73823 i Economic Mobility and the Rise of the Latin American Middle Class ECONOMIC MOBILITY AND THE RISE OF THE LATIN AMERICAN MIDDLE CLASS iii Economic Mobility and the Rise of the Latin American Middle Class Francisco H. G. Ferreira, Julian Messina, Jamele Rigolini, Luis-Felipe López-Calva, Maria Ana Lugo, and Renos Vakis © 2013 International Bank for Reconstruction and Development / The World Bank 1818 H Street NW, Washington DC 20433 Telephone: 202-473-1000; Internet: www.worldbank.org Some rights reserved 1 2 3 4 16 15 14 13 This work is a product of the staff of The World Bank with external contributions. Note that The World Bank does not necessarily own each component of the content included in the work. The World Bank therefore does not warrant that the use of the content contained in the work will not infringe on the rights of third parties. The risk of claims resulting from such infringement rests solely with you. The findings, interpretations, and conclusions expressed in this work do not necessarily reflect the views of The World Bank, its Board of Executive Directors, or the governments they represent. The World Bank does not guarantee the accuracy of the data included in this work. The boundaries, colors, denominations, and other information shown on any map in this work do not imply any judgment on the part of The World Bank concerning the legal status of any territory or the endorsement or acceptance of such boundaries. Nothing herein shall constitute or be considered to be a limitation upon or waiver of the privileges and immunities of The World Bank, all of which are specifically reserved. Rights and Permissions This work is available under the Creative Commons Attribution 3.0 Unported license (CC BY 3.0) http:// creativecommons.org/licenses/by/3.0. Under the Creative Commons Attribution license, you are free to copy, distribute, transmit, and adapt this work, including for commercial purposes, under the following conditions: Attribution—Please cite the work as follows: Ferreira, Francisco H. G., Julian Messina, Jamele Rigolini, Luis-Felipe López-Calva, Maria Ana Lugo, and Renos Vakis. 2013. Economic Mobility and the Rise of the Latin American Middle Class. Washington, DC: World Bank. doi: 10.1596/978-0-8213-9634-6. License: Creative Commons Attribution CC BY 3.0 Translations—If you create a translation of this work, please add the following disclaimer along with the attribution: This translation was not created by The World Bank and should not be considered an official World Bank translation. The World Bank shall not be liable for any content or error in this translation. All queries on rights and licenses should be addressed to the Office of the Publisher, The World Bank, 1818 H Street NW, Washington, DC 20433, USA; fax: 202-522-2625; e-mail: pubrights@worldbank.org. ISBN (paper): 978-0-8213-9634-6 ISBN (electronic): 978-0-8213-9723-7 DOI: 10.1596/978-0-8213-9634-6 Cover design: Naylor Design Library of Congress Cataloging-in-Publication Data Ferreira, Francisco H. G. Economic mobility and the rise of the Latin American middle class / Francisco H.G. Ferreira [and five others]. pages cm. — (World Bank Latin American and Caribbean studies) Includes bibliographical references. ISBN 978-0-8213-9634-6 — ISBN 978-0-8213-9723-7 (electronic) 1. Income—Latin America. 2. Middle class—Latin America. 3. Households—Economic aspects— Latin America. 4. Occupational mobility—Latin America. 5. Social mobility—Latin America. 6. Latin America—Economic conditions. I. World Bank. II. Title. HC130.I5F47 2012 305.5’5098—dc23 2012041229 Contents Foreword . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xi Acknowledgments. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiii Abbreviations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xv Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 A middle-income region on the way to becoming a middle-class region . . . . . . . . . . . . . . . . 1 Within generations, remarkable upward mobility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 Across generations, mobility remains low . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 A snapshot of the Latin American middle class . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 The middle class and the social contract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 Latin American “climbers” and “stayers”. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 The broad context . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18 Pursuing the questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22 2 Economic Mobility and the Middle Class: Concepts and Measurement. . . . . . . . . . . . . . . 23 Spaces, domains, and concepts of economic mobility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24 Defi ning the middle class . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29 Linking mobility and middle-class dynamics: A matrix decomposition . . . . . . . . . . . . . . . 37 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46 v vi CONTENTS 3 Mobility across Generations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49 Educational attainment: How important is parental background? . . . . . . . . . . . . . . . . . . . 53 The importance of educational achievement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60 From educational to income mobility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65 Policies and intergenerational educational mobility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67 Conclusions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88 4 Mobility within Generations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93 Using synthetic panels to study long-term mobility. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94 Income mobility in Latin America: The past two decades . . . . . . . . . . . . . . . . . . . . . . . . . 98 Unravelling the box: Exiting poverty and entering the middle class . . . . . . . . . . . . . . . . 101 Mobility profi les: Insights for policy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108 Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117 Annex 4.1 Data used for intragenerational mobility estimates. . . . . . . . . . . . . . . . . . . . . 124 Annex 4.2 Regional and country intragenerational mobility estimates and decomposition using synthetic panels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132 5 The Rising Latin American and Caribbean Middle Class . . . . . . . . . . . . . . . . . . . . . . . . 135 The middle class in Latin America and the Caribbean . . . . . . . . . . . . . . . . . . . . . . . . . . . 136 Recent middle-class growth trends . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137 Forecasts for poverty reduction and middle-class growth . . . . . . . . . . . . . . . . . . . . . . . . . 142 Who is middle class in Latin America and the Caribbean? . . . . . . . . . . . . . . . . . . . . . . . . 145 Broad class profiles from three exemplar countries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146 Middle-class characteristics, selected countries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158 6 The Middle Class and the Social Contract in Latin America . . . . . . . . . . . . . . . . . . . . . . 159 The middle class and the shaping of economic policy . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160 Values and beliefs of the Latin American middle classes. . . . . . . . . . . . . . . . . . . . . . . . . . 166 Overcoming a fragmented social contract. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179 Boxes 3.1 Assessing the association of socioeconomic status across generations . . . . . . . . . . . . . 52 3.2 Income mobility in high-income countries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66 3.3 Cross-country analysis of policies and institutions and intergenerational mobility . . . 68 3.4 Tuition loans in Chile: Is the alleviation of credit constraints a good policy to close the gap in educational attainment between rich and poor? . . . . . . . . . . . . . . . . 71 3.5 Conditional cash transfers and children’s educational outcomes . . . . . . . . . . . . . . . . . 77 3.6 Voucher systems in Chile and Colombia: Did they help the achievements of the poor? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80 4.1 Existing fi ndings on intragenerational mobility in Latin America . . . . . . . . . . . . . . . . 95 4.2 The welfare costs of downward mobility in Nicaragua . . . . . . . . . . . . . . . . . . . . . . . 108 4.3 “Calling in” long-term mobility: Did cell phones improve mobility in rural Peru? . . 119 CONTENTS vii F4.1 Validating the approach for the case of Latin America . . . . . . . . . . . . . . . . . . . . . . . 122 5.1 The (sustainable?) rise of the Brazilian middle class . . . . . . . . . . . . . . . . . . . . . . . . . . 140 6.1 Inequality, growth, and institutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162 6.2 A new data set on the world’s middle classes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163 6.3 Studying middle-class values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167 6.4 Individualization of public goods and lack of institutional trust in the Dominican Republic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176 Figures O.1 The distribution of income in Latin America and the Caribbean, 2009. . . . . . . . . . . . . 3 O.2 Trends in middle class, vulnerability, and poverty in Latin America and the Caribbean, 1995–2009 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 O.3 The growth and redistribution components of middle-class growth in Latin America and the Caribbean, 1995–2010 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 O.4 Association between parental education and children’s years of schooling, selected countries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 O.5 Relationship between average PISA test scores and intergenerational mobility across 65 countries, 2009. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 O.6 Impact of parental background on children’s educational gap at age 15 in Latin America, 1995–2009 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 O.7 Association between income inequality and intergenerational immobility . . . . . . . . . . 9 O.8 Average years of schooling (ages 25–65), selected Latin American countries, by income class, circa 2009 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 1.1 Average annual per capita GDP growth in Latin America and the Caribbean, 2000–10 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19 1.2 Change in the Gini index, selected Latin American countries, 2000–10 . . . . . . . . . . . 20 1.3 Moderate and extreme poverty in Latin America, 1995–2010. . . . . . . . . . . . . . . . . . . 20 2.1 Income-based vulnerability to poverty in Chile, Mexico, and Peru in the 2000s . . . . 33 2.2 Distribution of self-reported class status in Mexico, 2007 . . . . . . . . . . . . . . . . . . . . . . 35 2.3 Four economic classes, by income distribution, in selected Latin American countries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37 2.4 Horizontal decomposition of mobility in Peru, 2004–06 . . . . . . . . . . . . . . . . . . . . . . 38 2.5 Vertical decomposition of mobility in Peru, 2004–06 . . . . . . . . . . . . . . . . . . . . . . . . . 39 3.1 The intergenerational association between parental background and children’s income . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50 3.2 Impact of parental education on children’s years of education, selected countries . . . 54 3.3 Evolution of intergenerational persistence in education across birth cohorts in seven Latin American countries, 1930s–80s . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55 3.4 Evolution of intergenerational persistence in education across birth cohorts in Peru and Colombia, 1920s–80s: Decomposition between parental inequality and E . . 56 3.5 Average children’s educational gap in Latin America, 1995–2009 . . . . . . . . . . . . . . . . 57 3.6 Differences in the educational gap between the top and bottom income quintiles in Latin America, 1995–2009 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57 3.7 Impact of parental background on children’s educational gap at age 15 in Latin America, 1995–2009 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59 3.8 Impact of ethnic minority status on children’s educational gap in Brazil, Ecuador, and Guatemala . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60 3.9 Influence of parental background on secondary students’ PISA test scores across countries and economies, 2009 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61 viii CONTENTS 3.10 Relationship of average PISA test scores and intergenerational mobility across 65 countries and economies, 2009 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62 3.11 Enrollment and inequalities in reading test scores, selected countries, 2006 . . . . . . . . 64 3.12 Intergenerational earnings elasticity between fathers and sons and its relationship to earnings inequality. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65 3.13 Impact of public education expenditures on the schooling gap between rich and poor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69 B3.4.1 Tuition loans and school enrollment in Chile . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72 3.14 Direct and overall impact of parental background on children’s test scores . . . . . . . . . 73 3.15 Differences in school characteristics between the top and bottom quintile of the ESCS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75 3.16 School practices and reading test scores for high and low values of selected policies . . 79 F3.1A School enrollment rates, selected Latin American countries . . . . . . . . . . . . . . . . . . . . 83 F3.1B Inequalities in reading test scores of sixth-grade students, selected Latin American countries, 2006 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85 F3.1C Inequalities in reading test scores at age 15, selected Latin American countries, 2009 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86 4.1 Sliders, climbers, and stayers: Intragenerational mobility in Latin America, by country . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99 4.2 Intragenerational mobility in Latin America, by country . . . . . . . . . . . . . . . . . . . . . . 101 4.3 Mobility for whom? Contribution to overall mobility of initial economic status in Latin America, by country . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102 4.4 Upward mobility out of poverty: Origin and destination in Uruguay, 1989–2009 . . 103 4.5 Intragenerational upward mobility in Latin America: Origin and destination, by country . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104 4.6 Growth incidence curves for Costa Rica and El Salvador, using anonymous and non-anonymous information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105 4.7 Downward intragenerational mobility into poverty and out of middle class in Latin America, by country . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106 4.8 Downward mobility into poverty in Latin American revisited, by country . . . . . . . . 107 4.9 Economic class (circa 2010) and initial characteristics (circa 1995) in Latin America, by country . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109 4.10 Upward mobility conditional on initial characteristics in Latin America, by country . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111 4.11 GDP growth as a key correlate to upward mobility in Latin America . . . . . . . . . . . . 113 4.12 Mobility by decade in Latin America, 1990s versus 2000s . . . . . . . . . . . . . . . . . . . . 113 4.13 Mobility over time in Latin America, 1990s versus 2000s . . . . . . . . . . . . . . . . . . . . 114 4.14 Upward mobility and inequality in Latin America: A trade-off? . . . . . . . . . . . . . . . . 115 4.15 Educational expenditures and upward mobility in Latin America . . . . . . . . . . . . . . . 115 4.16 Overall and targeted social protection expenditures and upward mobility in Latin America . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116 4.17 Female labor force participation and upward mobility in Latin America. . . . . . . . . . 117 4.18 Informality and upward mobility in Latin America . . . . . . . . . . . . . . . . . . . . . . . . . . 118 B4.3.1 The effect of mobile phone coverage on extreme poverty in rural Peru . . . . . . . . . . . 119 F4.1 Poverty dynamics: Synthetic versus actual panel data for alternative poverty lines in Peru, 2008 and 2009 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123 5.1 Income distribution in Latin America and the Caribbean, selected countries, 2009 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137 CONTENTS ix 5.2 Class composition in Latin America by income percentile, selected countries, 2009 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138 5.3 Middle class, vulnerability, and poverty trends in Latin America, 1995–2009 . . . . . 139 5.4 Middle class versus economic growth in Latin America, selected countries, 2000–10 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139 B5.1A The Brazilian middle class under alternative defi nitions, 1990–2009 . . . . . . . . . . . . 140 B5.1B Consumer and mortgage credit relative to GDP in Brazil, 2001–09 . . . . . . . . . . . . . 141 5.5 Decomposition of class growth attributable to income growth versus redistributive policies in Latin America, by country, circa 1995–2010 . . . . . . . . . . . 143 5.6 Middle-class growth forecasts for Latin America, 2005–30 . . . . . . . . . . . . . . . . . . . 144 5.7 Middle-class growth in the BRICs, circa 1980–2010. . . . . . . . . . . . . . . . . . . . . . . . . 144 5.8 The emerging world’s middle-class growth forecasts, 2005 versus 2030 . . . . . . . . . . 145 5.9 Average years of schooling (ages 25–65), selected Latin American countries, by income class, circa 2009 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149 5.10 Percentage of households living in urban areas, by income class, selected Latin American countries, circa 2009 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149 5.11 Percentage of adults (25–65) living in a municipality other than place of birth, by income class, selected Latin American countries, circa 2009 . . . . . . . . . . . . . . . . 150 5.12 Female labor-force participation by class, ages 25–65, selected Latin American countries, circa 2009 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153 F5.1A Middle-class growth trends in Chile under two absolute defi nitions, 1992–2009 . . . 154 F5.1B Middle-class trends in Peru and Argentina under absolute and relative defi nitions, by income percentile, 1990s–2000s . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155 F5.1C Comparison of income polarization in selected countries of the world . . . . . . . . . . . 156 F5.1D Average income by occupation type in Chile, 2009 . . . . . . . . . . . . . . . . . . . . . . . . . 157 6.1 Education, class, and values, selected Latin American countries, 2007 . . . . . . . . . . . 168 6.2 Income versus country-specific values, selected Latin American countries, 2007 . . . 172 6.3 Class incidence of social policies, selected Latin American countries, circa 2007–10 . . 174 6.4 Incidence of tertiary public education spending, selected Latin American countries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175 6.5 Percentage of students 6–12 years old enrolled in private schools, by income group, selected Latin American countries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175 B6.4 Ownership of electrical inverters in the Dominican Republic, 2010 . . . . . . . . . . . . . 176 6.6 Sixth-grade reading test scores, by income group, selected Latin American countries, 2006 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177 6.7 Tax revenues by type, selected Latin American and other countries, 1990–2010 . . . 178 Focus Notes 2.1 Mobility concepts and measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42 3.1 Bounding the estimates of parental background on student achievement . . . . . . . . . . 83 4.1 Synthetic panels using repeated cross-sectional data . . . . . . . . . . . . . . . . . . . . . . . . . 121 5.1 The Latin American middle class under alternative defi nitions . . . . . . . . . . . . . . . . . 154 Tables O.1 Intragenerational mobility in Latin America over the past 15 years, circa 1995–2010 . . 5 2.1 How different mobility concepts rank the same vector transformation . . . . . . . . . . . 26 x CONTENTS 2.2 Key mobility concepts and domains under consideration: The main diagonal . . . . . . . 29 2.3 Income-based defi nitions of the middle class . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32 2.4 Middle-class thresholds from self-reported class status, selected Latin American countries, 2007 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35 2.5 Matrix decomposition of M 3: A schematic representation . . . . . . . . . . . . . . . . . . . . . . 39 2.6 Matrix decomposition of M 3 in Peru, 2004–06 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41 F2.1 Sample mobility functions and graphical representation of Peru, 2004–06 . . . . . . . . 42 3.1 Relationship between parental education and children’s average educational gap at age 15 in Latin America, 1995 versus 2009 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58 3.2 Interaction of school practices and parental background on reading test scores . . . . . 78 4.1 Intragenerational mobility in Latin America over past 15 years (circa 1995–2010) . . . 98 4.2 Intragenerational mobility in Latin America, by median income change, (circa 1995–2010) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100 4.3 Intragenerational mobility in Latin America, by median income change, (circa 1995–2010) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100 A4.1 Data sets used, years, and coverage, by country. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124 A4.2A Regional weighted intragenerational mobility decomposition . . . . . . . . . . . . . . . . . . 125 A4.2B Regional weighted intragenerational mobility decomposition . . . . . . . . . . . . . . . . . . 125 A4.2C Country-specific intragenerational mobility in Latin America . . . . . . . . . . . . . . . . . . 126 A4.2D Country-specific intragenerational mobility decomposition in Latin America, by country . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128 A4.2E Country-specific weighted intragenerational mobility decomposition in Latin America, by country . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130 5.1 Average class characteristics in El Salvador, Panama, and Argentina, 2009/10 . . . . . 146 5.2 Trends in middle-class characteristics in Latin America (pooled), 1992–2009 . . . . . 147 5.3 Average household characteristics, selected Latin American countries, circa 2009 . . 148 5.4 Employment sector by class, ages 25–65, selected Latin American countries, circa 2009 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151 5.5 Employment status by class, ages 25–65, selected Latin American countries, circa 2009 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152 5.6 Private and public employment by class, ages 25–65, selected Latin American countries, circa 2009 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153 6.1 Relationship between economic development and institutions . . . . . . . . . . . . . . . . . . 164 6.2 The middle-class effect on indicators of social policy, economic structure, and governance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165 Foreword A fter a decade marked by sustained eco- generations, to understand the drivers of nomic growth—despite the 2008–09 success in escaping poverty. global financial crisis—and declining The result is a nuanced picture. On the one inequality in many countries in Latin America hand, in most countries in the region, while and the Caribbean (LAC), it is time to take intergenerational mobility has improved, stock of the region’s broad socio-economic it remains limited: parents’ education and trends. Moderate poverty fell from more than income levels still substantially influence 40 percent in 2000 to less than 30 percent in their children’s outcomes, and this appears 2010. This decline in poverty implies that 50 to be true to a greater extent than in other million Latin Americans escaped poverty over regions. On the other hand, mobility within the decade. But which workers and households generations has been significant. At least succeeded in leaving poverty, and which did 40 percent of the region’s households are not? What happened to those who left poverty estimated to have moved upward in “socio- behind? Did they all join the region’s growing economic class” between 1995 and 2010. middle class? What are the implications for Most of the poor that moved up did not go public policy? directly to the middle class but rather joined To address these questions, Economic a group sandwiched between the poor and Mobility and the Rise of the Latin American the middle class, which the report calls the Middle Class exploits a unique combina- vulnerable class and is now the largest class tion of data sources, ranging from multiple in the region. household surveys and student achievement Still, the Latin American middle class did tests to surveys of attitudes, opinions, and grow and very substantially: from 100 mil- beliefs, to shed light on the social transfor- lion people in 2000 to around 150 million by mation going on in Latin America in this the end of the last decade. The emerging mid- new millennium. It proposes a new defi ni- dle class differs, of course, from country to tion of the middle class based on economic country, but there are a number of common security and applies it to most countries threads. Middle class entrants are more edu- in the region. The report also investigates cated than those they have left behind. They economic mobility, both within and across are also more likely to live in urban areas and xi xii FOREWORD to work in formal sector jobs. Middle class region, much remains to be done. Regional women are more likely to have fewer children leaders will need to continue to devote con- and to participate in the labor force than siderable policy attention to the one-third women in the poor or vulnerable groups. of Latin Americans who remain poor, while This report will certainly stimulate the seeking to promote the security and prosper- debate on the implications of these new ity of those who are vulnerable. trends—for the functioning of the economy, for policy priorities, and for the performance Hasan Tuluy of democratic institutions. While LAC is now Vice President on the path to becoming a middle-income Latin America and the Caribbean Region Acknowledgments This report is dedicated to the memory of Gonzalo Llorente. T his report was prepared by a team Crawford, Wendy Cunningham, Anna Frut- led by Francisco H.G. Ferreira, Julian tero, Rafael de Hoyos, and Alex Solis are Messina, and Jamele Rigolini, and gratefully acknowledged. comprising Luis Felipe López-Calva, Maria We are also grateful to the individuals and Ana Lugo, and Renos Vakis. Important addi- organizations that hosted a series of consulta- tional contributions were made by João Pedro tions undertaken in the spring 2011, includ- Azevedo, Nancy Birdsall, Maurizio Bussolo, ing (but not restricted to) Leonardo Gasparini Guillermo Cruces, Markus Jäntti, Peter Lan- (CEDLAS), Alejandro Gaviria (Universidad de jouw, Norman Loayza, Leonardo Lucchetti, los Andes), Miguel Jaramillo (GRADE), Edu- Nora Lustig, Bill Maloney, Eduardo Ortiz, ardo Lora (IDB), Patricio Meller (CIEPLAN), Harry Patrinos, Elizaveta Perova, Miguel Marcelo Neri (CPS-FGV), Rafael Rofman Sánchez, Roby Senderowitsch, Florencia (World Bank), Isidro Soloaga (El Colegio de Torche, and Mariana Viollaz. The team was México), and Miguel Székely (Instituto Tec- ably assisted by Manuel Fernández Sierra, nológico de Monterrey). Thanks are also due Gonzalo Llorente, Nathaly Rivera Casa- to our hosts at the Institute for Economic nova, and Cynthia van der Werf. The work Analysis (IAE), Barcelona, where a mid-term was conducted under the general guidance of conference was held: Joan Maria Estebán, Augusto de la Torre, LCR Chief Economist. Ada Ferrer-i- Carbonnel and Xavi Ramos. The team was fortunate to receive advice The team would like to acknowledge fi nan- and guidance from four distinguished peer cial support from the Government of Spain, reviewers: François Bourguignon, Gary under the SFLAC program. Book design, edit- Fields, Philip Keefer and Ana Revenga, as ing, and production were coordinated by the well as from a panel of advisers compris- World Bank’s Office of the Publisher, under ing Nancy Birdsall, Louise Cord, and James the supervision of Patricia Katayama, Nora Foster. While we are very grateful for the Ridolfi, and Dina Towbin. guidance received, these advisors and review- Last but not least, we thank Ruth Del- ers are not responsible for any remaining gado, Erika Bazan Lavanda, and Jacqueline errors, omissions or interpretations. Addi- Larrabure Rivero for unfailing administra- tional insights from Barbara Bruns, Michael tive support. xiii Abbreviations CCT conditional cash transfer ELTI mobility as equalizer of long-term incomes ESCS economic, social, and cultural status (PISA index) GDP gross domestic product GIC growth incidence curve IMD directional income movement IMND nondirectional income movement km kilometer(s) MOI mobility as origin independence OECD Organisation for Economic Co-operation and Development PISA Program for International Student Assessment PM positional movement PPP purchasing power parity SEDLAC Socioeconomic Database for Latin America and the Caribbean (by the Centro de Estudios Distributivos, Laborales y Sociales [CEDLAS] of the Universidad de la Plata in Argentina, and the World Bank) SERCE Second Regional Comparative and Explanatory Study SM share movement USAID U.S. Agency for International Development WDI World Development Indicators xv Overview A fter decades of stagnation, the size This report discusses the relevant concepts of the middle class in Latin Amer- and documents the facts about mobility in ica and the Caribbean recently Latin America and the Caribbean over the expanded by 50 percent—from 103 million past two decades, both within and between people in 2003 to 152 million (or 30 per- generations. In addition, it investigates the cent of the continent’s population) in 2009. rise of the Latin American middle class over Over the same period, as household incomes the past 10–15 years and explores the size, grew and inequality edged downward in nature, and composition of this pivotal new most countries, the proportion of people in social group. More speculatively, it also asks poverty fell markedly: from 44 percent to how the rising middle class may reshape the 30 percent. As a result, the middle class and region’s social contract. the poor now account for roughly the same share of Latin America’s population. This is A middle-income region on the in stark contrast to the situation prevailing way to becoming a middle-class (for a long period) until about 10 years ago, region when the share of the poor hovered around 2.5 times that of the middle class. This Defining the middle class is not a trivial mat- study investigates the nature, determinants, ter, and the choices depend on the perspec- and possible consequences of this remark- tive of the researcher. Sociologists and politi- able process of social transformation. (See cal scientists, for instance, usually define the figures O.1 and O.2.) middle class in terms of education (for exam- Such large changes in the size and com- ple, above secondary), occupation (typically position of social classes must, by definition, white collar), or asset ownership (including imply substantial economic mobility of some the ownership of basic consumer durables form. A large number of people who were or a house). Economists, by contrast, tend poor in the late 1990s are now no longer to focus on income levels. This study adopts poor. Others who were not yet middle class an economic perspective but, to arrive at a have now joined its ranks. But social and eco- more robust—less arbitrary—definition, it nomic mobility does not mean the same thing anchors the income-based definition on the to different people or in different contexts. crucial notion of economic security (that is, a 1 2 OVERVIEW low probability of falling back into poverty). Although US$10 per day (or US$3,650 per The thresholds chosen for per capita income person per year) may not sound like a par- and economic security arise from the analy- ticularly demanding requirement for a fam- sis of Latin American data and are there- ily to be considered middle class, this income fore broadly applicable to middle-income level corresponds to the 68th percentile of the countries. Latin American income distribution in 2009. The study applies this definition of the In other words, according to our definition, middle class consistently across a compre- 68 percent of the region’s population—over hensive, Latin America-wide set of house- two-thirds—lived below middle-class income hold surveys. It presents a profile of the new standards in 2009. Not all of these people middle class in the region, highlighting both were poor, of course. If we use US$4 per day objective characteristics—including demo- as a moderate poverty line for the region, as graphics, education, and occupation—and typically done by the World Bank, these 68.0 subjective values and beliefs. It also asks how percent are split into 30.5 percent of the pop- this middle class interacts with economic ulation living in poverty (US$0–US$4 per and social policy, both in terms of the past day) and 37.5 percent living between poverty policies that helped shape its growth and in and the middle class (US$4–US$10 per day). terms of what its views, opinions, and rising This second group is a segment of the popu- political weight might mean for future pol- lation that is at risk of falling into poverty, icy choices. Because policy choices and the with an estimated probability greater than growth of the middle class are jointly deter- 10 percent. mined, the study often documents correla- Above the vulnerable segment, about 30 tions. Only where special data circumstances percent of the Latin American population permit are causal effects between policies and are in the middle class (US$10–US$50 per income movements inferred. day) and some 2 percent are in the upper- The concept of economic security is cen- income class (living on more than US$50 per tral to our approach because a defining fea- day), to whom we will refer interchangeably ture of middle-class status is a certain degree as the rich or the elite. Figure O.1, which of economic stability and resilience to shocks. draws on harmonized household surveys We adopt a probability of falling into pov- from 15 countries in Latin America and erty over a five-year interval of 10 percent the Caribbean (accounting for 86 percent (approximately the average in countries such of the region’s population and representing as Argentina, Colombia, and Costa Rica) as 500 million people) depicts the continent- the maximum level of insecurity that may wide income distribution and indicates the reasonably be borne by a household that is three key per capita income thresholds in considered middle class. To map such a prob- our analysis: the poverty line at US$4 per ability to a household income range, we ask— day, the lower bound for the middle class in those countries for which the right kinds of at US$10 per day, and its upper bound at data are available—which income levels are US$50 per day.2 typically associated with that level of insecu- Figure O.1 illustrates one of the key results rity. This exercise yields an income threshold from this study: if one adopts a middle class of US$10 per day, at purchasing power par- definition based on the notion of economic ity (PPP) exchange rates, as our lower-bound security—and validated by self-perceptions— per capita household income for the middle as well as a standard moderate poverty line, class.1 The upper income threshold for the then there are four, not three, classes in Latin middle class is set at US$50 per capita per America and the Caribbean. Sandwiched day, based primarily on survey data consid- between the poor and the middle class, there erations. According to these thresholds, a lies a large group of people who appear to family of four would be considered middle make ends meet well enough so as not to class if its annual household income ranged be counted among the poor but who do not between US$14,600 and US$73,000. enjoy the economic security that would be OVERVIEW 3 FIGURE O.1 The distribution of income in population is much less attractive than being Latin America and the Caribbean, 2009 a middle-class continent, but it is clearly much better than being a predominantly poor continent. Moreover, the current situa- .04 tion in the region is as recent as it is unprec- edented—it is the result of a process of social transformation that began around 2003, in .03 which upward social mobility took place at a remarkable pace. Before 2005, as figure O.2 shows, poverty was still the most prevalent Density .02 condition in our four-way classification. In an almost mechanical sense, this trans- formation reflects both economic growth and .01 declining inequality in Latin America and the Caribbean over the period. Gross domestic product (GDP) per capita grew at an annual 0 rate of 2.2 percent between 2000 and 2010 4a 10b 50c 100 and somewhat faster over the crucial 2003– Per capita daily income, US$ PPP 09 period. Although these are not East Asian growth rates, they represent a substantial Source: Authors’ calculations on data from SEDLAC (Socio-Economic Data- base for Latin America and the Caribbean). improvement over the region’s own past Note: PPP = purchasing power parity. Countries include Argentina, Bolivia growth performance: negative 0.2 percent (2008), Brazil, Chile, Colombia, Costa Rica, the Dominican Republic, Ecua- dor, El Salvador, Honduras, Mexico (2010), Panama, Paraguay, Peru, and per year in the 1980s and positive 1.2 percent Uruguay. in the 1990s. And whereas in those earlier a. US$4 = moderate Latin American and Caribbean poverty line. b. US$10 = lower bound of Latin American middle class. c. US$50 = upper bound of Latin American middle class. FIGURE O.2 Trends in middle class, vulnerability, and poverty in Latin America and the Caribbean, 1995–2009 required for membership in the middle class. One might have called this group by various 50 names, such as the near-poor or the lower 45 middle class. Because, by virtue of our defini- 40 tion of the middle class, these are households Percentage of population with a relatively high probability of experi- 35 encing spells of poverty in the future, we call 30 them “the vulnerable.” 25 As shown in figure O.1, this vulnerable 20 class includes the modal Latin American household—the household whose income is 15 observed with the highest frequency in the 10 distribution. And as shown in figure O.2, it 5 is now the largest social class in the region, 0 accounting for 38 percent of the population. 1995 2000 2005 2010 As poverty fell and the middle class rose—to Year about 30 percent of the population each dur- Poor (US$0–US$4 a day) Vulnerable (US$4–US$10 a day) ing the past decade—the most common Latin Middle class (US$10–US$50 a day) American family is in a state of vulnerability. Source: Authors’ calculations on data from SEDLAC (Socio-Economic Database for Latin America and Yet there is no question that the dynamics the Caribbean). illustrated by figure O.2 are, on the whole, Note: PPP = purchasing power parity. Covered countries include Argentina, Bolivia, Brazil, Chile, Colombia, Costa Rica, the Dominican Republic, El Salvador, Ecuador, Guatemala, Honduras, Mexico, very encouraging. Being a continent where Nicaragua, Panama, Paraguay, Peru, Uruguay, and República Bolivariana de Venezuela. Poverty lines the vulnerable are the largest segment of the and incomes are expressed in 2005 US$ PPP per day. 4 OVERVIEW FIGURE O.3 The growth and redistribution components of middle-class growth in Latin America and the Caribbean, 1995–2010 US$10 to US$50 a day 20 Change in middle class (percentage points) 10 0 –10 s ht s a il a ca lic r r as ico a ay y do do rie ig trie ua az in bi m ile ru ur ub Ri gu Ar d ex nt m na Br ua lva nt ug Pe Ch e nd we un sta ep ra lo ge M ou Pa Ec Ur Sa Ho Pa un n co Co nR Co nc El ica ica ica in er er m Am Am Do tin tin La La Redistribution Growth Source: Azevedo and Sanfelice (2012) based on SEDLAC (Socio-Economic Database for Latin America and the Caribbean) data. Note: PPP = purchasing power parity. Middle-class per capita income is expressed in 2005 US$ PPP per day. decades inequality was either stable or rising, 15 years or so—which are used to account the 2000s saw declining income disparities for middle-class growth in figure O.3—are in 12 of the 15 countries for which data are themselves aggregate statistics that simply available (as further discussed in chapter 1). summarize changes in well-being for indi- Both of these factors—higher incomes and viduals and families. Behind these account- less income inequality—contributed to pov- ing decompositions are real individual tra- erty reduction and the growth in the middle jectories, which generally imply significant class. Statistically, however, economic growth churning in the distribution of incomes. In (growth in average per capita income) played any given year, some households earn more a much larger role, accounting for 66 percent than before while others earn less. Behind the of the reduction in poverty and 74 percent net changes in the size of each socioeconomic of the rise of the middle class in the 2000s class depicted in figure O.2, there are larger (with the remainder, in each case, associated gross flows, with many households moving with changes in inequality). Yet, as figure up while others move down. O.3 illustrates, the average hides significant To shed light on these dynamics, we adopt intercountry variation within Latin America a measure of economic mobility within gener- in these decompositions: in Argentina and ations (intragenerational mobility) that sum- Brazil, for example, falling income inequality marizes (directional) income movement. Put contributed substantially to the expansion of simply, this measure of directional income the middle class.3 movement captures the average growth rate in household-specific incomes.4 This mobility index, which is well known in the scholarly Within generations, remarkable literature, can be decomposed into “gainers” upward mobility and “losers” as well as by the original social In a deeper sense, the rise of the region’s mid- class of each household. This decomposition dle class also reflects substantial upward eco- allows various versions of the measure to be nomic mobility. The growth in mean incomes expressed in terms of transition matrices, and the changes in inequality over the past such as in table O.1. Considering that data OVERVIEW 5 TABLE O.1 Intragenerational mobility in Latin America over the past 15 years, circa 1995–2010 (percentage of population)     Destination (c. 2010) Poor Vulnerable Middle class Total Poor 22.5 21.0 2.2 45.7 Origin (c.1995) Vulnerable 0.9 14.3 18.2 33.4 Middle class 0.1 0.5 20.3 20.9 Total 23.4 35.9 40.7 100.0 Source: Authors’ calculations on data from SEDLAC (Socio-Economic Database for Latin America and the Caribbean). Note: “Poor” = individuals with a daily per capita income lower than US$4. “Vulnerable” = individuals with a daily per capita income of US$4–US$10. “Middle class” = individuals with a daily per capita income higher than US$10. Poverty lines and incomes are expressed in 2005 US$ PPP per day. PPP = purchas- ing power parity. The table shows lower-bound mobility estimates. Results are weighted averages for 18 Latin American and Caribbean countries using country-specific population estimates of the last available period (as detailed further in the notes to table 4.1, chapter 4). The bottom row does not match the numbers used above for describing figure O.1 because of sample differences in both countries and years. In addition, table O.1 conflates the middle class and elite into a single class. following the same individuals (that is, panel social classes between the mid-1990s and the data) for long time spans are rarely available end of the 2000s, and most of this move- in the region, directional income mobility ment was upward. In fact, only 2 percent was estimated using synthetic panels, and of the population experienced a downward we report here conservative (that is, lower- class transition, (although this is also a lower bound) measures of mobility.5 bound). Table O.1 provides a summary of eco- As one might expect, most class move- nomic mobility within generations between ment was gradual: most of the “climbers” circa 1995 and circa 2010 for Latin America moved either from poverty to vulnerability as a whole. The data are representative of 18 or from vulnerability to the middle class; few countries in the region. Each cell gives the made the jump directly from poverty to the proportion of the overall population that middle class during these 15 years. Rags- started out in the “origin” row of socioeco- to-riches stories capture the imagination nomic class in 1995 and ended up in the “des- precisely because they are, in reality, rather tination” column of class in 2010. For exam- rare—even in a high-mobility context such as ple, the first row tells us that, of the 45.7 Latin America in the 2000s. percent of the population who were poor in Naturally, these average statistics once 1995, fewer than half (22.5 percent) were still again hide considerable variation, both poor in 2010, while the rest mainly moved up within and across countries. The extent of to become vulnerable (21.0 percent) and, to a economic mobility captured by our measure substantially lesser extent, jumped directly to of directional income movement was much the middle class (2.2 percent). Analogously, higher in Brazil and Chile, for example, than of the 33.4 percent of the population who in Guatemala or Paraguay. There was also started out as vulnerable in 1995, more than variation in terms of where in the distribution half (18.2 percent) moved up and joined the the mobility was taking place, often associ- middle class.6 ated with the initial level of the country’s Table O.1 reveals an impressive degree income per capita: whereas most mobility in of income mobility in Latin America. The Ecuador and Peru came from the originally population shares along the main diagonal poor, in Argentina and Uruguay—countries represent the “stayers”: people whose income with a higher initial per capita income— movement over this period, upward or down- most of it was accounted for by the originally ward, was insufficient for them to cross a vulnerable. class threshold. Because these shares add up Within most Latin American countries, to 57.1 percent, we can conclude that at least households were more likely to experience 43.0 percent of all Latin Americans changed upward mobility if the household head had 6 OVERVIEW more years of schooling in the initial year. decreases as the correlation between initial Movements into the middle class, in particu- and final positions increases. In the present lar, were much likelier for people who had context, origin dependence would refer to some tertiary education. Being employed in the extent to which the family and socio- the formal sector and living in an urban area economic conditions into which a person is were also good predictors of upward mobil- born determine his or her future income and ity. Migration from rural to urban areas was socioeconomic class. A higher measure of ori- also associated with greater prospects of gin independence implies higher intergenera- upward movement, and more so for move- tional mobility. ments out of poverty than for transitions into As this discussion suggests, when the con- the middle class. cept of mobility as origin independence is Across Latin American and Caribbean applied to an intergenerational context, it is countries, there was a clear association closely related to the concept of equality of between faster GDP growth and higher opportunity. Equality of opportunity is now income mobility—not surprising in light of predominantly understood to refer to a hypo- our earlier comments on economic growth as thetical situation in which predetermined the principal driver of middle-class growth. circumstances—such as race, gender, birth- Overall economic mobility was also cor- place, or family background—have no effect related with public health and education on people’s life achievements. Perfect mobil- spending. Interestingly, mobility was not ity in an origin-independence sense means found to be correlated with total social pro- the same thing when one looks only at a tection expenditures, but when one disaggre- single circumstance variable, such as parental gates those expenditures by type, mobility schooling.7 turned out to be associated with measures The main message of this report in this of targeted, progressive social protection respect is that, sadly, despite substantial programs, including conditional cash trans- upward income movements within genera- fers. Although the extent of mobility into the tions, intergenerational mobility remains middle class was positively correlated with limited in Latin America. Because data on increases in female labor force participation, parental incomes for today’s working adults this was not true of mobility out of poverty. are impossible to obtain (and difficult to esti- All of these are, of course, purely descriptive mate) for most countries in the region, most correlations. On the basis of the evidence of our analysis of intergenerational mobil- presented in the report, the variables in ques- ity—or lack thereof—relies on educational tion should not be interpreted as causes of attainment (as measured by years of school- mobility. ing) and educational achievement (as mea- sured by standardized test scores). In particu- lar, we ask to what extent the education of Across generations, mobility a person’s parents appears to determine the remains low person’s own level of educational attainment The above evidence does not imply that Latin (or achievement). One way to make that com- America is a high-mobility society in every parison across countries is to consider the sense of the word. As noted earlier, mobility effect of one standard deviation in parental has different meanings in different contexts, years of schooling on the years of school- and one important such meaning—par- ing of the children. By this metric, as figure ticularly in an intergenerational context—is O.4 illustrates, there is much greater inter- that of “origin independence.” A measure of generational persistence—that is, much less mobility as origin independence reaches its mobility—in Latin American countries (such maximum when information on the origi- as Brazil, Ecuador, Panama, and Peru) than nal, or initial, period is useless in predict- in most other countries—rich or poor—for ing terminal (or final) position. The measure which data are available. OVERVIEW 7 FIGURE O.4 Association between parental education and children’s years of schooling, selected countries 3 Years of education 2 1 0 rg ral pia Un ther epu na d re c Ne Kin land Ze om ec No and ep ay nm ic Uk ark va Ma ine ep sia Fin blic Ea sto d st nia lg r Un P ium d nd Ne es es Sw pal M en So Ire ab) Ne h A nd Ph erla ca Sw lipp ds er s et d Slo nam n a i L ry kis a n Gh aly Ni one a ca sia lo ua t, A C ia ra hile B p. u l na r Pe a ru Ec razi Be imo Pa ado itz ine ite n I bli Hu eni Pa ank d n m ta E an Vi lan De ubl b Re Sr ga at th fri i n h R rw Co rag r R hi In a h ( ed ite ola ut la k R lay It Ky Ru hio m ra l w gd u at al l St v No yz C T b t lE ra Ru yp lad Slo Cz Eg ng Ba Source: Authors’ calculations based on data from Hertz et al. 2007. Note: Bars represent the impact of one standard deviation of parental years of schooling on the years of schooling of children. The impact is averaged across birth cohorts born between 1930 and 1980. A similar, if slightly less stark, picture FIGURE O.5 Relationship between average PISA test scores and arises if one considers the effect of parental intergenerational mobility across 65 countries, 2009 background (measured by an index of socio- economic status) on student achievement, 550 FIN KOR measured by standardized test scores in Pro- CAN gram for International Student Assessment 500 DEU (PISA) exams, illustrated in figure O.5.8 Most USA Latin American countries for which the rel- Average test score evant data are available also appear toward 450 CHL Better performance URU the right of the distribution of that impact MEX TTO COL estimate, suggesting that family background 400 IDN BRA ARG is a bigger determinant of student learning PAN PER in Latin America than in other regions. But 350 there is more variation in those estimates than in the attainment numbers shown in figure O.4: in Mexico, for example, parental 300 10 20 30 40 50 background appears to be much less closely Effect of socioeconomic background on reading test scores associated with PISA test scores than in other Latin American countries or in a number More mobility of nations in other regions. Crucially, how- ever, most Latin American countries display Source: PISA 2009 data. not only lower intergenerational mobility in Note: PISA = Program for International Student Assessment. The effect of socioeconomic back- ground on reading test scores is calculated using the PISA index of economic, social, and cultural educational achievement but also very low status. The horizontal line represents the average test score in the sample. The vertical line repre- levels of student learning—an unfortunate sents the average effect of socioeconomic background on scores in the sample. 8 OVERVIEW FIGURE O.6 Impact of parental background on children’s educational gap at age 15 in Latin America, 1995–2009 0.80 0.60 0.40 0.20 0 –0.20 –0.40 –0.60 –0.80 –1.00 –1.20 r il ia r ile a ico ca a as lic a ru ay B a y do do ua az bi m in gu ,R liv Pe ur ub Ri gu Ch ex nt m na Br ua lva ug ela ra Bo nd sta ep ra lo ge M Pa ca Ec Ur Sa zu Ho Pa Co nR Co Ar Ni ne El ica Ve in m Do c. 1995 c. 2009 c. 2009 – c. 1995 Source: Data from SEDLAC (Socio-Economic Database for Latin America and the Caribbean). Note: “Educational gap” is defined as the difference between potential years of education at a given age and the years of completed education at that age. The green and orange bars represent the expected reduction in the schooling gap associated with one standard deviation of parental education in 1995 and 2009, respectively. The red bar is the difference between the two. Other covariates in the regression are children’s gender, living in an urban area, and country fixed effects. The estimated effect of parental education on the educa- tional gap is always statistically different from zero and so are the differences between 1995 and 2009. combination that clearly leaves a great deal cognitive outcomes through better nutrition, of scope for policy interventions in this area. exposure to richer vocabulary, differences in There is also some evidence on the mecha- cognitive stimulation, material resources at nisms through which the intergenerational home, and so on. persistence of educational achievement There is some room for hope that these occurs. In particular, it appears that sort- abysmally low levels of intergenerational ing —the process whereby children from mobility in Latin America—that is, high lev- more-advantaged backgrounds concentrate els of inequality of opportunity—are begin- in the same schools, from which those from ning to change. Intergenerational mobility in less-privileged families are excluded—is a educational attainment appears to have been more important component of intergen- rising over the past decade or so in most of erational immobility in Latin America than the region. Figure O.6 shows estimates of elsewhere. Sorting matters in Latin America the effect of one standard deviation of paren- because of the usual peer effects and because tal education on children’s schooling gap the schools attended by rich children are (the difference between the highest grade a much better than those attended by the poor, child could be attending under normal cir- in terms of their governance and account- cumstances and the last or current grade ability as well as their physical infrastructure actually attended) in 1995 and 2009. The and teaching quality. Of course, in addition, red bars show that the differences are posi- parental background also affects children’s tive and substantial in most Latin American OVERVIEW 9 countries, suggesting a generally improving FIGURE O.7 Association between income inequality and trend. While this is encouraging, the result is intergenerational immobility restricted to educational attainment. There is no clear evidence of similar improvements 0.8 in educational achievement and, hence, no Intergenerational earnings elasticity room for complacency. Peru How likely is it that these measures of (low) intergenerational educational mobility 0.6 Brazil imply similarly limited mobility in incomes United Kingdom Italy United States Chile Argentina between generations? Although we did not Pakistan Switzerland conduct original analysis on intergenerational 0.4 France Spain Japan income transitions for this report, the schol- Germany arly literature suggests that Latin America is Sweden New Zealand Australia also a region of low intergenerational mobil- 0.2 Finland Norway Canada ity in terms of income, and that this goes Denmark hand in hand with the region’s (still) high 20 30 40 50 60 levels of income inequality. This relationship Inequality (Gini coefficient) is corroborated in figure O.7—which repro- duces a well-known positive association: the Source: Corak 2012. higher the inequality of income (as measured by the Gini coefficient), the higher the inter- generational immobility. in its subjective values and beliefs. Take first In sum, the region’s stubbornly low levels the common objective features: In all Latin of intergenerational mobility stand in con- American countries, the heads of middle-class trast to the recent sharp increase in intragen- households have substantially more years of erational mobility. The overall picture of eco- schooling than those in the poor or vulnerable nomic mobility in Latin America is therefore classes but fewer years than the rich (figure a mixed one. Mobility across generations—in O.8). Middle-class households are also more the sense that personal outcomes are inde- urbanized than poorer groups. In addition, pendent of family background and social ori- formal employment appears to be a distinc- gin—remains an elusive goal. In intergenera- tive sign of the middle class in Latin America; tional terms, Latin America is not a mobile the middle-class worker is typically a formal society, and the signs that it is becoming a employee rather than being self-employed, little more mobile are tentative and so far unemployed, or an employer. In contrast, the limited to educational attainment. This pic- poor and vulnerable rely on self-employment ture is consistent with what is known about (or suffer from unemployment) more often, the high degree of inequality of opportunity while the rich are more frequently employers that continues to characterize the region. and, in some countries, self-employed. In terms of sector of economic activity, middle-class workers are frequently found A snapshot of the Latin American in the services sector, including health, edu- middle class cation, and public services, but manufac- What are the main characteristics of this turing jobs are more frequent among the emerging middle class? How similar is it middle class (and the vulnerable) than they across countries? Does it hold different views are among the poor or the rich. There is no and opinions than other social groups? Our evidence that the middle class is overly depen- analysis suggests, perhaps surprisingly, that dent on—or employed by—the public sector. the rising Latin American middle class, while In most Latin American countries for which sharing some common objective features data exist, public sector employment was across the region, displays much less similarity more frequent among the rich than among the 10 OVERVIEW FIGURE O.8 Average years of schooling (ages 25–65), selected respectively. Middle-class households typi- Latin American countries, by income class, circa 2009 cally have fewer children as well as women who join the labor force more frequently: 73 percent of middle-class women ages 25–65 16 across Latin America are either employed or looking for work compared with a region- wide population average of 62 percent. Their 12 children are typically in school: virtually Years of schooling all 6- to 12-year-old middle-class children attend school, as do roughly three-quarters 8 of those who are 13–18. In summary, although there are evidently variations in the middle-class profile across 4 countries, the similarities dominate: the mid- dle class presents a set of distinctive demo- graphic and socioeconomic patterns that 0 are present in almost every Latin American il ile a ca lic as ico ru country. Would this mean that the middle az bi Pe ur ub Ri Ch ex m Br nd sta ep lo M class also systematically shares opinions and Ho Co nR Co ica beliefs about society that are different from in m other groups? Our research suggests this not Do Poor Vulnerable Middle class Upper class to be the case. An analysis of middle-class values and Source: Birdsall 2012. beliefs using opinion surveys shows that Note: “Poor” = individuals with a per capita daily income lower than US$4. “Vulnerable” = individu- als with a per capita daily income of US$4–US$10. “Middle class” = individuals with a per capita country characteristics account for a much daily income of US$10–US$50. “Upper class” = individuals with a per capita daily income exceeding larger share of the variance in people’s val- US$50. Poverty lines and incomes are expressed in 2005 US$ PPP per day. PPP= purchasing power parity. ues than class membership. In particular, there is no strong evidence of any “middle- class exceptionalism” in terms of values and middle class (although Mexico and Peru were beliefs. To be sure, middle-class respondents exceptions). The public sector employed more are generally likelier than their poorer coun- than one-fourth of middle-class workers in terparts to trust their countries’ institutions only one country: Honduras. It would appear, (including the government, political parties, therefore, that popular images of the middle and the police) and to report greater faith in class—as being made up of either intrepid the meritocracy of their societies, and they entrepreneurs (who start their own small are less likely to perceive political violence businesses and pull themselves up the ladder as legitimate. But most of these associa- by their own shoestrings) or lazy bureaucrats tions simply reflect positive correlations with (comfortably relying on a government pay- income and education rather than something check)—are inaccurate. Typically, the Latin to do specifically with middle-class status. American middle-class worker is a reasonably And, on the whole, income and class status educated service worker, formally employed account for only a small share of the overall by a private enterprise in an urban area. variance in values. Family dynamics and demographics pro- This contrasting reality may be simply vide, perhaps, the most interesting traits of described as follows: when it comes to socio- the Latin American middle-class profile. economic and demographic characteristics, a Between 1992 and 2009, the average size of a middle-class person in Peru has more in com- middle-class household in Latin America fell mon with a middle-class person in Mexico from 3.3 to 2.9 individuals. This compares than with a poorer person in Peru; but when with populationwide averages of 4.1 and 3.4, it comes to values and aspirations, the same OVERVIEW 11 middle-class person in Peru has more in com- and Brazil, the region is characterized by rel- mon with a poor person in Peru than with a atively low tax revenues overall. The average middle-class person in Mexico. total tax revenue in 2010 was 20.4 percent of GDP in Latin America, versus 33.7 per- cent in the Organisation for Economic The middle class and the social Co-operation and Development (OECD) contract countries, for example.10 In addition, the What, if any, are the implications of a ris- composition of these tax revenues tended ing middle class with these characteristics— to be skewed toward indirect (sales) taxes urban, better educated, largely privately and social security contributions, relative to employed, and with beliefs and opinions income and property taxes, leading to a sys- broadly in line with those of their poorer and tem that is not particularly progressive. less-educated fellow citizens—for social and On the benefit side, the middle class (and economic policy? In particular, is the growth the elite) participated disproportionately of the Latin American middle class likely to in the social security system (including old- spell any changes for the region’s fragmented age and disability pensions, unemployment social contract? protection, severance payments, and health A “social contract” may be broadly under- insurance). But it tended to opt out of public stood as the combination of implicit and education and health services, in particular. explicit arrangements that determine what Instead, the upper and middle classes in Latin each group contributes to and receives from America often resorted to private alternatives the state. In stylized terms, Latin America’s to obtain these latter services. This tendency social contract in the latter half of the 20th to opt out extended even to services where century was characterized by a small state, public provision should be the uncontested to which the elite (and the small middle class norm, such as electricity: in some Latin appended to it) contributed through low American countries, private ownership of taxes, and from which they benefited largely electricity generators is still observed to rise through a “truncated” set of in-cash benefits with household income. The same applies for such as retirement pensions, severance pay- public security, with private security in closed ments, and the like, for which only formal condominiums not uncommon in a number sector workers qualified.9 Little was left for of countries in the region. providing high-quality public services in the This picture has not remained static, how- areas of education, health, infrastructure, ever. Over the past 10–20 years—and, in and security, for example. Public services in particular, following redemocratization pro- these areas were therefore generally of low cesses in many Latin American countries— quality; while the vast majority of the (poor this political equilibrium has begun to shift, and vulnerable) population had no choice, albeit gradually. The spread of noncontribu- the rich and the small middle class opted out tory old-age pension and health insurance and chose privately provided alternatives. schemes and the growth of conditional cash The essence of this (implicit) contract was transfers has meant that redistributive trans- simple: the upper and middle classes were fers from the state now reach the poor to an not asked to pay much and did not expect to extent that was unheard of 20 years ago in receive much from public services either. The most of the region. At the same time, in most poor also paid little and received correspond- countries in the region, the extension of cash ingly little in terms of public benefits. benefits to the poor has not been matched One manifestation of this social contract by a return of the middle class to public was a state that was typically small as well as health and education services. Latin Amer- skewed toward the provision of formal sec- ica’s “welfare state” may have become less tor social security payments to the better-off. “truncated,” but its social contract remains To this day, with the exception of Argentina fragmented. 12 OVERVIEW It is natural to question whether Latin remains to be seen and will doubtless be the America will be able to continue its recent subject of much research in the future. Nev- run of “growth with equity” (or at least with ertheless, the report highlights three areas declining inequality) on the basis of such a where reforms may help to gain the support fragmented contract, which inherently gen- of the middle class for a fairer and more erates fewer opportunities for the bulk of legitimate social contract: the population. Whether in postwar West- ern Europe or postrevolutionary China, • Incorporate the goal of equal opportu- whether in the post-land-reform Republic nities more explicitly into public policy. of Korea or in the United States under the This is crucial for ensuring that the mid- New Deal, socioeconomic progress has often dle classes feel that they live in a society required a combination of economic freedom where effort pays and merit is rewarded and a sound foundation of public education, instead of one that is rigged in favor of health, and infrastructure. It is almost cer- privileged groups. It is also crucial for tain that most countries in Latin America broadening the access of those who and the Caribbean will require additional remain poor or vulnerable to good jobs reforms to their social contracts to enable and stable sources of income. Although their states to provide that foundation and this effort will require reforms in a wide sustain growth. range of fields, this report emphasizes the But can the rise of the middle class docu- need to improve the quality of public edu- mented in this study facilitate these reforms? cation, from the development of cognitive Or will it instead entrench the middle-class and social skills during early childhood choice of private services and further reduce all the way to better colleges and univer- its willingness to contribute to the public sities. Greater equality of opportunity purse to generate opportunities for those who would, in turn, enhance economic effi- remain poor? In a sense, as it evolves toward ciency, thus helping address Latin Amer- a more mature social structure, with a larger ica’s persistent low-growth problem and and more vocal middle class, Latin America improving the conditions for the region’s stands at a crossroads: will it break (further) private sector to generate better and more with the fragmented social contract it inher- stable jobs for all classes. ited from its colonial past and continue to • Embark on a second generation of pursue greater parity of opportunities, or will reforms to the social protection system, it embrace even more forcefully a perverse encompassing both social assistance and model where the middle class opts out and social insurance. Although the aforemen- fends for itself? tioned improvements in targeted social This study does not answer those big assistance during the past 10–15 years questions. It merely poses them, because contributed much to the observed reduc- they follow naturally from the recent tions in poverty and income inequality, trends in economic mobility and the size their expansion has not been well inte- of the middle class—trends that combine grated into the overall social protection the good news of recent income growth system, and this has led to new challenges and poverty reduction with the reality of for both efficiency and fairness. Increas- limited mobility between generations and ingly, the middle classes are asked to pay persistent inequality of opportunity. The for services that are provided to others study suggests, however, that the middle for free. A dual social protection system classes may not automatically become the based on targeted assistance for the poor much-hoped-for catalytic agent for reforms. and (subsidized) insurance for the middle Whether and how the new middle class will classes may also be poorly tailored to a help strengthen the region’s social contract large vulnerable population that is neither OVERVIEW 13 poor nor middle class and whose vulner- Notes ability will rise if the external environ- 1. This lower income threshold was indepen- ment becomes less favorable than in the dently validated by an alternative approach, past. The time is ripe for embarking on based on self-perceptions of class member- a second generation of social protection ship, that was separately applied to five reforms, in which fragmentation will be countries: Brazil, Chile, Colombia, Mexico, overcome in ways that enhance fairness, and Peru. Methodological details of both solidarity, and inclusion. approaches are documented in chapter 2 of • Break the vicious cycle of low taxation the main report and in references therein. and low quality of public services that 2. As is well known, the household surveys on leads the middle and upper classes to which figure O.1 is based commonly suffer opt out. Although there is some margin from compliance and reporting problems that to improve the quality of public services render them unrepresentative of the top tail of within the current budget envelopes, the distribution. We are therefore circumspect it will be challenging to do so without in our analysis of the “rich” in our sample. 3. As detailed in chapter 5, these decompositions strengthening the revenue base, which are for the 1995–2010 period. remains low practically everywhere except 4. Because each household’s growth rate is given in Argentina and Brazil. Improving the equal weight, the average of growth rates is perception of fairness in taxation and not the same as the growth in the average the redistributive effectiveness of public income. The latter involves income weights, spending will be key to any successful whereas the former uses population weights. reform. The middle classes will not buy 5. Our measure of directional mobility is applied into and contribute to an improved social to a set of “synthetic panels” constructed contract if the goods that they value highly from the region’s repeated cross-section (such as civil rights protection, education, household surveys. A key caveat is that the police, and health services) are deficiently statistical procedures used to construct these supplied by the state and if they do not synthetic panels can only generate upper- and perceive that the rich contribute fairly to lower-bound estimates of mobility rather than exact figures. Most of the analysis in this the social contract. report relies on the lower-bound estimates, which yield a conservative picture of mobility During most of the 2000s, Latin Amer- in either direction. In our results, therefore, ica’s improved policy framework allowed upward and downward mobility are both many countries to take advantage of a benign likely to be underestimated. external environment to begin an impressive 6. The bottom row does not match the numbers transition toward a middle-class society. This used above for describing figure O.1 because has created enormous expectations, which of sample differences in both countries and risk turning into frustration if this transition years. In addition, table O.1 conflates the stalls. But the region cannot count on the middle class and elite into a single class. external environment remaining as friendly Despite sampling differences, though, the as in the recent past to achieve further social overall picture is consistent with the cross- and economic gains. A much greater policy section analysis described earlier. 7. The concepts of equality and inequality of effort will thus be required to consolidate and opportunity are increasingly important to the deepen the process of upward mobility and World Bank’s work in Latin America. See, to make it more resilient to potential adverse for example, the World Development Report shocks. In the end, the onus will mainly rest 2006: Equity and Development (World on the shoulders of the political leaders and Bank 2006), the regional study on Measuring democratic institutions of the region: they Inequality of Opportunities in Latin America face the challenge of overhauling its social and the Caribbean (Barros et al. 2009), and contract. references in those two volumes. 14 OVERVIEW 8. The OECD’s Program of International Stu- Birdsall, Nancy. 2012. “A Note on the Middle dent Assessment (PISA) produces a set of Class in Latin America.” Unpublished manu- school-based surveys that administer identi- script, Center for Global Development, Wash- cal cognitive achievement tests to samples of ington, DC. students across a number of countries, as well Corak, Miles. Forthcoming. “Inequality from as collecting (reasonably) comparable infor- Generation to Generation: The United States in mation about the students’ families and the Comparison.” In The Economics of Inequal- schools they attend. ity, Poverty and Discrimination in the 21st 9. The capture of Latin America’s social security Century, ed. Robert Rycroft. ABC-CLIO. systems by (largely better-off) formal sector De Ferranti, David, Francisco H. G. Ferreira, workers, to the exclusion of most of the con- Guillermo E. Perry, and Michael Walton. tinent’s poor, was described as a “truncated 2004. Inequality in Latin America: Breaking welfare state” in a previous regional report with History? Washington, DC: World Bank. in this series, Inequality in Latin America: Hertz, Tom, Tamara Jayasundera, Patrizio Breaking with History? (de Ferranti et al. Piraino, Sibel Selcuk, Nicole Smith, and Alina 2004). Verashchagina. 2007. “The Inheritance of 10. In 2010, Brazil’s total tax revenues were 33.6 Educational Inequality: International Compar- percent of GDP, whereas in Argentina the fig- isons and Fifty-Year Trends.” The B.E. Journal ure was 33.3 percent. of Economic Analysis & Policy 7 (2). SEDLAC (Socio-Economic Database for Latin America and the Caribbean). 2011. Data- References base of the Center for Distributive, Labor and Azevedo, Joao P., and Viviane Sanfelice. 