E Economia 38 1 2011

ESTUDIOS DE ECONOMIA
VOLUMEN 38 N 1 / JUNIO 2011

GUEST EDITOR: DAVID BRAVO (CENTRO DE MICRODATOS, UNIVERSIDAD DE CHILE) SPONSORED BY: INTER-AMERICAN DEVELOPMENT BANK INICIATIVA CIENTFICA MILENIO CENTRO DE MICRODATOS, UNIVERSIDAD DE CHILE
Introduction to the issue David Bravo
ARTCULOS How much might human capital policies affect earnings inequalities and poverty? Jere R. Behrman The impact of income adjustments in the Casen Survey on the measurement of inequality in Chile David Bravo, Jos Valderrama The longer-term effects of human capital enrichment programs on poverty and inequality: Oportunidades in Mexico Douglas McKee, Petra E.Todd Alleviating extreme poverty in Chile: the short term effects of Chile Solidario Emanuela Galasso Evaluating the Chile Solidario program: results using the Chile Solidario panel and the administrative databases Fernando Hoces de la Guardia, Andrs Hojman, Osvaldo Larraaga Top incomes in Chile using 50 years of household surveys: 1957-2007 Claudia Sanhueza, Ricardo Mayer Intergenerational income and educational mobility in urban Chile Javier Nez, Leslie Miranda A cohort analysis of the income distribution in Chile Claudio Sapelli Getting ahead, falling behind and standing still. Income mobility in Chile Rodrigo Castro Examinando la prominente posicin de Chile a nivel mundial en cuanto a desigualdad de ingresos: comparaciones regionales Juan Pablo Valenzuela, Suzanne Duryea Estabilidad en la desigualdad. Chile 1990-2003 Osvaldo Larraaga, Juan Pablo Valenzuela
UNIVERSIDAD DE CHILE FACULTAD DE ECONOMIA Y NEGOCIOS DEPARTAMENTO DE ECONOMIA
U N I V E R S I D A D D E C H I L E F A C U L T A D D E E C O N O M I A Y N E G O C I O S
ESTUDIOSDEECONOMIA
VOLUMEN38-N1 ISSN0304-2758 JUNIO2011
GUESTEDITOR: DAVID BRAVO (CENTRO DE MICRODATOS, UNIVERSIDAD DE CHILE) SPONSOREDBY: INTER-AMERICAN DEVELOPMENT BANK INICIATIVA CIENTFICA MILENIO CENTRO DE MICRODATOS, UNIVERSIDAD DE CHILE
Introduction to the issue David Bravo ARTCULOS How much might human capital policies affect earnings inequalities and poverty? Jere R. Behrman TheimpactofincomeadjustmentsintheCasenSurveyonthemeasurementof inequality in Chile David Bravo, Jos Valderrama Thelonger-termeffectsofhumancapitalenrichmentprogramsonpovertyandinequality: OportunidadesinMexico Douglas McKee, Petra E.Todd AlleviatingextremepovertyinChile:theshorttermeffectsofChileSolidario Emanuela Galasso EvaluatingtheChileSolidarioprogram:resultsusingtheChileSolidariopanelandthe administrative databases Fernando Hoces de la Guardia, Andrs Hojman, Osvaldo Larraaga TopincomesinChileusing50yearsofhouseholdsurveys:1957-2007 Claudia Sanhueza, Ricardo Mayer Intergenerational income and educational mobility in urban Chile Javier Nez, Leslie Miranda A cohort analysis of the income distribution in Chile Claudio Sapelli Gettingahead,fallingbehindandstandingstill.IncomemobilityinChile Rodrigo Castro ExaminandolaprominenteposicindeChileanivelmundialencuantoadesigualdadde ingresos:comparacionesregionales Juan Pablo Valenzuela, Suzanne Duryea Estabilidadenladesigualdad.Chile1990-2003 Osvaldo Larraaga, Juan Pablo Valenzuela
43
67 101
129 169 195 223 243
259 295
Departamento de Economa
Estudios de to the issue / David Bravo IntroductionEconoma. Vol. 38 - N 1, Junio 2011. Pgs. 5-7
Introduction to the issue

David Bravo*
Chile has usually been considered one of the outstanding countries in the World in terms of its record in reducing poverty. At the same time, income inequality in Chile remains high by international standards; in fact, cross-country comparisons places Chile amongst those countries with the highest dispersion of monetary income (World Bank 1997, 2001; Bravo and Contreras, 2004). This volume originated from a Workshop on Income Inequality sponsored by the Centro de Microdatos and the Inter-American Development Bank in December 2006. Some of the papers and authors that were part of this seminar sent updated versions of their pieces for a refereeing process. The papers of this volume cover a wide range of poverty and inequality issues. Jere Behrman asks how much might Human Capital policies affect earnings inequalities and poverty? It reviews some recent benefit/cost estimations for human capital intervention in Latin America and the Caribbean, suggesting some investments in which the returns appear quite high. The contribution of schooling attainment targeted to the poor in reducing poverty and income inequality is also reported and illustrated using Chilean data. Alternative simulations suggest to Behrman significant impacts of well-targeted increases in schooling attainment on reducing poverty and inequality. How does the income distribution of Chile compare with other countries? On this topic David Bravo and Jos Valderrama show that Chile is one of the few countries that adjusts the information obtained from household surveys to make the figures compatible with National Accounts. This paper shows that this adjustment leads to important changes in the top-end of the distribution and to an overestimation in the main inequality indicators in Chile. Chile looks more unequal in international relative terms due to this adjustment. Bravo and Valderrama use Peru and Chile to illustrate their point. Juan Pablo Valenzuela and Suzanne Duryea compare the income distributions of Chile and Uruguay. They show that Chile in 2003 had a Gini Coefficient 8.5 points higher than Uruguay. Using microsimulations, they conclude that most of the difference comes from the wealthier household, particularly those belonging to the top 2%. Another fraction of the differences is explained by differences in returns to higher education in both countries. They also emphasize that the national account adjustment in Chile is overestimating the gap between the Gini coefficients of those countries.
Centro de Microdatos, Departamento de Economa, Universidad de Chile. E-mail: dbravo@ econ.uchile.cl
Estudios de Economa, Vol. 38 - N 1
Three papers look at strategies for alleviating poverty. Douglas McKee and Petra Todd analyze the Mexican conditional cash transfer program Oportunidades. They study the programs potential longer-term consequences on the poverty and inequality of disadvantaged children. Using nonparametric simulations of earnings distributions, with and without the program, they are able to predict that Oportunidades will increase future mean earnings but have only modest effects on poverty rates and earnings inequality. On the other hand, Galasso and Hoces, Hojman and Larraaga evaluate Chile Solidario the Chilean antipoverty program. Emanuela Galasso finds that the program tends to increase significantly the take-up of cash assistance programs and of social programs for housing and employment, and to improve education and health outcomes for participating households. However, she doesnt find evidence on improved employment or outcomes in the short term. She concludes providing evidence of the key role that the psycho-social support has had by increasing awareness of social services in the community as well as households orientation towards the future. Finally Hoces, Hojman and Larraaga offer a critical view on the reliability of the official results due to data shortcomings and, particularly, the lack of baseline data. Using a huge administrative data set they find small but clearly positive effects for several variables as the number of workers in the family, the percentage of workers in the family and the employment of the head of the household. Javier Nez and Leslie Miranda provide evidence on the degree and patterns of intergenerational income and educational mobility for Greater Santiago. They find intergenerational income elasticities of 0.52 to 0.54, which stands as fairly high with respect to comparable international evidence. They also find that intergenerational educational mobility is lower for the younger cohorts and a higher degree of intergenerational persistence of income at the two extremes of the income distribution, even if this is more accentuated at the top. Claudia Sanhueza and Ricardo Mayer focus their analysis on the top incomes in Chile for the period 1957-2007. They corroborate that top incomes concentration is countercyclical in the short run. Over the 50 years period they find that top incomes concentration follows an inverted U-shape, peaking at the end of the 80s. The authors observe important changes in the composition of top income groups related to greater relative importance of women, employees and college schooling levels since the 80s. Claudio Sapelli gives a different look at the data in his paper. Constructing a synthetic panel from 50 years of cross-section household surveys for Greater Santiago, he decomposes the evolution of the estimated distributions into age, year and cohort effects. The cohort effects found shows a period where inequality increases to then decrease. The author states that the rise can be explained by variables associated with education, while the fall appears to be the consequence of a flattening in the income-age profile and hence a reduction in the returns to experience. Rodrigo Castro analyses household income mobility in Chile between 1996 and 2001. He finds that mobility has been quite high compared to international standards. His analysis shows that moving from unemployment to employment significantly increases the probability of moving up and decreases the probability of moving down. Vocational education is promoting to move up on the relative
Introduction to the issue / David Bravo
income scale and it is protecting movement down. An important result is that high-school education decreases the probability of degradation. Finally, Osvaldo Larraaga and Juan Pablo Valenzuela examine why the income distribution in Chile as a whole did not change between 1990 and 2003. Using micro-simulations of income distribution they analyze the role of returns, participation rates, occupational choices, schooling endowments, subsidies, pensions and household size. They find that the inertia shown by inequality reflects the interplay of factors that cancel each other out, others that operate slowly over time, and the emergence of new developments that affect the distribution. I hope that from the range of topics covered and methodologies used in the articles of this volume, progress will continue to be made. I would like to thank the Millennium Science Initiative who made possible the publication of this volume through its funding to the Microdata Center (Project P07S-023-F). We owe a special debt of gratitude to Cintia Guimaraes and Suzanne Duryea of the Interamerican Development Bank, for all their support and efforts in making the conference a success and now this special issue a reality. Claudia Sanhueza was also key for this success back in 2006. Finally, I am grateful to the editor of Estudios de Economa; to the authors who provided their papers; and to Milena Essus and Marizza Espinoza from the Microdata Center who worked hard on the edition of this issue. References Bravo, D. and D. Contreras (2004). La distribucin del ingreso en Chile 19901996: anlisis del impacto del mercado de trabajo y las polticas sociales, in Banco Interamericano de Desarrollo, Reformas y Equidad Social en Amrica Latina y el Caribe, april. World Bank (1997). Chile: Poverty and Income Distribution in a High-Growth Economy 1987-1995, Report N 16377-CH, November 25, Washington DC. World Bank (2001). Chile: Poverty and Income Distribution in a High-Growth Economy. The case of Chile 1987-1998, Report N 22037-CH, August, Washington DC.
Estudios de Economa. Vol. 38 - policies Jere Pgs. 9-41 How much might human capital N 1, Junio/ 2011.R. Behrman
How much might human capital policies affect earnings inequalities and poverty?
Cunto afectan las polticas en capital humano a la desigualdad del ingreso y a la pobreza?
Jere R. Behrman* Abstract Economic inequality and poverty have persisted in Latin America despite important changes in political and policy regimes. This paper explores the relationship between various human capital programs aimed to reduced poverty and how improvements of those in poverty in the left tail of the earning income distribution are likely to reduce inequality. First it reviews some recent benefit/cost estimates for human capital intervention in LAC, suggesting some investments in which the returns appear quite high. Then it turns over to how much increases in schooling attainment targeted to the poor would reduce poverty and income inequality. This is illustrated empirically using the 2004 Chilean Social Protection Survey data. Alternative simulations suggest significant impacts of well targeted increases in schooling attainment on reducing poverty and inequality. Key words: Inequality, Poverty, Human capital.. Resumen La desigualdad econmica y la pobreza han persistido en Amrica Latina a pesar de importantes cambios en los regmenes polticos y en la poltica social. Este artculo explora la relacin entre diversos programas de capital humano
Jere R. Behrman is the William R. Kenan, Jr. Professor of Economics and Research Associate of the Population Studies Center, McNeil 160, 3718 Locust Walk, University of Pennsylvania, Philadelphia, PA 19104-6297, USA; telephone 1 215 898 7704, fax 1 215 898 2124, e-mail: jbehrman@econ.sas.upenn.edu. He acknowledges research support from 1NIA grant AG023774-01; awards from the Population Aging Research Center (PARC) of the University of Pennsylvania Population Studies Center, and the Boettner Center for Pensions and Retirement Security and Pension Research Council at the University of Pennsylvania. He thanks Carlos Eduardo Velez, Ricardo Ffrench-Davis and Emanuela Galasso for helpful comments at the workshop and Harold Alderman, John Hoddinott, Jim Knowles and Petra Todd for helpful collaboration on related projects on which this study builds in part.
10
destinados a reducir la pobreza y cmo una reduccin de la pobreza tiende a reducir la desigualdad. En primer lugar, se presentan estimaciones de costo/ beneficio recientes de distintas intervenciones en capital humano en Amrica Latina, sugiriendo que en algunas de estas inversiones los rendimientos parecen bastante altos. Luego, se profundiza en cmo un aumento en la escolaridad de las personas en situacin de pobreza reducira los niveles de pobreza y la desigualdad de ingresos. Para ilustrar empricamente este punto se utilizan los datos de la ronda 2004 de la Encuesta de Proteccin Social de Chile. Simulaciones alternativas sugieren un impacto significativo del aumento en los aos de escolaridad, focalizados en personas pobres, en la reduccin de la pobreza y la desigualdad. Palabras clave: Desigualdad, Pobreza, Capital humano. JEL Classification: I24, I3, J24.
As is well-known, economic inequality long has been relatively high in most of Latin America and the Caribbean (LAC) in comparison with most other areas of the world. These inequalities generally have persisted despite some variations due to changes in political and policy regimes and markets and significant policy innovations in recent decades in LAC countries from Chile to Mexico. At the start of the 1970s, for example, LAC income inequality measured by the income Gini coefficient ranged from a low of 0.43 in Argentina and Uruguay, to a high of 0.62 in Honduras (Table 1). Average income inequality in the region as a whole for this period was 0.53 compared with 0.36 for high-income OECD countries, and 0.43 for East Asia. Schooling inequality was also high in Latin America in the early 1970s, with the schooling attainment Gini coefficients averaging 0.46. The countries with the highest schooling inequality were also some of the poorest in the region (Guatemala, Honduras, Nicaragua and El Salvador). In contrast, Argentina and Chile were well below the LAC average on both measures. The average LAC income Gini coefficient subsequently declined to 0.50 in the early 1980s but increased to about 0.54 in 2000, though the Gini coefficient for schooling attainment stayed at about 0.46 for 1980 and then declined somewhat to about 0.42 in 2000. Conceptually distinct from the question about inequality is the question of poverty. Most governments profess an interest in reducing poverty and have undertaken programs that purportedly are directed towards doing so. In the short run, transfers to the poor are often thought to be among the more effective ways of reducing short-run poverty, though such transfers might be conditional on activities such as work in employment programs in order to screen beneficiaries or to limit negative effects on labor supplies. In the longer run, investments in human resources of the poor are usually thought to be among the most effective ways of reducing poverty because the basic resource of the poor is their time and
TABLE 1
LAC INCOME AND SCHOOLING GINI COEFFICIENTS FOR EARLy 1970s, EARLy 1980s AND AROUND 2000
Early 1970s Income Gini Coefficient 1980-1985 Income Gini Coefficient 1998-2000 0.52 0.63 0.61 0.60 0.57 0.50 0.56 0.54 0.60 0.53 0.54 0.56 0.52 0.51 0.45 0.46 0.46 0.54 0.41 n/a 0.57 0.53 0.53 0.48 0.44 0.53 0.54 0.57 0.50 n/a 0.45 0.51 0.44 0.48 0.50 0.29 0.52 0.48 0.37 0.47 0.40 0.39 0.49 0.63 0.57 0.50 0.62 0.38 0.41 0.36 0.43 Schooling Gini Coefficient (people over 15) 1980
Early 1980s
1998-2000 Schooling Gini Coefficient (people over 15) 2000 0.27 0.47 0.43 0.37 0.48 0.42 0.43 0.53 0.59 0.43 0.36 0.52 0.36 0.36 0.35 0.38 0.42
How much might human capital policies / Jere R. Behrman
Schooling Gini Income Gini Coefficient Coefficient 1970 (people over 15) 1970 0.29 0.52 0.48 0.37 0.47 0.40 0.39 0.49 0.63 0.57 0.50 0.62 0.38 0.41 0.36 0.43 0.46
Argentina Bolivia Brazil Chile Colombia Costa Rica Ecuador El Salvador Guatemala Honduras Mexico Nicaragua Paraguay Peru Uruguay Venezuela
0.49 0.46 0.61 0.50 0.57 0.52 n/a 0.47 0.54 0.62 0.54 n/a n/a 0.55 0.43 0.55
Group mean (unweighted)
0.53
All income Gini coefficient estimates are population based. For 1970 the data for Argentina, Mexico and Nicaragua are for 1960 and for Venezuela 1961. Source is Behrman, Birdsall and Pettersson (2006) based on data from WDI (2005), WIID2a (2005), Thomas, Wang and Fan (2001) and Deininger and Squire (1996).
11
12
enhancing their human resources is likely to increase the return to the use of their time. In recent years in LAC, for example, there has been considerable expansion of conditional cash transfer programs based on the Mexican PROGRESA/ Oportunidades example in which the transfers are conditional on indicators of investments in human capital, most notably in schooling attainment (but also on investments in health and nutrition)1. Feasible improvements in the positions of those in poverty in the left tail of the earnings income distribution, of course, are likely to reduce inequality. This paper first considers some recent benefit/cost estimates for human capital interventions in LAC (Section 1). These suggest that there are some human capital investments in which the estimated returns appear quite high. Then the paper turns to rough calculations of how much increases in schooling attainment the most emphasized form of human capital investments targeted towards the poorer members of society will reduce poverty and income inequalities (Section 2). To illustrate the empirical orders of magnitude, earnings and wage rate function estimates and the distribution of earnings from the 2004 Chilean Social Protection Survey are used. Section 3 then concludes regarding the implications of the previous two sections for the possibility of using human capital investments to reduce poverty and inequality in LAC. 1. Benefit/Cost Estimates for Human Capital Investments in Latin America There is a surprising paucity of good estimates of benefit/cost estimates for human capital interventions in Latin America and in other developing regions. This is in important part because of the difficulties in making good estimates2. On the benefit side, for example, there often are difficulties in: Obtaining good estimates of important impacts given behavioral choices in the presence of unobserved heterogeneities (so that, for example, schooling is not incorrectly credited in part with the effects of unobserved abilities, motivations and family background that in part cause more schooling and also directly affect outcomes that might be directly affected by schooling) and that impacts may be distributed over many decades over the life cycle; Placing values on all the relevant impacts so that they can be aggregated, which is particularly difficult (e.g., for adverted mortality) but also difficult
There are a number of studies that report positive impact of these programs on schooling, nutrition and other outcomes (e.g., Behrman and Hoddinott (2005); Behrman, Parker and Todd (2009a,b); Behrman, Sengupta and Todd (2005); Fiszbein and Schady (2009); Gertler (2004); Hoddinott and Adato (2009); Rivera et al. (2004); Schultz (2004). For discussion in much more detail of these difficulties and some illustrative estimates of benefit/cost ratios for developing countries, including some in LAC, see Behrman, Alderman and Hoddinott (2004) and Knowles and Behrman (2004, 2005).
13
if projects are large enough to have impact on market prices or if prices are very distorted by market or policy failures; Selecting appropriate discount rates to permit aggregation across time (Table 2 illustrates how much the choice of discounting matters, as well as that impacts of human capital investments depend on life-cycle patterns in survival probabilities); and Distinguishing between private and social (beyond the private) benefits, a distinction that is central for the efficiency motive for policies. TABLE 2
PDV OF $ 1000 RECEIVED WITH VARIOUS LAGS STARTING FROM BIRTH AT ALTERNATIVE DISCOUNT RATES, ADJUSTED FOR SURVIVAL PROBABILITIES*
Discount Rate Lag in years 10 40 60 100
3.00% $ 717 $ 281 $ 134 1
5.00% $ 592 $ 130 $ 42 0
10.00% $ 372 $ 20 $ 3 0
Survival Prob 0.964 0.915 0.787 0.007
Survival probabilities based on WHO Life Tables for Brazil at http://www.who.int/countries.
On the cost side, there often are difficulties in: Identifying the true private and public resource costs (not the governmental budgetary costs), including distortion costs but excluding transfers; Assessing the costs when a project is scaled-up, including not only the possibility of changing marginal costs at a micro level but changing marginal resource costs through impacts on markets; and Distinguishing between private and social costs, again important for efficiency considerations. However there are some recent estimates of benefit/cost ratios for human capital investments in developing countries that attempt to deal with some of these problems but at the same time illustrate some of the problems. Sections 1.1 summarizes selected estimates for educational interventions in LAC and Section 1.2 summarizes selected estimates for nutrition/health interventions with adjustments to place them in a LAC context. 1.1. Benefit/Cost Estimates for Three Recent LAC Education-Related Programs If education means learning, then education is a life-long process and not limited to schooling though at times the literature inappropriately seems to equate education with schooling. Recent estimates of adult cognitive achievement production functions in Guatemala, for example, illustrate that assuming
14
only schooling matters in the production of adult cognitive skills leads to a substantial overestimates of the impact of schooling and underestimates of the importance of pre- and post-schooling experiences (Behrman et al. (2008)). In part for this reason, I here summarize three interventions one pre-school, one related to schooling, and one post-schooling. 1.1.1.Bolivian PIDI (Proyecto Integral de Desarrollo Infantil) Pre-School Program3 Program: PIDI provided daycare, nutritional, and educational services to children between the ages of 6 months and 72 months who lived in poor, predominantly urban areas. The goals were to improve health and early cognitive/social development by providing children with better nutrition, adequate supervision, and stimulating environments. It was hoped that the program also eased the transition to elementary school, improved progression through elementary grades, and raised school performance, all of which were expected to increase post-school productivity. Through PIDI, children attended full-time child care centers located in the homes of women living in low-income areas targeted by the program. These women were given training in child care and loans and grants (up to $ 500) to upgrade facilities in their homes. Each PIDI center had up to 15 children and approximately one staff member per five children, with additional staff provided when there was a larger proportion of infants. The program provided food to supply 70% of the childrens nutritional needs as well as health and nutrition monitoring and educational activity programs. Program benefits: Longitudinal information is available on a sample of program participants, a sample of other children in the same age range living in program areas and a sample of children in the same age range living in roughly comparable areas that did not have the program. This includes information on pre-school child anthropometrics, cognitive development and psychosocial development. However baseline pre-program data on critical child attributes (e.g., anthropometrics) are not available. Based on a dynamic model of investing in children, a range of estimation methods were used to assess the robustness of estimated program impacts under alternative assumptions about the selection into the program standard cross-sectional regressions, propensity score matching with the two samples of children not in the program, and marginal propensity score matching (with the advantage of controlling for unobserved factors that determined program participation) focusing on changes in the outcomes with different program exposures and conditional on child age when entering the program. The estimates show that the program significantly increased cognitive achievement and psychosocial test scores, especially for children who participated in the program for at least seven months. The impact estimates are fairly robust to the use of alternative comparison groups and estimators. Estimates obtained by the marginal matching estimator tend to be larger, particularly at longer durations and for children aged 6-36 months, than those obtained using traditional econometric estimators that impose stronger functional form assumptions.
3
Based on Behrman, Cheng and Todd (2004).
15
To obtain the program benefits, the impacts of these pre-school characteristics need to be translated in the value of their impact on lifetime productivity. Four channels through which the preschool program can affect lifetime earnings are considered: (1) increasing cognitive skills as an adult (conditional on grades completed), which directly affects earnings, (2) increasing physical stature as an adult, which directly affects earnings, (3) increasing the number of grades completed, which directly affects earnings and the age of school completion, and (4) decreasing the age of school completion without changing the number of grades completed. Available data from Bolivia are used for some dimensions of these processes, such as grade repetition, distribution of completed schooling attainment and related earnings. In the absence of estimates from Bolivia on other needed components, estimates from previous studies on other developing countries are used (e.g., Thomas and Strauss (1997) estimate of the relationship between adult earnings and height in Brazil, the Alderman et al. (1996) estimate of the cognitive skills earnings relationship in Pakistan) under the assumption that increases in height and cognitive ability as a child have persistent effects that translate into equiproportional increases as an adult. This last assumption may result in an overstatement of the impacts. On the other hand, in some respects these estimates probably understate the program benefits because (1) they do not incorporate non-labor-market effects such as increased productivity in the production of health and nutrition and (2) they do not incorporate the value of time gained through less time in childcare of the childrens mothers or other caregivers due to the program. Program costs: The program cost was estimated by Ruiz (1996) to have been approximately $ 43 per beneficiary per month, which is substantial in a country where per capita annual GDP at the time was $ 800 in exchange-rate-converted pesos, or $ 2540 in purchasing power parity terms. Approximately 40% of the expenditure went to the nutritional component of the program. This study uses the Ruiz estimate, with an adjustment in alternative estimates of 25% for the distortion cost of raising revenues for the program (e.g., see Knowles and Behrman 2004). Such estimates may overstate somewhat the costs if part of the nutritional supplementation for the children effectively was a transfer that replaced food that the children otherwise would have been provided by their families, as suggested by some previous studies.4 Benefit/cost ratios: Table 3 summarizes the estimated benefit/cost ratios with alternative assumptions concerning discount rates (3% and 5%), schooling levels, and whether distortion costs are included. The included estimates range from 1.4 to 3.7, all of which mean that the program has benefits greater than its costs, but the range indicates considerable sensitivity to the alternative assumptions.
Alternatively, the benefits are underestimated because they in part are reaped by other household members.
16
TABLE 3
BENEFIT/COST RATIOS FOR BOLIVIAN PIDI PRE-SCHOOL PROGRAM* Mean Annual Earnings ($) From Intermediate Secondary Intermediate Secondary * (8) (11) (8) (11) 1224 1422 1224 1422 To 1352 1550 1352 1550 Cost 1394 1394 1743 1743 Discount Rate 3% Benefit 5107 3969 5107 3069 Benefit/ Cost Ratio 3.7 2.9 2.9 2.3 Cost 1301 1301 1626 1626 Discount Rate 5% Benefit 3230 2232 3230 2232 Benefit/ Cost Ratio 2.5 1.7 2.0 1.4
Schooling Level
Assumes that children participate in program for three years starting at age 2. Impact: shortens time to complete schooling by 1 year; increases schooling level by 1 grade; increases cognitive skills by 5%. Costs in first two rows are $ 516 per year based on Ruiz (1996); costs in the second two rows also include a 25% adjustment for distortions due to raising public revenues for the program.
1.1.2.Colombian Programa de Ampliacin de Cobertura de la Educacin Secundaria (PACES) Scholarship Program for Poor Urban Secondary School Students5 Program: The Colombian government created the PACES voucher program for poor secondary schoolage children in 1991 to enable them to attend private secondary schools. By 1996, the program was covering about 100,000 students. At the time the program was launched, enrollment in primary school (covering grades 1-5) was almost universal, while the enrollment rate in secondary school (i.e., covering grades 6-11) was 73% overall and 55% for the poorest quintile of students. The vouchers were designed to cover about one-half of the cost of private secondary schools. They could be renewed as long as students maintained satisfactory academic performance. The vouchers were targeted to children residing in poor urban residential areas and who previously had attended public primary schools. Program benefits: The PACES scholarship program (and scholarship programs generally) might be expected to have the following effects: Increased schooling attainment (however, there may be negative effects via school crowding). Earlier completion of given schooling levels. Improved cognitive achievement for a given level of schooling attainment (assuming that private schools attended by some voucher recipients provided
Based on Angrist et al. (2002) as summarized in Knowles and Behrman (2004).
17
better-quality schooling than that provided by the public schools that would have been attended in the absence of the vouchers). Reduced work effort by other family members (any reduction in work effort by the scholarship recipient is considered to be part of the opportunity cost of schooling). Evaluation of the PACES program was facilitated by the fact that vouchers were initially awarded by lottery in municipalities in which the number of students applying for vouchers exceeded the estimated shortfall in the number of spaces available in public schools that was used by local authorities to decide how many new vouchers to make available in a given school year. This in effect created a natural experiment that made it possible to estimate the effect of the vouchers on the target population much as one could do if the voucher scheme had been conducted as a randomized experiment6, 7. This evaluation did not find any significant impact on enrollment. However, it did find that lottery winners were 15 percent more likely to attend a private school8. After three years, lottery winners also had completed 0.12-0.16 more grades of schooling (primarily due to lower repetition rates) and were about 10 percentage points more likely to have completed the 8th grade at the end of the programs third year. Although the program had no effect on dropout rates, lottery winners scored 0.2 standard deviations higher on standardized tests. In addition, lottery winners reported working 2.2 hours per week less than non-winners and were less likely to be either married or cohabiting as teenagers (however, this last difference affected only about one percent of the sample). The evaluation study estimates that the additional 0.12-0.16 grades of schooling completed by lottery winners would raise their annual incomes by about $ 36-48 per year (based on an estimated rate of return to schooling of 10 percent in Colombia and predicted average annual earnings of $ 3,000).
The implicit assumption is that the excess demand for vouchers occurred randomly across municipalities (and therefore was not associated with other municipality-level factors that might affect education). If, for example, excess demand for vouchers more likely occurred where recent expectations were higher regarding the returns to schooling because of greater economic activity than elsewhere, the estimated impacts may not be applicable to other municipalities with lower expectations regarding the returns to schooling. Unfortunately, the study does not provide information with which to assess whether the excess-demand municipalities were similar to or different from other municipalities. In fact, the evaluation found that only about 90% of lottery winners actually used their vouchers, while about 24% of lottery users received scholarships from other sources. To correct for the effects of such contamination, the evaluation also used lottery success as an instrument to predict whether a given youth received a scholarship. The instrumental variable estimates of the grade completion effect were about 50% greater than the reducedform OLS estimates of being a lottery winner. The estimates of effectiveness used in the published study to obtain the cost-benefit estimates are the reduced-form estimates. In this case, the presumably more accurate instrumental variable estimates arguably should have been used. Because of the reported poor quality in some of the new secondary schools that emerged to serve voucher recipients, eligibility to participate in the program was limited to notfor-profit schools from 1996 on.
18
Program costs: The economic cost of the PACES program includes: Additional schooling costs (some of which may be incurred directly by households, including the opportunity cost of the additional time scholarship students are engaged in schooling-related activities). Program administration costs. Distortionary costs related to the financing of the program (e.g., the deadweight cost of collecting the necessary taxes to finance the program) or to reduced work effort on the part of other household members (due to a possible income effect). The costs of scholarship programs do not include the cost of the scholarships themselves (i.e., the cost of the voucher, in the PACES program), which is a transfer. Table 4 summarizes information on the costs of the PACES program as reported in the evaluation study. The table indicates that the reported social costs of the program (the last column in the table) are at least US$ 195 per lottery winner for three years (after adjusting for different rates of voucher take-up in each year of the program). However, several cost items are not reported, including the deadweight cost of financing the additional governmental expenditure of $ 109 per lottery winner, the possible distortionary cost on adult work effort in lottery-winning households, and the programs administrative costs. TABLE 4
THREE-yEAR COSTS PER LOTTERy WINNER IN THE COLOMBIA PACES PROGRAM (US$)
Item Cost of the vouchers Deadweight cost of government financing Effect of voucher on adult work effort Administrative costs Reduced public expenditure on secondary schooling Additional amount spent on schooling by lottery winners Opportunity cost of student timea Totals
Government 336
Lottery Winners 336
All Households
Society 0
? ? ? 227 236 186 109 86 0
? ? ? 227 236 186 195
Source: Angrist, et al. (2003). a Estimated on the basis of reported reductions in time worked by lottery winners as compared to non-winners.
19
Program benefit/cost estimates: The evaluation does not include a formal benefit/ cost analysis, concluding that the benefits clearly exceed the costs using any plausible discount rate. However Knowles and Behrman (2004) build on the evaluation study results to obtain benefit/cost estimates. A life-cycle approach is used with the following assumptions: The three-year costs of the program include the estimated project costs of $ 195 per lottery winner from Table 4 but also administrative and distortionary costs of $ 32.65 per lottery winner (i.e., 30% of the increase in government expenditure from Table 4)9 and that all program costs occur at ages 13-15 in equal annual installments. Lottery winners complete 0.12 additional grades of schooling, with each additional grade of schooling resulting in a 10% increase in annual earnings of $ 3,000 per year. Annual benefits occur at ages 16 to 60. The discount rate is alternatively 3, 5 or 10% per annum, and benefits and costs are discounted to age 13. Under these assumptions, the benefit-cost ratio is 3.8 with a discount rate of 3%, 2.7 with a discount rate of 5%, and 2.4 with a discount rate of 10%. The discount rate again, as in the pre-school program considered in Section 1.1.1 and the health and nutrition programs considered in Section 1.2, has a strong effect on the estimated benefit/cost ratio because the benefits are spread over an extended period of time. Table 5 presents the estimated benefits and costs at selected ages for illustrative purposes using a discount rate of 5%. The estimated benefits from the increased number of school years completed are presented in column 1, while the programs cost is presented in column 4 (the other columns are discussed below). The calculations show that the benefits are heavily discounted (comparing the last two rows of the table). These estimates of the benefit/cost ratio are conservative, however, not only because they assume the lower limit to the estimated gain in the number of school grades completed (0.12, as compared to the upper limit of 0.16)10 but also because they neglect the other estimated effects of the intervention, i.e., the finding that lottery winners were 10% more likely to have completed the 8th grade at the time of the evaluation and that they scored 0.2 standard deviations higher on standardized tests11.
9
10
11
The 30% estimate for administrative and distortionary costs is probably conservative, based on the discussion in Knowles and Behrman (2004, Section 3.2.3) where it is recommended that an estimate of 20-25% be used as an estimate of the cost of raising additional tax revenue. If the estimated impact of the voucher scheme on the number of school years completed is assumed to be 0.16, instead of 0.12, the estimated benefit/cost ratio increases from 2.7 to 3.6 with a discount rate of 5%. There may be other biases as well that are not explored here. On one hand, the estimate of a 10% rate of return to schooling attainment may be upward-biased due to the failure to control for unobserved endowments (e.g., innate ability, family connections, motivations) in the earnings function estimates. On the other hand, there may be non-labor-market returns that are additional to labor market returns.
20
TABLE 5
CALCULATIONS OF DISCOUNTED BENEFITS AND COSTS (US$) PER LOTTERy WINNER FOR THE COLOMBIA PACES VOUCHER SCHEME
Benefits Earlier completion of schooling Improved test scores 0.00 0.00 0.00 360.00 360.00 360.00 360.00 360.00 0.00 0.00 0.00 0.00 16,200.00 5,803.78 0.00 0.00 303.60 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 303.60 275.37 Cost
Age
Increased number of school grades completed 0.00 0.00 0.00 36.00 36.00 36.00 36.00 36.00 0.00 0.00 0.00 0.00
13 14 15 16 17 18 19-59 60 61 62 63-69 70
75.88 75.88 75.88 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 227.65 216.98
Totals 580.38
1,620.00
Discounted to age 13a
Source: See text. a Using a discount rate of 5%.
21
Considering first the effect of the voucher scheme on the age of completing the 8th grade, it is assumed that the estimated effect implies that lottery winners gain an additional 0.1 year of full-time work at age 15. This additional benefit is displayed in row 3, column 2 of Table 5. Like costs, this benefit is only slightly discounted because it occurs at age 15. The effect of adding this benefit is to increase the estimated benefit-cost ratio from 2.7 to 3.9 (using a discount rate of 5%). This is a significant increase, indicating that in the context of economic analysis earlier completion of a given grade or level of schooling is an important schooling outcome12. The PACES evaluation study estimates that the increase of 0.2 standard deviations in standardized test scores among lottery winners is equivalent to about one full additional grade of schooling completed (based on the mean test scores by grade of United States Hispanic students taking the same test). If correct, this translates into an additional annual earnings benefit of $ 300, compared to the previously included annual earnings benefit of $ 36 based on the number of additional grades completed. This additional benefit is displayed in column 3 of Table 5. Like the benefit associated with the increased number of school grades completed, this benefit is heavily discounted because it occurs during the persons entire assumed working life (ages 16-60). Not surprisingly, including this benefit increases the estimated benefit/cost ratio dramatically, from 3.9 (including the earlier completion benefit and using a discount rate of 5%) to 30.7. 1.1.3.Argentinean Programa Joven Youth Job Training Program13 Program: The target population of Programa Joven was the large number of poor youth, male and female, with limited education and without work experience and who were unemployed, under-employed or inactive. The selection criteria for the program were minimum age of 16 years, no more than secondary education completed, member of a poor household, and not currently employed. The program provided intensive training (200 hours over a period of 6 to 12 weeks) for positions in the productive sectors of the economy, including reimbursement of transportation expenses, a stipend for females with children under 5, medical checkups, books, materials, work clothing and an 8-week internship in a firm. Program benefits: Program impacts were estimated by using propensity score matching to compare program participants with one or more similar (in terms of observed characteristics) non-participants using three different samples/ sources of information:
12
13
Moreover, its relative importance increases with the discount rate. For example, with a 5% discount rate, the earlier completion benefit accounts for 32% of total estimated benefits in this example, whereas with a 10% discount rate it accounts for 46% of total estimated benefits. Of course the gain from completing a given level of schooling at a younger age has been noted in other studies, including the one in Section 1.1.1, going back at least to Glewwe and Jacoby (1995). Based on Aedo and Nuez (2001) as summarized in Knowles and Behrman (2004).
22
1. Administrative data on all individuals who registered and qualified to take training programs during the period March 1996 to December 1997 (about 140,000 persons). 2. The same administrative data as above but restricted to a sample of 3,340 individuals consisting of equal numbers of persons extracted from the above group in each of the following two groups: 1) persons who completed the program training, selected to be representative in terms of sex and region of residence, and 2) a comparison group of persons who were qualified to enter the training programs but who did not participate and who resembled those who did participate in terms of age, sex, level of education, labor force participation, socioeconomic level and whether they had children under age 5 years. 3. Additional survey data available only for the restricted sample of 3,340 persons in the second data base. Separate logit functions explaining participation (with participation defined as successful completion of the technical knowledge phase of the training) were estimated using the three different samples/information sources described above to obtain three different sets of propensity scores14. The authors conjectured that use of different propensity scores would result in significantly different estimates of program impact. The estimation was done separately for: adult males (21-35), young males (16-20), adult females (21-35) and young females (16-20) as well as for all four groups combined. The outcome variables considered were earnings and employment in the 12th month following completion of the program. Estimates of program effects on earnings were statistically significant only for young males (16-20) and adult females (21-35). For young males, the estimated impact of the program on monthly earnings varied from $ 17.17 to $ 23.75 depending on the assumptions15. For adult females, the corresponding estimates varied from $ 23.40 to $ 32.40. Estimates of the programs impact on monthly earnings for all four groups combined varied from $ 15.67 to $ 22.26. Estimates of the program effect on the probability of employment were statistically significant only for adult females, with the estimated effect varying from 0.104 to 0.135. The evaluation study concludes that these differences in estimated program impact among groups probably reflect mainly the different labor market conditions facing each group. Program costs: The evaluation study separated the programs costs into direct and indirect components. Direct costs were defined to include the cost of program training services provided by competitively-selected providers, the cost of employee insurance during the internships, and the cost of stipends provided to
14
15
The samples were re-weighted prior to estimation with the sample of 3,340 to correct for choice-based sampling (i.e., the fact that the sample was constructed to have equal numbers of participants and non-participants). Within the statistically significant groups, the impact estimates were not very sensitive either to the sample/information used to estimate the propensity score functions or to the number of nearest neighbors used (5,10, 20, or 30). This result was surprising in the case of the propensity scores since the three sets of propensity scores were not very highly correlated.
23
the trainees (for example, the stipends paid to female participants with children under 5)16. The study assumes that the opportunity cost of trainee time is zero, which is a strong assumption17. The study also assumes a zero deadweight loss for the governmental resources used to fund the program (although a 50% deadweight loss is alternatively assumed in sensitivity analysis reported in the study). Indirect costs were estimated to be 32.7% of direct costs (with an indirect cost ratio of 15% used in sensitivity analysis). Program benefit/cost estimates: Benefit/cost analysis was done for young males and adult females (representing a targeted program) and all four groups combined (representing the actual program). The estimated effect of the program on monthly earnings (the average of the estimates obtained for each group) was assumed to be the programs only benefit and was assumed to remain constant during a period of time that was alternatively assumed to last for 1, 3, 6, 9, 12, 15 years or for an unlimited period of time. Two alternative discount rates were used (5% and 10%). In the case of the four groups combined, representing the actual program, the estimated net present value of benefits was positive with a discount rate of 5% after 12 years (but was still negative with a discount rate of 10% after 15 years).18 With the analysis restricted to young males and adult females, representing a program targeted only to these two groups, the estimated net present benefits are positive after 9 years with a discount rate of 5% (or with an indirect cost ratio of 15%) and after 12 years with a discount rate of 10%. If a deadweight loss of 50% of the programs total expenditure is assumed as the distortionary cost of governmental financing, estimated net present benefits of the existing program are non-positive after 15 years, even with a discount rate of 5%, while for a targeted program (i.e., one restricted to young males and adult females), they are positive only after 15 years with a discount rate of 5% (but only when an unlimited time period is assumed with a discount rate of 10%). 1.2. Human Capital Investments to Reduce Hunger and Malnutrition Behrman, Alderman and Hoddinott (2004) present benefit/cost ratios for four sets of interventions related to hunger and malnutrition in a low-income country context, based on extensive discussion of the available information with which to assess benefits and costs and the sensitivity of the estimates to different assumptions. This was part of the Copenhagen Consensus in which a panel of eight distinguished economists, including four Nobel Laureates, ranked about 40 projects in ten broad topic areas on the basis in part of benefit/cost analyses
16 17
18
This last category is actually a transfer and should not have been included in the program resource costs. Presumably, many of the trainees would have been employed in the informal sector or in housework if they had not participated in the program. This assumption also implicitly assumes that interns did not contribute positively to the output of the firms in which they were working for 8 weeks. However, with an indirect cost rate of 15% estimated net present benefits were positive after 15 years even with a discount rate of 10%. In addition, under the basic assumptions used, the program had estimated positive net benefits over an infinite time period.
24
prepared by experts in each of the areas (Lomborg 2004). In the final ranking of the alternatives projects related to hunger and malnutrition were highly ranked in terms of priorities. Here, I present new estimates based on a similar approach, but using prices and incomes for the LAC 2005 average19, Brazilian survival probabilities20 and alternative means of valuing adverted mortality in terms of the value placed on value per DALy (Disability Adjusted Life years) of $1,000 (low) and $ 5,000 (high), which are the values suggested by the Copenhagen Consensus for new estimates in initial Problem Papers for a 2008 Copenhagen Consensus exercise21. I consider three sets of nutritional interventions: Reducing the Prevalence of low birth weight (LBW): Many of the 12 million LBW infants born each year die at young ages, contributing significantly to neonatal mortality, which makes up the largest proportion of infant mortality in many developing countries. Unfortunately, rates of LBW have remained relatively static in recent decades. Because LBW infants are 40% more likely to die in the neonatal period than their normal weight counterparts, addressing LBW is essential to achieve reductions in infant mortality. Moreover, many of the LBW children who survive infancy suffer cognitive and neurological impairment and are stunted as adolescents and adults. Thus, in addition to contributing to excess mortality, LBW is associated with lower productivity in a range of economic and other activities. LBW also may be important in light of new evidence that shows that LBW infants may have an increased risk of cardiovascular disease, diabetes and hypertension later in life. LBW may also be an intergenerational problem because LBW girls who survive tend to be undernourished when pregnant with relatively high incidence of LBW children. While the prevalence of LBW is particularly high in South Asia, there are substantial numbers of LBW babies in other parts of the developing world, including in LAC. The benefits from reducing LBW encompass the PDV of the six impacts over the lifecycle and the one across generations that are summarized in the previous paragraph. Under the assumptions used here, the most important effects are from adverting mortality (11 to 50% of the total), intergeneration effects (18-39%), chronic diseases (8-38%), and productivity (6-27%). Note that these relative contributions differ a lot depending on the discount rates (5% or 10%) and the DALyS ($ 1,000 or $ 5,000). The higher the discount rate, the more
19
20
21
In particular, I use the Gross National Income per capita (Atlas method) of $4008 for Latin America in 2005 from http://web.worldbank.org/WBSITE/EXTERNAL/DATASTATISTICS/0 (24 November 2006) relative to the $ 500 value for the target population in a context such as South Asia in Behrman, Alderman and Hoddinott (2004) to adjust all the prices and incomes in that study for the present purpose. Survival probabilities are relevant since some impacts occur only if the individual affected by the intervention lives long enough. The source used is the WHO Life Table for Brazil at http://www.who.int/countries as of 24 November 2006. In Behrman, Alderman and Hoddinott (2004) for our preferred estimates we followed Summers (1992, 1994) and used instead of DALys at such levels, the alternative resource cost that these societies had chosen to allocate to advert an infant mortality, which was estimated to be $ 1250 in PDV terms (from inoculations) a much lower value than implied by DALys of $ 1,000 or $ 5,000.
25
important is adverting infant mortality (and to a lesser extent, neonatal and infant illnesses). The larger the DALyS, the more important is the effect on chronic diseases, as well of adverting infant mortality. Tables 6 and 7 summarize the benefits and costs for three interventions to reduce LBW. The benefit-to-cost ratios range from 2.5 to 292.6 with higher values for the low (5%) discount rate, the high DALy ($ 5,000) and the intervention of providing drugs for pregnant women with poor obstetric histories. Improving Infant and Child Nutrition: The nutritional literature emphasizes that undernutrition is most common and severe during periods of greatest vulnerability (Martorell 1997, UNICEF 1998). One such period is in utero, that is addressed above with regard to LBW. A second vulnerable period is the first two years or so of life. young children have high nutritional requirements, in part because they are growing so fast. Unfortunately, the diets commonly offered to young children in developing countries to complement breast milk are of low quality (i.e., with low energy and nutrient density), and as a result, multiple nutrient deficiencies are common. young children are also very susceptible to infections because their immune systems are both developmentally immature and compromised by poor nutrition. In poor countries, foods and liquids are often contaminated and are thus key sources of frequent infections that both reduce appetite and increase metabolic demands. Furthermore, in many societies, suboptimal traditional remedies for childhood infections, including withholding of foods and breast milk, are common. Thus infection and malnutrition reinforce each other. Improving the nutrition of infants and young children has effects at most points of the life cycle described above (except for reduced neonatal care and reduced costs of chronic diseases associated with LBW). While the prevalence of stunting is relatively high in South Asia and Sub-Saharan Africa, there is considerable stunting in other parts of the developing world including LAC. Tables 6 and 7 summarize the benefits and costs for two interventions to reduce stunting. The benefit/cost ratios range from 25.6 to 930.3 with higher values for the low (5%) discount rate, the high DALy ($ 5,000) and the intervention of breastfeeding promotion in hospitals. Reducing the Prevalence of Iron, Iodine and Vitamin A Deficiencies: Deficiencies in iron, iodine, and Vitamin A all have both immediate and long-term consequences. Iodine deficiency adversely affects development of the central nervous system and individuals with an iodine deficiency have, on average, lower IQs. Approximately 2 billion people are affected by iodine deficiency including 285 million children aged 6 to 12 years. Adequate iron intake is also necessary for brain development. More than 40% of children age 0-4 years in developing countries suffer from anemia; further anemia in school-age children may also affect schooling whether or not there had been earlier impaired brain development. Vitamin A deficiencies, which are estimated to affect 140 million pre-school children, are associated with increased risk of infant and child mortality. Some of these micronutrient deficiencies are considerable in LAC. Lessening these micronutrient deficiencies has effects at most points of the life cycle noted above. Tables 6 and 7 summarize the benefits and costs for three such interventions. The benefit/cost ratios range
26
TABLE 6
OVERVIEW TABLE ON COSTS AND BENEFITS FOR SELECTED INTERVENTIONS GIVEN DISCOUNT RATES AND DALyS Discount Rate 5% Benefit Cost Benefit 10% Cost
Objectives/Interventions
DALy
1. Reducing LBW for pregnancies with high probabilities LBW 1a. Treatments for women with asymptomatic bacterial infections
1b. Treatment for women with presumptive STD
1c. Drugs for pregnant women with poor obstretic history
$ 1,000 $ 5,000 $ 1,000 $ 5,000 $ 1,000 $ 5,000
$ 13,135 $ 45,062 $ 13,135 $ 45,062 $ 13,135 $ 45,062
$ 1,100 $ 1,100 $ 276 $ 276 $ 154 $ 154
$ 2,766 $ 7,460 $ 2,766 $ 7,460 $ 2,766 $ 7,460
$ 1,100 $ 1,100 $ 276 $ 276 $ 154 $ 154
2. Improving infant and child nutrition in populations with high prevalence of child malnutrition (fairly widespread in poor populations in developing countries) 2a. Breastfeeding promotion in hospitals in which norm was promotion of infant formula 2b. Integrated child care programs (adding improved home child care practices) 3. Reducing micro nutrient deficiencies in populations in which they are prevalent 3a. Iodine (per woman of child bearing age) $ 1,000 $ 5,000 $ 1,000 $ 5,000 $ 142,219 $ 556,814 $ 2,839 $ 2,839
$ 599 $ 599 $ 40 $ 40
$ 26,542 $ 89,484 $ 1,024 $ 1,024
$ 599 $ 599 $ 40 $ 40
3b Vitamin A (pre child under six years)
3c. Iron (pregnant women and then over life cycle of children)
$ 1,000 $ 5,000 $ 1,000 $ 5,000 $ 1,000 $ 5,000
$ 12,244 $ 40,607 $ 520 $ 2,377 $ 14,831 $ 46,758
$3 $3 $6 $6 $ 16 $ 16
$ 2,304 $ 5,147 $ 257 $ 1,221 $ 3,261 $ 7,955
$3 $3 $6 $6 $ 14 $ 14
Benefits are calculated as described in Behrman, Alderman and Hoddinott (2004), but with the discount rates and DALys indicated at the column head of this table and with the survival probabilities implied by life tables from India. Costs are mid-points of costs from Behrman, Alderman and Hoddinott (2004, Table 6). For iron the costs are the mid-point for pregnant women initially with the per capital costs in subsequent years.
TABLE 7
OVERVIEW TABLE FOR BENEFIT/COST RATIOS 5% $ 1,000 $ 5,000 $ 1,000 10% $ 5,000
Discount Rate
DALy
1. Reducing LBW for pregnancies with high probabilities LBW 11.9 47.6 85.3 292.6 163.3 41.0 2.5 10.0 18.0 6.8 27.0 48.4
1a. Treatments for women with asymptomatic bacterial infections
1b. Treatment for women with presumptive STD
1c. Drugs for pregnant women with poor obstretic history
2. Improving infant and child nutrition in populations with high prevalence of child malnutrition (fairly widespread in poor populations in developing countries) 237.6 71.0 930.3 71.0 44.3 25.6 149.5 25.6
2a. Breastfeeding promotion in hospitals in which norm was promotion of infant formula
2b. Integrated child care programs (adding improved home child care practices)
3. Reducing micro nutrient deficiencies in populations in which they are prevalent 4,664.5 94.6 903.3 15,469.3 432.1 2,847.8 877.6 46.8 229.8 1,960.8 222.0 560.6
3a. Iodine (per woman of child bearing age)
3b Vitamin A (pre child under six years)
3c. Iron (pregnant women and then over life cycle of children)
27
28
from 46.8 to 15,469 with higher values for the low (5%) discount rate, the high DALy ($ 5,000) and for the provision of iodine. 1.3. Summary of Benefit/Cost Estimates for Human Capital Interventions in LAC To my knowledge, there are but a limited number of benefit/cost estimates for human capital interventions in LAC or even more broadly in the developing world, and a number of the estimates that do exist have problems with interpretation because, for example, the definition of benefits used is very limited or confounding resource costs with transfers or other estimation problems. The selected survey of benefit/cost estimates in this section includes studies that probably are better than most others in regard to the many estimation issues, but still have their flaws, as noted. And they appear quite sensitive to some critical assumptions, particularly regarding the appropriate discount rate and the value of adverting mortality (or increasing the number of healthy years of life). In a number of cases for the nutritional interventions, moreover, the benefit/cost ratios are so high as to strain credibility. If benefit/cost ratios are anywhere near this high, how is it possible that such opportunities have been bypassed? Part of the answer to this question may relate to the underlying estimates of impacts and costs (e.g., is scaling-up really captured adequately?) or to, particularly in the general LAC context and probably even more so for Chile in particular, there being only relatively small and dispersed groups for which the deficiencies exist for which the returns are high. Nevertheless, despite all their warts and flaws, the benefit/cost ratios reviewed suggest that even if these estimates might be high, there are likely to be some attractive human capital-related interventions in LAC more generally and in Chile in particular. 2. Schooling Attainment and Earnings Distribution in the 2004 Chilean Social Protection Survey Given the qualified, but nevertheless definitely positive assessment of a number of human capital-related interventions from the selected review in Section 1, it is of interest in the context of this volume to ask what would be the impact on the income distribution were there have been some targeted increases in human capital. The review in Section 1 would seem to suggest that some of these effects might be considerable. Therefore I explore this question in this section by simulating impacts of hypothetical changes in schooling attainment in Chile. I focus on schooling attainment because it is the most-emphasized among human capital investments indeed many studies empirically seem to equate schooling attainment with human capital. I focus on Chile not only because it is the focus of this volume and but also because I am involved, with some other colleagues in collection and analysis of the Chilean Social Protection Survey (SPS) that is described in more detail in Appendix A. The purpose of this exercise, however, is not so much to describe real-world policy options for Chile, but to use the SPS to generate some illustrative simulations about the impact of changed human capital distributions on the earnings distributions and to calculate suggestive orders of magnitude of the effects.
29
Basic description of inequality and poverty in the 2004 Chilean SPS: Table 8 summarizes some relevant sample-weighted statistics from the 2004 SPS regarding the distribution of schooling attainment, labor market earnings, and wage rates per hour for the 51,244 adults age 21 years or older for whom this information all is available. As is well-known, average schooling attainment levels in Chile are relatively high in comparison with many developing countries, including a number of others in LAC. These estimates suggest that the distribution of schooling attainment also is relatively equal, with a Gini coefficient of 0.23, in comparison numbers reported for other countries in LAC (Table 1)22. Almost a fifth of the sample (18%) had schooling attainment below six grades, and about three tenths had schooling attainment below eight grades. TABLE 8
CHILEAN 2004 SPS, BASIC STATISTICS FOR SCHOOLING, EARNINGS AND WAGE RATES FOR 51,244 ADULTS AGE 21+ yEARS*
Mean SD Schooling Attainment (Grades) Earnings (1000 Pesos per year) Wage Rate (Pesos per hour)
*
Gini Coefficient SE 0.23 0.001 0.53 0.003 0.49 0.004
Poverty Headcount Low SE 0.18 0.002 0.20 0.002 0.17 0.002 Higher SE 0.30 0.002 0.36 0.002 0.33 0.002
9.5 4.3 2149 3195 1193 1751
The low and higher poverty lines for schooling are < 6 and < 8 grades, for earnings < 500000 and < 1000000 pesos per year, and for wages < 400 and < 600 pesos per hour. The standard errors for the Gini coefficients are calculated by bootstrapping with 50 repetitions. All estimates are weighted to reflect the sample design.
The 2004 SPS earnings Gini coefficient is 0.53, significantly below the value for Chile given in Table 1 for around 2000 and at about the mean for LAC in that table. The 2004 wage rate Gini coefficient is somewhat lower at 0.49, suggesting that some of the inequality in actual earnings income is due to choices of those who have higher market wage rates to spend more time working. Therefore the Gini coefficient for the wage rate or, equivalently, for full income, is somewhat lower. For both earnings and the wage rate I define low and higher poverty cutoffs. These are not related to any official or unofficial efforts to define poverty consumption baskets in any meaningful way, for which reason I put poverty in quotation marks here. They are just convenient benchmarks at rounded values of pesos per year and per hour, respectively, that demarcate the points below which in the 2004 SPS distribution there are about a fifth (0.20 for earnings,
22
Including the Gini coefficient reported for Chile in 2000 in Table 1. I do not have an explanation for the difference.
30
0.17 for the wage rate) and about a third (0.36 for earnings, 0.33 for the wage rate) of the individuals with positive earnings and wage rates. Semilog earnings and wage rate functions based on the 2004 SPS: To be able to simulate how different hypothetical patterns of schooling attainment would affect the earnings and wage rate distributions, I need estimates of these relations. Of course there is considerable controversy about how best to estimate these relations and to what extent and in what direction OLS estimates are biased (e.g., Card 1999, Behrman and Rosenzweig 1999). For the illustrative purposes here, I nevertheless use OLS estimates, with splines in schooling to allow for nonlinearities with differential returns to the first six grades (primary) versus the second six grades (secondary) versus more than 12 grades (tertiary). I also include a quadratic in potential work experience (age-schooling attainment six) and allow for all the coefficient estimates to differ between males and females. Table 9 summarizes the resulting estimates. The first two rows of coefficient estimates refer to the semilog earnings function and the second two rows refer to the semilog wage rate function, with t values in italics beneath each point estimate. Within each pair of rows, the first row gives the estimates for females and the second row gives the extent to which the point estimate for males differs significantly (at the 5% level) if it does. TABLE 9
WEIGHTED SEMI-LOG EARNINGS AND WAGE FUNCTIONS FOR 51,244 ADULTS AGE 21+ IN 2004 CHILEAN SPS*
Spline in Schooling Prim ln Earnings xMale 0.040 5.31 0.036 4.16 0.039 9.06 0.041 30.42 Sec 0.102 22.90 0.015 2.84 0.104 49.16 Ter 0.235 36.69 0.038 4.20 0.227 77.89
Pot Experience Level 0.040 28.79 Square
Const
R sq F 0.201 1009
0.00048 12.412 22.50 286.00 0.169 4.05 0.00030 5.481 16.44 220.08
ln Wage xMale
0.022 21.37
0.339 2810
t-statistics in italics beneath coefficient estimates; potential experience is age schooling attainment 6; the row with xMale includes all significant coefficients for the interaction between male and the right-side variable in that column.
Overall the semilog earnings function is consistent with about a fifth of the sample variance and the semilog wage rate function is consistent with about a third of the sample variance. That means, of course, that there are considerable
31
shares of the variance that are not affected by schooling attainment four-fifths for ln earnings and two-thirds for ln wage rates. The quadratics in experience are highly significant, and suggest an increasing impact at a diminishing rate of experience with no significant differences between females and males. The schooling coefficient estimates also are highly significant and also suggest strong nonlinearities, with much larger effects for an additional grade of tertiary school attainment than for an additional grade of secondary school attainment and a much larger effect for an additional grade of secondary school attainment than for an additional grade of primary school attainment. Males have an estimated 16.9% greater earnings independent of schooling attainment and potential experience, and have significantly higher schooling coefficients for both primary and secondary schooling, though females have a significantly higher coefficient for tertiary schooling in the ln earnings function. However, except for the higher coefficient for males for primary schooling, the gender differences in the ln earnings function estimates basically reflect gender differences in hours worked in the paid labor force. Therefore, with the exception again of primary schooling, there are no significant gender differences in the ln wage rate function. Simulations of the impact of alternative targeted schooling attainment increases on inequality and poverty: Table 10 summarizes eight simulations of the impact of alternative targeted changes in schooling attainment on Gini coefficients for inequality and the low and higher poverty cutoff headcounts for earnings and for wage rates. These simulations assume that the semilog earnings and wage rate relations in Table 9 hold. That is, they assume that the coefficient estimates are unbiased estimates of the true causal effects of schooling attainment on ln earnings and ln wages23, that none of the simulation interventions is so large to change the relevant coefficients, and that each individual receives the same shock (i.e., has the same residual) as in the estimates in Table 924. The Gini coefficients and the poverty headcount measures are in bold, with the standard errors to the right in italics. Note that the standard errors generally are small relative to the simulated changes, so the simulated changes generally are significant. At the top of Table 10 the Gini coefficient estimates and poverty headcounts for the base data are given, as in Table 8, for easy reference. The question of interest is how much do each of the alternative simulations differ from these base values. The eight simulations that are included in the table are:
23
24
That is, that there are not omitted variable biases because these estimates do not control for unobserved ability and other aspects of family background and for school quality, though some available estimates suggest that the failure to control for such factors may result in substantial overestimates of the impact of schooling attainment because schooling attainment partially is proxying for individual and family background and school quality (e.g., Behrman and Birdsall (1983), Behrman and Rosenzweig (1999)). An alternative interpretation of the simulations with regard to school quality is that the underlying assumption is that school quality is positive correlated with schooling attainment (as suggested by the modeling and the empirical results in Behrman and Birdsall (1983)) and that in the simulations schooling quality is being increased together with schooling attainment in order to maintain the association between the two that exists in the sample data. From one cross-section one can not identify whether these are transitory or permanent shocks.
32
1. Increasing schooling attainment by one grade for all individuals whose wage rate is below 400 Pesos per hour in the SPS. 2. Increasing schooling attainment by three grades for all individuals whose wage rate is below 400 Pesos per hour in the SPS. 3. Increasing schooling attainment by five grades for all individuals whose wage rate is below 400 Pesos per hour in the SPS. 4. Increasing schooling attainment by three grades for all individuals whose wage rate is below 600 Pesos per hour in the SPS. 5. Increasing schooling attainment by three grades for all individuals whose schooling attainment is equal to or below six grades in the SPS. 6. Increasing schooling attainment by three grades for all individuals whose schooling attainment is equal to or below nine grades in the SPS. 7. Increasing schooling attainment to six grades for all individuals whose schooling is below six grades in the SPS. 8. Increasing schooling attainment to nine grades for all individuals whose schooling is below nine grades in the SPS. The first four simulations each have increases in schooling attainment for individuals who are low in the wage rate distribution. The second set of four simulations each have increases in schooling attainment for individuals low in the schooling distribution. Of course these eight simulations differ, perhaps substantially, with regard to the resource costs of the changes in schooling attainment that are considered. This is indicated by the column in Table 10 titled Added School Grades which gives the number of added school grades per 100 representative people that would be required for each simulation. In simulation 1, for example, the 17% of the population with wage rates below 400 pesos per hour are given one more grade of school, so 17 added school grades are needed per 100 people and in simulation 2 since the same individuals are given three more school grades each, 51 added school grades are needed per 100 people. Under the assumption that the cost per added grade of school attained is the same for all grades of primary and secondary school for all people whatever their actual schooling and that the inequality and poverty measures also are approximately linear over small ranges, the changes in Gini coefficients and in poverty headcounts can be standardized by the added school grades to obtain the changes in Gini coefficients and in poverty headcounts across the eight simulations (each in comparison with the baseline) relative to the same costs in terms of schooling attainment. Table 11 presents these estimates, which are more informative than are those in Table 10 because of the control for costs across alternatives. Some observations that come out of this exercise include: Schooling inequality: With the exception of simulation 1, these simulations reduce the Gini coefficient for schooling. These reductions are much larger (about twice as large) for the simulations targeted towards those with low schooling (i.e., simulations 5-8) rather than those targeted to those with low wage rates (i.e.,
TABLE 10
SIMULATIONS OF IMPACTS OF CHANGED SCHOOLING DISTRIBUTION ON INEQUALITy AND POVERTy BASED ON 2004 CHILEAN SOCIAL PROTECTION SURVEy*
Simulation
Schooling Change
Target Pop
Added School Grades
Base 17 51 85 100 81 141 56 212
1 grade
Wage <400
3 grades
Wage <400
5 grades
Wage <400
3 grades
wage <600
3 grades
sch <=6
3 grades
sch <=9
to 6 grades
sch < 6
to 9 grades
sch < 9
Schooling Earnings Wage Rates Schooling Earnings Wage Rates Schooling Earnings Wage Rates Schooling Earnings Wage Rates Schooling Earnings Wage Rates Schooling Earnings Wage Rates Schooling Earnings Wage Rates Schooling Earnings Wage Rates Schooling Earnings Wage Rates
Gini Coefficient Level SE 0.23 0.001 0.53 0.003 0.49 0.003 0.24 0.001 0.53 0.003 0.48 0.004 0.22 0.001 0.51 0.003 0.47 0.004 0.21 0.001 0.50 0.003 0.46 0.004 0.21 0.001 0.50 0.003 0.45 0.004 0.18 0.001 0.52 0.003 0.48 0.003 0.16 0.001 0.51 0.003 0.47 0.004 0.19 0.000 0.52 0.003 0.48 0.003 0.12 0.001 0.50 0.003 0.46 0.005
Poverty Headcount Low Cutoff Higher Cutoff Level SE Level SE 0.18 0.002 0.30 0.002 0.20 0.002 0.36 0.002 0.17 0.002 0.33 0.002 0.17 0.002 0.29 0.002 0.19 0.002 0.34 0.002 0.15 0.002 0.33 0.330 0.14 0.002 0.26 0.002 0.18 0.002 0.32 0.002 0.11 0.001 0.32 0.002 0.13 0.002 0.23 0.002 0.16 0.002 0.30 0.002 0.08 0.001 0.29 0.002 0.11 0.001 0.22 0.002 0.17 0.002 0.30 0.002 0.11 0.001 0.20 0.002 0.06 0.001 0.18 0.002 0.18 0.002 0.33 0.002 0.15 0.002 0.30 0.002 0.06 0.001 0.15 0.002 0.17 0.002 0.31 0.002 0.13 0.002 0.26 0.002 0.00 0.000 0.30 0.002 0.19 0.002 0.34 0.002 0.16 0.002 0.32 0.002 0.00 0.000 0.00 0.000 0.16 0.002 0.30 0.002 0.12 0.002 0.26 0.002
Based on the estimates in Table 9. The low and higher poverty lines for schooling are <6 and <8 grades, for earnings < 500000 and < 1000000 pesos per year, and for wages < 400 and < 600 pesos per hour. Standard errors are to the right of the estimates in italics. The standard errors for the Gini coefficients are calculated by bootstrapping with 50 repetitions. All estimates are weighted to reflect the sample design. Added school grades is the added school grades per 100 representative people required by the simulation.
33
34
TABLE 11
SIMULATIONS OF IMPACTS OF CHANGED SCHOOLING DISTRIBUTION ON INEQUALITy AND POVERTy HEADCOUNT MEASURES, NORMALIzED By REQUIRED SCHOOL GRADE INCREASES, BASED ON 2004 CHILEAN SPS*
Simulation Schooling Change 1 1 grade Target Pop wage <400 Schooling Earnings Wage Rates Schooling Earnings Wage Rates Schooling Earnings Wage Rates Schooling Earnings Wage Rates Schooling Earnings Wage Rates Schooling Earnings Wage Rates Schooling Earnings Wage Rates Schooling Earnings Wage Rates Gini Coefficient 0.01 0.03 0.03 0.03 0.03 0.03 0.03 0.03 0.04 0.03 0.03 0.04 0.07 0.01 0.01 0.06 0.02 0.02 0.08 0.01 0.02 0.06 0.01 0.01 Poverty Headcount Low 0.07 0.08 0.11 0.08 0.05 0.12 0.07 0.05 0.11 0.07 0.03 0.06 0.15 0.02 0.03 0.09 0.02 0.03 0.33 0.02 0.03 0.09 0.02 0.02 Higher 0.04 0.07 0.00 0.09 0.06 0.02 0.09 0.07 0.06 0.08 0.06 0.14 0.15 0.03 0.04 0.11 0.04 0.05 0.00 0.03 0.03 0.14 0.03 0.03
3 grades
wage<400
5 grades
wage <400
3 grades
wage <600
3 grades
sch <=6
3 grades
sch <=9
to 6 grades
sch < 6
to 9 grades
sch < 9
Based on the estimates in Table 10, normalized by the required additional grades of schooling for each simulation that are given in that table. The low and higher poverty lines for schooling are < 6 and < 8 grades, for earnings < 500000 and < 1000000 pesos per year, and for wages < 400 and < 600 pesos per hour.
simulations 1-4). This pattern emphasizes the point that while schooling attainment and wage rates (and earnings) are positively correlated, the correlations are far from one25. The normalized changes in the Gini coefficients for schooling are from 0.01 to 0.03 for the four simulations with low wage rates targeted and are from 0.06 to 0.08 for the four simulations with low schooling targeted. The largest simulated decline of 0.08 is for bringing all with less than six grades of completed schooling up to six grades of schooling attainment. In order to reduce schooling inequality, hardly surprisingly, the most effective policy is to focus on those in the left-hand part of the schooling distribution26.
25 26
The correlations are 0.36 between schooling attainment and either the wage rate or earnings. The correlation between the wage rate and earnings is 0.80. If the cost per added schooling grade attainment increases with schooling levels (e.g., is higher for secondary than for primary school), then the relative advantage in terms of reducing schooling inequality of focusing on the left-hand tail of the schooling distribution is even greater than indicated by these simulations.
35
Schooling poverty headcounts: The same general point holds, also not surprisingly, for the schooling poverty headcounts. For the simulations targeted towards those with low wage rates, the proportion of the population below the low cutoff of six grades of attained school drop by 0.07 or 0.08 and the proportion below the higher cutoff of eight grades drops by from 0.04 to 0.09. For the simulations targeted towards those with low schooling attainment, the proportion of the population below the low cutoff of six grades of attained school drop by 0.09 to 0.3327 and the proportion below the higher cutoff of eight grades drops by from 0.00 to 0.15. Notice, however, that there is a nuance. Policies targeted toward increasing schooling only below the low cutoff (e.g., simulation 7) reduce the proportion below the high cutoff less (by 0.0 in this case) than policies directed towards increasing the schooling attainment of those with low wage rates (simulations 1-4). Earnings inequality: All of these simulations reduce the Gini coefficient for earnings. These reductions are much larger (from 1.5 to three times as large) for the simulations targeted towards those with low wage rates (i.e., simulations 1-4) rather than those targeted to those with low schooling attainment (i.e., simulations 5-8). This pattern again reinforces the point that while schooling attainment and wage rates (and earnings) are positively correlated, the correlations are far from one. The normalized changes in the Gini coefficients for earnings are 0.03 for the four simulations with low wage rates targeted and are 0.01 to 0.02 for the four simulations with low schooling targeted. If the goal were only to reduce earnings inequality, thus, it would be desirable to target effectively those with low wage rates28 but in terms of this criterion alone it does not matter which of the four alternatives considered with low wage rate targeting (simulations 1-4) is used (though it does matter for other possible objectives such as lowering those below poverty cutoffs). Of course how to target those with low wage rates for interventions to increase schooling attainment is a tricker question than how to target those with low schooling given the concentration of schooling in childhood and youth and that wage outcomes are revealed only later in the life cycle. Earnings poverty headcounts: The same general point again holds, not surprisingly, for the earnings poverty headcounts. For the simulations targeted towards those with low wage rates, the proportion of the population below the low cutoff of 500000 pesos per year drop by from 0.03 to 0.08 and the proportion below the higher cutoff of 1000000 pesos per year drops by from 0.06 to 0.07. For the simulations targeted towards those with low schooling attainment, the proportion of the population below the low cutoff of 500000
27
28
That this drop is greater than the 0.18 in the base simulation is because of the normalization. Obviously this policy could be pursued only to the point at which the proportion below the cutoff became 0. To keep the number of simulations discussed to a reasonable limit, I do not present simulations that target low earnings; these are somewhat more effective than targeting low wage rate recipients in reducing earnings inequality, but the difference is not as great as between targeting low wage recipients versus those with low schooling (given the much higher correlation between earnings and wage rates noted above than between either and schooling attainment).
36
pesos per year drop by 0.02 and the proportion below the higher cutoff of 1000000 pesos per year drops by from 0.03 to 0.04. Though all four of the simulations directed towards low wage rate recipients have the same impact on reducing the earnings Gini coefficient, they differ a fair amount with respect to their impacts on the proportions below the poverty cutoffs. In particular increasing schooling attainment by one grade for all those with wage rates below 400 Pesos per hour is simulated to reduce the normalized proportion below the low poverty cutoff of 500000 pesos per year by 0.08 (simulation 1), while at the other extreme among the first four simulations, increasing schooling attainment by three grades for all those with wage rates below 600 pesos per hour is simulated to reduce the normalized proportion below the same low poverty cutoff by only 0.03 (simulation 4). This is an interesting illustration of why it is important not to presume that reducing inequality and reducing poverty are sufficiently similar objectives that they lead to the same policy priorities. Wage rate inequality: All of these simulations reduce the Gini coefficients for wage rates at least as much or slightly more (i.e., by an additional 0.01 in three of the eight simulations) than for earnings. That the simulations, if anything, are slightly more effective for reducing wage rate than earnings inequalities reflects that the simulations presented are targeted towards low wage rates rather than low earnings but that the differences are not all that great reflects the relatively high correlation between earnings and wage rates. As for earnings, these reductions are much larger (from 1.5 to four times as large) for the simulations targeted towards those with low wage rates (i.e., simulations 1-4) rather than those targeted to those with low schooling attainment (i.e., simulations 5-8). The normalized changes in the Gini coefficients for wage rates are 0.03 to 0.04 for the four simulations with low wage rates targeted and are 0.01 to 0.02 for the four simulations with low schooling targeted. If the goal were only to reduce wage rate or full income inequality, thus, it would be desirable to target effectively those with low wage rates. The question of how to target those with low wage rates for interventions to increase schooling attainment, of course, remains a challenge given life-cycle schooling-earnings sequencing patterns. Wage rate poverty headcounts: The same general point again holds, again not surprisingly, for the wage rate poverty headcounts. For the simulations targeted towards those with low wage rates, the proportion of the population below the low cutoff of 400 pesos per hour drop by 0.06 to 0.12 and the proportion below the higher cutoff of 600 pesos per hour drops by from 0.00 to 0.14. For the simulations targeted towards those with low schooling attainment, the proportion of the population below the low cutoff of 400 pesos per hour drop by 0.02 to 0.03 and the proportion below the higher cutoff of 1000000 pesos per year drops by from 0.03 to 0.05. Though all four of the simulations directed towards low wage rate recipients have about the same impact on reducing the wage rate Gini coefficient, they differ a fair amount with respect to their impacts on the proportions below the poverty cutoffs. In particular increasing schooling attainment alone for those with wage rates below 400 pesos per hour is simulated to reduce the normalized proportion below the low poverty cutoff of 400 pesos per hour by 0.11 or 0.12 (simulations 1-3), while increasing schooling attainment by three grades for all those with wage rates below 600 pesos per
37
hour is simulated to reduce the normalized proportion below the same low poverty cutoff by only about half as much at 0.06 (simulation 4). On the other hand the latter simulation is much more effective at reducing the proportion below the higher poverty cutoff of 600 pesos per hour (by 0.14) than the first three simulations (from 0.00 to 0.06). This is an interesting illustration of how sensitive policy effectiveness rankings might be to which part of the lower end of the wage rate distribution is of interest29. 3. Conclusions Human resource investments targeted towards the lower part of wage rate, earnings or income distributions are often thought to be major means through which the persistent high LAC inequalities in general, including those in Chile, might effectively be addressed, with concomitant effects of reducing poverty. Previous and new benefit/cost estimates for a range of human resource investments pertaining to different aspects of education and health/nutrition, though subject to a number of caveats because of the problems in making such estimates and their sensitivity to alternative assumptions such as regarding discount rates and the value of adverting mortality, suggest that there are likely to be a number of high rate-of-return options for such investments in LAC. These include different dimensions of education, including pre- and post-school experiences and not just schooling, and a number of possibly higher rate-of-return interventions in health/nutrition. These estimates suggest that there are opportunities that LAC governments should evaluate in their own particular contexts, including the extent of the subpopulations for which deficits are important and the nature of delivery mechanisms. This is likely to lead to some high rate-of-return options for human resource interventions. Will such human resource interventions, if well-targeted, reduce substantially inequality in the and poverty region? To partially explore this question, the Chilean 2004 SPS data are used to examine the impact of schooling attainment, the most emphasized human capital investment (and one through which in important part other interventions such as those related to early childhood health and nutrition may work), using wage rate and earnings functions that probably are optimistic about the impact of schooling (because they ignore such factors as ability bias and market-level effects). Alternative simulations suggest significant impacts of well-targeted increases in schooling attainment on reducing inequality and poverty headcounts in schooling, earnings, and wage rates. They also illustrate the desirability of targeting directly towards the outcome of interest (e.g., towards those with low wage rates, not low schooling, if full income is of primary concern) despite the possible difficulties in doing so (since wage experiences typically are not revealed until after most people have
29
This point is closely related, of course, to the question of what poverty measure to use i.e., headcount versus poverty gap versus poverty depth measures within the Foster, Greer and Thorbecke (1984) measures. For simplicity in this paper I present only the headcount measures, though as part of the analysis for the paper all three of the measures have been simulated.
38
finished their schooling), the possible complementarities and tradeoffs between reducing inequality and reducing poverty, that policies with the same impact on inequality may have very different effects on poverty, and that the poverty impact may depend importantly on what part of the lower end of distributions is of interest. But, though the magnitudes of some of the simulated impacts on the poverty headcounts are fairly large, the magnitudes of the reductions in the Gini coefficient estimates for earnings and for wage rates are not very large even though the simulated changes in schooling attainment are considerable. Of course the benefit/cost ratios suggest that some other human capital interventions may have higher rates of return than increasing schooling attainment, and this may carryover to greater impacts on inequality and on poverty. But all in all it would seem that these simulations suggest that while there is significant scope for reducing inequality and probably somewhat more poverty through human capital interventions, expectations should not be for massive changes through these mechanisms alone unless there really are massive improvements in the human capital of the poorer members of society.
APPEnDIx A The Chilean Social Protection Survey In 2002, the Microdata Center of the Department of Economics of the Universidad de Chile under the Directorship of David Bravo conducted a new household survey (2002 EPS, Encuesta de Proteccin Social, or Social Protection Survey). This is a very valuable research asset providing researchers a host of useful new individual-level information for addressing numerous research questions information that previously was unavailable30. The interview sample was drawn from a sampling frame of approximately 8.1 million current and former affiliates of the Chilean old-age systems compiled from official databases obtained from the Chilean Ministry of Labor and Social Security; the frame included about three-quarters of the population age 15+ in 2001. The survey was fielded between April and December of 2002, collecting data from individuals who were working, unemployed, out of the labor force, receiving pensions, or deceased (in the latter case, information was collected from surviving relatives). The 2002 EPS survey included socio-demographic information and current labor market data for each member of the household, detailed information about receipt of pensions and types of pension plan participation and retrospective labor market history going back to 1980. The 2002 survey contains data on 17,246 respondents (937 of them were reports by surviving relatives) affiliated with
30
The 2002 survey was initially called the 2002 History of Labor and Social Security (HLSS, Historia Laboral y Seguridad Social) survey. But the follow-up 2004, 2006 and 2009 longitudinal surveys are called Social Protection Surveys (SPS, Encuesta de Proteccin Social). The interested reader is referred to http://www.proteccionsocial.cl for access to the public use data, codebooks, and documentation of the survey.
39
the old or the new retirement system for at least one month at any time during the 1981-2001 time span31. Another round of the survey was administered in 2004, which included a second wave for previously surveyed respondents, plus new surveys for a subsample of individuals not affiliated with the social security system (individuals never employed in the formal sector) and also a subsample of new entrants into the AFP system between 2002 and 2004. In addition a host of new health and wealth questions were introduced for the first time32. Accordingly, the 2004 survey is representative of the entire Chilean population33. Furthermore in 2004, the research project received permission to merge responses to the sampled respondents with administrative records on pension contributions and earnings in the PAyGO and AFP systems since 1980, data on the amounts of recognition bonds; and monthly data on account changes in the individual investment accounts, switches between AFPs, AFP commissions charged, and investment returns earned on all accounts in the AFP system. The survey data has also been merged with monthly Social Security records, available since 198134. The 2004 EPS was the last round available when this paper was written for the December 2006 workshop. References Aedo, C. and S. Nez (2001). The Impact of Training Policies in Latin America and the Caribbean: The Case of Programa Joven. Processed, Georgetown University (May). Alderman, H., J.R. Behrman, D. Ross, and R. Sabot (1996). The Returns to Endogenous Human Capital in Pakistans Rural Wage Labor Market, Oxford Bulletin of Economics and Statistics 58: 1 (February); 29-55. Angrist, J.D., E. Bettinger, E. Bloom, E. King and M. Kremer (2002). Vouchers for Private Schooling in Colombia: Evidence from a Randomized Natural Experiment, American Economic Review 92: 5 (December); 1535-59. Barro, R., J. and J.-W. Lee (2000). International Data on Educational Attainment: Updates and Implications. CID Working Paper 42. Cambridge, MA: Center for International Development. Behrman, J.R., H. Alderman and J. Hoddinott (2004). Hunger and Malnutrition in ed. B. Lomborg, Global Crises, Global Solutions, Cambridge, UK: Cambridge University Press; 363-420.
31
32 33 34
Information on the methodology and extent of the survey can be found in Bravo (2004), Bravo et al. (2006) and Bravo et al. (2008). Members of the Armed Forces or police covered by separate government pension systems were excluded, as well as a very small percentage of the Chilean population residing in inaccessible or sparsely populated areas (e.g. islands). A number of the questions were adapted from the United States Health and Retirement Study (HRS) with the intention of providing cross-national comparisons. Weights are available to reweight to random sampling proportions. Additional rounds of the survey were administered in 2006 and 2009 and there are plans to undertake further rounds in the future (see www.proteccionsocial.cl).
40
Behrman, J.R. and N. Birdsall (1983). The Quality of Schooling: Quantity Alone is Misleading, American Economic Review 73; 928-946. Behrman, J.R., N. Birdsall, and G. Pettersson (2006). Inequality and Financial Reform in Latin America. Washington, DC: Center for Global Development. Behrman, J.R., y. Cheng and P. Todd (2004). Evaluating Preschool Programs when Length of Exposure to the Program Varies: A Nonparametric Approach, Review of Economics and Statistics 86: 1 (February 2004); 108-132. Behrman, J.R. and J. Hoddinott (2005). Program Evaluation with Unobserved Heterogeneity and Selective Implementation: The Mexican Progresa Impact on Child Nutrition, Oxford Bulletin of Economics and Statistics 67: 4; 547-569. Behrman, J.R., J. Hoddinott, J.A. Maluccio, E. Soler-Hampejsek, E.L. Behrman, R. Martorell, M. Ramirez and A.D. Stein (2008). What Determines Adult Cognitive Skills? Impacts of Pre-School, School-years and PostSchool Experiences in Guatemala, Washington, DC: IFPRI Discussion Paper N 826. Behrman, J.R., P. Sengupta and P. Todd (2005). Progressing through PROGRESA: An Impact Assessment of Mexicos School Subsidy Experiment, Economic Development and Cultural Change 54: 1 (October 2005); 237-275. Behrman, J.R., S.W. Parker and P.E. Todd (2009a) Medium-Term Impacts of the Oportunidades Conditional Cash Transfer Program on Rural youth in Mexico, in S. Klasen and F. Nowak-Lehmann, eds., Poverty, Inequality, and Policy in Latin America, Cambridge, MA: MIT Press; 219-270. Behrman, J.R., S.W. Parker, and P.E. Todd (2009). Schooling Impacts of Conditional Cash Transfers on young Children: Evidence from Mexico, Economic Development and Cultural Change 57: 3 (April); 439-477. Behrman, J.R. and M.R. Rosenzweig (1999). Ability Biases in Schooling Returns and Twins: A Test and New Estimates, Economics of Education Review 18; 159-167. 1 Bravo, D. (2004). Anlisis y Principales Resultados. Primera Encuesta de Proteccin Social. Centro de Microdatos, Universidad de Chile and Subsecretara de Previsin Social, Ministerio del Trabajo. Bravo, D., J. Behrman, O. Mitchell and P. Todd (2006). Encuesta de Proteccin Social 2004: Presentacin general y principales resultados. Centro de Microdatos, Universidad de Chile and Subsecretara de Previsin Social, Ministerio del Trabajo. Bravo, D., J. Vsquez, J. Behrman, O. Mitchell and P. Todd (2008). Encuesta de Proteccin Social 2006: Presentacin general y principales resultados. Centro de Microdatos, Universidad de Chile and Subsecretara de Previsin Social, Ministerio del Trabajo. Card, D.E. (1999). New Developments in the Economics of Schooling, in Orley Ashenfelter and David Card, eds., Handbook of Labor Economics, Vol. 4, Amsterdam: North-Holland Publishing Co. Deininger, K. and L. Squire (1996). A New Data Set Measuring Income Inequality, World Bank Economic Review 10: 3 (September); 565-591.
41
Fiszbein, A. and N. Schady (2009). Conditional Cash Transfers: Reducing Present and Future Poverty, Washington, DC: World Bank Policy Research Report. Foster, J.E., J. Greer and E. Thorbecke (1984). A Class of Decomposable Poverty Measures, Econometrica (May); 52, 761-6. Gertler, P.J. (2004). Do Conditional Cash Transfers Improve Child Health? Evidence from PROGRESAs Control Randomized Experiment, American Economic Review Papers and Proceedings 94: 2 (May); 336-341. Glewwe, P. and H. Jacoby (1995). An Economic Analysis of Delayed Primary School Enrollment and Childhood Malnutrition in a Low Income Country, Review of Economics and Statistics 77: 1 (February); 156-169. Hoddinott, J. and M. Adato (eds.) (2009). Conditional Cash Transfers Programs in Latin America: A Synthesis of Their Impacts on Education, Health and Nutrition, Washington, DC: International Food Policy Research Institute. Knowles, J.C. and J.R. Behrman (2004). A Practical Guide to Economic Analysis of youth Projects, Philadelphia, PA: University of Pennsylvania and Bangkok, for World Bank, mimeo. Knowles, J.C. and J.R. Behrman (2005). Economic Returns to Investing in youth, in J.R. Behrman, B. Cohen, C. Lloyd and N. Stromquist, eds., The Transition to Adulthood in Developing Countries: Selected Studies, Washington, DC: National Academy of Science-National Research Council; 424-490. Martorell, R. (1997). Undernutrition During Pregnancy and Early Childhood and Its Consequences for Cognitive and Behavioral Development, in M.E. young (ed.), Early Child Development: Investing in Our Childrens Future, 39-83, Elsevier, Amsterdam. Ruiz, F. (1996). Estudio de Costos del Proyecto Integral de Desarrollo Infantil (PIDI). La Paz, Bolivia: Ruiz Guissani Consultores, mimeo. Rivera, J.A., D. Sotres-Alvarez, J.P. Habicht, T. Shamah, S. Villalpando (2004). Impact of the Mexican Program for Education, Health, and Nutrition (Progresa) on Rates of Growth and Anemia in Infants and young Children. Journal of the American Medical Association 291; 2563-2570. Schultz, T.P. (2004). School Subsidies for the Poor: Evaluating the Mexican Progresa Poverty Program, Journal of Development Economics 74:2 (June); 199-250. Thomas, D., and J. Strauss (1997). Health and Wages: Evidence on Men and Women in Urban Brazil, Journal of Econometrics 77: 1 (March); 15987. Thomas, V., y. Wang, X. Fan (2001). Measuring Education Inequality: Gini Coefficients of Education for 140 countries, 1960-2000. World Bank Policy Research Working Paper 2525, later published in Journal of Education Planning and Administration 17 (1); 5-33. United Nations Childrens Fund (UNICEF) (1998). The State of The Worlds Children 1998. New york: Oxford University Press. WIID 2.0a. (2005). World Income Inequality Database V 2.0a, June 2005. Helsinki: World Institute for Development Economics Research. WDI (2005). World Development Indicators Database. Washington, D.C.: World Bank.
Estudios de Economa. Vol. 38 - N 1, David Bravo, Jos A. Valderrama Torres The impact of income adjustments /Junio 2011. Pgs. 43-65
43
The impact of income adjustments in the Casen Survey on the measurement of inequality in Chile
El impacto de los ajustes de ingresos realizados en la Encuesta casEn sobre la medicin de la desigualdad en chile*
David Bravo** Jos A. Valderrama Torres*** Abstract The adjustment of the information obtained from household surveys to make the figures compatible with National Accounts is a non-standard and potentially questionable practice given that it alters the structure of income distribution. This paper analyzes the sensitivity of inequality and poverty indicators to the adjustments made by ECLAC so as to enable a consistency between what is reported by the CASEN survey and the National Accounts figures in Chile. The results reveal that this leads to important changes in the top-end of the distribution and to an overestimation in the main inequality indicators in Chile. Chile looks more unequal in international relative terms due to this adjustment. Key words: Inequality, Poverty, Income adjustment, Chile. Resumen La prctica de ajustar los ingresos de datos provenientes de encuestas de hogares para que las cifras que provengan de stas sean compatibles con las Cuentas Nacionales es una prctica que no sigue estndares internacionales y potencialmente criticable por alterar la estructura de la distribucin de ingresos. En este artculo se analiza la sensibilidad de los indicadores de desigualdad y pobreza ante el ajuste realizado por la CEPAL para que exista consistencia
The authors would like to thank Henry Espinoza, Osvaldo Larraaga, Gustavo Yamada, Patricia Medrano, Esteban Puentes, and the participants at the Chilean Public Policies Society Meeting. We also thank MIDEPLAN for the information provided and Juan Carlos Feres (ECLAC) for valuable conversations. David Bravo thanks the Millennium Science Initiative (Project P07S-023-F) for financial support. The usual disclaimers apply. ** Centro de Microdatos, Departamento de Economa, Universidad de Chile. E-mail: dbravo@ econ.uchile.cl *** Direccin de Calidad del Gasto. Ministerio de Economa y Finanzas, Per. E-mail: jvalder@ulima.edu.pe
44
entre lo reportado por las Encuestas CASEN y las cuentas nacionales en Chile. Los resultados revelan que dichas imputaciones provocan cambios importantes en la cola superior de la distribucin, generando con ello una sobreestimacin en los principales indicadores de desigualdad de Chile. Esta situacin afecta las comparaciones internacionales mostrando a Chile ms desigual en trminos relativos, por cuanto el ajuste de ingresos no es realizado por las cifras oficiales de otros pases. Palabras clave: Desigualdad, Pobreza, Ajuste de ingresos, Chile. JEL Classification: C81, D3, I32, N36, O15.
1. Introduction The official figures in Chile reveal an important reduction of poverty, a tendency that can be observed from the beginning of the 90s and in the 19902006 period. According to official sources, poverty was at its lowest level in 2006. Extreme poverty or indigence in 1990 was six times higher than poverty observed in 2006, whereas poverty in general more than tripled this figure (see Figure 1). On the other hand, inequality has remained largely unchanged in the period 1990-2003. Only in 2006 there was a statistically significant drop in the indicators (see Figure 2)1. FIGUrE 1
POVErTY rATE IN CHILE
50% Indigence 40% 30% 20% 10% 0% 1990 1992 1994 1996 1998 2000 2003 2006 Poverty
See Bravo and Contreras (2004), Bravo and Marinovic (1997) and Larraaga (2001) for an analysis of wage inequality over a longer period of time. They show that inequality has been rather unstable with a high increase in wage inequality between 1974 and 1987.
The impact of income adjustments / David Bravo, Jos A. Valderrama Torres
45
FIGUrE 2
INCOME INEqUALITY IN CHILE (Gini Coefficient)
57%
55%
53%
51% 1990 1992 1994 1996 1998 2000 2003 2006
The official source for calculating inequality and poverty indicators in Chile is the National Socioeconomic Characterization Survey (CASEN) while the variable used is household income. Compared with common practices applied elsewhere, the information in Chile differs in that the data is adjusted to make the accumulated amounts consistent with those registered in the National Accounts. This adjustment, trying to correct for under-reporting, affects laborrelated income of specific occupational categories, social security receipts and also has an impact on the self-reported implicit rent from own-housing. An additional amount is imputed solely to income recipients in the richest quintile to account form capital income or property income in National Accounts. It is natural then to question the effect of these adjustments on the official inequality and poverty figures. The only study on this topic we are aware of is Pizzolito (2005). She found that the trends on inequality and poverty for the 1990-2000 period are not changed when ignoring the National Accounts adjustment (she had to use adjustment factors for aggregate items available for 1990 and 1996 instead of 2000). However, she warns that international comparisons for Chile using this data could be affected. Unfortunately, despite requests by researchers over time, the original unadjusted microdata have not been made available by Mideplan. There are at least two reasons why analysts would prefer to use unadjusted information. First, since adjustment depends on how National Accounts behave over time and the income distribution figures have a component that is sensible to the National Accounts figures and that could distort the real behavior of indicators derived from income distribution, such as inequality and poverty levels.
46
Second, the practice of imputing this kind of income is precisely one of the reasons why the information is not comparable on a global scale2 since this practice is neither recommended as an international standard nor used in other Latin American countries other than Chile. This also questions the veracity of the ECLAC statistics that indicate that inequality in Chile is higher, for example, than Peru, Mexico and Argentina (see Figure 3). FIGUrE 3
INEqUALITY IN LATIN AMErICA MEASUrED BY THE GINI COEFFICIENT Bolivia Brazil Honduras Colombia Nicaragua rep. Dominicana Panama Guatemala Paraguay Ecuador Chile Argentina Mexico Peru El Salvador Costa rica Venezuela Source: ECLAC (2007). 0.614 0.602 0.587 0.584 0.579 0.578 0.548 0.543 0.536 0.526 0.522 0.510 0.506 0.505 0.493 0.478 0.441
This paper analyzes the impact of adjustments practices based on National Accounts on the income distribution in Chile. Our study deepens Pizzolito (2005) estimating unadjusted data for 2000, 2003 and 2006 and using an array of inequality indicators as well as official poverty lines. The study is structured in five sections. In the second part we outline the methodological aspects that justify the adjustment of income to National Accounts and discuss its disadvantages. In the third section we look into the sources of information in Chile used to calculate the indicators and compare them to universally accepted methodologies used to calculate these same indicators. In part four we analyze how sensitive Chilean inequality indicators are when they are not based on adjustment to National Accounts. We also analyze Chile with strictly comparable data from Peru. Section five is dedicated to discussion and conclusions. An appendix is also included.
2
Other causes that affect international comparisons are how variables are defined (consumption or income) and the scope of the surveys (nationally representative; considering only urban areas, etc.).
47
2. Methodological Issues Although ideally the microeconomic (household surveys) and macroeconomic (National Accounts) statistics should be related, in practice, the construction of National Accounts figures is not based on information from relevant units (such as households), leading to serious discrepancies in these statistics. Similarly, ravallion (2001), referring to household consumption expenses, claims that the figures referring to those reported in national accounts are rarely based on household survey results and, in the best of cases, only some components of these are taken into account. These discrepancies have led the National Accounts System for 1993 to point out that the results of household income surveys should be adjusted to compensate for certain typical biases to make them compatible with the National Accounting figures. The Canberra Group (2001) has also pointed out that it is necessary to make these adjustments: It undoubtedly causes considerable harm to users when two sets of statistics known as household income produce very different results which will, in turn, have a significant impact on social policies. Despite this need, national statistics offices rarely look to reconcile their results. In an attempt to resolve this problem and under the assumption that the National Accounting information offers more trustworthy figures than household surveys, a variety of methodologies have been developed to adjust the information to the National Accounts figures. In any case, the methodologies depend substantially on the assumptions of each author in relation to the rules of allocation that give rise to the distribution of the value of the differences between the total income of households reported by income surveys among the households or groups of households and the National Accounting. The above mentioned allocation rules have been associated with the level of household income, composition by household income sources (wage-earning, self-employment, etc.) or other combinations of the level of household income and its composition according to different sources. Using the level of total household income as a guide, the simplest way of adjusting consists of correcting the figures by multiplying a constant factor corresponding to the ratio of household income for all the households in all the income levels according to National Accounts and the aggregate income of the surveyed households. The methodology suggested by Altimir (1987) is rather more elaborated since it emphasizes the discrepancies by income source as the cornerstone of the adjustment. Altimirs proposal, which the Economic Commission For Latin America and the Caribbean (ECLAC) has based its calculations, consists of adjusting the income of every household according to its composition, using specific adjustment factors for every income source; independently of the level of income of the household, except in what corresponds to property income, which has a zero adjustment factor for 80% of the households with less income and higher than one for 20% of richer households. The adjustment factors by source are obtained from dividing the total income reported for every category
48
of income from the National Accounts with those corresponding to the household survey. Altimir attributes the differences between the household survey and National Accounts to under-reporting from the survey participant, either voluntary or involuntary, and assumes that this measurement bias is more associated with the type of income than with the level of income. It also supposes that underreporting for each type of income can be estimated according to the discrepancy between what is reported by the survey and what is reported by the National Accounting figures, with the exception of the cases where the latter is less than the first. If what is reported in the survey is higher than what is shown in National Accounting, and there is no clear evidence of overestimation due to design, then the figures shown by the survey should be accepted as true. Using the Altimir method, two households with the same level of income might undergo different adjustments in magnitude if the types of income are different. Consequently, this method affects the structure of the income distribution and can increase or decrease inequality and poverty, depending on the composition of household income and on the specific adjustment factors for each income source. What indeed becomes clear is that, when considered as an isolated factor, the treatment of capital income tends to increase inequality, since the adjustment in this category is carried out only in 20% of the richest households. In addition to the selection of the percentile from which the adjustment factor for this variable (capital income) stops being neutral, other questionable characteristics of this type of adjustment is the fact that its application supposes compliance with the following assumptions: 1. That the incomes informed by the National Accounts are at least as credible as those from household income surveys. 2. That the differences between both sources are fundamentally due to problems of under-reporting (many of the survey participants say they have a lower income than the one they actually do) and not to problems of truncation (the richest households are not surveyed). 3. That there is a rule of ideal allocation that allows to distribute household income, at a macroeconomic level, to the (expanded) income of each household pertaining to the income survey sample (microeconomic level). The first assumption implies that the figures for household income of the National Accounting system are closer to reality, since they come from a wide variety of sources and that their design makes them necessarily compatible with the rest of the components of the accounting system. On the other hand, the second assumption refers to the fact that household survey participants report less income than they really receive and that this is the only reason behind the discrepancy between the total household income reported by the surveys and the income reported by the National Accounts. This means that it is assumed that both sources refer exactly to the same population. Nevertheless, in countries where the real inequality in income distribution is so high that an extremely small fraction of population participates with a notably high proportion of total income, it is unlikely that these few people and their incomes are represented in the survey sample. This, because in the
49
survey samples the probability of selection of households with extraordinarily high income is practically zero and also because the stratification of the sample is based on demographic variables estimated from population censuses and not based on income. Consequently, if a group of population, small in number but important insofar as income is concerned is not represented in the survey sample, the entire value of their expanded income, even without sub-reporting, must be lower than the one shown in the National Accounting figures, which, because of its methodology and coverage includes in principle the income of all the recipients, without exceptions. When defined this way, the truncation in the top end of income distribution is a characteristic of the sampling that might explain one part of what would be considered as under-reporting. This implies that if the adjustment to the National Accounts figures is made in such a way that it does not differentiate between the two components of the discrepancy, we would be statistically redistributing a higher quantity of money than the one in question between the sample households. That is to say, we would be artificially distributing the income of households that are so rich that it is highly unlikely that they would appear in the survey (the richest segment) in relation to the rest of the population, including some of the poorest households than do appear in the survey. This statistical correction of the figures, without compensating the real income received by all households, can lead to underestimations when measuring poverty. Taking into account the above, the compliance with the second assumption is not credible because we cannot distinguish which part corresponds to subreporting and which part to truncation. If we were able to distinguish which part of the discrepancy corresponds solely to sub-reporting, the adjustment according to the National Accounts would consist of re-assigning only this part among the households. Nevertheless, the distinction between the amount stated in sub-reports and the amount that corresponds to the difference for truncation is not negligible and it would need at least a sample representative of those who have a higher income. Finally, regarding the suppositions that refer to the allocation rule, by design all the available allocation rules are subjective. That is why any income distribution resulting from an adjustment to National Accounts figures is only a probable distribution whose verisimilitude depends on the validity of the assumptions that were initially chosen to make the microeconomic allocation of the macroeconomic discrepancy. Measurement error is a serious concern when using micro-level cross-sectional or longitudinal survey data. Therefore, a National Accounts adjustment like the one used in Chile could easily introduce additional noise to the original microdata when unadjusted records are not made available. To face mis-reporting in surveys one could develop specific studies to understand the direction and size of the biases (see Angrist and Krueger, 1999). 3. Data The Socio-Economic Characterization Survey (CASEN) is the principal survey used in Chile to diagnose and assess the incidence of social policies in
50
households and the most important programs that constitute social spending3. It is a multi-purpose survey that provides information about the socioeconomic conditions of the countrys different social sectors, its most important deficiencies, the dimensions and characteristics of poverty and income distribution of the households. Additionally, the survey reports on the coverage and profile of the beneficiaries of social programs, their monetary and non-monetary contributions to household incomes which are identical to social sectors that do not have access to the above-mentioned programs, which makes it possible to calculate the related assistance shortfalls. Such information guides the design of new projects and any modifications in benefit allocation systems to improve the focus of those that have a more selective character. The surveys have been implemented in the following years: 1985, 1987, 1990, 1992, 1994, 1996, 1998, 2000, 2003, 2006 and 2009. This is a crosssectional survey implemented regularly in November by the Ministry of Planning (MIDEPLAN). MIDEPLAN has usually hired the University of Chile for the implementation of the fieldwork and once the data has been collected the Economic Commission for Latin America and the Caribbean (ECLAC) is responsible for making the adjustments for response errors. 4. Assessing the Impact of the Income Adjustments in the Casen Survey 4.1. The process of income adjustment as carried out by ECLAC The adjustments made by ECLAC to the survey collected data come from two sources: non-response and income misreporting. To deal with the second problem, information from National Accounts provided by the Central Bank of Chile is used and, in particular, an estimation of the principal aggregates of the income and spending accounts of households prepared specially for this task. The imputation process is shown in Figure 4. The first stage of the adjustment process is the correction of information that has been omitted, that is to say, those people who have declared that they receive some type of income, but that have not declared the amount or the corresponding total. The correction process involves assigning some type of response to the group of people who say they receive a certain income, but do not assign any values. Three groups are considered in this process: 1. People who declare they are employed in a category different to a non-paid family member and who do not report the income received from a main occupation. 2. People who declare they are retired senior citizens or who receive pensions and who do not report this income. 3. Households that inhabit the homes that they own and do not declare the imputed rent.
This section is based on the methodological reports of the surveys carried out in 2003 and 2006 (Mideplan, 2005 and Mideplan 2006).
51
FIGUrE 4
CASEN 2006. INCOME ADJUSTMENT PrOCESS CArrIED OUT BY ECLAC
ORIGINAL DATA BASE REPORTED INCOMES
NON RESPONSE ADJUSTMENT
- Employed and Retired (average) - Imputed Rent from Own Housing (hot-deck)
NATIONAL ACCOUNTS ADJUSTMENT
- (Wages) X 1.1 - (Employer and Self-Employed) X 1.976 - (Social Security) X 1.126 - (Implicit Rent from Own Housing) X 0.437
Capital Incomes: only to the highest 20% it is added 0.035 x (Autonomous Income)
FINAL DATA BASE ADJUSTED INCOMES
This is enough to close the gap between CASEN and National Accounts
The adjustment process for the employed and retired persons is carried out according to the method of averages. According to this method populations with similar characteristics to persons who have not provided any answers are selected. The average income of this group is imputed to persons who did not provide any answers. In the case of housing (Income for Imputed rent), the Hot Deck methodology is used where households are selected according to their housing type and situation. In this case it is also necessary to correct the cases where a value for imputed rent is declared but the person is not the owner of the house. According to this method, in every group obtained according to the characteristics of the housing type and situation, those households that provided no response have the same distribution as those that did respond. This means that the selected variables must be related to the variable that one wants to impute thats why they must be variables related to the price of the house rental.
52
Once the base is corrected according to the non-declaration of certain incomes, it is possible to make the corresponding adjustments according to the National Accounts figures. The income is multiplied by a certain factor, so that the income figures obtained by CASEN are compatible with the information for the whole country delivered in the National Accounts System4. According to the type of income, adjustment is applied to variables related to the following: Salaries and wages Income of the employer and self-employed Social security benefits Imputed rent Property income
For all categories but the last, the adjustment is applied directly, and the only requirement is to know which factor was used and the variables related to each type of income. In the case of property/capital income the adjustment involves calculating the total capital income of the survey (rentals, interests and dividends) and the discrepancy between what is reported by CASEN and the National Accounts figures (always in per capita terms) is attributed to all the recipients of autonomous income belonging to the last quintile, in such a way that this gap is distributed proportionally to the received autonomous income. The autonomous income considered for this purpose is the one that is previously adjusted in all its components, including only the capital income that was declared. Finally, the additional income assigned as capital income imputed to people in the last quintile, corrects the value of the autonomous income. For example, for the 2006 figures, the recipients of income from the last quintile were imputed an additional income under the concept of capital income for the amount of 0.035 times the autonomous income registered in the survey. Table 1 shows the different values of the adjustment factors according to the type of income. Two characteristics stand out. First, the variability over the years and second, with the exception of the imputed rent, all other variables are underestimated by the people surveyed, which is why the adjustment factor exceeds the unit. (In the case of property income, the adjustment is additive, which explains why this factor is less than one).
The comparison is carried out at an average level of per capita income, expressed in relation to the entire population of the country, because in National Accounts the number of recipients of each income type is unknown.
53
TABLE 1
CASEN, 1990-2006 ADJUSTMENT FACTOrS
Wages and Salaries 1990 1992 1994 1996 1998 2000 2003 2006 1.208 1.071 1.071 0.990 1.004 0.957 1.000 1.010
Income from Selfemployment 1.980 1.992 1.513 2.043 1.955 1.826 1.976 1.976
Social Security Benefits 1.473 1.633 1.435 1.398 1.347 1.471 1.145 1.126
Property/ Capital Income 0.129 0.067 0.064 0.064 0.069 0.054 0.028 0.035
Imputed rent 0.664 0.548 0.475 0.454 0.439 0.449 0.437 0.437
Source: Mideplan.
4.2. Results for 2006 In this section we will review the 2006 CASEN survey to understand the impact of the mis-reporting imputations on income distribution5. We are particularly interested in investigating if the progressive character that the imputations for the implicit rent from own-housing would sufficiently compensate the regressive effects of all the other adjustments. As we have seen, in the first one there is an overestimation in what is declared and in the last there is under-reporting, which is why the net effect on the distribution is uncertain. Additionally, by decomposing the figures from these different sources, we can assess the relevance of each type of imputation when explaining total inequality. 4.2.1. Comparison between indicators with and without adjustment Table 2 shows the sensitivity of the per capita household income distribution to the adjustments to National Accounts figures. The results of the Gini coefficient indicate that the adjustment overstates this inequality index in 3.4 points (a 7%) for the total income. Nevertheless, as was expected, the most significant changes occur in the tail-end of the distributions, and reach an increase in inequality of up to 22% (as in the case of the decile ratio in the total income). The only indicator that reflects a descent in inequality as a consequence of the adjustment to the national accounts figures is the ratio of income deciles of the main occupation. This can be explained by the fact that self-employed workers are over-represented in the first 10%, and they are precisely the people for whom income is almost duplicated. This is why the ratio decreases after the adjustment as compared with the ratio without the adjustment.
All the estimates for 2006 were also applied to 2003, with similar results.
54
In the case of the poverty indicators, the extreme poverty rate changes from 3.2% with adjustment to 2.9% without adjustment; and the poverty rate would fall from 13.7% to 13.1% undoing the income adjustment. TABLE 2
CASEN 2006. IMPACT OF THE NATIONAL ACCOUNTS INCOME ADJUSTMENT With adjustment Average income ($) Main Employment Autonomous Total Gini coefficient Main Employment Autonomous Total Coefficient of variation Main Employment Autonomous Total Theil index Main Employment Autonomous Total Decile 10/decile 1 Main Employment Autonomous Total Poverty Indigence rate Poverty rate 3.2% 13.7% 2.9% 13.1% 0.003 0.006 10% 4% 3533 41 29 5232 38 24 1699 3 5 32% 9% 22% 0.687 0.615 0.569 0.609 0.553 0.483 0.078 0.062 0.086 13% 11% 18% 1.765 1.801 1.719 1.530 1.701 1.543 0.234 0.100 0.176 15% 6% 11% 0.579 0.543 0.522 0.555 0.520 0.488 0.024 0.023 0.034 4% 4% 7% 121,644 166,556 176,981 98,666 135,733 157,073 22,978 30,823 19,908 23% 23% 13% Without adjustment % variation
Variation
Note: All income is expressed in per capita terms. The calculations exclude income of domestic workers who live in their employers house. The table includes the possible zeros that might be present in some households. For the case of poverty indicators, total family income per capita has been employed.
55
4.2.2. Changes along the distribution As can be seen in Figure 5, there is an evident distortion at the tail-end of the distribution due to the adjustment. In the case of income from the main occupation and autonomous income (labor-related and non-labor related income), there is always an overestimation, especially in the lower end up to the 20 percentile and in the upper end from the 80 percentile onwards. In both cases the overestimation is higher than 20%. The picture for total income (autonomous income plus monetary transfers plus imputed rent) is somewhat different: in the first centiles the National Accounts Adjustment decreases the income; up to the 40th centile there is no significant change made by this adjustment; however starting from this centile the adjustment increases the income reaching values near 30%. From that we can conclude that the net effect of the imputation on the variable is a reduction in the income of the poorest (up to the 5th percentile of distribution), no alteration to the income of households located in the lower-middle part of the distribution curve (between the 5th and 40th centiles) and an increase in the income of those who earn more (centile 40 and upwards).
FIGUrE 5
PErCENTAGE OF DISCrEPANCY BETWEEN DISTrIBUTIONS WITH AND WITHOUT ADJUSTMENT
60
Main occupation income Autonomous income Total income
% change of average income
40
20
20 0 20 40 60 Centiles of per capita income 80 100
Note: The calculations exclude income of domestic workers who live in their employers house. Included are the possible zeros that might be present in some households.
56
If we consider that total income incorporates the imputed rent which is decreased with the adjustment and other incomes that are increased with the adjustment, there are other conclusions to be made. In the lower part of the total income distribution predominates the effect of the downward adjustments, probably explained by over-reporting of the value of the imputed rent by such households; whereas in the high part of the distribution the same effect is exceeded by other imputations of other variables, especially of the property income category, that, as we have previously stated, is imputed only to people from the last quintile of income. An additional detail from the graph that stands out is the U form shown by the autonomous income discrepancy curve and the main occupation income discrepancy curve. This can be explained by the aforementioned over-presentation of self-employed workers in the first centiles who, as we have seen, see their incomes nearly doubled by the adjustment whereas the income of wage earners scarcely changes. 4.2.3. Impact according to the significance of each component of total income The progressive or regressive character and the significance that each income source has in income inequality can be formalized by the decomposition proposed by Shorrocks (1982). According to this proposal and considering income as Y and its components are expressed generically by Y f , where Y = Y f , an indicator of the contribution of each component to inequality is given by: Sf = f *
Where f is the correlation coefficient between factor Y f , and total income Y and denotes the standard deviation. Similarly, S f is the regression slope of Y f on total income Y, where it is easy to show that S f = 1. Components with a positive value for S f have a de-equalizing contribution meanwhile components with a negative value have an equalizing contribution6. Considering the issue at hand, the exercise would be to establish the sign and the size of the contribution of the different types of imputations starting from the following identity: Y f = Yo + Dif1 + Dif2 + Dif3 + Dif4 + Dif5
For more details of the decomposition, see Shorrocks (1980). The appendix outlines the decomposition for two sources of income.
57
where Y f and Yo are the total income with all the adjustments for imputation applied and the income without any adjustment according to National Accounts figures, respectively. Other components are the changes in income caused by the different adjustments that were made which were grouped in the following five components: Dif1 Dif2 Dif3 Dif4 Dif5 : increase of labor-related income paid in cash. : increase of income due to adjustments to social security benefits. : decrease of income due to imputed rent. : increase of income due to property and capital income. : increase of income in other categories (self-consumption, previous work, etc.).
The results confirm the progressive character of the adjustment due to the rent imputed with an importance in total inequality of 2.2%. By far the most important component is the adjustment made to labor-related income paid in cash (19.9%) followed by the adjustment for property income (3.4%), imputed rent (2.2%) and finally other incomes with 1.9% and social security with 0.2% (see Figure 6). Similarly, if we only take into account the gap existing between income without adjustment and income with adjustment, we can see that imputed rent makes a negative 10% contribution whereas in cash labor-related income is the most important component in with 86% of participation (Figure 7).
FIGUrE 6
IMPOrTANCE OF THE COMPONENTS IN TOTAL INCOME FOr 2006 Shorrocks Decomposition 120% 90% 60% 30% 0% 30% Income without adjustment In cash labor income Social security benefits Imputed rent Capital income Others incomes Income with adjustment 19.9% 0.2% 2.2% 3.4% 1.9% 76.8%
Gap
100.0%
Note: Other income includes self-consumption, withdrawal of profits, previous work and other sources of income.
58
FIGUrE 7
IMPOrTANCE OF THE IMPUTATIONS IN THE 2006 GAP. Shorrocks Decomposition 120% 90% 60% 30% 1% 0% 30% In cash labor income Social security benefits 10% Imputed rent Capital income Others incomes Total gap 15% 86%
100%
8%
Note: Other income includes self-consumption, withdrawal of profits, previous work and other income.
4.3. Effects of adjustment over time Considering that the least important components when explaining inequality are social security and other incomes and that we cannot recover the databases for some of the years and the impact of other variables on the inequality of total income, we have excluded these two factors when carrying out an analysis for the trend in the 1990-2006 period Figures 8, 9 and 10 show this analysis. The comparison between the indicators shows that, with the exception of the Gini Coefficient which keeps a constant difference (7% on average), all other changes are not systematic and that the average income shows the most important changes (29% in 1990 and 6% in 2000). In the case of poverty, there is a heterogeneous behavior: before 1996 the effect of imputations on both indicators without adjustment is an underestimation that reaches its highest value in 1990 for the case of poverty, a year in which this indicator is underestimated by about 25%. After this year the indicator is overestimated reaching its highest value in 2000 for the indigent people category with 15%. In short, not having made any imputation would have reflected a bigger reduction in the levels of poverty that those presently known. On the other hand, Figure 10 shows that the component of income without adjustment has an increasing participation when explaining total inequality, going from 63% of total inequality in 1990 to 79% in 2006. At the same time, the importance of the imputations made in labor-related income has decreased moving from accounting for 27% in 1990 to 19% in 2006.
59
FIGUrE 8
INEqUALITY AND AVErAGE INCOME, rATIO BETWEEN ADJUSTED VArIABLES AND THOSE WITHOUT ADJUSTMENT
140%
130%
120%
110%
100% 1990 1992

