A partnership with academia

Building knowledge for trade and development

Vi Digital Library - Text Preview

New Century,old Disparities: Gender And Ethnic Earnings Gaps In Latin America And The Caribbean

Book by Hugo Ñopo/IDB/WB, 2012

Download original document (English)

This book adopts an econometric methodology for measuring earnings gaps and applies it consistently across and within countries to measure gender and racial or ethnic differences. It offers insights on economic and political strategies that could be adopted to reduce inequality.

l
New Century,
Old Disparities


GENDER AND ETHNIC
EARNINGS GAPS IN
LATIN AMERICA AND
THE CARIBBEAN


Hugo Ñopo


Pu
bl


ic
Di


sc
lo


su
re


A
ut


ho
riz


ed
Pu


bl
ic


Di
sc


lo
su


re
A


ut
ho


riz
ed


Pu
bl


ic
Di


sc
lo


su
re


A
ut


ho
riz


ed
Pu


bl
ic


Di
sc


lo
su


re
A


ut
ho


riz
ed






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


1 2 3 4 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




v


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






vii


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




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




ix


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).






xi


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




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




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


xvi contents




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




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


xviii contents




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


contents xix




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


xx contents




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


contents xxi




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


xxii contents




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


contents xxiii




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




xxv


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)






xxvii


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
Ángel-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.




xxviii


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




Part I


Overview, Methodology,
and Data






3


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.




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.




9


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.




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,




15


Table 2.1 Household Survey Data Used, by Country and Chapter


Country Survey


Gender chapters Ethnic chapters


Education Regional Subregional Country Regional Country


Argentina Encuesta Permanente de Hogares (EPH)
Encuesta Permanente de Hogares-Continua


(EPH-C)
2006


1992
2006


Barbados Continuous Labor Force Sample Survey
(CLFSS)


2004


Bolivia Encuesta Nacional de Empleo (ENE)
Encuesta Continua de Hogares (ECH)—


Mejoramiento de las Encuestas y Medición de
las Condiciones de Vida (MECOVI)


2007
1997


2006, 2007 2006


Brazil Pesquisa Nacional por Amostra de Domicilios
(PNAD)


2008 1992
2007, 2008


1996–2006 2007 1996–2006


Chile Encuesta de Caracterización Socioeconómica
Nacional (CASEN)


2006 1992
2006


1992, 1994,
1996, 1998,
2000, 2003,


2006, 2009


2006


Colombia Encuesta Nacional de Hogares — Fuerza de
Trabajo (ENH—FT)


Gran Encuesta Integrada de Hogares (GEIH) 2006


1992, 2005


2006


1994–2006


(continued next page)




16


Costa Rica Encuesta de Hogares de Propósitos Múltiples
(EHPM)


2007 1992
2006, 2007


1995, 2000,
2006


Dominican
Republic


Encuesta Nacional de Fuerza de Trabajo
(ENFT)


2007 2000
2003, 2007


Ecuador Encuesta de Condiciones de Vida (ECV)
Encuesta de Empleo, Desempleo, y Subempleo


(ENEMDU)


2006 1995, 2006
2007 2003–07 2007 2003–07


El Salvador Encuesta de Hogares de Propósitos Multiples
(EHPM)


2007 1991
2005, 2007


1995, 2000,
2005


Guatemala Encuesta Nacional de Condiciones de Vida
(ENCOVI) Encuesto Nacional de Empleo e
Ingresos (ENEI)


2006 2000, 2006 2000, 2006
2004


2006 2000, 2006
2004


Honduras Encuesta Permanente de Hogares de Propósitos
Multiples (EPHPM)


2007 1997
2007


1997, 2002,
2007


Jamaica Labor Force Survey undertaken by the
Statistical Institute (STATIN)


2003


Mexico Encuesta Nacional de Ingresos y Gastos de los
Hogares (ENIGH)


Encuesta Nacional de Empleo Urbano (ENEU)


2008 1992, 2008


2004


1994–2004


Table 2.1 Household Survey Data Used, by Country and Chapter (continued)


Country Survey


Gender chapters Ethnic chapters


Education Regional Subregional Country Regional Country




Nicaragua Encuesta Nacional de Hogares sobre Medición
de Nivel de Vida (EMNV)


2005 1993
2005


1998, 2001,
2005


Panama Encuesta de Hogares, Mano de Obra (EMO)
Encuesta de Hogares (EH)


2006 1991
2003, 2006


1997, 2002,
2006


Paraguay Encuesta de Hogares (Mano de Obra) (EH)
Encuesta Permanente de Hogares (EPH) 2007 2006, 2007


2006


Peru Encuesta Nacional de Hogares (ENAHO) 2007 1997
2006, 2007


1997–2009 2006


Uruguay Encuesta Continua de Hogares (ECH) 2007 1992
2005, 2007


Venezuela,
RB


Encuesta de Hogares Por Muestreo (EHM) 2006 1992
2004, 2006


Source: The data sources were compiled and harmonized by the Research Department of the Inter-American Development Bank and CEDLAS.


Table 2.1 Household Survey Data Used, by Country and Chapter (continued)


Country Survey


Gender chapters Ethnic chapters


Education Regional Subregional Country Regional Country


17




18 new century, old disparities


Table 2.2 Criteria for Classifying Ethnic Groups as “Minorities,”
by Country


Country Criterion


Percentage
of workers


12–65


Bolivia Self-declaration Self-declaration as Quechua,
Aymara, Guarani, Chiquitano,
Mojeño, or other


52.6


Brazil Skin color Self-declaration of skin color as
black or brown


48.5


Chile Self-declaration Self-declaration as Aymara,
Rapa nui, Quechua, Mapuche,
Atacameño, Coya, Kawaskar,
Yagan, or Diaguita


6.0


Ecuador Self-declaration Self-declaration as indigenous,
black, mulatto, or other


10.0


Guatemala Self-declaration Self-declaration as K´iche´,
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


35.1


Paraguay Self-declaration Self-declaration as Guarani
speaking


33.4


Peru Self-declaration Self-declaration as Quechua, or
Aymara; from Amazonia; or
black, mulatto, Zambo, or other


31.3


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.




21


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.




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).


Source: Based on data gathered from national household surveys, 2001–04.
Note: Figures are three-year moving averages.


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


–2
–1
0
1
2
3
4
5
6
7
8
9


10
11


19
40


19
42


19
44


19
46


19
48


19
50


19
52


19
54


19
56


19
58


19
60


19
62


19
64


19
66


19
68


19
70


19
72


19
74


19
76


19
78


19
80


19
82


19
84


av
er


ag
e


ye
ar


s
of


s
ch


oo
lin


g


birth cohort


women men gap




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)


Country
Gap for 1940
birth cohort


Birth cohort at
which the gap closes


Gap for 1984
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




Source: Based on data from national household surveys, 2001–04.


Figure 3.2 Educational Attainment of Men and Women in Labor Force in Latin America and the Caribbean
Born 1940–84


0
10
20
30
40
50
60
70


0
10
20
30
40
50
60
70


0
10
20
30
40
50
60
70


1 9
4 0


1 9
4 4


1 9
4 8


1 9
5 2


1 9
5 6


1 9
6 0


1 9
6 4


1 9
6 8


1 9
7 2


1 9
7 6


1 9
8 0


1 9
8 4


p


e


r


c


e


n


t


p


e


r


c


e


n


t


1 9
4 0


1 9
4 4


1 9
4 8


1 9
5 2


1 9
5 6


1 9
6 0


1 9
6 4


1 9
6 8


1 9
7 2


1 9
7 6


1 9
8 0


1 9
8 4


p


e


r


c


e


n


t


1 9
4 0


1 9
4 4


1 9
4 8


1 9
5 2


1 9
5 6


1 9
6 0


1 9
6 4


1 9
6 8


1 9
7 2


1 9
7 6


1 9
8 0


1 9
8 4


a. Percentage of population with no education or
incomplete primary school


b. Percentage of population with complete primary school or
incomplete secondary education


c. Percentage of population with complete secondary or
incomplete university education


p


e


r


c


e


n


t


0
10
20
30
40
50
60
70


1 9
4 0


1 9
4 4


1 9
4 8


1 9
5 2


1 9
5 6


1 9
6 0


1 9
6 4


1 9
6 8


1 9
7 2


1 9
7 6


1 9
8 0


1 9
8 4


birth cohort
women men


birth cohort
women men


d. Percentage of population with university degree


26




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


Source: Based on data from national household surveys, 2001–04.


–0.06
–0.10
–0.17


–0.39


0.49


–0.8
–0.6
–0.4
–0.2


0
0.2
0.4
0.6
0.8


19841940


ga
p


in
y


ea
rs


o
f e


du
ca


tio
n


(av
er


ag
e w


om
en


’s
ed


uc
ati


on


av
er


ag
e


m
en


’s
e


du
ca


tio
n)


–0.06
–0.13


0.23


birth cohort


no education or incomplete primary complete secondary or incomplete university


complete primary or incomplete secondary university degree




28 new century, old disparities


Source: Based on data from national household surveys, 2001–04.
Note: Figure shows change between cohorts born in 1940 and cohorts born


in 1984.


Figure 3.4 Decomposition of Difference in Educational
Gender Gap between Youngest and Oldest Cohort in Latin
America and the Caribbean, by Educational Level


–0.05
–0.10


0.36


–0.4
–0.2


0
0.2
0.4
0.6
0.8
1.0
1.2
1.4


1984/1940


0.88


ch
an


ge
in


e
du


ca
tio


na
l g


ap
(w


om
en



m


en
)


birth cohort


no education or incomplete primary complete secondary or incomplete university


complete primary or incomplete secondary university degree


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


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


Source: Based on data from national household surveys, 2001–04.


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


–2


–1


0


1


2


3


4


ga
p


in


ye
a


rs


o
f e


du
ca


tio
n


(av
er


ag
e w


o
m


en
’s


ed
uc


at
io


n



av


er
ag


e
m


en
’s


e
du


ca
tio


n)


Arg
ent


ina
Bo


livi
a


Bra
zil Ch


ile


Co
lom


bia


Co
sta


Ri
ca


Do
min


ica
n R


epu
blic


Ec
uad


or


Gu
ate


ma
la


Ho
ndu


ras
Me


xic
o


Nic
ara


gua


Pa
nam


a
Pe


ru


Pa
rag


uay


EI
Sa


lva
dor


Uru
gua


y


Ve
nez


uel
a, R


B


no education or primary incomplete


secondary complete or college incomplete college degree or more


primary complete or secondary incomplete




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.


Source: Based on data from national household surveys, 2001–04.


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


component


no education or incomplete primary


complete primary or incomplete secondary


complete secondary or incomplete
university
university degree


0.00
0.03


–0.4
–0.2


0
0.2
0.4
0.6
0.8
1.0
1.2
1.4


probability conditional expectations


–0.11
–0.19


0.33


0.87


0.14
0.02


ga
p


in
y


ea
rs


o
f e


du
ca


tio
n


(av
er


ag
e w


om
en


’s
ed


uc
ati


on


av
er


ag
e


m
en


’s
e


du
ca


tio
n)




gender differences in education 31


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.


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.


Figure 3.7 Decomposition of Changes in Educational
Gender Gap in Latin America and the Caribbean, by
Component and Country


ga
p


in


ye
a


rs


o
f e


du
ca


tio
n


(av
er


ag
e w


o
m


en
’s


ed
uc


at
io


n



av


er
ag


e
m


en
’s


e
du


ca
tio


n)


0.16


0.5
2


1.2
4


0.2
2


0.2
6


0.0
6


0.0
7


0.0
6


0.5
3


0.4
8


0.1
0


0.2
6


0.6
2


0.5
3


0.3
7


0.5
6


0.3
4


0.2
4


1.05


0.99


1.01
0.65


0.67
1.06


1.67


0.47




0.3
8


0.76


0.61


1.80
1.161.151.15 0.93


0.50


1.80


–1.0


–0.5


0


0.5


1.0


1.5


2.0


2.5


Arg
ent


ina
Bo


livi
a


Bra
zil Ch


ile


Co
lom


bia


Co
sta


Ri
ca


Do
min


ica
n R


epu
blic


Ec
uad


or


Gu
ate


ma
la


Ho
ndu


ras
Me


xic
o


Nic
ara


gua


Pa
nam


a
Pe


ru


Pa
rag


uay


EI
Sa


lva
dor


Uru
gua


y


Ve
nez


uel
a, R


B




conditional expectations probability




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


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,


Source: Based on data from national household surveys, circa 2003.


Figure 3.8 School Attendance Rates in Bolivia, Guatemala,
Mexico, and Peru, by Gender, Age, and Per Capita House-
hold Income Quintile, Circa 2003


0


10


20


30


40


50


60


70


80


90


100


6 7 8 9 10 11 12 13 14 15 16 17 18


age


pe
rc


en
ta


ge
a


tte
nd


in
g


sc
ho


ol


men, bottom quintile women, bottom quintile men, middle quintile


women, middle quintile men, top quintile women, top quintile




34 new century, old disparities


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.


Source: Based on data from national household surveys, circa 2003.


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


0


10


20


30


40


50


60


70


80


90


100


110


120


6 7 8 9 10 11 12 13 14 15 16 17 18


age


m
ea


n
y


ea
rs


o
f c


om
pl


et
ed


s
ch


oo
lin


g


men, bottom quintile women, bottom quintile men, middle quintile
women, middle quintile men, top quintile women, top quintile




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






39


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.




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


Composition
(percent)


Relative earnings
(base: average women’s


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


(continued next page)


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
technicians


9.6 15.1 208.7 182.2


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


Table 4.1 (continued)


Composition
(percent)


Relative earnings
(base: average women’s


earnings = 100)


Men Women Men Women




44 new century, old disparities


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


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.


Table 4.1 (continued)


Composition
(percent)


Relative earnings
(base: average women’s


earnings = 100)


Men Women Men Women




45


Table 4.2 Decomposition of Gender Earnings Gap in Latin America and the Caribbean after Controlling for
Demographic Characteristics, Circa 2007
(percent)


Age + Education
+ Presence of children


in the household


+ Presence of other
household member with


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).




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.




47


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
set


& Type of
employment


& Part
time & Formality & Sector & Occupation


& Small
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).