2012. Social Studies, Argentina, and World Bank, “The Rise of the Middle Class in Latin Amer- Washington, DC. http://sedlac.econo.unlp ica.” Draft, World Bank, Washington, DC. .edu.ar/eng/. Barros, Ricardo, Francisco H. G. Ferreira, José World Bank. 2006. World Development Report Molinas, and Jaime Saavedra. 2009. Mea- 2006: Equity and Development. Washington, suring Inequality of Opportunities in Latin DC: World Bank. America and the Caribbean. Washington, DC: World Bank. 1 Introduction I sabel’s life is nothing like those she likes to that sells combine harvesters, tractors, and follow in the evening telenovelas. For one other equipment to the sugar cane planta- thing, she has been married for 20 years to tions that have boomed in western São Paulo the same man, Roberto. For another, she and state, supplying ethanol to the whole coun- her husband do not drive an imported car or try. Together, they brought home US$4,380 live in a luxurious apartment with a sea view per month in 2010 at purchasing power par- in Ipanema. Yet although she likes to dream ity (PPP) exchange rates.1 For the family of about some of the glamorous aspects of life three, this translates into an annual income in a Brazilian soap opera, Isabel is conscious of US$52,560. It was enough to pay the fees that her family—like her country—has not for Patrícia’s private school, from which she been doing badly of late. has just graduated. (At 18 years of age, she Isabel, Roberto, and their only child, was accepted into São Paulo State University Patrícia, live in Presidente Prudente, a city of [UNESP], the excellent university based in some 210,000 people that lies 580 kilometers her home town, to study veterinary sciences.) (km) west of São Paulo, Brazil. They are real Although it would not occur to the family to locals: both were born here, and Isabel’s late think of their income in daily terms, their per father owned a small padaria (bakery) in one capita daily income was US$48 at PPP—an of the city’s older residential neighborhoods. income that places them near the top of what He used to say that the bakery’s opening in this volume will argue should be considered 1952 was the fourth happiest day in his life: Latin America’s middle class in 2010. coming after only his wedding day and the You don’t have to go far to find rather dif- birth dates of Isabel and her brother. Both of ferent living standards. Fabiano and Irene live Isabel’s parents had completed high school, in the small town of Quatá, some 80 km east and they encouraged her to attend a local col- of Isabel’s house. They have two children, lege. The family never experienced poverty Marisa and Ricardo. Unlike Isabel, Irene was during her childhood. not born in the state of São Paulo. She was At age 50, Isabel is a kindergarten teacher born in 1971, in Santa Quitéria, in the state at a small, private day care center in her of Maranhão in the Brazilian Northeast. own neighborhood. Roberto is a sales man- Her father had not finished primary school, ager at a local agricultural machinery store and her mother had no formal schooling at 15 16 INTRODUCTION all. They moved to western São Paulo state gone up roughly in line with the international in the mid-1970s, when Irene was a little price of sugar (although she was probably too girl, as demand for agricultural labor rose busy to notice the correlation). in the booming Southeast. Here, Irene was Now that Marisa has promised to stay sent to public school and completed all eight on at school next year, the only nagging years of primary school. At age 39, she is a worry that keeps Irene awake at night is the part-time waitress at a local snack bar. She fear that her boss at the bar might one day does not have a carteira de trabalho (formal choose to replace her with a younger, more work papers) and earns just over US$200 per energetic worker. The family’s modest relative month. bliss would not survive such a shock. Vulner- Fabiano works as a farmhand at one of the ability, rather than extreme poverty, is their region’s large sugar cane engenhos (planta- bane nowadays. The waitress’s family has a tions). For most of his working life, he, too, daily per capita income of US$6 at PPP—one- was an informal worker and, when he started eighth that of Isabel (the kindergarten teacher) out, they still cut the cane by hand. He has and her family but 50 percent more than the since been taught to drive the tractors and international per capita poverty line of US$4 operate other farm machinery and, in 2007, per day that is often applied to Latin America the plantation owners decided to regular- and the Caribbean (World Bank 2011). ize most of their labor force. Fabiano now Although Irene cannot afford it, Isabel earns two minimum wages and is entitled does have her nails done regularly, at a lit- to paid holidays and a minimum pension. 2 tle manicure shop near the day care center. Though the family made much less money Sônia is her favorite manicurist, and they see in 1995, things have steadily improved since each other often enough that Isabel knows then. Marisa, now 15, attends the local pub- her life story. Sônia was born in 1978 and lic school, and the plan is that she will start graduated from high school at age 18. She secondary school next year—the first person had her only child, a boy named Pedro, ever to do so in her family. Putting together shortly thereafter but was never married and the two minimum wages plus Irene’s income raises him on her own. Pedro was named at the snack bar, the family earns US$730 after his maternal grandfather, a construc- per month, for an annual family income of tion worker who had worked hard all his life US$8,760. and whose meager earnings had been barely As Irene and Fabiano are instinctively enough to support his four children and their aware, this is not enough to earn them a stay-at-home mom. place in the new Brazilian middle class, of With little child care available, times were which they have recently been hearing more tough for Sônia during her twenties. Leaving and more on the evening news. But when the boy with his grandmother, she worked Irene thinks back to what she remembers of mostly as a shop assistant, and the minimum her childhood in Maranhão, with no elec- wage she earned was just enough to keep her tricity in the house and occasionally not and her son above the poverty line. As Pedro enough food to go around for her and her grew older and enrolled in school, Sônia’s three brothers and sisters, she acknowledges hard work and dedication paid off. Noticing that things have improved. If her childhood the number of new beauty parlors opening to in Maranhão—or, for that matter, those of serve the growing number of affluent ladies some of her nephews and nieces who still live in Presidente Prudente, she took an evening “up north” today—characterized poverty, course in manicure and pedicure and landed then her own family is no longer poor. They the job she currently holds. Between the small are not quite middle-class yet, but they have monthly contribution the courts force Pedro’s escaped real poverty. The past few years have father to pay and her earnings at the salon, been good in Quatá. Fabiano is proud of his Sônia takes home just over US$709 PPP per official labor papers, and Irene’s tips have month, for a total annual family income of INTRODUCTION 17 US$9,490. Because hers is only a two-person southeastern Brazil, typify three of the four family, this implies a daily per capita income broad groups that account for the bulk of of US$13—more than twice what waitress Latin America’s population in terms of their Irene and her family live on but just over a economic mobility during the 2000s. Two quarter of the equivalent figure for her cli- broad groups of “climbers” are illustrated ent, Isabel. On the basis of these numbers, by the waitress-farmworker couple, who although Sônia was not middle-class in the move from poverty to vulnerability, and by late 1990s, she is now. the manicurist and her son, who move from vulnerability to the middle class. The kin- dergarten teacher-sales manager couple and Latin American “climbers” their daughter illustrate one of the two main and “stayers” groups of “stayers”: those who remained Although these three families are imaginary, in the middle class throughout the period. they are also eminently plausible in the sense Finally, the fourth main broad family type, that their characteristics (incomes, fam- which we have not named, would have ily sizes, educational attainment levels, and started out poor and stayed poor throughout occupations) are in line with “typical” fami- the period. In a sense, this last, least-happy lies in their broad social groups.3 Their sto- story is that of Irene’s brothers and sisters ries were chosen to illustrate both the broad who stayed up in Maranhão. Although still theme of this volume—the complex relation- poor, even they would have had some prog- ship between economic mobility and the rise ress to report. They would probably be of the Latin American middle class—and the receiving the “Bolsa Família” benefit now, nuances and apparent contradictions that a social welfare program reaching the less surround that relationship. favored in Brazil, and outright hunger (not All three families were better off in 2010 uncommon in their parents’ day) would now than their parents had been 30 to 40 years be unlikely. prior, but their family backgrounds still pow- The three families also exemplify some erfully shape where they are today: one could of the themes that will recur throughout the scarcely imagine a child of Irene’s parents volume: all three are very small families, living like Isabel and Roberto (the kindergar- by historical standards, but quite typical of ten teacher and sales manager) or vice versa. their social groups in Brazil today.4 Related All three families have also experienced real to falling fertility and family size, and to ris- income gains over the past decade, enabling ing levels of educational attainment, is a tale the richer family (Isabel and Roberto) to of increasing labor force participation among stay comfortably atop the country’s expand- women: none of our three leading ladies ing middle class, and Sônia and her son to chose to stay at home with their children, enter it from below. It has also seen Irene and as Sônia’s mother had. Occupational shifts Fabiano, the striving Quatá couple, cross that (including a movement toward working with imaginary threshold (the poverty line) that more capital and technology within agricul- separates the poor from the nonpoor. They ture) and growing access to formal employ- did not jump miraculously from poverty into ment have also been important. Even the link the middle class—a feat that, as we shall see, between these larger employment opportu- relatively few people across the continent nities and local economic growth—possibly have achieved. Instead, Fabiano and Irene fueled in part by higher commodity prices— now inhabit an intermediate group defined is not entirely fictional. primarily by its vulnerability. They may not As these illustrative stories suggest, eco- know it, but they are very close to the mode nomic mobility and the resulting transforma- of the Latin American income distribution. tions in the size and composition of social As this volume will show, these imaginary classes are a complex set of subjects. There is families living in western São Paulo state, in change, both across generations and within 18 INTRODUCTION them. There are thresholds defining the standards: by averaging 4.2 percent annual poor and the middle class. There are causes growth in per capita incomes, average GDP and correlates of various changes, both at a per person in Peru rose from US$5,543 PPP microeconomic household level and at a more in 2000 to US$8,555 PPP in 2010. But in the aggregate level. And these changes may well 2000s, to sustained economic growth was have implications for the future. added another—even rarer—achievement: Before we can make any headway in under- in most Latin American countries for which standing these issues, we need to remind our- data are available, income inequality fell. The selves of the backdrop to the social change of Gini coefficient, one of the most commonly the 2000s and state the precise questions that used measures of inequality, declined in 12 of we seek to answer in this volume. That back- the 15 countries reported in figure 1.2. These drop is the object of the next section. Finally, declines were all statistically significant, and in “Pursuing the Questions,” we summarize their magnitude was not trivial: The average the main objectives of the report in terms of total decline between 2000 and 2010 in the the questions we are about to ask. 12 countries where inequality fell was 5 Gini points. Across all 15 countries for which data are available, the decline was 3.5 Gini points. The broad context Inequality fell in the three largest econo- Economic growth in the Presidente Prudente mies in the region—Brazil, Mexico, and area during 2000–10 led to better conditions Argentina—by 5, 7, and 6 points of the Gini, in Fabiano’s farm job and opened up new respectively. As Barros et al. (2010) note for work opportunities for Sônia, the manicur- the case of Brazil, this process was the result ist. It was one important reason why the past of sustained German-like growth rates for decade was good for our three families. the top tenth of the income distribution, Moving from the illustrative to the sta- combined with Chinese-like growth rates tistical, the 2000s were also a good decade for the bottom tenth, over a 10-year period.6 for Latin America more broadly. Despite It is difficult to overstate the importance of the onset of the global financial crisis in late this achievement: ever since household sur- 2008, the continent’s gross domestic prod- vey data became available in the 1970s (and, uct (GDP) per capita grew at an average rate other data suggests, since long before), Latin of 2.2 percent per year between 2000 and America has been the world’s most unequal 2010.5 In six countries (including Argentina, region, rivaled only by certain countries in the Dominican Republic, Panama, and Peru), Sub-Saharan Africa. annual growth rates in excess of 3 percent per That remains true today: apart from South capita per year accumulated throughout the Africa and Swaziland, all other countries period. Although this is by no means a stel- with the 10 highest Gini coefficients available lar growth performance if compared with the for 2008–10 in the World Development Indi- likes of China (averaging 9.6 percent annual cators (WDI) database are in Latin America.7 growth in the same period) and India (aver- But whereas periods of stability and rising aging 5.7 percent) or with that of the East inequality had historically alternated, there Asian tigers in the 1990s, it represents a con- is no record of a previous period of similarly siderable improvement over the region’s own broad-based and sustained decline in income average annual growth rates in the 1980s disparities in the region. The same WDI (−0.2 percent) and 1990s (1.2 percent). Fig- database lists six Latin American countries ure 1.1 shows the distribution of annual GDP among those with the 10 largest drops in the per capita growth rates for the 32 countries Gini coefficient between 2000 and 2010.8 in the Latin America and Caribbean region, The picture of inequality dynamics in Latin as well as the region’s simple and population- America during 2000–10 evokes contradic- weighted averages. tory emotions: the levels remain unaccept- It is easy to forget the transformative power ably high, but the changes are undeniable and of compound economic growth to raise living point in the right direction. INTRODUCTION 19 FIGURE 1.1 Average annual per capita GDP growth in Latin America and the Caribbean, 2000–10 6 5 4 3 Percent 2 1 0 –1 –2 ge e a m os a go ica Pan ru e a rin lic e G cu a na or Ch s lo ile Br a ru il eig Co min y ht sta ica av ica nd e ht ar s av ua Bo ge Ve G livia ela a Gr , RB El Be a l e an rag or Ba ay St uda Gu ex a at ico itt Jam ala nd ica Ba rb is he iti ne ig Nic ura Do gua az n R am th E ntin bi zu an ad M ci Ho rag Sa liz Ba Nev Ar am Ha re ad ua Pa vad Pe Su pub d u ed ag a ,T ba ed R u sa a em ha ad m di ne uy en er rb .L e as To o U d C an ad in .K id we m d St tig nw in an Do un An Tr io nt an lat ce be u in p ib .V po ar St eC n ea th bb d i ar an eC ia er th Am d an tin ia La er Am tin La Source: World Development Indicators, World Bank. Note: GDP measured using purchasing power parity exchange rates. a. Fiscal year 2009. Naturally, the combination of sustained substantially. Even the 1990s were marked (even if still largely unspectacular) economic largely by stagnation, with the poverty head- growth with reductions in inequality resulted count ending the decade at 43 percent in in substantial drops in absolute poverty. 1999. There was something really different The incidence of moderate poverty in Latin about the 2000s, and figure 1.3 clearly illus- America fell from 41.4 percent in 2000 to trates the structural break in Latin American 28.0 percent in 2010 despite the global finan- poverty reduction trends around the turn of cial crisis in the last two years of the decade the millennium. The year 2003 stands out as (World Bank 2011). This decline in poverty a break in the series. implies that 50 million fewer Latin Ameri- cans were living in poverty in 2010 than 10 years earlier.9 If the comparison is made with Pursuing the questions 2003 instead, the decline in absolute numbers When poverty is reduced by almost a third is even larger: 75 million. in 10 years—or by 13 percentage points of This result contrasts with significantly the population—it is likely that real social worse performances during the previous change is taking place. What is happening to two decades. In the lost decade of the 1980s, all these people who are no longer officially moderate poverty in Latin America rose poor? Have they all joined the emerging 20 INTRODUCTION FIGURE 1.2 Change in the Gini index, selected Latin American countries, 2000–10 6 4 2 Gini coefficient points 0 –2 –4 –6 –8 –10 ru ico r a lic il ay r a a ia ile as y ca e e do do ua az in m bi ag ag liv Pe ur ub Ri gu Ch ex nt m na Br ua lva ug er er Bo nd sta ep ra lo ge M Pa av av Ec Ur Sa Ho Pa Co nR Co Ar ed ed El ica ht ht ig ig in we we m Do un un es es tri tri un un co co 15 12 Source: Socio-Economic Database for Latin America and the Caribbean (SEDLAC), updating figure 1.2 in López-Calva and Lustig 2010. Note: The Gini Index is a commonly used measure of inequality, and can be defined as half the relative mean difference. A Gini coefficient of zero expresses perfect equality (everyone has exactly the same income), whereas a Gini of one expresses maximal inequality (one person has all the income). FIGURE 1.3 Moderate and extreme poverty in Latin America, 1995–2010 50 11,000 45 10,500 constant 2005 international dollar 40 Poverty headcount, percent 10,000 GDP per capita, PPP, 35 9,500 30 9,000 25 8,500 20 15 8,000 10 7,500 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 Extreme poverty Moderate poverty GDP per capita PPP (constant 2005 international dollar) Source: World Bank 2011. Note: GDP = gross domestic product. PPP = purchasing power parity. INTRODUCTION 21 Latin American middle class, of which one does the new Latin American middle class often hears? How does one define that middle look like? class? Does it begin where poverty ends, at • Finally, what are the implications of the poverty line of US$4 per day? Where does these social changes—and of the rise of “middle class” end and the “elite” begin? Do the middle class in particular—for Latin these declines in poverty and inequality, com- America’s social contract in the future? Is bined with a growing middle class, mean that there any evidence that the middle class Latin America has become a more mobile set holds special values or beliefs that may of societies? How can that mobility be prop- lead to greater stability, better policy erly measured, and what does it presage for making, and faster economic progress? the region’s future? Can it be anchored to a cohesive social This volume is motivated by those ques- contract, under which taxes are paid and tions—questions about recent social change quality public services demanded and pro- in Latin America in a context of reasonably vided, so that those who have been left paced growth, a decline in economic inequal- behind retain a chance of completing their ity, and the resulting poverty reduction and own journeys out of poverty? middle-class growth. In particular, we ask the following: The volume is structured around those six groups of questions. Chapter 2, “Economic • Among the myriad concepts, domains, Mobility and the Middle Class: Concepts and measures of economic mobility avail- and Measurement,” addresses the concep- able in the literature, which are most tual issues and definitional questions. It appropriate to gauge the extent of mobil- discusses different concepts of mobility and ity in Latin America in the 2000s? different definitions of the middle class, and • Similarly, how is the middle class—a com- it lays out the arguments underpinning the plex concept, treated differently by differ- choice of concepts and measures used in the ent disciplines and heterogeneously within subsequent chapters. Although chapter 2 is them—best defined for this continent, and important for a full understanding of our at this point in its history? approach, it is also somewhat dryer and • When one looks across generations, how more technical than the rest of the report, mobile is Latin America today? How and some readers may prefer to skip it. important are one’s parents in determin- Chapter 3, “Mobility across Generations,” ing one’s chances of success, even in these reviews the evidence on intergenerational promising times? Is there any evidence of mobility in Latin America—some from the change in those patterns of intergenera- existing literature and some drawing on orig- tional persistence? inal work. • And how much mobility is there within Chapter 4, “Mobility within Genera- individual lifetimes? Behind headline tions,” relies on synthetic panel techniques to numbers for economic growth, falling shed light on the extent of mobility—viewed poverty, and growing middle classes, primarily as income growth for individual individual Latin American families must households—within lifetimes. It pays par- be experiencing economic progress. Who ticular attention to movements across classes: are they? How many are moving across out of poverty, into the middle class, and class barriers? How much vulnerability to in and out of the vulnerable group that lies reversals of fortune remains? between. • Once agreement is reached on a definition Chapter 5, “The Rising Latin American of the middle class, how has it evolved over and Caribbean Middle Class,” discusses the past decade or so? How have its size quantitative trends in the size and composi- and composition changed, and have these tion of the Latin American middle class and changes differed across countries? What presents its current profile. 22 INTRODUCTION Chapter 6, “The Middle Class and the (0.51). Note that these are the latest numbers Social Contract in Latin America,” dis- available for each country during 2008–10 cusses some international evidence on the and do not account for subsequent changes. association between middle-class size and 8. The other four countries in that list are Arme- nia, Côte d’Ivoire, Kazakhstan, and Moldova. policy choices, and asks whether those asso- 9. Moderate poverty is defined by comparing ciations are likely to hold in Latin America. household income per capita with a poverty Survey evidence on the values and beliefs line of US$4 per day at PPP exchange rates. held by individuals in different social groups Extreme poverty, which is defined with is brought to bear, and the implications for respect to a poverty line of US$2.50 per day the future of the region’s social contract are for Latin America, fell from 24.5 percent to examined. 14.0 percent in the same period (see World Bank 2011). Notes 1. If they were real people, Isabel and Roberto References would naturally think of their income in Barros, Ricardo P., M. Carvalho, S. Franco, and Brazilian reais. However, because they are R. Mendonça. 2010. “Markets, the State, fictional characters in this technical volume and the Dynamics of Inequality in Brazil.” about economic mobility and the middle class In Declining Inequality in Latin America: A in Latin America, we use PPP dollars to com- Decade of Progress?, ed. Luis F. López-Calva pare incomes across space and time. and Nora Lustig, 134–174. Washington, DC: 2. The Brazilian minimum wage was R$510.00 Brookings Institution Press. per month in 2010, equivalent to US$324.60 Ferreira, Francisco H. G., Phillippe Leite, and at purchasing power parity. Julie Litchfield. 2008. “The Rise and Fall of 3. Most of these characteristics are further Brazilian Inequality: 1981–2004.” Macroeco- described in tables in chapters 4 and 5. nomic Dynamics 12 (S2): 199–230. 4. The families might be just a little larger in López-Calva, Luis F., and Nora Lustig, eds. 2010. some other countries in the region. Declining Inequality in Latin America: A 5. These are population-weighted averages across Decade of Progress? Washington, DC: Brook- the 30 Latin American and Caribbean coun- ings Institution Press. tries shown in figure 1.1. GDP is measured in World Bank. 2011. “On the Edge of Uncertainty: constant 2005 international dollars, at PPP Poverty Reduction in Latin America and the exchange rates. Caribbean during the Great Recession and 6. See also Ferreira, Leite, and Litchfield (2008) Beyond.” Poverty and labor brief, World Bank, for an early account of declining inequality in Washington, DC. Brazil. World Bank. n.d. World Development Indica- 7. They are Honduras (0.57); Bolivia and Colom- tors. Online database. Washington, DC: World bia (0.56 each); Brazil (0.55); Chile, Panama, Bank. http://data.worldbank.org/data-catalog/ and Paraguay (0.52 each); and Costa Rica world-development-indicators. 2 Economic Mobility and the Middle Class: Concepts and Measurement S ocial and economic mobility are integral exploitation were even greater. In a famous components of economic development. paper, Hirschman and Rothschild (1973) sug- At its most basic, income growth is itself gested that the relationship between mobility a form of economic mobility (as we shall and political reaction might be more compli- discuss further below). But the nature and cated: the upward movement of others, when extent of mobility arguably affect societies in unaccompanied by one’s own, might be wel- ways that are both more subtle and more dra- come at first but subsequently resented (much matic than income growth. as the faster movement of cars in the lane Economic mobility, in the sense of changes next to one’s own in a tunnel). The relation- in relative rank and position, has often been ship between economic mobility, individual said to mitigate static inequality and contrib- preferences for redistribution, and economic ute to long-term fairness, as eloquently sug- outcomes remains of crucial interest to econ- gested by Milton Friedman (1962, p. 171): omists today (see, for example, Piketty 1995). “Consider two societies that have the same Mobility is also often seen as the ex post distribution of annual income. In one there realization of evenly spread economic oppor- is great mobility and change so that the posi- tunity ex ante, and some countries hold up tion of particular families in the income hier- that manifestation as an integral part of their archy varies widely from year to year. In the very national identity. A recent U.S. article other, there is great rigidity so that each fam- on “What Happened to Upward Mobility?” ily stays in the same position year after year. (Time 2011) opened thus: “America’s story, Clearly, in any meaningful sense, the second our national mythology, is built on the idea of would be the more unequal society.” being an opportunity society. From the tales Mobility also affects politics. Tocqueville of Horatio Alger to the real lives of Henry ([1856] 1986) argued that greater opportu- Ford and Mark Zuckerberg, we have defined nity for upward movement among the poor our country as a place where everyone, if he might make long-standing oppression and or she works hard enough, can get ahead.” inequality less, rather than more, tolerable There is also a long-standing and wide- and that this may explain why revolution spread view, going back at least to the Greek took place in France earlier than in other philosophers, that a robust middle class can European countries, where oppression and play a role beneficial for both economic and 23 24 E C O N O M I C M O B I L I T Y A N D T H E R I S E O F T H E L AT I N A M E R I C A N M I D D L E C L A S S : C O N C E P T S A N D M E A S U R E M E N T political development: “It is possible for those feature occasionally in subsequent chapters, states to be well governed that are of the kind the report analyzes mobility and class dynam- in which the middle-class is numerous, and ics primarily through an economic lens. preferably stronger than both the other two Accordingly, this chapter also focuses on the classes, or at all events than one of them, views of mobility and the middle class promi- for by throwing in its weight it sways the nent in the economics literature. Yet it is not balance and prevents the opposite extremes intended as a comprehensive review of that lit- from coming into existence . . . Surely the erature, which is not the remit of this volume. ideal of the state is to consist as much as pos- Instead, it seeks to offer a nontechnical over- sible of persons that are equal and alike, and view of the multiplicity of definitions currently this similarity is most found in the middle in use and to present and justify the definitions classes” (Aristotle [c. 350 BC] 1932). and approaches chosen for use in the analysis Scratch the surface of these grand state- that follows in subsequent chapters. ments, however, and it quickly becomes The chapter is structured as follows: The apparent that both concepts—economic next section discusses the “Spaces, Domains, mobility and the middle class—mean very and Concepts of Economic Mobility” and different things to different people. Mobility highlights those featured in chapters 3 and 4. has long been important in sociology and, at “Defining the Middle Class” turns to alter- least since the late 1970s, it has also been the native definitions of the middle class and subject of formal economic analysis. At the describes how we chose one particular defini- risk of considerable oversimplification, soci- tion for application to Latin America in the ologists may be said to see social mobility in early 21st century, particularly in chapters terms of changes in the class and occupational 5 and 6. The final section, “Linking Mobil- makeup of populations over time, largely as a ity and Middle-Class Dynamics,” illustrates result of technological and economic change. how one of our chosen measures of economic Economists, on the other hand, tend to think mobility is straightforwardly decomposed of mobility in terms of the transformation of into movements in and out of poverty, vul- a vector of incomes (or some other measure nerability, and the middle class. of well-being or economic achievement) in an initial period into another income vector in a second period, and possibly onward to subse- Spaces, domains, and concepts of quent periods.1 And, as we shall see, even this economic mobility apparently narrow economic framework for To impart meaning to the view of mobility as mobility comports with a number of distinct the transformation of a vector into another concepts and measures—distinct enough to over a period of time, economists must be be frequently inconsistent with one another. able to answer three questions quite precisely: The middle class is a similarly slippery, or The first question is: mobility of what? multifaceted, concept. Philosophers, politi- It concerns the space of economic mobil- cal scientists, sociologists, and economists ity and refers to the choice of variable in the have meant different things by the same term. vector (or distribution) under consideration: Before we can meaningfully assess the extent, Is it a vector of current incomes or perma- nature, and possible consequences of eco- nent incomes? Or is it a vector of just labor nomic mobility and the growth of the middle earnings, or perhaps of consumption expen- class in Latin America in the past decade or diture, or wealth? Is it a vector of educa- so, it is important to be clear about what we tional attainment (measured, for instance, by mean by each of these concepts. This chap- completed years of schooling), or of educa- ter therefore offers a brief review of the main tional achievement (measured, say, by perfor- definitions used in the economic analysis of mance on standardized tests)? In attempting mobility and the middle class, and identifies to patch together a comprehensive picture those that will be used in the remainder of of economic mobility in Latin America on the volume. Although sociological approaches the basis of the highly imperfect data that E C O N O M I C M O B I L I T Y A N D T H E R I S E O F T H E L AT I N A M E R I C A N M I D D L E C L A S S : C O N C E P T S A N D M E A S U R E M E N T 25 are available, the chapters that follow will influential taxonomy of Fields (2000), one draw on information from all of the above can identify three basic concepts of mobility, variables, except for physical and fi nancial as follows: wealth. In each instance, the mobility space being considered will be clearly indicated to • Mobility as movement. Informally, this avoid misleading the reader into comparing concept associates mobility with move- apples with oranges.2 In the remainder of this ment: the more movement we observe chapter, however, the term “income vector” between two distributions, the more is shorthand for the vector of interest in any mobile the society. Even this apparently well-defined economic space. simple concept can give rise to very dif- The second question concerns how far ferent indices, however, because an ele- apart in time the two (or more) income vec- ment in an income vector can be char- tors are from one another. In particular, one acterized by three different attributes: must distinguish between two very different its income level (y); its income share, domains of economic mobility: the intragen- s(y) (that is, relative to total or mean erational —when the identities of the elements income); and its position or rank in of the income vectors correspond to the the distribution (p = F(y)). In addition, same individuals over time—and the inter- when we consider movement in levels, generational —when those identities refer it matters whether we are interested to the same lineage across generations (that in “gross” movements or flux (so that is, fathers and sons, mothers and daughters, income falls are added to income gains) and so on). The domain distinction is funda- or in “net” or directional movement mental because it interacts powerfully with (so that, for instance, income falls are the different concepts of mobility described subtracted from income gains). So the below. It is far from obvious that the most “mobility as movement” concept really relevant concepts, or the set of key properties consists of four subconcepts: one would like a measure of mobility to have, ⅙ Direc tion al income move me nt are the same for mobility across generations (IMD) seeks to quantify the extent and for mobility over a person’s lifetime. It is of net upward (or downward) move- also possible for a given society, at a particu- ment in individual incomes. lar point in time, to exhibit a great deal of ⅙ Nondirectional income movement mobility within generations while remaining (IMND) seeks to measure the extent rather immobile across them, or vice versa. of gross movement in incomes, or For this reason, evidence on recent intergen- flux. erational and intragenerational mobility in ⅙ Share movement (SM) seeks to assess Latin America is presented in two separate the extent of movement in relative chapters, respectively chapters 3 and 4. incomes (that is, changes in individ- The third question economists must be ual shares of the overall income pie). able to answer to arrive at a meaningful mea- ⅙ Positional movement (PM) seeks to sure of economic mobility refers to the con- quantify the extent of reranking from cept of mobility one seeks to capture. The one distribution to another. applied literature contains at least 20 differ- • Mobilit y as origin independence ent empirical measures of mobility (Fields (MOI). The basic property underpinning 2010), and the differences are not merely these measures is that a more mobile soci- trivial issues of functional form or relative ety is one where one’s (or one’s parents’) weights. Information on (identity-preserving) initial position is a less important deter- changes between two vectors can be aggre- minant of one’s future position. In a two- gated in very different ways, and these often period setting, mobility as origin (or time) correspond to deep distinctions in the fun- independence can be seen as the converse damental conception of mobility one has in of the correlation between the initial and mind. Drawing on (and slightly adapting) the final vectors. 26 E C O N O M I C M O B I L I T Y A N D T H E R I S E O F T H E L AT I N A M E R I C A N M I D D L E C L A S S : C O N C E P T S A N D M E A S U R E M E N T • Mobility as equalizer of long-term correlated, so there is zero MOI. And for any incomes (ELTI). In this view, a more measure of relative inequality—including all mobile society is one in which inequality those that are Lorenz consistent—inequality in permanent incomes (where permanent in average incomes (1.5, 15, 150) is identical income is defi ned, say, as an individual’s to inequality in the initial and final vectors. average income across all periods) is less So there is no mobility as an ELTI. than the inequality at any particular point Scenario A illustrates the point that the in time (or, in an alternative specification, concepts of income movement are essen- at the initial period). tially different from the other mobility concepts. Naturally, measures of mobil- These different concepts of mobility corre- ity designed to capture share or positional spond to inherently distinct notions of what movement, or indeed MOI or ELTI, cannot mobility is. In general, although many indices be expected to accurately gauge the extent may be consistent with each concept, a good of IMD or IMND in a particular vector measure of one particular concept will be a transformation. poor measure of any other. Which is to say, to As Scenario B illustrates, IMD and IMND choose a particular index to measure mobil- need not always be aligned, either. In this ity, one must first decide which concept of case, a measure of the extent of churning mobility one is trying to capture. To see this, or nondirectional movement in the distribu- consider a simple example, illustrated in table tion (that is, an index where income falls and 2.1. Imagine a three-person economy, where income gains enter with the same sign) would the initial income vector (in pesos per day, say) record a high value, whereas a measure of is (1, 10, 100). Now consider three alternative growth (or directional movement) would “mobility scenarios.” In Scenario A, the final indicate no mobility. In this particular “rank- income vector is (2, 20, 200). In Scenario B, it reversal” example, there is also a large degree is (100, 10, 1); and in Scenario C, it is (36, 37, of PM and SM. And inequality in average 38). Which of these scenarios has the highest incomes (50.5, 10.0, 50.5) is also lower than mobility? And which has the least? in either initial or final points, so some ELTI The answer clearly depends on the mobil- takes place. ity concept. In Scenario A, all incomes have Scenario C is an example of transfor- doubled, so there is a good deal of income mations where SM and PM yield different movement, both directional (IMD) and non- degrees of mobility. In this case, there is posi- directional (IMND). But each person’s share tive IMND and SM, and there is also ELTI, of total income remains unchanged: person but there is no aggregate IMD (at least by 1 has 1/111; person 2 has 10/111, and per- some measures) or PM. son 3 has 100/111 of the total pie in both The broader point that table 2.1 seeks initial and final vectors. So there is no SM to convey is that the distinctions between at all! Similarly, the ranking of individuals the three main concepts of mobility (move- is unchanged between the initial and final ment, origin independence, and the long- vectors, so there is no PM either. Incomes in term equalization of incomes)—or indeed the initial and final vectors are also perfectly between the four subconcepts of mobility as TABLE 2.1 How different mobility concepts rank the same vector transformation Initial income vector Final income vector Mobility No mobility Scenario A: (1, 10, 100) (2, 20, 200) IMD, IMND SM, PM, MOI, ELTI Scenario B: (1, 10, 100) (100, 10, 1) IMND, PM, SM, ELTI IMD, MOI Scenario C: (1, 10, 100) (36, 37, 38) IMND, SM, ELTI IMD, PM Note: ELTI = mobility as equalizer of long-term incomes. IMD = directional income movement. IMND = nondirectional income movement. MOI = mobility as origin independence. PM = positional movement. SM = share movement. E C O N O M I C M O B I L I T Y A N D T H E R I S E O F T H E L AT I N A M E R I C A N M I D D L E C L A S S : C O N C E P T S A N D M E A S U R E M E N T 27 movement—are not small quibbles of lim- notion of equality of opportunity. A mobile ited practical interest. They go to the very society is one in which the children of law- heart of what mobility is. If we are trying to yers or doctors and those of farmers or con- determine whether mobility is higher in Peru struction workers have similar (income or or in Mexico at a certain point in time, or educational) prospects—one where a parent’s to ascertain whether it has grown or been income, occupation, education, or status reduced between the 1990s and the 2000s does not fully or substantially predetermine in Argentina, we might well get different the son’s (or daughter’s). answers, depending on which concept we are The mobility concept most closely associ- interested in and on what particular measure ated with this notion is that of origin inde- of mobility we choose to adopt. For to each pendence.4 The origin-independence axiom of the six concepts discussed above (IMD, (see Shorrocks 1978) requires a measure of IMND, SM, PM, MOI, and ELTI), there mobility to rise when the association between corresponds a number of different specific initial and final vectors falls. In practice, mobility indices. a social mobility index commonly used to For mobility as movement and as origin measure MOI is the complement of the cor- independence, the problem of measuring relation coefficient between initial and final overall mobility in a society can be decom- vectors: posed into two intuitively simple steps: First, one defines an individual mobility function. d(y0, y1) = 1 – ρ01.5 (2.1) Second, one aggregates across all individuals (or lineages) in the economy to obtain a social This measure of mobility as the converse mobility index. 3 Focus note 2.1 (at the end of correlation has a long tradition in statistics of this chapter) provides examples of some of and economics, going back to English stat- the most commonly used individual mobil- istician Sir Francis Galton (1886). Another ity functions corresponding to each concept frequent measure of intergenerational mobil- of mobility and the aggregate indices they ity is the complement of the gradient in the give rise to. It also illustrates each function regression of the final vector on the initial graphically for transformations in the vector one. Classic applications in economics include of household per capita incomes in Peru in Zimmerman (1992) and Solon (2002). This 2004–06. measure may be preferable to the correlation Given the broad scope of this report—cov- coefficient when one is interested not only ering mobility across a good number of coun- in how much parental background explains tries, over a relatively long period, and in dif- the outcomes in the children’s generation ferent spaces (such as income, consumption, but also in how unequal those outcomes are. and education)—attempting to systematically This is the primary measure of mobility that provide results for all mobility subconcepts in will be reported in the next chapter, which every instance seems unwise: a wide variety focuses on intergenerational mobility in Latin of measures for every example would prob- America. ably lead to a bewildering array of numbers, Within a given generation, on the other more likely to obscure the big picture than hand, it is not clear that one would prefer a to provide meaningful detail. We have there- society where people’s incomes today had no fore chosen to focus on a single mobility con- correlation at all with their incomes, say, 10 cept for each of our two domains of interest: years ago. The degree of churning and eco- across and within generations. nomic upheaval necessary to engender such As suggested above, however, it is not a (lack of) serial correlation would likely lead obvious that the same concept serves equally to a great deal of intertemporal variation in well to capture the fundamental properties of consumption and well-being (unless credit mobility in these two domains. Across gen- markets were perfect). In addition, there are erations, mobility is often associated with the both ethical and incentive-related reasons 28 E C O N O M I C M O B I L I T Y A N D T H E R I S E O F T H E L AT I N A M E R I C A N M I D D L E C L A S S : C O N C E P T S A N D M E A S U R E M E N T for welcoming a certain degree of temporal For these various reasons, this propor- persistence in the rewards to effort: a society tional measure of directional income move- where the economic benefits from complet- ment is our preferred index of mobility ing a demanding college education dissipated within generations, and it features promi- over a few years would be unlikely to be nently in chapter 4. either efficient or fair. These choices give rise to a simple 2 × 2 Provided there is equality of opportunity matrix of concepts by domains, whereof the (and intergenerational mobility in an origin- report will focus on a main diagonal: when independence sense), one might reasonably assessing the extent of mobility of individu- choose to focus on income growth as the key als within a generation, we will be primar- desideratum of mobility within generations. ily concerned with measures of directional And, as we have seen, the mobility concept income movement, while when measuring that most closely corresponds to the idea of mobility across generations we will focus on growth (at the individual level) is directional mobility as origin independence. income movement. Table 2.2 illustrates this conceptual Among various possible measures of direc- matrix. As it suggests, the focus on the main tional income movement, we chose the index diagonal does not imply absolute silence denoted M 3 in Fields and Ok (1999), which about the off-diagonal cells. It will sometimes is simply the average of the growth rates in make sense to investigate how much growth individual incomes between the initial and or progress took place between generations final vectors. This index is appealing for a in absolute terms, just as it may occasionally number of reasons: be interesting to assess measures of indepen- dence between circumstances at childhood • Intuitively, it captures the microfounda- and achievements at adulthood. In the main, tions of economic growth—at the level of however, mobility between generations will the individual household. be taken to mean mobility as origin inde- • Formally, it is the integral of the non- pendence, whereas individual mobility over anonymous growth incidence curve the course of a number of years will be seen (GIC), which plots income growth rates through the prism of directional income for each percentile of the initial income movement. distribution.6 As noted above, and in the bottom-right • It can be naturally interpreted as a “dem- cell in table 2.2, one attraction of the direc- ocratic” measure of economic growth, tional income movement concept in the which differs from the conventional mea- intragenerational domain is that it imme- sure by weighting households by their diately lends itself to the analysis of social population shares rather than income group dynamics. By means of GICs, or of shares.7 decompositions of our mobility measure, • It can also be decomposed in a number of it allows us to identify individuals leaving informative ways, including one that gen- or entering poverty as well as those arriv- erates a “transition matrix-like” portrait ing at or falling from the middle class. The of transitions into and out of poverty, the demographic, educational, and occupational middle class, and other social groups one characteristics of these individuals; their cares to defi ne. We return to this matrix access to or use of services; and the differ- decomposition in the concluding section, ent policy regimes under which they live and “Linking Mobility and Middle-Class work can be informative of the nature of Dynamics,” where we use it to link this upward mobility and the rise of the middle measure of intragenerational mobility to class—the virtues of which have appealed the definition of the middle class proposed to so many since Aristotle. As the following below. chapters suggest, they may even allow us to E C O N O M I C M O B I L I T Y A N D T H E R I S E O F T H E L AT I N A M E R I C A N M I D D L E C L A S S : C O N C E P T S A N D M E A S U R E M E N T 29 TABLE 2.2 Key mobility concepts and domains under consideration: The main diagonal Concept and Domain Intergenerational Intragenerational Origin Key desideratum is equality of opportunity. When looking at long-term life-cycle move- independence ments (for example, from childhood to adult- Main index is the gradient of regression across hood), concepts of origin independence and generations. equality of opportunity become relevant again. Objects of analysis include the extent to which parental characteristics affect educational achievement today and how policies might interact with that relationship. Directional income Absolute progress across generations, in Key desideratum is individual growth. movement income or education, will also be reported. Main index is the average of the growth rates in individual incomes (M3). Objects of analysis include movements in and out of poverty and in and out of the middle class. formulate and investigate certain hypotheses groups formally defined as “classes.” King about the causes and policy effects of these Servius Tullius of Rome, in a visionary plan changes. to enlarge his kingdom, extended the fran- But to do this, we need a definition of the chise to people outside the traditional limits social classes we are interested in, particu- of Rome and launched the first census of larly the middle class. Beyond that, given the the Western world (McGeough 2004; Cor- available data and the concepts of mobility nell 1995). During the 6th century BC, the we have chosen, it will help enormously if the king gathered demographic and socioeco- “middle class” definition we adopt, wher- nomic information about his subjects. The ever it may come from, can ultimately be purpose of the census was to classify citizens expressed in income terms. into income groups for tax purposes, called classis, and to establish the contributions of each family to the empire, according to their Defining the middle class declared means. Those in the poorest class, In Western civilization, the notion of social who were unable to contribute with financial class, like much else, goes back to Greco- resources, were nevertheless capable of con- Roman antiquity. Aristotle’s Politics, as we tributing by having children, who potentially have already hinted, noted that people with would serve as soldiers. That lowest classis different levels of wealth tended to have dif- would only contribute with prole, the Latin ferent political preferences and interests and word for children, and was named the pro- suggested that there might often be conflict letarius. The notion would evolve into the between the interests of the poor and those modern concept of proletariat, a class whose of the rich. Such conflict might be alleviated only means of production is their own labor by the existence of a large group of people and which is then subject to exploitation by “in the middle,” particularly if such people those who own capital, according to Marx- were “equal and alike” (Aristotle [c. 350 BC] ian analysis. 1932). The notion of class is, of course, central The Romans, often seen as more practical- in Karl Marx’s writings. Marx viewed class minded than the Greeks, are credited with a as being essentially defined by the ownership first operational classification of people into of the same factors of production (chiefly 30 E C O N O M I C M O B I L I T Y A N D T H E R I S E O F T H E L AT I N A M E R I C A N M I D D L E C L A S S : C O N C E P T S A N D M E A S U R E M E N T labor or capital). The factors one sold into the behavior of those who belong to the production process in turn engendered for group, as in Akerlof and Kranton (2002). each group a common position in a stratified Status is not limited by the market. social structure, characterized primarily by According to Weber, both propertied and the exploitation of workers by capitalists. But propertyless people can belong to status he did allow for the existence of a small, inde- groups—although there is a clear overlap pendent group of businessmen and profes- between status and class. The main differ- sionals who acquired skills, knowledge, and ence between both concepts is that while education to rely only on themselves and their classes relate to the production of goods, resources to achieve a better economic posi- “status groups” are stratified according to tion. This embryonic middle class was seen as the consumption of goods as represented a relatively narrow group known as the petty by particular “styles of life” and associ- bourgeoisie, composed largely of small entre- ated principles, values, and ideas.9 preneurs and bureaucrats (as opposed to the • Party, the third dimension of Weber’s haute bourgeoisie, the capitalists). social structure, is related to the notion But as the market economy in industrial of power in social relations. An individual Europe during the late 19th and early 20th holds more power to the degree that he or centuries evolved, and more complex pro- she controls resources that are important cesses in manufacturing and services also to others, inasmuch as this individual can demanded education and skills, a new class induce others to act in his or her own inter- of educated people emerged who did not est. As expressed in Weber’s words, power necessarily own capital and who sold their is “the probability that one actor within a labor in the market. This class did not fit eas- social relationship will be in a position to ily within the classic Marxist framework: it carry out his own will despite resistance, did not belong to the lumpenproletariat or to regardless of the basis on which this prob- the proletariat (the working class) in terms of ability rests” (Weber [1922] 1978). their role in the dynamics of class conflict, but neither did it own capital, as capitalists Together, and in their very different ways, did.8 Marx and Weber can be seen as the found- The more modern, nuanced, and complex ers of the modern sociological approach(es) concept of the middle class that was needed to class and, hence, more specifically to the has evolved, in large part, from the writings middle class. Although, as we have seen, of Max Weber (for example, Weber [1922] both Marx and Weber regarded economic 1978). In the Weberian tradition, the concept interests, and participation in economic pro- of social stratification contains three inter- cesses, as fundamental to the definition of twined notions: class, status, and party (or social classes, they also acknowledged the power, more broadly): importance of other aspects. Weber and his followers, in particular, noted the impor- • Class refers to the strictly economic aspect tance of political organization and collective of stratification. In Weber’s own words, action, patterns of consumption and lifestyle, “the factor that creates ‘class’ is unam- and finally beliefs and a system of ideas. biguously economic interest, and indeed, When Marx and Weber were writing, only those interests involved in the exis- however, household-level data capable of tence of the ‘market’ ” (Weber 1946). identifying individuals into social classes • Status, on the other hand, relates to the were extremely scarce, and the technology “lifestyle” of a group of people, the iden- for manipulating and analyzing them was tity and prestige associated with mem- rudimentary. As household data have become bership, and the expression of such con- more plentiful, and information technology ditions through cultural consumption revolutionized their analysis, it has become (Torche 2010). It relates to the expected possible to take definitions of the middle class E C O N O M I C M O B I L I T Y A N D T H E R I S E O F T H E L AT I N A M E R I C A N M I D D L E C L A S S : C O N C E P T S A N D M E A S U R E M E N T 31 to the data. Economists, who have recently of the middle class (in terms of population) is taken the lead in this process, have seldom naturally fixed by the very definition. These sought to identify the middle class in terms of measures seek, instead, to quantify the share its educational makeup, occupational compo- of total income appropriated by this group. sition, or system of beliefs. Predictably, per- We refer to both of these groups of stud- haps, most studies have opted for an income- ies—those that define the middle-class based definition. thresholds as multiples of median income and Income is a tempting variable on which to those that define thresholds based on certain base criteria for defining the middle class: it income quintiles or deciles—as using rela- provides a natural metric on a single dimen- tive, income-based definitions of the middle sion, facilitating the location of a “middle class. Table 2.3 summarizes the specific cut- group.” Choose two income thresholds, and off points used in some of the key studies in you could call those below the lower thresh- this group. old the “lower class,” those above the higher In comparing middle classes across coun- one the “upper class,” and in between them tries, the relative, income-based defini- you have the middle class, much as in Aris- tions—or at least those among them that rely totle’s quote at the beginning of this chapter. on the median—face the problem of a dif- Albeit stylized, this has been essentially how ferent median income in each country, and the economics literature on the middle class therefore different middle classes from place has evolved. to place. Imagine two countries, A and B, Studies differ from one another largely in with median per capita incomes of US$3 and terms of which two thresholds are chosen. US$8 a day, respectively. If the middle class is A first group of studies select thresholds in defined as those households with per capita relation to the median income of the distri- income ranging between 0.60 and 1.40 times bution. For example, Blackburn and Bloom the median income, a household living on (1985) identify the middle class as house- US$1.8 to US$4.2 a day in country A would holds with per capita income between 0.60 undoubtedly be part of the middle class in and 2.25 times the median income in the that country; however, this household would United States. Davis and Huston (1992) be part of the lower class in country B, where use a narrower range: between 0.50 and the income thresholds range between US$4.8 1.50 times the median, also for the United and US$11.2 a day. States. And Birdsall, Graham, and Pettinato An alternative is using an absolute, (2000) use a range between 0.75 and 1.25 income-based definition, which avoids the times the median for 30 countries, including previous shortcoming because it identifies high-income, transition, and Latin Ameri- the middle class as those households with can economies. Some authors also call these income or consumption in a specific range groups “social strata” precisely to emphasize of standardized international dollars (that is, the narrowly economic nature of the concept, at purchasing power parity [PPP] exchange avoiding the sociological discussion. rates). The fundamental question is how to Another set of studies sets the thresholds define such an absolute level. So far, most not on the income space itself but on the absolute thresholds appear to have been space of ranks or positions in their distribu- picked somewhat arbitrarily. In an influential tion: p = F(y). study, Milanovic and Yitzhaki (2002) divided For example, Alesina and Perotti (1996) the world population into three groups and use the income share of the third and fourth used household surveys to identify the mid- quintiles of the distribution; Partridge (1997) dle class as those households with per cap- uses the middle quintile; Barro (2000) and ita incomes between the average per capita Easterly (2001) use the middle three quin- incomes of Brazil and Italy (US$12–US$50 tiles; and Solimano (2008), the third to ninth a day). Banerjee and Duflo (2008) define the deciles. Under this latter approach, the size middle class as those households living with a 32 E C O N O M I C M O B I L I T Y A N D T H E R I S E O F T H E L AT I N A M E R I C A N M I D D L E C L A S S : C O N C E P T S A N D M E A S U R E M E N T TABLE 2.3 Income-based definitions of the middle class Relative definitions of the middle class Percentiles of the income distribution Birdsall, Graham, and Pettinato (2000) i  middle class 0.75 y ( p50 ) ≤ yi ≤ 1.25 y ( p50 ) Blackburn and Bloom (1985) 0.60 y ( p50 ) ≤ yi ≤ 2.25 y ( p50 ) Davis and Huston (1992) 0.50 y ( p50 ) ≤ yi ≤ 1.50 y ( p50 ) Alesina and Perotti (1996) p40 ≤ p(yi ) ≤ p80 Barro (1999) and Easterly (2001) P20 ≤ p(yi ) ≤ p80 Partridge (1997) p40 ≤ p(yi ) ≤ p60 Solimano (2008) P20 ≤ p(yi ) ≤ p90 Absolute definitions of the middle class Banerjee and Duflo (2008) $2 ≤ yi ≤ $10 a day Kharas (2010) $10 ≤ yi ≤ $100 a day López-Calva and Ortiz-Juarez (2011) $10 ≤ yi ≤ $50 a day Milanovic and Yitzhaki (2002) i  middle class $12 ≤ yi ≤ $50 a day Ravallion (2010) $2 ≤ yi ≤ $13 a day Note: All values expressed in US$ at purchasing power parity exchange rates. per capita expenditure of US$2–US$10 a day the middle class, these authors looked for an and analyze the consumption and employ- income value that corresponds to a minimum ment patterns of this group in 11 developing requirement for the functionings that define countries. the middle-class.10 One might see the inabil- Similarly, Ravallion (2010) recently pro- ity to attain adequate nourishment or to par- posed the concept of a “developing world’s ticipate meaningfully in a minimum set of middle class,” defined as a range between social activities as the (absence of) function- the developing countries’ median poverty line ings that define poverty, and a poverty line as and the U.S. poverty line—in other words, some demarcation in income space of what is the range between (a) those households required to attain those minimum function- with per capita consumption at or above ings and escape poverty, in a particular soci- the median poverty line for 70 developing ety and at a particular time. countries (US$2 a day per person), and (b) Analogously, one might search for the households at or below the U.S. poverty line set of functionings that are associated with (US$13 a day per person). Using household belonging to a middle class and then attempt surveys for almost 100 developing countries, to quantify an income level that permits their Ravallion showed that the developing world’s attainment in a given society at a given time. middle class increased from 32.8 percent of One advantage of this approach is that it the population in 1990 to 48.5 percent of the moves us a little closer, however slightly, to population in 2005. These figures suggest the concept of a common “lifestyle”—includ- that more than 1.2 billion people joined the ing certain consumption patterns and cul- middle class over 1990–2005, with China tural habits—that sociologists in the Webe- accounting for a startling half of this amount. rian tradition associate with class. Sensibly, In an even more recent study, López-Calva López-Calva and Ortiz-Juarez (2011) do and Ortiz-Juarez (2011) also proposed abso- not attempt to fully identify a vector of con- lute income-based thresholds to define the sumption goods associated with middle-class middle class. Like Banerjee and Duflo (2008) status. Instead, they choose one particular and Ravallion (2010), there are elements of “functioning,” namely economic security, an analogy with poverty measurement in as the defining characteristic of the middle how they go about this. But instead of choos- class. And economic security is measured, in ing a particular poverty line (the upper bound turn, as the converse of vulnerability to fall- on the set of the poor) as the lower bound on ing into poverty.11 E C O N O M I C M O B I L I T Y A N D T H E R I S E O F T H E L AT I N A M E R I C A N M I D D L E C L A S S : C O N C E P T S A N D M E A S U R E M E N T 33 Specifically, these authors estimate the FIGURE 2.1 Income-based vulnerability to poverty in Chile, probability of falling into poverty in three Mexico, and Peru in the 2000s Latin American countries for which longi- tudinal household data are available from 0.7 the early 2000s: namely Chile, Mexico, and Peru, conditional on a set of observed covari- 0.6 ates (including demographic indicators, Probability of fall into poverty, labor market resources, and household-level 0.5 below US$4 PPP/day shocks).12 Using these panels, poverty tran- sition matrices are constructed on the basis 0.4 of national poverty lines in each country, all of which are in the PPP US$4–US$5 per day 0.3 range.13 The results are shown in figure 2.1. Figure 2.1 depicts the inverse relation- 0.2 ship between initial incomes (on the hori- zontal axis) and the probability that house- 0.1 holds with those levels of (predicted) income would find themselves in poverty at the end 0 0 5 10 15 20 of the five-year interval in each country. Full economic security may well be thought to Initial per capita daily income, 2005 PPP correspond to a zero, or near-zero, poverty Chile Mexico Peru probability. However, the existence of (some unknown amount of) measurement error in Source: López-Calva and Ortiz-Juarez 2011. any panel survey implies that mobility in any Note: PPP = purchasing power parity. such transition matrix is likely to be overesti- mated and that taking a lower bound for the middle class at the 0–5 percent probability American countries specifically. Like pov- range may well be excessively conservative. erty, the notion of the middle class might be López-Calva and Ortiz-Juarez (2011) sug- absolute in some capability space but not in gest taking a 10 percent probability of falling income space. National poverty lines, which into poverty as an “operational” dividing line are used to inform national policy decisions, between economic security and vulnerability. vary with aggregate incomes and are not the As discussed in chapter 4, that probability is same in Argentina, China, and India (see slightly lower than the average upper-bound Chen and Ravallion 2001). Similarly, a lower estimate for downward mobility into poverty threshold of US$2 per day, as proposed by in the region as a whole and very close to Banerjee and Duflo (2008) and Ravallion the country estimates for Argentina, Colom- (2010), might (or might not) make sense in bia, and Costa Rica. The choice of a 10 per- the poorest countries, but it is unsuitable in cent probability of falling into poverty in a a Latin American context. It is well below five-year interval yields income thresholds most national poverty lines in the region, of PPP US$8.5 per capita per day in Chile, for example. Both the conceptual basis and PPP US$9.7 in Mexico, and PPP US$9.6 in the regional specificity of the López-Calva Peru. The authors furthermore present some and Ortiz-Juarez (2011) lower threshold are evidence that these thresholds are relatively therefore appealing for the purposes of this robust to changes in the specification of their report. conditioning models. One concern may remain, however, The anchoring of a middle-class defini- regarding the possibly arbitrary choice of a tion to economic security is conceptually 10 percent probability of falling into pov- appealing, as is the fact that these authors erty in a five-year interval as the dividing have applied their proposal to three Latin line between security and vulnerability, from 34 E C O N O M I C M O B I L I T Y A N D T H E R I S E O F T H E L AT I N A M E R I C A N M I D D L E C L A S S : C O N C E P T S A N D M E A S U R E M E N T which the monetary threshold follows.14 We countries was 18.5 percent and 0.8 percent, therefore asked whether an alternative, and respectively. One implication of this is that completely independent, approach to defin- a grouping of the three “middle-class” cat- ing the middle class would yield a very differ- egories dominates over the two extremes at ent lower threshold. Specifically, we adopted all income ranges, providing no meaningful a subjective approach based on self-reported insight. class membership. The idea—somewhat One obvious alternative is to consider the analogous to the Leyden school of subjec- lower and lower-middle categories jointly as tive poverty measurement—was to look for a group below the middle class, and the three the lowest income level around which more upper categories as a joint “middle class and people regard themselves as middle class than the elite” amalgamation. This approach will as poor or “lower class.”15 not help us to shed light on the upper thresh- The best set of nationally representative old of the middle class, to which we return household surveys that contain a question below, but it may help us understand where on social class as well as some objective mea- Latin Americans themselves perceive the sure of socioeconomic status for a number of lower bound of the middle class. In five of Latin American countries are the Encuestas the seven Ecosocial countries for which the de Cohesión Social en América Latina (Eco- analysis was possible, we therefore plotted social), fielded by the Corporación de Estu- the density functions of the income distribu- dios para Latinoamérica (CIEPLAN), an tion of all those who considered themselves influential Chilean think tank. In particular, as lower or lower-middle class, and sepa- we used the 2007 wave for seven countries, rately the densities of those who considered namely Argentina, Brazil, Chile, Colombia, themselves as “middle or upper class.” Fig- Guatemala, Mexico, and Peru. The Ecosocial ure 2.2 illustrates the results for Mexico. surveys do not ask individuals about actual Our proposed “subjective approach” would household income, but reasonably detailed treat the income at which the two functions information is available on a set of assets, cross—that is, the lowest income at which durable goods, and dwelling characteristics. more people see themselves as middle class This permits the application of a survey-to- than otherwise—as the lower threshold for survey income imputation method based on the middle class. the poverty mapping work of Elbers, Lan- For Mexico, this income level is PPP jouw, and Lanjouw (2003). US$9.6 per capita per day—remarkably close Once incomes are thus imputed from the to the US$9.7 line obtained by the vulnerabil- mainstream household surveys into the Eco- ity approach of López-Calva and Ortiz-Juarez social, we can observe, for each household, (2011), which was reported above. In Peru, it both a measure of “predicted income” and a is PPP US$10.5, also not far from the US$9.6 class self-report. The latter records the answer line yielded by the vulnerability approach. Of to the following question: “In our society, course, as is to be expected from the applica- people tend to place themselves within differ- tion of two completely different approaches ent social classes. Would you classify yourself such as these to a question as tenuous as the as belonging to one of these?”16 The ques- membership of the middle class, the two tion is asked identically in all seven countries. answers do not always coincide.17 Table 2.4 Answers fall into five categories: lower class, presents the lower middle-class thresholds lower middle class, middle class, upper mid- obtained from the subjective approach for dle class, and upper class. Perhaps unsurpris- all five countries where the exercise was pos- ingly, relatively few respondents self-describe sible, namely Brazil, Chile, Colombia, Mex- in the extreme categories, particularly the ico, and Peru. The lines are presented both upper class: the average density in the lower- in household per capita income terms and in and upper-class categories across the seven terms of income per earner; and, in each case, E C O N O M I C M O B I L I T Y A N D T H E R I S E O F T H E L AT I N A M E R I C A N M I D D L E C L A S S : C O N C E P T S A N D M E A S U R E M E N T 35 FIGURE 2.2 Distribution of self-reported class TABLE 2.4 Middle-class thresholds from self-reported class status in Mexico, 2007 status, selected Latin American countries, 2007 Lower threshold Lower threshold (per capita income) (income per earner) .04 US$ PPP Percentile US$ PPP Percentile Brazil 16.3 84 26.2 82 Chile 20.3 83 33.2 77 .03 Colombia 9.3 69 17.1 57 Mexico 9.6 68 19.9 66 Density Peru 10.5 76 18.1 74 .02 Sources: Ecosocial 2007 and SEDLAC harmonized data. Note: PPP = purchasing power parity. SEDLAC = Socioeconomic Database for Latin America and the Caribbean, jointly managed and maintained by the Centro de Estudios Distributivos, Labo- rales y Sociales (CEDLAS) of the Universidad de la Plata in Argentina, and the World Bank. .01 0 inevitably associated with the two proce- 0 20 40 60 80 dures, and the various imputations and esti- Predicted per capita mations that are carried out in each, we take daily income in US$ PPP these numbers as indicators of a broad order Middle to upper of magnitude rather than as precise point esti- Lower to lower middle mates. We do draw comfort from the fact that two conceptually appealing approaches— Sources: Encuestas de Cohesión Social en América Latina (Ecosocial) by the one based on the attainment of an objective Corporación de Estudios para Latinoamérica (CIEPLAN) 2007; Encuesta Nacional de Ingreso y Gastos de los Hogares 2008 by the Mexican Instituto functioning (economic security) and another Nacional de Estadística y Geografía (INEGI). based on self-perceptions of class—yield very Note: Densities are weighted by class size. PPP = purchasing power parity. similar lower bounds for the middle class in Latin America. But we do not read a great deal into decimal points and are happy to fol- the rank of that income in the correspond- low the recommendation in López-Calva and ing distribution is given. The latter are rather Ortiz-Juarez (2011) of adopting PPP US$10 similar across the two distributions (except in per capita per day as our operational lower the case of Colombia), suggesting robustness bound for the Latin American middle class in across income normalizations. the chapters that follow. In our preferred income concept, namely household per capita income, these subjective Determining the upper income thresholds range from US$9.3 (in Colombia) threshold to US$20.3 (in Chile) at PPP exchange rates. Though the number for Chile, in particular, In principle, of course, the upper bound of indicates that the subjective approach can the middle class should matter as much as generate lower bounds that are considerably the lower bound. Yet inspection of table 2.4 higher than those obtained from the vul- suggests that self-reported middle-class status nerability approach, the exercise also sug- is already associated with people fairly high gests that the lower envelope of subjective up in the continent’s income distribution. lower thresholds in our sample (of around As we will see in much more detail in chap- PPP US$9–US$10 per day) is remarkably ter 5 (which describes the nature, profile, close to the lines yielded by the vulnerability and trends of the regional middle class), our approach. US$10 line falls approximately on the seventh Given the scope for measurement error decile of the continent’s overall distribution, and imprecision of various kinds that are as obtained from the household surveys in 36 E C O N O M I C M O B I L I T Y A N D T H E R I S E O F T H E L AT I N A M E R I C A N M I D D L E C L A S S : C O N C E P T S A N D M E A S U R E M E N T SEDLAC.18 If one were to apply, for example, The practical implication is that the analy- Kharas’s (2010) suggested upper threshold sis of the nature and evolution of the Latin for the middle class of US$100 per capita American middle class in this volume will per day to that household-survey-based mix- perforce remain based on household survey ture of distributions, only 0.5 percent of the data. Given the uncertainty surrounding continent’s population would be counted as survey representativeness at the very top, being the elite, or “above the middle class.”19 two implications would seem to follow, in Naturally, this 0.5 percent figure reflects the turn, for the choice of an upper middle-class well-known—and severe—shortcomings threshold: First, less attention should be paid of household survey representativeness at to it, and less confidence placed on it, than on the top of the income distribution. Income the lower threshold. Second, it may be prefer- underreporting by richer households in sur- able to err on the side of a lower threshold so veys of this kind has long been known to be that a reasonable number of observations are a common problem. More important still are left above it, even in the poorer countries of the effects of survey noncompliance at the the region. top end of the distribution.20 With those considerations in mind, we Although the existence of these survey follow López-Calva and Ortiz-Juarez (2011) problems is widely acknowledged, estimates here, too, and adopt a PPP US$50 per capita of their extent in each country or methods to per day upper bound for the Latin American correct for them are much scarcer. In richer middle class. This is precisely half the line countries, a lot of work on the distribution suggested by Kharas (2010), and it leaves 2.2 of top incomes has recently relied on anony- percent of the (survey-based) Latin American mized tax record data (see, for example, Pik- population in the “elite,” rather than 0.5 per- etty and Saez 2003; Atkinson, Piketty, and cent. But the analysis that follows will place Saez 2011). A great deal has been learned much less emphasis on the class divide at from this approach, and it is encouraging the top than it will on the bottom threshold. that similar methods are now being applied That analysis will be concentrated in chap- to Latin America (Alvaredo 2010). Neverthe- ters 5 and 6, which focus, respectively, on a less, it is not clear that this work has evolved description of the size, nature, and evolution sufficiently to help us define a realistic upper of the middle class and on the implications threshold for the middle class in Latin Amer- for economic policy. ica at this time. This is for two reasons: First, the tax record data that are needed for ana- Distribution of four economic classes in lyzing top incomes have only recently been Latin America made available to researchers.21 Second, most of the top-income analysis so far has focused Our middle-class thresholds—PPP US$10 exclusively on the analysis of tax record data, and PPP US$50 per capita per day—are which, in the Latin American case, would shown in figure 2.3, which depicts the density clearly be inferior to household survey data function for the Latin America-wide income for the lower (and possible middle) income distribution. This continental distribution ranges, given the narrow coverage of the was constructed from the Socioeconomic income tax in most of the region as well as Database for Latin America and the Carib- problems of tax avoidance and evasion. Com- bean (SEDLAC) data set mentioned above bining tax record data and household sur- and represents 500 million individuals from vey data—although a promising avenue of 15 countries, or 86 percent of the region’s research for truly understanding the income population. We will return to the continen- distribution of middle-income countries— tal distribution in chapter 5, but it is included remains a frontier issue on which, to our here so that the income thresholds derived knowledge, little progress has been made. in this section can be pictured in context. In E C O N O M I C M O B I L I T Y A N D T H E R I S E O F T H E L AT I N A M E R I C A N M I D D L E C L A S S : C O N C E P T S A N D M E A S U R E M E N T 37 FIGURE 2.3 Four economic classes, by income is by no means a small group: it includes 37.6 distribution, in selected Latin American countries percent of the continent’s population, includ- ing its modal resident! The existence and characteristics of this .04 group provide a number of useful insights. At the most basic level, perhaps, it suggests that escaping poverty—as most countries .03 and international agencies define it—is not enough to join the ranks of the comfortable- sounding, economically secure middle class. Density .02 There is a narrow but populous purgatory between those two states, characterized by considerable vulnerability and a high risk of .01 falling back into poverty. As a group, they are likely to be central to the continent’s social policy design, political dynamics, and 0 broad social contract, and we will return to 4a 10b 50c 100 them often in subsequent chapters. Before we Per capita daily income, US$ PPP get there, however, the next section describes the analytical framework used to link the Source: SEDLAC (Socioeconomic Database for Latin America and the Caribbean) data. two (multifaceted) concepts discussed so far Note: PPP = purchasing power parity. Countries include Argentina, Bolivia in this chapter: economic mobility and the (2008), Brazil, Chile, Colombia, Costa Rica, Dominican Republic, Ecuador, El Salvador, Honduras, Mexico (2010), Panama, Paraguay, Peru, and Uruguay. middle class. a. US$4 PPP = moderate Latin American and Caribbean poverty line. b. US$10 PPP = lower bound of Latin American middle class. c. US$50 PPP = upper bound of Latin American middle class. Linking mobility and middle- class dynamics: A matrix addition to the lower and upper bounds of decomposition the middle class, figure 2.3 also indicates the As noted in chapter 1, this volume aims to Latin American moderate poverty line of PPP shed light on three aspects of the microeco- US$4 per capita per day. nomic dynamics underpinning the recent One consequence of having sought to growth process in Latin America and the define a lower threshold for the middle class, Caribbean: which is endogenously derived from the vul- First, as economies grow (and, in many nerability approach (and validated by the cases, become less unequal), are greater subjective approach) is that there is no rea- opportunities being seized by all Latin Amer- son why that threshold should then coincide icans or only by those whose families have with the poverty line. Indeed, in our case, the long hoarded the keys to prosperity? This is lower bound of the middle class is consider- largely a question of mobility across genera- ably higher than the moderate poverty line tions, and it is addressed in chapter 3, which of US$4 a day commonly used by the World investigates the extent to which success in Bank for the Latin American and Caribbean our societies today—in school or at work—is region. This implies, of course, that there are determined by who our parents were. four—not three—economic classes in our Second, how does growth manifest itself analysis. We refer to the people with incomes at the level of the individual worker or stu- between US$4 and US$10 per person per dent? How do the aggregate GDP growth fig- day, who are too well-off to be considered ures translate into growing incomes for indi- poor but too vulnerable to be regarded as viduals and their families? These questions middle-class, as the vulnerable class. 