8 37.8 37.8 37.8 37.8 Δ0 40.0 39.5 27.9 26.9 26.2 25.1 ΔW 0.0 0.0 1.4 2.4 3.6 3.5 ΔNW 0.0 0.0 –0.2 –0.4 –0.8 –0.6 ΔX –2.2 –1.7 8.7 8.9 8.8 9.8 Percentage of nonminorities in common support 100.0 100.0 98.0 95.9 93.3 89.6 Percentage of minorities in common support 100.0 100.0 99.7 99.3 98.1 95.7 Source: Based on pooled data from national household surveys from circa 2005. Note: ΔW(ΔNW) is the part of the earnings gap attributed to the existence of nonminorities (minorities) with combinations of characteristics that are not met by any minorities (nonminorities). ΔX is the part of the earnings gap attributed to differences in the observable characteristics of nonminorities and minorities over the “common support.� Δ0 is the part of the earnings gap that cannot be attributed to differences in charac- teristics of the individuals. It is typically attributed to a combination of both unobservable characteristics and discrimination. The sum of these 253 components equals the total earnings gap (ΔW + ΔNW + ΔX+Δ0 = Δ). 254 new century, old disparities portion of the gap still remains unexplained, suggesting that, like educa- tional attainment, returns to schooling are lower for ethnic minorities than for nonminorities.2 After education, the other demographic variables (presence of children and other income earners in the household) add little to the explanation of ethnic earnings gaps. Table 13.3 presents the results of the decompositions obtained after adding each of the six job characteristics. To facilitate the comparison of results, the first column of table 13.3 reports the last column of table 13.2, which reports results after matching on the six demographic characteris- tics. The last column of table 13.3 shows the earnings gap decompositions resulting from matching on the full set of variables (the six demographic and six job characteristics). The comparison of the six job characteristics reveals that, in contrast with the gender case, occupational segregation plays an important role in explaining ethnic earnings gaps. In fact, occupation is the characteristic that most reduces the earnings gap. When this characteristic is added to the demographic set of matching variables, the unexplained component decreases from 25 percent to 18 percent. Of the other five job-related covariates, three positively contribute to the ethnic earnings gaps but with small effects (2–3 percentage points): type of employment, formality, and economic sector. The other two (part-time and small firm) have almost no effect on ethnic earnings gaps. However, when all these covariates are considered together (last col- umn of table 13.3), the unexplained component of the ethnic earnings gap diminishes substantially, to just a third of the ethnic gap. Almost one-fourth of the gap can be explained by differences in the distribution of characteristics over the common support (ΔX), and an important part of the gap can be explained by the component that exists because nonmi- norities achieve certain combinations of human capital characteristics that minorities fail to reach (ΔW). Indeed, more than half of the ethnic earnings gap is attributable to the existence of these sorts of access barriers to high- paying segments of the labor markets. Not surprisingly, when the full set of demographic and job characteris- tics is used, only 43 percent of nonminorities and 51 percent of minorities lie on the common support of distributions of observable characteristics. Even greater segmentation of the labor market occurs along the gender divide, but with no substantial contribution to earnings gaps. Further analysis of the combinations of characteristics found among nonminori- ties but not among minorities promises to increase the understanding of ethnic earnings gaps. Disaggregation of the ethnic earnings gap by country for three sets of control variables reveals high cross-country heterogeneity (table 13.4). In Guatemala, for example, both the total gap and the unexplained gap after controlling for gender and age are more than twice as large as in Chile. The effect of controlling by education differs substantially from country Table 13.3 Decomposition of Ethnic Earnings Gap in Latin America after Controlling for Demographic, Job, and Full Set of Characteristics, Circa 2005 (percent) Demographic & Type of & Small set employment & Part time & Formality & Sector & Occupation firm Full set Δ 37.8 37.8 37.8 37.8 37.8 37.8 37.8 37.8 Δ0 25.1 22.7 25.9 22.4 22.8 18.0 25.1 12.9 ΔW 3.5 6.5 4.7 4.9 6.1 7.2 4.0 21.2 ΔNW –0.6 –1.4 –1.4 –0.8 –1.1 -1.0 –1.1 –7.3 ΔX 9.8 10.0 8.6 11.4 10.1 13.5 9.7 10.9 Percent of nonminorities in common support 89.6 83.5 85.9 84.8 73.2 74.8 85.0 43.0 Percent of minorities in common support 95.7 91.2 92.3 93.1 83.6 85.0 94.4 51.4 Source: Based on data from national household surveys from circa 2005. Note: ΔW(ΔNW) is the part of the earnings gap attributed to the existence of nonminorities (minorities) with combinations of characteristics that are not met by any minorities (nonminorities). ΔX is the part of the earnings gap attributed to differences in the observable characteristics of nonminorities and minorities over the “common support.� Δ0 is the part of the earnings gap that cannot be attributed to differences in characteristics of the individu- als. It is typically attributed to a combination of both unobservable characteristics and discrimination. The sum of these components equals the total earnings gap (ΔW + ΔNW + ΔX + Δ0 = Δ). 255 256 Table 13.4 Decomposition of Ethnic Earnings Gap by Demographic and Job Characteristics in Selected Countries in Latin America, Circa 2005 (percent) Δ0 + Presence of children in + Part time, formality, household, presence of occupation, economic sector, + other income earner in type of employment, Country Δ Gender and age Education household, and urban and small firm Bolivia 30.8 35.6* 16.5* 12.7* 21.2* Brazil 38.7 38.6* 30.0* 27.2* 13.9* Chile 30.8 29.3* 10.6* 8.4* 1.4 Ecuador 30.7 26.7* 3.9 2.6 0.7 Guatemala 67.7 67.4* 23.5* 21.0* 11.4* Peru 45.5 45.6* 20.9* 17.5* 14.4* Paraguay 59.6 58.0* 21.8* 12.3* 6.3 Latin America 37.8 39.5 27.9 25.1 12.9 Source: Based on data from national household surveys from circa 2005. Note: * p < 0.10. Δ corresponds to the total earnings gap. Δ0 is the part of the earnings gap that cannot be attributed to differences in characteristics of the individuals and is typically attributed to a combination of both unobservable characteristics and the existence of discrimination. overlapping disadvantages: ethnicity and earnings gaps 257 to country. In Ecuador, for example, the unexplained component is no longer significantly different from zero after accounting for differences in education, whereas in Brazil it falls from 39 percent to 30 percent. This result is driven by the fact that the gap in educational attainment differs substantially between these two countries. In Ecuador, the percentage of workers with university degrees is 16 percent among nonminorities and 6 percent among minorities. In Brazil, this difference is substantially smaller: 5 percent of nonminority and 4 percent of minority workers have university degrees.3 Figure 13.1 presents the four components of the earnings gap (sorted by the magnitude of the unexplained component) for the specification with the full set of control variables. As in the case of the gender gap, there are clear qualitative patterns across countries. First, ΔX is positive in every Figure 13.1 Decomposition of Ethnic Earnings Gap in Selected Countries in Latin America after Controlling for Demographic and Job Characteristics, Circa 2005 Bolivia (Δ = 30.6%) Peru (Δ = 45.5%) Brazil (Δ = 38.6%) Guatemala (Δ = 67.8%) Paraguay (Δ = 59.6%) Chile (Δ = 30.8%) Ecuador(Δ = 30.7%) –40 –20 0 20 40 60 80 100 percentage of average minorities’ earnings Δ0 ΔW ΔNW ΔX Source: Based on data from national household surveys from circa 2005. Note: ΔW (ΔNW) is the part of the earnings gap attributed to the existence of minorities (nonminorities) with combinations of characteristics that are not met by any minorities (nonminorities). ΔX is the part of the earnings gap attributed to differences in the observable characteristics of nonminorities and minorities over the “common support.� Δ0 is the part of the earnings gap that cannot be attributed to differences in characteristics of the individuals. It is typically attributed to a combination of both unobservable characteristics and discrimination. The sum of these components equals the total earnings gap (ΔW + ΔNW + ΔX + Δ0 = Δ). 258 new century, old disparities country, meaning that minorities in every country have combinations of characteristics that are associated with lower returns in the labor market (in particular, educational attainment). Second, ΔW is positive in all coun- tries, and it represents the largest component in most of them, suggesting that in every country, the existence of combinations of characteristics that are achieved only by nonminorities plays an important role in explaining part of the earnings gap. Access barriers—hypothesized here as an expla- nation for the earnings gaps—prevail in all countries. Unexplained ethnic earnings gaps (Δ0) are also positive in all countries (although they are not significantly different from zero in Chile, Ecuador, and Paraguay). Exploring the Unexplained Component of the Ethnic Earnings Gap Several interesting features are evident in the distribution of the unexplained ethnic earnings gaps across observable characteristics (figure 13.2). The gap is larger among men. This observation does not contradict the fact that minority women fare worst in labor markets: the earnings gap between minority women and nonminority men reaches an astonishing 60 percent when no control variables are used. Most of this gap cannot be explained on the basis of observable characteristics. Of the components attributable to observable characteristics, the largest is the one explained by combinations of characteristics that white men achieve but minority women do not (tables and figures corresponding to this decomposition are not reported). The gap is smallest among the youngest cohort. As discussed in the case of gender, where a similar finding was reported, this result may contain good news, but the evidence is not definitive. The good news would be that younger cohorts entering the labor market face less discrimination and therefore get closer to the “equal pay for equal productive character- istics.� The word of caution is that this finding may reflect the effect of unobservable characteristics correlated with age, such as experience. Four other important conclusions, which will not be described as the previous two, are still important. First, the gap is smaller among workers with other labor income generators in the household than among workers who are the sole income generator at home. Second, the gap is smaller in urban areas than in rural ones. Third, the gap is smaller and more dis- persed among part-time workers than among people who work full time. Fourth, the gap is more dispersed at both extremes of the educational attainment distribution. These results are consistent with previous find- ings in the gender earnings gaps and are relevant results as well in later chapters. overlapping disadvantages: ethnicity and earnings gaps 259 Figure 13.2 Confidence Intervals for Unexplained Ethnic Earnings Gap in Latin America after Controlling for Demographic and Job Characteristics, Circa 2005 a. Controlling for gender b. Controlling for age percentage of average percentage of average minorities’ earnings 20 minorities’ earnings 16 14 15 12 10 10 8 5 women men 24 4 4 4 54 –3 –4 –5 gender < > 24 34 44 range of age c. Controlling for education d. Controlling for children at home percentage of average minorities’ earnings 25 15 percentage of average minorities’ earnings 20 14 15 13 10 12 5 11 0 10 no yes co ple lete rim ne ri y co s on ry co ple o ry et r y te ry y m te p ar pl te ar ar in lete sec ma m ec da e tia children in the household m p p o m te nd rti co m te n in co ple m p co in education e. Controlling for other income earners in the household f. Controlling for urban percentage of average percentage of average location minorities’ earnings minorities’ earnings 18 25 16 20 14 15 12 10 10 no yes no yes other with income urban in the household g. Controlling for type h. Controlling for part-time percentage of average percentage of average of employment employment minorities’ earnings minorities’ earnings 30 15 20 10 10 0 5 no yes er ed er oy oy oy part time pl pl pl em em m lf-e se type of employment (continued next page) 260 new century, old disparities Figure 13.2 (continued) i. Controlling for formality j. Controlling for occupation 15 30 percentage of average percentage of average minorities’ earnings minorities’ earnings 14 20 13 10 12 0 11 10 −10 no yes m rec ls is rs na ricu vic se tive ic ra o rs ra o rs m o rs t c fo rs ss s ed la rce ad di na in to gr ltu w lle tu w e ar l w rke no ed rke ul l rk ifi tra formality o si es e of pr no ag ser occupation k. Controlling for sector l. Controlling for small firm size 20 100 percentage of average percentage of average minorities’ earnings minorities’ earnings 15 50 10 5 0 0 no yes el an m ture s, c city tur g an on su ing n re ru ly a r ur n rs al s ora s al rvi e rv s es pe nci t st ant se ce tri ac in fin spo sta ctio on e g tra d st pp ic ec uf in small firm ul ric ag m el ot ,h de sector tra Source: Based on data from national household surveys from circa 2005. Note: Figures show results after controlling for demographic and job- related characteristics. Boxes show 90 percent confidence intervals for unexplained earnings; whiskers show 99 percent confidence intervals. Figure 13.3 presents the unexplained ethnic earnings gap by percentile of the earnings distributions of minorities and nonminorities, in order to assess whether the unexplained component is concentrated, as in the case of the gender gap, in particular segments of the earnings distributions. After controlling only for gender and age, the unexplained gap is significantly larger among low-income workers. The gap is more than 100 percent at the overlapping disadvantages: ethnicity and earnings gaps 261 Figure 13.3 Unexplained Ethnic Earnings Gap in Latin America after Controlling for Demographic and Job Characteristics, by Percentile of Earnings Distribution, Circa 2005 120 percentage of average minorities’ earnings 100 80 60 40 20 0 0 10 20 30 40 50 60 70 80 90 100 earnings percentile gender and age gender, age, and education all demographic characteristics all demographic and job characteristics Source: Based on data from national household surveys from circa 2005. bottom of the distributions. It decreases sharply until the 30th percentile, where it is close to 27 percent. The gap then increases slightly, closing altogether only at the very right end of the distribution. When education is added as a matching variable, this overall pattern is almost maintained, with a reduction in the unexplained component of the ethnic earnings gap. However, the largest reductions in the gap occur at the lower percentiles of the distributions. Thus, educational attainment explains more of the differences in earnings of low-income workers than middle- or high-income workers. After controlling for demographic and job characteristics, the unexplained gap becomes roughly homogenous along the earnings distribution. A distinctive feature of ethnic earnings gaps is that they are smaller among part-time workers than full-time workers. Although no strong impact of economic sector segregation on earnings gaps was found, there is a link between the ethnic earnings gap and occupational segregation, in contrast with the results for gender earnings gaps. About 21 percentage points of the 39 percentage points of the earnings gap (that is, slightly more than half the earnings gap) is attributable to the existence of nonminorities 262 new century, old disparities with combinations of characteristics that are not realized by minorities. These are highly paid profiles of older, educated professionals, directors, or senior managers in specific sectors. In this sense, there is evidence that ethnic minorities in the region are confronted with glass ceilings. In sum, this chapter provides suggestive evidence that the region still faces major labor market disadvantages based on ethnicity. Policies aimed at reducing these inequalities are still needed, not only because of ethical considerations regarding equality but also as a major strategy to reduce poverty. Policies aimed at boosting school attendance for minorities are welcomed, but they should take into account the lower incentives minor- ities face to completing school given their lower returns to education in the labor market. Because ethnic minorities and women are particularly disadvantaged, indigenous girls should be given special attention. The next three chapters analyze ethnic earnings gaps in Brazil, Ecua- dor, and Guatemala. These countries are important to analyze individu- ally because they are representative of different situations for minorities. In Brazil (chapter 14), there is little difference between the educational attainment of minorities and nonminorities: both groups have poor attain- ment. In contrast, in Ecuador (chapter 15), minorities have many fewer years of education than nonminorities. Guatemala (chapter 16) has the widest ethnic earnings gaps in the region. Notes 1. For a description of the methodology used in this chapter, see chapter 2. 