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)


Country


Δ0


Δ
Age and


education


+ Presence of
children in the


household,
presence of other
income earner in
the household,


and urban


+ Part-time,
formality,


occupation,
economic


sector, type of
employment, and


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


–60 –40 –20 0
percentage of average women’s earnings


20 40 60
Ecuador (Δ = –3.2%)
Paraguay (Δ = 6.2%)


Colombia (Δ = –0.9%)
Panama (Δ = –8.6%)
Argentina (Δ = 0.5%)


El Salvador (Δ = 3.3%)
Venezuela, RB (Δ = 0.4%)


Chile (Δ = 10.9%)
Mexico (Δ = 2.6%)


Guatemala (Δ = –3.3%)
Bolivia (Δ = –5.5%)


Costa Rica (Δ = –5.8%)
Uruguay (Δ = 5.7%)


Peru (Δ = 18.3%)
Dominican Republic (Δ = –3.1%)


Honduras (Δ = 5.6%)
Brazil (Δ = 20.5%)


Nicaragua (Δ = 1.5%)


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).


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


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


30


25


30


15


0


10


20


30


40


10


age


a. Controlling for age b. Controlling for education


c. Controlling for presence of children
under 12 in household


d. Controlling for presence of other
household member with labor


income


<
24


24
–3


4
34


–4
4


44
–5


4
>


54


no
ne


pri
ma


ry
inc


om
ple


te


pri
ma


ry
co


mp
let


e


se
co


nd
ary


in
co


mp
let


e


se
co


nd
ary


co
mp


let
e


ter
tia


ry
inc


om
ple


te


ter
tia


ry
co


mp
let


e


17
no yes


18


19


20


21


22


17


no yes


24


20


22


pe
rc


en
ta


ge
o


f a
ve


ra
ge


w
om


en
’s


e
ar


ni
ng


s


pe
rc


en
ta


ge
o


f a
ve


ra
ge


w
om


en
’s


e
ar


ni
ng


s
pe


rc
en


ta
ge


o
f a


ve
ra


ge
w


om
en


’s
e


ar
ni


ng
s


pe
rc


en
ta


ge
o


f a
ve


ra
ge


w
om


en
’s


e
ar


ni
ng


s


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


(continued next page)




more schooling, lower earnings for women 53


16


5


10


15


20


25


30


e. Controlling for urban location


g. Controlling for part-time employment


f. Controlling for type of employment


h. Controlling for formality of employment


18


20


22


24


no yes
em


plo
ye


e


se
lf-e


mp
loy


ed


em
plo


ye
d


16


18


20


22


24


no yes
16


18


20


22


24


no yes


pe
rc


en
ta


ge
o


f a
ve


ra
ge


w
om


en
’s


e
ar


ni
ng


s


pe
rc


en
ta


ge
o


f a
ve


ra
ge


w
om


en
’s


e
ar


ni
ng


s


pe
rc


en
ta


ge
o


f a
ve


ra
ge


w
om


en
’s


e
ar


ni
ng


s


pe
rc


en
ta


ge
o


f a
ve


ra
ge


w
om


en
’s


e
ar


ni
ng


s


Figure 4.2 (continued)


(continued next page)




54 new century, old disparities


i. Controlling for occupation


k. Controlling for small firm


–20


0


20


10


no
t c


las
sifi


ed


ar
m


ed
fo


rce
s


no
na


gri
cul


tur
al w


ork
ers


ag
ricu


ltur
al w


ork
ers


se
rv


ice
wo


rke
rs


ad
mi


nis
tra


tive
se


ller
s


dir
ec


tor
s


pro
fes


sio
na


ls


10


no


15


20


25


yes


j. Controlling for sector


–60


40


20


0


–20


–40


pe
rso


na
l se


rvi
ce


s


tra
ns


po
rt,s


tor
ag


e


fin
an


cia
l se


rvi
ce


s


ag
ricu


ltur
e


m
inin


g


m
an


ufa
ctu


rin
g


ele
ctr


icit
y s


up
ply


co
ns


tuc
tio


n


tra
de


, h
ote


ls,
an


d r
es


tau
ran


ts


pe
rc


en
ta


ge
o


f a
ve


ra
ge


w
om


en
’s


e
ar


ni
ng


s
pe


rc
en


ta
ge


o
f a


ve
ra


ge
w


om
en


’s
e


ar
ni


ng
s


pe
rc


en
ta


ge
o


f a
ve


ra
ge


w
om


en
’s


e
ar


ni
ng


s


Figure 4.2 (continued)


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


age + education


0


10


20


30


40


50


60


70


0
0 10 20 30 40 50 60 70 80 90 100


0 10 20 30 40 50 60 70 80 90 100


10p
er


ce
nt


ag
e


of
a


ve
ra


ge
w


om
en


’s
ea


rn
in


gs
g


ap
pe


rc
en


ta
ge


o
f a


ve
ra


ge
w


om
en


’s
ea


rn
in


gs
g


ap


20


30


40


50


60


70


percentile


percentile


age + education + presence of children in household
+ presence of other household member with labor income
+ type of employment + part time


+ urban


a. Controlling for age and education


b. Controlling by set of observable characteristics


Figure 4.3 Unexplained Gender Earnings Gap in
Latin America and the Caribbean, by Percentiles of
Earnings Distribution, Circa 2007


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




59


Table 4.5 Correlation between Gender Earnings Gap and Economic Indicators in Latin America and the
Caribbean, Circa 2007


Variable Years
Number of


countries
Correlation


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
15–19) years old 2003–07 18 0.2


Health, nutrition, and
population statistics


Household final consumption expenditure per capita
(constant 2000 U.S. dollars) 2003–07 18 –0.1 World Development Indicators


(continued next page)




60


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


Table 4.5 (continued)


Variable Years
Number of


countries
Correlation


coefficient Source




61


Employees, agriculture, male (percentage of male
employment)


2003–07 17 0.3 Gender statistics


Employees, industry, female (percentage of female
employment)


2003–07 17 0.5* Gender statistics


Employees, industry, male (percentage of male
employment)


2003–07 17 0.1 Gender statistics


Employees, services, female (percentage of female
employment)


2003–07 17 –0.4 Gender statistics


Employees, services, male (percentage of male
employment)


2003–07 17 –0.4 Gender statistics


Employment to population ratio, 15+, female
(percentage)


2003–07 18 –0.2 Millenium Development Goals


Employment to population ratio, 15+, male
(percentage)


2003–07 18 0.2 Millenium Development Goals


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
female population 15-64)


2003–07 18 0.2 Gender statistics


Labor force participation rate, male (percentage of
male population 15-64)


2003–07 18 0.2 Gender statistics


Labor market liberalization index 2007 18 0.4* Economic Freedom of the
World


(continued next page)


Variable Years
Number of


countries
Correlation


coefficient Source




62


Table 4.5 (continued)


Variable Years
Number of


countries
Correlation


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
(percentage of total)


2003–07 16 0.4* World Development Indicators


Political transformation (Bertelsmann transformation
index)


2008 18 0.2 Bertelsmann Foundation


Proportion of seats held by women in national
parliaments (percent)


2003–07 18 0.1 World Development Indicators


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


Argentina


Brazil


Chile


Colombia


Costa Rica


Dominican
Republic


Ecuador


employees, industry, female (percentage of employment)


El Salvador


Guatemala


Honduras


Mexico


Nicaragua


Panama


Paraguay


Peru
Uruguay


Venezuela, RB


R 2 = 0.18


R2 = 0.18


5


10


15


20


25


30


u
n


e
x


pl
ai


n
e


d
ge


n
de


r
e


ar
n


in
gs



ga


p
(%


)


10 20 30 40 50


fitted valuesfitted values (Peru not included)Δ0 (full set)


Figure 4.4 Female Employment in Industry versus
Unexplained Gender Earnings Gaps, Circa 2007


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


Argentina


Bolivia


Brazil


Chile


Colombia


Costa Rica


Dominican Republic


Ecuador


El Salvador


Honduras


Mexico


Nicaragua


Panama


Peru
Uruguay


Venezuela, RB


R 2 = 0.18


5


10


15


20


25


30


u
n


e
x


pl
ai


n
e


d
ge


n
de


r
e


ar
n


in
gs


g
ap


(%
)


25 30 35 40 45
women legislators, senior officials, and managers (percenatge of total)


fitted valuesΔ0 (full set)


Figure 4.5 Female Legislators, Senior Officials, and Managers
versus Unexplained Gender Earnings Gaps, Circa 2007


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


Argentina


Bolivia


Brazil


Chile


Colombia


Costa Rica


Dominican Republic


Ecuador


El Salvador


Guatemala


Honduras


Mexico


Nicaragua


Panama


Paraguay


PeruUruguay


Venezuela, RB


R 2 = 0.16


5


10


15


20


25


30


u
n


ex
pl


ai
ne


d
ge


nd
er


e
ar


ni
ng


s
ga


p
(%


)


3 4 5 6 7
labor market liberalization index


fitted valuesΔ0 (full set)


Figure 4.6 Labor Market Liberalization Index versus
Unexplained Gender Earnings Gaps, Circa 2007


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


(continued next page)


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
(base: average women’s


earnings = 100)


Circa 2007
(base: average women’s


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
incomplete


106.4 83.6 87.9 73.0


Secondary
complete


148.0 124.2 116.2 90.7


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




more schooling, lower earnings for women 67


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.


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.


Table 4.6 (continued)


Circa 1992
(base: average women’s


earnings = 100)


Circa 2007
(base: average women’s


earnings = 100)


Men Women Men Women




68 new century, old disparities


(continued next page)


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




more schooling, lower earnings for women 69


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


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.


Table 4.7 (continued)


Circa 1992 Circa 2007


Men Women Men Women




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)


Age + Education


+ Presence of
children in the


household


+ Presence of
other household


member with labor
income + Urban


+ Type of
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




71


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.




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


10


15


20


25


30


35


pe
rc


e
n


ta
ge



o


f w
o


m
en


’s
e


ar
n


in
gs


age +education +children +other
labor


income
earner


+urban +type of
employment


+time
worked


circa 2007
circa 2007


circa 1997
circa 1997


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


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


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


Table 4.9 Decomposition of Changes in Unexplained Gender
Earnings Gap in Latin America and the Caribbean
between Circa 1992 and 2007
(percent)


Characteristics


Counterfactual
change if no change


in observable
characteristics


Part of the change
attributed to


changes in observable
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.




more schooling, lower earnings for women 75


0


0


10


pe
rc


en
ta


ge
o


f a
ve


ra
ge


w
om


en
’s


e
ar


ni
ng


s


20


30


40


50


10 20 30 40 50
earnings percentile


60 70 80 90 100


circa 1997 circa 2007


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


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


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.


15 to 29 in 1992


20


pe
rc


en
ta


ge
o


f a
v


er
ag


e
w


o
m


en
’s


e
ar


n
in


gs


25


30


35


40


45


30 to 44 in 1992
cohort


45 to 59 in 1992


circa 1997
circa 1997


circa 2007
circa 2007


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)


Characteristics


Age


15–29
in 1992


30–44
in 1992


45–59
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


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.


–10


Arg
ent


ina
Bo


livi
a


Bra
zil


Ch
ile


Co
lom


bia


Co
sta


Ri
ca


Do
m


inic
an


Re
pub


lic


Ec
uad


or


El
Sa


lva
dor


Gu
ate


ma
la


Ho
ndu


ra
s


Me
xic


o


Nic
ar


agu
a


Pa
na


ma


Pa
rag


uay Pe
ru


Uru
gua


y


Ve
nez


uel
a, R


B


0


10


20


30


40


50


pe
rc


en
ta


ge
o


f w
om


en
’s


e
ar


ni
ng


s


a. Original gender earnings gap


b. Unexplained gender earnings gap after controlling for full
set of demographic and job characteristics


–10


40


30


20


10


0


50


pe
rc


en
ta


ge
o


f w
om


en
’s


e
ar


ni
ng


s


Arg
ent


ina
Bo


livi
a


Bra
zil


Ch
ile


Co
lom


bia


Co
sta


Ri
ca


Do
min


ica
n R


epu
blic


Ec
uad


or


El
Sa


lva
dor


Gu
ate


ma
la


Ho
ndu


ra
s


Me
xic


o


Nic
ara


gua


Pa
nam


a


Pa
rag


uay Pe
ru


Uru
gua


y


Ve
nez


ue
la,


RB


circa 1997 circa 1997 circa 2007 circa 2007


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.




83


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.




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


Composition
(percent)


Relative earnings (average
women’s earnings for


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)


Composition
(percent)


Relative earnings (average
women’s earnings for


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


Source: Based on 1997–2009 data from ENAHO.


Figure 5.1 Average Years of Education of Men and Women
in Peru’s Labor Force, 1997–2009


6


7


8


9


10


11


12


av
er


ag
e


ye
ar


s
of


e
du


ca
tio


n


199
7


199
8


199
9


200
0


200
1


200
2


200
3


200
4


200
5


200
6


200
7


200
8


200
9


women men


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).




88 new century, old disparities


Source: Based on 1997–2009 data from ENAHO.


Figure 5.2 Educational Levels of Men and Women in Peru’s
Labor Force, 1997–2009


a. No or incomplete primary education


b. Complete primary or incomplete secondary education


c. Tertiary education


0


19
97


19
98


19
99


20
00


20
01


20
02


20
03


20
04


20
05


20
06


20
07


20
08


20
09


19
97


19
98


19
99


20
00


20
01


20
02


20
03


20
04


20
05


20
06


20
07


20
08


20
09


19
97


19
98


19
99


20
00


20
01


20
02


20
03


20
04


20
05


20
06


20
07


20
08


20
09


5
10
15
20
25
30
35
40


pe
rc


en
t


10
5
0


15
20
25
30
35
40


pe
rc


en
t


0
5


10
15
20
25
30
35
40


pe
rc


en
t


women men


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


Source: Based on 1997–2009 data from ENAHO.


Figure 5.3 Gender Gap in Hourly Earnings in Peru,
1997–2009


–20


–10


0


10


20


30


40


50


pe
rc


en
ta


ge
o


f a
ve


ra
ge


w
om


en
’s


ea
rn


in
gs


199
7


199
8


199
9


200
0


200
1


200
2


200
3


200
4


200
5


200
6


200
7


200
8


200
9


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


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.