22 This about income growth, or income movement 38 E C O N O M I C M O B I L I T Y A N D T H E R I S E O F T H E L AT I N A M E R I C A N M I D D L E C L A S S : C O N C E P T S A N D M E A S U R E M E N T within a person’s lifetime, are addressed in FIGURE 2.4 Horizontal decomposition of chapter 4. mobility in Peru, 2004–06 Third, as incomes grow at the lower end of the income distribution, raising millions of 2.0 Latin Americans out of poverty, is it true that the middle class is growing across the conti- 1.5 nent? How are these middle classes defined? Who was already part of them, and who are Density the new entrants? Do the “old” and “new” 1.0 middle classes look alike? Do they have simi- lar backgrounds? Do they think and act in 0.5 similar ways? What will a larger (and possi- bly different) middle class mean for savings, growth, and the shaping of economic policy 0 in Latin America? 0 0.2 0.4 0.6 0.8 1.0 The link between the last two questions— Percentile between mobility as income movement and gross upward mobility the growth of the middle class—is prob- gross downward mobility ably self-evident. If middle classes grow, it is because more people have incomes that are Source: Encuesta Nacional de Hogares (ENAHO) 2004 and 2006 by the Insti- tuto Nacional de Estadística e Informática (INEI) de Perú. large enough to earn them membership. Con- veniently, this obvious link can be intuitively formalized in terms of a matrix decomposi- tion of M 3: the measure of mobility as direc- na-GIC—measures total mobility as direc- tional income movement described early in tional income movement. this chapter (and more formally in focus note The same measure of social mobility can 2.1 at the end of the chapter). This decom- also be decomposed “vertically,” by social position is used for much of the analysis in class at origin. One could simply partition chapter 4. the initial income vector by deciles, or in any As noted earlier, that measure of social other way, and measure aggregate mobility mobility M 3 is simply an average of house- (both upward and downward) within each hold per capita income growth rates. It can subgroup of the initial population. Because a thus be decomposed as the sum of all pro- partition ensures that each household belongs portional income gains and all proportional to one and only one subgroup, the sum of all income losses. For any given income vector such measures of subgroup mobility would transformation, this “horizontal” decom- yield total mobility once again. position separates overall mobility into For our purposes, and given the income- that attributable to the “gainers” and that based definition of the middle class described associated with the “losers.” Exploiting the previously, it makes sense to partition the fact (discussed in more detail in focus note initial income vector into four groups. Using 2.1) that the measure corresponds graphi- the PPP US$4 per capita per day poverty line cally to the area under the non-anonymous commonly applied to Latin American and growth incidence curve (na-GIC), figure Caribbean countries in World Bank stud- 2.4 illustrates this decomposition for the ies, as well as the US$10–US$50 per day 2004–06 interval in Peru. In this figure, the middle-class thresholds, one could decom- area in green corresponds to a measure of pose economic mobility by group at origin gross upward mobility in Peru during the in the manner depicted in figure 2.5, once period, while the area in orange measures again for Peru (2004–06). The area under gross downward mobility. 23 The difference the na-GIC up until the percentile corre- between two—that is, the integral of the sponding to an income of US$4 a day yields E C O N O M I C M O B I L I T Y A N D T H E R I S E O F T H E L AT I N A M E R I C A N M I D D L E C L A S S : C O N C E P T S A N D M E A S U R E M E N T 39 FIGURE 2.5 Vertical decomposition of mobility group at origin, and the columns to the social in Peru, 2004–06 group at destination. Entries in each cell give the average income gain for that subgroup, 2.0 weighted by its population share. Rich Poor Vulnerable Middle Table 2.5 provides a schematic illustration class of how the elements in the decomposition 1.5 can be presented in a matrix or table format. To keep the picture as simple as possible, the Density 1.0 middle class and the elite have been grouped together, so that each panel has 9, rather than 16 cells. This is merely a presentational sim- 0.5 plification, although, in light of our earlier discussion of the nature of household survey 0 data for the rich in Latin America, it might be a sensible option in the actual analysis as 0 0.2 0.4 0.6 0.8 1.0 well. Percentile Table 2.5 shows that the overall amount of mobility in a particular society, over Source: ENAHO 2004 and 2006 by the INEI de Perú. a given period, can be separated out into the net income gains or losses among nine groups, or cells. These cells can be grouped according to different criteria, depending on overall mobility among the originally poor. the object of interest. We highlight three such The area under the GIC and between the per- possibilities: In a first cut, the nine cells can centiles corresponding to US$4 and US$10 be divided with respect to whether income in figure 2.5 measures mobility among those movement was sufficient for individuals to originally vulnerable in Peru. Similarly, the “change class.” Three cells (A, E, and I) rep- area between the percentiles corresponding resent “stayer groups”: people whose incomes to US$10 and US$50 per day shows mobility have not changed enough to move them among the middle class (at origin), and that across classes. They stay poor (A), stay vul- above US$50 measures mobility among the nerable (E), or stay middle or upper-class (I). rich elite. Another three groups (B, C, and F) are the To shed light on poverty dynamics, or on upwardly mobile “climbers”: their income the dynamics of the middle class, one can gains were enough for them to leave poverty combine the horizontal and vertical decom- or near-poverty behind and to join the ranks positions just described. Because M 3 is per- fectly additively separable, the decomposition into “winners and losers” and the decompo- TABLE 2.5 Matrix decomposition of M3: A schematic sition by social group at origin can be com- representation bined into what Ferreira and Lugo (2012) call Origin (rows) or Middle class a matrix decomposition of mobility as direc- destination (columns) Poor Vulnerable and above tional income movement. As those authors Poor A B C show, this decomposition is simply the sum of population-weighted average net income Vulnerable D E F gains (or losses) for each cell in a transition matrix. Indeed, the sum can be stacked so Middle class and G H I that all those who were initially poor are in above one row, while all those in the next group Source: Ferreira and Lugo 2012. Note: M3 = the measure of economic mobility as directional income movement. The rows cor- are immediately below, and so on. In such respond to the social group at origin, and the columns to the social group at destination. Entries in a matrix, the rows correspond to the social each cell represent the average income gain for that subgroup, weighted by its population share. 40 E C O N O M I C M O B I L I T Y A N D T H E R I S E O F T H E L AT I N A M E R I C A N M I D D L E C L A S S : C O N C E P T S A N D M E A S U R E M E N T of the near-poor or the middle class, respec- is conflated in the entries into the decom- tively. The final three groups (D, G, and H) position matrix in panel A but can easily be are the downwardly mobile “sliders”: they separated out. Panel B of table 2.6 presents live in households where income losses led to only the population shares for each cell. In falling back into vulnerability or, worse, into 2004, 34 percent of the population was poor poverty. in 2004, but by 2006, only 29 percent were A second cut focuses on transitions in and poor: 22 percent simply stayed poor, and out of poverty, and hence on a subset of five another 7 percent fell back into poverty. The cells in the matrix. This decomposition sug- Peruvian middle class (and elite) made up 26 gests a natural definition of chronic poverty: percent of the population in 2004, but almost those who started out and remain poor (A). It 12 percent joined them over the period—the also identifies those who left poverty (B and overwhelming majority not directly from the C) or who fell back into it (D and G). ranks of the poor. Because 8 percent of the A third possible cut narrows in on the five population fell from the middle class, that cells of greatest relevance for those interested class grew to almost 30 percent of the popu- in middle-class dynamics: Cell I contains lation in 2006. those who were and remain middle-class (or While panel B shows the population elite), while cells C and F include those who shares, panel C shows average income gains have recently joined the ranks of the middle (or losses) for each group. 24 One can see class (from vulnerability or directly from pov- that those few people who made it straight erty). Cells G and H consist of people whose from poverty to the middle class (2 percent falling income suggests that they have been of the population, from panel B) experienced displaced from the middle class. a remarkable 420 percent growth over the Naturally, the information presented in period. Conversely, the unfortunate 1 percent each of these cells may be the actual ele- who fell two classes did so by losing almost ments of the decomposition of M 3, or it may 80 percent of their incomes. And so on. be other information of interest about these Our preferred measure of intragenera- population subgroups. Table 2.6 illustrates tional mobility, M 3 , can therefore be used three possible alternatives. for much more than simply comparing the Panel A presents the actual decomposi- extent of directional income movement tion for Peru in 2004–06, which we have across countries or over time, important been using as an example. M 3 , which gives though that may be. By means of the matrix the average rate of growth in household per decomposition above, it can also shed light capita incomes across the Peruvian income on how growth was distributed across the distribution over this period, was 0.4 (or 40 population and what that means in terms percent). As the marginal distributions in of class dynamics. Using a standard poverty table 2.6 (panel A) show, all of this growth line, as well as the more original definition took place among those who were origi- proposed in the “Defining the Middle Class” nally poor or vulnerable (with 26 percent- section, the decomposition allows us to age points coming from the poor). But this investigate the extent of chronic poverty and growth moved enough people across social contrast it with the magnitude of movements classes that, if we look by destination, 24 of both out of and into poverty. It permits us the 40 percentage points of growth were for to investigate the stable middle class as well households that ended up in the middle class as those who have recently ascended to or in 2006! fallen from its ranks. The analysis in chap- But how many people moved across ters 4 and 5 draws on these and other tools social groups? And what were their average to gain a better understanding of the nature income gains and losses? This information of intragenerational mobility and the growth E C O N O M I C M O B I L I T Y A N D T H E R I S E O F T H E L AT I N A M E R I C A N M I D D L E C L A S S : C O N C E P T S A N D M E A S U R E M E N T 41 TABLE 2.6 Matrix decomposition of M3 in Peru, 2004–06 a. Decomposition of M3 2006 (destination) Poor Vulnerable Middle class + Total 2004 Poor 0.06 0.14 0.06 0.26 2004 Vulnerable −0.02 0.02 0.14 0.14 (origin) Middle class + −0.01 −0.03 0.04 0.00 Total 2006 0.03 0.13 0.24 0.40 b. Population shares in the transition matrix 2006 (destination) Poor Vulnerable Middle class + Total 2004 Poor 0.22 0.10 0.02 0.34 2004 Vulnerable 0.06 0.23 0.10 0.40 (origin) Middle class + 0.01 0.07 0.18 0.26 Total 2006 0.29 0.41 0.30 1.00 c. Average income growth in the transition matrix 2006 (destination) Poor Vulnerable Middle class + Total 2004 Poor 0.28 1.34 4.21 0.77 2004 Vulnerable −0.44 0.13 1.37 0.36 (origin) Middle class + −0.79 −0.47 0.21 −0.02 Total 2006 0.10 0.32 0.80 0.40 Source: ENAHO 2004 and 2006 by the INEI de Perú. Note: M3 = the measure of economic mobility as directional income movement. of the middle class in Latin America over the of social dynamics in Latin America but also past one or two decades. By identifying indi- to begin investigating their determinants viduals belonging to each of the cells in table and, in particular, how public policy in vari- 2.5 and analyzing their characteristics, we ous realms may have promoted or impeded hope not only to paint an accurate portrait upward mobility. 42 E C O N O M I C M O B I L I T Y A N D T H E R I S E O F T H E L AT I N A M E R I C A N M I D D L E C L A S S : C O N C E P T S A N D M E A S U R E M E N T Focus Note 2.1 Mobility concepts and measures Most measures of mobility as movement or origin 2. The second step is aggregation, wherein informa- independence can be constructed in two steps: tion across all individual mobility functions in the population is combined into a single summary 1. First, define an individual mobility function as index. One simple and appealing aggregator is some measure of the distance between an indi- arithmetic averaging: vidual’s income in the initial and final vectors, 1 respectively, y 0, y1. If we denote each “individual” M(Y0 , Y1) = ³ d (y0 (p0 ), y1 (p0 ))dp0 . (F2.1.2) 0 by the position he or she occupies in the initial vector or distribution, p 0 = F 0 (y), then the indi- Table F2.1 provides a simple example of individual vidual mobility function can be written in general mobility functions for each of the subconcepts 1 (a–d) terms as and 2, as well as a graphical depiction of what the function or profile looks like in an actual recent vec- m(p0 ) = d (y0 (p0 ), y1 (p0 )).25 (F2.1.1) tor transformation in Latin America, namely that in Peru, between 2004 and 2006. TABLE F2.1 Sample mobility functions and graphical representation of Peru, 2004–06 Individual mobility Graphical representation of the Concept function: an example profile: Peru, 2004–06 Directional income movement d ( y0 , y1 ) = y1 − y0 Differences in income 1,100 700 300 –100 0 0.2 0.4 0.6 0.8 1.0 Percentile Nondirectional income movement d ( y0 , y1 ) = y1 − y0 Absolute differences in income 1,100 700 300 –100 0 0.2 0.4 0.6 0.8 1.0 Percentile E C O N O M I C M O B I L I T Y A N D T H E R I S E O F T H E L AT I N A M E R I C A N M I D D L E C L A S S : C O N C E P T S A N D M E A S U R E M E N T 43 Focus Note 2.1 (continued) Individual mobility Graphical representation of the Concept function: an example profile: Peru, 2004–06 Share movement y1 y0 Differences in income shares d ( y0 , y1 ) = − 0.2 m1 m0 μt: mean income in time t 0 –0.2 –0.4 –0.6 –0.8 0 0.2 0.4 0.6 0.8 1.0 Percentile Positional movement d ( y0 , y1 ) = rank 1 − rank 0 Differences in rank 0.10 0.05 0 –0.05 –0.10 –0.15 0 0.2 0.4 0.6 0.8 1.0 Percentile Mobility as origin 1 ⎛ y − m0 y1 − m1 ⎞ Square differences in d( y0 , y1 ) = ⎜ 0 − independence 2 ⎝ so si ⎟ ⎠ standardized income 5 μt: mean income in time t 4 st: standard deviation of income in time t 3 2 1 0 0 0.2 0.4 0.6 0.8 1.0 Percentile (focus note continued next page) 44 E C O N O M I C M O B I L I T Y A N D T H E R I S E O F T H E L AT I N A M E R I C A N M I D D L E C L A S S : C O N C E P T S A N D M E A S U R E M E N T Focus Note 2.1 Mobility concepts and measures (continued) Visual inspection of the income mobility function If expressed as a function of the original percentile diagrams (in the third column of table F2.1) clearly p0 , this is simply a non-anonymous growth incidence reveals how each particular measure is sensitive to curve, g (p0 ) = y1 (p0 ) − y0 (p0 ) , described by Grimm (2007) different ranges of the distribution and maps the same y0 (p0 ) underlying distributional change differently into its and Bourguignon (2011): it gives the growth rate of own metric. Although each concept (and index) sum- individual incomes (between periods 0 and 1) for marizes a complex distributional change in a different those people initially in position p 0 of the original way, and thus contributes to one’s overall understand- distribution. This is an identity-preserving (that is, ing of the process, it is infeasible to present all such non-anonymous) version of the well-known growth indices for all mobility episodes examined in this vol- incidence curve introduced by Ravallion and Chen ume. Choices had to be made and, as argued in the text, we have chosen to focus on mobility as direc- (2003): g (p ) = y1( p ) − y 0( p ) , which considers the propor- tional income movement for the intragenerational y0 ( p ) domain and on mobility as origin independence for tional income differences between those in percentile the intergenerational. p in the final distribution and those in the same per- The specific index we use for mobility as direc- centile in the initial distribution. tional income movement is a simple, yet interesting, transformation of the individual mobility function in Aggregating equation (F2.1.3) across individuals the first row of table F2.1. If one takes the individ- to obtain a measure of social mobility yields ual income distance function as a proportion of the 1 y1 (p0 )  y0 (p0 ) 1 initial income, rather than as the absolute difference, M(Y0 , Y11) = ³ dp0 ³ g (p0 )dp0 (F2.1.4) 0 y0 (p0 ) 0 we have y1( p0 )− y0 ( p0 ) . Equation (F2.1.4) is a well-known mobility index. d (y0 ,y1) = (F2.1.3) Its log-approximation is the M 3 measure in Fields and y0 ( p0 ) Ok (1999). It has the appealing feature that it corre- sponds to the area under the non-anonymous growth incidence curve. E C O N O M I C M O B I L I T Y A N D T H E R I S E O F T H E L AT I N A M E R I C A N M I D D L E C L A S S : C O N C E P T S A N D M E A S U R E M E N T 45 Notes she actually enjoys or exercises, is often called one’s “capabilities.” 1. To assess mobility, economists must keep 11. In another link with the sociological litera- track of the individual identity of income ture, López-Calva and Ortiz-Juarez (2011) recipients as the income vector (or distribu- refer to the view that class positions are inher- tion) evolves though time. ently intertwined with vulnerability and risk, 2. However, given differences in the way ques- as discussed, for instance, in Goldthorpe and tions are asked and surveys collected across McKnight (2004). countries and over time, comparing different 12. The data sets used are the Socioeconomic varieties of apples will often be inevitable. Characterization Survey (CASEN Panel) for 3. When the individual mobility function is writ- 2001 and 2006 for Chile; the Mexican Fam- ten as a function of the individual’s position ily Life Survey (MxFLS) for 2002 and 2005 in the initial income vector, it corresponds to for Mexico; and the National Household what van Kerm (2009) calls an income mobil- Survey (ENAHO Panel) for 2002 and 2006 ity profile. for Peru. 4. Ferreira and Gignoux (2011) show that this 13. The same vector of covariates is also used as family of intergenerational mobility measures independent variables in a household income is isomorphic to a widely used measure of ex regression. The results allow the authors to ante inequality of opportunity. map predicted conditional probabilities of 5. As shown by D’Agostino and Dardanoni falling into poverty (in an interval of roughly (2006), this social mobility function can be five years early in the last decade) to (pre- obtained as an aggregation of the individual dicted) average initial household incomes. mobility function shown in the last row of See López-Calva and Ortiz-Juarez (2011) for focus note 2.1 (at the end of the chapter): details. 2 1 1 ⎛ y0 − m0 y1 − m1 ⎞ 14. As noted, however, the threshold is anchored d (y0 , y1) = ∑ ⎝ s0 − s1 ⎟ 2n ⎜ ⎠ = 1 − r01 . by being close to the average upper-bound estimate of actual vulnerability to poverty in 6. Both the anonymous and non-anonymous the region (see chapter 4). growth incidence curves (GICs) are briefly 15. In the so-called Leyden approach to subjec- introduced in focus note 2.1, at the end of the tive poverty identification, households with chapter. certain demographic characteristics were 7. Omitting subscripts for simplicity: asked what income they felt a household Δy like theirs needed to “make ends meet.” The M 3 = ∫ y f (y0 )dy0 , whereas growth in mean answers were typically found to increase with Δm my y household (objective) incomes, and the level at incomes is m = ∫ y m f (y0 )dy0 . See Klasen which the two (actual and answered) incomes (1994) for a related discussion. crossed was taken as the poverty line (see, for 8. The origin of these concepts is in Marx’s example, Hagenaars and van Praag [1985]). classic writings, such as Marx and Engels’ 16. The question in Spanish is: “En nuestra socie- The German Ideology ([1845] [1932] 1998), dad la gente tiende a ubicarse en distintas where they propose the concept of the clases sociales. ¿Se siente Ud. perteneciente a lumpenproletariat as the lowest class among alguna de estas clases?” the working class or proletariat. 17. In particular, they are very different for Chile, 9. The importance of “thoughts, perceptions, where the vulnerability approach yields a expressions, and actions” to the “symbolic threshold of PPP US$8.50 per day compared aspect of class structure” has been famously with PPP US$20.30 per day for the subjective emphasized more recently by Bourdieu (1980; approach. 1987) and others. 18. SEDLAC , (Socioeconomic Database for 10. The term “functionings” is commonly used Latin America and the Caribbean), is jointly in development economics to denote the set managed and maintained by the Centro de of activities and achievements (“beings and Estudios Distributivos, Laborales y Sociales doings”) that a person is capable of, following (CEDLAS) of the Universidad de la Plata in Sen (1985). The set of feasible functionings, Argentina, and the World Bank. See http:// from which a person chooses those that he or cedlas.econo.unlp.edu.ar. 46 E C O N O M I C M O B I L I T Y A N D T H E R I S E O F T H E L AT I N A M E R I C A N M I D D L E C L A S S : C O N C E P T S A N D M E A S U R E M E N T 19. Kharas’s (2010) proposal was motivated by Aristotle. (c. 350 BC) 1932. Politics. Translated by the fact that average daily per capita income H. Rackham. Cambridge, MA: Harvard Uni- in the United States in 2009 was $98.77. versity Press. 20. See, for example, Korinek, Mistiaen, and Atkinson, Anthony B., Thomas Piketty, and Ravallion (2006). Emmanuel Saez. 2011. “Top Incomes in the 21. So far, the only analysis of tax record data Long Run of History.” Journal of Economic in Latin America and the Caribbean is for Literature 49 (1): 3–71. Argentina (Alvaredo 2010). Other coun- Banerjee, Abhijit V., and Esther Duflo. 2008. tries in the region for which there is ongoing “What Is Middle Class about the Middle work include Belize, Brazil, Chile, Colombia, Classes around the World?” Journal of Eco- Guyana, Jamaica, St. Vincent, and Trinidad nomic Perspectives 22 (2): 3–28. and Tobago. For more details, see the World Barro, Robert J. 1999. “Determinants of Democ- Top I ncomes Database, ht tp: //g-mond racy.” Journal of Political Economy 107 (6): .parisschoolofeconomics.eu/topincomes/. 158–83. 22. One might also call them the “near-poor” or ———. 2000. “Inequality and Growth in a Panel the “lower middle class.” The latter terminol- of Countries.” Journal of Economic Growth ogy is more consistent with the domestic clas- 5 (1): 5–32. sification chosen, for example, by Brazil. Birdsall, Nancy, Carol Graham, and Stefano Pet- 23. The figure has been constructed from the tinato. 2000. “Stuck in the Tunnel: Is Global- percentile distribution in the initial period. ization Muddling the Middle Class?” Work- Hence, each point in the line formally repre- ing Paper 14, Center on Social and Economic sents the average income growth within each Dynamics, Brookings Institution, Washington, percentile. DC. 24. As the attentive reader will have guessed, Blackburn, McKinley L., and David E. Bloom. the products of the cells in panels B and C of 1985. “What Is Happening to the Middle table 2.6 yield the corresponding entries in Class?” American Demographics 7 (1): 18–25. panel A, up to an error of approximation. Bourdieu, Pierre. 1980. The Logic of Practice. 25. Just as “income” is used here as shorthand for Translated by Richard Nice. Stanford, CA: whichever variable is appropriate to capture Stanford University Press. the mobility space of interest, so “individual” ———. 1987. “What Makes a Social Class? On is used as shorthand for the identity of the ele- the Theoretical and Practical Existence of ments in the income vector. In the intragen- Groups.” Berkeley Journal of Sociology 32: erational domain, these would generally cor- 1–27. respond to actual individuals, whereas in the Bourguignon, François. 2011. “Non-anonymous intergenerational domain they would usually Growth Incidence Curves, Income Mobility, denote lineages: parents and their children. and Social Welfare Dominance.” Journal of Economic Inequality 9 (4): 605–27. Chen, Shaohua, and Martin Ravallion. 2001. “How Did the World’s Poor Fare in the References 1990s?”Review of Income and Wealth 47 (3): Akerlof, George A., and Rachel E. Kranton. 283–300. 2002. “Identity and Schooling: Some Lessons Cornell, Tim J. 1995. The Beginnings of Rome. for the Economics of Education.” Journal of New York: Routledge. Economic Literature 40 (4): 1167–201. D’Agostino, Marcello, and Valentino Dardanoni. Alesina, Alberto, and Roberto Perotti. 1996. 2006. “The Measurement of Mobility: A Class “Income Distribution, Political Instability, and of Distance Measures.” Unpublished manu- Investment.” European Economic Review 40 script, University of Palermo, Italy. (6): 1203–28. Davis, Joe C., and John H. Huston. 1992. “The Alvaredo, Facundo. 2010. “The Rich in Argentina Shrinking Middle-Income Class: A Multivari- over the Twentieth Century 1932–2004.” In ate Analysis.” Eastern Economic Journal 18 Top Incomes over the Twentieth Century vol. (3): 277–85. II: A Global Perspective, ed. A. Atkinson and Easterly, William. 2001. “The Middle Class Con- T. Piketty, 253–98. Oxford: Oxford University sensus and Economic Development.” Journal Press. of Economic Growth 6 (4): 317–35. E C O N O M I C M O B I L I T Y A N D T H E R I S E O F T H E L AT I N A M E R I C A N M I D D L E C L A S S : C O N C E P T S A N D M E A S U R E M E N T 47 Elbers, Chris, Jean O. Lanjouw, and Peter Lan- Klasen, Stephan. 1994. “Growth and Well-Being: jouw. 2003. “Micro-Level Estimation of Wel- Introducing Distribution-Weighted Growth fare.” Econometrica 71(1): 355–64. Rates to Reevaluate U.S. Postwar Economic Ferreira, Francisco H. G., and Jérémie Gig- Performance.” Review of Income and Wealth noux. 2011. “The Measurement of Educa- 40 (3): 251–72. tional Inequality: Achievement and Opportu- Korinek, Anton, Johan A. Mistiaen, and Martin nity.” Policy Research Working Paper 5873, Ravallion. 2006. “Survey Non-Response and World Bank, Washington, DC. the Distribution of Income.” Policy Research Ferreira, Francisco H. G., and Maria A. Lugo. Working Paper 3543, World Bank, Washing- 2012. “Decomposition of Measures of Income ton, DC. Movement.” Unpublished manuscript, World López-Calva, Luis F., and Eduardo Ortiz-Juarez. Bank, Washington, DC. 2011. “A Vulnerability Approach to the Defi- Fields, Gary S. 2000. “Income Mobility: Con- nition of the Middle Class.” Policy Research cepts and Measures.” In New Markets, New Working Paper 5902, World Bank, Washing- Opportunities? Economic and Social Mobil- ton, DC. ity in a Changing World, ed. Nancy Birdsall Marx, Karl, and Friedrich Engels. (1845) (1932) and Carol Graham, 101–33. Washington, DC: 1998. The German Ideology. Amherst, NY: Brookings Institution and Carnegie Endow- Prometheus. ment Press. McGeough, Kevin M. 2004. The Romans: An ———. 2010. “Does Income Mobility Equalize Introduction . Oxford: Oxford University Longer-Term Incomes? New Measures of an Press. Old Concept.” Journal of Economic Inequal- Milanovic, Branko, and Shlomo Yitzhaki. 2002. ity 8 (4): 409–27. “Decomposing World Income Distribu- Fields, Gary S., and Efe A. Ok. 1999. “Measuring tion: Does the World Have a Middle Class?” Movement of Incomes.” Economica 66 (264): Review of Income and Wealth 48 (2): 155–78. 455–71. Partridge, Mark D. 1997. “Is Inequality Harmful Friedman, Milton. 1962. Capitalism and Free- For Growth? Comment.” American Economic dom. Chicago: University of Chicago Press. Review 87 (5): 1019–32. Galton, Francis. 1886. “Regression toward Medi- Piketty, Thomas. 1995. “Social Mobility and ocrity in Hereditary Stature.” Journal of the Redistributive Politics.” Quarterly Journal of Anthropological Institute of Great Britain and Economics 110 (3): 551–84. Ireland 15: 246–63. Piketty, Thomas, and Emmanuel Saez. 2003. Goldthorpe, John H., and Abigail McKnight. “Income Inequality in the United States, 1913– 20 0 4. “T he E conom ic Basis of S ocial 1998.” Quarterly Journal of Economics 118 Class.” CASE paper 80, Centre for Analysis of (1): 1–39. Social Exclusion, London School of Economics Ravallion, Martin. 2010. “The Developing and Political Science, London. World’s Bulging (but Vulnerable) Middle Grimm, Michael. 2007. “Removing the Anonym- Class.” World Development 38 (4): 445–54. ity Axiom in Assessing Pro-Poor Growth.” Ravallion, Martin, and Shaohua Chen. 2003. Journal of Economic Inequality 5 (2): 179–97. “Measuring Pro-Poor Growth.” Economics Hagenaars, Aldi J. M., and Bernard van Praag. Letters 78 (1): 93–99. 1985. “A Synthesis of Poverty Line Defini- Sen, Amartya K. 1985. Commodities and Capa- tions.” Review of Income and Wealth 31 (2): bilities. Oxford: Oxford University Press. 139–54. Shorrocks, Anthony F. 1978. “Income Inequality Hirschman, Albert, and Michael Rothschild. and Income Mobility.” Journal of Economic 1973. “The Changing Tolerance for Income Theory 19 (2): 376–93. Inequality in the Course of Economic Devel- Solimano, Andres. 2008. “The Middle Class opment.” Quarterly Journal of Economics 87 and the Development Process.” Serie Macro- (4): 544–66. economía del Desarrollo 65, United Nations Kharas, Homi. 2010. “The Emerging Middle and Economic Commission for Latin America Class in Developing Countries.” Working and the Caribbean, Santiago, Chile. Paper 285, Development Centre, Organisation Solon, Gary. 2002. “Cross-Country Differences for Economic Co-operation and Development, in Intergenerational Earnings Mobility.” Jour- Paris. nal of Economic Perspectives 16 (3): 59–66. 48 E C O N O M I C M O B I L I T Y A N D T H E R I S E O F T H E L AT I N A M E R I C A N M I D D L E C L A S S : C O N C E P T S A N D M E A S U R E M E N T Time. 2011. “What Happened to Upward Mobil- Weber, Max. 1946. “Class, Status, Party.” In ity?” November 14. From Max Weber: Essays in Sociology, ed. Tocqueville, Alexis de. (1856) 1986. L’Ancien Hans H. Gerth and C. Wright Mills. New Régime et la Révolution. Paris: Robert Lafont. York: Oxford University Press. Torche, Florencia. 2010. “Social Status and Pub- ———. (1922) 1978. Economy and Society. Ed. lic Cultural Consumption: Chile in Compara- Guenther Roth and Claus Wittich. Berkeley: tive Perspective.” In Social Status and Cultural University of California Press. Consumption, ed. T. W. Chan and J. H. Gold- Zimmerman, David. 1992. “Regression toward thorpe, 109–38. Cambridge, U.K.: Cambridge Mediocrity in Economic Stature.” American University Press. Economic Review 82 (3): 409–29. Van Kerm, Philippe. 2009. “Income Mobility Pro- files.” Economics Letters 102 (2): 93–95. 3 Mobility across Generations If income mobility were very high, the degree of inequality in any given year would be unimportant, because the distribution of lifetime income would be very even. . . . An increase in income mobility tends to make the distribution of lifetime income more equal. —Paul Krugman (1992), “The Rich, the Right, and the Facts.” I n spite of the remarkable achievements Parental background influences chil- obtained during the past decade, income dren’s outcomes through a variety of chan- inequality in Latin America remains high nels. Even before children are born, maternal by international standards and certainly a nutrition and health during gestation have major concern for policy makers. Arguably, an impact on children endowments at birth however, high inequality might be socially (Currie 2009). In turn, there is increasing acceptable if coupled with strong social empirical evidence suggesting that these birth mobility. This is especially true in the case endowments have an influence on adult out- of mobility across generations. To the extent comes, including educational attainment and that equal opportunities are provided to chil- incomes (Currie 2011). A schematic simplifi- dren from different parental backgrounds, cation of the complex relationship between some inequality of outcomes might be socially parental background and their children’s acceptable, or even desirable, because it may income is presented in figure 3.1, which provide the right incentives for exerting effort draws from Haveman and Wolfe (1995). and, through this channel, foster economic Parents affect children through heredity efficiency and future growth. Indeed, it has of genetic endowments, which in turn affects been argued that individuals living in a soci- children’s schooling and income, an aspect ety characterized by a great degree of gen- first formalized by the seminal work of erational mobility are more likely to accept Becker and Tomes (1979). In addition, paren- existing inequalities than individuals living tal ability influences their own educational in a world where their fortunes are highly attainment and thus their income. Together, dependent on the socioeconomic statuses of these determine the level of “home invest- their parents (Bénabou and Ok 2001). ments” in offspring (including time spent 49 50 MOBILIT Y ACROSS GENERATIONS FIGURE 3.1 The intergenerational association between parental background and children’s income Heredity Ability Quality of time inputs Parents’ abilities Quantity of time imputs Final schooling Quality of goods inputs Home investments level Income Parents’ education Quantity of goods imputs Postschool Family income investment Source: Haveman and Wolfe 1995. with the children and the quality and quan- among schools in how they treat differences tity of goods and services delivered to them), in children’s endowments. Some schools which, in turn, will affect the final schooling have more inclusive policies and try to level. Furthermore, parental income exerts a bring the worst-performing kids closer to direct influence on final schooling (through the average. Others put more emphasis on the choice of school) and on the children’s the better-endowed kids, trying to enhance eventual income (through networks and con- and develop their full potential. Second, the nections in the labor market). Finally, the importance of the school in promoting equal schooling level attained by the children will opportunity is confounded with the role of affect their income later in life and further parental background. Most naturally, par- experience through the labor market (post- ents do not choose schools for their chil- school investments). All of these factors, in dren randomly. There is instead substantial turn, affect their own children’s earnings and positive sorting: that is, parents with more income. resources send their children to better- Haveman and Wolfe’s (1995) diagram endowed schools. (figure 3.1) focuses on the direct and indi- The government is the second funda- rect links between parental background and mental actor that shapes the complex rela- children’s income. In doing so, it pays little tionship between parental background and attention to the external factors that shape children’s outcomes. Governments indeed parental influences on children’s outcomes. have the capacity to alter this relationship Three key actors mediate the process of inter- through a number of different channels. A generational mobility: the schooling system, primary intervention is through the provi- the government, and the labor market. sion of public schools. Access to high-quality It is generally understood that better- universal education can certainly help to level managed, better-endowed schools are more the playing field. Additionally, the govern- likely to succeed in bringing up children’s ment can influence the sorting process that human capital. However, the role of schools characterizes schooling choice—for instance, in promoting intergenerational mobility through fellowships and voucher programs. is far from being settled for at least two As emphasized by Solon (2004), the progres- reasons: First, there is great heterogeneity sivity of public investment in human capital MOBILIT Y ACROSS GENERATIONS 51 is a key government intervention to enhance are completely determined by the socioeco- intergenerational income mobility. nomic status of their parents. Poor children The final mediator between parental are born from poor families, and rich chil- background and children’s well-being is the dren are born from rich families. This soci- labor market. An inefficient labor market ety is thought to display no intergenerational that favors connections and nepotism instead mobility. At the other extreme, in the second of rewarding talent will hinder intergenera- society, the relative socioeconomic position tional income mobility. On the other hand, of parents is completely uninformative of the higher returns to schooling are expected to income or education of their offspring. The be associated with a higher investment effort probability that children born from poor of parents in the human capital of their chil- families will go to college and end up rich is dren, increasing the association between par- equal to that of children born from rich fami- ents and children’s incomes (Solon 2004). lies. In this case, there is perfect intergenera- tional mobility. The main goal of this chapter is to understand where Latin America stands Chapter focus and objectives in the continuum of possibilities delimited by In this chapter, we discuss the impact of these two extremes. How do Latin Ameri- parental background on a variety of out- can countries compare with middle-income comes for their children, including educa- and rich countries in other regions? Can tional attainment; educational achievement; we provide some tentative evidence on the and, in the few cases where data for Latin determinants of the relative position of these America are available, income. An important countries? limitation of the analysis is that data sets fea- This chapter pays special attention to the turing the same measure of socioeconomic influence of parental background on the edu- status across generations (for example, per- cational outcomes of Latin American chil- taining to education, occupation, or income) dren. This emphasis contrasts with most of are generally not available for Latin America. the literature on intergenerational mobility, For this reason, we use different approxi- in which income or earnings are central to mations of the complex concept of parental the analyses. The rationale for putting educa- background, including proxies for the per- tion center stage is twofold: First, for Latin manent income of the parents, their ethnic- American countries, there is greater data ity, their schooling, and their occupational availability on education across generations status. In a few cases we also try to evaluate than on income or earnings. These microdata the role of policies and institutions in shap- sets are also available for a large number of ing differences over time and across countries countries, allowing benchmarking of Latin in the observed patterns of intergenerational America with respect to other developed and mobility. developing countries. The second rationale is As discussed in chapter 2, the particular of substance. Among observable and measur- concept of mobility adopted throughout the able human characteristics, education is the chapter is that of origin independence. (See most important determinant of income, as box 3.1 for a discussion of the chosen mea- emphasized by the enormous literature on sure of origin independence.) This is the con- the returns from education.1 In addition, the cept of choice of most intergenerational stud- increasing availability of data sets measur- ies (Solon 2002; Zimmerman 1992). ing cognitive test scores in a large number of To illustrate this concept, we will focus countries has shown a strong positive effect on two polar cases, following Solon (1999): of educational achievement on labor-market Imagine two societies with the same level performance (Neal and Johnson 1996; McIn- and distribution of income and thus the same tosh and Vignoles 2001; Currie and Thomas proportion of rich and poor. In the first soci- 2000). Moreover, the education of a person ety, children’s income and level of education affects not only his or her wages but also his 52 MOBILIT Y ACROSS GENERATIONS BOX 3.1 Assessing the association of socioeconomic status across generations There are different ways to measure the association that it is less prone to classical measurement error in in outcomes across generations. Perhaps the simplest the outcomes of the children. Perhaps more impor- and most common application relates an outcome tant, the gradient is to be preferred if the researcher variable of the parent generation, denoted by the is interested not only in how much parental back- subscript 0, with the same outcome for their chil- ground explains the outcomes in the children’s gen- dren, denoted with subscript 1, in the following lin- eration but also in how unequal those outcomes are. ear fashion: To provide an idea of what a “typical” degree of inequality in the parental generation represents for y 1 = D + Ey 0 + H, (B3.1a) their children’s outcomes, our preferred measure dis- cussed in the chapter will be E × V(y 0). The reading where E is a measure of the persistence in of this measure is simple. It states by how much chil- incomes across generations; and dren’s outcomes change (in the units in which y 1 is measured) when parental background increases by 1 – E is a measure of intergenerational one standard deviation. mobility. We will consider a variety of outcomes for the children’s generation throughout the chapter, and in The outcomes differ across studies, but the most most cases we will not be able to measure the same common applications consider the log of incomes in outcomes for their parents, as suggested in equation each generation (see Black and Devereux 2011 and (B3.1a), but instead we can measure some indicators Björklund and Salvanes 2011 for recent overviews) that are likely to be correlated with such outcomes. so that E represents the intergenerational elasticity In the case of educational quality, for instance, we of income. An alternative measure is the correla- can measure cognitive tests only for the kids, while tion coeffi cient (U) between the vectors y 1 and y 2 , parental background is approximated by different which standardizes the intergenerational gradient indicators of socioeconomic status, including indices by the ratio of the standard deviations in the two of physical asset holdings, education, and occupation. generations: We view these outcomes of the parents as proxies for a latent variable, the test score, which is not observed σ(y0) in the data, as in Ferreira and Gignoux (2011). ρ=β . (B3.1b) σ(y1) To facilitate the interpretation of the results, in some cases we will discuss the differences in chil- The gradient and the correlation might provide dren’s outcomes between two representative families different pictures in specific instances and countries. at the extremes in the distribution of socioeconomic For instance, the gradient in one country might be status. Thus, we often will discuss differences in reduced over time simply because there has been outcomes of the children among parents who have a reduction in the inequality of outcomes in the tertiary education against parents who did not fi n- children’s generation, while the correlation would ish primary. Sometimes, the index of socioeconomic remain unaffected. Both metrics have advantages status is multidimensional. In these cases, we label and disadvantages, and there is no clear ranking “poor” and “rich” families as those that are at the between the two. One advantage of the gradient is bottom and top quintile of the distribution of the outcome variable (or set of variables), respectively. or her probability of employment. Hence, An important caveat should be highlighted a society that is not displaying substantial at the outset: correlations do not necessar- mobility in education across generations is ily imply causality. The patterns discussed unlikely to have a great deal of intergenera- here may not reflect a direct causal effect of tional income mobility, unless markets work parental socioeconomic status on the edu- in a very inefficient manner (for example, by cation or socioeconomic status of the chil- not rewarding education). dren. Instead, they will include both direct MOBILIT Y ACROSS GENERATIONS 53 and indirect effects of growing up in a more facts, the last part of this chapter discusses advantageous background. Both educational the policies and institutions that appear to be and socioeconomic outcomes may be deter- related to the different degrees of intergenera- mined, at least in part, by other unobserved tional mobility found both across and within individual or family characteristics such as countries. hard-wired genetic traits and environmental factors related to where individuals live (their neighborhoods, housing, schools, and so Educational attainment: on). Children from rich and better-educated How important is parental families tend to live in richer neighborhoods background? where unemployment, crime, and violence The reduction of poverty and inequality dur- are lower and the overall quality of services ing the 2000s, alluded to in chapter 2, was is higher. In addition, rich children are more anticipated by a rapid expansion of educa- likely to attend better schools, with better tional attainment. Has this expansion at the teachers and more inputs, and to interact same time become more egalitarian? Has with other kids from better-educated parents. the schooling gap between rich and poor All these factors are most likely to affect the narrowed? Are the sons and daughters of performance of children in school as well as less-favored households today more likely to later in life. Many of these factors are unob- finish primary and secondary schooling on served in most of the data sets used in this time than they were 20 years ago? How does chapter and will thus be partially captured by Latin America compare with other regions parental background. regarding the intergenerational association of This chapter aims to uncover some empiri- education? cal regularities in Latin America, but in most cases does not attempt to assess causality. Parental influence on years of schooling Hence, drawing specific policy conclusions from the analysis will often be difficult, pre- We start by examining the last question, cisely because we cannot isolate the specific and the message obtained is quite clear: the channel through which the intergenerational Latin American region is characterized by association emerges. In an attempt to over- substantial educational persistence across come this limitation, we discuss in boxes generations. Recent estimates of the correla- throughout the chapter specific examples tion in schooling across generations in differ- where either natural experiments or specific ent countries of the world suggest that Latin characteristics of programs in one coun- America presents the highest persistence of try have allowed researchers to identify the education across generations. Hertz et al. causal impact of policies on the extent of (2007) provides an excellent overview of intergenerational mobility. We complement these differences across countries. The study the analysis with cross-county regressions, produced an impressive data set of associa- which exploit the variability of policies across tions in years of schooling between parents countries to provide an indication of what and children in 42 countries, for different seems to work or not at the aggregate level. birth cohorts spanning the last 50 years. Our analysis starts with the evolution of Figure 3.2 is based on their estimates, educational attainment in Latin America and showing the average effect of one stan- differences among rich and poor children in dard deviation of parental years of school- years of schooling. We next study the influ- ing on children’s schooling (see box 3.1 for ences of parental background on educa- details on measurement issues). According tional achievement, measured by test scores to this metric, Latin American countries are in international assessments. We then briefly undoubtedly among the less educationally discuss the link between educational mobility mobile regions in the sample. In the extreme, and income mobility. Having established the in Peru, one standard deviation in parental 54 MOBILIT Y ACROSS GENERATIONS FIGURE 3.2 Impact of parental education on children’s years of education, selected countries 3 Years of education 2 1 0 rg ral pia Un ther epu na d re c Ne Kin land Ze om ec No and ep ay nm ic Uk ark va Ma ine ep sia Fin blic Ea sto d st nia lg r Un P ium d nd Ne es es Sw pal M en So Ire ab) Ne h A nd Ph erla ca Sw lipp ds er s et d Slo nam n a i L ry kis a n Gh aly Ni one a ca sia lo ua t, A C ia ra hile B p. u l na r Pe a ru Ec razi Be imo Pa ado itz ine ite n I bli Hu eni Pa ank d n m ta E an Vi lan De ubl b Re Sr ga at th fri i n h R rw Co rag r R hi In a h ( ed ite ola ut la k R lay It Ky Ru hio m ra l w gd u at al l St v No yz C T b t lE ra Ru yp lad Slo Cz Eg ng Ba Source: Data from Hertz et al. 2007. Note: Bars represent the impact of one standard deviation of parental years of schooling on the years of schooling of children. The impact is averaged across birth cohorts born between 1930 and 1980. education (about 3.5 additional years of that, in some low-income countries such as education) is associated with more than 3.0 rural Ethiopia, persistence in attainment additional years of schooling in the next gen- is low not because of low grade persistence eration. Peru is closely followed by Panama, (the regression coefficient) but because of the Ecuador, Brazil, Chile, Colombia, and Nica- low parental schooling level even among the ragua. The only non-Latin American coun- country’s most highly educated. The stan- try in the study group that displays a similar dard deviation of years of schooling among level of educational persistence across genera- parents in this case is extremely low, at an tions is the Arab Republic of Egypt. average of 0.5. The fact that Latin American countries However, educational mobility in Latin cluster at the top of the persistence of educa- America has improved in the past two tion across generations is especially remark- decades. Across cohorts born during different able, considering that the sample includes periods, we observe a mild increase in inter- both developing and developed economies. generational mobility of education in Latin As expected, Scandinavian countries tend America, especially during the past couple to display low educational persistence. Per- of decades. Figure 3.3 shows the evolution of haps more surprising is the low association the association between education of the par- between the years of schooling of parents ents and children in the seven Latin American and their children in Northern Ireland and countries, included in Hertz et al. (2007). In the United Kingdom, which contrasts with all Latin American countries, with the excep- the result found later in the chapter, where tion of Nicaragua, intergenerational mobility Anglo-Saxon countries display substantial increased between those born in the 1920s persistence when the outcome considered is and 1930s and those born in the 1970s. In student achievement (test scores) rather than some cases such as Chile, the reduction in the attainment (schooling). It should be noted relationship between parental and children’s MOBILIT Y ACROSS GENERATIONS 55 FIGURE 3.3 Evolution of intergenerational persistence in education across birth cohorts in seven Latin American countries, 1930s–80s a. Brazil b. Chile c. Colombia 3.5 3.5 3.5 Years of education Years of education Years of education 3.0 3.0 3.0 2.5 2.5 2.5 2.0 2.0 2.0 1.5 1.5 1.5 1930 1940 1950 1960 1970 1930 1940 1950 1960 1970 1980 1930 1940 1950 1960 1970 1980 d. Ecuador e. Nicaragua f. Panama 3.5 3.5 3.5 Years of education Years of education Years of education 3.0 3.0 3.0 2.5 2.5 2.5 2.0 2.0 2.0 1.5 1.5 1.5 1930 1940 1950 1960 1970 1930 1940 1950 1960 1970 1980 1940 1950 1960 1970 1980 g. Peru h. High income i. Low-middle income 3.5 1.6 1.6 Years of education Years of education 1.5 Years of education 1.5 3.0 1.4 1.4 1.3 1.3 2.5 1.2 1.2 2.0 1.1 1.1 1.0 1.0 1.5 0.9 0.9 1920 1930 1940 1950 1960 1940 1950 1960 1970 1980 1930 1940 1950 1960 1970 1980 Source: Data from Hertz et al. 2007. Note: Lines represent the estimated effect of one standard deviation of parental years of schooling on the years of schooling of children, across cohorts. education is quite impressive: the impact of of a regression of children’s education on one standard deviation of parental schooling parental education and changes in the stan- level on children’s years of education more dard deviation of parental education (see than halves during the period. It should be box 3.1). When we unbundle the two effects, noted, however, that regardless of the mod- we find that, in most countries, the E’s have erate improvements in mobility, the levels of declined, driving the increase in mobility we educational intergenerational persistence in are documenting. Latin America remain substantially higher Figure 3.4 shows this pattern in two coun- even among the youngest cohorts, compared tries: Colombia and Peru. In Colombia, we with either developed countries or developing observe a sharp decline in the E coefficient countries. (leaving the standard deviation fixed), which Naturally, changes in our estimates of the drives the overall increase in mobility we doc- association between parental and children’s ument in spite of the hump-shaped evolution education, as shown in figure 3.3, are driven in the inequality of education in the parents’ by two factors: changes in the E coefficients generation. In Peru, the hump is even more 56 MOBILIT Y ACROSS GENERATIONS FIGURE 3.4 Evolution of intergenerational persistence in education across birth cohorts in Peru and Colombia, 1920s–80s: Decomposition between parental inequality and ␤ a. Peru b. Colombia 4.0 4.0 3.5 3.5 Years of education Years of education 3.0 3.0 2.5 2.5 2.0 2.0 1.7 1.7 1920 1930 1940 1950 1960 1930 1940 1950 1960 1970 1980 Year Year Overall Beta fixed S.D. fixed Overall Beta fixed S.D. fixed Source: Data from Hertz et al. 2007. Note: S.D. = standard deviation. The overall effect represents the estimated effect of one standard deviation of parental years of schooling on the years of schooling of children, across cohorts. “Beta fixed” is the effect of a changing standard deviation in the education of the parents, keeping the beta fixed at the level obtained in the first available cohort in each country. “Standard deviation fixed” does the opposite exercise: it keeps the standard deviation of parental education fixed at the level of the first birth cohort available and allows the betas to change over time. pronounced. Educational inequality among school at the required age but dropped out parents (leaving E fixed) increases steadily up after completing only the first grade, his or to the cohort of children born in 1955, and her educational gap would be equal to 3. The then it reverses. This explains the very mild advantage of using this indicator relative to increase in overall mobility despite a steady the previous one is that it allows us to look decline in the E. at the gap at different schooling ages in the 1990s and 2000s and thus obtain a more recent picture of intergenerational mobility. Parental influence on the educational The educational gap of children at school- gap ing ages in Latin America is large, but it has The next few figures show the evolution of unambiguously declined during the past two an alternative measure of educational attain- decades. Figure 3.5 shows the educational ment: the educational gap.2 This indicator is gap at ages 10, 15, and 18 for the Latin defined as the difference between potential American average.3 On average, the school- years of education and the years of completed ing gap in the mid-1990s was 1.3 years for education. For instance, if the starting age of kids aged 10, and the gap was 5.0 years for primary education is 6, a 10-year-old child kids aged 18. By 2009, the gap more than should have, by the end of the school year, halved for both ages. four years of schooling (the exact number What is the impact of parental back- depends also on the month of birth and of the ground on the children’s schooling gap? interview). This is the maximum potential Because we rely on household surveys, we years completed. If this is the case for a given now have two possible proxies for paren- child, his or her educational gap would be tal background: education and income. The equal to 0. If, instead, the child just finished broad messages with the two measures are third grade, his or her educational gap would fairly consistent, so we concentrate on edu- be equal to 1. Similarly, if the child started cation to avoid repetition. Across all ages, MOBILIT Y ACROSS GENERATIONS 57 FIGURE 3.5 Average children’s educational gap FIGURE 3.6 Differences in the educational gap in Latin America, 1995–2009 between the top and bottom income quintiles in Latin America, 1995–2009 5 0 4 –0.2 Years of education –0.4 3 Years of education –0.6 2 –0.8 –1.0 1 –1.2 –1.4 0 95 97 99 01 03 05 07 09 –1.6 19 19 19 20 20 20 20 20 95 97 99 01 03 05 07 09 Year 19 19 19 20 20 20 20 20 Age 10 Age 15 Age 18 Year Age 10 Age 15 Age 18 Source: Data from SEDLAC (Socioeconomic Database for Latin America and the Caribbean). Note: Lines represent the (population weighted) average schooling gap Source: Data from SEDLAC (Socioeconomic Database for Latin America across Latin American countries, for children aged 10, 15, and 18, respec- and the Caribbean). tively. The sample includes a maximum of 15 countries in the region. In Note: Lines represent the expected reduction in the schooling gap asso - some years, however, some countries have no survey so they are excluded ciated with one standard deviation of parental education. Other covariates from the calculation. The minimum number of countries is 10, for the first in the regression are children’s gender, living in an urban area, and country three years of the period considered. Similar trends are found if the addi- fixed effects. The estimated effect of parental education on the educa- tional countries are excluded. tional gap is always statistically different from zero (confidence bands are omitted for clarity). the children of more-educated parents pres- studies show that a constant-elasticity rela- ent a lower schooling gap than the children tionship between the child’s and parent’s of less-educated parents, but the differences socioeconomic status is not supported by the associated with parental background in the data (for example, Behrman and Taubman children’s schooling gap narrowed during 1990; Zimmerman 1992; Dearden, Machin, the past 15 years (as shown in figure 3.6). and Reed 1997). Most of these studies find In 1995, one standard deviation of parental that there is more upward mobility at the education was associated with an additional bottom than downward mobility at the top 0.5-year gap at age 10; with a 1.0-year gap of the distribution. However, the importance at age 15; and with a 1.5-year gap at age 18. of poverty traps in Latin America can turn By 2009, the gap had declined to 0.3, 0.6, around these results. and 1.0 year for each of the age categories, Table 3.1 presents cross-tabulations of respectively. schooling gaps (in columns) as a function of the level of completed education of the par- ents (in rows) for the first (circa 1995) and Mobility at the top and mobility at the last (circa 2009) years in our sample in the bottom region. For simplicity, we focus our discus- The discussion so far has focused on a single sion on children aged 15. average parameter for the whole population Several aspects are worth noting: in each point in time, but there is no reason to think that the extent of mobility in a society • First, there has been great mobility at is equal at different points of the distribution the bottom. The share of children whose of parental background. Indeed, numerous parents had less than primary education 58 MOBILIT Y ACROSS GENERATIONS TABLE 3.1 Relationship between parental education and children’s average educational gap at age 15 in Latin America, 1995 versus 2009 a. Gap for 15-year-olds, circa 1995 (%) b. Gap for 15-year-olds, circa 2009 (%) Gaps in years Gaps in years 0 1 [2, 3] ≥4 0 1 [2, 3] ≥4 Less than primary 8.4 10.0 1 0 29.7 51.9 64.6 Less than primary 24.4 . 19.9 3 .6 31.6 24.1 36.5 Primary completed 23.9 21.6 38.8 1 . 15.8 18.1 Primary completed 44.8 23.7 23.8 7.7 33.0 Secondary completed 8.3 38.3 25.2 24.8 . 11.7 11.6 Secondary completed 59.8 23.2 . 13.1 3.9 21.0 Tertiary completed 58.3 23.7 1 . 13.6 4.4 5.7 Tertiary completed 73.0 19 0 19.0 6.1 1.8 9.5 [0, 10) [10, 20] [20, 30] [0, 10] [10, 20] [20, 30] [30, 40] [40, 50] > 50 [30, 40] [40, 50] > 50 Source: Data from SEDLAC (Socioeconomic Database for Latin America and the Caribbean). Note: Each row (of green-shaded squares) adds up to approximately 100 horizontally. The red column on the right represents the distribution of parental education and thus adds up to 100 vertically. “Educational gap” is defined as the difference between potential years of education at a given age and the years of completed education at that age. almost halved between 1995 and 2009, that an important factor in the reduction of from 65 percent to 36 percent. this gap is a genuine reduction in the depen- • Although 50 percent of the children in dence of children’s outcomes on parental the bottom group of parental background background.4 presented more than four years of edu- cational gap in 1995, by 2009 this share Cross-country heterogeneity dropped to less than 25 percent. • The latter percentage remains far from The evolution across countries of the asso- the share of children with a similar gap ciation between parental education and chil- among parents with tertiary education (a dren’s educational gap at age 15 is presented mere 1.8 percent in 2009), but the abso- in figure 3.7. The common denominator is lute distance between children with highly that, in all countries and periods, the school- educated parents and children with low- ing gap is larger for those children raised in educated parents fell dramatically during households with a low parental background the period. than for those children raised in households with a high parental background. This is true We should note that the closing of the whether we measure parental background schooling gap between the rich and the poor through the parents’ education or their observed in Latin America is not driven income. That is, no country shows complete solely by the general educational improve- independence of children’s educational out- ment in the population but also by the fact comes with respect to parental background. that the educational gap today is less depen- Encouragingly, for most of the countries dent on parental background than it was a and children’s ages we considered, the dif- decade ago. It is naturally the case that, as ferences in the schooling gap associated with poorer children are able to attend and finish parental education in 2009 are lower than school, the schooling gap between rich and in 1995. We also observe some convergence poor children is bound to drop because richer across countries. Ecuador, Brazil, and Bolivia children cannot go beyond their grade. How- (in that order) are the countries that made the ever, using different metrics, we have found greatest progress in reducing the children’s MOBILIT Y ACROSS GENERATIONS 59 FIGURE 3.7 Impact of parental background on children’s educational gap at age 15 in Latin America, 1995–2009 0.80 0.60 0.40 0.20 Years of education 0 –0.20 –0.40 –0.60 –0.80 –1.00 –1.20 r il ia r ile a ico ca a as lic a ru ay B a y do do ua az bi m in gu ,R liv Pe ur ub Ri gu Ch ex nt m na Br ua lva ug ela ra Bo nd sta ep ra lo ge M Pa ca Ec Ur Sa zu Ho Pa Co nR Co Ar Ni ne El ica Ve in m Do circa 1995 circa 2009 circa 2009 – circa 1995 Source: Data from SEDLAC (Socioeconomic Database for Latin America and the Caribbean). Note: “Educational gap” is defined as the difference between potential years of education at a given age and the years of completed education at that age. The green and orange bars represent the expected reduction in the schooling gap associated with one standard deviation of parental education in 1995 and 2009, respectively. The red bar is the difference between the two. Other covariates in the regression are children’s gender, living in an urban area, and country fixed effects. The estimated effect of parental education on the educa- tional gap is always statistically different from zero and so are the differences between 1995 and 2009. educational gap associated with parental edu- captured by the income or education of the cation, but their starting levels of inequality parents. Ethnicity has been and remains a of opportunity in this particular dimension significant source of disparity in Latin Amer- were among the highest in 1995. Uruguay is ican countries (Justino and Acharya 2003; the country that made the least progress dur- Busso, Cicowiez, and Gasparini 2005; Chong ing the period, but it had started from a fairly and Ñopo 2008) and hence is likely to be an low level of inequality associated with paren- important determinant of the children’s edu- tal background. Most saliently, throughout cational attainment (Cruces et al. 2011). this period, Chile managed to reduce the dif- Our next exercise concentrates on three ferences in schooling gaps across socioeco- countries where existing microdata sets nomic groups almost completely: by 2009, allow for a similar definition of ethnicity: one standard deviation in parental education Brazil, Ecuador, and Guatemala. In the three is associated with less than 0.1 years of addi- cases, we identify ethnic minorities as those tional schooling gap. who define themselves as nonwhite.5 Ethnic minority households tend to be concentrated at the bottom of the income and education The role of ethnicity distribution. Therefore, taking into consider- The role of parental background in the ation the previous analysis, we would expect determination of their children’s educational ethnic minority children to present lower attainment probably goes well beyond that educational attainment levels. Our interest 60 MOBILIT Y ACROSS GENERATIONS FIGURE 3.8 Impact of ethnic minority status on children’s lies, however, in determining whether chil- educational gap in Brazil, Ecuador, and Guatemala dren of ethnic minority groups are doing worse in schools than white children, once a. Brazil differences associated with parental income 0.5 and education have been taken into account. 0.4 Hence, our results show the impact of eth- 0.3 nicity above and beyond those of parental Years of schooling 0.2 income and education. 0.1 0 Once we control for parental schooling –0.1 and income, ethnic minority children are less –0.2 likely to succeed in school. Yet the impor- –0.3 tance of ethnicity in the determination of –0.4 children’s schooling gaps has declined over –0.5 1990 2009 Difference the past few years in the three countries stud- Year ied (see figure 3.8). In Brazil, the educational Age 10 Age 15 Age 18 gap associated with ethnic minorities was cut by about 50 percent within all age categories b. Ecuador between 1990 and 2009. The case of Guate- 0.8 mala is even more remarkable: between 2000 0.6 and 2006 the ethnic gap almost disappeared 0.4 for children aged 10 and 15. But perhaps the Years of schooling 0.2 most striking case is that of Ecuador, which 0 displays in 2009 no significant additional –0.2 penalty associated with ethnic minorities –0.4 within the three age categories considered. –0.6 –0.8 2003 2009 Difference The importance of educational Year achievement Age 10 Age 15 Age 18 Closing the schooling gap between children living in poor households and those from rich households is an important step towards c. Guatemala 1.2 achieving more equal opportunities. How- 1.0 ever, substantial differences may remain 0.8 in the quality of schooling received by chil- Years of schooling 0.6 dren with different parental backgrounds. 