2. It could also be the case that lower returns to schooling for ethnic minori- ties create incentives for them to drop out of the educational system or exert less effort while in school. 3. This is not to say that Brazil has actually been successful in closing the gap in educational attainment between minorities and nonminorities, but that educa- tional attainment is low for both minorities and nonminorities. References Atal, J. P., H. Ñopo, and N. Winder, 2009. “New Century, Old Disparities: Gender and Ethnic Wage Gaps in Latin America.� RES Working Paper 4640, Inter- American Development Bank, Research Department, Washington, DC. Calla, R. 2007. “La mujer indígena en Bolivia, Brasil, Ecuador, Guatemala, y Panama: un panorama de base a partir de la ronda de censos de 2000.� Serie Mujer y Desarrollo, Consejo Económico para América Latina y el Caribe, Santiago, Chile. Chiswick, B. R., H. A. Patrinos, and M. E. Hurst. 2000. “Indigenous Language Skills and the Labor Market in a Developing Economy: Bolivia.� Economic Development and Cultural Change 48 (2): 349–67. overlapping disadvantages: ethnicity and earnings gaps 263 Contreras, D., and M. Galván. 2003. “¿Ha disminuido la discriminación salarial por género y etnia en Bolivia? evidencia del periodo 1994–1999.� http://www .depeco.econo.unlp.edu.ar/reunion_desigualdad/trabajo3.pdf. Fazio, M. V. 2007. “Economic Opportunities.� In Economic Opportunities for Indigenous Peoples in Latin America, Conference Edition, 9–20. Washington, DC: World Bank. Hall, G., and H. A. Patrinos, eds. 2006. Indigenous Peoples, Poverty and Human Development in Latin America. London: Palgrave Macmillan. Hopenhayn, M., and A. Bello. 2001. “Discriminación étnico-racial y xenofobia en América Latina y el Caribe.� Serie Políticas Sociales 47, Comisión Económica para América Latina y el Caribe (CEPAL), Santiago, Chile. ILO (International Labour Organization). 2007. Modelo de tendencias mundiales del empleo. Geneva: ILO. Jiménez Pozo, W., F. L. Casazola, and E. Yáñez Aguilar. 2006. “Bolivia.� In Indig- enous Peoples, Poverty and Human Development in Latin America, ed. G. Hall and H. A. Patrinos, 40–66. London: Palgrave Macmillan. Lovell, P. 2000. “Race, Gender and Regional Labour Market Inequalities in Brazil.� Review of Social Economy 58 (3): 277–93. Ñopo, H., and A. Hoyos. 2010. “Evolution of Gender Wage Gaps in Latin America at the Turn of the Twentieth Century: An Addendum to ‘New Century, Old Disparities.’� IZA Discussion Paper 5086, Institute for the Study of Labor, Bonn, Germany. Patrinos, H. A. 2000. “The Cost of Discrimination in Latin America.� Studies in Comparative International Development 35 (2): 3–17. Piras, C. 2004. “An Overview of the Challenges and Policy Issues Facing Women in the Labor Force.� In Women at Work: Challenges for Latin America, ed. C. Piras, 3–24. Washington, DC: Inter-American Development Bank. Psacharopoulos, G. 1992. “Ethnicity, Education, and Earnings in Bolivia and Gua- temala.� Policy Research Working Paper 1014, World Bank, Washington, DC. Psacharopoulos, G., and H. A. Patrinos. 1994. Indigenous People and Poverty in Latin America: An Empirical Analysis. Washington, DC: World Bank. Rangel, M. 1998. “Raza y género en Brasil: las regiones metropolitanas de Rio de Janeiro y de São Paulo.� Acta Sociologica 23, Universidad Nacional Autónoma de México, Facultad de Ciencias Políticas y Sociales, Mexico City. 14 Promoting Ethnic Equality: Brazil 1996–2006 As in other countries in the region, Brazil’s history includes several centu- ries of slavery involving both indigenous peoples and Afro-descendents. The legacy of slavery persists in more and less subtle forms of discrimina- tion. Although grassroots movements have denounced these problems for decades, only recently has the federal government launched an innova- tive and coordinated National Policy for the Promotion of Gender and Race Equality. For the first time, the multiyear plan for 2004–07 included “social inclusion and reduction of social inequalities� in its goals. The central objective of the national policy is to reduce gender and ethnic inequalities in Brazil, with emphasis on the Afro-descendant population. The policy’s success will depend on coordinated action and commitment by all spheres of government and society. This chapter uses data from the Pesquisa Nacional por Amostra de Domi- cilios (National Survey of Sample Households, PNAD) for 1996–2006 to analyze and decompose the ethnic earnings gap based on the methodology described in chapter 2.1 Attention is restricted to people 15–65 years old with positive earnings at the primary occupation (measured as hourly earnings). What Does the Literature Show? López-Calva and Lustig (2009) report a decline in inequality across Latin America. They focus on four countries: Argentina, Brazil, Mexico, and Peru. This chapter was adapted from “Gender and Racial Wage Gaps in Brazil 1996–2006: Evidence Using a Matching Comparisons Approach,� Luana Marquez Garcia, Hugo Ñopo, and Paola Salardi, RES Working Paper 4626, Inter-American Development Bank, 2009. Luana Marques Garcia is a young professional at the Inter-American Devel- opment Bank. Paola Salardi is a research fellow in the Economics Group at the University of Sussex, in Brighton, United Kingdom. 265 266 new century, old disparities In Brazil, they report a steady fall in the Gini coefficient since 1998 and a decline in poverty and extreme poverty between 2001 and 2007.2,3 During this period, annual per capita income of the poorest grew at a much faster rate (7.0 percent) than per capita income of the richest decile (1.1 percent), which defines Brazil’s growth pattern as pro-poor. Reductions in overall inequality and poverty are caused by the decline in labor income inequality, which occurred thanks to an accelerated expansion of access to education in Brazil and a drop in the returns to education. Labor earnings differen- tials by education level have declined at all levels in Brazil, particularly for secondary and tertiary education (López-Calva and Lustig 2009). Changes in education account for half the reduction in labor income inequality; the other half is accounted for by a number of factors, among which Barros et al. (2009) include changes in gender and ethnic discrimination and labor force participation rates. A popular perception in Brazil is that racism does not affect a person’s life and that study, hard work, and initiative are the main factors leading to success. There is an emerging popular belief, how- ever, that class differences prevent people from progressing. Research suggests that earnings gaps between whites (nonminorities) and nonwhites (minorities) were about 50 percent for men and 45 percent for women in the mid-2000s (that is, white men earned 50 percent more than Afro-descendant men, and white women earned 45 percent than Afro-descendant women [De Carvalho, Néri, and Britz do Nascimento Silva 2006; Guerreiro 2008]). Race and gender significantly affect income, even when education, experience, and labor market characteristics are taken into account. One of the most comprehensive analyses of gender and ethnic earn- ings differentials in Brazil is Soares (2000). He documents that, since the 1980s, ethnic earnings gaps have been larger than gender earnings gaps. White women earn 79 percent and Afro-descendant men only 46 percent of white men’s earnings. Using the Blinder-Oaxaca decomposition yields very different patterns for gender and ethnic differentials in earnings Although gender earnings gaps decreased over time, ethnic differentials remained constant. Most of the earnings differentials by race can be explained by differences in observ- able characteristics between ethnic groups, whereas the unexplained com- ponent by gender is constantly larger than the explained component. De Carvalho, Néri, and Britz do Nascimento Silva (2006) ana- lyze gender and ethnic earnings gaps by applying the Blinder-Oaxaca decomposition and correcting for selection bias as proposed by Heck- man (1979).4 Correcting for labor market participation reduces the unexplained component of the ethnic gender gap from 37 percent to 30 percent and the unexplained component of the gender gap from 33 per- cent to 18 percent. It increases the earnings gap between white men and Afro-descendant women, however, from 78 percent to 95 percent. Lovell (1994, 2000, and 2006) analyzes gender and ethnic differences in earnings using census data instead of national household surveys. In promoting ethnic equality: brazil 1996–2006 267 her empirical applications, she adopts a modified version of the standard Blinder-Oaxaca decomposition as proposed by Jones and Kelly (1984). Drawing on sample data from the 1960 and 1980 censuses, Lovell (1994) finds that gender earnings gaps are larger than ethnic earnings gaps. Lovell and Wood (1998) highlight how the unexplained component of both gender and ethnic earnings gaps has increased. Lovell (2000) focuses on regional differences in earnings gaps, considering only the states of São Paulo and Bahia. The wealthier state, São Paulo, shows larger earnings differentials and a larger unexplained component. Lovell (2006) focuses on earnings gaps in São Paulo, covering a longer time period. Her finding that ethnic differentials are stable whereas gender differentials diminished over time is in line with previous studies. She finds that the unexplained component of both gaps increased. Calvalieri and Fernandes (1998) also report earnings gaps that are larger along gender than ethnic lines. Using the PNAD for 1989, they estimate earnings equations. They find that after controlling for a large set of characteristics, the gender earnings gap becomes larger than the ethnic earnings gap, probably because of the greater variation in the ethnic earn- ings gap than in the gender earnings gap, which is captured by regional dummies included in the regression equations. The 1980 study by Silva represents a pioneering analysis of ethnic earn- ings gaps using the Blinder-Oaxaca decomposition technique. He employs a 1.27 percent subsample of the 1960 census, restricting his analysis to male workers living in the Rio de Janeiro area. He examines three ethnic groups: whites; “mulattoes� (people of brown complexion, presumably of mixed European and African ancestry); and “negroes� (darker-skinned people appearing to be primarily or exclusively of African ancestry). Silva finds a larger earnings gap for negroes than for mulattoes with respect to white male workers and finds that the explained component is larger than the unexplained component. Silva’s seminal work was not updated until 2004, when Arias, Yamada, and Tejerina examined the entire earnings distribution, using the quantile regression methodology developed by Koenker and Bassett (1978). Their findings support the importance of examining different points of the earn- ings distribution, not simply average values, as in the Blinder-Oaxaca decomposition technique. They find that the bottom decile of nonwhites earns 24 percent less than comparable whites, whereas the top decile of nonwhites earns 56 percent less. Overall, nonwhites earn 46 percent less than whites, and people of mixed race earn 42 percent less. The earnings of people of mixed race at the bottom of the earnings distribution are similar to those of nonwhites. In contrast, the earnings of people of mixed race at the upper end of the income distribution are similar to those of whites. Arcand and D’Hombres (2004) enrich the study of ethnic earnings differentials based on the Blinder-Oaxaca decomposition and quantile regression by considering the selection bias correction for occupational attachment. The explained component accounts for most of the gaps for 268 new century, old disparities both nonwhites and people of mixed race; the unexplained component is larger for nonwhites. Expanding on Soares (2000), Campante, Crespo, and Leite (2004) focus on differences between the North-East and South-East regions. In the South-East, the ethnic gap exceeds the national average, and the unex- plained component tends to be larger than elsewhere in Brazil. Leite (2005) shows that the unexplained component is higher in the South-East than the North-East. This finding holds after controlling for the endogeneity of individuals’ schooling, which reduces the size of the unex- plained component. Reis and Crespo (2005) show how ethnic earnings differentials are not constant over time, as claimed by previous studies. They decompose the unexplained component into age, period, and cohort effects and show that ethnic earnings gaps are smaller for younger cohorts. Taking as a point of departure Campante, Crespo, and Leite (2004) and Soares (2000), Guimarães (2006) adds controls for region and sector of activity. She finds that unexplained differences represent 30 percent of total differentials and that pay gaps between whites and nonwhites are larger in the North and North-East regions than elsewhere. In summary, ethnic earnings gaps in Brazil were larger than gender earnings gaps in recent decades (Soares 2000); only before the 1980s were gender earnings gaps more important (Lovell 1994; Lovell and Wood 1998). Gender earnings gaps tend to be more homogenous across regions than ethnic gaps (Calvalieri and Fernandes 1998). Ethnic gaps are wider in the South-East region than in the North-East; they are also wider in urban than rural areas (Lovell 2000; Campante, Crespo, and Leite 2004; Loureiro, Carneiro, and Sachsida 2004; Leite 2005). Over time, gender earnings gaps have decreased significantly; ethnic gaps have not. Nonetheless, work on cohorts by Reis and Crespo (2005) finds that ethnic earnings gaps are shrinking for the younger generation. The explained component of the ethnic gap is