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).


Figure 5.4 Decomposition of Gender Earnings Gap in Peru
after Controlling for Demographic Characteristics, 1997–2009


–30


–20


–10


0


10


20


30


40


50


60


19
97


19
98


19
99


20
00


20
01


20
02


20
03


20
04


20
05


20
06


20
07


20
08


20
09


pe
rc


en
ta


ge
o


f a
ve


ra
ge


w
om


en
’s


e
ar


ni
ng


s


ΔX ΔF ΔM ΔO




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


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).


Figure 5.5 Decomposition of Gender Earnings Gap in Peru
after Controlling for Demographic and Job Characteristics,
1997–2009


−30


−20


−10


0


10


20


30


40


50


60


19
97


19
98


19
99


20
00


20
01


20
02


20
03


20
04


20
05


20
06


20
07


20
08


20
09


pe
rc


en
ta


ge
o


f a
ve


ra
ge


w
om


en
’s


e
ar


ni
ng


s


ΔX ΔF ΔM Δ0




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




93 (continued next page)


Table 5.2 Demographic and Job Characteristics of Matched and Unmatched Samples of Men and Women in Peru’s
Labor Force, 1997–2009
(percent)


Characteristics


Age, education, marital status, and
urban area


Age, education, firm size, occupation, and
economic sector


Matched
women and


men
Unmatched


men
Unmatched


women


Matched
women and


men
Unmatched


men
Unmatched


women


Average hourly earnings
(constant 1994
Peruvian soles)


2.6 2.4 4.8 5.5


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
incomplete


18.2 59.0 21.5 18.20 29.6 12.9


Primary incomplete or
secondary complete


26.4 12.8 34.2 26.50 19.2 33.9


Secondary complete or
tertiary incomplete


34.7 16.9 28.4 34.70 29.7 34.4


Tertiary complete 20.7 11.2 15.9 20.60 21.5 18.8




94


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)


Characteristics


Age, education, marital status, and
urban area


Age, education, firm size, occupation, and
economic sector


Matched
women and


men
Unmatched


men
Unmatched


women


Matched
women and


men
Unmatched


men
Unmatched


women




95


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


Source: Based on 1997–2009 data from ENAHO.
Note: Blank cells indicate that the variable is not being controlled for.


Table 5.2 (continued)


Characteristics


Age, education, marital status, and
urban area


Age, education, firm size, occupation, and
economic sector


Matched
women and


men
Unmatched


men
Unmatched


women


Matched
women and


men
Unmatched


men
Unmatched


women




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.


Source: Based on 1997–2009 data from ENAHO.


Figure 5.6 Unexplained Gender Earnings Gap in Peru after
Controlling for Demographic Characteristics, by Percentiles
of Earnings Distribution, 1997–2009


earnings percentile


−40


−20


0


20


40


60


pe
rc


en
ta


ge
o


f a
ve


ra
ge


w
om


en
’s


e
ar


ni
ng


s


0 10 20 30 40 50 60 70 80 90 100


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)




high unexplained gaps: peru, 1997–2009 97


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.


Source: Based on 1997–2009 data from ENAHO.


Figure 5.7 Unexplained Gender Earnings Gap in Peru after
Controlling for Demographic and Job Characteristics, by
Percentiles of Earnings Distribution, 1997–2009


0


20


40


60


80


pe
rc


en
ta


ge
o


f a
ve


ra
ge


w
om


en
’s


e
ar


ni
ng


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)
+ occupation


set 1 + full labor set (firm size, occupation)




98 new century, old disparities


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 5.8 Confidence Intervals for Unexplained Gender
Earnings Gap in Peru after Controlling for Demographic
Characteristics, 1997–2009


−30


−20


−10


0


10


20


30


40


50


60


70


199
7


199
8


199
9


200
0


200
1


200
2


200
3


200
4


200
5


200
6


200
7


200
8


200
9


pe
rc


en
ta


ge
o


f a
ve


ra
ge


w
om


en
’s


e
ar


ni
ng


s


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


−40


−20


0


20


40


60


80


199
7


199
8


199
9


200
0


200
1


200
2


200
3


200
4


200
5


200
6


200
7


200
8


200
9


pe
rc


en
ta


ge
o


f a
ve


ra
ge


w
om


en
’s


e
ar


ni
ng


s




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


Source: Based on 1997–2009 data from ENAHO.


Figure 5.10 Labor Force Participation Rates of Men and
Women in Peru, 1997–2009


40
45
50
55
60
65
70
75
80
85
90


pa
rti


ci
pa


tio
n


ra
te


(%
)


199
7


199
8


199
9


200
0


200
1


200
2


200
3


200
4


200
5


200
6


200
7


200
8


200
9


women men




100 new century, old disparities


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.


Source: Based on 1997–2009 data from ENAHO.


Figure 5.11 Unemployment Rates of Men and Women in
Peru, 1997–2009


u
n


em
pl


oy
m


en
t r


at
e


(%
)


1


1.5


2


2.5


3


3.5


4


4.5


199
7


199
8


199
9


200
0


200
1


200
2


200
3


200
4


200
5


200
6


200
7


200
8


200
9


women men




101


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.




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


Characteristics
Occupational
segregation


Hierarchical
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


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.


Table 6.1 (continued)


Characteristics
Occupational
segregation


Hierarchical
segregation


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


quarter


quarter


quarter


a. Matching by age, education, and marital status


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


original matching on age, education, and marital status


in
de


x


0.20


0.25


0.30


0.35


0.40


b. Matching by hierarchies


original matching on hierarchies


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


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


in
de


x


0.20


0.25


0.30


0.35


0.40


c. Matching by age, education, marital status, and hierarchies


original matching on age, education, marital status, and hierarchies


in
de


x


0.20


0.25


0.30


0.35


0.40


Source: Based on data from 1994–2004 ENEU.


Figure 6.1 Estimated Counterfactual Duncan Indexes of
Occupational Segregation in Mexico, 1994–2004


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


Source: Based on data from 1994–2004 ENEU.


Figure 6.2 Estimated Counterfactual Duncan Indexes of
Hierarchical Segregation in Mexico, 1994–2004


a. Matching by age, education, and marital status


0.00


0.05


0.10


0.15


0.20


94
-III 95


-II 96
-I


96
-IV 97


-III 98
-II 99


-I
99


-IV 00
-III 01


-II 02
-I


02
-IV 03


-III 04
-II


94
-III 95


-II 96
-I


96
-IV 97


-III 98
-II 99


-I
99


-IV 00
-III 01


-II 02
-I


02
-IV 03


-III 04
-II


94
-III 95


-II 96
-I


96
-IV 97


-III 98
-II 99


-I
99


-IV 00
-III 01


-II 02
-I


02
-IV 03


-III 04
-II


quarter


in
de


x


b. Matching by occupation


0.00


0.05


0.10


0.15


0.20


quarter


in
de


x


c. Matching by age, education, marital status, and occupations


0.00


0.05


0.10


0.15


0.20


quarter


in
de


x


original matching on age, schooling, and marital status


original matching on occupations


original matching on age, schooling, marital status, and occupations




gaps and workplace segregation: mexico, 1994–2004 109


Source: Based on data from 1994–2004 ENEU.


Figure 6.3 Estimated Counterfactual Gender Earnings Gaps
in Mexico, 1994–2004


a. Matching by age, education, and marital status


0


10


20


30


quarter


pe
rc


en
ta


ge
o


f a
ve


ra
ge


w
om


en
’s


e
ar


ni
ng


s


b. Matching by hierarchies


0


10


20


30


94
-III 95


-II 96
-I


96
-IV 97


-III 98
-II 99


-I
99


-IV 00
-III 01


-II 02
-I


02
-IV 03


-III 04
-II


94
-III 95


-II 96
-I


96
-IV 97


-III 98
-II 99


-I
99


-IV 00
-III 01


-II 02
-I


02
-IV 03


-III 04
-II


94
-III 95


-II 96
-I


96
-IV 97


-III 98
-II 99


-I
99


-IV 00
-III 01


-II 02
-I


02
-IV 03


-III 04
-II


quarter


pe
rc


en
ta


ge
o


f a
ve


ra
ge


w
om


en
’s


e
ar


ni
ng


s


c. Matching by occupations


0


10


20


30


quarter


pe
rc


en
ta


ge
o


f a
ve


ra
ge


w
om


en
’s


e
ar


ni
ng


s


original matching on age, schooling, and marital status


original matching on hierarchies


original matching on occupations




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


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.


Source: Based on data from 1994–2004 ENEU.


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


b. Estimated changes in earnings gap as a result of changes hierarchical segregation


–60


–50


–40


–30


–20


–10


0


pe
rc


en
ta


ge
o


f a
ve


ra
ge


w
om


en
’s


ea
rn


in
gs


0


40


80


120


160


III-
94 I-9


5
III-


95 I-9
6


III-
96 I-9


7
III-


97 I-9
8


III-
98 I-9


9
III-


99 I-0
0


III-
00 I-0


1
III-


01 I-0
2


III-
02 I-0


3
III-


03 I-0
4


III-
04


III-
94 I-9


5
III-


95 I-9
6


III-
96 I-9


7
III-


97 I-9
8


III-
98 I-9


9
III-


99 I-0
0


III-
00 I-0


1
III-


01 I-0
2


III-
02 I-0


3
III-


03 I-0
4


III-
04


quarter


quarter


pe
rc


en
ta


ge
o


f a
ve


ra
ge


w
om


en
’s


ea
rn


in
gs


men component (MC) women component (FC) total (MC+FC)




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.






115


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.




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


Source: Based on data from 1992–2009 CASEN.


Figure 7.1 Average Years of Education of Men and Women
in Chile’s Labor Force, 1992–2009


0


2


4


6


8


10


12


14


1992 1994 1996 1998 2000 2003 2006 2009


av
er


ag
e


ye
ar


s
of


e
du


ca
tio


n


women men




120 new century, old disparities


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


Source: Based on data from 1992–2009 CASEN.


Figure 7.2 Percentage of Men and Women in Chile’s Labor
Force with University Degrees, 1992–2009


0


2


4


6


8


10


12


14


16


pe
rc


e
n


t


women men


1992 1994 1996 1998 2000 2003 2006 2009




gaps and uneven participation: chile, 1992–2009 121


Source: Based on data from 1992–2009 CASEN.


Figure 7.3 Percentage of Men and Women in Chile’s Labor
Force with Less Than Secondary Education, 1992–2009


0


10


20


30


40


50


60


70


pe
rc


e
n


t


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


0


10


20


30


40


50


60


1992 1994 1996 1998 2000 2003 2006 2009


pe
rc


e
n


t


women men




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


Source: Based on data from 1992–2009 CASEN.


Figure 7.5 Unemployment Rates of Men and Women in
Chile, 1992–2009


0


1


2


3


4


5


6


1992 1994 1996 1998 2000 2003 2006 2009


pe
rc


e
n


t


women men




gaps and uneven participation: chile, 1992–2009 123


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


Source: Based on data from 1992–2009 CASEN.


Figure 7.6 Unemployment Rates of Men and Women in
Chile, by Educational Level, 1992–2009


0


1


2


3


4


5


6


7


8


1992 1994 1996 1998 2000 2003 2006 2009


pe
rc


en
t


a. Women


b. Men


0


1


2


3


4


5


6


7


8


1992 1994 1996 1998 2000 2003 2006 2009


pe
rc


en
t


no education or incomplete primary complete secondary or incomplete university
complete primary or incomplete secondary tertiary education


no education or incomplete primary complete secondary or incomplete university
complete primary or incomplete secondary tertiary education




124 new century, old disparities


Source: Based on data from 1992–2009 CASEN.


Figure 7.8 Average Weekly Hours Worked by Men and
Women in Chile, 1992–2009


0


10


20


30


40


50


60


1992 1994 1996 1998 2000 2003 2006 2009


n
u


m
be


r o
f h


ou
rs


women men


Source: Based on data from 2000–09 CASEN.


Figure 7.7 Average Years at Same Job by Men and Women
in Chile, 2000–09


2000
0


1


2


3


4


5


6


7


8


9


2003 2006 2009


n
u


m
be


r o
f y


ea
rs


women men




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


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.


Source: Based on data from 1992–2004 CASEN.


Figure 7.9 Average Weekly Hours Worked by Men and
Women in Chile by Educational Level, 1992–2009


0


10


20


30


40


50


60


1992 1994 1996 1998 2000 2003 2006 2009


n
u


m
be


r o
f h


ou
rs


0


10


20


30


40


50


60


1992 1994 1996 1998 2000 2003 2006 2009



n


u
m


be
r o


f h
ou


rs


no education or incomplete primary complete secondary or incomplete university
complete primary or incomplete secondary teritary education


no education or incomplete primary complete secondary or incomplete university
complete primary or incomplete secondary teritary education


a. Women


b. Men




gaps and uneven participation: chile, 1992–2009 127


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


Source: Based on data from 1992–2009 CASEN.


Figure 7.10 Gender Gap in Hourly Earnings in Chile,
1992–2009


0


5


10


15


20


25


30


35


40


1992 1994 1996 1998 2000 2003 2006 2009


pe
rc


en
ta


ge
o


f a
ve


ra
ge


w
om


en
’s


e
ar


ni
ng


s




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)


Characteristics


Age, education, marital status,
and occupation


Age, education, marital status, time
worked, and occupation


Age, education,marital status, and
years at the same job


Matched
sample


Unmatched
women


Unmatched
men


Matched
sample


Unmatched
women


Unmatched
men


Matched
sample


Unmatched
women


Unmatched
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




129


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.


Table 7.1 (continued)


Characteristics


Age, education, marital status,
and occupation


Age, education, marital status, time
worked, and occupation


Age, education,marital status, and
years at the same job


Matched
sample


Unmatched
women


Unmatched
men


Matched
sample


Unmatched
women


Unmatched
men


Matched
sample


Unmatched
women


Unmatched
men




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


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).