Avg. income Theil
1994
1996
Gini
1998
2000
2003
2006
Coef. variation
Decile 10/1 ratio
FIGUrE 9
POVErTY, rATIO BETWEEN ADJUSTED AND UNADJUSTED VArIABLES
1.2 1.1 1.0 0.9 0.8 0.7 1990 1992 1994

Indigence
1996
1998
2000
Poverty
2003
2006
60
FIGUrE 10
IMPOrTANCE OF THE COMPONENTS FOr TOTAL INCOME IN THE 1990-2006 PErIOD Shorrocks Decomposition
79.7 63.4 66.6 65.6 68.3 70.1
75.0
79.4
27.1 11.2
30.4
29.4
29.2
26.2 17.4 6.4 5.1
24.5
19.4 3.4 2.2 2006
6.1
6.0 0.9 1994
5.9
2.7 2.2 2003
1.7 1990
3.1 1992
3.4 1996
2.7 1998
2.3 2000
Income without adjustment Imputed rent
In cash labor income Capital income
4.4. Comparative analysis between distributions in Peru and Chile This section compares the distributions of income in Peru and Chile using the same methodology, that is to say using per capita household income7 adjusted only for non-response as an analysis variable. The most important results appear in the area of inequality. Assuming that the order of the other countries in Figure 3 does not change, Chile becomes one of the countries with lower inequality in the region, overtaken only by Venezuela and Costa rica, whereas Peru increases its level of inequality moving from a country with a low level of inequality to one that can be considered in the middle range of the regional ranking. Due to the fact that the same methodology is used for both countries, Chile and Peru exchange positions in relation to the ranking presented by ECLAC in its statistical 2007 yearbook (see Figures 3 and 11). Table 3 shows the levels of inequality according to the sources of total income. From the table we can draw the following conclusions: it is clear that inequality of income received by self-employed workers is higher in Chile than in Peru. The situation is the opposite for wages. When we consider the net effect of both sources we see that income inequality from the main occupation is lower in Chile than in Peru.
7
In Peru the official information on poverty and inequality are estimated from the National Household surveys (ENAHO) using the per capita consumption as the variable of analysis; nevertheless the survey also allows the income per capita to be calculated.
61
FIGUrE 11
INEqUALITY IN LATIN AMErICA MEASUrED BY THE GINI COEFFICIENT Bolivia Brazil Honduras Colombia Nicaragua rep. Dominicana Panama Guatemala Paraguay Ecuador Peru Argentina Mexico El Salvador Chile Costa rica Venezuela 0.614 0.602 0.587 0.584 0.579 0.578 0.548 0.543 0.536 0.526 0.511 0.510 0.506 0.493 0.488 0.478 0.441
Source: CEPAL 2007, ENAHO 2006 and CASEN 2006.
The component others that brings together all non-labor related income generated by the household, has a more unequal distribution in Chile than in Peru; whereas autonomous income, which is the sum of the main occupation income plus the rest displays a similar behavior in both countries. Finally, considering the total income, both the Gini and the Theil coefficients indicate that inequality is higher in Peru, whereas the Variation Coefficient, more sensitive to the top-end of distribution, indicates the contrary. TABLE 3
POVErTY AND INEqUALITY INDICATOrS FOr PErU AND CHILE IN 2006
Indicator Extreme poverty: FGT0 FGT1 FGT2 Total poverty: FGT0 FGT1 FGT2
Peru (A)
Chile (B)
Difference (B) (A)
19.8% 6.8% 3.2%
3.1% 1.0% 0.6%
0.17 0.06 0.03
43.4% 17.2% 9.1%
13.8% 4.3% 2.1%
0.30 0.13 0.07
62
Estudios de Economa, Vol. 38 - N 1 Peru (A) Chile (B) Difference (B) (A)
Indicator Gini coefficient Self-employment Income Salaries Main Occupation Income +rest Autonomous Income +Subsidies +Imputed rent Total Income Coefficient of Variation: Self-employment Income Salaries Main Occupation Income +rest Autonomous Income +Subsidies +Imputed rent Total Income Theil index: Self-employment Income Salaries Main Occupation Income +rest Autonomous Income +Subsidies +Imputed rent Total Income
0.74 0.73 0.59 0.61 0.52 0.70 0.72 0.511
0.89 0.61 0.56 0.74 0.52 0.82 0.61 0.488
0.15 0.13 0.03 0.13 0.00 0.12 0.10 0.02
2.57 2.20 1.65 2.19 1.52 4.31 2.42 1.480
4.01 1.64 1.54 4.03 1.71 2.63 1.54 1.550
1.44 0.56 0.11 1.84 0.19 1.68 0.88 0.07
1.17 1.11 0.67 0.80 0.53 1.11 1.09 0.515
2.03 0.73 0.61 1.27 0.56 1.52 0.73 0.487
0.86 0.39 0.06 0.47 0.02 0.41 0.36 0.03
Note: FGT is the Foster-Greer-Thorbecke metric, a generalized measure of poverty. If z: poverty line; N: the number of people in the country; H: the number of poor (those with incomes at or below z); yi : individual incomes; = a sensitivity parameter. Then: FGT = 1 N
z yi z i =1
H
FGT0: FGT if = 0 (the Headcount ratio or poverty rate used in Chile). FGT1: FGT if = 1 (the average poverty gap measuring intensity of poverty). FGT2: FGT if = 2 (an index that combines information on both poverty and income inequality among the poor). See Foster, Greer and Thorbecke (1984).
63
5. Discussion and Conclusions This paper has shown that inequality indicators and poverty in Chile are overestimated by virtue of the imputations for adjustment to National Accounts figures. In the case of inequality, the overestimation is nearly 22% (as in the case of the ratio between the deciles in 2006), whereas the most well-known indicator, the Gini coefficient, is overestimated in about 7%, an overestimation that is a constant in the period 1990-2006. In the case of poverty, both extreme and non-extreme, the overestimation is about 10% and 4% respectively, which can be explained by the downward adjustment of the value of imputed rent, the only variable that has an imputation in this direction (and that according to the results would have a higher net impact on the households with a lower income). Additionally, poverty is the only characteristic that has displayed a heterogeneous behavior in the 19902006 period: before 1996, when the levels of poverty were higher than the ones observed in the last few years, the levels of poverty were underestimated and after 1996 they are overestimated. This allows us to conclude that if there had been no adjustment applied to income, the reduction of the poverty in Chile in the period under analysis would have been more pronounced. The average income was overestimated in each year analyzed and this variable experienced the most important temporary changes, given that it moved from an overestimation of 29% in 1990 to 9% in 2006. Once the structure of the imputations according to national accounts figures was known, the CASEN figures were deconstructed and the same methodology was applied both to Peru and Chile so as to obtain poverty and inequality indicators that would enable an adequate international comparison. The results indicate that income inequality is higher in Peru than in Chile. It is important to point out that although we can recover the original information, when and if all the adjustment factors and variables involved in the adjustment are known (which does not happen in certain years), for the sake of transparency and efficiency, since the process of recovery is not negligible, the official databases should include the non-adjusted variables. This is especially important when making international comparisons, since the practice of income adjustment is not universal. Furthermore, the advantage of congruence between the micro and the macro accounts can be overlooked for at least two reasons: First, changes in the social indicators can reflect the adjustments applied and not necessarily the real dynamics of the socioeconomic indicators. Second, the fact that the National Accounts figures and the Household Surveys refer to different universes since the sample of people with a higher income is typically underestimated in any conventional household survey (truncation in the top-end of distribution), makes adjustment between the macro and the micro accounts necessary. Since this requires specific knowledge of what part of the discrepancy of the Household Survey with the National Accounts corresponds to sub-reporting and what part corresponds to truncation and since this is not known, the allocation to the households in the survey of the total discrepancy leads to an over-correction of income in all or part of the households surveyed.
64
APPEnDIx
Shorrocks Decomposition: In its simplest version (which can be generalized more easily) when the total income (YT) has two sources, such as (YT = Y1 + Y2), we can deduce that: Var(YT) = Var(Y1) + Var(Y2) + 2Cov(Y1,Y2) Var(YT) = Var(Y1) + Cov(Y1,Y2) + Var(Y2) + Cov(Y1,Y2) Var(YT) = Cov(Y1,YT) + Cov(Y2,YT) 1 1 = [Cov(Y1,YT) / Var(YT)] + [Cov(Y2,YT) / Var(YT)] = S1y + S2y
References Altimir, O. (1987). Income Distribution Statistics in Latin America and their reliability. Review of Income and Wealth, Series 33, N 2, June. Angrist, J. and A. Krueger (1999). Empirical Strategies in Labor Economics, in O. Ashenfelter and D. Card, Handbook of Labor Economics, Volume 3, Chapter 23. Elsevier Science B.V. Bravo, D. and A. Marinovic (1997). Wage Inequality in Chile: 40 Years of Evidence. Departamento de Economa, Universidad de Chile, August. Bravo, D. and D. Contreras (2004). La distribucin del ingreso en Chile 19901996: anlisis del impacto del mercado de trabajo y las polticas sociales, in Banco Interamericano de Desarrollo, Reformas y Equidad Social en Amrica Latina y el Caribe, abril. Canberra Group Expert on Household Income Statistics (2001). The Canberra Group, Final Report and Recommendations (Ottawa). Chez-Crespo, S. (1972). El tratamiento de preguntas de carcter ntimo: modelo de respuesta aleatorizada. Revista Estadstica Espaola, Instituto Nacional de Estadstica N 55. Perodo: Abr.-Jun.-72. Commission of the European Communities, International Monetary Fund, OECD, United Nations and World Bank (1993). System of National Accounts 1993, Brussels/Luxemburg, New York, Paris, Washington D.C. ECLAC (2007). Anuario Estadstico de Amrica Latina y el Caribe. Comisin Econmica para Amrica Latina y el Caribe. Foster, J., J. Greer and E. Thorbecke (1984). A class of decomposable poverty measures. Econometrica 52, 761-776. Greenberg, B., r. Kuebler Jr., J. Abernathy and D. Horvitz (1971). Application of the randomized response Technique in Obtaining quantitative Data.
65
Journal of the American Statistical Association, vol. 66, nm. 334, 243-50. INEI (2004). Ficha Tcnica ENAHO 2003, INEI, 1-18. Larraaga, O. (2001). Distribucin de Ingresos en Chile: 1958-2001. Documento de Trabajo N 178, Departamento de Economa, Universidad de Chile. Mideplan (2005). Marco metodolgico CASEN 2003, Mideplan, 1-25, (2005). Mideplan (2006). Marco metodolgico CASEN 2006, Mideplan, 1-97, (2006). Ministerio de Economa y Produccin (MEP) de Argentina (2007). Comunicado de prensa: Distribucin Funcional del Ingreso Cuenta de Generacin del Ingreso e Insumo de mano de obra. Anexo B Correcciones de ingresos a la Encuesta Permanente de Hogares, 2007. Pizzolito, G. (2005). Poverty and Inequality in Chile: Methodological Issues and a Literature review. Centro de Estudios Distributivos, Laborales y Sociales 20: 1-30. ravallion, M. (2001). Measuring Aggregate Welfare in Developing Countries: How well do National Accounts and Surveys Agree?. Working Paper, The World Bank, Washington, D.C. roca, J. and M. Hernndez (2004). Evasin Tributaria e Informalidad en el Per: una Aproximacin a Partir del Enfoque de Discrepancias en el Consumo. CIES. Shorrocks, A.F. (1980). The class of additively decomposable inequality measures, Econometrica 48 (1): 613-625. Shorrocks, A.F. (1982). Inequality decomposition by factor components, Econometrica 50 (1): 193-211.
Estudios de Economa. of human capital / Douglas McKee, Petra The longer-term effectsVol. 38 - N 1, Junio 2011. Pgs. 67-100 E. Todd
67
The longer-term effects of human capital enrichment programs on poverty and inequality: Oportunidades in Mexico*
Los efectos a largo plazo de programas de incremento en el capital humano sobre la pobreza y la desigualdad: Oportunidades en Mxico
Douglas McKee** Petra E. Todd*** Abstract Previous empirical research has shown that Mexicos Oportunidades program has succeeded in increasing schooling and improving health of disadvantaged children. This paper studies the programs potential longer-term consequences for the poverty and inequality of these children. It adapts methods developed in DiNardo, Fortin and Lemieux (1996) and incorporates existing experimental estimates of the programs effects on human capital to analyze how Oportunidades will affect future earnings of program participants. We nonparametrically simulate earnings distributions, with and without the program, and predict that Oportunidades will increase future mean earnings but have only modest effects on poverty rates and earnings inequality. Key words: Oportunidades, Human capital, Schooling, Health, Poverty, Inequality. Resumen La investigacin emprica previa ha mostrado que el programa mexicano Oportunidades ha tenido xito en aumentar la escolaridad y mejorar la salud de nios desfavorecidos. Este artculo estudia las potenciales consecuencias de largo plazo en la pobreza y desigualdad que afectan a estos nios. Adapta modelos desarrollados por DiNardo, Fortin y Lemieux (1996) e incorpora estimadores experimentales existentes de los efectos del programa en el capital
This paper was presented at the 2008 UNDP (United Nations Development Program) Conference on Inequality in Latin America and at the 2008 annual meeting of the Population Association of America. We thank Jere Behrman and Estela Rivero-Fuentes for helpful comments. ** Post-doctoral scholar at the Economic Growth Center at Yale University. *** Professor of Economics at the University of Pennsylvania.
68
humano para analizar cmo Oportunidades afectar los ingresos futuros de los participantes del programa. Simulamos de manera no-paramtrica las distribuciones de ingreso, con y sin el programa, y predecimos que Oportunidades aumentar los ingresos medios futuros pero slo tiene efectos modestos en la tasa de pobreza y la desigualdad de ingresos. Palabras clave: Oportunidades, Capital humano, Escolaridad, Salud, Pobreza, Desigualdad. JEL Classification: H50, I00, J24, O12, O54, O15. 1. Introduction In recent years, governments in many Latin American countries have adopted conditional cash transfer (CCT) programs as a primary strategy for alleviating poverty and stimulating investment in human capital. These programs typically provide cash grants to poor families if they send their age-eligible children to school and subsidies for regularly visiting health clinics. Mexico and Brazil first adopted CCT programs in the 1990s. Since then, programs with similar incentives have been introduced in Argentina, Chile, Colombia, Costa Rica, El Salvador, Ecuador, Honduras, Nicaragua, Peru and Uruguay1. The Mexican Oportunidades program (formerly called PROGRESA) was rigorously evaluated using both experimental and non-experimental evaluation designs. In the first two years (1998-1999) of its implementation in rural areas, the program was evaluated using a place-based social experiment that randomized 506 villages in or out of the program. The experimental results demonstrated statistically significant program impacts on increasing schooling enrollment and attainment, reducing child labor, improving health and nutrition outcomes and reducing poverty2. Partly on the basis of these observed positive program impacts, the Mexican government expanded the program into urban areas in 2002. By 2005, the program covered five million families and had an annual budget of U.S. $2.1 billion. A non-experimental evaluation carried out in urban areas found statistically significant program impacts similar in magnitude to those found in rural areas. As noted, previous evaluation studies of the Oportunidades program documented the programs short-term impacts. This paper takes as a point of departure the observed impacts on education and nutrition and estimates the effects of these changes on the future earnings distributions of the children currently participating in the program. The question we consider is how the programs
1 2
Some similar programs also have been introduced in Asian countries, such as Bangladesh and Pakistan. See, e.g., Schultz (2000,2004), Gertler (2000), Behrman, Sengupta and Todd (2005), Parker and Skoufias (2000), Buddelmeyer and Skoufias (2003), Todd and Wolpin (2006) and Freije, Bando and Arce (2006).
The longer-term effects of human capital / Douglas McKee, Petra E. Todd
69
impacts on human capital, as measured by years of schooling attained and increases in height (interpreted as an indicator of long-term nutritional status), will affect future earnings inequality and poverty of the younger generation. In the last decade, Mexico has ranked among the countries in Latin America with the highest income inequality. A study by Lopez-Acevedo (2004) finds that educational inequality accounts for the largest share of Mexicos earnings inequality, suggesting that human capital enrichment programs could be an effective instrument for reducing inequality. Freije, Bando and Arce (2006) show that Oportunidades has significantly decreased the poverty rate among the current generation of recipients, but little is known about the longer-term effects of the program on poverty. Our approach to simulating program impacts on earnings distributions adapts for use in program evaluation a nonparametric decomposition method originally developed in DiNardo, Fortin and Lemieux (1996) and extends this method to allow for probability mass at zero in earnings distributions. Existing micro-simulation approaches for predicting effects of conditional cash transfer programs have focused on the short-term and are mainly based on parametric modeling frameworks3. The parametric models can be quite rich, but they typically impose strong functional form assumptions. The goal in microsimulation studies is usually to forecast the effects of programs prior to their implementation, whereas our aim is to understand how program impacts that have already been estimated will affect future earnings and poverty. Our approach is fully nonparametric and does not impose any functional form assumptions on the earnings-height-education-work experience relationship, other than continuity and differentiability. We find evidence of nonlinearities in the estimated relationship that shows the benefits of flexibility with regard to model specification. We use the nonparametric simulation method to compare the earnings and employment distributions with and without the program and to compare our inferences to those that would be obtained using more standard parametric approaches. The key implications from the analysis are that the programs impacts on education and height will increase mean future earnings of beneficiaries but will likely have little impact on earnings inequality. The modest overall observed impacts on inequality are attributable to two main factors. First, the program targets children from poor family backgrounds, and family background is an imperfect predictor of future earnings. Children from poor backgrounds ultimately get distributed throughout the adult earnings distribution due to substantial intergenerational mobility. Second, we find important nonlinearities in the relationship between earnings, education and height, the most notable being that the returns to education are greater for post-primary years of education. Such nonlinearities imply that people who would obtain higher levels of schooling in the absence of the program tend to benefit more from the intervention, which contributes to widening rather than lessening inequality.
3
See, e.g., Freije, Bando and Arce (2006) and Bourguignon, Ferreira and Leite (2003).
70
Our empirical analysis is based on the first wave of the Mexican Family Life Survey (MxFLS-1) which was collected in 2002. The survey collected data for all members of 8,440 households and includes information about labor force participation, income for both primary and secondary jobs (including self-employment), education, and health. It also contains measures of family background, that we use to simulate program targeting. Our final analyses use a subsample of 5,171 individuals age 25 to 40. This paper proceeds as follows. Section two describes the nonparametric simulation method and how we adapt and use it to study how Oportunidades affects employment and the overall earnings distribution. Section three describes the Mexican Family Life Survey and our analysis samples. The empirical results are presented in section four. Section five concludes. 2. Methodology for Simulating Program Effects on Population Earnings Distributions The simulation method that we use to study program effects on earnings and employment outcomes is adapted from a wage decomposition method originally proposed in DiNardo et al. (1996). Their study uses the method to investigate the effects of institutional and labor market factors on changes in the U.S. wage distribution over time. Their approach writes the overall wage density at time t, fw = ( w | t ) in terms of the conditional wage densities, where conditioning is on a set of labor market or institutional factors, z, whose effects on earnings they analyze: fw ( w | t ) = fw ( w | z , t ) fz ( z | t )dz.
z
In their study, z includes variables indicating union status, industrial sector, and whether the wage falls above or below the minimum wage. Counterfactual wage densities are constructed by replacing fz ( z | t ) by a different hypothetical conditional density, gz ( z | t ). We apply the DiNardo et al. (1996) method to simulate earnings densities with and without a program intervention, where the program intervention changes the distribution of z. We extend the method to account for simultaneous analysis of both employment and earnings by permitting the earnings distribution to have a mass point at zero due to nonparticipation. In this section, we first describe the simulation approach in general terms, and then how it applies to evaluating the effects of the Oportunidades program. 2.1. Basic method Denote some outcome of interest (earnings) by y and define its density in terms of its conditional density (conditional on some observed characteristics x): f ( y) = f ( y, x )dx = f ( y | x ) f ( x )dx.
x x
71
Suppose that the program intervention changes the distribution of x from f(x) to f ( x ) but that the distribution of y conditional on x stays the same ( y | x ) = f ( y | x ) . The new unconditional distribution of y would be given f by:
f ( y) = f ( y | x ) f ( x )dx.
x
We wish to simulate the effect of the program intervention on the outcome y as it operates through changing x. For example, suppose that the variable x represents years of schooling attained and height and that the program intervention increases schooling attainment and height by some amount, i.e. x = x + x . Suppose also that we have a sample of size n drawn from the unconditional density, f(x). If we know x we can generate for each individual xi = xi + x . We can simulate the post-program earnings density f ( y) at a i point y0 by the average: 1 f ( y0 ) = f ( y0 | xi = xi + x ) i n x X
i
f ( y0 , xi ) 1 f ( x ) , n x X i i
where f ( y, xi ) and f ( xi ) are nonparametric estimators of the unconditional densities computed from the original (pre-program) sample: f ( y0 , xi ) = =
n y j y0 x j xi 1 K K f ( xi ) y x n j =1 y x
1 n x j xi K . x n j =1 x
y and x are bandwidths that are assumed to satisfy the usual requirements for consistent kernel density estimation4. The MxFLS data are a stratified sample, so sampling weights are required to reweight the sample back to population proportions. Incorporating sampling weights into the simulation method is straightforward. Assume each observation has a sampling weight, i, and that the weights are scaled so that i = n . The weights can be incorporated into the estimation of f ( y) as follows: f ( y, x i ) 1 f ( y) = i , n x x f ( xi ) i
4
We require ax 0, ay 0, as n and ayaxn .
72
and also into the estimation of the unconditional kernel densities: f ( y0 , xi ) =

n y j y0 x j xi 1 jK K y x n j =1 y x
x j xi 1 n f ( xi ) = jK . x n j =1 x For expositional clarity, we suppress the weights in the remainder of the discussion, although we incorporate them in the estimation. 2.2. Accounting for probability mass at zero Kernel density estimation can approximate well the distributions of continuous random variables, but in our data many people (especially women) report zero earnings. The program intervention could increase earnings among workers as well as change the probability of having positive earnings. We accommodate the mass point at zero in the earnings distribution by estimating the density of earnings as a mixture, where with some probability individuals earn zero and with the remaining probability they earn income drawn from the density of income conditional on its being positive, fy >0 ( y). Both the probability of having positive earnings and the magnitude of earnings are potentially affected by the program. Let y be the random variable representing the distribution of income implied by the counterfactual distribution of x. Again, we assume the distribution of y conditional on x stays constant; in other words that the density of earnings conditional on schooling attainment and height is the same whether or not the program is in place. This implies that Pr ( y = 0 | x ) = Pr ( y = 0 | x ) f ( y | x) = f ( y | x)
y> 0 y> 0
The stability assumption implicitly rules out general equilibrium effects, because it assumes that increases in the population in schooling attainment or height do not affect the earnings premium for those characteristics. We can obtain the probability of zero earnings, Pr( y = 0), with the program ( x ) ) using the following: intervention (under the counterfactual f Pr ( y = 0) = Pr ( y = 0 | x ) f ( x )dx ,
x
1 x Pr( y = 0 | i = xi + x ), n x X
i
where X is the support of xi and where
73
x j xi x ( y = 0 | x ) = x j X Pr . i x j xi K x j X x
1( y j = 0)K
In the last equation, 1(yi = 0) is an indicator that denotes whether the individual has zero earnings. Let f y>0 ( y) be the density of income conditional on its being positive. The counterfactual distribution of y conditional on y being positive is given by: f y > 0 ( y) = f y > 0 ( y | x ) f ( x | y > 0 )
x x
= fy> 0 ( y | x)
Pr ( y > 0 | x ) f ( x ). Pr ( y > 0)
We estimate the conditional density by: Pr ( y > 0 | xi = xi + x ) 1 f y > 0 ( y) = f y > 0 ( y | x i ) ( y > 0) n x X Pr

i
f y > 0 ( y, xi ) Pr ( y > 0 | xi ) 1 . n x X f y > 0 ( xi ) Pr ( y > 0) i
We estimate the conditional densities at a point (y0, x0) using the standard kernel density estimator applied to the subset of data for which income is positive: f y > 0 ( y0 , x 0 ) =
n y y x x 1 1( yi > 0)K i 0 K i 0 y x 1( yi > 0) i =1 y x i
n x x 1 f y > 0 ( x0 ) = 1( yi > 0)K i 0 . x 1( yi > 0) i =1 x i
We now have all the ingredients to simulate the post-intervention earn ings distribution. Earnings is 0 with probability Pr ( y = 0) and is drawn from f ( y) with Pr ( y > 0).
y> 0
2.3. Measures of poverty and inequality After simulating the distribution of earnings with and without actual and hypothetical program impacts, it is possible to examine the effect of that the
74
program intervention has on poverty and inequality using standard measures considered in the poverty measurement literature. Below, we briefly summarize the measures that we use in the empirical analysis as functions of the estimated densities, taking into account that densities may have probability mass at zero. For a recent discussion of the relative merits of alternative poverty and inequality measures, see Foster and Szekely (2007). Headcount Ratio. The headcount ratio is the fraction of the population below a predefined poverty line. Denote the value of the poverty line by L. HCR = Pr ( y = 0) + (1 Pr ( y = 0)) F| y > 0 ( L ) Average Poverty Gap Ratio. The average poverty gap ratio is the mean shortfall between an individuals income and the poverty line (with those above the poverty line having no shortfall) expressed as a fraction of the poverty line: APGR = Pr ( y = 0) + (1 Pr ( y = 0))
L
Ly f ( y)dy L | y> 0
Foster-Greer-Thorbecke (1984) Index. The Foster-Greer-Thorbecke (1984) index is a weighted version of the average poverty gap ratio that gives more weight to poorer individuals:
L L y FGT = Pr ( y = 0) + (1 Pr ( y = 0)) f ( y)dy 0 L | y> 0 2
Coefficient of variation. The coefficient of variation is a common measure of dispersion of a distribution, defined as coef of variation = Var[ y] E[ y] where
0
E[ y] = Pr ( y = 0) 0 + (1 Pr ( y = 0)) y f| y > 0 ( y)dy = (1 Pr ( y = 0)) y f| y > 0 ( y)dy