smaller for nonwhites than for people of mixed race; people of mixed race also earn more than nonwhites (Arcand and D’Hombres 2004; Arias, Yamada, and Tejerina 2004). How Do Ethnic Minorities and Nonminorities in the Work Force Differ? Age tends to be homogeneous across people who matched and people who do not (that is, people in and out of the “common support� [see chapter 2]), as well as over time (table 14.1). Among whites (nonminorities) who do not match nonwhites(minorities), the share that had more than 15 years of schooling was 19.9 percent in 1996 and 28.3 percent in 2006; among unmatched nonwhites, these shares were just 2.8 percent in 1996 and Table 14.1 Demographic and Job Characteristics of Matched and Unmatched Samples of Whites and Nonwhites in Brazil, 1996 and 2006 (percent) 1996 2006 Matched Matched Unmatched Unmatched whites and Unmatched Unmatched whites and nonwhites whites nonwhites nonwhites whites nonwhites Personal characteristics Age 15–24 28.9 23.2 28.3 24.1 19.2 25.3 25–34 27.4 27.7 30.3 28.1 25.3 28.9 35–44 23.0 24.8 24.6 23.1 24.5 25.8 45–54 13.7 16.0 12.4 16.7 20.5 14.9 55–65 7.0 8.3 4.4 8.1 10.5 5.1 Years of education Less than 4 39.6 19.4 31.9 30.0 13.1 21.7 4–10 56.4 56.9 61.4 62.9 52.0 64.7 11–15 1.1 3.9 0.7 1.8 6.6 1.8 More than 15 2.8 19.9 6.0 5.4 28.3 11.9 269 (continued next page) 270 Table 14.1 (continued) 1996 2006 Matched Matched Unmatched Unmatched whites and Unmatched Unmatched whites and nonwhites whites nonwhites nonwhites whites nonwhites Gender (male) 70.1 64.9 58.1 70.0 63.1 54.7 Urban 87.8 90.6 83.2 87.1 92.4 86.5 Regions: North 19.8 4.0 4.4 26.0 6.1 10.5 North-East 39.9 9.2 31.5 35.1 9.8 31.4 South-East 18.4 33.2 42.8 16.8 30.1 35.4 South 4.4 42.5 12.1 4.9 42.0 12.6 Central West 17.5 10.8 9.2 17.4 12.1 10.2 Job characteristics Type of occupation: Professionals 7.1 14.9 10.7 16.1 38.2 18.3 Intermediate 43.3 52.0 44.8 41.7 35.7 45.7 Blue collar 49.6 33.2 44.6 42.2 26.1 36.1 Table 14.1 (continued) 1996 2006 Matched Matched Unmatched Unmatched whites and Unmatched Unmatched whites and nonwhites whites nonwhites nonwhites whites nonwhites Formal 49.1 51.9 45.7 47.2 50.9 48.2 Agriculture 9.4 7.3 14.9 9.1 6.0 11.7 Construction 11.1 5.6 7.1 11.0 5.5 7.1 Social services 35.2 35.8 44.6 25.3 28.1 41.2 Source: Based on data from 1996 and 2006 PNAD. 271 272 new century, old disparities 5.4 percent in 2006. Unmatched nonwhites are more likely to be men. There seems to be a geographical concentration of unmatched nonwhites in the North-East and of unmatched whites in the South. This pattern reflects Brazilian regional disparities by ethnic groups. Reflecting educational attainment patterns, unmatched whites are more likely to be professionals: in 2006, 38.2 percent of unmatched whites—and just 16.1 percent of unmatched nonwhites and 18.3 per- cent of matched white and nonwhites—were professionals. Unmatched nonwhites are employed mainly as blue-collar workers and are more likely to work in the informal sector. For economic activities, differ- ences in and out of the common support are smaller for race than for gender, although unmatched nonwhites are more likely to work in sec- tors with a higher density of low-skilled workers, such as agriculture and construction. The Role of Individual Characteristics in Explaining the Ethnic Earnings Gap This subsection describes the matching conducted, based on six sets of human capital and job characteristics. The first set includes only the num- ber of years of schooling. The second set adds age and education, and the third set adds the region.5 Job variables are then added. The ethnic earnings gap in Brazil is large, and it has been decreas- ing slowly (figure 14.1). Starting from a value of 96 percent in 1996, it declined from 18 percent to 78 percent in 2006. The unexplained component, the part of the earnings gap that cannot be attributed to differences in characteristics of the individuals, is small: after controlling for the wider set of characteristics, Δ0 accounts for about 18 percent of the total gap. The bulk of the gap is given by the explained component, ΔX—the part of the earnings gap attributed to differences in the observ- able characteristics of whites and nonwhites over the common support. The unexplained component is responsible for most of the drop in the total gap between 1996 and 2006 (15.2 percentage points of the 18.0 percentage point decline), however. For unmatched individuals, ΔNW represents the portion of the earnings gap for which there are nonwhites who cannot be matched with whites, and it is negative. Interestingly, ΔW (the part of the earnings gap attributed to the fact that there are whites with characteristics that are not matched by nonwhites) is larger than ΔX and fairly stable over time. This result may reflect that fact that a consistent portion of white workers has stronger human capital characteristics than nonwhites and may hold very high-paid positions. promoting ethnic equality: brazil 1996–2006 273 Exploring the Unexplained Component of the Ethnic Earnings Gap Table 14.2 reports ethnic earnings gaps by various demographic and job characteristics, considering only the first year (1996) and the last year (2006) of the period under study.6 Ethnic earnings gaps increase with age and education; they are large for high-paid positions. The gap for the youngest age group is far smaller than the gap for other groups. The Figure 14.1 Decomposition of Ethnic Earnings Gap in Brazil, 1996–2006 percentage of average earnings 120 a. Controlling for education 100 of nonwhites 80 60 40 20 0 1996 1997 1998 1999 2001 2002 2003 2004 2005 2006 120 b. Controlling for education and age percentage of average earnings 100 80 of nonwhites 60 40 20 0 –20 1996 1997 1998 1999 2001 2002 2003 2004 2005 2006 c. Controlling for education, age, and region percentage of average earnings 120 100 of nonwhites 80 60 40 20 0 –20 1996 1997 1998 1999 2001 2002 2003 2004 2005 2006 (continued next page) 274 new century, old disparities Figure 14.1 (continued) d. Controlling for education, age, region, and occupation percentage of average earnings 100 80 of nonwhites 60 40 20 0 –20 1996 1997 1998 1999 2001 2002 2003 2004 2005 2006 e. Controlling for education, age, region, 120 occupation, and sector percentage of average earnings 100 80 of nonwhites 60 40 20 0 –20 1996 1997 1998 1999 2001 2002 2003 2004 2005 2006 f. Controlling for education, age, region, occupation, percentage of average earnings sector, and employment in formal sector 120 100 of nonwhites 80 60 40 20 0 –20 1996 1997 1998 1999 2001 2002 2003 2004 2005 2006 ΔX ΔNW ΔW Δ0 Source: Based on data from the 1996–2006 PNAD. Note: ΔW (ΔNW) is the part of the earnings gap attributed to the existence of whites (nonwhites) with combinations of characteristics that are not met by any nonwhites (whites). ΔX is the part of the earnings gap attributed to differences in the observable characteristics of whites and nonwhites over the “common support.� Δ0 is the part of the earnings gap that cannot be attributed to differences in characteristics of the individuals. It is typically attributed to a combination of both unobservable characteristics and discrimination. The sum of these components equals the total earnings gap (ΔW + ΔNW + ΔX + Δ0 = Δ). promoting ethnic equality: brazil 1996–2006 275 Table 14.2 Original and Unexplained Ethnic Earnings Gaps in Brazil, by Demographic and Job Characteristics, 1996 and 2006 (percent) 1996 2006 Δ Δ0 Δ Δ0 Demographic characteristics Age groups: 15–24 33.85 8.26 25.53 5.53 25–34 91.31 20.42 67.44 15.29 35–44 125.98 22.76 88.43 15.05 45–54 141.84 20.92 121.05 26.26 55–65 109.8 12.23 123.72 19.21 Years of education: Less than 4 26.38 6.08 17.52 3.81 4–10 41.78 16.21 29.29 8.47 11–15 75.21 52.15 54.73 38.98 More than 15 146.09 61.25 130.35 80.91 Men 114.19 20.22 94.74 15.45 Urban 99.77 20.28 81.07 16.6 Regions: North 71.86 7.16 53.8 16.01 North-East 74.5 8.7 53.15 9.47 South-East 106.91 28.17 87.75 19.83 South 82.39 17.87 70.38 19.85 Central West 92.71 16.36 84.33 12.54 Labor characteristics Type of occupation: Professionals 153.41 23.92 130.55 45.17 Intermediate 118.81 19.65 29.48 7.83 Blue collar 43.88 13.52 40.32 11.06 Formal 80.27 19.57 61.81 14.93 Agriculture 64.81 10.84 60.68 8.43 Construction 64.48 15.39 49.78 16.14 Social services 99.19 13.95 90.8 16.06 Source: Based on data from 1996 and 2006 PNAD. Note: Δ is the total earnings gap. Δ0 is the part of the gap attributed to differences between whites and nonwhites that cannot be explained by observable characteristics. 276 new century, old disparities geographical distribution of ethnic earnings gaps, which are larger in the South-East, confirms the crucial role played by this variable. The analysis is enriched by considering unexplained earnings differ- entials in individual income. Data sets were pooled, rescaling earnings so that the average earnings of ethnic minorities are normalized to 100 in each year. In this way, changes in earnings in the economy over time are ignored, in order to focus on earnings gaps. At each percentile of the Figure 14.2 Original and Unexplained Ethnic Earnings Gap in Brazil after Matching, by Percentile of Earnings Distribu- tion, 1996–2006 a. Original ethnic earnings gap 120 100 earnings of nonwhites percent of average 80 60 40 20 0 0 10 20 30 40 50 60 70 80 90 100 percentile of earnings distribution b. Unexplained ethnic earnings gap after controlling for 120 demographic and job characteristics 100 earnings of nonwhites percent of average 80 60 40 20 0 0 10 20 30 40 50 60 70 80 90 100 percentile of earnings distribution Source: Based on data from the 1996–2006 PNAD. promoting ethnic equality: brazil 1996–2006 277 earnings distribution of whites and nonwhites, the earnings of represen- tative individuals in each distribution are compared and the gap between the two is computed (figure 14.2). The difference between the total gap and the gap that remains after controlling for the full set of observable characteristics is large. The total gap increases at the upper end of the earnings distribution. Although the unexplained gap is considerably smaller than the total, it shows larger differentials for better-paid workers, a result similar to that found by Crespo (2003). Ethnic earnings gaps are significantly larger than gender gaps; after controlling for observable individual characteristics, however, the situa- tion is reversed. Observable individual characteristics play an important role in explaining earnings differentials between whites and nonwhites but a smaller role in gender earnings gaps. Among these characteristics, edu- cation plays a prominent role; labor market characteristics (occupation, economic sector, and formality) are also significant in explaining ethnic earnings differentials. The data suggest that the way in which these labor market characteristics operate takes the form of some sort of access bar- rier (as the ΔW components are largest). Almost half of the ethnic earnings differential can be explained by the fact that whites have greater access to certain occupations, in certain sectors, with a certain degree of formality than nonwhites. Education matters, but segregation in labor markets does too. Unexplained ethnic earnings gaps increase with workers’ age and education; they are larger among professionals and in the South-East. Unexplained gaps increase monotonically, albeit only slightly, with income. Notes 1. For 2000, census data were used. Asians and unnidentified ethnic minorities were dropped because of their negligible sample sizes. 2. The Gini coefficient is a measure of inequality in a society. A Gini coefficient of 0 expresses perfect equality; a Gini coefficient of 1 expresses perfect inequality. 3. Extreme poverty is the absence of one or more factors enabling individuals or households to assume basic responsibilities and enjoy fundamental rights. 4. This study also controls for the use of complex sample surveys without find- ing any significant alterations in the estimated coefficients. 5. The regions are North (Rondônia, Acre, Amazonas, Roraima, Parà, Amapà, Tocantins); North-East (Maranhão, Piauì, Cearà, Rio Grande do Norte, Paraiba, Pernambuco, Alagoas, Sergipe, Bahia); South-East (Minas Gerais, Espìrito Santo, Rio de Janeiro, São Paulo); South (Paraná, Santa Catarina, Rio Grande do Sul); and Central-West (Mato Grasso do Sul, Mato Grosso, Goiás, Distrito Federal). 6. Only the results for the first and last year are reported, because the trend over the decade is stable and smoothly decreasing. For all subsamples of the population, both explained and unexplained earnings gaps decreased over time. 278 new century, old disparities References Arcand, J. L., and B. D’Hombres. 2004. “Racial Discrimination in the Brazilian Labour Market: Wage, Employment and Segregation Effects.� Journal of Inter- national Development 16: 1053–66. Arias, O., G. Yamada, and L. Tejerina. 2004. “Education, Family Background and Racial Earnings Inequality in Brazil.� International Journal of Manpower 25 (3/4): 355–74. Barros, R., F. H. G. Ferreira, J. Molinas Vega, and J. Saavedra Chanduvi. 2009. Measuring Inequality of Opportunities in Latin America and the Caribbean. Washington, DC: World Bank. Calvalieri, C., and R. Fernandes. 1998. “Diferenciais de salarios por genero e por cor: uma comparação entre as regiões metropolitanas Brasileiras.� Revista de Economia Politica 18 (1): 158–15. Campante, F. R., A. R. V. Crespo, and P. G. Leite. 2004. “Desigualdade salarial entre raças nomercado de trabalho urbano Brasileiro: aspectos regionais.� Revi- sta Brasileira de Economia 58 (2): 185–210. Crespo, A. 2003. “Desigualdade entre racas e generos no Brasol: uma analise com simulacoes contra-factuais� Dissertação de mestrado, Pontifícia Universidade Católica do. De Carvalho, A. P., M. Néri, and D. Britz do Nascimento Silva. 2006. Diferen- ciais de saláriospor raça e gênero no Brasil: aplicação dos procedimentos de Oaxaca e Heckman em pesquisas amostrais complexas. Rio de Janeiro: Insti- tuto Brasileiro de Geografia e Estatística. Guerreiro, R. 2008. “Is All Socioeconomic Inequality among Racial Groups in Bra- zil Caused by Racial Discrimination?� Working Paper 43, International Policy Centre for Inclusive Growth, Brasilia. Guimarães, R. 2006. “Desigualdade salarial entre negros e brancos no Brasil: dis- criminação ou exclusão?� Econômica 8 (2): 227–51. Heckman, J. 1979. “Sample Selection Bias as a Specification Error.� Econometrica 47 (1): 153–61. Jones, F. L., and J. Kelley. 1984. “Decomposing Differences between Groups. A Cautionary Note on Measuring Discrimination.� Sociological Methods and Research 12: 323–43. Koenker, R., and G. Bassett. 1978. “Regression Quantiles.� Econometrica 46 (1): 33–50. Leite, P. G. 2005. “Race Discrimination or Inequality of Opportunities: The Brazilian Case.� Universität Göttingen Discussion Paper 118, Göttingen, Germany. López-Calva, L. F., and N. Lustig, 2009. “The Recent Decline of Inequality in Latin America: Argentina, Brazil, Mexico and Peru.� Working Paper 140, ECINEQ (Society for the Study of Economic Inequality), Palma de Mallorca, Spain. Loureiro, P. R. A., F. G. Carneiro, and A. Sachsida. 2004. “Race and Gender Discrimination in the Labor Market: An Urban and Rural Sector Analysis for Brazil.� Journal of Economic Studies 31 (2): 129–43. Lovell, P. 1994. “Race, Gender and Development in Brazil.� Latin American Research Review 29(3): 1–35. ––––. 2000. “Race, Gender and Regional Labour Market Inequalities in Brazil.� Review of Social Economy 58 (3): 277–93. promoting ethnic equality: brazil 1996–2006 279 ––––. 2006. “Race, Gender, and Work in São Paolo, Brazil, 1960–2000.