Figure 7.11 Decomposition of Gender Earnings Gap in
Chile after Controlling for Demographic and Job
Characteristics, 1992–2009


–0.3


–0.2


–0.1


0


0.1


0.2


0.3


0.4


0.5


0.6


1992 1994 1996 1998 2000 2003 2006 2009


pe
rc


en
ta


ge
o


f a
ve


ra
ge


w
om


en
’s


e
ar


ni
ng


s


Δ0ΔMΔFΔX




gaps and uneven participation: chile, 1992–2009 131


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.


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).


Figure 7.12 Decomposition of Gender Earnings Gap in Chile
after Controlling for Age, Marital Status, Education, and
Years in Same Occupation, 2000–09


–0.2


–0.1


0


0.1


0.2


0.3


0.4


0.5


2000 2003 2006 2009


pe
rc


en
ta


ge
o


f a
ve


ra
ge


w
om


en
’s


e
ar


ni
ng


s


Δ0ΔMΔFΔX




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


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


Source: Based on data from 1992–2009 CASEN.
Note: Boxes show 90 percent confidence intervals for unexplained earnings;


whiskers show 99 percent confidence intervals.


Figure 7.13 Confidence Intervals for Unexplained Gender
Earnings Gap in Chile, 1992–2009


0


10


20


30


40


50


60


1992 1994 1996


a. Unexplained earnings gap after controlling for age, marital status,
education, time worked, and occupation, 1992–2009


pe
rc


en
ta


ge
o


f a
ve


ra
ge


w
om


en
’s


e
ar


ni
ng


s
pe


rc
en


ta
ge


o
f a


ve
ra


ge
w


om
en


’s
e


ar
ni


ng
s


b. Unexplained earnings gap after controlling for age, marital status,
education, and years at the same job, 2000–09


1998 2000 2003 2006 2009


0


5


10


15


20


25


30


2000 2003 2006 2009




134 new century, old disparities


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.


Source: Based on data from 1992–2009 CASEN.


Figure 7.14 Unexplained Gender Earnings Gap in Chile, by
Percentiles of Earnings Distribution, 1992–2009


80


60


40


20


0


0 10 20 30 40 50


earnings percentile


pe
rc


en
ta


ge


o
f a


v
er


ag
e


w
o


m
en


’s
e


ar
n


in
gs


60 70 80 90 100


year, age, marital status, and
education


set 1 (year, age, marital status, and
education) + occupation


set 1 (year, age, marital status, and
education) + time worked
full set (year, age, marital status, education,
time worked, and occupation)




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.




137


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.




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.




140


Table 8.1 Relative Hourly Earnings of Men and Women in Colombia, 1994–2006


Characteristics


1994–98 2000–01 2002–06


(Base: average women’s earnings
for each year = 100)


(Base: average women’s
earnings for each year = 100)


(Base: average women’s
earnings for each year = 100)


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




141


Table 8.1 (continued)


Characteristics


1994–98 2000–01 2002–06


(Base: average women’s earnings
for each year = 100)


(Base: average women’s
earnings for each year = 100)


(Base: average women’s
earnings for each year = 100)


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)




142


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.


Table 8.1 (continued)


Characteristics


1994–98 2000–01 2002–06


(Base: average women’s earnings
for each year = 100)


(Base: average women’s
earnings for each year = 100)


(Base: average women’s
earnings for each year = 100)


Women Men Women Men Women Men




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).




144


Table 8.2 Demographic and Job Characteristics of Men and Women in Colombia’s Labor Force, 1994–2006
(percent)


Characteristics


1994–98 2000–01 2002–06


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




145


Table 8.2 (continued)


Characteristics


1994–98 2000–01 2002–06


Women Men Women Men Women Men


Tertiary complete 12.2 11.1 12.7 11.2 16.5 14.1


Presence of children
(6 years or younger in
the household)


22.7 29.6 21.3 26.6 19.2 23.5


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,
or separated


20.6 4.6 23.4 6.7 22.3 7.0


Single (never married) 35.6 29.3 31.8 27.4 33.3 28.3


Presence of other
household member
with labor income


79.5 64.5 73.3 62.1 73.5 63.8


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)




146


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
or less)


46.3 45.8 52.5 52.3 52.2 48.7


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.


Table 8.2 (continued)


Characteristics


1994–98 2000–01 2002–06


Women Men Women Men Women Men




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.




Table 8.3 Decomposition of Gender Earnings Gap in Colombia after Controlling for Demographic
Characteristics, 1994–2006
(percent)


Year, metropolitan
area, and age + Education


+ Presence of
children in the


household
+ Marital


status


+ Presence of other
household member


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


148




Table 8.3 (continued)


Year, metropolitan
area, and age + Education


+ Presence of
children in the


household
+ Marital


status


+ Presence of other
household member


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


men in common
support


99.9 96.9 92.3 73.6 55.9


Percentage of
women in
common support


100.0 98.7 96.1 74.1 61.2


149


(continued next page)




Table 8.3 (continued)


Year, metropolitan
area, and age + Education


+ Presence of
children in the


household
+ Marital


status


+ Presence of other
household member


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


men in common
support


99.9 97.6 93.6 75.9 57.9


Percentage of
women in
common support


100.0 98.7 96.3 74.5 61.6


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).


150




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.




152 Table 8.4 Decomposition of Gender Earnings Gap in Colombia after Controlling for Demographic and Job
Characteristics, 1994–2006
(percent)


Demographic
set


& Small
firm & Sector & Occupation


& Type of
employment & Formality


& Time
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)




153


Demographic
set


& Small
firm & Sector & Occupation


& Type of
employment & Formality


& Time
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


Table 8.4 (continued)


(continued next page)




Table 8.4 (continued)


Demographic
set


& Small
firm & Sector & Occupation


& Type of
employment & Formality


& Time
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).


154




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


men
Unmatched


women
Unmatched


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


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.


Table 8.5 (continued)


Matched
women and


men
Unmatched


women
Unmatched


men


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


–10 0 10 20 30


2006 (Δ = 10.6%)


2005 (Δ = 13.4%)


2004 (Δ = 14.4%)


2003 (Δ = 15.5%)


2002 (Δ = 14.8%)


2001 (Δ = 14.1%)


2000 (Δ = 13.3%)


1998 (Δ = 19.8%)


1996 (Δ = 18.4%)


1994 (Δ = 17.2%)


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).


Figure 8.1 Decomposition of Gender Earnings Gap in
Colombia after Controlling for Demographic and Job
Characteristics, 1994–2006


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


0


10


20


30


40


50


60


70


earnings percentile


pe
rc


en
ta


ge
o


f a
ve


ra
ge


w
om


en
’s


e
ar


ni
ng


s


original gap


after controlling for demographic and job characteristics
after controlling for demographic characteristics


0 10 20 30 40 50 60 70 80 90 100


Source: Based on data from household surveys for 1994–2006 conducted
by the Colombian National Statistical Agency.


Figure 8.2 Gender Earnings Gaps in Colombia after
Controlling for Demographic and Job Characteristics,
by Percentiles of Earnings Distribution, 1994–2006


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.






163


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.




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




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)


Characteristics


1996 2006


Unmatched
women


Unmatched
men


Matched
women and men


Unmatched
women


Matched
men


Matched
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


166




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


Table 9.1 (continued)


Characteristics


1996 2006


Unmatched
women


Unmatched
men


Matched
women and men


Unmatched
women


Matched
men


Matched
women and men




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


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).


Figure 9.1 Decomposition of Gender Earnings Gaps in Brazil
after Controlling for Demographic and Job Characteristics,
1996–2006


−40


−20


10


20


40


60


80


100


pe
rc


en
ta


ge
o


f a
ve


ra
ge


w
om


en
’s


e
ar


ni
ng


s


ΔMΔF Δ0ΔX


199919971996 1998 2001 2002 2003 2004 2005 2006




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


Source: Based on data from 1996–2006 PNAD.


Figure 9.2 Original and Unexplained Gender Earnings Gap
in Brazil, by Percentiles of Earnings Distribution, 1996–2006


b. Unexplained component of gender earnings gap after controlling
for demographic and job characteristics


percentile of earnings distribution


percentile of earnings distribution


pe
rc


en
ta


ge
o


f a
ve


ra
ge


w
om


en
’s


e
ar


ni
ng


s


0


20


40


60


80


100


120


0 10 20 30 40 50 60 70 80 90 100


a. Original gender earnings gap


0


20


40


60


80


100


120


0 10 20 30 40 50 60 70 80 90 100


pe
rc


en
ta


ge
o


f a
ve


ra
ge


w
om


en
’s


e
ar


ni
ng


s


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.




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.






175


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.




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
education


2003 2007


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


−20


−15


−10
−5


0


5


10
15
20


25


30


−20


−15


−10
−5


0


5


10
15
20


25


30


−20
−15
−10
−5
0
5


10
15
20
25
30


2003 2004 2005 2006 2007


2003 2004 2005 2006 2007


2003 2004 2005 2006 2007


pe
rc


en
ta


ge
o


f a
ve


ra
ge


w
om


en
’s


ea
rn


in
gs


pe
rc


en
ta


ge
o


f a
ve


ra
ge


w
om


en
’s


ea
rn


in
gs


pe
rc


en
ta


ge
o


f a
ve


ra
ge


w
om


en
’s


ea
rn


in
gs


a. Controlling for area, education,
ethnicity, and age


b. Controlling for area, education, ethnicity,
age, and head of household


c. Controlling for area, education, ethnicity,
age, head of household, and occupation


Δ0 ΔF ΔM ΔX


(continued next page)




gaps and indigenous populations: ecuador, 2003–07 179


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).


−20
−15
−10
−5
0
5


10
15
20
25
30


2003 2004 2005 2006 2007


pe
rc


en
ta


ge


o
f a


ve
ra


ge
w


om
en


’s
ea


rn
in


gs


d. Controlling for area, education, ethnicity, age,
head of household, occupation, and remittances


Δ0 ΔF ΔM ΔX


Figure 10.1 (continued)


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.


Source: Based on data from 2003–07 ENEMDU.


Figure 10.2 Unexplained Gender Earnings Gap in Ecuador
after Controlling for Demographic and Job Characteristics,
by Percentile of Earnings Distribution, 2003–07


0


10


20


30


pe
rc


en
ta


ge
o


f a
ve


ra
ge


w
o


m
en


’s
e


ar
n


in
gs


percentile of earnings distribution


40


50


60


0 10 20 30 40 50 60 70 80 90 100


area, education, ethnicity, and age (set 1)
set 1, household
set 1, household and occupation
set 1, household, occupation, and remittances




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.






183


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.




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




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
earnings in each year and


country = 100


Base: average women’s
earnings in each year and


country = 100


Base: average women’s
earnings in each year and


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


186




187 (continued next page)


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


Table 11.1 (continued)


Circa 1997 Circa 2001 Circa 2006


Base: average women’s
earnings in each year and


country = 100


Base: average women’s
earnings in each year and


country = 100


Base: average women’s
earnings in each year and


country = 100


Women Men Women Men Women Men




188


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
earnings in each year and


country = 100


Base: average women’s
earnings in each year and


country = 100


Base: average women’s
earnings in each year and


country = 100


Women Men Women Men Women Men




189


(continued next page)


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


Table 11.1 (continued)


Circa 1997 Circa 2001 Circa 2006


Base: average women’s
earnings in each year and


country = 100


Base: average women’s
earnings in each year and


country = 100


Base: average women’s
earnings in each year and


country = 100


Women Men Women Men Women Men




190


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 11.1 (continued)


Circa 1997 Circa 2001 Circa 2006


Base: average women’s
earnings in each year and


country = 100


Base: average women’s
earnings in each year and


country = 100


Base: average women’s
earnings in each year and


country = 100


Women Men Women Men Women Men




191


(continued next page)


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




192


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




193


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)


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


194 Table 11.2 (continued)


Circa 1997 Circa 2001 Circa 2006


Women Men Women Men Women Men




195


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.


Table 11.2 (continued)


Circa 1997 Circa 2001 Circa 2006


Women Men Women Men Women Men




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


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


Source: Based on 1995–2007 national household surveys of Costa Rica, El
Salvador, Honduras, and Nicaragua.


Figure 11.1 Gender Earnings Gap in Central American
Countries, Circa 1997–2006


−15


−10


−5


0


5


10


15


20


pe
rc


en
ta


ge
o


f a
ve


ra
ge


w
om


en
’s


e
ar


ni
ng


s


(circa 2006)(circa 1997) (circa 2001)


Costa Rica
Honduras Nicaragua


El Salvador




Table 11.3 Decomposition of Gender Earnings Gap in Central America after Controlling for Demographic
Characteristics, Circa 1997


(percent)


Circa 1997—Guatemala not included


Age + Education


+ Presence of
children in the


household


+ Presence of
other household


member with
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


common support
100.0 98.5 96.0 89.2 72.9 59.5


Percentage of women in
the common support


100.0 99.5 98.7 95.8 83.6 74.0


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).


198




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




Table 11.4 Decomposition of Gender Earnings Gaps in Central American Countries after Controlling for
Demographic Characteristics, Various Years
(percent)


Period


Costa Rica El Salvador Honduras Nicaragua


Area and
education


Urban,
education,


age,
head of


household,
marital


status, and
occupation


Area and
education


Urban,
education,


age,
head of


household,
marital


status, and
occupation


Area and
education


Urban,
education,


age,
head of


household,
marital


status, and
occupation


Area and
education


Urban,
education,


age,
head of


household,
marital


status, and
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


200




201


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).


Table 11.4 (continued)


Period


Costa Rica El Salvador Honduras Nicaragua


Area and
education


Urban,
education,


age,
head of


household,
marital


status, and
occupation


Area and
education


Urban,
education,


age,
head of


household,
marital


status, and
occupation


Area and
education


Urban,
education,


age,
head of


household,
marital


status, and
occupation


Area and
education


Urban,
education,


age,
head of


household,
marital


status, and
occupation




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


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).