0.4 To understand the influence of parents on 0.2 0 student achievement, we use two similar –0.2 cross-country harmonized data sets that –0.3 include detailed information about paren- –0.4 tal background and student test scores. The –0.5 most comprehensive is the 2009 Program 2000 2006 Difference Year for International Student Assessment (PISA), Age 10 Age 15 Age 18 which tests student at age 15 in 65 countries, including Organisation for Economic Co- Source: Data from SEDLAC (Socioeconomic Database for Latin America and the Caribbean). operation and Development (OECD) coun- Note: “Educational gap” is defined as the difference between potential years of education at a given age and the years of completed education at that age. The first two sets of bars represent the tries, nine Latin American countries, and coefficient associated with ethnicity in a schooling gap regression for each country, children’s age, other rich and poor countries outside Latin and year. The last set of bars is the difference between the two years. The regressions include as control variables the maximum education of the parents (and its square), household income (and its America and the OECD (OECD 2011). The square), sex, urban/rural dummy, and regions dummies. second data set used is the Second Regional MOBILIT Y ACROSS GENERATIONS 61 Student Achievement Test (SERCE), an constructing the index and include 11 house- assessment sponsored by the United Nations hold asset holdings and dwelling characteris- Educational, Scientific, and Cultural Organi- tics as well as parental schooling.6 zation (UNESCO) and carried out in 2006 Intergenerational persistence in achieve- in 17 Latin American countries (UNESCO ment in Latin America is fairly high.7 With 2009). The harmonized questionnaires tested the exception of Mexico, the region’s coun- children attending the sixth grade of primary tries do not fare well in terms of independence school. of achievements in secondary education from Both the PISA and SERCE data sets share parental background, either compared with various characteristics and enable the assess- more-developed countries or with countries ment of the relationship between children’s at a similar level of development from other test scores and their parents’ socioeconomic regions. backgrounds. Test scores in both data sets Figure 3.9 presents estimates of the effect are standardized to have an average across of parental background on children’s test countries of 500 points and a standard devia- scores in all 65 countries and economies tion of 100. Parental background is measured included in the PISA sample. In Argentina, through an index of economic, social, and Peru, and Uruguay, an improvement of one cultural status (ESCS). In the case of PISA, standard deviation of the ESCS of the parents the index includes information on household is associated with an increase of around 45 asset holdings, occupation, and educational points in the test scores, which is about half attainment of the parents (see OECD 2011). a standard deviation of test scores included In SERCE, we follow a similar procedure for in the sample. Brazil, Chile, Colombia, and FIGURE 3.9 Influence of parental background on secondary students’ PISA test scores across countries and economies, 2009 50 40 30 Test score 20 10 0 ng do ina A , C ia ch bai a te jan Q in Es atar Fin nia Ice and Ko Tun nd a, a La p. No tvia ail y Jo and Ca dan Se da M rbia De Jap o nm an Al ark ss Ka Cro nia Ky Fe hs tia yz ra n pu n th Spa ic er in hu ds It a Sw Tu aly er y Gr and Tr Cz R ed e in ec om en Slo and pu ia va To blic ep go ve c Po nia rtu d Au Ch l str ile Ire alia d B nd Co gdo il lo m Ge ust a rm ria S an ny x a a it o e Ne ed S urg Ze tes Be land m ug el ng y Ar P y ge eru Bu ina ia ga Slo bli Th wa itz rke Hu ua ar n z Lie zer hin re isi i A bi Lu ing am Sw eec Un embpor ic rg de ta Re tio Po lan bl R s an ad R n ar Ur Isra Re te iu Ki ra Lit lan P a na ian za a k R ba la la SA ne ex w ta Ko In , Ch to ba id h a nt m u lg l r l lg ns r a k e R SA Ne ite ao Un ac M Ru ng Ho Source: PISA 2009 data. Note: ESCS = PISA index of economic, social, and cultural status; PISA = Program for International Student Assessment. The bars represent the effect on reading test scores of one standard deviation of change in the ESCS index. Other variables included in the regression as controls include gender of the pupil, urban/rural dummy, and immigration status (first- or second-generation). Standard errors are clustered at the school level, and in all cases, the estimated effects are significant at the 5% level. 62 MOBILIT Y ACROSS GENERATIONS Panama show slightly more intergenerational parental ESCS on children’s reading tests mobility but are well below the cross-country across countries and economies. PISA average. This result is consistent with There is no clear association between the findings in Ferreira and Gignoux (2011), two measures. In other words, there does which uses the R-squared of a similar regres- not seem to be a clear trade-off between sion to gauge the extent of (in)equality of efficiency, as measured through average test opportunities in Latin America with respect score, and equity. The vertical and horizon- to other regions. tal lines show the cross-country averages for The message regarding achievements is both axes. With the exception of Mexico, particularly worrying for two reasons: First, Latin American countries are clustered in the not only is equity low in Latin America but southeast quadrant of the graph, character- average performance is also quite poor by ized by lower-than-average performance and international standards. Second, the esti- lower-than-average intergenerational mobil- mates of parental background in the stu- ity (in other words, higher-than-average dent test scores from developing countries in impact of parental ESCS on the children’s data sets like PISA or SERCE are likely to be test scores). downward-biased. We elaborate on both of The importance of looking at these two these points in turn. dimensions together can be illustrated with The high influence of parental background a couple of examples. In absolute terms, the on student achievements in Latin America is impact of one standard deviation of parental coupled with low levels of efficiency. This background on the test scores of the chil- combination amplifies the troubling nature dren in Panama and Germany is similar. of intergenerational mobility in the region. However, the average test score in Panama is Figure 3.10 shows the bivariate association around 360, while in Germany it is almost between average scores and the impact of 500. Hence, in relative terms, the impact of parental background on the test scores is much larger in Panama. Similarly, the impact FIGURE 3.10 Relationship of average PISA test scores and of parental background on the test score in intergenerational mobility across 65 countries and economies, 2009 Argentina is almost three times as large as the one measured in Indonesia, even if both 550 FIN KOR countries present similar average test scores. CAN Hence, in relative terms, the mobility gap 500 DEU USA between the two countries is larger than in absolute terms. Average test score 450 CHL In the previous section, we saw that in Better performance MEX URU all Latin American countries there is a posi- TTO IDN COL tive impact of parental background on chil- 400 BRA ARG dren’s educational attainment, as measured PAN PER by schooling gaps. This positive impact has 350 two components: children from better-off backgrounds are less likely to drop out from 300 school and, among those who remain in 10 20 30 40 50 school, they are also less likely to fall behind. Effect of socioeconomic background on reading test scores Both effects are confirmed by the data. This More mobility poses a serious challenge to the investigation of educational achievement using standard Source: PISA 2009 data. surveys such as PISA and SERCE because Note: PISA = Program for International Student Assessment. The effect of socioeconomic back- these data are representative of the popula- ground on reading test scores is calculated as described in figure 3.9. The horizontal line represents the average test score in the sample. The vertical line represents the average effect of parental tion of children attending school at a given background on test scores in the sample. age (in PISA) or in a given grade (in SERCE) MOBILIT Y ACROSS GENERATIONS 63 but not of the entire child population of that In countries where almost all children age.8 When working solely with OECD coun- attend school, the lower and upper bounds tries in the PISA sample, this is not a great of the differences in sixth-grade test scores limitation because enrollment rates at age 15 are close to each other. In Chile, for instance, are high (Hanushek and Woessman 2011). children whose parents completed tertiary However, in developing countries, the enroll- education score around 120 points more ment rates are much lower, especially in sec- (that is, more than one standard deviation) ondary education. than those whose parents have no education The implication is that estimates of the (figure 3.11, panel b). Although this test score association between parental background difference is large, the bounds around it are and test scores for developing countries are quite tight. likely to be downward-biased due to selec- In contrast, in countries where a significant tion. As a result, the gap in equity between proportion of children do not attend school, high-income and low-income countries is the distances between the lower bound of the most likely larger than observed. Indeed, this gap and the upper bound can be quite large. will be the case if the following three condi- The extreme example is Guatemala, where a tions are satisfied: 12-year-old child born from tertiary-educated parents will most certainly attend school at 1. Enrollment rates in developing countries age 12, as is the case in Chile for any kid. are lower than in developed countries. However, if the child had parents with no 2. The probability of attending school education, he or she would have only a 60 increases with parental background. percent chance of being at school (figure 3.11, 3. Children who do not attend school per- panel a). As a consequence, the difference in form no better than those with similar performance between children from parents backgrounds who do attend school. with no education and those with tertiary education lies somewhere between 80 and Conditions 1 and 2 are indeed confirmed 180 test-score points (that is, between one by the data. We have no direct evidence of and two standard deviations). Importantly, condition 3, but it is reasonable to believe it although the estimate from the distribution of is also met. those children attending school indicates that Lugo and Messina (2012) discuss the prob- the gap in Guatemala is among the smallest lem of selection into schools and propose a in the region, the bounds clearly suggest that correction based on building bounds around it is possible that, instead, the gap between the estimated effects of parental background the highest and lowest parental background is on children’s test scores in the case of Latin indeed the largest of all (figure 3.11, panel b). American countries.9 The authors combined For PISA estimates, bounds are much information (from national household sur- wider than in SERCE. However, in most of veys, PISA, and SERCE) on enrollment rates the cases, the estimated effects without tak- and test score data for children from differ- ing into account the sample selection are very ent socioeconomic backgrounds to construct close to the lower bound. Considering that reasonable lower and upper bounds of test enrollment rates are lower in Latin American scores for the nonobserved population. (For countries than in most OECD countries, this a detailed explanation and results for PISA, implies that the distance in intergenerational see focus note 3.1 at the end of this chapter.) mobility between the two groups of countries We present here results for SERCE, which is even larger than what we have discussed measures cognitive development in children so far. attending sixth grade of primary school, and Do we observe similar improvements focus on the gap in test scores between chil- in intergenerational educational mobility dren with high-educated parents (tertiary) when we consider achievements rather than and low-educated parents (no education). educational attainment? Unfortunately, the 64 MOBILIT Y ACROSS GENERATIONS FIGURE 3.11 Enrollment and inequalities in reading test scores, selected countries, 2006 a. Proportion of children (age 11–12) attending third grade b. Difference in reading test score between median child or above, by parental background whose parents have tertiary education and median child whose parents have less than primary education Chile 200 Argentina 180 Uruguay 160 Panama 140 Mexico 120 Test score Dominican Republic Costa Rica 100 Paraguay 80 Ecuador 60 Colombia 40 El Salvador 20 Brazil 0 Nicaragua Ni Cu ic ca ba El lom ua lv ia Co cuad r sta or M ica ra co Ar Ch y ge ile na a Pe a Br ru Gu rug zil em y ala E ado a at ua Pa in m l Sa b U a ub gu Pa exi Co g R nt a ep r Peru nR ica Guatemala in m Do 0 0.2 0.4 0.6 0.8 1.0 Total Tertiary Secondary Range (lower and upper bounds) Point estimate Primary Less than primary Source: Lugo and Messina (2012); SEDLAC (Socioeconomic Database for Latin America and the Source: Lugo and Messina (2012), based on SERCE 2006 and SEDLAC (Socioeco- Caribbean) data. nomic Database for Latin America and the Caribbean) data. Note: Range represents upper and lower bounds. The dots are the point estimates. evidence is much scarcer. The PISA study • There is some indication that the scores started collecting information in 2000, so we from the worst-performing children in could in principle discuss evidence of the evo- PISA 2000 were left censored, diminish- lution of the socioeconomic gradient in test ing the variance in the test scores. scores in the past decade. However, several data limitations make this comparison quite The first and last concerns are likely problematic: to bias the comparison of 2000 and 2009 scores against finding improvements because • We have observed an important improve- they suggest that the impact of parental ment in educational attainment during background on test scores in PISA 2000 is the past decade in Latin America. Hence, likely to be downward-biased with respect raw estimates of the changes in the socio- to 2009. Keeping all these caveats in mind, economic gradient of student achieve- there is no evidence of an improvement in the ments are likely to be affected by changes socioeconomic gradient of student achieve- in selection over time. ments in the past decade in Latin America. • The index of socioeconomic status was Estimates relying on the same methods used constructed differently in the 2000 and to construct figure 3.9 for the few countries 2009 waves of PISA. where data is available show very limited MOBILIT Y ACROSS GENERATIONS 65 changes over time, and sometimes even nega- FIGURE 3.12 Intergenerational earnings elasticity between tive ones.10 fathers and sons and its relationship to earnings inequality From educational to income a. Intergenerational earnings elasticity mobility Peru China In the previous sections we have seen that Brazil intergenerational mobility in educational Chile attainment and achievement in the Latin United Kingdom Italy American region is generally low compared Argentina with other regions—although, at least in United States the case of attainment, it has improved in Switzerland the past decades. To the extent that the level Pakistan Singapore of education attained is related to incomes France and wealth later in life, one would wonder Spain whether similar improvements in income Japan mobility are observed. The present section Germany New Zealand tries to shed some light on this question. Sweden To study intergenerational persistence of Australia income, one would ideally need income data Canada for parents and their children (once they are Finland Norway adults). Indeed, these types of data have been Denmark used to study mobility in high-income coun- 0 0.20 0.40 0.60 0.80 tries, as described in box 3.2. Unfortunately, Intergenerational earnings elasticity no Latin American country has the similar long-term panel surveys needed to perform similar analyses. One possibility to overcome b. Relationship of intergenerational earnings this limitation is to use retrospective infor- elasticity to earnings inequality 0.8 mation on parental background such as edu- Intergenerational earnings elasticity cation and occupation to build estimates of parental income, which can be related in a Peru second stage to the income of the children. 0.6 Brazil In a recent paper, Corak (forthcoming) com- Italy United States Chile United Kingdom piled methodologically comparable estimates Pakistan Switzerland Argentina of fathers’ and sons’ earnings mobility across 0.4 France Spain a large host of countries, using data from 19 Japan Germany studies. In most cases, estimates of paren- Sweden New Zealand Australia tal income are obtained using retrospective 0.2 Finland Norway Canada information on parental education and occu- Denmark pation. Earnings mobility is computed as the 20 30 40 50 60 elasticity of earnings between the two gen- Inequality (Gini coefficient) erations when both are of similar age. How does Latin American generational Source: Corak, forthcoming. income mobility compare with that of other Note: Bars shaded in orange are not OECD member countries in panel a. countries? Figure 3.12 presents country esti- mates from Corak (forthcoming) in panel a, together with its relationship to the cur- among the economies with the highest lev- rent level of earnings inequality (panel b). els of intergenerational dependence. In Peru, Two striking features emerge from these for instance, if one father earns 100 per- figures. First, Latin American countries are cent more than another, then the son of the 66 MOBILIT Y ACROSS GENERATIONS BOX 3.2 Income mobility in high-income countries The importance of family background for chil- siblings can thus be considered as a lower bound dren’s economic status in developed countries has on the impact of family background on economic been studied intensively in Nordic countries and status. When analyzing international sibling cor- in the United States, relying mostly on three dif- relations of earnings in Nordic countries and the ferent approaches: intergenerational associations United States, Björklund and Jäntti (2009) show that of income, sibling similarities, and the relationship Norway stands out as having much smaller correla- between inequality of income and unequal opportu- tions than Denmark, Finland, and Sweden: 0.14 for nities. The results of these studies, summarized by Norway compared with around 0.25 for the other Jäntti (2012), show a similar pattern across meth- countries. At the opposite extreme, the correlation odologies, revealing evidence against the traditional in the U.S. is approximately 0.50, that is, more than notion of “American exceptionalism”—which con- double the Nordic countries’ average. sists of, among other aspects, a belief in a greater rate An alternative approach to examining the impor- of upward social mobility in the United States than tance of family background on people’s income is in other countries. The discomfort with the “end of present in the literature of equality of opportunities. the American dream,” made explicit by the Occupy This research has been inspired by developments in Movement, is probably not all that surprising. political and social philosophy (see Arneson 1989; Analyzing the joint distribution of fathers’ Cohen 1989; and Roemer 1993, 1998). In differ- and sons’ income (intergenerational associations) ent ways, these authors argue that not all inequal- through quintile group mobility matrices, Jäntti ity need be ethically unacceptable (see Almås et al. (2012) shows that, for the Nordic countries, approx- 2011). Individuals should be held accountable for imately 25 percent of sons born into the poorest outcomes for which they can be held responsible, quintile remain in that position, while around 10 to such as level of effort exerted at work, but not for 15 percent reach the very top quintile. In contrast, those outcomes for which they are not responsible, the author fi nds that more than 40 percent of U.S. such as the color of their skin or the place of birth. males born in the poorest quintile remain there, Consequently, inequalities due to effort are viewed reflecting a much lower upward mobility. Similarly, as ethically acceptable, whereas inequalities related the probability that the son of a lowest-quintile par- to circumstances beyond individuals’ control are, in ent makes it into the top-quintile group is lower in turn, ethically unacceptable. the United States than in all of the Nordic coun- Often, groups are defined by a few observable tries, and the top-to-bottom mobility is also lower parental traits—typically parental income, occu- in the United Kingdom and the United States than in pation, and education (see Roemer et al. 2003)— Nordic countries, as shown by Atkinson (1981) and to compute the extent to which tax-and-transfer more recently by Jäntti (2012). Fewer than 10 per- regimes in 11 rich countries equalize opportunities cent of U.S. males born into the richest quintile fall among citizens for income acquisition. In this con- all the way down to the bottom quintile, while this text, equality of opportunity in incomes is achieved is typically the case for around 15 percent of Nordic when the distributions of postfi scal income are the males. In more central parts of the income distribu- same for different “types” of citizen (“types” being tion, all countries are remarkably similar. defined according to parental socioeconomic sta- A fuller account of the impact of family back- tus). The authors find that high-income countries ground on economic status can be found by study- tax income at close to, and sometimes possibly in ing the extent to which siblings’ economic charac- excess of, equality-of-opportunity norms. These teristics resemble each other. Siblings share part of results are particularly present among northern the attributes that parents transfer to their children, European economies—such as Sweden and Den- which are partially related to income, such as values mark—which tended to do well in achieving equality and aspirations. The correlation in income between of opportunity. MOBILIT Y ACROSS GENERATIONS 67 high-income father will make, as an adult, 3.1 to estimate the E coefficient as a measure 67 percent more than the son of the rela- of intergenerational persistence.11 tively lower-income father. This is in sharp Confirming previous evidence, the authors contrast with the elasticity found, say, in find that the association between socioeco- Norway (0.17) but also in Spain (0.40) or the nomic status of parents and children in both United States (0.47). countries is fairly high, with an intergen- The second interesting result from Corak erational elasticity of 0.56 in Colombia and is that low generational mobility goes hand 0.48 in Mexico. Both associations are highly in hand with high inequality. Countries significant. However, the two countries with low mobility (high intergenerational have followed different trends. In the case of earnings elasticity) tend to have high lev- Colombia, we observe an increase in intergen- els of inequality. (See, in particular, that all erational mobility across cohorts. The elas- four of the Latin American countries in the ticity of the index of socioeconomic status is sample appear in the far right side of panel 0.66 for the oldest cohort (ages 56–64), while b.) In contrast, highly mobile societies are the elasticity for those age 41–55 is 10.0 per- also the ones presenting the lowest levels of centage points lower, at 0.56, and in the case cross-sectional inequality (not only lower of the youngest cohort it declines to 0.47. In lifetime inequality). In the next section, we the case of Mexico, however, the association will describe some of the potential underly- in the asset index is larger for the younger ing causes of this observed relationship. We cohort, at 0.56, against 0.43 for the older one. examine whether specific market structures and policies that are correlated with chil- dren’s unequal access to education interact Policies and intergenerational with the level of education or wealth of their educational mobility parents. What are the determinants of cross-country Examining changes over time in the inter- differences in intergenerational mobility? Are generational association of income is an even there institutional factors, policies, and mac- harder task. In an attempt to shed some light roeconomic environments that favor mobility on this issue, Azevedo et al. (2012) use retro- across generations? These are extremely dif- spective information from household surveys ficult questions, but they remain at the core in Colombia and Mexico. These surveys ask of the policy analysis of intergenerational all household heads and their spouses about mobility. As we discussed in the introduction the households’ possession of a number of to this chapter, the transmission of socioeco- assets generally associated with household nomic status from one generation to the next wealth at two points in time: when they is a combination of exogenous biological fac- were 10 years old and at the time the survey tors, endogenous optimizing behavior of par- is administered. Combined with the educa- ents, macroeconomic or environmental con- tional attainment of adults in the household ditions, and collective policies (Solon 2004). and their parents, one can construct two However, none of these factors operates in indices of socioeconomic status, one for each isolation. The optimizing behavior of par- generation, using a methodology similar to ents and policies, for instance, interact with the one described in previous sections (see each other, complicating the interpretation of Azevedo et al. 2012 for details). This method estimates of intergenerational mobility in a allows the researchers to assess the extent to cross-country context. which the position of children in the overall The literature on the determinants of wealth distribution is associated with the intergenerational mobility is rather thin, position their parents held. Such an associa- which is not surprising considering the tion is examined employing a similar meth- extreme complexity of simply measuring odology to the one previously outlined in box this social phenomenon. There is a limited 68 MOBILIT Y ACROSS GENERATIONS literature that looks at correlations of poli- to identify the causal impact of a particu- cies and institutional variables and different lar policy intervention on intergenerational proxies of intergenerational mobility across mobility. It should be noted that even this countries, including Woessmann et al. (2009) analysis is not exempt from the typical limi- and Causa and Chapuis (2009), both focused tations of quasi-experimental settings. The on OECD countries, and Ferreira and Gig- fact that some policy worked in a particu- noux (2011) for both OECD and non-OECD lar context does not guarantee that it would countries. In this section, we extend this lit- have the same impact in a different macro erature to consider a larger set of countries or institutional environment. For all of these and indicators of intergenerational mobility. reasons, rather than providing definitive Box 3.3 briefly outlines the methodology. answers, the aim of this section is to high- We complement the general overview of light particular aspects of the institutional the role of policies (provided with the cross- and policy environments that appear to be country analysis) with selected examples important for the determination of mobility where sharp policy changes have helped across generations. BOX 3.3 Cross-country analysis of policies and institutions and intergenerational mobility Cross-country comparisons can shed light on the ESCSic is an index of socioeconomic status that cap- importance of policies and institutions in the deter- tures a wide array of parental background charac- mination of intergenerational mobility, but they teristics, including education, occupation, and some present clear limitations. Countries are very dif- tangible asset holdings (the exact variables changing ferent in many respects. Moreover, institutional with the data set). We will consider two outcome aspects are highly correlated within countries. variables (yic): test scores from PISA and the school- Both aspects make it almost impossible to isolate ing gap. Importantly, data on schooling gaps are the impact of a particular policy on the outcome of available for repeated cross-sections in each country. interest. Moreover, in many instances, institutional In this case, we add to equation (B3.3a) time dum- indicators are imperfect measures of the complex mies that capture common factors that shock coun- real-life phenomena they are meant to summarize. tries. For schooling gap regressions, standard errors The increasing availability of panel data can help are robust to country-year. In PISA, instead, stan- us start to compare apples to apples. Relying on dard errors are robust to correlation within schools. changes in policies and institutions within countries Our interest lies in the impacts of policies (Pc) on brings us closer to identifying their actual impact, the socioeconomic gradient of the test score or edu- even if they are not the definitive answer since cational gap. Thus, we are interested in comovement in other confounding factors limits the ∂yic = γ + γ P . power of identification. (B3.3b) ∂ESCSic 1 2 c Here, we estimate cross-country regressions of the following format: This equation indicates that socioeconomic back- ground is associated with the educational outcome yic = D + EX ic + J1 ESCSic directly and through the level of the policy vari- + J2(ESCSic × Pc) + Pc + Hic , (B3.3a) able or institution we are considering. A positive (negative) J2 implies that in countries with higher where subscript ic represents individual i living in provision of the policy, the educational outcomes country c, Xic is a set of control variables, and Pc is a of children are more (less) strongly associated with country fi xed effect. parental background. MOBILIT Y ACROSS GENERATIONS 69 Expenditures on education and FIGURE 3.13 Impact of public education expenditures on the educational outcomes schooling gap between rich and poor We start the analysis by focusing on expen- ditures. There is an important debate in the 0.5 literature on the impact of public education expenditures on educational outcomes. Har- 0 bison and Hanushek (1992) review 12 case –0.5 studies in developing countries, and only half of them reported a positive association. Raj- –1.0 kumar and Swaroop (2008) argue that dif- ferences across countries may be related to –1.5 differences in governance. In countries where corruption is high and the quality of bureau- –2.0 cracy is low, higher spending in education or health may not need to be translated into bet- –2.5 ter outcomes. They also offer cross-country Age 10 Age 15 Age 18 evidence that is in line with the proposed SD of low expenditure Difference hypothesis. SD of high expenditure The debate about the efficacy of public spending in education is important for the Source: Data from SEDLAC (Socioeconomic Database for Latin America and the Caribbean). Latin American region because even if public Note: SD = standard deviation. expenditures in education are still low today The green and orange bars represent the effect of a one-SD increase in parental education on schooling gaps in countries with low and high levels of public expenditure on education per pupil relative to other developed and developing as a percentage of GDP, respectively. The estimated effects are obtained from three independent countries, it is undeniable that an important regressions (one per age), including Latin American household surveys from the period 1995–2009. The dependent variable is the student gap, and the coefficients of interest are the highest level of effort was devoted to “catch up” during the parental education and its interaction with public education expenditures (see box 3.3 for details). past decade. According to UNESCO data, Regressions include as control variables gender of the student, urban status, country, and time fixed effects. Standard errors are clustered by country. All estimates are statistically different from zero at between 1999 and 2009, expenditures per standard levels of testing. student in primary and secondary education as a percentage of gross domestic product (GDP) per capita in the region grew by almost The differences in the gap between a high- 50 percent—from 10.5 percent to 15.1 percent spending country (see the orange bar, corre- in the case of primary education, and from sponding to Costa Rica) and a low-spending 11.9 percent to 16.9 percent in secondary edu- one (the green bar, corresponding to Peru) is cation. Has this increase in resources helped 0.08 years of schooling at age 10. At age 15, close the gap in attainment and achievements differences in the gap between low-spending between the poor and the rich? (Dominican Republic) and high-spending We pool individual data from different (Brazil) country-years are 0.18, and at age countries and years (1990–2009) with pub- 18 the gap differs by 0.28 years of education. lic expenditure data on education expendi- Although these differences are sizable (and tures in primary and secondary education, statistically significant at 1 percent), it is clear measured at the country level to estimate that they explain only a small part of the the association between public expenditures reduction of the gap observed in the region and individual schooling gaps, following throughout the period. the methodology explained in box 3.3. We The effects of parental background are find that public expenditures per student in evaluated at different levels of public expen- primary and secondary school have indeed ditures. In the case of children aged 10, helped to reduce the schooling gap between expenditure levels in primary education rich and poor children (figure 3.13). in 2006 are considered. Low expenditures 70 MOBILIT Y ACROSS GENERATIONS correspond to Peru, and high expenditures as Latin America gradually closes the educa- correspond to Costa Rica. For children aged tional gap in primary and secondary school- 15 and 18, expenditures in secondary educa- ing, the importance of credit constraints in tion in 2006 are considered. High and low limiting intergenerational educational mobil- expenditures correspond to Brazil and the ity is likely to increase. Moreover, the recent Dominican Republic, respectively success in Latin America in closing the gap Higher public educational expenditure in secondary education, and the resulting might be associated with a lower schooling increase of workers with high school cre- gap for several reasons. Schools receiving dentials in the labor market, is reducing the more funding can devote more resources to returns to secondary education (see Aedo the poor and those students lagging behind, and Walker 2011). Hence, limited access to either in the form of physical inputs (such as credit, by limiting access to tertiary educa- books and stationery) or by providing special tion, is likely to be hindering those human tuition and extra-support lessons. Higher capital investments that present the highest investments also can result in better school marginal returns in Latin America. infrastructure in remote areas, where poor The impact of public educational expen- families are likely to be concentrated. This is ditures on educational achievement—as probably particularly important for enroll- opposed to educational attainment—is ment and attainment in secondary schools, ambiguous. In this case, we work with where the opportunity cost of studying is cross-sectional data only (PISA) and the full higher, especially in families where credit sample, which includes the 50 countries for constraints are important. which we have expenditure data. On aver- Gauging evidence of the particular chan- age, we find that higher public expenditures nels through which additional public spend- per pupil are associated with a larger gap in ing influences educational attainment is test scores between the rich and the poor, but hard at the macro level, but some lessons this result is not uniform across countries. can be learned. Regarding the importance In developing countries, public expenditures of resources, we have included in our cross- appear to be mildly progressive,12 whereas in country regressions pupil-teacher ratios developed countries, larger gaps are associ- and found that assigning fewer students per ated with more spending. teacher helped to reduce the gap between the rich and the poor in primary schools, The importance of school quality and although not in secondary schools. Box 3.4, children’s sorting in intergenerational drawn from Solis (2011), presents evidence mobility from a quasi-natural experiment in Chile on the importance of credit constraints for Affluent parents interested in providing the enrollment in tertiary education. The impact best possible education for their children are of having access to student loans for the poor likely to send them to better schools even is hard to dispute: among those eligible for when this implies paying an additional fee, a student loan, having access to credit after an option that may not be available for par- completion of high school completely elimi- ents with fewer resources and limited access nates the enrollment gap in college between to credits. The implication for analyses using the rich and the poor. assessment data such as PISA or SERCE As we argued before, we expect credit is that it will be difficult to disentangle the constraints to be a more important factor impact of school and parental inputs because limiting access to education as we move up children are not randomly allocated across the educational ladder. Higher education is schools. Instead, in all educational systems more expensive than secondary education, where parents have some freedom of choice, and the opportunity cost of studying at age there is sorting across schools according to 18 is higher than at age 14. This implies that parents’ and children’s preferences. Indirect MOBILIT Y ACROSS GENERATIONS 71 BOX 3.4 Tuition loans in Chile: Is the alleviation of credit constraints a good policy to close the gap in educational attainment between rich and poor? Tertiary education is costly for the students. Costs students scoring in the vicinity of the threshold, include not only student fees, which vary greatly say, one or two points above or below, are likely on across countries, but also the forgone income associ- average to be very similar. If students immediately ated with the loss of hours worked. Naturally, fami- above the threshold present a higher probability of lies in need may not be able to afford this income enrollment in college than those immediately below, loss, and difficult access to credit can make the this can be interpreted as a strong sign of credit option of a college education simply unaffordable for constraints. the most vulnerable households. However, assessing In fi gure B3.4.1, panel a, the author shows that the importance of credit constraints for tertiary edu- the program was properly implemented; only those cation is not straightforward, and the evidence until students who obtained a PSU above 475 were eli- recently has been highly inconclusive. Keane and gible for the loan. Importantly, the loan take-up Wolpin (2001); Card (2001); Carneiro and Heckman among those who were eligible was fairly high (2002); and Brown, Scholz, and Seshadri (2009), (panel b), at 30 percent, and increasing with the among others, rely on different identifi cation tech- PSU score. Panel c shows that college enrollment niques to obtain little agreement on the importance increases sharply, on average, for those who obtain of credit constraints for the intergenerational mobil- a PSU score above 475. The probability of enroll- ity of education. Still, none of these papers has been ment is around 17 percent for the students who are able to observe credit constraints directly. immediately below the 475 threshold. In contrast, The difficulty in assessing the importance of the probability of enrollment jumps to 35 percent credit constraints lies in the fact that the intention of for those immediately above the threshold. Hence, accessing credit is generally not observable. Credit- having access to the loan doubles the probability of constrained families might not apply for a loan if enrollment. Solís (2011) interprets this as strong evi- they think the likelihood of obtaining it to be very dence of binding credit constraints. low. In parallel, children from poor families might Most crucially, the loan appears to help reduce also be less likely to enroll in college because they the gap between the rich and the poor. Panel d lack the necessary qualifi cations or hold different shows the probability of enrollment by income quin- values transmitted by their parents, among other tile for two groups of students: those who score 475 reasons. A recent paper by Solís (2011) takes advan- or 476 on the PSU and thus are eligible for the loan tage of a natural experiment to isolate the impact (the treated group) and those who obtained a PSU of credit constraints on the probability of attending score of 474 or 473 and hence are ineligible (the college and to study differences in such probabilities control group). Several aspects are worth noting: across poor and more affluent families. Enrollment rates among the control group increase The experiment is simple. A fi nancing program monotonically with income. In contrast, enrollment in Chile offers tuition loans to students who fulfi ll rates among those with access to the loan seem to be three criteria: (a) apply for the loan, (b) belong to the independent of their income level. Hence, for those lowest four income quintiles of the income distribu- students who are around the cutoff—that is, those tion, and (c) obtain a score above 475 points on the who obtain a PSU score of around 475—the inclu- College Admission Test (Prueba de Selección Uni- sion into this program completely eliminates the versitaria , PSU). The structure of the loan is such gap of college enrollment by family income. This, of that it creates a sharp discontinuity. Similar stu- course, does not mean that the program wipes out dents who apply for the loan might be successful by all influences of family background into children’s just one-point difference in the PSU score. Students access to higher education in the population because obtaining a very high score (say, 550) and students children from affluent families are more likely to obtaining a very low score (say, 400) are likely to obtain a higher PSU score than are children from be very different in many observed and unobserved poorer backgrounds. Indeed, the mean test score of characteristics and would have also different incen- children in the top quintile was 527, whereas it was tives to go to college. Instead, one may argue that 468 for those in the bottom quintile. (Box continues next page) 72 MOBILIT Y ACROSS GENERATIONS BOX 3.4 Tuition loans in Chile: is the alleviation of credit constraints a good policy to close the gap in educational attainment between rich and poor? (continued) FIGURE B3.4.1 Tuition loans and school enrollment in Chile a. Loan eligibility all years b. Loan taken up among eligible all years 0.6 0.4 f college loan eligibility Probability of loan take up 0.3 0.4 0.2 0.2 Probability of 0.1 0 0 450 460 470 480 490 500 450 460 470 480 490 500 PSU score PSU score c. College enrollment all years d. Enrollment treat versus control by quintile, 2007–09 0.5 0.5 f college enrollment 0.4 Enrollment rate 0.4 0.3 0.3 0.2 Probability of 0.1 0.2 0 1 2 3 4 5 0.1 Income quintile 450 460 470 480 490 500 Enroll control Enroll treat PSU score Source: Solís 2011. Note: PSU = Prueba de Selección Universitaria (College Admission Test). evidence about the extent of parental influ- that are inheritable, effort, and time spent ence that is channeled through the schooling with the children, but also (b) indirect effects system can be obtained by comparing the through parental choices and investments, amount of sorting in different regions. including the type of schools the children are Is there more sorting among schools in attending. It should be noted that although Latin America than in other economic areas? the direct impact estimates the effect of par- A first take at this question compares the ents on children’s test scores that is not con- impact of the socioeconomic status (ESCS) founded with the school, this estimate cannot on children’s test scores as estimated in fig- be interpreted as the “true” impact, precisely ure 3.9 with the impact resulting from add- because the allocation of kids to schools ing to the regression school fixed effects. is not random. For convenience, we group This latter estimation will be labeled as the countries in seven groups:13 “direct” impact of ESCS, as opposed to the first result, labeled “overall” impact, which • Latin America and the Caribbean includes not only (a) the direct association • Anglo-Saxon (excluding United Kingdom from, among other factors, genetic traits and United States) MOBILIT Y ACROSS GENERATIONS 73 • Nordic FIGURE 3.14 Direct and overall impact of parental background on • Continental Europe children’s test scores • United Kingdom and United States • Remaining low- and middle-income countries 40 • Remaining high-income countries. 30 The green bars in figure 3.14 show the Test score overall impact of ESCS on children’s test scores. As we discussed earlier, Latin Ameri- 20 can countries show a high degree of intergen- erational persistence: the impact of one stan- 10 dard deviation in ESCS is higher only in the combined group of the United Kingdom and the United States. At the other extreme of the 0 graph, Nordic countries present the highest m tate ing ic e e e ica m m om op rd do dl s) d lud er co No ur levels of mobility. The orange bars show the nc ng Am in c lE ei ni (ex Ki S gh ta tin direct impact of parents, once school effects d en –U n Hi e te id La m xo t in ni do -sa nt –U w- have been controlled for. Note that Latin Co ng lo Lo es Ki Ang at American countries now fall to the opposite St d ite extreme of the distribution, showing the low- d Un ite Un est influence of parental ESCS on the chil- dren’s test scores. Overall: not controlling for school effects Interestingly, once we control for school Direct: controlling for school effects effects, the Nordic countries appear to have the highest level of persistence across genera- Source: Authors’ calculations from PISA 2009 data (OECD 2011). tions in the six groups of countries consid- Note: ESCS = PISA index of economic, social, and cultural status; PISA = Program for International Student Assessment. The bars represent the effect of one standard deviation in ESCS on children’s ered. Similarly, the overall and direct effects test scores, for each group of countries. Group-specific regressions include as control variables in the United Kingdom and United States are gender of the pupil, urban dummy, immigration status, and country dummies. The orange bar regressions include school fixed effects, whereas the green bars do not. not very different. These findings have two possible interpretations: (a) The first is that, in Latin American countries, there is much more sorting of children across schools than mitigating the existing socioeconomic dispar- in Nordic countries. In this candidate expla- ities, the schooling system in Latin America nation, schooling inputs might even be the would exacerbate them. same across schools. The differences between To shed further light on the importance the green and orange bars simply reflect sort- of sorting versus school inequality, our next ing and not differences in schooling quality. exercise assesses the differences in the inputs (b) Alternatively, the distance between the of the typical schools attended by the rich green and orange bars may be a sign not only and the poor. Our conclusion is that dif- of sorting but also of differences in school ferences between schools attended by the quality between the schools attended by rich rich and those attended by the poor are and poor children. In this alternative expla- larger in Latin America than in any other nation, the fact that there is a much larger group of countries, suggesting that inequal- gap between the overall and direct effect in ity in schooling inputs in Latin America is Latin America than in the Nordic countries likely to play a more important role than in would be a sign not only of higher sorting other regions in widening the achievement but also of the importance of the school- gap between children from different back- ing system in the determination of mobility grounds. Figure 3.15 shows differences in across generations. In this case, rather than the characteristics of the schools frequented 74 MOBILIT Y ACROSS GENERATIONS by the rich (top quintile of the ESCS distribu- This similarity might be related to the fact tion) and the poor (bottom quintile) in Latin that the publicly owned and operated schools American countries compared with those in these two countries operate under rules of schools attended by rich and poor in the and manners similar to those of private United Kingdom and United States (taken schools—a likeness that suggests a role for together) as well as those in other low- and governments in homogenizing standards middle-income countries. that appears to be missing in Latin America. The metric used is the ratio between Some differences are detected in the case of the quintiles; hence, an outcome equal to other low- and middle-income countries, but 1 means that there is an equal proportion these tend to be smaller than those observed of poor and rich children attending schools in Latin America. Nordic countries (not with a particular characteristic. The further shown in figure 3.15) constitute an interest- away from 1 the outcome is in any direction, ing example of equality of inputs: although the greater the difference in that character- rich children outnumber the poor in atten- istic in schools attended by the rich and the dance at privately operated schools, there are poor. no significant differences between private Latin American countries clearly stand and public schools in any of the dimensions out as presenting an unequal distribution of of educational inputs. school characteristics between the rich and the poor. Rich children have a probability of School systems and the influence of attending private-independent schools that is parental background on schooling 22 times greater than that of poor children. If outcomes the school is government-dependent but pri- vately operated, the difference is by a factor Are these schooling characteristics associ- of 10. Such ratios are much lower in all the ated with more intergenerational mobility other regions. The only exception that comes or with less? We have established that school close is in the United Kingdom and United characteristics are not uniformly distributed States. There is an important difference among children with different socioeconomic between Latin America and these two coun- backgrounds, at least in Latin America. In a tries, however. In Latin American countries, different dimension, the importance of each the differences between public and private of these characteristics varies greatly across provision translate into important inequali- countries. Some countries such as the Neth- ties in a wide range of observable school erlands allow for great autonomy of schools characteristics: in hiring and firing teachers. Others, such as Colombia, are much more restrictive in • Schools attended by the rich have much the role they assign to schools in personnel more autonomy in hiring and fi ring teach- policies. Keeping the caveat that the same ers as well as in selecting teachers’ pay. incidence of a given policy in two countries • Schools attended by the rich also have might hide different patterns (for example, more autonomy in selecting the course con- depending on the level of sorting), it is of tents and in administering their budgets. interest to understand the association, on • The percentage of fully certified teach- average, of each of the schooling dimensions ers, a measure of teaching quality, in the on intergenerational mobility. schools attended by the rich is 50 percent To that end, we have built country-level larger than in the schools the poor attend. indicators from PISA using a wide range of the school dimensions from the data set. In contrast, the observable inputs in U.K. Naturally, these variables tend to be highly and U.S. schools are much more similar correlated across countries. To limit the across income groups, with no particular dimensionality problem in the data, we con- aspect standing out as a major difference. ducted a factor analysis of all the institutional MOBILIT Y ACROSS GENERATIONS 75 FIGURE 3.15 Differences in school characteristics between the top and bottom quintile of the ESCS Percentage of private schools Autonomy in establishing teachers’ starting salaries Autonomy in establishing teachers’ salaries increases Percentage of publicly funded private schools Autonomy in firing teachers Autonomy in selecting teachers for hire Influence on staffing decisions Autonomy in determining which courses are offered Autonomy in formulating budget Percentage of teachers fully certified by the appropriate authority Autonomy in determining course content Autonomy in deciding on budget allocations Use of teacher-developed tests at least monthly Autonomy in choosing which textbooks are used Autonomy in establishing student assessment policies Use of student assignments/projects/homework Assessments to improve curriculum 0 1 2 3 Ratio Latin America Low-middle income United States–United Kingdom Source: Data from PISA 2009 (OECD 2011). Note: ESCS = PISA index of economic, social, and cultural status; PISA = Program for International Student Assessment. Each bar shows the ratio in the characteristics of the schools frequented by the rich (top quintile of the ESCS distribution) and the poor (bottom quintile) in Latin American countries as opposed to the United Kingdom and United States (considered jointly). variables in PISA except for three variables examine the impact of parental background that appeared to be crucial, hence warrant- across countries and time as a function of the ing separate appearances in the regressions: age when tracking takes place and the length (a) the share of students attending private of tracking systems in a variety of student independent schools, (b) the share of students outcomes, including educational attainment, attending private but partially government- enrollment in college, employment, train- subsidized private schools, and (c) the amount ing, and wages. They find that tracking has of tracking in the system. a detrimental impact on educational attain- School tracking has been found to be ment among students from lower parental an important obstacle to intergenerational background because it limits the possibility mobility. Using international data sets similar of attending tertiary education. However, to ours from primary and secondary schools, the specialization induced by tracking sys- Hanushek and Woessmann (2006) find that tems appears to reduce the impact of paren- early tracking systems lead to more educa- tal background on learning outcomes among tional inequality because tracking accentu- adults. ates the role of family background on student Most previous studies have looked at a performance. Ammermüller (2005) studies particular form of tracking, namely, the early the importance of the number of school types separation of children among well-defined available in the system, obtaining similar segments in the educational process, typi- conclusions. Brunello and Checchi (2007) cally specializing in general and vocational 76 MOBILIT Y ACROSS GENERATIONS education. This form of tracking is not par- progress from year to year, (c) comparing ticularly prevalent in Latin America. How- the school with others schools, (d) inform- ever, educational systems also differ regard- ing parents about their child’s progress, ing the importance of grouping students by (e) judging teachers’ effectiveness, and (f) ability within schools, and hence this is the identifying aspects of instruction or the dimension of tracking we discuss here. For curriculum that could be improved. this purpose, we construct two variables: • Autonomy regarding students and course (a) class tracking, the percentage of students contents. Higher scores indicate more in each system who are divided by ability autonomy of schools in (a) establishing into different classes within schools for at student assessment policies, (b) determin- least some subjects and (b) grade tracking, ing course content, (c) choosing textbooks the percentage of schools that use student used, (d) determining the courses offered, assessments to decide students’ retention and (e) establishing student disciplinary poli- promotion. cies, and (f) approving students for admis- Considering the importance of sorting sion to the school. across economic background between pub- • Autonomy regarding staff. Higher scores lic and private schools, we also keep the two indicate more autonomy in the follow- variables characterizing the share of private ing areas that involve the management schools in the country separate in the analy- of school staff: (a) establishing teachers’ sis. As before, we distinguish between purely starting salaries, (b) establishing teacher’s private schools (those that are fully funded by salaries increase, (c) firing teachers, (d) hir- private sources) and privately managed but ing teachers, (e) deciding on budget allo- publicly subsidized schools (those that receive cations, and (f) formulating the school some funding from the government). budget. The rest of the variables in the regression analysis are constructed using factor analysis, As comprehensive as this list of variables which reduces the dimensions of the institu- may seem at first sight, our study of the asso- tional variables. The factor analysis of the ciation of policies and the socioeconomic gra- remaining 22 variables suggests the existence dient of educational achievement cover only of four well-identified groups of indicators. a subset of the policies that may be related In what follows, we describe them and list to cross-country differences. In particular, the variables with the larger loadings in each because of data availability, all the poli- of the categories: cies analyzed pertain to interventions at the school level. Even within this domain, some • High frequency of assessment practices. potentially important variables such as those A high score indicates that student assess- related to key infrastructure and equipment ments are done frequently in the schools. are not considered. In rural areas in poor In particular, three variables are summa- countries, school availability of basic infra- rized by this factor: (a) the incidence of structure (such as well-kept roofs and heat- at least monthly teacher-developed tests, ing) and inputs (such as textbooks) may be an (b) incidence of at least monthly teacher’s important constraint on children’s learning. judgment ratings, and (c) incidence of Moreover, our analysis leaves out impor- at least monthly student assignments or tant influences that governments may have projects or homework. on the socioeconomic gradient of education • Accountability. A high score indicates by acting directly at the family level. These higher importance of student assessments include not only taxes and transfers but in the system. Six variables are behind also targeted programs to directly increase this factor: (a) assessments to compare parents’ competencies in an attempt to the school with district or national per- indirectly improve children’s behavior and formance, (b) monitoring of the school’s development. Little is known regarding the MOBILIT Y ACROSS GENERATIONS 77 impact of such programs on the educational their impact on educational achievement is outcomes of the children.14 On the side of still subject to debate (see box 3.5). subsidies, the prominent importance of We proceed to analyze the association conditional cash transfers (CCTs) in Latin between each of these variables and the socio- America during the past decade is undeni- economic gradient of student achievement, able. Although there is a consensus that following the methodology described in box CCTs have been an effective tool in increas- 3.3. In table 3.2, we show regression results, ing educational attainment of poor children, highlighting the interaction term between the BOX 3.5 Conditional cash transfers and children’s educational outcomes Conditional cash transfers (CCTs) are programs include positive effects on schooling, reductions in that transfer cash, generally to poor households, on work for younger youth (consistent with postpon- the condition that those households make specified ing labor force entry), increases in work for older investments in the human capital of their children. girls, and shifts from agricultural to nonagricultural Health and nutrition conditions generally require employment (Behrman, Parker, and Todd 2010). periodic checkups, growth monitoring, and vaccina- In terms of the program’s effect on occupational tions for children less than five years of age, prena- mobility, Oportunidades has a positive effect on job tal care for mothers, and attendance by mothers at insertion because it increases benefi ciaries’ educa- periodic health information talks. Education condi- tional attainment. In addition, those who received tions usually include school enrollment, attendance the transfer for more than six years through primary on 80–85 percent of school days, and occasionally and secondary education received salaries that were some measure of performance. Most CCT programs 12 percent and 14 percent higher, respectively, than transfer the money to the mother of the household the nonbeneficiaries (Rodríguez-Oreggia and Freije or to the student in some circumstances. Interest in 2008). An updated version of this study shows that CCTs has grown enormously in the past 10 years if migrants are included in the analysis, the effect on (Fiszbein et al. 2009). earning could be as high as 44 percent (Rodriguez- In terms of educational outcomes, adults with Oreggia 2011). more exposure to CCT programs have completed In Colombia, an interesting randomized trial more years of schooling than have those with less compared three designs: exposure. There is also some evidence that CCT pro- grams promote cognitive development in early child- • A standard design hood. Rigorous impact evaluations of the Mexican • A design that postponed part of the monthly CCT program Oportunidades (originally known as transfers until children reenroll in school Progresa) indicate that it has significantly increased • A design that lowered the reward for attendance the enrollment of children, particularly girls and but rewarded graduation and tertiary enrollment. especially at the secondary school level. The results imply that children will have an average of 0.7 The two nonstandard designs significantly years of extra schooling because of Oportunidades, increased enrollment rates at both the secondary and although this effect may increase if children are tertiary levels while delivering the same attendance more likely to go on to upper secondary school as a gains as the standard design. Postponing some of the result of the program. Using panel data for Mexico attendance transfers until the time of reenrollment for 1997–99 (Behrman, Gaviria, and Székely 2001; appeared particularly effective for the most at-risk Skoufias and Parker 2001; Schultz 2004), it is shown children (Barrera-Osorio et al. 2011). that Oportunidades resulted in higher school attain- A number of evaluations have concluded that the ment among indigenous children and a signifi cant higher enrollment levels have not resulted in bet- reduction in the gap between indigenous and nonin- ter performance on achievement tests, even after digenous children in other outcomes (Bando, López- accounting for selection into school (see, for exam- Calva, and Patrinos 2005). Longer-run impacts ple, Fiszbein et al. 2009; Ponce and Bedi 2010). (Box continues next page) 78 MOBILIT Y ACROSS GENERATIONS BOX 3.5 Conditional cash transfers and children’s educational outcomes (continued) Thus, the potential for CCTs to improve learning possibility is that CCTs, as currently designed, on their own may be limited. To be clear, CCTs are do not address some important constraints at the not designed to improve children’s performance at household level. These constraints could include school but rather to increase enrollment. Still, it is poor parenting practices, inadequate information, or interesting to see whether CCTs have had a positive other inputs (or lack thereof) into the production of indirect effect on achievement or whether other tools education. Another possibility is that the quality of should be used for that purpose. services is so low—perhaps especially for the poor— There are various reasons why CCTs may have that increased use of CCTs alone does not yield large had only modest effects on “fi nal” outcomes. One benefits in terms of student achievement. TABLE 3.2 Interaction of school practices and parental background on reading test scores Interaction of policy variable with Parental education Home possessions ESCS (1) (2) (3) Highest education of parents 6.987*** (0.489) Home possessions 26.37*** (1.388) Index of socioeconomic status (ESCS) 34.67*** (1.448) Teacher qualification −0.00821* −0.0731*** −0.0564*** (0.00422) (0.0144) (0.0141) Incidence of private schools 0.00895 0.0504 −0.0582 (0.0280) (0.104) (0.0998) Incidence of publicly funded private schools −0.0591*** −0.113*** −0.214*** (0.00573) (0.0200) (0.0222) Grade tracking 0.00987* 0.0777*** 0.0314** (0.00536) (0.0151) (0.0158) Class tracking 0.0148*** 0.00741 0.0707*** (0.00489) (0.0140) (0.0155) High frequency of assessments −0.559*** 0.809* −0.157 (0.177) (0.446) (0.556) Accountability −0.824*** −2.553*** −3.205*** (0.175) (0.490) (0.613) Autonomy in selecting courses and determining student 1.050*** 2.404*** 3.583*** (0.149) (0.446) (0.497) Autonomy in selecting staff and budget 0.703*** −0.698 1.723*** (0.141) (0.462) (0.476) Constant 331.4*** 434.0*** 436.2*** (2.105) (1.767) (1.693) Observations 422,332 428,984 428,516 R-squared 0.330 0.340 0.375 Source: Data from PISA 2009. Note: PISA = Program for International Student Assessment. ESCS = PISA index of economic, social, and cultural status. Results correspond to a pooled regres- sion for all countries, with country-fixed effects and as control variables gender, urban dummy, and immigration status. Standard errors (in parentheses) are clustered at the school level. *** p < 0.01, ** p < 0.05, * p < 0.1 MOBILIT Y ACROSS GENERATIONS 79 institutional variable of interest and paren- FIGURE 3.16 School practices and reading test scores for high and tal background. For robustness, we run the low values of selected policies regressions for three alternative measures of parental background: parental education, an index of home asset holdings, and the overall Teacher qualification ESCS index. Figure 3.16 then highlights the impact of the variables that proved to be sig- Incidence of publicly nificant in the ESCS regression by depicting funded private schools the effect of one standard deviation of paren- tal background on the average test score in Grade tracking two scenarios: in a representative country where the level of the policy variable is high Class tracking and in a representative country where such a level is low. Accountability Countries with a larger share of publicly funded private schools tend to have a smaller Autonomy in selecting socioeconomic gradient, controlling for other staff and budget school factors. One standard deviation in Autonomy in the average ESCS in a country like Indonesia selecting courses (where 34 percent of all schools are publicly 0 5 10 15 20 25 30 35 40 45 funded private schools) results in test scores High Low that are nine points lower than in a coun- try like Uruguay, where no private school Source: Data from PISA 2009. received government subsidies. A possible Note: ESCS = PISA index of economic, social, and cultural status; PISA = Program for International interpretation of this result is that govern- Student Assessment. Bars represent the effect of one standard deviation of ESCS on test scores when the policy is set at a high level (green bar) and a low one (orange bar). School variables in the ment funding of private schools increases the graph include only those whose interaction with ESCS is significant in the regression and for which choice of schools available to worse-off fami- the difference between high and low are statistically significant at 5 percent level. Results cor- respond to a pooled regression for all countries, with country-fixed effects and as control variables lies, hence decreasing inequality of educa- gender, urban dummy, and immigration status. Standard errors are clustered at the school level. tional opportunities. This result is consistent with Woessmann et al. (2009). On the other hand, there is no significant A priori, the impact of teacher quality on difference in the socioeconomic gradient for the achievement gap between rich and poor countries with either a high or low incidence kids is not clear. Better-educated teachers of private nonsubsidized schools. Hence, it is might be frustrated with poor-performing the combination of private management but students and concentrate on bringing those guaranteed public financing that is associated more capable upward, increasing the gap with a lower gap in achievements between between poor and high performers. On the the rich and the poor. This combination of other hand, better-educated teachers are publicly funded but privately operated sys- more likely to be better trained to deal with tems has been proven to be causally related classes that are more heterogeneous and to higher performance in other contexts, for hence more able to increase the pace with- example in the Netherlands, where more out leaving behind the worst-performing than 80 percent of the students are enrolled students. Our results suggest that the latter in these types of schools, but education is free effect appears to be dominant, inasmuch as for the compulsory first 10 years of schooling countries with a higher proportion of teach- (Patrinos 2012). In Latin America, prominent ers with an ISCED5A (International Stan- examples of this mixed system are the voucher dard Classification of Education) certifica- programs in Chile and Colombia, which also tion, an indicator of teacher quality, present appear to be related to less inequality in edu- a lower gap in test scores between rich and cational achievements (see box 3.6) poor students. 