� Latin American Research Review 41 (3): 63–87. Lovell, P., and C. H. Wood. 1998. “Skin Colour, Racial Identity and Life Chances in Brazil.� Latin American Perspectives 25 (3): 90–109. Marquez Garcia, L., H. Ñopo, and P. Salardi. 2009. “Gender and Racial Wage Gaps in Brazil 1996–2006: Evidence Using a Matching Comparisons Approach.� RES Working Paper 4626, Inter-American Development Bank, Research Depart- ment, Washington, DC. Reis, M. C., and A. R. V. Crespo. 2005. “Race Discrimination in Brazil: An Analy- sis of the Age, Period and Cohort Effects.� IPEA Texto para Discussão 1114, Instituto de Pesquisa Econômica Aplicada, Rio de Janeiro. Silva, N. D. V. 1980. “O preço da cor: diferenciais raciais na distribuição de renda no Brasil.� Pesquisa e Planejamento Econômico 10 (1): 57–67. Soares, S. D. S. 2000. “O perfil da discriminação no mercado de trabalho: homens negros, mulheres brancas e mulheres negras.� IPEA Texto para Discussão 769, Instituto de Pesquisa Econômica Aplicada, Brasilia. 15 No Good Jobs and Lower Earnings: Ecuador 2000–07 Within Latin America, Ecuador can be regarded as paradigmatic, with one of the largest shares of indigenous people and a very high incidence of poverty among them and Afro-descendants. Despite the economic poten- tial that this cultural diversity and social capital could represent, socioeco- nomic differences persist. The empirical analysis of ethnic earnings gaps reported in this chapter was conducted using annual data from the national labor survey (Encuesta de Empleo, Desempleo, y Subempleo [ENEMDU]) collected by the Insti- tuto Nacional de Estadísticas y Censos de Ecuador (INEC) for 2003–07. The sample includes labor income earners and the self-employed reporting positive earnings (measured in hourly earnings) who were 15–65 years old and lived in the coastal, highland, and Amazon regions of Ecuador. What Does the Literature Show? García-Aracil and Winter (2006) measure the extent to which earnings dif- ferentials can be attributed to differences in human capital or to discrimi- nation for labor income earners 12–65 years old. They identify indigenous people as people who live in a household in which there is at least one inhabitant who speaks an indigenous language. They use variables such as age and family composition (number of older and younger siblings in the household) as instruments for labor market participation in order to reduce bias caused by selection into the labor markets. Their decomposi- tion results, using the nonindigenous pay structure as reference, yield a This chapter was adapted from “Ethnic and Gender Wage Gaps in Ecuador,� Lourdes Gallardo and Hugo Ñopo, RES Working Paper 4625, Inter-American Development Bank, 2009. Lourdes Gallardo is an investment officer at the Inter-American Development Bank. 281 282 new century, old disparities total earnings difference of 104 percent between indigenous and nonin- digenous workers, of which 46 percent reflects difference in endowments and 58 percent reflects “unexplained� differences. Larrea and Montenegro (2005) calculate separate regressions of labor earnings for indigenous (minorities) and nonindigenous (nonminorities) workers using data from the 1998 Survey on Living Conditions (Encuesta de Condiciones de Vida [ECV]), which approximates ethnicity through language. Using traditional Blinder-Oaxaca decompositions, they report a total earnings differential between indigenous and nonindigenous workers of 69 percent, of which 17.4 percent reflects endowment differences and 82.6 percent is unexplained. The difference between García-Aracil and Winter (2006) and Larrea and Montenegro (2005) is considerable given that both use data from the same source, collected only one year apart. The language-based definition of ethnicity used by both García- Aracil and Winter (2006) and Larrea and Montenegro (2005) has a limitation, as it includes Spanish-speaking indigenous workers among nonindigenous workers. Doing so could underestimate earnings differ- entials, because the lower earnings of indigenous workers narrow the earnings gap as well as the differences caused by endowments and the differences that are left unexplained. Furthermore, this language-based approach includes other minority groups, such as Afro-descendants and people of mixed race who are Spanish speakers. There is consistent anecdotal evidence that points to discriminatory treatments of these people in everyday activities, possibly leading to biases and underesti- mates in the decomposition outcomes. Including nonindigenous people with indigenous language speakers within indigenous households will likewise negatively bias estimates of differences. Both studies use monthly earnings as the dependent variable. It can be argued that monthly earnings do not accurately capture the return to productivity based on each worker’s human capital endowments, because they are affected by workers’ decision on how many hours to allocate to their job throughout a month, not just the return to their labor. Monthly earnings are useful when measuring income inequality between two groups. Hourly earnings are a better measure of pay differentials between groups, as they are compensation rates per unit of time worked. In this way, differences in hourly earnings can show pay differentials for equal productivity. Gallardo (2006) analyzes labor market differentials of the indigenous and Afro-descendant population in Ecuador. Unlike the previous two studies, this study uses ethnic self-identification, as reported in the 2000 Household and Childhood Measurement Indicators Survey (Encuesta de Medición de Indicadores sobre la Niñez y los Hogares [EMEDINHO]) survey. Another difference between this study and the other two is the extended earnings differential decomposition model for labor income earn- ers, based on the traditional Blinder-Oaxaca methodology and a system of no good jobs, and lower earnings: ecuador 2000–07 283 simultaneous equations. This extension contributes to the analysis by rec- ognizing that ethnicity and the intergenerational transmission of human capital may influence educational investments, sector of employment, and area of residence (Black, Devereux, and Salvanes 2003). By decomposing these three variables separately using the Blinder-Oaxaca method, Gal- lardo captures direct and indirect paths through which discrimination may affect earnings in the labor market. Gallardo finds that low levels of educational attainment accompany higher rates of informal sector employment and that returns to educa- tion in the labor market for both indigenous and Afro-descendant labor income earners are lower than those of the mixed-race and white popu- lations. The author also finds evidence that the transmission of human capital from parents to children has negative education and labor market outcomes for the indigenous and Afro-descendant populations. Among male workers, the direct effect on earnings differentials between indig- enous, Afro-descendant, and mixed-race employees and white employees with similar endowments accounts for 27.1 percent of overall earnings differences. Indirect channels through schooling, sector of employment, and area of residence account for 39.9 percent of the earnings differential. For women, unexplained differences in pay account for 23.5 percent of the difference in earnings between the two ethnic clusters, and indirect channels account for 56.9 percent. Ethnic minorities in Ecuador are concentrated largely in rural areas, where they are employed mostly in the agricultural sector; on-farm employment constitutes the main source of income for most indigenous families (World Bank 2004). Poverty in Ecuador affects predominantly rural areas. Ethnic minorities still have limited or no access to land owner- ship and work mostly low-productivity land (De Ferranti et al. 2003). This unequal distribution of land reflects the historical and institutional legacy dating back to colonial times. MacIsaac and Rama (1997) find that the largest earnings gap in Ecuador is between workers in agriculture and workers in the rest of the economy. The income of rural poor indigenous workers is still tied to agriculture, a sector characterized by lower economic outcomes for all workers than other sectors of the economy. The authors also find that ethnic minorities in Ecuador are overrepresented in agriculture and in informal nonunion- ized activities and that hourly earnings in agriculture are 30 percent lower than in the informal sector. How Do Ethnic Minorities and Nonminorities in the Work Force Differ? Table 15.1 presents the proportion of the Ecuadorian population that reports being indigenous or Afro-descendant (black or mulatto). These 284 new century, old disparities Table 15.1 Ethnic Minorities in Ecuador, by Gender, 2003–07 (percent) National 2003 2004 2005 2006 2007 Men 14.5 12.1 12.3 12.5 12.8 Women 14.0 11.7 12.4 12.6 11.4 Urban Men 9.4 7.2 8.4 8.3 9.1 Women 8.7 6.9 7.8 7.6 8.0 Source: Based on data from 2003–07 ENEMDU. populations are referred to as ethnic minorities. One of the traditional concerns attending the use of self-identification rather than native lan- guage to determine ethnicity is the “self-whitening� phenomenon, in which minorities deny their “indigeneity.� This phenomenon leads to sta- tistical underreporting. In recent years, underreporting seems unlikely, as the identity of the indigenous population has been empowered in Ecuador through social mobilization and political events. Ethnic minorities in Ecuador have traditionally been predominantly rural; in 2003, 63 percent of the indigenous population was concentrated in rural areas. Based on the ENEMDU/EMEDINHO data, Gallardo (2006) estimates that about 78 percent of the indigenous population was concentrated in rural areas in 2000. This figure declined to 58 percent by 2007. The proportion of ethnic minorities also declined nationally and in urban settings, possibly influenced by the effects of the 1999 financial crisis, which stimulated internal and international migration. Both the reduction of the proportion of ethnic minorities and their growing concentration in urban areas are important for understanding the evolution of these populations’ well-being in Ecuador. Many observers believe that these phenomena have generated new forms of discrimination against emigrants and their families in Ecuador, many of them, indigenous. Ecuadorian society views emigrants and their families who stay behind as irrational, unproductive, and dysfunctional for the national economy. The families who stay behind usually consist of households headed by women, as men have higher emigration rates. Furthermore, emigrants’ children have lower educational outcomes than nonemigrant children. They are inclined to leave the countryside as their parents did, which encourages dropping out of high school and university (Soruco, Piani, and Rossi 2008). If emigration-based discrimination spills over to labor markets, women and indigenous people related to emigrants could suffer adverse labor outcomes as a consequence of this phenomenon. no good jobs, and lower earnings: ecuador 2000–07 285 The ethnic educational gap is still wide, particularly at higher levels of education, but it has been narrowing, because enrollment of ethnic minorities in secondary and higher education has slightly increased while enrollment of nonminorities has stayed roughly constant (table 15.2). Between 2003 and 2007, the percentage of ethnic minorities with no education also declined slightly. This trend suggests that there were higher enrollment rates for ethnic minority children, as total net primary enroll- ment in 2006 was 94.3 percent, up from 90.3 percent in 1999. The participation of ethnic minorities in low-income occupations such as day work, domestic employment, and self-employment, which predom- inantly includes informal sector workers, is high (table 15.3). However, between 2003 and 2007, male labor force participation in self-employment decreased considerably, and male participation as day laborers increased. Meanwhile, the relative proportions of the self-employed decreased from 39 percent women versus 30 percent men in 2003 to 36 percent women versus 25 percent men in 2007. Among the self-employed, the proportion of women is higher among ethnic minorities than nonminorities. Women from ethnic minorities are highly concentrated in domestic employment and self-employment. Unemployment is particularly high among Afro-descendants. Accord- ing to the INEC, in 2007, the unemployment rate in Ecuador was 7.9 percent for the general population, 11.0 percent for Afro-descendants, and 17.5 percent among women. Among the indigenous population, the unemployment rate was 6.0 percent. The Role of Individual Characteristics in Explaining the Ethnic Earnings Gap Figure 15.1 illustrates the ethnic earnings gap decompositions for four combinations of observable characteristics used as controls.1 The first combination includes area (rural or urban), education, gender, and age. The second adds to the previous list a dummy variable that identifies whether the respondent is the head of household. The third combination builds on the second one by adding occupation (coded at the one-digit level). The fourth combination adds a variable that reports whether the respondent’s income is complemented by remittances from abroad. The earnings difference between ethnic minority and nonminority groups fluctuated around 45 percent during the period of analysis. The ΔW (the portion of the earrings gap attributed to characteristics of nonminorities that are not met by minorities) is positive and larger when the occupation variable is introduced, suggesting the existence of glass-ceiling effects in the form of barriers to access to certain human capital profiles. Furthermore, nonminorities with combinations of observable characteristics that are not 286 Table 15.2 Educational Attainment in Ecuador’s Labor Force, by Gender and Minority Status, 2003 and 2007 (percent) Men Women Year/education level Ethnic minorities Nonminorities Total Ethnic minorities Nonminorities Total 2003 None 12.4 4.1 5.3 21.9 5.4 7.8 Pre-school 0.7 0.2 0.3 0.9 0.2 0.3 Basic 61.5 50.6 52.2 54.3 47.3 48.3 a Bachillerato 20.4 30.2 28.7 17.9 30.7 28.8 Tertiary 5.0 15.0 13.5 5.0 16.3 14.7 2007 None 9.0 3.4 4.1 17.2 4.5 6.1 Pre-school 0.8 0.2 0.3 1.4 0.2 0.4 Basic 63.4 51.8 53.3 56.7 48.9 49.9 Bachilleratoa 21.2 28.5 27.5 18.6 28.9 27.6 Tertiary 5.6 16.2 14.8 6.1 17.4 16.0 Source: Based on data from 2003–07 ENEMDU. a. Equivalent to last three years of high school. Table 15.3 Occupational Distribution in Ecuador, by Gender and Minority Status, 2003 and 2007 (percent) Men Women Year/employment status Ethnic minorities Nonminorities Total Ethnic minorities Nonminorities Total 2003 Government employee 5.7 9.4 8.9 7.4 14.7 13.6 Private employee 21.9 28.9 27.8 16.2 27.3 25.8 Day laborer 26.8 27.2 27.1 12.5 5.9 6.8 Boss or employer 3.8 6.0 5.6 3.2 3.9 3.8 Self-employed 41.4 28.4 30.2 47.6 37.8 39.2 Domestic employee 0.5 0.2 0.3 13.1 10.4 10.7 2007 Government employee 6.7 9.2 8.9 9.4 14.1 13.5 Private employee 24.5 34.1 32.9 23.1 32.0 31.0 Day laborer 32.0 26.1 26.8 12.2 5.7 6.5 Boss or employer 3.1 6.4 6.0 1.7 3.9 3.7 Self-employed 33.4 23.8 25.0 42.3 35.2 36.0 Domestic employee 0.3 0.3 0.3 11.2 9.1 9.3 287 Source: Based on data from 2003–07 ENEMDU. Figure 15.1 Decomposition of Ethnic Earnings Gap in Ecuador, 2003–07 288 a. Controlling for area, education, b. Controlling for area, education, gender, gender, and age age, and head of household percentage of average earnings percentage of average earnings 48 48 38 38 of minorities of minorities 28 28 18 18 8 8 –2 –2 2003 2004 2005 2006 2007 2003 2004 2005 2006 2007 c. Controlling for area, education, gender, d. Controlling for area, education, gender, age, age, head of household, and occupation head of household, occupation, and remittances percentage of average earnings percentage of average earnings 48 48 38 38 of minorities of minorities 28 28 18 18 8 8 –2 –2 2003 2004 2005 2006 2007 2003 2004 2005 2006 2007 Δ0 ΔNW ΔX ΔW Source: Based on data from 2003–07 ENEMDU. Note: ΔW (ΔNW) is the part of the earnings gap attributed to the existence of nonminorities (minorities) with combinations of characteris- tics that are not met by any minorities (nonminorities). ΔX is the part of the earnings gap attributed to differences in the observable character- istics of nonminorities and minorities over the “common support.