Figure 11.2 Decomposition of Gender Earnings Gap in
Guatemala after Controlling for Age, Marital Status,
and Education, 2000–06


−5


0


5


10


15


20


25


30


2000 2004 2006


pe
rc


en
t o


f a
ve


ra
ge


w
om


en
’s


ea
rn


in
gs


Δ0ΔM ΔF ΔX




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


0


5


10


15


20


25


1995 2000


a. Costa Rica, 1995–2006


2006


pe
rc


en
ta


ge
o


f a
ve


ra
ge


w
om


en
’s


ea
rn


in
gs


b. El Salvador, 1995–2006


1995 2000 2005
0


5


10


15


20


25


30


35


pe
rc


en
ta


ge
o


f a
ve


ra
ge


w
om


en
’s


ea
rn


in
gs


(continued next page)




204 new century, old disparities


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 (continued)


d. Nicaragua, 1998–2005


–30
–20
–10


0
10
20
30
40
50
60


pe
rc


en
ta


ge
o


f a
ve


ra
ge


w
om


en
’s


ea
rn


in
gs


1995 2000 2005


c. Honduras, 1997–2007


1997 2002 2007
–10


–5


0


5


10


15


20


25
pe


rc
en


ta
ge


o
f a


ve
ra


ge
w


om
en


’s
ea


rn
in


gs


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


Source: Based on data from Costa Rica’s 2006 national household surveys.


Figure 11.4 Unexplained Gender Earnings Gap in
Costa Rica after Controlling for Demographic and Job
Characteristics, by Percentile of Earnings Distribution, 2006


–20


0


20


40


60


80


100


120


pe
rc


en
ta


ge
o


f a
ve


ra
ge


w
om


en
’s


e
ar


ni
ng


s


area area, education, and age
area, education, age, head of household, marital status, and occupation


percentile of earnings distribution


0 10 20 30 40 50 60 70 80 90 100




206 new century, old disparities


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


Source: Based on data from 2005 El Salvador’s national household surveys.


Figure 11.5 Unexplained Gender Earnings Gap in El
Salvador after Controlling for Demographic and Job
Characteristics, by Percentile of Earnings Distribution, 2005


0


20


40


60


80


100


120


140


160


180


percentile of earnings distribution


pe
rc


en
ta


ge
o


f a
ve


ra
ge


w
om


en
’s


e
ar


ni
ng


s


area area, education, and age
area, education, age, head of household, marital status, and occupation


0 10 20 30 40 50 60 70 80 90 100




heterogeneity within central america, 1997–2006 207


Source: Based on data from Guatemala’s 2000–06 national household
surveys.


Figure 11.6 Unexplained Gender Earnings Gap in Guatemala
after Controlling for Demographic and Job Characteristics,
by Percentile of Earnings Distribution, 2000–06


–10


10


30


50


70


90


110


0 10 20 30 40 50 60 70 80 90 100
percentile of earnings distribution


pe
rc


en
ta


ge
o


f a
ve


ra
ge


w
om


en
’s


e
ar


ni
ng


s


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


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


Source: Based on data from Honduras’ 2007 national household survey.


Figure 11.7 Unexplained Gender Earnings Gap in Honduras
after Controlling for Demographic and Job Characteristics,
by Percentile of Earnings Distribution, 2007


–20


0


20


40


60


80


100


120


0 10 20 30 40 50 60 70 80 90 100
percentile of earnings distribution


pe
rc


en
ta


ge
o


f a
ve


ra
ge


w
om


en
’s


e
ar


ni
ng


s


area area, education, and age
area, education, age, head of household, marital status, and occupation




heterogeneity within central america, 1997–2006 209


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


Source: Based on 2005 data from Nicaragua’s national household survey.


Figure 11.8 Unexplained Gender Earnings Gap in Nicaragua
after Controlling for Demographic and Job Characteristics,
by Percentile of Earnings Distribution, 2005


–60


–40


–20


0


20


40


60


80


100


120


0 10 20 30 40 50 60 70 80 90 100
percentile of earnings distribution


pe
rc


en
ta


ge
o


f a
ve


ra
ge


w
om


en
’s


e
ar


ni
ng


s


area area, education, and age
area, education, age, head of household, marital status, and occupation




210 new century, old disparities


Figure 11.9 Confidence Intervals for Unexplained Earnings
Gaps in Central America after Controlling for Demographic
Characteristics, Circa 2006


0
02
04
06
08
10
12
14
16
18
20


area


a. By area


pe
rc


en
ta


ge
o


f w
om


en
’s


e
ar


ni
ng


s


urban rural


c. By years of education


0


5


10


15


20


25


education


pe
rc


en
ta


ge
o


f w
om


en
’s


e
ar


ni
ng


s


low medium high


b. By age


–30


–20


–10


0


10


20


30


40


age


pe
rc


en
ta


ge
o


f w
om


en
’s


e
ar


ni
ng


s


0–14 15–24 25–40 41–64 65 or older


(continued next page)




heterogeneity within central america, 1997–2006 211


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.


Figure 11.9 (continued)


f. By occupation


pe
rc


en
ta


ge


o
f w


om
en


’s
e


ar
ni


ng
s


pe
rc


en
ta


ge


o
f w


om
en


’s
e


ar
ni


ng
s


–2


–1


0


1


2


3


4


5


6


occupation


d. By marital status


e. By head of household status


–70
–60
–50
–40
–30
–20
–10


0
10
20
30
40


years


single formal or
informal partners


divorced or
separated


widow


pro
fes


sio
na


ls


dir
ec


tor
s


ad
mi


nis
tra


tive
s


se
ller


s a
nd


m
erc


ha
nts


se
rv


ice
wo


rke
rs


ag
ricu


ltur
al w


ork
ers


no
na


gri
cul


tur
al


wo
rke


rs
oth


ers


0


5


10


15


20


25


30


head of the household


not household head household head


pe
rc


en
ta


ge


o
f w


om
en


’s
e


ar
ni


ng
s




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.






215


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.




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


Source: Based on data from 2004 CLFSS.


Figure 12.1 Distribution of Weekly Earnings of Men and
Women in Barbados, by Earnings Interval, 2004


a. Men b. Women
30


20


pe
rc


en
t


pe
rc


en
t


10


0


$20
0–$


299


un
de


r $2
00


$30
0–$


399


$40
0–$


499


$50
0–$


599


$60
0–$


699


$70
0–$


799


$80
0–$


899


$90
0–$


999


$1,
00


0–
$1,


300


ov
er


$1
,30


0


30


20


10


0


$20
0–$


299


un
de


r $2
00


$30
0–$


399


$40
0–$


499


$50
0–$


599


$60
0–$


699


$70
0–$


799


$80
0–$


899


$90
0–$


999


$1,
00


0–
$1,


300


ov
er


$1
,30


0


Source: Based on data from 2004 CLFSS.


Figure 12.2 Women’s Participation in the Labor Force in
Barbados, by Earnings Interval, 2004


$20
0–$


299


un
de


r $2
00


30


40


50


60


70


pe
rc


en
t


$30
0–$


399


$40
0–$


499


$50
0–$


599


$60
0–$


699


$70
0–$


799


$80
0–$


899


$90
0–$


999


$1,
000


–$1
,30


0


ov
er


$1
,30


0


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


(continued next page)


Table 12.1 Demographic and Job Characteristics and Relative
Earnings of Men and Women in Labor Force in Barbados, 2004


Composition (%)


Earnings index
(Base: average


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




220 new century, old disparities


(continued next page)


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


Table 12.1 (continued)


Composition (%)


Earnings index
(Base: average


women’s earnings = 100)


Men Women Men Women




understudied gaps: barbados, 2004 and jamaica, 2003 221


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,


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.


Table 12.1 (continued)


Composition (%)


Earnings index
(Base: average


women’s earnings = 100)


Men Women Men Women




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




223


Table 12.2 Decomposition of Earnings Gap in Barbados after Controlling for Demographic
Characteristics, 2004
(percent)


Age + Education
+ Presence of children


in the household


+ Presence of other
household member with


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).




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


Source: Based on data from 2004 CLFSS.


Figure 12.3 Proportion of Unmatched Women in Barbados,
by Earnings Interval, 2004


$20
0–$


299


un
de


r $2
00


10


20


30


40


50


pe
rc


en
t


$30
0–$


399


$40
0–$


499


$50
0–$


599


$60
0–$


699


$70
0–$


799


$80
0–$


899


$90
0–$


999


$1,
000


–$1
,30


0


ov
er


$1
,30


0


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).




227


Table 12.4 Unexplained Gender Earnings Gap in Barbados after Controlling for Demographic and Job
Characteristics, 2004
(percent)


Demographic
set


& Type of
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


Yes 14.6 13.6 20.6 11.2 19.8 23.9


(continued next page)




228


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


Table 12.4 (continued)


Demographic
set


& Type of
employment


&
Occupation


&
Sector


&
Experience Full set




229


Technicians and associate
professionals


16.6 4.8


Clerks 13.1 15.5


Service, shop, and market
sales workers


43.0 27.5


Skilled agricultural and
fishery workers


8.4 8.4


Craft and related trades
workers


52.1 23.1


Plant and machine operators
and assemblers


76.9 197.7


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
hotels and restaurants


19.0 22.9


(continued next page)


Demographic
set


& Type of
employment


&
Occupation


&
Sector


&
Experience Full set




230


Transport, storage, and
communication


20.5 20.9


Finance, insurance, real
estate, and business services


11.8 21.3


Community, social, and
personal services


5.2 1.5


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.


Table 12.4 (continued)


Demographic
set


& Type of
employment


&
Occupation


&
Sector


&
Experience Full set




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


(continued next page)


Table 12.5 Demographic and Job Characteristics and Relative
Hourly Earnings of Men and Women in Jamaica’s Labor Force, 2003


Composition
(%)


Earnings Index
(Base: average


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




234 new century, old disparities


Occupation


Armed forces 0.3 0.0 161.3 —


Legislators and senior
officials


2.6 7.6 166.5 121.1


Professionals 3.9 10.3 212.6 208.5


Technicians and associate
professionals


3.6 5.8 175.8 124.3


Clerks 2.5 12.1 133.5 122.2


Service, shop, and
market sales workers


11.1 24.4 108.9 86.3


Skilled agricultural and
fishery workers


30.8 8.1 58.1 47.2


Craft and related trades
workers


22.2 4.0 116.8 70.5


Plant and machine
operators and assemblers


8.5 1.7 113.7 78.6


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
hotels and restaurants


13.8 32.3 94.3 81.9


Transport, storage, and
communication


8.0 2.7 119.9 157.1


Finance, insurance, real
estate, and business services


4.4 5.7 133.1 177.2


Community, social, and
personal services


16.3 43.8 136.1 111.8


Experience


Less than 3 months 1.9 2.3 84.3 80.2


Table 12.5 (continued)


Composition
(%)


Earnings index
(Base: average


women’s earnings =
100)


Men Women Men Women


(continued next page)




understudied gaps: barbados, 2004 and jamaica, 2003 235


Table 12.5 (continued)


Composition
(%)


Earnings index
(Base: average


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)


Age + Education
+ Presence of children


in the household


+ Presence of other
household member with


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


in common support
100.0 99.5 98.0 94.8 88.7


Percentage of men in
common support


99.9 98.3 96.9 94.3 88.7


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).




237


Table 12.7 Decomposition of Gender Earnings Gap in Jamaica after Controlling for Demographic and
Job Characteristics, 2003
(percent)


Demographic
set


& Type of
employment


&
Occupation


&
Sector


&
Experience


&
Firm


size


&
Time


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


in common support
88.7 69.4 37.4 41.3 70.7 79.7 72.4 5.8


Percentage of men in
common support


88.7 72.6 46.3 52.0 64.9 82.5 74.0 5.5


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).




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


Source: Based on data from 2003 Labor Force Survey.


Figure 12.4 Unexplained Gender Earnings Gap in Jamaica,
after Controlling for Demographic Characteristics, by
Percentile of Earnings Distribution, 2003


1009080706050403020100
−50


0


50


pe
rc


en
ta


ge
o


f a
v


er
ag


e
w


o
m


en
’s


ea
rn


in
gs


100


earnings percentile


original gap age and education
full set of demographic characteristics




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






245


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.




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




249


Table 13.1 Demographic and Job Characteristics and Relative Earnings of Nonminority and Minority Workers
in Latin America, Circa 2005


Composition Earnings index


(percentage)
(Base: average minority


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)




250


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


(percentage)
(Base: average minority


earnings = 100)


Nonminorities Minorities Nonminorities Minorities




251


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


Table 13.1 (continued)


Composition Earnings index


(percentage)
(Base: average minority


earnings = 100)


Nonminorities Minorities Nonminorities Minorities


(continued next page)




252


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 13.1 (continued)


Composition Earnings index


(percentage)
(Base: average minority


earnings = 100)


Nonminorities Minorities Nonminorities Minorities




Table 13.2 Decomposition of Ethnic Earnings Gap in Latin America after Controlling for Demographic
Characteristics, Circa 2005
(percent)


Gender + Age + Education


+ Presence of
children in the


household


+ Presence of other
household member with


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
components equals the total earnings gap (ΔW + ΔNW + ΔX+Δ0 = Δ).


253




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
set


& Type of
employment & Part time & Formality & Sector & Occupation


& Small
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




Table 13.4 Decomposition of Ethnic Earnings Gap by Demographic and Job Characteristics in Selected
Countries in Latin America, Circa 2005
(percent)


Country Δ


Δ0


Gender and age
+


Education


+ Presence of children in
household, presence of


other income earner in
household, and urban


+ Part time, formality,
occupation, economic sector,


type of employment,
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.


256




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


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).