0
Var[ y] = Pr ( y = 0)E[ y]2 + (1 Pr ( y = 0)) (y E[ y])2 f| y > 0 ( y)dy.

0
Inter-quantile ranges. Another common measure of the dispersion of a distribution is the interquartile range. The differences between quantiles of y can be computed directly from the empirical cdf:
75
F ( y) = Pr ( y = 0) if y = 0 = Pr ( y = 0) + (1 Pr ( y = 0)) F| y > 0 ( y) if y > 0 Gini Coefficient. The Gini coefficient is widely used as a measure of inequality of a distribution of income. Its values range between 0 and 1, with 0 corresponding to perfect equality and 1 corresponding to perfect inequality (one person has all the income). G = 1 1 (1 F ( y))2 dy E[ y] 0
Theil Entropy Coefficient. The Theil entropy coefficient can be computed from a set of observations by: T= 1 n yi yi ln y n i =1 y
If everyone has the same (i.e., mean) income, then the index equals 0. If one person has all the income, then the index equals ln n. Taking the limit as n , we get the following formula in terms of the density, conditional on y > 0: T| y > 0 =
0
y y f ( y)dy log E[ y] E[ y] | y > 0
Generalizing this measure to the case where there can be probability mass at 0 gives the following: T = (1 Pr ( y = 0))
0
y y log f ( y)dy E[ y] E[ y] | y > 0
Below, we report how the program affects each of these alternative measures of poverty. 2.4. Applying the simulation method to evaluation of Oportunidades We next describe how the nonparametric simulation method is applied in the context of evaluating Oportunidades. y represents labor earnings, and is modeled as a function of three covariates: e denotes years of schooling attainment, h denotes height in centimeters (a measure of long-term nutritional status), and x denotes years of potential labor market experience5. The conditional density of labor market earnings is
The MxFLS data contain information on recent labor histories, but these are not long enough to construct a measure of actual experience. For this reason, we use the standard Mincer potential experience measure: Age minus years of schooling minus 6.
76
f ( y | e, h , x ) The overall income distribution integrates over the observed schooling attainment, height and experience distribution in the population: f ( y) =
(e ,h , x )A
f ( y | e, h, x )dFe,h , x (e, h, x ).
The Oportunidades program is known to impact schooling attainment levels (e) and height (h) and we want to assess how these impacts translate into changes in the earnings distribution. If participation in the program was universal, we could nonparametrically simulate the effect of the program on the income distribution simply by augmenting schooling attainment and height values by the expected program impacts. Let e denote the expected impact on schooling attainment and h the impact on height. f ( y) =
(e ,h , x )A
f ( y | e, h, x )dFe,h , x (e + e , h + h , x | (e, h, x ) S ).
Because nonparametric estimation methods do not extrapolate beyond the observed support (A), this simulation can only be performed for the subset of people for whom (e + e , h + h , x ) A, which we denote by S. The above equation assumes that everyone experiences a program effect of the magnitude (e , h), but Oportunidades was targeted to a subset of the population based on poverty-related criteria that are discussed in detail below. Let D = 1 for the subset of individuals targeted by the program and D = 0 for those not targeted. The overall income distribution that results, g(y), reflects that of the combined targeted and nontargeted subgroups: g( y) = Pr( D = 0) f ( y | D = 0) + Pr( D = 1) f ( y | D = 1) Suppose the nontargeted subgroup experiences no effect of the program6. The larger the subgroup targeted by the program (Pr(D = 1)), the larger will be the potential effect on the overall earnings distribution. Using this methodology, we can explore the relative contribution of impacts on schooling attainment and height in changing the overall income distribution, by considering the case where (i) we set e = 0 and the only effect is through h , and (ii) where h = 0 and the only effect comes through e . Implementing the simulation estimator of the previous section requires nonparametrically estimating the conditional density f ( y | e, h, x ) and the unconditional density fe,h , x (e, h, x ). We estimate the latter using a three dimensional kernel density estimator:
This assumption rules out spillover effects of the program onto the nontargeted population. See Angelucci et al. (2008) for a discussion of potential spillover effects of Oportunidades.
77
fe,h , x (e0 , h0 , x 0 ) =
1 nae ah ax
K
i =1
ei e0 hi h0 xi x 0 K K , ae ah ax
where ae , ah and ax are the bandwidth choices. In our analysis below, we use a Gaussian kernel and apply Silvermans rule for univariate distributions to each dimension of the data (Silverman, 1986). We also experimented with other bandwidth choices and found our main results were quite robust. To estimate the conditional density f ( y | e, h, x ) , observe that the conditional density can be expressed as the ratio of two joint unconditional densities: f ( y | e, h , x ) = f ( y, e, h , x ) , f (e, h, x )
each of which can be nonparametrically estimated by standard kernel density estimators. The convergence rate of pointwise nonparametric density estimators slows down as the dimensionality increases, a problem known as the curse of dimensionality. However, the proposed estimators average over the nonparametric estimates and therefore converge at a faster rate. 3. Description of the Analysis Subsamples We analyze data from the Mexican Family Life Survey (MxFLS-1), which conducted interviews with 8,440 households in 150 communities in 2002. Every household member age 15 or older was interviewed, yielding about 38,000 individual interviews. 16 of Mexicos 32 states/districts are represented (roughly 70% of the population resides in these states). Weights are provided to make the sample nationally representative. The survey includes comprehensive information on employment and income for both primary and secondary jobs in the formal and informal sectors. The survey also includes information on household structure, education, and health. The key variables used in simulating counterfactual outcomes are income, employment, education level, height and labor market experience. Appendix A describes in more detail how we construct each of these variables from the data. Table 1 presents descriptive statistics for our two main analysis samples: Adult men and women age 25 to 40. About 10% of men and 64% of women report zero labor income. Mean monthly earnings for males are 3,945 pesos and for women 1,140 pesos, where the means include zeros for nonworkers7. The average education level for men is 8.8 years, which is about one year higher than the average for women of 7.7 years. Men are on average 166 centimeters tall, and women are on average 153 centimeters tall. The Gini coefficient for earnings
In 2002 the average daily exchange rate was 1 USD equals 9.68 pesos. Because a small number of the the earnings values seemed to be outliers, we implemented a trimming procedure and omitted all individuals who reported income higher than 40,000 pesos/ month. This corresponded to 9 of 5,180 observations or the top 0.2%.
78
of men is 0.483 and for earnings of women is 0.819. The higher coefficient for women reflects the fact that a large fraction of women in Mexico do not work, so the earnings distribution for women is more unequal than that for men8. TABLE 1
DESCRIPTIVE STATISTICS Men and Women, age 25-40
Men Proportion with zero earnings Mean monthly earnings (1000s pesos) Median earnings Interquartile range of earnings Coefficient of Variation Gini Coefficient Theil Index Headcount Ratio (FGT, = 0)* Average Poverty Gap Ratio (FGT, = 1)* Foster-Greer-Thorbecke Index (FGT, = 2)* Mean schooling level (last grade completed) Mean height (cms) Mean potential labor market experience Sample Size 0.099 3.945 (0.187) 3.000 3.300 1.123 0.483 0.443 0.227 0.148 0.126 8.8 (0.27) 166 (0.52) 17.3 (0.36) 1950
Women 0.638 1.140 (0.127) 0.000 3.600 2.276 0.819 1.459 0.763 0.702 0.681 7.7 (0.20) 153 (0.41) 18.5 (0.26) 3221
* The three poverty measures are computed using poverty line of 1,452 pesos/mth (= 5 USD/day).
4. Empirical Results We use the methods described in section two to simulate the effect of the Oportunidades program on the earnings distribution as it operates through changing education and height levels of the younger generation. We infer the relationship between earnings, education, height and labor market experience from information on adults who are age 25 to 40 population and then use that estimated relationship to draw inferences about how increases in schooling and height would affect earnings distributions. Experimental evaluations of the Oportunidades program (as well as of its predecessor, the PROGRESA program) have found that the program increases schooling attainment levels by 0.6 years
As a point of reference, most developed European nations tend to have Gini coefficients for household income between 0.24 and 0.36. For household income, the United States Gini coefficient is around 0.45 and for Mexico is 0.55 (in 2003).
79
on average and adds about one cm to height for both men and women9. We consider the following hypothetical combinations of impacts and their effect on the earnings outcome distribution: (a) an increase in schooling attainment of 0.6 years, (b) an increase in height of one cm, (c) a combined increase in schooling attainment and height in the magnitudes specified in (a) and (b), (d) an increase in schooling attainment by three years, and (e) an increase in height of three centimeters. An increase of three years of education or an increase in height of three centimeters is a very large impact that is much greater than what was observed under the program, but we include these hypothetical impacts simply for purposes of comparison. 4.1. Program targeting Our goal is to simulate the longer-term effects of Oportunidades on earnings inequality and poverty. Ideally, we would compare two groups: The treatment group would be the population targeted as children by the program observed 20 years later and the control group would be the same people in a world where the program did not exist. Unfortunately, we cannot currently observe either group. The program was implemented relatively recently (in the late 1990s), so many of the children who participated are still too young to observe their longer-term labor market outcomes. Additionally, although we can observe which families are currently participating in the program, it is likely that children from todays Oportunidades households may not themselves meet the program eligibility criteria as adults. In fact, one of the primary goals of the program is to reduce the intergenerational transmission of poverty. Our simulation is therefore based on a synthetic cohort approach that assumes stability in earnings relationships for neighboring cohorts. In particular, it assumes that individuals age 25 to 40 can be used to represent the future earnings of children in families currently participating in the program. We simulate the effects of Oportunidades by identifying the 40% of current 25-40 year-olds that would have been most likely to be targeted when young had the program been available, making use of the observed family background characteristics. We analyze the effects of the program by changing this groups observed characteristics (education, height, and potential experience) in a way that is consistent with the impacts that have been measured in recent program evaluation studies. The MxFLS-1 dataset does not contain information on all the criteria used to determine eligibility for Oportunidades, and in fact the exact eligibility criteria are not public. However, from program officials we have learned the approximate criteria and use the most closely related variables from the MxFLS-1 dataset to approximate eligibility. Specifically, we estimate a probit model for program participation using data on children (age 9 to 12) who are currently participating in Oportunidades as a nonlinear function of several variables: mothers education, fathers education, whether the household has indoor plumbing, and the number of children age 0-10 residing in the household.
See Behrman and Hoddinott (2005) for discussion of the impacts of PROGRESA on height, and Schultz (2000, 2004), Behrman, Sengupta and Todd (2005) and Todd and Wolpin (2006) for discussion of impacts on years of schooling.
80
Table 2 presents descriptive statistics for these variables. In the sample, 37% of children participate in Oportunidades. The program is most active in the poorer southern states (Chiapas, Oaxaca, Guerrero, Michoacan, and Puebla), where 31% of the children live. On average, the children in the sample have mothers with 4.7 years of schooling attained and fathers with 5.2 years. Only 46% of these children live in households with indoor plumbing. Table 3 shows the estimated coefficients from the probit model for program participation10. As expected, parental education, indoor plumbing, and the presence of young children in the household are highly significant predictors of program participation. TABLE 2
DESCRIPTIVE STATISTICS FOR VARIABLES USED IN TARGETING ANALYSIS Children Age 9-12
Children age 9-12 Participates in Oportunidades Mothers schooling Fathers schooling Maximum of parents schooling Household has indoor plumbing Number of children age 0-10 in household Lives in Poor Southern State Sample Size
Chiapas, Oaxaca, Guerrero, Michoacan, or Puebla.
0.37 (0.05) 4.7 (0.28) 5.2 (0.21) 6.1 (0.22) 0.46 (0.05) 2.1 (0.08) 0.31 (0.07) 1699
Next, we compute a propensity score (the predicted probability of being eligible and participating in the program) for each adult age 25 to 40 using the estimated probit model coefficients and measures of their family background (parental education, characteristics of the household when they were age 12, and an approximation of the number of children age 0 to 10 in the household at that time). Although the actual targeting of Oportunidades is based on several additional variables, we have to restrict the analysis to the subset available in the dataset for both children and adults, which fortunately includes the major
10
The participation model is estimated only for children in rural and semi-urban areas, because in 2002 (the time of our data collection) the program had not been significantly extended to urban areas. The data contain information pertaining to interviews with the parents of 1,970 children age 9-12 in rural areas. After dropping observations with missing variables, we are left with 1,699 observations.
81
determinants of program eligibility. We classify the 40% with the highest predicted probabilities of participation as the target group and the remaining 60% as the non-target group. TABLE 3
ESTIMATED PROBIT MODEL FOR PROBABILITY OF PARTICIPATING IN OPOrTuNIDADES
Variable Mothers schooling less than 6 grades (omitted) Mothers schooling 6 grades Mothers schooling 7 to 9 grades Mothers schooling 10 to 12 grades Mothers schooling 13 or more grades Fathers schooling less than 6 grades (omitted) Fathers schooling 6 grades Fathers schooling 7 to 9 grades Fathers schooling 10 to 12 grades Fathers schooling 13 or more grades Max parents schooling less than 6 grades (omitted) Max parents schooling 6 grades Max parents schooling 7 to 9 grades Max parents schooling 10 to 12 grades Max parents schooling 13 or more grades Indoor plumbing 0 or 1 young children in household (omitted) 2 to 4 young children in household 5 young children in household 6 or more young children in household Living in poor southern state Constant term Sample Size Pseudo R-squared
Chiapas, Oaxaca, Guerrero, Michoacan, or Puebla.
Coefficient 0.624 0.914 1.286 0.652 0.592 0.836 1.284 0.317 0.750 0.916 1.211 0.370 0.291 0.139 0.466 1.159 0.257 0.214 1699 0.11
p-value 0.000 0.001 0.006 0.264 0.000 0.000 0.015 0.453 0.000 0.000 0.034 0.551 0.029 0.160 0.037 0.023 0.201 0.302
Table 4 compares the characteristics of the target and non-target groups, separately for men and women. For both men and women, the target group has much lower maternal and paternal education levels. Individuals in the target groups also grew up with more young children in households that were much less likely to have indoor plumbing. For both men and women, there is a two year schooling attainment gap between the target and non-target groups as well as a two cm difference in height. The labor market experience measure we use is Mincer potential experience, which equals age minus years of education minus six. The target group has more experience under this measure, mainly because of having lower schooling attainment11.
11
The MxFLS data do not include years of actual labor market experience.
82
TABLE 4
DESCRIPTIVE STATISTICS FOR MEN AND WOMEN, AGE 25-40, BY PROJECTED OPOrTuNIDADES PARTICIPATION
Men 40% Target Mothers schooling Fathers schooling Max Parental education Indoor plumbing # children age 0-10 in household Living in poor southern state Mean monthy earnings (in 1000s of pesos) Schooling Height Experience Sample Size 1.9 (0.14) 2.7 (0.17) 3.2 (0.19) 0.18 (0.03) 2.5 (0.12) 0.34 (0.06) 3.3 (0.28) 7.4 (0.25) 164.4 (0.56) 19.5 (0.40) 867 60% Non-target 4.4 (0.22) 5.1 (0.26) 5.7 (0.24) 0.85 (0.03) 1.1 (0.07) 0.11 (0.05) 4.3 (0.25) 9.6 (0.30) 166.9 (0.55) 16.1 (0.42) 1083 40% Target 1.9 (0.12) 2.6 (0.13) 3.2 (0.14) 0.18 (0.03) 2.5 (0.10) 0.35 (0.06) 0.7 (0.07) 6.3 (0.20) 152.1 (0.50) 20.7 (0.26) 1629
Women 60% Non-target 4.1 (0.21) 4.6 (0.25) 5.3 (0.24) 0.80 (0.03) 1.2 (0.06) 0.11 (0.04) 1.5 (0.20) 8.7 (0.22) 154.3 (0.43) 16.9 (0.33) 1592
The mean levels in Table 4 shows that the target group is less advantaged than the non-target group. In particular, mean monthly earnings are 3,300 pesos per month for targeted men and 4,300 pesos per month for non-targeted men. Targeted women can expect about half (700 pesos per month) the labor income of non-targeted women (1,500 pesos per month). But there is still substantial overlap in the two earnings distributions, as shown in Figure 1. The top panel describes mens labor income while the bottom panel describes womens. The solid line in each panel is a nonparametric estimate of the density of positive earnings, while the two dashed lines correspond to the densities of positive earnings in the target and nontarget groups12. Again, the mean of the the target subsample is clearly lower than that of the nontarget, but a substantial proportion of the target group can expect to receive earnings above the population mean and a large proportion of the nontarget group receives very little income13.
12 13
The target density has been scaled by a factor of 0.4 and the non-target by a factor of 0.6 so that together they add up to equal the total population density. The fraction of men receiving no labor income differs very little between the target (11.4%) and nontarget (10.6%) groups, but the difference is actually quite large among women where 71% of targeted women receive no labor income compared to 58% of the non-target group.
83
FIGURE 1
DENSITIES OF INCOME FOR MEN AND WOMEN 0.2 0.18 0.16 0.14 0.12 0.1 0.08 0.06 0.04 0.02 0 0 5 10 15 Mens Monthly Income (1000s pesos) 20
Total Sample (Men) Scaled Target Subsample Scaled Non-target Subsample
0.25
Total Sample (Men) Scaled Target Subsample Scaled Non-target Subsample
0.2
0.15
0.1
0.05
5 10 15 Womens Monthly Income (1000s pesos)
20
Source: MxFLS 2002. All densities are nonparametrically estimated using non-zero values of income.
84
4.2. Simulating counterfactual distributions We next use the estimated earnings-schooling-height relationships to simulate the longer-term effects of the Oportunidades program on labor income. Figures 2 and 3 illustrate the nonlinearities in the density of non-zero labor income, conditional on schooling attainment and height; Figure 2 graphs the conditional density for men and Figure 3 for women. It is evident from the figures that higher levels of schooling attainment are associated with relatively larger increases in marginal earnings. The marginal earnings benefit is more homogeneous with respect to increases in height, but there also appears to be some nonlinearity near the upper end of the height distribution. Tables 5a and 5b show the results of our main simulation experiments for men and women. The first column displays characteristics of the income distribution without any program impacts. This income distribution is equal to the original income distribution with the addition of a small amount of error introduced by the nonparametric smoothing. The other columns of Table 5a and 5b each represent a different set of hypothetical program impacts, given by (a)-(e), where we give the stated program impact to each individual in the target group and calculate the implied income distribution for the combined target and non-target groups. For example, case (a) augments each individuals education level by 0.6 years. We use the nonparametric simulation method described above to simulate a counterfactual earnings distribution whose features can be compared to the original no-program earnings distribution. As previously noted, we simulate changes in employment along with changes in the distribution of positive earnings. That is, the earnings distribution includes a mass point at zero for nonworkers and the fraction of nonworkers can be affected by the program. Monthly earnings are measured in thousands of pesos. Table 5a indicates that the program would not significantly affect the fraction of men participating in the labor market, which remains around 90% across all the simulations. Also, impacts (a)-(c) have modest effects on mean earnings for men and almost no effect on earnings inequality, regardless of the measure. The effect of a 0.6 year impact on schooling attainment (in columns (b) and (c)) is larger for women than it is for men; however, the changes in income inequality are relatively minor for both men and women. The hypothetical three year increase in schooling attainment, shown in column (d), leads to substantially higher mean earnings and a reduction in poverty as measured by the Headcount ratio and the average poverty gap. While income inequality actually increases slightly for men, it declines somewhat for women due to the large induced increase in female employment. A one cm increase in height leads to about a 30 peso increase in mean monthly earnings for men but no substantial difference for women. The height impact has almost no influence on earnings inequality, but a large hypothetical increase in height of three centimeters slightly increases mean earnings and inequality, without having much effect on poverty. Even though Oportunidades significantly increased the human capital of children from disadvantaged families and substantially raised mean earnings, we have found its effects on earnings inequality to be minimal for two main reasons. First, returns to schooling in this environment are highly non-linear and in particular, we observe increasing returns at higher schooling levels. Those individuals in the target group that would have higher educational at-
85
FIGURE 2
DENSITY OF MENS INCOME CONDITIONAL ON SCHOOLING AND HEIGHT
0.3 0.25 0.2 0.15 0.1 0.05 0 14 16 18
10 Monthly Income (1000s pesos) 15
20 0
12 10 Schooling (years)
0.25 0.2 0.15 0.1 0.05 0 180 175
10 Monthly Income (1000s pesos)
15
20 140
145
160 155 150
165
170
Height (cms)
Source: MxFLS 2002 All densities are nonparametrically estimated using non-zero values of income.
tainment in the absence of the program experience relatively larger increases in income as a result of the program, so it is not the case that the poorest of the target group experience the largest benefit. The second factor that dampens the programs effect on inequality is that targeting children from poor backgrounds only imperfectly targets future low-earning adults, because of substantial intergenerational mobility.
86
FIGURE 3
DENSITY OF WOMENS INCOME CONDITIONAL ON SCHOOLING AND HEIGHT 0.35 0.3 0.25 0.2 0.15 0.1 0.05 0 14 16 18
15
20 0
12 10 Schooling (years)
0.3 0.25 0.2 0.15 0.1 0.05 0 170 180 175
15
150 145 140 20
155
160
165
Height (cms)
Source: MxFLS 2002 All densities are nonparametrically estimated using non-zero values of income.
We measure the influence of the nonlinearity in returns to schooling by estimating and simulating parametric models of the employment and earnings processes and comparing these results to those found in our nonparametric simulations. Tables 6a and 6b present estimated coefficients for a probit model of employment and a linear regression model of log earnings. Each model contains a linear term for years of schooling and quadratics for height and potential experience. To simulate the employment process, we augment schooling attainment, height, and/or experience under the same program scenarios evaluated
TABLE 5A
SIMULATED EFFECTS OF OPOrTuNIDADES IMPACTS ON INCOME DISTRIBUTION BASED ON NONPARAMETRIC EARNINGS DENSITY ESTIMATIONS Men, Age 25-40
Original
Schooling (+0.6 yrs) (a) (b) 0.099 3.961 4.455 3.012 3.338 1.125 0.487 0.451 0.235 0.158 0.126 0.099 3.974 4.456 3.021 3.339 1.121 0.487 0.450 0.235 0.157 0.126 (c) (d) 0.097 4.255 4.857 3.180 3.443 1.142 0.491 0.458 0.219 0.149 0.121
Height (+1 cm)
Schooling (+0.6 yrs) Height (+1 cm) Schooling (+3 yrs)
Height (+3 cm) (e) 0.101 4.037 4.631 3.030 3.371 1.147 0.493 0.462 0.233 0.158 0.127
Proportion with zero earnings Mean earnings Std. Dev. earnings Median earnings Interquartile Range Coefficient of Variation Gini Coefficient Theil Index Headcount Ratio (FGT, = 0)* Average Poverty Gap Ratio (FGT, = 1) * Foster-Greer-Thorbecke Index (FGT, = 2)*
0.098 3.931 4.382 3.003 3.331 1.115 0.485 0.446 0.237 0.158 0.126
0.099 3.944 4.382 3.013 3.337 1.111 0.485 0.445 0.236 0.158 0.126
Sample size is 1950.
87
88
TABLE 5B
SIMULATED EFFECTS OF OPOrTuNIDADES IMPACTS ON INCOME DISTRIBUTION BASED ON NONPARAMETRIC EARNINGS DENSITY ESTIMATIONS Women, Age 25-40
Original
Height (+1 cm)
Height (+3 cm) (e) 0.640 1.155 2.626 0.000 1.335 2.274 0.819 1.480 0.759 0.708 0.678
0.637 1.147 2.618 0.000 1.335 2.283 0.819 1.478 0.759 0.706 0.675 0.670
0.632 1.173 2.644 0.000 1.394 2.254 0.816 1.460 0.755 0.701
89
above and predict employment using draws from the probit error distribution that are consistent with the observed choices. To simulate earnings, we make the same augmentations to the human capital variables and for each worker incorporate the original earnings residual if it was observed and draw from the earnings residual distribution for those who were not working in the original sample. These simulation results are shown in Tables 7a and 7b. TABLE 6A
ESTIMATED PARAMETRIC PROBIT MODELS FOR EMPLOYMENT
Variables Schooling (years) Height Height2 Experience Experience2 Constant Sample Size Pseudo R-squared
Standard errors are in parentheses.
Men 0.009 (0.022) 0.141 (0.212) 0.000 (0.001) 0.067 (0.038) 0.002 (0.001) 9.914 (17.628) 1950 0.0145
Women 0.107 (0.013) 0.070 (0.145) 0.000 (0.0005) 0.000 (0.029) 0.000 (0.0007) 6.030 (11.178) 3221 0.0513
TABLE 6B
ESTIMATED PARAMETRIC REGRESSION MODELS FOR LOG INCOME
Variables Schooling (years) Height Height2 Experience Experience2 Constant Sample Size R-squared
Standard errors are in parentheses.
Men 0.087 (0.011) 0.073 (0.085) 0.000 (0.0002) 0.040 (0.019) 0.001 (0.001) 4.298 (7.300) 1720 0.1966
Women 0.154 (0.015) 0.090 (0.139) 0.000 (0.0004) 0.002 (0.037) 0.001 (0.001) 9.251 (10.768) 1044 0.2712
90
TABLE 7A
SIMULATED EFFECTS OF OPOrTuNIDADES IMPACTS ON INCOME DISTRIBUTION BASED ON PARAMETRIC EARNINGS MODELS Men, Age 25-40
Original
Height (+1 cm)
Height (+3 cm) (e) 0.102 4.015 4.497 3.000 3.440 1.120 0.483 0.444 0.222 0.149 0.128
0.099 3.945 4.432 3.000 3.300 1.123 0.483 0.443 0.227 0.148 0.126
0.098 4.006 4.472 3.000 3.417 1.116 0.481 0.439 0.220 0.145 0.124
TABLE 7B
SIMULATED EFFECTS OF OPOrTuNIDADES IMPACTS ON INCOME DISTRIBUTION BASED ON PARAMETRIC EARNINGS MODEL Women, Age 25-40
Original
Height (+1 cm)
Height (+3 cm) (e) 0.644 1.133 2.583 0.000 1.265 2.279 0.821 1.467 0.764 0.706 0.686
0.638 1.140 2.595 0.000 1.200 2.276 0.819 1.459 0.763 0.702 0.681 0.670
0.627 1.196 2.662 0.000 1.400 2.225 0.813 1.423 0.753 0.691
91
92
Because there is near universal employment of men across the human capital distribution we find little effect of schooling and height on employment with a small positive effect of experience. The story is quite different for women where a year of schooling has a strong and significant positive effect on the probability of employment. When schooling is constrained to have a linear effect on log earnings, a year of schooling increases earnings by 8.7% for men and 15.4% for women. The linear and quadratic terms for height are jointly significant ( < 0.05 for both men and women while the experience terms are only jointly significant for women. A comparison of Table 7a with Table 5a shows that for men, the parametric simulation approach tends to predict small reductions in inequality relative to the small increases in inequality predicted by the nonparametric approach. These differences are almost entirely due to the fact that the parametric model constrains log earnings to be a linear function of schooling and does not capture the fact that schooling has increasing returns. For women (Tables 5b and 7b) the parametric model predicts a smaller reduction in inequality because of the differences in how schooling affects employment. In particular, the nonparametric model predicts that targeted increases in schooling will increase womens employment more than a parametric model that includes schooling as a linear term. This difference in the effect on employment overpowers the impact of imposing constant returns to schooling in the earnings process. The second major factor explaining Oportunidades modest effect on inequality is that it targets children from poor families and these children are not necessarily the future poor adults. That is, the program increases the completed schooling of some children from an already high level to an even higher level. To explore the importance of targeting, we performed another set of simulations using our nonparametric earnings model where we target the same fraction of individuals with the program but choose them on the basis of low adult education levels. Specifically, we give the program to those who would otherwise form the bottom of the education distribution. This targeting is of course not feasible in practice, because it is impossible to know which children would eventually complete the least amount of schooling. Nevertheless, the simulation results reported in Tables 8a and 8b give an upper bound for improving earnings and inequality through more precise targeting. A comparison with Tables 5a and 5b shows that targeting individuals at the bottom of the education distribution would be more effective in reducing inequality than the current targeting mechanism, but at the cost of lower mean earnings, because it does not take advantage of the larger program impacts at higher schooling levels. In addition to the simulations we report, which assume constant treatment effects, we also carried out all the simulations under alternative scenarios of heterogeneous program effects. For example, we assigned half the target population impacts that were twice as high and half zero impact, keeping the average treatment effect the same. The half of the target group that received the double impact was alternatively chosen to be the less or more advantaged subgroup. Our findings with regard to effects on mean earnings and earnings inequality under
TABLE 8A
SIMULATED EFFECTS OF OPOrTuNIDADES IMPACTS ON INCOME DISTRIBUTION BASED ON NONPARAMETRIC EARNINGS DENSITY ESTIMATIONS PERFECT TARGETING Men, Age 25-40
Original
Height (+1 cm)
Height (+3 cm) (e) 0.101 3.958 4.451 3.031 3.308 1.124 0.486 0.451 0.234 0.159 0.128
0.098 3.931 4.382 3.003 3.331 1.115 0.485 0.446 0.237 0.158 0.126 0.125
0.097 3.945 4.381 3.015 3.344 1.111 0.485 0.444 0.236 0.157
93
94
TABLE 8B
SIMULATED EFFECTS OF OPOrTuNIDADES IMPACTS ON INCOME DISTRIBUTION BASED ON NONPARAMETRIC EARNINGS DENSITY ESTIMATIONS PERFECT TARGETING Women, Age 25-40
Original
Height (+1 cm)
Height (+3 cm) (e) 0.640 1.152 2.638 0.000 1.332 2.290 0.820 1.484 0.759 0.708 0.678
0.637 1.147 2.618 0.000 1.335 2.283 0.819 1.478 0.759 0.706 0.675 0.673
0.635 1.157 2.624 0.000 1.366 2.267 0.817 1.468 0.757 0.704
95
the heterogeneous treatment impact simulations were very similar to those discussed previously, so we omit them for the sake of brevity14. 5. Conclusions The Oportunidades program aims to reduce poverty of the current generation through transfers and to alleviate poverty of the next generation through human capital investment. A number of experimental and nonexperimental evaluation studies have documented that the program significantly improves schooling attainment, health and nutrition over the short-term. This paper develops and applies a nonparametric simulation method for the purpose of studying how increases in schooling attainment and in height, as a measure of long-term nutritional status, will affect the distribution of earnings in the next generation. Our empirical findings suggest that the human capital investment in todays youth will increase their mean earnings levels, but will have only a modest effect on earnings inequality. Behrman (2006) comes to a similar conclusion in a survey of human capital policies and from an empirical study of how increasing education affects earnings inequality in Chile. The key factors underlying the modest effects on inequality that we observe are the difficulty in predicting which children will become future low earning adults and nonlinearities in how health and education are priced in the labor market. With regard to the first factor, childhood poverty is a strong predictor of future low earnings, but there is also substantial intergenerational mobility that makes it difficult to target low adult earners on the basis of childhood characteristics. With regard to the second factor, we found evidence of important nonlinearities in how height and education influence earnings. Most notably, an additional year of secondary school has a higher monetary return than an additional year of primary school. Because of these nonlinearities, people at the upper deciles of the targeted population tend to benefit more from the program intervention. We conclude by considering some limitations of the simulation method studied in this paper. First, the method assumes that the observed relationship between earnings and the covariates of education, height, and work experience is causal. This raises concern about potential bias due to unobserved ability, which is the subject of a large labor economics literature. Previous attempts to control for ability bias have relied mainly on instrumental variables or natural experiments (e.g. twins with different levels of schooling).15 Although there is variation in reported estimates, most estimates of the rate of return to schooling that purport to control for ability bias through the use of instrumental variables exceed those obtained by ordinary least squares. The variation in estimates is partly accounted for by heterogeneity in returns to education on earnings that requires a LATE (local average treatment effect) interpretation
14 15
The estimates are available from the authors on request. e.g., Behrman, Rosenzweig and Taubman (1994), Ashenfelter and Krueger (1994), Ashenfelter and Rouse (1998), Card (1995, 1999).
96
of the instrumental variables estimates.16 Estimates that account for ability bias using variation in twin pairs, on the other hand, tend to be somewhat lower than cross-sectional OLS estimates. Because the literature finds that OLS estimates do not necessarily overstate the causal effect, we have no reason to believe that our nonparametric procedure necessarily overstates the true return to schooling attainment. Also, much of the instrumental variables literature operates within a parametric framework and does not easily allow for the nonlinearities in the earnings-schooling-height-experience relationship that we find to be quantitatively important. Nevertheless, further exploration of how the simulation method could be modified to account for unobserved ability and endogenous covariates would be useful. A second critical assumption of the simulation method is the usual synthetic cohort assumption, namely that the characteristics of todays 25 to 40 year olds, observed in 2002, are representative of the future adulthood of todays children. Extrapolating from current trends, children today would likely attain more education than current 25 to 40 year olds in the absence of the program intervention. Our estimates indicate that the marginal effect of education on earnings is increasing in years of education, so overall rising education levels could lead the simulation to understate somewhat the impact of Oportunidades on earnings. Third, the simulation method does not account for the general equilibrium effects of increasing the education levels of a large segment of the future labor force, which would tend to decrease returns to education. Any decline, though, is at least partially mitigated by the fact that Mexico is an open economy. Fourth, this study focused on individual level earnings for men and women, although household-level earnings inequality may be more relevant to policy makers. It is also not clear how to interpret high income inequality in a group (like women) where a large proportion choose not to work, because they have a partner who provides enough money for the household. The simulation method could be extended to model household formation by incorporating a marriage outcome, where marriage opportunities and outcomes potentially also depend on variables influenced by the program. Our method could similarly be extended to account for the influence of improving human capital on internal and external migration. Lastly, improvements in future earnings are only one of the long-term benefits expected from the program. For example, there is a substantial literature documenting how upgrading mothers education increases child test scores (e.g., Rosenzweig and Wolpin, 1994). Female program beneficiaries who choose not to work may be more effective mothers and may choose to have fewer children and to invest more in them. The simulation methodology in this paper could conceivably be extended to examine changes in fertility.
16
Card (1999, 2001).
97
References Angelucci, M., G. De Giorgi, M. Rangel, and I. Rasul (2008). Family Networks and School Enrollment: Evidence from a Randomized Social Experiment, Working Paper. Ashenfelter, O. and A. B. Krueger (1994). Estimates of the Economic Return to Schooling from a New Sample of Twins, Quarterly Journal of Economics, 84 (5): 1157-73. Ashenfelter, O. and C. Rouse (1998). Income, Schooling and Ability: Evidence from a New Sample of Identical Twins, Quarterly Journal of Economics, 113 (1): 253-84. Behrman, J. and J. Hoddinott (2005). Programme Evaluation with Unobserved Heterogeneity and Selective Implementation: The Mexican PROGRESA Impact on Child Nutrition, Oxford Bulletin of Economics and Statistics, Vol. 67, N 4, pp. 547-569. Behrman, J., P. Sengupta and P. Todd (2004). Progressing through PROGRESA: An Impact Assessment of a School Subsidy Experiment, Philadelphia: University of Pennsylvania, forthcoming in Economic Development and Cultural Change. Behrman, J. and E. Skoufias (2004). Evaluation of PROGRESA/Oportunidades: Mexicos Anti-Poverty and Human Resource Investment Program, in Jere R. Behrman, Douglas Massey and Magaly Sanchez R, eds., The Social Consequences of Structural Adjustment in Latin America, book manuscript. Behrman, J. (2006). How Much Might Human Capital policies Affect Earnings Inequalities and Poverty?, Working Paper. Behrman, J., M. Rosenzweig and P. Taubman (1994). Endowments and the Allocation of Schooling in the Family and in the Marriage Market: The Twins Experiment, in Journal of Political Economy, volume 102, N 6, 1131-1174. Bourguignon, F., F. Ferreira and P. Leite (2003). Conditional Cash Transfers, Schooling and Child Labor: Micro-Simulating Brazils Bolsa Escola Program. World Bank Economic review 17 (2): 229-54. Buddelmeyer, H., and E. Skoufias (2003). An Evaluation of the Performance of Regression Discontinuity Design on PROGRESA, IZA Discussion Paper N 827 (July), Institute for the Study of Labor (IZA), Bonn, Germany. Card, D. (1999). The Causal Effect of Education on Earnings. In Handbook of Labor Economics, Vol. 3A, edited by Orley Ashenfelter and David Card. (Amsterdam: North Holland). Card, D. (2001). Estimating the Return to Schooling: Progress on Some Persistent Econometric Problems, Econometrica 69 (5), 1127-1160. Chakravarty, S. (1990). Ethical Social Index Numbers. New York: SpringerVerlag. DiNardo, J., N. Fortin, T. Lemieux (1996). Labor Market Institutions and the Distribution of Wages, 1973-1992: A Semiparametric Approach, Econometrica, Vol. 64, N 5, pp. 1001-1044. Foster, J., J. Greer and E. Thorbecke (1984). A Class of Decomposable Poverty Measures, in Econometrica, Vol. 52, N 3, pp. 761-766.
98
Freije, S., R. Bando and F. Arce (2006). Conditional Transfers, Labor Supply, and Poverty: Microsimulating Oportunidades, in Economia, Vol. 7, Issue 1, 73-124. Gertler, P. (2000). Final Report: The Impact of PROGRESA on Health, International Food Policy research Institute, Washington, D.C. Lopez-Acevedo, G. (2004). Mexico: Evolution of Earnings Inequality and Rates of Returns to Education (1988-2002), Work Bank Report N 19945. Parker, S. and E. Skoufias, 2000, The impact of PROGRESA on work, leisure and time allocation, October. Report submitted to PROGRESA. International Food Policy Research Institute, Washington, D.C. <http:// www.ifpri.org/themes/progresa.htm> Persico, N., A. Postlewaite, and D. Silverman (2004). The Effect of Adolescent Experience on Labor Market Outcomes: The Case of Height, Journal of Political Economy. Rosenzweig, M. and K. Wolpin (1994). Are There Increasing Returns to the Intergenerational Production of Human Capital? Maternal Schooling and Child Intellectual Achievement, The Journal of Human resources, Vol. 29, N 2, Special Issue: Womens Work, Wages, and Well-Being (Spring, 1994), pp. 670-693. Schultz, T. P. (2000). Impact of PROGRESA on school attendance rates in the sampled population, February. Report submitted to PROGRESA. International Food Policy Research Institute, Washington, D.C. Schultz, T. P. (2004). School subsidies for the poor: Evaluating a Mexican strategy for reducing poverty, Journal of Development Economics, (Revision of June 2000 Report submitted to PROGRESA. International Food Policy Research Institute, Washington, D.C. <http://www.ifpri.org/ themes/progresa.htm> Silverman, B. (1986). Density Estimation for Statistics and Data Analysis. Chapman & Hall/CRC. Skoufias, E. and B. McClafferty (2001). Is PROGRESA working? Summary of the results of an Evaluation by IFPRI, Report submitted to PROGRESA. Washington, D.C.: International Food Policy Research Institute, <http:// www.ifpri.org/themes/progresa.htm> Strauss, J. (1986). Does Better Nutrition Raise Farm Productivity? Journal of Political Economy, 94, April, p. 297-320. Strauss, J. and Duncan T. (1998). Health, Nutrition and Economic Development, Journal of Economic Literature, Vol. 36 (2), 766-817. Todd, P. and K. Wolpin (2006). Assessing the Impact of a School Subsidy Program in Mexico,with Kenneth I. Wolpin, American Economic Review, 2006, December. Todd, P. (2004). Technical Note on Using Matching Estimators to Evaluate the Oportunidades Program For Six Year Follow-up Evaluation of Oportunidades in Rural Areas, Philadelphia: University of Pennsylvania, mimeo.
99
APPEnDIx A Construction of Samples and Variables This appendix describes how each of the variables for the empirical analysis was constructed. The data analysis has three parts. First, we estimate a probability of participating in the Oportunidades program and use the estimated model to simulate program targeting for men and women between age 25 and 40. Second, we estimate nonparametrically the relationship between income, education, height, and work experience for men and women between age 25 and 40. Third, we compute the counterfactual income distribution under assumptions of how the program affects education, height, and work experience that are consistent with recent evaluations of short-term program impacts. Sample Construction The initial sample of MxFLS respondents between age 25 and 40 contains 6,564 observations. When we drop the individuals who worked but did not report their income, the number goes down to 5,871. It drops further to 5,180 (79% of the original sample) when we drop those individuals who did not report their education or whose height was not measured. Finally, we drop an additional 9 outlier observations for individuals who report receiving more than 40,000 pesos in the previous month. This leaves a final sample size of 5,171. Construction of Variables Income. Income is measured as total labor income earned (including net profits for the self-employed) in the previous month. It includes zeros for those individuals who dont work. About 6% of individuals who reported working in the previous week are recorded as being peasants on their plot. 40% of these individuals report zero income in the last month. This seems plausible for subsistence farmers. Only 2% of other individuals who report working report zero income. Income is measured in thousands of pesos and in 2002 the average daily exchange rate was 1 USD = 9.68 pesos. We do not use proxy reports on income, because it is not clear how to combine this data with the first-person reports and weight the data correctly. The proxy reports also have more missing data. Schooling. The MxFLS collects the type of the last school attended and, for most individuals, the number of years that the individual completed at that level. We do not include years of technical education in our measure, because wage returns to technical education (based on our own linear regressions) are much lower than the returns of conventional schooling.
100
Height. Height is not self-reported but instead is measured by trained survey personnel. Experience MxFLS did not collect information on actual labor force experience, so we use the standard Mincer measure of potential experience equal to age minus years of schooling minus six.
Estudios de Economa. Vol. in - N 1, Junio 2011. Galasso Alleviating extreme poverty 38 Chile / EmanuelaPgs. 101-127
101
Alleviating extreme poverty in Chile: the short term effects of Chile Solidario*
Aliviando la extrema pobreza en Chile: efectos a corto plazo de Chile Solidario
Emanuela Galasso** Abstract This paper evaluates the effect of an anti-poverty program, Chile Solidario, during its first two years of operation. We find that the program tends to increases significantly their take-up of cash assistance programs and of social programs for housing and employment, and to improve education and health outcomes for participating households. There is no evidence that the participation to employment program translates into improved employment or income outcomes in the short term. Finally, we provide suggestive evidence of the key role that the psycho-social support had in enabling this change, by increasing awareness of social services in the community as well as households orientation towards the future. Key words: Program evaluation, Matching estimators, Extreme poverty, Chile. Resumen En este estudio se evala el efecto de un programa de lucha contra la pobreza, Chile Solidario, durante sus primeros dos aos de funcionamiento. Encontramos que el programa tiende a aumentar significativamente su asimilacin para los programas de asistencia en efectivo y los programas de vivienda sociales y empleo, as como a mejorar los resultados de la educacin y la salud de los hogares participantes. No hay evidencia de que la participacin en programas de empleo se traduzca en mejoras en el empleo o los ingresos en el corto plazo.
The work reported in this paper is part of the technical assistance of the World Banks Social Protection Sector Adjustment Loan in Chile. The work has been done in close collaboration with Evaluation Office in the Ministry of Planning (MIDEPLAN), Government of Chile, under the supervision of the Executive Secretary of the program. These are the views of the author, and do not necessarily reflect those of the Government of Chile or the World Bank. ** Development Research Group, World Bank. E-mail: egalasso@worldbank.org.
102
Finalmente, la evidencia sugiere que el apoyo psicosocial ha tenido un rol fundamental en generar este cambio, tanto por medio de aumentar el conocimiento acerca de los servicios sociales en la comunidad como a travs de la orientacin de los hogares hacia el futuro. Palabras clave: Evaluacin de programas, Estimadores de matching, Pobreza extrema, Chile. JEL Classification: c14, c51, I31, I38, O15.
We cannot be content, when we know that 6% of the population lives in conditions of indigence. () We are going to go where they live. We want not only to provide subsidies, we want their children to study, to have health assistance, and we want to include them into social networks and into the society in its entirety. We are going to build a bridge between them and their rights, so that they can exercise them to defeat their conditions of extreme poverty. Ricardo Lagos, President of the Republic of Chile. Presidential address, May 2002.
1. Introduction The perception that poverty is associated with social exclusion is subject of public debate in many countries. There is a general agreement that household in extreme poverty are deprived along multiple dimensions, which reinforce each other to jointly lock them into indigence. Yet there are very few examples of policy interventions that take this multi-dimensionality seriously, so as to help the extreme poor to escape deprivation in a sustained way by simultaneously addressing different structural constraints. An important exception approach might come from a new program aimed at tackling extreme poverty in Chile. The country has experienced years of sustained income growth during the 1990s, with an average per capita GDP growth of 4.5 per cent between 1990 and 2002. As a result, in the context of a stable income distribution (Ferreira and Litchfield, 1999),1 economic growth has translated into a reduction in the incidence of overall poverty in the country (from 33 per cent to around 15 per cent), but without much changes in extreme poverty (stable at around 5.6 per cent) over the same period (World Bank, 2001). The benefits from growth did not trickle down to the poorest segments of the population despite a large array of social services, targeted to the poor.2
Between 1987 and 1994, the shape of the income distribution has only slightly changed, with a small compression at the bottom and a small increase in the upper tail (Ferreira and Litchfield, 1999; Litchfield, 2001). As of 1998, the first five ventiles of household income were receiving 54% of all cash assistance programs, up from 40% at the beginning of the 1990s (MIDEPLAN, 2002).
Alleviating extreme poverty in Chile / Emanuela Galasso
103
The poorest segments are often unaware of their eligibility to certain programs or do not know how to activate the process of accessing them. As a response, the government of Chile has proactively introduced in 2002 a program, Chile Solidario, which aims reaching households in indigence in the country with an approach that goes beyond improving the targeting performance of public programs or simply providing recipients with cash assistance. The first component of the program reaches households in extreme poverty (through a proxy means testing) and provides them with a two year period of psycho-social support through a local social worker. During this period, the social worker works with the household to assess their needs and to help them devise a strategy to exit extreme poverty in the short run, by providing direct cash transfers at a decreasing rate over time and by connecting households to various social programs. After the two year intensive period, households are ensured a direct cash transfer and preferential access to assistance programs for an additional period of three years. At the same time, the program aims at helping households to progressively sustain their exit from extreme poverty in the long run by improving their human capital assets, their housing and their income generation capacity. The second component works on the supply side, by ensuring coordination among different programs. The rational comes from the recognition that an approach with isolated and sectoral programs does not lend itself to face the multiple and interrelated material as well as psycho-emotional deprivation of the extreme poor. The objective in the long term is to move away from an approach based on single programs towards a system of social protection, where the supply side provides bundles of programs that are tailored to meet the specific needs of households that are hard to reach. The program scaled up and expanded a pilot program called Puente, previously operating in 4 provinces. The program was phased in four waves, from 2002 to 2005 to cover a target population of 225,000 households, the estimated number of households in indigence in the country. The program has subsequently evolved to become a building block of the system of social protection in the country. Even though the primary objective is to alleviate extreme poverty, over time it added a complementary social protection focus, with the objective of protecting households from falling back into poverty when faced by uninsured risk. This document summarizes the results of the first quantitative short term evaluation of the program (Galasso, 2006). The data used in this paper for the purpose of the evaluation uses a subset of participating households and matched non-participants interviewed in the nationwide socio-economic survey (CASEN) in 2003 and followed up longitudinally in 2004. The results of the evaluation cover only the short term impact for the first cohorts of beneficiaries up to 2004, the majority of whom are still part of the two-year phase of psycho-social support by the social worker. The scope for identification comes from the design features of the program. The program assignment is based on a proxy-means score (CAS), related to unsatisfied basic needs. In the empirical analysis, we will exploit the exogenous geographic variation in the distribution of the CAS score, as well as in eligibility to estimate the effect of the program on a wide array of outcomes. The results from the first two years of intervention of the program show gains along different dimensions of education (preschool enrolment, enrollment into
104
school for 6-15, adult literacy) and milder effects on health outcomes (enrolment in the public health system, as well as preventive health visits for children under 6 and women). The results show also a strong take-up of employment programs, though this participation is not (yet) translated into employment effects. There are no significant effects on household income per capita, though participating households are significantly more likely to be receiving social assistance transfers. There is also evidence that on average Chile Solidario participants have increased their awareness of social services in the community and are more likely to be more optimistic about their future socio-economic situation. The structure of the paper is as follows. We start in section 2 with a detailed description of the Chile Solidario program and its assignment mechanism. Section 3 presents the methodology we apply, and discuss the identification assumptions. Section 4 describes the data and section 5 presents the results. Concluding comments will be provided in section 6. 2. Background on the Program The objective of alleviating extreme poverty is achieved through a two-pronged strategy, working on both the demand and the supply side of public services. The first component of Chile Solidario provides participating households with a two year period of psycho-social support implemented through the outreach activities of a local social worker. The social worker has a dual role of helping households to create or restore their basic capabilities and functions and helping them to create links and get connected to a local network of social services. He/she conveys information and represents a catalyst for households to elicit unexpressed demand for those public programs that meet their needs. The psycho-social support has been recognized by law as an integral component of the intervention and represents the key distinctive feature of this approach. The multidimensional aspect of deprivation is operationalized in terms of a set of minimum conditions, which aim at measuring a minimally acceptable level of well-being along different dimensions or pillars (identification/ legal documentation, family dynamics, education, health, housing, employment, income). During this intensive phase, each family works with the worker to get familiarized with the minimum conditions, and identifies the key priority areas to work on during the intervention. The families then commit to put their effort in meeting those unmet priority conditions, by signing partial contracts with the social worker. The program includes also a small cash transfer (bono de proteccin), which is transferred to participating households after having signed their partial contracts. The bonos value is tapered over time, with the idea that households should progressively improve their standards of living as a result of the program.3 The short-term income support in the case of Chile Solidario besides the
The direct transfer is set at Ch$ 10,500 per month for the first six months of the Puente program; decreases to Ch$ 8,000 in the second six months of the program. In the second year it decreases to Ch$ 5,500 and finally to Ch$ 3,500 for the last six months, an amount equivalent to the family allowance (SUF), one of the main cash assistance transfers.
105
bono takes the form of accessing existing cash assistance program to which participating households were already eligible to. The transfer is not conditional on any behavioral requirement on school enrolment or health visits, though it is terminated if households interrupt their participation to the program.4 The conditionality relates to the partial contracts that households signed during the intensive phase: households are expected to show efforts in working on those conditions that are recognized by the family itself as structural bottlenecks and to which they have committed to. After the two years of psycho-social support, households receive an unconditional exit bonus (bono de egreso) for additional three years, of an amount comparable to the last transfer of the bono de proteccion. The underlying rational for Puente and Chile Solidario was the realization that households in indigence were unable to formulate and activate their demand for social services, due to constraints to take-up that have been well documented in the US labor literature (Currie, 2004), such as information, feeling of helplessness and discrimination. The social worker conveys information and helps the households activate their demand for those public programs that meet their needs. At the same time the social worker assists the households in realizing what their needs and priorities are helps them devise a strategy (their life-time project), by developing a set of endowments (assets, skills, abilities, information, autonomy and self-efficacy) that allow them to autonomously sustain their exit from extreme poverty in the long-run. The second component of the intervention focuses on strengthening the supply side of public services. Public programs and services were previously available for their respective eligible population upon demand. Chile Solidario works directly with the municipalities, which are the local providers of public services, by making sure that the supply side is locally organized to attend the needs of this specific target population and bridge the demand gap. This process is facilitated by the fact that the activities of the social workers are institutionally grounded in each municipality (UIF, Unidad de Intervencin Familiar). Their work and performance is supervised and coordinated by a municipal employee (head of the UIF). Until 2004, the social workers had to work with the existing supply of social services within the municipalities, and the supply component of the program worked mainly through facilitating the coordination and networking of the different actors. The supply side response in terms of volume, size and targeting performance of new programs was activated after 2004, which is outside the scope of the current analysis.
The drop-out rate is estimated to be very low, around 3 per cent of all the households invited to participate.
106
3. Methodology 3.1. The assignment mechanism Impact is defined as the change in the outcome of interest that can be causally attributed to the program. It is important to bear in mind that the program CHS is targeted to households in extreme poverty. As a consequence, differences in outcomes between participants and non-participants could be attributed either to differences in initial conditions (such as their educational attainment, or ethnicity) or to non-observable characteristics (such as motivation or ability). As in other examples of social programs and conditional cash transfers programs in developing countries, the program is assigned on the basis of a proxy-means score calculated on the basis of a card (CAS ficha).5 All households whose scores are below a predetermined threshold are considered eligible to participate. In order to ensure a wide geographical coverage of the program, a decision was made to allow thresholds to vary across communes and regions, with the aim of reflecting differences in the poverty rates across different geographic areas. Households within municipalities are sequentially invited to participate to the program, by starting from the bottom up of their CAS distribution.6 These design features are such that two potentially eligible and observationally equivalent households can have been differentially exposed to the program. The results reported in this document use the method of matching to estimate the counterfactual of no-program outcomes and estimate the impact of the program. This method estimates the counterfactual of no-program outcomes by matching on the observable characteristics used in practice during the assignment mechanism, i.e. the CAS score. Households are assigned according to the CAS. Households are assigned to participate in strict ordering depending on their score. Moreover, given that households are invited to participate, they are assumed not to self-select into the program based on the expected gains. In this setting, participation (treatment in the evaluation literature), is assumed to be independent of the outcomes of interest, conditional on the score. We follow Abadie and Imbens (2006) to estimate the effect of the program on participants, by matching on the CAS score (more methodological details are
The score is a summary index of unsatisfied basic needs that is used as pre-requisite for participation to Chile Solidario and a wide-array of other social programs in Chile, from income transfers (e.g. family allowance SUF, old age public pension PASIS) as well as subsidies to health utilization(FONASA), water subsidies SAP, access to public housing and childcare centers. Even though in principle households could refuse the invitation to participate, in practice, the proportion of households who refused to is too small to meaningfully model the selection process.
107
available in Galasso, 2006). We also estimate the effect by matching on the CAS score and adjusting the difference within the matches in their covariates.7 We perform separate estimations for rural and urban areas: the incidence of poverty,8 the infrastructure, the supply side of public services and the labor markets faced by households living in rural and urban areas are very different. Note that this identification strategy compares participating and nonparticipating households with similar scores (and household characteristics). One potential concern is that it assumes that the effect of the treatment does not vary across regions and/or municipalities, which might be a strong assumption in the context of the program, which gives such an important role to municipalities. Different municipalities might face a different supply of social services. We address these concerns by presenting the results with an additional specification where we allow for community effects, in addition to household characteristics. This will control for any time-invariant differences in the initial conditions of the supply side, as well as for unobserved characteristics of the local labor market.9 An alternative identification strategy would rely on the exogenous variation in the thresholds for eligibility. The method of regression discontinuity design (van der Klaaw (2002) and Chay, McEwan, and Urquiola (2005)) allows to estimate the effect the program by comparing a large set of outcomes for households just above and below the cutoffs. However, given that municipalities started from the bottom-up of their CAS distribution, the bulk of the distribution of participants for the first cohorts is concentrated to the left of the graph, away from the cut-offs. For this reason, the application of this alternative method is deemed to be more relevant to subsequent cohorts, as it would allow making inferences applicable to a larger share of the target population (see Carneiro, Galasso and Ginja (2009)). 4. Data In order to document the evolution of impact of the program over time, the evaluation was implemented through a longitudinal survey (planned at
7
The bias-correction introduced by Abadie and Imbens (2006) removes the conditional bias that arises when matching is performed on more than two variables are used. In our framework, only one matching variable is used (i.e. the CAS score). We use their approach to estimate the conditional treatment effect on the treated. The incidence of rural poverty was found to be double the incidence of urban poverty in 1987, although the differences across urban and rural areas have converged over time, especially for the incidence of extreme poverty (Litchfield, 2001). The underlying assumption behind this specification is that the supply side is given at any given point in time. The assumption seems relevant in the first two years of operation of the program. Over time, to the extent that the supply side responds differentially depending on the local unsatisfied demand, this approach will need to be modified to ensure identification.
108
one year intervals between 2003 and 2007). The longitudinal survey identifies a sample of participants and comparable non-participants identified from the the nationally representative household survey the Caracterizacin Socioeconmica Nacional CASEN in 2003. The survey is multi-topic, ranging from questions on demographics, employment, income, education, health status and utilization of services to access to public subsidies and transfers. MIDEPLAN, the ministry in charge of the survey as well as of the program, planned to add questions on program participation to the CASEN administered in 2003. The sample size has been augmented to over-sample Chile Solidario beneficiaries. In Tables 1 we report weighted means for demographic, socio-economic characteristics, household income and intermediate indicators used in the analysis. The descriptive statistics (Table 1a) confirm that program has indeed been well targeted. Participants households come from larger households, where both the head and the spouse have lower educational attainments (about 2/3 have not completed primary education), have lower labor force attachment, and lower assets (durables). They are also more likely to come from rural areas, and from ethnic minorities. Participants are twice more likely than non-participants to have at least one member with disabilities. In order to allow for the possibility of following up the impact of the program over time, while keeping low the survey cost, MIDEPLAN agreed to interviewed only a subset of participants together their matched comparison one year apart (November 2004). Two subsequent rounds of the longitudinal survey (2006 and 2007) were collected to eventually form a four-year longitudinal panel. The 2004 questionnaire has newly added questions on participation to various social programs. It has also new modules on intergenerational mobility (with questions on the education and background of the parents), subjective welfare, as well as a short module on perceptions (problem solving, perceived social support and expectations about the future). The descriptive statistics, by participation status, on the intergenerational and the perception questions are suggestive. These underlying differences in the socio-economic conditions of participants and non-participants are also reflected in their subjective measures of well-being, with more than 2/3 of the participants whom consider themselves to belong to the lower ladder of socio-economic well-being compared to 1/2 of the non-participants. Finally, our identification strategy requires that we observe the actual CAS (proxy-means) score used to select households into the program. 2004 panel sample for the rest of the analysis. As shown in Figure 2, the distribution of CAS scores for participation is strictly to the left of that of non-participants. Participants are much more likely to be eligible (having their CAS below the relevant threshold). Next section will provide the results obtained by applying the empirical methods outlined in section 3.
109
TABLE 1A
DESCRIPTIVE STATISTICS, BY PARTICIPATION STATUS CASEN 2003 Non participants characteristics head household mean Rural area Hhld income p.c.* Hhld ingreso autnomo p.c. (excl. public transfers)* Hhld ingreso autnomo p.c. [adjusted] Durables: refrigerator Durables: calefont Household size Presence disabled member male age age<30 Education level: no education Incomplete primary complete primary Incomplete secondary Complete secondary Higher education Marital status: married union single Employment status: employed in 2000 housework in 2000 unemployed in 2000 currently employed currently unemployed currently inactive Relationship to the head: head spouse child other No.obs 0.130 128,927 125,925 171,010 0.864 0.626 3.744 0.135 0.739 49.52 0.098 0.033 0.203 0.139 0.181 0.234 0.205 0.579 0.128 0.101 0.714 0.086 0.038 0.718 0.041 0.241 st.dev. 0.337 304,879 304,494 439,221 0.343 0.484 1.749 0.397 0.439 15.27 0.297 0.180 0.403 0.346 0.385 0.423 0.403 0.494 0.334 0.301 0.452 0.281 0.192 0.450 0.199 0.428 Participants characteristics head household mean 0.307 33,330 27,945 38,339 0.470 0.088 4.716 0.251 0.665 47.55 0.125 0.124 0.515 0.176 0.119 0.054 0.008 0.445 0.247 0.108 0.619 0.172 0.070 0.620 0.079 0.302 st.dev. 0.461 30,789 30,519 51,926 0.499 0.283 2.083 0.531 0.472 15.39 0.331 0.330 0.500 0.381 0.324 0.226 0.088 0.497 0.431 0.311 0.486 0.378 0.256 0.486 0.269 0.459 characteristics Beneficiaries mean st.dev.
0.134 0.341 42.20 14.78 0.243 0.429 0.106 0.511 0.170 0.308 0.500 0.376
0.411 0.272 0.147 0.295 0.581 0.037 0.365 0.054 0.580
0.492 0.445 0.355 0.456 0.493 0.190 0.481 0.226 0.494
65,628
5,608
0.400 0.490 0.483 0.500 0.080 0.272 0.037 0.189 5,360
Note: Own calculation from the CASEN 2003 survey. Variables are weighted using sampling weights. (*) Income figures are not adjusted, i.e. without the application of the CEPAL correction. Hhld ingreso autonomo p.c. (adjusted) is used by MIDEPLAN to calculate poverty indexes.
110
TABLE 1B
DESCRIPTIVE STATISTICS, BY PARTICIPATION STATUS. PANEL SAMPLE 2003-2004
2003 Beneficiaries households Mean Demographics household size Share hhld members 0-5 Share hhld members 6-17 Share hhld members 18-64 Share hhld members 65+ presence disabled member both parents present head is male head is married ethnic minority age head<30 age head (30, 40) age head (40, 50) age spouse <30 age spouse (30, 40) age spouse (40, 50) rural area Family background head - no education head - incomplete primary Spouse - no education Spouse - incomplete primary durables: refrigerator durables: calefont CAS score =1 if CAS score is missing Eligibility (CAS<CAS cutoff) Housing No. rooms Water: public network Sewage: public network Walls: concrete, breaks Ownership: own house Ownership: rented house Roof conditions: interior ceiling OUTCOMES: Household Income (*) Hhld income p.c. Hhld labor income p.c. Hhld non labor income p.c. Hhld public transfers p.c. Hhld ingreso autnomo p.c. (excl. public transfers) Intermediate indicators if disabled: enrolled in Nat. Registry if disabled: children in education sys. st.dev. non-particip. households Mean Diff. Std. st.dev Means 2004 Beneficiaries households Mean st.dev. non-particip. households mean st.dev
Variables:
4.756 0.123 0.267 0.543 0.067 0.252 0.694 0.663 0.448 0.105 0.134 0.279 0.228 0.168 0.221 0.151 0.211 0.125 0.522 0.049 0.334 0.493 0.080 466.32 0.128 0.953 3.09 0.810 0.500 0.164 0.439 0.081 0.544
1.995 0.146 0.198 0.205 0.168 0.536 0.461 0.473 0.497 0.306 0.340 0.449 0.419 0.374 0.415 0.358 0.408
4.340 0.091 0.233 0.577 0.098 0.203 0.713 0.710 0.559 0.076 0.086 0.243 0.257 0.115 0.217 0.184 0.152
1.900 0.136 0.206 0.238 0.230 0.493 0.453 0.454 0.497 0.266 0.280 0.429 0.437 0.319 0.412 0.387 0.359 0.231 0.472 0.176 0.411 0.416 0.484 48.33 0.497 0.352 1.14 0.286 0.379 0.497 0.474 0.295 0.369
0.276 0.232 0.161 0.144 0.133 0.106 0.038 0.098 0.214 0.101 0.134 0.056 0.060 0.144 0.009 0.079 0.177 0.253 0.376 0.097 0.245 0.584 0.714 1.396 0.689 1.620 0.345 0.302 0.665 0.626 0.386 0.109 0.617
4.748 0.112 0.271 0.544 0.073 0.267 0.662 0.633 0.405 0.133 0.108 0.266 0.238 0.133 0.218 0.153 0.230 0.127 0.528 0.048 0.348 0.475 0.066
2.123 0.139 0.201 0.209 0.177 0.567 0.473 0.482 0.491 0.340 0.310 0.442 0.426 0.339 0.413 0.360 0.421 0.333 0.499 0.213 0.476 0.499 0.248
4.305 0.084 0.232 0.575 0.109 0.199 0.698 0.696 0.544 0.082 0.070 0.224 0.268 0.097 0.201 0.190 0.149 0.053 0.344 0.029 0.206 0.788 0.364
1.920 0.131 0.207 0.244 0.244 0.475 0.459 0.460 0.498 0.275 0.255 0.417 0.443 0.296 0.401 0.392 0.356 0.223 0.475 0.167 0.405 0.409 0.481
0.331 0.056 0.500 0.334 0.216 0.032 0.472 0.215 0.500 0.778 0.271 0.374 27.61 540.31 0.335 0.450 0.211 0.114 1.36 0.392 0.500 0.370 0.496 0.273 0.498 3.54 0.910 0.826 0.441 0.658 0.096 0.837
3.07 0.838 0.490 0.199 0.461 0.077 0.572
1.21 0.368 0.500 0.399 0.499 0.266 0.495
3.49 0.921 0.832 0.455 0.662 0.092 0.818
1.10 0.269 0.374 0.498 0.473 0.289 0.386
31,175 21,337 1,814 8,023 25,747
23,596 51,843 41,282 21,411 39,458 40,078 8,692 3,201 11,763 10,559 9,185 17,445 23,433 49,803 41,967
33,155 23,144 1,252 8,759 27,111
25,060 24,132 5,030 11,752 25,071
55,913 41,945 3,484 10,483 53,635
52,358 48,632 18,717 22,987 52,981
0.273 0.797
0.446 0.404
0.296 0.722
0.457 0.449
0.299 0.717
0.458 0.451
0.240 0.748
0.427 0.435

2003 Beneficiaries households Mean OUTCOMES (cont.) Health (unconditional): inscribed in SAPS (public system) pregnant women w/regular check-up all children<6 w/regular check-ups all women>35 w/regular pap smear all elderly w/regular check-up Education (unconditional): all children 4-5 attending pre-school All children<15 enrolled in school children in school receiving assistance all children 12-18 can read/write all adults can read/write adults enrolled in adult literacy pgm/ nivelacin competencia Employment: at least one member working Children 8-15 not in school&working at least one member with stable job+ at least one member enrolled in microemprendimiento pgm+ at least one member enrolled in employment programs+ at least one member enrolled in the local employment office+ at least one member enrolled in a training program+ Income: hhld receiving SUF hhld receiving asignacin familiar hhld receving PASIS Housing: hhld receving SAP received material to protect house from rain/cold+ received equipment for kitchen/ bedroom+ hhld applying for public housing No. Obs. st.dev. non-particip. households Mean Diff. Std. st.dev Means 2004 Beneficiaries households Mean st.dev.
111
Variables:
non-particip. households mean st.dev
0.983 0.044 0.211 0.385 0.077 0.169 0.674 0.559 0.166 0.704 0.114
0.129 0.204 0.408 0.487 0.267 0.374 0.469 0.497 0.372 0.457 0.318
0.914 0.036 0.163 0.384 0.100 0.110 0.582 0.481 0.158 0.855 0.064
0.280 0.185 0.370 0.486 0.299 0.313 0.493 0.500 0.365 0.352 0.244
0.987 0.045 0.168 0.424 0.082 0.153 0.652 0.530 0.165 0.678 0.067
0.113 0.208 0.374 0.494 0.275 0.360 0.476 0.499 0.371 0.467 0.250
0.933 0.035 0.152 0.411 0.117 0.109 0.575 0.471 0.176 0.853 0.041
0.250 0.184 0.359 0.492 0.321 0.311 0.494 0.499 0.381 0.354 0.198
0.835 0.0008
0.371 0.874 0.031 0.0001
0.332 0.013
0.831 0.0001 0.751 0.306 0.096 0.459 0.182
0.374 0.886 0.000 0.0008 0.433 0.867 0.461 0.027 0.295 0.498 0.386 0.040 0.231 0.166
0.318 0.029 0.339 0.161 0.195 0.422 0.372
0.606 0.116 0.115 0.131
0.489 0.320 0.319 0.337
0.187 0.291 0.061 0.199
0.390 0.454 0.238 0.400
0.639 0.117 0.124 0.110 0.248 0.233
0.480 0.322 0.330 0.313 0.432 0.423
0.189 0.285 0.071 0.176 0.045 0.013
0.391 0.452 0.257 0.381 0.207 0.113
0.329 0.470 3,495
0.241 0.428 9,509
0.373 0.484 3,495
0.230 0.421 9,509
Note: Own calculation from the longitudinal sample CASEN 2003-Encuesta Panel 2004. Variables are weighted using sampling weights. The column diff. std. means reports the difference in means of the variables between the beneficiary and non-beneficiary households, after they have been normalized to have mean zero and unit variance. (*) Income figures are not adjusted, i.e. without the application of the CEPAL correction. (+) indicates that the variable available only in 2004. The summary statistics for the intermediate indicators are calculated on the entire sample, irrespectively of whether the indicator applies to the population of interest (ex. Having children less than six, having at least one disabled member, etc.).
112
TABLE 1C
DESCRIPTIVE STATISTICS ON PERCEPTION qUESTIONS+, PANEL SAMPLE 2004 Head of household Beneficiaries Mean Situacin econmica en su infancia mejor Subjective welfare scale (5 ladder): pertenece al grupo socioeconmico bajo pertenece al grupo socioeconmico mediobajo Hizo algn trabajo por la comunidad - 2 ltimos aos Nadie lo ayudara a solucionar su problema, si tuviera un problema importante (social support) Situacin econmica en el futuro mejor que ahora (economic situation in the future will be better) Ha ido buscar por iniciatica propia ayuda a una institucin (looked for help out of own initiative) Aware of public programs in the community Satisfaction with life index: good or very good 0.398 st.dev. 0.49 nonparticipants Mean 0.349 st.dev. 0.477 Spouse head Beneficiaries Mean 0.359 st.dev. 0.48 nonparticipants Mean 0.336 st.dev. 0.472
0.704
0.456
0.479
0.499
0.655
0.475
0.400
0.489
0.230
0.421
0.343
0.474
0.273
0.445
0.397
0.489
0.251
0.434
0.197
0.398
0.238
0.426
0.206
0.404
0.165
0.371
0.162
0.369
0.153
0.360
0.156
0.362
0.599
0.49
0.527
0.499
0.628
0.483
0.562
0.496
0.428 0.543 0.718
0.495 0.498 0.45
0.216 0.481 0.783
0.412 0.5 0.412
0.448 0.529 0.736
0.497 0.499 0.441
0.212 0.493 0.814
0.408 0.5 0.389
Note: Own calculation from the longitudinal sample Encuesta Panel 2004. Variables are weighted using sampling weights. The subjective welfare questions (first 3 lines) were administered only to the head and his/ her spouse, only when present.
113
FIGURE 1
SUPPORT OF THE CAS DISTRIBUTION (PANEL SAMPLE), BY REGION
I II III IV V VI VII VIII IX X XI XII R.M. 300 400 500 600 700
Note: Own calculations: sample of all households with information on their CAS score (panel 2003-2004).
FIGURE 2
SUPPORT OF THE CAS DISTRIBUTION
CAS scores and cutoff scores .02 0 .005 kdensity CAS .01 .015
300
400
500 x
600
700
Note: Own calculations from the panel sample 2003-04. Sample of all households with information on their CAS score. Vertical lines represents the ranges of the municipal (dashed) and regional (solid) cutoff scores.
114
5. Results 5.1. Income and employment effects Results on the income and employment dimension are reported in Table 2, using the estimated program effects on the CAS score, separately by urban and rural and by year. For each outcome of interest, outcomes are based on matching on the CAS score, unconditionally, conditionally on a large set of household characteristics10, and finally conditioning on both household characteristics as well as on municipality fixed effects. Income effects: The short impact of the program on total income and labor income is overall small and mainly non significant across alternative specifications. Participating households have a preferential access to public transfers such as the family allowance (SUF, Subsidio Unico Familiar), the old age and disability pension (PASIS) and the potable water subsidy (SAP). The results in Table 2 suggest that households in Chile Solidario are more likely to have received the SUF, especially in urban areas, but less likely to be receiving the water subsidy, the old-age pension and another form of family allowance (in urban areas). The negative sign cannot be interpreted as a negative impact. These programs are assigned strictly on the basis of the within commune ranking of the CAS score among the applicant households. The fact that we observe negative effects might simply reflect the fact that participating households lagged behind in activating their demand even before the program (which is not controlled for by the observed covariates), and that Chile Solidario did not help bridge such differences in take-up for such programs. Labor Supply effects: Chile Solidario households exhibit very strong take-up of labor market programs: they are more likely to be participating to programs aimed at supporting self-employed and more likely to be participating to public employment/labor re-insertion and training programs. Participation rates increase by around 30 percentage points in urban areas, and about 14 percentage points in rural areas for self-employment programs. The same pattern is observed for public employment program (increased by about 6% points in urban areas, and 4% points in rural areas), while the effect on take-up training programs is significant only in urban areas. There is also a very strong effect in increasing the likelihood of household members to be enrolled in the local employment office (OMIL), one of the minimum conditions previewed by the Chile Solidario program for unemployed members. Being enrolled in such offices not only should facilitate the process of looking for a job, but also represents a pre-condition for eligibility to various public training programs. While the program activated a significant take-up on programs that might increase the employment prospects for the participating households in the medium term, the results do not translate into current gains in their labor supply. There
10
Namely, family composition, age, sex marital status and education of the head of the household, presence/age of the spouse, indicators for ethnic minority and indicators of basic asset ownership.
115
TABLE 2
IMPACT OF THE PROGRAM ON INTERMEDIATE INDICATORS: EMPLOYMENT, HOUSING AND INCOME
Matching on CAS 2003 Panel a: urban Unconditional dimension Identification if disabled: enrolled in Nat. Registry (ident6) dimension Employment: at least one member working at least one member with stable job+ (trab1) Share of members employed Share of members active children<15 not in school and working (trab2) at least one member enrolled in the local empl office (trab3) + at least one member enrolled in programas de micro-emprend.+ at least one member enrolled in a training program+ at least one member enrolled in programas de empleo+ dimension Housing received material to protect house from rain/cold+ received equipamiento for kitchen/bedroom+ hhld receiving subsidio agua potable (SAP) hhld postulando a vivienda dimension income hhld receiving subsidio nico familiar (SUF) hhld receving public pensions (PASIS) Hhld income p.c. Hhld labor income p.c. Hhld non labor income p.c. Hhld public transfers p.c. Conditional (hhld) Conditional (hhld+ Commune) 0.003 (0.060) 0.030 (0.030) 0.033 (0.027) 0.029 (0.025) 0.001 (0.002) Unconditional Conditional (hhld) Conditional (hhld+ Commune) 0.060 (0.053) 0.053* (0.028) 0.046 (0.037) 0.004 (0.027) 0.005 (0.026) 0.186*** (0.038) 0.294*** (0.027) 0.113*** (0.027) 0.057** (0.022) 0.127*** (0.029) 0.211*** (0.025) 0.071*** (.029) 0.110*** (0.033) 0.113*** (0.032) 0.024 (0.022) 2,454 (1,861) 4,839*** (1,720) 386 (487) 2,771*** (754) 2004
0.042 (0.067) 0.035 (0.033) 0.003 (0.028) 0.017 (0.026) 0.001 (0.002)
0.063 (0.067) 0.017 (0.027) 0.011 (0.028) 0.005 (0.026) 0.001 (0.002)
0.047 (0.062) 0.017 (0.032) 0.030 (0.040) 0.005 (0.029) 0.005 (0.027) 0.282*** (0.040) 0.322*** (0.028) 0.135*** (0.027) 0.077*** (0.022) 0.176*** (0.030) 0.239*** (0.026) 0.085** (0.031) 0.168*** (0.035) 0.154*** (0.038) 0.079*** (0.027) 1,695*** (1,975) 1,732 (1,904) 226 (474) 2,019* (1,109)
0.028 (0.061) 0.045 (0.029) 0.041 (0.038) 0.002 (0.027) 0.006 (0.026) 0.271*** (0.040) 0.321*** (0.028) 0.133*** (0.027) 0.064*** (0.022) 0.145*** (0.029) 0.234*** (0.026) 0.067** (0.031) 0.136*** (0.034) 0.081** (0.033) 0.009 (0.022) 169 (1,794) 2,245 (1,666) 74 (478) 2,151*** (711)
0.057** (0.031) 0.102*** (0.037) 0.178*** (0.039) 0.082*** (0.026) 3,457*** (1,826) 1,783 (551) 438 (670) 1,236 (966)
0.070** (0.031) 0.077** (0.036) 0.100*** (0.033) 0.009 (0.020) 407 (1,617) 2,226 (1,567) 623 (664) 2,442*** (613)
0.080** (0.029) 0.063** (0.035) 0.125*** (0.032) 0.024 (0.020) 4,187 (1,683) 5,753*** (1,607) 997 (667) 2,564*** (642)
116
Table 2 (cont.)
Matching on CAS Panel B: Rural Unconditional dimension Identification if disabled: enrolled 0.064 in Nat. Registry (0.052) (ident6) dimension Employment: at least one member 0.011 working (0.029) at least one member with stable job+ (trab1) Share of members 0.028 employed (0.023) Share of members 0.049** active (0.022) children<15 not in 0.001 school and working (0.002) (trab2) at least one member enrolled in local empl. office (trab3)+ enrolled in microemprendimiento pgm+ enrolled in a training program+ enrolled in programa de empleo+ dimension Housing received material to protect house from rain/cold+ received equipment for kitchen/bedroom+ hhld receiving water 0.012 subsidy (SAP) (0.017) hhld applying for 0.045 public housing (0.028) dimension income hhld receiving sub0.111*** sidio nico familiar (0.034) (SUF) hhld receving public 0.057** pensions (PASIS) (0.027) Hhld income p.c. 1,648 (1,455) Hhld labor income 1810 p.c. (1,374) Hhld non labor 331 income p.c. (367) Hhld public trans494.882 fers p.c. (767) 2003 Conditional (hhld) Conditional (hhld+ Commune) 0.019 (0.051) 0.017 (0.027) Unconditional 2004 Conditional (hhld) Conditional (hhld+ Commune) 0.110** (0.043) 0.028 (0.026) 0.004 (0.033) 0.005 (0.022) 0.001 (0.021) 0.001 (0.002) 0.110*** (0.025) 0.123*** (0.028) 0.018 (0.020) 0.031** (0.014) 0.098*** (0.027) 0.219*** (0.021) 0.012 (0.016) 0.021 (0.029) 0.074*** (0.027) 0.025 (0.022) 806 (1,373) 564 (1,205) 293 (495) 1,077* (602)
0.019 (0.051) 0.017 (0.027)
0.087* (0.052) 0.002 (0.028) 0.012 (0.036) 0.012 (0.024) 0.010 (0.023) 0.001 (0.002) 0.147*** (0.025) 0.136*** (0.029) 0.023 (0.021) 0.044*** (0.014) 0.118*** (0.028) 0.240*** (0.022) 0.019 (0.015) 0.014 (0.030) 0.063** (0.027) 0.009 (0.022) 1,604 (1,340) 102 (1,243) 192 (492) 1,311** (548)
.0110** (0.051) 0.029 (0.027) 0.006 (0.034) 0.010 (0.023) 0.007 (0.022) 0.002 (0.002) 0.134*** (0.025) 0.139*** (0.029) 0.032 (0.021) 0.040**** (0.014) 0.114*** (0.028) 0.236*** (0.022) 0.003 (0.015) 0.002 (0.027) 0.068** (0.028) 0.035 (0.021) 1,352 (1,365) 45 (1,337) 35 (370) 1,272** (527)
0.028 (0.022) 0.047** (0.021) 0.001 (0.002)
0.028 0.022 0.047 (0.021) 0.001 (0.002)
0.003 (0.017) 0.006 (0.030) 0.108*** (0.033 0.090** (0.027) 691 (1,452) 2,428 (1294) 377 (489) 1360* (818)
0.001 (0.016) 0.013 (0.029) 0.060** (0.028) 0.026 (0.022) 2,403* (1,311) 694 (1,304) 73 (371) 1,636*** (507)
Note: Matching estimator: matching on the CAS score, with replacement. 3 nearest neighbors. Standard errors in parentheses based on the estimated variance of the sample average treatment effect (as in Abadie and Imbens (2006)). * significant at 10%; ** significant at 5%, *** significant at 1%. Household characteristics include: household size and composition (share of males and females 0-, 6-17, 18-64, older than 65 is the excluded category), whether both head and spouse are in the household, sex, age, marital status and education dummies of the head, age spouse, whether the household is an ethnic minority, ownership of durables (refrigerator, calefont).
117
is no sign of improvements of the share of members who are employed, nor on the share of members who have a stable employment (self-reported). The only positive and significant effects on labor force participation are observed in rural areas, with gains in the share of members who are active. On the one hand, the inconclusive evidence leaves pending important questions on the ability of the program in helping households achieve a sustained exit from extreme poverty. qualitative work11 clearly suggests that improvements in employment (especially those related to having a stable source of income in the household) and housing are among the most important aspirations of participating families and those conditions that are perceived as structural factors preventing households to escape extreme poverty. In this light, they are also perceived as the most difficult minimal conditions to meet. On the other hand, the short term horizon of the current analysis might not be sufficient to observe any impact along these dimensions. In principle, the employment and earnings trajectories of those households who have participated to self-employment/public works/training programs and/or those members who had adult literacy program might improve in the medium run. This short term effects are potentially consistent with the logic of the program to satisfy some basic needs in the short term, while at the same time building the assets to allow households to improve their welfare in a sustained way in the medium-long run. However, the evidence on the effectiveness of active labor market program in North American and European studies is mixed, with modest impact on increasing employment rates, though not much impact on earnings (Heckman, LaLonde and Smith (1999)). If any, the literature shows that some of the estimated gains are not sustained over time when longer follow-up data are available. 5.2. Housing effects Having their own house (casa propia) and improving its basic infrastructure also feature as a very important dimension among the aspirations of participating families (Asesoras para el Desarrollo (2005), Universidad de Chile (2004b, c), FOSIS (2004b).12 Owning a house reinforces the identity of the household as an independent unit and represents a capital that can be bequeathed to their children, together with the investment in human capital. Besides the ownership status, basic sanitary and housing infrastructure are important correlates of household welfare (FOSIS (2004a)): having basic infrastructure has potential complementarities with health outcomes (access to safe water and sanitation) as well as family dynamics (in terms of a space that allows for better roles and interactions among different household members). The results in Table 2 show significant effects on the enrollment in housing programs in urban areas. The estimated effect ranges from 7% in 2003
11 12
See footnote 6. Textual analysis of the life projects of those households exiting the two-years period of psycho-social support (ficha final Puente) also confirms that the modal combination of words in the aspirations of the participating households relates to having their own house, as well improving their own house (Silvas presentation at MIDEPLAN, October 2005).
118
to double to 14% in 2004. Compared to an average take-up of 24% of nonparticipants (stable over time), this amounts to an estimated sizable increase ranging between 30 to 60%. Enrolment in such public programs requires that the households have set some minimum amount of savings to be eligible. Possibly the cash received through public transfers (either though the bono or through the Subsidio Unico Familiar) has allowed the participating households to save towards this objective. Housing is one of the dimensions along which there might have been some rationing of the supply side. The results with municipality fixed effects, which control for time-invariant differences in the initial availability of housing, seem to rule out the rationing explanation. The results with community effects do not differ significantly from the other specifications within areas (rural/urban), (though it might still be possible that the rationing applies uniformly to all rural areas). Table 2 also shows significant effects of the program on the receipt of basic housing equipment (of about 23 percentage points) as well as basic material to protect the house from rain/cold (ranging from 10-15 percentage points). These results are also robust to controlling for community effects. Overall, the results provide evidence that participating households are more likely to activate themselves to connect to the social protection network to bridge the initial gaps in their housing situation. 5.3. Impact on human capital outcomes Table 3 reports all the results that relate to various dimensions of health and education. The choice of the outcomes of interest follows the list of intermediate indicators that are set by the program as minimum conditions to be achieved by participating households and that can be measured in the CASEN survey. All variables are computed as averages of individual outcomes at the household level (independently of whether the condition apply to specific subgroups of households or not), having in mind the objective to obtain an average effect of the program on the overall population of participating households. The only exception is given by outcomes that refer to households with disabled members, for whom the baseline characteristics and the expected behavioral response are expected to be substantially different. Education effects: Overall, the results suggest significant and consistent increases in the likelihood of having all children aged 4-5 year olds enrolled in a pre-school. The effects for pre-school enrolment are in the range of 4-6 percentage points, consistently found in both urban and rural areas, as well as across different methods. Availability of preschools or financial constraints are not perceived to be an issue: cultural perceptions that the child is too young, or that he/she is better off taken care at home account for 90% of the self-reported reasons for non-enrolment (MIDEPLAN, analysis of CASEN 2003). On the supply side, there are different pre-school programs that have been adapted to reach the target population by providing free access as well as flexible hours to meet the needs of working mothers, even with temporary jobs or households where the head of the household is unemployed and the mother is looking for work.
119
TABLE 3
IMPACT OF THE PROGRAM ON INTERMEDIATE INDICATORS: HEALTH AND EDUCATION
2003 Uncondit (1) dimension Education all children 4-5 attending pre-school (educ1) children<15 enrolled in school (educ3) children in school receiving assistance (educ4) all children 12-18 can read/write (educ5) adults enrolled in adult literacy program/ nivelacin competencia dimension Health inscribed in SAPS (salud1) pregnant women w/regular check-up (salud2) all children<6 w/regular check-ups (salud4) all women>35 w/regular pap smear (salud5) all elderly w/regular check-up (salud7) Sample size Panel B: Rural dimension Education all children 4-5 attending pre-school (educ1) children<15 enrolled in school (educ3) children in school receiving assistance (educ4) all children 12-18 can read/write (educ5) adults enrolled in adult literacy program/ nivelacin competencia dimension Health inscribed in SAPS (salud1) pregnant women w/regular check-up (salud2) all children<6 w/regular check-ups (salud4) all women>35 w/regular pap smear (salud5) all elderly w/regular check-up (salud7) Sample size (1) (2) (3) (1) 0.004 (0.022) 0.054 0.034 0.066* (0.034) 0.022 (0.025) 0.019 0.015 (2) (3) w/hhld controls (2) w/hhld controls+ communeFE (3) Uncondit (1) 2004 w/hhld w/hhld controls+ controls communeFE (2) (3)
Panel A: Urban
0.062** (0.027) 0.096** (0.038) 0.090** (0.038) 0.005 (0.027) 0.025 (0.024)
0.027 (0.023) 0.023 (0.025) 0.040 (0.033) 0.009 (0.027) 0.001 (0.024)
0.026 (0.023) 0.019 (0.025) 0.063* (0.033) 0.005 (0.028) 0.002 (0.024)
0.071** (0.027) 0.089** (0.038) 0.033 (0.039) 0.010 (0.028) 0.039** (0.019) 0.019 (0.012) 0.012 (0.017) 0.040 (0.029) 0.024 (0.038) 0.038 (0.024)
0.042* (0.023) 0.000 (0.024) 0.025 (0.033) 0.025 (0.027) 0.025 (0.019)
0.035 (0.023) 0.005 (0.024) 0.006 (0.033) 0.014 0.027 0.026 (0.018)
0.051*** 0.036** 0.038** (0.017) (0.017) (0.016) 0.022 0.019 0.013 (0.015) (0.015) (0.015) 0.059** 0.022 0.025 (0.029) (0.027) (0.026) 0.060 0.040 0.025 (0.037) (0.031) (0.031) 0.110*** 0.060** 0.051** (0.026) (0.021) 0.021 5563
0.001 0.010 0.012 (0.012) 0.035** 0.044** (0.017) (0.017) 0.015 0.029 (0.027) (0.026) 0.029 0.037 (0.032) 0.033 0.003 0.000 (0.021) (0.021) 5563
0.047** 0.051** 0.046** (0.021) (0.019) (0.018) 0.046 0.036 0.040* 0.034 (0.023) (0.024) 0.046 0.011 0.006 0.034 (0.026) (0.026) 0.022 0.010 0.006 (0.023) (0.022) (0.022) 0.053*** 0.054*** 0.050*** (0.017) (0.017) (0.017) 0.033*** (0.012) 0.022* (0.012) 0.035 (0.027) 0.041 (0.032) 0.029 (0.022) 0.030** (0.012) 0.016 (0.012) 0.000 (0.024) 0.009 (0.028) 0.008 (0.020) 3270
0.017 0.013 (0.019) (0.019) 0.044* 0.052** (0.022) (0.023) 0.008 0.005 (0.029) (0.028) 0.006 0.003 (0.025) (0.024) 0.015 0.015 (0.015) (0.015) 0.005 (0.011) 0.009 (0.011) 0.015 (0.023) 0.016 (0.028) 0.018 (0.019) 3270 0.002 (0.010) 0.009 (0.011) 0.022 (0.022) 0.026 (0.028) 0.010 (0.019)
0.030** 0.010 (0.012) (0.012) 0.015 0.005 (0.011) (0.012) 0.009 0.026 (0.023) (0.024) 0.015 0.067** (0.027) (0.034) 0.001 0.053** (0.020) (0.022)
Note: Panel Sample 2003-2004. Columns (1-3): Matching on the CAS score, with replacement, 3 nearest neighbors. See footnote on Table 2 for more details. Columns (4-5): Reported coefficients (Huber-White standard errors are in parentheses) from a regression of the outcome variable on an indicator of participation and a cubic polynomial in the CAS score.
120
School enrolment of children from 6 to 15 has mildly improved (with estimates ranging between 7-9%), although the results are not robust across all matching specifications. Household in urban areas are also more likely to have taken-up complementary programs of school materials, meals, and dental care directed to subsidize direct costs of schooling for households with lower socioeconomic conditions. There are no fees for public schools in Chile, so most of costs of enrolment are indirect (opportunity cost of the childs time). There are no significant differences in terms of literacy of children aged 12-18. As part of its comprehensive strategy, the program also targets illiterate adults or adults that would like to complete their elementary middle school levels. On the benefit side, literacy or improvements in the educational attainment can increase the adults self-esteem, and help process information about services/ jobs and be instrumental to supporting the children in their educational learning. The costs of participation are not only measured in terms of opportunity cost of their leisure time after work, but also in terms of psychological costs. In this respect, the psycho-social support by the social worker is instrumental to discuss the potential benefits of such programs, and to encourage the potential participants to feel motivated and capable of attaining such an objective. The results show a statistically significant take-up of adult literacy and education completion programs of around 4% in urban areas and 5 percentage points in rural areas. Health effects: The impact of the program on health outcomes is more muted than the one on educational outcomes. The only consistent result is that participating households are more likely to be enrolled in the public health system (SAPS) (2-3% in urban areas, 3% in rural areas). The impact on health visits for preventive care is found on some subgroups (for health visits for children below six years of age of the order of 4-6 percentage points, only in rural areas, and for women aged 35 or older for their pap smear of the order of 6-7%, mostly in 2004 for rural areas and in 2003 in urban areas). The results on elderly are not significant in urban areas and often negative in rural areas. We believe that the negative effects are more of a reflection of the differences in the composition of the elderly population in our sample and of the lack of sufficient covariates that are specific to this age group13 rather than credible negative estimates of impact. 5.4. Evidence on perceptions and orientation towards the future The 2004 questionnaire includes some basic perception questions administered to the head of the household and/or her spouse. Although some of the differences might be capturing underlying differences in personality traits and personal attitudes and preferences, we assume that the distribution of these unobserved characteristics is uniformly distributed across households with similar socio-economic characteristics and is unrelated to program participation under our identification assumptions.
13
Namely we control for the share of male and female elderly in the households but fail to account for their initial differences in health status.
121
TABLE 4
IMPACT OF THE PROGRAM ON PERCEPTIONS (2004)
+ hhld controls, commune FE (3)
Unconditional (1) Panel A: Urban Situacin econmica en su infancia mejor Subjective welfare scale (5 ladder): - pertenece al grupo socioeconmico bajo Hizo algn trabajo por la comunidad - 2 ltimos aos Nadie lo ayudara a solucionar su problema, si tuviera un problema importante Situacin econmica en el futuro mejor que ahora Ha ido buscar por iniciativa propia ayuda a una institucin Aware of social services in the community Satisfaction with life index: good or very good Panel B: Rural Situacin econmica en su infancia mejor Subjective welfare scale (5 ladder): - pertenece al grupo socioeconmico bajo Hizo algn trabajo por la comunidad - 2 ltimos aos Nadie lo ayudara a solucionar su problema, si tuviera un problema importante Situacin econmica en el futuro mejor que ahora Ha ido buscar por iniciativa propia ayuda a una institucin Aware of social services in the community Satisfaction with life index: good or very good 0.000 0.034 0.012 0.035 0.013 0.030 0.027 (0.027) 0.136*** (0.035) 0.050 (0.032) 0.110*** (0.034) 0.043 (0.029) 0.019 (0.043) 0.008 (0.040) 0.056*** (0.029) 0.064** (0.031) 0.176*** 0.040 0.099** (0.037) 0.189*** (0.038) 0.061* 0.036
+ hhld controls (2)
0.008 (0.043) 0.016 (0.039) 0.047 (0.030) 0.034 (0.031) 0.114*** (0.040) 0.070* (0.037) 0.161*** (0.038) 0.015 (0.036)
0.067 (0.042) 0.068* (0.038) 0.060** (0.030) 0.029 (0.030) 0.097** (0.039) 0.041 (0.036) 0.129*** (0.037) 0.017 (0.035)
0.017 (0.034) 0.000 (0.034) 0.005 (0.029) 0.019 (0.027) 0.077** (0.035) 0.028 (0.031) 0.100*** (0.034) 0.056* (0.029)
0.030 (0.033) 0.010 (0.033) 0.007 (0.028) 0.011 (0.026) 0.068** (0.034) 0.024 (0.031) 0.100*** (0.032) 0.050* (0.028)
Note: Panel Sample 2003-2004. Columns (1-3): Matching on the CAS score, with replacement, 3 nearest neighbors. See footnote on Table 2 for more details on the set of covariates.
Results are presented in Table 4. Households in Chile Solidario are more likely to be aware of social services in the community (10 percentage points in rural areas and 13-16 percentage points in urban areas, corresponding to an increase of the order of 20-30% relative to the non-participants). This result is in line with the main objectives of bridging the demand gap. Households in urban areas are also reporting to be more likely to proactively look for help from local institutions (7 percentage points).
122
Finally, households seem to be more optimistic about their future socioeconomic status (7-8 percentage points in rural areas and 10-11 percentage points in urban areas, corresponding to an increase of about 15-20% relative to non-participation). This improved outlook, even if it is measured with a basic perception question, is likely to be correlated with their orientation towards the future, and their willingness to invest in assets that improve their likelihood to eventually escape extreme poverty over time. 6. Conclusions This paper provides estimates of the short run effects of Chile Solidario on a large array of household outcomes. The results focus on the first two cohorts of participants, where most of the beneficiaries are still under the psychosocial support of the social worker assigned to them. The main results show a significant and substantial effect on the take-up of cash assistance and social services, which was one of the main objectives of the program in its inception. The rationale for the intervention was that households in extreme poverty were previously observed to be disconnected from the public network of social services, and the program seemed to have bridged part of this gap. Second, we find that the program, in its two first years of operations, improves educational and some health outcomes of the participating households, though there are no effects on labor supply or income. Finally, we describe suggestive evidence that the psycho-social support was an important factor in enabling this change, by increasing awareness of social services in the community as well as households orientation towards the future. The comparison with other conditional transfers programs comes naturally to mind, though it should be exercised with extreme caution. The scope of the program (reaching 5% of the population), the institutional strength of local municipalities and the vast array of social services available in Chile makes it hard to extrapolate the results to other countries in the Latin American region. Nonetheless, the methodological approach that works jointly on the demand and supply side of social services is an innovation with respect to traditional conditional cash transfers and has already attracted attention from other countries in Central America (such as Guatemala, Honduras, and Colombia). Both types of approaches show gains in human capital indicators with increased health and education visits. Gertler et al (2005) provide preliminary evidence that some of the substantial cash transfer received by Mexican families in the context of Progresa (now Oportunidades) has been saved and used to finance in micro-enterprise activities and increased investments in farm assets and agricultural activities. In the case of Chile Solidario, future analysis beyond the short term will provide important insights as of whether the income and employment gains are going to be achieved through a different strategy of intervention. In the medium term, the program aims at removing structural bottlenecks by strengthening the human capital of adults and expanding their employment opportunities and productive activities (via education completion/training/public employment or self-employment programs). This expansion of opportunities through this second component is deemed essential to help sustain the participants exit from
123
extreme poverty. The program impact on poverty may take time to materialize, and exit from extreme poverty, when achieved, may or may not be sustained over time. Furthermore, participating households that have been linked to the supply side of social services (red local de proteccion social) though the program are expected to access the red independently in the future when faced by idiosyncratic shocks. The program over time has become a building block of the social protection system for the poor. It will be important to document its effectiveness to prevent households from falling back into poverty when faced by uninsured risk. Finally, one of the crucial innovations of the program is to bring the psychological dimension at the center of a large scale poverty intervention. The paramount importance of the psycho-social support, well documented in beneficiaries assessments and in the qualitative work, has been only touched upon in the current analysis by looking at a few isolated perception questions. There is scope for improving our understanding on such important dimensions by enriching the quality of instruments for measurement and study how gains in the psychological dimension correlate with changes in socio-economic conditions. References Appadurai, A. (2004). The Capacity to Aspire: Culture and the Terms of Recognition, Chapter 3 in Culture and Public Action, V. Rao and M. Walton eds., Stanford University Press, Stanford. Abadie, A. and G. Imbens (2006). Large Sample Properties of Matching Estimators for Average Treatment Effects, Econometrica, 74 (1); 235-267. Asesoras para el Desarrollo (2005). Necesidades y Aspiraciones Prioritarias de las Familias que han finalizado la etapa de apoyo psicosocial del sistema de proteccin social Chile Solidario, Santiago, Chile. Bourguignon, F., and S. Chakravarty (2003). The measurement of multidimensional poverty, Journal of Economic Inequality, 1; 25-49. Blundell, R., M. Costa Dias, C. Meghir and J. van Reenen (2004). Evaluating the Employment Impact of a Mandatory Job Search Program, Journal of the European Economic Association, 2 (4); 569-606. Carneiro, P., E. Galasso and R. Ginja (2009). The Impact of Providing PsychoSocial Support to Indigent Families and Increasing their Access to Social Services: Evaluating Chile Solidario, mimeo, Development Research Group, The World Bank. Clert, C. and q. Wodon (2001). The Targeting of Government Programs in Chile, Background Paper for Poverty and Income Distribution in a High Growth Economy - The Case of Chile 1987-98, The World Bank. Chay, K., P. McEwan and M. Urquiola (2005). The Central Role of Noise in Evaluating Interventions that use Test Scores to Rank Schools, American Economic Review, 95(4), pp. 1237-58. Currie, J. (2004), The Take Up of Social Benefits, NBER Working Papers 10488, National Bureau of Economic Research, Inc. De Janvry, A. and E. Sadoulet (2005). Making Conditional Cash Transfer Programs More Efficient: Designing for Maximum Effect of the Conditionality, processed, University of California at Berkeley, Department of Agricultural Economics.
124
Finan, F., A. de Janvry, E. Sadoulet, and R. Vakis (2005). Can Conditional Cash Transfers Serve as Safety Nets to Keep Children at School and Out of the Labor Market?, processed, University of California at Berkeley, Department of Agricultural Economics. Forthcoming in Journal of Development Economics. Galasso, E. (2006). With their effort and one opportunity: Alleviating extreme poverty in Chile (2006), mimeo, Development Research Group, The World Bank. Gertler, P., and S. Boyce (2001). An experiment in incentive-based welfare: The impact of PROGRESA on health in Mexico. Mimeo, University of California, Berkeley. Gertler, P. (2004). Do Conditional Cash Transfers Improve Child Health? Evidence from PROGRESAs Control Randomized Experiment, American Economic Review, Papers and Proceedings, 94 (2); 336-341. Gertler, P., S. Martinez and G. Rubio (2005). Investing Cash Transfers to Raise Long Term Living Standards, processed, The World Bank. Heckman, J., R. LaLonde and J. Smith (1999). The Economics and Econometrics of Active Labor Market Programs. In Orley Ashenfelter and David Card, eds., Handbook of Labor Economics, Volume 3A. Amsterdam: North-Holland; 1865-2097. Universidad de Chile, Facultad de Ciencias Sociales (2004a). Informe. Evaluacin del Estado de Avance del Sistema Chile Solidario, Santiago, Chile. Universidad de Chile, Facultad de Ciencias Sociales (2004b). Efectos de la Intervencin Psicosocial en Mujeres que participan directamente en el Sistema Chile Solidario, Santiago, Chile. Universidad de Chile, Facultad de Ciencias Sociales (2004c). Resultados Estudio Evaluativo Programa Puente y Chile Solidario, Santiago, Chile. Universidad de Chile, Facultad de Ciencias Sociales (2005). Trayectorias Laborales en Familias del Programa Puente, Programa de Estudios Desarrollo y Sociedad (PREDES), Santiago, Chile. FOSIS (2004a). Avance de las Obras, Cuadernillo de Trabajo N 1, Series Reflexiones Desde el Puente, Santiago, Chile. FOSIS (2004b). Las Condiciones Mnimas para la Construccin del Puente, Cuadernillo de Trabajo N 3, Series Reflexiones Desde el Puente, Santiago, Chile. FOSIS (2004c). Los Apoyos Familiares. Los otros Constructores del Puente, Cuadernillo de Trabajo N 4, Series Reflexiones Desde el Puente, Santiago, Chile. FOSIS (2005a). El Plano de los Servicios para Emplazar el Puente: Las Redes Locales de Intervencin, Cuadernillo de Trabajo N 5, Series Reflexiones Desde el Puente, Santiago, Chile. FOSIS (2005b). Con su Esfuerzo y una Oportunidad, Historias de vida de familias que participan en el programa Puente, Santiago, Chile. MIDEPLAN (2002). Estrategia de Intervencin Integral a favor de Familias en Extrema Pobreza, Divisin Social. Santiago, Chile. MIDEPLAN (2004). Pobreza, Distribucin del Ingreso e Impacto Distributivo del Gasto Social, volumen 1, serie CASEN 2003, Divisin Social. Santiago, Chile.
125
Hahn, J., P. Todd and W. van der Klauww (2001). Identification of Treatment Effects by Regression-Discontinuity Design, Econometrica; 201-209. Lindert, K., E. Skoufias, and J. Shapiro (2005). Redistributing Income to the Poor and the Rich: Public Transfers in Latin America and the Caribbean, paper presented at the Annual meetings of the Latin American and Caribbean Economic Association, Paris. Ravallion, M. (2005). Evaluating Anti-Poverty Programs, chapter prepared for the Handbook of Agricultural Economics, World Bank Policy Research Working Paper N 3625, The World Bank. Van der Klaauw (2002). Estimating the Effect of Financial Aid Offers on College Enrollment: A Regression Discontinuity Approach, International Economic Review, 43 (3); 1249-1287.
126
APPEnDIx A1. Structure of the longitudinal panel 2003-2004