� Δ0 is the part of the earnings gap that cannot be attributed to differences in characteristics of the individuals. It is typically attributed to a combination of both unobservable characteristics and discrimination. The sum of these components equals the total earnings gap (ΔW + ΔNW + ΔX + Δ0 = Δ). no good jobs, and lower earnings: ecuador 2000–07 289 matchable to those of ethnic minorities have earnings that are, on aver- age, higher than in the rest of the economy. The ΔNW (the portion of the earnings gap attributed to characteristics of minorities that are not met by nonminorities) is small and almost negative; whether positive or negative, they do not play an important role. The part of the gap attributable to differences in observable charac- teristics, ΔX, becomes smaller as variables are added to the matching, particularly in the occupational category, which is also associated with an increase in ΔW. This tendency accounts for the fact that certain combina- tions of human capital characteristics are achieved by nonminorities but not ethnic minorities. Decompositions controlling for whether the house- hold received remittances from abroad do not change the earnings gap decompositions between these two groups. The unexplained component of the decomposition, Δ0—the part of the earnings gap that cannot be attributed to differences in characteristics of the individuals—accounts for about a fifth of the difference in earnings between minorities and nonminorities. Δ0 is smaller when matching com- parisons are used than when the traditional Blinder-Oaxaca methodology is used. This finding is relevant, as the Blinder-Oaxaca methodology has been found to overestimate the unexplained earnings differences because of its failure to take into account differences in the supports of the dis- tributions of observable characteristics (see chapter 2). Differences in the supports account for an important part of the gap (in the full set of char- acteristics, it accounts for almost one-third of the total gap). Exploring the Unexplained Component of the Ethnic Earnings Gap Figure 15.2 shows the unexplained component of the ethnic earnings gap for different percentiles of the income distribution of minorities and non- minorities when the pooled data set for the five years under study is used and the earnings each year are normalized such that average earnings of minorities are constant over time. At the lower deciles of the income dis- tribution, occupation is the most important variable explaining earnings differentials, accounting for almost a third of the difference. This outcome likely reflects the facts that ethnic minorities are clustered in agriculture and in informal sector employment and that the largest earnings gaps in Ecuador are still between jobs in agriculture and in the rest of the economy. Moreover, the income of ethnic minority workers is tied to agricultural output in a sector characterized by lower economic outcomes than in other sectors of the economy. Unexplained differences in earnings between the two groups decrease as income increases; Δ0 is smallest between the 50th and 90th percentile of the distribution. Occupation itself does not account 290 new century, old disparities Figure 15.2 Unexplained Ethnic Earnings Gap in Ecuador after Controlling for Demographic and Job Characteristics, by Percentile of Earnings Distribution, 2003–07 percentage of average earnings of minorities 60 50 40 30 20 10 0 0 10 20 30 40 50 60 70 80 90 100 area, education, gender, and age area, education, gender, age, and household area, education, gender, age, area, education, gender, age, household, household, and occupation occupation, and remittances Source: Based on data from 2003–07 ENEMDU. for any more of the earnings difference than area, education, gender, and age within those percentiles. However, toward the high end of the income distribution, Δ0 increases; none of the control variables seems to account for the ethnic earnings gap. In general, results for Ecuador are similar to those found in Brazil. In both countries, ethnic gaps are larger than gender gaps, and ethnic earn- ings gaps are larger among men than among women. Whereas differences in human capital characteristics help explain almost half of ethnic earnings gaps, they account for only a very small fraction of gender earnings gaps. Likewise, occupational segregation is important for explaining ethnic but not gender earnings gaps. Ethnic minorities in Ecuador are concentrated in agricultural and informal employment, segments of the labor markets with lower productivity than the rest of the economy. Both gender and ethnic earnings gaps are more pronounced at the lower percentiles of the earnings distribution. On the basis of these results, it can be inferred that policies aimed at reducing ethnic and gender disparities in earnings should also reduce poverty. no good jobs, and lower earnings: ecuador 2000–07 291 Note 1. For a description of the methodology used in this chapter, see chapter 2. References Black, S. E., P. J. Devereux, and K. G. Salvanes. 2003. “Is Education Inherited? Understanding Intergenerational Transmission of Human Capital.� Norwegian School of Economics and Business Administration, Oslo, and Institute for the Study of Labor, Bonn, Germany. De Ferranti, D., G. E. Perry, F. H. G. Ferreira, M. Walton, D. Coady, W. Cunningham, L. Gasparini, J. Jacobsen, Y. Matsuda, J. Robinson, K. Sokoloff, and Q. Wodon. 2003. Inequality in Latin America and the Caribbean: Breaking with History? Washington, DC: World Bank. Gallardo, M. L. 2006. “Ethnicity-Based Wage Differentials in Ecuador’s Labor Mar- ket.� Master’s thesis, Cornell University, Department of Economics, Ithaca, NY. Gallardo, M. L., and H. Ñopo, 2009. “Ethnic and Gender Wage Gaps in Ecua- dor.� RES Working Paper 4625, Inter-American Development Bank, Research Department, Washington, DC. García-Aracil, A., and C. Winter. 2006. “Gender and Ethnicity Differentials in School Attainment and Labor Market Earnings in Ecuador.� World Develop- ment 34 (2): 289–307. Larrea, C., and F. Montenegro Torres. 2005. “Ecuador.� In Indigenous Peoples, Poverty and Human Development in Latin America: 1994–2004, ed. H. A. Patrinos and G. Hall, 67–105. Washington, DC: World Bank. MacIsaac, D., and M. Rama. 1997. “Determinants of Hourly Earnings in Ecuador: The Role of Labor Market Regulations.� Policy Research Working Paper 1717, World Bank, Washington, DC. Soruco, X., G. Piani, and M. Rossi. 2008. “What Emigration Leaves Behind: The Situation of Emigrants and Their Families in Ecuador.� Latin American Research Network Working Paper R–542, Inter-American Development Bank, Washington, DC. World Bank. 2004. Ecuador Poverty Assessment. World Bank: Washington, DC. 16 Ethnic Earnings Gaps for Large Minorities: Guatemala 2000–06 Guatemala is one of the countries with the highest ethnic diversity, not only in Latin America but also in the world. However, the economic well- being of the different ethnic groups is far from homogenous. Indigenous groups represent 41 percent of Guatemala’s population. They are concentrated in rural and poor areas. Furthermore, the inci- dence of poverty in Guatemala is twice as high among indigenous people (72 percent) as nonindigenous people (36 percent) (Sauma 2004). Along the same lines, the indigenous population amounts for less than one- quarter of national consumption (Fazio 2007). As Guatemalans generate about 90 percent of their family income in labor markets (Fazio 2007), the analysis of the role of ethnic differences in earnings is important for an understanding of Guatemalans’ general well-being. To some extent, earnings gaps reflect differences in human capital characteristics. Indeed, differences in average human capital characteristics (age, education, marital status, migrant status) between indigenous and nonindigenous groups explain a little more than half of the ethnic earnings gap (Romero 2007). This chapter analyzes ethnic earnings gaps in Guatemala using data from the 2000 and 2006 National Survey of Living Conditions (Encuesta Nacio- nal de Condiciones de Vida [ENCOVI]) and the 2004 National Survey of Employment and Income (Encuesta Nacional de Empleo e Ingresos [ENEI]). The population under consideration is all employed individuals between the ages of 18 and 65; earnings are measured as hourly earnings in the main occupation. This chapter was adapted from the following source: “Gender and Ethnic Wage Gaps in Guatemala from a Matching Comparisons Perspective,� Hugo Ñopo and Alberto Gonzales, RES Working Paper 4587, Inter-American Development Bank, 2008; and Hugo Ñopo and Alberto Gonzales, “Brechas salariales por género y etni- cidad,� in Más crecimiento, más equidad, ed. Ernesto Stein, Osmel Manzano, Hector Morena, and Fernando Straface, Banco Interamericano de Desarrollo, 265–98, 2009. Alberto Gonzales is a PhD student in the department of economics at the University of Virginia, Charlottesville. 293 294 new century, old disparities The ethnic variable comes from individuals’ self-identification in surveys. Surveyed individuals were asked “To which of the following ethnic groups do you belong?� The list included 22 ethnic indigenous Mayan and 2 non-Mayan groups. Respondents who reported belong- ing to one of these ethnic groups were regarded as indigenous. Mestizos (Ladinos) and foreigners were considered nonindigenous. How Do Ethnic Minorities and Nonminorities in the Work Force Differ? Real earnings of the indigenous (minorities) population remained roughly constant during the period under review, while real earnings of nonindig- enous (nonminorities) people fell slightly, especially in urban areas (figure 16.1). The earnings gaps favored nonindigenous workers in both urban and rural areas, but the gap was larger in urban areas: whereas in urban areas, the average earnings of nonindigenous people were twice those of indigenous people, in rural areas they were 1.4 times as great (for graphs reporting on urban and rural areas, see Ñopo and Gonzales 2008). The earnings gap between low-educated and better-educated workers is enormous. The average earnings of a person with higher education is four times the average earnings of a person who did not complete second- ary education. Table 16.1 shows the average years of education for indig- enous and nonindigenous workers for 2000–06. Nonindigenous work- ers have about three more years of education than indigenous workers. Figure 16.1 Real Monthly Earnings of Indigenous and Nonindigenous Workers in Guatemala, 2000–06 2,400 2,151 2,000 1,949 1,859 quetzales (2006) 1,600 1,200 1,061 1,073 1,059 800 400 0 2000 2004 2006 nonindigenous indigenous Source: Based on data from Guatemala’s 2000 and 2006 ENCOVI and 2004 ENEI. Table 16.1 Highest Educational Level Begun or Completed by Indigenous and Nonindigenous Workers in Guatemala, 2000, 2004, 2006 (percent) 2000 2004 2006 Education Nonindigenous Indigenous Nonindigenous Indigenous Nonindigenous Indigenous Average years 6.6 3.4 6.5 3.8 7.3 4.6 Level Less than high school 74.8 91.5 67.1 84.3 76.8 89.0 University degree or more 5.2 1.1 4.6 1.4 3.3 0.7 Source: Based on data from Guatemala’s 2000 and 2006 ENCOVI and 2004 ENEI. 295 296 new century, old disparities In rural areas, where the majority of the population is indigenous, edu- cational levels achieved are systematically lower than in the urban areas. Whereas in rural areas the schooling gap by ethnicity is about one year, in urban areas it is nearly four years. During the period studied, almost 9 out of 10 indigenous workers and 7 out of 10 nonindigenous workers had not completed secondary education. These figures were higher in rural than in urban areas. Thus, the ethnic gap in education is wider in urban areas. The share of indigenous workers with higher education is very low, at just 1 percent nationally and virtually zero in rural areas. The Role of Individual Characteristics in Explaining the Ethnic Earnings Gap Based on these figures, one would expect earnings gaps to be at least partly explained by differences in human capital characteristics of dif- ferent groups. The rest of the chapter analyzes the decomposition of the ethnic earnings gap, in order to identify the part of the gap that results from educational gaps and other differences in characteristics between indigenous and nonindigenous populations.1 Matching is done based on four combinations of characteristics. The first combination includes age, marital status, and years of education. The second combination adds gender to the variables set. The third combina- tion adds migrant status. The fourth combination adds whether the person lives in the capital or not. The gaps are measured as percentages of the average earnings of the lowest income group (in this case, the indigenous group). ΔW denotes the component of the gap that can be explained by the existence of cer- tain profiles of nonindigenous workers that cannot be met by indigenous workers. ΔNW denotes the component of the gap caused by the presence of certain profiles of indigenous workers that cannot be met in the sample of nonindigenous workers. Figure 16.2 shows the decomposition at the national level, using the full set of matching characteristics. Ethnic gaps are 50–80 percent of average indigenous earnings—that is, on average, nonindigenous workers earn 50–80 percent more than indigenous workers with the same characteristics. In rural areas, the unexplained earn- ings gap is larger. ΔNW plays a significant role in both urban and rural areas: the existence of certain human capital profiles present only in the indigenous population increases the ethnic earnings gap by about 10 percentage points. Exploring the Unexplained Component of the Ethnic Earnings Gap Figure 16.3 reports the unexplained component of the ethnic earnings gap that cannot be attributed to differences in characteristics of the individuals earnings gaps and minorities: guatemala 2000–06 297 Figure 16.2 Decomposition of Ethnic Earnings Gap in Guatemala after Controlling for Demographic and Job Characteristics, 2000–06 100 80 percentage of average earnings 60 of indigenous 40 20 0 –20 2000 2004 2006 ΔX ΔNW ΔW Δ0 Source: Based on data from Guatemala’s 2000 and 2006 ENCOVI and 2004 ENEI. Note: ΔW (ΔNW) is the part of the earnings gap attributed to the existence of nonindigenous (indigenous) with combinations of characteristics that are not met by any indigenous (nonindigenous). ΔX is the part of the earnings gap attributed to differences in the observable characteristics of nonindigenous and indigenous over the “common support.