–40 –20 0 20 40 60 80 100


Ecuador(Δ = 30.7%)


Chile (Δ = 30.8%)


Paraguay (Δ = 59.6%)


Guatemala (Δ = 67.8%)


Brazil (Δ = 38.6%)


Peru (Δ = 45.5%)


Bolivia (Δ = 30.6%)


Δ0 ΔW ΔNW ΔX


percentage of average minorities’ earnings




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


children in the household


education


0 10
no yes


11
12
13
14
15


5
10
15
20
25




inc
om


ple
te


pri
ma


ry
no


ne




co
m


ple
te


pri
ma


ry




inc
om


ple
te


se
co


nd
ary




co
m


ple
te


se
co


nd
ary




inc
om


ple
te


ter
tia


ry




co
m


ple
te


ter
tia


ry


c. Controlling for education d. Controlling for children at home


pe
rc


en
ta


ge
o


f a
ve


ra
ge


m
in


or
iti


es
’ e


ar
ni


ng
s


pe
rc


en
ta


ge
o


f a
ve


ra
ge


m
in


or
iti


es
’ e


ar
ni


ng
s


women
gender


men
5


10


8


12


14


16


10


15


20
a. Controlling for gender b. Controlling for age


pe
rc


en
ta


ge
o


f a
ve


ra
ge


m
in


or
iti


es
’ e


ar
ni


ng
s


pe
rc


en
ta


ge
o


f a
ve


ra
ge


m
in


or
iti


es
’ e


ar
ni


ng
s


range of age
<


24
24


–3
4


34
–4


4
44


–5
4


>
54


18


16


14


12


pe
rc


en
ta


ge
o


f a
ve


ra
ge


m
in


or
iti


es
’ e


ar
ni


ng
s


pe
rc


en
ta


ge
o


f a
ve


ra
ge


m
in


or
iti


es
’ e


ar
ni


ng
s


10


25


20


15


10
no yesno


other with income
in the household


urban
yes


e. Controlling for other income
earners in the household f. Controlling for urban


location


pe
rc


en
ta


ge
o


f a
ve


ra
ge


m
in


or
iti


es
’ e


ar
ni


ng
s


pe
rc


en
ta


ge
o


f a
ve


ra
ge


m
in


or
iti


es
’ e


ar
ni


ng
s 15


10


5


part time
yesno


30


20


10


0


type of employment


em
plo


ye
r


se
lf-e


mp
loy


ed


em
plo


ye
r


g. Controlling for type
of employment


h. Controlling for part-time
employment


(continued next page)




260 new century, old disparities


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 13.2 (continued)


15 30


20


10


0


−10


14


13


12


pe
rc


en
ta


ge
o


f a
ve


ra
ge


m
in


or
iti


es
’ e


ar
ni


ng
s


pe
rc


en
ta


ge
o


f a
ve


ra
ge


m
in


or
iti


es
’ e


ar
ni


ng
s


11


10
no


formality


occupation


pro
fes


sio
na


ls


dir
ec


tor
s


ad
mi


nis
tra


tive
se


ller
s


se
rv


ice
wo


rke
rs


ag
ricu


ltur
al w


ork
ers




no
na


gri
cul


tur
al w


ork
ers


ar
m


ed
fo


rce
s


no
t c


las
sifi


ed
yes


i. Controlling for formality j. Controlling for occupation


20


15


10


5


0


no yes
small firm


sector


pe
rc


en
ta


ge
o


f a
ve


ra
ge


m
in


or
iti


es
’ e


ar
ni


ng
s


pe
rc


en
ta


ge
o


f a
ve


ra
ge


m
in


or
iti


es
’ e


ar
ni


ng
s


ag
ricu


ltur
e


m
inin


g


m
an


ufa
ctu


rin
g


ele
ctr


icit
y s


up
ply


co
ns


tru
ctio


n


tra
de


, h
ote


ls,
an


d r
es


tau
ran


ts


tra
ns


po
rt s


tor
ag


e


fin
an


cia
l se


rvi
ce


s


pe
rso


na
l se


rvi
ce


s


100


50


0


k. Controlling for sector l. Controlling for small firm size


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


Source: Based on data from national household surveys from circa 2005.


Figure 13.3 Unexplained Ethnic Earnings Gap in Latin
America after Controlling for Demographic and Job
Characteristics, by Percentile of Earnings Distribution,
Circa 2005


0


20


40


60


pe
rc


en
ta


ge
o


f a
ve


ra
ge


m
in


or
iti


es
’ e


ar
ni


ng
s


80


100


120


earnings percentile


gender and age
all demographic characteristics all demographic and job characteristics


gender, age, and education


0 10 20 30 40 50 60 70 80 90 100


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.






265


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.




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


Unmatched
nonwhites


Unmatched
whites


Matched
whites and


nonwhites
Unmatched


nonwhites
Unmatched


whites


Matched
whites and


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)




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


Unmatched
nonwhites


Unmatched
whites


Matched
whites and


nonwhites
Unmatched


nonwhites
Unmatched


whites


Matched
whites and


nonwhites


270




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.


Table 14.1 (continued)


1996 2006


Unmatched
nonwhites


Unmatched
whites


Matched
whites and


nonwhites
Unmatched


nonwhites
Unmatched


whites


Matched
whites and


nonwhites


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


b. Controlling for education and age


–20


0


20


40


60


80


100


120


pe
rc


en
ta


ge
o


f a
ve


ra
ge


e
ar


ni
ng


s
o


f n
on


w
hi


te
s


c. Controlling for education, age, and region


–20


0


20


40


60


80


100


120


pe
rc


en
ta


ge
o


f a
ve


ra
ge


e
ar


ni
ng


s
o


f n
on


w
hi


te
s


a. Controlling for education


0


20


40


60


80


100


120


pe
rc


en
ta


ge
o


f a
ve


ra
ge


e
ar


ni
ng


s
o


f n
on


w
hi


te
s


1996 1997 1998 1999 2001 2002 2003 2004 2005 2006


1996 1997 1998 1999 2001 2002 2003 2004 2005 2006


1996 1997 1998 1999 2001 2002 2003 2004 2005 2006


Figure 14.1 Decomposition of Ethnic Earnings Gap in Brazil,
1996–2006


(continued next page)




274 new century, old disparities


d. Controlling for education, age, region, and occupation


–20


0


20


40


60


80


100


pe
rc


en
ta


ge
o


f a
ve


ra
ge


e
ar


ni
ng


s
o


f n
on


w
hi


te
s


e. Controlling for education, age, region,
occupation, and sector


–20


0


20


40


60


80


100


120


pe
rc


en
ta


ge
o


f a
ve


ra
ge


e
ar


ni
ng


s
o


f n
on


w
hi


te
s


f. Controlling for education, age, region, occupation,
sector, and employment in formal sector


–20


0


20


40


60


80


100


120


pe
rc


en
ta


ge
o


f a
ve


ra
ge


e
ar


ni
ng


s
o


f n
on


w
hi


te
s


ΔX ΔNW ΔW Δ0


1996 1997 1998 1999 2001 2002 2003 2004 2005 2006


1996 1997 1998 1999 2001 2002 2003 2004 2005 2006


1996 1997 1998 1999 2001 2002 2003 2004 2005 2006


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).


Figure 14.1 (continued)




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


Source: Based on data from the 1996–2006 PNAD.


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


percentile of earnings distribution


pe
rc


en
t o


f a
ve


ra
ge


ea
rn


in
gs


o
f n


on
w


hi
te


s


0


20


40


60


80


100


120


b. Unexplained ethnic earnings gap after controlling for
demographic and job characteristics


percentile of earnings distribution


pe
rc


en
t o


f a
ve


ra
ge


ea
rn


in
gs


o
f n


on
w


hi
te


s


0


20


40


60


80


100


120


0 10 20 30 40 50 60 70 80 90 100


0 10 20 30 40 50 60 70 80 90 100




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.






281


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.




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


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.


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.




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)


Year/education
level


Men Women


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


Bachilleratoa 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.




287


Table 15.3 Occupational Distribution in Ecuador, by Gender and Minority Status, 2003 and 2007
(percent)


Year/employment status


Men Women


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


Source: Based on data from 2003–07 ENEMDU.




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).


Figure 15.1 Decomposition of Ethnic Earnings Gap in Ecuador, 2003–07


–2


8


18


28


38


48


2003 2004 2005 2006 2007


p


e


r


c


e


n


t


a


g


e




o


f




a


v


e


r


a


g


e




e


a


r


n


i


n


g


s


o


f




m


i


n


o


r


i


t


i


e


s


b. Controlling for area, education, gender,
age, and head of household


a. Controlling for area, education,
gender, and age


–2


8


18


28


38


48


2003 2004 2005 2006 2007


p


e


r


c


e


n


t


a


g


e




o


f




a


v


e


r


a


g


e




e


a


r


n


i


n


g


s


o


f




m


i


n


o


r


i


t


i


e


s


c. Controlling for area, education, gender,
age, head of household, and occupation


–2


8


18


28


38


48


2003 2004 2005 2006 2007


p


e


r


c


e


n


t


a


g


e




o


f




a


v


e


r


a


g


e




e


a


r


n


i


n


g


s


o


f




m


i


n


o


r


i


t


i


e


s


d. Controlling for area, education, gender, age,
head of household, occupation, and remittances


–2


8


18


28


38


48


2003 2004 2005 2006 2007


p


e


r


c


e


n


t


a


g


e




o


f




a


v


e


r


a


g


e




e


a


r


n


i


n


g


s


o


f




m


i


n


o


r


i


t


i


e


s


ΔWΔXΔNWΔ0


288




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


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.


Figure 15.2 Unexplained Ethnic Earnings Gap in Ecuador
after Controlling for Demographic and Job Characteristics,
by Percentile of Earnings Distribution, 2003–07


Source: Based on data from 2003–07 ENEMDU.


0


10


20


pe
rc


en
ta


ge
o


f a
ve


ra
ge


e
ar


ni
ng


s
of


m
in


or
iti


es


30


40


50


60


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,
household, and occupation


area, education, gender, age, household,
occupation, and remittances




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.






293


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.




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.


Source: Based on data from Guatemala’s 2000 and 2006 ENCOVI and
2004 ENEI.


Figure 16.1 Real Monthly Earnings of Indigenous and
Nonindigenous Workers in Guatemala, 2000–06


2,151
1,949


1,859


1,061 1,073 1,059


400


0


800


1,200


1,600


2,000


2,400


2000 2004 2006


qu
et


za
le


s
(20


06
)


nonindigenous indigenous




Table 16.1 Highest Educational Level Begun or Completed by Indigenous and Nonindigenous Workers in
Guatemala, 2000, 2004, 2006
(percent)


Education


2000 2004 2006


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


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).


Figure 16.2 Decomposition of Ethnic Earnings Gap in
Guatemala after Controlling for Demographic and Job
Characteristics, 2000–06


–20


0


20


40


60


80


100


2000 2004 2006


pe
rc


en
ta


ge
o


f a
ve


ra
ge


e
ar


ni
ng


s
o


f i
nd


ig
en


ou
s


ΔWΔNW Δ0ΔX


(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


Table 16.2 Unexplained Ethnic Earnings Gap in Guatemala
Controlling for Demographic Characteristics, 2000–06
(percent)


Characteristics


Age,
education,
and marital


status
+


gender


+
migrant


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)


Source: Based on data from Guatemala’s 2000–06 ENCOVI.


Figure 16.3 Ethnic Earnings Gap in Guatemala after
Controlling for Demographic and Job Characteristics, by
Percentile of Earnings Distribution, 2000–06


0


10


20


30


40


50


60


70


pe
rc


en
ta


ge
o


f a
ve


ra
ge


e
ar


ni
ng


s
of


in
di


ge
no


us


0 10 20 30 40 50 60 70 80 90 100


percentile of earnings distribution




earnings gaps and minorities: guatemala 2000–06 299


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
capital city


9.8 14.8 19.5 20.3


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.


Table 16.2 (continued)


Characteristics


Age,
education,
and marital


status
+


gender


+
migrant


condition
+


residence




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






303


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.




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.




309


A


Abadía, L. K., 138
age of workers


ethnic earnings gap, 249t, 258.
See also specific countries


gender earnings gap. See gender
earnings gap; specific
countries


agriculture sector. See also
economic sectors


ethnic earnings gap, 283
gender earnings gap, 48,


80n1, 209
Altonji, J., 306
Alves, M. C., 164
Amador, D., 143
Angel-Urdinola, D., 138
Arabsheibani, G. R., 164
Arcand, J. L., 267
Argentina


educational attainment
birth cohort in which


gender parity achieved,
24, 25t


change in educational gender
gap 1940–84, 29, 29f


fertility rates, 5
gender earnings gap, 49t,


50, 51f


unexplained components of,
50, 80


survey data, 15t
Arias, O., 267
armed forces, 55
Artana, D., 196
Astudillo, A., 118
Atal, J. P., 41
Auguste, S., 196


B


Barbados
educational attainment, 219t,


221, 227t
gender earnings gap


(1992–2009)
age of workers, 219, 219t,


222, 227t
children in household, 219t,


227t
demographic and job


characteristics, 219–21t
earnings gaps


decompositions, 219–21t,
222–25, 223t, 225f,
226–30t


economic sectors, 220t, 222,
229–30t, 231


Boxes, figures, notes, and tables are indicated by b, f, n, and t following
the page number.