Admin. Data (base Puente) Household survey data CASEN 03 PANEL 04
2002
2003
2004
2005
225,000 participants
73,000 hhlds
13,000 hhlds
A2. Construction of the panel sample 2003-2004 The objective of the panel survey was to follow up over time only a sample of the CS beneficiaries and their matched comparison group. The selected longitudinal sample was composed of about 3,400 participating households (comprising of 60% of the beneficiaries interviewed in the CASEN 2003, and of 186 new beneficiaries of the 2004 cohort identified by cross-checking the administrative list of beneficiaries and with the names/addresses of the original CASEN). The matched comparison group was constructed by estimating a propensity score of participation into the program separately for four broad geographic areas.14 The list of covariates included household size and age composition, whether the household belongs to an ethnic minority or speaks a minority language, head characteristics (age dummies, education dummies, marital status dummies,
14
The four geographic areas selected by Mideplan are: regions I-IV, regions V-VII and XIII, regions VIII-X, regions XI-XII.
127
labor force history in 2000), housing characteristics, asset indicators, household non-labor income per capita, a rural indicator and dummies for the regions, and interactions between housing indicators and rural. The matching was done among households who reported having filled in a ficha CAS. The prediction of the propensity score and the balancing of the covariates performed better than in the case where the comparison group was drawn from all the households sampled in the CASEN). The matching was done choosing the 3 nearest neighbors for each beneficiary within each geographic area. Matching was done with replacement, based on the log of the odds ratios15 and imposing a common support in the propensity score among both beneficiaries and non-participants. Comparison households were forced to be chosen within the same geographic area and zone (rural/urban) for practical convenience. The final sample of original non-participants selected by MIDEPLAN for the panel includes 9,500 households.
15
Heckman and Todd (1995) show that in the case of sampling situations where program participants are oversampled (choice-based sampled data), matching can still be applied when matching is done on the odds ratio rather than on the propensity score. Matching on p/(1-p), the odds ratios, which are a monotonic transformation of the propensity score p, overcomes the problem of over-sampling.
Estudios de Economa. F. Hoces de la Guardia, A. Hojman, O. Larraaga Evaluating the Chile /Vol. 38 - N 1, Junio 2011. Pgs. 129-168
129
Evaluating the Chile Solidario program: results using the Chile Solidario panel and the administrative databases*
Evaluando el programa Chile Solidario: resultados usando el panel Chile Solidario y datos administrativos
Fernando Hoces de la Guardia** Andrs Hojman***, Osvaldo Larraaga**** Abstract This paper presents the results of a three-year evaluation performed on the first cohorts of Chile Solidario, the most important anti-poverty program in Chile. The paper presents a description of the program, emphasizing the mechanism by which people were admitted into the program. We then propose evaluation strategies and discuss their validity. The final evaluation is conducted using a Matching estimator, and we discuss the principles surrounding the potential for this to be a valid evaluation method. The initial results using the Chile Solidario Panel suggest that the program had positive effects on psychosocial welfare and on take-up of subsidies and social programs. However, it is not possible to obtain reliable results due to data shortcomings, particularly the lack of baseline data. In order to solve the problem, we generated a database using six years of administrative data, including around 1,000,000 family records per year. A method for overcoming the treatment substitution problem is discussed and implemented. Results are much more robust than those of the Chile Solidario Panel and show small, but clearly positive effects for several variables, especially the number of workers in the family, the percentage of workers in the family and the employment of the head of the family. Key words: Matching, Extreme poverty, Impact evaluation, Treatment substitution, Administrative data.
Valuable contributions to this work have been provided by Jos Miguel Benavente, David Bravo, Paula Castro, Dante Contreras, Paz Garcs, Rodrigo Herrera, Jorge Manzi, Toms Rau, Mara de la Luz Ramrez, Jaime Ruiz-Tagle, Vernica Silva, Sergio Urza y Ana Mara Vliz. We thank the support from the Centro de Microdatos, Iniciativa Cientfica Milenio (Project P075-023-F). The results and opinions are exclusive responsibility of the authors. ** Centro de Microdatos, Universidad de Chile and University of Pennsylvania Student. *** Centro de Microdatos, Universidad de Chile and University of Chicago (PhD Student). **** Universidad de Chile Professor and PNUD Program Officer.
130
Resumen Este artculo presenta los resultados de una evaluacin de tres aos realizada en las primeras cohortes del Programa Chile Solidario, el programa de reduccin de la pobreza ms importante de Chile. El artculo presenta una descripcin del programa, haciendo hincapi en el mecanismo por el cual las personas fueron admitidas en el programa. A continuacin, propone estrategias de evaluacin y discute su validez. La evaluacin final se realiza mediante un estimador de Matching (pareo), y se discuten los principios en base a los cuales se tratara de un mtodo de evaluacin vlida. Los resultados iniciales usando el Panel de Chile Solidario sugieren que el programa ha tenido efectos positivos sobre el bienestar psicosocial y en la adopcin de subsidios y programas sociales. Sin embargo, no es posible obtener resultados fiables debido a las deficiencias de datos, en particular la falta de una lnea de base. Con el fin de resolver el problema, hemos generado una base de datos utilizando seis aos de datos administrativos, incluyendo alrededor de 1.000.000 de registros de familias por ao. Un mtodo para superar el problema de sustitucin de tratamiento es discutido y aplicado. Los resultados son mucho ms robustos que los del Panel de Chile Solidario y muestran pequeos, pero evidentes efectos positivos de varias variables, especialmente relativos al nmero de trabajadores en el hogar, el porcentaje de trabajadores en el hogar y el empleo del jefe de hogar. Key words: Pareo, Pobreza extrema, Evaluacin de impacto, Sustitucin de tratamiento, Datos administrativos. JEL Classification: C14, I30, I38.
1. Introduction Chile Solidario is the most important anti-poverty program in Chile. It is a joint coordination effort between all public sectors, so as to strongly focus social benefits on families living in extreme poverty conditions1. It consists of three main components: psychosocial support, in the form of a subprogram called Puente (Bridge); preferential access to the social services and programs offered by the State; and guaranteed access to the subsidies offered by the State, including a small amount of money provided exclusively to program beneficiaries. In the last year covered by this evaluation (2006) there were more than 250,000 participating families.
1
The executives of Chile Solidario refer to it as a Social Protection System, which coordinates different programs. Within the Impact Evaluation literature it can be considered a program, only that its benefits are selected from a menu instead of being identical to all of those in the treatment group.
Evaluating the Chile / F. Hoces de la Guardia, A. Hojman, O. Larraaga
131
The impact evaluation was designed over a period beginning one year after the programs implementation. The instrument designed for this purpose was the Chile Solidario Panel Survey (PCHS). This survey presents some sampling difficulties. Since the evaluation was designed after the implementation began, there are no baseline data, thereby causing lower credibility in the estimations. In order to solve this problem, and with the goal of collecting the variables that determine the mechanism by which beneficiaries are admitted into the program (hereafter referred to as the admission mechanism), apart from PCHS the evaluation was complemented with administrative data from the Puente Program and from the CAS2 databases, an instrument used at the time to assign State benefits (administered yearly to more than 25% of the population). The evaluation conducted using the administrative data is the primary focus of the paper. The evaluation design is non-experimental, based on the Matching technique. In order to estimate the impact via Matching in a valid form, it is necessary to identify all of the variables determining participation in a program. In this particular case, there are characteristics of the families and of the districts3 in which these families live that determine the probability these families have to be invited to participate in the program. After carrying out an extensive analysis on the admission mechanism it was determined that nearly all of the relevant variables had been captured. Results from the Chile Solidario Panel are obtained using the Matching approach. In most cases it is possible to carry out differences-in-differences analyses with the Chile Solidario Panel. However, in some cases, especially in the take-up of social programs and psychosocial results, it is only possible to find results using cross-sectional estimations for 2006. These results are mostly positive. Also, some negative results for areas such as poverty and indigence in cross-sectional estimations have been found for 2006. However, the results show that when it is possible to contrast the similarities prior to the implementation of the program between the treatment group and the controls using administrative data, in a big proportion of the cases important differences can be found before the start of the program between both groups. This implies that estimations performed using only the Chile Solidario Panel will likely be seriously questionable. In order to overcome the problems of the Chile Solidario Panel described above, a new evaluation was carried out which merged all of the available administrative databases. A panel with seven rounds was generated, with around 1,000,000 records for each round. However, there are also shortcomings to this approach, such as the non-random attrition and the measurement error, but the evaluation overall is much more reliable. There is a substantial rate of participation in Chile Solidario in the control group, so different strategies to overcome this problem are discussed. A simple extension to the standard solution substitu-
2 3
CAS measures resource availability and it breaks down into Housing, Income, Employment and Education scores. The program has a decentralized management model, so it is executed at the comuna (Translated here as district) level.
132
tion bias is adopted. The results show a mixture of small positive and negative effects on some variables, while other important variables, such as those related to employment, have small but clear positive effects. The evaluation of Chile Solidario presented several challenges, but these challenges also provided valuable lessons to consider when implementing impact evaluations in developing countries. Those lessons include the need to create evaluations as soon as programs are designed, the need to extensively document all long-term decisions carried out, the importance of keeping the questions on the questionnaires over time, the usefulness of verifying IDs in the location where the survey was administered, the need to deeply understand the program when designing the instruments to be used, the opportunities that are given by the use of administrative data, etc. All of the problems encountered may appear in future impact evaluations; therefore, systematizing these problems may prove a useful guide to organizations wanting to evaluate their programs. In addition to these lessons, the information available to program executives makes the Impact Evaluation a valuable exercise. The rest of the paper is structured as follows: In section 2 the Chile Solidario Protection System is described. In section 3 the admission mechanism to the system is thoroughly explained, so as to justify the non-experimental techniques used in the evaluation. Section 4 assesses the validity of the identification strategy. Section 5 describes the data that are used. Section 6 provides a description of the methodology used in the evaluation. Section 7 presents the results using the Chile Solidario Panel database. Section 8 describes the Administrative Panel Database. Section 9 describes the criteria for choosing the control group from that database. Section 10 describes the results obtained from the Administrative Panel. Section 11 concludes and systematizes the lessons for future evaluations. 2. Chile Solidario Social Protection System During the 1990s, Chile went through a fast decline in poverty levels. The total poverty index decreased 15.4 points in six years. Extreme poverty in the same period was reduced by more than half. However, from 1996 to 2000 and to 2003 there was relative stagnation in the poverty figures. Chile Solidario was created as a response to the determination that the pace and volume of the extreme poverty figures were stagnated since 1996. Fortunately, from 2003 to 2006 there was a new important reduction in non-extreme poverty levels, going from 18.7% to 13.7% and of extreme poverty, from 4.7% to 3.2%. It is of major public interest to quantify in what way Chile Solidario was responsible for this important progress. The Social Protection System Chile Solidario (from now on CHS) was a pioneering initiative in Chile in two dimensions. It was the first governmental program to address poverty in a multidimensional way, and it was also the first large-scale public policy to undergo an impact evaluation. Besides, social programs and benefits were not focused, and there was a lack of mechanisms and safety nets that would help to avoid loss of income and to mitigate the effects of unforeseen circumstances suffered by the poorest families. The program is run by MIDEPLAN, the anti-poverty department of the Government of Chile. The families remain in the System for a 5-year period. During this time an intervention is carried out, which incorporates three components:
133
FIGURE 1
EVOLUTION OF POVERTy AND ExTREME POVERTy 1990-2006 45 40 35 30 25 20 15 10 5 0 1990 1990 1990 1990 1990 1990 1990
Total Poverty 13 9 7.6 25.6 23.8 20.1 17.5 16.1 14.6 5.6 38.6 32.9 27.6 23.2
21.7
20.2
18.7 13.7 14 4.7 10.5 3.2
5.8
5.6
1990
Extreme Poverty
Non Extreme Poverty
Source: MIDEPLAN.
i) Psychosocial Support and Protection Bonus Psychosocial Support consists of personalized mentoring, in 21 sessions over a two year period, given to enrolled families by a specially trained professional (Family Support). During this mentoring, mutual commitments are established. Moreover, the professional helps with the procedures for the respective bonuses and subsidies to which the beneficiaries are entitled, and also helps them to access a varied group of locally implemented social programs that the State has put at their disposal. Additionally, the law contemplates that the families enrolled in Chile Solidario will receive a monetary bonus, whose amount decreases from US$ 21 per month during the first year to approximately US$ 12 per month in the fifth year. ii) Guaranteed Access to Monetary Subsidies Monetary subsidies are guaranteed, as long as they remain in the system, for everyone fulfilling the requirements. There are 4 different money transfers people may apply to4, each with different requirements.
People may apply to SUF (around US$10 a month, given for each child, pregnant woman); SAP (Drinkable Water Subsidy); PASIS (around US$100 a month, given to elderly or sick people) and the Subsidy for School Retention.
134
iii) Preferential Access to Assistance and Social Promotion Programs The beneficiaries of Chile Solidario have priority of access to these social services. In order to ensure the preferential access, MIDEPLAN establishes assistance agreements with many public services for enrolled families. As an example, some benefits5 are specified: Health assistance free of charge in public health services; preferential access to Training Programs; preferential access to the Support Program for Small Businesses. Chile Solidario was created in order to reduce extreme poverty in Chile. The program has defined a set of intermediate objectives in seven different areas for each family. This evaluation focuses on two of the most difficult areas from which to obtain results: employment and income. Costs Associated with the Chile Solidario System In terms of budget, Chile Solidario has grown strongly, growing from around US$ 5 million in 2002 to more than US$ 140 million in 2007, using total resources equivalent to around 0.1% of that years GDP. 3. Admission Mechanism A complete understanding of the mechanism by which families enter the program is fundamental to determine feasible identification strategies and the assumptions that must be used in the evaluation. Chile Solidario was created as a Social Protection System targeting people living in conditions of extreme poverty. As a consequence, 225,0816 program slots were originally assigned for the period 2002-2005 (a number that was thought to be enough to cover the total number of indigent families). Later, more slots were opened in the system. As of November 2006, the last date this evaluation covers, the program included a total of 252,586 families7. Since the plan was to cover the population living in extreme poverty, and due to the fact that there was no clear way to distinguish between indigent and non-indigent families using the CAS score, it was decided to use the district and regional indigence percentages obtained through the CASEN survey (a
5
6 7
The broken down detail for 2005 corresponds to the information contained in the Chile Solidario Information book N 2 (Cuaderno de Informacin N 2 Chile Solidario) (MIDEPLAN). Only a few examples are named, but there are a wide variety of agreements between MIDEPLAN and other public sector institutions regarding the preferential access of the program beneficiaries Cobertura de Chile Solidario por Regin 2002-2005 Final. Secretara Ejecutiva de Chile Solidario, Excel File. Base Puente Personas, May 2007. Includes Families and One-Person Households constituted by Old Age People. It includes active families (families that completed the program) and interrupted families (those that did not complete the program.
135
national households survey) in order to determine a CAS cutoff such that a similar percentage of the district population was eligible for the program. This led to the adoption of different CAS cutoff scores by district. Families whose CAS score was less than the maximum of either their districts or their regions cutoff scores were eligible for the program. Since different cutoff scores were established in districts all over the country8, it is possible to use non-experimental strategies for the evaluation, such as matching, in which it is mandatory that the treatment and the control group are comparable. Both this impact evaluation of Chile Solidario and the previous ones have exploited the fact that families having identical CAS scores are or are not eligible for the program, depending on the district in which they live. Once the effective cutoff scores by district have been established, it is possible to know who is eligible for Chile Solidario. However, there were not necessarily program slots available for all eligible families. The number of slots available for each district was decided in the following manner: Based on the effective cutoff scores, the potential demand of Chile Solidario beneficiaries in CAS databases was estimated; this corresponded to 298,409 families 9. Since the budget was only available for 225,081 slots, in most cases a lower amount to the potential demand was assigned to each district. The way the total number of places by district was assigned, and their distribution through time, were the result of a discretional analysis process, taken case by case in each district, based on the real possibilities they would have to give psychosocial support to a certain amount of families. Some of the factors considered were the number of public employees counted on by the district administration10, the geographical characteristics of the district, etc. Once the cutoff scores and the distribution of the yearly slots were determined, the families with the smallest CAS score had priority. Given this information, we should see more people with higher CAS scores being admitted into the program over time. Figure 2 shows such a pattern. Each mark represents a family entering the program. The x-axis displays the entry date. The y-axis presents the CAS score for the family. There is a marked increase from the first to the later years. However, this does not happen in all cases, due to two reasons: 1) Each districts CAS database is continuously updated. 2) The law establishes that in those cases where the families in the initial list are too dispersed in location inside the district, it will be possible to distribute families between the Familiar Supports at geographical convenience, even though they were not families with smaller scores. In practice, it can be observed that the scores of people admitted to the program can even
8 9 10
Chile has 346 districts. Amount obtained in 2001 CAS Conglomerate of the addition through the communes of the families eligible for the program. The administration of the Comunas (districts) are called on this paper Municipalidades (district administrations).
136
show a uniform pattern through time (Figure 3). This phenomenon is relevant for the evaluation, as it explains why after many years of the system being in practice there is still common support (that in fact makes matching possible). It is worth mentioning that the districts strictly respect the prohibition on including beneficiaries whose score is over the cutoff score, as is shown by the fact that no more that 2% of beneficiaries appear in the database with CAS scores exceeding the effective cutoff of their district. It can be concluded that, for a family, the variables having influence on their program participation are their CAS score, the cutoff score obtained by their district, the amount of slots assigned for that district, and the rate of assigned slots/original slots, which is related to the management capacities of each district. Another variable that might be relevant is the geographic isolation of the family and the capacity of that family to have influence on the district administrations employees to be included in the program. Chile Solidario executives believe this last variable may not be relevant since there has never been a massive demand for inclusion in the program.
FIGURE 2
BENEFICIARy ADMISSION PATTERN, LOS ANGELES DISTRICT
Los Angeles
District Cutoff: 468.974 550 Regional Cutoff: 472.03
500 CAS Score
450
400
350
01 Jan 02 01 Jan 03 01 Jan 04 01 Jan 05 01 Jan 06 01 Jan 07
Date of Entry Source: Own elaboration using Puente Base, May 2007.
137
FIGURE 3
ADMISSION PATTERN FOR BENEFICIARIES, TEMUCO DISTRICT
Temuco
District Cutoff: 483.679 550 Regional Cutoff: 457.073
500 CAS Score
450
400
01 Jan 03 01 Jan 04 01 Jan 05 01 Jan 06 01 Jan 07
Date of Entry Source: Own elaboration using Puente Base, May 2007.
4. Validity of the Identification Strategy The identification strategy used to measure CHS impact is based on the existence of different families that have the same probability, conditioned on a group of variables, of entering the program, some of them are admitted and some others are not, due to exogenous reasons. There are three situations where different families with a similar CAS are not all included in the program: 1) the treatment group has a marginally lower CAS score; 2) the treatment group and controls have the same score but they live in different districts with different cutoff scores; and 3) both families have the same CAS score and live in the same district, but due to unknown reasons, one of them receives treatment and the other does not. This third situation presents a new issue as to whether the reasons families are admitted into the program are indeed external. This evaluation is going to be carried out by Matching (see section 6). The fundamental assumption of that methodology is unconfoundedness. Under this assumption, all factors relating treatment assignment with the potential outcome are observable11. As an example, lets suppose those in the treatment group come mainly from districts where the public services provided are better-than-average
11
Estimation of Average Treatment Effects Under Unconfoundedness. Imbens, G. en Imbens & Wooldridge (Whats new in econometrics, Lecture Notes 1, Summer 2007).
138
quality. On the other hand, lets suppose those in the control group come from all kind of districts with equal opportunity. In case of not being able to measure the quality of the public services, the mentioned effect would contaminate the programs estimated impact because what would be measured would be better quality of public services + program impact. Next, it is discussed whether that assumption is fulfilled for this evaluation. The factors having an influence over the probability of treatment have been classified intofour groups. Individual and district variables are considered, since both can affect the probability of treatment. If on top of that they have an influence over the potential outcome, any of these variables can be a source of bias if the program impact is not controlled. In the following table there are examples of variables in each of the categories that could have an influence over the likelihood of admission to the program: TABLE 1
FACTORS THAT COULD DETERMINE THE PARTICIPATION ON THE PROGRAM
Observable Individuals Districts CA score Cutoff score and slots available
Unobservable Motivation; Closeness to the district administration District administration management capabilities
Observable individual variables: The only individual-level characteristic explaining his/her participation in the program, according to the rules, is the CAS score. CAS score is related to the socioeconomic level, hence it is important to control for when evaluating the program. Observable communal variables: The only district characteristics explaining the probability that an individual living in that district takes part in the program are the district cutoff score and the number of program slots available in the district. Both variables can be related to the potential outcome. The number of places has been decided considering the quality of the public service. Hence, besides controlling for the number of slots it is necessary to capture, in some way, the quality of public service in the district that is unobservable on the data. Unobservable individual variables: When families are invited to participate in the program they can choose to accept or not. This can be a first cause of violation to the assumption of unconfoundedness (self-selection). However, the rate of rejection of enrolment in the program is around 1%, so even if that factor exists it is not very relevant. The invitations to the program within each district were based on the CAS score, with the exception of those cases where in geographical terms it was more efficient to group beneficiary families. For this reason, a familys geographical accessibility is an unobservable variable in the data that determines the probability of being treated. This could be a second violation of the assumption of unconfoundedness. We do not have data
139
to control for this potential bias, so it will only be possible to isolate it assuming that it did not change in time and estimating through differences-in-differences, as explained in the methodology section. The third potential violation of the assumption of unconfoundedness occurs when the individuals motivation can affect the probability of being invited to participate in the program. This could bias the evaluation outcomes. According to Chile Solidario workers, until the end of year 2003, there was scarce public awareness of the programs existence, so people could hardly go to the Municipalities and ask to be included. Unobservable communal variables: The only moment there was a subjective process affecting the chance to participate was when the number of slots per district was assigned (See section 3). As described in section 3 of this document, the potential demand had been set at 298,409 families, whereas for budgetary reasons 225,000 slots were given. The manner in which the total of these places per district was assigned and their distribution in time was the result of a case-by-case analysis of the districts and their real possibilities to provide psychosocial support to a high number of families. This could be a fourth violation of the unconfoundedness assumption. This last selection process is important in magnitude, given that 30% of the potential beneficiaries were not included. The problem that may arise from this procedure would be, for example, that the beneficiaries were grouped in districts with better public management capabilities. In this case, the beneficiaries would be in districts with characteristics that are different from the controls and this difference would not be noticeable. As has been noted, the only moment there were decisions based on unobservables that could have an influence at the same time in the assignment of places and in the quality of services rendered by the district was when the assignment of places per district was decided. One possibility of isolating that decision process is to control for the manner in which the decision-maker determines the capability of the district to provide quality support to families: taking the total of available slots for the program (from the CAS databases) and calculating the proportion of those slots effectively given to the district, we can quantify the subjective aspects the program workers considered. Given that this measure was the only way the program workers had an influence on the probability of participation, by controlling for this variable, the identification strategy becomes valid. 5. Data Puente Databases: The Planning Ministry (MIDEPLAN) is in charge of managing the Chile Solidario System. Puente databases are used as a tool to handle program management, which includes all the beneficiaries of the program, their identification numbers and their CAS scores, among other variables. The Chile Solidario Panel Survey (CHSP) is the survey on which this evaluation is based, and was specifically designed for the impact evaluation. The panel is a longitudinal survey, composed of beneficiaries and non-beneficiaries of the system. The information gathered contains questions about demographic characteristics, education, health, employment, income, housing, family information and a psychosocial module. In general, the questionnaire is very detailed, its only flaw being that it is not specific enough for Chile Solidario, as it does not have many
140
questions on the use of benefits specific to the program. The rounds used in this evaluation are for the years 2003, 2004 and 200612. Despite being a longitudinal survey, the follow-up unit and protocols to minimize attrition were only determined in 2006. The main characteristics of the sample are detailed below. The Puente program started in mid-year 2002. Only a year after, planning for the program evaluation was initiated, and it was decided to use the CASEN 2003 survey as the first round of CHSP. In order to identify the treatment group for the Chile Solidario Panel the question from the CASEN survey Does your family participate in the Chile Solidario program? was used. After an extensive process intended to correct the Identification Numbers13, it became clear that that variable had an important level of measurement error when compared to the administrative data from Chile Solidario, with an inclusion error of about 6% and an exclusion error of about 15%. Another problem was that the ID numbers were not checked at the time the data were collected, which led to a high probability that those numbers were not correct. Additionally, a complementary sample of homes that benefitted from Chile Solidario was added using data obtained from Puente Bases. For the years 2004 and 2006 beneficiaries belonging to the programs most recent cohorts were added, who had been surveyed in CASEN 2003. The control group was selected by matching, but there were no appropriate administrative variables available at that time and some regional restrictions were imposed on the matching. Due to budgetary reasons all those observations having missings in the variables used in the propensity score estimation were excluded as well as those of beneficiaries having no control families with a similar probability of program participation (those being outside the common support). The sample was stratified in 2004, according to the demographic distribution of the beneficiaries included in Puente Base in July 2004. In the end, for the 2006 round, enrolled families that did not have a CAS record were excluded. Due to the successive stratifications and to the fact that the new cohorts were not added in a way that is proportional to their magnitude, it is not clear whether this panel is representative of the actual CHS population. Of note, a series of changes in the questions included in the PCHS questionnaire were made over the three rounds. It was necessary for evaluation purposes to carry out an extensive data standardization process in order to make the three rounds compatible. The scale of the changes was such that the questionnaire in the psychosocial section is not comparable between the years 2004 and 2006. TABLE 2
TOTAL NUMBER OF INDIVIDUALS, CHILE SOLIDARIO PANEL
2003 Beneficiaries Non Beneficiaries 13,705 18,538
2004 13,627 18,326
2006 14,913 18,225
Source: Chile Solidario Panel 2003-2006.
12 13
Currently the 2007 round of the panel is available. In Chile, the identification number is known as RUN or RUT.
141
CAS record: The CAS record was the instrument used up to 2006 to select the beneficiaries of the targeted social programs. It contains relevant information including ID number, zone, district, current months income, employment, housing, education, etc. The CAS records expired after two years. The CAS record with the information of the families is stored in the CAS database. These bases contain information from all records that were still valid up to December of the calendar year of the CAS database. As an example, the CAS database 2001 contains information for all families that applied for a CAS record between January 1999 and December 2001. Each CAS database covers about 6 million individuals, including approximately 75% of families living in extreme poverty conditions14. The main problem of Chile Solidario Panel is the lack of baseline data. As a way of partially addressing this problem, in this evaluation the databases from the years 2000 to 2005 have been put together, in order to have pre-treatment data available on the program. 6. Methodology Used in the Evaluation When Chile Solidario was designed, the creation of an experimental control group15 was not considered. The most proper way, statistically, to evaluate a programs impact is to perform an evaluation with a control group of that kind. When not being able to use experimental methods, some kind of non-experimental method to estimate the impact is required. Considering that the characteristics determining eligibility in Chile Solidario are not random, the Propensity Score Matching is used to build up a counterfactual scenario. Matching methodology has a long history in non-experimental studies of impact evaluation. (Rosenbaum & Rubin (1983); LaLonde (1986); Dehejia & Wahba (1998); Abadie and Imbens (2006)) and nowadays it is still considered a valid way of estimation that is being continuously improved Blundell & Costa Dias (2002); Imbens (2004); Heckman, Ichimura & Todd (1997)). Matching is applied to identify an adequate control group whenever this group cannot be built up from randomization of the intervention. Conceptually, the propensity score matching method consists of looking for a control for each treatment group member that is identical in terms of his/her probability of participation in the program. Once the sample of non-beneficiaries has been chosen, and having fulfilled the fundamental assumptions of the method, the conditions for a natural experiment are re-established. The use of matching to determine a program effect requires two assumptions:
14 15
Estimation by Osvaldo Larraaga based on CASEN 2003. A group of experimental controls is one that has been randomly selected between the potential participants in a system, so that, statistically, it has the same characteristics as the group under treatment in the system.
142
Unconfoundedness assumption: The assignment to treatment (D) is orthogonal to the potential results conditioned on a group of noticeable x variables. The conditioning is performed only on observable variables, which have to be collected before the program may alter them. In other words, if the unconfoundedness assumption is fulfilled, the results of the non-treated families behave similarly to those of treated families for the cases when they have not participated in the program. This assumption is not possible to prove, since the potential result (Yi) is not known before the intervention. In the section 4 (Validity of the Identification Strategy) an extensive discussion was presented about the scenarios where it is possible to consider this assumption as true. Given such discussion, the variables to which we had access and that would influence admission to the program would be: the familys CAS score16, the cutoff score of the district, the number of places available in the district at the time of evaluation and the proxy measurement regarding management capacity. This basic theoretical specification was empirically tested, as the difference in pre-treatment17 variables between the treatment group and the control group was measured. The unconfoundedness assumption cannot be easily assessed using the Chile Solidario Panel, so it was necessary to use the pre-treatment variables available on the CAS records to have a more solid comparison point between treated and control families. The matching specification using the theoretical model attained a medium level of similarity between the treatment group and the control group. A pre-treatment imbalance on the Number of persons in the family variable was confirmed after successive tests, among other variables18. Tests with more and with fewer variables were carried out, modifying the number of neighbours and using kernel and calliper matching. No specification had a better match than the basic model. Considering the importance of the number of people in a family, it was decided this would be included in the p-score identification. In total, 13 models were tested, but the original theoretical specification was kept, with the addition of the number of people in each family. These first tests are just a beginning for checking pre-treatment consistency. Effort is put on the Results Section to make it possible to determine whether this assumption may be feasible, using more pre-treatment variables and more outcomes. If after matching there are still significant differences in important pre-treatment variables, the method may not be valid fordetermining program impacts.
16
17 18
The CAS score used was the latest score before the entry to the program for the treatment group. For each control family we chose their CAS record that was closest to the average survey date of the treatment group. This is equivalent to testing whether the control group and the treatment group are statistically equal before the program. Head of the family education level and the zone also appeared with pre-treatment differences.
143
Common Support Assumption: Let S be the set defined for all beneficiary families that have a propensity score close enough19 to the score of at least one control family. There is common support if S is non-void. This second assumption guarantees that the required comparison group exists, i.e. at least a subset of the treated families has a match in the group of non-treated families, where each one of the control group members is a potential candidate for participating in the program. The range of the Propensity Score for which there are treatment and control families is called Common Support and the matching-estimated parameters will only be valid for the on-common support group, not for all the sample. The principal reason why there is an overlap on the distributions of the propensity scores of both groups (treatment and control) is that the cutoff scores for Chile Solidario were set by district, using criteria that made it possible for two identical families to have different eligibility status. Taking into account all of the previous precautions the propensity score for the treatment and control group was estimated. The result is presented here in Figure 4: Methodological Decisions The identification strategy used to measure the impact of CHS is based on the fact that there are many families that had the same probability (conditioned to a group of variables) of being enrolled in the program; some of them are enrolled and some are not due to external reasons. Considering the method by which people were enrolled in the program, in order to build up the counterfactual group and to determine the impact on homes participating in the System, the Propensity Score Matching20 methodology is used. The fundamental assumption of matching is that all the factors relating the treatment assignment to the potential result are observed as in Imbens and Wooldridge (2007). Nearest Neighbour Matching with 4 neighbours is being used, following the suggestions given by Imbens (2004). The regression discontinuity approach was also considered, but since the number of observations from Chile Solidario Panel is limited, the estimations were extremely unstable.
19 20
The definition of what is close enough often is a decision that comes up from the empirical work.
Despite the fact that the non-beneficiary sample had been originally obtained by matching, those data would not necessarily match with the new control group, due to many reasons, including: the original matching was regionally severely restricted; the variables used in the original p-score matching were CASEN only, in this case however RUTs (IDs) were available in order to use the CAS score; in the original matching no communal variables were used; and finally, in the original matching self report was used to define the condition of the treatment group instead of the administrative records of the program.
144
FIGURE 4
DISTRIBUTION OF PROPENSITy SCORE FOR TREATMENT AND CONTROL GROUP 3
0 0 .2 .4 k density Treated Source: Own elaboration using Chile Solidario Panel. x .6 .8 k density Control 1
a. Parameter to Be Estimated Inside the Chile Solidario Panel it is possible to estimate program results using the differences-in-differences or cross-sectional approaches. Given the lack of a baseline, both approaches present problems. In order to select the appropriate analysis approach, we will assume that it is possible to divide the program impact into two parts, the impact before the baseline is taken (beginning with the family entering the program to November 2003, when the first survey was conducted) and the impact from November 2003 until November 2006 (date of the last panel round used). If only one cross-section is measured, this will show the final difference between the two groups, relying on matching to assume that both would be identical before the treatment. In the results section it will be shown that this is not necessarily true. If differences-in-differences is used we can only obtain the progress from one point to another in the treatment. This can be interesting, but it strongly depends on the sequence in which the program impacts the beneficiaries. For example, if people greatly improve for a given result at the beginning of the program, but this improvement is partly temporal, the result the differences-in-differences estimation may deliver can be negative, even though the program has a positive effect overall. Whenever possible, both approaches will be used on this paper.
145
b. Grouping the participants and non-participants according to survey date and date of admission to the program When estimating the program impact it is necessary to clearly determine the definition used for participating in the program that is being considered, so that the impacts found can be correctly interpreted, thus enabling policy decisionmaking. The whole Chile Solidario program contemplates a two-year Puente program, with active psychosocial support and then, a three-year follow up, where people still receives the benefits of preferential access to public programs and the monetary bonus, but the psychosocial support is taken away. The treatment group that, at the moment of the poll during the year 2006, had been for about four years in the program (cohorts 1 to 10 of the program) will be the group considered in the analysis. Those people surveyed between March 2002 and March 2003 were included in the analysis, since the Chile Solidario Panel rounds are taken in November and a minimum amount of time must take place before the families start to receive the program benefits. Given this definition and the way the CAS databases are grouped, enrolled families included in this evaluation spent an average of 38 months in the program. In order to make the treatment and control groups possible to compare, some proximity was sought between the survey date on the CAS record for the treatment group, with which they entered to the program, and the CAS survey date of the controls. In short, for each control family, the CAS record that is nearest to the average date of the treatment group survey is taken. c. Definition of Treated The Puente databases were used for the administrative identification of the treatment group. All beneficiaries enrolled in Chile Solidario are identified by their RUT (ID). These people and their families were excluded as potential controls in this evaluation. However, in order to be included in the treatment group the administrative definition of the program was applied: Treated families were those in which the recipient of the Chile Solidario Bonus lives. 7. Results using the Chile Solidario Panel In section 7.1 the results in the variables of Chile Solidario Panel are shown, which are also available in CAS records21. The variables on the CAS record are the best available proxy of a baseline. The variables being used here are the amount of people in each family, the labour income and employment. For each of them an average group of results is presented: years 2003 and 2006 of Chile Solidario Panel, which this evaluation is based on; the average in both groups in CAS records 2001 and 2002, in order to show if there are any pre-treatment differences between the groups that are not attributable to the program; the result of CAS groups 2005, because it is comparable to our pre-treatment variables. Then, the CAS 2003 result is presented in order to examine the comparability between PCHS and CAS record variables. In the end, the differences for CAS records are presented between the years 2005 and 2002 and for Chile Solidario
21
They are not totally comparable, because the questions are not identical.
146
Panel between the years 2006 and 2003. In some cases, when it is possible to sum up the results of a variable without showing the tables, these are omitted to make the presentation shorter. The review of this first group of variables will set the basis to evaluate the credibility of this evaluation: if important pre-treatment differences are found, the results in the variables that are not on CAS record will be, by extension, hardly credible. In section 7.2 the results in the variables that are not available on CAS records are shown. Among these variables we can find the take-up of subsidies and social programs, and also variables including income by subsidies, such as Extreme poverty and poverty. In these cases only the results for the rounds 2006 and 2003 of Chile Solidario Panel and the differences between both of them are presented In section 7.3 the results in perception variables are shown, which are only available for the year 2006 in Chile Solidario Panel. 7.1. Variables available on the CAS record Number of people in the family: This is one of the most important variables for the evaluation, since a lot of the calculated outputs are formulated based on the per capita or family total. As can be seen in the selection of variables for the propensity score (section 6), Number of People in the Family is included in the propensity score. However, as is shown in Table 3, this is not enough to ensure there is no pre-treatment impact. Significant impacts can be observed both in the valid22 number of people in the family as in almost every year. However, the differences-in-differences results are not significant, suggesting that the differences were not the result of the program. Due to this persistent difference between the treatment group and the control group, both per capita variables and per family variables may be biased. TABLE 3
# of people in the family CAS 2001 CAS 2002 CAS 2003 CAS 2005 PCHS 2003 PCHS 2006 Dif-Dif PCHS 2006-2003 Dif-DifCAS2005-2002 Impact 0.162 0.297 0.341 0.480 0.590 0.119 0.085 0.110 t 1.531 2.735 3.454 5.068 6.106 1.169 0.900 1.625 Mean T 4.324 4.303 4.318 4.665 4.600 3.511 0.167 0.065 Mean C 4.163 4.007 3.977 4.185 4.009 3.392 0.252 0.176 N 1368 1320 1505 1698 1698 1047 1320 1698
Head of the family employment: Significant pre-treatment differences are observed favoring the treatment group. The differences-in-differences estimation
22
The valid survey refers to the number of people the family had in their latest CAS record before entering the program, for the case of beneficiaries, and to the people the family had in their CAS record, which is closest to the average date the survey was taken for the treated, in case of controls. This is the number of people used in the Matching.
147
in the CAS record is significant against the treatment group. While employment for the control group rose by 5% in the period, employment in the treatment group slightly decreased. Income of the Head of the family: It is possible to estimate the impact on income with greater confidence, since there are no pre-treatment impacts in any of the years. Surprisingly, considering the previous result for employment, a positive and significant effect can be found in the head of the familys income for the difference in incomes in Chile Solidario Panel. TABLE 4
Job Income Head of HhId CAS 2001 CAS 2002 CAS 2003 CAS 2005 PCHS 2003 PCHS 2006 Dif-Dif CAS 2005-2002 Dif-Dif PCHS 2006-2003 Impact $ 574 $ 2,313 $ 7,041 $ 133 $ 8,220 $ 2,331 $ 2,180 $ 10,551 t 0.097 0.473 1.420 0.031 1.932 0.581 0.412 2.390 Mean T $ 68,786 $ 66,156 $ 69,787 $ 69,205 $ 50,501 $ 58,789 $ 3,049 $ 8,288 Mean C $ 69,360 $ 68,469 $ 62,746 $ 69,338 $ 58,721 $ 56,458 $ 869 $ 2,263 N 1239 1132 1365 1132 1698 1698 1132 1698
Number of people employed in the family: There are no pre-treatment impacts, so the number of people employed by family can be estimated with relative confidence. Both panels coincide in 2003, which is the only year where significant differences can be found between both groups; however, these differences are not maintained in subsequent years. Percentage of employed people: There are pre-treatment impacts in all measurements during the first years, so it is not possible to obtain accurate conclusions. Family Income: While possible to estimate this impact, no significant impacts were found. The most notorious point is the difference between the impact in the years 2003 and 2006 favouring the treated group. However, this difference of $12,325 does not translate into a significant impact. TABLE 5
Labor Income CAS 2001 CAS 2002 CAS 2003 CAS 2005 PCHS 2003 PCHS 2006 Dif-Dif CAS 2005-2002 Dif-Dif PCHS 2006-2003 Impact $ 3,773 $ 12,786 $ 3,587 $ 10,768 $ 11,514 $ 811 $ 2,018 $ 12,325 t 0.535 1.878 0.637 1.626 1.860 0.117 0.256 1.745 Mean T $ 94,242 $ 90,850 $ 92,685 $ 98,194 $ 97,695 $ 122,378 $ 7,344 $ 24,683 Mean C $ 98,015 $ 103,637 $ 89,097 $ 108,962 $ 109,209 $ 121,568 $ 5,325 $ 12,359 N 1285 1143 1413 1143 1698 1698 1143 1698
148
Number of working women: There are no pre-treatment impacts and there is a certain effect in year 2003 on the treatment group that disappears with time. This effect leads to a negative differences-in-differences estimate. Percentage of working women: There are pre-treatment effects for this impact; however, the differences are considerable in year 2003 in the Chile Solidario Panel, while after this period the effects disappear. CAS Score: There are significant pre-treatment differences. These differences appear to be widening with time, but conclusions at this point cannot be clearly drawn. 7.2. Variables only available in Chile Solidario Panel Given the results presented in the previous part, the results presented here must be handled with a lot of caution, since it is likely that the initial levels between treatment group and the controls are not identical. This is problematic, because most important variables in the program, such as Extreme Poverty and Poverty are included in this category. In the case of variables related to the take-up of subsidies and social programs, the calculations are based on the percentage of eligible individuals receiving the benefit. Indigence and Poverty: The same logic is given for both variables. For both variables, strong negative impacts can be found within the cross-section. However, it is highly probable that those impacts are due to pre-treatment differences between the treatment group and the controls. TABLE 6
% in Extreme Poverty PCHS 2003 PCHS 2006 Dif-Dif PCHS 2006-2003 % in Poverty PCHS 2003 PCHS 2006 Dif-Dif PCHS 2006-2003 0.156 0.185 0.029 6.600 7.630 1.001 0.730 0.638 0.092 0.574 0.453 0.121 1698 1698 1698 Impact 0.115 0.093 0.022 t 5.435 4.610 0.852 Mean T 0.341 0.290 0.051 Mean C 0.226 0.196 0.029 N 1698 1698 1698
Income through Subsidy: The differences-in-differences result is negative, but it is composed of two positive results for cross section of the treatment group. This is likely due to the fact that people in the treatment group received more subsidies than did the controls during the psychosocial support stage, but that difference disappears once they are out of that stage.
149
TABLE 7
Income through Subsidy PCHS 2003 PCHS 2006 Dif-Dif PCHS 2006-2003 Impact $ 6,005 $ 2,948 $ 3,057 t 5.609 2.243 2.758 Mean T $ 27,868 $ 26,321 $ 1,547 Mean C $ 21,863 $ 23,373 $ 1,510 N 1698 1698 1698
Total income and net income of the Chile Solidario bonus: This comparison table provides a good example as to how a temporary benefit given by the program can provide wrong signs when baselines are not available. The first table shows $ 8,621 as the impact of the program. However, this impact is reduced with the retirement of the Chile Solidario cash transfer, which decreases its amount with time. Despite the fact that in this case the impacts in general do not reach a significant level, it is worth mentioning, so as to be aware of what can happen when benefits are given to the beneficiaries and then are taken away. TABLE 8
Total Income PCHS 2003 PCHS 2006 Dif-Dif PCHS 2006-2003 Net Income (of CHS bonus) PCHS 2003 PCHS 2006 Dif-Dif PCHS 2006-2003 $ 18,276 2.711 $ 138,071 $ 5,587 0.780 $ 164,826 $ 12,688 1.629 $ 26,755 $ 156,346 $ 170,413 $ 14,067 1698 1698 1698 Impact t Mean T Mean C $ 156,702 $ 170,531 $ 13,829 N 1698 1698 1698
$ 11,052 1.639 $ 145,651 $ 2,431 0.339 $ 168,100 $ 8,621 1.107 $ 22,449
Training: We can see an important change in tendency, where the treatment group started with more training and finished with less. This is a clear example of the need to have information on previous years, since it is highly probable that the treatment group received more training in the years of the Puente intensive phase, and thereafter received less training, which can be incorrectly seen as a negative dif-dif. TABLE 9
% in Job Training PCHS 2003 PCHS 2006 Dif-Dif PCHS 2006-2003 Impact 0.044 0.025 0.069 t 4.926 2.262 5.088 Mean T 0.094 0.062 0.032 Mean C 0.050 0.087 0.037 N 1698 1698 1698
150
Preschool Coverage, School Coverage, and Adult Preventive Health Control: For all these variables, no significant results were found. Registration in the Health System, Medical Practice, Gynaecological Examination, Adult Preventive Control and Basic Coverage: These variables present some impacts using the cross-sectional analysis, but they do not present impacts in the differences-in-differences analysis. Since we do not have pre-treatment variables it is not clear whether it is possible to confer causality to these results. Dental Attention: There are positive impacts of the program in a very clear way, whether analyzed using the differences-in-differences or cross-sectional approach. 7.3. Variables that are only present in the 2006 round Take-up of social programs: These variables are only available for the 2006 round of the Panel. There is a big impact in OMIL registration and in the participation in the employment programs. Despite of not counting on the groups initial statistics, these are powerful signals on the program being effective in contacting the families with the public programs. TABLE 10
% usage of public ss (2006) Health Benefits OMIL registration Participation in the employment programs Impact 0.016 0.135 0.015 t 0.394 10.979 2.677 Mean T Mean C 0.268 0.217 0.035 0.253 0.082 0.020 N 315 1698 1698
Psychosocial Welfare: One of the most important components of Chile Solidario is the psychosocial support. For this reason, this is one of the aspects where the impact of the program should be most strongly noticed. Unfortunately, there is no way to clearly determine the validity of these results, since we do not count on pre-treatment statistics for these variables. In addition, they are only available for the year 2006. The variables shown are the addition of the positive answers at scales of 4 or 5 questions each. The scales have good metric properties, understood as Cronbach alpha coefficients of 0.7 or superior. Manzi (2007) interprets the psychosocial results for the Chile Solidario Panel. The results shown on the table below are very positive, with the exceptions of the Social Network variables. Especially important are the results in Vision of Future and Expectations of Change, which have been consistent in all estimations made in the Chile Solidario Panel. These show a much more positive attitude when facing future challenges from the treatment group than from the controls. It can be appreciated that controls are less active when facing their problems than the treated.
151
TABLE 11
Psychosocial Variables (2006) Vision of Future Expectations of Change Family Network Friends Network Self-Control (internal) Self-Control (external) Efficacy on the Job Efficacy on the childs school Mental Health Self-image Impact 0.116 0.138 0.101 0.111 0.136 0.057 0.083 0.054 0.006 0.121 t 4.333 5.540 2.851 3.154 4.587 1.265 3.258 3.063 0.275 3.540 Mean T Mean C 2.361 1.992 2.226 1.683 3.601 2.920 2.563 2.810 3.075 4.265 2.245 1.854 2.327 1.794 3.465 2.863 2.480 2.756 3.070 4.144 N 1729 1728 1730 1729 1723 1723 1723 1700 1730 1728
8. A Possible Solution to the Data Issues: The Administrative Panel Database As a potential solution for all the issues of the evaluation based on the Chile Solidario Panel, an evaluation based con administrative record was implemented. An effort was made to gather different sources of administrative information in order to generate a Panel database made of Administrative records to evaluate the impact of Chile Solidario. Through an individuals Identification Number, available in each database, different records for the same individual were gathered. The CAS database, the Ficha de Proteccin Social (FPS) database and the Puente Database are the most important pieces of the Administrative Panel. Each familys CAS record information is yearly stored in databases. These databases count on the information of all the records completed by a family and that are still valid by December of the CAS database year in question. As an example, CAS database 2001 contains information of all families completing a CAS record between December 1999 and December 2001. The Social Protection Record (FPS) replaced the CAS record in 2006. It is a different instrument, aiming to depict the potential families have to generate incomes. From November 2006 until April 2007 a massive poll was carried out at the national level that covered 1,375,041 homes (62.7% of the total polled homes) by May 1, 2007. (See Figure 5). Thus, information from CAS records between the years 2000 and 2005 were put together and with them, the information from the Social Protection record and information from the Chile Solidario administrative records (Puente Bases). Some concerns about this database are assessed below. Unit of Analysis: The CAS record considered the family as a reference unit, describing a family as the group of people living together and acknowledging themselves as a family group whether they have a kinship link between them or not, who have the intention of remaining together, and have some kind of monetary income. The definition of family for the FPS is similar to that of CAS, only that it specifically states that the group must share a common budget to be considered a family.
152
FIGURE 5
DATE OF FPS RECORDS
Percent
0
01nov2006 01feb2007 01may2007 01aug2007 01nov2007 01feb2008
Date Source: Own elaboration using FPS Conglomerate.
Questions Comparability: FPS data are not necessarily equivalent to those of the CAS Record. In particular, some of the questions are different. An example of this is that the CAS record inquired about incomes in only one question, whereas FPS did so in many, trying to identify all possible sources of income. The below table presents comparisons between the value of a group of variables in CAS record 2005 and FPS 2006, so as to show the magnitude of the comparability problems. Table 13 presents the number of observations (after having left only one observation per family) available on the panel used for the analysis. They are separated in Puente and No Puente depending on whether the family took part in Chile Solidario at some point or not. Quality of the data and measurement error: As has been mentioned before, the data available is fundamentally used to target subsidies and access to public programs. Due to this reason, it could be expected that the data suffer from strong underreporting, for example, of incomes. On the other hand, the amount of handled data is so high that this type of systematic failure in the dependent variables should be less worrisome. The main variables used in the study (CAS and Program participation) are obtained from administrative sources, so they do not suffer from measurement errors. In general, the main estimations made for this paper are double differences, so any measurement error problem affecting the treatment group should affect the controls in a similar way, thus voiding each other. In the same way, if one of the groups has a different type of measu-
153
TABLE 12
COMPARABILITy BETWEEN CAS 2005 AND FPS
Comparability of Income Between CAS and FPS (2006 pesos) Family Total CAS 2005 Family Total FPS Family Head CAS 2005 Family Head FPS # workers in the Hhld CAS 2005 #workers in the Hhld FPS % of Hhld Head with job CAS 2005 % of Hhld Head with job FPS $ 120,124 $ 157,239 $ 86,673 $ 108,219 0.94 1.13 0.62 0.60
TABLE 13
OBSERVATIONS IN EACH yEAR FOR CAS AND FPS
year 2000 2001 2002 2003 2004 2005 FPS
In Puente 78,389 97,271 136,039 110,180 104,515 107,672 102,722
Not in Puente 529,643 717,159 1,044,694 727,860 603,051 635,079 628,600
rement error from the other, but the error is constant in time, that error will be eliminated in the differences analysis. Attrition:On this panel, attrition can arise for reasons that are different from those arising in a normal panel poll (they did not want to participate in the CAS record completion process; they do not have a current valid ID, etc). If a family appears in a CAS database and that family is a Chile Solidario program beneficiary, it is highly probable that this fact has an influence on the likelihood that they will appear in the next CAS database, contact with the council administration, minimum condition of having an ID, etc). This phenomenon must be studied, since if this is the case, it could bias the impact estimations of the program. Tests for attrition randomness have been carried out and it seem like there is a correlation between attrition and some observable characteristics, including some related to the participation in Chile Solidario23. It is not possible
23
A first attrition randomness test is to measure whether some of the characteristics of the families (particularly participation in the program) determine the probability that the observation is being lost. In order to avoid the mechanical fulfilment of the relation between the program and the attrition (whoever entered the program in 2004 would have necessarily had to have a valid CAS record for that year), attrition is studied in years 2004 or 2005 explained by variables from the year 2002. The binomial variable indicating whether the family entered Chile Solidario during the first year of the program is significant and negative.
154
to quantify the magnitude of this effect without specific work on the subject, which has not been mentioned in this study. As a consequence of the previous situation, it is necessary to look for a way to control the attrition effects in this panel, which can become very important. Thus, it is necessary to choose a follow up unit that simplifies as much as possible the control of this phenomenon. The analysis units to be followed up in time are the families to which those head of the families reported in CAS 2002 (pre-treatment situation) belong. In this case, during the successive measurements, the anchor taken will be the head of the family and/or his or her spouse. The family or home can be the same over time, or it can be the case that its members have changed. Only the anchor cannot change. In particular, in bi-parental families the anchor will be the woman and in mono-parental families it will be the man or the woman that is the head of the family. This is because in most cases those appointed to receive the bonus are the people within the family, and given the case that the family composition changes, Chile Solidario payments remain with the owner of the bonus. As a consequence, choosing this unit would provide much more stability for the person to be followed in respect to his or her participation in the program. Additionally, this analysis unit is the one that most agrees with the administrative definitions of Chile Solidario. 9. Treated and Control Groups This evaluation covers those families that entered the program during the first 18 months, separating them into corresponding cohorts of three months each. The following graphic presents the families admission pace into the program, as well as the groups that are going to be evaluated (delimited by vertical lines). These groups were selected because they had spent enough time in the program by 2005. The division into six cohorts was the result of an interest in being able to isolate the effects of different levels of treatment. The controls are chosen, as in the Chile Solidario Panel, in such a way that their mean survey date is similar that of the treatment group. In the below table, we show the most important characteristics of selected groups. The second and the third columns are the months where the cohort entered the program. The fourth and fifth columns are the average date of the survey (CAS record) for the treatment and control groups. Because it was one of the selection criteria for the comparison group, the dates are very similar. The following two columns, which are closely related, are the total time of observation (the time between the first CAS record to the CAS record 2005, which is always the last observation used). The total time of observation is also very similar between the treatment and the control groups, something that wasnt granted, given the nature of the database. The final column is the total time of treatment for the treatment cohort, which is between 2.4 and 1.4 years. Another, more important decision that had to be made for the evaluation was whether the families that were not part of the cohort being evaluated, but
155
FIGURE 6
FAMILIES ADMISSION PACE TO CHILE SOLIDARIO 8000
6000 Frequency
4000
2000
0 15500 15000 16500 date_entry Source: Own elaboration based on Puente Base. 17000 17500
TABLE 14
From Cohort 1 Cohort 2 Cohort 3 Cohort 4 Cohort 5 Cohort 6 jan-02 apr-02 jul-02 oct-02 jan-02 apr-03 To mar-02 jun-02 sep-02 dec-02 mar-03 Jun-03 Date treated apr-01 jul-01 nov-01 jan-02 apr-02 jun-02 Date controls aug-01 sep-01 nov-01 dec-01 feb-02 may-02 Time ohs T 3.4 3.1 2.8 2.7 2.5 2.3 Time ohs C 3.1 3.0 2.9 2.8 2.6 2.4 Time treatment 2.4 2.2 2.0 1.8 1.6 1.4
who had been or would be participants of Chile Solidario, should be part of the control group. The problem is important given that, even though Chile Solidario participants represent a small fraction of the total panel sample (between 13% and 16%, see Table 13), they were closer in characteristics to the treatment group, thus having a higher probability of being chosen as part of the control group. Three alternatives were considered when estimating for each cohort: 1. To leave all other past and future Chile Solidario participants into the sample, allowing them to be considered part of the control group. This method implied
156
an important underestimation of the impacts of the program, as there were an important proportion of families on the control group (after matching) that participated in the program during the evaluation period. The table below shows, for each cohort, the percentage of the controls (after having been chosen by matching and weighted for their participation using the nearest neighbor estimation) that participated in Chile Solidario at any time during the period of this evaluation. It also shows the mean time of treatment for the treatment group on each cohort (naturally, decreasing with the cohort); the mean time of treatment for all controls; and the time of treatment for the controls that were part of Chile Solidario. It can be seen that, on average, a control group family had around a third of the mean time of intervention of a treated family. The Chile Solidario families on the control group, however, have almost the same time of intervention as the treated families. TABLE 15
% CHS control Cohort 1 Cohort 2 Cohort 3 Cohort 4 Cohort 5 Cohort 6 39% 37% 38% 39% 37% 37% Time Treat T 878 816 738 653 577 494 Time Treat Cont 226 207 224 226 222 220 Time Treat CHS Cont 587 559 586 586 597 592
2. To take all other past and future Chile Solidario participants out of the sample, thus denying the possibility for them to be included in the control group. This alternative seemed the most obvious. However, if this were done, we would have strongly biased the control group. The matching occurs before the treatment. However, for the next cohorts, their entrance into the program occurs after the matching, and is likely not to happen randomly. Chile Solidario is targeted to the poorest people. It is probable that the worstperforming controls after the matching will end up entering the program, so we would have underestimated the impact of the program by introducing ex-post selection bias on the control group. There is no way to correct this problem, after people have been taken out of the sample. 3. The standard correction for substitution and dropout bias when the treatment status is binary is widely known, as explained in Heckman, LaLonde and Smith (1999) or in Heckman et al. (2000). It suggests carrying out the usual estimation of the parameter, as the average of all the treatment group (including individuals dropping out) less the average of all the control group (including individuals taking substitute training), but then scaling up the result for the inverse of the difference in probability of participation. This is the same correction that is generally used in instrumental variables or in fuzzy regression discontinuity.
157
However, in Chile Solidario the treatment is not binary; in any given date there are very different periods of exposure to the program across the beneficiaries. The correction we propose (and the estimation method we finally chose) is a very simple extension to that technique: YT Yc TT Tc T T
Where is the impact of the program, YT Yc is the usual estimator for matching, once the control and treatment groups have been selected. TT is the average treatment time in the treated group and Tc is the average treatment time in the control group. In practice, the estimator rescales the estimated impact, accounting for the fact that the difference between the means of the groups is due to just a subgroup, that are the families whose matches are not participating on the program. This way, we dont have to introduce selection bias and we are able to estimate even while having different individuals in the control group both participating and not participating on the program24. The necessary assumptions for this estimator, besides the standard assumptions for matching, are that the potential outcomes of the group taking substitute training are equal to the outcomes for the treatment group. The condition is trivially accomplished here, as the program they are taking is the same. The decisions taken in this section are implemented in the results section, where the estimated impacts of the program are presented. 10. Results Using the Administrative Panel Database In this section, the results are presented in two ways, graphically and grouped in tables. The same cohorts explained before are used as treatment groups. On one hand, they are all represented in graphs by series, all referring to the variable that gives name to the graph. The x-axis has the different measurements across the time, from the CAS record of 2000 to the FPS 2006. All the graphs show on the y-axis the impacts for each year that is, the differences between treatment and control families- from the pretreatment years, where it is interesting to check the quality of the matching, to the later years, where the impacts of the program should be found. The cohorts are named by the familys month of entry into the program.
24
The analytic distribution in not known for this estimator, and it is not clear if bootstrap is a valid method when using matching, so we rely on the same standard errors as before, but amplifying the t-test and the impact for the factor mentioned.
158
On the other hand, the exact numeric results are shown in tables, in particular, two results are shown: the pre-treatment differences-in-differences and the post-treatment differences-in-differences. The first is presented as a check that the matching is good enough on that variable, the latter, to estimate the effects of the program. To ensure objectivity, the following sequence is followed in the commentaries: If there are negative and significant effects pre-treatment, the result is only highlighted if its positive and significant. If there are positive and significant effects pre-treatment, the result is only highlighted if its negative and significant. Finally, if there arent pre-treatment effects, the result is highlighted whenever its significant. A total of eight variables are analyzed: CAS score, three labor income variables (income of the head of family, income per capita and total family income), three employment variables (employment of the head of family, number of workers in the family and percent of workers in the family) and number of people in the family, which is analyzed because most other results depend on this variable. The tables show the estimated impact and the t-test (both already multiplied by the scaling factor), the factor itself, the mean of the 2005-2002 difference for the treated group, and the same mean for the control group. The number of treated families and the (non-weighted) number of control families are also reported. 10.1. Labor income of the head of the family TABLE 16
Family Head Labor Income Jan-Mar 2002 DD_PreTreatment DD DD_PreTreatment DD DD_PreTreatment DD DD_PreTreatment DD DD_PreTreatment DD DD_PreTreatment DD Impact $ 1,602 $ 2,385 t 1.72 1.92 Factor 1.35 1.35 1.35 1.35 1.44 1.44 1.53 1.53 1.64 1.64 1.86 1.86 Mean T Mean C NC NT
$ 4,504 $ 5,687 10573 3012 $ 5,592 $ 3,831 10573 3012
Apr-Jul 2002
$ 1,551 1.99 $ 384 0.46 $ 1,670 1.33 $ 3,385 2.95 $ 4,671 3.19 $ 3,053 2.92 $ 1,204 1.10 $ 2,724 2.60 $ 2,077 2.09 $ 3,962 3.99
$ 6,624 $ 5,475 16070 4904 $ 3,421 $ 3,705 16070 4904 $ 7,498 $ 6,338 $ 5,916 $ 3,565 $ 8,344 $ 5,290 $ 4,746 $ 2,750 9983 2756 9983 2756 9486 2587 9486 2587
Jul-Sep 2002
Oct-Dec 2002
Jan-Mar 2003
$ 6,249 $ 5,515 11109 3058 $ 1,336 $ 2,998 11109 3058 $ 5,301 $ 6,420 12331 3442 $ 691 $ 2,825 12331 3442
Apr-Jun 2003
There is a pattern in the results; we obtain negative results for the two less-treated cohorts, while obtaining positive results for the previous two. The multiplying factor has a key role here, permitting two of the results to become significant.
159
FIGURE 7
LABOR INCOME OF THE HEAD OF THE FAMILy
10.2. Total family labor income TABLE 17

Total Family Labor Income Jan-Mar 2002 DD_PreTreatment DD DD_PreTreatment DD DD_PreTreatment DD DD_PreTreatment DD DD_PreTreatment DD DD_PreTreatment DD Impact $ 79 $ 10,502 $ 4,710 $ 665 $ 8,225 $ 4,102 $ 10,818 $ 5,195 $ 4,348 $ 1,040 $ 4,033 $ 397 t 0.06 5.84 4.40 0.60 4.89 2.77 5.99 3.78 2.97 0.77 3.03 0.30 Factor 1.35 1.35 1.34 1.34 1.43 1.43 1.52 1.52 1.64 1.64 1.86 1.86 Mean T Mean C $ 5,371 $ 17,064 $ 8,828 $ 10,058 $ 12,422 $ 13,104 $ 11,771 $11,270 $ 9,039 $ 8,254 $ 8,277 $ 8,639 $ 5,430 $ 9,278 NC NT
9661 2726 9661 2726
Apr-Jul 2002
$ 5,316 15230 4659 $ 9,562 15230 4659 $ 6,688 $ 10,245 $ 4,671 $ 7,860 9296 2571 9296 2571 9008 2456 9008 2456
Jul-Sep 2002
Oct-Dec 2002
Jan-Mar 2003
$ 6,384 10501 2890 $ 8,889 10501 2890 $ 6,112 11769 3297 $ 8,426 11769 3297
Apr-Jun 2003
Three positive results are found, which is remarkable given the negative pre-treatment impact on the same cohorts.
160
FIGURE 8
TOTAL FAMILy LABOR INCOME
10.3. Per capita labor income TABLE 18

Per Capita Labor Income Jan-Mar 2002 DD_PreTreatment DD DD_PreTreatment DD DD_PreTreatment DD DD_PreTreatment DD DD_PreTreatment DD DD_PreTreatment DD Impact t Factor 1.35 1.35 1.34 1.34 1.43 1.43 1.52 1.52 1.64 1.64 1.86 1.86 Mean T $ 700 $ 6,124 $ 810 $ 4,913 $ 1,482 $ 5,340 $ 1,353 $ 4,517 $ 511 $ 3,662 $ 169 $ 3,823 Mean C $ 674 $ 5,681 NC 9662 9662 NT 2726 2726 4661 4661 2571 2571 2456 2456 2890 2890 3297 3297
$ 36 0.12 $ 597 1.37 $ 183 0.77 $ 522 1.43 $ 1,146 2.92 $ 327 0.82 $ 1,602 3.78 $ 208 0.57 $ 275 0.81 $ 2,314 6.52 $ 819 2.82 $ 1,923 5.78
Apr-Jul 2002
$ 673 15232 $ 5,303 15232 $ 684 $ 5,112 $ 302 $ 4,653 9296 9296 9008 9008
Jul-Sep 2002
Oct-Dec 2002
Jan-Mar 2003
$ 680 10501 $ 5,075 10501 $ 608 11769 $ 4,856 11769
Apr-Jun 2003
The two results found keep the pattern seen in the first graph: the lesstreated cohorts received the worst results. Again, negative results are observed on income variables.
161
FIGURE 9
PER CAPITA LABOR INCOME
10.4. Percentage of workers in the family TABLE 19

% of Workers in the Family Jan-Mar 2002 Impact t 2.49 2.36 3.94 2.05 5.73 4.63 1.16 1.40 1.15 3.07 0.06 3.39 Factor 1.35 1.35 1.34 1.34 1.43 1.43 1.53 1.53 1.63 1.63 1.80 1.80 Mean T Mean C 0.015 0.025 0.015 0.016 0.017 0.019 0.010 0.014 0.011 0.011 0.008 0.012 0.009 0.018 0.007 0.011 0.002 0.006 0.007 0.011 0.008 0.004 0.008 0.005 NC 11753 11753 18000 18000 10948 10948 10455 10455 12326 12326 13897 13897 NT 3333 3333 5492 5492 3017 3017 2843 2843 3406 3406 3876 3876
DD_PreTreatment 0.008 DD 0.010 DD_PreTreatment 0.011 DD 0.006 DD_PreTreatment 0.021 DD 0.019 DD_PreTreatment 0.005 DD 0.006 DD_PreTreatment 0.004 DD 0.011 DD_PreTreatment DD 0.000 0.012
Apr-Ju1 2002
Jul-Sep 2002
Oct-Dec 2002
Jan-Mar 2003
Apr-Jun 2003
Five of the six possible results are positive and significant, which is very impressive. Although the results are not large in magnitude, since there are a large number of observations, it is statistically possible to capture effects of that size. It is worth remarking that the mean values of the variables are not high, so the impact attributable to the program represents an important part of that change.
162
FIGURE 10
PERCENTAGE OF WORKERS IN THE FAMILy
10.5. Number of workers in the family TABLE 20

Number of Workers in the Family Jan-Mar 2002 DD_PreTreatment DD DD_PreTreatment DD DD_PreTreatment DD DD_PreTreatment DD DD_PreTreatment DD DD_PreTreatment DD Impact 0.043 0.113 0.063 0.087 0.085 0.097 0.060 0.094 0.058 0.096 0.064 0.095 t 3.63 7.89 6.20 7.79 6.09 6.78 4.09 6.41 4.25 7.26 5.17 7.79 Factor 1.35 1.35 1.34 1.34 1.43 1.43 1.53 1.53 1.63 1.63 1.80 1.80 Mean T Mean C 0.047 0.077 0.060 0.040 0.074 0.055 0.064 0.053 0.057 0.028 0.056 0.030 0.016 0.007 0.013 0.025 0.014 0.013 0.024 0.008 0.021 0.031 0.020 0.023 NC 11753 11753 18000 18000 10948 10948 10455 10455 12326 12326 13897 13897 NT 3333 3333 5492 5492 3017 3017 2843 2843 3406 3406 3876 3876
Apr-Jul 2002
Jul-Sep 2002
Oct-Dec 2002
Jan-Mar 2003
Apr-Jun 2003
This presents the most important evidence as to the effects of the program. Not only are all the results positive, but also all had pre-treatment negative effects. The evidence is very strong in this case, and consistent with the women-oriented Chile Solidario policies.
163
FIGURE 11
NUMBER OF WORKERS IN THE FAMILy
10.6. CAS score TABLE 21

CAS Score Jan-Mar2002 DD_PreTreatment DD DD_PreTreatment DD DD_PreTreatment DD DD_PreTreatment DD DD_PreTreatment DD DD_PreTreatment DD Impact 1.558 7.795 0.159 2.908 2.571 1.845 0.519 0.599 2.640 1.991 0.880 2.992 t 2.91 11.37 0.35 5.49 4.26 2.76 0.84 0.93 4.47 3.26 1.64 5.33 Factor 1.35 1.35 1.34 1.34 1.43 1.43 1.53 1.53 1.63 1.63 1.80 1.80 Mean T Mean C 7.092 28.542 8.365 25.300 9.428 23.008 7.808 20.051 8.962 18.817 8.445 17.309 8.249 22.753 8.484 23.130 7.636 21.722 8.147 20.443 7.338 20.042 7.956 18.970 NC 11753 11753 18000 18000 10948 10948 10455 10455 12326 12326 13897 13897 NT 3333 3333 5492 5492 3017 3017 2843 2843 3406 3406 3876 3876
Apr-Jul2002
Jul-Sep2002
Oct-Dec2002
Jan-Mar2003
Apr-Jun2003
A reiterative pattern is present here, with the cohorts around the middle obtaining the most positive impacts and the less-treated cohorts the worst. Nevertheless, the magnitudes of the impacts are very small.
164
FIGURE 12
CAS SCORE
10.7. Employment of the head of the family TABLE 22

Employment of the Family Head Jan-Mar 2002 DD_PreTreatment DD Apr-Jul 2002 DD_PreTreatment DD Jul-Sep 2002 DD_PreTreatment DD Oct-Dec 2002 DD_PreTreatment DD Jan-Mar 2003 DD_PreTreatment DD Apr-Jun 2003 DDPreTreatment DD Impact 0.006 0.019 0.000 0.019 0.005 0.029 0.009 0.020 0.018 0.011 0.017 0.036 t 0.79 2.20 0.02 2.83 0.53 3.50 1.04 2.29 2.30 1.33 2.18 4.94 Factor 1.35 1.35 1.34 1.34 1.43 1.43 1.53 1.53 1.63 1.63 1.80 1.80 Mean T Mean C 0.020 0.020 0.009 0.028 0.004 0.031 0.013 0.027 0.004 0.039 0.010 0.017 0.015 0.034 0.010 0.042 0.001 0.052 0.007 0.040 0.015 0.046 0.019 0.038 NC 11753 11753 18000 18000 10948 10948 10455 10455 12326 12326 13897 13897 NT 3333 3333 5492 5492 3017 3017 2843 2843 3406 3406 3876 3876
Again, there is a clear, positive pattern on the employment of the head of the family. The magnitudes are all around 2%, which is not huge, but this variable is very important for the well-being of the families.
165
FIGURE 13
EMPLOyMENT OF THE HEAD OF THE FAMILy
10.8. Number of people in the family TABLE 23

Number of People in the Family Jan-Mar 2002 DD_PreTreatment DD DD_PreTreatment DD DD_PreTreatment DD DD_PreTreatment DD DD_PreTreatment DD DD_PreTreatment DD Impact 0.022 0.368 0.106 0.266 0.087 0.215 0.170 0.241 0.153 0.219 0.207 0.281 t 0.92 13.31 4.98 11.90 2.92 7.88 5.34 8.87 5.41 8.79 7.54 12.22 Factor 1.35 1.35 1.34 1.34 1.43 1.43 1.53 1.53 1.63 1.63 1.80 1.80 mean T 0.022 0.089 0.053 0.127 0.086 0.083 0.125 0.083 0.098 0.106 0.114 0.092 mean C 0.038 0.362 0.026 0.325 0.025 0.232 0.013 0.240 0.004 0.241 0.001 0.248 NC 11753 11753 18000 18000 10948 10948 10455 10455 12326 12326 13897 13897 NT 3333 3333 5492 5492 3017 3017 2843 2843 3406 3406 3876 3876
Apr-Jul 2002
Jul-Sep 2002
Oct-Dec 2002
Jan-Mar 2003
Apr-Jun 2003
This is the clearest pattern of all, and the reason for this phenomenon is not clear. Possible hypotheses include people moving in with their close families or friends, to get the benefits of the program. Given these results, it is necessary to be very careful with the interpretation of the previous results, especially with the total labor income of the family or the number of workers, which will tend to be overestimated- and with the per capita variables, which will tend to be underestimated.
166
FIGURE 14
NUMBER OF PEOPLE IN THE FAMILy
11. Conclusions The Impact Evaluation of Chile Solidario is a pioneer initiative in Chile to perform a follow-up over time of the families that benefitted and families that did not benefit from the system. The lessons learned in this process of more than four years of investigation provide knowledge that is unique in Chile about the design and evaluation of future public policies. In spite of this, there are a number of serious problems coming from a careful analysis of the data that cause the results to be less reliable than might be deemed necessary for making future policy decisions. One problem was that the Chile Solidario Panel sample did not have an adequate sample selection, nor was it transparent enough, which is evident in the difficulty in creating an adequate control group. In addition, the unit to be followed through time was never formally established. The attrition rate up to 2004 was not documented and no measures to reduce it had been taken. Through all the rounds there were serious problems in obtaining beneficiary IDs. With respect to the original sample design, serious mistakes were made when choosing the controls. Finally, and most importantly, it was not possible to count with a baseline. In this work, an analysis of the feasibility of carrying out an impact evaluation for the Chile Solidario program was made. The process by which slots by district are assigned and the mechanism by which families are taking the slots in the program has been fully described. Both aspects are essential to achieving a valid identification strategy. The group of variables that appear to have an influence in the participation in the system have been described, and the possibility of carrying out an impact evaluation by Matching was discussed. Using this group of variables in the propensity score, the quality of the matching was tested on pre-treatment variables. The Chile Solidario Panel was put to the test, estimating the impacts on a group of variables for which there is pre-treatment information. The results are not very encouraging, since there are many variables showing pre-treatment differences between the treatment and control groups. It is probable that previous Chile Solidario impact studies
167
that did not use administrative variables had strong pre-treatment differences between the treatment group and the control group before the first participants entered the program. As a result of these problems, the results obtained in this evaluation should be qualified as not conclusive. From a total of 10 related variables related to employment and income the only ones allowing a comparison with pre-treatment CAS variables four are not possible to estimate, because of large pre-treatment differences. The results for the income variables are not very illustrative, since the variable for the number of people in the family is unbalanced. The variables for the number of employed people in the family, the number of working women, and the percentage of working women have negative or zero impact using the differences-in-differences approach. Head of the family employment has, instead, positive differencesin-differences results. Among variables without baseline, in terms of indigence and poverty we observed strongly negative impacts in the cross-sectional analysis. However, the impact when using differences-in-differences is zero; it is highly probable that this is due to differences in the pre-treatment situation of the groups. In the psychosocial area and recruitment to public offerings there is an important group of variables for which cross-sectional impacts are found, but these same impacts are not found in differences-in-differences. It is likely that these impacts are real, but in a strict sense there is no way to distinguish if the groups had these differences prior to the process or if they are the result of the program. Inside this group are the expectations of change, employment registration and vision of future variables, for example. Finally, take-up of post-natal care and dental examinations do have differences-in-differences impacts. In order to solve the main problems of the Chile Solidario Panel, an Administrative Panel Database was generated, using over 800,000 registries per year, across seven years. This database had, however, some problems, such as selective attrition and measurement error. One of the most important problems was the big proportion of families on the control group enrolling in Chile Solidario. A simple extension of the standard correction for substitution programs was adopted, and the impacts were estimated. There are a number of positive significant effects, which are especially robust on the employment variables. Even though the magnitudes are not huge, achieving impacts on important variables such as employment of the head of the family may make substantial changes in the quality of life for those families. These results are much more robust than those obtained from the Chile Solidario Panel. The overall lesson of this evaluation for future social programs is clear: evaluation and management of the program should go hand in hand from the moment the program is being designed. In this sense, generalizing the usage of independent evaluations and professionally evaluated pilot programs is highly recommended. This may be the only way to rectify in forthcoming evaluations most of the difficulties faced by the evaluation of the Chile Solidario program. All large-scale public programs should follow this trend, improving every time the way in which the evaluations are planned and organized.
168
12. References Abadie, A. and G. Imbens (2006). Large Sample Properties of Matching Estimators for Average Treatment Effects, Econometrica. Blundell, R. and M. Costa Dias (2002). Alternative Approaches to Evaluation In Empirical Microeconomics. Working Paper, CENMAP. Carneiro, P. and E. Galasso (2007). Lessons from the evaluation of Chile Solidario. World Bank. Dehejia, R. and S. Wahba (1998). Causal Effects in Non-Experimental Studies: Re-Evaluating the Evaluation of Training Programs, The Review of Economics and Statistics. Galasso, E. (2006). With their effort and an opportunity. World Bank. Heckman, J., N. Hohmann, J. Smith and M. Khoo (2000) Substitution And Dropout Bias in Social Experiments: A Study of an Influential Social Experiment. Heckman, J., H. Ichimura and P. Todd (1997). Matching as an econometric evaluation estimator: Evidence from evaluating a job training programme, Review of Economic Studies, 64, 605-654. Heckman, J., R. LaLonde and J. Smith (1999). The economics and econometrics of active labor market programs. In: Handbook of Labor Economics. Hoces de la Guardia, F., A. Hojman (2007). Revisin al Diseo de las Muestras en las rondas del Panel Chile Solidario. Imbens, G. (2007). Estimation of Average Treatment Effects Under Unconfoundedness, In: Imbens & Wooldridge, Whats new in econometrics, Lecture Notes 1, Summer 2007. Imbens, G. (2004). Nonparametric Estimation of Average Treatment Effects Under Exogeneity, a Review, The Review of Economics and Statistics. Imbens, G. (2004). Implementing Matching Estimators for Average Treatment Effects in STATA, STATA Journal. LaLonde, R. (1986). Evaluating The Econometric Evaluation of Training Programs with Experimental Data, American Economic Review. Manzi, J. y R. Gonzlez (2007). Evaluacin de Impacto de Chile Solidario, Documento N 1: Apoyo Psicosocial; Divisin Social, MIDEPLAN. Perticar, M. (2006). Aplicacin y toma de encuestas a hogares beneficiarios y controles del Sistema de Proteccin Social Chile Solidario: Anlisis Cuantitativo de Impacto. Rosenbaum, P. R. and D.B. Rubin (1983). The central role of the propensity score in observational studies for causal effects, Biometrika, 70, 41-55. Secretara Ejecutiva Chile Solidario, Ministerio de Planificacin (2003). Seleccin de las familias beneficiarias final. Secretara Ejecutiva de Chile Solidario, Ministerio de Planificacin Cobertura de Chile Solidario por Regin 2002-2005 Final. Archivo Excel. Ministerio de Planificacin (2006). Cuaderno de Informacin N 1: La creacin del Sistema de Proteccin Social Chile Solidario: Debate Previo y Propuesta. Secretara Ejecutiva de Chile Solidario, Ministerio de Planificacin (2006). Cuadernos de Informacin N 2. Secretara Ejecutiva Chile Solidario, Ministerio de Planificacin. Seleccin de las familias beneficiarias final.
Estudios de Economa. / Claudia Sanhueza, Ricardo Mayer Top incomes in ChileVol. 38 - N 1, Junio 2011. Pgs. 169-193
169
Top Incomes in Chile using 50 years of household surveys: 1957-2007*