� Δ0 is the part of the earnings gap that cannot be attributed to differences in characteristics of the individuals. It is typically attributed to a combination of both unobservable characteristics and discrimination. The sum of these components equals the total earnings gap (ΔW + ΔNW + ΔX + Δ0 = Δ). (Δ0) by percentile of the income distribution after controlling for the full set of observable characteristics. The unexplained gaps are larger for low- income workers; the decline in Δ0 related to higher income percentiles reverts in the highest income decile, where Δ0 increases. Table 16.2 reports the unexplained earnings gaps for different segments of the working population. Unexplained ethnic earnings gaps are smaller for younger workers (ages 18–25). They are larger for married workers, more educated workers, and men. This exploration of earnings gaps in Guatemala yields several results, suggesting some guidelines for policy discussion (see next chapter). Earnings gaps favoring men and nonindigenous workers are very large in Guatemala (Chong and Ñopo [2007] report that they are among the highest in Latin America). Differences in observable human capital 298 new century, old disparities Figure 16.3 Ethnic Earnings Gap in Guatemala after Controlling for Demographic and Job Characteristics, by Percentile of Earnings Distribution, 2000–06 70 percentage of average earnings of indigenous 60 50 40 30 20 10 0 0 10 20 30 40 50 60 70 80 90 100 percentile of earnings distribution Source: Based on data from Guatemala’s 2000–06 ENCOVI. Table 16.2 Unexplained Ethnic Earnings Gap in Guatemala Controlling for Demographic Characteristics, 2000–06 (percent) Age, education, + and marital + migrant + Characteristics status gender condition residence Age 18–25 17.2 17.7 17.5 15.6 26–35 25.7 29.0 27.9 24.4 36–45 20.2 25.6 26.3 23.8 46–55 24.3 31.1 30.7 27.7 56 or more 24.8 26.6 21.5 19.8 (continued next page) earnings gaps and minorities: guatemala 2000–06 299 Table 16.2 (continued) Age, education, + and marital + migrant + Characteristics status gender condition residence Education Nothing 22.1 20.8 20.3 19.5 Primary 21.9 25.6 24.3 22.6 Secondary 21.0 26.1 25.7 22.2 Tertiary 73.9 80.4 78.8 45.4 Marital status Married 22.9 26.9 26.7 23.7 Separated 10.1 10.7 12.7 11.8 Single 19.6 18.2 17.4 15.3 Migrant condition Nonmigrant 20.8 23.6 24.1 21.4 Migrant 15.3 21.9 21.3 19.5 Residence In capital city 18.6 21.2 20.7 21.4 Outside of 9.8 14.8 19.5 20.3 capital city Gender Women 17.3 17.5 15.4 12.7 Men 24.5 25.6 26.1 23.2 Area Urban 19.8 24.0 23.9 20.2 Rural 24.6 26.3 22.3 22.8 Total sample 21.2 24.3 23.9 21.3 Source: Based on data from Guatemala’s 2000, 2006 ENCOVI and 2004 ENEI. 300 new century, old disparities characteristics of workers, particularly education, explain about half of these earnings gaps in Guatemala. According to Latinobarometro, a polling organization, Guatemalans believe that lack of education is the principal cause of discrimination. This result is in line with the findings reported in chapter 3: educational gaps in Guatemala are among the highest in Latin America. Note 1. For a description of the methadology used in this chapter, see chapter 2. References Chong, A., and H. Ñopo. 2007. “Discrimination in Latin America: An Elephant in the Room?� Research Department Working Paper 614, Inter-American Devel- opment Bank, Washington, DC. Fazio, M. V. 2007. Economic Opportunities for Indigenous Peoples in Guatemala. Conference Edition. Washington, DC: World Bank. Ñopo, H., and A. Gonzales. 2008. “Brechas salariales por género y etnicidad en Guatemala desde una perspectiva de comparaciones emparejadas.� RES Working Paper 4588, Inter-American Development Bank, Research Depart- ment, Washington, DC. ———. 2009. “Brechas salariales por género y etnicidad.� In Más crecimiento, más equidad, ed. Ernesto Stein, Osmel Manzano, Hector Morena, and Fernando Straface, 265–98. Banco Interamericano de Desarrollo. Romero, W. 2007. “Los costos de la discriminación étnica en Guatemala.� In Diagnóstico del racismo en Guatemala, vol. 1, 69–95. Vicepresidencia de la República de Guatemala, Guatemala City. Sauma, P. 2004. “Las desigualdades étnicas y de género en el mercado de trabajo de Guatemala.� Working Paper 27/2004, International Labour Organization, Geneva. Part IV Policy Options 17 Policy Options Despite substantial improvements in human capital indicators during the past few decades, women and ethnic minorities still lag men and whites in labor markets, especially in labor earnings. Women, indigenous people, and Afro-descendants are participating more in labor markets and bring- ing greater human capital to their jobs—but labor markets still fail to reward them appropriately. This book documents the extent to which earnings disparities cor- respond to gender and ethnic differences in observable demographic and job characteristics in Latin America and the Caribbean. The first result it highlights is the role of education in explaining earnings differentials. Despite completing more years of schooling than men, women still earn less. In fact, earnings gaps for men and women the same age and with the same number of years of schooling are actually wider than the gaps observed in the data overall. Regarding ethnicity, the situation is even more problematic, as indigenous people and Afro-descendants still lag the rest of the population in years of education. Another variable that plays an important role in the analysis of gender earnings gaps is part-time work. Including this variable increases the gen- der earnings gap significantly. Comparing earnings of men and women with the same demographic and job characteristics reveals that the gender earnings gap was 34 percent in the 1990s and 30 percent by the mid- 2000s. These values are more than twice as high as the unconditional gender earnings gap. These average gender earnings gaps mask considerable heterogene- ity. The gap is more pronounced among poor and low-educated work- ers, workers employed by small firms or self-employed, people working part time, and people without formal labor contracts. The good news is that the segments of the labor market in which gender disparities are This chapter was adapted from the following source: “Pushing for Progress: Women, Work and Gender Roles in Latin America,� Hugo Ñopo, Harvard Inter- national Review 33 (2): 315–28, 2011. 303 304 new century, old disparities more pronounced have experienced the largest reductions in the earnings gaps. Brazil, for example, has the widest gender earnings disparities in the region, but it also experienced the largest declines. Ethnic earnings dispari- ties have also declined over the past few decades. Gender and ethnic gaps are narrowing, particularly in countries where they are—or were—widest. The pace at which they are doing so, however, does not seem commensurate with the pace at which women, indigenous people, and Afro-descendants have been acquiring education and human capital. Much work remains to be done to close these gaps. Policies aimed at reducing these disparities are still needed. Policies that do so should also reduce poverty, as earnings differentials are larger among the poor. Four sets of policies may be effective. Investing in Education Early in Life Girls’ educational attainment is at least as strong as boys’ in most coun- tries in the region. Attainment by minorities is well below that of non- minorities, however. More needs to be done to improve the educational attainment of minorities by providing equal access to education. Inclu- sive educational methods in the region have included bilingual educa- tion (in Bolivia, Ecuador, and Honduras); the expansion of physical access and use of innovative teaching methods that allow people with disabilities to attend regular classes (in Mexico’s Inclusion in Higher Education program); the incorporation and adaptation of curricula to emphasize multicultural heritage and the contributions of indigenous groups and people of African descent to national culture and history (in Colombia); and the linkage of education and school attendance with programs aimed at eradicating the worst forms of child labor (in Central America), Márquez et al. (2007). Policy makers could consider adopting any of these interventions. The earlier in the life cycle an intervention is made, the more effective it is (Carneiro and Heckman 2003; Heckman 2011). For this reason, some researchers and policy experts support interventions that stimulate devel- opment in early childhood—through, for example, conditional cash trans- fers complemented by quality and quantity improvements in education. Gender and ethnicity have a synergistically negative effect on indi- viduals’ labor market performance. Consequently, it makes sense for a long-run strategy to focus on indigenous girls, who underperform boys on a series of educational indicators. Policy needs to create incentives for household heads to send their girls to school, and increases in enroll- ment have to be paired with improvements in the supply of educational services. Policies aimed at boosting school attendance and improving the quality of education for minorities should take into account the lower incentives to policy options 305 completing schooling the labor market provides them, because of their con- centration in agriculture and informal labor activities, where the returns to education are lower than in other sectors. Although training in the skills required by the modern economy may induce workers to move out of these sectors, it is not clear that labor markets will absorb the workers in the short or medium run if their quality of education is not improved. Boosting Productivity and Reducing Labor Market Segregation Ethnic earnings gaps—and their unexplained component—are larger in rural areas than in urban areas. To address this problem, it is necessary to boost productivity in underperforming rural sectors, by facilitating stron- ger links with other participants in production chains and adding value to them. Localities need to develop skills relevant to their environment and respect local customs. Investments are necessary not only in infrastructure but particularly in individuals’ accumulation of human capital. Additionally, for workers at the bottom of the earnings distribution, policies aimed at reducing occupational segregation seem to be effective in reducing ethnic and gender earnings gaps. This reduction of segregation would not only reduce disparities; it would also make better use of human capital resources, improving overall economic productivity (Hsieh et al. 2012). Labor intermediation services and information campaigns (in both labor and education markets) have proven fruitful in both the developed and the developing world. Expansion of these types of programs would be useful (Autor 2009). Fostering a More Equitable Division of Household Responsibilities Unequal relations between men and women within households have important social consequences. More evenly balanced bargaining power increases employment opportunities for women and improves the nutri- tional status of household members (Calderón 2007). Part of the gender gap in earnings stems from women’s dual roles as workers and homemakers, which reduces their labor market attachment and bargaining power at work. Family-friendly policies may have the potential to reduce this gap. Policy makers could, for example, expand early childhood development facilities and extend school schedules for primary school students. Longer schools days would not only allow more women to work full time, but they would also increase the human capital of the next generation, improving its labor market outcomes. 306 new century, old disparities Most countries in the region have antidiscriminatory laws and legal pro- tections for women. Such legislation is full of good intentions—but most of these laws incentivize behaviors that reduce rather than increase gender parity. A law that mandates equal pay for men and women performing the same job, for example, may encourage employers to avoid hiring or promoting women, who are more likely than men to leave the work force. Legislation promoting parity should therefore be analyzed for both intended and unintended consequences. Equalization of maternity and paternity leave could help level the playing field regarding hiring decisions for men and women. It could also have posi- tive consequences outside the labor market. Encouraging men and women to devote the same amount of time to their newborns could help create more harmonious households, with more equitable intrahousehold bargaining and decision making. Such a rebalancing could help nurture more equitable divisions of responsibilities, time, and opportunities within households. Over time, equalization of leave following the birth of a child, together with a host of other measures, could help create a more egalitarian society. Diminishing Stereotyping The findings in this book that gender disparities in earnings are wid- est among the self-employed suggest that employer discrimination may not be a major factor accounting for such disparities in the region. To the extent that employers do discriminate, however, information can reduce it. Altonji and Pierret (2001) pioneered this notion by positing that discrimination declines with job tenure, a hypothesis they validated with U.S. data. Job tenure reveals information about a worker’s real productivity, which leads people to abandon assessments of capabilities based on stereotypes. Torero, Castillo, and Petrie (2008) show that in the absence of other information, people in Peru—like people elsewhere—use observable char- acteristics, such as gender, skin color, and height, as proxies for productiv- ity. When information about actual productivity is revealed, they replace these proxies with data. Information thus displaces discrimination. Initiatives to improve information on labor markets (employment bureaus and job intermediation, for example) can help change atti- tudes, stereotypes, and social norms. These instruments can and should compensate for the disadvantages women and minorities face, par- ticularly in terms of network building and the development of core competencies. Information can also be used to effect cultural and attitudinal changes. One fruitful avenue in this regard has been the tying of job placement with mentoring and networking programs. The entrance of women into policy options 307 the workplace has helped change the perceptions of (male and female) employers and coworkers, replacing stereotypes with facts. Tools that reach mass markets are also needed. Chong and La Ferrara (2009) illustrate how the subtle introduction of role models in Brazilian soap operas over the course of decades induced changes in fertility and divorce rates in Brazil’s middle class. Mass media campaigns that make people more aware of misperceptions about gender roles may also play an important role. But egalitarian values take time to be nurtured. Such nurturing has to start at home, during the early years, and continues at school. School systems can nurture gender stereotypes. Researchers found that two-thirds of the images of children in fourth and sixth grade textbooks in Peru were of boys (GRADE 2005). In addition, images of women were related largely to leisure and domestic work, whereas images of men were linked to work and schooling. This subtle, and most likely unconscious, communication of stereotypes needs to be eliminated. Changes in attitudes may take more than a generation to effect. Ensur- ing that they do so will require the active participation of current and future members of society, not only employers and job seekers. References Altonji, J., and C. Pierret. 2001. “Employer Learning and Statistical Discrimina- tion.� Quarterly Journal of Economics 116 (1): 313–50. Autor, David H., ed. 2009. Studies of Labor Market Intermediation. Chicago: University of Chicago Press. GRADE (Grupo de Análisis para el Desarrollo). 