80 MOBILIT Y ACROSS GENERATIONS BOX 3.6 Voucher systems in Chile and Colombia: Did they help the achievements of the poor? Chile was one of the first countries to implement 1991, the program provided the poorest third of its a school choice program for the stated purpose of population with access to secondary education. Run- improving efficiency in education. But as important ning until 1997, PACES covered more than a quar- as the 1981 program was, the nationwide imple- ter of a million students. The vouchers were renew- mentation presents two important difficulties for able through to the end of high school as long as evaluating its impact. First, participation was not the student continued to progress. More than three- randomized; therefore, it is diffi cult to disentangle quarters of the beneficiaries renewed their vouchers. the effects of the program from the inherent differ- The vouchers could be used at private academic and ence in populations due to self-selection. Second, it vocational schools, and about 40 percent of private is a universal program, implying that the counterfac- schools accepted them. The unit costs for participat- tual of no vouchers is difficult to construct (Hoxby ing private schools were 40 percent lower than for 2003). nonparticipating private schools. In spite of these difficulties, the school choice Due to oversubscription in the program, available program in Chile has been subject to a high level places were allocated by lottery. This created a natu- of scrutiny. The earlier literature presents mixed ral, randomized experiment that enabled research- results, to say the least, but identification of causal ers to undertake rigorous impact evaluations of the impacts is difficult for the aforementioned reasons program and test several hypotheses. The results (see, for example, Aedo 1997; Aedo and Larrañaga for this targeted voucher program are encourag- 1994; Contreras 2001; Rodríguez 1988; Gallegos ing. Researchers found that voucher beneficiaries 2002; Mizala and Romaguera 2000). More recently, had higher educational attainment: they were 10 several strategies to overcome the problem of self- percent more likely to fi nish the eighth grade three selection using the Heckman correction method years after they won the vouchers. They were also (see, for example, Sapelli and Vial 2004) and instru- 5–6 percent less likely to repeat a grade. They scored mental variables (IV) approaches (see, for example, 0.2 standard deviations higher on achievement tests Auguste and Valenzuela 2004; Gallegos 2002; Hsieh than nonvoucher students. And they were 20 per- and Urquiola 2006) have been proposed. According cent more likely to take the college entrance exam to an evaluation by Henríquez et al. (2012), voucher than students who had not won a voucher in the lot- schools run by Sociedad de Instrucción Primaria tery. They were also 0.6–1.0 percent less likely to be (SIP) that serve low-income students obtain test married and 2.5–3.0 percent less likely to be work- scores that are up to one standard deviation higher ing (Angrist et al. 2002). In a study of longer-term than those obtained by public schools, and up to 70 effects, Angrist, Bettinger, and Kremer (2006) found percent of one standard deviation higher than pri- that the program improved scores for both average vate voucher schools in Santiago. Furthermore, the students and those over the 90th percentile. performance of SIP schools is similar to that of pri- Yet another study tested whether vouchers vate nonvoucher schools, which typically serve the increased educational productivity or were purely elite families in Chile. Long-term benefits of vouch- redistributive, benefiting recipients by giving them ers have been examined with IV estimates and show access to more desirable peers at others’ expense. significant effects in the labor market (Bravo, Muk- Among the voucher applicants to vocational schools, hopadhyay, and Todd 2010; Patrinos and Sakellar- lottery winners were less likely to attend academic iou 2011). secondary schools and thus had peers with less- Colombia, in an effort to increase access to sec- desirable observable characteristics. Despite this, ondary schools, offered funding to private schools lottery winners had better educational outcomes. that enrolled students from poor families. This Hence, in this population, vouchers improved edu- became known as the secondary school voucher pro- cational outcomes through channels beyond redis- gram, the Programa de Ampliación de Cobertura tribution of desirable peers (Bettinger, Kremer, and de la Educación Secundaria (PACES). Launched in Saavedra 2010). MOBILIT Y ACROSS GENERATIONS 81 School accountability is a last factor asso- between high- and low-autonomy countries is ciated with a lower socioeconomic gradient about 7 points in the PISA test score. in student achievements. Countries where students’ assessments are used to compare schools with others and where year-to-year Conclusions performance is monitored are, on average, This chapter has documented the extent of more equitable (once we control for other intergenerational mobility in Latin American factors). In randomized evaluations that countries and compared it with other devel- can identify the causal effects of increasing oped and developing economies. The com- accountability in schools, the evidence sug- parative analysis allowed us to draw several gests a positive role for accountability in pro- conclusions: moting the performance of the system. Consistent with the literature summa- • On the positive side, the 2000s showed a rized above, in countries prone to use pupils’ notable decline in the inequality of oppor- assessments to assign them to grades and tunities related to educational attainment. classes, the test score gap between rich and The children born to households that are poor kids is larger. The differences in the disadvantaged (whether due to the par- impact of one standard deviation in ESCS on ents’ lower education or lower income) are test scores in the United States—where class less likely to be delayed in schools today tracking is prevalent (88 percent of schools than they were in the 1990s. use this method in one way or another)—is • Similarly, delays in school attainment more than 6 percentage points larger than in associated with ethnic minority groups other countries such as Brazil, where only 10 are less prevalent today than they were in percent of the schools use such tracking. the recent past. Regarding school autonomy, the expected • In educational achievement, the evidence impact on the socioeconomic gradient of the is scarcer and more complicated to evalu- children’s test scores is, in principle, ambigu- ate, but it suggests very limited, if any, ous. On the one hand, school autonomy improvements during the 2000s. may allow greater influence of parents in transforming children’s potential into higher • The negative note comes from the long achievements (Ammermüller 2005). To the road ahead that must be traveled to extent that sorting according to parental achieve an equal-opportunity environ- preferences or resources is important in a ment for Latin American children. In spite country, it may lead to larger inequalities of the progress made, Latin American in achievement. On the other hand, greater countries display some of the lowest levels school autonomy may allow schools to adapt of intergenerational mobility in income or their curriculum and structure to try to miti- education in the world. gate possible learning difficulties or delays of less-well-off children. The chapter has also provided some tenta- The two variables capturing autonomy of tive evidence of the correlates of cross-country schools (autonomy regarding students and differences in educational achievement, and course contents and autonomy regarding has reviewed some of the key findings from staff) yielded similar results in the analysis. In the impact evaluation literature in this area: both cases, higher school autonomy is associ- ated with a higher degree of intergenerational • Better teachers, more accountable and persistence, but the magnitude of the effect is transparent schools, and a mixed system larger in the case of more autonomy regard- of public funding with private provision ing students and course contents. In this case, are associated with more intergenera- the difference in the socioeconomic gradient tional mobility in the sense that children’s 82 MOBILIT Y ACROSS GENERATIONS educational outcomes are less affected by cational persistence in Latin America is parental background. the high inequality of the region’s edu- • School tracking (that is, the grouping of cational system. The differences in the students according to their ability or per- characteristics of the schools attended by formance) and school autonomy appear children from advantageous backgrounds instead to work to the detriment of stu- and those attended by children from poor dents from poor socioeconomic back- socioeconomic backgrounds are larger in grounds and in favor of those with better- Latin America than in most of the other educated or higher-income parents. regions covered in this chapter. • An important factor that appears to be behind the strong intergenerational edu- MOBILIT Y ACROSS GENERATIONS 83 Focus Note 3.1 Bounding the estimates of parental background on student achievement Test-score data sets such as SERCE and PISA, as used age 12, as is the case in Uruguay for any child, but if in this chapter, allow us to estimate the gap in school he or she had parents with no education, the child has achievement according to pupils’ parental back- only a 60 percent chance of being in school at age 12. ground. The problem is that, most often, social scien- The problem in calculating the importance of tists and policy makers are interested in the degree of parental background on children’s achievements for interdependency for the society as a whole rather than the whole population using schooled-based data sets only among those attending school. This becomes is that information on the test scores that the nonen- particularly relevant when comparing countries with rolled children would have had were they in school varying rates of school enrollments. For instance, is not available. Additionally, selection into school is almost all children aged 15 in Chile are enrolled in most likely not random, so that the observed distribu- seventh grade or above, whereas in Peru only 6 out tion is unlikely to be a good indication of test scores of 10 are (see figure F3.1A). Similarly, almost all for the nonenrolled children. One alternative is to 12-year-old Uruguayans attend third grade or above, construct reasonable lower and upper bounds on the while 70 percent of Guatemalans do. More impor- distribution of test scores for the nonobserved pop- tant, these proportions vary significantly, depending ulation and combine it with the observed one. This on the socioeconomic status of the child’s family. Case approach to missing data has been proposed by Man- in point: a Guatemalan child born from secondary- ski (1994) and Manski and Pepper (2000) and used educated parents will most certainly attend school at by, among others, Blundell et al. (2007) to study wage FIGURE F3.1A School enrollment rates, selected Latin American countries percentage a. Proportion of children (age 11–12) attending 3rd grade b. Proportion of children (age 15) attending 7th grade or above, by parental background or above, by parental background Chile Chile Argentina Uruguay Argentina Panama Mexicoa Panama Dominican Republic Costa Rica Uruguay Paraguay Ecuador Colombia Colombia El Salvador Mexicoa Brazil Nicaragua Brazil Peru Peru Guatemala 0 0.2 0.4 0.6 0.8 1.0 0 0.2 0.4 0.6 0.8 1.0 Total Tertiary Secondary Total Tertiary Secondary Primary Less than primary Primary Less than primary Source: Lugo and Messina 2012. a. For Mexico, the data is from the state of Nuevo León (NLE). (Box continues next page) 84 MOBILIT Y ACROSS GENERATIONS Focus Note 3.1 Bounding the estimates of parental background on student achievement (continued) inequality and educational and gender differentials For each of these two distributions on either side of in the United Kingdom, to allow for the nonrandom the inequality, one can compute the median test score selection into work. for each level of parental education. The median Lugo and Messina (2012) apply Manski’s derived from the distribution on the left will give approach to test score data from SERCE and PISA for the upper bound for that level of parental education, children aged 11–12 and 15, respectively. Bounds are denoted by ts(u)(pe), while the median derived from constructed, imposing positive selection into school the right side of the inequality will give the lower in a weak sense; that is, children currently not in bound of test score, ts(l)(pe). school are assumed to perform no better than those From here, one can compute the differences in school with similar backgrounds. Strictly speak- between test scores of children from different paren- ing, the assumption is that, conditional on parental tal education (for example, tertiary versus primary) background, the test score distribution of nonenrolled as follows: individuals is first-order, stochastically dominated by the distribution of enrolled children. For each level of ts(l)(pe = tertiary) – ts(u)(pe = primary) parental education, the observed distribution of test d D d ts(u)(pe = tertiary) scores is combined with the two extreme distributions – ts(l)(pe = primary). (F3.1d) for the nonenrolled to compute the lower and upper bounds. These results are presented in figure F3.1b. Formally, the method can be described as follows. As explained in the chapter, in countries where The cumulative distribution function of interest is the selection is small, the lower and upper bounds that of children’s test scores, conditional on parental of the differences in sixth-grade test scores are very education, denoted F (ts/pe), where ts is the test score close to each other. In these cases, one can get a fairly and pe is the level of parental education, expressed accurate estimate of the true gap in performance. See, as a categorical variable. Selection is indicated by the for instance, Argentina, Chile, or Uruguay, where indicator function E. When a child attends school, although the difference in test scores is significantly E = 1, whereas when he or she does not, E = 0. While high (above one standard deviation), the bounds are F(ts/pe) is not observed, because of nonrandom selec- extremely tight so that the point estimate is a fairly tion, we can write good approximation of the true difference. Instead, in countries where a significant propor- F (ts/pe) = F (ts/pe, E = 1)P(pe) tion of children do not attend school, the distances + F (ts/pe, E = 0)[1 – P(pe)], (F3.1a) between the lower estimates of the gap and the upper bound can be quite large. An extreme case is that of where P(pe) is the probability of attending school, Guatemala when comparing the differences in test conditional on parental education. Lugo and Messina scores for children of tertiary-educated parents with (2012) calculate these probabilities (enrollment rates) those whose parents had no education. The bounds from national household surveys. indicate that the true test score difference lies some- The assumption of positive selection into school where between 80 and 180 (that is, between one and implies that two standard deviations). Importantly, although the estimate from the observed distribution (black dot) F (ts/pe, E = 1) d will indicate that the gap in Guatemala is among the F (ts/pe, E = 0) ts, pe (F3.1b) smallest in the region, the bounds will indicate that it is possible that, instead, the gap between the highest Under this assumption, the bounds to the distribution and lowest parental background is indeed the larg- of test scores is est of all. Similar results, though to a lesser extent, are found in Brazil, Colombia, and Nicaragua, where F (ts/pe, E = 1) d F (ts/pe) d F (ts/pe, E = 1) their position in the region in terms of degree of inter- P(pe) + [1 – P(pe)]. (F3.1c) generational dependence differ once we incorporate information on the nonenrolled individuals. MOBILIT Y ACROSS GENERATIONS 85 Focus Note 3.1 (continued) FIGURE F3.1B Inequalities in reading test scores of sixth-grade students, selected Latin American countries, 2006 a. Test score (parental education = tertiary) – b. Test score (parental education = tertiary) – test score (parental education ≤ primary) test score (parental education = primary) 200 200 180 180 160 160 140 140 120 120 Test score Test score 100 100 80 80 60 60 40 40 20 20 0 0 Ni Cu c ra a a lv a Co uad r sta or M ica ra o ay ge ile na a Pe a B u Gu gu il at ay ala ca lic El lom a lva a Gu cua r em r ala r a sta ay M ica Pa ico a ge ile a Ur Peru Br y il Ec ado i E do at do Ur raz ca b Co gu Sa bi Pa tin m Pa exic r ua az Co ragu Sa bi Pa Cub m in bl gu Ar Ch Ni pub Co agu Ar Ch R em El lom R ex nt na pu n ug u Re e nR n ica ica in in m Range (lower and upper bounds) Range (lower and upper bounds) m Do Do Point estimate Point estimate Sources: Lugo and Messina 2012; SERCE and SEDLAC (Socioeconomic Database for Latin America and the Caribbean) data. The problem of nonenrollment is, naturally, socioeconomic background, the enrollment rate is 97 more acute for older children. PISA surveys chil- percent. According to these results, one can be con- dren aged 15 who are enrolled in school in seventh fident that the test-score gap between children from grade or above. As shown in figure F3.1c, the pro- primary and tertiary parental education is larger in portion of children of that age who are not in school Chile, Peru, and Uruguay than in Brazil, Colombia, or fall behind significantly is, in some countries, and Panama. Similarly, irrespective of performance of quite large. This leads to much larger bounds in the nonattendant children, the differences in test scores estimates of the performance gap, with the notable by parental education are larger in Uruguay than in exception of Chile, where, even for children of lower Chile. 86 MOBILIT Y ACROSS GENERATIONS Focus Note 3.1 Bounding the estimates of parental background on student achievement (continued) FIGURE F3.1C Inequalities in reading test scores at age 15, selected Latin American countries, 2009 a. Test score (parental education = tertiary) – b. Test score (parental education = tertiary) – test score (parental education ≤ primary) test score (parental education = primary) 260 260 240 240 220 220 200 200 180 180 160 160 Test score Test score 140 140 120 120 100 100 80 80 60 60 40 40 20 20 0 0 ile il a a a ico ru y a il a ile ico a y ru ua ua az az bi m in m bi in Pe Pe Ch Ch ex ex nt nt m m na na Br Br ug ug lo lo ge ge M M Pa Pa Ur Ur Co Co Ar Ar Range (lower and upper bounds) Point estimate Range (lower and upper bounds) Point estimate Sources: Lugo and Messina 2012; PISA and SEDLAC (Socioeconomic Database for Latin America and the Caribbean) data. MOBILIT Y ACROSS GENERATIONS 87 Notes comparisons, considering that the distribu- tion of the population across urban and rural 1. See Card (2001) and the references therein for areas varies greatly across countries, and in an overview. particular between developing and developed 2. The educational gap and its relationship with countries. parental background are studied in Andersen 9. A related approach is proposed by Ferreira (2001) for a cross-section of Latin Ameri- and Gignoux (2011). can countries and in Behrman, Gaviria, and 10. Aedo and Walker (2011) study the changes in Székely (2001) for 16 Latin American coun- the gradient of socioeconomic status on PISA tries during the period 1980–96. test scores in seven Latin American countries 3. The schooling gap is calculated for children between 2000 and 2009, also finding mixed living with their parents because other infor- results. mation is not available in the household sur- 11. The main difference is that we need to control veys. For this reason, we do not calculate the for life-cycle effects because we are compar- schooling gap for individuals older than 18. ing adults of different ages with parents of 4. Fields (1996) proposes a decomposition of different ages. This is done by including in the the R-squared in a regression that allows regressions a full set of dummy variables for separating the contributions of each different fathers’ and children’s ages. variable (or set of variables) in the explana- 12. We define developing countries as those with tion of the overall variance. Like the correla- a GDP per capita purchasing power parity tion coefficient, this measure is invariant to below US$20,000 in 2009. the variance in either parental background or 13. Each group is composed of the following children’s schooling gaps. Using this proposed countries: “Nordic countries” include Den- methodology, we find that the percentage of mark, Finland, Iceland, Norway, and Sweden. the variance in the schooling gap explained “Anglo-Saxon” countries include Australia, by parental income and education declines Canada, Ireland, and New Zealand. “Con- steadily in the Latin American region. tinental Europe” includes Austria, Belgium, 5. Evaluating the evolution of the educational France, Germany, Greece, Italy, Liechten- gap associated with ethnic minorities is par- stein, Luxembourg, the Netherlands, Portu- ticularly challenging when the information gal, Spain, and Switzerland. “High-Income” comes from household surveys and hence is economies include Croatia; the Czech Repub- self-declared. The decision to declare belong- lic; Estonia; Hong Kong SAR, China; Hun- ing to a certain ethnic minority group (or not) gary; Indonesia; Israel; Japan; Kazakhstan; the depends on social norms and attitudes toward Republic of Korea; Jordan; Kyrgyzstan; Lat- ethnic minorities, which are likely to change via; Macao SAR, China; Poland; Qatar; the over time. This in turn introduces a problem Slovak Republic; Slovenia; Shanghai, China; of selection that we are not addressing in this Taiwan, China; Singapore; and Dubai (United chapter. For this reason, the results regard- Arab Emirates). “Low-Middle Income” coun- ing differences in educational gaps associ- tries include Albania, Azerbaijan, Bulgaria, ated with ethnicity should be interpreted with Lithuania, Montenegro, Romania, the Rus- caution. sian Federation, Serbia, Tunisia, Thailand, and 6. The index was constructed using princi- Turkey. “Latin America” includes Argentina, pal component analysis in two steps: first, Brazil, Chile, Colombia, Mexico, Panama, on asset holdings and dwelling characteris- Peru, Trinidad and Tobago, and Uruguay. tics; second, combining the resulting index “United Kingdom-United States” includes the with the maximum years of schooling of the United Kingdom and United States. parents. 14. The School Management Support program 7. In the chapter, we will restrict the attention (Apoyo a La Gestión Escolar; AGE) in Mexico to results from the reading literacy test, which is a randomized trial that doubled resources was the focus of the 2009 PISA survey, but and hence the school responsibilities of par- results are qualitatively similar for math. ent associations in highly disadvantaged rural 8. An additional challenge from PISA is that it is communities in four high-poverty states with representative only of urban areas. This intro- a high concentration of indigenous peoples. duces serious limitations in cross-country Gertler, Patrinos, and Rodríguez-Oreggia 88 MOBILIT Y ACROSS GENERATIONS (2012) found that AGE improved learning Arneson, Richard. 1989. “Equality and Equality outcomes by almost a quarter of a standard of Opportunity for Welfare.” Philosophical deviation. A separate component designed to Studies 56 (1): 77–93. test the impact of training parents in organiz- Atkinson, Anthony B. 1981. “On Intergenera- ing themselves (but with no cash grant) also tional Income Mobility in Britain.” Journal of proved successful compared with a group of Post Keynesian Economics 3 (2): 194–218. schools receiving neither grants nor train- Auguste, Sebastian, and Juan P. Valenzuela. 2004. ing. The effects of training alone are slightly “Do Students Benefit from School Competi- higher than the cash grant, though the schools tion? Evidence from Chile.” Doctoral disserta- are not directly comparable. tion, University of Michigan, Ann Arbor. Azevedo, João Pedro, Alejandro Gaviria, Roberto Angulo, and Gustavo Nicolás Paez. 2012. References “Social Mobility in Colombia.” Unpublished Aedo, Cristian. 1997. “Organización Industrial de paper, World Bank, Washington, DC. la Prestación de Servicios Sociales.” Research Bando, Rosagenla, Luis F. López-Calva, and Network Working Paper 3001, Inter-American Harry A. Patrinos. 2005. “Child Labor, School Development Bank, Washington, DC. Attendance, and Indigenous Households: Evi- Aedo, Cristian, and Osvaldo Larrañaga. 1994. dence from Mexico.” Policy Research Working “Educación Privada vs. Pública en Chile: Cali- Paper 3487, World Bank, Washington, DC. dad y Sesgo de Selección.” Unpublished paper, Barrera-Osorio, Felipe, Marianne Bertrand, postgraduate program in economics, Latin Leight L. Linden, and Francisco Perez-Calle. American Institute of Doctrine and Social 2011. “Improving the Design of Conditional Studies and Georgetown University, Washing- Transfer Programs: Evidence from a Random- ton, DC. ized Education Experiment in Colombia.” Aedo, Christian, and Ian Walker. 2011. Skills for American Economic Journal: Applied Eco- the 21st Century in Latin America and the nomics 3 (2): 167–95. Caribbean. Directions in Development Series. Becker, Gary S., and Nigel Tomes. 1979. “An Washington DC: World Bank. Equilibrium Theory of the Distribution of Almås, Ingvild, Alexander W. Cappelen, Thori Income and Intergenerational Mobility.” Jour- J. Lind, Erik Sorensen, and Bertil Tungodden. nal of Political Economy 87 (6): 1153–89. 2011. “Measuring Unfair (In)Equality.” Jour- Behrman, Jere R., Alejandro Gaviria, and Miguel nal of Public Economics 95 (7–8): 488–99. Székely. 2001. “Intergenerational Mobility in Ammermüller, Andreas. 2005. “Educational Latin America.” Journal of the Latin Ameri- Opportunities and the Role of Institutions.” can and Caribbean Economic Association 2 Research Memoranda 004, ROA, Research (1): 1–44. Centre for Education and the Labour Market, Behrman, Jere R., Susan W. Parker, and Petra Maastricht, Netherlands. Todd. 2010. “Do Conditional Cash Transfers Andersen, Lykke. 2001. “Social Mobility in Latin for Schooling Generate Lasting Benefits? A America: Links with Adolescent Schooling.” Five-Year Followup of PROGRESA/Oportuni- Research Network Working Paper 3130, Inter- dades.” Journal of Human Resources 46 (1): American Development Bank, Washington, 93–122. DC. Behrman, Jere R., and Paul Taubman. 1990. Angrist, Joshua, Eric Bettinger, Erik Bloom, Eliza- “The Intergenerational Correlation between beth King, and Michal Kremer. 2002. “Vouch- Children’s Adult Earnings and Their Parents’ ers for Private Schooling in Colombia: Evidence Income: Results from the Michigan Panel Sur- from a Randomized Natural Experiment.” vey of Income Dynamics.” Review of Income American Economic Review 92 (5): 1535–58. and Wealth 36 (2): 115–27. Angrist, Joshua, Eric Bettinger, and Michael Bénabou, Roland, and Efe A. Ok. 2001. “Social Kremer. 2006. “Long-Term Educational Mobility and the Demand for Redistribution: Consequences of Secondary School Vouch- The Poum Hypothesis.” The Quarterly Jour- ers: Evidence from Administrative Records in nal of Economics 116 (2): 447–87. Colombia.” American Economic Review 96 Bettinger, Eric, Michael Kremer, and Juan E. Saa- (3): 847–62. vedra. 2010 “Are Educational Vouchers Only MOBILIT Y ACROSS GENERATIONS 89 Redistributive?” Economic Journal 120 (546): Post-Secondary Schooling.” The Economic F204–F228. Journal 112 (482): 705–34. Björklund, Anders, and Markus Jäntti. 2009. Causa, Orsetta, and Catherine Chapuis. 2009. “Intergenerational Income Mobility and the “Equity in Student Achievement Across OECD Role of Family Background.” In The Oxford Countries: An Investigation of the Role of Poli- Handbook of Economic Inequality, ed. W. cies.” Economics Department Working Paper Salverda, B. Nolan, and T. M. Smeeding, 491– 708, Organisation for Economic Co-operation 521. Oxford, U.K.: Oxford University Press. and Development, Paris. Björklund, Anders, and Kjell G. Salvanes. 2011. Chong, Alberto, and Hugo R. Ñopo. 2008. “The “Education and Family Background: Mecha- Mystery of Discrimination in Latin America.” nisms and Policies.” In Handbook of the Eco- Journal of the Latin American and Caribbean nomics of Education, Vol. 3, ed. E. Hanushek, Economic Association 8 (2): 79–115. S. Machin, and L. Woessmann, 201–47. San Cohen, Gerald A. 1989. “On the Currency of Diego: Elsevier Egalitarian Justice.” Ethics 99 (4): 906–944. Black, Sandra E., and Paul J. Devereux. 2011. Contreras, Dante. 2001. “Evaluating a Voucher “Recent Developments in Intergenerational System in Chile: Individual, Family, and School Mobility.” In Handbook of Labor Economics, Characteristics.” Working Paper 175, Facultad ed. O. Ashenfelter, R. Layard, and D. Card, de Ciencias Económicas y Administrativas, 1487–1542. San Diego: Elsevier. Universidad de Chile, Santiago. Blundell, Richard, Amanda Gosling, Hidehiko Corak, Miles. Forthcoming. “Inequality from Ichimura, and Costas Meghir. 2007. “Changes Generation to Generation: The United States in in the Distribution of Male and Female Wages Comparison.” In The Economics of Inequal- Accounting for Employment Composition ity, Poverty and Discrimination in the 21st Using Bounds.” Econometrica 75 (2): 323–63. Century, ed. Robert Rycroft. Santa Barbara, Bravo, David, Sankar Mukhopadhyay, and Petra CA: ABC-CLIO. E. Todd. 2010. “Effects of School Reform on Cruces, Guillermo, Marcelo Bergolo, Fedora Car- Education and Labor Market Performance: bajal, Adriana Conconi, and Andres Ham. Evidence from Chile’s Universal Voucher Sys- 2011. “Are There Ethnic Inequality Traps in tem.” Quantitative Economics 1 (1): 47–95. Education? Empirical Evidence for Brazil and Brown, Meta, John K. Scholz, and Anath Ses- Chile.” Joint publication of the Center for hadri. 2009. “A New Test of Borrowing Con- Distributive, Labor and Social Studies of the straints for Education.” Working Paper 14879, Universidad Nacional de La Plata and Consejo National Bureau of Economic Research, Cam- Nacional de Investigaciones Científicas y Téc- bridge, MA. nicas, Argentina. Brunello, Giorgio, and Daniele Checchi. 2007. Currie, Janet. 2009. “Healthy, Wealthy, and “Does School Tracking Affect Equality of Wise: Socioeconomic Status, Poor Health in Opportunity? New International Evidence.” Childhood, and Human Capital Develop- Economic Policy 22 (52): 781–861. ment.” Journal of Economic Literature 47 (1): Busso, Matias, Martin Cicowiez, and Leonardo 87–122. Gasparini. 2005. “Ethnicity and the Millen- ———. 2011. “Inequality at Birth: Some Causes nium Development Goals in Latin America and Consequences.” American Economic and the Caribbean.” Joint publication of Eco- Review: Papers and Proceedings 101: 1–22 nomic Commission for Latin America and Currie, Janet, and Duncan Thomas. 2000. the Caribbean (of the United Nations), Inter- “School Quality and the Longer-Term Effects American Development Bank, United Nations of Head Start.” The Journal of Human Development Programme, and World Bank, Resources 35 (4): 755–74. Washington, DC. Dearden, Lorraine, Steve Machin, and Howard Card, David. 2001. “Estimating the Return Reed. 1997. “Intergenerational Mobility in to Schooling: Progress on Some Persistent Britain.” The Economic Journal 107 (440): Econometric Problems.” Econometrica 69 (5): 47–66. 1127–60. Ferreira, Francisco H. G., and Jérémie Gig- Carneiro, Pedro, and James J. Heckman. 2002. noux. 2011. “The Measurement of Edu- “The Evidence on Credit Constraints in c at iona l I nequ a l it y: Ach ievement a nd 90 MOBILIT Y ACROSS GENERATIONS Opportunity.” Policy Research Working Paper Hoxby, Caroline M. 2003. “School Choice and 5873, World Bank, Washington, DC. School Competition: Evidence from the United Fields, Gary S. 1996. “Accounting for Income States.” Swedish Economic Policy Review 10 Inequality and Its Change: A New Method, (2): 9–65. with Application to the Distribution of Earn- Hsieh, Chang-Tai, and Miguel Urquiola. 2006. ings in the United States.” In Research in “The Effects of Generalized School Choice on Labor Economics, ed. S. Polachek. Bingley, Achievement and Stratification: Evidence from U.K.: Emerald Group Publishing Limited. Chile’s Voucher Program.” Journal of Public Fiszbein, Ariel, Norbert Schady, Francisco H. G. Economics 90 (8–9): 1477–503. Ferreira, Margaret Grosh, Nial Kelleher, Pedro Jäntti, Markus. 2012. Family Associations in Olinto, and Emmanuel Skoufias. 2009. “Con- Economic Status in Developed Countries: ditional Cash Transfers: Reducing Present A Review of Approaches. Washington, DC: and Future Poverty.” Policy Research Report, World Bank. World Bank, Washington DC. Justino, Patricia, and Arnab Acharya. 2003. Gallegos, Francisco. 2002. “Competencia y “Inequality in Latin America: Processes and Resultados Educativos: Teoría y Evidencia Inputs.” Working Paper 22, Poverty Research para Chile.” Latin American Journal of Eco- Unit, University of Sussex, U.K. nomics (formerly Cuadernos de Economía) 39 Keane, Michael, and Kenneth I. Wolpin. 2001. (118): 309–52. “The Effect of Parental Transfers and Bor- Gertler, Paul, Harry Patrinos, and Eduardo rowing Constraints on Educational Attain- Rodríguez-Oreggia. 2012. “Parental Empow- ment.” International Economic Review 42 (4): erment in Mexico: Randomized Experiment of 1051–103. the Apoyo a La Gestion Escolar (AGE) Pro- Krugman, Paul. 1992. “The Rich, the Right, and gram in Rural Primary Schools in Mexico.” the Facts.” The American Prospect 11: 19–31. Research report, Society for Research on Edu- Lugo, Maria Ana, and Julian Messina. 2012. cational Effectiveness, Evanston, IL. “Student Achievement: Correcting for Selec- Hanushek, Eric A., and Ludger Woessmann. tion into School Using Non-Parametric 2006. “Does Educational Tracking Affect Bounds.” Unpublished background paper for Performance and Inequality? Differences-in- the current volume of Economic Mobility and Differences Evidence across Countries.” Eco- the Rise of the Latin Middle Class , World nomic Journal 116 (510): C63–C76. Bank, Washington, DC. ———. 2011. “How Much Do Educational Out- Manski, Charles. 1994. “The Selection Problem.” comes Matter in OECD Countries?” Eco- In Advances in Econometrics , Sixth World nomic Policy 26 (67): 427–91. Congress, vol. 1, ed. C. Sims, 143–170. Cam- Harbison, Ralph W., and Eric A. Hanushek. bridge, U.K.: Cambridge University Press. 1992. Educational Performance of the Poor: Manski, Charles, and John Pepper. 2000. “Mono- Lessons from Rural Northeast Brazil. Oxford, tone Instrumental Variables: With Application U.K.: Oxford University Press. to the Returns to Schooling.” Econometrica Haveman, Robert, and Barbara Wolfe. 1995. 68 (4): 997–1010. “The Determinants of Children’s Attainments: McIntosh, Steven, and Anna Vignoles. 2001. A Review of Methods and Findings.” Journal “Measuring and Assessing the Impact of Basic of Economic Literature 33 (4): 1829–78. Skills on Labour Market Outcomes.” Oxford Henríquez, Francisco, Bernardo Lara, Alejandra Economic Papers 53 (3): 453–81. Mizala, and Andrea Repetto. 2012. “Effective Mizala, Alejandra, and Pilar Romaguera. 2000. Schools Do Exist: Low-Income Children’s Aca- “School Performance and Choice: The Chilean demic Performance in Chile.” Applied Eco- Experience.” Journal of Human Resources 35 nomics Letters 19 (5): 445–51. (2): 392–417. Hertz, Tom, Tamara Jayasundera, Patrizio Neal, Derek A., and William R. Johnson. 1996. Piraino, Sibel Selcuk, Nicole Smith, and Alina “The Role of Premarket Factors in Black- Verashchagina. 2007. “The Inheritance of White Wage Differences.” Journal of Political Educational Inequality: International Compar- Economy 104 (5): 869–95. isons and Fifty-Year Trends.” The B.E. Journal OECD (Organisation for Economic Co-opera- of Economic Analysis and Policy 7 (2). tion and Development). 2011. “PISA 2009 MOBILIT Y ACROSS GENERATIONS 91 Technical Report (Preliminary Version).” Sapelli, Claudio, and Bernardita Vial. 2004. “Peer OECD, Paris. Effects and Relative Performance of Voucher Patrinos, Harry A. 2012. “Private Education Pro- Schools in Chile.” Working Paper 256, Ponti- vision and Public Finance: The Netherlands.” ficia Universidad Católica de Chile, Santiago. In Education Economics, forthcoming. Online Schultz, T. Paul. 2004. “School Subsidies for the publication at http://www.tandfonline.com/ Poor: Evaluating the Mexican Progresa Pov- doi/abs/10.1080/09645292.2011.568696. erty Program.” Journal of Development Eco- Patrinos, Harry A., and Christos Sakellariou. nomics 74 (1): 199–250. 2011. “Quality of Schooling, Returns to SEDLAC (Socio-Economic Database for Latin Schooling, and the 1981 Vouchers Reform in America and the Caribbean). 2011. Data- Chile.” World Development 39 (12): 2245–56. base of the Center for Distributive, Labor and Ponce, Juan, and Arjun S. Bedi. 2010. “The Social Studies (CEDLAS), University of La Impact of a Cash Transfer Program on Cog- Plata, Argentina, and World Bank, Washing- nitive Achievement: The Bono de Desarrollo ton, DC. http://sedlac.econo.unlp.edu.ar/eng/. Humano of Ecuador.” Economics of Educa- Skoufias, Emmanuel, and Susan W. Parker. 2001. tion Review 29 (1): 116–25. “Conditional Cash Transfers and their Impact Rajkumar, Andrew S., and Vinaya Swaroop. on Child Work and Schooling.” Journal of the 2008. “Public Spending and Outcomes: Does Latin American and Caribbean Economic Governance Matter?” Journal of Development Association 2 (1): 45–96. Economics 86 (1): 96–111. Solís, Alex. 2011. “Credit Access and College Rodríguez, J. 1988. “School Achievement and Enrollment.” Unpublished paper, University of Decentralization Policy: The Chilean Case.” California, Berkeley. Revista de Análisis Económico 3 (1): 75–88. Solon, Gary. 1999. “Intergenerational Mobility Rodríguez-Oreggia, Eduardo. 2011. “Movilidad in the Labor Market.” In Handbook of Labor Social Intergeneracional de los Jóvenes Benefi- Economics, vol. 3, ed. O. Ashenfelter and D. ciarios de Oportunidades Provenientes de Ho- Card, 1761–800. Amsterdam: North-Holland. gares en Zonas Rurales.” Unpublished external ———. 2002. “Cross-Country Differences in evaluation, Oportunidades. http://evaluacion Intergenerational Earnings Mobility.” Journal .oportunidades.gob.mx:8010/es/anuncios.php. of Economic Perspectives 16 (3): 59–66. Rodríguez-Oreggia, Eduardo, and Freije, Samuel. ———. 2004. “A Model of Intergenerational 2008. “An Impact Evaluation of Oportuni- Mobility Variation over Time and Place.” dades on Rural Employment, Wages and In Generational Income Mobility in North Intergenerational Occupational Mobility.” America and Europe, ed. M. Corak, 38–47. In External Evaluation of Oportunidades Cambridge, U.K.: Cambridge University Press. 2008: 10 Years of Intervention in Rural Areas UNESCO (United Nations Educational, Sci- (1997–2007). Mexico D.F.: Secretaria de entific, and Cultural Organization). 2009. Desarrollo Social. “Reporte técnico del segundo estudio regional Roemer, John E. 1993. “A Pragmatic Theory of comparativo y explicativo: los aprendizajes de Responsibility for the Egalitarian Planner.” los estudiantes en America Latina y el Caribe.” Philosophy and Public Affairs 22 (2): 146–66. Oficina Regional de Educación para América ———. 1998. Equality of Opportunity. New Latina y el Caribe/UNESCO, Santiago. York: Harvard University Press. Woessmann, Ludger, Elke Luedemann, Gabriela Roemer, John E., Rolf Aaberge, Ugo Colombino, Schuetz, and Martin R. West. 2009. School Johan Fritzell, Stephen P. Jenkins, Arnand Accountability, Autonomy and Choice around Lefranc, Ive Marx, Marianne Page, Evert Pom- the World. Cheltenham, U.K.: Edward Elgar. mer, and Javier Ruiz-Castillo. 2003. “To What Zimmerman, David. 1992. “Regression toward Extent Do Fiscal Regimes Equalize Oppor- Mediocrity in Economic Stature.” American tunities for Income Acquisition among Citi- Economic Review 82 (3): 409–29. zens?” Journal of Public Economics 87 (3–4) 539–65. 4 Mobility within Generations All mankind is divided into three classes: those that are immovable, those that are movable, and those that move. —Benjamin Franklin C hapter 3 explored how mobility out of poverty, where do those people go? across generations has evolved in the Is it true that the middle class is growing region. In doing so, it asked whether across the continent? Who was already part children’s opportunities have been improv- of it, and who are the new entrants? Using ing over time relative to those of their par- the mobility measures and decomposition ents. By contrast, this chapter explores how developed in chapter 2, we can explore the an individual can seize opportunities within nature of social dynamics in the region and his or her own lifetime—specifically focus- begin an investigation of their determinants ing on long-term directional intragenera- and, in particular, of how public policy tional mobility. As chapter 2 discussed, the in various realms may have promoted or concept of directional income movement in impeded upward mobility. the intragenerational domain is of particular Of course, to do this entails studying the interest if we want to shed light on the micro- “gross” flow of movements for specific indi- economic dynamics underpinning the growth viduals over time, as opposed to the typical process in Latin America. poverty analysis over time that focuses on For example, how does growth manifest “net” flows and trends of groups at different itself at the individual or household level? parts of the income distribution. In particu- How do the aggregate gross domestic prod- lar, panel data that follow individuals over uct (GDP) growth figures translate into time are needed. Although short-term panel growing incomes for individuals and their data are widely available in Latin America as families? Those are questions about income elsewhere, they rarely cover more than three growth, or income movement, within a per- to four years between rounds, making the son’s lifetime. On the other hand, as incomes study of long-term intragenerational mobility grow at the lower end of the income distri- for the same individual (over 10 or even 20 bution, raising millions of Latin Americans years) impossible. 93 94 MOBILIT Y WITHIN GENERATIONS This is what we aim to do here. To over- that recent years have seen the study of intra- come the lack of long-term panels, we apply generational mobility increasingly capture a recently developed approach—validated the attention of policy makers and research- in three Latin American countries—to con- ers in developing countries.1 Latin America is struct synthetic panels for specific individu- not the exception; a growing number of stud- als, hence allowing the analysis of long-term ies on intragenerational mobility have been dynamics. Specifically, the chapter aims to developed in several countries in the region. understand income dynamics during the past Most of these studies are based on short-term 20 years across Latin America by exploring panels or use pseudo-panel techniques.2 three broad sets of questions: The existing literature in the region employs a variety of methods, time periods, 1. How much directional intragenerational data sets, and measures to gauge the notion mobility has there been in the past 15 to of intragenerational mobility. Moreover, the 20 years? measurement of mobility is based on sev- 2. Who benefited from upward mobility, and eral welfare aggregates, ranging from earn- who suffered from downward mobility ings to household total income. All of this, (for example, who exited poverty or joined while useful, makes it difficult to synthesize the middle class, and who fell behind)? the existing evidence on the magnitude of 3. W hat a re t he me cha n ism s b eh i nd mobility in Latin America or to compare observed results? (In other words, is there mobility levels across different countries in evidence, at least descriptive evidence, the region.3 Despite these differences, the lit- that different policy regimes are associ- erature reveals certain commonalities. Box ated with different degrees of mobility— 4.1 summarizes the main findings of the lit- for example, as policy relates to macro- erature on intragenerational mobility in Latin economic performance, labor markets, or America. social policies?) When measuring intragenerational mobil- ity, it is desirable to work with panel data sets As the chapter shows, the answers to these that follow individuals or households over three questions are telling: time. Unfortunately, such surveys pose at least four empirical challenges: 1. The region has experienced high levels of intragenerational mobility during the past • In Latin America in particular and in the 20 years, especially upward mobility. developing world in general, panels are 2. Those who are poor and or near poverty usually limited to urban areas or to small have benefited the most. samples and therefore they are not repre- 3. Growth (especially during the past decade) sentative of the entire population of the has played a big role in helping people country (Fields et al. 2007). move, especially upward, and other fac- • Because they are typically costly and com- tors such as improvements in education, plex to administer, panel data sets that labor markets, and social policies may track individuals or households over long have also facilitated mobility. periods of time (10 to 20 years) are still rare in Latin America. This scarcity lim- The rest of the chapter explains how we its the generalization of results across the arrive at these insights. region (Fields et al. 2007). • Connected to the previous point, it is usu- ally difficult to revisit households that Using synthetic panels to study physically move or drop out from panel long-term mobility data surveys. As such, nonrandom attri- With the proliferation of short-term panel tion may significantly bias results, leading data in the past decade or so, it is no surprise to an underestimation of the actual mobil- MOBILIT Y WITHIN GENERATIONS 95 ity in the general population (Antman and understand the long-term trends related to McKenzie 2007). movements across economic classes. These • Finally, measurement error will also intro- limitations raise serious concerns regarding duce bias in the mobility estimates. the validity of the policy implications drawn from the short-term panel literature. Although the existing literature may allow Learning about the levels of long-term us to understand the correlates of short-term intragenerational mobility is crucial for the intragenerational mobility (which can be design of effective social policy interven- especially useful in a context of volatility and tions. The type of policies needed to attack crises), short-term panel data do not help us long-term persistent poverty may be quite BOX 4.1 Existing findings on intragenerational mobility in Latin America Most of the earlier studies describe substantial positive coefficients for both regressions in several directional mobility in terms of discrete welfare years, suggesting that a convergence (or nondiver- trajectories by which households in Latin America gence) exists between the earnings of the rich and move across income classes and into and out of pov- the poor. Moreover, the authors fi nd that relatively erty and the middle class (see, for instance, Scott more convergence exists when using initial earnings and Litchfield [1994] and Paredes and Zubizarreta than when using predicted earnings. This result sug- [2005], both regarding Chile in the 1990s). gests that most of the observed changes in incomes Apart from considerable mobility based on dis- are transitory and do not affect longer-term posi- crete welfare trajectories, many studies also find tions in the income distribution. substantial mobility in terms of the magnitude of Comparisons of mobility levels between dif- income changes. More important, these studies gen- ferent country-specific studies are also difficult erally fi nd that income changes are heterogeneous to gauge because of the differences in methodolo- across income groups, the most disadvantaged being gies. However, a few studies provide cross-country those who experienced in general the largest gains mobility comparisons in Latin America as a region, in the region. In this sense, Fields et al. (2007) pres- showing that mobility differs between countries. ent compelling evidence that, conditional on observ- For instance, Calónico (2006) uses pseudo-panels able characteristics, households with the lowest spanning 1992 to 2003 for eight countries and con- initial incomes tend to gain relatively more (see, for cludes that Argentina, Brazil, and Uruguay have instance, Duval Hernandez [2006] for the case of very low levels of mobility, while Chile, Mexico, Mexico). and República Bolivariana de Venezuela are among Because not many long-term panels are available the most mobile countries on the continent. Ñopo in Latin America, it is difficult to study whether (2011) estimates mobility for 14 countries, also short-term transitory movements affect long-term using pseudo-panels for 1992 to 2003. The author positions in the income distribution. To answer fi nds low income mobility only in Brazil, Colombia, this question, Fields et al. (2006) estimate two and Costa Rica; Chile and Argentina show modest regressions using panels from Argentina, Mexico, levels of mobility; and the rest of the region is con- and República Bolivariana de Venezuela. First, the sidered relatively mobile. authors regress earning changes on initial earn- Finally, a large body of the literature studies the ings. Then, they regress earning changes on pre- correlates of intragenerational mobility: gender, edu- dicted earnings, where predicted earnings come cation, employment status, household composition, from regressing observed earnings on a set of time- and the quality of housing systematically appear to invariant observable variables. The authors fi nd non- be related to mobility in Latin America.a a. Regarding gender, see, for instance, Ñopo (2011); Glewwe and Hall (1998); McKenzie (2004); and Corbacho, Garcia-Escribano, and Ichauste (2007). Regarding edu- cation, see, for example, Beccaria and Groisman (2006); Cruces and Wodon (2006); Beneke de Sanfeliu and Shi (2003); and Herrera (1999). Regarding employment, see, for instance, Cruces and Wodon (2006); Corbacho, Garcia-Escribiano, and Ichauste (2007); McKenzie (2004); Duval Hernandez (2006); and Fields et al. (2003). Regard- ing household composition, see, for example, Beneke de Sanfeliu and Shi (2003); Glewwe and Hall (1998); Herrera (1999); and Fields et al. (2003). Finally, regarding the quality of housing, see for instance, Paredes and Zubizarreta (2005). 96 MOBILIT Y WITHIN GENERATIONS different from those required to address tran- Mobility, measured in relation to the cor- sient poverty; the former requires skills and relation of incomes over time, 6 can then asset creation, while the later must focus on be directly related to the parameters of the social protection to cope with risks. There- income process, which can then be estimated fore, availability of longer-term panels is an through standard econometric techniques. indispensable tool for policy makers con- This framework has the merit of draw- cerned about persistent poverty. Unfortu- ing out the links between income volatility, nately, due to the lack of these types of data income mobility, distribution, and social sets, no long-term income mobility estimates welfare in a simple and transparent man- that are comparable across countries exist in ner, allowing for a clearer analytical and Latin America. quantitative discussion of these interrelated Because of the growing concern that exists concepts than has generally been possible in regarding the evaluation of long-term transi- the past. The approach also permits a sin- tions into and out of poverty, an emerging gle measure of welfare, comparable across body of the literature has developed tech- countries, that encompasses these interre- niques to overcome the major limitations of lated phenomena and disaggregates mea- panel data sets by employing cross-sectional sured income mobility into the three com- surveys. A vast array of the literature has ponents above. mainly focused on what is commonly called Their results from Argentina and Mexico the “pseudo-panel” approach, which tracks offer some striking insights: cohorts of individuals over several periods of time.4 This methodology helps to overcome • More than half of measured mobility in the main limitation of panel data sets; it these countries is estimated to be driven can be used to understand long-term mobil- by transitory shocks to income. ity across economic classes. However, stud- • Approximately half of the residual (mobil- ies that use pseudo-panels usually need to ity in permanent income) is driven by impose significant structural assumptions social-welfare-reducing persistent income to yield mobility measures out of repeated shocks. cross-sectional surveys (Dang et al. 2011). In • Finally, the share of measured mobility addition, by aggregating average trends for a that corresponds to “good” mobility is given group (or cohort), this technique does fairly small. not consider intragroup mobility, which may be equally or even more relevant than aggre- Despite the potential of this approach, it gate mobility. relies on multiple rounds of panel data fol- Taking another route, Krebs, Krishna, and lowing specific individuals. These data Maloney (2011) develop a framework linking requirements render it impractical for appli- individual income dynamics, social mobility, cation to a large number of Latin American and welfare. In doing so, they define mobility countries.7 as a composite of three elements: (a) “good” In the absence of long-term panel data mobility, which is a convergence of individu- in the region, and to overcome these limita- als toward some appropriately defined level tions, we employ an alternative approach to of income; (b) risk (“bad” mobility), cor- study directional intragenerational mobility. responding to the variance of permanent Specifically, we apply an innovative exten- shocks; and (c) transitory shocks and mea- sion of poverty mapping techniques that surement error. construct “synthetic panels” using repeated Based on this refinement, they offer a trac- cross-sectional data (Dang et al. 2011). The table analytical framework based on stan- approach builds on an “out-of-sample” impu- dard income processes that provides a closed- tation methodology described in Elbers, Lan- form link between the welfare theory and the jouw, and Lanjouw (2002, 2003) for small- empirics of income dynamic measurement.5 area estimation of poverty (“poverty maps”).8 MOBILIT Y WITHIN GENERATIONS 97 T he method converts t wo or more approach, the synthetic panel approach rounds of cross-sectional data into a panel predicts the income of the same individual (of individuals or households) by predicting or household in different time periods. income for the same households in future • The same approach can be replicated for (or past) periods. Because of the estimation most Latin American countries for the involved, lower- and upper-bound estimates same underlying period and using consis- of income are constructed and can be used tent concepts of income measures, data to subsequently estimate bounds of mobil- sets, and mobility measures, thus consid- ity measures. The bounds produced—the erably enlarging the universe of estimates validation exercises by Cruces et al. (2011) of mobility in the region. in three countries in Latin America (Chile, • The biggest contribution of this method- Nicaragua, and Peru) also confirm this—are ology is that it allows for the estimation expected to sandwich true mobility estimates of long-term intragenerational mobility obtained from actual panel data sets. Focus measures for the 18 Latin American coun- note 4.1 at the end of this chapter provides tries we study, in some cases spanning 20 a detailed technical note of this method and years.9 discusses both its advantages and limitations. One important caveat: for the sake of sim- To do so, we use the Socio-Economic plicity, this chapter will focus on the lower- Database for Latin America and the Carib- bound estimates (the story line is consistent bean (SEDLAC) database of Latin American when we look at the upper-bound results). household surveys, compiled by the Univer- One additional advantage beyond simplicity sidad de la Plata in Argentina (CEDLAS) in is that focusing on the lower-bound estimates partnership with the World Bank. The data- allows us to provide conservative estimates base consists of a set of harmonized surveys of long-term intragenerational mobility that include income, labor market, and other because these estimates do not suffer from socioeconomic information spanning more traditional measurement error. This decision than 20 years of nationally representative is not without a drawback: the lower-bound cross-sectional surveys, totaling more than estimates will underestimate upward mobil- 250 surveys for the region.10 Data and periods ity, painting a less-rosy picture in terms of used, by country, are presented in annex 4.1. welfare improvements. But focusing on the We focus on three measures of directional lower-bound estimates will also underesti- mobility for each country and the region (as mate downward mobility and, as such, the chapter 2 discusses in more detail): role of risk and vulnerability will be inher- ently underplayed. The results should be • The proportion of people who move interpreted accordingly. In fact, to reinforce across thresholds (poverty and the middle this point—for the case of downward mobil- class), which provides an estimate of over- ity—we also discuss the upper-bound results. all mobility (how much) as in traditional Although the technique is no substitute for transition matrices. having actual panel data, it has a number of • Aggregate mobility as the sum of all advantages that enable great strides in both income changes (in levels and percent- describing and understanding socioeconomic ages), which gives us a sense of the magni- mobility in Latin America: tude (how far) of long-term mobility. • Transitions out of poverty (US$4 pur- • It constructs a synthetic panel for every chasing power parity [PPP] per day) and individual in the sample; therefore, it into the middle class (US$10 PPP per day) overcomes the attrition problem of actual in their own right. This allows us to study panel data sets. dynamics for each country related to these • Instead of constructing panels of cohort two thresholds and describe the types of averages, as in the classical pseudo-panel groups that are more (or less) successful 98 MOBILIT Y WITHIN GENERATIONS in moving upward (or downward) and across classes (the sum of the off-diagonal to explore the types of policies that may cells in the matrix).11 have contributed to the observed patterns. As the table 4.1 suggests, Latin America has experienced dramatic mobility in the past The mobility estimates (as well as the 15 years. A number of striking results emerge: decomposed measures in annex 4.2) permit us to identify how far the poor move out of • Out of every 100 Latin Americans, 43 poverty (or the vulnerable move into middle changed their economic status during the class) and whether these movements vary period. across the income distribution. • There is considerably more upward than downward mobility: out of the 43 people changing economic status, only two expe- Income mobility in Latin rienced a worsening of their status (into America: The past two decades poverty or out of the middle class). Note that, as discussed above, these are lower- Overall long-term mobility bound estimates (we come back to this Using synthetic panels as described above, point later). we can estimate intragenerational mobil- • Despite the large levels of mobility, table ity measures as in chapter 2 for each indi- 4.1 also suggests that a large part of the vidual between two periods, from which we population is immobile. For example, can obtain aggregate mobility measures for more than one in five Latin Americans a country or the region as a whole. Table remained chronically poor throughout 4.1 presents an aggregate regional transi- the whole period, while approximately the tion matrix for Latin America in terms of same proportion remained steadily in the the three economic groups discussed earlier: middle class. the poor (those with incomes below US$4 PPP per day), the vulnerable (with incomes These trends vary across countries. For between US$4–10 PPP per day), and those in example, 50–60 percent of the population the middle and upper classes (with incomes in countries such as Brazil, Chile, Colom- above US$10 PPP per day). In this context, bia, and Costa Rica moved across one of the a measure of intragenerational mobility is three economic groups during the past 15 the share of the total population that moved years, as shown in figure 4.1). This compares TABLE 4.1 Intragenerational mobility in Latin America over past 15 years (circa 1995–2010) percentage of population Destination (c. 2010) Poor Vulnerable Middle class Total Poor 22.5 21.0 2.2 45.7 Origin (c.1995) Vulnerable 0.9 14.3 18.2 33.4 Middle class 0.1 0.5 20.3 20.9 Total 23.4 35.9 40.7 100.0 Source: Data from SEDLAC (Socio-Economic Database for Latin America and the Caribbean). Note: “Poor” = individuals with a per capita income lower than US$4. “Vulnerable” = individuals with a per capita income of US$4–US$10. “Middle class” = individuals with a per capita income higher than US$10. Poverty lines and incomes are expressed in 2005 US$ PPP per day. PPP = purchasing power parity. Years vary across countries as follows: Argentina 1994 and 2009; Bolivia 1992 and 2007; Brazil 1990 and 2009; Chile 1992 and 2009; Colombia 1992 and 2008; Costa Rica 1989 and 2009; Dominican Republic 1996 and 2009; Ecuador 1995 and 2009; El Salvador 1991 and 2008; Guatemala 2000 and 2006; Honduras 1994 and 2009; Mexico 2000 and 2008; Nicaragua 1998 and 2005; Panama 1994 and 2009; Peru 1999 and 2009; Paraguay 1999 and 2009; Uruguay 1989 and 2009; and República Bolivariana de Venezuela 1992 and 2006. The table shows lower-bound mobility estimates using the Dang et al. (2011) technique. Results are weighted using country-specific population estimates of the last available period. MOBILIT Y WITHIN GENERATIONS 99 FIGURE 4.1 Sliders, climbers, and stayers: Intragenerational mobility in Latin America, by country 100 80 Percentage of population 60 40 20 0 9 9 09 08 09 9 7 08 9 06 09 9 09 9 09 8 05 6 00 00 00 00 00 00 00 −0 −0 20 20 20 20 20 20 20 20 20 −2 −2 −2 −2 −2 −2 −2 00 00 0− 2− 5− 1− 2− 5− 9− 4− 8− 92 89 94 92 99 96 99 20 20 99 99 99 99 99 99 98 99 99 19 19 19 19 19 19 19 ico ala il 1 a1 r1 r1 B1 a1 y1 a1 a1 ile ca as ia ru lic ay em ex do do ua az bi m in gu ,R liv Pe ur ub Ri gu Ch M nt m na Br ua lva ug at ela ra Bo nd sta ep ra lo ge Gu Pa ca Ec Ur Sa zu Ho Pa Co nR Co Ar Ni ne El ica Ve in m Do Sliders Climbers Stayers Source: Data from SEDLAC (Socio-Economic Database for Latin America and the Caribbean). Note: The figure shows lower-bound mobility estimates using the Dang et al. (2011) technique. “Sliders” refers to those individuals who move downward. “Climbers” refers to those who move upward. “Stayers” refers to those who did not change status. “Poor” = individuals with a per capita income lower than US$4. “Vulnerable” = individuals with a per capita income of US$4–US$10. “Middle class” = individuals with a per capita income higher than US$10. Poverty lines and incomes are expressed in 2005 US$ PPP per day. PPP = purchasing power parity. with significantly lower mobility (of around whole region) was US$3.30 PPP per day per or less than 20 percent) in the populations of capita. In addition, the marginal distribu- Argentina, Guatemala, Mexico, Nicaragua, tions (last column of table 4.2) show that the and Paraguay. Although downward mobility net income change was different for different (“sliders”) is also part of the overall mobility parts of the distribution: among those who measure, as the figure shows, upward mobil- escaped poverty (into vulnerability), income ity (“climbers”) is driving the mobility results; grew by US$2.80, while among the vulnerable downward mobility affects only a small who entered the middle class, income grew by number of countries (such as the Dominican US$6.90. Similarly, households that remained Republic, Paraguay, and República Bolivari- poor over the period experienced a small ana de Venezuela). increase in median income of US$1 PPP, while The synthetic panels can allow us to esti- those who remained in the vulnerable class mate additional measures of directional saw their net incomes grow by US$2.60 PPP. mobility (described in chapter 2). For exam- Among the sliders, those who entered poverty ple, we may want to know how much individ- experienced a net income decline of US$0.80, uals’ incomes grew (in levels or percentages) while those who fell out of the middle class over the period for the whole distribution or saw a net income loss of US$1.80 PPP. for specific groups.12 Table 4.2 indicates that Table 4.3 presents similar trends in total mobility (as net income change for the growth rates over the past 15 years. For the 100 MOBILIT Y WITHIN GENERATIONS TABLE 4.2 Intragenerational mobility in Latin America, by median income change, (circa 1995–2010) US$ PPP per capita per day Destination (c. 2010) Poor Vulnerable Middle class Total Poor 1.0 2.8 8.4 1.8 Origin (c.1995) Vulnerable –0.8 2.6 6.9 4.9 Middle class –1.2 –1.8 11.6 11.3 Total 0.9 2.6 7.9 3.3 Source: Data from SEDLAC. Note: SEDLAC = Socio-Economic Database for Latin America and the Caribbean. “Poor” = individuals with a per capita income lower than US$4. “Vulner- able” = individuals with a per capita income of US$4–US$10. “Middle class” = individuals with a per capita income higher than US$10. Poverty lines and incomes are expressed in 2005 US$ PPP (purchasing power parity) per day. Years vary across countries as follows: Argentina 1994 and 2009; Bolivia 1992 and 2007; Brazil 1990 and 2009; Chile 1992 and 2009; Colombia 1992 and 2008; Costa Rica 1989 and 2009; Dominican Republic 1996 and 2009; Ecuador 1995 and 2009; El Salvador 1991 and 2008; Guatemala 2000 and 2006; Honduras 1994 and 2009; Mexico 2000 and 2008; Nicaragua 1998 and 2005; Panama 1994 and 2009; Peru 1999 and 2009; Paraguay 1999 and 2009; Uruguay 1989 and 2009; and República Bolivariana de Venezuela 1992 and 2006. The table shows lower-bound mobility estimates using the Dang et al. (2011) technique. Results are weighted using country-specific population estimates of the last avail- able period. TABLE 4.3 Intragenerational mobility in Latin America, by median income change, (circa 1995–2010) mecdian percentage income change Destination (c. 2010) Poor Vulnerable Middle Class Total Poor 86.6 110.0 269.8 99.5 Origin (c. 1995) Vulnerable –15.7 51.7 106.4 81.1 Middle Class –9.2 –15.4 65.1 63.5 Total 84.6 88.2 89.2 87.4 Source: Data from SEDLAC. Note: “Poor” = individuals with a per capita income lower than US$4. “Vulnerable” = individuals with a per capita income of US$4–US$10. “Middle class” = individuals with a per capita income higher than US$10. Poverty lines and incomes are expressed in 2005 US$ PPP per day. PPP = purchasing power parity. SEDLAC = Socio-Economic Database for Latin America and the Caribbean. Years vary across countries as follows: Argentina 1994 and 2009; Bolivia 1992 and 2007; Brazil 1990 and 2009; Chile 1992 and 2009; Colombia 1992 and 2008; Costa Rica 1989 and 2009; Dominican Republic 1996 and 2009; Ecuador 1995 and 2009; El Salvador 1991 and 2008; Guatemala 2000 and 2006; Honduras 1994 and 2009; Mexico 2000 and 2008; Nicaragua 1998 and 2005; Panama 1994 and 2009; Peru 1999 and 2009; Paraguay 1999 and 2009; Uruguay 1989 and 2009; and República Bolivariana de Venezuela 1992 and 2006. The table shows lower-bound mobility estimates using the Dang et al. (2011) technique. Results are weighted using country-specific population estimates of the last avail- able period. period covering the study, incomes grew by poor, compared with 110 percent for those an average of almost 90 percent (with an who entered the vulnerable group and 270 annualized rate of 6 percent). The results percent for the few who made it into the mid- across groups show a progressive trend: much dle class (2.2 percent of the total population, of this growth occurred among those origi- as in table 4.1). Among the “sliders,” those nally poor or vulnerable. Specifically, while originally vulnerable who fell into poverty incomes among the originally poor doubled, saw an average income reduction of almost the originally vulnerable experienced an 16 percent. increase in incomes of 81 percent, compared These trends also vary across countries, with 64 percent for those originally in the as shown in figure 4.2. For example, Brazil, middle class (last column of table 4.3). As Chile, and Honduras have had the highest before, additional differences within specific median income growth since the early 1990s groups exist. For the originally poor, incomes (of almost 150 percent), while Guatemala, grew by 87 percent for those who remained Paraguay, and República Bolivariana de MOBILIT Y WITHIN GENERATIONS 101 FIGURE 4.2 Intragenerational mobility in Latin America, by country percentage median income change 150 Percentage income change 100 50 0 09 9 9 9 08 09 7 08 9 09 09 9 05 09 8 6 6 9 00 00 00 00 00 00 00 00 00 00 20 20 20 20 20 20 20 20 −2 −2 −2 −2 −2 −2 −2 −2 −2 −2 0− 2− 5− 1− 9− 5− 8− 4− 92 94 89 92 99 96 00 92 00 99 99 99 99 99 98 99 99 99 19 19 19 19 19 19 20 19 20 19 il 1 a1 r1 r1 y1 a1 a1 a1 ile as ca ia ru lic ico ela ala ay do do ua az bi m gu in liv Pe ur ub Ri gu Ch em ex zu nt m na Br ua lva ug ra Bo nd sta ep ra lo ge M ne Pa ca Ec at Ur Sa Ho Pa Co nR Co Ar Ve Ni Gu El ica in m Do Source: Data from SEDLAC. Note: Figure shows the synthetic panel median growth rate in incomes. Horizontal dashed line shows overall growth rates, weighted using country-specific population estimates of the last period from SEDLAC (Socio-Economic Database for Latin America and the Caribbean). Figure based on lower-bound mobil- ity estimates using the Dang et al. (2011) technique. Venezuela were among the worst perform- Latin America who changed status during the ers (with median income growth over their period, 23 were originally poor who exited respective periods of less than 10 percent). poverty, 18 were vulnerable who entered the In sum, these results suggest large income middle class, while 2 fell into poverty or out mobility in the region, driven by upward of the middle class. mobility out of poverty and into the vulner- Again, this varies across countries. Among able or middle classes. For these long-term those countries with higher overall mobility mobility trends, the results suggest very (Brazil, Chile, Colombia, and Costa Rica), little downward mobility. The next section those exiting poverty contribute at least explores these results further. equally to overall mobility trends relative to those who were originally vulnerable and entered the middle class, as shown in figure Unravelling the box: 4.3. By contrast, mobility in countries with Exiting poverty and entering lower overall mobility is driven by people the middle class moving into the middle class from the vul- Is this economic mobility similar across dif- nerable group (Argentina, Uruguay, and ferent parts of the distribution? The results República Bolivariana de Venezuela). For above suggest that this is not the case. Focus- countries such as Paraguay and República ing on initial economic status, table 4.1 sug- Bolivariana de Venezuela, downward mobil- gests that out of the 43 people for each 100 in ity (into poverty) is also more pronounced. 102 MOBILIT Y WITHIN GENERATIONS FIGURE 4.3 Mobility for whom? Contribution to overall mobility of initial economic status in Latin America, by country Percentage of people moving 60 40 20 0 9 9 09 08 09 9 7 08 9 06 09 9 09 9 09 8 05 6 00 00 00 00 00 00 00 00 00 20 20 20 20 20 20 20 20 20 −2 −2 −2 −2 −2 −2 −2 −2 −2 0− 2− 5− 1− 2− 5− 9− 4− 8− 92 89 94 92 99 96 99 00 00 99 99 99 99 99 99 98 99 99 19 19 19 19 19 19 19 20 20 il 1 a1 r1 r1 B1 a1 y1 a1 a1 ile ca as ia ru lic ay ico ala do do ua az bi m in gu ,R liv Pe ur ub Ri gu Ch em ex nt m na Br ua lva ug ela ra Bo nd sta ep ra lo ge M Pa ca Ec at Ur Sa zu Ho Pa Co nR Co Ar Ni Gu ne El ica Ve in m Do Initial poor Initial vulnerable (downward) Initial vulnerable (upward) Initial middle class Source: Data from SEDLAC. Note: The figure shows lower-bound mobility estimates using the Dang et al. (2011) technique. “Initial poor” refers to those who were poor in the first period. “Initial middle class” refers to those who belonged to the middle class in the first period. “Initial vulnerable (downward)” refers to those who were initially vulnerable in the first period and became poor in the second period. “Initial vulnerable (upward)” refers to those who were initially vulnerable in the first period and became middle class in the second period. “Poor” = individuals with a per capita income lower than US$4. “Vulnerable” = individuals with a per capita income of US$4–US$10. “Middle class” = individuals with a per capita income higher than US$10. Poverty lines and incomes are expressed in 2005 US$ PPP per day. PPP = purchasing power parity. The horizontal dashed line shows overall mobility in Latin America, weighted using country-specific population estimates of the last period from SEDLAC = Socio-Economic Database for Latin America and the Caribbean. How large are these mobility changes? c). Once they entered the middle class, most One way to answer this question is by look- of them were in the range of US$10–15 PPP ing at the initial and final income distribution per day (panel d). among those who exited poverty or entered These trends are generalized for most the middle class. This is summarized in figure countries in the region, despite some differ- 4.4 for the case of Uruguay. As can be seen, ences. For example, those who exited poverty most of those who exited poverty between originally had a median income of between 1989 and 2009 were near the poverty line in US$2 and US$3 PPP per capita per day as 1989 in terms of income per capita (panel a). shown in figure 4.5, panel a. In most coun- In addition, most of those who exited poverty tries, the net income change was large enough hover above the poverty line with a median to bring those individuals’ median income up income of around US$6 PPP per day (panel to around US$6 PPP per capita per day. This b). Few households crossed the middle-class increase corresponds to more than double the threshold. Indeed, most of the population per capita incomes of those exiting poverty who entered the middle class between 1989 in these countries (for their respective peri- and 2009 in Uruguay were near the middle- ods). And consistent with the earlier results, class line (US$10 PPP per day) in 1989 (panel in none of the countries in the region did the MOBILIT Y WITHIN GENERATIONS 103 FIGURE 4.4 Upward mobility out of poverty: Origin and destination in Uruguay, 1989–2009 a. Initial income distribution: b. Final income distribution: poor in 1989 who are nonpoor in 2009 poor in 1989 who are nonpoor in 2009 1.0 0.4 0.8 0.3 0.6 0.2 0.4 0.1 0.2 0 2 4 6 0 2 4 6 8 10 12 14 16 Per capita income, 2005 US$ PPP/day, year 1989 Per capita income, 2005 US$ PPP/day, year 2009 c. Initial income distribution: poor and vulnerable in 1989 d. Final income distribution: poor and vulnerable in 1989 who entered middle class in 2009 who entered middle class in 2009 0.4 .20 0.3 .15 0.2 .10 0.1 .05 0 4 8 10 12 0 4 8 10 12 16 20 24 28 Per capita income, 2005 US$ PPP/day, year 1989 Per capita income, 2005 US$ PPP/day, year 2009 Source: Data from SEDLAC. Note: The figure shows lower-bound mobility estimates using the Dang et al. (2011) technique. Panels a and b show the initial and final income distributions of those originally poor who escaped poverty, respectively. Panels c and d show the initial and final income distributions of those originally poor or vulnerable who entered middle class, respectively. “Poor” = individuals with a per capita income lower than US$4. “Vulnerable” = individuals with a per capita income of US$4–US$10. “Middle class” = individuals with a per capita income higher than US$10. Dashed vertical lines represent the US$4 poverty and the US$10 vulnerable poverty lines. Poverty lines and incomes are expressed in 2005 US$ PPP per day. PPP = purchasing power parity. SEDLAC = Socio-Economic Database for Latin America and the Caribbean. income gains among the poor allow them to period were around US$6 PPP per capita as surpass US$10 PPP (median), thus putting shown in figure 4.5, panel b. The net income them in the middle class. In other words, change brought those individuals’ median those who exited poverty in Latin America income up to around US$14 PPP, suggesting in the past two decades are neither poor that the median person in this group doubled nor in the middle class: they are vulnerable. his or her per capita income. Similarly, for those who entered the middle Instead of looking at levels, one final exer- class, median incomes at the beginning of the cise explores whether income growth differs 104 MOBILIT Y WITHIN GENERATIONS FIGURE 4.5 Intragenerational upward mobility in Latin America: Origin and destination, by country median income, 2005 US$ PPP per capita per day a. Out of poverty 8 2005 US$ PPP/day 6 4 2 0 09 sta 199 009 Ur a 19 009 9 ua 992 9 lva 199 08 Ve agu 91− 9 ela 999 08 9 06 ep 199 09 19 007 09 9 o 2 05 Gu a 19 −08 20 9 6 Co ay 1 200 0 r 1 200 B 1 200 a 1 200 00 −0 20 0 0 20 0 0 ru −20 20 2 2 bi 9−2 −2 m 2−2 2 2 em 4−2 in 000 00 0− ur 92− 4− − do 5− − ica oliv 95− 2− gu 9− 8− 89 6 99 98 9 9 9 99 9 19 9 9 19 19 ala il 1 a1 1 Ar exic ile s r ay a a lic do a az c i ,R Pe ub Ri u Ch M nt m na Br ug at ra nd lo ge B Pa ca Ec r Sa zu Ho Pa nR Co Ni ne El in m Do Median initial income Median net income change b. Into middle class 15 2005 US$ PPP/day 10 4 0 9 09 09 6 09 9 9 08 09 08 7 09 9 9 09 8 05 6 00 00 00 00 00 00 00 −0 −0 20 20 20 20 20 20 20 20 20 −2 −2 −2 −2 −2 −2 −2 00 00 0− 9− 9− 2− 5− 1− 4− 5− 8− 92 92 94 89 92 99 96 20 20 99 98 99 99 99 99 99 99 99 19 19 19 19 19 19 19 ico ala il 1 y1 y1 a1 r1 r1 a1 a1 a1 ile RB as ca ia ru lic m ex do do ua ua az bi in am gu liv Pe ur ub Ri Ch te M nt m la, Br ua lva ug ag ra Bo nd n a sta ep lo ge ue Gu Pa ca Ec r Ur Sa Ho Pa Co nR Co Ar ez Ni El n ica Ve in m Do Median initial income Median net income change Source: Data from SEDLAC. Note: The figure shows lower-bound mobility estimates using the Dang et al. (2011) technique. Panel a shows the median initial income and median income change of those originally poor who escaped poverty. Panel b shows the median initial income and median income change of those originally poor or vul- nerable who entered the middle class. “Poor” = individuals with a per capita income lower than US$4. “Vulnerable” = individuals with a per capita income of US$4–US$10. “Middle class” = individuals with a per capita income higher than US$10. Poverty lines and incomes are expressed in 2005 US$ PPP per day. PPP = purchasing power parity. Horizontal dashed lines show overall mobility in Latin America, weighted using country-specific population estimates of the last period from SEDLAC = Socio-Economic Database for Latin America and the Caribbean. across the income distribution. Figure 4.6 • For Costa Rica, the traditional static GIC shows growth incidence curves (GIC) across (anonymous as it looks at changes in deciles in two countries using two approaches: mean incomes for a specific part of the anonymous and non-anonymous: distribution) indicates a regressive story MOBILIT Y WITHIN GENERATIONS 105 FIGURE 4.6 Growth incidence curves for Costa Rica and El Salvador, using anonymous and non-anonymous information percentage income change, by decile a. Costa Rica b. El Salvador 130 110 Percentage income change between Percentage income change between 120 100 1989 and 2009 1991 and 2008 110 90 100 90 80 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10 Deciles Deciles Anonymous Non−anonymous Anonymous Non−anonymous Source: Data from SEDLAC. Note: All figures present the growth incidence curves (GICs), which show percentage income changes by decile of the per capita income distribution. First-round incomes are actual incomes in surveys, while second-round incomes come from lower-bound estimates using the Dang et al. (2011) technique. Anonymous GICs treat first and second round as if they were cross-sectional surveys, while non-anonymous estimates come from respecting the synthetic panel structure. The left vertical solid line represents the proportion of poor, while the right solid vertical line represents the proportion of vulnerable. “Poor” = individuals with a per capita income lower than US$4. “Vulnerable” = individuals with a per capita income of US$4–US$10. “Middle class” = individuals with a per capita income higher than US$10. Poverty lines and incomes are expressed in 2005 US$ PPP per day. PPP = purchasing power parity. SEDLAC = Socio-Economic Database for Latin America and the Caribbean. where income growth was higher among because one does not follow the same indi- the richer part of the distribution (the vidual or household over time. If what we middle class). By contrast, the non-anony- want to understand is an individual’s welfare mous GIC (based on synthetic panels that trajectory over time (true or synthetic), panel follow the same households over time and data are essential. look at mean changes in incomes for a specific part of the distribution) show that Downward mobility revisited the poor actually had at least the same performance in income growth over the An additional insight from looking at the same period. short-term mobility trends is in helping us • In El Salvador, the non-anonymous results understand one of the most striking results also suggest very progressive changes in of long-term mobility in Latin America: the the income distribution (the poorer parts extremely low downward mobility, both in of the distribution growing faster) relative terms of the new poor as well as those who to the conclusions one would reach from exit the middle class. As figure 4.7 summa- the anonymous trends. rizes, with the exception of the Dominican Republic, Paraguay, and República Boli- Similar exercises in other countries reveal variana de Venezuela, the population that similar patterns. Taken together, the results fell below their original class over the past suggest heterogeneity across countries in 15-year period is small. Does this suggest terms of mobility but also emphasize how that we should not care about interventions the use of traditional incidence curves can oriented to prevent people from falling into hide the true nature of mobility trajectories poverty or out of the middle class (such as 106 MOBILIT Y WITHIN GENERATIONS FIGURE 4.7 Downward intragenerational mobility into poverty and out of middle class in Latin America, by country a. Population entering poverty (% of originally not poor) 20 Percentage of people moving 15 10 5 0 06 99 09 99 09 20 5 99 08 Gu a 19 009 a 2 007 Ur s 19 −06 ua 989 9 Pa r 19 009 09 lva 992 9 sta 199 09 Br 989 8 9 9 99 09 08 0 y 1 200 0 0 00 Co ru 1 200 20 c 1 20 0 20 − a 1 20 20 0 20 0 20 gu 6−2 2 −2 −2 −2 −2 2 00 0 2− ub 99− 8− 4− − − Ch 95− 1− Pe 90− 9− 2− 00 92 94 95 99 99 19 9 9 ge ico B1 a1 a1 1 1 il 1 a1 Ho mal ay ile r ca x do do a i ua az in i m bi ,R e l liv ur Ri gu e M nt m na ug at ela ra Bo nd ep ra lo ca Ec Sa zu Pa nR Co Ar Ni ne El ica Ve in m Do b. Population exiting middle class (% of originally in middle class) 25 Percentage of people moving 20 15 10 5 0 at 994 7 ug a 20 9 ua 992 8 6 9 nt 996 8 99 09 5 6 lva 999 09 9 Pa r 19 009 Br 995 9 sta 199 009 89 9 99 09 08 09 0 00 0 00 00 0 Ho ia 1 200 −0 r 1 200 00 0 as −20 ile −20 0− a 1 20 0 ca −20 0 20 Ni ina −20 −2 −2 ex 9−2 2 −2 m 5−2 −2 −2 00 gu 94− 8− Pe 89− − 2− ub 200 2 1 92 0 99 99 9 9 19 19 19 9 9 ico 1 y1 1 1 a1 1 a1 1 al RB ay ru Co zil Ar lic em do do ua bi liv ur a Ri u Ch M m la, na ag ra Bo nd ep lo ge ue Gu ca Ec r Ur Sa Pa Co nR ez El n ica Ve in m Do Source: Data from SEDLAC. Note: The figure shows lower-bound mobility estimates using the Dang et al. (2011) technique. Panel a shows the proportion of those originally not poor who enter poverty. Panel b shows the proportion of those originally in the middle class who became poor or vulnerable. “Poor” = individuals with a per capita income lower than US$4. “Vulnerable” = individuals with a per capita income of US$4–US$10. “Middle class” = individuals with a per capita income higher than US$10. Both lines are expressed in 2005 US$ PPP per day. PPP = purchasing power parity. Horizontal dashed lines show overall mobility in Latin America, weighted using country-specific population estimates of the last period from SEDLAC = Socio-Economic Database for Latin America and the Caribbean. social insurance or risk management poli- this chapter use the lower bound. The choice cies)? There is reason to be cautious. was based on two reasons, as discussed ear- As was described earlier, the synthetic lier: (a) it is based on fewer technical assump- panel approach allows us to estimate lower tions (the strongest being that it does not and upper bounds for the mobility measures include measurement error); and (b) by con- we discuss here. All the results presented in struction, it estimates a lower-bound estimate MOBILIT Y WITHIN GENERATIONS 107 FIGURE 4.8 Downward mobility into poverty in Latin American revisited, by country percentage of those originally not poor 50 Percentage of people moving 40 30 20 10 0 05 6 06 9 7 9 9 08 8 09 9 08 9 09 09 9 09 09 −0 00 00 00 00 −0 00 00 00 20 20 20 20 20 20 20 20 20 −2 −2 −2 −2 −2 −2 −2 00 00 8− 2− 1− 5− 2− 4− 5− 9− 0− 20 99 92 94 96 20 99 89 92 99 99 99 99 99 99 99 98 99 19 19 19 19 19 19 19 ala ico a1 B1 r1 r1 a1 a1 a1 y1 il 1 ay ia as lic ru ca ile em ex do do ua gu bi in m az ,R liv Pe ur ub Ri gu Ch M nt m na lva ua Br ug at ela ra Bo nd sta ep ra lo ge Gu Pa ca Ec Ur Sa zu Ho Pa Co nR Co Ar Ni ne El ica Ve in m Do Source: Data from SEDLAC. Note: The figure shows upper-bound mobility estimates using the Dang et al. (2011) technique. It presents the proportion of those originally not poor who enter poverty. “Poor” = individuals with a per capita income lower than US$4. The poverty line is expressed in 2005 US$PPP per day. PPP = purchasing power parity. The horizontal dashed line shows overall Latin American mobility, weighted using country-specific population estimates of the last period from SEDLAC (Socio-Economic Database for Latin America and the Caribbean). that allows us to discuss the most conservative mobility is an important issue to consider scenario of directional mobility (both upward from a policy point of view. and downward). In this sense, we expect the Using the upper bound to discuss down- “true” mobility estimates to be higher. ward mobility also suggests another interest- Therefore, we also present the upper- ing result: the overall average probability in bound estimates of downward mobility in fig- Latin America of becoming poor is slightly ure 4.8. As can be seen, at the upper bound, higher than 10 percent. As discussed in more there is considerable downward mobility into detail in chapter 2, we define the middle-class poverty. For countries such as Guatemala, threshold at US$10 PPP per day using a vul- Nicaragua, and República Bolivariana de nerability concept of the probability of fall- Venezuela, there is a 30 percent probability of ing into poverty (based on a 10 percent prob- falling into poverty, while for Bolivia, Hon- ability). As figure 4.8 shows, 10 percent turns duras, and Paraguay, it is 20 percent. At the out to be close to the average probability in other end, countries such as Brazil, Chile, and the region, and in fact it is the prevailing Uruguay still exhibit low rates of downward probability for middle-class countries such mobility (also see box 4.2). To the extent that as Argentina, Colombia, and Costa Rica. In the “true” mobility estimates are somewhere this sense, this probability provides empirical between the lower and upper bounds, these validation of the justification of the “middle results do indicate that long-term downward class” definition used in this volume. 