Ingresos ms altos en Chile usando 50 aos de encuestas de hogares: 1957-2007
Claudia Sanhueza** Ricardo Mayer*** Abstract Using household surveys that cover more than 50 years of the political and economic history of Chile, we investigate changes in the shape of the distribution of income in Chile, and in the composition of top 10% and top 1% incomes. In line with international evidence, top incomes concentration appears to be countercyclical in the short run. For the entire length of this survey, this concentration shows roughly an inverted U-shape, peaking at the end of the 80s. These changes correspond approximately with different economic policy regimes prevailing in Chile. We observe important changes in the composition of top income groups related to greater relative importance of women, employees and college schooling levels. These changes are stronger for the top 10% than the top 1% of incomes. Additionally, using a national level panel of households for the period 1996-2006 we explore correlations between probabilities of permanence and arrival to the top decile with variables such as composition of the household, ownership of physical and human assets, job quality and changes in the numbers of household members working in the labor market. Key words: Income distribution, Income mobility. Resumen Usando encuestas de hogares que cubren ms de 50 aos de la historia poltica y econmica de Chile, investigamos los cambios en la forma de la distribucin del ingreso en Chile y en la composicin del 10% y 1% superior. De acuerdo con la evidencia internacional, la concentracin de los ingresos ms altos parece ser contracclica en el corto plazo. Para todo el perodo de estudio,
* We thank Sergio Ros for his excellent research assistance. ** Instituto de Polticas Pblicas, Universidad Diego Portales, Ejrcito 260, Santiago, Chile. E-mail: claudia.sanhueza@udp.cl. *** Facultad de Economa y Empresa, Universidad Diego Portales, Manuel Rodrguez Sur 0253, Santiago, Chile. E-mail: ricardo.mayer@udp.cl.
170
esta concentracin muestra una forma de U invertida, alcanzando un mximo a finales de los aos 80. Estos cambios corresponden a diferentes regmenes de poltica econmica vigentes en Chile. Observamos cambios importantes en la composicin de los grupos de ingresos ms altos: mayor importancia relativa de mujeres, empleados y personas con educacin universitaria. Estos cambios son ms fuertes para el 10% que para el 1% ms rico. Adems, utilizando un panel nacional de hogares para el perodo 1996-2006 se exploran las correlaciones entre las probabilidades de permanencia y llegada al decil 10 con variables tales como la composicin de la familia, la propiedad de activos fsicos y humanos, la calidad del empleo y los cambios en el nmero de miembros del hogar que trabajan en el mercado de trabajo. Palabras clave: Distribucin del ingreso, Movilidad del ingreso. JEL Classification: D31, J6.
1. Introduction An income distribution more concentrated at the top has significant implications for the economy and politics. Leigh (2009) argues that if a small elite gets a big share of societys income, it could influence certain industries and, through their campaign contributions, certain politicians. Moreover, Frank (2007) notes that the increase in spending of high-income individuals can affect the middle class because of a contagion effect on the rest of the population. He argues that the welfare evaluation depends on context, and therefore consumption choices also depend on the comparison made by the individual with respect to those around him. Finally, Tawney (1913) argues that understanding the concentration of incomes at the top of distribution tells us something about the bottom part of it, since the concentration of income at the top is highly correlated with relative poverty. With this in mind, several authors have analyzed the evolution of top incomes by building long time series of top income shares during the twentieth century. This includes Piketty (2003), Piketty and Saez (2006), Atkinson (2002), Saez and Veall (2005), Atkinson and Leigh (2005, 2007), Atkinson and Piketty (2007), Atkinson and Salverda (2005), Banerjee and Piketty (2005) among others. On the other hand, Saez and Veall (2005) and Kopczuk, Saez and Song (2007) have also studied the welfare consequences of income mobility at the top of the income distribution and the effect of the increase in female labor participation. These topics are especially important in Latin American countries, given that most of them present the worlds worst indicators of income inequality and the factors behind this high and stable inequality are in permanent discussion. Chile, the subject of this paper, ranks among the most unequal countries in the region,1 only bellow Brazil, Guatemala and Colombia.
1
World Bank Report, 2003, Inequality in Latin America: Breaking with History?, Chapter 1.
Top incomes in Chile / Claudia Sanhueza, Ricardo Mayer
171
During the last 50 years Chile has faced various types of economic situations as a result of international events or policies developed within the country. Including conservator, social-democratic and socialist governments and a 17 years strong dictatorship. In the last two decades, after the dictatorship, Chile has shown high growth rates, greater macroeconomic stability, significant reduction in poverty rates, all generating well-being of the population. However, with all this progress the issue of income inequality shows no improvement, and Chile is among the countries with the highest levels of inequality in Latin America. The richest 10 percent of individuals in Chile receive 47 percent of total income and the bottom 20 percent receive 3.4 percent. By way of comparison, in the United States the richest 10 percent receive 31 percent of total income, and the bottom 20 percent receive 5.2 percent of the total income. The evolution of income inequality has attracted the attention of both economic and political studies, especially as the economy grows. This evolution has been affected by different changes: developments in technology, policy interventions, political shocks, changes in social indicators and other factors. Although, Chile has grown there is the doubt about whether this has led to benefit a small group or owners of the capital, leaving behind a large group of the population. Studies done for Chile show that in the short term, the distribution of income does not appear to be affected, so that one can sense that specific policies do not have permanent or long term consequences and therefore they will not be effective to combat the problem of income inequality. Thus, the analysis is to be returned to a long-term horizon, and that is why it is necessary to have historical series, which allows us to understand the problem. The aim of this paper is to study the evolution of top income shares in Chile and the composition of this group during the last 50 years. In addition, using panel data we study income mobility at the top of the distribution in recent years and its determinants. The available information allows us to study the evolution of top incomes shares, distinguishing between individual and family income, and hence to investigate if the increase of female participation rate in the labor market has contributed to greater income concentration due to the interrelation between spouses income. To study the evolution of top incomes in the last 50 years we use the Employment and Unemployment Survey of the University of Chile. This survey contains considerable information about incomes for a sample of households in Greater Santiago between 1957 even today. With this information the following indicators are constructed: average real income of decile 10, percentile 1, percentiles 10 to 2, growth of those average incomes, participation relative to the entire revenue of every year, distances between average income and decile 10s average income and others. All these indicators are constructed for both household and individual incomes. Additionally, we analyze their composition in terms of wages and other incomes, education, gender and types of occupation. In turn, for the analysis of mobility in decile 10, we use CASEN panel surveys. This allows us to study transitions in the upper part of the distribution, answering the question of what is the probability of remaining in decile 10 and the probability of arriving to the top decile 10. Moreover, the wealth of information readily available in household surveys allows us to study the variables correlated with the probability of remaining in decile 10 and of arrival to decile 10, using initial conditions as explanatory variables, including household composition variables, household assets and shocks.
172
This paper contributes to the debate on the income distribution in Chile in several ways. First, to our knowledge this is the first time that the long series of top incomes has been analyzed for the Chilean case, which is key to getting insights about the effects of different economic policy regimes. Ruiz-Tagle (1998) uses the same survey to build indicators of income distribution over 40 years, however, he did not analyze what happened to the top income shares. Second, the determinants of the probability of arriving and remaining at the top of the income distribution have not been analyzed before. Contreras et al. (2004) study the dynamic of poverty using one of the waves of Panel Casen surveys. They do not include the study of the dynamic of the top of the distribution. It is worth noting that we use survey data rather than income tax data. Most of top income studies have made use of tax data. Saez (2004) argues that surveys information is available only in the recent years and that, at least in the United States, the household surveys present information on codified form or by stretches. On the other hand, tax data also suffer from certain problems. First of all, income information is based on self-reported information therefore problems of evasion and elusion can slant the results. Second, taxes statistics cover only a fraction of the population. Historically, the fraction of the population who declares income is small and there is a big part that is exempt of it or where informality conditions prevail in the labor market. These facts are especially important in the case of Chile. For this research, we maintain that the use of household surveys allows us to address these issues. First, the extraordinary series of 50 years of the Employment and Unemployment Survey allows us to circumvent the issue of short lengths of time. Second, income data is captured as a continuous variable, and not coded into income sections. Also the information in the survey registers the identity of income sources and whether they are individual or family figures. Third, household incomes in the surveys may come from informal mechanisms or be exempt from taxes. The structure of the papers is as follows. After this introduction we present a brief literature review. Then we present the data and methodology in section 3. In section 4 we present the evolution of top incomes in Chile and in section 5 the mobility analysis at the top of the income distribution. Finally, section 6 concludes. 2. Literature Review A review of recent literature is available in Saez (2004), Piketty and Saez (2006) and Leigh (2009). Series of top incomes have been produced for various developed countries, including Australia (Atkinson and Leigh, 2007), Canada (Saez and Veall, 2005), Finland (Riihel, Sullstrm and Tuomala, 2005), France (Piketty, 2003), Germany (Dell, 2007), Ireland (Nolan, 2007), Japan (Moriguchi and Saez, 2008), Holland (Atkinson and Salverda, 2005), New Zealand (Atkinson and Leigh, 2005), Spain (Alvaredo and Saez, 2006) Switzerland (Dell 2005, Dell, Piketty and Saez, 2007), United Kingdom (Atkinson, 2002, 2007) and U.S. (Piketty and Saez, 2003). Piketty (2004) and Legih (2009) emphasize that international comparisons find a significant decrease in top income share during the first half of the twentieth
173
century in all countries except Switzerland, with a later increase of this shares in the second part of the century mainly in Anglo-Saxon countries but not in Japan and continental Europe. Indeed, top income shares make a full recovery in the U.S., a significant one in England and Canada, and none in France. This fall in the first part is attributed to the incorporation of highly progressive tax systems after the Second World War and the subsequent growing importance of salaries in the composition of top incomes, which in turn have won profitability due to technological progress. Moreover, Leigh (2009) noted that differences between countries are not due to institutional differences in the labor market, such as levels of centralization of collective bargaining. Atkinson (2002) studies the evolution of top incomes in the United Kingdom. In his research advances beyond what previous authors had developed for the UK since it tries to identify the amount of aggregate income and aggregate population and argues that his data is a unique source of evidence on the distribution of higher incomes which allows him to cover the twentieth century. He shows that the First and Second World Wars conveyed a significant drop in the income shares of the top 0.05% and 1% incomes. Piketty (2003), in turn, studies the same series for France. In particular, he concludes that the decline in France of income inequality is largely accidental. For the United States, Piketty and Saez (2003) work with a database with information about the concentration of wealth and income. They acknowledge that working with this type of information has important limitations. In particular they mention that their long term series have little information on the bottom incomes, but because of being homogeneous across the countries and decomposed in different income sources, they are the only opportunity to understand the dynamics of the distributions of income and wealth. They mention that the general pattern, across the century, for decil 10s income has a U shape, that it experienced a substantial decrease, greater than 30% during WWII, and that remained above 31 and 32 per cent until 1970. After decades of stability in the post-war period, the share of the richest decile increased dramatically on the last 25 years, reaching its pre-War levels, but with a different composition in which the labor income is now the main income source. Additionally, Saez and Veall (2005) studied the evolution of high-income families and individuals, concluding that the historical evolution of both series follow the same pattern. This indicates that in spite of increasing incorporation of women into the labor market, this does not improve or deteriorate the concentration of incomes, probably due to the correlation between the earnings of spouses. They also study the consequences in terms of welfare for income mobility. They find that there has been an increase in mobility in Canada at the top of the income distribution. In each of the works mentioned the methodology is very similar, and so are the main conclusions, apparently because developed countries have followed the same trend in the implementation of tax policies. For example, Piketty (2003) argues that many authors have said that the dramatic increase in progressive taxation taking place in the interwar period has been the main factor that prevent income and wealth shares return to their previously high levels. This taxation trend would explain also the generally observed decrease in the relative importance of capital revenues and an increase in the relative importance of labor income as determinant of total income, at least in recent decades.
174
A short background about Chile Chile has been pointed out as an example of successful economic development among developing economics. During the last decades has shown an increasing and currently high per capita GDP, a significant reduction in poverty rate and economic and political stability. A set of innovative reforms implemented during the last decades have made of Chile a reference for countries in similar, lower and even upper stages of development. However, and despite the success, probably the single most important failure of the Chilean experience has been the high and apparently permanent income inequality. Ruiz-Tagle (1998) uses the Employment and Unemployment Survey of the University of Chile and constructs series of indicators of income distribution in Santiago (Chile). He concludes that inequality has generally increased since 1957, peaking during the 1980s and improving a little over the following decade. The indicators he uses exhibit high levels of persistence over time, leading him to conclude that income distribution cannot be modified significantly over short periods of time. Larraaga (1999) studies the relationship between economic growth and income distribution in Chile in the 1987-1996 period. To this end, he uses CASEN household surveys from years 1987, 1990, 1992, 1994 and 1996, disaggregating the dynamics of growth and income distribution in a sector-bysector basis. His main findings are an increasing and concave relationship between mean income and sectors inequality, and a positive correlation between mean income and changes in levels of inequality. These two findings hold for short run (two years) as well as for long run (ten years) analyses. Larraaga and Valenzuela (2011) study the factors behind the stable and high-income inequality in Chile, given the great changes realized in the country from 1990 to 2003. They concluded that although several factors changed over the period their effects over inequality cancelled out. Valenzuela y Duryea (2011) using micro-simulations compare the income distribution of Chile with respect to Uruguay. They find that the main differences between the two countries are present at the top of the distribution, and they are due to the composition of income in those deciles. In Chile the proportion of income that comes from employers income is higher and more unequal. Finally, Sapelli (2011) looks at the income distribution by cohorts in Chile by constructing a synthetic panel and estimate the income distribution for cohorts born between 1902 and 1978. The cohort effects show a period where inequality increases and then decreases. The rise can be explained by variables associated with education, while the fall appears to be the consequence of a flattening of the income-age profile and hence a reduction in the returns to experience. Engel and Eberhard (2007) show evidence that inequality has been decreasing since 1990s due to a decrease in the college skill-premium that was due to the deregulation of the college market during 1980s. Regarding short-run mobility analysis Contreras, et al. (2004) analyzes the dynamics of the relationship between poverty and poverty mobility using CASEN panel survey for 1996 and 2001. They look at probabilities of arriving to and departing from poverty and the factors behind these movements in and out of poverty. They find high mobility in all deciles from deciles 1 trough 9, which implies that more than half of the population is potentially in risk of falling into poverty. The opposite holds for the transition between deciles 9 and 10, for which there is no such high mobility. The bottom forty percent of the income
175
distribution does not have the resources to deal with illness or health hazard affecting the household head. On the other hand, long-term mobility estimates show high-income elasticity and therefore a country with low social mobility (Nuez and Risco, 2004 and Celhay et al., 2010). 3. Data and Methodology Regarding the analysis of the evolution of top incomes we use the Employment and Unemployment Survey of the University of Chile. This is the oldest survey available in Chile and has rich relatively information on income of households in Greater Santiago from 1957 up to current date. Another key feature of this database is its homogeneity, as the survey format has remained virtually the same over all these years, therefore information is similar throughout the period, making it easier to validate comparisons over time. This survey allows us to distinguish between total household income, individual income and per capita household income. Figure 1 shows the monthly per capita GDP between 1957 and 2004 taken from Diaz, Luders, and Wagner (2007) and per capita household income from the Employment and Unemployment Survey of the University of Chile. Both series are in real Chilean pesos. We can observe similar trends in both series. However, the income measure from the survey is below the range of the GDP. There are several reasons for that. First, GDP includes production activities than are carried out within the boundaries of the country that are not a property of nationals. Second, the Survey includes only income from Greater Santiago and GDP is a national measure. In the Appendix we show the table underlying this chart FIGURE 1
COMPARISON OF TOTAL INCOME AND GDP PER CAPITA: CHILE 1957-2007
250000 200000 150000 $ 100000 50000 0 1957 1959 1961 1963 1965 1967 1969 1971 1973 1975 1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 Per capita household income Per capita monthly GDP (Diaz, Luders y Wagner)
176
and also a comparison between individual and total income of the household. Incomes from the survey were corrected by CPI to be left in real Chilean pesos, and moreover, the different currencies used during the period of study were made equivalent. The sample size corresponds, roughly, to 10,000 people and 5,000 households on average per year. Table 1 shows some descriptive statistics for the year 2007. We focus on the decile 10 and the 99th percentile of income distribution. We see that the average household income was 511,929 Chilean pesos (930 dollars). The median on the other hand, is 338,929 (615 dollars). In the case of income distributions with a long tail, the median is a better indicator of mean income than the average income. Moreover, the average total household income in decile 10 is 1,895,859 Chilean pesos (3,450 dollars), and within the richest 1% is 4,333,647 Chilean pesos (7,880 dollars). The minimum household income of decile 10 is 1,060,429 (1,930 dollars) and the minimum household income of percentile 99 is 3,305,993 (6,010 dollars). This shows the great gap between incomes of the richest deciles and average households in Chile. TABLE 1
AVERAGE INCOME THRESHOLDS AND TOP 10% AND TOP 1%, CHILE, 2007 Average population Total household income Individual income Per capita household income 511,929 246,201 130,713 Average top 10% 1,895,859 1,028,191 315,574 Average top 1% 4,333,647 2,948,340 1,254,873
Median
p90
p99
338,969 157,767 82,357
1,060,429 518,915 273,407
3,305,993 1,740,877 841,144
Source: Employment and Unemployment Survey, Universidad de Chile, 2007. 8,870 individuals. 4,934 households.
Income proportions were calculated using as a numerator the sum of incomes of all individuals or households in decile 10 (and top 1 %), divided by the sum of the income of all individuals or households in the sample. Another way of doing it would be to use as denominator the monthly GDP of the country. Nevertheless, as we said previously this one includes incomes that are not a property of nationals. In case of the individual income they include only the income of individuals who work or have personal individual revenue. On the other hand, the survey has information on the sources of individual income. This allows us to distinguish the differences in the evolution of the different sources for top incomes: wages, capital gains or other incomes. Income is divided in the following way: i) salaries and wages, ii) independent income, originated from industrial, agricultural, commercial and professional activities, iii) pensions and iv) other incomes, which includes capital revenues in addition to other non welfare income. We also take other individual information, such as the type of occupation and gender.
177
To perform mobility analysis in the high part of the income distribution, however, we need longitudinal information. The survey panel that covers the longest time period in Chile is the survey Panel CASEN 1996-2006. Using three periods of the survey we calculate matrices of mobility for the top decile between 1996-2001 and 2001-2006. Also, we estimate two models of discreet dependent variable in which we identify variables correlated with the probability of permanence in the decile 10 from the rest of the distribution and variables correlated with the probability of arrival to the 10th decile. The permanence in the top decile is studied by means of the construction of a discreet variable that takes the value of 1 if the household is observed in the 10th decile in year t conditional on being in decile 10 in year t1, and 0 if person is not observed in decile 10 in year t conditional on being in decile 10 in year t1. The model incorporates characteristics of the household in year t1, and changes produced between t1 and t. For the case of arriving to the decile 10 we generate a discreet variable that take the value of 1 if the household is observed in decile 10 in year t conditional on been in decile 10 in year t1, and 0 if the household is not observed in decile 10 in year t conditional on not been in decile 10 in year t1. The model incorporates characteristics of the household in year t1, and changes produced between t1 and t. The explanatory variables for both models include composition of the household, physical and human capital, households geographic characteristics and shocks. Shocks include health problems, changes of the numbers of persons in the household, and changes of the number of persons in the household that work. 4. Top Incomes Evolution in Chile Figure 2 shows the evolution of the top deciles share of total households income in Chile between the years 1957 and 2007. Just like the available evidence for other countries has suggested elsewhere, the behavior of this series appears to be countercyclical. During a economic crisis, for example during the crisis of 1982, the populations elite who enjoy major capital endowment will increase his economic differences with those more disadvantaged. On the other hand, during periods of economic expansion as those that happened from 1987 up to 1996, the participation of decile 10 was diminishing. When the country began to slow down its economic growth, the income share of the rich began rising again. The evidence suggests that for periods of rapid economic growth wage gaps tends to diminish and the opposite holds for periods of slower or negative economic growth. Also we can observe a inverted-U shape for top income shares in the period under study, which reaches a maximum at the end of the 80s. Figure 3 shows the series of income shares of the household grouped in income ranges p90-95, p95-99 and p99-100. We note that in periods of growth of the top decile ( p90-100) it was primarily due to growth in the richest part of the top decile: p95-99 and p99-100. The indicator for p90-95 percentile even decreases gradually in the period under study. This shows that the concentration in the upper part of the distribution is mainly determined by what happens in the richest 5% of the population.
178
FIGURE 2
PROPORTION OF INCOME DECILE 10 50.00 45.00 40.00 35.00 30.00 25.00 20.00 1957 1959 1961 1963 1965 1967 1969 1971 1973 1975 1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 Decile 10 %
FIGURE 3
PROPORTION OF INCOME p90-95, p95-99, p99-100 30.00 25.00 20.00 15.00 10.00 5.00 0.00
1955 1957 1959 1961 1963 1965 1967 1969 1971 1973 1975 1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 2009 %
p99-100
p90-95
p95-99
179
Total household income, per capita family income and individual income In this section we present the income share of the top decile and the top percentile using three different measures: total household income, per capita household income and individual income. Changes in labor force participation of married women may have introduced differences in income shares based on whether we look at total household income or individual income. We therefore compare both series. Figure 4 shows this comparison and we can se that although very similar, income distribution appears less concentrated when measured on total household income than when using individual income. This greater inequality of individual income vis-a-vis total household income may be reflecting that increases in married womens labor participation was more important for low level families than for rich ones, making comparisons of households more egalitarian, but also could be a reflection of the fact that the majority of women newly incorporated to the labor market earned relatively low salaries, increasing the numbers of low-income individuals reporting earned incomes more rapidly than that of high-income individuals. In addition, per capita family income allows us to incorporate the effects of household size. As households in the richest deciles have less people, when we calculate the proportion of income using per capita household income will observe higher or at least equal percentages. Figure 5 shows this comparison. In a large majority of years, this is in fact true in the data. Exceptions include some years at the beginning of the sample period and some years where the top decile looses participation on total income. FIGURE 4
PROPORTION OF INCOME DECILE 10 INDIVIDUAL INCOME VS TOTAL HOUSEHOLD INCOME 55.00 50.00 45.00 40.00
%
35.00 30.00 25.00 20.00

1955 1957 1959 1961 1963 1965 1967 1969 1971 1973 1975 1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 2009
Total household income
Individual income
180
FIGURE 5
PROPORTION OF INCOME DECILE 10 TOTAL HOUSEHOLD INCOME VS PER CAPITA HOUSEHOLD INCOME
55.00 50.00 45.00 % 40.00 35.00 30.00 25.00 20.00 1955 1957 1959 1961 1963 1965 1967 1969 1971 1973 1975 1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 2009 Total household income Per capita household income
But the differences, when present, are small. This suggests that no significant changes arise whether we use total household income, per capita household income or individual income. Figure 6 shows the comparison of the types of income for the income share of the individuals or households belonging to percentile 99-100. We notice a similar pattern in all three series.
FIGURE 6
COMPARISON TOTAL HOUSEHOLD INCOME, PER CAPITA HOUSEHOLD INCOME, INDIVIDUAL INCOME
20.00 18.00 16.00 14.00 12.00 10.00 8.00 6.00 4.00 2.00 1957 1959 1961 1963 1965 1967 1969 1971 1973 1975 1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 Total household income Per capita household income Individual income 0.00 %
181
Income composition One big advantage that gives us the survey is to decompose the income of individuals in their different sources of origin: i) salaries and wages, ii) independent income from industrial activities, agricultural, commercial and professional iii ) retirement and iv) other income, which includes the capital rental income plus other income not specified. The decomposition was carried out on individual income. Figure 7 shows the series for the top decile. Most of the incomes of individuals are from wages; this proportion has increased steadily over time. The evolution of independent FIGURE 7
SOURCES OF INCOME IN DECILE 10
100 90 80 70 60 50 40 30 20 10 0
1961 1962 1963 1964 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007
Other income
Pensions
Wages
Independent
FIGURE 8
SOURCES OF INCOME IN p99-100
100 90 80 70 60 50 40 30 20 10 0
1961 1962 1963 1964 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 2001 2002 2003 2004 2005 2006 2007
Other income
Pensions
Wages
Independent
182
income, which may include income from holding any capital or business, shows a downward trend. In the same way we show the composition of income for the richest 1% of the sample. We note that, for this small segment of the population, its compositional ranking is different from that of the top decile, obtaining most of its income in the form of independent income from holding any capital or business. Education When looking at the data series, we can see two trends that are more or less clear: throughout the period the coverage of secondary education and higher education have increased steadily, but while the category of individuals with tertiary education has been increasing its share among the highest income, the opposite happens with those with only secondary education. People with higher education went from representing just fewer than 40% of the top decile in 1957 to 80% in 2007, while in this same group the participation of high school graduates fell from an average above 40% in the first years of the sample to average about 20% in the last years of the survey. The evolution of the presence of these educational groups in percentile 100 is quite similar to that observed in decile 10: in the late 50s about 40% of the members of this decile reported to have higher education, but in recent years this group includes more than 80% of the members of this percentile. The story for the group with high school education exhibits a sharp drop in its participation in the richest percentile, from about 40% in the first half of the survey to under 10% on average for the decade 1996-2006. FIGURE 9
PROPORTION OF INDIVIDUALS WITH TERTIARy EDUCATION
120 100 80 60 40 20 0 1957 1961 1966 1969 1972 1975 1978 1981 1984 1987 1990 1993 1996 1999 2002 2005
Total
Top 10
Top 1
183
FIGURE 10
PROPORTION OF INDIVIDUALS WITH SECONDARy EDUCATION
60 50 40 30 20 10 0 1957 1961 1966 1969 1972 1975 1978 1981 1984 1987 1990 1993 1996 1999 2002 2005
Total
Top 10
Top 1
Gender Regarding gender differences, the evidence shows that for individuals belonging to decile 10, the gap in percentage of male versus female members had no significant changes between 1957 and 1982, where about 10% of people FIGURE 11
PROPORTION OF MEN AND WOMEN IN DECILE 10
100 90 80 70 60 50 40 30 20 10 0 1957 1960 1963 1966 1969 1972 1975 1978 1981 1984 1987 1990 1993 1996 1999 2002 2005
Men
Women
184
receiving these high incomes were women. But from that year, this gap between the percentages of women versus men has been declining steadily, with women reaching a rate of over 30% of this decile. In the case of percentile 100, although this gap has been decreasing since 1982, the increase in the percentage of members of this income group who are women has grown much less than in decile 10. Instead of reaching over 30% of participation in this income group, the average over the last decade is close to 15%.
FIGURE 12
PROPORTION OF MEN AND WOMEN IN PERCENTIL 99-100
120 100 80 60 40 20 0 1957 1960 1963 1966 1969 1972 1975 1978 1981 1984 1987 1990 1993 1996 1999 2002 2005
Men
Women
Type of occupation As discussed in the two graphs of this subsection, there are three occupational groups in the survey which concentrate most of the composition of these high-income groups: employees, employers and self-employed, so we comment what data shows on the evolution of these three groups. According to the survey, one of the most clear trends is the increase, both in the top decile and the top percentile, of those who declare themselves employed, especially since the 1990s. In this decade, in case of decile 10, the relative progress of the group of employees corresponds to a slight decline of both the employers and self employed persons. For the top percentile variability exists in the 1990s, but it is possible to observe, however, a long-term tendency of self-employed workers to disappear from the top 1% of income. This decrease in the relative presence of self-employed persons is also observable inside the decile 10, but the fall is less dramatic. Finally, in the case of the category of employers, it represents about 20% of decile 10 since the mid 70s until 1990, and begin to fluctuate around approximately 13% by the end of the 2000s.
185
Less clear appears to be the history of employers in the top 1%, but it is still possible to note that its share amounted to less than 20% in several years after 1990, a phenomenon that occurs only once in history before 1990 and very mildly. Also in the period corresponding from mid-70s to 1989, it is common to find that over 40% of members were employers, a phenomenon that occurs much less frequently in the past 16 years of the sample. FIGURE 13
TyPES OF OCCUPATIONS IN DECILE 10
80 70 60 50 40 30 20 10 0 1957 1960 1963 1966 1969 1972 1975 1978 1981 1984 1987 1990 1993 1996 1999 2002 2002 2005 2005
Employers
Self-employed
Employees
FIGURE 14
TyPES OF OCCUPATIONS IN PERCENTILE 100
100 90 80 70 60 50 40 30 20 10 0 1957 1960 1963 1966 1969 1972 1975 1978 1981 1984 1987 1990 1993 1996 1999
Employers
Self-employed
Employees
186
Ratios The analysis of the relative differences between the highest incomes and middle-income population reveals that this particular measure of inequality itself has undergone changes since 1957 until today. Whether we look at personal income and per capita household income or total family income, changes over time are basically the same: relatively low levels of this ratio, and with little variation until 1967, then we have a moderate increase until 1970 where the distance between these high-income and middle income falls visibly during the Popular Unity government, to begin to grow steadily since 1974, with further and notorious increase after the crisis of 1982, reaching its peak around 1987, where it begins to fall to achieve some stability in the nineties and early 2000s around levels slightly lower than those of the 1980s but still higher than those observed at the beginning of the sample. If we look specifically at individual incomes, we see that at the beginning of the sample, the average of the top 1% percentile income was about 13 times the median income. Then it goes down to levels close to 8 times the size of the median income for the first three years of the 70s and is still depressed for the first year of the dictatorship (1974), only to increase up to 25 times the median income in 1981 and 35 times in 1987. The nineties moderated a bit this distance, with ratios around 19 and a further apparent decline towards the end of the sample with ratios closer to 14. However, after 1979 levels seem to fluctuate around levels clearly greater than those of 1957-1979.
FIGURE 15
RATIOS DECILE 10 AND PERCENTIL 100 ON MEDIAN INCOME
40 35 30 25 20 15 10 5 0 1957 1959 1961 1963 1965 1967 1969 1971 1973 1975 1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 Decile 10 to median Percentil 100 to median
187
5. Mobility Analysis at the Top of the Income Distribution Transition matrices Table 2 shows transitions between 1996 and 2001. From all households belonging to decile 10 in the year 1996, 48.4% of them remained in 2001s decile 10. And of the households that were not in decile 10 in the year 1996, 6.4% arrived to decile 10 in 2001. Transition probabilities for the years 20012006 are not statistically different than those of 1996-2001. This speaks of little or no change in mobility for this decile. Retention and arrival rates are the same for both periods. Additionally, we noted that retention rate in decile 10 is significantly higher than in the rest of the distribution. In decile 1, for example, between 1996 and 2001, 32.1% remained there, and between 2001 and 2006 the retention rate was 29.7%. TABLE 2
TRANSITION MATRICES BETWEEN 1996 AND 2001
Transition matrices 2001 1996 Decile 1-9 Decile 10 Decile 1-9 93.6% 51.6% 2006 2001 Decile 1-9 Decile 10 Decile 1-9 93.7% 52.2% Decile 10 6.3% 47.8% Decile 10 6.4% 48.4%
Source: Panel Casen 1996-2001-2006, Household total income.
Regression analysis Regression analysis aims to identify socio-economic variables that are correlated with the probabilities of stay and arrival in decile 10. To perform this analysis, we use longitudinal data from CASEN Panel Survey 1996-2006, which is the longest survey panel for Chile. Using data for 1996, 2001 and 2006, we estimate transition matrices for two periods. In addition, two discrete dependent variable models are estimated, identifying variables which correlate with the likelihood of remaining in decile 10 and of arriving to it from somewhere below in the income distribution. To study retention probabilities, we construct a discrete variable equal to 1 if the household is observed in decile 10 in year t conditional on being in decile 10 in t1, and 0 if not found in decile 10 in the year t conditional on being in decile
188
10 in year t1. This variable is modeled as a function of household characteristics and environment for year t1, and changes between t1 and t. Explanatory variables for both models include household composition, physical capital, human capital, working capital, home environment and shocks. Among shocks, we include health problems, reduction and increasing numbers of people at home. Table 3, column (1) shows the results for the probability of staying in decile 10. We see that household composition considering children 5 years or less has positive effect. Education plays an important role in remaining in decile 10, and so the education of the household head has a positive effect, like that of a spouse if such spouse has college or graduate studies. The set of workrelated variables reveals that people who have a permanent contract are more likely to remain in the decile 10. On the other hand, individuals who have other occupations, in addition to their main one, appear with a negative effect. The proportion of people working at home also plays an important role, if there is a greater number of people working, and then the associated staying probability in decile 10 is also greater. When a family faces a shock in household composition, i.e. there is an increase or decrease in the number of members, this also affects staying probabilities. In particular, if the number of member decreases, the staying probability rises, and when the number of members increases, the staying probability falls. In the case where a household member enters the labor market, this is shown to be associated with higher permanence probability of the household in decile 10. TABLE 3
PROBIT REGRESSION, REPORTING MARGINAL EFFECTS (1) Permanence Number of persons in the household Average age of the household Gender of the household head (Men=1) Biparental home Proportion of people<=5 aos Proportion of people>=6 & <=15 aos Proportion of people>=16 & <=65 aos Proportion of people>=66 aos Housing ownership years of Schooling of the head of the household 0.014 (0.011) 0.002 (0.002) 0.119 (0.033)** 0.008 (0.036) 0.382 (0.293) 0.652 (0.273)* 0.566 (0.284)* 0.510 (0.304) 0.004 (0.030) 0.028 (0.004)** (2) Arrival 0.007 (0.001)** 0.001 (0.000)** 0.007 (0.004)* 0.007 (0.003)* 0.062 (0.022)** 0.057 (0.021)** 0.022 (0.022) 0.028 (0.026) 0.014 (0.002)** 0.006 (0.000)**

(1) Permanence years of Schooling of the spouse Head of the household with University or Postgraduate Education Spouse with University or Postgraduate Education Employer=1 Self-employed=1 Working in a public firm Working in a private firm Permanent contract=1 Second occupation=1 Proportion of people working in the Household Region III Region VII Region VIII Metropolitan Region Urban Zone=1 Health Problems Number of people in the household decrease Number of people in the household increase Number of people working in the household decrease Number of people working in the household increase Observations 0.076 (0.043) 0.059 (0.042) 0.101 (0.070) 0.057 (0.029) 0.163 (0.032)** 0.259 (0.033)** 0.266 (0.031)** 0.050 (0.037) 2076 0.004 (0.004) 0.128 (0.041)** 0.194 (0.042)** 0.075 (0.058) 0.075 (0.046) 0.000 (0.056) 0.005 (0.040) 0.044 (0.034) 0.040 (0.072) 0.237 (0.058)** 0.191 (0.051)** (2) Arrival
189
0.003 (0.000)** 0.048 (0.013)** 0.034 (0.014)* 0.066 (0.024)** 0.008 (0.003)* 0.016 (0.008)* 0.004 (0.003) 0.016 (0.004)** 0.008 (0.008) 0.081 (0.007)**
0.018 (0.004)** 0.021 (0.004)** 0.001 (0.005) 0.025 (0.002)** 0.008 (0.003)* 0.043 (0.004)** 0.029 (0.002)** 0.019 (0.002)** 0.036 (0.004)** 23221
Notes: Standard errors in parentheses. * significant at 5%; ** significant at 1%. Dependent variable of permanence takes value 1 if the household is observed in decile 10 in year t conditional on being in decile 10 in t1, and 0 if it is not observed in decil 10 in year t conditional on being in decile 10 in year t1. The dependent variable is a discrete variable that takes the value 1 if the household is observed in decile 10 in year t conditional on not being in decile 10 in t1, and 0 if not found in decile 10 in year t conditional on not being in decile 10 in year t1.
190
Table 3, column (2) shows the results for the probability of arrival. With respect to the variables correlated with the probability of arrival at the decile 10, the education of the household head as well as the spouse, plays an important positive role, if education levels correspond to higher education or graduate studies. If the family has paid their housing, is positively correlated with the probability of arriving to the 10th decile. Just as in the case of permanence in the 10th decile, the proportion of people working at home also plays an important part in arriving at decile 10, i.e., if there is a greater number of people working at home then are more likely to ascend to the top decile. The increase in the number of household members is negatively correlated with the arrival in decile 10. If any household member leaves the labor market, i.e., become unemployed, it will have a negative effect on arriving to decile 10. In the other hand, if the proportion of people working increases, this will help significantly and positively to the household in its way to decile 10. 6. Conclusions
The study of the top incomes in Chile during 1957-2007 in Chile reveals changes in the shape of income distribution, as well as in the occupational composition, gender and educational status of this income group. Changes in the shape of the distribution can be seen in the evolution of the upper part of the distribution versus the median. The distance between the top decile and the richest percentile from the median has grown less permanently after 1975-1978. In terms of domestic policies that coincide with a large change in the Chilean economic model we can mention trade liberalization, financial liberalization, price liberalization, relative loss of power of unions among other changes. After 1990s there is a decrease of this distance in the final two years of the sample, but we should wait a little longer to confirm if it is relatively permanent. In addition, a significant change in this measure took place between 1970 and 1974, suggesting that this is an aspect of income distribution in Chile that it is sensitive to important changes in the economic model. In a shorter term perspective, we note that top incomes share seems to be countercyclical. This latter feature is similar to what international evidence has pointed out for developed countries. The composition of the highest income group has changed to incorporate a greater proportion of women in this group starting from 1982, although this effect is much lower in the richest percentile compared with the full top 10%. There is also a gradual and continuous fall in the fraction of people with secondary education to be found in the upper tail of the distribution, being replaced by people with higher education. In the top 10% of higher income has grown over time the relative importance of the group of employees and so the importance of salaries and wages for income decile 10. The top 1% of higher income increase is less noticeable and the category of independent incomes still retains a significant fraction relative to other sources. With respect to mobility in the top 10% of the income distribution, we detected no changes in our measurements in the decade 1996 to 2006. The
191
probabilities of arrival and departure of this decile are basically the same as in 1996-2001 as in 2001-2006. In each of these periods there is relative stability: high probability of remaining in the top decile and low probability of reaching this decile. A person who was part of this decile in 1996 had about 50% chance of continuing in this income group and 5 years after the same is true of someone who in 2001 belonged to this group. In contrast, the probability of arrival in that income group is close to 6%. To get a perspective, the probability of remaining in the bottom decile is approximately 30%. Among the variables most correlated with the probability of stay and arrival in the richest decile of the population are observed the possession of physical assets such as housing, graduate studies, the proportion of working household members and workers with permanent contracts. Future research should incorporate the analysis of other surveys such as CASEN, the Financial Survey and Survey of Social Protection. Also, we do not rule out the possibility of incorporating administrative data from tax. References Alvaredo, F., E. Saez (2006). Income and Wealth Concentration in Spain in a Historical and Fiscal Perspective, CEPR Discussion Paper N 5836, Centre for Economic Policy Research, London. Atkinson, A. (2002). Top Incomes in the United Kingdom over the Twentieth Century, Discussion Papers in Economic and Social History, N 43, University of Oxford. Atkinson, A., A. Leigh (2005). The Distribution of Top Incomes in New Zealand, Discussion Paper N 503, Centre for Economic Policy Research, Australian National University. Atkinson, A., A. Leigh (2007). The Distribution of Top Incomes in Australia, Economic Record, 83: 247-261. Atkinson, A., A. Leigh (2007). The Distribution of Top Incomes in Five Anglo-Saxon Countries Over the Twentieth Century, mimeo, Australian National University. Atkinson, A., T. Piketty (2007). Top Incomes over the Twentieth Century: A Contrast Between Continental European and English Speaking Countries, Oxford: Oxford University Press. Atkinson, A., W. Salverda (2005). Top Incomes in the Netherlands and the United Kingdom Over the 20th Century, Journal of the European Economic Association, 3: 883-913. Banerjee, A., T. Piketty (2005). Top Indian Incomes: 1922-2000, The World Bank Economic Review, 19: 1-20. Celhay, P., C. Sanhueza and J.R. Zubizarreta (2010). Intergenerational Mobility of Income and Schooling: Chile 1996-2006. Revista de Anlisis Econmico - Economic Analysis Review, Vol 25, N 2. Contreras, Cooper, Hermann and Neilson. Poverty Dynamics and Relative Income Mobility: Chile 1996 and 2001, University of Chile, Manuscript. Dell, F. (2007). Top Incomes in Germany Throughout the Twentieth Century: 1891-1998, in A. Atkinson and T. Piketty (eds.), Top Incomes over the
192
Twentieth Century: A Contrast Between Continental European and English Speaking Countries. Oxford: Oxford University Press, 365-425. Dell, F., T. Piketty, E. Saez (2007). Income and Wealth Concentration in Switzerland over the Twentieth Century, in A. Atkinson and T. Piketty (eds.), Top Incomes over the Twentieth Century: A Contrast Between Continental European and English Speaking Countries. Oxford: Oxford University Press, 472-500. Engel, E. and J. Eberhard (2008). The Educational Transition and Decreasing Wage Inequality in Chile. Unpublished Manuscript. yale University. Larraaga, O. (1999). Distribucin de Ingresos y Crecimiento Econmico en Chile, Serie Reformas Econmicas N 35, Cepal. Larraaga, O. and J.P. Valenzuela (2011). Estabilidad en la desigualdad. Chile 1990-2003, Revista Estudios de Economa, Vol. 38, N 1, junio. Leigh, A. (2009). Top Incomes, in W. Salverda, B. Nolan, and T. Smeeding eds., The Oxford Handbook of Economic Inequality. Frank, R. (2007). Falling Behind: How Rising Inequality Harms the Middle Class, Berkeley: University of California Press. Kopczuk, W., E. Saez, J. Song (2007). Uncovering the American Dream: Inequality and Mobilit in Social Security Earnings Data since 1937, NBER Working Paper N 13345. Moriguchi, C., E. Saez, 2008. The Evolution of Income Concentration in Japan, 1885-2002: Evidence From Income Tax Statistics, Review of Economics and Statistics, forthcoming. Nolan, B., 2007. Long-Term Trends in Top Income Shares in Ireland, in A. Atkinson and T. Piketty (eds.), Top Incomes over the Twentieth Century: A Contrast Between Continental European and English Speaking Countries. Oxford: Oxford University Press, 501-530. Nuez, J. and C. Risco (2004) Movilidad intergeneracional del ingreso en un pas en desarrollo: el caso de Chile. Documento de Trabajo, Departamento de Economa, Universidad de Chile. Pags, C., C. Montenegro (1999). Job Security and the Age Composition of Employment: Evidence from Chile, Research Department Working Paper 398. Washington, DC, United States: Inter-American Development Bank. Piketty, T. (2003). Income Inequality in France, 1901-1998, Journal of Political Economy, 111: 1004-1042. Piketty, T., E. Saez (2003). Income Inequality in the United States, 1913-1998, Quarterly Journal of Economics, 118:1-39. Piketty, T., E. Saez (2006). The Evolution of Top Incomes: A historical and International Perspective, American Economic Review, Papers and Proceedings, 96: 200-205. Riihel, M., R. Sullstrm, M. Tuomala (2005). Trends in Top Income Shares in Finland, The Government Institute for Economic Research (VATT), Discussion Papers 371. Ruiz-Tagle, J. (1998). Chile: 40 aos de desigualdad de ingresos, University of Chile, Manuscript. Saez, E. (2004). Income and Wealth Concentration in a Historical and International Perspective, mimeo, University of Berkeley.
193
Saez, E., M. Veall (2005). The Evolution of High Incomes in Northern America: Lessons from Canadian Evidence, The American Economic Review, 95: 831-849. Sapelli, C. (2011). A Cohort Analysis of the Income Distribution in Chile, Revista Estudios de Economa, Vol. 38, N 1, junio. Tawney, R. (1913). Poverty as an Industrial Problem, London School of Economics: London. Valenzuela, J.P., S. Duryea (2011). Examining the prominent position of Chile in the world in terms of income inequality: Regional comparisons, Revista Estudios de Economa, Vol. 38, N 1, junio 2011.
Estudios de Economa. Vol. 38 - N 1, Junio / Javier Nez, Leslie Intergenerational income and educational 2011. Pgs. 195-221 Miranda
195
Intergenerational income and educational mobility in urban Chile*

Movilidad intergeneracional del ingreso y la educacin en zonas urbanas de Chile
Javier Nez** Leslie Miranda*** Abstract This paper provides evidence on the degree and patterns of intergenerational income and educational mobility in urban Chile. We find intergenerational income elasticities for Greater Santiago in Chile in the range of 0.52 to 0.54. This is lower than recent nation-wide elasticities for Chile of about 0.6-0.7, but still stands as fairly high in comparison with the comparable international evidence. We also find that intergenerational educational mobility is lower for the younger cohorts, which however does not necessarily imply an increase of intergenerational educational mobility in the last decades, as life-cycle effects may be at work. Finally, we find evidence of a higher degree of intergenerational persistence of income at the two extremes of the income distribution, which is more accentuated at the top centiles of the distribution. We suggest that this may mirror the unusually high concentration of income at the top of the income distribution in Chile, a hypothesis that requires further research. Key words: Intergenerational mobility, Schooling, Mobility patterns. Resumen Este paper proporciona evidencia sobre el grado y los patrones de movilidad intergeneracional del ingreso y la educacin en zonas urbanas de Chile. Encontramos elasticidades intergeneracionales del ingreso en el Gran Santiago
We thank David Bravo, Dante Contreras, Marcelo Fuenzalida, Osvaldo Larraaga, Marcela Perticara, Cristina Risco, Claudia Sanhueza, Claudio Sapelli, Andrea Tartakowsky and Florencia Torche for previous discussions, comments and insights. All posible errors are the authors responsibility. The authors thank the support of the Interamerican Development Bank. Javier Nez thanks the support of Proyecto Anillos SOC12 funded by Conicyt, Chile, as well as the support of the Centre for Latin American Studies at Cambridge University. Leslie Miranda thanks the support of Centro de Microdatos, Iniciativa Cientfica Milenio (Project P075-023-F). ** Department of Economics, Universidad de Chile. *** Centro de Microdatos, Universidad de Chile.
196
del orden de 0.52-0.54, las que son menores que valores reportados para Chile en el rango de 0.6-0.7, pero son an elevadas en relacin a la evidencia internacional. Tambin encontramos mayor movilidad en los cohortes ms jvenes, lo que no debe interpretarse necesariamente como aumentos en la movilidad del ingreso, pues efectos de ciclo de vida pueden estar presentes. Finalmente, encontramos evidencia de mayor persistencia intergeneracional del ingreso en los dos extremos de la distribucin, la cual es particularmente acentuada en los centiles de ingreso ms altos de la poblacin. Sugerimos que este hallazgo puede reflejar la gran concentracin de ingresos en los centiles altos que caracteriza a la distribucin del ingreso en Chile, hiptesis que amerita investigacin a futuro. Palabras clave: Movilidad intergeneracional, Escolaridad, Patrones de movilidad. JEL Classification: D3, I2, J6.
1. Introduction Much of the literature on inequality in Chile has focused on the inequality of outcomes, typically the distribution of incomes, and little is still known about the countrys levels of inequality of opportunity, its assessment in comparative perspective and its evolution in recent times. A common approach to empirically assess a countrys degree of equality of opportunity is the notion of intergenerational economic mobility: a higher level of equality of opportunity is expected to decrease the effect of an individuals early socioeconomic background on his economic achievement in adulthood, yielding therefore a higher level of intergenerational economic mobility. This paper examines the degree of intergenerational income and educational mobility in urban Chile in comparative perspective, and some of its salient patterns. Since the research in intergenerational income mobility is fairly recent and limited in Chile, this work also proposes and discusses some exploratory hypothesis and some avenues for future research. The study of intergenerational income mobility requires income data for pairs of fathers (or parents) and their offspring. As it is common in most of the developing world, such data is often limited or unavailable. In this work we follow the methodology often known as Two-Stage Two-Sample Least Squares (TSTSLS) developed originally by Bjrklud and Jntti (1997) and widely applied, whereby fathers incomes are predicted from data on income determinants such as fathers schooling and education provided by their sons. We apply this methodology to Greater Santiago, Chiles main urban center, and analyze its results in comparative perspective. We find a lower intergenerational mobility in Greater Santiago compared with estimates using nationwide data, which may indicate that rural and small urban areas may exhibit lower educational and occupational opportunities than a large urban area such as Greater Santiago, an issue open for future research. Yet, the results for Greater Santiago are still relatively high compared to international evidence. This paper also explores
Intergenerational income and educational / Javier Nez, Leslie Miranda
197
how intergenerational income mobility varies along parents relative position in the income distribution in urban Chile. This paper is structured as follows. The next section motivates the paper by providing evidence on the conceptual and empirical distinctions between the notions of inequality of income vs. inequality of opportunity, and by discussing the need to increase the understanding of inequality of opportunity, a task for which the notion of intergenerational mobility is well suited. Section 3 presents the theoretical framework and the empirical strategy followed. Section 4 describes the dataset employed, and Section 5 presents and discusses the main results. Finally Section 6 concludes. 2. Inequality of Outcomes vs. inequality of Opportunity in Chile There has been a long debate about whether inequality, and the policies designed to deal with it, must tackle the inequality of outcomes or alternatively the inequality of opportunities. Advocates of the latter stress that inequality of outcomes (typically incomes) depend not only on the circumstances that are beyond the control of individuals, such as parental background, but also on factors that are (presumably) under their control, such as effort and choices. Accordingly, some authors have suggested that from a moral standpoint, public policies should focus on equalizing opportunities instead of outcomes (incomes)1. In a seminal contribution, Bourguignon, Ferreira and Menndez (2005, 2007) have developed a methodology to attempt to measure the proportion of the observed income distribution that is associated with inequality of (uncontrolled) circumstances of origin such as parental education and occupation and race in Brazil. The methodology distinguishes the direct effect that circumstances have on earnings in adulthood (the partial effect) from the indirect effect of circumstances on earnings through the accumulation of human capital (schooling). They found that the Gini coefficient is reduced in up to about 10 percentage points (about 20 per cent) after equalizing the set of circumstances mentioned earlier. Table 1 shows the effect on the Gini coefficient for Greater Santiago and Chile of equalizing across individuals various circumstances of origin including parental schooling, size and composition of the household during infancy (single versus biparental), the age of the household head, and a measure of parents job vulnerability. Even though many relevant circumstances may certainly remain unobserved, these results do suggest that the important circumstances mentioned above play an important yet limited role in determining the income distribution, results that are similar to those reported for Brazil. These findings suggest that income distribution indicators seem only partially associated to a societys degree of equality of opportunity, and that income distribution may be also affected by other factor presumably unrelated to exogenous circumstances. In this context, a closer and perhaps more direct way of assessing a countrys degree
For a discussion of these issues, see for example Bourguignon, Ferreira and Menndez (2007).
198
of equality of opportunity is to examine its level of intergenerational income and educational mobility, issue that we address next. TABLE 1
EFFECT OF EqUALIzING OBSERVED CIRCUMSTANCES OF ORIGIN ON GINI COEFFICIENT, ChILE, MEN AGED 23-65
Greater Santiago employment survey, 2004 Gini Coefficient 0.503
Nationwide Casen survey 2006 0.535
Gini after equalizing observed circumstances of origin Partial Effect Total Effect 0.433 0.420 0.491 0.455
Source: Nez and Tartakowsky (2007, 2009).
3. Theoretical Framework and Empirical Strategy This paper motivates and illustrates the intergenerational transmission of the socioeconomic status and the related concept of intergenerational income mobility by means of a simplified version of the model suggested by Becker and Tomes (1979). This model assumes that a family only consists of one individual at each generation. Consider two generations within a given family, that is, father and child. Individual permanent income Y is assumed to derive from two components: the individual endowment of human capital and individual ability denoted by A. Becker and Tomes (1979) assume that a childs endowment of human capital is a result of his fathers optimal allocation of his permanent income, where the fathers utility depends of his own consumption and the childs permanent income. This framework yields the following relationship between the fathers and his childs permanent incomes: (1) Y child = Y father + Achild
Equation (1) indicates that the fathers permanent income has a positive causal influence on the childs income captured by parameter . Equation (1) would also imply a second source of correlation of the fathers and the childs income if the childs ability is correlated (as can be expected) with the fathers ability. Parameter can be interpreted as a causal effect of the previous generation on the next, which is independent of the fathers investment decisions and budget constraints. Thus, this parameter encompasses all aspects of earnings determinants that money cannot buy, such as innate cognitive abilities, preferences or access to social networks, among others. It is important to note that a regression of the childs income on fathers income would capture both transmission mechanisms. hence, a standard OLS
199
estimate of the intergenerational earnings transmission coefficient would provide an upward biased estimate of the pure causal effect of parental income on childs income. In this work, we do not attempt to separate both effects. Instead, and in line with the related literature, we are interested in the estimation of a reduced-form intergenerational earnings regression to measure the degree of intergenerational income mobility. From the previous framework, if long-rung economic status were directly observed, the following log-linear relationship between the permanent incomes of father and child can be estimated by OLS: (2) Yichild = 0 + 1Yi father + i
where Yichild denotes the log of childs permanent income in family i and Yi father the log of his fathers permanent income, and i is an error term independent of Yi father usually assumed to be distributed as N(0,2). Our parameter of interest 1 represents the elasticity of a childs long-run income with respect to his fathers long-run income. There are two extreme cases. First, 1 = 0 would depict a situation involving full intergenerational mobility, as the permanent income of the child in adulthood would show no statistical association whatsoever with the fathers permanent income. At the other extreme, 1 = 1 would indicate a situation of very low mobility, whereby a child born from a parent with an income, say, x per cent above the mean of parents incomes will have an expected income located x per cent above the mean of his own generation. Alternatively, 11 can be interpreted as a regression-to-the-mean effect: If 1 = 0, for example, the regression-to-the-mean effect is maximum, as sons will have the same expected value of income of adulthood regardless of their respective fathers relative socioeconomic status. however, long-run incomes are not directly observed. Instead, most data sets only provide measures of current incomes or earnings. Solon (1992) and zimmerman (1992) have shown that the use of incomes in a single year can underestimate the true intergenerational transmission coefficient due to the presence of transitory components in current income, especially in combination with the use of a homogeneous sample. An alternative to address this bias involves using panel data on fathers income to obtain an average of fathers current income over several years as a proxy for their permanent income. Solon (1992) shows that the inconsistency of the transmission coefficient diminishes with the number of years over which incomes are averaged. Another problem emerges when, as in this paper, income information of father-child pairs is not available. In this context, a solution proposed by Arellano and Meghir (1992) and Angrist and Krueger (1992) and followed by Bjrklud and Jntti (1997) for intergenerational mobility studies is to use information from two separate samples: first, earnings equations can be estimated using an older sample of men in order to obtain coefficients of some key earnings determinants, such as schooling, experience and occupation, for example. Then, these coefficients can be employed to predict the income of the fathers of a
200
sample of sons who have reported the required information about their fathers. This technique is often known as two-sample instrumental variables estimation (TSIV) or two-sample two-stage least squares (TSTSLS)2. More formally, assume that the log of the fathers and sons current income at date t can be written as: (3) (4) Yitfather = Yi father + itfather
child Yitchild = Yichild + it
father child where it and it incorporates transitory fluctuations in the father and childs current income and measurement errors. Let Zifather denote a set of sociodemographic characteristics (like education, occupation, among others) of fathers from a sample of families i I and assume that Yitfather can be written as:
(5) Yitfather = Zifather + vifather + itfather
where vifather is independent of Zifather . Term Yitfather is not observed in sample I. Yet, if there exists a separate sample J from the same population as I, it can be used to provide an estimate of , namely , which would be derived from estimation of equation (6) using the sample of adult men J: (6) Y jt = Z j + v j + jt
with j J. An OLS estimation of (6) would provide predictions of the fathers earnings in sample I: Yitfather = Zifather . This prediction can in turn be used to estimate the intergenerational income elasticity coefficient 1 since equations (2), (3), (4) and (6) imply: (7) where Yitchild = 0 + 1 Zifather + it
child it = i + it + 1vifather + 1 Zifather ( ) .
In this work, the estimates of 1 are based on the estimation of equations (6) and (7) on separate samples as we describe in the following section. In particular, in a first stage we estimate a Mincer version of equation (6) that allows for different schooling returns by educational level3:
2 3
See for example Dunn (2004). In Chile, elementary education consists of the first eight years and secondary school consists of four additional years.

father Y js = 0 + 1S js + 2 d1 ( S js 8) + 3 d 2 ( S js 12) 2 + 4 Experjs + 5 Experjs + js
201
(8)
where Sjs represents the years of schooling of father in year s4, Experjs stands for fathers potential experience5 and js is a random error term. In addition, dummy variables for educational levels are defined as: 1 if S > 8 d1 = 0 otherwise 1 if S > 12 d2 = 0 otherwise
In another specification we also include information of fathers occupation under the assumption that occupation is a good instrument to estimate the fathers permanent income. This information comes from a second survey undertaken in June 2006 to a sub-sample of the June 2004 version of Employment and Unemployment Survey of Greater Santiago. In a second stage, we use the estimated parameters in (8) and fathers information reported by the sons to predict the fathers income, as follows: (9) father = + S + d ( S 8) + d ( S 12) Y is 0 1 is 2 1 is 3 2 is + Exper + Exper 2
4 is 5 is
hence, we obtain the intergenerational income elasticity 1 from: (10) Yitchild = 0 + 1Y is

father 2 + 2 ageit + 3 ageit + it
where ageit stands for childs age and controls for life-cycle profiles in childs earnings. The methodology described above is subject to some well-known biases that have been identified in the related literature, which are worth pointing out. As shown in Solon (1992, 2002), a first bias may arise if the fathers schooling and occupation, apart from being correlated with the fathers earnings, are also positive predictors of the sons earnings in their own right. Thus, in the second-stage regression, where schooling and occupation are used to predict the fathers earnings but are not included as separate explanatory variables of the sons earnings, the resulting omitted-variable problem would yield an upward bias in the intergenerational income elasticity.
4 5
The s year corresponds to time when father were taken investment decisions on his childs human capital. Potential experience is defined as: age minus years of schooling minus 6.
202
Another source of bias is related to the ages of sons being considered for estimating equation (7). In particular, various studies have found that the estimated intergenerational elasticities increase substantially as sons earnings are observed further on in their careers. Accordingly, studies that use earnings data of sons in the early stages of their life-cycle tend to underestimate the intergenerational income elasticity. This arises if the measurement error in the sons early earnings is negatively correlated with the long-run income, as can be expected.6 Due to the existence of these potential biases, in this paper we compare the results obtained for urban Chile with the results of international studies that follow a similar methodology, and are accordingly subject to the same kind of biases. We also report the age brackets of sons ages employed in all studies. 4. Data Our dataset comes from the Employment and Unemployment Survey for the Greater Santiago conducted annually by Universidad de Chile since 1957, which is applied to approximately 4,000 households. Greater Santiago is the largest urban centre in Chile, home to approximately 40 per cent of the countrys population. In order to avoid selectivity issues associated with female participation in the labor market, we focus only on fathers-sons intergenerational income mobility. The analysis of intergenerational income mobility involving mothers and daughters remains as future research. however, we consider sons as well as daughters when we examine intergenerational educational mobility in section 5.2. The Greater Santiago Employment and Unemployment Survey provides information on gender, age, educational level, employment status, occupational position, economic sectors and monthly income from wages, salaries and selfemployment. All this information is employed to estimate the coefficients of Mincer equations that are employed to predict the unobserved income of fathers. Mincer equations like (8) were estimated for the male labor force in 15-65 age range with positive income and working at least 30 hours per week. Our sample of sons comes from the June 2004 version of the survey. In this year, additional to demographic and economic data, respondents were asked to provide information about educational and individual characteristics of their parents. We consider sons in the 23-65 age range to control for potential selectivity problems in individuals outside this age range. We eliminate unemployed and inactive individuals, those with no positive incomes and those missing parental information. Our sample was composed by 649 pairs of fathers and sons in the corresponding age range. Fathers predicted incomes were estimated dividing the sons sample in three sub-samples by age groups: 23-34, 35-44, 45-54 and 55-65 years old. We select the corresponding fathers samples by assuming that the fathers investment decisions in his sons human capital, which are expected to be a major source of socioeconomic transmission across generations, were taken approximately when the child was between 6 and 18 years old. These years correspond to the 1987,
6
See for example Solon (2002), haider and Solon (2006), Grawe (2006) and Dunn (2007).
203
1977, 1967 and 1958 versions of the Employment and Unemployment Survey for the 23-34, 35-44, 45-54 and 55-65 cohorts, respectively. Those are periods of relative economic stability; hence, estimated coefficients of Mincer equations for mentioned years are similar to those obtained from adjacent years. A second data source comes from a new survey realized in June 2006 to a sub-sample of males previously surveyed in the Employment and Unemployment Survey of June 2004. In this new survey, individuals were asked to provide additional information about specific occupation and other individual characteristics of their parents, as well as diverse family background information corresponding to period when they were about 15 years old. We use the information about the fathers occupations7 to estimate a second Mincer equation specification under the assumption that occupation is a good instrument to estimate the fathers permanent income, in addition to fathers schooling. 5. Results 5.1. Estimates of intergenerational income elasticity for urban Chile Table 2 reports intergenerational regression coefficients for labor incomes8. Results are reported for the full sample in the 23-65 age range. Estimates in this table are obtained using fathers education, potential experience and occupation as predictors of fathers income, as described earlier. First-step income regressions are provided in Tables A.1 and A.2 in the appendix. TABLE 2
ESTIMATES OF INTERGENERATIONAL LABOR INCOME ELASTICITY BY TWO-SAMPLE TWO-STAGE LEAST SqUARES (TSTSLS) Greater Santiago, Chile
Fathers income predicted from: Cohort 24-65 Schooling and Experience 0.54 Schooling, Experience and Occupation 0.52
Table 2 indicates that fathers predicted log income has a significant positive effect on their sons log income. For the whole sample, the estimated elasticity is around 0.52-0.54 depending on the predictors employed9. Table 3 reports the results of other intergenerational income mobility studies for urban Chile. It is interesting to note that employing only the SIALS10 data
7 8 9 10
The 5-level occupational categories are: employer (1); self-employed (2); employee (3); blue-collar worker (4) (reference) and domestic (household) workers (5). The estimates using personal incomes yield the same global elasticity. This number is a weighted average of elasticities of each age group. Second International Adult Literacy Survey.
TABLE 3
204
TSTSLS INTERGENERATIONAL INCOME MOBILITY IN ChILE, USING URBAN AND NATIONWIDE DATA
Study
Database
Population
Key fathers income predictors Sons ages Potential experience, schooling Schooling 23-55
Elasticity
Nez and Risco (2004)
Employment and Unemployment Survey (U. of Chile SIALS Greater Santiago Urban Population Nationwide Data Nationwide Data Greater Santiago Greater Santiago Schooling
Greater Santiago
0.55
Contreras, Fuenzalida and Nez (2006) SIALS
23-55
0.58
Contreras, Fuenzalida and Nez (2006) CASEN
23-55
0.67
Nez and Miranda (2010)
Schooling
25-40
0.74
Nez and Miranda (2010)
CASEN
Schooling and Occupation Schooling
25-40
0.57
This study
Employment/Unemployment Survey (U. of Chile)
23-65
0.54
This study
Employment/Unemployment Survey (U. of Chile)
Schooling and Occupation
23-65
0.52
205
for the Metropolitan Region (similar but slightly larger than Greater Santiago) yields fairly similar results than the Greater Santiago studies although lower than the 0.67 elasticity obtained from the national urban SIALS database, and lower as well than the nationwide CASEN studies, which yield elasticities in the 0.6-0.7 range. This evidence suggests a lower intergenerational income elasticity for the large, more prosperous Greater Santiago urban area in comparison with the rest of the country.11 This may be an indication of a lower intergenerational mobility in rural, semi-rural and small urban areas, where educational and occupational opportunities might be lower than in a large urban centre like Greater Santiago. Table 4 summarizes some selected international evidence on (nationwide) intergenerational income mobility. As can be seen, urban Chile presents relatively low intergenerational income mobility in comparison with other developing and developed countries. This could be even more so for nationwide mobility in Chile if the previous discussion on the lower elasticity values for urban areas is considered. Levels of intergenerational mobility in Chile seem somewhat
TABLE 4
INTERNATIONAL EVIDENCE ON INTERGENERATIONAL INCOME ELASTICITY Method OLS 0.53 0.23 0.19-0.22 0.54 0.13 0.36-0.43 0.34 0.48 0.39-0.59 0.29-0.39 0.45-0.53 0.52 0.28 0.44 0.46 IV-TSTSLS 0.2-0.3 0.69 0.58
Country Australia Brazil Brazil Canada Canada Malaysia Finland France Germany Italy United Kingdom United States United States United States Sweden Nepal Pakistan
Study Leigh (2007) Dunn (2004) Ferreira and Veloso (2004) Corak and heisz (1999) Fortin and Lefebvre (1998) Grawe (2001) Osterbacka (2001) Lefrane and Trannoy (2004) Wiegand (1997) Piraino (2007) Dearden, Machin, Reed (1997) Solon (1992) Solon (1992) Bjrklud and Jntii (1997) Bjrklud and Jntii (1997) Grawe (2001) Grawe (2001)
Sons ages 25-54 25-34 25-64 29-32 17-59 25-45 30-40 27-33 30-45 33 25-33 25-33 28-36 29-38
11
This seems consistent with of evidence on the pattern of intergenerational income mobility in Brazil, according to Ferreira and Veloso (2004): The more prosperous and more urban Brazilian Southwest has an intergenerational income elasticity of 0.54, which is lower than the rest of the country, and much lower than the poorer Northeast region (0.73).
206
similar to Brazils12. Some authors have suggested and provided theoretical arguments and some evidence in favor of an overall positive relationship between cross-sectional income inequality and intergenerational persistence of inequality13. From this perspective, the evidence for Chile would be consistent with this hypothesis, considering that Chile has a particularly unequal distribution of income in compassion with the international evidence14. 5.2. Intergenerational income and educational mobility across cohorts Table 5 presents the intergenerational income elasticities by cohort employing the fathers predicted income from schooling and experience only. As can be seen, the elasticity coefficient is monotonically decreasing for the three younger cohorts. A possible explanation of this phenomenon is that intergenerational mobility could have increased in the last decades. however, as mentioned earlier, this pattern can also be the result of life-cycle effects on earnings, and therefore whether income mobility has increased in time cannot be concluded15. however, a way round this problem is that unlike earnings, schooling levels usually exhibit little or negligible life-cycle effects after the time when most individuals stop studying and enter the labor market, usually around the mid twenties, after which schooling levels remain largely fixed for most individuals (as we shall see below). Following this reasoning we attempt to examine whether intergenerational educational mobility has exhibited changes in the recent decades by comparing educational mobility values across cohorts. TABLE 5
ESTIMATES OF ThE COEFFICIENT OF INTERGENERATIONAL INCOME MOBILITY
Cohort 23-34 35-44 45-54 55-65 All sample
Personal Income 0.46 0.54 0.63 0.59 0.54
Labor Income 0.46 0.52 0.65 0.58 0.54
In order to substantiate this claim we explore the accumulation of schooling in Chile by age groups using the 1996-2001 CASEN Panel data. Figure 1 shows the average individual accumulation of schooling in years in Chile by age groups between 1996 and 2001. Figure 1 shows that while schooling accumulation is important before age 20, it decreases thereafter, and in fact, it
12 13 14 15
In this context, for comparison purposes it would be useful to obtain intergenerational incomes elasticities for large urban areas in Brazil. See for example Dunn (2004) and Andrews and Leigh (2008). See for example De Ferranti, D., Perry, G., Ferreira F., and M. Walton (2003). See Dunn (2004) for a discussion on the role of life-cycle effects on intergenerational income mobility elasticities.
207
FIGURE 1
1996-2001 AVERAGE INDIVIDUAL DIFFERENTIAL IN YEARS OF SChOOLING BY AGE Average differential in years of schooling 5.0 4.0 3.0 2.0 1.0 0.0 -1.0 5-9 10-14 15-19 20-24 25-29 30-34 35-39 40-44 45-49 50-54 Age Cohort Source: Panel CASEN 1996-2001. > 54
TABLE 6
SChOOLING ELASTICITY BY COhORT
Cohort 23-34 35-44 45-54 55-65 All sample
Sons and Daughters 0.15 (7.48)** 0.15 (6.48)** 0.29 (9.75)** 0.37 (7.97)** 0.21
Sons 0.15 (7.70)** 0.15 (4.12)** 0.24 (6.59)** 0.41 (5.34)** 0.21
Daughters 0.14 (4.14)** 0.15 (5.55)** 0.37 (7.57)** 0.32 (6.82)** 0.23
Robust t statistics in parentheses.
* significant at 5%; ** significant at 1%.
remains negligible after age 25 for the population as a whole. This suggests that life-cycle effects in years of schooling are unimportant for the age group 23-65 considered in this study. Therefore, finding evidence of higher intergenerational educational mobility for the younger cohorts would be suggestive of an increase of educational mobility in time. Table 6 presents intergenerational schooling elasticities by cohorts. The evidence indicates lower values for the younger cohorts. Based on the argument presented above, this would be suggestive of an overall increase in intergenerational educational mobility in the last decades in Greater Santiago, for both men and women.
208
TABLE 7
COhORT EFFECTS IN YEARS OF SChOOLING. DEPENDENT VARIABLE: SONS AND DAUGhTERS SChOOLING
Variables Fathers schooling Fathers schooling*Cohort Cohort Constant Observations Adj. R-squared
All 17.183 (5.21)** 0.009 (5.09)** 0.119 (7.10)** 223.716 (6.83)** 1197 0.33
Sons 14.440 (3.06)** 0.007 (2.98)** 0.022 (7.00)** 40.527 (6.65)** 649 0.29
Daughters 19.945 (4.29)** 0.010 (4.21)** 0.089 (3.55)** 165.007 (3.36)** 548 0.38
Note: Cohort is defined as offsprings year of birth. Robust t statistics in parentheses. * significant at 5%; ** significant at 1%.
TABLE 8
COhORT EFFECTS IN INTERGENERATIONAL SChOOLING ELASTICITY Dependent variable: Sons log schooling
Variables Fathers log schooling Fathers log schooling*Cohort Cohort Constant Observations Adj. R-squared
All 15.156 (5.67)** 0.008 (5.60)** 0.022 (4.11)** 40.567 (3.90)** 1197 0.32
Sons 16.548 (3.65)** 0.008 (3.62)** 0.153 (7.11)** 292.412 (6.91)** 649 0.29
Daughters 13.554 (4.66)** 0.007 (4.59)** 0.022 (6.68)** 41.082 (6.35)** 548 0.35
Note: Cohort is defined as offsprings year of birth. Robust t statistics in parentheses. * significant at 5%; ** significant at 1%.
This is further confirmed in Tables 7 and 8. While these tables indicate a positive association between fathers and sons schooling levels, as well as strong evidence of an expansion in schooling in time, particularly for daughters (as indicated by the coefficients of the cohort variable), they also provide evidence of a lower association between fathers and sons schooling for the younger cohorts, as indicated by the coefficients of the interactive term in both specifications. To what extent this evidence of lower statistical association between fathers and sons education in the younger cohorts is related or causes the lower intergenerational income elasticities of the younger cohorts reported earlier
209
in Table 5 remains open for future research. In particular, it seems plausible that factors associated with socioeconomic background such as the quality of education, social capital and access to social networks, or the existence of classdiscrimination in the labor market, for example, may yield different returns to schooling for individuals according to their social background16. Likewise, these factors may limit the transformation of the higher degrees of educational mobility of the younger cohorts reported above into higher levels of intergenerational income mobility. This hypothesis, however, remains open for future research. 5.3. Patterns of intergenerational income mobility in urban Chile We finally examine some patterns of intergenerational income mobility in urban Chile. In particular, in this section we study whether the degree of the intergenerational transmission of the socioeconomic status varies across different population segments of the income distribution. Table 9 reports an estimates of quintile transition matrix for labor income using fathers education and potential experience as predictors of fathers income, and deriving fathers quintiles from the real corresponding income distribution. As a robustness check, Table 10 reports an equivalent transition matrix in which fathers quintiles are obtained from the distribution of the fathers predicted incomes instead.17 Yet, both transition matrices yield similar patterns. In both cases the bottom-to-bottom and the topto-top transition probabilities are large (37-30 and 47-57, respectively), a pattern that is, in fact, also observed for other countries. In addition, the probabilities of transiting from the lowest to the highest quintiles and vice versa are fairly low, around 0 to 8 per cent, also consistent with the international evidence. TABLE 9
qUINTILE TRANSITION MATRIx Fathers quintiles obtained from real income distribution
Father
Son
Bottom 2nd 3rd 4th Top 0.00 0.07 0.11 0.23 0.47
Bottom 2nd 3rd 4th Top Immobility Index
0.37 0.23 0.21 0.15 0.08 0.30
0.35 0.29 0.31 0.31 0.14
0.14 0.15 0.14 0.10 0.07
0.14 0.26 0.24 0.20 0.24
16
17
For example Nez and Gutirrez (2004) report a 50 per cent gap in earnings of Chilean professionals from different socioeconomic background of origin after controlling for academic performance, experience, schools academic performance, postgraduate studies, command of a second language, among other controls. quartile transition matrices are provided in Tables A3 and A4 in the appendix, also illustrated in Figures A1 and A2.
210
TABLE 10
qUINTILE TRANSITION MATRIx Fathers quintile obtained from distribution of fathers predicted incomes
Father Bottom 2nd 3rd 4th Top Immobility Index
Son Bottom 0.30 0.17 0.21 0.12 0.06 0.30 2nd 0.34 0.29 0.33 0.18 0.13 3rd 0.10 0.22 0.10 0.11 0.04 4th 0.22 0.22 0.24 0.26 0.20
Top 0.04 0.10 0.12 0.33 0.57
The transition matrixes in Tables 9 and 10 suggest two important disparities in the intergenerational income transmission mechanism. First, they show more income persistence at both extremes of the fathers income distribution, together with a substantial level of intergenerational mobility in the intermediate fathers quintiles. Second, they suggest an asymmetry in the degree of intergenerational persistence at the two extremes of the fathers income distribution, in particular a higher degree of persistence at the top quintile versus the bottom quintile. In order to explore these issues further, regression equations of the fathers centiles versus sons centiles are reported in Table 11. Table 11 shows that the association between fathers and their sons income centiles is increasing (as expected) but not linearly. In fact, specifications 2 and 6 of Table 11 show that when a quadratic functional form is imposed it yields a robust overall convex pattern, indicating that the intergenerational income persistence is asymmetric, being higher in the upper part of the fathers income distribution. Yet, the cubic specifications in 3, 4, 7 and 8 of Table 11 outperform the quadratic models, indicating that intergenerational persistence is higher at the two extremes of the income distribution than in the intermediate segments of the fathers income distribution, consistent with the transition matrixes 18. Table 12 also indicates by means of a structural change test that the slope of fathers vs. sons centiles is statistically steeper at the top than at the bottom of the fathers income distribution. This pattern is depicted in Figure 2, which shows the profile of cubic OLS and quantile regressions of fathers vs. sons income centiles. But Figure 2 also illustrates how broad is the spectrum of income centiles where sons are likely to end up in adulthood, conditional on their fathers income centiles. Note that while for the most part of the fathers income distribution the sons can end up as adults in income centiles often quite different from their parents, at the top of the fathers income distribution chances are that sons will occupy relative income positions similar to those of their fathers.
18
A similar pattern is observed when the specification is regressed separately for the 23-44 and the 45-65 age groups (see figures A3 and A4 in the appendix).
TABLE 11
ALTERNATIVE FUNCTIONAL FORMS FOR INTERGENERATIONAL INCOME MOBILITY IN URBAN ChILE (Centiles of labor income)
Variables (3) 1.622 (2.99)*** 0.035 (3.09)*** 0.0003 (3.71)*** 3.324 (0.20) 649 0.21 0.18 649 649 0.15 (2.80)*** (0.34) 22.394 5.312 (3.73)*** 12.151 (0.79) 649 0.17 0.0003 (3.09)*** (4.14)*** 0.035 0.005 (2.97)*** (11.00)*** (1.15) 1.62 0.379 0.157 1.027 (3.09)*** 0.024 (3.08)*** 0.0002 (3.81)*** 3.506 (0.23) 649 0.19 (4) (5) (6) (7)
Fathers centile from real income distribution
Fathers centile from predicted income distribution (8) 1.050 (3.12)*** 0.025 (3.11)*** 0.0002 (3.83)*** 34.479 (9.01)*** 649 0.18
(1)
(2)
Fathers centile
0.453
0.236
(10.97)***
(1.18)
Fathers centile^2
0.006
(3.55)***
Fathers centile^3
Constant
1.607
16.473
(0.11)
(1.04)
Observations
649
649
R-squared
0.17
0.19
Note: Models (1), (2), (3), (5), (6) and (7) include controls for son s life-cycle effect (age and age^2). Robust t statistics in parentheses. * significant at 10%,** significant at 5%; *** significant at 1%.
211
212
TABLE 12
STRUCTURAL ChANGE IN OLS COEFFICIENT OF FAThERS VS. SONS CENTILES IN UPPER AND LOWER FAThERS qUANTILE
Estimates Top quartile Bottom Top quintile Bottom 1.46 (0.27)** 0.7 (0.26)** 1.88 (0.37)** 0.91 (0.39)**
F-test of structural change 4.244
P-value 0.0023
4.231
0.0025
Robust standard errors in parentheses. * significant at 5%; ** significant at 1%.
FIGURE 2
qUANTILE CUBIC REGRESSIONS 100 90 80 70 60 50 40 30 20 10 0
1 4 7 10 14 17 20 23 27 29 33 36 39 42 46 49 52 56 59 62 65 69 72 75 78 82 85 88 92 94 98
Expected Son's Centile
Father's Centile 90th 75th median mean 25th 10th
Table 13 indicates that the expected income centiles of sons of fathers in the bottom decile is 30, that is, at the border of the 3rd and 4th deciles. Although this shows an important degree of intergenerational persistence, it also shows an important degree of the regression-to-the-mean effect for this group. however, sons of fathers in the top income decile can expect to be, on average, on the 86th centile, that is well into the 9th decile and close to the top centile: the regression-to-the-mean effect does not seem to be very important in this case. This pattern repeats itself in a milder version when comparing the expected
213
TABLE 13
SONS ExPECTED CENTILE IN ADULThOOD GIVEN FAThERS DECILE
Fathers Decile 1 2 3 4 5 6 7 8 9 10
Sons Expected Centile (Fathers Centile from Real Distribution) 30.04 39.90 45.20 47.49 48.35 49.32 51.98 57.88 68.59 85.66
income centiles of sons of fathers in the 2nd and 9th deciles, namely 40 and 69, respectively: the former sons are only 10 centiles below the median, while the latter are 19 centiles above it. As for the sons of fathers in deciles 3 to 8, their expected income centile is only a few percentage points away from the median, confirming an important degree of intergenerational income mobility at the centre of the fathers income distribution. It is interesting to note that the latter result is consistent with evidence on intergenerational occupational mobility for Chile. In particular, Torche (2005) finds that Chile exhibits a high level of persistence in the occupations with highest social status in compassion with the international evidence, but a significant degree of mobility in the remaining occupations. As a hypothesis waiting for research, this converging evidence may be associated with Chiles particular income distribution, characterized by an unusually large share of the national income being held by the top decile or so of the population, the remaining part of the population being fairly egalitarian19. Put quite simply, it seems plausible that the top decile or so of Chiles income distribution may be largely responsible for Chiles unequal distribution of income as well as for shaping Chiles social mobility patterns, a hypothesis that is open for future research. 6. Conclusions This paper has presented some findings on intergenerational income and educational mobility in urban Chile. A constraint to study intergenerational mobility in Chile, as in most of the developing world, is the lack of income panels where both fathers and their offsprings income can be observed. In this context, we follow the widely employed methodology known as two-sample two-stage
19
See for example De Ferranti et al. (2003).
214
least squares, whereby fathers incomes are predicted from income determinants reported by their sons, such as fathers schooling, occupation and age. We find intergenerational income elasticities for Greater Santiago in Chile in the range of 0.52 to 0.54. These figures seem lower than nationwide intergenerational income elasticities in the range of 0.6-0.7, but still high in comparison with the international evidence. The lower values for Greater Santiago may reflect the higher educational and occupational opportunities than can be expected in a large urban centre, compared to rural and small urban areas of the country, yet this hypothesis requires explicit investigation. The intergenerational income elasticities are somewhat lower for the younger cohorts. While this may suggest an increasing intergenerational mobility in time, it can also be associated with life-cycle effects on earnings, as indeed suggested by the literature. however, we find that intergenerational educational mobility is also lower for the younger cohorts for both sons and daughters. Based on the fact that life-cycle effects are unimportant as schooling remains fixed for most individuals along the life cycle after the mid twenties, this finding is suggestive of and increase of intergenerational educational mobility in urban Chile in the last decades. This seems consistent with the significant expansion of school and tertiary education observed in the last decades in Chile. Whether this phenomenon has indeed translated into higher intergenerational income mobility in recent decades is open for future research. We also examine some patterns of intergenerational income mobility by studying how the degree of mobility varies along the fathers relative income position. We find evidence of a higher degree of intergenerational income persistence (lower mobility) at the bottom-to-bottom and top-to-top transition probabilities, and a higher degree of intergenerational mobility in the intermediate segments of the fathers income distribution a pattern that is consistent with the international evidence. however, we find also that intergenerational mobility is substantially lower at the top of fathers income distribution. While this may be an expected pattern in other countries, this evidence is consistent with recent findings of intergenerational occupational mobility for Chile by Torche (2005), who reports a high degree of intergenerational persistence in occupations associated with high social status compared with a variety of other countries. It is tempting to propose, as hypothesis awaiting explicit research that this findings may be associated with Chiles particular income distribution, known to be quite egalitarian within the bottom 80 to 90 percent of the population, but relatively unequal for the country as a whole as a consequence of a high concentration of income in the top centiles of the distribution. 7. References Andrews, D. and A. Leigh (2008). More inequality, less social mobility, Applied Economics Letters; 1466-4291. Angrist, J. and A. Krueger (1992). The Effect of Age at School Entry on Educational Attainment: An Application of Instrumental Variables with Moments from Two Samples, Journal of American Statistical Association, 87 (418); 328-36.
215
Arellano, M. and C. Meghir (1992). Female Labour Supply and On the Job Search: An Empirical Model Estimated Using Complementary Data Set, Review of Economic Studies 59 (3); 537-59. Becker, G. and N. Tomes (1979). An Equilibrium Theory of the Distribution of Income and Intergenerational Mobility, Journal of Political Economy 87; 1153-1189. Bjrklud, A. and M. Jntti (1997). Intergenerational Income Mobility in Sweden Compared to the United States, American Economic Review 87 (5); p. 1009-18. Bourguignon, Franois, Ferreira, Francisco and Marta Menndez (2005). Inequality of Opportunity in Brazil? World Bank. Washington, D.C. Bourguignon, Franois, Ferreira, Francisco and Marta Menndez (2007). Inequality of Opportunity in Brazil, Review of Income and Wealth, Vol. 53, Issue 4; 585-618. Contreras, D., M. Fuenzalida and J. Nez (2006). Persistencia Intergeneracional del Ingreso en Chile y el Rol de la habilidad de los hijos, Economics Department, Universidad de Chile. Corak, M. and A. heisz (1999). The Intergenerational Earnings and Income Mobility of Canadian Men: Evidence from Longitudinal Income Tax Data, Journal of Human Resources 34 (3); 504-33. De Ferranti, D., Perry, G., Ferreira F., and M. Walton (2003). Inequality in Latin America and the Caribbean. Breaking with History?, The International Bank for Reconstruction and Development/The World Bank, Washington D.C. Dearden L., Machin S. and h. Reed (1997). Intergenerational Mobility in Britain, The Economic Journal 107(440); 47-66. Dunn C. (2007). The Intergenerational Transmission of Lifetime Earnings: Evidence from Brazil, Contributions to Economic Analysis & Policy, vol. 7 (2). Ferreira, S. and F. Veloso (2004). Intergenerational Mobility of Wages in Brazil. Unpublished paper. Fortin N. and S. Lefebvre (1998). Intergenerational Income Mobility in Canada, in Labor Markets, Social Institutions and the Future of Canadas Children, Miles Corak (ed.). Statistics of Canada, Ottawa. Grawe N. (2001). Intergenerational Mobility in the US and Abroad: quantile and Mean Regression Measures, PhD Thesis, University of Chicago. Lefrane A. and A. Trannoy (2004). Intergenerational Earnings mobility in France: Is France more mobile than the US?, Institut dconomie publique (IDEP) Working Paper N 401. Leigh, Andrew (2007). Intergenerational Mobility in Australia, The B.E. Journal of Economic Analysis & Policy, vol. 7 (2). article 6. Nez, J. and C. Risco (2004). Movilidad Intergeneracional de Ingresos en un Pas en Desarrollo: El Caso de Chile, Working Paper N 210, Department of Economics, Universidad de Chile. Nez, J. and R. Gutirrez (2004). Class Discrimination and Meritocracy in the Labor Market: evidence from Chile, Estudios de Economa, Vol. 31: 2; 113-132.
216
Nez, J. and A. Tartakowsky (2007). Inequality of Outcomes vs. Inequality of Opportunities in a developing country: an exploratory analysis for Chile, Estudios de Economa, Vol. 34: 2; 185-202. Nez, J. and A. Tartakowsky (2009). The relationship between Inequality of Outcomes and Inequality of Opportunities in a high-inequality country: The case of Chile, Departamento de Economa, Working Paper N 292. sterbacka, E. (2001). Family Background and Economic Status in Finland, Scand J. of Economics 103 (3); 467-484. Piraino, P. (2007). Comparable Estimates of Intergenerational Income Mobility in Italy, The B.E. Journal of Economic Analysis & Policy, vol. 7 (2). Solon, G. (1992). Intergenerational Income Mobility in the United States, American Economic Review 82 (3); 393-408. Solon, G. (2002). Cross-countries Differences in Intergenerational Earnings Mobility, Journal of Economic Perspectives 16 (3); 59-66. Torche, F. (2005). Unequal but fluid: social mobility in Chile in comparative perspective, American Sociological Review 70 (3); 422-450. Wiegand, J. (1997). Intergenerational Earnings Mobility in Germany, unpublished manuscript. zimmerman, D. (1992). Regression toward Mediocrity in Economic Stature, American Economic Review 82; 409-429.
Appendix TABLE A.1
ESTIMATES OF MINCER EqUATIONS USING EDUCATION AND POTENTIAL ExPERIENCE AS REGRESSORS
1958 Personal Income Labor Income Personal Income Labor Income Personal Income
1967
1977
1987 Labor Income
Log Income
Personal Income
Labor Income
(S-8)*d1
(S-12)*d2
Exper
Exper2
Constant
Observations Adj. R-squared
0.1098 (14.76)** 0.0839 (4.73)** 0.0675 (3.00)** 0.0604 (14.20)** 0.0008 (10.47)** 8.9524 (134.89)** 1747 0.51
0.1043 (14.20)** 0.0873 (4.99)** 0.0722 (3.25)** 0.0605 (14.43)** 0.0009 (11.14)** 8.9958 (137.61)** 1736 0.50
0.0768 (11.38)** 0.1252 (8.71)** 0.0249 (1.54) 0.0620 (18.73)** 0.0009 (13.79)** 11.4925 (210.52)** 2700 0.53
0.0725 (10.78)** 0.1288 (8.98)** 0.0329 (2.04)* 0.0617 (18.63)** 0.0009 (14.12)** 11.5125 (211.27)** 2691 0.52
0.0784 (8.35)** 0.1266 (7.20)** 0.0322 (1.88) 0.0636 (16.60)** 0.0009 (11.52)** 6.0278 (79.81)** 2325 0.48
0.0808 (8.60)** 0.1240 (7.04)** 0.0343 (2.01)* 0.0651 (17.00)** 0.0009 (12.33)** 5.9930 (79.34)** 2321 0.48
0.0638 (5.59)** 0.1038 (5.55)** 0.1152 (7.19)** 0.0539 (13.11)** 0.0007 (7.87)** 8.3213 (92.97)** 2070 0.59
0.0626 (5.43)** 0.1066 (5.65)** 0.1112 (6.89)** 0.0545 (13.17)** 0.0007 (8.23)** 8.3199 (92.11)** 2068 0.59
Absolute value of t statistics in parentheses. * significant at 5%; ** significant at 1%.
217
TABLE A.2
218
ESTIMATES OF MINCER EqUATIONS USING EDUCATION, POTENTIAL ExPERIENCE AND OCCUPATION AS REGRESSORS
1958 Personal Income Labor Income Personal Income Labor Income Personal Income
1967
1977
1987 Labor Income
Log
Personal Income
Labor Income
(S-8)*d1
(S-12)*d2
Exper
Exper2
Employer = 1
Self-employed = 1
Employee = 1
Domestic servants=1
Constant
Observations Adj. R-squared
0.0852 (11.46)** 0.0496 (2.89)** 0.0083 (0.38) 0.0548 (13.37)** 0.0008 (10.30)** 0.7414 (10.82)** 0.1677 (4.44)** 0.3755 (9.96)** 0.7535 (1.97)* 9.0698 (141.07)** 1747 0.55
0.0805 (10.97)** 0.0530 (3.12)** 0.0127 (0.58) 0.0550 (13.59)** 0.0008 (10.90)** 0.7137 (10.49)** 0.1220 (3.27)** 0.3734 (10.05)** 0.7570 (2.01)* 9.1128 (143.92)** 1736 0.54
0.0526 (7.84)** 0.1033 (7.47)** 0.0199 (1.26) 0.0550 (17.16)** 0.0008 (13.17)** 1.0662 (15.37)** 0.2197 (6.69)** 0.2997 (9.95)** 0.3714 (1.89) 11.6337 (219.63)** 2700 0.58
0.0478 (7.16)** 0.1087 (7.89)** 0.0100 (0.64) 0.0545 (17.10)** 0.0008 (13.52)** 1.0558 (15.23)** 0.2033 (6.21)** 0.2962 (9.87)** 0.7523 (3.85)** 11.6607 (221.22)** 2691 0.57
0.0584 (6.59)** 0.0808 (4.87)** 0.0195 (1.21) 0.0527 (14.57)** 0.0007 (10.48)** 1.3777 (18.79)** 0.2582 (7.16)** 0.3794 (10.59)** 0.0025 (0.01) 6.1962 (87.39)** 2325 0.55
0.0606 (6.87)** 0.0759 (4.58)** 0.0196 (1.22) 0.0538 (14.96)** 0.0008 (11.29)** 1.4163 (19.29)** 0.2637 (7.35)** 0.3891 (10.89)** 0.3188 (0.91) 6.1659 (87.34)** 2321 0.56
0.0511 (4.94)** 0.0629 (3.64)** 0.1418 (9.73)** 0.0427 (11.32)** 0.0005 (6.94)** 1.5265 (21.22)** 0.0983 (2.71)** 0.3284 (9.08)** 8.5123 (104.32)** 2070 0.67
0.0499 (4.77)** 0.0646 (3.71)** 0.1383 (9.42)** 0.0432 (11.38)** 0.0006 (7.34)** 1.5306 (21.12)** 0.0994 (2.72)** 0.3389 (9.30)** 8.5107 (103.31)** 2068 0.66
Absolute value of t statistics in parentheses. * significant at 5%; ** significant at 1%.
219
TABLE A3
qUARTILE TRANSITION MATRIx Fathers quartile from real income distribution
Father Bottom 2nd 3rd Top Immobility Index
Son Bottom 0.50 0.30 0.29 0.14 0.38 2nd 0.27 0.22 0.24 0.14 3rd 0.19 0.33 0.28 0.18 Top 0.04 0.15 0.19 0.54
TABLE A4
qUARTILE TRANSITION MATRIx Fathers quartile from distribution of predicted income
Father Bottom 2nd 3rd Top Immobility Index
Son Bottom 0.39 0.32 0.26 0.12 0.35 2nd 0.24 0.23 0.26 0.12 3rd 0.30 0.30 0.22 0.20 Top 0.08 0.15 0.26 0.55
220
FIGURE A1
INTERGENERATIONAL INCOME TRANSITION PROBABILITIES (quintiles)
0.60
Probability Son from Lowest Income Quintile Family is in Lowest Quintile is 37%
Probability Son from Highest Income Quintile Family is in Highest Quintile is 47%
0.50
0.40
0.30
0.20
Bottom 2nd
C
Lowest Quintile Father Highest Quintile Son, 0% Highest Quintile Father Lowest Quintile Son, 8%
0.10
0.00
3rd
Fathers Quintile
4th Top Top 4th
3rd
2nd
Bottom
Sons Quintile
FIGURE A2
INTERGENERATIONAL INCOME TRANSITION PROBABILITIES (quartiles)
B
Probability Son from Highest Income Quartile Family is in Highest Quartile is 54%
Probability Son from Lowest Income Quartile Family is in Lowest Quartile is 50%
0.60
0.50
0.40
0.30
0.20
D C
Bottom 2nd
Lowest Quartile Father Highest Quartile Son, 4%
0.10
Highest Quartile Father Lowest Quartile Son, 14%
0.00
Fathers Quartil
3rd Top 3rd Top
Bottom 2nd
Sons Quartil
221
FIGURE A3
qUANTILE CUBIC REGRESSIONS FOR 23-44 COhORT 100 90 Expected Son's Centile 80 70 60 50 40 30 20 10 0
1 5 10 15 19 24 29 34 39 44 48 53 58 62 67 73 77 82 87 92 96
Father's Centile
90th 75th median mean 25th 10th
FIGURE A4
qUANTILE CUBIC REGRESSIONS FOR 45-65 COhORT 100 90 80 70 60 50 40 30 20 10 0
1 5 10 15 21 26 31 35 41 46 51 56 62 67 72 77 82 87 92 98
Expected Son's Centile
Fathers Centile
90th 75th median mean 25th 10th
Estudios analysis of the income N 1, Junio in Chile / Claudio Sapelli A cohort de Economa. Vol. 38 - distribution 2011. Pgs. 223-242
223
A cohort analysis of the income distribution in Chile*