2005. Educación de las niñas: lecciones del proceso peruano. http://educacion-nosexista.org/repo/educdelas niasleccionesdelprocesoperuano.pdf. Calderón, M. C. 2007. “Discrimination, Marital Bargaining Power and Intra- household Allocation in Guatemala.� Population Studies Center, University of Pennsylvania, Philadelphia. Carneiro, P., and J. Heckman. 2003. “Human Capital Policy.� IZA Discussion Paper 821, Institute for the Study of Labor, Bonn, Germany. Chong, A., and E. La Ferrara. 2009. “Television and Divorce: Evidence from Brazilian Novelas.� RES Working Paper 4611, Inter-American Development Bank, Research Department, Washington, DC. Heckman, J. 2011. “The Economics of Inequality. The Value of Early Childhood Education.� American Educator 35 (1): 31–47. Hsieh, C.-T., E. Hurst, C. Jones, and P. Klenow. 2012. “The Allocation of Talent and U.S. Economics Growth.� http://klenow.com/HHJK.pdf. Márquez, G., A. Chong, S. Duryea, J. Mazza, and H. Ñopo, eds. 2007. ¿Los de afuera? patrones cambiantes de exclusión en América Latina y el Caribe. Informe de progreso económico y social en América Latina. Washington, DC: Inter-American Development Bank. 308 new century, old disparities Ñopo, H. 2011. “Pushing for Progress: Women, Work and Gender Roles in Latin America� Harvard International Review 33 (2): 315–28. Torero, M., M. Castillo, and R. Petrie. 2008. “Ethnic and Social Barriers to Cooperation: Experiments Studying the Extent and Nature of Discrimination in Urban Peru.� RES Working Paper 3246, Inter-American Development Bank, Research Department, Washington, DC. Index Boxes, figures, notes, and tables are indicated by b, f, n, and t following the page number. A unexplained components of, 50, 80 Abadía, L. K., 138 survey data, 15t age of workers Arias, O., 267 ethnic earnings gap, 249t, 258. armed forces, 55 See also specific countries Artana, D., 196 gender earnings gap. See gender Astudillo, A., 118 earnings gap; specific Atal, J. P., 41 countries Auguste, S., 196 agriculture sector. See also economic sectors ethnic earnings gap, 283 B gender earnings gap, 48, 80n1, 209 Barbados Altonji, J., 306 educational attainment, 219t, Alves, M. C., 164 221, 227t Amador, D., 143 gender earnings gap Angel-Urdinola, D., 138 (1992–2009) Arabsheibani, G. R., 164 age of workers, 219, 219t, Arcand, J. L., 267 222, 227t Argentina children in household, 219t, educational attainment 227t birth cohort in which demographic and job gender parity achieved, characteristics, 219–21t 24, 25t earnings gaps change in educational gender decompositions, 219–21t, gap 1940–84, 29, 29f 222–25, 223t, 225f, fertility rates, 5 226–30t gender earnings gap, 49t, economic sectors, 220t, 222, 50, 51f 229–30t, 231 309 310 index Barbados (continued) school attendance rates, literature review, 215–16 32–33, 33f occupational differences, 222 ethnic earnings gap by occupational experience, demographic and job 221t, 230t, 241 characteristics, 256t occupational segregation, 224 fertility rates, 5 other household member gender earnings gap, 48, with labor income, 49t, 51f 219t, 228t unexplained components overtime work, 217–18 of, 50 scope of, 217–22, 218f health indicators, 7n3 type of employment, 219t, survey data, 15t 224, 228t, 231 Bravo, D., 116 unexplained components of, Brazil 225–31, 227–30t educational attainment urban vs. rural birth cohort in which gender areas, 221–22 parity achieved, 24, 25t history and development of, ethnic earnings gap, 216–17 269t, 275t labor force participation of gender earnings gap, women, 217, 218f 166t, 170t survey data, 15t returns to education, 164, Barros, R., 164, 266 266, 277 Behrman, R., 22–23 ethnic earnings gap Bernal, R., 143 (1996–2006), Bernat, L. F., 139 265–79, 304 Birdsall, N., 164 age of workers, 269t, 275t Blinder-Oaxaca decomposition, demographic and job 10–13, 115, 116, 131, characteristics, 256t, 165, 266, 267 268–72, 269–71t, 275t Bolivia earnings gaps educational attainment, 4 decompositions, 272, 273f birth cohort in which each economic sectors, 271t, 275t country achieved gender literature review, 265–68 parity, 24, 25t occupational segregation, 272 change in educational regional, 270t, 275t gender gap 1940–84, scope of, 268–72 28, 29f type of employment, conditional expectations, 270t, 275t 31, 31f unexplained component of, ethnicity as factor, 32–34 273–77, 275t, 276f household income level and, urban vs. rural areas, 32, 33–34f 270t, 275t inclusive educational gender earnings gap methods, 304 (1996–2006), 48, 49t, probability component, 51f, 163–73, 304 31, 31f age of workers, 166t, 170t index 311 compared to ethnic earnings children in household, gap, 268 187t, 192t demographic and job confidence intervals, 203–4f, characteristics, 166–67t, 204, 208–9, 210–11f 170–71t demographic and job earnings gaps characteristics, 186–95t decompositions, 165–69, earnings gaps decompositions, 166–67t, 168f 196–203, 197f, 198t, formal vs. informal 200–201t, 202f workers, 170t economic sectors, 190t, literature review, 163–65 194–95t occupational segregation, firm size, 189t, 194t 41, 170t literature review, 184–85 regional, 166t, 170t other household member unexplained component of, with labor income, 79, 169–72, 170–71t, 187t, 192t 171f part-time workers, 189t, 193t urban, 166t, 170t scope of, 186–96 survey data, 15t type of employment, unemployment, 285 189t, 193t Brendan, D., 216 unexplained component of, Britz do Nascimento Silva, D., 266 203–4f, 203–12 Brown, R. S., 164 urban vs. rural areas, 188t, Bueno, I., 117 193t, 203, 209 labor force participation of women, 185, 212 C marital arrangements, 196, 209 Calderón, V., 23 unemployment, 183 Calvalieri, C., 267 Centro de Estudios Distributivos Camargo, J. M., 163 Laborales y Sociales Campante, F. R., 268 (CEDLAS), 14 Caribbean countries, gender CEO effect, 12, 50, 58, 64f, 179 earnings gap in, 215–42. childbearing age of women, 5 See also specific countries childcare affordability, 41 history and development of, children in household 216–17 effect of educational attainment literature review, 215–16 of women on, 22 Carneiro, F. G., 164, 165 ethnic earnings gap. See specific Central America. See also specific countries countries gender earnings gap. See gender educational attainment, 183, earnings gap; specific 185, 186–87t, 191t, 196 countries gender earnings gap Chile (1997–2006), educational attainment 183–213, 186–95t birth cohort in which each age of workers, 186t, country achieved gender 191t, 209 parity, 24, 25t 312 index Chile (continued) inclusive educational change in educational gender methods, 304 gap 1940–84, 28, 29f gender earnings gap gender earnings gap (1994–2006), 49t, 50, 51f, and, 116, 118–20, 137–61 119–21f, 128t age of workers, 140t, 144t, ethnic earnings gap by 156t demographic and job children in household, 140t, characteristics, 256t 156t gender earnings gap demographic and job (1992–2009), 49t, characteristics, 144–46t, 51f, 115–36 156–57t age of workers, 128t earnings gaps decompositions, average weekly hours 140–42t, 144–46t, worked, 118, 125, 147–55, 148–50t, 126–27f, 129t 152–54t, 156–57t, 159f confidence intervals, economic sector, 142t, 132, 133f 146t, 157t demographic and job family formation and characteristics, 128t dissolution and, 143 earnings gaps firm size, 142t, 157t decompositions, 126–31, formal vs. informal workers, 128–29t, 130–31f 142t, 143, 146t, 157t job tenure, 129t, 134–35 literature review, 138–39 labor force participation of marital status, 141t, 143, women, 118, 121, 121f 145t, 156t literature review, 115–18 metropolitan areas marital status, 128t studied, 138 occupational segregation, 127 other household member unemployment, 117, 120–25, with labor income, 122–24f 141t, 147 unexplained component of, part-time workers, 142t, 143, 132–35, 134f 146t, 157t urban vs. rural areas, 125 public vs. private workers, 143 survey data, 15t scope of, 139–47 Chilean Social Protection Survey statistical discrimination 2002 (SPS02), 116 against women, 138 Chong, A., 297, 307 type of employment, 141–46t Colmenares, G., 102 147, 156–57t Colombia unexplained component of, educational attainment 155–60 birth cohort in which each maternity laws (Law 50 of country achieved gender 1990), 138 parity, 24, 25t survey data, 15t gender earnings gap and, unemployment, 137 140t, 143, 144–45t, Colombian National Statistical 147, 156t Agency, 137 index 313 community, social, and personal stereotyping, programs to services, 168. See also diminish effects economic sectors of, 306–7 ethnic earnings gap, 252t Cunningham, W., 6 gender earnings gap, 44 “curse of dimensionality,� 13, construction sector, 231. See also 160n3, 224, 242n4 economic sectors ethnic earnings gap, 252t D gender earnings gap, 44 Household Sample Survey (CHSS, data. See methodology and data Barbados), 216 Dávila, A., 12, 184 Continuous Labor Force De Carvalho, A. P., 266 Sample Survey (CLFSS, demographic changes. Barbados), 217 Latin America and Continuous Sample Survey of Caribbean, generally, the Population (CSSP, 41, 42–45t, 46, 47t, Trinidad & Tobago), 216 67, 68–69t Contreras, D., 116 Deutsch, R., 184 Coppin, A., 216 D’Hombres, B., 267 Corley, M., 184 domestic employment, 102, 285. Correia, M., 176 See also maid effect; Costa Rica occupational segregation educational attainment, 24, 25t domestic violence prevention, 41 birth cohort in which each Dominican Republic country achieved gender educational attainment, 24, 25t parity, 24, 25t birth cohort in which each gender earnings gap, 49t, country achieved gender 50, 51f parity, 24, 25t occupational segregation, 12, formal vs. informal 40, 184 workers, 80n5 public vs. private gender earnings gap, 49t, 51f workers, 184 survey data, 16t unexplained component Duncan index of hierarchical/ of, 203f, 205, 205f, occupational segregation, 208–12 103, 105–6, 106f labor force participation of Duryea, S., 22–23 women, 185 survey data, 16t E Creole, use of, 216–17 Crespo, A. R. V., 268, 277 earnings gap cross-country heterogeneity ethnic. See ethnic earnings gap; ethnic earnings gap and, 246 specific countries gender earnings gap and, 48, gender. See gender earnings gap; 49t, 80n7 specific countries Cuevas, M., 196 methodological approach, 9–20. cultural bias favoring men’s role in See also methodology and society, 63 data 314 index economic sectors. See also literature review, 176 occupational segregation; occupational segregation, 40 specific countries scope of, 176–77 ethnic earnings gap. See specific unexplained component of, countries 179–81 gender earnings gap, 13, 43–44t, labor force participation of 44, 54f, 55, 58, 63f, 80n1 women, 175, 176 survey data, 14 survey data, 16t Ecuador educational attainment, ethnic educational attainment minority differences in, birth cohort in which each 32–34, 303, 304. See also country achieved gender specific countries parity, 24, 25t policy options early in life, change in educational gender 304–5 gap 1940–84, 28, 29f returns to schooling, 116, conditional expectations, 262n3, 294, 295t 31, 31f educational attainment, gender gender differences, 176, 176t differences in, 4, 6, inclusive educational 21–36, 41, 50, methods, 304 303, 304 probability component, birth cohort in which each 31, 31f country achieved gender ethnic earnings gap (2000–07), parity, 24, 25t 281–91 changes by cohorts, 23–24, 23f demographic and job decomposing changes in characteristics, 256t education, 25–32 earnings gaps complete primary education decompositions, 285–89, or incomplete secondary 286–87t, 288f education, 26f emigrants’ families, 284 complete secondary or language-based definition of incomplete university ethnicity, 282 education, 26f, 27 literature review, 281–83 by country, 28–29, 29f occupational segregation, 290 no schooling or incomplete scope of, 283–85, 284t primary education, “self-whitening� 25, 26f phenomenon, 284 between oldest and type of employment, 285 youngest cohort, 27–28, unemployment, 285 27–28f unexplained component of, conditional expectations 257, 289–90, 290f component, 30–31, urban vs. rural areas, 283, 284 30–31f gender earnings gap (2003–07), probability component, 49t, 50, 51f, 175–81 30–31, 30–31f earnings gaps school attendance rates, decompositions, 177–79, 32–33, 33f 178–80f university degree, 26f, 27 index 315 household income level and, demographic and job 32–33, 33–34f characteristics, 248, Latin America and Caribbean, 249–53t generally, 42t, 45t, 46, earnings gaps decompositions, 51–52, 52f, 67, 68t, 248–58, 249–53t, 70–71t, 73, 78t 255–56t, 257f literature review, 22–23 economic sectors, 252t policy options early in life, Ecuador (2000–07), 281–91. See 304–5 also Ecuador school attendance profiles, educational attainment. See 32–34, 33f ethnic minorities survey data, 14 firm size, 251t El Salvador formal vs. informal educational attainment workers, 251t birth cohort in which each Guatemala (2000–06), 293–300. country achieved gender See also Guatemala parity, 24, 25t interplay with gender, 247, 258 change in educational gender literature review, 246–47 gap 1940–84, 28, 29f occupational segregation, 247, conditional expectations, 254 31, 31f other household member with probability component, labor income, 258 31, 31f part-time workers, 251t gender earnings gap, 49t, 50, policy options, 303–4 51f, 79 type of employment, 250t earnings gaps unexplained component of, decompositions, 258–62, 261f 196–203, 197f urban vs. rural areas, 250t, occupational segregation, 184 258, 283 public vs. private workers, 184 ethnic minorities unexplained component Afro-descendent workers, 247 of, 203f, 205–6, 206f, changes affecting, 5–6 208–12 criteria for classifying ethnic labor force participation of groups as “minorities� by women, 185 country, 18t survey data, 16t defined, 19n1, 282 Encuesta Nacional de Empleo earnings gap. See ethnic earnings (ENE, Mexico), 112n1 gap Encuesta Nacional de Hogares educational attainment and, (ENAHO), 84 32–34, 248, 249t, 258, ethnic earnings gap, 245–300 262n3 age of workers, 249t, 258 indigenous population, 7n2, 247 Brazil (1996–2006), 265–79. See also Brazil F children in household, 250t confidence intervals, 258, female-headed households, 4 259–60f Fernandes, R., 267 316 index Fernández, P., 139 confidence intervals, 52–54f, fertility rates decline, 5, 22, 69 72, 72f, 75, 77f, 79f firm size, 18. See also specific cross-country heterogeneity, countries 48, 49t, 80n7 formal vs. informal workers, 18, demographic and job 80n5. See also informal characteristics, 41, sector; specific countries 42–44t, 46, 47t, 48, Fox, M. L., 164 67–68t, 70–71t demographic changes, 41, 42–45t, 46, 47t, 65, 67, G 68–69t Gallardo, M. L., 6, 282 domestic violence Ganguli, I., 5 prevention, 41 García, J., 116 earnings gaps García-Aracil, A., 176, 281–82 decompositions, 44–45, García Durán, B., 117 45t, 51t, 70–71t gender-based segregation economic sectors, 13, 43–44t, in workplace. See 44, 54f, 55, 58, 63f, 80n1 occupational segregation educational attainment and, gender earnings gap, 39–213 42t, 45t, 46, 51–52, Brazil (1996–2006), 163–73. See 52f, 66t, 67, 68t, 70–71t, also Brazil 73, 78t Central America (1997–2006), evolution at turn of 20th 183–213. See also Central century, 65–75, 66–67t America fertility and. See fertility rates Chile (1992–2009), 115–36. See decline also Chile firm size, 43t, 54–55 Colombia (1994–2006), 137–61. formal vs. informal workers, See also Colombia 41, 43t, 53f, 54–56 economic sectors, 13, 43–44t, labor market liberalization 44, 54f, 55, 58, 63f, 80n1 index vs., 65f Ecuador (2003–07), 175–81. See macroeconomic, social, and also Ecuador governance indicators, interplay with ethnicity, 247 linkages with 57–64, Latin America and Caribbean, 59–62t generally, 39–82 literature review, 40–41 age of workers, 42t, 45t, “matching-after-matching� 46, 50, 52f, 66t, 67, 68t, exercise, 73, 74t 70–71t, 78t maternity laws, 41, 138 changes (1992–2007), 64–80 men’s role in society, 63 childcare affordability, 41 microeconomic causes, 63 children in household, occupational segregation. 