108 MOBILIT Y WITHIN GENERATIONS BOX 4.2 The welfare costs of downward mobility in Nicaragua Premand and Vakis (2010) use three rounds of panel by 13 percent and increased the probability of falling data from Nicaragua between 1998 and 2005 to out of the middle class (the top tercile) by 25 percent. study downward mobility. They fi nd that more than A novel fi nding in the study is that weather shocks 25 percent of the vulnerable (using a defi nition com- also increase the probability of poverty persistence: parable to that used in this chapter, based on con- poor households affected by weather shocks have a sumption terciles) became poor over the period they 10 percent higher probability of staying poor in sub- studied, attributed partly to uninsured risks. sequent periods. This fi nding provides new evidence Using matching and double difference techniques, of how shocks can also prevent upward mobility and they also fi nd causal evidence of the impact of past perpetuate poverty traps.a weather shocks in triggering downward mobility into Taken together, these results point to large poten- poverty. Specifically, they fi nd that a severe drought tial gains from social risk management policies, five years earlier increased the probability of the vul- targeting both the near-poor (vulnerable) and the nerable (those in the second tercile) becoming poor extreme poor. a. Cruces, Glüzmann, and López-Calva (2011) also find evidence that economic crises in Argentina have permanent impacts on increased mortality and long-term impacts through deteriorating health outcomes (low birth weight), with substantial consequences in terms of future income-generating capacity. Mobility profiles: Insights for had 10 years of education 15 years earlier (in policy 1995), compared with only 6 years among heads of households that today are poor. How much do initial conditions matter? And although the levels differ (also because the beginning and end of the period Having established the long-term mobility observed differ across countries), these trends trends and stylized facts in the Latin Ameri- are consistent across countries. Interestingly, can region, we explore descriptively some there seems to be a general trend that today’s of the potential channels that are associated middle-class households had twice the initial with mobility. A first way to do this is to education levels of today’s poor households, understand the correlates of upward mobil- while the vulnerable class falls in between. ity with household characteristics. Given the Similar trends emerge in other indicators. synthetic panel approach, one limitation is For example, today’s middle class was more that we can only explain the extent to which likely to have a household head working in the socioeconomic characteristics at the ini- formal sector at the initial period relative to tial period are correlated with mobility (as the other two classes. In the case of Brazil, 80 opposed to changes in those characteristics). percent of household heads that are today in We first explore how today’s households the middle class were already working in the across the three economic classes (poor, vul- formal sector in 1990, compared with only 40 nerable, and middle class) looked 15 years percent among today’s poor. Again, the vul- earlier (figure 4.9). The results, while not sur- nerable households fall in between, and the prising, suggest that households that had bet- trends are consistent across countries. Simi- ter initial socioeconomic indicators are more lar trends for location of residence (middle- likely to end up in a higher economic class 15 class households are more likely to have years later. For example, households that are resided in urban areas earlier) and access to today in the middle class were, on average, services like water and electricity (households more educated than those in the vulnerable in today’s higher economic classes already class who, in turn, had more education than had better access to basic services 15 years the poor. In Ecuador, for example, a typical earlier). There is no difference, however, con- middle-class household head (today) already cerning the gender of household heads. MOBILIT Y WITHIN GENERATIONS 109 FIGURE 4.9 Economic class (circa 2010) and initial characteristics (circa 1995) in Latin America, by country a. Gender b. Sector of work 100 100 90 90 80 80 Proportion of formal 70 70 Average male 60 60 50 50 40 40 30 30 20 20 10 10 0 0 zu ma Ec xico ca Bo B a a ca Ur ile at ia ca la sta a ep u Sa na ico ala a Ho blic g r Br s M ay lo y ile ru ay r ra l r Ar ado il bi a do in Pa azi Co agu ,R gu Co ua r do Gu liv Ni ma Ri n R Pe az Ri ur Pe Ch gu gu El enti Ch ne na m em ex nt ela u lva ug e ra lva Br nd sta u e r ge ra M Ve Pa ca at Sa Pa Co Ar Ni Gu El ica in m Do Poor Vulnerable Middle class Poor Vulnerable Middle class c. Education d. Area of residence 10.0 100 90 Average years of education 80 7.5 Proportion of urban 70 60 5.0 50 40 30 2.5 20 10 0 0 Ur bia ra a Pa , RB ica Me ile B ep a a zu la a Ve Rep co Ec ivia ala lo a Ho Peru Ve ate ca in ica o ile n a Sa s Pa blic zu lic Pa ay sta r Sa ras ca ru at or il ge r y sta y M il Co ado Pa agu El ura n R gu m in Co am Ar ado Ho Ric ,R m N exic az ua Co gua ne ma az Gu Ri Ch n xi ne ub Ch Ni Pe Gu lvad gu em m nt na El du l ela ela Br u ica ra Br Bo ug n nd lv r u ra in m Do Do Poor Vulnerable Middle class Poor Vulnerable Middle class Source: Data from SEDLAC (Socio-Economic Database for Latin America and the Caribbean). Note: The figure shows average characteristics (gender, sector of work, education, and area of residence) defined circa 1995 by economic status defined circa 2010 (poor, vulnerable, and middle class). “Poor” = individuals with a per capita income lower than US$4. “Vulnerable” = individuals with a per capita income of US$4–US$10. “Middle class” = individuals with a per capita income higher than US$10. Both lines and incomes are expressed in 2005 US$PPP/Per Day. PPP = purchasing power parity. What drives upward mobility? To try and upward mobility. A few insights stand out. answer this question, we estimate conditional For example, gender is not associated with mobility measures out of poverty and into the different levels of mobility. As figure 4.10, middle class for various subpopulations using panel a, indicates, the only variation regard- different initial characteristics. This allows us ing gender is across countries. to say something about how much the initial However, education strongly predicts up- levels of assets and endowments matter for ward mobility, both out of poverty and into 110 MOBILIT Y WITHIN GENERATIONS the middle class. Specifically, three messages access to the formal sector is generally asso- emerge on education: ciated with slightly larger probabilities of upward mobility into the middle class.14 • Secondary or university education is asso- The results with respect to geography and ciated with larger probabilities of upward long-term mobility are also interesting, at mobility than primary education (figure least for two reasons: Residing in urban areas 4.10, panel b). For example, in Costa is in general associated with higher levels of Rica, 8 out of 10 originally poor with mobility out of poverty or into the middle higher education left poverty during the class (see figure 4.10, panel d). For example, period, compared with only 6 out of 10 households exiting poverty in countries such for those with primary education. as Brazil, Honduras, Panama, Peru, or Gua- • These results are consistent across coun- temala had up to 50 percent higher probabil- tries with different overall mobility ity of doing it if they resided in urban areas trends. (we find similar trends for those households • Even if suggestive, the figure also indi- that entered the middle class). The data in cates that a university degree is much some countries also allow us to identify more important for entries into the mid- recent migrants from rural to urban areas. As dle class than to exiting poverty. Specifi- such, we compare whether households that cally, although having more than primary moved to urban areas at the initial period (15 education is sufficient to improve one’s years earlier) were more likely to experience odds of upward mobility out of poverty, upward mobility than those that lived in rural university education provides an addi- areas. The results seem to indicate that this tional increase in the probability to enter is the case, at least for some countries. For the middle class (beyond the increase pro- example, households in Brazil, Guatemala, vided by primary and secondary educa- and Honduras that migrated to urban areas tion).13 For example, only 20 percent of were more likely to exit poverty during the originally vulnerable Hondurans with pri- period than those that lived in rural areas (see mary education entered the middle class figure 4.10, panel e). To the extent that these during the period. This compares with 40 results capture the ability to take advantage of percent among those with secondary edu- local opportunities as a channel for upward cation and more than 60 percent among mobility, they seem to highlight the role of those with a university degree. We might economic opportunities and geography. expect such returns from the steeper age- income profi le of those with higher edu- How important is economic growth for cation, but the correlation nonetheless long-term mobility? indicates the high premium of education. Again, these trends vary across countries. Do countries that managed to grow faster over the period also have higher (directional) A similar analysis can be done for charac- mobility? It turns out, yes. Although a com- teristics relating to labor market access. For plex analysis is difficult to do because of data example, Figure 4.10, panel c, shows condi- requirements, some simple correlations are tional mobility measures out of poverty and informative, even if they should be taken with into the middle class based on the households a grain of salt. Figure 4.11 presents correla- head’s sector of work. As the results indicate, tions between the conditional mobility out of with respect to poverty exits, having access poverty (left panel) and into the middle class to the formal work sector provides a small (right panel) with annualized GDP growth advantage over those in the informal sector rates across the region. For both poverty in only a few of the countries, while for the exit and middle-class entry, countries with rest there is not a difference. By contrast, higher growth rates are strongly associated Percentage of people moving Percentage of people moving Percentage of people moving 0 20 40 60 80 100 0 10 20 30 40 50 60 70 80 0 20 40 60 80 100 Ch C C ile Co hile Co hile lo lo Co Co mb m sta Ur bia sta ia u Ri ca R Co gua Ur ica sta y ug ua R Pe Do y Ec ica Br ua ru m do in Ec azil r Primary ica u Do Br n R ado m az Br ep r in az ub ica Bo il il l n R liv Male Formal El Bo ic ep ia Sa li ub lva Pa via Pa lic do n na r Ho am m Ar n a a g Secondary en El dur Ho Peru t Sa as Out of poverty Out of poverty Out of poverty in lva nd a do El ur r Sa as Female M lv Informal ex Ar Peru Ar ado ico Ve ge Ve ge r ne nti ne nti Pa zu na zu na ra ela ela gu ,R ,R University ay M B M B e e Ni Pa xico Pa xico ca ra ra ra Ni gua Ni gua gu a ca y ca y Gu Gu ragu Gu ragu at at a at a em em em ala ala ala a. Gender Percentage of people moving Percentage of people moving Percentage of people moving b. Education c. Sector of work Co Ur Ur 0 20 40 60 80 100 0 10 20 30 40 50 60 70 80 0 20 40 60 80 100 sta ug ug Ri ua ua ca y y Co Chi Co Chil sta le sta e Ch Ri ile c Co Ric Br a lo a m Co azil bi Br lo a az Br il Ar mbi a ge a Pa zil Primary El nt na Sa lva Do Pa ina Do Ec ma do m na ua r m in Arg dor Male Formal in Ec ma ica u ica en n R ad n R tin Pe ep or ep a ru El ub Sa lic Ho ubli Ar lv n c g Secondary FIGURE 4.10 Upward mobility conditional on initial characteristics in Latin America, by country en Ho ado El dur tin Sa as Into middle class nd r Into middle class a lva  Into middle class  ur do Ve B as Female M Bo r Informal ex ne oliv liv ico zu ia Ve ia ela Gu ,R ne Pe at B zu ru ela em Pe ,R University ala M ru M B e e Ni Pa xico Pa xico ca ra ra ra MOBILIT Y WITHIN GENERATIONS gu Ni gua Ni gua a ca y ca y Pa Gu ragu Gu ragu ra at a at a gu em em ay ala ala 111 (Figure continues next page) 112 MOBILIT Y WITHIN GENERATIONS FIGURE 4.10 Upward mobility conditional on initial characteristics in Latin America, by country (Continued) d. Area of residence Out of poverty Into middle class 80 80 Percentage of people moving Percentage of people moving 60 60 40 40 20 20 0 0 a as B a ru ala a a ca ca ile as B Gu ico zu o ala ile ru r zu y Ni ay r lic lic Ho il il gu do m gu m do ,R ,R c a az az ur Pe ur Ri Ri Pe Ch i Ch ub ub Ve agu gu em em ex ex na na ela lva ra ela ra lva Br Br nd nd sta sta ep ep ra M M ca Pa Pa ca at at r Sa Sa Ho Pa Pa Co Co nR nR Ni Gu ne ne El El ica ica Ve in in m m Do Do Urban Rural Urban Rural e. Migration from rural to urban areas Out of poverty Into middle class 80 80 70 70 Percentage of people moving Percentage of people moving 60 60 50 50 40 40 30 30 20 20 10 10 0 0 ala ala ca B B as ca lic lic ay ay as il il ,R ,R az az Ri ur Ri ur ub ub gu gu em em ela ela Br Br nd nd sta sta ep ep ra ra at at zu zu Ho Pa Pa Ho Co Co nR nR Gu Gu ne ne ica ica Ve Ve in in m m Do Do Urban migrant Rural Urban migrant Rural Source: Data from SEDLAC. Note: The figure shows lower-bound mobility estimates using the Dang et al. (2011) technique. Left panels show the proportion of those originally poor who escaped poverty, while right panels show the proportion of those originally poor or vulnerable who entered the middle class. Mobility estimates are conditional on these initial characteristics of the household head and household: (a) gender (male versus female); (b) education (primary, secondary, or university); (c) sector of work (formal [that is, contributing to a pension] versus informal [that is, not contributing to a pension]); (d) area of residence (urban versus rural area); and (e) migration from rural to urban areas (recent migrants who are currently living in urban areas versus individuals living in rural areas). “Poor” = individuals with a per capita income lower than US$4. “Vulnerable” = individuals with a per capita income of US$4– US$10. “Middle class” = individuals with a per capita income higher than US$10. Poverty lines and incomes are expressed in 2005 US$ PPP per day. PPP = purchasing power parity. SEDLAC = Socio-Economic Database for Latin America and the Caribbean. with higher mobility. Although this is only an for different periods and different year spans association—and not unexpected—it is nev- if cross-sectional data are available. Using the ertheless an important finding. SEDLAC data and the fact that all countries Additional analysis of these results sug- in Latin America have had periodic cross- gests that much of the correlation above is sectional surveys since the 1990s, we cre- particularly linked to the roaring 2000s. Spe- ated synthetic panels covering five-year spans cifically, an advantage of the synthetic panel in the 1990s and the 2000s. Although this approach is that one can estimate mobility analysis deviates from the chapter’s overall MOBILIT Y WITHIN GENERATIONS 113 FIGURE 4.11 GDP growth as a key correlate to upward mobility in Latin America a. Out of poverty b. Into middle class 5 5 Peru Ecuador Chile Annualized mobility, percent Colombia Annualized mobility, percent 4 Dominican 4 Mexico Republic Bolivia Nicaragua Costa Rica Brazil Panama 3 Honduras Uruguay 3 Argentina El Salvador Paraguay Uruguay Chile Venezuela, RB Guatemala Colombia Panama 2 2 Ecuador Costa Rica Brazil Argentina Peru Mexico Dominican Venezuela, RB Honduras Republic Bolivia El Salvador 1 1 Paraguay Nicaragua Guatemala 0 1 2 3 4 0 1 2 3 4 Annualized percentage change in GDP per capita Annualized percentage change in GDP per capita Source: Data from SEDLAC and the World Bank’s World Development Indicators (WDI). Note: The figure shows lower-bound mobility estimates using the Dang et al. (2011) technique. Panel a shows the correlation between annualized GDP growth and the annualized proportion of those originally poor who escaped poverty. Panel b shows the correlation between annualized GDP growth and the annualized proportion of those originally poor or vulnerable who entered the middle class. “Poor” = individuals with a per capita income lower than US$4. “Vulnerable” = individuals with a per capita income of US$4–US$10. “Middle class” = individuals with a per capita income higher than US$10. Poverty lines and incomes are expressed in 2005 US$ PPP per day. GDP = gross domestic product. PPP = purchasing power parity. SEDLAC = Socio-Economic Database for Latin America and the Caribbean. Dashed lines show the ordinary least square estimation. focus on long-term mobility, it allows us to FIGURE 4.12 Mobility by decade in Latin America, 1990s versus compare mobility trends between the two 2000s decades. percentage of population changing economic status Two results stand out. First, for the large 50 majority of countries in Latin America, the 2000s was a decade with higher levels of mobility than the 1990s. For example, 30 40 percent of the people in Uruguay changed Mobility in 1990s economic status in the 2000s, compared with 30 Venezuela, RB only 10 percent in the 1990s (see figure 4.12). Dominican Republic On the other end, countries like Nicaragua Nicaragua Paraguay Peru 20 Mexico Honduras experienced similar mobility in each decade Bolivia Colombia (about 20 percent in each). El Salvador Ecuador Uruguay Second, even among those countries where 10 Guatemala Costa Chile Panama Argentina Rica overall mobility is similar between the two Brazil decades, further distinguishing between downward and upper mobility suggests that 0 10 20 30 40 50 the two decades were indeed different. As Mobility in 2000s Figure 4.13 shows, the 2000s was a period Source: Data from SEDLAC. of dramatic upward mobility (panels a and Note: Years vary across countries. The table shows lower-bound mobility estimates using the Dang c) and very little downward mobility (panels et al. (2011) technique. Within each decade, periods span about five years. Economic status refers to poor, vulnerable, and middle class. “Poor” = individuals with a per capita income lower than US$4. b and d). By contrast, the 1990s exhibited “Vulnerable” = individuals with a per capita income of US$4–US$10. “Middle Class” = individuals much lower levels of upward mobility while with a per capita income higher than US$10. Poverty lines and incomes are expressed in 2005 US$ PPP per day. PPP = purchasing power parity. SEDLAC = Socio-Economic Database for Latin America higher downward mobility. and the Caribbean. The solid line represents the 45-degree line. 114 MOBILIT Y WITHIN GENERATIONS FIGURE 4.13 Mobility over time in Latin America, 1990s versus 2000s percentage of population changing economic status a. Upward out of poverty b. Downward into poverty 70 30 Mobility out of poverty in 1990s  Mobility into poverty in 1990s  60 50 Colombia 20 Bolivia 40 Venezuela, RB El Salvador 30 Dominican Republic Ecuador Mexico Peru Paraguay Nicaragua Honduras Uruguay 10 Guatemala 20 Costa Rica Chile Dominican Republic Paraguay Panama Honduras Nicaragua 10 Guatemala Ecuador Panama Argentina Argentina Bolivia Mexico El Salvador Brazil Colombia Peru Brazil 0 10 20 30 40 50 60 70 0 10 20 30 Mobility out of poverty in 2000s Mobility into poverty in 2000s c. Upward into middle class d. Downward out of middle class Mobility out of middle class in 1990s  Mobility into middle class in 1990s  30 30 Colombia Paraguay 20 20 Bolivia Uruguay El Salvador Venezuela, RB Dominican Republic Ecuador Dominican Republic 10 Mexico 10 Peru Honduras Nicaragua Honduras Costa Rica Chile Panama Mexico Nicaragua Paraguay Argentina Argentina Guatemala El Salvador Panama Brazil Guatemala Bolivia Ecuador Brazil Colombia Peru 0 10 20 30 0 10 20 30 Mobility into middle class in 2000s Mobility out of middle class in 2000s Source: Data from SEDLAC. Note: Panels a and c show upward mobility out of poverty and into the middle class, respectively. Panels b and d show downward mobility into poverty and out of the middle class, respectively. Years vary across countries. The table shows lower-bound mobility estimates using the Dang et al. (2011) technique. Within each decade, periods span about five years. Economic status refers to poor, vulnerable, and middle class. “Poor” = individuals with a per capita income lower than US$4. “Vulnerable” = individuals with a per capita income of US$4–US$10. “Middle class” = individuals with a per capita income higher than US$10. Poverty lines and incomes are expressed in 2005 US$ PPP per day. PPP = purchasing power parity. SEDLAC = Socio-Economic Database for Latin America and the Caribbean. The solid line represents the 45-degree line. Beyond growth: Mobility, policies, in policies and other relevant characteristics and labor markets beyond the importance of growth. For example, when we look at mobility Is long-term mobility related only to growth? and changes in inequality (measured by the Additional exercises indicate that policies income Gini coefficient), we find an interest- have an ample role in shaping long-term ing trend: Mobility out of poverty is strongly mobility. To explore this, we look at the cor- and negatively associated with increases relation between the conditional mobility in inequality, suggesting that mobility was out of poverty and into the middle class with higher among those countries that managed changes (over the same period) of key indica- to reduce inequality (see figure 4.14). How- tors and policies for each country. The analy- ever, a rising inequality is weakly positively sis nets out the role of GDP growth—the correlated with upward mobility into the idea being to explore, to the extent possible, middle class. Despite the caveats of these cor- whether changes in these conditional mobil- relations, they suggest a potential trade-off of ity probabilities are associated with changes policies that reduce inequality because they MOBILIT Y WITHIN GENERATIONS 115 FIGURE 4.14 Upward mobility and inequality in Latin America: A trade-off? a. Out of poverty b. Into middle class 1.5 1.5 Annualized mobility, percent Annualized mobility, percent Ecuador 1.0 Colombia 1.0 Mexico Peru Uruguay Chile Colombia 0.5 Chile 0.5 Ecuador Mexico Brazil Costa Rica Costa Panama Argentina 0 Brazil Honduras 0 Venezuela, RB Rica Nicaragua Uruguay Venezuela, RB Peru Honduras Paraguay Dominican Panama Dominican Paraguay Republic –0.5 Argentina Republic –0.5 El Salvador Guatemala Guatemala Nicaragua El Salvador –1.0 –1.0 –1.5 –1.0 –0.5 0 0.5 1.0 –1.5 –1.0 –0.5 0 0.5 1.0 Annualized percentage change in Gini coefficient Annualized percentage change in Gini coefficient Source: Data from SEDLAC and the World Bank’s WDI. Note: The panels show the correlation between annualized changes in inequality (measured using the Gini coefficient), estimated from cross-sectional surveys matching the specific periods for each country used for the synthetic panels and (panel a) the annualized proportion of those originally poor who escaped poverty and (panel b) the annualized proportion of those originally poor or vulnerable who entered middle class. The figure shows lower-bound mobility estimates using the Dang et al. (2011) technique. All figures show regressions controlling for annualized GDP growth. “Poor” = individuals with a per capita income lower than US$4. “Vulnerable” = individuals with a per capita income of US$4–US$10. “Middle class” = individuals with a per capita income higher than US$10. Poverty lines and incomes are expressed in 2005 US$ PPP per day. GDP = gross domestic product. PPP = purchasing power parity. SEDLAC = Socio-Economic Database for Latin America and the Caribbean. Dashed lines show the ordinary least square estimation. FIGURE 4.15 Educational expenditures and upward mobility in Latin America a. Out of poverty b. Into middle class 1.5 1.5 Annualized mobility, percent Annualized mobility, percent 1.0 1.0 Mexico Uruguay 0.5 0.5 Mexico Argentina Costa Rica Bolivia 0 0 Panama Costa Rica Bolivia Dominican Republic Uruguay Panama Dominican Republic –0.5 Argentina –0.5 Guatemala El Salvador Guatemala –1.0 El Salvador –1.0 –10.0 –5.0 0 5.0 –10.0 –5.0 0 5.0 Annualized percentage change in educational expenditure Annualized percentage change in educational expenditure Source: Data from SEDLAC and the World Bank’s WDI. Note: The figure shows lower-bound mobility estimates using the Dang et al. (2011) technique. Panel a shows the correlation between annualized changes in public expenditures in education with the annualized proportion of those originally poor who escaped poverty. Panel b shows the annualized proportion of those originally poor or vulnerable who entered middle class. All figures show regressions controlling for annualized GDP growth. “Poor” = individuals with a per capita income lower than US$4. “Vulnerable” = individuals with a per capita income of US$4–US$10. “Middle class” = individuals with a per capita income higher than US$10. Poverty lines and incomes are expressed in 2005 US$ PPP per day. GDP = gross domestic product. PPP = purchasing power parity. SEDLAC = Socio-Economic Database for Latin America and the Caribbean. Dashed lines show the ordinary least square estimation. could be affecting differentially distinct parts of government spending for different types of the distribution. of expenditures. Again, these results con- In terms of policies, we explore whether trol for GDP growth and, as such, they sug- different policies are correlated with higher gest the role of the changes in these policies levels of upward mobility. We proxy this by beyond the GDP growth in these countries. using the annualized changes in GDP shares For example, as figure 4.15 shows, countries 116 MOBILIT Y WITHIN GENERATIONS FIGURE 4.16 Overall and targeted social protection expenditures and upward mobility in Latin America a. Overall a.1. Out of poverty a.2. Into middle class 1.5 1.5 Annualized mobility, percent Annualized mobility, percent 1.0 1.0 Mexico Uruguay 0.5 0.5 Mexico Costa Rica Argentina 0 Bolivia 0 Panama Costa Rica Uruguay Bolivia Dominican Republic Panama Dominican Republic –0.5 Argentina –0.5 Guatemala El Salvador Guatemala El Salvador –1.0 –1.0 –40 –20 0 20 40 60 –40 –20 0 20 40 60 Annualized percentage change in social protection expenditure Annualized percentage change in social protection expenditure b. Targeted b.1. Out of poverty b.2. Into middle class 1.5 1.5 Annualized mobility, percent Annualized mobility, percent Ecuador 1.0 Colombia 1.0 Peru Mexico Colombia Chile 0.5 Chile 0.5 Costa Rica Brazil Ecuador Argentina Mexico Honduras Brazil Panama 0 0 Venezuela, RB Nicaragua Costa Rica Honduras Peru Dominican Republic Paraguay Dominican Republic Paraguay Venezuela, RB Panama –0.5 Argentina –0.5 El Salvador Nicaragua El Salvador –1.0 –1.0 –4 –2 0 2 4 6 –4 –2 0 2 4 6 Annualized percentage of poor receiving CCT Annualized percentage of poor receiving CCT Source: Data from SEDLAC, the World Bank’s WDI, and the International Food Policy Research Institute’s (IFPRI) Statistics of Public Expenditure for Economic Development (SPEED) data sets. Note: Panel a shows the correlation between the annualized social protection expenditure and the annualized upward mobility. Panel b shows the correlation between the annual- ized proportion of poor receiving conditional cash transfers and the annualized upward mobility. The figure shows lower-bound mobility estimates using the Dang et al. (2011) tech- nique. All figures show regressions controlling for annualized GDP growth. “Poor” = individuals with a per capita income lower than US$4. “Vulnerable” = individuals with a per capita income of US$4–US$10. “Middle class” = individuals with a per capita income higher than US$10. Poverty lines and incomes are expressed in 2005 US$ PPP per day. CCT = conditional cash transfers. GDP = gross domestic product. PPP = purchasing power parity. SEDLAC = Socio-Economic Database for Latin America and the Caribbean . Dashed lines show the ordinary least square estimation. that increased spending in education over the mobility (see figure 4.16, panel a). This is period exhibit higher levels of upward mobil- consistent with the fact that many social pro- ity for both poverty exits and entries into tection systems in the region, although they the middle class, confirming in a sense the play a key role in supporting beneficiaries, important role of education investments. are regressive in the sense that most of those The insights from the same analysis on receiving benefits (pensions, unemployment social protection spending are particularly schemes, and so on) are in the formal sector. telling. When we look at changes in over- This makes it less likely for traditional social all spending adding all the components of protection to reach the poor or the vulnerable social protection (pensions, unemployment, classes (who tend to work in the informal and safety nets like conditional cash trans- sector). As such, although such schemes can fers), there is little correlation with upward be critical in reducing downward mobility, MOBILIT Y WITHIN GENERATIONS 117 FIGURE 4.17 Female labor force participation and upward mobility in Latin America a. Out of poverty b. Into middle class 1.5 1.5 Annualized mobility, percent Annualized mobility, percent Ecuador 1.0 Colombia 1.0 Mexico Peru Uruguay Colombia Chile 0.5 Chile 0.5 Ecuador Mexico Costa Rica Brazil Argentina Brazil Panama 0 Bolivia Costa Rica 0 Bolivia Nicaragua Honduras Peru Venezuela, RB Dominican Republic UruguayPanama Dominican Republic Honduras –0.5 Paraguay Argentina –0.5 Paraguay Guatemala Venezuela, RB El Salvador Guatemala Nicaragua El Salvador –1.0 –1.0 –1.5 –1.0 –0.5 0 0.5 1.0 –1.5 –1.0 –0.5 0 0.5 1.0 Annualized percentage change in female labor force participation Annualized percentage change in female labor force participation Source: Data from SEDLAC. N ote: The figure shows lower-bound mobility estimates using the Dang et al. (2011) technique. Panel a shows the correlation between annualized change in female labor force par- ticipation and the annualized proportion of those originally poor who escaped poverty. Panel b shows the correlation between annualized change in female labor force participation and the annualized proportion of those originally poor or vulnerable who entered middle class. All figures show regressions controlling for annualized GDP growth. “Poor” = individuals with a per capita income lower than US$4. “Vulnerable” = individuals with a per capita income of US$4–US$10. “Middle class” = individuals with a per capita income higher than US$10. Poverty lines and incomes are expressed in 2005 US$ PPP per day. GDP = gross domestic product. PPP = purchasing power parity. SEDLAC = Socio-Economic Database for Latin America and the Caribbean. Dashed lines show the ordinary least square estimation. they do not seem to be conducive to upward with respect to moving out of poverty, there mobility. is no evidence of any correlation with women Interestingly, when the analysis focuses on entering the labor market (see figure 4.17). targeted interventions (see figure 4.16, panel By contrast, there seems to be (at least, less b)—and, specifically, conditional cash trans- weak) support that the additional increases of fers (captured by the annualized change in women in the labor force are positively asso- the size of the programs as a share of the poor ciated with middle-class entries. Finally, with over the period we study)—we find a strong respect to changes in country-level informal- reversal of the results above: countries that ity rates, the results do not show any corre- increased their program coverage over the lation with mobility, suggesting that faster period are significantly more likely to have formal sector growth in some countries is not improved the probability of upward mobility, necessarily associated with higher long-term both out of poverty and into the middle class. mobility (see figure 4.18). Although, again, this is not a causal attribu- tion, it suggests the potential role of targeted interventions in promoting upward mobility. Concluding remarks We also explore the role of labor markets This chapter explored directional intragener- in promoting mobility. Specifically, we focus ational mobility. Because we are interested in on the role of female labor force participation long-term movements—and to overcome the and informality. In the case of female labor problem of the lack of long-term panel data force participation, the past two decades have in the region—we construct synthetic pan- seen a significant entry of women in the labor els that rely on two or more cross-sectional force in the region. A recent study suggests surveys. This allows the analysis of long-term that more than 70 million women entered the dynamics and the calculation of mobility labor force since the 1980s (Chioda 2011). As estimates for 18 countries in the region cover- such, we explore whether this is associated ing the past two decades. The main results with mobility. The results are mixed. First, are as follows: 118 MOBILIT Y WITHIN GENERATIONS FIGURE 4.18 Informality and upward mobility in Latin America a. Out of poverty b. Into middle class 1.5 1.5 Annualized mobility, percent Annualized mobility, percent 1.0 Ecuador 1.0 Mexico Peru Uruguay 0.5 Chile 0.5 Chile Mexico Ecuador Brazil Costa Rica Brazil Argentina Panama Venezuela, RB 0 Bolivia 0 Costa Rica Nicaragua Honduras Venezuela, RB Peru Honduras Uruguay Bolivia Dominican Paraguay Panama Paraguay Republic –0.5 Argentina Dominican –0.5 Guatemala Republic El Salvador El Salvador Nicaragua Guatemala –1.0 –1.0 –1.0 –5.0 0 5.0 1.0 1.5 –1.0 –5.0 0 5.0 1.0 1.5 Annualized percentage change in informality Annualized percentage change in informality Source: Data from SEDLAC. Note: The figure shows lower-bound mobility estimates using the Dang et al. (2011) technique. Panel a shows the correlation between annualized change in informality and the annualized proportion of those originally poor who escaped poverty. Panel b shows the correlation between annualized change in informality and the annualized proportion of those originally poor or vulnerable who entered middle class. All figures show regressions controlling for annualized GDP growth. “Poor” = individuals with a per capita income lower than US$4. “Vulnerable” = individuals with a per capita income of US$4–US$10. “Middle class” = individuals with a per capita income higher than US$10. Poverty lines and incomes are expressed in 2005 US$ PPP per day. PPP = purchasing power parity. SEDLAC = Socio-Economic Database for Latin America and the Caribbean. Dashed lines show the ordinary least square estimation. • Latin America has experienced dramatic moved out of poverty), on average they mobility in the past two decades. Out do not enter the middle class but instead of every 100 Latin Americans, 43 have remain vulnerable to poverty. Only 5 per- changed their economic status during cent of those exiting poverty entered the the period. There is considerably more middle class. In fact, for every 10 people upward than downward mobility: out who entered the middle class, only one of the 43 people changing economic sta- was originally poor. By contrast, the tus, 23 exited poverty and 18 entered the number of those who entered the middle middle class, while only 2 experienced a class from the vulnerable was higher than worsening of their status. And despite the the number who remained vulnerable— large levels of mobility, more than 1 in a trend that shows considerable upward 5 Latin Americans remained chronically mobility to the middle class. poor throughout the whole period. These • Despite the low levels of long-term down- trends vary across countries. ward mobility using the more conserva- • T h is mobi l it y was especia l ly pro - tive estimate, alternative (less conserva- nounced at the bottom of the distribu- tive) estimates using the upper-bound tion. Although overall median incomes synthetic panels suggest that downward increased by US$3.3 PPP per day per mobility is of some concern, even in the capita (or almost 90 percent during this long run. Using these estimates, the analy- period) across the region, incomes among sis suggests that up to 13 percent of those the originally poor doubled (an increase originally not poor fell into poverty. This of US$1.8), compared with an 81 percent supports the idea that exploring policy (US$4.9) increase among the originally options to reduce long-term downward vulnerable and a 64 percent (US$11.3) mobility for the vulnerable (but not poor) increase for those originally in the middle may be an important direction for further class. work. • Although the poor are moving up (half • Various key correlates with mobility of those who were originally poor have emerge at the individual level. Education, MOBILIT Y WITHIN GENERATIONS 119 BOX 4.3 “Calling in” long-term mobility: Did cell phones improve mobility in rural Peru? Beuermann and Vakis (forthcoming) estimate percentage points (see figure B4.3.1). These benefits the effects of mobile phone expansion over the increased over time: villages that received mobile past 15 years on poverty. They exploit the timing coverage nine years earlier have extreme poverty of the arrival of mobile phone coverage at the vil- rates that are almost 15 percentage points lower lage level in rural Peru, which allows them to test than those villages that did not receive mobile cov- causally whether extreme poverty in villages that erage. Equally important, those benefits appear to received mobile coverage early on is lower. Their have been shared by all households in the villages, main fi ndings are striking: mobile phone expansion regardless of mobile ownership, suggesting strong increases household real consumption by 11 per- spillover effects and equalizing opportunities. cent and decreases extreme poverty by more than 5 FIGURE B4.3.1 The effect of mobile phone coverage on extreme poverty in rural Peru 0.2 0.1 percentage points 0 –0.1 –0.2 –5 –4 –3 –2 –1 0 1 2 3 4 5 6 7 8 9 Years since cellphone coverage Source: Beuermann and Vakis, forthcoming. Note: The figure shows double difference village-level estimates (and confidence intervals) of the additional extreme poverty reduction that can be attributed to the arrival of mobile phone coverage in a village as a function of the years since its arrival. for example, strongly predicts upward generally associated with slightly larger mobility, both out of poverty and into the probabilities of upward mobility into the middle class. This is especially true for the middle class. Finally, households that correlation between university degrees moved to urban areas were more likely to and entries into the middle class—sug- experience upward mobility than those gesting the high premium of education. who lived (stayed) in rural areas, high- With respect to labor market access, hav- lighting the role of economic opportuni- ing access to the formal sector provides a ties and geography. small advantage over those in the infor- • Sustained economic growth matters for mal sector in only a few of the countries, long-term mobility. For both poverty exit while for the rest there is not a difference. and middle-class entry, countries with By contrast, access to the formal sector is higher growth rates over the period are 120 MOBILIT Y WITHIN GENERATIONS strongly associated with higher mobil- interventions like conditional cash trans- ity. This is particularly telling when one fer programs are associated with upward compares mobility in the 1990s (a period mobility. of mixed growth results) with that of the • With respect to labor market outcomes, 2000s (a period of high sustained growth the results are mixed. For female labor for much of the region). Growth was force participation, there is no evidence indeed pro-poor! of any correlation between women enter- • Long-term mobility is not just about ing the labor market and mobility among growth but also relates to macroeconomic the poor. The analysis also shows no cor- stability and social policy. For example, relation between mobility and formal sec- countries that reduced inflation or income tor expansions, suggesting that countries inequality are more associated with where the formal sector grew faster are mobility out of poverty. Similarly, coun- not necessarily associated with higher tries that increased spending in education long-term mobility. By contrast, there over the period exhibited higher levels of seems to be (at least, less weak) support upward mobility for both poverty exits for the proposition that the additional and entries into the middle class. In social increases of women in the labor force are protection, although increases in overall positively associated with middle-class social protection spending are not associ- entries. ated with mobility, increases in targeted MOBILIT Y WITHIN GENERATIONS 121 Focus Note 4.1 Synthetic panels using repeated cross-sectional data This section summarizes the technique proposed (OLS) estimates of parameters E ˆ , where the super- 1 by Dang et al. (2011) to estimate intragenerational script refers to observations of households surveyed mobility by converting two or more rounds of cross- in the second round. Because we do not know the sectional data into a synthetic panel. A model of empirical correlation between the error term between income (or consumption) is estimated from cross- the two rounds, lower- and upper-bound estimates section data in year K, using a specification that of mobility are derived using two different sets of includes only time-invariant covariates.a Parameter assumptions about the correlation. estimates from this model are then applied to the Specifically, Lanjouw et al. (2011) argue that same time-invariant regressors in a cross-sectional the correlation between both error terms is likely survey from year L to predict an income estimate to be non-negative.b Then, if we assume zero cor- for households from year L in year K, thus creating relation between the first-round and second-round a “synthetic panel.” Analysis of mobility can then error terms, Lanjouw et al. (2011) propose to predict be done based on the households from year L, using income in the first round by randomly drawing with their actual income observed in year L along with replacement for each household i in the second round their predicted income from year K. from the empirical distribution of first-round esti- Formally, assume that we have two rounds of mated residuals (denoted by H˜ i21) as follows: cross-sectional surveys (denoted as round 1 and round 2). Calling yit round t household log per capita ˆ i2 y U ˆ c x2 + H ˜ i21. 1 = E1 i1 (F4.1b) consumption or income (where t =1, 2) of household i and z the poverty line, we are interested in estimat- Equation (F4.1b) allows us then to compute estimates ing (a) the fraction of poor households in the first of movements in and out of poverty. For example, the round of the survey that escaped poverty (Pr(y i 2 > fraction of poor households in the first round that z|yi1 < z)) or remained poor (Pr(yi2 < z|yi1 < z)) in the escaped poverty in the second round is given by second round of the survey; and (b) the fraction of nonpoor households in the first round of the survey Pr(yi2 ˆ i2 2 > z|y U 1 < z). (F4.1c) who became poor (Pr(y i 2 < z |y i1 > z)) or remained nonpoor (Pr(yi 2 > z |yi1 > z)) in the second round of Because we are randomly drawing from the empiri- the survey. This task cannot be performed directly cal distribution of estimated errors, we need to repeat by using repeated cross-sectional surveys because all the procedure R times and take average of equation households are interviewed only once, in either in the (F4.1c) to estimate movements in and out of poverty. first or second round of the survey. In all likelihood, however, the correlation between However, we can straightforwardly estimate the error terms will be positive. By assuming no corre- relationship between income and time-invariant char- lation, equation (F4.1c) will provide an upper-bound acteristics in each round: estimate of the mobility in and out of poverty. Dang et al. (2011) propose estimating also a lower bound yit = Etc xit + Hit t = 1,2 (F4.1a) on mobility by now assuming a perfect positive cor- relation between error terms. In this particular case, where xit is a vector of time-invariant characteristics estimates of residuals from the second round (ˆ Hi22) can (or characteristics that can be easily recalled from one be directly used to predict income in the first round round to the other one) of household i in round t of as follows: the survey and Hit is an error term. Using observations from the second round, we can predict consumption ˆ i2 y L ˆ c x2 + H ˆ i22. 1 = E1 i1 (F4.1d) in the first round (yˆ i21) by means of the same observed vector of time-invariant or retrospective character- Equation (F4.1d) allows us to compute lower-bound istics (xi21) and the first round ordinary least squares estimates of movements in and out of poverty. For a. The analysis presented in this chapter is based on the sample of households whose heads are between 25 and 65 years old. Results are then weighted using household-level survey sampling weights. b. Correlation between error terms will be non-zero in two cases: (a) the error term includes an individual fixed effect, and (b) shocks to consumption persist over time. Lanjouw et al. (2011) argue that correlation between error terms will almost certainly be positive if the condition (b) holds. In their study using Vietnamese and Indonesian data, they present empirical support in favor of this assumption. (Box continues next page) 122 MOBILIT Y WITHIN GENERATIONS Focus Note 4.1 Synthetic panels using repeated cross-sectional data (continued) example, the fraction of poor households in the first error and thereby provides a lower-bound estimate of round that escaped poverty in the second time is “true” mobility. It is for this reason that we report given by these estimates in the report: it allows a more conser- vative estimate of mobility trends. Pr(yi2 ˆ i2 2 > z|y L 1 < z). (F4.1e) Any new methodology would make little sense without validating it, especially in a context of inter- Because we are not drawing from the empirical est. Cruces et al. (2011) conduct a validation of this distribution of estimated errors, we do not need to approach by implementing a wide range of sensitiv- repeat the procedure R times as in the upper-bound ity analyses and robustness checks in three countries approach. In fact, this last approach provides a in Latin America where different lengths of panel clean underestimate of true mobility because we are data are available (Chile, Nicaragua, and Peru). The using household-specific error terms (from the sec- authors show that the methodology performs well in ond round in this example). In other words, because predicting actual mobility in and out of poverty by mobility is estimated across two survey rounds in means of two rounds of cross-sectional data; true which the same disturbance term applies to both mobility lies within the two bounds most of the time, consumption measures, the lower-bound measure of and the results are robust to additional tests. Box F4.1 mobility has been “purged” of classical measurement summarizes the paper’s key findings. BOX F4.1 Validating the approach for the case of Latin America A recent paper by Cruces et al. (2011) validates transitions into and out of poverty or the middle the synthetic panel approach in three different class in general and for a diverse set of subgroups settings in Latin America where panel data also are impressively similar (such as female-headed exist (Chile, Nicaragua, and Peru). This allows the households, households residing in urban areas, authors to compare true panel estimates of intra- and household-head education levels). The results generational mobility using the three panel data are also robust to alternative thresholds defi ni- sets, with mobility estimates based on the Dang et tions (see box figure F4.1). More important, the al. (2011) synthetic panel approach. In the process, technique does equally well in predicting short- they carry out a number of refi nements and test and long-term mobility patterns and is robust to how well the procedure does. a broad set of additional “stress” and sensitivity The results are encouraging: the methodol- tests. As such, the paper offers solid empirical ogy performs really well in predicting a range of validation to apply the approach to settings where mobility measures in all three settings, especially panel data are absent by expanding this work to in cases where richer model specifi cations can be the 18 countries in Latin America as we do here. estimated. For example, estimates for mobility MOBILIT Y WITHIN GENERATIONS 123 Focus Note 4.1 (continued) BOX FIGURE F4.1 Poverty dynamics: Synthetic versus actual panel data for alternative poverty lines in Peru, 2008 and 2009 a. Poor in 2008 and poor in 2009 b. Poor in 2008 and nonpoor in 2009 100 100 80 80 Poor, nonpoor Poor, poor 60 60 40 40 20 20 0 0 2 3 4 5 6 7 8 9 10 11 12 13 14 15 2 3 4 5 6 7 8 9 10 11 12 13 14 15 Poverty line, US$ 2005 PPP Poverty line, US$ 2005 PPP c. Nonpoor in 2008 and poor in 2009 d. Nonpoor in 2008 and nonpoor in 2009 100 100 80 80 Nonpoor, nonpoor Nonpoor, poor 60 60 40 40 20 20 0 0 2 3 4 5 6 7 8 9 10 11 12 13 14 15 2 3 4 5 6 7 8 9 10 11 12 13 14 15 Poverty line, US$ 2005 PPP Poverty line, US$ 2005 PPP Lower bound Upper bound Actual Source: Cruces et al. 2011. Note: Results are constrained to the panel sample of households whose heads are between 25 and 65 years old. Results are weighted using household-level survey-sampling weights. Upper-bound estimations are based on 50 repetitions. 124 MOBILIT Y WITHIN GENERATIONS Annex 4.1 Data used for intragenerational mobility estimates TABLE A4.1 Data sets used, years, and coverage, by country First Last Country Data set year year Coverage Argentina Encuesta Permanente de Hogares 1994 2009 Urban: 31 cities Bolivia Encuesta Continua de Hogares 1992 2007 National Brazil Pesquisa Nacional por Amostra de Domicilios 1990 2009 National Chile Encuesta de Caracterización Socioeconómica Nacional 1992 2009 National Colombia Gran Encuesta Integrada de Hogares 1992 2008 National Costa Rica Encuesta de Hogares de Propósitos Múltiples 1989 2009 National Dominican Republic Encuesta Nacional de Fuerza de Trabajo 1996 2009 National Ecuador Encuesta de Empleo, Desempleo y Subempleo 1995 2009 National El Salvador Encuesta de Hogares de Propósitos Múltiples 1991 2008 National Guatemala Encuesta Nacional de Condiciones de Vida 2000 2006 National Honduras Encuesta Permanente de Hogares de Propósitos Múltiples 1994 2009 National Mexico Encuesta Nacional de Ingresos y Gastos de los Hogares 2000 2008 National Nicaragua Encuesta Nacional de Medición de Vida 1998 2005 National Panama Encuesta de Hogares 1995 2009 National Paraguay Encuesta Permanente de Hogares 1999 2009 National Peru Encuesta Nacional de Hogares 1999 2009 National Uruguay Encuesta Continua de Hogares 1989 2009 National Venezuela, RB Encuesta de Hogares Por Muestreo 1992 2006 National MOBILIT Y WITHIN GENERATIONS 125 Annex 4.2 Regional and country intragenerational mobility estimates and decomposition using synthetic panels TABLE A4.2A Regional weighted intragenerational mobility decomposition median per capita income changes in levels (US$ PPP ) Destination Poor Vulnerable Middle class Total Poor 0.22 0.60 0.18 0.99 Origin Vulnerable −0.01 0.37 1.26 1.62 Middle class 0.00 −0.01 2.36 2.34 Total 0.21 0.95 3.79 4.96 Source: Data from SEDLAC. Note: Years vary across countries. Years used are: Argentina 1994 and 2009; Bolivia 1992 and 2007; Brazil 1990 and 2009; Chile 1992 and 2009; Colombia 1992 and 2008; Costa Rica 1989 and 2009; Dominican Republic 1996 and 2009; Ecuador 1995 and 2009; Guatemala 2000 and 2006; Honduras 1994 and 2009; Mexico 2000 and 2008; Nicaragua 1998 and 2005; Panama 1994 and 2009; Peru 1999 and 2009; Paraguay 1999 and 2009; El Salvador 1991 and 2008; Uru- guay 1989 and 2009; and República Bolivariana de Venezuela 1992 and 2006. The table shows lower-bound mobility estimates using the Dang et al. (2011) technique. Each cell shows median income changes in levels weighted using the proportion of the population in each cell. “Poor” = individuals with a per capita income lower than US$4. “Vulnerable” = individuals with a per capita income of US$4–US$10. “Middle class” = individuals with a per capita income higher than US$10. Poverty lines and incomes are expressed in 2005 US$ PPP per day. PPP = purchasing power parity. SEDLAC = Socio-Economic Database for Latin America and the Caribbean. TABLE A4.2B Regional weighted intragenerational mobility decomposition percentage median income growth Destination Poor Vulnerable Middle class Total Poor 19.48 23.13 5.84 48.45 Origin Vulnerable −0.13 7.42 19.36 26.64 Middle class –0.01 −0.08 13.22 13.13 Total 19.34 30.47 38.42 88.22 Source: Data from SEDLAC. Note: Years vary across countries. Years used are: Argentina 1994 and 2009; Bolivia 1992 and 2007; Brazil 1990 and 2009; Chile 1992 and 2009; Colombia 1992 and 2008; Costa Rica 1989 and 2009; Dominican Republic 1996 and 2009; Ecuador 1995 and 2009; Guatemala 2000 and 2006; Honduras 1994 and 2009; Mexico 2000 and 2008; Nicaragua 1998 and 2005; Panama 1994 and 2009; Peru 1999 and 2009; Paraguay 1999 and 2009; El Salvador 1991 and 2008; Uru- guay 1989 and 2009; and República Bolivariana de Venezuela 1992 and 2006. The table shows lower-bound mobility estimates using the Dang et al. (2011) technique. Each cell shows median income changes in levels weighted using the proportion of the population in each cell. “Poor” = individuals with a per capita income lower than US$4. “Vulnerable” = individuals with a per capita income of US$4–US$10. “Middle class” = individuals with a per capita income higher than US$10. Poverty lines and incomes are expressed in 2005 US$ PPP per day. PPP = purchasing power parity. SEDLAC = Socio-Economic Database for Latin America and the Caribbean. 126 MOBILIT Y WITHIN GENERATIONS TABLE A4.2C Country-specific intragenerational mobility in Latin America percentage of population Argentina Bolivia Brazil Destination Destination Destination P V MC Total P V MC Total P V MC Total P 10.6 7.6 0.1 18.3 34.0 28.1 3.1 65.2 20.8 27.8 4.5 53.1 Origin V 0.8 22.6 14.7 38.2 0.3 10.8 15.1 26.3 0.0 4.8 22.5 27.3 MC 0.0 1.1 42.4 43.5 0.0 0.1 8.4 8.5 0.0 0.0 19.6 19.6 Total 11.4 31.4 57.2 100.0 34.4 39.0 26.7 100.0 20.8 32.5 46.6 100.0 Costa Rica Chile Colombia Destination Destination Destination P V MC Total P V MC Total P V MC Total P 13.6 24.8 1.0 39.3 11.6 27.3 1.9 40.9 12.9 26.5 0.8 40.1 Origin V 0.0 11.3 30.4 41.7 0.0 7.9 31.4 39.3 0.0 11.9 26.5 38.4 MC 0.0 0.0 18.9 18.9 0.0 0.0 19.8 19.8 0.0 0.0 21.5 21.5 Total 13.6 36.0 50.4 100.0 11.6 35.3 53.1 100.0 12.9 38.4 48.7 100.0 Dominican Republic Ecuador Guatemala Destination Destination Destination P V MC Total P V MC Total P V MC Total P 17.8 14.8 1.2 33.8 20.5 30.2 2.4 53.1 50.2 8.1 0.0 58.3 Origin V 2.4 23.3 16.6 42.3 0.0 11.5 20.9 32.5 0.3 24.9 3.8 28.9 MC 0.1 0.9 22.8 23.8 0.0 0.0 14.4 14.4 0.0 0.0 12.7 12.8 Total 20.4 39.0 40.6 100.0 20.6 41.8 37.7 100.0 50.4 33.1 16.5 100.0 Honduras Mexico Nicaragua Destination Destination Destination P V MC Total P V MC Total P V MC Total P 37.3 28.4 4.0 69.7 24.9 11.1 0.2 36.2 54.3 15.5 0.4 70.1 Origin V 0.1 6.1 15.7 22.0 0.9 27.9 10.7 39.5 0.5 16.8 5.6 22.8 MC 0.0 0.1 8.3 8.4 0.0 1.2 23.1 24.3 0.0 0.1 7.0 7.1 Total 37.5 34.6 27.9 100.0 25.8 40.2 34.0 100.0 54.7 32.4 12.9 100.0 MOBILIT Y WITHIN GENERATIONS 127 TABLE A4.2C (continued) Panama Peru Paraguay Destination Destination Destination P V MC Total P V MC Total P V MC Total P 19.4 17.4 0.7 37.5 31.0 25.7 0.8 57.5 33.4 9.3 1.4 44.2 Origin V 0.0 14.4 19.5 33.9 0.0 14.4 15.1 29.5 4.5 22.7 5.4 32.6 MC 0.0 0.0 28.5 28.5 0.0 0.0 13.1 13.1 0.1 3.8 19.4 23.2 Total 19.4 31.9 48.7 100.0 31.0 40.1 28.9 100.0 38.0 35.8 26.2 100.0 El Salvador Uruguay Venezuela, RB Destination Destination Destination P V MC Total P V MC Total P V MC Total P 31.2 24.9 0.4 56.4 4.3 7.3 0.5 12.1 22.2 9.6 1.5 33.3 Origin V 0.0 13.6 17.6 31.1 0.1 13.8 23.9 37.8 10.1 24.1 12.7 46.9 MC 0.0 0.0 12.4 12.4 0.0 0.1 50.1 50.1 1.5 3.2 15.1 19.7 Total 31.2 38.4 30.4 100.0 4.5 21.1 74.4 100.0 33.8 37.0 29.2 100.0 Source: Data from SEDLAC. Notes: P = poor. V = vulnerable. MC = middle class. Years vary across countries. Years used are: Argentina 1994 and 2009; Bolivia 1992 and 2007; Brazil 1990 and 2009; Chile 1992 and 2009; Colombia 1992 and 2008; Costa Rica 1989 and 2009; Dominican Republic 1996 and 2009; Ecuador 1995 and 2009; Guate- mala 2000 and 2006; Honduras 1994 and 2009; Mexico 2000 and 2008; Nicaragua 1998 and 2005; Panama 1994 and 2009; Peru 1999 and 2009; Paraguay 1999 and 2009; El Salvador 1991 and 2008; Uruguay 1989 and 2009; and República Bolivariana de Venezuela 1992 and 2006. The table shows lower-bound mobility estimates using the Dang et al. (2011) technique. “Poor” = individuals with a per capita income lower than US$4. “Vulnerable” = individuals with a per capita income of US$4–US$10. “Middle class” = individuals with a per capita income higher than US$10. Poverty lines and incomes are expressed in 2005 US$ PPP per day. PPP = purchasing power parity. SEDLAC = Socio-Economic Database for Latin America and the Caribbean. 128 MOBILIT Y WITHIN GENERATIONS TABLE A4.2D Country-specific intragenerational mobility decomposition in Latin America, by country median per capita income changes in levels (US$ PPP) Argentina Bolivia Brazil Destination Destination Destination P V MC Total P V MC Total P V MC Total P 0.03 0.12 0.01 0.16 0.33 0.75 0.30 1.38 0.31 1.04 0.38 1.72 Origin V −0.01 0.26 0.75 0.99 −0.01 0.34 1.03 1.36 0.00 0.17 1.99 2.16 MC 0.00 −0.03 2.91 2.88 0.00 0.00 1.22 1.22 0.00 0.00 3.26 3.26 Total 0.02 0.35 3.66 4.03 0.32 1.09 2.55 3.96 0.31 1.21 5.62 7.14 Costa Rica Chile Colombia Destination Destination Destination P V MC Total P V MC Total P V MC Total P 0.15 0.73 0.09 0.97 0.15 0.89 0.15 1.19 0.16 0.77 0.06 0.99 Origin V 0.00 0.40 2.18 2.59 0.00 0.29 2.78 3.07 0.00 0.44 1.91 2.36 MC 0.00 0.00 3.15 3.15 0.00 0.00 4.94 4.94 0.00 0.00 3.58 3.58 Total 0.15 1.13 5.42 6.71 0.15 1.18 7.88 9.20 0.16 1.21 5.55 6.92 Dominican Republic Ecuador Guatemala Destination Destination Destination P V MC Total P V MC Total P V MC Total P 0.11 0.34 0.11 0.57 0.24 0.87 0.20 1.31 0.06 0.08 0.00 0.14 Origin V −0.05 0.40 0.93 1.29 0.00 0.37 1.47 1.85 0.00 0.24 0.11 0.35 MC −0.01 −0.04 1.91 1.86 0.00 0.00 2.17 2.17 0.00 0.00 0.58 0.58 Total 0.06 0.70 2.96 3.72 0.24 1.24 3.85 5.33 0.06 0.32 0.69 1.07 Honduras Mexico Nicaragua Destination Destination Destination P V MC Total P V MC Total P V MC Total P 0.38 0.96 0.36 1.70 0.11 0.21 0.01 0.32 0.20 0.34 0.03 0.58 Origin V 0.00 0.17 1.29 1.46 −0.01 0.31 0.48 0.78 0.00 0.14 0.32 0.45 MC 0.00 0.00 1.25 1.25 0.00 −0.02 0.83 0.81 0.00 −0.01 0.18 0.17 Total 0.38 1.13 2.91 4.41 0.10 0.50 1.32 1.92 0.20 0.48 0.53 1.20 MOBILIT Y WITHIN GENERATIONS 129 TABLE A4.2D (continued) Panama Peru Paraguay Destination Destination Destination P V MC Total P V MC Total P V MC Total P 0.17 0.43 0.06 0.67 0.33 0.69 0.07 1.09 0.06 0.23 0.19 0.48 Origin V 0.00 0.39 1.06 1.45 0.00 0.39 0.87 1.26 −0.07 −0.03 0.33 0.24 MC 0.00 0.00 2.64 2.64 0.00 0.00 1.15 1.15 −0.01 −0.11 0.12 0.01 Total 0.17 0.82 3.77 4.77 0.33 1.08 2.09 3.49 −0.02 0.09 0.65 0.72 El Salvador Uruguay Venezuela, RB Destination Destination Destination P V MC Total P V MC Total P V MC Total P 0.28 0.69 0.03 1.00 0.03 0.19 0.04 0.26 0.02 0.22 0.20 0.44 Origin V 0.00 0.33 1.15 1.48 0.00 0.30 1.55 1.84 −0.36 0.14 0.87 0.64 MC 0.00 0.00 1.53 1.53 0.00 0.00 6.29 6.29 −0.18 −0.13 0.45 0.13 Total 0.28 1.02 2.71 4.01 0.03 0.49 7.87 8.39 −0.53 0.22 1.53 1.22 Source: Data from SEDLAC. Note: P = poor. V = vulnerable. MC = middle class. Years vary across countries. Years used are: Argentina 1994 and 2009; Bolivia 1992 and 2007; Brazil 1990 and 2009; Chile 1992 and 2009; Colombia 1992 and 2008; Costa Rica 1989 and 2009; Dominican Republic 1996 and 2009; Ecuador 1995 and 2009; Guate- mala 2000 and 2006; Honduras 1994 and 2009; Mexico 2000 and 2008; Nicaragua 1998 and 2005; Panama 1994 and 2009; Peru 1999 and 2009; Paraguay 1999 and 2009; El Salvador 1991 and 2008; Uruguay 1989 and 2009; and República Bolivariana de Venezuela 1992 and 2006. The table shows lower-bound mobility estimates using the Dang et al. (2011) technique. Each cell show median income changes in levels. “Poor” = individuals with a per capita income lower than US$4. “Vulnerable” = individuals with a per capita income of US$4–US$10. “Middle class” = individuals with a per capita income higher than US$10. Poverty lines and incomes are expressed in 2005 US$ PPP per day. PPP = purchasing power parity. SEDLAC = Socio-Economic Database for Latin America and the Caribbean. 130 MOBILIT Y WITHIN GENERATIONS TABLE A4.2E Country-specific weighted intragenerational mobility decomposition in Latin America, by country median income growth in percentages Argentina Bolivia Brazil Destination Destination Destination P V MC Total P V MC Total P V MC Total P 1.38 3.88 0.22 5.48 22.53 27.07 9.31 58.91 32.63 45.42 11.67 89.72 Origin V −0.23 4.48 9.78 14.03 −0.12 6.66 15.85 22.40 0.00 3.62 32.63 36.25 MC 0.00 −0.25 17.22 16.98 0.00 −0.02 8.16 8.14 0.00 0.00 17.04 17.04 Total 3.88 0.22 0.00 36.49 22.42 33.70 33.33 89.45 32.63 49.05 61.34 143.02 Costa Rica Chile Colombia Destination Destination Destination P V MC Total P V MC Total P V MC Total P 11.43 25.29 2.82 39.55 10.09 32.63 4.47 47.19 10.38 27.26 1.58 39.23 Origin V 0.00 8.42 33.21 41.62 −0.01 6.13 45.30 51.42 0.00 9.27 29.37 38.64 MC 0.00 0.00 21.44 21.44 0.00 0.00 30.81 30.80 0.00 0.00 20.83 20.83 Total 11.43 33.71 57.47 102.61 10.08 38.76 80.57 129.41 10.38 36.53 51.78 98.69 Dominican Republic Ecuador Guatemala Destination Destination Destination P V MC Total P V MC Total P V MC Total P 6.67 10.70 3.54 20.91 16.94 31.56 5.97 54.47 3.77 2.39 0.00 6.15 Origin V −0.95 7.02 13.03 19.10 −0.01 7.88 22.17 30.04 −0.02 4.33 1.30 5.62 MC −0.08 −0.33 12.85 12.44 0.00 0.00 13.77 13.76 0.00 0.00 3.80 3.79 Total 5.64 17.39 29.42 52.45 16.93 39.44 41.90 98.26 3.75 6.71 5.10 15.57 Honduras Mexico Nicaragua Destination Destination Destination P V MC Total P V MC Total P V MC Total P 39.66 39.20 11.90 90.76 5.65 6.55 0.37 12.58 16.36 12.05 0.86 29.27 Origin V −0.05 3.17 21.59 24.71 −0.20 5.30 6.13 11.23 −0.07 2.72 4.76 7.42 MC 0.00 −0.01 7.65 7.64 0.00 −0.19 4.20 4.02 0.00 -0.05 1.04 1.00 Total 39.61 42.36 41.15 123.11 5.45 11.66 10.71 27.82 16.29 14.73 6.66 37.68 MOBILIT Y WITHIN GENERATIONS 131 TABLE A4.2E (continued) Panama Peru Paraguay Destination Destination Destination P V MC Total P V MC Total P V MC Total P 15.00 14.34 2.28 31.62 29.21 25.71 2.01 56.94 5.17 7.90 9.79 22.86 Origin V −0.01 7.43 14.54 21.97 0.00 7.51 12.65 20.15 −1.45 −0.46 4.53 2.63 MC 0.00 0.00 14.77 14.76 0.00 0.00 7.19 7.19 −0.05 −0.89 0.57 −0.37 Total 14.99 21.77 31.59 68.35 29.21 33.22 21.85 84.28 3.68 6.55 14.89 25.11 El Salvador Uruguay Venezuela, RB Destination Destination Destination P V MC Total P V MC Total P V MC Total P 25.33 26.26 0.91 52.50 1.47 6.37 1.03 8.87 0.84 7.19 7.05 15.08 Origin V 0.00 6.12 18.71 24.83 −0.01 5.50 21.84 27.32 −6.28 2.45 14.16 10.33 MC 0.00 0.00 9.75 9.74 0.00 -0.01 35.60 35.60 −1.43 −1.09 3.14 0.62 Total 25.33 32.38 29.36 87.07 1.46 11.86 58.47 71.79 −6.88 8.55 24.35 26.02 Source: Data from SEDLAC. Note: P = poor. V = vulnerable. MC = middle class. Years vary across countries. Years used are: Argentina 1994 and 2009; Bolivia 1992 and 2007; Brazil 1990 and 2009; Chile 1992 and 2009; Colombia 1992 and 2008; Costa Rica 1989 and 2009; Dominican Republic 1996 and 2009; Ecuador 1995 and 2009; Guate- mala 2000 and 2006; Honduras 1994 and 2009; Mexico 2000 and 2008; Nicaragua 1998 and 2005; Panama 1994 and 2009; Peru 1999 and 2009; Paraguay 1999 and 2009; El Salvador 1991 and 2008; Uruguay 1989 and 2009; and República Bolivariana de Venezuela 1992 and 2006. The table shows lower-bound mobility estimates using the Dang et al. (2011) technique. Each cell show median income growth in percentage weighted using the proportion of the population in each cell. “Poor” = individuals with a per capita income lower than US$4. “Vulnerable” = individuals with a per capita income of US$4–US$10. “Middle class” = individuals with a per capita income higher than US$10. Poverty lines and incomes are expressed in 2005 US$ PPP per day. PPP = purchas- ing power parity. SEDLAC = Socio-Economic Database for Latin America and the Caribbean. 132 MOBILIT Y WITHIN GENERATIONS Notes 12. As chapter 2 discusses, these mobility mea- sures can be decomposed linearly, which pro- 1. This section benefits from the excellent vides us with a number of additional insights review carried out in Fields et al. (2007). (see annex 4.2 for the full set of results by 2. Pseudo-panel methods construct panels country). of cohort averages, tracking these cohorts 13. The population that was originally poor with through multiple rounds of cross-section sur- a secondary degree is generally small (less vey data. than 10 percent, with some variations across 3. As Fields et al. (2007) conclude, “While countries). there is a vast array of results on mobility for 14. Neri (2010) argues that the recent increase of Latin American economies, the large meth- the middle class in Brazil is directly linked to odological disparities across studies limits the equally large increase of the formal sector their usefulness in contributing to a regional during the past decade. understanding.” 4. Recent developments on pseudo-panel anal- References ysis include Bourguignon, Goh, and Kim (2004) and Antman and McKenzie (2007). Antman, Francisca, and David McKenzie. 2007. 5. As is well known, general versions of such “Earnings Mobility and Measurement Error: models are difficult to solve, and most work A Pseudo-Panel Approach.” Economic Devel- in the literature has therefore been computa- opment and Cultural Change 56 (1): 125–62. tionally intensive (Hugget 1993; Krusell and Beccaria, Luis, and Fernando Groisman. 2006. Smith 1998). In contrast, this work relies “Inestabilidad, Movilidad y Distribución del upon an extended version of the incomplete- Ingreso en Argentina.” Revista de la CEPAL markets model recently developed and ana- 89 (August): 133–52. lyzed by Constantinides and Duffie (1996) Beneke de Sanfeliu, Margarita, and Mauricio Shi. and Krebs (2004) that is highly tractable 2003. Dinámica del Ingreso Rural en El Sal- but still rich enough to allow for tight links vador. San Salvador: Fundación Salvadoreña between the econometric framework and the para el Desarrollo Económico y Social. welfare-theoretic model. Beuermann, Diether, and Renos Vakis. Forthcom- 6. Specifically, the Hart index is used, which is ing. “Mobile Phones and Economic Develop- the complement of the correlation between ment in Rural Peru.” Journal of Development the logarithm of incomes over time (see Hart Studies. [1981]). Bourguignon, Francois, Chor-ching Goh and Dae 7. The results of Krebs, Krishna, and Maloney Il Kim. 2004. “Estimating Individual Vulner- (2011) are based on five rounds of panel data ability to Poverty with Pseudo-panel Data.” spanning two years. Policy Research Working Paper Series 3375, 8. See Hoogeveen, Emwanu, and Okwi (2003) World Bank, Washington, DC. for an early application of the Elbers, Lan- Calónico, Sebastian. 2006. “Pseudo-Panel Anal- jouw, and Lanjouw (2002, 2003) approach to ysis of Earnings Dynamics and Mobility in the construction of a “pseudo-panel” poverty Latin America.” Discussion paper, Inter-Amer- map. ican Development Bank, Washington, DC. 9. The validation exercises done by Cruces et al. Chioda, Laura. 2011. “Work and Family: Latin (2011) show that the synthetic panel approach American and Caribbean Women in Search of performs well in predicting both short- and a New Balance.” Regional study, World Bank, long-term intragenerational mobility Washington, DC. 10. For more information, see SEDLAC at http:// Constantinides, George, and Darrell Duffie. sedlac.econo.unlp.edu.ar/eng. 1996. “Asset Pricing with Heterogeneous Con- 11. To construct mobility measures for the whole sumers.” Journal of Political Economy 104 region, we are constrained to use the respec- (2): 219–40. tive periods of data from each country that Corbacho, Ana, Mercedes Garcia-Escribano, and are available. Because the start and end peri- Gabriela Inchauste. 2007. “Argentina: Mac- ods differ across countries, although we find roeconomic Crisis and Household Vulnerabil- these aggregate mobility results informative, ity.” Review of Development Economics 11 they should be interpreted accordingly. (1): 92–106. MOBILIT Y WITHIN GENERATIONS 133 Cruces, Guillermo, Pablo Glüzmann, and Luis F. Glewwe, Paul, and Gillette Hall. 1998. “Are Some López-Calva. 2011. “Economic Crises, Mater- Groups More Vulnerable to Macroeconomic nal and Infant Mortality, Low Birth Weight Shocks Than Others? Hypothesis Tests Based and Enrollment Rates: Evidence from Argen- on Panel Data from Peru.” Journal of Develop- tina’s Downturns.” Working Paper 121, Center ment Economics 56 (1): 181–206. for Distributive, Labor and Social Studies: Uni- Hart, Peter E. 1981. “The Statics and Dynamics versidad de La Plata, Argentina. of Income Distributions: A Survey.” In The Cruces, Guillermo, Peter Lanjouw, Leonardo Statics and Dynamics of Income, ed. N. A. Lucchetti, Elizaveta Perova, Renos Vakis, and Klevmarket, J. A. Lybeck, and C. Tieto, 108- Mariana Viollaz. 2011. “Intragenerational 25. Clevedon, U.K.: Tieto. Mobility and Repeated Cross-Sections: A Herrera, Javier. 1999. “Ajuste Económico, Three-Country Validation Exercise.” Policy Desigualdad, y Movilidad.” In Pobreza y Research Working Paper 5916, World Bank, Economía Social: Análisis de una Encuesta Washington, DC. ENNIV-1997, ed. Richard Webb and Moises Cruces, Guillermo, and Quentin Wodon. 2006. Ventocilla, 101–42. Lima: Instituto Cuanto, “Risk-Adjusted Poverty in Argentina: Mea- United Nations Children’s Fund and U.S. surement and Determinants.” Financiamiento Agency for International Development. del Desarrollo Paper 182, United Nations Eco- Hoogeveen, J., T. Emwanu, and P. Okwi. 2003. nomic Commission for Latin America and the “Updating Small Area Welfare Indicators in Caribbean, Santiago. the Absence of a New Census.” Unpublished Dang, Hai-Anh, Peter Lanjouw, Jill Luoto, and manuscript, World Bank, Washington, DC. David McKenzie. 2011. “Using Repeated Huggett, Mark. 1993. “The Risk-Free Rate in Cross-Sections to Explore Movements in and Heterogeneous-Agent Incomplete-Market out of Poverty.” Policy Resarch Working Paper Economies.” Journal of Economic Dynamics 550, World Bank, Washington, DC. and Control 17 (5–6): 953–69. Duval Hernández, Robert. 2006. “Dynamics of IFPRI (International Food Policy Research Insti- Labor Market Earnings and Sector of Employ- tute). Online database. Statistics of Public ment in Urban Mexico, 1987–2002.” PhD dis- Expenditure for Economic Development sertation, Cornell University, Ithaca, NY. (SPEED). IFPRI, Washington, DC. http:// Elbers, Chris, Jean O. Lanjouw, and Peter Lan- www.ifpri.org/book-39/ourwork/programs/ jouw. 2002. “Micro-Level Estimation of Wel- priorities-public-investment/speed-database. fare.” Policy Research Working Paper 2911, Krebs, Tom. 2004. “Testable Implications of Development Research Group and World Consumption-Based Asset Pricing Models Bank, Washington, DC. with Incomplete Markets.” Journal of Math- ———. 2003. “Micro-Level Estimation of Pov- ematical Economics 40 (1–2): 191–206. erty and Inequality.” Econometrica 71 (1): Krebs, Tom, Pravin Krishna, and William Malo- 355–64. ney. 2011. “Income Dynamics, Mobility and Fields, Gary S., Paul Cichello, Samuel Freije, Welfare in Developing Countries.” Discussion Marta Men é ndez, and David Newhouse. Paper, World Bank, Washington, DC. 2003. “For Richer or Poorer? Evidence from Krusell, Per, and Anthony A. Smith. 1998. Indonesia, South Africa, Spain, and Venezu- “Income and Wealth Heterogeneity in the ela.” Journal of Economic Inequality 1 (1): Macroeconomy.” Journal of Political Econ- 67–99. omy 106 (5): 867–96. Fields, Gary S., Robert Duval Hernandez, et al. McKenzie, David J. 2004. “Aggregate Shocks 2006. “Earnings Mobility in Argentina, Mex- and Urban Labor Market Responses: Evidence ico, and Venezuela: Testing the Divergence from Argentina’s Financial Crisis.” Economic of Earnings and the Symmetry of Mobility Development and Cultural Change, 52 (4): Hypothesis.” Cornell University, Ithaca, NY. 719–58. Fields, Gary S., Robert D. Hernández, Samuel Neri, Marcelo. 2010. The New Middle Class: The Freije, and Maria L. Sánchez. 2007. “Intra- Bright Side of the Poor. Rio de Janeiro: Funda- generational Income Mobility in Latin Amer- çäo Getulio Vargas Press. ica.” Journal of the Latin American and Carib- Ñopo, Hugo. 2011. “Using Pseudo-Panels to bean Economic Association 7 (2): 101–54. Measure Income Mobility in Latin America.” 134 MOBILIT Y WITHIN GENERATIONS Discussion Paper 5449, Institute for the Study Income among the Rural Poor in Chile, 1968– of Labor, Bonn. 1986.” Discussion Paper 53, London School of Paredes, Ricardo, and José Ramos Zubizarreta. Economics and Political Science, Suntory and 2005. “Focusing on the Extremely Poor: Toyota International Centers for Economics Income Dynamics and Policies in Chile.” and Related Disciplines, London. Working Paper 183, Departamento de Ingeni- SEDLAC (Socio-Economic Database for Latin ería Industrial y Sistemas, Pontifica Universi- America and the Caribbean). Center for Dis- dad Católica de Chile, Santiago. tributive, Labor and Social Studies (CEDLAS) Premand, Patrick, and Renos Vakis. 2010. “Do of Universidad de La Plata, Argentina, and Shocks Affect Poverty Persistence? Evidence World Bank, Washington, DC. http://sedlac Using Welfare Trajectories from Nicaragua.” .econo.unlp.edu.ar/eng. Well-Being and Social Policy 6 (1): 95–129. World Bank. Online database. World Develop- Scott, Christopher D., and Julie Litchfield. 1994. ment Indicators. Washington, DC: World “Inequality, Mobility, and the Determinants of Bank. http://data.worldbank.org/indicator/. 