Index




310 index


Barbados (continued)
literature review, 215–16
occupational differences, 222
occupational experience,


221t, 230t, 241
occupational segregation, 224
other household member


with labor income,
219t, 228t


overtime work, 217–18
scope of, 217–22, 218f
type of employment, 219t,


224, 228t, 231
unexplained components of,


225–31, 227–30t
urban vs. rural


areas, 221–22
history and development of,


216–17
labor force participation of


women, 217, 218f
survey data, 15t


Barros, R., 164, 266
Behrman, R., 22–23
Bernal, R., 143
Bernat, L. F., 139
Birdsall, N., 164
Blinder-Oaxaca decomposition,


10–13, 115, 116, 131,
165, 266, 267


Bolivia
educational attainment, 4


birth cohort in which each
country achieved gender
parity, 24, 25t


change in educational
gender gap 1940–84,
28, 29f


conditional expectations,
31, 31f


ethnicity as factor, 32–34
household income level and,


32, 33–34f
inclusive educational


methods, 304
probability component,


31, 31f


school attendance rates,
32–33, 33f


ethnic earnings gap by
demographic and job
characteristics, 256t


fertility rates, 5
gender earnings gap, 48,


49t, 51f
unexplained components


of, 50
health indicators, 7n3
survey data, 15t


Bravo, D., 116
Brazil


educational attainment
birth cohort in which gender


parity achieved, 24, 25t
ethnic earnings gap,


269t, 275t
gender earnings gap,


166t, 170t
returns to education, 164,


266, 277
ethnic earnings gap


(1996–2006),
265–79, 304


age of workers, 269t, 275t
demographic and job


characteristics, 256t,
268–72, 269–71t, 275t


earnings gaps
decompositions, 272, 273f


economic sectors, 271t, 275t
literature review, 265–68
occupational segregation, 272
regional, 270t, 275t
scope of, 268–72
type of employment,


270t, 275t
unexplained component of,


273–77, 275t, 276f
urban vs. rural areas,


270t, 275t
gender earnings gap


(1996–2006), 48, 49t,
51f, 163–73, 304


age of workers, 166t, 170t




index 311


compared to ethnic earnings
gap, 268


demographic and job
characteristics, 166–67t,
170–71t


earnings gaps
decompositions, 165–69,
166–67t, 168f


formal vs. informal
workers, 170t


literature review, 163–65
occupational segregation,


41, 170t
regional, 166t, 170t
unexplained component of,


79, 169–72, 170–71t,
171f


urban, 166t, 170t
survey data, 15t
unemployment, 285


Brendan, D., 216
Britz do Nascimento Silva, D., 266
Brown, R. S., 164
Bueno, I., 117


C


Calderón, V., 23
Calvalieri, C., 267
Camargo, J. M., 163
Campante, F. R., 268
Caribbean countries, gender


earnings gap in, 215–42.
See also specific countries


history and development of,
216–17


literature review, 215–16
Carneiro, F. G., 164, 165
Central America. See also specific


countries
educational attainment, 183,


185, 186–87t, 191t, 196
gender earnings gap


(1997–2006),
183–213, 186–95t


age of workers, 186t,
191t, 209


children in household,
187t, 192t


confidence intervals, 203–4f,
204, 208–9, 210–11f


demographic and job
characteristics, 186–95t


earnings gaps decompositions,
196–203, 197f, 198t,
200–201t, 202f


economic sectors, 190t,
194–95t


firm size, 189t, 194t
literature review, 184–85
other household member


with labor income,
187t, 192t


part-time workers, 189t, 193t
scope of, 186–96
type of employment,


189t, 193t
unexplained component of,


203–4f, 203–12
urban vs. rural areas, 188t,


193t, 203, 209
labor force participation of


women, 185, 212
marital arrangements, 196, 209
unemployment, 183


Centro de Estudios Distributivos
Laborales y Sociales
(CEDLAS), 14


CEO effect, 12, 50, 58, 64f, 179
childbearing age of women, 5
childcare affordability, 41
children in household


effect of educational attainment
of women on, 22


ethnic earnings gap. See specific
countries


gender earnings gap. See gender
earnings gap; specific
countries


Chile
educational attainment


birth cohort in which each
country achieved gender
parity, 24, 25t




312 index


Chile (continued)
change in educational gender


gap 1940–84, 28, 29f
gender earnings gap


and, 116, 118–20,
119–21f, 128t


ethnic earnings gap by
demographic and job
characteristics, 256t


gender earnings gap
(1992–2009), 49t,
51f, 115–36


age of workers, 128t
average weekly hours


worked, 118, 125,
126–27f, 129t


confidence intervals,
132, 133f


demographic and job
characteristics, 128t


earnings gaps
decompositions, 126–31,
128–29t, 130–31f


job tenure, 129t, 134–35
labor force participation of


women, 118, 121, 121f
literature review, 115–18
marital status, 128t


occupational segregation, 127
unemployment, 117, 120–25,


122–24f
unexplained component of,


132–35, 134f
urban vs. rural areas, 125


survey data, 15t
Chilean Social Protection Survey


2002 (SPS02), 116
Chong, A., 297, 307
Colmenares, G., 102
Colombia


educational attainment
birth cohort in which each


country achieved gender
parity, 24, 25t


gender earnings gap and,
140t, 143, 144–45t,
147, 156t


inclusive educational
methods, 304


gender earnings gap
(1994–2006), 49t, 50, 51f,
137–61


age of workers, 140t, 144t,
156t


children in household, 140t,
156t


demographic and job
characteristics, 144–46t,
156–57t


earnings gaps decompositions,
140–42t, 144–46t,
147–55, 148–50t,
152–54t, 156–57t, 159f


economic sector, 142t,
146t, 157t


family formation and
dissolution and, 143


firm size, 142t, 157t
formal vs. informal workers,


142t, 143, 146t, 157t
literature review, 138–39
marital status, 141t, 143,


145t, 156t
metropolitan areas


studied, 138
other household member


with labor income,
141t, 147


part-time workers, 142t, 143,
146t, 157t


public vs. private workers, 143
scope of, 139–47
statistical discrimination


against women, 138
type of employment, 141–46t


147, 156–57t
unexplained component of,


155–60
maternity laws (Law 50 of


1990), 138
survey data, 15t
unemployment, 137


Colombian National Statistical
Agency, 137




index 313


community, social, and personal
services, 168. See also
economic sectors


ethnic earnings gap, 252t
gender earnings gap, 44


construction sector, 231. See also
economic sectors


ethnic earnings gap, 252t
gender earnings gap, 44


Household Sample Survey (CHSS,
Barbados), 216


Continuous Labor Force
Sample Survey (CLFSS,
Barbados), 217


Continuous Sample Survey of
the Population (CSSP,
Trinidad & Tobago), 216


Contreras, D., 116
Coppin, A., 216
Corley, M., 184
Correia, M., 176
Costa Rica


educational attainment, 24, 25t
birth cohort in which each


country achieved gender
parity, 24, 25t


gender earnings gap, 49t,
50, 51f


occupational segregation, 12,
40, 184


public vs. private
workers, 184


unexplained component
of, 203f, 205, 205f,
208–12


labor force participation of
women, 185


survey data, 16t
Creole, use of, 216–17
Crespo, A. R. V., 268, 277
cross-country heterogeneity


ethnic earnings gap and, 246
gender earnings gap and, 48,


49t, 80n7
Cuevas, M., 196
cultural bias favoring men’s role in


society, 63


stereotyping, programs to
diminish effects
of, 306–7


Cunningham, W., 6
“curse of dimensionality,” 13,


160n3, 224, 242n4


D


data. See methodology and data
Dávila, A., 12, 184
De Carvalho, A. P., 266
demographic changes.


Latin America and
Caribbean, generally,
41, 42–45t, 46, 47t,
67, 68–69t


Deutsch, R., 184
D’Hombres, B., 267
domestic employment, 102, 285.


See also maid effect;
occupational segregation


domestic violence prevention, 41
Dominican Republic


educational attainment, 24, 25t
birth cohort in which each


country achieved gender
parity, 24, 25t


formal vs. informal
workers, 80n5


gender earnings gap, 49t, 51f
survey data, 16t


Duncan index of hierarchical/
occupational segregation,
103, 105–6, 106f


Duryea, S., 22–23


E


earnings gap
ethnic. See ethnic earnings gap;


specific countries
gender. See gender earnings gap;


specific countries
methodological approach, 9–20.


See also methodology and
data




314 index


economic sectors. See also
occupational segregation;
specific countries


ethnic earnings gap. See specific
countries


gender earnings gap, 13, 43–44t,
44, 54f, 55, 58, 63f, 80n1


survey data, 14
Ecuador


educational attainment
birth cohort in which each


country achieved gender
parity, 24, 25t


change in educational gender
gap 1940–84, 28, 29f


conditional expectations,
31, 31f


gender differences, 176, 176t
inclusive educational


methods, 304
probability component,


31, 31f
ethnic earnings gap (2000–07),


281–91
demographic and job


characteristics, 256t
earnings gaps


decompositions, 285–89,
286–87t, 288f


emigrants’ families, 284
language-based definition of


ethnicity, 282
literature review, 281–83
occupational segregation, 290
scope of, 283–85, 284t
“self-whitening”


phenomenon, 284
type of employment, 285
unemployment, 285
unexplained component of,


257, 289–90, 290f
urban vs. rural areas, 283, 284


gender earnings gap (2003–07),
49t, 50, 51f, 175–81


earnings gaps
decompositions, 177–79,
178–80f


literature review, 176
occupational segregation, 40
scope of, 176–77
unexplained component of,


179–81
labor force participation of


women, 175, 176
survey data, 16t


educational attainment, ethnic
minority differences in,
32–34, 303, 304. See also
specific countries


policy options early in life,
304–5


returns to schooling, 116,
262n3, 294, 295t


educational attainment, gender
differences in, 4, 6,
21–36, 41, 50,
303, 304


birth cohort in which each
country achieved gender
parity, 24, 25t


changes by cohorts, 23–24, 23f
decomposing changes in


education, 25–32
complete primary education


or incomplete secondary
education, 26f


complete secondary or
incomplete university
education, 26f, 27


by country, 28–29, 29f
no schooling or incomplete


primary education,
25, 26f


between oldest and
youngest cohort, 27–28,
27–28f


conditional expectations
component, 30–31,
30–31f


probability component,
30–31, 30–31f


school attendance rates,
32–33, 33f


university degree, 26f, 27




index 315


household income level and,
32–33, 33–34f


Latin America and Caribbean,
generally, 42t, 45t, 46,
51–52, 52f, 67, 68t,
70–71t, 73, 78t


literature review, 22–23
policy options early in life,


304–5
school attendance profiles,


32–34, 33f
survey data, 14


El Salvador
educational attainment


birth cohort in which each
country achieved gender
parity, 24, 25t


change in educational gender
gap 1940–84, 28, 29f


conditional expectations,
31, 31f


probability component,
31, 31f


gender earnings gap, 49t, 50,
51f, 79


earnings gaps
decompositions,
196–203, 197f


occupational segregation, 184
public vs. private workers, 184
unexplained component


of, 203f, 205–6, 206f,
208–12


labor force participation of
women, 185


survey data, 16t
Encuesta Nacional de Empleo


(ENE, Mexico), 112n1
Encuesta Nacional de Hogares


(ENAHO), 84
ethnic earnings gap, 245–300


age of workers, 249t, 258
Brazil (1996–2006), 265–79. See


also Brazil
children in household, 250t
confidence intervals, 258,


259–60f


demographic and job
characteristics, 248,
249–53t


earnings gaps decompositions,
248–58, 249–53t,
255–56t, 257f


economic sectors, 252t
Ecuador (2000–07), 281–91. See


also Ecuador
educational attainment. See


ethnic minorities
firm size, 251t
formal vs. informal


workers, 251t
Guatemala (2000–06), 293–300.


See also Guatemala
interplay with gender, 247, 258
literature review, 246–47
occupational segregation, 247,


254
other household member with


labor income, 258
part-time workers, 251t
policy options, 303–4
type of employment, 250t
unexplained component of,


258–62, 261f
urban vs. rural areas, 250t,


258, 283
ethnic minorities


Afro-descendent workers, 247
changes affecting, 5–6
criteria for classifying ethnic


groups as “minorities” by
country, 18t


defined, 19n1, 282
earnings gap. See ethnic earnings


gap
educational attainment and,


32–34, 248, 249t, 258,
262n3


indigenous population, 7n2, 247


F


female-headed households, 4
Fernandes, R., 267




316 index


Fernández, P., 139
fertility rates decline, 5, 22, 69
firm size, 18. See also specific


countries
formal vs. informal workers, 18,


80n5. See also informal
sector; specific countries


Fox, M. L., 164


G


Gallardo, M. L., 6, 282
Ganguli, I., 5
García, J., 116
García-Aracil, A., 176, 281–82
García Durán, B., 117
gender-based segregation


in workplace. See
occupational segregation


gender earnings gap, 39–213
Brazil (1996–2006), 163–73. See


also Brazil
Central America (1997–2006),


183–213. See also Central
America


Chile (1992–2009), 115–36. See
also Chile


Colombia (1994–2006), 137–61.
See also Colombia


economic sectors, 13, 43–44t,
44, 54f, 55, 58, 63f, 80n1


Ecuador (2003–07), 175–81. See
also Ecuador


interplay with ethnicity, 247
Latin America and Caribbean,


generally, 39–82
age of workers, 42t, 45t,


46, 50, 52f, 66t, 67, 68t,
70–71t, 78t


changes (1992–2007), 64–80
childcare affordability, 41
children in household,


42t, 52f, 65, 66t, 68t,
70–71t, 78t


cohort approach to
unexplained changes,
75–80, 77f, 78t


confidence intervals, 52–54f,
72, 72f, 75, 77f, 79f


cross-country heterogeneity,
48, 49t, 80n7


demographic and job
characteristics, 41,
42–44t, 46, 47t, 48,
67–68t, 70–71t


demographic changes, 41,
42–45t, 46, 47t, 65, 67,
68–69t


domestic violence
prevention, 41


earnings gaps
decompositions, 44–45,
45t, 51t, 70–71t


economic sectors, 13, 43–44t,
44, 54f, 55, 58, 63f, 80n1


educational attainment and,
42t, 45t, 46, 51–52,
52f, 66t, 67, 68t, 70–71t,
73, 78t


evolution at turn of 20th
century, 65–75, 66–67t


fertility and. See fertility rates
decline


firm size, 43t, 54–55
formal vs. informal workers,


41, 43t, 53f, 54–56
labor market liberalization


index vs., 65f
macroeconomic, social, and


governance indicators,
linkages with 57–64,
59–62t


literature review, 40–41
“matching-after-matching”


exercise, 73, 74t
maternity laws, 41, 138
men’s role in society, 63
microeconomic causes, 63
occupational segregation.