42t, 52f, 65, 66t, 68t, See occupational 70–71t, 78t segregation cohort approach to other household member with unexplained changes, labor income, 42t, 52f, 75–80, 77f, 78t 65, 66t, 68t, 70–71t, 78t index 317 part-time workers, 43t, 48, age of workers, 298–99t 53f, 55, 56–57, 67, 67t, demographic and job 69t, 70–71t, 78t, 303 characteristics, 256t policy options, 303–4 earnings gaps public vs. private workers, 40 decompositions, 296, 297f scope of, 41–57 educational attainment and, type of employment, 41, 294, 295t, 298–99t 42t, 53f, 55, 56, 67–69t, marital status, 298–99t 70–71t, 78t scope of, 254, 294–96, 294f unexplained components of, unexplained component of, 49t, 50–55, 56f, 57–64, 296–300, 298–99t, 298f 69, 72, 73–75, 75f urban vs. rural areas, 294, 296 urban vs. rural areas, 40, fertility rates, 5 41, 53f, 66t, 67, 68t, 69, gender earnings gap 70–71t, 78t (1997–2006), 48, 49t, well-paid jobs, access to, 50, 50, 51f, 79 70. See also CEO effect earnings gaps women’s role in society, 41 decompositions, 196–203, methodological approach, 9, 39 197f, 202–3, 202f Mexico, 101–13. See also public vs. private Mexico workers, 196 Peru (1997–2009), 49t, 50, 51f, unexplained component of, 83–100. See also Peru 206–7, 207f, 209 Gini coefficient, 277n2 labor market participation of glass ceiling effect, 58, 134, women vs. men, 196 139, 177 survey data, 16t governance, linkage with Guimarães, R., 268 unexplained gender earnings gaps, 58, 62t H Guatemala educational attainment, 4, 196 Hall, G., 6, 246 birth cohort in which each Hausmann, R., 5 country achieved gender Heckman, J., 266 parity, 24, 25t Henley, A., 164 change in educational gender Hernández, P. J., 116 gap 1940–84, 28, 29f Hernández-Zavala, M., 6 conditional expectations, Hertz, T., 12 31, 31f hierarchical segregation, 5, 103–4t, ethnicity as factor, 32–34 103–5 household income level and, Hoffmann, R., 164 32, 33–34f Honduras probability component, educational attainment 31, 31f birth cohort in which each school attendance rates, country achieved gender 32–33, 33f parity, 24, 25t ethnic earnings gap (2000–06), change in educational gender 293–300 gap 1940–84, 28, 29f 318 index Honduras (continued) demographic and job inclusive educational characteristics, 233–35t methods, 304 earnings gaps fertility rates, 5 decompositions, 232–35t, gender earnings gap, 49t, 51f 235–38, 236–37t earnings gaps economic sectors, 234t decompositions, firm size, 235t 196–203, 197f job tenure, 234–35t unexplained component of, occupational segregation, 204f, 207, 208–12, 208f 232, 238 labor force participation of other household member women, 185 with labor income, 233t survey data, 16t overtime work, 235t, 240 Hotchkiss, J., 216 part-time workers, 240 Household and Childhood public vs. private Measurement Indicators workers, 232 Survey (Encuesta de type of employment, 232, 233t Medición de Indicadores unexplained components of, sobre la Niñez y los 238–40, 239f Hogares [EMEDINHO], urban vs. rural areas, 233t Ecuador), 282 history and development of, household responsibilities, 305–6 216–17 survey data, 16t job tenure I discrimination lessened by, 306 indigenous populations. See ethnic Jamaica, 232, 240, 241 minorities Jones, F. L., 267 industries and sectors. See Jones, G., 22 economic sectors informal sector and gender earnings K gap, 41, 43t, 53f, 54–55. See also specific countries Kassouf, A. L., 172n1 Inter-American Development Kelley, J., 267 Bank, 6 Knodel, J., 22 Inter-American Development Bank’s Research Department, 14 L labor force participation J of ethnic groups, 6, 246 Jacobsen, J.P., 6 reducing segregation in Jamaica and improving educational attainment, 232, 233t productivity, 305 gender earnings gap (2003), rewarding men and whites 231–40 characteristics in labor age of workers, 232, 233t markets, 9 children in household, 232, sectoral differences between men 233t, 239 and women, 13 index 319 of women, 4, 6n1, 39, 112. See marital status. See also specific also specific countries countries linkage with unexplained change in marital and gender earnings gaps, 57, cohabitation 60–61t arrangements, effect on reducing household gender earnings gap, 69 responsibilities for, 305–6 of skilled women, 5 Labor Force Survey (Jamaica), 231 Marshall, J., 23 Labor market liberalization Mata, J., 118 index, 65f maternity laws, 306 La Ferrara, E., 307 men’s role in society, 63. See also Larrea, C., 282 stereotyping Latin America. See also specific mentoring programs, 306 countries methodology and data, 9–20 educational attainment. See Blinder-Oaxaca decomposition educational attainment, extension of, 10–13 ethnic minority differences shortcomings of, 10 in; educational attainment, “curse of dimensionality,� 13 gender differences in earnings gaps decompositions ethnic groups. See ethnic using matching, 11, 12, 13 earnings gap; ethnic “matching-after-matching� minorities exercise, 73, 74t gender earnings gap. See gender nationally representative earnings gap household survey data, inequality, 3, 6 14–19 labor force participation. See by country, 15–17t labor force participation criteria for classifying ethnic women’s progress, 4–5 groups as “minorities� by legislators, women as, 58, 64f country, 18t Leite, P. G., 268 ethnic minorities, countries Leme, M. C., 164 covered, 14 Living Standards Measurement working population as Survey (Colombia), 139 focus, 14 López, A., 116 regression-based vs. matching- López-Calva, L. F., 265 based approach, 13, 50 Loureiro, P. R. A., 165 sectoral differences between men Lovell, P., 266, 267 and women, 13 Lustig, N., 265 unexplained gender earnings gaps, 10, 13, 58, 59–62t Mexico M educational attainment, Machado, J. A. F., 118 23, 103t MacIsaac, D., 283 birth cohort in which each maid effect, 12, 50, 179 country achieved gender Maldonnado, V. M., 102 parity, 24, 25t managerial positions. See CEO change in educational gender effect gap 1940–84, 28, 29f 320 index Mexico (continued) National Socioeconomic household income level and, Characterization 32, 33–34f Survey (Encuesta inclusive educational de Caracterización methods, 304 Socioeconómico Nacional school attendance rates, [CASEN], Chile), 115, 32–33, 33f 116, 117, 123 gender earnings gap (1994–2004), National Survey of Employment 49t, 51f, 101–13 and Income (Encuesta age of workers, 103t, 105–7, Nacional de Empleo 106f, 108–9f e Ingresos [ENEI], demographic and job Guatemala), 212n3, 293 characteristics, 103t National Survey of Living earnings gaps Conditions (Encuesta decompositions, 105–12 Nacional de Condiciones economic sector, 104t de Vida [ENCOVI], firm size, 104t Guatemala), 212n3, 293 literature review, 102 National Survey of Urban marital status, 104t Employment (Encuesta occupational and hierarchical Nacional de Empleo segregation, 41, 58, 101, Urbano [ENEU], Mexico), 102, 103–4t, 103–12, 101, 102, 103, 105, 106f, 108–9f, 111f 112n1 public vs. private Néri, M., 266 workers, 104t Nicaragua unexplained components of, 80 educational attainment labor force participation of birth cohort in which each women, 101 country achieved gender survey data, 16t parity, 24, 25t microeconomic linkages with change in educational gender earnings gap, 63 gender gap 1940–84, mining sector. See also economic 29, 29f sectors fertility rates, 5 ethnic earnings gap, 55 gender earnings gap, 49t, 50, gender earnings gap, 252t 51f, 79 Montenegro, C., 116 earnings gaps Montenegro Torres, F., 282 decompositions, 196–203, Moon, M., 164 197f Moore, R., 216 occupational differences, 184 unexplained component of, N 204f, 207–8, 208–12, National Household Sample Survey 209f (PNAD, Brazil), 163 labor force participation of National Institute of Statistics women, 185 and Informatics (INEI, survey data, 17t Peru), 84 Ñopo, H., 10, 11 index 321 O Pederzini, C., 23 Peña, X., 139 occupational categories, 19, 43t, Perardel, Y., 184 85–86t Perticará, M., 117, 118 occupational segregation. See also Peru specific countries educational attainment ethnic earnings gap, 101–12, birth cohort in which each 254, 261, 262 country achieved gender gender earnings gap, 5, 12, parity, 24, 25t 40–41, 43t, 48, 54f, change in educational gender 69, 241 gap 1940–84, 28, 29f policy options to reduce, 305 ethnicity as factor, 32–34 Ometto, A. M. H., 164 household income level and, 32, 33–34f P school attendance rates, Pagán, J., 12, 185 32–33, 33f Panama ethnic earnings gap by educational attainment demographic and job birth cohort in which each characteristics, 256t country achieved gender gender earnings gap parity, 24, 25t (1997–2009), 49t, 50, conditional expectations, 51f, 83–100 31, 31f age of workers, 84, 85t, probability component, 89, 93t 31, 31f confidence intervals, 97, 98f gender earnings gap, 49t, 50, 51f demographic and job survey data, 17t characteristics, 84, Panizza, U., 184 85t, 93t Paraguay earnings gaps educational attainment, 24, 25t decompositions, 89–92, ethnic earnings gap by 90–91f, 93t, 96f demographic and job economic sectors, 86t, characteristics, 256t 92, 95t fertility rates, 5 educational attainment gender earnings gap, 49t, and, 84, 85t, 87–88f, 50, 51f 89–90, 93t survey data, 17t firm size, 85t, 94t Paredes, R., 115, 116 formal vs. informal Parker, S., 23, 102 workers, 83 part-time workers. See also specific labor force participation of countries women, 99–100, 99f ethnic earnings gap, 258, labor market reform, 83 261, 262 marital status, 94t gender earnings gap, 48, 53f, occupational segregation, 83, 55, 56–57, 67, 69t, 85–86t, 94t 70–71t, 303 part-time workers, 85t Patrinos, H. A., 6, 246 scope of, 84–89 322 index Peru (continued) Scott, K., 216 unemployment, 97–100, 100f Sectors, economic. See economic unexplained components of, sectorsself-employment, 50, 89, 92–97 41, 55, 303. See also urban vs. rural areas, 85t, 94t specific countries for gender stereotyping, 307 “type of employment� survey data, 17t “self-whitening� phenomenon, unemployment, 97–100, 100f 284 Pesquisa Nacional por Amostra senior officials. See Type of de Domicilios (National employment Survey of Sample Serrano, F., 163 Households, PNAD, Silva, N. D. V., 172n1, 267 Brazil), 265, 267 Soares, S. D. S., 266, 268 Pierret, C., 306 sociodemographic characteristics, Pisani, M. J., 185 linkage with unexplained Popova, K., 184 gender earnings productivity improvements, 305 gaps, 57 Psacharopoulos, G., 6, 40, 184 Solano, E., 6 public vs. private workers, 40. See Sookram, S., 216 also specific countries statistical discrimination against Puentes, E., 116 women, 138, 160n1 Stelcner, M., 164 stereotyping, 63 Q programs to diminish effects of, Qiang, C. Z.-W., 184 306–7 survey data. See methodology and data R Survey on Employment, Rama, M., 283 Unemployment, and Ramos, L., 164 Underemployment Rangel, M., 246 (Encuesta de Empleo, Reis, M. C., 268 Desempleo, y Rendón, T., 102 Subempleo [ENEMDU], Riveros, L., 115 Ecuador), 175 Survey on Living Conditions (Encuesta de Condiciones S de Vida [ECV], Ecuador), Saavedra, J., 6 282 Sabogal, A., 138 Székely, M., 22–23 Sachsida, A., 165 Sanhueza, C., 116 T Santos, E., 164 school attendance. See educational Tejerina, L., 267 attainment, ethnic Terrell, K., 216 minority differences in; Tiefenthaler, J., 164 educational attainment, traditional gender roles, 5 gender differences in Tzannatos, Z., 40, 184 index 323 U Venezuela, República Bolivariana de educational attainment unemployment, 80n1. See also birth cohort in which each specific countries country achieved gender unexplained components parity, 24, 25t ethnic earnings gap. See ethnic change in educational earnings gap gender gap 1940–84, gender earnings gap. See gender 29, 29f earnings gap gender earnings gap, 49t, 51f, 80 United Nations Human survey data, 17t Development Index, 217 Viarengo, M., 5 University of the West Indies (UWI), 217 urban vs. rural areas. See also W specific countries Wajnman, S., 164 ethnic earnings gap, 258, 283 Watson, P., 216 gender earnings gap, 40, 41, well-paid jobs. See also CEO 42t, 53f, 67, 68t, 69, effect 70–71t, 78t ethnic minorities’ access to, 254 Uruguay women’s access to, 50, 70 educational attainment, 24, 25t Winder, N., 41 birth cohort in which each Winter, C., 176, 281–82 country achieved gender Wodon, Q., 138 parity, 24, 25t Wood, C. H., 267 fertility rates, 5 World Bank, 3 gender earnings gap, 49t, 51f occupational segregation, 40 Y survey data, 17t Urzúa, S., 116 Yamada, G., 267 V Z Van Bronkhorst, B., 176 Zoloth, B. S., 164 ECO-AUDIT Environmental Benefits Statement The World Bank is committed to preserving Saved: endangered forests and natural resources. • 10 trees The Office of the Publisher has chosen • 4 million British thermal to print New Century, Old Disparities: units of total energy Gender and Ethnic Earnings Gaps in Latin • 1,025 pounds of net America and the Caribbean on recycled greenhouse gases paper with 50 percent postconsumer fiber in • 4,621 gallons of accordance with the recommended standards waste water for paper usage set by the Green Press Ini- • 293 pounds of solid tiative, a nonprofit program supporting pub- waste lishers in using fiber that is not sourced from endangered forests. For more information, visit www.greenpressinitiative.org. LAT IN AM ERICA N D EVELO P ME NT FORUM S E RI E S “The inequities suffered by women, particularly indigenous women and Afro-descendants in Latin America, are once again exposed with clarity and rigor in this study. Women receive lower wages for their work than men, and that difference is almost double in the case of persons belonging to an ethnic group. The evidence is increasingly clear; now it’s time for governments and international organizations to be more responsive and efficient in our actions.� — MICHELLE BACHELET Under-Secretary-General and Executive Director, UN Women; President of Chile (2006–10) “Hugo Ñopo’s book should focus us all on what is perhaps the most fundamental issue facing societies around the world: persistent gaps in opportunity and achievement based on ethnic, gender, and socioeconomic background. We must understand these gaps and their root causes and come together around an aggressive plan to address them. From what I have seen in classrooms and schools that serve some of society’s most marginalized children, we can effect changes in educational and life paths that are not merely incremental but that are in fact transformational for children, families, and ultimately communities. The first step is identifying the gaps and resolving to tackle them; this book is an invaluable resource in this regard.� — WENDY KOPP CEO and founder, Teach for America; CEO and co-founder, Teach for All; Time magazine’s list of the world’s 100 most influential people (2008) “New Century, Old Disparities adopts a sophisticated econometric methodology for measuring earnings gaps and applies it consistently across and within countries to measure gender and racial or ethnic differences. The analysis includes a dynamic dimension that sheds light on the evolution of earnings gaps over time. The book offers important insights on economic and political strategies that could be adopted to reduce inequality. As such, it is a must for any academic or policy maker interested in understanding and correcting inequality, with respect to not only Latin America and the Caribbean but also anywhere in the world.� — RONALD L. OAXACA University of Arizona and Institute for the Study of Labor (IZA) ISBN 978-0-8213-8686-6 SKU 18686