Un anlisis de cohorte de la distribucin del ingreso en Chile
Claudio Sapelli** Abstract In this paper we look at the income distribution by cohort in Chile. We construct a synthetic panel from cross section surveys and estimate the income distribution for cohorts born between 1902 and 1978. We then decompose the evolution of these distributions into age, year and cohort effects. The cohort effects show a period where inequality increases, to then decrease. We attempt to explain this evolution. The rise can be explained by variables associated with education, while the fall appears to be the consequence of a flattening of the income-age profile and hence a reduction in the returns to experience. Key words: Synthetic cohorts, Income distribution, Rates of return, Education. Resumen En este artculo analizamos la distribucin del ingreso en Chile por cohortes. Construimos un panel sinttico a partir de encuestas de corte transversal y estimamos la distribucin de ingresos para los cohortes nacidos entre 1902 y 1978. Luego, descomponemos la evolucin de estas distribuciones en efectos de edad, ao y cohorte. El efecto de cohorte muestra un perodo donde la desigualdad aumenta, para luego decrecer. Intentamos explicar esta evolucin. El incremento puede se explicado por variables asociadas a la educacin, mientras la cada parece ser consecuencia de un aplanamiento del perfil de ingreso-edad y por lo tanto a una reduccin de los retornos a la experiencia. Palabras clave: Cohortes sintticas, Distribucin del ingreso, Tasas de retorno, Educacin. JEL Classification: I20, J24, J31.
I would like to thank Alejandra Candia, Rafael Sanchez, and in particular Will Mullins for research assistance in the production of this paper. Comments by Harald Beyer, Dante Contreras, Kevin Cowan, Eduardo Engel, Walter Sosa, Bernardita Vial, and the participants at the SECHI 2005/2006 Meetings and the LACEA 2005/2006 meetings are gratefully acknowledged. All remaining errors are my sole responsibility. ** Instituto de Economa, Pontificia Universidad Catlica de Chile.
224
1. Introduction The central contribution of this paper is its use of synthetic cohorts to analyze the issue of income distribution, providing a new perspective on the topic. The paper studies the evolution of income distribution for cohorts born between 1902 and 1978, using synthetic cohorts constructed from successive cross section surveys (with data taken from the Employment surveys of the University of Chile - 1957 to 2004). It decomposes this evolution into cohort, age and year effects and then analyzes the pattern of cohort effects. In particular it looks into whether these patterns can be explained by trends in the mean and dispersion of both years of education and returns to education in the cohort. To do this we estimate the rates of return to schooling in Chile for those cohorts (see Sapelli and Mullins (2006) and Sapelli (2009)). A cohort analysis of the issue of income distribution opens up a new perspective on the topic. The evolution of cohort income inequality during the period 1902-1978 shows an interesting dynamics: inequality first increases and later decreases1. However, the public policy discussion in Chile regarding income distribution is made on the basis of the cross section income distribution (based on one cross section survey, be it the U. de Chile survey or the CASEN survey) which show inequality basically flat for many decades now. However, from the point of view of designing public policy this appears to be a mistake, since it is very hard for public policy to act on the stock (the sum of all cohorts) income distribution, since it would imply addressing a multiplicity of causes. But public policy can act on the income distribution of recent cohorts (or marginal cohorts) with a much wider range of policies (by improving the quality of education, decreasing desertion, increasing tuition credit). Since these policies act at the margin, they should not be judged according to what happens to the stock. Hence this paper argues that the evolution of income distribution that is described here should be considered more indicative of what is currently happening than estimates based on cross section data for one year. Since both methods tell different stories, the analysis of inequality using data from one cross section survey (the methodology that is most commonly used), may lead to incorrect public policy decisions. Panel data would probably be more precise, but in Chile, as is the case in many developing countries they do not exist yet: cross section surveys are all that is available. We need to look at whether policy is having an effect on the inequality of recent cohorts. Changes in the marginal distribution that are sustained over time will eventually affect the stock distribution. Moreover, by looking at these changes we could predict what the stock income distribution will look like in the future (however, this is not done here). Another angle from
Inequality is measured by the Gini index. The method of estimation of these Ginis is described bellow.
A cohort analysis of the income distribution in Chile / Claudio Sapelli
225
which this dichotomy can be viewed is that of the difference between intragenerational inequality and inter-generational inequality. We will not look into inter-generational inequality in this paper2, but we will look at intra-generational inequality and its evolution. 2. The Data We use the Universidad de Chile Encuesta de Ocupacin y Desocupacin en el Gran Santiago surveys of Gran Santiago called for the period June 1957-2004. The sample is composed of men and women in the greater Santiago area who report non-zero work income, are aged between 18 and 65 inclusive, and who were born between 1904 (on occasion 1902) and 1978 inclusive3. The individuals are grouped into cells of individuals that were all born in the same year and are all observed at the same age e.g. men and women born in 1940 observed at age 20 (in 1960). The tables below detail the means, standard deviations, maximums and minimums of the data thusly organized. Hence we construct synthetic panels of the population born between 1904 and 1978.
Variable Age Work Income Education in years Gini Education x cell Gini Education x cohort Income Gini x cell Mean Education x cohort Mean Education x cell Obs. 161,827 161,827 161,827 161,827 161,827 161,827 161,827 161,827 Mean. 36.732 131326.1 9.389971 0.2445167 0.2545859 0.4652665 9.389971 9.389971 Std. Dev. 11.55452 238897 4.399738 0.0667459 0.0623036 0.0911754 1.449989 1.787015 Min. 18 84.5 0 0 0.1191496 0.1521739 6.605911 4.153846 Max. 65 1.22e+07 20 0.5460317 0.4082756 0.8339984 12.14029 13.57282
Inter-generational inequality is marked by the existence of older generations with a low average number of years of education. Since they are present in the stock distribution along with generations with more education, and since over time the number of college graduates has increased, and university rates of return have also risen, it is unsurprising that the tendency of the stock income distribution has been to deteriorate or at best to stay the same. However this group does not exist in the most recent cohorts. Some individuals were dropped from the sample, They were individuals that had either: 1) Zero work income; 2) Missing/wrongly coded work income data; 3) Missing/wrongly coded educational data (This excludes all data from the 1958, 1963 and 1964 surveys: they did not collect educational data); 4) Younger than 18 or Older than 65; 5) Born before 1904 (in some cases 1902) or after 1978; 6) Belong in a cohort-age cell (i.e. born in year 19wx aged yz) composed of less than 15 individuals. This group is composed of approximately 630 individuals.
226
3. Income Distribution by Cohort: What Does the Data Say?4 To analyze the data and estimate the cohort effects we first estimate the Gini coefficients by cohort and year. For example, we estimate the Gini coefficient for the cohort born in 1939 in every year we observe them within our sample period of 1957 to 2004. This is done for all cohorts born from 1902 to 1978 inclusive, and these Gini coefficients, calculated from an average of 80 individuals incomes, are used as the dependent variable. We set the cut-off points at the 1902 and 1978 cohorts because working with individuals between 18 and 65 years of age implies that we observe these cohorts at least in nine different years, allowing a sufficiently trustworthy estimate of the cohort effect. With the data so generated we run a regression to separate cohort, age and year effects. Because of the well-known estimation difficulties that arise from the perfect co-linearity between cohort birth year, age, and year of survey, we use the method developed by Deaton (1997) to sort between these three effects. The idea behind this methodology is to run a regression between these Gini coefficients and dummies by cohort, by year and by age (regression results available from the author). The assumption used by the Deaton method to identify these three effects is that the year effects have no trend and therefore add to zero. Another assumption is that there are no interactions between the three effects. The Deaton method, among other things, permits us to abstract from the fact that our cohorts have different age compositions. This is important since the earlier cohorts we observe only for their later years and the later cohorts only for their earlier years. Figure 1 shows the results of applying this method5. We graph only the cohort effects, i.e. the coefficients of the cohort dummies. In this paper we abstract from the results of the year effect (that show that inequality increases at times of high growth and decreases with recessions) and from the results of the age effect (that show the standard result that inequality grows with age, at a declining rate). The cohort effects we show in Figure 1 are what later we call the effects estimated with a pure Deaton regression (i.e. only with the dummies for the three effects), to distinguish them from regressions where we attempt to explain the evolution of the cohort effects described here. We obtain that the cohort effects explain changes of 9 points in the Gini index. Most recently cohort effects decrease systematically for cohorts born from the fifties onwards (and in particular since 1959). We can see in Figure 1 that the cohort effects describe an inverted U shaped curve from the cohort born in 1929 to that born in 1978. Starting with cohorts born in 1929 (which entered the labor market in the late forties/early fifties) there is an increase in cohort
There are several changes between the data used here and that used for previous versions of this paper that are worth noting. First, the surveys of 1959, 1963, and 1964 are included in some regressions. They were excluded in the past because they lacked data on education. Data from 2003 and 2004 (which we did not have) were also added to the sample. All cohorts from 1902 to 1978 are now studied; previously we worked only with 1945-1978. Some changes relevant to the rate of return calculations were made to the inflation deflator (see Sapelli and Mullins 2006). Detailed results for all regressions cited in this paper can be examined in Sapelli (2007).
227
income inequality that culminates in 1951. From 1951 on, and in particular from 1959 on, there is a downward trend. FIGURE 1
COHORT EFFECTS FROM A PURE DEATON REGRESSION (With a 95% confidence interval)
0.14 0.12 0.1 0.08 0.06 0.04 0.02 0 0.02

1906 1909 1912 1915 1918 1921 1924 1927 1930 1933 1936 1939 1942 1945 1948 1951 1954 1957 1960 1963 1966 1969 1972 1975
0.06 0.08 0.1
In Figure 1 we can appreciate that standard errors6 are large and cohort effects are not significantly different from zero for most of the cohorts. However, there is a period (for cohorts born from 1939 to 1961) where most (20 out of 23) of the cohort effects are significantly different from zero. Thereafter they return to being statistically zero. Hence one could say that while there are no cohort effects at the beginning and the end, they exist for the mid period and hence income inequality is significantly larger for cohorts born from 1939 to 1961. However, given the low mass in each cell, standard errors are large, and most cohort effects are not significantly different from 0.04. To diminish the problem of large standard errors, we use a standard methodology in cohort analysis to increase mass in each cell and achieve more precision in our estimates7: we work with moving averages of several cohorts instead of with each cohort individually. We therefore work with moving averages of 3 (MA3) and 5 cohorts
We report here results with normal standard errors. Robust standard errors were also used to correct for some indications of heteroskedasticity caused by the fact that our dependent variable (Ginis by cohort and year) are being calculated from different numbers of individuals incomes e.g. some are calculated from 200 individuals incomes, while others were based on 15 individuals (the average number of individuals used to calculate the Ginis is 80). However, results with normal SE or robust SE do not differ much. This is standard procedure in the literature (see, for example, Eckstein and Nagypal The Evolution of U.S. Earnings Inequality: 1961-2002, Federal Reserve Bank of Minneapolis Quarterly Review 28 (2), 2004).
1978
1903
0.04
228
(MA5)8. In each moving average we include all the data of 3 or 5 generations to characterize the cohort at the centre of the interval. We only report here the results for the MA5. The MA5 estimates are shown in Figure 2. They show basically the same evolution seen in Figure 1, however the inverted U now covers the whole period: first a rise in inequality and then a fall. The statistical significance rises and the period where cohort effects are significantly different from zero is now more extensive and ALL cohort effects are significant within this period. The period with significant cohort effects is now for the generations born between 1935 and 1965. Most importantly, the cohort effects show a statistically significant pattern, in particular regarding the rise in inequality which is significant at the 5% level. Regarding the fall towards the end, it is not significant at the 5% level, but it is significant at the 10% level. Regarding the magnitude of the changes in inequality, in the case of the estimation without moving averages, the cohort effects rise 7 points between 1929 and 1951, then are stable from 1951 to 1959, and then drop 8 points. In the case of the MA5 estimates, the rise occurs from 1911 to 1947 (and is of 5 points) and then there is a drop of about the same magnitude (4 points)9. As is usually the case, trends are softened when working with moving averages. FIGURE 2
COHORT EFFECTS FOR A PURE DEATON REGRESSION WITH COHORTS DEFINED AS A MOVING AVERAGE OF 5 COHORTS (Shown together with a 95% confidence interval)
0.06 0.05 0.04 0.03 0.02 0.01 0 1902 1905 1908 1911 1914 1917 1920 1923 1926 1929 1932 1935 1938 1941 1944 1947 1950 1953 1956 1959 1962 1965 1968 1971 1974 1977 0.01 0.02 0.03 0.04
We do this only for the pure Deaton regression but not for other regressions in this paper since the generation of variables by cohort in this definition is cumbersome (especially for the rates of return) and interpretation of results is not straightforward. Quantitatively these falls can be appreciated in the following manner. If we take into account a Gini of between 0.5 and 0.55 (which is the result from cross section estimates) a decrease of 8-9 point would imply a fall of 16-18% in the index. One of 4-5 points would imply a fall of about 8-10 percent.
229
It is important to note that the evolution of inequality we have described (i.e. the inverted U shape) is robust to working with other inequality indexes. In particular we have verified this using both the Theil 0 and Theil 1 indexes and several percentile indices of dispersion, such as the interquartile range, as the dependent variable. The evolution of the different percentiles tells an interesting story that is discussed in the following section. 4. Robustness of the Evolution of Inequality In this section we report the cohort effects estimated with the Theil 1 and Theil 0 indexes and compare them with the Gini index. We also analyze the evolution of different percentile measures. In Figure 3 we can see that the evolution of the Theil 1, Theil 0 and Gini indexes are very similar. The three show similar behavior throughout. In all there is a period of increase in income inequality, followed by a period of decreasing inequality. In all the period of increase includes the 1929-1951 cohorts, though in the case of the Theil 1 it is possible to argue that the trend to increase started beforehand. The three show a drop, but the duration of the drop varies: it is longer for the Gini index (up until the end of the series) but shorter in the other two series (up to the 1964 cohort). FIGURE 3
COMPARISON OF THE EVOLUTION OF THE THEIL 1, THEIL 0 AND GINI INDExES OF INCOME DISTRIBUTION
0.25 0.225 0.2 0.175 0.15 0.125 0.1 0.075 0.05 0.025
1903 1906 1909 1912 1915 1918 1921 1924 1927 1930 1933 1936 1939 1942 1945 1948 1951 1954 1957 1960 1963 1966 1969 1972 1975
Gini
Theil 0
Theil 1
0.025 0.05
Cohort
Since we observe a somewhat different behavior between the three measures we look deeper into the evolution of the different parts of the distribution to try to assess why they differ. Possibly the most notable similarity is in the peaks and troughs of the three series. However, there is a clear difference in the mag-
1978
230
nitude of the fall in the later years. Excluding both the 1909 cohort and the last cohort, we find that in the case of the Gini index, the fall is larger in magnitude than the rise. However in the case of both Theil indexes the fall is smaller than the rise (two thirds of the rise in the case of the Theil 1 and three quarters in the case of the Theil 0). It is useful to examine the evolution of the P90-P50 difference (the upper half of the distribution), of the interquartile range (the middle of the distribution) and the P50-P10 difference (the lower half of the distribution). The interquartile range behaves pretty much like the P90-P50 difference but the upper half of the distribution and the lower half behave very differently. While there is a systematic decrease of the inequality in the bottom half of the distribution (up until the 1972 cohort), inequality in the upper half rose between the 1929 cohort and the 1941 cohort, then fell between the 1941 and 1962 cohorts and then rose again. The rise in inequality in the upper half of the distribution explains most of the raise in overall inequality and then the fall occurs when both halves of the distribution are following the same trend and are decreasing in inequality. However, the periods when the percentiles fall and when the overall indexes fall do not coincide. Hence we take a more detailed look at the different parts of the distribution, looking at the differences between the following percentiles: 95th and 80th, 80th and 65th, 65th and 50th, 50th and 35th, 35th and 20th, and finally 20th and 5th. We omit their presentation not to clutter the paper with too many numbers. However, the conclusions from their analysis follow. The best way to analyze these data is to separate the period into three subperiods (according to the evolution of the Theil and Gini indexes), and see which parts of the distribution contribute to explain the evolution of the overall indexes in each of the sub-periods. We separate the period into the 1903-29, 1929-1951 and 1951-78 sub-periods, which correspond to periods of: trend stability (but with great variation), rise, and fall of the indexes, respectively. We analyze the evolution of these percentiles by looking at the simple correlations between the different measures. In the first period it is mostly the upper part of the distribution that leads the evolution of the overall indexes. The lowest part (P20-P5) has a life of its own, since the correlation with most other percentile differences is negative. In the second period the same is true again. The overall indexes are lead by the upper part, while the lower part has the opposite behavior (there is an improvement in inequality in the lower part while there is deterioration in the upper part). Again the P20-P5 is an outlier, this time having the same behavior as the upper part of the distribution (deterioration). Hence the period when inequality worsens is a period where this evolution is mostly explained by an increase in inequality in the upper half of the distribution. The P95-P80 measure has a steep increase in the period 1923-1941 and then stays stable. These cohorts entered the labor market in the forties and fifties. If we take into account that most of the present cross section inequality is explained by inequality in the upper quintile (see Torche (2005)), explaining this deterioration that is lead by the upper part of the distribution is key in explaining the high level of inequality in Chile today. In the last period (the period where the decline occurs) we observe that the overall indexes are highly correlated with the bottom of the distribution. Hence the reason behind the increase is deterioration at the top, but the reason behind
231
the decrease is an improvement at the bottom. That leaves the distribution exactly where Torche describes it: with very good distribution at the bottom, but unequal at the top. In the section that follows we will attempt to explain the evolution of the overall indexes through the evolution of human capital variables. 5. How Education Evolved Over the Period What we have found shows that there is an interesting dynamic in the behavior of income distribution. This runs counter to the common observation that income distribution has not changed in Chile for many years. This stagnation is frustrating, specially given the fact that during the 20th century the population systematically increased its average education level, and the dispersion of the education level also fell steadily, as can be seen in Figures 4 and 5. These trends make clear that education cannot explain both the increase and the decrease in inequality, since these variables increase or decrease continuously over the period. We will address this issue later. Figure 4 shows that mean education, in particular starting in 1939 has grown steadily. It climbed at a higher rate in the period 1939-1958, which roughly coincides with the period in which the income distribution deteriorated. Figure 5 shows how the dispersion of education within cohorts (as measured by the Gini index of education) has decreased steadily. In this case acceleration in the rate of decrease is observed for the cohorts born between 1947 and 1961 (those going through the education system starting in the mid fifties to the late sixties). However, this does not coincide with the period where income inequality decreases. FIGURE 4
MEAN EDUCATION BY COHORT (1902-1978) 14 13 12 11 10 9 8 7 6 1902 1906 1910 1914 1918 1922 1926 1930 1934 1938 1942 1946 1950 1954 1958 1962 1966 1970 1974 1978
232
FIGURE 5
DISPERSION OF EDUCATION BY COHORT (Cohorts born 1902-1978) 0.45 0.4 0.35 0.3 0.25 0.2 0.15 0.1
Only indivs with income > 0 All individuals
To link the evolution of education to that of income inequality we use the human capital model and in particular what is known as the Mincer equation. Hence our theoretical framework is one in which the evolution of income is determined by changes in human capital and in the rate of return to human capital. As a result, the evolution of the income distribution is determined by the means and standard deviations of the distribution of education and of the rates of return to education for each cohort. The traditional Mincer equation (where we ignore experience) is: (1) ln Y = A + RS
Where Y is income, S is education, and A is a constant and R can be interpreted as the rate of return to education. If we take the variance of equation (1), assuming S and R are independent10 we obtain: (2) V(lnY) = [mean (R)]2V(S)+[mean(S)]2V(R) + V(S)V(R).
Where V(S) and V(R) are the variance of education and the variance of the rate of return, respectively. Hence, all else equal, the partial derivatives of the variance of log income with respect to mean education, variance of education, mean returns to education and variance of rates of return, are all positive. This provides an explanation for the fact that income inequality increased at the time when mean education rose more rapidly. From the point of view of the human capital Mincerian framework, during the 20th century, the evolution of the distribution of education
10
19 2 0 19 5 0 19 8 1 19 1 1 19 4 1 19 7 2 19 0 23 19 2 19 6 2 19 9 3 19 2 3 19 5 3 19 8 41 19 4 19 4 4 19 7 50 19 5 19 3 5 19 6 5 19 9 6 19 2 6 19 5 6 19 8 7 19 1 7 19 4 77
Cohort
19
This is a reasonable assumption if the returns to education are mainly driven by demand shocks rather than by supply shocks.
233
provided two countervailing forces: one force increasing income inequality (the increase in the mean) and the other force decreasing inequality (the decrease in dispersion)11. Since the inverted U shape of income distribution cannot be explained by variables that follow basically a linear trend, it is possible that the returns to education play a part in this story the topic of the following sections. 6. How Rates of Return Evolved Here we will only report findings and discuss how they may explain the evolution of income inequality. Further on we will run a more formal test of the influence of education and rates of return on income inequality12. Alternative estimates of rates of return are reported and discussed in Sapelli and Mullins (2006). The results are shown in Figure 6. There we include the returns to complete university education (i.e. the number of percentile points that persons climb in the income distribution if they have complete university education as opposed to having complete secondary education) and complete secondary education (the same, comparing complete secondary with complete primary). We omitted primary education (returns for the seventh and eighth years here have been steady at about five percentile points). As can be seen in the Figure 6, returns for secondary school were higher than those of university up until cohorts born in 1927, when both lines cross for the first time. Both returns continue at similar levels until 1943. From cohorts born in 1944 and after returns diverge sharply, with returns to complete university education implying changes in the income distribution of up to 45 percentile points, and with returns to complete secondary education falling to about 5 percentile points. The maximum for one series and the minimum for the other coincide in
11
12
This process will end with the mean reaching a plateau, as it has in developed countries, at which point the dispersion of education will be the only force driving income inequality. Nonetheless, we are not there yet, and we will probably have to wait for most people to complete secondary education before this occurs. What we have here are rates of return estimated as how many percentile points in the income distribution a person ascends within his own cohort with additional education. These rates of return we can estimate for the full sample of cohorts. What we do is basically a Mincer regression in which we do not use income as the dependant variable but the percentile where the person is in the income distribution of his cohort. The coefficient of education gives us a measure of how many percentiles an additional year of education will lift a person in the income distribution of his own cohort. We work with a spline and interactions by cohort to estimate returns to different levels of education and sheepskin effects for different cohorts. See Sapelli and Mullins (2006) for a more detailed discussion of the methodology. This measure of the return to education is not sensitive to either the correction used for inflation (very relevant for periods of high inflation and dubious price statistics e.g. Chile 1970-78), or highly determined by the first five or so income flows, both problems faced by traditional rates of return. Moreover, the latter approach requires data on a cohort from age 12 onwards, making it impossible to estimate rates of return for cohorts born before 1945 (our earliest data is the 1957 survey), a problem sidestepped by the regression structure of the percentile approach. For alternative measures of the rates of return to education see Sapelli (2009).
234
time, for cohorts born in the early seventies. Then there is a sharp reversal in the trend and both returns tend to equality at about 30 percentile points. As we will see, the beginning of the period, when returns diverge, coincides with the period when income distribution deteriorates. However returns converge too late to explain the decline in inequality that starts with the cohorts born in the fifties. FIGURE 6
RATES OF RETURN TO UNIVERSITY EDUCATION (LIGHT) AND SECONDARY EDUCATION (DARK)
50 45 40 35 30 25 20 15 10 5
1904 1907 1910 1913 1916 1919 1922 1925 1928 1931 1934 1937 1940 1943 1946 1949 1952 1955 1958 1961 1964 1967 1970 1973 1976
High School
College
Figure 7 shows the relative returns to different levels of education. This is estimated as the ratio of the number of percentiles you climb by completing one level over the number of percentiles you climb by completing the other. In the case of primary school we use the number of percentiles a person that completes the last two years of primary (7th and 8th grades) climbs in the income distribution. The dark line tells us the same story as Figure 6: relative returns of university compared to secondary schooling are approximately flat (and similar) for most of the period, and start rising (at an increasing pace) with the cohorts born in the mid forties. They then fall sharply towards the end of the period. The ratios of university and secondary to primary tell a new story. Both ratios rise until the late forties and fall after that, but the ratio of secondary to primary falls more sharply. Since these ratios are an indication of how dispersed the rates of return are, and that the dispersion of returns increase income inequality, they may be part of the explanation of the trends in cohort income inequality. This is more probable since these broad trends do coincide with those in income inequality, an issue we will examine more formally later.
235
FIGURE 7
RATIO OF RETURNS: UNIVERSITY/SECONDARY (DIAMONDS), UNIVERSITY/PRIMARY (SQUARES) AND SECONDARY/PRIMARY (TRIANGLES) 16
RELATIVE RATES OF RETURN
14 12 10 8 6 4 2 0
19 0 19 4 0 19 6 0 19 8 1 19 0 1 19 2 1 19 4 1 19 6 1 19 8 2 19 0 2 19 2 2 19 4 26 19 2 19 8 3 19 0 3 19 2 3 19 4 3 19 6 3 19 8 4 19 0 4 19 2 4 19 4 4 19 6 4 19 8 5 19 0 5 19 2 5 19 4 5 19 6 5 19 8 6 19 0 6 19 2 6 19 4 6 19 6 6 19 8 7 19 0 7 19 2 7 19 4 7 19 6 78
Ratio Return to College/Return to High School Ratio Return to College/Return to Primary School Ratio Return to High School/Return to Primary School
FIGURE 8
MEAN RETURN TO EDUCATION MEASURED IN PERCENTILES (Weighted average of returns to the three levels) 21 19 17 Percentiles 15 13 1 9 7 5
1904 1906 1908 1910 1912 1914 1916 1918 1920 1922 1924 1926 1928 1930 1932 1934 1936 1938 1940 1942 1944 1946 1948 1950 1952 1954 1956 1958 1960 1962 1964 1966 1968 1970 1972 1974 1976 1978
Cohort
Equation (2) above tells us that average returns (given dispersion) will increase income inequality. The data shows average returns increasing for most of the century and stabilizing towards the end.
236
In sum, most of the indicators we have looked at could contribute to explain the increase in inequality, but the only one that appears to contribute to the explanation of the most recent downward trend is the dispersion of returns. To take a more detailed look at these, we look at sheepskin effects (the returns to finishing a level, be it primary, secondary or university, the actual values are not reported here), the existence of which will contribute to increasing the dispersion of rates of return per year of education. The estimates show positive sheepskin effects for most of the period. They are of a similar magnitude for about a decade starting with cohorts born in 1918. They then diverge slowly, and starting in the late thirties they diverge strongly, to converge again towards the end of the period (starting in the mid sixties). This confirms that there may be an important contribution of the evolution of the dispersion of rates of return in the reduction in income inequality. 7. Do Education Variables Explain the Trend in Cohort Income Inequality?: A Formal Test In this section we add variables to the pure Deaton regression to see if the trend in cohort income inequality is explained by the evolution of variables related to the level of education of the cohort, its dispersion, the rates of return of the cohort and its dispersion. If when we add them, the cohort effects disappear, then these variables explain the cohort effects. TABLE 1
SUMMARY OF REGRESSIONS (ExCLUDING COEFFICIENTS OF DUMMY VARIABLES COHORT, AGE AND YEAR EFFECTS)
1 Coefficient Rate of return (mean) College/High School College/Primary mean education Gini education Number of obs R-squared Root MSE 0.0067596 0.0091804 0.0008937 0.0241826 0.5091556 2025 0.6719 0.05804
2 Coefficient na na na 0.0241826 0.5091556 0.6719 0.05804
3 Coefficient 0.0068581 0.0122137 0.0009814 0.0035959 dropped 2025 0.6479 0.0601
4 Coefficient na na na 0.0062458 dropped 0.6479 0.0601
All coefficients significant at 1% level. Column 1 Column 2 Column 3 Column 4 DEATON PLUS EDUCATION PLUS RETURNS (CELL DEFINITION) DEATON PLUS EDUCATION (CELL DEFINITION) DEATON PLUS EDUCATION PLUS RETURNS (COHORT DEFINITION) DEATON PLUS MEAN EDUCATION AND GINI EDUCATION (BY COHORT)
237
We estimate the additional variables both by cohort and by age-cohort cell. In the first case we look into variables that could explain the cohort effect as estimated by a pure Deaton regression. In the second case we use variables that could explain part of both the cohort effect and the age effect. Hence we implicitly introduce an interaction term that implies modifying slightly the traditional Deaton methodology. We do this because, as we have seen, trends in education are continuous during the century (in particular for the Gini of years of education and the mean level of education achieved), and hence cohorts are systematically different in these respects. The relatively crude separation technique offered by Deaton may confound what in reality are the effects of the changes in education (and hence cohort effects) with age effects. Since older persons in every year have a higher degree of education inequality, the method may attribute this to pure age effects when in reality they are cohort effects. The point is that since there is a correlation between what happens by cohort and what happens by age, it is possible that the Deaton methodology is not correctly separating cohort and age effects. More to the point, it is possible that the reduction in inequality due to cohort effects is even larger than the one showed by the pure Deaton regression, since part of the effect of the reduction in the Gini index of education may incorrectly be attributed to age (being younger) rather than cohort. What one would expect if this is true is that the age effects would be appreciably reduced when using the age-cohort characterization of education variables (only the mean and dispersion of the years of education is changed in the age-cohort specification, not the variables associated with the returns to education which are necessarily estimated by cohort). Regressions results with variables defined by cohort are not discussed here though they are reported in Table 2. Variables defined by cohort do not help us explain the evolution of cohort inequality. TABLE 2
COHORT EFFECTS FROM TROUGH TO PEAK AND FROM PEAK TO TROUGH FOR REGRESSIONS INCLUDING DIFFERENT VARIABLES (Changes in points of the Gini index) Variables defined by cohort Regression Inequality increases 6 (1936-1951) 5 (1937-1951) 5 (1936-1951) Inequality falls 9 (1952-1978) 11 (1952-1978) 8 (1952-1978) 12 (1952-1977) Variables def. by cohort-age cell Inequality increases 6 (1936-1951) 3 (1909-1951) 0 (1909-1951) Inequality falls 9 (1952-1978) 12 (1952-1978) 8 (1952-1978) 13 (1952-1977)
Pure Deaton Plus education Plus education and returns
8. Regressions with Variables Defined by Cohort-Age Cell Figures 9 and 10 show the cohort effects after we control only for mean years of education and Gini of education (Figure 9) and for those variables plus variables that characterize the distribution of returns (Figure 10). Regressions are summarized in Table 1.
238
FIGURE 9
COHORT EFFECTS, WITH MEAN AND GINI EDUCATION ADDED TO PURE DEATON REGRESSION, VARIABLES DEFINED BY CELL 0,15 0,1 0,05 0
1903 1908 1913 1918 1923 1928 1933 1938 1943 1948 1953 1958 1963 1968 1973 1973 1978 1977
-0,05 -0,1 -0,15
Cohort
FIGURE 10
COHORT EFFECTS, REGRESSION INCLUDING EDUCATION AND RETURNS TO PURE DEATON, EDUCATION VARIABLES DEFINED BY CELL 0.15 0.1 0.05 0 -0.05 -0.1 -0.15 Cohort
1905
1909
1913
1917
1921
1925
1929
1933
1937
1941
1945
1949
1953
1957
1961
1965
The key conclusions from this exercise are as follows. When controlling for the number of years of education (mean and dispersion) the increase in inequality disappears and the drop in inequality is accentuated, showing that education variables fully explain the increase in inequality but do not explain the reduction. Actually, the evolution of the education variables would have justified a further increase in inequality that is not observed. Hence the unexplained fall
1969
239
once education variables are incorporated is larger. When we also control for returns (mean and dispersion, see Figure 10) then we observe that returns do explain part of the decline in inequality. What is interesting is that the evolution of the number of years of education and of returns completely explains the increase in inequality from 1936 to 1951. However, they would have lead to a further increase in inequality during the period when cohort income inequality falls, hence the drop is accentuated when one includes education variables. This is so with the exception of the last few years (in particular the last-1978) where returns do appear to explain the drop in inequality. It is also interesting to note that when we use the data by cell, the education variables explain 10 points of the Gini index that in the other specifications are explained by the age effects. Age effects go from zero to 35 points as a cohort ages when education variables specified by cell are not included, but go from zero to 25 points when they are. This proves that there is a possible absorption of cohort effects by the age effects in specifications such as the pure Deaton regression. We can conclude that all the specifications we have tried, even though they do a fair job of explaining the rise, fail to explain the fall. Up to now we have constrained the regressions to have the same coefficients for the three periods we are attempting to explain. We will now relax this assumption. 9. What Explains the Drop in Income Inequality? If education variables do not explain it, then what is it that explains the drop in cohort income inequality? One candidate that we have left out of our human capital framework is experience. A drop in the returns to experience would lead to a decrease in cohort inequality. And there is evidence that this has occurred in the data. People born between 1953 and 1957 entered the labor market in the early to mid seventies. At that time other studies (Haindl (1987), Sapelli (1990), Gill, Haindl, Montenegro and Sapelli (2002), and Lima and Paredes (2004)) show an important change in the behavior of the labor market with a significant increase in turnover. A higher rate of turnover implies less accumulation of specific human capital and hence less increase of income with age. This is very good evidence to proceed to the next step, that is, to split the sample. What we are saying here is that the age effects may change as cohorts become younger. But in our methodology we are forcing the age effects to be the same across all generations. We split the sample into 3: the first period when cohort effects are constant (1902-1929), the period when they are rising (1929-1959) and the period when they are falling (1959-1978)13. Hence we run the same regressions we were running beforehand (with number of years of
13
We use the pure Deaton cohort effects for the regression without moving averages to define these periods. We tried with a split in 1951 instead of 1959 but the latter seams to approximate better the structural break.
240
education and returns) for the three subperiods (i.e. three separate regressions). These are our final regressions. FIGURE 11
COHORT EFFECTS FROM SPLIT SAMPLE
0.1 0.08 0.06 0.04 0.02 0 0.02 0.04 0.06 0.08 0.1 Cohort
1909
1913
1917
1921
1925
1929
1933
1937
1941
1945
1949
1953
1957
1961
1965
1969
1973
As can be seen in Figure 11, this specification manages to eliminate most of the cohort effects, and the ones that continue to be there do not show any pattern. This specification, therefore, has achieved to explain both the rise and the drop: the rise through education variables and the drop through the returns to experience. However it should be noted that splitting the sample diminishes the power of our tests. This can be seen by the wide standard errors for the first subperiod or the wild change from significant negative to significant positive cohort effects in the case of our third subperiod. We proceed to look at the differences in age effects for the three periods in Figure 12. Figure 12 shows that the age inequality profile was much steeper during the intermediate period (when inequality was rising) than in the other two periods. In the intermediate period the inequality due to age effects added up to 15 points to the Gini index. However in the other two periods they did not add more than 10 points. These five points are key to the explanation in the fall of inequality. Although the higher education of the latter sample would justify a pattern of much steeper increase in income by age, we observe a less steep increase than that observed in the mid period. This shows that returns to experience must have decreased. Since then the fanning out of income with age is lessened, cohorts are more equal. The cause of this decrease in the returns to experience could be the increase in turnover (but this is a hypothesis that requires testing, we leave this for future work).
1905
1977
241
FIGURE 12
AGE EFFECTS FOR 3 PERIODS IN SPLIT SAMPLE 0.2 0.175 0.15 0.125 0.1 0.075 0.05 0.025
19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63 65
0 0.025 0.05
Age
Cohorts born 1904-1929 Cohorts born 1929-59 Cohorts born 1959-78
10. Conclusion For cohorts born from 1902 to 1978 one observes three periods in the evolution of cohort effects in income distribution. At first it is flat. Starting with cohorts born in the early thirties there is a deterioration that continues up to cohorts born in the fifties. The income distribution then undergoes an improvement, starting with cohorts born in the early fifties and especially after the cohort born in 195914. The improvement is statistically significant and numerically important: of about 9 points in the Gini index (or about 20% considering a Gini index of 0.50). We attempt to explain this evolution controlling for variables that are important in the human capital framework: mean and deviation of education levels and rates of return. This evolution does help understand the evolution of inequality, and in particular it explains fully the increase from the thirties to the fifties. Once one includes these variables, the cohort effects in those decades do not show deterioration in income inequality. However, these variables do not explain the recent improvement. In some regressions the drop even increases, implying that education variables in the period would have contributed to a further deterioration. Hence, once one controls for education variables, the gap to explain is even larger. The only variable that helps in explaining the decrease
14
About half of the working population was born before this cohort. Hence this improvement at the margin is not yet discernable in the stock. But the improvement is already present, and visible if one looks at the marginal instead of the stock distribution.
242
is the evolution of rates of return, specially the drop in means and dispersion that has occurred recently (though this does not explain what happened when the decline in inequality started). What appears to be driving the improvement is a reduction in the return to experience and hence a decrease in the fanning out that occurs after persons with different educational levels leave school. There has been a reduction in the increase of inequality with age. References Deaton, A. (1997). The Analysis of Household Surveys: A Microeconomic Approach to Development Policy, Baltimore, Johns Hopkins University Press. Eckstein, Z. and E. Nagypal (2004). The Evolution of U.S. Earnings Inequality: 1961-2002, Federal Reserve Bank of Minneapolis Quarterly Review, Vol. 28, N 2, December 2004, pp. 10-29. Gill, I. S., E. Haindl, C. E. Montenegro and C. Sapelli (2000). Chile: Has Employment Become more Precarious?, in I. S. Gill and C. E. Montenegro (eds.), Readdressing Latin Americas Forgotten Reform: Quantifying Policy Changes in Argentina, Brazil, and Chile, World Bank. Mincer, J. (1974). Schooling, Experience and Earnings, New York: National Bureau of Economic Research. Paredes, R. & V. Lima (2004). Labor Market Regimes and Mobility through a Markov Chain in Chile, Econometric Society 2004 Latin American Meetings 317, Econometric Society. Sapelli, C. (2007). A Cohort Analysis of the Income Distribution in Chile. Working Paper 290, Department of Economics, Pontificia Universidad Catlica de Chile. Sapelli, C. (2009). Los Retornos a la Educacin en Chile: Estimaciones por Corte Transversal y por Cohortes. Working Paper 349, Department of Economics, Pontificia Universidad Catlica de Chile. Sapelli, C. and W. Mullins (2006), Returns to Schooling by Cohort in the Twentieth Century in Chile. An Analysis Based On Synthetic Panel Data. Mimeo, 2006. Torche, F. (2005). Unequal but Fluid: Social Mobility in Chile in Comparative Perspective. American Sociological Review 70 (3): 422-450.
Estudios de Economa. Vol. 38 - N 1, Junio 2011. Pgs. Castro Getting ahead, falling behind and standing still / Rodrigo243-258
243
Getting ahead, falling behind and standing still. Income mobility in Chile*
El ascensor social. Movilidad de ingresos en Chile
Rodrigo Castro** Abstract This paper analyses household income mobility in Chile between 1996 and 2001. Compared to industrialized and most developing countries, mobility has been quite high. The purpose of this paper is to apply a binomial probit model and split analysis into assessment of individuals and households on the relative income distribution. Main results are that moving from unemployment to employment significantly increases probability of moving up and decreases probability of moving down. Technical-professional education is promoting move up on the relative income scale and it is protecting movement down. An important result is that high-school education decreases probability of degradation. Key words: Mobility, Poverty, Household structure, Chile. Resumen Este artculo analiza la movilidad de ingresos de los hogares en Chile entre los aos 1996 y 2001. En comparacin con los pases desarrollados e industrializados, la movilidad ha sido bastante alta. El propsito de este artculo es aplicar un modelo probit binomial y dividir el anlisis de la distribucin de ingresos relativa entre los individuos y los hogares. Los principales resultados son que el movimiento del desempleo al empleo aumenta significativamente la probabilidad de movilidad ascendente y disminuye la probabilidad de movilidad descendente. La educacin tcnico-profesional promueve la movilidad ascendente en la escala de ingresos relativos y protege el movimiento descendente. Un importante resultado es que la escolaridad secundaria disminuye la probabilidad de descenso. Palabras clave: Movilidad, Pobreza, Estructura de hogares, Chile. JEL Classification: D63, J12, J15, J6.
I would like to thank Professor Aristide Torche and Francisco Ferreira for their helpful comments. Usual disclaimers apply. ** Universidad del Desarrollo.
244
1. Introduction As an upper middle income country with a per capita GDP of $10,874 in 2004 PPP, Chile scores well on international comparisons of social indicators and remains one of the outstanding countries in Latin America in terms of its record in reducing poverty. The human development index is 0.859, which gives it a rank of 38th out of 177 countries. A combination of strong growth, sound macroeconomic policies and well directed social programs have combined to reduce headcount index in more than half during sixtheen years. Despite large advances in poverty reduction, and significant increases in social expenditure, income inequality remains stubbornly stagnant. Furthermore, cross country comparisons place Chile amongst those countries with the highest gini coefficient. Even though poverty and income distribution are key to economic development, an issue that is discussed less, is intertemporal income mobility. This is of particular relevance in the Chilean context. Just comparing income distribution across time cannot answer questions like are the poor getting poorer and the rich richer? or is economic growth benefiting individuals that were initially poor? In order to answer such questions, it is necessary to perform income mobility analyses, tracking the evolution of individual incomes over time and seeing who are the winners and losers during the growth process (Fields et al., 2006). Since according to generally accepted methodology developed by Atkinson (1970) higher levels of inequality decrease welfare, this trend is still a concern for policy makers. This tendency might not be as harmful if we take into account income mobility, which has slightly different impact on welfare. One of the interesting features of mobility is that, long run inequality can be only lower than short run inequality and also when it is assumed usual social welfare function, then higher mobility increases welfare (Shorrocks, 1978). This has implications on international comparisons of income inequality since even though one country might have higher cross section inequality, it can be offset by higher income mobility and exhibit lower long run inequality (Grodner, 2000). Another issue of income distribution dynamics is related to the fact that society might prefer higher probabilities of transitions out from poverty. Also, income mobility is related to the governments policies to reduce poverty and inequality gaps. The importance of demographic and economic events on mobility should be carefully examined. The structure of this paper is the following. Section 2 describes the data and methodology. Section 3 presents the results regarding income mobility in the short run. Section 4 assess the determinants of short run income mobility. Section 5 concludes. 2. Data and Methodology In contrast to the vast theoretical and applied income inequality literature, the literature on the measurement and interpretation of mobility is more limited and generally more ad hoc (Fields and Ok, 1999) Important distinctions are made between relative and absolute mobility. The former examines changes in rank of households between two periods and is thus mainly concerned with the
Getting ahead, falling behind and standing still / Rodrigo Castro
245
ability of individuals to move up (and down) in the rankings of incomes while the latter examines absolute changes in income between two periods and thus is additionally concerned with changes in absolute well-being (and poverty). As far as measures of mobility are concerned, one first needs to distinguish between what Cowell and Schluter (1998) call single-stage and two-stage indices. Single-stage indices consider the entire distribution in both years and examine mobility using that entire distribution, while two-stage indices first allocate individuals to income groups and then examines mobility between these groups. Examples of single-stage indices are the correlation coefficient of incomes between two periods, Shorrocks rigidity index, Fields and Oks measures and Kings measure (Fields, 2001; Cowell and Schluter, 1998 and Woolard and Klasen, 2004) They have the advantage of using all available information inherent in the actual distributions and thus give the most comprehensive assessment of mobility. However, they have the disadvantage of being particularly sensitive to measurement error which is a particular problem when data from only two waves are available1. In simulation studies, Shorrocks rigidity index was least sensitive to measurement error (Cowell and Schluter, 1998). This index compares the Gini of the average income between the periods with the weighted average of the Gini in each period. A value of one would mean no mobility at all, while 0 would indicate perfect mobility. Regarding two-stage indices, the most commonly used measure is the transition matrix and indices derived from it. For a transition matrix, the data are divided into n equally sized income classes (e.g. quintiles or deciles) which are endogenously determined for each year. While sometimes the brackets of a transition matrix are exogenously fixed income classes, the more common method are endogenously determined income groups based on quantiles of the distribution in a given year (such as quintiles, deciles or poverty transitions). The advantage of the transition matrix is that it can summarize mobility at various points in the distribution which is harder to gauge from a single index. It also turns out to be more robust to measurement error (Cowell and Schluter, 1998). There are serious costs as well, including the disregard of important information, such as income changes within a bracket and the different absolute income changes that underly a change in income bracket (Fields and Ok, 1999). This issue can be important in international comparisons of mobility. In a country with low inequality, the same transition matrix may mean much smaller changes in absolute income levels compared to a country with very high inequality. To the extent one wants to capture these absolute changes as well, a transition matrix may not be the right tool. Despite these problems, the advantages of transition matrices are considerable. The choice of income groups in these transition matrices is largely arbitrary and, in general, tends to take the form prevalent in the literature to allow for the comparison of results. The most popular choices seem to be quintiles and deciles. Nevertheless, the choice of groups influences the results. The smaller (in terms of income range) the brackets, the more likely that people will move between brackets and thus
As happens to be the case in the Casen Panel Survey 1996-2001.
246
mobility will appear larger. Thus using deciles usually will generate higher perceived mobility than quintiles. There are relatively few studies on income mobility in developing countries and even fewer that are roughly comparable. The reason why this issue had not been widely studied in developing countries and particular in Chile until recently was the lack of suitable data. In order to study mobility, it is necessary to have longitudinal data tracking individuals or households over time. Collecting this type of data is expensive and for many years it was not generally done in most Latin American countries. Yacub (2000) stated that only 5 of the 44 low human development countries, and 7 of the 66 countries with intermediate human development, according to the UNDP classification, had available panel data. Baulch and Hoddinot (2000) confirmed this panel data shortage. However, the extreme diversity of the panels used (in terms of geographic extension, reference timeframe, type of sampling, welfare indicators, poverty line, etc.) considerably limits the analytical scope of these different case studies, particularly with respect to their comparison dimension. It is difficult to draw general conclusions from prior research, due to the heterogeneity of the samples, data and methodological choices, which restricts the sphere of comparison among countries. Most analyses focus particularly on poverty dynamics rather than on household income mobility more generally (e.g. Jalan and Ravallion 2000; Dercon and Krishnan, 2000; Scott, 2000; Justino and Lichfield, 2002, McCulloch and Calandrino, 2002; Woolard and Klasen, 2004; Fields et al., 2006). This paper will address household income mobility. These studies generally suggest that income mobility in developing countries is higher than in industrialized countries, particularly at the botton end of the distribution (e.g. Dercon and Krishnan, 2000; Fields, 2001; Contreras et al., 2004; Paredes and Zubizarreta, 2005). They also seem to suggest increasing mobility over time in most places. Panel data from Peru based on expenditures points to increased mobility in the 1990s (Fields, 2001). Data from rural China point towards rapidly increasing mobility from very low levels in the 1980s (Nee, 1994) and generally very high mobility at the low end of the distribution (McCulloch and Calandrino, 2002). These studies as well as studies from Malaysia suggest that education, changes in employment and demographic composition of the household play a key role in explaining existing mobility and in distinguising between the transient and the chronic poor (Fields, 2001). In Chile, income mobility studies have been conducted for rural sector by Scott and Litchfield (1994) and Scott (2000). Both papers are based on a small longitudinal study of rural households between 1968 and 1986. The authors analyze mobility of household per capita income, with and without government transfers. The panel consists of only two observations in time, but those capture the impact of Chiles liberalization reforms after 1974. Scott and Litchfield (1994) study income mobility and the evolution of inequality over time. The study shows that not only there were more upward than downward movers, but the extent of upward mobility (in terms of number of classes transited) was greater than the extent of downward mobility. Also, the authors model the determinants of directional income movement by using a linear regression and an ordered logit model (in which the dependent variable is whether the household moved to a higher income class, stayed in the same income class, or moved to a lower income class). The variables found to be significant determinants of
247
upward income movement are age and education of the household head, amount of land owned, and per capita household income in the base year (the richer the household in 1968, the smaller the growth of income from 1968 to 1986). Scott (2000) complements the previous findings by analyzing the extent of movements out of poverty for the households in the sample. The results show that, while there was upward mobility during those years, around 70% of the initially poor households were below the poverty line in 1986. Similarly, 64% of the non-poor households stayed above this line eighteen years later. Moreover, on 2001 the Ministry of Planning conducted the first panel survey on a national basis in Chile. This survey collects follow up data on 4700 Chilean households in the III, VII, VIII and Metropolitans regions between 1996 and 2001. The data is representative for the four regions surveyed , which accounts for 60% of the population and 64% of national GDP. This survey has been used in different studies. Aguilar (2002), uses a descriptive analysis and highlights poverty dynamics. Castro and Cheyre (2004) using a multivariate analysis find four types of poverty traps, associated with large initial household size, poor initial education, poor initial asset endowment and poor initial employment access. Contreras et al. (2004) analyze income mobility and assess the determinants of poverty dynamics. Their main results are that there is high mobility within all income distribution, but top decile. Finally, Paredes and Zubizarreta (2005) analyze the policy performance and determinants of mobility of the extremely poor, as compared with those poor but not extremely poor. The authors conclude that mobility of some extremely poor is higher than usually thought and determinants to leave and come into extreme poverty are different to those which determine mobility to and from non extreme poverty. 3. Income Mobility in the Short Run I begin by reporting Shorrocks rigidity index using the Gini coefficient. The Ginis for the two years are presented as well as those for the average income and rigidity index (Table 1). The rigidity index indicates a fairly high degree of mobility, when compared to developed countries where the rigidity index is usually around 0.95 or above for countries such as the US, United Kingdom, Germany or Sweden (e.g. Jarvis and Jenkins, 1998; Eriksson and Pettersson, 2000) it is closer to countries such as Spain in the 1990s (Canto, 2000). TABLE 1
INDICES OF RIGIDITY
Deciles 1-10 Gini 1996 Gini 2001 Average Gini Average Income 1996 ($) Average Income 2001 ($) Rigidity Index 0.56 0.53 0.50 93,067 143,160 0.92
Deciles 1-9 0.39 0.38 0.35 55,407 112,204 0.91
Deciles 1-5 0.26 0.25 0.23 27,026 80,814 0.90
248
While these statistics give a lot of information, is not enough. We want a more disaggregated view, using transition matrices. The quintile mobility matrix (Table 2) shows the distribution of households by quintile for 1996 and 2001. It can be seen that 49.9% of households who were in the richest quintile in 1996 remained there in 2001 and 23% moved down just one quintile. Likewise, 40.6% of those who began in the poorest quintile were still there 5 years later and another 25.8% had moved up just one quintile. It is immediately evident that there is less mobility in the top and bottom quintile than in the middle of the distribution2. This is, however, unsurprising given that the bottom (top) quintile can only stay in the same quintile or move up (down) also, furthermore the income range that make up the quintile is much larger for the richest quintile where the right hand tail is particularly large which is the reason why persistence in that group is particularly high. TABLE 2
TRANSITION MATRIx BY QUINTILES (%)
Quintile 1996 1 2 3 4 5 Total
Quintile 2001 1 7.87 5.41 2.92 1.40 0.94 18.55 2 5.00 5.50 5.09 3.12 1.37 20.08 3 3.50 4.72 4.88 4.13 3.15 20.39 4 1.70 2.02 4.83 7.30 4.64 20.48 5 1.32 2.54 2.28 4.29 10.07 20.49 Total 19.39 20.20 20.01 20.24 20.17 100.00
These figures also suggest quite a high degree of short term income mobility among Chileans households which is certainly higher than that observed in most industrialized countries (e.g Jarvis and Jenkins, 1997), but also higher than in rural China between 1978 and 1983, Malaysia between 1967 and 1976, South Africa between 1993 and 1998 and Peru in the 1980s (Fields, 2001; Woolard and Klasen, 2004) It is quite similar, however, to rural China between 1983 and 1989 although the structure of mobility appears to be somewhat different. In rural China, (downward) mobility from the top quintile is higher than in Chile. This may partly be due to the fact that overall income inequality in rural areas was much lower to begin with so that the income change required to change income bracket is smaller than in Chile.
Even though some individuals are be dropped in matching the files, we decided to use full files in order to be able use proper weights - in this way assignment to quintiles is correct relative to other demographic groups. Thus, because of this attrition our transition matrix does not necessarily show that there are 20% in each group.
249
TABLE 3
DIFFERENCES IN QUINTILE BETWEEN SECOND AND FIRST YEAR (%)
Differences in quintiles 4 3 2 1 0 1 2 3 4
Chile 1996-2001 0.69 2.31 5.29 21.83 41.09 19.45 7.64 1.37 0.33
Germany 1985-1987 0.22 0.83 1.78 11.99 68.86 13.21 2.18 0.72 0.21
US 1985-1987 0.22 0.57 2.16 12.13 68.03 14.39 2.04 0.4 0.06
Table 3 compares relative income mobility between Chile, Germany and US. We use a simple measure of immobility which is just a proportion of individuals on the diagonal. For Chile it is 41%, while Germany and US are 68.8% and 68.0% respectively. Also, we can compute share of individuals moving by one quintile and by more than two (any direction) this might indicate patterns of dynamics. For Chile move by one (up or down) is 41.2%, while for Germany and US are 25.2% and 26.5% respectively. For moves higher than two quintiles, Chile it is 17.6%, Germany 5.9% and for US 5.4%. Thus, from those robust numbers we can conclude that unconditional relative mobility was higher in Chile (between 1996 and 2001) than Germany and US (between 1985 and 1987). Table 4 shows a transition matrix by income deciles. It evidences a high mobility in all income distribution, but the highest and the lowest decile. Only 20.7% of the households remained in its original income decile after five years. This matrix shows that Chilean income distribution in the short run is very mobile. A key issue is that this income mobility does not include the highest decile. Becker (1980) found that being rich is sticky, even though being poor is not. A poor face higher probability to leave his economic condition, but a rich has a higher probability to keep his economic condition. Only 27.3% of who were in lowest decile in 1996 were in the same decile in 2001. That means that the remaining households were richer in 2001. In fact, 22% of them moved up to the upper half of the distribution (from deciles six to ten). The richest (highest decile) had a higher probability to keep in that decile (42.5%). However, 15.9% of them move down to the lower half of the distribution (deciles one to five). This shows that a rich household could become poor between five years. In Germany and US between 1990 and 1995 it was found that 23% of the population remained in the diagonal. Therefore, the above confirms that Chile has higher income mobility, even more than developed countries.
TABLE 4
TRANSITION MATRIx BY DECILES (%)
250
Decile 1996 4 0.81 1.54 1.95 1.46 1.15 0.73 1.09 0.45 0.64 0.24 10.08 9.92 10.01 10.11 9.88 10.02 1.16 0.83 0.84 1.48 1.05 1.19 1.38 0.54 1.01 0.43 0.78 0.72 1.35 0.92 0.84 1.67 1.13 0.97 0.63 1.00 0.45 0.59 0.71 0.53 1.22 1.49 2.07 1.43 0.85 0.77 0.40 0.28 0.33 0.40 0.73 1.26 0.93 2.66 2.25 0.63 0.41 0.41 0.45 1.33 0.53 0.81 1.08 1.41 1.85 1.74 0.16 0.35 0.31 0.38 0.35 0.52 0.54 1.15 1.97 4.24 9.98 5 6 7 8 9 10
Decile 2001 Total 10.00 10.01 10.00 10.10 9.90 10.02 9.98 10.06 9.97 9.97 100.00
1 2 3 4 5 6 7 8 9 10
2.73 1.96 0.95 1.11 0.80 0.76 0.53 0.34 0.36 0.48
1.73 2.04 2.14 1.51 0.96 0.78 0.55 0.29 0.20 0.19
1.38 1.27 0.95 0.98 2.26 0.81 0.68 0.81 0.20 0.25
Total
10.02
10.39
9.59
251
4. Determinants of Income Mobility in the Short Run In this section I apply a methodology developed by Finnie and Gray (1998) who used binomial probit model and split analysis into assessment of individuals and households on the relative income distribution (quintiles). This framework is applied to equivalised household income to measure the degree of mobility observed between 1996 and 2001 for Chilean households in four different regions. Given that we have only two observations per household, measurement error might well influence the results. Consequently, we use a variety of procedures to test and, to the extent possible, correct for measurement error and examine the robustness of these results. Our controls for measurement error do not necessarily provide unbiased estimates but will help us get a sense of the magnitude of possible biases and thus the robustness of results, a strategy suggested among others by Bound et al. (2001) Almost 70% of the sample reported that household income had increased over the period. Real median adult equivalent income increased by 41% over the five year period. While some of this discrepancy can be real and relates to the timing of the survey (seasonality and business cycle) changes in perceptions of permanent incomes and the large role of transitory incomes, this large discrepancy in levels and trends raises some questions about the reliability of the data. These discrepancies could also indicate that measurement error is significant. Therefore, I purge the 1996 and 2001 labor income data by specifying an income regressions of hourly income on gender, location, industry, age, age square, education and throwing out all observations that are outside two standard deviations from the point estimate of this income regression. The income regressions have a good fit (adjusted R2 around 0.42) and confirm the usual findings from the human capital literature. Using this procedure, I end up eliminating about 3% of observations. Also, I use an instrumental variable approach to measurement error. Using a regression of household adult equivalent income on household size, demographic structure, average education, age of household head, female headship, location, and other asset ownership, and the employment and unemployment situation of adults, I predict household incomes in 1996 and 2001 and assess mobility using these predicted income. Therefore, I throw away quite a lot of true mobility that would not be captured by these regressions but this approach should give me sense of the maximum extent to which measurement error affects incomes. Carter and May (2001) interpreted these differences between predicted and actual incomes as stochastic features of income that can make households stochastically poor or non-poor. After the above procedure, I use probit model for studying influence on probability of movements up and down. For movement up we have individuals who either stayed in quintile or moved up, and the same for movers down (stayers are a comparison group in both cases). Thus, I will estimate two models for moving up and down using standard probit specification: Pr(m) = ( + Xi + Zij) where is a standard normal cumulative distribution function, i indexes first period and j second period; m denotes indicator of whether individual moved
252
up or stayed, or whether moved down or stayed; Xi represents characteristics of an individual/household which do not change over time and are taken at first period; Zij represent characteristics of an individual/household which do change over time. For continous variables marginal effects are computed at mean values of all variables with: (a + bXi + cZij) / xk = (a + bXi+cZij) bk where is standard normal density function; k indexes independent variable; a,b,c are estimates of , , ; and Xi, Zij are sample means of Xi, Zij. For discrete variables I compute difference in probability (those are indicated with stars in output tables) with all other variables being evaluated at mean values:
[(a + bXi + cZij)] | x(k) = 1 [(a + bXi + cZij)] | x (k) = 0

where xk is being set to 0 and 1 respectively. I select the same individuals and there is clear correlation between individuals since these are the same individuals whenever they are in the consecutive year. Since in this case standard errors are incorrect, I computed robust standard errors3. The main dependent variable is difference in quintiles between 1996 and 2001. In order to construct it, I first compute quintiles for each year using all individuals who had positive total income in each single year. The analyses focus on four classes of variables: labor market experience; changes in family composition; family characteristics; and individual characteristics. I also compute distance from quintiles boundaries from lower and upper, as well as dummies for starting quintile and transition between self-employment and other employment. I create a variable which indicates difference between being in quintile of first year and quintile in the second year, and compute mean of this variable for different independent variables. The results of the marginal effects from the estimated probit models for moving up and down are presented in Tables 5 and 6 respectively (in Appendix). All estimates are computed at sample means, while for dichotomous variables I compute difference in probability between variable having value 0 or 1. Results for changes in employment status seem to reveal some effects which were not apparent when I use means of difference in quintiles (ANOVA tests). The effect of moving from unemployment to employment, increases probability of moving up by 9.5% and decreases probability of moving down by 5.3%. The highest effect on moving down is experienced by individuals who change from employment to unemployment, as this increases probability of degradation by 22.6%. Regarding age effect, the group between 35 and 44 faces the higher probability of moving up.
3
In all estimations we use analytical weights which use original weights from first year, but preserve the same number of observations.
253
Higher education is promoting move up on the relative income scale by almost 8%, and it is protecting movement down by 9%. It can also be seen that high-school education does not help to promote positive mobility. On the other hand, high-school education decreases probability of degradation by 6%. It is likely to move up if the household lives in the urban sector by 14%, while it decreases the probability to move down. On the other hand, if the head of household is married on both years, increases the probability of moving up by 10%. If the head of the household is male in 1996 and 2001, it is less likely that the household moves up by 9%, but also decreases the probability of moving down. However, if the head of the household is female on both years, it is more likely that the household moves up by 9%. Finally, as long the share of children in household increases, it is less likely to move up and more likely to move down. There are some challenges ahead regarding the methodology used in this section and its main findings. The results could be improved by better specification of the set of independent variables. For example, barely significant differences of impact by age and education groups might reveal more information when interacted with gender (as suggested by Finnie and Gray, 1998). However, future research should probably be continued on longer periods of time as that might reveal more stable patterns. On 2007 it will be available the third wave of the original panel dataset which will cover 1996-2001-2006. 5. Conclusions The motivation of this paper is the evidence that despite large advances in poverty reduction and significant increases in social expenditure, income inequality remains stubbornly stable. However, this phenomena, Chile experiences high level of short run income mobility. The purpose of this paper is twofold. First, to present evidence about the short run mobility and second analyze its determinants. We find that the effect of moving from unemployment to employment, significantly increases probability of moving up and decreases probability of moving down. Higher education is promoting move up on the relative income scale and it is protecting movement down. An important result is that high-school education decreases probability of degradation. References Aaberge, Bjoklund, Janti, Palme, Pedersen, Smith and Wennemo (1996). Income inequality mobility and income mobility in the Scandinavian countries compared to the United States. University of Stockholm Discussion Paper N 98. University of Stockholm. Aguilar (2002). Dinmica de la Pobreza: Resultados de la Encuesta Panel 1996-2001. Documento no publicado, Mideplan. Baulch and Hoddinott (eds.) (2000). Economic mobility and poverty dynamics in developing countries. Frank Cass Publishers.
254
Bound, Brown, and Methiowetz (2001). Measurement Error in Survey Data. Handbook of Econometrics, Volume 5. Amsterdam: North-Holland. Bound and Krueger (1991). The Extent of Measurement Error in Longitudinal Earnings Data: Do two Wrongs make a Right?. Journal of Labor Economics 9. Burkhauser and Poupore (1996). A cross-national comparison of permanent inequality in the United States and Germany. Review of Economics and Statistics LxxIx (1). Cant (2000). Income mobility in Spain: how much is there?. Review of Income and Wealth 46. Castro y Cheyre (2006). Midiendo la movilidad de ingresos y la dinmica de la pobreza en Chile, ed. Rosita Camhi y Rodrigo Castro, en La Nueva Realidad de la Pobreza en Chile, Libertad y Desarrollo. Contreras, Cooper, Herman y Neilson (2004). Dinmica de la Pobreza y Movilidad Social: Chile 1996-2001. Documento de Trabajo, Universidad de Chile. Cowell and Schluter (1998). Income Mobility: A Robust Approach. STICERD Discussion Paper N 37. London: LSE. Cowell and Schluter (1998). Measuring Income Mobility with Dirty Data. CASE/16. London: LSE. Deaton (1997). The Analysis of Household Surveys. Baltimore: Johns Hopkins University Press. Eriksson and Pettersson (2000). Income Distribution and Income MobilityRecent Trends in Sweden. In Hauser, R. and I. Becker (eds.). The Personal Distribution of Income in an Historical Perspective. Berlin: Springer. Fabig (2000). Labor Income Mobility- Germany, the USA and Great Britain Compared. In Hauser, R. and I. Becker (eds.). The Personal Distribution of Income in an Historical Perspective. Berlin: Springer. Fields (1998). Accounting for income inequality and its change. Mimeo. Ithaca: Cornell University. Fields (2001). Distribution and development: a new look at the developing world. Cambridge: MIT Press. Fields, Duval, Freije and Sanchez (2006). Income Mobility in Latin America, draft. Fields and Ok (1999). The Measurement of Income Mobility: An introduction to the literature. In Silber, J. (ed.). Handbook of Income Inequality Measurement. Boston: Kluwer. Fields and Ok (2003). Household Income Dynamics. A Four Country Study. Journal of Development Studies. Forbes (2000). A Reassessment of the Relationship between Inequality and Growth. American Economic Review 90. Gottschalk (1982). Earnings mobility: permanent change or transitory fluctuations?. Review of Economics and Statistics 64. Grodner (2000). Factors influencing earnings mobility in USA and Germany (1985-87). A Bivariate Probit Model. Center for Policy Research, Syracuse University. Hauser and Fabig (1997). Labor Earnings and Household Income Mobility in Reunified Germany: A comparison of the Eastern and Western States.
255
Income Security Policy Series Paper N15. Center for Policy Research. Syracuse University. Heckman and Carneiro (2003). Human Capital Policy. NBER Working Papers 9495. Heckman and Dimitry (2004). Skill Policies for Scotland. University of Chicago. The Allander Series. Jalan and Ravallion (2000). Is Transient Poverty Different? Evidence for Rural China. Journal of Development Studies 36. Jarvis and Jenkins (1995). Do the poor stay poor? New evidence about income dynamics from the British household panel survey. Occassional Paper 95-2, ESRC Research Centre on Micro-Social Change. Jarvis and Jenkins (1997). Low income dynamics in 1990s Britain. Fiscal Studies 18. Jarvis and Jenkins (1998). How much income mobility is there in Britain?. Economic Journal, 108. Jenkins and Rigg (2001). The Dynamics of Poverty in Britain. Department for Work and Pensions Research Report N 157. London: Corporate Document Services. Justino and Lichfield (2002). Poverty Dynamics in Rural Vietnam: Winners and Losers during Reform. Paper presented at the 27th Biannual Conference of the IARIW, Stockholm. Lillard and Willis (1978). Dynamic aspects of earnings mobility. Econometrica 46. McCulloch and Calandrino (2003). Vulnerability and Chronic Poverty in Rural Sichuan. World Development. Nee (1994). The Emergence of a Market Society: Changing Mechanisms of Stratification in China. Working Papers on Transitions from State Socialism, Cornell University. Neumark and Wascher (1995). Minimum Wage Effects on Employment and School Enrollment. Journal of Business and Economic Statistics. Paredes and Zubizarreta (2005). Focusing on the Extremely Poor: Income Dynamics and Policies in Chile. Documento de Trabajo. Departamento de Ingeniera Industrial. Pontificia Universidad Catlica de Chile. Scott and Litchfield (1994). Inequality, Mobility and the Determinants of Income among the Rural Poor in Chile, 1968-1986. Development Economics. Scott (2000). Mixed Fortunes: A Study of Poverty Mobility among Small Farm Households in Chile, 1968-1986. Journal of Development Studies 36. Van Kerm (2002). Tools for the analysis of income mobility in Stata. 2nd Dutch Stata Users Meeting, Maastricht. Woolard and Klasen (2004). Determinants of Income Mobility and Household Poverty Dynamics in South Africa. IZA Discussion Paper N 1030. Yaqub (2000). Poverty dynamics in developing countries, mimeo, University of Sussex. Yaqub (2002). Chronic poverty: scrutinizing estimates, patterns, correlates and explanations. Working Paper 21, CPRC, Manchester.
AppenDIx
256
TABLE 5
z 10.11 19.82 52.86 19.97 11.58 11.80 21.55 22.04 6.14 13.71 16.85 4.77 18.52 11.02 13.15 11.61 3.76 5.76 4.76 22.01 21.25 204.59 46.00 22.55 1.26 40.81 18.08 4.14 2.66 38.13 21.90 15.58 8.38 10.68 10.69 158.00 31.87 93.20 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.207 0.000 0.000 0.000 0.008 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 .774092 470.259 .858061 .123525 .313857 .532448 .251007 .248779 .603082 .208033 .034428 .020965 .094813 .009571 .026114 .001518 .034108 .183114 .273002 .204321 .157258 149.944 .502604 .417188 .055981 .631676 .042052 .003215 .010441 .093377 .009715 .004063 .037359 .756092 .178764 .063228 .562219 .334232 .104578 .004491 .140224 .11949 .071327 .051295 .043913 .046559 .19474 .318971 .318681 .150312 .350955 .256004 .280564 .287374 .073928 .080637 .024074 .130143 .087536 .222323 .26308 .131216 .004489 .103431 .091155 .142675 .006318 .159852 .172846 .209361 .073667 .113173 .090648 .127049 .058599 .51009 P>|z| xbar [ 95% C.I. ] .151715 .005477 .15011 .099296 .050921 .07164 .052773 .055733 4 .100402 .252394 .278771 .067786 .308645 .197025 .22868 .231534 .024203 .040394 .058174 .156006 .10567 .218112 .242304 .110587 .020537 .113624 .113864 .055675 .043182 .177439 .207069 .170991 .11996 .077672 .13221 .123925 .066229 .489042
MARGINAL EFFECTS IN PROBIT MODEL (MOVING UP) MODEL 1
Up
dF/dx
Robust Std. Err.
gender* agehh urban* r7* r8* r13* q96_2* q96_3* employed* inact_1* unemp_2* emp_2* inact_2* inact_3* emp_3* unemp_3* agecat1* agecat2* agecat3* agecat4* agecat5* Pssdbii edss_01* edss_02* edss_04* marriage* divorce* marriage_1* divorce_1* widow* widow_1* maless* femass* maless_1* femass_1* Chhs howner* sharech obs. P pred. P
.1281465 .0049842 .1451672 .1093931 .0611241 .0614676 .0483433 .051146 .147573 .2856827 .2987263 .0948491 .3298001 .2265142 .0056219 .259454 .0490653 .0605154 .0411241 .1430747 .0966031 .2202174 .252692 .1209015 .0802421 .1085273 .0025096 .0991752 .0247503 .1686456 .1899574 .1901761 .0968138 .0954226 .111429 .1254869 .0624142 .4995657 .3955202 .3639907
.012025 .0002515 .002522 .0051517 .0052056 .00519 .0022603 .0023401 .0240673 .0169841 .0101814 .0210529 .0107936 .015046 .0132359 .014245 .0126853 .0102661 .008699 .0065978 .0046262 .0010741 .0052999 .0052627 .0063843 .0026003 .005793 .0221943 .0094043 .0044865 .0087306 .0097885 .0118097 .0090566 .010603 .0007971 .0019466 .0053695
(at xbar)
(*) dF/dx is for discrete change of dummy variable from 0 to 1. z and P>|z| correspond to the test of the underlying coefficient being 0.
TABLE 6
MARGINAL EFFECTS IN PROBIT MODEL (MOVING DOWN) MODEL 2