See occupational
segregation


other household member with
labor income, 42t, 52f,
65, 66t, 68t, 70–71t, 78t




index 317


part-time workers, 43t, 48,
53f, 55, 56–57, 67, 67t,
69t, 70–71t, 78t, 303


policy options, 303–4
public vs. private workers, 40
scope of, 41–57
type of employment, 41,


42t, 53f, 55, 56, 67–69t,
70–71t, 78t


unexplained components of,
49t, 50–55, 56f, 57–64,
69, 72, 73–75, 75f


urban vs. rural areas, 40,
41, 53f, 66t, 67, 68t, 69,
70–71t, 78t


well-paid jobs, access to, 50,
70. See also CEO effect


women’s role in society, 41
methodological approach, 9, 39
Mexico, 101–13. See also


Mexico
Peru (1997–2009), 49t, 50, 51f,


83–100. See also Peru
Gini coefficient, 277n2
glass ceiling effect, 58, 134,


139, 177
governance, linkage with


unexplained gender
earnings gaps, 58, 62t


Guatemala
educational attainment, 4, 196


birth cohort in which each
country achieved gender
parity, 24, 25t


change in educational gender
gap 1940–84, 28, 29f


conditional expectations,
31, 31f


ethnicity as factor, 32–34
household income level and,


32, 33–34f
probability component,


31, 31f
school attendance rates,


32–33, 33f
ethnic earnings gap (2000–06),


293–300


age of workers, 298–99t
demographic and job


characteristics, 256t
earnings gaps


decompositions, 296, 297f
educational attainment and,


294, 295t, 298–99t
marital status, 298–99t
scope of, 254, 294–96, 294f
unexplained component of,


296–300, 298–99t, 298f
urban vs. rural areas, 294, 296


fertility rates, 5
gender earnings gap


(1997–2006), 48, 49t,
50, 51f, 79


earnings gaps
decompositions, 196–203,
197f, 202–3, 202f


public vs. private
workers, 196


unexplained component of,
206–7, 207f, 209


labor market participation of
women vs. men, 196


survey data, 16t
Guimarães, R., 268


H


Hall, G., 6, 246
Hausmann, R., 5
Heckman, J., 266
Henley, A., 164
Hernández, P. J., 116
Hernández-Zavala, M., 6
Hertz, T., 12
hierarchical segregation, 5, 103–4t,


103–5
Hoffmann, R., 164
Honduras


educational attainment
birth cohort in which each


country achieved gender
parity, 24, 25t


change in educational gender
gap 1940–84, 28, 29f




318 index


Honduras (continued)
inclusive educational


methods, 304
fertility rates, 5
gender earnings gap, 49t, 51f


earnings gaps
decompositions,
196–203, 197f


unexplained component of,
204f, 207, 208–12, 208f


labor force participation of
women, 185


survey data, 16t
Hotchkiss, J., 216
Household and Childhood


Measurement Indicators
Survey (Encuesta de
Medición de Indicadores
sobre la Niñez y los
Hogares [EMEDINHO],
Ecuador), 282


household responsibilities, 305–6


I


indigenous populations. See ethnic
minorities


industries and sectors. See
economic sectors


informal sector and gender earnings
gap, 41, 43t, 53f, 54–55.
See also specific countries


Inter-American Development
Bank, 6


Inter-American Development Bank’s
Research Department, 14


J


Jacobsen, J.P., 6
Jamaica


educational attainment, 232, 233t
gender earnings gap (2003),


231–40
age of workers, 232, 233t
children in household, 232,


233t, 239


demographic and job
characteristics, 233–35t


earnings gaps
decompositions, 232–35t,
235–38, 236–37t


economic sectors, 234t
firm size, 235t
job tenure, 234–35t
occupational segregation,


232, 238
other household member


with labor income, 233t
overtime work, 235t, 240
part-time workers, 240
public vs. private


workers, 232
type of employment, 232, 233t
unexplained components of,


238–40, 239f
urban vs. rural areas, 233t


history and development of,
216–17


survey data, 16t
job tenure


discrimination lessened by, 306
Jamaica, 232, 240, 241


Jones, F. L., 267
Jones, G., 22


K


Kassouf, A. L., 172n1
Kelley, J., 267
Knodel, J., 22


L


labor force participation
of ethnic groups, 6, 246
reducing segregation in


and improving
productivity, 305


rewarding men and whites
characteristics in labor
markets, 9


sectoral differences between men
and women, 13




index 319


of women, 4, 6n1, 39, 112. See
also specific countries


linkage with unexplained
gender earnings gaps, 57,
60–61t


reducing household
responsibilities for, 305–6


Labor Force Survey (Jamaica), 231
Labor market liberalization


index, 65f
La Ferrara, E., 307
Larrea, C., 282
Latin America. See also specific


countries
educational attainment. See


educational attainment,
ethnic minority differences
in; educational attainment,
gender differences in


ethnic groups. See ethnic
earnings gap; ethnic
minorities


gender earnings gap. See gender
earnings gap


inequality, 3, 6
labor force participation. See


labor force participation
women’s progress, 4–5


legislators, women as, 58, 64f
Leite, P. G., 268
Leme, M. C., 164
Living Standards Measurement


Survey (Colombia), 139
López, A., 116
López-Calva, L. F., 265
Loureiro, P. R. A., 165
Lovell, P., 266, 267
Lustig, N., 265


M


Machado, J. A. F., 118
MacIsaac, D., 283
maid effect, 12, 50, 179
Maldonnado, V. M., 102
managerial positions. See CEO


effect


marital status. See also specific
countries


change in marital and
cohabitation
arrangements, effect on
gender earnings gap, 69


of skilled women, 5
Marshall, J., 23
Mata, J., 118
maternity laws, 306
men’s role in society, 63. See also


stereotyping
mentoring programs, 306
methodology and data, 9–20


Blinder-Oaxaca decomposition
extension of, 10–13
shortcomings of, 10


“curse of dimensionality,” 13
earnings gaps decompositions


using matching, 11, 12, 13
“matching-after-matching”


exercise, 73, 74t
nationally representative


household survey data,
14–19


by country, 15–17t
criteria for classifying ethnic


groups as “minorities” by
country, 18t


ethnic minorities, countries
covered, 14


working population as
focus, 14


regression-based vs. matching-
based approach, 13, 50


sectoral differences between men
and women, 13


unexplained gender earnings
gaps, 10, 13, 58, 59–62t


Mexico
educational attainment,


23, 103t
birth cohort in which each


country achieved gender
parity, 24, 25t


change in educational gender
gap 1940–84, 28, 29f




320 index


Mexico (continued)
household income level and,


32, 33–34f
inclusive educational


methods, 304
school attendance rates,


32–33, 33f
gender earnings gap (1994–2004),


49t, 51f, 101–13
age of workers, 103t, 105–7,


106f, 108–9f
demographic and job


characteristics, 103t
earnings gaps


decompositions, 105–12
economic sector, 104t
firm size, 104t
literature review, 102
marital status, 104t
occupational and hierarchical


segregation, 41, 58, 101,
102, 103–4t, 103–12,
106f, 108–9f, 111f


public vs. private
workers, 104t


unexplained components of, 80
labor force participation of


women, 101
survey data, 16t


microeconomic linkages with
gender earnings gap, 63


mining sector. See also economic
sectors


ethnic earnings gap, 55
gender earnings gap, 252t


Montenegro, C., 116
Montenegro Torres, F., 282
Moon, M., 164
Moore, R., 216


N


National Household Sample Survey
(PNAD, Brazil), 163


National Institute of Statistics
and Informatics (INEI,
Peru), 84


National Socioeconomic
Characterization
Survey (Encuesta
de Caracterización
Socioeconómico Nacional
[CASEN], Chile), 115,
116, 117, 123


National Survey of Employment
and Income (Encuesta
Nacional de Empleo
e Ingresos [ENEI],
Guatemala), 212n3, 293


National Survey of Living
Conditions (Encuesta
Nacional de Condiciones
de Vida [ENCOVI],
Guatemala), 212n3, 293


National Survey of Urban
Employment (Encuesta
Nacional de Empleo
Urbano [ENEU], Mexico),
101, 102, 103, 105,
112n1


Néri, M., 266
Nicaragua


educational attainment
birth cohort in which each


country achieved gender
parity, 24, 25t


change in educational
gender gap 1940–84,
29, 29f


fertility rates, 5
gender earnings gap, 49t, 50,


51f, 79
earnings gaps


decompositions, 196–203,
197f


occupational differences,
184


unexplained component of,
204f, 207–8, 208–12,
209f


labor force participation of
women, 185


survey data, 17t
Ñopo, H., 10, 11




index 321


O


occupational categories, 19, 43t,
85–86t


occupational segregation. See also
specific countries


ethnic earnings gap, 101–12,
254, 261, 262


gender earnings gap, 5, 12,
40–41, 43t, 48, 54f,
69, 241


policy options to reduce, 305
Ometto, A. M. H., 164


P


Pagán, J., 12, 185
Panama


educational attainment
birth cohort in which each


country achieved gender
parity, 24, 25t


conditional expectations,
31, 31f


probability component,
31, 31f


gender earnings gap, 49t, 50, 51f
survey data, 17t


Panizza, U., 184
Paraguay


educational attainment, 24, 25t
ethnic earnings gap by


demographic and job
characteristics, 256t


fertility rates, 5
gender earnings gap, 49t,


50, 51f
survey data, 17t


Paredes, R., 115, 116
Parker, S., 23, 102
part-time workers. See also specific


countries
ethnic earnings gap, 258,


261, 262
gender earnings gap, 48, 53f,


55, 56–57, 67, 69t,
70–71t, 303


Patrinos, H. A., 6, 246


Pederzini, C., 23
Peña, X., 139
Perardel, Y., 184
Perticará, M., 117, 118
Peru


educational attainment
birth cohort in which each


country achieved gender
parity, 24, 25t


change in educational gender
gap 1940–84, 28, 29f


ethnicity as factor, 32–34
household income level and,


32, 33–34f
school attendance rates,


32–33, 33f
ethnic earnings gap by


demographic and job
characteristics, 256t


gender earnings gap
(1997–2009), 49t, 50,
51f, 83–100


age of workers, 84, 85t,
89, 93t


confidence intervals, 97, 98f
demographic and job


characteristics, 84,
85t, 93t


earnings gaps
decompositions, 89–92,
90–91f, 93t, 96f


economic sectors, 86t,
92, 95t


educational attainment
and, 84, 85t, 87–88f,
89–90, 93t


firm size, 85t, 94t
formal vs. informal


workers, 83
labor force participation of


women, 99–100, 99f
labor market reform, 83
marital status, 94t
occupational segregation, 83,


85–86t, 94t
part-time workers, 85t
scope of, 84–89




322 index


Peru (continued)
unemployment, 97–100, 100f
unexplained components of,


50, 89, 92–97
urban vs. rural areas, 85t, 94t


gender stereotyping, 307
survey data, 17t
unemployment, 97–100, 100f


Pesquisa Nacional por Amostra
de Domicilios (National
Survey of Sample
Households, PNAD,
Brazil), 265, 267


Pierret, C., 306
Pisani, M. J., 185
Popova, K., 184
productivity improvements, 305
Psacharopoulos, G., 6, 40, 184
public vs. private workers, 40. See


also specific countries
Puentes, E., 116


Q


Qiang, C. Z.-W., 184


R


Rama, M., 283
Ramos, L., 164
Rangel, M., 246
Reis, M. C., 268
Rendón, T., 102
Riveros, L., 115


S


Saavedra, J., 6
Sabogal, A., 138
Sachsida, A., 165
Sanhueza, C., 116
Santos, E., 164
school attendance. See educational


attainment, ethnic
minority differences in;
educational attainment,
gender differences in


Scott, K., 216
Sectors, economic. See economic


sectorsself-employment,
41, 55, 303. See also
specific countries for
“type of employment”


“self-whitening” phenomenon,
284


senior officials. See Type of
employment


Serrano, F., 163
Silva, N. D. V., 172n1, 267
Soares, S. D. S., 266, 268
sociodemographic characteristics,


linkage with unexplained
gender earnings
gaps, 57


Solano, E., 6
Sookram, S., 216
statistical discrimination against


women, 138, 160n1
Stelcner, M., 164
stereotyping, 63


programs to diminish effects of,
306–7


survey data. See methodology and
data


Survey on Employment,
Unemployment, and
Underemployment
(Encuesta de Empleo,
Desempleo, y
Subempleo [ENEMDU],
Ecuador), 175


Survey on Living Conditions
(Encuesta de Condiciones
de Vida [ECV], Ecuador),
282


Székely, M., 22–23


T


Tejerina, L., 267
Terrell, K., 216
Tiefenthaler, J., 164
traditional gender roles, 5
Tzannatos, Z., 40, 184




index 323


U


unemployment, 80n1. See also
specific countries


unexplained components
ethnic earnings gap. See ethnic


earnings gap
gender earnings gap. See gender


earnings gap
United Nations Human


Development Index, 217
University of the West Indies


(UWI), 217
urban vs. rural areas. See also


specific countries
ethnic earnings gap, 258, 283
gender earnings gap, 40, 41,


42t, 53f, 67, 68t, 69,
70–71t, 78t


Uruguay
educational attainment, 24, 25t


birth cohort in which each
country achieved gender
parity, 24, 25t


fertility rates, 5
gender earnings gap, 49t, 51f


occupational segregation, 40
survey data, 17t


Urzúa, S., 116


V


Van Bronkhorst, B., 176


Venezuela, República Bolivariana de
educational attainment


birth cohort in which each
country achieved gender
parity, 24, 25t


change in educational
gender gap 1940–84,
29, 29f


gender earnings gap, 49t, 51f, 80
survey data, 17t


Viarengo, M., 5


W


Wajnman, S., 164
Watson, P., 216
well-paid jobs. See also CEO


effect
ethnic minorities’ access to, 254
women’s access to, 50, 70


Winder, N., 41
Winter, C., 176, 281–82
Wodon, Q., 138
Wood, C. H., 267
World Bank, 3


Y


Yamada, G., 267


Z


Zoloth, B. S., 164




ECO-AUDIT
Environmental Benefits Statement


The World Bank is committed to preserving
endangered forests and natural resources.
The Office of the Publisher has chosen
to print New Century, Old Disparities:
Gender and Ethnic Earnings Gaps in Latin
America and the Caribbean on recycled
paper with 50 percent postconsumer fiber in
accordance with the recommended standards
for paper usage set by the Green Press Ini-
tiative, a nonprofit program supporting pub-
lishers in using fiber that is not sourced from
endangered forests. For more information,
visit www.greenpressinitiative.org.


Saved:
• 10 trees
• 4 million British thermal


units of total energy
• 1,025 pounds of net


greenhouse gases
• 4,621 gallons of


waste water
• 293 pounds of solid


waste






LATIN AMERICAN DEVELOPMENT FORUM SERIES


“The inequities suff ered 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 diff erence 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 effi cient 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
eff ect changes in educational and life paths that are not merely incremental but that are in fact
transformational for children, families, and ultimately communities. The fi rst 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 diff erences. The analysis includes a dynamic dimension that sheds light on
the evolution of earnings gaps over time. The book off ers 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)


SKU 18686


ISBN 978-0-8213-8686-6




Login