5 The Rising Latin American and Caribbean Middle Class I n the past two decades, most of Latin population. It shows that where economic America was characterized by a consider- growth was able to translate into higher able degree of upward income movement. household incomes, it was the principal Such dynamics helped move a large number source of the middle-class expansion, rein- of families into the middle class, although forced by a reduction in income inequality. many others stayed in a vulnerable condition. Nevertheless, despite impressive growth in What did this process mean for the size and the ranks of the middle class in most of the composition of different income groups or region’s countries (on average, by 10 per- classes in the region? centage points in less than a decade), Latin Chapter 4 has documented that the transi- America and the Caribbean remains for tion from poverty into the middle class was the most part a “vulnerable” society, with not automatic. There are characteristics asso- many households that escaped poverty fac- ciated with class transitions, such as educa- ing a nonnegligible risk of falling back into tion, job stability, and area of residence. And it. Social protection policies aimed at the as the poor move upward, in most cases, they poor are thus likely to remain crucial in the do not jump all the way into the middle class. medium term. In fact, given the nonnegligible Instead, they remain vulnerable to poverty, likelihood of the vulnerable to fall back into and it may take time for them to accumulate poverty, it may be worth exploring how best assets or reach a combination of characteris- to address the vulnerabilities of this class, tics that allows them to move into the middle which currently is likely to be excluded from class. Thus, despite the dramatic movements social assistance programs targeted to the out of poverty, the “new” middle classes may poor but, at the same time, may not be able not be that different from the “old” ones. to fully benefit from social insurance pro- The first part of this chapter documents grams designed for the middle class. the size and growth of the Latin American The second part of the chapter profiles and Caribbean middle class, which, after the region’s middle class. Although the syn- an impressive growth spurt in the early thetic panels discussed in chapter 4 allow 2000s, now represents a third of the region’s us to identify time-invariant characteristics 135 136 THE RISING LATIN AMERICAN AND CARIBBEAN MIDDLE CLASS associated with class transitions, the absence population. We also converted per capita of panel datasets still makes it difficult to household income from local currencies to profile “new” members of the middle class 2005 U.S. dollars in purchasing power parity as opposed to “old” ones. To cope with this (PPP) terms. The resulting income distribu- challenge, we conclude instead by reviewing tion includes 15 out of 41 countries (includ- how much the profile of the middle class has ing overseas territories) in Latin America and changed over the past 20 years. the Caribbean, covering 86 percent of the The definition of middle-class status used region’s population. in chapters 5 and 6 echoes the concept of Figure 5.1 shows that both the poverty economic security, which translates into the and middle-class lines (US$4 per capita or income thresholds discussed in chapter 2 (per less and US$10 per capita or more per day, capita income between US$10 and US$50 a respectively) intersect the region’s income day). In some analyses, because of data con- distribution close to its mode. This is part of straints, we shall, however, group the middle the reason why, as we shall document, we are and upper classes together. At the end of this observing both dramatic decreases of pov- chapter, focus note 5.1 discusses how class erty and increases of the middle class: any levels and trends change under alternative small shift in the mean of the income distri- definitions. It shows that middle-class defini- bution is accompanied by many people exit- tions ought to be context-specific and that, ing poverty and entering the middle class— for the purposes of this review, the absolute much more movement than occurs, say, at definition we have adopted appears to per- the upper middle-class threshold of US$50 form better than relative ones. The analysis dollars a day. Corroborating the evidence in this chapter is mostly based on harmo- discussed in chapter 2, the figure also shows nized survey data from the Socio-Economic that the middle class in Latin America and Database for Latin America and the Carib- the Caribbean remains relatively wealthy: bean (SEDLAC), a collaboration between the the middle class starts at the 68th percen- Universidad Nacional de La Plata’s Center for tile, way above the median, and what we Distributive, Labor and Social Studies (CED- define as the upper class (which, for a family LAS) in Argentina and the World Bank. of three, corresponds to a monthly house- hold income of approximately US$4,500) represents around 2 percent of the region’s The middle class in Latin America population. and the Caribbean In fact, about two-thirds of the region’s In 2009, for the first time in history, one out population remains concentrated in the poor of three individuals in Latin America and and vulnerable classes. This suggests that, the Caribbean was living with a per capita despite positive trends, the region is not yet income above US$10 a day, joining the ranks a “middle-class society” where most people of the middle class. This achievement not- earn a sufficiently high income to consume, withstanding, being middle class in Latin live, and behave like middle-class citizens. America is still, in relative terms, a privileged Although people leaving poverty status rep- status. resents a positive trend, vulnerability to poverty remains a serious concern for the majority, and social policies will continue to Income distribution play an important role in the lives of many Figure 5.1 shows the distribution of income households for the foreseeable future. The in 2009 for the region. To construct figure large proportion of people who escaped pov- 5.1, we merged available household surveys erty but did not join the ranks of the middle from Latin America and the Caribbean, class is so high, in fact, that it may be worth weighting each observation by a country’s exploring the extent to which the vulnerable THE RISING LATIN AMERICAN AND CARIBBEAN MIDDLE CLASS 137 FIGURE 5.1 Income distribution in Latin America GDP and other drivers of heterogeneity and the Caribbean, selected countries, 2009 Although the size of the middle class does relate to overall economic development, the .04 relationship is far from perfect: the correla- tion between the size of the middle class and gross domestic product (GDP) per capita in .03 PPP terms is in fact only equal to 0.65 for the region. Other factors, such as income inequality, are also important determinants Density .02 of the size of the middle class: for example, in 2009, the size of the middle class differed between Brazil and Peru by only 7 percentage .01 points despite GDP per capita being 20 per- cent higher in Brazil, partly because of higher income inequality there. 0 4a 10b 50c 100 Per capita daily income, US$ PPP Recent middle-class growth trends Source: Data from SEDLAC (Socio-Economic Database for Latin America and the Caribbean). The appearance of a strong middle class is, for many countries, a relatively new phenom- are adequately protected. Unfortunately such enon. Between 2003 and 2009, the Latin analyses, to reach a good level of accuracy, American middle class grew at an annualized would require the use of “true” household rate of 6.7 percent, from slightly above 100 panel data and fall beyond the scope of this million people to more than 150 million (see report. But future advances in the analysis figure 5.3). In 2008, for the first time, there and design of social protection programs were almost as many people in the middle will likely require dynamic studies of poverty class as in poverty (152 million and 158 mil- patterns. lion, respectively). Despite the global finan- cial crisis, the trend reverted only minimally in 2009. This dramatic increase in the middle Regional heterogeneity of income class contrasts strongly with the lagging per- distribution formance of the 1990s—a “lost decade” for The regional distribution of income in figure the middle class, during which its size fluctu- 5.1 hides strong heterogeneities within the ated at around 21 percent of the population region. Although in Uruguay, for instance, for most of the decade, barely keeping pace more than 50 percent of society is of middle- with population growth. class status, the proportion drops to around a third for countries such as Brazil and Heterogeneity of trends Panama, and to less than a fifth in El Salva- dor and Honduras, as figure 5.2 illustrates. As with the magnitudes (see figure 5.2), the Almost symmetrically, more than half of the overall class-related trends also hide heteroge- population still lives in poverty (per capita neities across countries. In Argentina, Chile, income of less than US$4 a day) in Hondu- and Peru, the middle class increased by more ras. And even in wealthier countries such as than 10 percentage points between 2000 and Colombia, Mexico, and Panama, the propor- 2010, while in the Dominican Republic, El tion of those in poverty is about a third of the Salvador, and Uruguay, it actually shrank (as population. shown in figure 5.4). Overall, however, most 138 THE RISING LATIN AMERICAN AND CARIBBEAN MIDDLE CLASS FIGURE 5.2 Class composition in Latin America by income percentile, selected countries, 2009 100 80 60 Percentile 40 20 0 r ia as or ay lic u ico a a il ca ile a y do ua az bi m in r liv ad ur Pe ub Ri gu Ch ex nt om na Br lva ug Bo nd u sta ep ra ge M Pa Ec Ur Sa l Pa Ho Co nR Co Ar El ica in m Do Upper class Middle class Vulnerable Poor Source: Data from SEDLAC. Note: Class composition in Bolivia is for 2008, and in Mexico is for 2010. “Poor” = individuals with a per capita daily income lower than US$4. “Vulnerable” = individuals with a per capita daily income of US$4–US$10. “Middle class” = individuals with a per capita daily income of US$10–US$50. “Upper class” = indi- viduals with a per capita daily income exceeding US$50. Poverty lines and incomes are expressed in 2005 US$ PPP per day. PPP = purchasing power parity. SEDLAC = Socio-Economic Database for Latin America and the Caribbean. countries experienced a large surge in their Republic, for instance, experienced a higher middle classes, so that the aggregate trend for growth rate than Ecuador between 2000 Latin America observed in figure 5.3 did not and 2010, but its middle class shrank, while hinge only on the massive increase of the Bra- Ecuador’s grew by more than 15 percentage zilian middle class, which alone contributed points. This difference clearly indicates that more than 40 percent of the overall increase several other factors influence the growth of in the region (see also box 5.1). the middle class. A purely mechanical factor that is often overlooked concerns differences in initial Influential factors in middle-class conditions (Bourguignon 2002). How much growth the middle class increases for each percent- Although important, as previously men- age point of growth depends on where the tioned, economic growth is not the only middle class threshold of US$10 per capita driver of the increase in middle class: figure per day crosses each country’s income dis- 5.4 shows that countries with similar growth tribution. By simple “mechanics,” in poorer rates at times differed significantly in terms countries such as Honduras, 1 percentage of middle-class growth. The Dominican point of growth will bring smaller growth THE RISING LATIN AMERICAN AND CARIBBEAN MIDDLE CLASS 139 of the middle class than it will in wealthier FIGURE 5.3 Middle class, vulnerability, and poverty trends in Latin countries such as Uruguay, where the middle America, 1995–2009 class threshold crosses the income distribu- tion nearer the mode (where population den- 50 sity is higher and, thus, more people change 45 class for the same growth performance). 40 Similarly, initial levels of income inequality Percentage of population also influence the extent to which the size 35 of the middle class responds to economic 30 growth. 25 In addition to initial conditions, changes 20 in the size of the middle class are influenced by redistributive policies. Using a methodol- 15 ogy based on Datt and Ravallion (1992), Aze- 10 vedo and Sanfelice (2012) have decomposed 5 changes in the shares of population in each 0 class between 1995 and 2010 into those that 1995 2000 2005 2010 can be attributed to (a) growth in average per Year capita income, or (b) changes in the shape Poor (US$0–US$4 a day) Vulnerable (US$4–US$10 a day) of the income distribution (that is, inequal- Middle class (US$10–US$50 a day) ity). They find that per capita income growth Source: Data from SEDLAC. and redistributive policies play different roles Note: Covered countries include Argentina, Bolivia, Brazil, Chile, Colombia, Costa Rica, the Domini- across classes. On average, across the sample can Republic, El Salvador, Ecuador, Guatemala, Honduras, Mexico, Nicaragua, Panama, Paraguay, Peru, Uruguay, and República Bolivariana de Venezuela. Poverty lines and incomes are expressed in of countries in figure 5.5, redistributive poli- 2005 US$ PPP per day. PPP= purchasing power parity. SEDLAC = Socio-Economic Database for Latin cies played a substantial role in decreasing America and the Caribbean. poverty: 34 percent of the decrease in poverty can be attributed to redistributive policies, FIGURE 5.4 Middle class versus economic growth in Latin against 66 percent attributable to growth in America, selected countries, 2000–10 average per capita income. The high contribution of falling inequal- ity to falling poverty corroborates the effec- 20 Growth of the middle class, percentage points tiveness of the dramatic expansion of social 15 Ecuador programs in most Latin American and Carib- Argentina Colombia Panama bean countries during the 2000s. Lustig, Chile Peru 10 López-Calva, and Ortiz-Juarez (2011), for Brazil Costa Rica instance, do an in-depth analysis of the 5 Mexico Paraguay causes underlying the decline in inequality Bolivia Honduras in Argentina, Brazil, Mexico, and Peru—a 0 representative sample of the region’s diversity El Salvador Dominican in terms of initial inequality and economic –5 Republic growth. They find that policy interventions in Uruguay the social sector played a key role. In Brazil, –10 the authors estimate, the Benefício de Presta- 0 5 10 15 20 25 30 35 40 45 50 ção Continuada and Bolsa Família programs GDP per capita growth, percentage PPP explain more than 20 percent of the decline in household income inequality. In Mexico, Source: Data from SEDLAC and WDI. Note: “Middle class” = individuals with a per capita daily income of US$10–US$50 in 2005 US$ PPP the Oportunidades social assistance program per day. GDP = gross domestic product. PPP = purchasing power parity. SEDLAC = Socio-Economic accounts for 18 percent of the change in the Database for Latin America and the Caribbean. 140 THE RISING LATIN AMERICAN AND CARIBBEAN MIDDLE CLASS BOX 5.1 The (sustainable?) rise of the Brazilian middle class The rise of the Brazilian middle class is receiving idiosyncratic risks that may draw them back into increased attention of both academics and policy poverty, would also be at risk under a worsening of makers. According to Neri (2010), the middle class macroeconomic conditions. now represents more than half of the population in The second cautionary note regards the sustain- Brazil. It rose thanks to strong economic perfor- ability of the current consumption boom. Although mance but also because the workforce became more middle-class households do have stronger purchas- educated, which commands higher wages. The rise ing power, in Brazil many households fi nance a dis- in households with strong purchasing power is spur- proportionate share of their consumption through ring a consumption boom, which, given the contin- credit, and the current consumption boom is as ued entrance of a more educated workforce in the much driven by the large demand of households labor market and the expansion of formal labor, is joining the middle class as by the rapid growth in expected to continue. consumer credit due to microeconomic reforms that Our analysis broadly confi rms these trends, albeit have facilitated credit-risk screening and the provi- with cautionary notes. Neri (2010) classifies house- sion and recovery of collateral. holds into five classes (from A, the richest, to E, the A question remains about the extent to which poorest). a The large size of the middle class (class the new middle classes have the financial literacy C) stems in part from the fact that, with lower- and needed to avoid getting themselves into excessive upper-income thresholds of approximately US$6.1 debt. Figure B5.1b shows trends in consumer and and US$26.2 a day per capita, class C comprises mortgage credit relative to gross domestic policy many of both our vulnerable and middle-class (GDP). It compares Brazil with the other six larg- households (see figure B5.1a). Under the defi nition of est Latin American economies (in aggregate) as well Neri (2010), many middle-class households therefore with an international benchmark that takes into remain close to poverty, and, in addition to facing account GDP and other factors that are exogenous FIGURE B5.1A The Brazilian middle class under alternative definitions, 1990–2009 60 50 Percentage of population 40 30 20 10 0 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 Year Middle class US$10–US$50 a day Class C circa US$6.10–US$26.20 Vulnerable US$4–US$10 Class D circa US$3.80–US$6.10 Source: Data from SEDLAC and the World Bank’s World Development Indicators (WDI); Class C and D definitions from Neri 2010. Note: Poverty lines and incomes are expressed in 2005 US$ PPP per day. PPP= purchasing power parity. SEDLAC = Socio-Economic Database for Latin America and the Caribbean. THE RISING LATIN AMERICAN AND CARIBBEAN MIDDLE CLASS 141 BOX 5.1 The (sustainable?) rise of the Brazilian middle class (continued) to economic performance, such as demography and generating fi nance that leads to asset accumulation. country size. The figure shows that Brazil is an out- Brazilian middle-class households may thus be over- lier: countries of similar characteristics tend to have indebted and investing too little in asset accumula- half the consumer credit of Brazil and twice the tion, which may pose relatively few risks under the mortgage credit (Didier and Schmukler 2011; De current high-growth scenario but could be a source la Torre, Ize, and Schmukler 2012). Although Bra- of vulnerability in the long term. zil managed to foster consumer fi nance, it lags in FIGURE B5.1B Consumer and mortgage credit relative to GDP in Brazil, 2001–09 a. Consumer credit b. Mortgage credit 16 16 12 12 Percentage of GDP Percentage of GDP 8 8 4 4 0 0 2001 2002 2003 2004 2005 2006 2007 2008 2009 2001 2002 2003 2004 2005 2006 2007 2008 2009 Year Year Brazil Benchmark LAC 6 Brazil Benchmark LAC 6 Sources: Adapted from Didier and Schmukler 2011; De la Torre, Ize, and Schmukler 2012. Note: LAC 6 = Argentina, Brazil, Chile, Colombia, Mexico, and Peru. The international benchmark is based on regressing the variable of interest on country structural characteristics. a. Neri’s (2010) classification is close to the more widely known “Brazil Criterion,” which uses access to and number of durable goods, as well as the education of the household head, to classify households into income categories (for more details, see Neri 2010). pre- and post-transfers difference in the Gini both redistributive policies and growth in coefficient. Observe that pro-poor targeting average income affected the vulnerable class went beyond targeted cash transfers: spend- to a lesser extent. This is because the vulner- ing on health, education, nutrition, and basic able class faced both strong entry and exit infrastructure also became more pro-poor. flows. Thus, while that segment’s absolute In contrast, the new generation of social size may have remained relatively unchanged, programs had a lower incidence in middle- people belonging to the vulnerable class now- class households. Growth in average house- adays are not the same people who belonged hold income played a more important role to it 15 years ago. in inflating the ranks of the middle class: for Observe that GDP growth does not neces- the same set of countries, 74 percent of the sarily translate into higher household income. growth in the middle class (shown in the bot- The Dominican Republic and Uruguay, for tom panel of figure 5.5) can be attributed to instance, saw the size of their middle classes growth in average income, while reductions decline despite sustained economic growth in inequality accounted for only 26 percent (figure 5.4). In-depth, country-specific analy- of middle-class growth. Observe, also, that ses fall beyond the scope of this report. One 142 THE RISING LATIN AMERICAN AND CARIBBEAN MIDDLE CLASS should not forget, however, that to the extent • It is linked across regions, which allows that capital, rather than labor, may benefit for the influence of openness (through disproportionately from economic growth, trade and fi nance) on domestic variables incomes of the lower and middle classes may such as output and wages. not rise as much. It could also be that the • It has a more diverse set of productive fac- incomes of many poor and vulnerable grew, tors, including land, natural resources, but not enough to lead to a class transition. and skilled and unskilled labor. Moreover, household surveys substantially fail to capture incomes at the top. Thus, On the other hand, the GIDD simulation they are not informative in assessing how is based on microsimulation methodologies much the very rich (as opposed to households developed in recent literature (Bourguignon captured by the survey) benefited from the and Pereira da Silva 2003; Ferreira and Leite growth spurt of the past decade. Our find- 2003, 2004; Ravallion and Chen 2003; and ings therefore do not imply that growth was Bussolo, Lay, and Van der Mensbrugghe necessarily inclusive; rather, they suggest that 2006, among others). The authors’ start- where economic growth trickled down into ing point is the global income distribution in higher average household incomes, it was the 2000, assembled with data from household principal source of the middle-class expan- surveys that cover 91 percent of the world sion, reinforced by a reduction in income population. They then combine a set of price inequality. and volume changes from the LINKAGE model with expected changes in demographic structure to create a simulated distribution Forecasts for poverty reduction of income in 2030. Notably, they apply three and middle-class growth main changes to the initial distribution: demo- Poverty reduction and the rise of the middle graphic changes (including aging and shifts in class are expected to continue for the next the skill composition of the population); shifts two decades, albeit at a slower pace. Bus- in the sectoral composition of employment; solo and Murard (2011) forecast poverty and and economic growth (including changes in middle-class levels in 2030 for both Latin relative wages across skills and sectors). America and the emerging world. They base their forecasts on two tools developed by the Outlook for 2030 in Latin America Development Economic Prospects Group of the World Bank: (a) a LINKAGE global By 2030, 42 percent of Latin Americans computable general equilibrium (CGE) model are expected to be in the middle class, up that feeds into a (b) Global Income Distribu- from 29 percent in 2009 (as shown in fig- tion Dynamics (GIDD) simulation. ure 5.6). However, almost a fifth (18 per- cent) will remain in poverty. Over the next two decades, poverty is thus expected to fall New forecasting tools by approximately 14 percentage points— At its core, the LINKAGE CGE model is a slower decline than the recent one, where essentially a neoclassical growth model, with poverty fell by more than 10 percentage aggregate growth predicted on assumptions points during the 2000s. Lower rates of pov- regarding the growth of the labor force, sav- erty reduction are expected, both because ings and investment decisions (and there- the poverty gap remains relatively high in the fore capital accumulation), and productiv- region (hence, some of the remaining poor ity. Unlike simpler growth models, however, are far from the poverty line of US$4 per LINKAGE has considerably more structure: capita a day), and because of lower long-term growth forecasts with respect to the recent • It is multisectoral, which allows for dif- boom. Observe, also, that the proportion of ferentiating productivity growth between people in the vulnerable group is expected to agriculture, manufacturing, and services. remain at current levels until at least 2030. THE RISING LATIN AMERICAN AND CARIBBEAN MIDDLE CLASS 143 FIGURE 5.5 Decomposition of class growth attributable to income growth versus redistributive policies in Latin America, by country, circa 1995–2010 20      US$0–US$4 a day Change in percentage points 10 0 – 10 –20 –30 s ht s a il a ca ic r r as ico a y y do do rie ig trie ua ua az in bi m bl ile ru ur Ri Ar d ex nt m na Br ua lva nt u g ug Pe Ch e nd we un sta ep ra lo ge M ou Pa Ec Ur Sa Ho Pa un co Co nR Co nc El n ica ica ica in er er m Am Am Do tin tin La La Redistribution Growth 20      US$4–US$10 a day Change in percentage points 10 0 – 10 –20 –30 s ht s a il a ca lic r r as ico a ay y do do rie ig trie ua az in bi m ile ru ur ub Ri gu Ar d ex nt m na Br ua lva nt ug Pe Ch e nd we un sta ep ra lo ge M ou Pa Ec Ur Sa Ho Pa un n co Co nR Co nc El ica ica ica in er er m Am Am Do tin tin La La Redistribution Growth 20 US$10–US$50 a day Change in percentage points 10 0 –10 –20 –30 es d a il ile a ica ic r r as ico a y ru y do do a ua az in bi m te bl Pe ur tri gu Ch aR ex nt m na gh Br ua lva pu ug nd un ra lo ge M Pa st Ec ei Re Ur Sa Ho Pa co Co Co Ar w El n un ica ica er C in LA Am m Do tin La Redistribution Growth Source: Azevedo and Sanfelice 2012, using data from SEDLAC. Note: “Poor” = individuals with a per capita daily income lower than US$4. “Vulnerable” = individuals with a per capita daily income of US$4–US$10. “Middle class” = individuals with a per capita daily income of US$10–US$50. Poverty lines and other income thresholds are expressed in 2005 US$ PPP per day. PPP= purchasing power parity. SEDLAC = Socio- economic Database for Latin America and the Caribbean. 144 THE RISING LATIN AMERICAN AND CARIBBEAN MIDDLE CLASS FIGURE 5.6 Middle-class growth forecasts for all around the emerging world, especially in Latin America, 2005–30 countries that have faced long spells of sus- tained economic growth. Figure 5.7 shows 50 the evolution of the middle class in Latin America as a whole relative to the BRIC Percentage of population 40 countries (Brazil, the Russian Federation, India, and China), both as a percentage of the 30 population and in absolute terms. In Brazil, China, and Russia, the middle 20 class has gained dramatic relevance in just the 10 past 10 to 15 years. Around 2009, the middle class consisted of 61 million people in Brazil, 0 83 million in China, and 75 million in Russia. Poor Vulnerable Middle class US$0–4 a day US$4–10 a day US$10–50 a day When measured as a percentage of the popu- 2005 lation, however, the same countries appear 2030 to be at different stages. In Brazil, the emer- Source: Bussolo and Murard 2011. gence of a middle class is not an entirely new Note: Poverty lines and incomes are expressed in 2005 US$ PPP per day. phenomenon: already in the early 1980s, the PPP = purchasing power parity. SEDLAC = Socio-Economic Database for middle class consisted of more than 15 per- Latin America and the Caribbean. cent of the population, although now it con- sists of almost a third. The same can be said for Russia, where the middle class currently Outlook for 2030 throughout the comprises more than half of the population. emerging world In contrast, in India, with 8.8 million people, The steady growth of the middle class is not the middle class still remains fairly mod- specific to Latin America; it can be observed est in both absolute and relative terms. The FIGURE 5.7 Middle-class growth in the BRICs, circa 1980–2010 Millions of people . . . . . . and as a percentage of the population 180 60 Middle class, percentage of population 160 50 140 Middle class, millions 120 40 100 30 80 60 20 40 10 20 0 0 1975 1980 1985 1990 1995 2000 2005 2010 1975 1980 1985 1990 1995 2000 2005 2010 Brazil China Brazil China India Latin America and India Latin America and Russian Federation the Caribbean Russian Federation the Caribbean Sources: Data from PovcalNet, SEDLAC, and nationally representative household surveys. Note: BRICs = Brazil, the Russian Federation, India, and China. Covered Latin American countries include Argentina, Bolivia, Brazil, Chile, Colombia, Costa Rica, the Dominican Republic, El Salvador, Ecuador, Guatemala, Honduras, Mexico, Nicaragua, Panama, Paraguay, Peru, Uruguay, and República Bolivariana de Venezuela. “Middle class” = individuals with a per capita daily income of US$10–US$50, expressed in 2005 US$ PPP per day. PPP = purchasing power parity. SEDLAC = Socio-Economic Database for Latin America and the Caribbean. THE RISING LATIN AMERICAN AND CARIBBEAN MIDDLE CLASS 145 largest contributor to the growth of the mid- FIGURE 5.8 The emerging world’s middle-class dle classes in emerging countries, however, is growth forecasts, 2005 versus 2030 China, where sustained economic growth led to a stunning eightfold increase of the middle 2000 class in the past decade, surpassing both Bra- 1800 zil and Russia. Observe, however, that the middle class 1600 in China still represents a mere 6.3 percent 1400 of its population; hence its growth potential Million people 1200 remains enormous. Accordingly, Bussolo and Murard (2011) predict that most of the 1000 growth of the emerging world’s middle class 800 in the next two decades will stem from China 600 (see figure 5.8), where they forecast that the 400 middle class will grow from 54 million peo- ple in 2005 to more than 1 billion in 2030. 200 In contrast, although still growing in 0 2005 2030 absolute terms, the Latin American and Year Caribbean middle classes will gradually lose Central Asia and Middle East and predominance. In 2005, the region’s middle Eastern Europe North Africa classes represented more than 40 percent of China South Asia the entire middle class in low- and middle- East Asia Sub-Saharan Africa Latin America and income countries, but given the dramatic rise the Caribbean of China, that share is expected to drop to less than 20 percent in 2030. Source: Bussolo and Murard 2011. Overall, the growth of the emerging Note: “Middle class” = individuals with a per capita daily income of US$10– world’s middle classes will extend beyond US$50, expressed in 2005 US$ PPP per day. PPP = purchasing power parity. China and Latin America. The next two decades will be characterized by a massive increase of middle-class households through- shared identity? If so, is such an identity war- out emerging countries, from around 300 ranted in terms of economic and political million households in 2005 to almost 1.9 interests, as opposed to ethnic, racial, or reli- billion in 2030—approximately six times gious interests? the current population of the United States. Profiles of the middle class across coun- Of course, as with any projections about an tries and over time can help address some uncertain future, these numbers should be relevant questions. They tell us how middle- taken with a grain of salt. Forecasting is as class households differ from both poorer much an art as a science, and in two decades and richer households in terms of education, many factors could affect, in one way or employment, and other characteristics. They another, the parameters underlying the help us assess whether middle-class people forecasts. In particular, an average Chinese have particular attributes beyond their place economic growth rate of 7 percent between in the income distribution. The extent of 2005 and 2030 is a key driving assumption commonality across countries in middle-class behind these results. household characteristics, beyond the associ- ation with income, is also worth exploring: Does a middle-class household in Honduras Who is middle class in Latin look the same as a middle-class household America and the Caribbean? in Chile? And have the characteristics of Do members of the Latin American and the middle class in Latin America and the Caribbean middle class have a sense of Caribbean changed over time? This section, 146 THE RISING LATIN AMERICAN AND CARIBBEAN MIDDLE CLASS which draws on Birdsall (2012), describes the claiming statistical representativeness, we middle classes of eight Latin American coun- shall begin by looking at average characteris- tries: Brazil, Chile, Colombia, Costa Rica, tics of a poor household in El Salvador, a vul- the Dominican Republic, Honduras, Mexico, nerable household in Panama, and a middle- and Peru. class household in Argentina (see table 5.1). The countries, which differ from the eight receiving a more comprehensive analysis Broad class profiles from three below, are chosen to some extent because of exemplar countries the preponderance of the respective classes in Before plunging into the detailed country- these countries. We find marked differences specific profiles, we summarize the broader in households’ profiles across classes. regional profiles of poor, vulnerable, and middle-class households as well as the main Poor households in El Salvador trends arising from the analysis. Without An average poor household in El Salvador earns a daily income, for the whole family, of TABLE 5.1 Average class characteristics in El Salvador, Panama, and US$10.30 a day in PPP terms (US$3,760 per Argentina, 2009/10 year). It has 4.6 members, and the household head has 3.9 years of education. Only 40 per- Poor Vulnerable Middle class cent of working-age women (ages 25–65) are El Salvador Panama Argentina in the labor force. Workers from poor house- (2009) (2010) (2010) holds are roughly equally split between wage Household characteristics work and self-employment (around 40 percent Household income (daily US$) 10.3 26.5 54.3 of the workers in each employment category). Household income per capita Few are employers, and unemployment rates (daily US$) 2.3 6.8 20.9 remain high, although the latter may reflect, Household size 4.6 3.9 2.8 in part, structural characteristics of the coun- Age of household head 46.9 49.0 52.5 try. Almost no poor worker is employed by the Number of children 2.2 1.5 0.5 public sector, and most work in agriculture. Years of education (household head) 3.9 7.8 11.3 Female labor participation (25–65) 0.40 0.49 0.72 Vulnerable households in Panama Labor force characteristics (25–65) With a household income of US$26.50 a day Employer (%) 2 2 6 in PPP terms (US$9,670 per year), the aver- Employee (%) 40 64 75 age vulnerable household in Panama is richer. Self-Employed (%) 43 27 16 It has 3.9 members, and the household head Unpaid worker (%) 5 2 0 has 7.8 years of education. Female labor-force Unemployed (%) 10 6 3 participation is slightly higher (around 50 percent), and wage employment is now much Employment sector higher than self-employment (64 percent and Private sector (%) 99 88 79 27 percent, respectively). Among vulnerable Public sector (%) 1 12 21 workers, 12 percent are employed by the pub- Primary (%) 81 15 1 lic sector, and only a few work in agriculture Health, education, and services (%) 11 14 26 (although this feature also reflects structural Manufacturing (%) 5 11 14 characteristics of the country). Construction (%) 2 14 6 Source: Data from SEDLAC. Note: “Poor” = individuals with a per capita daily income lower than US$4. “Vulnerable” = individuals Middle-class households in Argentina with a per capita daily income of US$4–US$10. “Middle class” = individuals with a per capita daily income of US$10–US$50. Poverty lines and incomes are expressed in 2005 US$ PPP per day. PPP = Finally, at US$54.30 a day (US$19,820 per purchasing power parity. SEDLAC = Socio-Economic Database for Latin America and the Caribbean. year), the average household income of a THE RISING LATIN AMERICAN AND CARIBBEAN MIDDLE CLASS 147 middle-class family in Argentina is twice as TABLE 5.2 Trends in middle-class characteristics in Latin America much as the income of a vulnerable house- (pooled), 1992–2009 hold in Panama, and five times as much as   1992 2000 2009 the income of a poor household in El Sal- Middle class (% population) 15.5 21.2 29.9 vador. Household size is also smaller (2.8), and the household head is much more edu- Daily household income per capita (2005 US$ PPP) 18.9 19.6 19.3 cated (11.3 years). In the middle-class labor Years of education, adults 25–65 9.4 9.8 10.1 force, 6 percent are employers, 75 percent are wage workers, and only 16 percent are Age of household head 45.5 47.2 50.3 self-employed. At 72 percent, female labor- Age of children 0–17 8.7 9.3 9.4 force participation is significantly higher. Household size 3.3 3.1 2.9 The likelihood of working for the public Children per household 0.9 0.8 0.6 sector is also higher (21 percent), and only Source: Based on Birdsall 2012. 1 percent of middle-class workers work in Note: Pooled, population-weighted averages for Brazil, Chile, Costa Rica, Honduras, and Mexico. “Middle class” = individuals with a per capita daily income of US$10–US$50, expressed in 2005 US$ agriculture. PPP per day. PPP = purchasing power parity. Middle-class characteristics, selected countries among the poor increased by 1.9 and 2.2 To be sure, it could be rightly objected that years in Brazil and Mexico, respectively. these differences may reflect different stages Overall, the impressive changes at the coun- of economic development in the countries we try and regional levels do not concern class examine. Yet, at least for the middle classes, characteristics but rather the massive move- profiles differ surprisingly little across the ments of households along the income scale, eight countries we investigate below. This leading to dramatic increases in the middle is partly because PPP income thresholds are class. Although the middle class of today may applied across countries to categorize house- remain similar to the middle class of 20 years holds as middle class. Once income has been ago, many more households have reached the controlled for, however, marked differences standards that enable them to belong to it. could still subsist. Instead, the profiles reveal The rise of the middle class raises important that, with a few exceptions, being middle questions for policy making: Has anything class in Latin America and the Caribbean changed in the socioeconomic structure of carries common characteristics. society? And what are the implications of the In addition to differing little across coun- strengthening of the middle classes for the tries, the profile of the middle class also political process? We explore these questions seems to change little over time. Table 5.2 in chapter 6. Before doing so, however, we looks at trends in middle-class characteristics end this chapter by discussing country-specific for the pooled middle classes of Brazil, Chile, middle-class profiles. Costa Rica, Honduras, and Mexico. Between 1992 and 2009, schooling rose by less than a Demographics year. The average age of the household head rose by five years; average household size fell Middle-class households tend to have about by slightly less than 0.5 people, and average three to four people (more in poorer Hondu- children per household fell by 0.3. Although ras, fewer in richer Brazil and Chile) and an we do observe changes that reflect the major average of less than one child per household, ongoing economic and demographic shifts as shown in table 5.3. Brazilian households in the region, these changes tend to remain have an average of just 2.7 people and 0.3 relatively modest compared with changes in children. The average age of all middle-class characteristics of other groups: during the adults is 39 (younger in Honduras; older in same period, for instance, years of schooling Chile and Brazil), approaching age 40. 148 THE RISING LATIN AMERICAN AND CARIBBEAN MIDDLE CLASS As we would expect, the countries’ house- therefore, 80 percent of Brazilian children hold characteristics tend to vary monoton- are growing up in households that are not ically by class. Household size drops by one middle or upper class. to two individuals between the poor and the upper class, and the average adult age Education increases by four to seven years (with the exception of Honduras). The proportion of The average years of schooling of adults (ages elderly, in contrast, remains fairly constant. 25–65) increase with income class, as shown The differences in average household size in figure 5.9. The poor have not completed, and number of children accumulate across on average, the basic curriculum, while, in households in the different groups in a way virtually every country, the average adult in a that adds up: in Brazil, for instance, half of middle-class household has attended at least all children live in households that are below some secondary school. At the other extreme the US$4-per-day poverty line, and another of the spectrum, adults from the upper classes 30 percent live in vulnerable households; are far more likely to have attended, and even TABLE 5.3 Average household characteristics, selected Latin American countries, circa 2009   Poor   Vulnerable Age of Age of Household Children Children Adults adults Household Children Children Adults adults   size 0–12 13–18 over 70 18+   size 0–12 13–18 over 70 18+ Brazil 3.8 1.4 0.5 0.04 35 3.1 0.6 0.4 0.2 38 Chile 4.2 1.2 0.6 0.1 37 3.8 0.8 0.5 0.3 38 Colombia 4.2 1.5 0.6 0.2 37 3.8 1.0 0.5 0.1 37 Costa Rica 4.0 1.3 0.6 0.2 37 3.8 0.9 0.5 0.2 37 Dominican Republic 4.5 1.5 0.7 0.2 36 3.7 0.8 0.5 0.2 36 Honduras 5.0 1.8 0.8 0.2 36 4.5 1.2 0.7 0.2 35 Mexico 4.6 1.6 0.6 0.2 36 4.2 1.1 0.6 0.2 36 Peru 5.0 1.9 0.7 0.2 37 4.3 1.1 0.6 0.2 37   Middle class   Upper class Age of Age of Household Children Children Adults adults Household Children Children Adults adults   size 0–12 13–18 over 70 18+   size 0–12 13–18 over 70 18+ Brazil 2.7 0.3 0.2 0.2 40 2.2 0.2 0.1 0.2 42 Chile 3.2 0.5 0.3 0.3 40 2.6 0.3 0.1 0.1 41 Colombia 3.0 0.5 0.3 0.2 38 2.3 0.3 0.2 0.1 42 Costa Rica 3.3 0.5 0.3 0.1 38 2.5 0.2 0.1 0.1 41 Dominican Republic 3 0.5 0.3 0.2 37 2.2 0.2 0.1 0.2 41 Honduras 4.0 0.8 0.6 0.1 36 3.1 0.5 0.4 0.2 38 Mexico 3.4 0.5 0.4 0.2 38 2.4 0.2 0.2 0.2 42 Peru 3.4 0.5 0.4 0.3 39 2.4 0.2 0.2 0.3 42 Source: Based on Birdsall 2012. Note: “Poor” = individuals with a per capita daily income lower than US$4. “Vulnerable” = individuals with a per capita daily income of US$4–US$10. “Middle class” = individuals with a per capita daily income of US$10–US$50. “Upper class” = individuals with a per capita daily income exceeding US$50. Poverty lines and incomes are expressed in 2005 US$ PPP per day. PPP = purchasing power parity. THE RISING LATIN AMERICAN AND CARIBBEAN MIDDLE CLASS 149 completed, university. Three points about FIGURE 5.9 Average years of schooling (ages 25–65), selected Latin education are noteworthy: American countries, by income class, circa 2009 • Among middle-class adults, average years 16 of schooling vary little across countries; there is thus constancy in the crude rela- tionship between income (US$10–50 a 12 Years of schooling day) and schooling of adults throughout the region. 8 • Apart from Chile, the average schooling years of the middle class are about 30–50 4 percent higher than the schooling of the vulnerable, and 80–250 percent higher than the schooling of the poor. 0 • Within each category, however, there is il ile a ca lic as ico ru az bi Pe ur ub Ri Ch ex m Br nd sta ep lo M considerable variation, suggesting that Ho Co nR Co ica even if the association between class and in m years of schooling is strong, other factors Do influence class status as well. Poor Vulnerable Middle class Upper class Source: Based on Birdsall 2012. Note: “Poor” = individuals with a per capita daily income lower than US$4. “Vulnerable” = individuals Geography with a per capita daily income of US$4–US$10. “Middle class” = individuals with a per capita daily income of US$10–US$50. “Upper class” = individuals with a per capita daily income exceeding US$50. Figure 5.10 shows the percentage of house- Poverty lines and incomes are expressed in 2005 US$ PPP per day. PPP = purchasing power parity. holds living in urban areas, by country. Overall, the region is highly urbanized. Com- parisons across countries are not possible, FIGURE 5.10 Percentage of households living in urban areas, by however, because the definition of “urban” income class, selected Latin American countries, circa 2009 varies, but within each country, the middle classes are more likely than poorer classes 100 to live in urban areas, which is consistent with economic activity being concentrated in 80 urban areas. Overall, the region also shows a great Percentage 60 deal of internal migration: even among the poor, around half of the adults (ages 25–65) 40 migrated out of the municipality where they grew up (figure 5.11). The proportion of 20 migrants tends to increase with income in all countries, especially for the upper class. But 0 il ile a ca ic s ico ru how much the middle class differs from the a az bi l Pe ur ub Ri Ch ex m Br nd sta ep lo M poor and vulnerable, in terms of migration, is Ho Co nR Co ica very much country-specific. in m Do Poor Vulnerable Middle class Upper class Employment Source: Based on Birdsall 2012. Table 5.4 provides a breakdown of workers Note: “Poor” = individuals with a per capita daily income lower than US$4. “Vulnerable” = individuals by class and employment sector. The catego- with a per capita daily income between US$4–US$10. “Middle class” = individuals with a per capita daily income of US$10–US$50. “Upper class” = individuals with a per capita daily income exceeding ries aggregate across 17 sectors; “primary US$50. Poverty lines and incomes are expressed in 2005 US$ PPP per day. PPP = purchasing power activities” include agriculture, mining, and parity. 150 THE RISING LATIN AMERICAN AND CARIBBEAN MIDDLE CLASS FIGURE 5.11 Percentage of adults (25–65) living in a municipality The middle class also differs by employ- other than place of birth, by income class, selected Latin American ment status, as shown in table 5.5. Consistent countries, circa 2009 with the reality that many middle-class work- ers benefit from a regular wage or salary, they are more likely than poor and vulnerable 100 workers to be employees and less likely to be 80 self-employed—though a considerable num- ber of middle-class workers remain as such. Middle-class workers are also more likely to Percentage 60 be employers. However, in that respect, they 40 remain more similar to the poor and vulner- able than to the upper class, where the like- 20 lihood of being an employer is significantly higher. 0 Table 5.6 shows private and public il ile ca lic as ru employment by class. Casual observation az Pe ur ub Ri Ch Br nd sta ep might suggest that middle-class workers are Ho nR Co ica concentrated in public sector jobs, includ- in m ing those in state-owned enterprises. That Do Poor Vulnerable Middle class Upper class is true to some extent: between 9 percent (in Colombia) and 34 percent (in Honduras) of Source: Based on Birdsall 2012. middle-class workers are in the public sector. Note: “Poor” = individuals with a per capita daily income lower than US$4. “Vulnerable” = individuals On the other hand, in many countries, upper- with a per capita daily income of US$4–US$10. “Middle class” = individuals with a per capita daily income of US$10–US$50. “Upper class” = individuals with a per capita daily income exceeding US$50. class workers are as much or even more con- Poverty lines and incomes are expressed in 2005 US$ PPP per day. PPP = purchasing power parity. centrated in the public sector. Among private firms’ workers, the poor are more concen- trated in small firms, the upper class in large fishing, and “other” comprises mostly private firms. However, there is considerable varia- activities such as real estate and hotels and tion across countries. restaurants. To be sure, within each category, there are more- and less-skilled jobs com- Female labor-force participation manding more and less pay. Hence, it is not surprising that, for example, some workers Female labor-force participation is relatively in poor households work in the public sector high across the board (as shown in figure and some in rich households work in primary 5.12), which is consistent with rising levels of activities. education and urbanization as well as with At the same time, some broad patterns declining fertility. In middle-class house- emerge: middle-class workers are less likely holds, 60–70 percent of women are in the to work in the primary sectors and more labor force. This fits into a monotonic rela- likely to work in health, education, and pub- tion between female labor-force participation lic services (in both the public and private and income: for the most part, the higher the sectors) than their poorer counterparts. On income per capita of the household, the more this dimension, middle-class workers look far likely women are to be in the labor force. more like the typical “richer” worker than This association may arise from a num- the typical “poorer” one. This finding is con- ber of sources: Women’s contributions to sistent with (a) our data on schooling, where household incomes may move their house- differences are greater between the poorer holds into higher income categories. Women and middle groups than between the middle in more-affluent households also tend to and richer groups, and (b) the Latin Ameri- be more educated and may thus exhibit a can middle class’s concentration in the top higher likelihood of working. The notable two or three income deciles. exception is Peru, where female labor-force THE RISING LATIN AMERICAN AND CARIBBEAN MIDDLE CLASS 151 TABLE 5.4 Employment sector by class, ages 25–65, selected Latin American countries, circa 2009 percentage   Poor   Vulnerable Health, Health, Primary education, and Manu- Primary education, and Manu-   activities public services facturing Construction Other   activities public services facturing Construction Other Brazil 35.6 7.2 9.4 9.5 38.3 16.1 12.0 14.6 9.5 47.9 Chile 22.6 9.1 10.8 13.2 44.4 17.8 11.7 11.8 11.8 46.9 Colombia 36.1 8.9 9.7 5.8 39.5 18.6 11.0 13.7 6.9 49.8 Costa Rica 23.0 7.6 9.6 8.0 52.0 15.5 10.1 13.6 8.1 52.8 Dominican Republic 22.3 14.5 7.7 6.1 49.5 14.2 15.5 11.2 6.7 52.5 Honduras 57.3 5.3 11.2 4.8 21.4 17.1 10.6 18.0 8.8 45.5 Mexico 34.6 4.4 14.4 8.8 37.7 9.5 9.7 18.5 10.6 51.7 Peru 67.1 4.5 6.5 2.6 19.3 23.4 11.7 11.2 6.0 47.7   Middle class   Upper class Health, Health, Primary education, and Manu- Primary education, and Manu-   activities public services facturing Construction Other   activities public services facturing Construction Other Brazil 8.1 20.0 15.4 6.2 50.3 4.0 29.2 10.0 2.5 54.4 Chile 11.1 19.8 10.3 7.3 51.5 7.2 28.2 6.2 7.1 51.3 Colombia 7.9 19.3 15.6 4.4 52.8 4.7 28.7 10.3 2.1 54.2 Costa Rica 5.8 20.7 11.8 5.7 56.0 2.7 29.5 7.1 2.8 58.0 Dominican Republic 7.0 20.5 11.4 7.1 54.1 6.9 29.8 6.5 3.5 53.3 Honduras 7.2 19.8 13.4 6.6 53.1 10.4 24.8 8.3 2.9 53.6 Mexico 4.2 22.9 14.6 6.5 51.8 6.7 26.9 11.2 4.7 50.6 Peru 8.6 19.6 12.2 5.2 54.4 7.6 15.2 14.8 3.5 58.9 Source: Based on Birdsall 2012. Note: “Primary activities” include agriculture, mining, and fishing. “Other” comprises mostly private activities such as real estate and hotels and restaurants. “Poor” = individuals with a per capita daily income lower than US$4. “Vulnerable” = individuals with a per capita daily income of US$4–US$10. “Middle class” = individuals with a per capita daily income of US$10– US$50. “Upper class” = individuals with a per capita daily income exceeding US$50. Poverty lines and incomes are expressed in 2005 US$ PPP per day. PPP = purchasing power parity. participation is not only among the high- Consistently, being more educated and into est but is also flat across classes. This could wage employment, the middle-class profile reflect both greater pressure to maintain seems to be more stable, which suggests that high income and cultural differences. people need to reach certain levels of socio- economic characteristics to become less vul- nerable to poverty and be considered as mid- Summing up: Is Latin America a middle- dle-class households. class society? Although Latin America is in the process Although a third of the population in Latin of becoming a middle-class society, the trans- America and the Caribbean is middle class, formation is not yet complete. Some of the many people who left poverty are still in a social and political foundations for the sus- condition of vulnerability and require poli- tainability of the current trends are the theme cies that protect them from falling back. of the next chapter. 152 THE RISING LATIN AMERICAN AND CARIBBEAN MIDDLE CLASS TABLE 5.5 Employment status by class, ages 25–65, selected Latin American countries, circa 2009 percentage   Poor   Vulnerable Working Working Self- without Self- without Country Employer Employee employed salary Unemployed   Employer Employee employed salary Unemployed Brazil 1.2 46.8 25.2 12.9 13.9 2.2 65.3 21.4 5.6 5.6 Chile 1.4 54.0 15.6 0.5 28.4 0.8 74.2 14.9 0.3 9.8 Colombia 3.6 22.6 54.8 4.3 14.7 3.9 39.4 44.5 3.0 9.2 Costa Rica 6.0 48.5 27.5 1.0 17.0 5.6 66.1 21.8 1.2 5.3 Dominican Republic 2.2 43.0 48.6 0.8 5.4 3.8 50.5 42.8 0.7 2.1 Honduras 15.6 31.0 46.6 4.6 2.2 13.3 46.9 34.0 3.1 2.7 Mexico 5.1 51.2 31.1 7.0 5.6 3.8 71.5 18.6 3.3 2.8 Peru 4.1 18.8 51.7 23.1 2.3 5.7 44.2 39.6 7.5 3.1   Middle class   Upper class Working Working Self- without Self- without   Employer Employee employed salary Unemployed   Employer Employee employed salary Unemployed Brazil 7.4 66.7 20.0 3.4 2.5 20.9 60.0 16.4 1.3 1.4 Chile 3.0 70.7 21.9 0.4 4.0 15.0 60.0 21.5 0.2 3.4 Colombia 6.4 54.6 31.6 1.9 5.4 14.3 58.7 23.9 0.7 2.4 Costa Rica 8.6 72.5 16.0 0.9 2.0 17.6 73.6 7.4 0.3 1.0 Dominican Republic 8.2 52.8 36.0 1.1 1.9 17.2 46.8 34.5 0.0 1.6 Honduras 15.4 56.2 22.9 3.4 2.2 30.6 52.8 14.4 1.0 1.2 Mexico 6.9 76.3 12.9 2.2 1.8 21.1 65.0 8.7 3.5 1.6 Peru 8.6 56.3 28.1 4.2 2.8 20.5 63.8 12.3 0.6 2.8 Source: Based on Birdsall 2012. Note: “Poor” = individuals with a per capita daily income lower than US$4. “Vulnerable” = individuals with a per capita daily income between US$4–US$10. “Middle class” = individu- als with a per capita daily income of US$10–US$50. “Upper class” = individuals with a per capita daily income exceeding US$50. Poverty lines and incomes are expressed in 2005 US$ PPP per day. PPP = purchasing power parity. THE RISING LATIN AMERICAN AND CARIBBEAN MIDDLE CLASS 153 TABLE 5.6 Private and public employment by class, ages 25–65, selected Latin American countries, circa 2009 percentage   Poor   Vulnerable   Private firm   Private firm   Small Large Public firm   Small Large Public firm Brazil 75.4 19.2 5.4 56.5 33.6 9.9 Chile 51.9 38.8 9.3 39.0 50.7 10.3 Colombia 87.2 11.5 1.3 71.9 26.3 1.8 Costa Rica 66.0 27.9 6.1 50.6 40.0 9.4 Dominican Republic 67.1 21.6 11.2 61.1 27.0 11.9 Honduras 68.8 26.4 4.9 29.4 54.7 15.9 Mexico 75.7 20.9 3.5 54.6 36.1 9.4 Peru 81.9 15.4 2.7 66.7 22.9 10.3   Middle class   Upper class   Private firm   Private firm   Small Large Public firm   Small Large Public firm Brazil 43.7 37.7 18.6 34.8 35.4 29.8 Chile 36.6 47.7 15.7 29.9 49.5 20.6 Colombia 48.8 42.0 9.2 28.7 51.5 19.9 Costa Rica 35.4 39.7 24.9 20.6 43.3 36.1 Dominican Republic 49.2 33.4 17.4 40.3 44.2 15.5 Honduras 11.6 54.3 34.2 2.2 50.1 47.7 Mexico 37.3 38.7 24.0 27.5 51.4 21.1 Peru 50.2 31.6 18.2 29.2 53.4 17.4 Source: Based on Birdsall 2012. Note: Small firms have fewer than five employees. “Poor” = individuals with a per capita daily income lower than US$4. “Vulnerable” = individuals with a per capita daily income of US$4–US$10. “Middle class” = individuals with a per capita daily income of US$10–US$50. “Upper class” = individuals with a per capita daily income exceeding US$50. Poverty lines and incomes are expressed in 2005 US$ PPP per day. PPP = purchasing power parity. FIGURE 5.12 Female labor-force participation by class, ages 25–65, selected Latin American countries, circa 2009 100 80 Percentage 60 40 20 0 il ile a ca ic s ico ru a az bi l Pe ur ub Ri Ch ex m Br nd sta ep lo M Ho Co nR Co ica in m Do Poor Vulnerable Middle class Upper class Source: Based on Birdsall 2012. Note: Female labor-force participation is defined as the percentage of women between 25 and 65 who worked in a paid or unpaid job during the previous week (or month, for some surveys), provided a service, or looked for a job. “Poor” = individuals with a per capita daily income lower than US$4. “Vulnerable” = individuals with a per capita daily income of US$4–US$10. “Middle class” = individuals with a per capita daily income of US$10–US$50. “Upper class” = indi- viduals with a per capita daily income exceeding US$50. Poverty lines and incomes are expressed in 2005 US$ PPP per day. PPP = purchasing power parity. 154 THE RISING LATIN AMERICAN AND CARIBBEAN MIDDLE CLASS Focus Note 5.1 The Latin American middle class under alternative definitions There is no one unique definition of the middle class. but no upper threshold because otherwise counter- Rather, because levels and trends of the middle class intuitive trends, such as those in figure F5.1a, may over time are likely to be sensitive to the concept appear. behind them, the literature suggests that the concept • When possible, lower-income thresholds for the of (middle) class to be used must be linked to the middle class ought to be adjusted to a country or objective of the analysis (Sorensen 2005). region’s level of economic development by run- In what follows, we review how the main trends ning, for instance, multiple vulnerability analyses would vary for Latin America and the Caribbean along the lines of chapter 2. Although, for Latin under alternative definitions. In doing so, we explain America, US$10 a day seems to be an appropriate why we believe that the definition of the middle class lower threshold (it distinguishes the middle class we have adopted is the most appropriate for the pur- from the poor while also including a significant poses of this report. We focus attention on three share of the population), the same amount does definitions: not provide the same resilience to poverty in Sub- Saharan Africa or Organisation for Economic • An alternative, absolute definition of the middle Co-operation and Development countries. Even class that uses lower income thresholds (between absolute measures of the middle class ought to US$2 and US$13 a day), following Ravallion depend, in part, on considerations about the rela- (2009) tive position of the middle class, in full similarity • A relative definition of the middle class by Cruces, with poverty lines in high-income countries being López-Calva, and Battiston (2011) set higher than in low- and middle-income ones. • A sociological definition of middle-class status based on occupation, from Erikson and Gold- thorpe (1992). FIGURE F5.1A Middle-class growth trends in Chile under two absolute definitions, 1992–2009 Absolute definitions 0.8 We begin by comparing absolute definitions. Figure 0.7 F5.1a compares trends in the evolution of the middle Percentage of population 0.6 class in Chile using the Ravallion (2009) definition, with trends using the definition of this report. The 0.5 two measures show opposing trends: while, under our definition, the Chilean middle class has grown sub- 0.4 stantially in two decades; it decreased under the defi- 0.3 nition used in Ravallion (2009), which caps the mid- dle class at relatively low levels (US$13 a day), given 0.2 Chile’s per capita income. Hence, with continued 0.1 growth, more people leave the middle class to reach the upper-class level than extremely poor people join 0.0 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 the ranks of the middle class. Year The comparison presents two relevant lessons: Middle class US$2–US$13 a day • Middle-class measurements remain sensitive to Middle class US$10–US$50 a day the upper threshold. When attempting to make Source: Data from SEDLAC. Note: Income ranges are expressed in 2005 US$ PPP per day. PPP = purchasing international comparisons, therefore, it may be power parity. SEDLAC = Socio-Economic Database for Latin America and the preferable to only use a common lower threshold Caribbean. THE RISING LATIN AMERICAN AND CARIBBEAN MIDDLE CLASS 155 Focus Note 5.1 (continued) Relative definitions that the two definitions were capturing totally different households. In contrast, the overlap in We turn, next, to relative definitions of the middle Argentina, a wealthier country, is much greater, class. Although we choose, as a point of comparison, although it dropped during the 2002 South Ameri- the definition of Cruces, López-Calva, and Battiston can economic crisis. Absolute and relative con- (2011), other relative definitions tend to deliver simi- cepts of the middle class thus remain substantially lar results. Their definition draws from the polariza- different, at times capturing fully different strata tion literature (in particular, the work of Esteban and of the population. This is why the findings of our Ray [1994] and Esteban, Gradín, and Ray [2007]) report may present marked differences from other and relates the middle class to a measure of polar- studies (such as OECD [2010]) that have adopted a ization of income. Specifically, income thresholds relative definition. across classes are computed using a numerical pro- • The relative definitions exhibit extreme stability. cedure that maximizes income inequality (captured The Latin American middle class, when measured by the Gini coefficient) across classes while minimiz- in relative terms, has faced virtually no growth ing within-class inequality. Two interesting findings in the past two decades. It also did not signal emerge: any decline during the downturn in Argentina at the beginning of the 2000s. This stability occurs • Absolute and relative measures of the middle class because the relative definition relates to the shape do not necessarily overlap. In fact, in poor coun- of the income distribution, which is much more tries, they may fail to overlap at all. Figure F5.1b persistent than its mean (to which the absolute shows the evolution of the middle class in Peru and definition relates). Because this report explores Argentina, measured both in absolute terms (using directional income movements across classes, the the definition of this report) and in relative ones. use of an absolute measure therefore seems more Until 2005, there was no overlap in Peru, implying appropriate. FIGURE F5.1B Middle-class trends in Peru and Argentina under absolute and relative definitions, by income percentile, 1990s–2000s b. Argentina 1.0 1.0 0.8 0.8 Income percentiles Income percentiles 0.6 0.6 0.4 0.4 0.2 0.2 0 0 1995 2000 2005 2010 1990 1995 2000 2005 2010 Year Year Absolute middle class Relative middle class Absolute middle class Relative middle class Source: Data from SEDLAC. Note: Pairs of solid and dotted lines in each figure panel represent the upper and lower end of the range covered by their respective definitions. Under the absolute definition (used throughout this volume), “middle class” = individuals with a per capita daily income of US$10–US$50. Incomes are expressed in 2005 US$ PPP per day. PPP = purchasing power parity. SEDLAC = Socio-Economic Database for Latin America and the Caribbean. (Box continues next page) 156 THE RISING LATIN AMERICAN AND CARIBBEAN MIDDLE CLASS Focus Note 5.1 The Latin American middle class under alternative definitions (continued) Observe also, that the Cruces, López-Calva, and A sociological definition Battiston (2011) measure of the middle class sug- gests that Latin America is not only highly unequal Finally, a comparison of our absolute definition with but also highly polarized. Figure F5.1c shows the per- a sociological definition based on occupational sta- centiles of the income distribution where the relative tus also shows marked differences. We base our com- middle class starts and ends, using the polarization- parison on the Goldthorpe class schema often used in based definition, compared with a more traditional sociological class analyses (Erikson and Goldthorpe measure that draws from the literature on inequality 1992). According to its underlying theory, industrial- (specifically, people whose income is 75–125 percent ized societies are stratified because of an increase in of median income). It shows that although many the differentiation of labor. Differentiation gave rise Latin American and Caribbean countries do not have to a multiplication of scarce, yet desirable, techni- necessarily a much smaller relative middle class than cal and professional skills and to the emergence of do countries in other regions (with some exceptions, a middle class. The diversification of occupations such as Chile and Colombia), most Latin American can be classified according to the relations they form countries stand out for having a lower class extend- with each other (see also Bergman and Joye 2005). ing until above the median (that is, consisting of half Erikson and Goldthorpe (1992) identify 11 main or more of the population), at the expense of a much categories: higher- and lower-grade professionals; narrower upper class. administrators, officials, and technicians; routine FIGURE F5.1C Comparison of income polarization in selected countries of the world 1.0 0.8 0.6 Percentile 0.4 0.2 0 Se tes ite aki l Ur ey sta il te ico d a ian Eth an Au nds de ia Br c ut bia Ge alia Tu p. ca ca Eg d Ki stan d ra m M ay Bo e th nd Un ala n za y itz p. ep d e in Th cco ia Un P ga li Co az ite ysi Ka man ir Co hil lan M io n R ilan Fe iop Re Sw , Re liv t, A do ub rk Ri fri t u Ivo a Cô Mex Ne ola ne So lom hs t str la ug o C St ra hA er yp ng b or ica a er k a d’ r P re Ko ss m Ru Do EGR middle stratum Population between 0.75–1.25 median income Source: Data from SEDLAC. Note: “EGR middle stratum” = approach to polarization in Esteban, Gradín, and Ray 2007. SEDLAC = Socio-Economic Database for Latin America and the Caribbean. THE RISING LATIN AMERICAN AND CARIBBEAN MIDDLE CLASS 157 Focus Note 5.1 (continued) nonmanual employees (higher and lower grades); The adoption of an absolute measure also comple- small proprietors (with and without employees); ments a predominant emphasis on relative measures farmers and smallholders; lower-grade technicians; in some of the recent academic and policy literature skilled manual workers; semiskilled and unskilled (OECD 2010). manual workers; and workers in primary production. Figure F5.1d shows the association between income and occupational class in Chile. The dots show average income for each occupation, while the FIGURE F5.1D Average income by occupation type in bars show the standard deviation. Although there is Chile, 2009 some association between occupational status and income, the income within each category appears to 120 vary dramatically. Part of the variation within cat- Daily income per capita, 2055 US$ PPP 100 egories can be attributed to difficulties in relating information from the survey about occupation to 80 the Goldthorpe classification. But even within occu- 60 pational categories defined by the survey, variation in income remains significant. Observe, also, that 40 although they remain extremely informative, occu- 20 pational definitions of middle-class status make com- parisons across time challenging, and across coun- 0 tries almost impossible. They are thus poorly suited –20 for the purposes of this report. Summing up, although there is clearly no one –40 up yees unique ideal measure of middle-class status, one based an em es s ce fa r e s th mp al m l pl rm l er o a ille isor em Fa ua Lo rvic e u u em loy e n ervi oy lab rm ou loy ed ith an an an ua plo v on absolute income thresholds appears to be the most se s er m m w gh pl ed on- ed d led e appropriate for this report for two main reasons: Hi ls t kil Sk w ns in em out i-u wi M lf- m R • It is fully consistent with our focus on directional Se Se p oy income movement in the analysis of economic lf- Se lf- mobility within generations. Se • It lends itself well to a comparison of trends across Source: Income data from Chile’s 2009 Encuesta de Caracterización Socio- a group of countries characterized by a reasonable Económica Nacional (CASEN). Note: Occupational categories are from the Goldthorpe Occupational Classifica- degree of common cultural and economic identity tion (Erikson and Goldthorpe 1992). Dots show average income for each occu- (despite important income-level differences). pation. Bars represent one standard deviation. PPP = purchasing power parity. 158 THE RISING LATIN AMERICAN AND CARIBBEAN MIDDLE CLASS References for Latin America and the Caribbean.” Back- ground document, World Bank, Washington, DC. Azevedo, Joao P., and Viviane Sanfelice. 2012. Erikson, Robert, and John H. Goldthorpe. 1992. “The Rise of the Middle Class in Latin Amer- The Constant Flux: A Study of Class Mobility ica.” Unpublished manuscript, World Bank, in Industrial Societies. Oxford: Clarendon Press. Washington, DC. Esteban, Joan, Carlos Gradín, and Debraj Ray. Bergman, Manfred M., and Dominique Joye. 2007. “An Extension of a Measure of Polariza- 2005. “Comparing Social Stratification Sche- tion, with an Application to the Income Distri- mata: CAMSIS, CSP-CH, Goldthorpe, ISCO- bution of Five OECD Countries.” Journal of 88, Treiman, and Wright.” Working paper, Economic Inequality 5 (1): 1–19. Cambridge Studies in Social Research 10, Esteban, Joan, and Debraj Ray. 1994. “On the Social Science Research Group Publications, Measurement of Polarization.” Econometrica Cambridge, U.K. 62 (4): 819–51. Birdsall, Nancy. 2012. “A Note on the Middle Ferreira, Francisco H. G., and Phillippe G. Leite. Class in Latin America.” Unpublished manu- 2003. “Meeting the Millennium Development script, Center for Global Development, Wash- Goals in Brazil: Can Microeconomic Simula- ington, DC. tions Help?” Economía: Journal of the Latin Bourguignon, François. 2002. “The Growth American and Caribbean Economic Associa- Elasticity of Poverty Reduction: Explaining tion 3 (2): 235–79. Heterogeneity across Countries and Time ———. 2004. “Educational Expansion and Income Periods.” DELTA Working Paper 2002-03, Distribution: A Microsimulation for Ceará.” In DELTA (Ecole normale supérieure), Paris. Growth, Inequality, and Poverty: Prospects for Bourguignon, François, and Luiz A. Pereira da Pro-Poor Economic Development, ed. Anthony Silva. 2003. The Impact of Economic Policies Shorrocks and Rolph van der Hoeven, 222–50. on Poverty and Income Distribution. Washing- Oxford: Oxford University Press. ton DC: World Bank; New York: Oxford Uni- Lustig, Nora, Luis F. López-Calva, and Eduardo versity Press. Ortiz-Juarez. 2011. “The Decline in Inequal- Bussolo, Maurizio, Jann Lay, and Dominique van ity in Latin America: How Much, Since When der Mensbrugghe. 2006. “Structural Change and Why?” Economics Working Paper 1118, and Poverty Reduction in Brazil: The Impact Tulane University, New Orleans. of the Doha Round.” Policy Research Working Neri, Marcelo. 2010. The New Middle Class: The Paper 3833, World Bank, Washington, DC. Bright Side of the Poor. Rio de Janeiro: Funda- Bussolo, Maurizio, and Elie Murard. 2011. “The ção Getúlio Vargas Press. Evolution of the Middle Class in Latin Amer- OECD (Organisation for Economic Co-operation ica: 2005–2030.” Unpublished manuscript, and Development). 2010. Latin American Eco- World Bank, Washington, DC. nomic Outlook 2011: How Middle-Class Is CASEN (Encuesta de Caracterización Socioeconó- Latin America? Paris: OECD. mica Nacional), http://observatorio.ministerio PovcalNet (database). Online poverty analysis desarrollosocial.gob.cl. tool. World Bank, Washington, DC. http://ire Cruces, Guillermo, Luis F. López-Calva, and Diego search.worldbank.org/povcalnet. Battiston. 2011. “Down and Out or Up and In? Ravallion, Martin, 2009. “The Developing Polarization-Based Measures of the Middle World’s Bulging (buy Vulnerable) “Middle Class for Latin America.” Working Paper 113, Class”.” Policy Research Working Paper Series Center for Distributive, Labor and Social Stud- 4816, World Bank, Washington, DC. ies, Universidad de La Plata, Argentina. Ravallion, Martin, and Shaohua Chen. 2003. Datt, Gaurav, and Martin Ravallion. 1992. “Measuring Pro-Poor Growth.” Economics “Growth and Redistribution Components of Letters 78 (1): 93–99. Changes in Poverty Measures: A Decomposi- SEDLAC (Socio-Economic Database for Latin tion with Applications to Brazil and India in America and the Caribbean). Center for Dis- the 1980s.” Journal of Development Econom- tributive, Labor and Social Studies, Universidad ics 38 (2): 275–96. de La Plata, Argentina, and World Bank, Wash- De la Torre, Augusto, Alain Ize, and Sergio L. ington, DC. Schmukler. 2012. Financial Development in Sorensen, Aage B. 2005. “Foundations of a Rent- Latin America and the Caribbean: The Road Based Class Analysis.” In Approaches to Class Ahead. World Bank Latin American and Carib- Analysis, ed. Erik Olin Wright, 119–48. Cam- bean Studies. Washington, DC: World Bank. bridge, UK: Cambridge University Press. Didier, Tatiana, and Sergio L. Schmukler. 