z 15.95 0.51 62.41 4.25 2.41 15.98 164.02 167.18 150.08 14.22 8.84 11.40 2.82 10.97 4.07 38.51 13.60 5.99 10.08 14.06 197.92 47.51 35.65 20.39 13.81 12.50 6.89 29.75 8.13 9.58 11.99 38.29 23.06 33.91 151.97 25.55 69.56 0.000 0.612 0.000 0.000 0.016 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.005 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 47.87 .217839 .876135 .111427 .290743 .566693 .201448 .19966 .200874 .619091 .201164 .032133 .099646 .008893 .022163 .165442 .270077 .212591 .161603 .161722 200.276 .450576 .421738 .088971 .633519 .045093 .0038 .009173 .096101 .010473 .003332 .032657 .767117 .177179 .020768 .587439 .30804 .000549 .002673 .042307 .004354 .002938 .012495 .996614 .900898 .317136 .027233 .013406 .019347 .001073 .015131 .008425 .038902 .010701 .004419 .012235 .026191 .113195 .027959 .019287 .010978 .005305 .007668 .011369 .019592 .003431 .009368 .017444 .021909 .044662 .028053 .015918 .006916 .041961 P>|z| x-bar [ 95% C.I. ] .000433 .004504 .038167 .001746 .000344 .009755 .997359 .908725 .324382 .020576 .009575 .032628 .008223 .012626 .003673 .046177 .015032 .009276 .019668 .038213 .120048 .031899 .022564 .009375 .003892 .005809 .024718 .017981 .005857 .006666 .01515 .020002 .036364 .025147 .017299 .005814 .046389
Down
dF/dx
Robust Std. Err.
agehh genderhh* urban* r7* r8* r13* q96_2* q96_3* q96_4* employed* inact_1* unemp_2* inact_2* inact_3* emp_2* agecat2* agecat3* agecat4* agecat5* agecat6* pssdbii edss_01* edss_02* edss_04* marriage* divorce* marriage_1* divorce_1* widow* widow_1* maless* femass* maless_1* femass_1* chhs howner* sharech obs. P pred. P
.0004914 .0009155 .0402368 .0030502 .0016407 .011125 .9969862 .9048116 .3207587 .0239045 .0114906 .2259877 .0046477 .0138786 .0534861 .0425395 .0128664 .0068475 .0159513 .0322018 .1166214 .0299293 .0209256 .0911764 .0045985 .0067385 .0180434 .0187865 .0046439 .0080172 .0162967 .0209555 .0405126 .0266004 .0166087 .0063652 .0441748 .329458 .019567
.0000296 .0018307 .0010561 .0006652 .0006617 .0006988 .00019 .0019968 .0018486 .0016982 .0009773 .0033881 .0018241 .0006391 .0012122 .0018557 .0011048 .0012393 .0018961 .0030669 .0017484 .0010052 .0008361 .0004088 .0003603 .0004743 .0034053 .0004108 .0006189 .0006894 .0005852 .0004863 .0021169 .0007413 .0003523 .0002811 .0011297
(at xbar)
(*) dF/dx is for discrete change of dummy variable from 0 to 1. z and P>|z| correspond to the test of the underlying coefficient being.
257
258 Variables agehh genderhh urban r7 r8 r13 q96_2 q96_2 q96_3 q96_4 employed inact_1 unemp_2 inact_2 inact_3 emp_3 agecat2 agecat3 agecat4 agecat5 agecat6 pssdbii pssdtii edss_01 edss_02 edss_04 marriage divorce marriage_1 divorce_1 widow widow_1 maless femass maless_1 femass_1 chhs howner sharech : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
head of household age dummy, head of household gender (male=1, female=0) dummy, zone (urban=1; rural=0) dummy, VII Region dummy, VIII Region dummy, RM Region dummy, beginning at quintile 1 dummy, beginning at quintile 2 dummy, beginning at quintile 3 dummy, beginning at quintile 4 dummy, employed in 1996 and 2001 dummy, employed in 1996 and out of labor force in 2001 dummy, employed in 1996 and unemployed in 2001 dummy, out of labor force in 1996 and 2001 dummy, unemployed in 1996 and out of labor force in 2001 dummy, out of labor force in 1996 and employed in 2001 dummy, age group, 25-34 dummy, age group, 35-44 dummy, age group, 45-54 dummy, age group, 55-64 dummy, age group, +65 distance from bottom of quintile distance from top of quintile dummy, primary education dummy, secondary education dummy, tertiary education dummy, married both years dummy, married in 1996, divorce in 2001 dummy, single in 1996, married in 2001 dummy, divorce in 1996, divorce in 2001 dummy, married in 1996, widow in 2001 dummy, widow in 1996, widow in 2001 dummy, head of household male in 1996 and 2001 dummy, head of household female in 1996 and 2001 dummy, head of household male in 1996 and female in 2001 dummy, head of household female in 1996 and male in 2001 difference number of people in household between 1996 and 2001 dummy, owner of dwelling share of children in household in 1996
Estudios de la prominente Juan Pablo Valenzuela, Suzanne Duryea ExaminandoEconoma. Vol. 38/ - N 1, Junio 2011. Pgs. 259-293
259
Examinando la prominente posicin de Chile a nivel mundial en cuanto a desigualdad de ingresos: comparaciones regionales*
Examining the prominent position of Chile in the world in terms of income inequality: regional comparisons
Juan Pablo Valenzuela** Suzanne Duryea*** Resumen En el contexto latinoamericano, Chile tiene el nivel ms alto de ingreso per cpita y del ndice de desarrollo humano, aunque la distribucin del ingreso es bastante desigual. A diferencia de Uruguay, Chile tiene una de las distribuciones de ingreso ms desiguales de la regin. En el 2003, Chile tena un coeficiente de Gini de 8,5 puntos ms altos que Uruguay. Usando microsimulaciones, el anlisis muestra que la mayor parte de la diferencia relativa a la distribucin del ingreso proviene de los hogares ms ricos, particularmente aquellos que pertenecen al 2% ms alto. Dichos hogares obtienen una mayor proporcin de recursos provenientes de ingresos no laborales. Al mismo tiempo, la diferencia en retornos a la educacin superior explica otro 20% de las diferencias de ingreso entre Chile y Uruguay. Las condiciones sociales como los beneficios de seguridad social y la participacin de la mujer en el mercado del trabajo no son significativas para explicar las diferencias entre ambos pases. Finalmente, este artculo muestra que el ajuste de la informacin de los ingresos de la encuesta de hogares chilena a las cuentas nacionales explica un tercio de la brecha entre los coeficientes de Gini de Chile y Uruguay. Sin el ajuste a las cuentas nacionales, la brecha en la distribucin de ingresos entre ambos pases disminuira en tres puntos la estimacin anterior. A pesar de esta significativa reduccin las razones para explicar la brecha permaneceran idnticas frente a anlisis anteriores. Palabras clave: Microsimulacin, Distribucin del ingreso, Desigualdad, Pensiones, Participacin laboral, Retornos a la educacin.
Los autores agradecen a Mara Victoria Rodrguez por su excepcional asistencia investigativa. Este documento refleja las opiniones de los autores y no representa las opiniones del BID. Cualquier error es responsabilidad de los autores. ** Universidad de Chile. *** Banco Interamericano de Desarrollo.
260
Abstract In the Latin American context, Chile has the highest level of per capita income and the human development index, though the distribution of income is quite unequal. Unlike Uruguay, Chile has one of the more unequal income distributions of the region. In 2003, Chile had Gini Coefficient of 8.5 points higher than Uruguay. Using micro/simulations, the analysis shows that most of the difference regarding income distribution comes from the wealthier households, particularly those that belong to the top 2%. Those households get the greatest proportion of resources coming from non-labor income. At the same time, the difference in returns to higher education explains another 20% of the income differences between Chile and Uruguay. Social conditions such as social security benefits and the participation of women in the labor market are not significant to explain the differences between these countries. Finally, this paper shows that national account adjustment to income information in Chilean households survey explains a third of the Gini coefficient gap between Chile and Uruguay, without the national account adjustment, the income distribution gap among both countries would diminish in three points from earlier estimation. Even though this significative reduction the reasons to explain the gap would remain identical than previous analysis. Key words: Microsimulation, Income distribution, Inequality, Pensions, Labor Participation, Returns to schooling. Clasificacin JEL: D3, J2, J3. 1. Introduccin La preocupacin por el alto nivel de desigualdad de ingreso per cpita en Chile ha ido ganando impulso poltico as como atencin acadmica en los ltimos aos (Ferreira y Litchfield, 1999; Contreras, 2003; Torche, 2005; Larraaga y Valenzuela, 2006). A pesar de que Chile ha dado avances importantes en reducir la tasa de la pobreza, la desigualdad de ingresos ha permanecido obstinadamente persistente. Este documento trata de tomar una perspectiva comparativa sobre el problema mediante el examen de la distribucin de diferentes fuentes de ingreso en Chile con los pases vecinos. Se hace especial hincapi en la comparacin entre Chile y Uruguay con anlisis de microsimulaciones empleados para identificar los principales factores que contribuyen a la diferencia en las distribuciones. Las distribuciones de la oferta laboral femenina, retornos a la escolaridad, pensiones y otros ingresos no laborales son considerados. Aunque numerosos estudios examinan la metodologa de microsimulacin para los cambios intertemporales en la distribucin del ingreso en Chile (Ferreira y Litchfield, 1999; Urza et al., 2002, y Larraaga y Valenzuela, 2006), este documento es el primer estudio en realizar una comparacin entre pases para examinar las causas de la alta concentracin de ingresos en Chile. La metodologa de microsimulaciones es aplicada para comparar la distribucin del ingreso entre ambos pases. Esta metodologa, desarrollada por Bourguignon, Fournier
Examinando la prominente / Juan Pablo Valenzuela, Suzanne Duryea
261
y Gurgand (2000) tiene dos ventajas (Urza et al., 2002): i) analizar los cambios ocurridos durante toda la distribucin, ii) permitir el anlisis de los efectos de los diversos componentes de la distribucin de ingresos. La seccin 2 proporciona una descripcin de las encuestas de hogares y fuentes de ingresos analizadas en este documento. La seccin 3 presenta los hechos estilizados respecto de la desigualdad en forma comparada para varios pases de la regin; en la seccin 4 se compara la distribucin del ingreso entre Chile y Uruguay y en la seccin 5 se profundiza en las caractersticas de la distribucin del ingreso de ambos pases. El modelo de microsimulaciones y sus estimaciones se presentan en la seccin 6 y el anlisis de estos resultados en la seccin 7. La seccin 8 plantea la sensibilidad de las estimaciones a las correcciones de ingresos de la base de datos de Chile y sus implicancias para el anlisis anterior. Finalmente las conclusiones se presentan en la seccin 9. 2. Datos Las estimaciones presentadas en este estudio vienen de la encuesta de hogares de seis pases de Latinoamrica como se muestra en el Cuadro 1. Estos pases fueron seleccionados sobre la base de puntajes similares en el ndice de Desarrollo Humano (IDH) de 2004 todos ellos en un rango de entre 0,82 y 0,86 en el IDH, as como su proximidad fsica con Chile. La cercana comparacin entre Chile y Uruguay est relacionada con la similitud en sus caractersticas sociodemogrficas, tales como altos niveles de educacin, esperanza de vida, urbanizacin y la avanzada fase de transicin demogrfica, lo cual se presenta en el Anexo I. Es importante resaltar que Uruguay experiment una importante recesin entre 1999-2003, con la cada del PIB en 11% (Cepal, 2006), por lo cual, para las microsimulaciones que comparan Chile y Uruguay se ha preferido utilizar la encuesta de hogares del 2005 para Uruguay, mientras que para Chile se consider la del ao 2003. CUADRo 1
ENCUESTAS DE HoGARES UTILIzADAS
Pas Argentina Brasil Chile Costa Rica Mxico Uruguay
Aos 1992, 2002 1992, 2003 1992, 2003 1992, 2004 1992, 2004 1992, 2005
Nombre Encuesta Permanente de Hogares Pesquisa Nacional de Muestra de Domicilios Encuesta de Caracterizacin Socioeconmica Encuesta de Hogares de Propsitos Mltiples Encuesta Nacional de Ingresos y Gastos de los Hogares Encuesta Continua de Hogares
Estas encuestas, levantadas por el Instituto Nacional de Estadsticas, han sido procesadas y homologadas por el departamento de investigacin del Banco Interamericano de Desarrollo (BID) para facilitar la comparacin entre pases. El proceso de homologacin implic un cuidadoso tratamiento de las distintas
262
variables de las encuestas, particularmente para las variables de ingreso. Como los instrumentos de las encuestas varan entre pases, as como intertemporalmente dentro de los pases, el proceso de homologacin se ha destinado a crear definiciones similares. Los ingresos individuales estn compuestos por cuatro componentes bsicos: ingreso monetario del trabajo, ingreso monetario no laboral, ingreso del trabajo no monetario e ingresos no monetarios no asociados al trabajo. Debido a que este documento enfatiza la comparacin entre los pases, la construccin de variables se enfoca en la uniformidad a travs de los seis pases en el ltimo perodo en lugar de la uniformidad dentro de los pases1. De todas formas, no es posible realizar una homologacin completa, ya que las encuestas documentan de manera levemente distinta las fuentes de ingreso debido a que utilizan preguntas diferentes. Este documento considera el ingreso monetario (desde fuentes laborales y no laborales) recibido de los miembros de hogar y excluye el ingreso no monetario, as como todos los ingresos derivados de terceros al hogar. El salario de los empleadores (o patrn segn la nomenclatura de algunas de las encuestas) est tratado como ingreso monetario no laboral. Esto se debe a la dificultad para distinguir entre los recursos provenientes del lucro del negocio y del ingreso pagado por el trabajo desarrollado por el empleador en la administracin de su negocio. Una diferencia importante entre las encuestas es el tratamiento de los datos de ingreso mal reportados. Los datos de ingreso para Mxico y Chile carecen de no respuesta, o valores extremos ingresados, porque estos datos han sido previamente ajustados. Adems de los ajustes a las respuestas individuales en los datos originales, los datos para Chile han sido ajustados a las cuentas nacionales. Esto genera obviamente cierto grado de aprensin respecto a la comparacin entre pases. Si bien los ajustes aplicados por la Cepal a los datos de la Casen para Chile no pueden ser eliminados, dada la carencia de informacin necesaria para ello (Feres 1996, 1997). Un anlisis de la metodologa descrita fue llevado a cabo de tal forma que los ajustes llevan a una ligera reduccin en las medidas de desigualdad (Pizzolitto 2005)2. La limpieza de la encuesta de hogares es claramente un rea donde un intercambio de las mejores prcticas entre los institutos estadsticos de la regin podra ser til. 3. Estadsticas Descriptivas de la Desigualdad de Ingresos en Chile Comparadas con los Otros Pases Exploramos algunas estadsticas comparativas en esta seccin para actualizar la ubicacin de Chile respecto a la desigualdad de ingresos de Argentina, Brasil, Costa Rica, Mxico y Uruguay. Muchas de las diferencias encontradas ya han sido examinadas en estudios recientes. Lo novedoso es que aqu se distingue
Un enfoque alternativo sera la creacin de una medida de los ingresos que hace hincapi en las consistencias dentro de un pas a travs del tiempo. Este es el enfoque adoptado en el Socimetro (www.iadb.org/sociometro) que tiene un fuerte enfoque intertemporal. En este escenario, al comparar la distribucin de los ingresos entre Uruguay y Chile, se hace un esfuerzo metodolgico para analizar el impacto de esta situacin. Para las estimaciones ver seccin VIII.
263
entre ingreso laboral monetario, ingreso laboral no monetario, as como las pensiones, que son un componente especfico del ingreso laboral monetario (ver Anexo II para ms detalle de qu se incluye en la categora pensiones). Adems restringimos las muestras de todos los pases a las zonas urbanas, ya que el marco de muestreo para Uruguay no cubre las zonas rurales y las comparaciones con Uruguay son un eje central de este trabajo. El Cuadro 2 presenta las medidas de desigualdad estndar tales como el coeficiente de Gini, el Theil-L y el Theil-2 para mencionadas fuentes de ingresos y muestreo. En trminos generales se puede decir que Chile y Brasil son vistos como los ms desiguales, en funcin de la ponderacin implcita de la medida de desigualdad. Para el coeficiente de Gini basado en ingresos laborales (primeras dos columnas), estos dos pases tienen medidas similares. Sin embargo, el Theil-L, tambin conocido como la media de la desviacin logartmica, que da mayor peso a los ingresos ms bajos muestra que Brasil tiene la mayor desigualdad en el salario por hora y salario mensual. Con respecto a Theil-2 que da mayor ponderacin a los ingresos ms altos, Chile tiene la ms desigual distribucin de los ingresos laborales. En todas las medidas Uruguay figura entre los pases con menor desigualdad de ingresos laborales monetarios3. CUADRo 2
NDICES DE GINI, THEIL y THEIL-2 PARA EL INGRESo MoNETARIo LABoRAL y No LABoRAL (Muestra: Hombres Urbanos de entre 15 y 65 aos empleados a tiempo completo) Salario Horarioa Salario Mensuala Ingreso No Laborala, d Gini 0,54 0,52 0,49 0,63 0,86 0,87 0,57 0,64 0,52 0,70 0,54 0,53 Theil-Lb Theil-2c 0,49 0,48 0,89 0,90 2,05 2,08 0,65 0,81 0,54 1,04 0,54 0,56 1,28 1,79 2,62 2,21 6,75 12,84 1,42 2,17 1,04 7,64 1,20 1,10
Ao 1992 2002 1992 2003 1992 2003 1992 2003 1992 2004 1992 2005
Gini Theil-Lb Theil-2c Gini Theil-Lb Theil-2c 0,39 0,44 0,53 0,52 0,48 0,48 0,39 0,41 0,45 0,45 0,41 0,46 0,25 0,34 0,51 0,46 0,39 0,39 0,26 0,29 0,34 0,35 0,29 0,36 0,49 0,52 1,46 1,14 0,96 1,70 0,46 0,52 0,65 0,87 0,66 0,70 0,38 0,43 0,52 0,50 0,46 0,47 0,39 0,40 0,44 0,44 0,42 0,45 0,24 0,32 0,47 0,43 0,36 0,37 0,25 0,27 0,33 0,34 0,30 0,36 0,45 0,50 1,38 1,02 0,90 2,00 0,47 0,49 0,66 0,95 0,71 0,67
Fuente: Elaboracin propia en base a Encuestas de Hogares. a Los clculos no incluyen ceros. b La definicin de Theil utilizada es la siguiente: Medida de entropa generalizada con parmetro de sensibilidad igual cero. c La definicin de Theil-2 utilizada es la siguiente: Medida de entropa generalizada con parmetro de sensibilidad igual dos. El valor dos significa que a las diferencias en el extremo superior de la distribucin del ingreso se le asigna ms importancia que a las diferencias en el extremo inferior. d El ingreso de los patrones es considerado como ingreso no laboral.
Estos no son necesariamente bajos con respecto a los ingresos europeos.
264
El Grfico 1 examina el ingreso laboral para hombres urbanos entre 15-65 aos, y muestra la proporcin de ingresos promedios entre deciles adyacentes4. Chile acta de forma similar a otros pases: para los ratios de los deciles 2:1 y hasta el 9:8 todos los promedios estn por debajo de 1,5. Sin embargo, aunque los ratios de todos los pases aumentan en la parte superior de la distribucin, este salto es mayor en Chile. El promedio del ingreso laboral monetario en el dcimo decil es 2,4 veces el promedio del noveno decil. Esta diferencia es de 2,0 para Uruguay5. GRFICo 1
RATIo DE LA MEDIA DEL INGRESo LABoRAL MoNETARIo ENTRE DECILES (Muestra: Hombres urbanos de entre 15 y 65 aos empleados a tiempo completo) 2,6 2,4 Ratio de las medias de ingreso 2,2 2 1,8 1,6 1,4 1,2 1 2:1 3:2 ARG 4:3 BRA 5:4 CHL 6:5 Decil de Ingreso CRI MX URY 7:6 8:7 9:8 10:9
Fuente: Elaboracin propia en base a encuestas de hogar. Nota: Ingreso laboral monetario no incluye el ingreso de los patrones.
El Grfico 2 muestra el ratio del ingreso no laboral monetario a travs de los deciles para individuos en reas urbanas. Una vez ms, los coeficientes son moderados. Estn por debajo de 1,4 hasta la parte alta de la distribucin, donde la proporcin se incrementa a 5 para Chile. La desigualdad excesiva de esta fuente de ingresos puede tambin ser vista en las medidas de desigualdad mostradas en el Cuadro 2. Para todos los indicadores, Gini, Theil-L y Theil-2, Chile tiene la desigualdad ms alta entre los seis pases, mientras que Uruguay se encuentra entre los pases con menor desigualdad en los resultados de estas mediciones. El Grfico 3 combina los efectos de ingresos laborales y no laborales y tambin los efectos del tamao de los hogares y muestra el ratio del ingreso monetario per cpita a travs de los deciles. Este ratio es extremo en los deciles superiores de Chile (3,1) y bajo en los mismos deciles de Uruguay (2,2).
4 5
Los ranking son similares si la mediana de los ingresos a travs de los deciles son utilizados ms que la media. Chile excede la desigualdad en la parte superior de la distribucin de los ingresos. Es tambin evidente si el Grfico 1 es estimado por la mediana en lugar de la media.
265
GRFICo 2
RATIo DE LA MEDIA DEL INGRESo No LABoRAL MoNETARIo ENTRE DECILES (Muestra: Urbana) 5,5 Ratio entre medias de ingreso 5 4,5 4 3,5 3 2,5 2 1,5 1 2:1 3:2 ARG 4:3 BRA 5:4 6:5 Deciles CHL CRI 7:6 8:7 MX 9:8 URY 10:9
Fuente: Elaboracin propia en base a encuestas de hogar.
GRFICo 3
RATIo DEL PRoMEDIo DE INGRESo ToTAL PER CPITA ENTRE DECILES (Muestra: Urbana) 3,5 Ratio entre promedios de ingreso per cpita 3 2,5 2 1,5 1 2:1 3:2 ARG 4:3 BRA 5:4 CHL 6:5 Deciles CRI MX URY 7:6 8:7 9:8 10:9
Fuente: Elaboracin propia en base a encuestas de hogar.
Dada la dispersin observada en la parte superior de la distribucin para Chile, un ejercicio interesante es explorar las propiedades de la distribucin si los percentiles superiores son excluidos de los clculos (IPES, 1998). El Grfico 4 presenta los resultados comparando Chile y Uruguay. Cuando el 1% superior de la distribucin es eliminado, el Gini para ingreso monetario entre los dos
266
pases se equipara y para el momento que se elimin el 5% superior, Uruguay alcanza a ser ms desigual que Chile. Sin embargo, cuando hacemos la misma comparacin para el ingreso no laboral monetario, el Gini para ambos pases disminuye, ya que la parte superior de la distribucin es descartada, pero persiste una gran disparidad entre los dos pases, inclusive cuando el 5% superior de la distribucin es eliminado. La desigualdad del ingreso no laboral se encuentra en 0,6 en Chile y 0,4 en Uruguay. GRFICo 4
CoEFICIENTES DE GINI PARA EL INGRESo LABoRAL MoNETARIo (Muestra: Hombres urbanos de entre 15 y 65 aos empleados a tiempo completo) 0,48 0,46 0,44 0,42 0,4 0,38 0,36 0,34 0,32 0,3
Gini
1-100
1-99
1-98 1-97 Centiles incluidos Chile 2003 Uruguay 2005
1-96
1-95
Fuente: Elaboracin propia en base a encuestas de hogar. Nota: Los ceros no estn incluidos en los clculos.
Hemos observado hasta el momento una alta desigualdad entre las dos fuentes de ingreso monetario a nivel individual: ingresos laborales e ingresos no laborales. A continuacin examinaremos directamente la distribucin del ingreso per cpita del hogar. El Cuadro 3 presenta algunas medidas de desigualdad estndar para este tipo de ingreso. En trminos del coeficiente de Gini, en donde todas las partes de la distribucin se ponderan igual, Chile y Brasil tienen un alto nivel de desigualdad de 0,56 y 0,58, respectivamente. En Uruguay, la desigualdad del ingreso per cpita es mucho ms moderada con un Gini de 0,47. Si utilizamos un indicador que pone mayor peso a la parte superior de la distribucin como es el Theil-1, Brasil se ubica en el primer lugar. Sin embargo, con el Theil-2 que tiene ponderaciones ms pronunciadas en la parte superior de la distribucin, tanto Chile como Mxico sobresalen. Incluso quitando los centiles ms altos de la distribucin, Chile no alcanza a reducir su coeficiente de Gini a los valores de Uruguay (Grfico 5). El anlisis de la distribucin del ingreso es complejo, no slo porque las diferentes fuentes de ingreso tienen diferentes propiedades sino porque la composicin del hogar juega un rol crtico. La composicin del hogar a travs de patrones de emparejamiento selectivo (assortative mating), patrones de fertilidad y las condiciones de vida de los adultos, tiene implicancias importantes para la distribucin del ingreso per cpita. El Cuadro 4 examina los patrones de
267
CUADRo 3
DISTRIBUCIN DEL INGRESo PER CPITA DE LoS HoGARES (Muestra: Urbana)
Ao 1992 2002 1992 2003 1992 2003 1992 2003 1992 2004 1992 2003
Ingreso per cpita de los hogaresa Gini 0,502 0,571 0,561 0,575 0,524 0,563 0,503 0,518 0,520 0,499 0,431 0,471 Theil-Lb 0,366 0,542 0,553 0,597 0,496 0,575 0,382 0,443 0,465 0,425 0,328 0,397 Theil2c 0,600 0,916 1,560 1,355 1,255 2,801 0,650 0,832 1,713 3,811 0,547 0,676
Fuente: Elaboracin propia en base a encuestas de hogares. a Los clculos incluyen ceros. b La definicin de Theil utilizada es la siguiente: Medida de entropa generalizada con parmetro de sensibilidad igual cero. c La definicin de Theil2 utilizada es la siguiente: Medida de entropa generalizada con parmetro de sensibilidad igual dos. El valor dos significa que a las diferencias en el extremo superior de la distribucin del ingreso se les asigna ms importancia que a las diferencias en el extremo inferior.
GRFICo 5
CoEFICIENTES DE GINI PARA EL INGRESo MoNETARIo PER CPITA DE LoS HoGARES (Muestra: Urbana) 0,6 0,55 0,5 Gini 0,45 0,4 0,35 0,3 1-100 1-99 1-98 Chile 2003 1-97 Uruguay 2005 1-96 1-95
Centiles incluidos
268
emparejamiento selectivo entre parejas en reas urbanas. Mientras Costa Rica, Argentina y Brasil tienen coeficientes de baja correlacin entre los seis pases, Chile tiene el ms alto. Chile tiene tambin una alta distribucin de parejas con el mismo nivel de educacin6, 7. Respecto a un reciente documento de investigacin de la movilidad intergeneracional en Europa, las personas en Chile son mucho ms similares a las personas del Reino Unido y Alemania. La correlacin entre la educacin de las personas en el Reino Unido es de 0,41 y en Alemania es de 0,52. Nuestras estimaciones para la correlacin en Chile son de 0,71. Mientras el 50% de las parejas comparte el mismo nivel de educacin en Chile, slo el 30% de las parejas britnicas comparte una condicin de escolaridad similar (Ermisch, Francesconi y Siedler, 2006). CUADRo 4
EMPAREJAMIENTo SELECTIVo ENTRE PAREJAS URBANAS
Ao 1992 2002 1992 2003 1992 2003 1992 2003 1992 2004 1992 2003
Coeficiente de correlacin na 0,655 0,705 0,653 0,717 0,714 0,662 0,649 0,684 0,704 0,608 0,671
Porcentaje de parejas con la misma educacin 0,466 0,470 0,438 0,435 0,533 0,501 0,436 0,429 0,420 0,484 0,436 0,469
Fuente: Elaboracin propia en base a encuestas de hogares.
Es bien conocido que la educacin desempea un rol importante en la determinacin de la desigualdad de ingresos. Como han encontrado muchos autores, los retornos a la educacin en Chile son altos, especialmente para los individuos que completan la educacin superior. El Cuadro 5 muestra los retornos anuales por completar un ao adicional de escolaridad en cada nivel educativo8. A pesar
6 7
La cohabitacin informal es incluida en la definicin de parejas. Siete diferentes niveles se definieron: Sin escolaridad, primaria incompleta, primaria completa, secundaria incompleta, secundaria completa, terciaria incompleta, terciaria completa o ms. Los retornos fueron calculados de una regresin del logaritmo del salario por hora del trabajo principal, su experiencia potencial y su cuadrado y niveles de escolaridad. La muestra contiene la edad de los hombres entre 15-65 aos, quienes trabajan tiempo completo en la semana de referencia (al menos las 30 horas). Los retornos fueron anualizados dividiendo por el nmero de aos necesarios para completar los niveles de educacin. La
269
de que el efecto de un alto retorno a la escolaridad sobre la desigualdad ha sido bien examinado (Contreras, 2003), se reafirma aqu por su relevancia para un futuro contrafactual. CUADRo 5
REToRNoS A LA EDUCACIN ESTIMADoS MEDIANTE EL MToDo DE SPLINE (Muestra: Hombres urbanos de entre 15 y 65 aos empleados a tiempo completo)
Pas ARG ARG BRA BRA CHL CHL CRI CRI MEX MEX URy URy
Ao 1992 2002 1992 2003 1992 2003 1992 2003 1992 2004 1992 2005
Primaria na 0,096 0,151 0,115 0,054 0,055 0,057 0,056 0,068 0,048 0,074 0,032
Secundaria na 0,099 0,140 0,124 0,107 0,091 0,108 0,085 0,119 0,088 0,092 0,133
Terciaria na 0,217 0,198 0,226 0,206 0,235 0,114 0,128 0,161 0,150 0,206 0,194
La participacin de la fuerza laboral femenina ha tenido implicancias a travs de los ingresos individuales, as como a travs de los patrones de emparejamiento selectivo. El Grfico 6 demuestra que el empleo femenino es ms bajo en Chile que en otros pases. Mientras que las mujeres en Uruguay demuestran una alta tasa de empleo (o participacin), sus homlogas chilenas tienen la ms baja tasa de empleo en todos los niveles de educacin, excepto en la categora de sin estudios. Cancian y Reed (2001) encuentran que cambios en el empleo y salarios de las esposas disminuyen la desigualdad de ingresos en los EE.UU. desde 1969 a 1999. El efecto de las bajas tasas de participacin entre mujeres chilenas es examinado en las microsimulaciones en la seccin IV. Chile y Uruguay tienen un sistema de pensiones bien desarrollado, al mismo tiempo que en la actualidad el sistema chileno se encuentra considerando la implementacin de una importante reforma en el sector. Como se ha visto, el ingreso monetario no laboral est altamente concentrado en Chile y en este estudio examinamos la contribucin especfica de las pensiones a la distribucin
interpretacin adecuada para los retornos de la escolaridad calculados a partir de este tipo de especificacin es el incremento porcentual en el salario de completar un ao adicional de escolaridad asumiendo que los niveles son completados.
270
de ingresos. El Cuadro 6 muestra que la recepcin de los ingresos no laborales es ms generalizada en Uruguay, aunque bastante ms comn en los deciles superiores en Chile. Con respecto al ingreso de pensiones, la cobertura es ms amplia en el Uruguay, donde un 40% de los hogares reciben ingresos por pensiones, mientras que en Chile lo hacen slo un 25%.
GRFICo 6
MUJERES EMPLEADAS PoR PAS y EDUCACIN (Muestra: Mujeres urbanas de entre 15 y 64 aos) 60
50
40 % empleado
30
20
10
Sin escuela
Primaria inc.
Primaria comp. Secundaria inc.
Secundaria comp.
Terciaria
URY
ARG
BRA
CRI
MX
CHL
La importancia de las pensiones en el ingreso familiar total tambin vara a travs de los dos pases. En Chile las pensiones desempean un rol importante en el ingreso de los individuos de los deciles ms bajos, en particular en los del decil 1, especialmente a travs de pensiones asistenciales y mnimas. En Uruguay las pensiones comprenden una mayor proporcin del ingreso total que en Chile para los deciles 3-10. Asimismo, el valor de las pensiones promedio en trminos reales es ms alto en Uruguay que en Chile. La funcin especfica de las pensiones en la explicacin de las diferencias en la desigualdad de ingresos per cpita tambin ser examinada con las microsimulaciones.
271
CUADRo 6
INGRESo No LABoRAL A TRAVS DE DECILES
Pas
Decil
Porcentaje de hogares con ingreso no laboral monetario per cpita positivo 56,030 39,370 55,420 56,150 59,830 61,640 66,330 68,470 76,620 87,060 97,660 71,160 74,290 66,520 64,610 67,500 68,140 72,090 73,360 72,900 73,220 73,540
Porcentaje del ingreso no laboral monetario per cpita sobre el ingreso total per cpita del hogar 0,157 0,150 0,116 0,109 0,109 0,109 0,117 0,135 0,162 0,195 0,340 0,573 0,548 0,491 0,502 0,521 0,550 0,583 0,606 0,629 0,612 0,646
Porcentaje de hogares que reciben pensiones
Porcentaje de las pensiones per cpita sobre el ingreso total per cpita del hogar 0,108 0,376 0,141 0,106 0,089 0,069 0,057 0,053 0,045 0,038 0,021 0,185 0,040 0,089 0,124 0,167 0,186 0,228 0,249 0,256 0,237 0,238
Chile
Uruguay
Total 1 2 3 4 5 6 7 8 9 10 Total 1 2 3 4 5 6 7 8 9 10
25,310 24,110 19,090 18,020 18,930 17,390 16,220 17,050 16,880 16,540 13,740 39,750 7,140 16,610 25,040 33,420 38,270 47,240 49,340 49,680 46,770 45,860
4. Comparacin de la Distribucin del Ingreso entre Chile y Uruguay A pesar de que la economa de Uruguay experiment una importante recesin entre 1999-2003, con un mximo en el 2002, cuando el PIB del pas cay 11,0% (Cepal, 2006), este pas ha presentando una situacin menos desigual en la distribucin de sus ingresos que la mayora de los pases de la regin. Por ejemplo, para zonas urbanas, en el ao 2002 Uruguay presentaba un coeficiente de Gini de 45,5, lo cual implicaba ocupar el primer lugar entre 18 pases del continente (Cepal, 2006)9. En este mismo ranking de pases Chile se encontraba en
9
Sin embargo, desde mediados de la dcada de los 90 Uruguay ha presentado un gradual incremento en todos los indicadores de desigualdad en la distribucin de los ingresos de los hogares (Bucheli y Furtado, 2004).
272
la posicin 13 con un Gini de 54,7 en zonas urbanas en el ao 2003, superado por Repblica Dominicana, Nicaragua, Bolivia, Colombia y Brasil. La comparacin de la distribucin del ingreso de Chile respecto a Uruguay no slo es interesante por las diferencias que en este mbito presentan ambas economas, sino tambin, como indicamos anteriormente, por la similitud en algunas caractersticas sociodemogrficas, como el alto porcentaje de poblacin urbana, su avanzada transicin demogrfica, altas expectativas de vida, as como una elevada cobertura educacional y un sistema relativamente masivo de seguridad social. Por otra parte, las causas que explican esta diferencia distributiva entre ambos pases pueden estar asociadas a la distribucin de las dotaciones de factores productivos y demogrficos, tambin a diferentes retornos a estos componentes10 o a la diferencia en las decisiones laborales de los individuos. El objetivo de este trabajo es indagar en las explicaciones detrs de este diferencial en el ndice de Gini entre ambas economas. Metodolgicamente, para comparar la distribucin del ingreso entre ambos pases se utilizan microsimulaciones basadas en la descomposicin de los cambios en la distribucin de diversos componentes asociados a los ingresos monetarios, la cual fue inicialmente desarrollada por Bourguignon, Fournier y Gurgand (2000) y aplicada en la comparacin entre diferentes pases por Bourguignon, Ferreira y Leite (2002). La metodologa de microsimulacin permite analizar los efectos de componentes especficos sobre la distribucin de los ingresos. Aunque existen diversas investigaciones que aplican la metodologa de microsimulaciones a la evolucin en la distribucin del ingreso entre los hogares chilenos (Ferreira y Litchfield, 1999; Urza et al., 2002; Larraaga y Valenzuela, 2006), este trabajo es el primero que realiza comparaciones entre diferentes economas para determinar las causas de la alta concentracin del ingreso en Chile. Para las comparaciones de la distribucin del ingreso entre las economas de Chile y Uruguay se utilizan las encuestas de hogares de ambos pases. En el caso de Uruguay se utilizar la base del 2005 y en el caso de Chile la base del ao 2003. La decisin de utilizar la base del 2005 para Uruguay se justifica debido a que esta base de datos refleja de mejor forma la economa de largo plazo del pas, puesto que en el perodo anterior Uruguay haba experimentado una fuerte recesin, con una cada de su economa de 11% durante el 2002 lo cual llev a duplicar la incidencia de la pobreza en las zonas urbanas del pas entre 2000-2005, a pesar de las altas tasas de crecimiento del 2004 y 2005, lo que podra genera mayores inconsistencias en la comparacin de ambas economas.
10
Lo cual podra explicarse por diferenciales en la oferta relativa de estos factores, calidad de los mismos, factores institucionales, entre otros.
273
CUADRo 7
ALGUNAS CARACTERSTICAS SoCIoECoNMICAS RECIENTES DE CHILE y URUGUAy 2000 Variacin del PIB Chile Uruguay Tasa de Desempleoa Chile Uruguay Tasa de Pobrezab Chile Uruguayc
a b c
2001
2002 2,2 11,0
2003 3,9 2,2 9,5 16,9 18,5
2004 6,2 11,8 10,0 13,1
2005 6,3 6,6 9,2 12,2 18,8
2006 4,4 7,5 7,9 11,6 14,0
4,5 1,4 9,7 13,6 19,7 9,4 9,9 15,3
9,8 17,0
Se refiere slo a zonas urbanas para Uruguay. Se refiere slo a zonas urbanas en ambos pases. Corresponde al ao 1999. Fuentes: Anuario Estadstico de Amrica Latina y el Caribe, 2006 (Cepal, 2007) y Serie de Anlisis de la Encuesta Casen 2006 (Mideplan, 2007).
5. Caractersticas Generales de la Distribucin del Ingreso en Ambos Pases Los anlisis de Chile y Uruguay estn basados en las encuestas de hogares del ao 2003 y 2005, respectivamente. Con el objetivo de hacer comparables ambas encuestas y dado que, para el caso de Uruguay, sta slo considera la poblacin urbana de la ciudad de Montevideo, se ha excluido la poblacin que habita en zonas rurales para el caso chileno11. CUADRo 8
INDICADoRES DISTRIBUCIN DEL INGRESo PER CPITA PoR HoGAR Chile 55,67 68,64 10,20 3,61 0,35 3,21 45,01 34,50 8,81 3,17 0,36 3,07 Uruguay 47,12 40,50 9,72 3,07 0,32 3,26 41,35 27,84 8,74 2,78 0,32 3,11
Gini Theil (Entropa) 90/10 90/50 10/50 75/25 Efectos al excluir al 2% de mayores ingresos Gini Theil (Entropa) 90/10 90/50 10/50 75/25
Fuente: Anlisis propios basados en MECoVI, BID. Encuestas de Hogares Uruguay y Chile, 2005 y 2003, respectivamente.
11
Es importante destacar que los ingresos provenientes de la encuesta Casen han sido corregidos por subdeclaracin, de tal forma que estos son consistentes con los antecedentes de las cuentas nacionales, situacin que no se presenta para el caso de Uruguay; sin embargo, ms adelante se realiza un ejercicio excluyendo parte de estas correcciones para el caso chileno.
274
5.1. Distribucin del ingreso Los coeficientes de desigualdad en la distribucin del ingreso para ambos pases indican una mayor desigualdad para el caso de Chile (con un coeficiente de Gini de 55,67 para Chile respecto a uno de 47,12 para Uruguay), diferencia que es an ms sustantiva utilizando el ndice de Theil, ms sensible a la concentracin del ingreso en la parte superior de la distribucin (Cuadro 8). Sin embargo, al comparar la relacin de diversos centiles se aprecia una situacin donde Chile presenta condiciones algo menos desiguales en la distribucin de los centiles inferiores. Estos resultados son consistentes con los Grficos 7 y 8, donde se aprecia que los ingresos promedio per cpita por hogar entre Chile y Uruguay son muy similares entre los centiles 5-80 a pesar de que el ingreso promedio per cpita en los hogares chilenos es 21% superior que el de los hogares uruguayos, con ingresos promedio ms bajos para Chile entre los hogares de los centiles inferiores, pero tambin con una mayor concentracin del ingreso en los ltimos centiles de la distribucin lo cual se refleja en el Grfico 7 sobre diferencias de los logaritmos de ingreso per cpita del hogar entre ambas economas, lo que es compatible con las curvas de Lorenz para ambas economas, donde la mejor distribucin del ingreso en Uruguay presenta una dominancia estocstica de segundo orden sobre la distribucin de Chile. GRFICo 7
DISTRIBUCIN DEL INGRESo CHILE y URUGUAy LN INGRESo DEL TRABAJo
Ln ingreso per cpita por hogar Chile (2003) y Uruguay (2005) urbano
(Centiles de ingreso) 10 log del ingreso per cpita por hogar
2 0 20 40 60 80 100
Centiles de ingreso per cpita por hogar Uruguay Chile
275
GRFICo 8
DISTRIBUCIN DEL INGRESo CHILE y URUGUAy CURVA DE LoRENz
1,0
Curva de Lorenz de ingresos per cpita
0,8
0,6
0,44
0,2
0,0 0,0 0,2 0,4 0,6 0,8 1,0
Proporcin de la poblacin acumulada Uruguay Chile
La importancia de la concentracin del ingreso en los ltimos centiles de la distribucin de los hogares es graficada al excluir de la distribucin al 2% de las personas de mayores ingresos en ambos pases12. El Cuadro 8 indica que los diferentes ndices de desigualdad se reducen considerablemente en Chile y tienden a homogeneizarse entre ambos pases. GRFICo 9
DIFERENCIA EFECTIVA EN INGRESo PER CPITA PoR HoGAR CHILE (2003) zoNAS URBANAS
1 0,5
Diferencia logaritmo
0 -0,5 -1 -1,5 -2 -2,5
9 13 17 21 25 29 33 37 41 45 49 53 57 61 65 69 73 77 81 85 89 93 97
Centiles de ingreso per cpita por hogar
12
Un 46,4% del total de ingresos monetarios del centil 99 provienen de ingresos no laborales o asociados a la condicin de empleadores, mientras que para el centil 100 estos son un 71,8%.
276
6. Metodologa y Estimacin de Parmetros 6.1. Metodologa de microsimulaciones Desde una perspectiva microeconmica la distribucin del ingreso corresponde a cambios en las decisiones de participacin laboral, en las caractersticas demogrficas o en el retorno (precio) que el mercado asigna a estos factores13. Si Yip es el ingreso monetario del trabajo para un individuo i en un perodo t y en un pas p, el cual puede ser descrito por una funcin F del vector de variables observables Xip, las cuales afectan el salario y la participacin en el mercado del trabajo, el vector ip de caractersticas no observadas, el vector p de precios que afectan los salarios y p de parmetros que afectan la participacin. (1) Yip = F X ip , ip , p , p
i=1,..,N
La distribucin del ingreso individual del ingreso del trabajo puede ser representada por: (2) D p = Y1 p ,......, YNp
De tal forma que es posible simular los ingresos del trabajo modificando uno o ms argumentos de la ecuacin (1), por ejemplo podemos pensar cmo cambiara la distribucin del ingreso de Chile si se utilizara el vector de precios para los salarios de Uruguay: (3)
YiCH ( UR = F ( X iCH , iCH , UR , CH
i=1, , N
De esta forma, el cambio en la distribucin del ingreso de Chile debido a cambios en cualquiera de los k componentes de la distribucin de Uruguay, podr ser expresado en trminos de ndices de desigualdad (I) como: (4) I [ DCH ] I [ DCH ( kUR )]
De igual forma podemos estimar los ingresos monetarios per cpita provenientes de los hogares h de un pas p, donde n corresponde al nmero de personas i en el hogar h. La primera sumatoria representa los ingresos provenientes del mercado del trabajo, donde el suprandice j representa el tipo de ingresos del trabajo y el subndice i representa el perceptor i de ingresos en el hogar. La variable I es una funcin indicadora que toma el valor 1 si el ingreso asociado tiene valor positivo y el valor cero en otro caso. La segunda sumatoria incluye el ingreso proveniente de pensiones, representados por el suprandice s. El tercer
13
A continuacin se utiliza la descripcin planteada por Gasparini et al. (2004).
277
tipo de ingreso yo representa otro tipo de ingresos monetarios, que se suponen exgenos en el ejercicio 1 n J j j I y k + nhp i =1 j =1 hip hip p yohp
(5)
( )
s s I hip yhip ( k p ) + n S i =1 s =1
Las modificaciones a alguno de los componentes de la distribucin de los ingresos de los hogares podrn ser estimadas en forma similar a (4). 6.2. Estrategia de estimacin de parmetros y resultados La estrategia de microsimulacin contempla tres etapas, la primera corresponde a la estimacin de los parmetros relevantes asociados a las ecuaciones de ingresos (Cuadro 9) y participacin laboral (Cuadro 10); la segunda, implica modificar la distribucin de las variables observables exgenas (X, z) que tienen efectos directos e indirectos, va participacin laboral, en los ndices de distribucin del ingreso. La tercera corresponde a la estimacin de la distribucin de los ingresos no-laborales de los hogares. 6.3. Estimacin de y La asignacin del individuo i a una actividad laboral j, la cual est modelada a travs de una estimacin multilogit: (6) Pr j = s = P s ( Z hi , ) =
e e
Z hi s
Z hi s Z hi s
+ e
js
donde Ps() es la probabilidad que la persona i en el hogar h est en la actividad ocupacional s ={inactivo, trabajo dependiente o trabajo por cuenta propia}. Similares estimaciones son realizadas separadamente para mujeres y hombres de entre 15-65 aos de edad. Las variables observadas z ={constante, experiencia, experiencia cuadrado, dummies para la educacin de la persona y para el promedio de educacin de los mayores de 14 aos del hogar, dummies para regiones en Chile, dummy para el jefe del hogar, si es pareja del jefe del hogar, si est casado o convive, si estudia, promedio de edad de los adultos y el promedio de nios menores de 15 aos}. Para determinar si el individuo es asignado a una determinada posicin laboral se asume que ste seleccionar la actividad que le genere mayor utilidad entre las s alternativas disponibles, es decir, el individuo resuelve sobre una funcin U j de utilidad, U hi = Z hi j + hi j , donde el trmino de error puede ser definido
278
para cada alternativa s asumiendo que sigue una distribucin determinada14, y seleccionando los valores que sean consistentes con las preferencias laborales observadas. El vector corresponde a los coeficientes de la estimacin de participacin laboral. Los resultados de estas estimaciones, diferenciadas por pas y sexo, se encuentran en el Cuadro 9. Respecto a las estimaciones de los ingresos monetarios del trabajo, se calculan ecuaciones a nivel individual para el logaritmo de todos los ingresos monetarios de todos los trabajos realizados por cada individuo, separadamente para mujeres y hombres. (7)
j j log yhi = j + X hi j + i = log yhi = j + X hi j + i
donde X={dummies para educacin, experiencia y su cuadrado, empleo de jornada parcial, regiones en Chile}; para simplificar los anlisis se incluye una dummy para los trabajadores por cuenta propia, asumiendo que su nica diferencia con los trabajadores dependientes ser un cambio en el intercepto de la estimacin. Estas estimaciones permitirn obtener valores para los residuos de las ecuaciones de hombres y mujeres con ingresos positivos observados y estimaciones de sus desviaciones estndar. Sin embargo, ante cambios en diferentes factores exgenos los individuos podran modificar su estatus ocupacional, por ejemplo pasando de inactivo a un trabajo asalariado. Ello implica contar con valores de los residuos para las ecuaciones de ingresos laborales (7) para los casos donde no se observan ingresos, lo cual es resuelto a travs de la seleccin de trminos aleatorios para estos residuos de una distribucin normal y que cumplan con las decisiones laborales observadas. Los Cuadros 9 y 10 presentan los coeficientes de las ecuaciones de ingreso y participacin para ambas economas. Los resultados indican que los retornos a la educacin son convexos en ambas economas, pero an ms en la chilena, debido a que los retornos a la educacin secundaria incompleta y completa son mayores en Uruguay que en Chile. Por otra parte, los retornos a la experiencia son ms altos en Uruguay, en particular para las mujeres, mientras que en Chile los ingresos de los trabajadores por cuenta propia son mayores que los obtenidos por asalariados. Esta situacin se revierte en el caso de Uruguay, lo cual podra estar asociado a la precaria condicin de su economa durante el perodo analizado; esta misma causa podra estar explicando que los ingresos laborales de los trabajadores uruguayos presenten una mayor heterogeneidad en los retornos asociados a factores no observables, ello reflejado en los mayores valores de la desviacin estndar de los residuos de las ecuaciones de Mincer (m y h).
14
Una alternativa es que siga una distribucin log[-log(x)], donde x es un trmino aleatorio con una distribucin uniforme entre [0,1] (Bourguignon y Ferreira, 2004).
CUADRo 9
ESTIMACIN DE ECUACIN DE INGRESoS MoNETARIoS DEL TRABAJo CHILE URBANo 2003 y URUGUAy URBANo 2005 Chile Mujer Hombre Mujer Uruguay Hombre
Primaria (1-6 aos)
Primaria Completa (7-8 aos)
Secundaria Incomp. (9-10 aos)
Secundaria Completa (11-12 aos)
Superior Incompleta (13-16 aos)
Superior Completa (17 o ms)
Exp
Exp2
Empleo Parcial
Cuenta Propia 40,93*** (0,088) 0,663 0,406
0,27*** (0,085) 0,42*** (0,088) 0,54*** (0,089) 0,85*** (0,088) 10,29*** (0,089) 20,01*** (0,090) 0,019*** (0,0019) 0,0002*** (0,00004) 10,47*** (0,054) 0,23*** (0,024) 50,01*** (0,056) 0,614 0,417 30,89*** (0,151) 0,792 0,387
0,26*** (0,053) 0,46*** (0,055) 0,58*** (0,055) 0,82*** (0,055) 10,28*** (0,057) 20,06*** (0,059) 0,040*** (0,0015) 0,0005*** (0,00003) 10,31*** (0,076) 0,41*** (0,014)
0,53*** (0,150) 0,65*** (0,153) 0,98*** (0,151) 10,38*** (0,151) 10,85*** (0,151) 20,45*** (0,153) 0,056*** (0,0026) 0,0007*** (0,00005) 0,28*** (0,024) 0,56*** (0,028)
0,46*** (0,169) 0,69*** (0,170) 0,93*** (0,170) 10,25*** (0,170) 10,65*** (0,171) 20,27*** (0,172) 0,076*** (0,0022) 0,001*** (0,00004) 0,34*** (0,042) 0,47*** (0,19) 40,16*** (0,170) 0,739 0,380
Constante
Desviacin Estndar de Residuos R2
***, **, * representa que el coeficiente es estadsticamente significativo a < 0,01, <0,05 y < 0,1, respectivamente. Muestra considera a asalariados y trabajadores por cuenta propia de entre 15-65 aos de edad y con ingresos monetarios del trabajo positivos.
279
CUADRo 10
Chile Hombre Depend, 0,84*** 0,70*** 0,70*** 1,24*** 1,92*** 2,18*** 0,12*** 0,003*** 0,82*** 2,25*** 0,01 0,03*** 0,04*** 0,02*** 0,42*** 0,06*** 0,024*** 0,05*** 0,54*** 0,40*** 0,242 0,242 0,02*** 0,01*** 0,47*** 2,36*** 0,04*** 0,02*** 0,02*** 0,01*** 0,29*** 0,00 0,07*** 0,06*** 0,09*** 0,13*** 0,07*** 0,06** 0,01*** 0,20*** 0,58*** 1,88*** 0,122 0,69*** 1,94*** 0,17*** 0,003*** 0,13*** 0,003*** 0,13*** 0,28*** 1,15*** 0,12*** 0,002*** 0,24*** 0,00 1,23*** 0,03* 0,01 0,03** 0,22*** 0,22*** 0,36*** 0,005*** 0,12*** 0,99*** 3,47*** 0,122 0,57*** 0,50*** 0,47*** 0,76*** 1,18*** 1,06*** 1,06*** 1,18*** 1,31*** 1,81*** 2,64*** 2,97*** 0,48*** 0,52*** 0,64*** 1,10*** 1,43*** 2,90*** 1,81*** 1,89*** 2,03*** 2,45*** 2,95*** 2,94*** 0,13*** 0,003*** 0,70*** 1,41*** 0,00 0,02* 0,06*** 0,15*** 0,26*** 0,11*** 0,02*** 0,08*** 0,74*** 1,61*** 0,164 Cuenta Propia Depend. Cuenta Propia Depend. Mujer Hombre Cuenta Propia 0,78*** 0,75*** 0,95*** 1,18*** 1,58*** 1,98*** 0,16*** 0,003*** 0,61*** 1,59*** 0,10*** 0,05*** 0,15*** 0,16*** 0,19*** 0,01 0,02*** 0,03*** 0,56*** 2,04*** 0,164 Uruguay
280
ESTIMACIN MULTILoGIT SoBRE DECISIoNES DE PARTICIPACIN LABoRAL zoNAS URBANAS DE CHILE 2003 y URUGUAy 2005
Mujer
Depend, 0,24*** 0,23*** 0,43*** 0,60*** 1,18*** 1,65*** 0,13*** 0,002*** 0,12*** 0,21*** 1,14*** 0,32*** 0,03*** 0,07*** 0,02*** 0,27*** 0,10*** 0,002*** 0,03*** 0,95*** 4,09*** 0,137
Cuenta Propia
Nivel Educacional Primaria (1-6 aos) Primaria Completa (7-8 aos) Secundaria Incomp. (9-10 aos) Secundaria Completa (11-12 aos) Superior Incompleta (13-16 aos) Superior Completa (17 o ms)
0,10*** 0,03* 0,05*** 0,48*** 1,37*** 2,27***
Exp Exp2 Es pareja del jefe del hogar? Convive Estudia
0,11*** 0,002*** 0,17*** 0,62*** 1,87***
Educacin Promedio de Adultos 1-6 7-8 9-10 11-12 13-16 17 o ms
0,11*** 0,05*** 0,17*** 0,06*** 0,23*** 0,00
Promedio edad de adultos Nmero de menores de 15 aos Jefe del Hogar Constante
0,007*** 0,21*** 0,62*** 0,89***
Pseudo-R2
0,137
En todas las estimaciones se incluy condicin del hogar si es unipersonal y si el jefe del hogar es activo laboralmente. ***, **, * representa que el coeficiente es estadsticamente significativo a < 0,01, < 0,05 y < 0,1, respectivamente.
281
6.4. Cambios en las dotaciones demogrficas (X, Z) Este estudio estima la distribucin condicional de la educacin de los mayores de 14 aos de edad (Educ) y del nmero de nios menores de 15 aos (N) a nivel de hogares. Para Educ15 se estima una regresin multilogit sobre sexo, cohortes de edad, edad y el nmero de adultos en el hogar. Para determinar la distribucin condicional de N16 se estima una regresin multilogit sobre edad, categoras de educacin y nmero de adultos en el hogar. Ambas distribuciones condicionales de Uruguay son impuestas a la distribucin de los hogares chilenos siguiendo la estructura de (6). 6.5. Cambios en las pensiones La simulacin de los cambios en ingresos provenientes de pensiones (ys) utiliza una metodologa mixta, similar a la aplicada por Larraaga y Valenzuela (2006). En trminos generales, la poblacin de 60 o ms aos de edad es dividida en 24 grupos demogrficos excluyentes de acuerdo a variables de sexo, educacin (0-6 aos de educacin, 7-12 y 13+) y edad (60-64 aos, 65-69 aos, 70-74 aos y 75+), los cuales son analizados para determinar los cambios ocurridos en el acceso y el monto de las pensiones recibidas, de tal forma de simular la distribucin de Uruguay para el caso chileno. Para determinar el monto a ser imputado en la simulacin se realiza una transformacin no paramtrica de la distribucin observada de ys dentro de cada subgrupo, ordenados de acuerdo al valor de ys, asignndoles el valor observado en el mismo percentil de la distribucin de Uruguay a la distribucin de Chile. Para simular la distribucin en el acceso a la transferencia se estima el modelo probit para cada uno de los 24 subgrupos y para cada pas, donde las variables independientes son: sexo, jefatura del hogar, 3 categoras del nivel educacional, si la persona es viuda y la edad y su cuadrado. Si la proporcin de personas que reciben pensin, en el respectivo subgrupo, de Chile es mayor que en Uruguay, aquellas personas con la menor probabilidad estimada de recibir la pensin son imputadas con un valor 0 hasta alcanzar un porcentaje similar en ambos pases. Para el caso contrario, las personas con la ms alta probabilidad de recibir pensin en Chile, pero que tienen valores observados iguales a 0, son modificados con el valor proveniente de una regresin lineal del monto de la pensin basado en: sexo, jefatura del hogar, viudez, edad y su cuadrado, as como una variable continua sobre escolaridad. 6.6. Cambios en otros ingresos Para el caso de la simulacin de otros ingresos (ingresos de empleadores, subsidios, ganancias de capital, del trabajo para poblacin menor de 15 aos y mayor de 65 aos, entre otros) se utiliza un mtodo simple no paramtrico, el cual ordena a los hogares de acuerdo al ingreso monetario per cpita, descontados
15 16
Se utilizan las mismas categoras definidas para las ecuaciones de participacin y salarios: 1-6, 7-8, 9-10, 11-12, 13-16 y17+), donde 0 es la categora omitida. Categorizado por 1, 2, 3 y 4+, donde 0 es la categora omitida.
282
los ingresos laborales y de pensiones. Los hogares son ordenados por percentiles y el valor observado en la distribucin de Chile es reemplazado por el valor de Uruguay para el mismo percentil. La principal ventaja de este mecanismo es su simplicidad, aunque presenta la debilidad de no reconocer la diversidad de fuentes de este componente. 7. Anlisis de Resultados 7.1. Educacin Las diferentes tasas de retorno a la educacin formal en Chile y Uruguay dan cuenta de 1,5 de los 8,5 puntos de diferencia del ndice de Gini entre ambos pases, es decir, esta variable explica alrededor de un 20% de las diferencias distributivas entre ambos pases. Esta diferencia se concentra en los mayores retornos a la educacin secundaria en Uruguay que en Chile, lo cual ocurre tanto para aquellos trabajadores con educacin secundaria incompleta y completa y para ambos sexos. 7.2. Participacin laboral de la mujer La mayor tasa de participacin laboral de la mujer en Uruguay que en Chile (por ejemplo, un 49,7% versus 39,7% de las mujeres de Uruguay y Chile tienen empleos con remuneraciones positivas) es una caracterstica para todos los centiles de ingreso per cpita por hogar, aunque esta diferencia se hace mayor entre las colas de la distribucin, as como tambin tiende a atenuarse hacia los centiles de mayores ingresos de Chile a pesar de ello, la participacin laboral de la mujer uruguaya supera sistemticamente la participacin laboral de las mujeres chilenas pertenecientes al 20% de los hogares de mayores ingresos. Esta situacin es descrita en el Grfico 10, donde la tasa de participacin femenina incluye slo aquellas mujeres que trabajan y perciben ingresos monetarios. Para ambos pases se puede apreciar una tendencia similar respecto a la mayor participacin laboral de las mujeres, a medida que pertenecen a mayores centiles de ingresos, aunque tiende a estabilizarse a partir del centil 80. La literatura es bastante clara en que esta situacin es endgena, puesto que es la misma participacin laboral de las mujeres la que permite que los ingresos de los hogares se incremente. La simulacin de la mayor participacin laboral de las mujeres uruguayas sobre la distribucin de ingresos de Chile implica una reduccin en 0,5 puntos en el ndice de Gini, indicando que este componente tiene una reducida capacidad de explicacin en la elevada brecha que separa la distribucin del ingreso entre ambos pases. Esta menor relacin entre participacin laboral y recepcin de ingresos monetarios del trabajo entre las mujeres se ve potenciada entre los hombres, puesto que el porcentaje de hombres de 15-65 aos y que son perceptores de ingresos laborales es slo 57,2% en Uruguay, mientras que en Chile es cercana al 70%. Este factor implica que al simular la participacin laboral de hombres y mujeres de Uruguay en Chile, el Gini chileno alcanzara un ndice de 58,3 (simulacin 8 en Cuadro 11), 3 puntos ms alto que la situacin inicial.
283
GRFICo 10
TASA DE PARTICIPACIN LABoRAL MUJER 15-65 AoS PoR CENTIL DE INGRESoS (Slo aquellas con ingresos positivos) 0,8 0,7 Porcentaje de mujeres 0,6 0,5 0,4 0,3 0,2 0,1 0
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63 65 67 69 71 73 75 75 77 79 81 83 85 87 89 91 93 95 97 99
Centil de ingresos per cpita por hogar Part. muj. Uruguay Part. muj. Chile
7.3. Pensiones y jubilaciones En el Cuadro 11 las simulaciones 10-12 indican los efectos separados, por precio y participacin, y el efecto total de simular la distribucin del sistema de pensiones uruguayo en Chile. Aunque el promedio total de las pensiones recibidas por personas de 60 o ms aos en Uruguay son mayores que aquellas recibidas por su pares chilenos, stas no presentan un patrn muy claro respecto a la diferencia relativa, por centil de ingreso (Grfico 11); sin embargo, para los centiles de menores ingresos las pensiones y jubilaciones promedio en Chile son entre un 5%-20% ms altas que en Uruguay, mientras que entre los hogares de centiles ms altos las pensiones promedio de Uruguay son alrededor de un 5%. Sobre la base de estos antecedentes es posible anticipar cambios redistributivos menores en la desigualdad de Chile debido a efectos de precio o monto de las subvenciones percibidas, lo cual es consistente con lo observado en los resultados de la simulacin 10. Una situacin similar es observada al incluir el efecto de cobertura y el efecto total (simulaciones 13 y 14), donde la simulacin de pensiones de Uruguay no altera la situacin distributiva chilena, lo cual se asocia a que son los hogares uruguayos de mayores centiles los que tienen una mayor cobertura social y que la dispersin en el monto de las pensiones promedio y por centil de ingreso es bastante similar a la situacin chilena.
284
GRFICo 11
DIFERENCIA DE PENSIoNES PRoMEDIo PoR CENTILES DE INGRESo DEL HoGAR (CHILE-URUGUAy) Slo poblacin mayor de 65 aos
0,5 0,4 0,3
Diferencia en logaritmo
0,2 0,1 0 0,1 0,2 0,3 0,4 Centil de ingreso per cpita por hogar
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63 65 67 69 71 73 75 77 79 81 83 85 87 89 91 93 95 97
7.4. Otros ingresos no laborales (excluye pensiones) La simulacin del resto de los ingresos no laborales, los cuales incluyen los ingresos de los empleadores, explica por s solo el 70% de la diferencia en el ndice de Gini entre ambos pases (ver simulacin 13 en Cuadro 11), inclusive su efecto marginal es an ms alto al comparar la simulacin con todos los factores que afectan la distribucin del ingreso e incluir adicionalmente este componente17. Este resultado es consistente con el fuerte impacto que tiene la estimacin de indicadores de desigualdad cuando se excluye de las estimaciones al 2% de la poblacin de mayores ingresos per cpita (Cuadro 8), que son aquellos que concentran parte importante de los otros ingresos no laborales. Los Grficos 12 y 13 indican el logaritmo del ingreso per cpita por hogar total y proveniente de ingresos no laborales (incluye pensiones) por centil, donde se aprecia que entre los centiles 2-95 los ingresos no laborales son mayores en Uruguay que en Chile en Uruguay un 44,7% del total de los ingresos per cpita por hogar corresponden a otros ingresos, mientras que en Chile este porcentaje es slo 35,6% y que, en ambos pases, la participacin relativa de este componente del ingreso es relativamente estable. Sin embargo, las diferencias entre los pases se presentan en el extremo superior de la distribucin. Entre los centiles 95-100 no slo el ingreso per cpita de los hogares chilenos es mayor que el de los uruguayos, sino que paulatinamente los ingresos no laborales se hacen ms relevantes, con los casos extremos de los centiles 99 y 100, donde los ingresos no laborales en Chile son mayores que en Uruguay y dan cuenta de una alta concentracin del total de estos ingresos en este grupo (38,6% del
17
Para ello se pueden comparar los resultados de las simulaciones 9 y 14 del Cuadro 11.
CUADRo 11
SIMULACIoNES SoBRE LA DISTRIBUCIN DEL INGRESo CASEN 2003 zoNAS URBANAS UTILIzANDo CoRRECCIoNES DE URUGUAy 2005
Ingreso per cpita por Hogar US$PPP Gini 55,67 47,12 9,72 3,07 0,32 10,20 3,61 0,35 3,21 3,26 90/10 90/50 10/50 75/25 484,0 390,1
Theil 68,64 40,50
Estimacin Efectiva Chile
Estimacin Efectiva Uruguay
Coeficientes de Precio 622,6 810,1 427,1 54,10 53,44 56,31 10,82 11,24 10,53 3,55 3,51 3,63 0,33 0,31 0,35 3,32 3,35 3,28 61,75 59,23 72,22
3 4 5
ed, Mujer y Hombre todos, Mujer y Hombre todos y , Mujer y Hombre
Coeficientes de Participacin Laboral 500,9 511,9 454,3 55,18 55,03 55,83 10,26 10,32 10,76 3,59 3,62 3,63 0,35 0,35 0,34 3,20 3,23 3,33 66,98 66,21 69,99
6 7 8
, slo Mujer , Mujer y Hombre , todos y , Mujer y Hombre
Variables Demogrficas 457,4 56,37 11,12 3,67 0,33 3,43 70,92
Xed, Xsize, , todos y : Mujer y Hombre
Pensiones y Otros Ingresos No Laborales 490,1 486,1 496,2 409,9 398,6 55,38 55,60 55,34 49,78 48,56 10,17 10,20 10,28 9,00 10,05 3,60 3,62 3,59 3,26 3,28 0,35 0,36 0,35 0,36 0,33 3,18 3,19 3,18 3,08 3,31 67,60 68,37 67,35 49,49 48,01
10 11
12 13 14
Pensiones/Jubilaciones Efecto Precio Pensiones/Jubilaciones Efecto Participacin Pensiones/Jubilaciones Efecto Total otros Ingresos No Pensiones Pensiones y otros Ingresos Xed, Xsize, , todos y
285
286
total de los ingresos no laborales del pas, donde 30,0% corresponde al centil 100). Mientras que entre los hogares uruguayos las jubilaciones y pensiones son una fraccin creciente de los ingresos no laborales de los hogares de mayores centiles, en Chile su relevancia disminuye sistemticamente a medida que se avanza en el centil de ingresos de los hogares, pasando a representar slo entre el 2%-5% entre los centiles de mayores ingresos. GRFICo 12
INGRESo PER CPITA ToTAL y DE INGRESoS No LABoRALES CHILE y URUGUAy zoNAS URBANAS
10 9
Ingreso promedio per cpita
8 7 6 5 4 3 2 1 0
1 5 9 13 17 21 25 29 33 37 41 45 49 53 57 61 65 69 73 77 81 85 89 93 97
Centil ingreso per cpita por hogar lningpcUR lningpcCH lningNOLpcUR lningNOLpcCH
GRFICo 13
INGRESo PER CPITA ToTAL y DE INGRESoS No LABoRALES CHILE y URUGUAy zoNAS URBANAS CENTILES 95-100
9,5 9
Ingreso promedio per cpita
8,5 8 7,5 7 6,5 6 5,5 5 95 96 lningpcUR 97 98 99 Centil ingreso per cpita por hogar lningpcCH lningnolabUR 100 lningnolabCH
287
Debido a la diversidad de fuentes que conforman este ingreso laboral no monetario, es interesante analizar separadamente los efectos que cada una de estas fuentes tiene sobre la distribucin del ingreso, lo cual se propone como seguimiento de este proyecto de investigacin. 8. Sensibilizacin de las Estimaciones Debido a Correcciones de los Ingresos Declarados por los Hogares Todas las estimaciones realizadas con la encuesta Casen para Chile consideran ingresos que han sido corregidos por Cepal en funcin de una estimacin de la sub o sobredeclaracin de diversas fuentes de ingresos de los hogares, basadas en las cuentas nacionales preparadas anualmente por el Banco Central de Chile. Para la encuesta Casen del ao 2003 los factores de ajuste promedio para los diversos componentes del ingreso del hogar fueron los siguientes (Mideplan, 2005): i) Sueldos y salarios (1,000), lo que implica que no hubo correccin en su media; ii) Ingresos del trabajo independiente y de los ingresos de los empleadores (1,976); iii) Prestaciones de la Seguridad Social (1,145); iv) Rentas de la Propiedad (1,028), y v) Alquiler Imputado (0,437). Los resultados anteriores dan cuenta que en general la correccin es por subdeclaracin, puesto que slo los montos indicados por alquiler imputado estn sobredeclarados por las familias (este componente representa un 15% del total de ingresos familiares). Por otra parte, la concentracin de la correccin de Cepal afecta los ingresos provenientes del trabajo de trabajadores independientes y empleadores, puesto que la correccin implica casi duplicar los ingresos declarados por este componente, lo cual puede afectar considerablemente la distribucin del ingreso estimada para el pas. Lamentablemente no es posible acceder a las bases de informacin de Casen sin las correcciones de Cepal, por lo cual no sabemos cul sera la distribucin del ingreso del pas en base a los datos declarados por los hogares, metodologa habitual para el resto de los pases de Amrica Latina, ni tampoco sabemos si estos factores de correccin por sub o sobredeclaracin de ingresos es diferenciado por grupos socioeconmicos o los afecta diferenciadamente. Las autoridades chilenas tienen el desafo urgente de facilitar estas bases de informacin, lo cual permitir hacer ms transparente el proceso de correccin de los datos Casen, as como realizar comparaciones internacionales con metodologas similares. Debido a los resultados obtenidos a partir de las microsimulaciones realizadas en nuestro trabajo, se presenta la hiptesis que una parte importante de las diferencias en la distribucin del ingreso entre Chile y Uruguay sean explicadas por los factores de ajuste aplicados por Casen a los diferentes componentes del ingreso de los hogares, en particular, debido a que la mayor parte de la explicacin de las diferencias entre ambos pases estn asociadas a otros ingresos monetarios, donde la principal fuente son los ingresos de los empleadores estos ltimos consideran un factor de ajuste que casi duplica su valor respecto al declarado por los hogares encuestados. Con el fin de resolver esta pregunta realizamos una correccin de los ingresos de los empleadores de Chile para el ao 2003 por medio de la divisin de este componente de ingresos por el factor de ajuste utilizado por Cepal para su
288
correccin (1,976), de tal forma de utilizar esta fuente de ingresos de acuerdo a lo declarado efectivamente por los entrevistados. El resto de los ingresos monetarios de los hogares no fueron modificados. Los resultados de esta simulacin se encuentran en el Cuadro 1 del Anexo III. Los principales efectos de esta simulacin son la reduccin de un 10% en el ingreso per cpita por hogar y en la cada de 3 puntos en el ndice de Gini estimado para Chile en el 2003. Esto significa que slo esta correccin implica que la diferencia en este indicador entre Chile y Uruguay se reduce desde 8,5 puntos a 5,5 puntos en el perodo analizado, es decir, que un tercio de esta diferencia es explicada slo por una correccin metodolgica que slo se aplica en Chile. Este resultado se vuelve muy sensible para la realizacin de cualquier comparacin internacional de la distribucin del ingreso como la que realizamos en este estudio puesto que el resto de los pases de Amrica Latina no utiliza la metodologa de correccin de ingresos familiares utilizada por Chile, generando una sobrestimacin de las diferencias de Chile con otros pases de la regin o, como contrapartida, una potencial subestimacin de los indicadores de la distribucin del ingreso del resto de los pases de la regin. Sin embargo, al reestimar todas las microsimulaciones realizadas en el captulo anterior es posible mantener las principales conclusiones ya analizadas, donde sobresale el hecho de que la principal causa que explica la diferencia en la distribucin del ingreso entre ambas economas, con cerca del 50% de sta, se concentra en los mayores ingresos provenientes de otras fuentes de ingresos principalmente compuestos por los ingresos de los empleadores que perciben las familias chilenas de los centiles superiores de la distribucin, donde factores de la economa uruguaya, tales como la mayor participacin de la mujer en el mercado del trabajo, los mayores retornos a la educacin o la poltica de pensiones, no tienen una principal incidencia en esta diferencia. 9. Conclusiones Finales Las simulaciones de los ingresos de los hogares de Uruguay en la economa chilena permitieron explicar siete de los ocho puntos y medio de diferencia entre los ndices de Gini. Estas replicaron bastante bien las diferencias en los ingresos monetarios de los hogares pertenecientes a los centiles 6-98 (Grfico 14), requiriendo mejorar las especificaciones para explicar los extremos de las colas de la distribucin del ingreso. A diferencia de lo esperado, la mayor participacin laboral de las mujeres en Uruguay estuvo asociada marginalmente a una mejor distribucin del ingreso, puesto que slo medio punto del total de la diferencia fue explicada por esta variable. Sin embargo, tal como se esperaba, si Chile presentara la menor convexidad de los retornos educacionales de Uruguay, la economa chilena reducira en 1,5 puntos su ndice de Gini, es decir, cerca del 20% del total de la diferencia en este indicador es explicada por esta variable. Esta magnitud
289
no es mayor, pues la mayor equidad educacional en Uruguay, respecto a Chile, se concentra en sus mayores retornos relativos a la educacin secundaria, siendo una caracterstica de ambos pases el alto premio relativo a la educacin postsecundaria. A su vez, el sistema de pensiones uruguayo, de mayor cobertura y mayores pensiones relativas, no explica las diferencias distributivas entre ambos pases. Ello puesto que los efectos de la mayor cobertura en Uruguay, mayor entre los centiles de mayores ingresos, tenderan a neutralizar los efectos positivos en la distribucin por cambios en los montos de stas. A pesar de lo anterior, la principal explicacin para la mayor parte de distribucin ms inequitativa del ingreso en Chile se vincula a los ingresos monetarios no laborales, excluidas las pensiones, los cuales estn mucho ms concentrados en Chile que en Uruguay. En Chile ms del 40% del total de estos ingresos son percibidos por el 2% de los hogares de mayores ingresos, mientras que en Uruguay este porcentaje es poco menos del 20%. Este componente explica casi 6 de los 8,5 puntos de diferencia en el coeficiente de Gini. Por otra parte, parece indispensable seguir profundizando en el anlisis ms detallado acerca de los efectos diferenciados que pueden tener los diferentes componentes de estos otros ingresos, subsidios monetarios, ingresos de empleadores, etc., en la distribucin. Respecto al punto anterior, el estudio realiza una sensibilizacin de los resultados respecto de la correccin que se realiza a los datos de ingresos monetarios declarados por los hogares chilenos, los cuales son corregidos en Cepal por la subdeclaracin para diferentes componentes de los ingresos en funcin de su consistencia con las cuentas nacionales preparadas por el Banco Central. Aunque no fue posible contar con la base de datos original aquella sin correcciones, antecedentes agregados indicados por Mideplan permitieron ajustar los ingresos provenientes del trabajo de los empleadores fuente principal del tem otros ingresos y que adems concentra la mayora de la magnitud de recursos monetarios corregidos por la metodologa aplicada por Cepal. Los resultados dan cuenta de dos efectos sustantivos en la distribucin del ingreso en Chile y su comparacin internacional. El primero da cuenta que slo la correccin a este factor explica casi 3 puntos del coeficiente de Gini de Chile, es decir el Gini estimado sin la correccin correspondera a 52,68 vs el 55,67 estimado con los datos corregidos. Esto implica que dado que el resto de los pases latinoamericanos no corrige sus encuestas de hogares incluyendo las encuestas utilizadas para comparar en caso de Uruguay podra existir una sobrestimacin de la desigualdad en Chile respecto a sus pares o una subestimacin de la desigualdad efectiva en el resto del continente. Para el actual estudio esto significa que la diferencia en el coeficiente de Gini en Chile para el perodo analizado sera de 5,5 puntos adicionales respecto de Uruguay y no de 8,5 puntos. El segundo efecto se refiere a que la reestimacin de las microsimulaciones da cuenta que an la principal fuente que explica esta diferencia distributiva entre ambos pases se concentra en los otros ingresos monetarios, puesto que el 50% de la diferencia entre ambos pases proviene de esta fuente, es decir, las diferencias an estn concentradas principalmente en los mayores ingresos de los deciles superiores de la distribucin de Chile.
290
GRFICo 14
DIFERENCIA DE INGRESo PER CPITA PoR HoGAR CHILE (2003) y URUGUAy URBANo (2005) (Efectiva y simulada) 1 0,8 0,6 Diferencia log 0,4 0,2 0
54 58 62 66 70 74 78 82 86 90 94
-0,2 -0,4 -0,6
Centiles ingreso per cpita por hogar Diferencia estimada Diferencia observada
Bibliografa Bucheli, M. y M. Furtado (2004). Uruguay 1998-2002: caractersticas de los cambios en el perfil de la distribucin del ingreso. Serie Estudios y Perspectivas, Ministerio de Economa y Finanzas de Uruguay y Cepal. Bourguignon, F., N. Fournier y M. Gurgand (2000). Distribution, development and education: Taiwan, 1997-1994. Working Paper 2000-7, cole Normale Superieure-Delta France. Bourguignon, F., F. Ferreira y P. Leite (2002). Beyond oaxaca-Blinder: Accounting for Differences in Household Income Distributions across Countries. Working Paper 2002-04, cole Normale Superieure, Delta, France. Bravo, D., D. Contreras y S. Urza (2002). Poverty and Inequality in Chile 1990-1998: Learning from Microeconomic Simulations. Borrador Universidad de Chile. Cancian, M. y D. Reed (2001). Assessing the Effects of Wives Earnings on Family Income Inequality, Review of Economic Statistics. Cepal (2007). Anuario Estadstico de Amrica Latina y el Caribe, 2006, julio, 2007. Contreras, D. (2003). Poverty and Inequality in a Rapid Growth Economy: Chile 1990-1996, Journal of Development Studies. Ermisch, J., M. Francesconi y T. Siedler (2004). Intergenerational Economic Mobility and Assortative Mating. Discussion Paper 338. German Institute for Economic Research. Feres, J. C. (2007). Confiabilidad de la Medicin del Ingreso en las Encuestas de Hogares. Comisin Econmica y Social para Amrica Latina y el Caribe (Cepal).
98
14
22
38
18
34
42
10
30
50
26
46
291
Ferreira, F. y J. Litchfield (1999). Calm after the Storms: Income Distribution and Welfare in Chile, 1987-94. The World Bank Economic Review. Sept. 1999; 31, 3. Gasparini, L., M. Marchionni y W. S. Escudero (2004). Characterization of Inequality Changes through Microeconometric Decompositions: The Case of Greater Buenos Aires. En: Bourguignon, Ferreira y Leite, The Microeconomics of Income Distribution Dynamics in East Asia and Latin America, The World Bank and oxford University Press. IMF (2004). World Economic Outlook Database. INE (2003). Encuesta Continua de Hogares, Instituto Nacional de Estadsticas de Uruguay. Larraaga, o. y J. P. Valenzuela (2006). Por qu no ha cambiado la distribucin del Ingreso en Chile?. Borrador para ser presentado en el Seminario sobre Distribucin del Ingreso en Santiago de Chile. Mideplan (2003). Encuesta de Caracterizacin Social y Econmica de Hogares, Ministerio de Planificacin de Chile. Mideplan (2004). Encuesta de Caracterizacin Social y Econmica de Hogares: Marco Metodolgico, Ministerio de Planificacin de Chile. Mideplan (2004). Encuesta de Caracterizacin Social y Econmica de Hogares Diccionario de Variables, Ministerio de Planificacin de Chile. Mideplan (2005). Encuesta de Caracterizacin Socioeconmica Nacional Casen 2003: Marco Metodolgico. Departamento de Informacin Social, Divisin Social. Julio 2005. Mideplan (2007). Serie de Anlisis de la Encuesta Casen 2006. N 1 La Situacin de Pobreza en Chile 2006. Pizzolitto, Georgina (2005). Poverty and Inequality in Chile: Methodological Issues and a Literature Review, Working Papers 0020, Cedlas, Universidad Nacional de La Plata. Torche, Florencia (2005). Unequal but Fluid Social Mobility in Chile in Comparative Perspective. American Sociological Review 70 (3): 422-450. www.bps.gub.uy Banco Previsin Social de Uruguay.
292
ANEXOS ANEXo I
CARACTERSTICAS SoCIoDEMoGRFICAS, CIRCA 2005