2011. World Bank. Various years. World Development “Financial Globalization: Some Basic Indicators Indicators. Washington, DC: World Bank. 6 The Middle Class and the Social Contract in Latin America T he Marxian sociologist Erik Olin the middle classes never lost momentum. Yet, Wright tells the story of a heated despite all the hopes that have been placed debate on British Broadcasting Cor- on the middle class as an agent of stability poration radio after the new seven-category and prosperity, and the attention the middle class scheme was introduced in the Brit- classes have received in both the academic ish census in 2001 (Wright 2005). A police and policy worlds (ADB 2010; OECD 2011; inspector talked about being classified in AfDB 2011), there is surprisingly little empir- class I, along with lawyers, doctors, and even ical evidence backing most assertions. The executives of private companies. “Does it social, political, and economic implications mean,” he reportedly asked, “that now I have of the rise of the middle classes in middle- to wear tennis whites when I go out to do income countries remain to be understood. my gardening?” Professor David Rose from In this chapter, we explore the potential the University of Essex, the author of the systemic implications of a larger middle class new categories, was challenged by one per- for the nature and quality of policy mak- son from the audience: “How can you have ing in Latin America and the Caribbean. a sense of solidarity and consciousness when Although we do not aim at providing a com- you are [category] ‘five’ or ‘seven’? . . . Can prehensive picture of these complex relation- you imagine the Communist Manifesto writ- ships, broad trends emerge that are relevant ten by the University of Essex?” for the region. Using cross-country analyses Indeed, many people associate the concept that include countries beyond Latin America of class with dimensions that go beyond its and the Caribbean, we find evidence of an economic basis and link it to aspects such association between larger middle classes and as identity, status, consumption patterns, better governance, deeper credit markets, and political beliefs. The middle class, spe- and greater spending on the social sectors, in cifically, has itself been associated in the lit- particular, public health and education. erature with a sense of cohesion around a set Such an association can have various of values that determine political attitudes. roots. It has often been claimed that the mid- Since the times of the classic Greek philoso- dle classes carry specific beliefs and values phers, the debate on the values and virtues of that lead to political, economic, and social 159 160 THE MIDDLE CLASS AND THE SOCIAL CONTRAC T IN LATIN AMERICA reforms. Middle classes, however, do not empirical associations, and which arguments need to carry “good values” to push for these still remain wishful working hypotheses. The reforms. Their higher incomes may simply chapter has three parts: give them greater voice to push for reforms that are beneficial for them. The two views • A review of the theory and evidence relat- are not mutually exclusive but carry dif- ing the middle classes to policy making ferent implications for policy: if the middle • A look at the values and beliefs of the classes have intrinsically “good values,” their Latin American middle classes growth is unambiguously beneficial to soci- • An examination of the fragmentation of ety. On the other hand, if they push agendas the Latin American social contract. that are beneficial for them, their growth is beneficial to society only to the extent that their needs are aligned with those of other The middle class and the shaping classes. It could be imaginable, for instance, of economic policy that growing middle classes may want to The emergence of a class of consumers that slow social spending targeted to the poor to has higher purchasing power is shifting limit the fiscal burden associated with it and, demand from basic goods to more sophis- in turn, to push for more public expenditures ticated ones, owing to larger incomes and, in services from which they benefit. possibly, changing consumer preferences. To answer these questions, we look next Demand for cars, personal computers, house- at the relationship between class and values hold appliances, and international tourism is in seven Latin American countries. We fail to booming all over the middle-income world. find values that distinguish the middle class But this mere economic effect is almost a from other classes in a particular way. With tautology. Wealthier households not only a couple of exceptions, values seem to asso- consume more but also want different goods, ciate monotonically with income. Moreover, and with sustained growth, consumption income accounts for only a small fraction of patterns in middle-income countries are the overall variation in values, while other bound to become closer to those of the rich factors, such as country effects, account for a developed world. much larger proportion. The more interesting, much speculated But even if economic class does not relate upon but less explored, aspect of growing significantly to values, is there a way to lever- middle classes pertains to their impact on the age the greater voice of the middle classes shaping of economic policy and the social toward higher inclusion and better gover- contract. Despite the plethora of case and nance in the region? We argue, in a conclud- historical studies attributing to the middle ing section, that an important obstacle to classes the merits of social cohesion and eco- social reforms is a historically fragmented nomic growth—including persuasive theo- social contract. Achieving a more inclu- retical arguments—the socioeconomic impli- sive social contract will require changing cations of a rising number of citizens with the framework in which the region thinks stronger economic means remains an open and operates—from a world of uncoordi- empirical question.1 nated patchwork approaches, each aimed at Along the history of social and economic addressing vulnerabilities and needs of spe- thought, the middle class has been assumed cific groups, to a more inclusive social con- to positively affect growth through various tract with which the poor, the vulnerable, channels. A first channel stresses how fixed and the middle class can all identify. costs in human and physical capital invest- The chapter may raise more questions ments may give the middle class a greater than it answers. In the process, however, we ability to make investments that lead to hope that the reader will acquire a clearer greater long-term returns. Under imperfect picture of which arguments are supported by credit markets, the poor may not earn enough THE MIDDLE CLASS AND THE SOCIAL CONTRAC T IN LATIN AMERICA 161 income or own enough savings to overcome a however, remains an open question. They fixed investment in the acquisition of human may, for instance, be influenced by the capital or productive physical capital. In this occupations of middle-class workers, which context, the larger the middle class, the more require skills and experience and thus may people who may be able to overcome these help in developing work ethics and patience fixed investments and contribute to economic (Doepke and Zilibotti 2008). There is also growth.2 a relevant literature that explores the links A second channel highlights the impor- between inequality, political economy, and tance of domestic markets for growth, which macroeconomic outcomes such as aggregate may be boosted by the middle classes’ higher investment and growth rates (see box 6.1). purchasing power and their demand for qual- Data do show a clear correlation between ity goods. Domestic markets may be particu- gross domestic product (GDP) per capita, larly important when world trade is costly democracy, and good institutions, but there is (Murphy, Shleifer, and Vishny 1989) and an ongoing debate on the extent to which the when middle-class consumers are able to play relationship can be interpreted causally (Ben- a catalytic role by providing a large market habib, Corvalan, and Spiegel 2011; Acemoglu for innovation that helps lower the prices of et al. 2008, 2009; Epstein et al. 2006; Glaeser new goods (Matsuyama 2002). From a global et al. 2004). Moreover, because of data con- perspective, the growth of the middle classes straints, most cross-country studies base their in emerging countries (in particular, China) results on GDP per capita, a good proxy for may also drive global consumption in the overall development that, however, embeds future and offset the falling trend in demand too many factors to properly distinguish the by American and European consumers (Kha- impact of the middle class from other char- ras 2010). acteristics of the economy. Studies looking The middle classes have not only been directly at the socioeconomic implications given a role as drivers of growth but have also of growing middle classes remain thus more often been perceived as agents of institutional qualitative in nature. Among the few, Birdsall change and democratization, as follows: (2010) argues that, though the middle class is key for government accountability and to • Particular attention has been given to the sustain good institutions, it may not benefit “modernization theory” (Lipset 1959), from the recent development of social poli- which looks at the extent to which more cies targeting the poor, which raises political affluent societies favor the creation and economy considerations on how best to sus- consolidation of democracies and, more tain inclusive growth—an argument that we generally, good institutions. will further develop in this chapter. • Conceptually, higher incomes may reduce In sum, despite a plethora of theoretical confl ict over income distribution because studies postulating that a strong middle class preferences for democracy and stability should bring stability and prosperity, rigorous may overcome the benefits from redis- statistical analyses remain scant. Next, there- tributive and expropriative activities (Ben- fore, we turn our attention to the data and habib and Przeworski 2006). investigate, by means of a newly developed • Citizens with higher human capital may data set, the extent to which middle classes be more effective in sustaining good insti- may be associated with good institutions. tutions (Glaeser et al. 2004). Despite the relevance and strong inter- est in understanding how the middle classes In addition to higher income and human may shape institutions, data constraints have capital, the middle classes may also hold limited the ability to conduct robust statisti- values that foster economic activity and the cal analyses. Only a handful of cross-coun- development of good institutions (Weber try data sets report headcount indexes for 1905 [2003]). How these values develop, income thresholds above US$4 a day, and 162 THE MIDDLE CLASS AND THE SOCIAL CONTRAC T IN LATIN AMERICA BOX 6.1 Inequality, growth, and institutions A large literature investigates how lower inequal- as variations in inequality are likely to be correlated ity and a larger class that “sits in the middle” of the with many unobservable factors associated with the income distribution may bring social cohesion and variable of interest. Moreover, the relationship could institutional change. This literature relates more also be nonlinear (Banerjee and Duflo 2003), and to a relative concept of the middle class and had income inequality may be a poor predictor of some a strong influence on the economic thought of the outcomes because other characteristics of society, past two decades. Early studies postulated that, in such as ethnic and religious fractionalization and democracies, “swing” voters who are at the median polarization, may be better related to outcomes such of the income distribution should drive redistribu- as instability and confl ict (Esteban and Ray 2008; tive decisions (Downs 1957; Roberts 1977; Meltzer Esteban and Schneider 2008). and Richard 1981). The higher the income inequal- Accordingly, a few studies also attempt to look ity (which implies a smaller middle class measured in at how equality and group homogeneity may lead relative terms), the more the income of the median to greater stability and better institutions. Easterly voter moves away from average income; hence, the (2001), for instance, looks at the extent to which a more demand there should be for redistribution, “middle class consensus” (defi ned as a high share which, the studies postulated, may lower economic of income in the hands of the middle three quintiles growth because of distortive taxation (Persson and combined with low ethnic fractionalization) affects Tabellini 1991; Alesina and Rodrik 1994). And growth and socioeconomic outcomes. He fi nds that even in nondemocratic regimes, income inequality, both the level and growth rate of per capita income by exacerbating the distance between median and are affected by the middle class consensus. He also average income, may foster social confl ict because looks at the extent to which the middle class consen- pressure for redistribution increases (Benhabib and sus is related to human capital, a variety of public Rustichini 1996). goods, and economic policies. Although the associa- Although these channels have strong theoretical tion with human capital and public goods tends to be foundations, most studies fail in fi nding a causally generally positive, controlling for per capita income clear-cut empirical relationship between inequality, lowers the significance of many results. In a related redistribution, social confl ict, and other variables paper, Easterly, Ritzen, and Woolcock (2006) also of interest. a The biggest challenge facing empirical fi nd that the middle class consensus affects institu- investigations are biases caused by omitted variables, tional quality. a. The advent of new cross-country data sets measuring income inequality has, in fact, spurred a plethora of studies looking at, among other things, the relationship between (a) inequality and growth (Persson and Tabellini 1991; Alesina and Rodrik 1994; Benhabib and Spiegel 1994; Forbes 2000; Barro 2000, 2008; Banerjee and Duflo 2003); (b) democracy (Barro 1999; Przeworski et al. 2000); (c) sociopolitical instability and conflict (Alesina and Perotti 1996; Perotti 1996; Esteban and Ray 2008; Esteban and Schneider 2008); and (d) corruption (You and Khagram 2005). the ones that do exist span time periods that classes and institutions, using better data (see are too short to exploit both cross-country box 6.2). The analysis focuses on the propor- and time series variations. As a result, cur- tion of individuals in extreme poverty (below rent analyses tend to use GDP per capita as US$2.5 a day in purchasing power parity opposed to actual household income, and [PPP] terms) and the proportion of individu- they fail to investigate directly whether, als who have reached middle-class status as postulated by the literature, “critical (above US$10 a day). The authors refrain masses” of people overcoming the middle- from using an upper income ceiling because class income threshold can affect institu- they pool together countries from all income tional outcomes. levels; hence, an income ceiling may lead to To cope with some of these pitfalls, strange measurements by which the middle Loayza, Rigolini, and Llorente (2012) build class in rich countries may be artificially too a new cross-country panel data set spanning small. The findings should, correspondingly, 672 yearly observations across 128 countries be interpreted as the impacts of a growing to revisit the relationship between the middle proportion of people with sufficient income THE MIDDLE CLASS AND THE SOCIAL CONTRAC T IN LATIN AMERICA 163 BOX 6.2 A new data set on the world’s middle classes Empirical analyses of the socioeconomic implica- PovcalNet database.a The data set is fairly balanced tions of growing middle classes suffer from a major across levels of economic development: 21 percent challenge: data availability, particularly data that of the observations are from high-income countries, are comparable across countries. There is currently 37 percent from upper-middle-income countries, 30 no data set that reports absolute measures of the percent from lower-middle-income countries, and 11 middle class (which, for many economic implica- percent from low-income countries. Because of the tions, may be the most appropriate variable to use) nature of the primary data, 17 percent of the coun- and that also has large enough cross-sections and tries and 38 percent of the annual observations are, long-enough time series to perform meaningful however, from Latin America. Because surveys tend cross-country comparisons. to report information either for income or expendi- To cope with this challenge, Loayza, Rigolini, tures, the data set reports, for each country, only one and Llorente (2012) have developed a cross-coun- of the two measures. try panel data set that contains information about All income and expenditures data are in 2005 the proportion of people living in extreme poverty U.S. dollars PPP. For each survey, current units are (below US$2.5 a day in per capita purchasing power fi rst corrected for infl ation using the national con- parity [PPP] terms), the percentage of the population sumer price indexes and then converted into 2005 that lives on more than US$10 a day, and overall U.S. dollars PPP using the International Compari- income inequality as measured by the Gini coef- son Program PPP conversions. Where possible, the fi cient. The data set spans 672 yearly observations conversion, weights, and methodology are the same across 128 countries, from 1967 to 2009. (Around as those used to compute internationally comparable 90 percent of the observations are, however, from poverty data. the 1990s and 2000s.) For the cross-country analysis, yearly observa- To compute the headcount indexes, the data tions are collapsed into five-year averages. Coun- set draws from (a) various World Bank collections tries with populations of less than 2 million are also of harmonized, nationally representative house- dropped. The fi nal data set thus contains 343 obser- hold surveys that contain information on income vations over 110 countries. By taking averages, the or expenditures, and (b) simulated distributions of proportion of observations from Latin America also income and expenditures from the World Bank’s is reduced to 24 percent. a. PovcalNet is an online poverty analysis tool, available at http://iresearch.worldbank.org/povcalnet. to undertake activities beyond constantly Differencing the regression equation also con- fighting poverty. trols for potential level effects caused by some An attempt is made to correct for reverse countries reporting income data while others causality and omitted-variable biases in the report expenditures. The method relies on association between the size of the middle similar instrumental variables to control for class and institutions. To do so, the cross- joint endogeneity (see also Loayza, Rigolini, country comparison draws from the general- and Llorente 2012). To be sure, GMM may ized method of moments (GMM) estimator not help in assessing causality under certain for panel data developed by Arellano and circumstances—for instance, if error terms Bond (1991) and Arellano and Bover (1995). suffer from higher order serial correlation. The GMM estimator takes advantage of the A Hansen-type test confirms the validity of panel nature of the data set in dealing with the moment conditions and their underlying country-specific effects and endogenous assumptions of the following analysis. But explanatory variables. Unobserved country- even if some skepticism may remain in inter- specific effects are controlled for by dif- preting the results causally, the mere associa- ferencing the regression equation and using tions we observe may be of interest. instrumental variables based on previous The cross-country comparison suggests observations of the explanatory variables. that looking at the association between GDP 164 THE MIDDLE CLASS AND THE SOCIAL CONTRAC T IN LATIN AMERICA TABLE 6.1 Relationship between economic development and institutions Social policy Economic structure Governance Public health Public education expenditures expenditures Mean applied Credit mkt Observations (% GDP) (% GDP) tariff liberalization Polity score Corruption Output per capita 1.209*** 0.717*** –2.179*** 0.585*** 2.727*** –0.656*** (Log of GDP per capita) [6.447] [3.183] [–3.882] [3.897] [4.563] [–4.423] Observations (5 year averages) 269 192 265 294 318 285 Number of countries 107 97 103 100 106 92 Hansen Test – p value 0.0349 0.330 0.0340 0.308 0.230 0.0450 Source: Loayza, Rigolini, and Llorente 2012. Note: z-statistics are in parentheses. GDP per capita is in PPP adjusted, constant 2005 prices. GDP = gross domestic product. PPP = purchasing power parity. *** p < 0.01, ** p < 0.05, * p < 0.1. and institutions, something commonly done, well specified. On conceptual considerations, may deliver an overly simplistic picture. Con- we would like to know which aspect of the trolling for the share of population below income distribution is most relevant: average and above given income thresholds gives, in output, income inequality, the prevalence of fact, a much richer picture and highlights the poverty, or, more specifically, the size of the association of the middle class with institu- middle class? tional reforms. Table 6.2 presents the results of the set of For these purposes, we look first at the regressions on the indicators of social policy, simple relationship between economic devel- economic structure, and governance, con- opment and institutions (table 6.1). We also sidering as explanatory variables not only represent a country’s economic development GDP per capita but also measures of poverty, by its GDP per capita and divide policies inequality, and the middle class. When con- and institutions into three broad categories: trolling for the size of the middle class, the social policies (public expenditures in health coefficients corresponding to GDP per cap- and education), market-oriented economic ita lose their significance, size, or even sign, structure (international trade and finance), depending on the regression. At the same and quality of governance (democratic par- time, the size of the middle class appears to ticipation and absence of official corruption). now carry the coefficients’ sign and signifi- A clear result emerges: GDP per capita sig- cance that GDP per capita used to have when nificantly and beneficially affects the indica- it was the only explanatory variable. tors of social policy, economic structure, and It is plausible, therefore, that the beneficial governance. Specifically, an increase in GDP effect that had been attributed to changes in per capita induces a rise in public health and GDP per capita actually corresponds to the education expenditures, a reduction in tariff evolution of the middle class. An expansion rates on international trade, a liberalization of the middle class has a significant impact of credit markets, an improvement in demo- on social policy by inducing an increase of cratic participation, and a reduction in offi- public health and education expenditures as cial corruption. a share of GDP. A larger middle class does The association found between GDP and not, however, necessarily mean a more state- institutions, however, may summarize more driven economy. An increase in the size of nuanced associations related to the distribu- the middle class reduces tariffs on interna- tion of income. From a statistical perspective, tional trade and liberalizes the financial it is unlikely that a regression model with only sector. No less remarkable is the effect on GDP per capita as an explanatory variable is the quality of governance. An expansion of THE MIDDLE CLASS AND THE SOCIAL CONTRAC T IN LATIN AMERICA 165 TABLE 6.2 The middle-class effect on indicators of social policy, economic structure, and governance Social policy Economic structure Governance Public health Public education expenditures expenditures Mean applied Credit market (% GDP) (% GDP) tariff liberalization Polity score Corruption Middle class 2.054*** 2.918** –10,945*** 1.357*** 6.431*** –1.764*** (% of population with income above 10 USD) [3.849] [2.337] [–3.072] [2.799] [4.068] [–4.767] Poverty –0.019** –0.042** 0.203*** 0.047*** –0.042** –0.011* (US$2.5 a day poverty headcount) [–2.411] [–2.472] [2.874] [3.383] [–2.345] [–1.825] Inequality –3.716** 3.028 20.736*** –0.209 9.262** 4.083*** (Gini Index) [–2.456] [1.360] [3.373] [–0.165] [2.164] [3.866] Output per capita 0.121 –0.922 5.485** 1.292*** 0.280 –0.470** (Log of GDP per capita) [0.416] [–1.310] [2.439] [3.009] [0.334] [–2.218] Observations (5 year averages) 269 192 265 294 318 285 Number of countries 107 97 103 100 106 92 Hansen Test – p value 0.174 0.640 0.934 0.469 0.701 0.451 Source: Loayza, Rigolini, and Llorente 2012. Note: z-statistics are in parentheses. GDP per capita is in PPP adjusted, constant 2005 prices. PPP = purchasing power parity. *** p < 0.01, ** p < 0.05, * p < 0.1. the middle class induces an improvement in hold for Latin America? And if they do, what democratic participation and a decline in are the reasons behind these reforms? Is the official corruption. middle class entrusted with specific values, or The indicators of poverty and inequal- does it act based upon self-interest? Under- ity are also relevant determinants for social standing well the reasons why the middle policies, economic structure, and governance class is associated with more social expen- quality, but not always in the expected way ditures, lower tariffs, more liberalized credit or with the consistency shown by the middle- markets, better functioning democracies, and class measure. For instance, a decrease in lower corruption is a challenging but essen- income inequality seems to produce not only tial exercise for sound policy analysis. For a decline in official corruption (as possibly one thing, if the middle class intrinsically expected) but also a reduction in democratic favors democracy and abhors corruption, it participation (which is hard to explain). Simi- can be an agent of change and reforms that larly, a decrease in the poverty headcount go beyond these specific issues. But it could appears to induce not only a liberalization of also be the case that an “income effect” international trade but also, surprisingly, a makes it simply more expensive to buy the constriction of credit markets. votes and favors of the middle class. Under The findings suggest that growing middle this alternative scenario, which has gained classes have enough voice to exert pressure traction in recent years (Fukuyama 2012), for reforms. When the size of the middle class the middle class may be a much less likely increases, social policy on health and educa- agent of change. tion becomes more active, and the quality of This report does not pretend to provide governance regarding democratic participa- comprehensive answers to these difficult tion and official corruption improves. questions. It aims, however, to shed some These regressions, however, include coun- light on how the growing Latin Ameri- tries from all over the world. Do the results can middle class may influence social and 166 THE MIDDLE CLASS AND THE SOCIAL CONTRAC T IN LATIN AMERICA economic policy along two dimensions: • How do political and social values vary First, it looks at the extent to which the Latin across income and class? American middle class may hold values that • To what extent does class, as opposed to distinguish it from other classes and may education and social origins, have a net favor institutional development. Second, it association with values? looks at the nature of the Latin American • Does the Latin American middle class social contract and at how growing middle hold specific values that distinguish it classes may affect it. from both upper and lower classes, or is the relationship between social classes and values, if any, a monotonic one? Values and beliefs of the Latin American middle classes It is important to state at the outset that Theories of middle-class values and beliefs the analysis will not claim to assert causal- contrast with the scarcity of empirical ity in the relationship between class and research on the association between income values. If the middle class is found to hold a (or occupation) and values, attitudes, and particular set of values, the analysis cannot behavior. Yet even from a theoretical perspec- establish that the level or sources of economic tive, the relationship is not necessarily obvi- well-being that characterize this class are the ous. Even if higher wealth and specific occu- cause of observed values, as implicitly sug- pations may lead to adopting a particular gested by theories on the role of the middle set of values—a hypothesis that is far from class in economic development or political proven—cultural and societal factors also stability. Endogeneity due to reverse causal- influence values, which may lead to tenuous ity or omitted-variable biases prevents such differences in the values profile with respect causal interpretation. But given the current to other classes. In this section, therefore, we status of research and the relevance of the review the association between income and question, documenting systematic variations values in Latin America. in values and orientations across education, Most empirical studies looking at middle- income, and occupation levels in Latin Amer- class values in emerging countries classify ica represents a first, necessary step in under- people based on self-perception of either standing how the emergence of new middle status or position in the income distribution classes may affect future growth and devel- (PRC 2008; Amoranto, Chun, and Deolalikar opment prospects. 2010; OECD 2011), but self-reported status Figure 6.1 shows the association between is a poor predictor of someone’s income, edu- values and beliefs, years of education, and cation, or occupation. In addition, attempts income class in a regression that also con- to use income measures in values surveys, trols for individual characteristics (age, gen- such as Cárdenas, Kharas, and Henao (2011), der, ethnicity) and country effects. To better are limited by the lack of accurate income compare the magnitude of the associations, information, which is either absent or classi- the effects are all expressed in terms of the fied into broad categories. Many studies also values’ standard deviation, and the education fail to compare income effects with relevant coefficient is also multiplied by its standard individual characteristics that could affect deviation. In addition, because of the difficul- values (such as education or occupation) and ties in including enough people from upper that could be in part captured by income. classes (those earning more than US$50 a day The analysis in this section, based on a per capita), the authors investigate differences study by López-Calva, Rigolini, and Torche in values between “lower-middle classes” (2011)—whose approach is described in box (with per capita incomes of US$10–US$20 6.3—attempts to solve some of these con- a day) and “upper-middle classes” (with per ceptual and technical issues. It addresses, for capita incomes above US$20 a day). Observe Latin America, the following questions: also that the associations with income shown THE MIDDLE CLASS AND THE SOCIAL CONTRAC T IN LATIN AMERICA 167 BOX 6.3 Studying middle-class values Middle-class values remain a challenging fi eld for Specifically, as in most values surveys, data on investigations. The first challenge to surmount is households’ income are unavailable in Ecosocial sur- conceptual: what measure of class to use? Results are veys. Therefore, they use information about house- likely to differ significantly if the measure of class is holds’ assets to construct a measure of households’ based on income rather than, say, occupation. It is permanent income—the long-term level of economic therefore important to clarify that values relate to a well-being, purged from short-term volatility and mea- specific aspect of the middle class as captured by the surement error (Torche 2009). To do so, they match measure that is used (in our case, absolute income). assets in Ecosocial with assets from an “external” The second challenge to surmount is statistical: household survey in each country that contains infor- analyses remain flawed by omitted-variable and mation on both assets and households’ income. Using reverse-causality biases. The latter can be quite these external surveys, they run a regression model worrying: Are values dictated by reaching a certain predicting the log of per capita household income by income, or did individuals manage to reach a certain means of the set of household goods and assets (con- income because they had specific values? trolling for the household head’s education) and the The third challenge regards the availability of log of household size. The coefficients obtained for the good data. The sampling rigor of some values sur- household goods and assets are then used in Ecosocial veys has been questioned, and most of them do a to predict, using the same set of assets and household poor job in capturing households’ income. This is characteristics, (the log of) per capita income for each why many studies of middle-class values use self- household. To achieve comparability across countries, reported status (PRC 2008; Amoranto, Chun, and they convert each income variable into 2005 U.S. dol- Deolalikar 2010; OECD 2011) or peoples’ perceived lar purchasing power parity terms. relative position in the income distribution (Fischer Finally, to investigate the association between and Torgler 2007) as indicators. Both are, however, income and values, they create “values indexes,” as poor substitutes for actual income. follows. First, they select a series of survey questions The analysis in López-Calva, Rigolini, and capturing orientations that are plausibly related with Torche (2011) does not resolve possible biases caused each other. They then extract the weight of each by omitted variables and, more important, reverse- variable in the fi rst principal component (the linear causality effects. The fi ndings thus only document combination that accounts for the largest propor- associations between income and values without tion of the variance across all items) and compute implying any causal link. On the other hand, the predicted values of the principal component for analysis makes a serious effort to address the chal- each observation in the data set. These new, sum- lenge of poor income data in values surveys. The mary variables constitute the dependent variables of analysis draws on the 2007 Ecosocial values surveys. the analysis. (For example, the value index “trust in These values surveys were implemented by the Cor- institutions” is based on five items, ascertaining trust poration for Latin American Studies (Corporación in the following institutions: the national govern- de Estudios para Latinoamérica ; CIEPLAN), a ment, congress, political parties, the mayor, and the Latin American think tank, in seven Latin American police.) This technique allows for substantive deci- countries: Argentina, Brazil, Chile, Colombia, Gua- sion making in terms of the items selected to identify temala, Mexico, and Peru. (However, the analysis each value index, while at the same time preventing does not use Argentinean data because of diffi cul- arbitrary combination of items that are only weakly ties in imputing income). The surveys are representa- correlated. Together, they investigate 11 dependent tive of the adult population (18 years or older) living variables: trust in institutions; political alienation; in large urban centers in each country. The authors perception of mobility and opportunity; support for choose to use the Ecosocial surveys because of their individual rights under any circumstances; legiti- rigorous sampling methodology, the information mization of political violence; voting; social toler- they collect on a variety of values, and the informa- ance; nationalism; political ideology; interpersonal tion they collect about households’ assets, which will trust; and interpersonal alienation. For a complete allow them to construct a measure of households’ list of indicators and details on the methodology, see permanent income. López-Calva, Rigolini, and Torche (2011). 168 THE MIDDLE CLASS AND THE SOCIAL CONTRAC T IN LATIN AMERICA FIGURE 6.1 Education, class, and values, selected Latin American countries, 2007 a. Trust in institutions b. Political alienation 0.3 0.3 Value indicator’s standard deviation Value indicator’s standard deviation 0.2 0.2 0.152 0.1 0.065 0.1 0.034 0.049 0.049 0.012 0 0 –0.021 –0.1 –0.1 –0.134 –0.2 –0.2 –0.3 –0.3 Education Vulnerable Lower- Upper- Education Vulnerable Lower- Upper- (US$4–US$10 middle class middle class (US$4–US$10 middle class middle class a day) (US$10–US$20 (US$20+ a day) (US$10–US$20 (US$20+ a day) a day) a day) a day) c. Perception of opportunity d. Support of individual rights under any circumstances 0.3 0.3 Value indicator’s standard deviation Value indicator’s standard deviation 0.2 0.176 0.2 0.150 0.121 0.1 0.1 0.053 0.053 0 0 –0.1 –0.1 –0.065 –0.120 –0.120 –0.2 –0.2 –0.3 –0.3 Education Vulnerable Lower- Upper- Education Vulnerable Lower- Upper- (US$4–US$10 middle class middle class (US$4–US$10 middle class middle class a day) (US$10–US$20 (US$20+ a day) (US$10–US$20 (US$20+ a day) a day) a day) a day) e. Legitimization of political violence f. Voted 0.3 0.3 Value indicator’s standard deviation Value indicator’s standard deviation 0.223 0.2 0.2 0.175 0.150 0.1 0.1 0.098 0 0 –0.032 –0.1 –0.072 –0.1 –0.103 –0.132 –0.2 –0.2 –0.3 –0.3 Education Vulnerable Lower- Upper- Education Vulnerable Lower- Upper- (US$4–US$10 middle class middle class (US$4–US$10 middle class middle class a day) (US$10–US$20 (US$20+ a day) (US$10–US$20 (US$20+ a day) a day) a day) a day) THE MIDDLE CLASS AND THE SOCIAL CONTRAC T IN LATIN AMERICA 169 FIGURE 6.1 Education, class, and values, selected Latin American countries, 2007 (continued) g. Social tolerance h. Nationalism 0.3 0.3 Value indicator’s standard deviation Value indicator’s standard deviation 0.2 0.2 0.1 0.099 0.1 0.067 0.036 0.006 0 0 –0.003 –0.050 –0.043 –0.1 –0.1 –0.096 –0.2 –0.2 –0.3 –0.3 Education Vulnerable Lower- Upper- Education Vulnerable Lower- Upper- (US$4–US$10 middle class middle class (US$4–US$10 middle class middle class a day) (US$10–US$20 (US$20+ a day) (US$10–US$20 (US$20+ a day) a day) a day) a day) i. Left-right political ideology (1 = left, 10 = right) j. Interpersonal trust 0.3 0.3 Value indicator’s standard deviation Value indicator’s standard deviation 0.2 0.2 0.141 0.1 0.059 0.1 0.081 0.010 0.009 0 0     –0.028 –0.006 –0.047 –0.1 –0.1 –0.2 –0.2 –0.3 –0.3 Education Vulnerable Lower- Upper- Education Vulnerable Lower- Upper- (US$4–US$10 middle class middle class (US$4–US$10 middle class middle class a day) (US$10–US$20 (US$20+ a day) (US$10–US$20 (US$20+ a day) a day) a day) a day) k. Interpersonal alienation 0.3 Value indicator’s standard deviation 0.2 0.1 –0.272 0 –0.1 –0.054 –0.2 –0.163 –0.170 –0.3 Education Vulnerable Lower- Upper- (US$4–US$10 middle class middle class a day) (US$10–US$20 (US$20+ a day) a day) Source: López-Calva, Rigolini, and Torche 2011. Note: Orange columns are statistically insignificant at the 10 percent level. Effects are all expressed in terms of the values’ standard deviation. Education is multiplied by its standard deviation. Class dummies refer to the difference from the poor (per capita income of US$0–US$4 a day in 2005 PPP terms). PPP = purchasing power parity. 170 THE MIDDLE CLASS AND THE SOCIAL CONTRAC T IN LATIN AMERICA in figure 6.1 should be interpreted as the dimensions. Higher education shows a posi- additional association of a given class with tive association with support of individual respect to the poor. rights under any circumstances (for example, Several findings emerge from the analysis. criminals should have the same rights as hon- Statistically, income is robustly associated est people), while the association with income with most values. Higher income is associ- classes appears to be nonlinear. And the ated with income-class variable shows a positive but insignificant association with a right-wing • More trust in institutions (an index based ideology, whereas the association between on how much individuals trust the gov- education and right-wing ideology remains ernment, congress, political parties, the negative. mayor, and the police) Overall, the analysis summarized in • Lower political alienation (the perceived figure 6.1 provides little support for theo- extent to which people in power care ries attributing special merits to the middle about people similar to the respondents as classes. Most values and beliefs tend to vary opposed to taking advantage of them) monotonically with income, and those of the • Stronger perception of opportunities (the middle class tend to be between those of the degree of perceived meritocracy in society poor and the rich. The only two exceptions and the perceived easiness of overcoming are the support for individual rights under poverty) any circumstances—where the two classes in • Less legitimization of political violence the middle seem to exhibit less support than (the use of violence to achieve socially the poorer and richer classes—and social desirable goals) tolerance (captured by tolerance for indi- • Higher likelihood of voting vidual traits such as race and homosexual- • Lower nationalistic beliefs ity), where the middle classes exhibit higher • Stronger belief that people can, in gen- tolerance than the poorer and richer classes. eral, be trusted If anything, the only value that consistently • Lower interpersonal alienation (a belief emerges from the analysis is moderation. that what happens to a person does not Values of the middle class are repeatedly count much). and consistently more moderate than those of people at the extremes of the income (and Unfortunately, it is impossible to interpret education) distribution. That in itself is a these differences causally. On the one hand, noteworthy finding that may affect the shap- values may drive success: for instance, people ing of social and economic policies because who believe in meritocracy may perform bet- moderation can indeed represent a force of ter in life. On the other hand, differences in social cohesion that intermediates between values may reflect the reality of Latin Ameri- the rich and the poor. can societies, where wealthier classes live in Nonetheless, we should avoid inferring too a different world that is closer to the politi- much from these associations for several rea- cal process and that may bear more influence sons. First, despite the statistically significant on social and political decisions. Such a view associations, the magnitude of the income may be reinforced by the fact that the class effects remains fairly small, suggesting that that seems to most distinguish itself from the income explains only a small fraction of the poor (people with incomes below US$4 a variation in values. In fact, even by look- day, the “omitted” class in the analysis) com- ing at all the explanatory variables together prises people earning between US$20 and (which include, in addition to income, indi- US$50 a day. vidual characteristics and country effects), Observe that although, overall, income the proportion of the overall variation in classes follow a pattern similar to that of values that is explained remains fairly small. education (although their association is con- The R-squared of the regressions—a mea- trolling for it), they differ in a few important sure of how much the regressions are able to THE MIDDLE CLASS AND THE SOCIAL CONTRAC T IN LATIN AMERICA 171 “capture” the variation in values—remain values of the middle class show moderation, overall very low, on the order of 2–15 per- they do so only within boundaries dictated cent. That means that 85 percent or more of by society. the variation in values remains unexplained. This should not be surprising at all. Over This poor performance of income and edu- the past century, the middle class has been cation (and of other individual characteris- tolerating extremist movements as much as tics, such as age, gender, and ethnicity that it supported reforms. For instance, an influ- are included in the regressions) suggests that ential 1967 article about the relationship of other factors must also influence values. To middle classes and military regimes asserts address this possibility, the authors run the that the middle class aspires to become part same regressions as in the baseline analysis, of the elite and is willing to abandon demo- but adding people’s occupation. Using occu- cratic values when it perceives a threat to its pational status available in Ecosocial, they own class status (Nun 1967). The moderation classify occupations into unskilled jobs, self- of middle-class values may, therefore, reflect employed, manual skilled, clerical (low), cler- a pragmatic attitude. Using our indicators, ical (high), professional independents, high for example, economic success may engender professional executives, workers in the home, in the middle class more trust in institutions students, and people not in the labor force. and a stronger perception of opportunities in Although they do find that some occupations life, but, at the same time, the middle classes (in particular, high professional executives) are not ready to support individual rights show some association with values, the asso- of criminals that have undermined law and ciation for most categories (after correcting order, a necessity for prosperous economic for individual characteristics, income, and activity. education) remains weaker than the one for The lack of strong values leading to greater income. Moreover, adding occupation only stability and cohesion, however, may further marginally improves the R-squared. The undermine a Latin American social contract exercise suggests, therefore, that values are that appears to be already under stress. We difficult to explain for many definitions of conclude the chapter by looking at these the middle class, not only income-related issues in deeper detail. ones. Second, the variation in values that is driven by income and education remains sig- Overcoming a fragmented nificantly smaller than the variation in val- social contract ues across countries. Figure 6.2 compares, The Latin American middle classes do not for selected values, the magnitude of their appear to hold exceptional values that may association with income and country effects. lead to greater stability and social cohesion. To ease the comparison, the income regres- In fact, as this concluding chapter suggests, sion coefficient is multiplied by the standard they appear to be rather pragmatic, support- deviation of income. It shows that income ing policies that are good for them, and, in has a relatively small association with val- some areas, may be opting out from a social ues such as trust in institutions, perceptions contract from which they see little benefit. of opportunities, and social tolerance when Many middle-class Latin Americans do not compared with the variation across coun- rely on the state for basic services such as tries. The strong cross-country variation education and health, and, in some cases, undermines further the supposition that the even for core public services such as the pro- middle classes share common values across vision of electricity and security. Latin America that can lead to greater social Although the middle classes may opt cohesion and economic prosperity. Values out from some basic services (see box 6.4), appear to be, to some extent, circumstan- they also benefit disproportionately from tial and driven by changing challenges and other services. The middle classes, for socioeconomic environments. And while the instance, benefit disproportionately from 172 THE MIDDLE CLASS AND THE SOCIAL CONTRAC T IN LATIN AMERICA FIGURE 6.2 Income versus country-specific values, selected Latin American countries, 2007 a. Trust in institutions b. Perceptions of opportunity 0.8 0.8 0.6 0.6 Value indicator’s standard deviation Value indicator’s standard deviation 0.4 0.311 0.4 0.189 0.226 0.2 0.166 0.2 0.114 0.059 0.040 0 0 –0.061 –0.2 –0.2 –0.194 –0.228 –0.4 –0.4 –0.331 –0.6 –0.6 –0.526 –0.8 –0.8 Ln Brazil Chile Colombia Mexico Peru Ln Brazil Chile Colombia Mexico Peru (income) (income) c. Legitimization of political violence d. Social tolerance 0.8 0.8 0.6 0.6 Value indicator’s standard deviation Value indicator’s standard deviation 0.4 0.4 0.288 0.259 0.2 0.2 0.108 0.078 –0.774 0.015 0 0 –0.046 –0.015 –0.2 –0.2 –0.232 –0.263 –0.4 –0.4 –0.472 –0.6 –0.526 –0.6 –0.8 –0.8 Ln Brazil Chile Colombia Mexico Peru Ln Brazil Chile Colombia Mexico Peru (income) (income) Source: Lopez-Calva, Rigolini, and Torche 2011. Note: Ln(income) = natural logarithm of income. Orange columns are statistically insignificant at the 10 percent level. Effects are all expressed in terms of the values’ standard devia- tion. Country dummies refer to the difference with respect to Guatemala. Income is multiplied by its standard deviation. public tertiary education and subsidized a report on its own. Rather, it wants to high- social insurance schemes, which only margin- light some features that may affect the mobil- ally address the needs of the poor. The Latin ity prospects of the remaining poor and may American and Caribbean social contract is a prevent, at the same time, achievement of a fragmented one, with facets of it benefiting more cohesive and stable contract. It shall do different classes and being only loosely con- so by focusing attention on two areas of the nected with one another. welfare state that have received strong atten- This section does not aim to provide a tion: cash transfers and education. comprehensive picture of the Latin American As with many institutions, the welfare social contract, which could be the subject of state reflects the spirit and nature of a social THE MIDDLE CLASS AND THE SOCIAL CONTRAC T IN LATIN AMERICA 173 contract. Rather than following a clear vision, programs are directed to the poor and vul- however, most Latin American welfare states nerable, in countries like Bolivia and Brazil, are the result of a buildup of features, each the middle class receives as much in cash aimed at addressing specific challenges. transfers as the poor (see figure 6.3). The Many welfare states thus remain highly frag- lack of progressiveness of cash transfers in mented, providing through various channels these countries stems from either the intro- differentiated services to various population duction of universal benefits or the legacy of groups, often separated along the divide of social insurance systems for the formal sec- labor formality (Ribe, Robalino, and Walker tor (often pensions and unemployment insur- 2010). History may be the culprit. Tradition- ance) that were financed out of general taxa- ally, social security systems in Latin America tion. These schemes may pose a challenge in were designed for middle-class, formal-sector expanding benefits for the poor because they workers under the belief that coverage would consume large portions of the governments’ have expanded as the economies formalized. budgets. Once it became clear that labor informality The fragmentation of the welfare state was not disappearing, however, many coun- may pose several challenges: tries began implementing parallel systems to fill the coverage gaps by offering protection • By having specific programs tailored to to informal workers (Kaplan and Levy 2012; each class, it may contribute to promot- Antón, Hernández, and Levy 2012). ing competition among classes for limited To be sure, the rapid developments of tar- resources. These tensions can be particu- geted cash transfer programs, combined with larly strong if needs differ across classes. heavy investments in primary and secondary • It may distort labor markets. If the social education, have borne fruit. Between 2000 insurance system is too generous, it may and 2009, around 10 percent of the Latin exacerbate the insiders-outsiders divide American population was lifted out of mod- and hence the vulnerability of informal erate poverty (measured as the proportion of labor. On the other hand, as the study of people with incomes below US$4 per capita Seguro Popular in Mexico seems to sug- per day; see World Bank 2010), and while gest, generous social assistance programs this would not have been possible without that compete with social insurance may sustained economic growth, the develop- stimulate informal labor and create an ment of targeted cash transfer programs and unjust situation, where formal workers the improvement in social spending toward are forced to pay to receive benefits that more progressivity played an important role are similar to those given to informal (López-Calva and Lustig 2010). workers for free (Levy 2008). Yet although these efforts did contribute • It may also pose a challenge in provid- significantly to poverty reduction, they were ing effective protection because people for the most part built on top of existing are more prone to fall through the cracks welfare states, partly because it would have (Ribe, Robalino, and Walker 2010). been politically difficult to introduce drastic overhauls of social protection systems. The Fragmentation and truncation also extend result of this uncoordinated development is a to education. On first impression, the region quite heterogeneous picture of social policies made significant efforts to increase educa- across countries, with some having managed tion spending for the poor, which brought to achieve a fair amount of progressiveness results: in Chile, Costa Rica, and Mexico, in their social spending, while others still for instance, net secondary enrollment rates distribute benefits in a fairly flat manner. among children from the poorest income An analysis by Lustig (2011) suggests, for quintile rose by 24, 53, and 38 percentage instance, that although, in countries such points, respectively, between the 1990s and as Argentina and Peru, most cash transfers 2009 (SEDLAC 2011). These advances are 174 THE MIDDLE CLASS AND THE SOCIAL CONTRAC T IN LATIN AMERICA FIGURE 6.3 Class incidence of social policies, selected Latin American countries, circa 2007–10 a. Argentina b. Bolivia 150 150 Spending per capita, extreme poor = 100 Spending per capita, extreme poor = 100 120 120 90 90 60 60 30 30 0 0 Extreme poor Poor Vulnerable Middle class Extreme poor Poor Vulnerable Middle class (0–2.5) (2.5–4) (4–10) (10–50) (0–2.5) (2.5–4) (4–10) (10–50) Main cash transfer programs Main cash transfer programs Nontertiary public education spending Nontertiary public education spending c. Brazil d. Peru 150 150 Spending per capita, extreme poor = 100 Spending per capita, extreme poor = 100 120 120 90 90 60 60 30 30 0 0 Extreme poor Poor Vulnerable Middle class Extreme poor Poor Vulnerable Middle class (0–2.5) (2.5–4) (4–10) (10–50) (0–2.5) (2.5–4) (4–10) (10–50) Main cash transfer programs Main cash transfer programs Nontertiary public education spending Nontertiary public education spending Source: Lustig 2011. Note: Class status is based on net market income. Upper class is omitted. “Extreme poor” = per capita income of US$0–US$2.50. “Poor” = per capita income of US$2.50–US$4. “Vulner- able” = per capita income of US$4–US$10. “Middle Class” = per capita income of US$10–US$50. Poverty lines and incomes are expressed in 2005 US$ PPP per day. PPP = purchasing power parity. Main cash transfer programs include Jefas y Jefes de Hogar, Familias, unemployment insurance, scholarships, noncontributory pensions, food, and Asignación Universal Por Hijo (Argentina); Bono Juancito Pinto, school feeding, PAN, Bono Sol, lactation subsidy, Bono de Natalidad, and pensions (Sistema de Reparto) (Bolivia); Bolsa Família, other scholarships, Benefício de Prestação Continuada, unemployment benefits, special circumstances pensions from INSS, and other social programs (Brazil); and Juntos and food transfers (Peru). reflected in the incidence of non-tertiary class in Argentina, Bolivia, Brazil, and Peru education spending, which also suggests is, respectively, 4.3, 1.8, 2.7, and 3.0. quite a progressive picture (the right-side These figures, however, may be as much bars in figure 6.3): the ratio of spending per the result of past successes as the source of capita in primary and secondary education future challenges. They indeed hide a high between the extreme poor and the middle degree of fragmentation of service provision. THE MIDDLE CLASS AND THE SOCIAL CONTRAC T IN LATIN AMERICA 175 Even within the education sector, the picture FIGURE 6.4 Incidence of tertiary public education spending, almost reverses in all countries for tertiary selected Latin American countries education. In the four countries analyzed by Lustig (2011), the middle classes benefit 1000 disproportionately from public tertiary edu- Spending per capita, extreme poor = 100 cation spending (see figure 6.4). A distribu- 800 tional distortion may have emerged in some Latin American and Caribbean countries, where poor children who attend low-quality 600 public schools cannot access high-quality public universities, which generally establish 400 high standards for admission. The poor— whose willingness to pay for education is 200 high and whose capacity to pay has increased in recent years—may end up paying tuition 0 in low-quality tertiary education institutions, Extreme poor Poor Vulnerable Middle class which have expanded dramatically in the (0–2.5) (2.5–4) (4–10) (10–50) past 15 years (UNESCO 2008). At the same Argentina Bolivia Brazil Peru time, better-educated middle-class children may receive free higher education in the best Source: Lustig 2011. Note: Class status is based on net market income. Upper class is omitted. “Extreme poor” = per public universities. Not surprisingly, political capita income of US$0–US$2.50. “Poor” = per capita income of US$2.50–US$4. “Vulnerable” = per movements related to public higher educa- capita income of US$4–US$10. “Middle Class” = per capita income of US$10–US$50. Poverty lines and incomes are expressed in 2005 US$ PPP per day. PPP = purchasing power parity. tion in Brazil, Chile, Mexico, and other Latin American countries have been led and sup- ported by the middle classes (Arocena and FIGURE 6.5 Percentage of students 6–12 years old enrolled in Sutz 2005; Lustig, Mizala, and Silva 2012). private schools, by income group, selected Latin American countries To add to the fragmentation, the middle and upper classes seem to opt out dispropor- 100 tionately from publicly provided primary and secondary education. Figure 6.5 shows the percentage of students 6 to 12 years old who 80 are enrolled in private schools, by income group. In most countries, with the exception 60 Percentage of Chile (which has fostered private school enrollment through vouchers), the figure 40 shows a sharp contrast between enrollment in private schools of the poor and vulnerable 20 and such enrollment of the middle and upper classes: in Brazil, for instance, only 13 per- 0 cent of children in the vulnerable class attend ile a ico ca il ru ic bi az l private schools at primary-school age, while Ri Ch Pe ub ex m Br sta lo ep M Co Co the proportion for the middle class is almost nR ica half (45 percent). And even in Costa Rica, one in m Do of the countries with the highest educational achievements in Latin America and the Carib- 0–25 2.5–4 4–10 10–50 50+ bean, only 2 percent of vulnerable children Source: Data from SEDLAC. attend private schools at primary-school age, Note: “Extreme poor” = per capita income of US$0–US$2.50. “Poor” = per capita income of US$2.50– while the proportion jumps to 25.3 percent US$4 per day. “Vulnerable” = per capita income of US$4–US$10 per day. “Middle Class” = per capita income of US$10–US$50 per day. “Upper class” = per capita income exceeding US$50 per day. for the middle class. The picture does not Poverty lines and incomes are expressed in 2005 US$ PPP per day. PPP = purchasing power parity. change much at the secondary level. SEDLAC = Socio-Economic Database for Latin America and the Caribbean. 176 THE MIDDLE CLASS AND THE SOCIAL CONTRAC T IN LATIN AMERICA BOX 6.4 Individualization of public goods and lack of institutional trust in the Dominican Republic Sánchez and Senderowitsch (2011) study how the mid- dle classes now have electrical inverters and genera- dle class in the Dominican Republic has resorted to tors at home (fi gure B6.4), much more so than the individualized solutions to substitute for faulty public poor and vulnerable. The charge needs of the invert- goods. Examples of this “adaptive behavior” include ers’ batteries represent 63 percent of the average elec- the heavy reliance on domestic generation of electric- tricity consumption by a middle-income household ity, the digging of wells to get running water at home, (453 kilowatt-hours a month), and approximately the use of private companies to report car accidents 246 gigawatts-hours a month are consumed just to (instead of doing it at the police station), and the use support individual private autogeneration capacity. of private services for education and health—these last This inefficiency in the system is a significant oppor- two also common in other countries of the region. tunity cost, which burdens both the state and fi nal Consider the electricity sector, for instance. In consumers, who pay around US$240 million of their the Dominican Republic, it has been characterized annual electricity bill to keep the inverters working. by large energy losses caused by deficient infrastruc- The authors also provide suggestive evidence that tures and maintenance, poorly targeted subsidies, low levels of institutional trust tend to reinforce the and widespread nonpayment of bills. Reforms have individualization of services and weaken the demand proven to be slow and not always successful: for for better services. This may have generated a vicious instance, the privatization process of the distribution cycle of poor service delivery, which self-reinforces companies had to be partially reversed because the low trust in institutions because “weak spending mentioned inefficiencies resulted in persistent losses institutions and low quality services . . . spawn dis- for private operators (Reinstein and Cayo 2010). To satisfaction and may underpin . . . low level of trust in cope with unstable provision of electricity, the mid- public institutions” (Ferroni, Mateo, and Payne 2008) FIGURE B6.4 Ownership of electrical inverters in the Dominican Republic, 2010 70 61.2 60 Households with inverter, percent 50 44.9 40 30 20.1 20 10.3 10 6.0 0 Extreme poor Poor Vulnerable Middle class Upper class (0–2.5) (2.5–4) (4–10) (10–50) (50+) Source: Data from Sánchez and Senderowitsch 2011. Although significant investments have Consider, for instance, the evidence emerging been put forward to improve coverage of from evaluations of conditional cash trans- services, quality issues may lie behind the fers (CCTs), reviewed in chapter 3. Although opting out of the middle and upper classes. cash transfers have improved the lives of poor THE MIDDLE CLASS AND THE SOCIAL CONTRAC T IN LATIN AMERICA 177 people and positively affected the schooling Study [SERCE] assessments). The “bounds” of beneficiaries, the impact on educational in figure 6.6 attempt to correct for the fact achievements remains limited, and gains in that, in some countries, dropout rates remain employment, wages, and intergenerational significant and may bias the observed esti- occupational mobility may not be sufficient mates upward (see focus note 3.1 at the end to break, by themselves, the intergenera- of chapter 3). tional cycle of poverty (Rodríguez-Oreggia Figure 6.6 shows that there is almost as and Freije 2012). The latter result (or lack much variation across classes as across coun- thereof), however, is not a flaw of the CCT tries: for instance, children from poor and programs themselves, which have achieved vulnerable families in Chile (the second-best- their purpose—increasing school enroll- performing country in the sample after Costa ment and attainment. Complementary poli- Rica) appear to perform as poorly as children cies related to quality on the supply side and from the middle and upper classes in Panama employment generation seem to be missing in (the fifth-worst-performing country). And the many cases (Fiszbein and Schady 2009). difference between test scores of poor and I n education, for instance, despite middle-class children in Uruguay (the third- improved attendance from the lower quin- best-performing country) is almost as wide as tiles, there remain marked differences in the difference between test scores of middle- learning achievements across classes. Fig- class children in Uruguay and Ecuador (the ure 6.6 shows reading test scores of sixth- second-worst-performing country). Unless grade children by income group (based on quality is improved, greater demand for ser- the nationally representative 2006 Second vices will serve little to improve mobility and Regional Comparative and Explanatory the cohesiveness of the social contract. FIGURE 6.6 Sixth-grade reading test scores, by income group, selected Latin American countries, 2006 600 550 500 Test scores 450 400 350 300 a lic r ala ay a ua ru r a a zil y ile ca ico do do ua m tin bi Pe a ub Ri gu Ch ag em m na Br ua lva ex ug n sta ep ra r lo ge Pa ca Ec M at Ur Sa Pa Co nR Co Ar Ni Gu El ica in m Do Poor Vulnerable Middle class + Median TS Source: Data from SERCE 2006 and household surveys. Note: SERCE = Second Regional Comparative and Explanatory Study. TS = test score. The bounds stem from estimated biases caused by missing observations. “Poor” = per capita income of US$0–US$4 per day. “Vulnerable” = per capita income of US$4–US$10 per day. “Middle Class” = per capita income of US$10– US$50 per day. NLE = State of Nuevo Leon, Mexico. Poverty lines and incomes are expressed in 2005 US$ PPP per day. PPP = purchasing power parity. a. For Mexico, the data is from the state of Nuevo León. 178 THE MIDDLE CLASS AND THE SOCIAL CONTRAC T IN LATIN AMERICA However, improving quality is not an easy of more inclusive programs financed out of task. And apart from a few countries—such general taxation. The lower the revenues, the as Brazil, where revenues have been histori- more likely that expansion of a program must cally high, or Argentina, which achieved a come at the expense of another one, gener- dramatic increase in revenues—fiscal rev- ating tensions across groups and classes that enues tend to be low in the region, as shown may undermine the social contract. in figure 6.7. In many cases, low revenues do The good news is that low revenues stem not stem from low tax rates (according to the mostly from a compliance challenge. Tech- 2009 U.S. Agency for International Develop- nically, it would thus be possible to increase ment [USAID] Fiscal Reform and Economic revenues through administrative reforms. Governance Project, the top marginal tax Data from USAID’s Fiscal Reform and Eco- rates of personal income in Colombia and nomic Governance Project suggest that in Chile, for instance, are close to Organisation Ecuador, for instance, value added tax (VAT) for Economic Co-operation and Develop- receipts out of each VAT percentage point ment [OECD] levels) but rather from evasion increased from 0.22 percent of GDP in 2004 and narrow tax bases because many firms to 0.53 percent in 2009, mostly thanks to and workers operate informally. administrative reforms. Low fiscal revenues limit the ability of gov- The main challenge in raising revenues, ernments to improve quality of services. They however, may not be technical. To enforce also make it more difficult to expand coverage taxation, there must be the political will FIGURE 6.7 Tax revenues by type, selected Latin American and other countries, 1990–2010 50 45 40 35 Percentage of GDP 30 25 20 15 10 5 0 20 0 20 0 19 0 20 0 20 0 19 0 90 20 0 19 0 20 0 20 0 10 20 0 00 19 0 20 0 20 0 10 20 0 20 0 19 0 90 20 0 10 20 0 20 0 19 0 90 20 0 10 20 0 20 0 19 0 20 0 20 0 10 9 0 1 9 0 1 0 1 9 0 9 1 9 0 9 0 1 0 9 0 1 0 9 0 1 9 0 19 20 19 20 19 20 19 20 19 Guatemala Venezuela, RB Honduras Panama Chile United States Turkey Poland Argentina Brazil OECD Hungary Direct tax revenues Indirect tax revenues Social contributions Other taxes Sources: OECD.Stat (stats.oecd.org) and ECLACSTAT (www.eclac.org) databases. Note: ECLAC = Economic Commission for Latin America and the Caribbean; OECD = Organisation for Economic Co-operation and Development. THE MIDDLE CLASS AND THE SOCIAL CONTRAC T IN LATIN AMERICA 179 to do so, and unless the middle and upper Notes classes (which contribute the largest share of revenues) benefit from the increase in spend- 1. See, among many others, Weber (1905), Adel- ing, it may be difficult to gain their support man and Morris (1967), and Landes (1998) for higher taxation. for Western Europe, and Pike (1963), Parker (1998), Barr-Melej (2001), and Adamovsky We end the section—and the chapter—on (2009) for Latin America. a positive note. We see two ways out of this 2. These arguments have mostly been studied vicious cycle in which some countries may be in the context of the literature on inequality trapped: and growth (Banerjee and Newman 1993; Galor and Zeira 1993), but they also apply to • First, it is not all about money. In many the middle classes. See also Galor and Moav countries, the design of social policies (2004), Voitchovsky (2005), and Foellmi and presents serious flaws, and redressing Oechslin (2008). institutional and individual incentives in a fl at budgetary environment may go a long way toward improving quality. References A recent study about teaching practices in the region, for instance, suggests that Adamovski, Ezequiel. 2009. Historia de la Clase learning achievements can be raised by Media Argentina: Apogeo y Decadencia providing the right incentives to teachers de una Ilusión, 1919–2003. Buenos Aires: to improve their mastery of content and Planeta. Acemoglu, Daron, Simon Johnson, James A. Rob- teaching practices (Bruns, Evans, and inson, and Pierre Yared. 2008. “Income and Luque 2012). In Mexico, a low-cost inter- Democracy.” American Economic Review 98 vention to empower parents in the man- (3): 808–42. agement of schools in disadvantaged rural ———. 2009. “Reevaluating the Modernization areas has had an impact on grade failure Hypothesis.” Journal of Monetary Economics and grade repetition (Gertler, Patrinos, 56 (8): 1043–58. and Rubio-Codina 2012). ADB (Asian Development Bank). 2010. Annual • Second, the recent boom in commodity Report 2010. Manila: ADB. prices, coupled with new oil and com- Adelman, Irma, and Cynthia T. Morris. 1967. modities discoveries, and improved mac- Society, Politics, and Economic Development: roeconomic management has given to A Quantitative Approach. Baltimore, MD: Johns Hopkins University Press. many countries the fiscal space necessary AfDB (African Development Bank). 2011. The to invest in the quality of services with- Africa Competitiveness Report 2011. Geneva: out engaging into a zero-sum competition The World Economic Forum. for a limited pool of resources between Alesina, Alberto, and Roberto Perotti. 1996. the poorer and the wealthier segments of “Income Distribution, Political Instability and society. Investment.” European Economic Review 40 (6): 1203–28. If the middle classes are more pragmatic, Alesina, Alberto, and Dani Rodrik. 1994. “Dis- rather than particularly value-oriented, as tributive Politics and Economic Growth.” this chapter suggests, building coalitions Quarterly Journal of Economics 109 (2): around the right policies may require less 465–90. Amoranto, Glenita, Natalie Chun, and Anil Deo- of a normative discussion and more of an lalikar. 2010. “Who Are the Middle Class and effort to design the right incentive-compat- What Values Do They Hold? Evidence from ible political platform. In any case, how to the World Values Survey.” Economics Work- improve services and achieve greater buy-in ing Paper 229, Asian Development Bank, of the wealthier segments of society is likely Manila. to remain at the center of the social policy Antón, Arturo, Fausto Hernández, and San- debate for the foreseeable future. tiago Levy. 2012. “The End of Informality in 180 THE MIDDLE CLASS AND THE SOCIAL CONTRAC T IN LATIN AMERICA Mexico? Fiscal Reform for Universal Social Data.” Journal of Monetary Economics 34 (2): Insurance.” Working paper, Centro de Inves- 143–73. tigación y Docencia Económicas, Mexico City; Birdsall, Nancy. 2010. “The (Indispensable) Inter-American Development Bank, Washing- Middle Class in Developing Countries; Or, the ton, DC. Rich and the Rest, Not the Poor and the Rest.” Arellano, Manuel, and Stephen Bond. 1991. Working Paper 207, Center for Global Devel- “Some Tests of Specification for Panel Data: opment, Washington, DC. Monte Carlo Evidence and an Application to Bruns, Barbara, Dave Evans, and Javier Luque. Employment Equations.” Review of Economic 2012. Building Better Teachers in Latin Studies, 58 (2): 277–97. America and the Caribbean. Washington, DC: Arellano, Manuel, and Olympia Bover. 1995. World Bank. “Another Look at the Instrumental-Variable Cárdenas, Mauricio, Homi Kharas, and Camila Estimation of Error-Components Models.” Henao. 2011. “Latin America’s Global Middle Journal of Econometrics 68 (1): 29–52. Class.” Working Paper, Global Economy and Arocena, R., and J. Sutz. 2005. “Latin American Development, Brookings Institution, Washing- Universities: From an Original Revolution to ton, DC. an Uncertain Transition.” Higher Education Doepke, Matthias, and Fabrizio Zilibotti. 2008. 50 (4): 573–92. “Occupational Choice and the Spirit of Capi- Banerjee, Abhijit V., and Esther Duflo. 2003. talism.” Quarterly Journal of Economics 123 “Inequality and Growth: What Can the Data (2): 747–93. Say?” Journal of Economic Growth 8 (3): Downs, Anthony. 1957. “An Economic Theory of 267–99. Political Action in a Democracy.” Journal of Banerjee, Abhijit V., and Andrew F. Newman. Political Economy 65 (2): 135–50. 1993. “Occupational Choice and the Process Easterly, William. 2001. “The Middle Class Con- of Development.” Journal of Political Econ- sensus and Economic Development.” Journal omy 101 (2): 274–98. of Economic Growth 6 (4): 317–35. Barr-Melej, Patrick. 2001. Reforming Chile: Cul- Easterly, William, Jozef Ritzen, and Michael tural Politics, Nationalism and the Rise of the Woolcock. 2006. “Social Cohesion, Institu- Middle Class. Chapel Hill: University of North tions, and Growth.” Economics and Politics Carolina Press. 18 (2): 103–20. Barro, Robert J. 1999. “Determinants of Democ- ECLACSTAT (database). Statistics of the Eco- racy.” Journal of Political Economy 107 (6): nomic Commission for Latin America and the 158–83. Caribbean (ECLAC; in Spanish, CEPAL) of ———. 2000. “Inequality and Growth in a Panel the United Nations. http://www.eclac.org/esta of Countries.” Journal of Economic Growth disticas/default.asp. 5 (1): 5–32. Epstein, David, Robert Bates, Jack A. Goldstone, ———. 2008. “Inequality and Growth Revisited.” Ida Kristensen, and Sharyn O’Halloran. 2006. Working Paper on Regional Economic Integra- “Democratic Transitions.” American Journal tion 11, Asian Development Bank, Manila. of Political Science 50 (3): 551–69. Benhabib, Jess, Alejandro Corvalan, and Mark Esteban, Joan, and Debraj Ray. 2008. “Polariza- M. Spiegel. 2011. “Reestablishing the Income- tion, Fractionalization and Confl ict.” Journal Democracy Nexus.” Working Paper 16832, of Peace Research 45 (2): 163–82. National Bureau of Economic Research, Cam- Esteban, Joan, and Gerald Schneider. 2008. bridge, MA. “Polarization and Conflict: Theoretical and Benhabib, Jess, and Adam Przeworski. 2006. Empirical Issues.” Journal of Peace Research “The Political Economy of Redistribution 45 (2): 131–41. under Democracy.” Economic Theory 29 (2): Ferroni, Marco, Mercedes Mateo, and Mark 271–90. Payne. 2008. “Development under Conditions Benhabib, Jess, and Aldo Rustichini. 1996. of Inequality and Distrust.” Discussion paper “Social Conflict and Growth.” Journal of Eco- 777, International Food Policy Research Insti- nomic Growth 1 (1): 125–42. tute, Washington, DC. Benhabib, Jess, and Mark M. Spiegel. 1994. “The Fischer, Justina A. V., and Benno Torgler. 2007. Role of Human Capital in Economic Develop- “Social Capital and Relative Income Concerns: ment: Evidence from Aggregate Cross-Country Evidence from 26 Countries.” Working Paper THE MIDDLE CLASS AND THE SOCIAL CONTRAC T IN LATIN AMERICA 181 Series, Berkeley Olin Program in Law & Eco- Institutional Reforms?” Policy Research nomics, University of California–Berkeley. Working Paper 6015, World Bank, Washing- Fiszbein, Ariel, and Norbert Schady. 2009. Con- ton, DC. ditional Cash Transfers: Reducing Present and López-Calva, Luis F., and Nora Lustig, eds. 2010. Future Poverty. Washington, DC: World Bank. Declining Inequality in Latin America: A Foellmi, Reto, and Manuel Oechslin. 2008. Decade of Progress? Washington, DC: Brook- “Why Progressive Redistribution Can Hurt the ings Institution Press. Poor.” Journal of Public Economics 92 (3–4): López-Calva, Luis F., Jamele Rigolini, and Flor- 738–47. encia Torche. 2011. “Is There Such a Thing as Forbes, Kristin J. 2000. “A Reassessment of the Middle Class Values? Class Differences, Val- Relationship between Inequality and Growth.” ues, and Political Orientations in Latin Amer- American Economic Review 90 (4): 869–87. ica.” Policy Research Working Paper 5874, Fukuyama, Francis. 2012. “The Politics of Latin World Bank, Washington, DC. America’s New Middle Class.” Remarks at Lustig, Nora. 2011. “Fiscal Policy, ‘Fiscal Mobil- 2012 Sol M. Linowitz Forum, The Inter-Amer- ity,’ the Poor, the Vulnerable and the Middle ican Dialogue, Washington, DC, June 8–9. Class in Latin America.” Background paper Galor, Oded, and Omer Moav. 2004. “From prepared for this volume, The Inter-American Physical to Human Capital Accumulation: Dialogue, Washington, DC; Tulane University, Inequality and the Process of Development.” New Orleans. Review of Economic Studies 71 (4): 1001–26. Lustig, N., A. Mizala, and G. E. Silva. 2012. Galor, Oded, and Joseph Zeira. 1993. “Income “¡Basta YA! Chilean Students Say ‘Enough’.” Distribution and Macroeconomics.” Review In The Occupy Handbook, ed. J. Byrne. Back of Economic Studies 60 (1): 35–52. Bay, MA: Little, Brown & Company. Gertler, Paul, Harry Patrinos, and Marta Rubio- Matsuyama, Kiminori. 2002. “The Rise of Mass Codina. 2012. “Empowering Parents to Consumption Societies.” Journal of Political Improve Education: Evidence from Rural Economy 110 (5): 1035–70. Mexico.” Journal of Development Economics Meltzer, Allan H., and Scott F. Richard. 1981. “A 99 (1): 68–79. Rational Theory of the Size of Government.” Glaeser, Edward L., Rafael La Porta, Florencio Journal of Political Economy 89 (5): 914–27. López-de-Silanes, and Andrei Shleifer. 2004. Murphy, Kevin M., Andrei Shleifer, and Robert “Do Institutions Cause Growth?” Journal of W. Vishny. 1989. “Income Distribution, Mar- Economic Growth 9 (3): 271–303. ket Size, and Industrialization.” Quarterly Kaplan, David S., and Santiago Levy. 2012. “The Journal of Economics 104 (3): 537–64. Evolution of Social Security Systems in Latin Nun, José. 1967. “The Middle-Class Military America .” Working paper, Inter-American Coup.” In The Politics of Conformity in Latin Development Bank, Washington, DC. America, ed. Claudio Veliz, 66-118. London: Kharas, Homi. 2010. “The Emerging Middle Oxford University Press. Class in Developing Countries.” Working OECD (Organisation for Economic Co-operation Paper 285, Development Centre, Organisation and Development). 2011. Latin American Eco- for Economic Co-operation and Development, nomic Outlook 2011: How Middle-Class Is Paris. Latin America? Paris: OECD. Landes, David S. 1998. The Wealth and Poverty OECD.Stat (database). Data and metadata for of Nations. New York, London: W. W. Norton Organisation for Economic Co-operation & Company. and Development countries and selected non- Levy, Santiago. 2008. Good Intentions, Bad Out- member economies. stats.oecd.org. comes: Social Policy, Informality, and Eco- Parker, Danny S. 1998. The Idea of the Middle nomic Growth in Mexico. Washington, DC: Class: White-Collar Workers and Peruvian Brookings Institution Press. Society, 1900–1950. University Park: Pennsyl- Lipset, Seymour M. 1959. “Some Social Requi- vania State University Press. sites of Democracy: Economic Development Perotti, Roberto. 1996. “Growth, Income Distri- and Political Legitimacy.” American Political bution, and Democracy: What the Data Say.” Science Review 53 (1): 69–105. Journal of Economic Growth 1 (2): 149–87. Loayza, Norman, Jamele Rigolini, and Gonzalo Persson, Torsten, and Guido Tabellini. 1991. “Is Llorente. 2012. “Do Middle Classes Bring Inequality Harmful for Growth? Theory and 182 THE MIDDLE CLASS AND THE SOCIAL CONTRAC T IN LATIN AMERICA Evidence.” Discussion Paper 581, Centre for Trust, and Weak Collective Action.” Policy Economic Policy Research, London. Research Working Paper 6049, World Bank, Pike, Fredrick B. 1963. “Aspects of Class Rela- Washington, DC. tions in Chile, 1850–1960.” Hispanic Ameri- SEDLAC (Socio-Economic Database for Latin can Historical Review 43 (1): 14–33. America and the Caribbean). 2011. Center PovcalNet. Online poverty analysis tool. World for Distributive, Labor and Social Studies, Bank, Washington, DC. http://iresearch.world Argentina, and World Bank, Washington, DC. bank.org/povcalnet. http://sedlac.econo.unlp.edu.ar/eng. PRC (Pew Research Center). 2008. Inside the SERCE (Segundo Estudio Regional Comparativo Middle Class: Bad Times Hit the Good Life. y Explicativo (UNESCO)). Washington, DC: PRC. Torche, Florencia. 2009. “Sociological and Eco- Przeworksi, Adam, Michael E. Alvarez, Jose nomic Approaches to the Intergenerational A. Cheibub, and Fernando Limongi. 2000. Transmission of Inequality in Latin America.” Democracy and Development: Political Insti- Working Paper HD-09-2009-UNDP, United tutions and Well-Being in the World: 1950– Nations Development Programme, New York. 1990. New York: Cambridge University Press. UNESCO (United Nations Educational, Sci- Reinstein, David, and Juan M. Cayo. 2010. “El entific, and Cultural Organization). 2008. Sector Eléctrico.” In República Dominicana: Trends in Higher Education in Latin America de la crisis financiera internacional al cre- and the Caribbean , ed. Lúcia Gazzola and cimiento para todos, ed. Roby Senderowitsch Axel Didriksson. Bogotá: Panamericana. and Yvonne M. Tsikata, 77–87. Dominican Voitchovsky, Sarah. 2005. “Does the Profile Republic: World Bank. of Income Inequality Matter for Economic Ribe, Helena, David Robalino, and Ian Walker. Growth? Distinguishing Between the Effects 2010. From Right to Reality: Achieving Social of Inequality in Different Parts of the Income Protection for All in Latin America and the Distribution.” Journal of Economic Growth Caribbean. Washington, DC: World Bank. 10 (3): 273–96. Roberts, Kevin, W.S., 1977. “Voting Over Income Weber, Max. (1905) 2003. The Protestant Ethic Tax Schedules.” Journal of Public Economics and the Spirit of Capitalism. Mineola, NY: vol 8 (3): 329–40. Dover. Rodríguez-Oreggia, Eduardo, and Samuel Freije. World Bank. 2010. Did Latin America Learn to 2012. “Long-Term Impact of a Cash-Transfers Shield Its Poor from Economic Shocks? Wash- Program on Labor Outcomes of the Rural ington, DC: World Bank. Youth.” Working Paper 230, Center for Inter- Wright, Erik O. 2005. Approaches to Class Anal- national Development, Harvard University, ysis. Cambridge, U.K.: Cambridge University Cambridge, MA. Press. Sánchez, Miguel E., and Roby Senderowitsch. You, Jong-Sung, and Sanjeev Khagram. 2005. “A 2011. “The Political Economy of the Middle Comparative Study of Inequality and Corrup- Class in the Dominican Republic: Individual- tion.” American Sociological Review 70 (1): ization of Public Goods, Lack of Institutional 136–57. ECO-AUDIT Environmental Benefits Statement The World Bank is committed to preserving Saved: endangered forests and natural resources. • 6 trees The Office of the Publisher has chosen to • 3 million BTUs of print the Economic Mobility and the Rise of total energy the Latin American Middle Class on recy- • 545 lbs. of net cled paper with 25 percent postconsumer greenhouse gases fiber in accordance with the recommended • 2,954 gallons of standards for paper usage set by the Green waste water Press Initiative, a nonprofit program sup- • 198 lbs. of porting publishers in using fiber that is not solid waste sourced from endangered forests. For more information, visit www.greenpressinitiative. org.