* Poblacin urbana.
Expectativa de vida 74,8 71,2 78,2 79,0 75,4 75,6
% de la poblacin en edad de trabajar 63,4 66,0 67,0 65,8 63,7 62,5
Aos de educacin 10,7* 6,7 10,5 8,3 8,0 9,7*
% de poblacin urbana 91,4 84,2 87,6 61,7 76,3 92,0
ANEXo II
El componente de Pensiones y Jubilaciones de este estudio contempla las siguientes fuentes de recursos para el caso de Uruguay: Jubilacin comn: existen dos regmenes, uno es de transicin para los trabajadores afiliados al Banco Previsional Social (BPS) y otro mixto, en el que los trabajadores estn afiliados a la BPS y los de mayores recursos a una Asociacin de Fondos de Pensiones (AFAP). La jubilacin es un derecho a los 60 aos con 35 aos de trabajo. La jubilacin por edad avanzada es para personas de 70 aos y con mnimo 15 aos de trabajo reconocido, pero que no cumplen con las condiciones para jubilar. Pensiones de sobrevivencia: ayuda econmica generada a partir del fallecimiento de un afiliado, tienen derecho las personas viudas, los hijos solteros menores de 21 aos (o 18 si ingresan al mercado laboral), los hijos mayores de 21 completamente incapacitados para todo tipo de trabajo; padres totalmente incapacitados y las personas divorciadas que reciben pensin alimenticia. Pensin por vejez: BPS la entrega a las personas mayores de 70 aos que carezcan de recursos. Jubilacin por incapacidad absoluta y permanente para todo trabajo.
El correlato con el caso chileno es el siguiente Uruguay Jubilacin comn Jubilacin por edad avanzada Pensiones de sobrevivencia Pensin por vejez Jubilacin por incapacidad absoluta y permanente para todo trabajo Chile Jubilaciones Jubilaciones Montepos o pensin de viudez Pensin Asistencial (PASIS) Pensin de invalidez
293
MICRoSIMULACIoNES SIN CoNSIDERAR CoRRECCIoNES DE CEPAL PoR CUENTAS NACIoNALES EN INGRESoS LABoRALES DE EMPLEADoRES EN ENCUESTA CASEN 2003 - CHILE
ANEXo III
CUADRo 1
SIMULACIoNES SoBRE LA DISTRIBUCIN DEL INGRESo No considera correccin por subdeclaracin de ingresos monetarios laborales para empleadores en Chile1 Casen 2003 zonas urbanas utilizando correcciones de Uruguay 2005
Ingreso per cpita por hogar US$PPP 1 2 Estimacin Efectiva Chile Estimacin Efectiva Uruguay Coeficientes de Precio 3 4 5 ed, Mujer y Hombre todos, Mujer y Hombre todos y , Mujer y Hombre Coeficientes de Participacin Laboral 6 7 8 , slo Mujer , Mujer y Hombre , todos y , Mujer y Hombre Variables Demogrficas 9 Xed, Xsize, , todos y : Mujer y Hombre Pensiones y Otros Ingresos No Laborales 10 11 12 13 14 Pensiones/Jubilaciones Efecto Precio Pensiones/Jubilaciones Efecto Participacin Pensiones/Jubilaciones Efecto Total otros Ingresos No Pensiones Pensiones y otros Ingresos Xed, Xsize, , todos y 450,5 446,4 456,6 410,0 398,5 52,44 52,64 52,45 50,01 49,78 9,66 9,77 9,79 9,07 10,11 3,45 3,49 3,46 3,31 3,33 0,36 0,36 0,35 0,37 0,33 3,15 3,15 3,13 3,07 3,31 57,19 57,81 57,09 50,10 48,63 415,9 53,00 10,62 3,54 0,33 3,35 58,56 461,3 472,3 414,6 52,29 52,22 52,57 9,85 9,90 10,24 3,47 3,50 3,39 0,35 0,35 0,34 3,13 3,15 3,26 56,64 56,40 58,15 583,0 770,5 387,4 51,96 52,02 52,82 10,35 10,95 9,97 3,42 3,43 3,47 0,33 0,31 0,35 3,28 3,29 3,19 54,70 54,91 59,32 441,6 390,1
Gini
90/10
90/50
10/50
75/25
Theil
52,68 47,12
9,73 9,72
3,46 3,07
0,36 0,32
3,16 3,26
57,97 40,50
Se ha utilizado el factor de ajuste para ingresos del trabajo independiente (trabajadores por cuenta propia y empleadores), estimado por Cepal para el ao 2003, el cual alcanz a 1,976, por lo cual los ingresos monetarios del trabajo de los empleadores indicados por Casen 2003 fueron divididos por este factor.
Estudios de Economa. Vol. 38 Osvaldo Larraaga, Juan Pablo Estabilidad en la desigualdad- /N 1, Junio 2011. Pgs. 295-329 Valenzuela
295
Estabilidad en la desigualdad. Chile 1990-2003*

Inequality stability. Chile 1990-2003
Osvaldo Larraaga** Juan Pablo Valenzuela*** Resumen Este estudio mide los impactos de los cambios en los determinantes de los ingresos en la desigualdad en el perodo entre 1990 y 2003, con el objeto de responder la pregunta sobre por qu la distribucin del ingreso como un todo no ha cambiado. La metodologa utilizada es microsimulaciones de la distribucin de ingresos, que es la tcnica ms apropiada para analizar la relacin entre cambios en los factores determinantes y cambios en la desigualdad de ingresos. Se analiza el rol de los retornos, las tasas de participacin, decisiones ocupacionales, dotaciones de escolaridad, subsidios, pensiones y el tamao de los hogares. La inercia mostrada por la desigualdad refleja la interrelacin de factores que se cancelan unos a otros, otros que operan lentamente en el tiempo, y la emergencia de nuevos desarrollos que afectan la distribucin. Adems, no se observan claras indicaciones de que esta situacin cambie en los prximos aos. Un progreso en esta rea requerir una poltica pblica ms activa que en el pasado. Palabras clave: Microsimulacin, Distribucin del ingreso, Desigualdad, Pensiones, Participacin laboral, Retornos a la educacin. Abstract This study measures the impact of changes in the income determinants on inequality in the 1990 to 2003 period, in order to answer the question of why income distribution as a whole has not changed. The methodology utilized are micro-simulations of income distribution, which is the most appropriate technique for analyzing the relationship between changes in determinant factors and
Se agradecen los comentarios recibidos en seminarios en el Departamento de Economa de la Universidad de Chile, el Instituto de Economa de la Universidad Catlica y del seminario de Desigualdad del Ingreso en Chile organizado por el Centro de Microdatos de la Universidad de Chile y el BID. Como es habitual, la responsabilidad de los contenidos de esta investigacin es exclusiva de los autores. ** Departamento de Economa, Universidad de Chile y PNUD. *** Programa de Investigacin en Educacin, Centro de Estudios Avanzados en Educacin, Universidad de Chile.
296
changes in income inequality. It is analyzed the role of returns, participation rates, occupational choices, schooling endowments, subsidies, pensions and household size. The inertia shown by inequality reflects the interplay of factors that cancel each other out, others that operate slowly over time, and the emergence of new developments that affect distribution. Furthermore, there are no clear indications that this situation will change over the next few years. Progress in this area will require a more active public policy than in the past. Key words: Microsimulation, Income distribution, Inequality, Pensions, Labor participation, Returns to schooling. JEL Classification: D3, J2, J3. 1. Introduccin A partir de 1990 los gobiernos en Chile han impulsado una poltica de crecimiento con equidad. sta tiene por objetivo la reduccin de la desigualdad, as como de la pobreza y la exclusin social, en el marco de una economa que deposita en el funcionamiento de los mercados y la iniciativa privada los procesos de asignacin de recursos y de crecimiento econmico. Para lograr los objetivos de equidad se ha incrementado fuertemente el gasto social en servicios tales como educacin, salud y vivienda, as como en numerosos programas de inversin social orientados a apoyar a los grupos vulnerables de la poblacin. Entre 1990 y 2003 el gasto social medido en trminos reales ha aumentado en un 125%, equivalente a una tasa promedio anual de 6,45%1. Los resultados a la fecha han sido mixtos. La economa chilena creci a una tasa promedio de 5,5% entre 1990 y 2003, posibilitando importantes aumentos del empleo y de los salarios reales. Ello ha facilitado la reduccin del porcentaje de pobreza, desde un nivel inicial de 38% en 1990 a un 18,8% en el ao 2003 (encuestas Casen, aos respectivos). Los citados logros en crecimiento econmico y reduccin de la pobreza han situado al pas como el caso de desarrollo econmico reciente ms exitoso de Amrica Latina, as como uno altamente destacado a nivel mundial. Por su parte, el aumento en el gasto social se traduce en una mayor disponibilidad de servicios sociales en educacin, vivienda y salud, representando aumentos en el bienestar a travs de dimensiones complementarias al ingreso. Sin embargo, la desigualdad de ingresos se ha mantenido prcticamente inalterada en el perodo 1990-2003. El coeficiente de Gini es el mismo en ambos aos (0,56); la distancia entre los percentiles 90 y 10 cae slo algunas dcimas (11,14 a 10,61), a la vez que la razn entre los quintiles quinto y primero es 17,9 y 17,6 en los respectivos aos. Al interior del perodo no hay una tendencia bien definida, aun cuando algunos indicadores presentan un aumento hasta el ao 2000 seguido de un cierto aflojamiento en la medicin del 2003.
Incluye el gasto en educacin, salud, vivienda y proteccin social. Direccin de Presupuestos: Estadsticas de las Finanzas Pblicas 1987-2003.
Estabilidad en la desigualdad / Osvaldo Larraaga, Juan Pablo Valenzuela
297
La relativa estabilidad que presenta la desigualdad en el perodo es sorprendente si se considera que ha habido cambios significativos en los determinantes prximos de la distribucin de ingresos. Este ha sido el caso de la participacin laboral de la mujer, los niveles de educacin de la fuerza de trabajo, la demografa de los hogares, la estructura de retornos de la educacin y los ingresos provenientes de las pensiones y de los subsidios monetarios. En un lapso de quince aos Chile ha experimentado niveles de cambios en estas variables que en otros pases pueden tomar bastante ms tiempo. Bajo tal contexto se podran haber esperado modificaciones en la distribucin de ingresos, si bien esta variable tiende a ser estable en el corto plazo. En este trabajo se mide el impacto de los cambios en los determinantes de los ingresos sobre la desigualdad en el perodo 1990 a 2003, de manera de responder a la pregunta del por qu no cambi la distribucin de ingresos en su conjunto. Se debe a una compensacin entre los factores determinantes, unos empujando al alza y otros a la baja de la desigualdad? O bien estamos en presencia de una estructura de desigualdad que es resistente frente a cambios en los determinantes de los ingresos? Tambin interesa identificar aquellos factores relacionados con polticas pblicas que pudieran jugar un rol importante a futuro en la reduccin de la desigualdad. La metodologa utilizada son microsimulaciones de la distribucin de ingresos, que representa el estado del arte en las tcnicas disponibles para analizar la relacin entre cambios en factores determinantes y cambios en la desigualdad de los ingresos. Un artculo pionero en la materia es Juhn, Murphy y Pierce (1994), donde se utiliza este tipo de tcnica para analizar la distribucin de los salarios en los Estados Unidos. Posteriormente, la tcnica se generaliza para analizar los cambios en la distribucin del ingreso de los hogares. El reciente volumen de Bourguignon, Ferreira y Lustig (2005) contiene un conjunto de aplicaciones de la microsimulacin del ingreso de los hogares para pases de Amrica Latina y el sudeste de Asia, as como una recopilacin terica de la metodologa y sus antecedentes. Para Chile se dispone de un trabajo previo, que analiza los cambios en la distribucin entre el perodo 1990 y 1998 (Bravo, Contreras y Urza, 2002). El presente trabajo contiene una innovacin metodolgica respecto de los trabajos citados, ms all de representar una extensin temporal al trabajo citado para Chile. Esta consiste en la simulacin de los ingresos de pensiones y de subsidios monetarios, en adicin a los ingresos laborales que conforman el eje de los anlisis previamente realizados. El trabajo se organiza en cinco secciones, aparte de esta introduccin. La siguiente seccin presenta hechos estilizados en materia de distribucin de ingresos y sus determinantes prximos en el perodo 1990 a 2003, incluyendo una descomposicin contable del cambio en el ingreso per cpita de los deciles en trminos de las diferentes fuentes de tal ingreso. La tercera seccin presenta la metodologa de microsimulaciones de la distribucin de los ingresos. La cuarta seccin presenta los datos y las estimaciones en que se basan los resultados del trabajo. La quinta seccin presenta los resultados de las microsimulaciones de la distribucin de los ingresos. Una ltima seccin presenta las conclusiones del trabajo.
298
2. Estructura y Cambios en la Distribucin de los Ingresos 2.1. El perodo 1990-2003 La desigualdad de ingresos en el ao 2003 es similar a la exhibida en 1990 (Cuadro 1). El coeficiente de Gini es 0,56 en ambos aos; la distancia entre los percentiles 90 y 10 cae algunas dcimas en el perodo, de 11,14 a 10,61, a la vez que la razn entre los quintiles quinto y primero es 17,9 y 17,6 en los respectivos aos. El ndice de Theil es el nico indicador que muestra un deterioro, el cual se asociara a cambios en la parte superior de la distribucin que este ndice refleja con mayor fuerza. La evolucin de la desigualdad al interior del perodo no muestra tendencias bien definidas, aun cuando la mayor parte de los indicadores presenta un aumento hasta el ao 2000, para luego descender en la medicin del 2003. Habr que esperar las prximas mediciones para conocer si este ltimo movimiento representa un quiebre de tendencia o si es una fluctuacin aleatoria en torno a un nivel promedio que se mantiene alto y estable. La caracterizacin de la desigualdad se realiza sobre la base del ingreso monetario de los hogares expresado en trminos per cpita. Los indicadores de desigualdad se calculan sobre la base de individuos, de modo que cada una tiene un peso equivalente al nmero de personas que all viven2. El servicio domstico que reside en el hogar del empleador se considera como unidad independiente. La inercia que presentan los indicadores de desigualdad contrasta con los fuertes cambios que caracterizan a los determinantes prximos de los ingresos (Cuadro A-1 en Anexo)3. En el perodo 1990 a 2003 el volumen de actividad econmica medida por Producto Interno Bruto prcticamente se duplica, el empleo crece en un 33% y los salarios reales en un 51%. La poblacin crece a una tasa anual de 1,43%, acumulando un crecimiento de 20,3% en el perodo 1990-2003. La estructura de edades presenta una tendencia hacia el envejecimiento: en el ao 2003 hay mayor proporcin de adultos mayores y menor proporcin de nios respecto del ao inicial. En el perodo se produce un aumento importante del nivel educacional de los ocupados, disminuyendo la fraccin de ocupados con educacin bsica e incrementndose aquella con estudios secundarios completos y la que ha cursado estudios de nivel superior. No obstante lo anterior, los salarios aumentan proporcionalmente ms para personas con mayor nivel de escolaridad. Los ocupados con estudios universitarios presentan incrementos reales superiores al 50% en sus remuneraciones en el perodo 1990-2003, mientras que los egresados de la educacin media muestran aumentos de ingresos en el entorno del 20% (Cuadro A-2 del Anexo).
2 3
Este procedimiento considera que la unidad ltima del bienestar es el individuo (Deaton, 1997). Por determinantes prximos se entienden aquellos que influencian directamente a los ingresos, como es el nivel de escolaridad. A su vez, estos factores tienen un conjunto de determinantes (ingreso de los padres, acceso a establecimientos educacionales, etc.). El trabajo analiza el impacto del primer grupo de factores.
299
CUADrO 1
DISTrIBUCIN DEL INGrESO PEr CPITA DE LOS HOGArES, 1990 y 2003
Promedio M$, 2003 1990 2003 91,0 142,1
Gini 56,0 56,0
Q5/Q1 17,9 17,6
90/10 11,14 10,61
90/50 3,85 3,64
10/50 0,35 0,34
75/25 3,32 3,26
Theil 66,8 69,7
Fuente: Encuestas Casen, aos respectivos. Nota: Corresponde a la distribucin del ingreso per cpita monetario del hogar a nivel de individuos (incluye ingresos igual a cero).
La tasa de participacin de la mujer se incrementa en ms de 11 puntos porcentuales en el perodo bajo estudio (de 35,0% en 1990 a 46,3% en el ao 2003). El aumento en la participacin no es homogneo, sino que se concentra en los grupos que presentan mayores espacios de crecimiento como son las mujeres de escolaridad baja y media. En cambio, las mujeres con educacin superior ya presentaban tasas de participacin cercanas a un 80% en el ao 1990, de modo que tiene poco espacio para seguir creciendo. En el perodo se produce tambin una reduccin en el tamao de los hogares. Ello responde en parte a la cada en la tasa de natalidad, la cual se inscribe en un proceso que se inicia en los aos 60, y que es ms importante en el caso de las mujeres de menor estrato socioeconmico (Larraaga, 2005 b). La reduccin en el tamao del hogar tambin se asocia con un crecimiento acelerado de viviendas, cuyo incremento de 30% en el perodo supera con creces el aumento en la poblacin de personas (21%). Ello posibilita la creacin de nuevos hogares y la cada en el tamao medio del hogar. Sin embargo, no hay una reduccin en la participacin relativa de los hogares con ms de un ncleo, contrario a lo esperado en el contexto de la fuerte expansin en el nmero de viviendas que se produce en el perodo (Cuadro A-3). Es posible que ello refleje el aumento en los ncleos secundarios asociado al incremento en el porcentaje de mujeres que son madres solteras, las cuales tienden a conformar un segundo ncleo al interior de sus hogares paternos (Larraaga, 2006 b). 2.2. La poltica social en Chile La poltica social en el pas consiste en un conjunto de prestaciones, transferencias y programas que pueden agruparse en cuatro categoras: subsidios monetarios, seguridad social, servicios sociales y programas de desarrollo social. El primer grupo incluye los subsidios que se entregan a hogares pobres; el segundo grupo son los pagos y transferencias que descansan en una lgica contributiva; el tercer grupo representa la poltica social tradicional en los sectores de educacin, salud y vivienda, mientras que la cuarta categora incluye a
300
una variada gama de programas de ms reciente creacin y orientada a grupos especficos de la poblacin4. En trminos cuantitativos, el gasto en subsidios monetarios representaba en el ao 2002 el 3,7% del gasto social; los pagos de seguridad social un 39,8%; los servicios sociales un 50,4% y los programas de inversin social un 6,1%. La estrategia de crecimiento con equidad ha sido exitosa en elevar los recursos que se destinan a los programas sociales: entre los aos 1990 y 2003 el gasto social prcticamente crece en un 145%. No obstante lo anterior, la poltica social ha sido poco efectiva para lograr los objetivos de equidad (ver Larraaga, 2006c). 2.3. La estructura de los ingresos Las principales fuentes de ingreso de los hogares son los salarios, el trabajo independiente y el ingreso de los empleadores, que en conjunto representaban un 80,8% del ingreso monetario de los hogares en el ao 2003 (Cuadro 2). Las pensiones, los subsidios monetarios y el resto de los ingresos representan en su conjunto el 19,2% restante del ingreso de los hogares. Debe tenerse presente que los ingresos reportados en las encuestas de hogares no incluyen partidas como las utilidades retenidas de las empresas, y que captura solo de modo imperfecto las ganancias de capital e ingresos relacionados. La partida de ingreso ms importante son los salarios, cuya participacin en el total asciende a 47,7%. Los asalariados representaban un 73% de los ocupados en el ao 2003. El salario promedio asciende a M$ 246,0, considerando aquellos con jornada laboral de 30 y ms horas semanales. Un 50% de los asalariados con jornada completa gana M$ 150,0 o menos al mes. La desigualdad salarial es elevada; el coeficiente de Gini sita al pas entre los pases de la regin con mayor desigualdad salarial5. Los ingresos del trabajo por cuenta propia son la segunda partida en importancia en el ingreso de los hogares, aun cuando son solo la tercera parte de los salarios. Los trabajadores por cuenta propia constituyen un 20% del total de ocupados, obtienen un ingreso medio que es 45% ms alto que los asalariados, presentando mayores niveles de desigualdad. El coeficiente de Gini para los ingresos del trabajo independiente es 47,4%, mientras que la distancia entre los percentiles 90 y 10 es alrededor de siete veces.
4
Hay distintas maneras de clasificar las polticas sociales. De acuerdo a raczynski y Serrano (2005), las funciones de las polticas sociales incluyen un grupo de acciones asistenciales y de proteccin social que compensan problemas originados en el funcionamiento del mercado o de las familias; un segundo grupo de funciones tiene por objetivo la promocin del bienestar, basado en la construccin de capacidades antes que en la reparacin del dao; un tercer grupo de funciones se asocia con la realizacin de derechos sociales de la ciudadana. La combinacin de estos tres grupos de funciones da lugar a una cuarta funcin, referida a la generacin y preservacin de un espacio social de pertenencia de la comunidad. Chile ocupa el lugar cuarto segn el coeficiente de Gini despus de Brasil, Bolivia y Guatemala (De Ferranti, 2004). Sin embargo, hay escasa distancia respecto de la mayora de los pases de la regin, tratndose de un grupo con desigualdad alta y relativamente pareja. Por su parte, la desigualdad medida por la distancia entre los percentiles 90 y 10 es prcticamente el doble de la desigualdad respecto de los pases de Europa del Norte y Europa Central (OECD, 1996).
301
Por su parte, el ingreso de los empleadores representa el 15,7% del ingreso monetario de los hogares. Se trata de una partida de ingresos sustantiva para explicar la desigualdad de ingresos, puesto que representan ms de la cuarta parte del ingreso del decil ms rico. Alrededor de tres cuartas partes del ingreso de los empleadores estn concentradas en el 5% de hogares con mayor ingreso per cpita. Las pensiones representan algo menos del 7% del ingreso de los hogares. Estos pagos incluyen a las pensiones por vejez, viudez, invalidez y orfandad, representando las dos primeras un 94% del pago de pensiones. En su mayor parte corresponden a pagos del antiguo sistema de seguridad social, cuyos montos estn en lnea con las remuneraciones que reciban los pensionados en su ciclo laboral. Ello responde a la segmentacin segn estatus ocupacional del antiguo sistema de pensiones, a pesar de que su denominacin de sistema de reparto sugiere un mayor grado de redistribucin6. De manera que el coeficiente de Gini de los perceptores individuales de pensiones de vejez y viudez pagadas por el Estado asciende a 0,3957, solo algunos puntos por debajo del coeficiente de Gini de los salarios (0,474). CUADrO 2
ESTrUCTUrA DEL INGrESO PEr CPITA DEL HOGAr (%), 2003
Decil 1 2 3 4 5 6 7 8 9 10 Total
Ingreso empleador 0,1 0,3 0,3 0,8 1,0 2,2 2,4 3,6 8,3 26,2 15,0
Cuenta Subsidios Otros Salarios Pensiones propia monetarios ingresos 9,6 11,3 15,5 15,0 17,5 18,3 20,6 22,0 20,1 16,7 17,4 49,3 61,5 62,2 62,5 61,7 59,6 58,0 54,5 51,8 38,3 47,7 8,5 8,9 8,7 9,0 9,2 10,0 9,4 8,4 8,0 4,3 6,7 14,2 7,3 4,6 3,1 2,3 1,7 0,9 0,5 0,2 0,0 1,2 18,3 10,7 8,7 9,6 8,3 8,2 8,6 11,0 11,7 14,4 12,0
Total 100,0 100,0 100,0 100,0 100,0 100,0 100,0 100,0 100,0 100,0 100,0
Fuente: En base a encuesta Casen 2003.
Alrededor del 48% de los pagos de pensiones registrados en la encuesta Casen de 2003 corresponden a pagos del antiguo sistema y solo el 28% corresponde a pagos del nuevo sistema (otro 24% corresponde a las Fuerzas Armadas y otro tipo de instituciones). Ms an, una parte de las pensiones pagadas en el nuevo sistema son fondos pblicos, por concepto de subsidios (pensin mnima) y transferencias en el caso de personas que realizaron contribuciones cuando jvenes en el antiguo sistema (bono de reconocimiento). Los citados coeficientes de Gini estn referidos a la encuesta Casen 2003.
302
Por su parte, los subsidios monetarios son transferencias focalizadas que tienen por objetivo aliviar la situacin de pobreza de personas que no estn en condicin de trabajar. El monto total gastado en subsidios monetarios representa una fraccin marginal del gasto social y equivale a slo el 1,2% del ingreso de los hogares, aun cuando su contribucin es alrededor del 10% del ingreso en el caso de los deciles inferiores8. Desde la perspectiva de la presente investigacin tiene particular relevancia la contribucin que realizan las distintas fuentes del ingreso al crecimiento del ingreso per cpita de los hogares. El Cuadro 3 presenta tal informacin a nivel de deciles de hogares, incluyendo a los cambios en el tamao del hogar. La primera columna del Cuadro 3 presenta la tasa de crecimiento del ingreso per cpita en el perodo 1990-2003. El patrn de crecimiento acumulado en estos aos es bastante parejo entre deciles, fluctuando entre 51,9% y 60,0%, consistente con indicadores de desigualdad estables. La contribucin de los distintos componentes del ingreso per cpita se reporta en las respectivas columnas del citado cuadro. La contribucin de cada componente es igual a la tasa de crecimiento de la respectiva fuente de ingreso multiplicada por su participacin en el ingreso del decil9. El crecimiento en los salarios explica casi la mitad del incremento en el ingreso per cpita de los hogares (47%). La contribucin de los salarios es sorprendentemente uniforme a nivel de los deciles segundo a noveno, representando en estos casos alrededor del 60% del crecimiento del ingreso total. La contribucin de los salarios es ms reducida en los deciles extremos, producto de la mayor importancia de otras fuentes de ingresos en esos hogares. Alrededor de la cuarta parte del crecimiento del ingreso per cpita se explica por aumentos en el ingreso de los empleadores. Estos ingresos se concentran en los deciles nueve y diez, representando un 45% del incremento del ingreso per cpita en el decil superior. Esta ltima cifra es clave en la contribucin de la partida a nivel agregado, dada la fuerte ponderacin del decil diez en el ingreso total. La siguiente partida en orden de importancia es el tamao de los hogares, cuya reduccin a lo largo del perodo explica un 12,3% del crecimiento en el ingreso per cpita de los hogares. El patrn de crecimiento es relativamente parejo a nivel de los deciles de hogares, sugiriendo que hay una compensacin entre factores subyacentes que operan en direcciones contrarias en materia de desigualdad: la reduccin de la natalidad y la creacin de nuevos hogares. Las otras fuentes de ingresos contribuyen marginalmente en el crecimiento del ingreso per cpita de los hogares. No obstante lo anterior, su contribucin a
8
Los subsidios incluyen la pensin asistencial (Pasis), que se paga por vejez o invalidez a personas pobres sin acceso a seguridad social; la asignacin familiar, que se paga a los dependientes de los trabajadores asalariados con menores ingresos; el subsidio nico familiar (SUF) es una asignacin familiar para nios y madres embarazadas de hogares pobres; el subsidio de agua potable paga una fraccin de la cuenta del servicio a los hogares pobres y un subsidio monetario asociado al programa Chile Solidario. La ponderacin referida es el promedio de la tasa de participacin en los aos considerados (1990 y 2003).
CUADrO 3
CrECIMIENTO DEL INGrESO PEr CPITA DEL HOGAr, 1990-2003
Participacin % en crecimiento del ingreso per cpita 1990-2003 Cuenta propia Salarios 48,8 61,7 68,6 70,7 66,5 56,6 68,8 61,2 61,9 29,5 47,0 2,1 1,3 0,0 2,4 4,9 5,3 3,6 6,4 6,1 1,6 1,8 1,5 0,7 28,8 12,4 7,1 4,5 3,1 1,6 0,4 0,0 0,2 0,1 11,4 4,8 1,5 0,4 1,3 2,1 2,3 0,5 1,8 0,6 11,7 10,8 9,7 10,8 9,5 16,5 13,0 10,2 12,0 18,2 12,3 Pensin Subsidio 0,9 3,9 10,5 7,6 12,7 11,4 15,1 20,9 9,8 6,1 8,8 Otros ingresos Tamao hogar Total 100,0 100,0 100,0 100,0 100,0 100,0 100,0 100,0 100,0 100,0 100,0
Decil
Crecimiento 1990-2003 Empleadores
1 2 3 4 5 6 7 8 9 10
0,600 0,583 0,591 0,585 0,584 0,570 0,554 0,552 0,519 0,539
0,0 0,0 2,0 6,7 3,6 12,0 3,7 9,7 27,2 45,3
Total
0,562
26,5
Fuente: En base a encuestas Casen 1990 y 2003.
303
304
nivel de deciles de ingreso puede ser relevante. Este es el caso de los subsidios monetarios para los deciles primero y segundo, el ingreso de los trabajadores por cuenta propia para los deciles medio-altos y el ingreso de las pensiones en el caso de los deciles medios. 3. Metodologa La metodologa a aplicar est basada en microsimulaciones de la distribucin de ingresos de los hogares, de manera de descomponer los cambios observados entre dos perodos en trminos de los distintos determinantes de estos ingresos. Ello posibilita identificar el impacto que cada uno de los factores determinantes tuvo sobre la distribucin de los ingresos de los hogares en el perodo bajo estudio. Para tal efecto considere que el ingreso per cpita del hogar h puede escribirse como: (1) yh = 1 nh n J j j I hi yhi + i =1 j =1
s s I hi yhi + i =1 s =1 n S
yoh
La primera sumatoria representa los ingresos provenientes del mercado del trabajo, donde el suprandice j representa el tipo de ingresos del trabajo y el subndice i representa el perceptor i de ingresos en el hogar. La variable I es una funcin indicador que toma el valor 1 si el ingreso asociado tiene valor positivo y el valor cero en otro caso. La segunda sumatoria incluye el ingreso proveniente de pensiones y subsidios monetarios, representados por el suprandice s. El tercer tipo de ingreso yo representa otro tipo de ingresos, que se suponen exgenos en el ejercicio. El nmero de personas en el hogar es nh. 3.1. Ingresos laborales La descomposicin de los ingresos provenientes del mercado del trabajo sigue cercanamente la metodologa aplicada el texto de Bourguignon, Ferreira y Lustig (2005), que incluye aplicaciones para un conjunto de pases de Amrica Latina y Asia del Este. El procedimiento implica descomponer los cambios que experimenta la distribucin del ingreso laboral entre dos perodos en trminos de tres tipos de efecto: (i) efecto precio, que entrega el cambio en la distribucin del ao base de acuerdo a los precios o retornos que rigen en el ao final; (ii) efecto dotacin, que trata sobre los cambios en la distribucin que se originan en el nuevo vector poblacional de dotaciones o activos laborales que generan ingresos; (iii) efecto participacin, que da cuenta de los cambios en la distribucin de ingresos generados por cambios en la participacin laboral de los miembros del hogar. El efecto precio requiere estimar los retornos de la educacin y dems determinantes de los ingresos laborales en el ao de trmino, para luego simular el efecto que el cambio en los retornos tiene en la distribucin del ao base. La estimacin de los retornos se realiza sobre la base de una ecuacin tradicional de ingresos para cada miembro del hogar y tipo de ingreso laboral:
305
(2)
j j log yhi = X hi j + uhi
para i = 1,......, nh
La participacin laboral se trabaja a travs de un modelo de eleccin mltiple, en el cual se postula que el individuo elige la alternativa laboral que maximiza su nivel de utilidad. El objetivo del modelo es la estimacin de los parmetros que inciden en la decisin laboral (), de tal modo de simular el efecto que sobre la distribucin del ao base tiene el comportamiento del ao final. Concretamente, el modelo a estimar es:
j I hi = 1 si Z hi j + ij j I hi = 0
> Max (0, Z hi j + ij ) si Z hi j + ij 0
(3)
para j = 1,....., J
j = 1,......., J Donde Z es un vector de variables determinantes de la participacin laboral, es el trmino de error de la ecuacin de participacin y j las alternativas laborales (trabaja dependiente, trabaja independiente, no trabaja)10. Es posible que el modelo indique una persona inactiva en el perodo base que hubiese participado en el mercado del trabajo segn los parmetros del ao final. En este caso se precisa imputar el ingreso laboral que esta persona hubiese obtenido en caso de estar ocupada en el primer perodo. Ello se realiza en base a la especificacin de la ecuacin (2) correspondiente al tipo de ocupacin elegida segn (3), incluyendo un componente especfico del individuo (ui) que se deriva de una distribucin aleatoria del residuo11. Finalmente, el efecto del cambio en las dotaciones (X, Z) es estimado a travs de un procedimiento no paramtrico. Considere para tal caso la variable Xj. El mtodo requiere la particin de los individuos en K subgrupos exclusivos y excluyentes, que se relacionan con la diferencia en la tenencia de la respectiva variable. A modo de ejemplo, en el caso de la escolaridad podra dividirse la muestra de individuos pertenecientes a la fuerza de trabajo en subgrupos segn gnero, tramos de edad y zona geogrfica de residencia. De este modo, la poblacin de tamao N se divide en K subgrupos: N ={N1 , N 2 ,.......... N K } Para cada subgrupo k se construye la funcin de distribucin de la dotacin Xj , que asocia a cada valor de la variable el percentil de la distribucin, para luego invertir esta funcin de modo tal que:
10
11
Dado que los anlisis de ingresos laborales corresponden a las personas con valores positivos para esta variable, todas las personas entre 15-65 aos de edad que se encuentran con ingresos laborales iguales a 0 o desempleadas se consideran como no trabajando. Ver Bourguignon y Ferreira (2005).
306
(4)
tjk = F ( X tjk ) k = 1,.....K ; t = 1, 2

X tjk = Ft1 (tjk )
Para cada procedimiento se realiza para cada uno de los perodos bajo estudio, t = 1, 2. Finalmente, se simula en el perodo inicial la dotacin de X del perodo final utilizando los respectivos percentiles de la distribucin. Esto es, se sustituyen los valores observados de X de t = 2 a t = 1, condicional en un mismo percentil. Mientras ms detallada sea la particin de la poblacin en trminos de los subgrupos k, ms aproximada ser la distribucin simulada en t = 1 a la distribucin efectiva en t = 2. El procedimiento supone que existe preservacin de rango en el ordenamiento de los hogares en la distribucin de X a lo largo del tiempo12. (5) X1jk = F11 ( 2jk )
3.2. Ingreso de pensiones y subsidios La simulacin de los cambios en pensiones y subsidios sigue una combinacin de los elementos anteriores. Considere para tal efecto el componente s de este tipo de ingresos, denominado ys. El problema a resolver es simular la distribucin del perodo 2 en la base de datos del perodo 1. Las distribuciones de ys entre ambos perodos pueden cambiar en trminos de la distribucin de los montos de transferencia recibidos as como por la distribucin en el acceso a las mismas. En este caso no se modelan decisiones de comportamiento, dado que el acceso a pensiones y subsidios no responde, por lo general, a decisiones que los individuos u hogares tomen en tiempo presente. Para realizar la simulacin se particiona la poblacin en trminos de subgrupos que representen distintos tipos de acceso y/o monto de la transferencia. A nivel de cada subgrupo se simula la distribucin del perodo 2 en el perodo 1, utilizando un procedimiento anlogo a lo expresado en (4) y (5) anteriores, el cual entrega por resultado la simulacin de la distribucin del monto de la transferencia. Por otra parte, para simular la distribucin en el acceso a la transferencia se estima el modelo: (6) (7) Pr ( y s > 0) = (V + v ) log( y s ) = W + w
Donde (6) representa un modelo probit que estima la probabilidad de acceso a la transferencia ys , mientras que (7) es una ecuacin que explica el monto de la transferencia por medio de un conjunto de factores relacionados W. Sea 1k
12
Alternativamente, puede utilizarse una funcin paramtrica para predecir los valores de X sobre la base de un conjunto de factores relacionados que se observen en ambos perodos.
307
la proporcin de la poblacin en el subgrupo k que accede al beneficio en el perodo 1 y 2k similar fraccin en el perodo 2. La imputacin de los ingresos en t = 1 se realiza sobre la siguiente regla:
k k si 1 2
(8)
s s y1 = F11 ( 2 ) si
s y1 > 0
s s y1 = y1
si
s k k y1 = 0 y G ( ) > (1 ( 2 1 ))
k k si 1 > 2
(9)
s s y1 = F11 ( 2 ) si
s y1 > 0
k y G ( ) 2
s y1 = 0 si
s k y1 > 0 y G ( ) > 2
Si la proporcin de receptores de la transferencia ys en el subgrupo k en el perodo 1 es menor o igual a similar proporcin del perodo 2, se procede a imputar a cada observacin con transferencia positiva en t = 1 el monto correspondiente a su percentil equivalente en el perodo 2. A ello se adicionan individuos que no reciban transferencias en t = 1, de modo de completar la fraccin que accede en el perodo 2, seleccionando a aquellos con mayor probabilidad de acceso predicha a partir de (6). Para estos casos se imputa un valor predicho de la transferencia segn (7). En caso de que la proporcin de receptores de la transferencia ys en el subgrupo k en el perodo 1 sea mayor que la fraccin en el perodo 2, se procede a imputar un monto igual a cero en t = 1 a aquellos individuos que habiendo accedido a una transferencia en el perodo inicial presentan la menor probabilidad de acceso segn (6). El resto de las observaciones con transferencia positiva en t = 1 se le imputa el monto correspondiente a su percentil equivalente en el perodo 2. 4. Datos y Estimacin del Modelo 4.1. Datos Las estimaciones y las simulaciones estn basadas en la base de datos de las encuestas Casen de 1990 y 2003. Esta es una encuesta multitpico de hogares, que incluye mdulos de empleo, ingresos, educacin, salud, vivienda y patrimonio. Se realiza desde el ao 1987 con periodicidad de cada dos o tres aos. La encuesta tiene representatividad nacional, regional y urbano-rural. En el ao 2003 la muestra alcanz a alrededor de 60.000 hogares. Todas las estadsticas de pobreza, distribucin de ingresos y acceso a servicios sociales utilizadas en el pas provienen de las bases de datos de las encuestas Casen. 4.2. Ecuaciones de ingresos Las estimaciones de ingresos, as como las relativas a la participacin en el mercado del trabajo fueron realizadas considerando a las personas de entre 15
308
y 65 aos de edad. La estimacin de los retornos a las competencias laborales se realiza a travs de ecuaciones de ingresos, diferenciando entre trabajadores dependientes e independientes13, as como hombres y mujeres. La ecuacin estimada corresponde a una especificacin de tipo Mincer, donde el logaritmo del ingreso laboral se regresiona en la escolaridad, la experiencia potencial y su cuadrado, y variables dicotmicas para la condicin de ruralidad, la regin de residencia y los trabajadores de jornada parcial. La escolaridad se incluye a travs de diez tramos de acuerdo al ciclo de educacin cursado (sin educacin, primaria incompleta, primaria completa, secundaria incompleta, secundaria humanista completa, secundaria tcnico-profesional completa, educacin tcnica superior incompleta, tcnica superior completa, universidad incompleta y universidad completa), diferenciando entre egreso o no egreso del mismo. El Cuadro A-4 en el Anexo presenta los resultados obtenidos para los aos 1990 y 2003. Los resultados ms interesantes son los retornos de la escolaridad, puesto que una de las principales causas detrs de los aumentos que ha experimentado la desigualdad de ingresos en diversos pases a partir de los aos 80 ha sido una mayor brecha de salarios entre trabajadores calificados y no calificados. En nuestro caso se constata un aumento en el premio salarial para los asalariados hombres con educacin superior y entre aquellos con escolaridad ms baja, a la vez que cae el premio en el caso de quienes tienen estudios de nivel secundario. Los trabajadores independientes hombres muestran un comportamiento anlogo, aunque se incrementa el retorno a la educacin secundaria para los trabajadores por cuenta propia. Hay que considerar que la fuerza de trabajo con escolaridad secundaria presenta la mayor tasa de crecimiento en el perodo de anlisis, lo cual puede estar detrs del patrn descrito de los retornos laborales. En el caso de las mujeres asalariadas hay incrementos en el retorno del grupo con mayor escolaridad, mientras que disminuye el premio salarial en los grupos con educacin bsica y media, as como entre quienes tienen estudios superiores incompletos. En cambio, las trabajadores independientes mujeres presentan aumentos en el premio salarial a lo largo de los distintos niveles de escolaridad. 4.3. Factores no observables Los factores no observables que afectan los salarios son incorporados en las simulaciones por medio de una correccin al trmino de error de las ecuaciones de salarios estimadas en la ecuacin (2). Para simular los efectos de los cambios en los factores no observables ocurridos en el 2003 en el perodo base del ao 1990 se reescalan los residuos de cada estimacin de salarios de 1990 por la relacin de la desviacin estndar estimada para cada ao (2003/1990). 4.4. Ecuaciones de participacin laboral Para estimar la participacin laboral de los trabajadores se consideraron estimaciones multilogit (inactivo, trabajador asalariado o trabajador por cuenta
13
Los empleadores no son considerados dentro de las estimaciones de salarios ni participacin laboral y sus ingresos son estimados como parte del componente de otros ingresos (yoh).
309
propia) por separado para hombres y mujeres, y considerando el rol que cumple la persona en la estructura del hogar (jefe del hogar, pareja14 u otro miembro del hogar). Un conjunto de variables independientes fueron similares en todas las estimaciones, mientras que algunas adicionales se incluyeron para los casos de parejas y otros miembros del hogar (Cuadros A-5 y A-6 en Anexo). Las variables comunes fueron las diez alternativas de educacin formal, la experiencia y su cuadrado, la condicin de ruralidad, as como dummies para las diversas regiones del pas. Tambin se incluyen dummies que sealan si la persona est casada y otra para indicar que tiene una relacin de convivencia; y si la persona est en edad escolar (menor de 25 aos de edad), as como variables continuas respecto al promedio de educacin y de edad de los adultos del hogar. Los controles independientes para el jefe del hogar son una variable que indica que en el hogar vive una sola persona; tambin se incluyen variables demogrficas de los miembros del hogar (el nmero de nios de entre 0 y 8 aos; el nmero de nios de entre 9 y 14 aos; nmero de personas de entre 15 y 65 aos y el nmero de los mayores de 65 aos), adems se incluye el porcentaje de mujeres dentro del hogar. Para las estimaciones de participacin laboral de las parejas se elimina la condicin de hogar unitario y se incorpora una dummy que refleja el sexo de la persona; asimismo se incluye una dummy que indica la condicin laboral del jefe del hogar y una variable continua respecto al salario del jefe del hogar para los casos cuando ste trabaja. Finalmente, para las personas que son otros miembros del hogar se incluyen, adicionalmente, variables relativas a la decisin laboral y el ingreso de la pareja, cuando sta existe. 4.5. Dotaciones i) Educacin Todas las personas de 15 o ms aos de edad son distribuidas en ocho grupos demogrficos, los cuales estn definidos por sexo, condicin de urbanidad y separados entre los ms jvenes (15 y 29 aos de edad) y el resto. En cada subgrupo se utiliza el mtodo no paramtrico indicado en (5), donde los valores observados en la distribucin de 1990 son reemplazados por los valores observados en la distribucin del 2003 corregido por c03/c90, donde c indica el centil de la distribucin del nivel de educacin, ordenado por los valores observados de la variable para cada ao. ii) Composicin etaria del hogar El total de hogares de cada perodo es subdividido en los mismos ocho grupos demogrficos descritos previamente, considerando las tres variables a nivel del jefe
14
Para el caso de las estimaciones relativas a las parejas del jefe del hogar se consider un anlisis conjunto de hombres y mujeres, puesto que el nmero de hombres definidos como pareja del jefe del hogar fue muy bajo. De esta forma, esta estimacin incluy una variable dummy para definir el sexo de la pareja.
310
del hogar. Dentro de cada grupo se analiza la distribucin de las cuatro variables asociadas a la composicin etaria del hogar, es decir, el nmero de miembros del hogar en edades: 0 a 8 aos, 9 a 14, 15 a 65 y 66 y ms. Nuevamente se realiza una imputacin a la distribucin observada en 1990 de acuerdo a los valores observados en la distribucin del 2003 corregida por la relacin de las medias observadas por cada distribucin y en cada grupo de edad. La sumatoria de los nuevos valores estimados de la composicin etaria de cada hogar para el perodo 1990, de acuerdo a la distribucin del ao 2003, permite determinar el tamao simulado de cada uno de los hogares observados en 1990. 4.6. Pensiones Para simular los cambios en la distribucin de pensiones se divide la poblacin de individuos en 24 subgrupos, segn gnero, tramos de edad (menor a 65, 65 a 74 y mayor que 74) y tramos de aos de escolaridad (0 a 3, 4 a 6, 7 a 11, 12 y ms). Los subgrupos resultantes difieren en trminos del porcentaje de su poblacin que accede a pago de pensiones, as como los montos promedios percibidos (Cuadro A-7 del Anexo). La pensin analizada corresponde a la suma de las pensiones pagadas por concepto de vejez, viudez, invalidez y orfandad, puesto que no se dispone para 1990 una desagregacin de estos componentes. Las pensiones de vejez y viudez representaban en el ao 2003 un 94% del agregado, de modo que la pensin bajo anlisis corresponde bsicamente a pagos que se realizan a adultos mayores. Hay una cada en la cobertura de las pensiones en prcticamente todos los subgrupos de poblacin. La cada es importante en magnitud, superando los diez puntos porcentuales en casi una tercera parte de los casos. Es probable que tal tendencia se origine en la reforma del sistema previsional, considerando que las cohortes de adultos mayores en el ao 2003 tienen mayor probabilidad de haber estado expuestos al sistema de capitalizacin individual, y que en tal sistema las personas pueden postergar la edad de jubilacin si la pensin que recibiran es muy baja o pueden quedar excluidos si no alcanzan un periodo mnimo de cotizaciones. Por su parte, el valor medio de la pensin se incrementa en 38,3% en trminos reales entre los aos 1990 y 2003. El incremento es superior a un 50% en el caso de las mujeres con baja escolaridad. Este efecto est relacionado con el reajuste en la pensin mnima, que represent un aumento real de 151% en el perodo 1990 a 2002 (Arellano, 2004). Para efectos de la simulacin posterior se requiere estimar las expresiones (6) y (7) anteriores, esto es, la probabilidad de acceso a pensiones a nivel de cada uno de los subgrupos de poblacin referidos. El procedimiento se realiza a travs de una estimacin probit, regresionando la variable dicotmica accede o no accede a pago de pensin en trminos de la edad, la condicin urbana y rural, el nmero de personas en el hogar y variables dummies regionales. Este mismo conjunto de variables se utiliza para predecir el ingreso de la pensin en la expresin (7). Los resultados de estas regresiones no se reportan por consideraciones de espacio, pero estn disponibles para los interesados.
311
4.7. Subsidios A diferencia de las dems partidas de ingresos, el tratamiento de los subsidios considera a los hogares como unidad de anlisis. Ello puesto que la condicin socioeconmica del hogar es el principal criterio en la seleccin de los beneficiarios de los subsidios monetarios. Este proceso se realiza a travs de un instrumento de focalizacin denominado Ficha Cas, que es un test relacionado de medios que entrega un puntaje segn el nivel de recursos econmicos que existen en el hogar. La racionalidad subyacente es que la familia (hogar) tiene la responsabilidad principal en el bienestar de sus miembros; la poltica de subsidios cumple un rol subsidiario puesto que apoya a los individuos cuando el hogar no tiene recursos suficientes. El tratamiento de los subsidios se realiza agregando los distintos pagos que se reciben en el hogar, habida cuenta de la baja participacin que tienen los subsidios en el ingreso monetario de los hogares. Al igual que en el caso de las pensiones, la simulacin incluye un componente no paramtrico basado en subgrupos de poblacin que difieren en la probabilidad de acceso a los subsidios y en el monto del beneficio recibido. En esta ocasin se trabaja con los hogares segn centiles de la distribucin de ingreso autnomo per cpita (ingreso monetario que excluye a los subsidios monetarios), de acuerdo a lo establecido en el prrafo anterior. Para cada uno de los subgrupos se computa el porcentaje de hogares que accede a subsidios durante los aos 1990 y 2003, as como el monto promedio del beneficio recibido. Los resultados de la estimacin muestran fuertes variaciones en la estructura de acceso a los subsidios a lo largo del perodo (Grfico A-1 del Anexo). En 1990 el patrn de participacin presentaba una pendiente moderada entre los centiles de ingreso per cpita del hogar, desde porcentajes en el entorno del 60% en los grupos ms pobres a tasas cercanas al 40% en los centiles superiores. En el ao 2003 la pendiente es bastante ms pronunciada, desde un 80% a un 15%, reflejando un patrn marcadamente focalizado de los subsidios. Ello resulta de polticas que focalizan subsidios que previamente tenan una estructura plana, como es el caso de la asignacin familiar por dependiente de los trabajadores asalariados y la introduccin de subsidios que benefician a la poblacin ms pobre, como es el caso de Chile Solidario y el subsidio de agua potable. Una tendencia similar presenta el monto promedio del beneficio por hogar (Grfico A-2 del Anexo). Considerando slo los hogares que reciben subsidios, el valor promedio del beneficio se incrementa en un 88% en trminos reales en el perodo entre 1990 y 2003. Los grupos ms pobres presentan los incrementos ms importantes, en el entorno de 160% en el decil inferior, mientras que los hogares del decil superior exhiben un incremento porcentual de alrededor de 55%. Estos resultados reflejan tanto la introduccin de nuevos subsidios orientados a los grupos pobres como un fuerte aumento en subsidios ya existentes focalizados en los pobres (pensin asistencial). 4.8. Efecto demogrfico El efecto demogrfico se analiza sobre la base de una clasificacin de los hogares en 21 subgrupos segn la escolaridad, edad y gnero del jefe, as como
312
del nmero de ncleos (Cuadros A-8 y A-9 en Anexo). Esta asociacin debe interpretarse como una forma reducida antes que un modelo estructural de la variable demogrfica. La escolaridad clasifica a los hogares en dos grupos segn si el jefe tiene entre 0 y 12 aos o ms de 13 aos de escolaridad; la edad clasifica a los hogares en cuatro tramos: 15-29, 30-44, 45-64 y 65 y ms. A ello se suman dos categoras segn sexo del jefe y nmero de ncleos (1, 2 y ms). Algunas de estas categoras se agregan cuando el tamao del subgrupo es muy reducido, resultando los 21 subgrupos referidos. La clasificacin utilizada explica un 37% de la varianza del tamao de los hogares. En trminos generales, el tamao del hogar es ms reducido en los hogares con jefe con alta educacin, edad joven o adulto mayor, sexo mujer y solo un ncleo por hogar. Entre 1990 y 2003 se producen cambios en la participacin relativa de los 21 subgrupos en el total de poblacin. En particular, se produce un incremento en los hogares con jefes con alta educacin, para los distintos tramos de edad y sexo del jefe, lo cual hace caer el tamao del hogar por efecto composicin. 5. Simulaciones de la Distribucin de Ingresos En esta seccin se presentan los resultados de las microsimulaciones de la distribucin del ingreso per cpita de los hogares. La pregunta bsica a responder es cul habra sido el cambio en la distribucin del ao base ante el cambio en cada uno de los componentes del ingreso de los hogares, desagregando en trminos de los respectivos efectos precio, dotaciones y participacin. Tambin se presenta el efecto conjunto de los cambios en los componentes individuales, as como un anlisis preliminar de tendencias en movilidad de ingresos. 5.1. Ingresos laborales La presentacin de los resultados se realiza para el conjunto de los ingresos del trabajo asalariado y cuenta propia. El Cuadro 4 muestra cul hubiese sido la distribucin del ingreso per cpita del ao 1990 si se sustituyen los ingresos del trabajo de hombres y mujeres por los correspondientes a la estructura del ao 2003. En el Cuadro 5 se presenta un ejercicio anlogo considerando slo el ingreso del trabajo de las mujeres, habida cuenta que se trata del grupo que experimenta los mayores cambios en su tasa de participacin laboral en el perodo analizado. El cambio en la estructura de los retornos a la educacin incluido en el efecto precio es el factor que induce mayor desigualdad en la distribucin del ingreso per cpita (Cuadro 4). Por s solo, habra inducido un incremento de 1,5 puntos en el Gini y ms de 4 puntos en el coeficiente de Theil; la razn de quintiles habra tambin aumentado en dos puntos, desde 17,6 a 19,6. Estos son efectos cuantitativos importantes y reflejan cambios en la demanda por trabajo calificado. Esta tendencia es compensada por la reduccin de la desigualdad asociada al comportamiento de los componentes no observables en las ecuaciones de salarios. Por su parte, los cambios en participacin laboral y en las dotaciones presionan en la direccin de una mayor desigualdad, pero su impacto no es cuantitativamente importante. Considerados en conjunto, los efectos citados se
313
compensan y resultan en una distribucin del ingreso per cpita que es prcticamente similar a la del ao inicial (ltima lnea del Cuadro 4). La simulacin del ingreso laboral de las mujeres resulta en incrementos pequeos en la desigualdad del ingreso per cpita, explicados en lo fundamental por los cambios en la estructura de los retornos o efecto precio. Esta vez no hay cambios compensatorios en el comportamiento de los no observables. El resultado ms interesante es el efecto de la participacin laboral, que resulta ser prcticamente neutral desde una perspectiva distributiva. Ello parecera contradictorio con los hechos estilizados del perodo que muestran un importante aumento en la tasa de participacin de las mujeres de escolaridad media y baja, situacin que tendra un efecto igualizador en la distribucin de los ingresos. CUADrO 4
SIMULACIN DEL CAMBIO EN EL INGrESO LABOrAL EN LA DISTrIBUCIN DEL INGrESO DEL HOGAr
Gini 1990 Efectiva 2003 Efectiva Efecto precio No observables Participacin laboral Dotacin de educacin Todos 56,0 56,0 57,5 54,9 56,0 55,9 55,2
Q5/Q1 90/10 17,9 17,6 19,9 16,8 18,5 18,6 18,5 10,9 10,4 11,7 10,5 11,4 11,6 11,5
90/50 3,9 3,6 3,9 3,8 3,9 3,9 3,7
10/50 0,35 0,35 0,33 0,36 0,34 0,33 0,33
75/25 3,3 3,3 3,4 3,3 3,4 3,5 3,5
Theil 66,8 69,7 71,1 63,2 65,8 64,9 62,5
CUADrO 5
SIMULACIN DEL CAMBIO EN INGrESO LABOrAL DE MUJErES EN DISTrIBUCIN INGrESO HOGAr
Gini 1990 Efectiva 2003 Efectiva Efecto precio No observables Participacin laboral Dotacin de educacin Todos 56,0 56,0 56,4 56,0 55,8 56,1 55,7
Q5/Q1 90/10 17,9 17,6 18,7 18,0 17,8 18,4 18,4 10,9 10,4 11,4 11,0 10,9 11,3 11,5
90/50 3,9 3,6 3,9 3,9 3,8 3,9 3,9
10/50 0,35 0,35 0,34 0,35 0,35 0,34 0,33
75/25 3,3 3,3 3,4 3,3 3,3 3,4 3,5
Theil 66,8 69,7 66,9 66,5 65,9 66,1 64,2
El Grfico A-3 en el Anexo muestra que la tasa de ocupacin de las mujeres crece en forma pareja a lo largo de la distribucin de los ingresos, lo cual es consistente con su citada neutralidad en el plano distributivo. La interpretacin de tales resultados debe distinguir entre movilidad y desigualdad. El hecho
314
de que una fraccin de las mujeres de escolaridad media y baja se adscriba al mercado del trabajo es consistente con una mayor movilidad de ingresos, pero no necesariamente con mayor igualdad de ingresos. Esto ltimo puesto que la tendencia a la mayor igualdad originada en la elevacin de los ingresos de hogares de quienes se insertan en el mercado del trabajo puede ser compensada por la mayor distancia que se produce entre ese grupo y quienes no modifican su situacin ocupacional. 5.2. Pensiones La simulacin de la distribucin de las pensiones del ao 1990 de acuerdo a los parmetros del ao 2003 se realiza bajo tres escenarios. En primer trmino, se simula la distribucin de 1990 con la estructura de montos pagados en el ao 2003, suponiendo constante la participacin o acceso de la poblacin al beneficio. Segundo, se simula la distribucin de 1990 con la estructura de participacin del ao 2003, manteniendo constantes los montos pagados en el primer ao. Tercero, se simula la distribucin de 1990 con la estructura de pagos y acceso del 2003. En todos los casos se aplica la metodologa antes descrita, trabajando a nivel de los 21 subgrupos de poblacin especificados. Los resultados de la distribucin de las pensiones individuales se presentan en el Cuadro A-10 del Anexo. La distribucin efectiva experimenta cambios importantes en el perodo, creciendo por debajo del resto de los ingresos y hacindose ms igualitaria. La simulacin del cambio en el monto de las pensiones (efecto precio) replica cercanamente el cambio en la distribucin efectiva de las pensiones; en cambio, la simulacin del cambio en el acceso (efecto participacin) presenta efectos slo marginales en el cambio de las pensiones. Los cambios en la distribucin de las pensiones se ilustran en el Grfico A-4, que presenta el crecimiento efectivo y aquel atribuible a los efectos precio y participacin. En el Cuadro 6 se presentan los efectos de los cambios en pensiones sobre la distribucin del ingreso per cpita de los hogares. El efecto simulado de los efectos precio y participacin induce cambios marginales en la distribucin del ingreso per cpita del hogar. As, mientras el ingreso promedio crece en un 56,2% en el perodo analizado, la modificacin que experimenta la partida de pensiones hace crecer el ingreso promedio en solo 1,2% (M$ 92,1 vs M$ 91,0). Por su parte, el cambio en las pensiones causa un incremento en la desigualdad del ingreso per cpita, aun cuando la magnitud del efecto sea marginal. A modo de ilustracin, la razn de quintiles habra crecido de 17,9 en 1990 a 18,6 en 2003, en caso de que el cambio en las pensiones hubiese sido la nica partida de ingresos que se modificara en el perodo. La tendencia sealada tiene su origen en el efecto participacin, que hace caer el ingreso promedio e incrementa los indicadores de desigualdad; en cambio, el efecto precio opera en direccin contraria, aumentando el ingreso per cpita y reduciendo la desigualdad. Una menor cobertura de las pensiones en el ao 2003 respecto del ao base explica la cada en el ingreso per cpita que se deriva de tal efecto. El incremento en desigualdad se debera a que una menor cobertura de las pensiones aumenta la distancia de ingresos entre hogares, puesto que disminuye uno de los componentes del ingreso total.
315
CUADrO 6
INGrESO PEr CPITA DEL HOGAr ($ 2003), SIMULACIN EFECTO PENSIONES
Media 1990 efectiva 2003 efectiva Simulacin Precio 03 Participacin 03 Ambos 03 93,1 90,0 92,1 91,0 142,1
Gini 56,0 56,0 55,7 56,5 56,2
Q5/Q1 17,9 17,6 17,7 18,8 18,6
90/10 11,1 10,6 10,8 11,6 11,4
90/50 3,9 3,6 3,8 3,9 3,9
10/50 0,35 0,34 0,35 0,33 0,34
75/25 3,3 3,3 3,3 3,4 3,4
Theil 66,8 69,7 65,7 67,9 66,8
5.3. Subsidios La poltica de subsidios monetarios tiene distinto tipo de efectos sobre la distribucin del ingreso per cpita de los hogares (Cuadro 7). Por una parte, su impacto en el valor medio del ingreso del hogar es bastante marginal, haciendo crecer tal monto desde el valor inicial de M$ 91,0 en 1990 a $ 93,1 en el ao 2003 (2,3%). Casi la totalidad del impacto se debe al efecto precio, mientras que el efecto participacin es casi nulo. En cualquier caso se trata de una contribucin que representa menos de la vigsima parte del incremento efectivo que experimenta el ingreso per cpita en el perodo, y que se explica por la baja participacin que tienen los subsidios en el ingreso de los hogares. No obstante lo anterior, la poltica de subsidios monetarios tiene un impacto ms marcado en materia de los indicadores de desigualdad. Por s solo, el cambio en los subsidios monetarios habra causado una cada de dos puntos porcentuales en el coeficiente de Gini (de 56,0 a 54,0), y una reduccin an mayor en la razn de quintiles 5/1, desde un 17,9 a un 14,4. Nuevamente, el peso del impacto recae sobre el efecto precio, de modo que son los incrementos en el monto de los subsidios focalizados en los pobres antes que los incrementos en participacin en los subsidios los que causan la cada en la desigualdad. Sin embargo, la desigualdad efectiva en el ao 2003 es similar a la del ao de inicio, de manera que la tendencia a la reduccin de la desigualdad es compensada por la tendencia en sentido contrario de otros componentes de la distribucin de los ingresos. CUADrO 7
INGrESO PEr CPITA DE LOS HOGArES, SIMULACIN EFECTO DE LOS SUBSIDIOS, 1990 y 2003
Media 1990 efectiva 2003 efectiva Simulacin subsidios Precio 03 Participacin 03 Ambos 03 92,7 91,2 93,1 91,0 142,1
Gini 56,0 56,0 54,7 55,7 54,0
Q5/Q1 90/10 17,9 17,6 15,6 17,2 14,4 11,1 10,6 9,7 10,8 9,1
90/50 3,9 3,6 3,7 3,8 3,7
10/50 0,35 0,34 0,38 0,35 0,40
75/25 3,3 3,3 3,1 3,3 3,0
Theil 66,8 69,7 64,1 66,2 62,8
316
5.4. Efecto demogrfico La simulacin de los efectos demogrficos en la distribucin de los ingresos se presenta en el Cuadro 8. El efecto demogrfico corresponde a los cambios en el tamao relativo de los subgrupos poblacionales y los cambios en el tamao de los hogares. Los resultados muestran que el efecto demogrfico induce un significativo crecimiento en el ingreso medio de los hogares; todo lo dems constante, la reduccin en el tamao de los hogares hace aumentar el ingreso per cpita en un 14%15. Este es un cambio relativamente parejo a lo largo de la distribucin de los ingresos, el cual no produce modificaciones en los indicadores de desigualdad del ingreso per cpita de los hogares. La distribucin del ingreso simulada con la estructura demogrfica del ao 2003 es prcticamente igual a la distribucin de 1990, con la citada excepcin del cambio en el nivel promedio de los ingresos. En este caso se est en presencia de factores que operan en direcciones opuestas y que se compensaran entre s: la cada en la tasa de natalidad es ms importante en las mujeres de baja condicin socioeconmica; la creacin de nuevos hogares vinculada a la construccin de viviendas sociales, y la creacin de nuevos ncleos secundarios asociados a un incremento en las madres solteras en estratos medios y bajos. CUADrO 8
INGrESO PEr CPITA DE LOS HOGArES, SIMULACIN DEL EFECTO DEMOGrFICO, 1990 y 2003
Media 1990 efectiva 2003 efectiva Simulacin Tamao hogar Participacin Efecto demogrfico 103,7 93,4 104,8 91,0 142,1
Gini 56,0 56,0 56,1 55,9 56,4
Q5/Q1 17,9 17,6 18,0 17,9 17,9
90/10 11,1 10,6 10,8 11,5 11,3
90/50 3,9 3,6 3,8 4,0 4,0
10/50 0,35 0,34 0,35 0,35 0,35
75/25 3,3 3,3 3,3 3,4 3,5
Theil 66,8 69,7 67,2 65,7 18,2
5.5. Otros ingresos En el caso del ingreso de los empleadores y del resto de los ingresos se realiza una simulacin de carcter contable, modificando el ingreso del ao inicial por el crecimiento acumulado en el perodo 1990-2003 a nivel de cada centil del ingreso per cpita. Este procedimiento tiene por objetivo contar con series que puedan ser homologadas a las simulaciones de las dems partidas de ingresos, pero no hay elementos endgenos aqu considerados.
15
Esta cifra es algo mayor respecto de la entregada por la descomposicin contable (Cuadro 3).
317
Los resultados de la simulacin se presentan en el Cuadro 9. Los cambios en el ingreso de los empleadores habran ejercido una considerable presin al alza en la desigualdad, con aumento de 4 puntos en el coeficiente de Gini, 3,5 puntos en la razn de quintiles y casi 18 puntos en el coeficiente de Theil. Por su parte, el resto de los ingresos incluye un conjunto heterogneo de elementos, incluyendo rentas del capital, donaciones, autoconsumo y otros. El efecto del conjunto de estos ingresos es prcticamente neutral en la distribucin del ingreso de los hogares, debido a la probable compensacin de efectos en distintas direcciones. CUADrO 9
INGrESO PEr CPITA DE LOS HOGArES, SIMULACIN OTrOS INGrESOS, 1990 y 2003
Media 1990 efectiva 2003 efectiva Simulacin Empleadores resto ingresos 101,9 94,2 91,0 142,1
Gini 56,0 56,0 60,2 56,1
Q5/Q1 17,9 17,6 21,4 17,6
90/10 11,1 10,6 11,8 10,8
90/50 3,9 3,6 4,1 3,7
10/50 0,35 0,34 0,34 0,35
75/25 3,3 3,3 3,4 3,4
Theil 66,8 69,7 84,4 68,0
6. A Modo de Conclusin La estabilidad que presenta la desigualdad de ingreso en la economa chilena en el perodo 1990-2003 representa un hecho sorprendente considerando los grandes cambios que experimentan los factores determinantes de los ingresos. En el perodo citado el empleo aument en un 33%, los salarios reales en un 51%, la participacin de la mujer en 11 puntos y el gasto social en 145%. La poblacin aumenta su nivel de escolaridad a la vez que se envejece, producto de la cada en la tasa de natalidad; el tamao promedio de los hogares se reduce en 8% en el perodo. La microsimulacin de la distribucin de los ingresos permite identificar los factores que subyacen detrs de la inercia que presentan los indicadores de desigualdad. Hay factores que tienen un impacto parejo a nivel de los ingresos de los distintos hogares, provocando cambios que son distributivamente neutros. Entre estos destacan la insercin laboral de la mujer y la reduccin del tamao de los hogares. En cierto sentido se trata de un resultado sorprendente, puesto que los antecedentes previos hacan suponer que se trataba de factores que reduciran la desigualdad en la medida que su efecto se concentrara en hogares situados en la parte media y baja de la distribucin. Los resultados obtenidos apuntan a nuevas hiptesis. As, la reduccin en el tamao del hogar de bajos ingresos producto de la cada en la tasa de natalidad y la expansin de la vivienda social habra sido compensada por un aumento de los ncleos secundarios asociado a los nacimientos fuera del matrimonio. La mayor participacin laboral de la
318
mujer de sectores medios y bajos reduce la brecha de ingresos respecto de los sectores ms acomodados, al tiempo que induce desigualdad respecto de quienes no se insertan en el mercado del trabajo. Por otra parte, hay factores que presionan hacia una mayor desigualdad de los ingresos. Entre estos destacan los cambios en los retornos de la educacin y el incremento en el ingreso de los empleadores, que por s solos habran aumentado todos los indicadores de desigualdad considerados en el anlisis. Estos factores estn cercanamente relacionados con el funcionamiento de los mercados en el contexto de una economa en fuerte expansin y como tal pone una nota de alerta respecto del rol que a la fecha han tenido tales determinantes en la distribucin de los ingresos. La fuerte reduccin de la desigualdad que experimentaron los pases hoy desarrollados en la primera mitad del siglo XX tuvo como causa principal la reduccin de la brecha laboral entre trabajadores calificados y no calificados. Tal desarrollo est an pendiente en el caso de la economa chilena. En el perodo se produce un importante aumento de los subsidios monetarios dirigidos a los grupos ms pobres. A pesar de que se trata de transferencias que representan una fraccin marginal del gasto social y del ingreso de los hogares, la focalizacin del gasto en los hogares pobres representa la principal fuerza que presiona a la baja la desigualdad de ingresos durante el perodo. Por s solo habra reducido en 2 puntos el coeficiente de Gini y en 3,5 puntos la razn de quintiles. El rol que juegan los subsidios monetarios contrasta con otros factores relacionados con la poltica pblica. Este es el caso de las pensiones y de la educacin, que no tienen impacto en la desigualdad en el perodo a pesar de que se trata de determinantes principales de los ingresos de los hogares. Las pensiones siguen siendo pagadas mayoritariamente por el Estado, no obstante la reforma que privatiza el sistema de pensiones a inicios de los 80. Tal situacin se explica porque la mayor parte de los pensionados pertenecen al antiguo sistema de reparto y por los aportes pblicos a los nuevos pensionados (bonos de reconocimiento y pensin mnima). Sin embargo, el antiguo sistema de pensiones se basa en un patrn contributivo, de modo que la distribucin de las pensiones sigue cercanamente la distribucin de los salarios, no conteniendo elementos redistributivos de importancia. Por su parte, el aumento de la escolaridad tiene efectos ambiguos sobre la distribucin de los ingresos, empujando al alza a algunos indicadores de desigualdad y a la baja a otros. Los incrementos en cobertura educacional debieran ir beneficiando cada vez ms a los grupos de bajos ingresos, dado que se trata de una variable acotada en su nivel superior, pero se trata de un proceso que es lento en el tiempo, dado el largo ciclo de estudios y el peso de las cohortes antiguas en el mercado del trabajo. Ello sin considerar los aspectos de calidad de la educacin, que se reflejan en la estructura de los retornos. En suma, la inercia que presenta la desigualdad refleja la accin de factores que se compensan entre s, otros que tienen ritmos lentos en el tiempo y la aparicin de nuevos desarrollos que afectan la distribucin. Progresos en la materia pueden requerir una accin ms asertiva de la poltica pblica, modificando su actual posicin pasiva en el mbito distributivo. Un importante paso en tal direccin es el establecimiento de una pensin bsica solidaria recientemente aprobada en el Congreso, que se convertir a futuro en la principal transferen-
319
cia monetaria destinada a grupos de bajos ingresos en el pas. La cobertura y los montos involucrados en esta iniciativa conforman una nueva posicin de la poltica social del pas, ms acorde con el nivel de desarrollo que ha alcanzado la economa chilena en el perodo reciente. Referencias Bibliogrficas Arellano, Jos Pablo (1985). Polticas Sociales y Desarrollo. Chile 1924-1984, Ediciones Cieplan, Santiago. Bravo, D., D. Contreras y S. Urza (2002). Poverty and Inequality in Chile 1990-1996: Learning from Microeconomic Simulations, Documento de Trabajo 196, Departamento de Economa, Universidad de Chile. Bourguignon, F., F. Ferreira and P. Leite (2002). Beyond Oaxaca-Blinder: Accounting for Differences in Household Income Distributions across Countries. PPR Working Paper 2828, The World Bank, Washington D.C. Bourguignon, F., F. Ferreira and N. Lustig (2005). Decomposing changes in the distribution of household income. Methodological Aspects, en Bourguignon, F., F. Ferreira and N. Lustig. (eds.), The Microeconomics of Income Distribution Dynamics in East Asia and Latin America, Washington: World Bank and Oxford University Press. De Ferranti, D., G. Perry, F. Ferreira and M. Walton (2003). Inequality in Latin America and the Caribbean: Breaking with History?, World Bank. Deaton, A. (1997). The Analysis of Household Surveys. A Microeconometric Approach to Development Policy, cap. 1 (The design and content of household surveys), Johns Hopkins. Fields, G. and S. Soares (2005). The Microeconomic of Changing Income Distribution in Malaysia, en Bourguignon, F., F. Ferreira and N. Lustig (eds.), op. cit. Juhn, C., K. Murphy and B. Pierce (1993). Wage Inequality and the rise in return to Skill, Journal of Political Economy 101. Larraaga, O. (2006 a). Participacin laboral de la mujer. Chile 1958-2003, en J. Samuel Valenzuela, Eugenio Tironi y Timothy Scully (eds.), El Eslabn Perdido. Familia, modernizacin y bienestar en Chile. Taurus, Santiago. Larraaga, O. (2006 b). Comportamientos reproductivos y fertilidad, 1960-2003, en J. Samuel Valenzuela, Eugenio Tironi y Timothy Scully, op. cit. Larraaga, O. (2006 c). Qu puede esperarse de la poltica social en Chile. Documento de Investigacin, Departamento de Economa, Universidad de Chile. raczynski, Dagmar y Claudia Serrano (2005). Las polticas y estrategias de desarrollo social: aportes de los aos 90 y desafos futuros, en P. Meller (ed). La Paradoja Aparente. Equidad y Eficiencia: resolviendo el dilema. Editorial Taurus, Santiago.
320
ANEXO CUADrO A-1

DETErMINANTES PrXIMOS DEL INGrESO PEr CPITA HOGArES
Indicadores Poblacin Total (en millones) % rural Distribucin % de poblacin segn edades 0-14 15-30 31-50 51-65 66 y ms ndices de empleo y salarios Empleo Salarios reales Distribucin % ocupados nivel educacional Primaria o ninguna Secundaria, incompleta Secundaria, completa Estudios superior no universitario Estudios universitarios Participacin Laboral (%) Mujeres Hombres Tasa de Desempleo % Mujeres Hombres Tipo de Empleo % Dependientes Cuenta Propia+Empleador Tamao Promedio del Hogar
Nota:
1990 12,93 18,5 28,5 31,2 24,0 10,5 5,8 100,0 100,0 38,2 16,0 24,0 7,0 14,8 35,0 77,9 9,7 7,7 75,6 24,4 4,08
2003 15,55 13,4 25,5 26,7 28,4 12,1 7,3 133,0 151,0 24,5 15,9 31,9 10,1 17,7 46,3 77,5 12,4 8,4 76,9 23,1 3,78
Todas las estadsticas provienen de las encuestas Casen de los aos 1990 y 2003, con excepcin de los datos de poblacin (INE).
321
CUADrO A-2
TASAS DE CrECIMIENTO SALArIOS rEALES 1990-2003, OCUPADOS 25-40 AOS
Nivel educacional Primaria o ninguna Secundaria, incompleta Secundaria, completa CFT, IP incompleta CFT, IP completa Universitario incompleto Universitario completo Total
Hombres 14,8 24,5 17,7 17,5 20,4 31,7 47,9 42,0
Mujeres 27,6 27,0 29,5 36,8 37,4 8,8 86,7 53,0
Total 16,4 23,1 19,7 21,8 30,5 15,9 57,9 43,3
CUADrO A-3
FACTOrES DEMOGrFICOS
Ao 1990 2003
Tamao promedio hogar 4,08 3,78
Promedio de ncleos por hogares 1,19 1,20
Promedio hijos menores de 20 por ncleo 1,18 0,98
Promedio hijos < 20, en ncleos con hijos 1,96 1,77
Fuente: Encuestas Casen, aos respectivos.
CUADrO A-4
322
ESTIMACIN DE ECUACIN DE INGrESOS DEL TrABAJO

Mujer Cuenta Propia 1990 2003 1990 2003 1990 Hombre Dependiente Hombre Cuenta Propia 2003
Mujer Dependiente 2003
1990
Primaria Incomp.
Primaria Completa
Secundaria Incomp.
Secundaria Humanista Comp.
Secundaria TP Comp.
TP/IP Incomp.
TP/IP Completa
Universidad Incomp.
Universidad Completa
Exp
Exp2
rural
Constante 0,580 0,481 0,908 0,254
0,14 (0,090) 0,28*** (0,094) 0,42*** (0,093) 0,72*** (0,093) 0,82*** (0,096) 0,95*** (0,112) 10,17*** (0,098) 10,27*** (0,096) 10,56*** (0,096) 0,030*** (0,002) 0,0005*** (0,0000) 0,01 (0,025) 10,60*** (0,097) 0,861 0,336 0,615 0,402 0,566 0,485 0,832 0,254
0,07 (0,063) 0,20*** (0,066) 0,30*** (0,066) 0,60*** (0,066) 0,69*** (0,067) 0,71*** (0,089) 10,08*** (0,068) 10,24*** (0,071) 10,73*** (0,068) 0,023*** (0,002) 0,0003*** (0,00004) 0,09*** (0,015) 10,73*** (0,088)
0,29** (0,141) 0,28* (0,168) 0,65*** (0,156) 0,80*** (0,162) 0,88*** (0,187) 10,29*** (0,268) 10,30*** (0,210) 10,50*** (0,197) 10,96*** (0,211) 0,020 (0,008) 0,0002 (0,0001) 0,21*** (0,072) 110,05*** (0,203)
0,40** (0,166) 0,64*** (0,176) 0,67*** (0,177) 10,04*** (0,177) 0,89*** (0,188) 10,29*** (0,302) 10,28*** (0,201) 10,49*** (0,208) 20,29*** (0,195) 0,024*** (0,006) 0,0002** (0,0001) 0,13*** (0,044) 10,78*** (0,215)
0,19*** (0,036) 0,31*** (0,040) 0,46*** (0,038) 0,72*** (0,041) 0,81*** (0,044) 0,93*** (0,088) 10,17*** (0,054) 10,26*** (0,054) 10,85*** (0,050) 0,040*** (0,002) 0,0006*** (0,0000) 0,19*** (0,013) 10,82*** (0,050)
0,21*** (0,040) 0,37*** (0,042) 0,49*** (0,043) 0,70*** (0,043) 0,81*** (0,045) 0,90*** (0,071) 10,22*** (0,047) 10,30*** (0,054***) 10,95*** (0,048) 0,042*** (0,001) 0,0006*** (0,00003) 0,16*** (0,011) 10,80*** (0,060)
0,08 (0,076) 0,17* (0,086) 0,37*** (0,083) 0,65*** (0,088) 0,72*** (0,103) 10,04*** (0,146) 0,91*** (0,121) 10,03*** (0,146) 10,72*** (0,116) 0,043*** (0,004) 0,0006*** (0,0001) 0,29*** (0,032) 110,34*** (0,109)
0,20*** (0,074) 0,47*** (0,077) 0,55*** (0,079) 0,80*** (0,080) 0,84*** (0,084) 0,90*** (0,344) 10,24*** (0,102) 10,21*** (0,102) 20,02*** (0,106) 0,038*** (0,003) 0,0005*** (0,0001) 0,24*** (0,023) 110,26*** (0,102) 0,724 0,377
Desviacin Estndar de residuos r2
0,577 0,423
Todas las estimaciones incluyen dummies para regiones y para trabajadores de tiempo parcial. ***, **, * representa que el coeficiente es estadsticamente significativo a < 0,01, < 0,05 y < 0,1, respectivamente.
CUADrO A-5
ESTIMACIN MULTILOGIT SOBrE DECISIONES DE PArTICIPACIN LABOrAL, 1990 y 2003

Mujer Otro Parentesco Dependiente 1990 10,63*** 10,97*** 10,81*** 30,03*** 30,42*** 20,91*** 30,79*** 30,10*** 40,66*** 0,243*** 0,005*** 0,42*** 0,87*** 0,98*** 0,03*** 0,04*** 0,03*** 0,02*** 0,23*** 0,06*** 40,45*** 0,184 30,82*** 0,189 60,77*** 0,184 60,70*** 0,189 0,42*** 0,31*** 0,32*** 0,06*** 0,02*** 0,12*** 0,04*** 20,58*** 20,58*** 0,119 30,45*** 30,10*** 30,56*** 20,79*** 30,89*** 0,206*** 0,004*** 0,15*** 0,86*** 0,49*** 0,29*** 0,04*** 0,07*** 0,03*** 0,30*** 0,17*** 20,99*** 30,31*** 30,63*** 20,45*** 40,20*** 0,216*** 0,004*** 0,56*** 0,79*** 0,13*** 0,53*** 0,05*** 0,12*** 0,10*** 0,19*** 0,34*** 20,44*** 10,78*** 20,75*** 20,48*** 20,87*** 0,19*** 0,003*** 0,10*** 0,47*** 0,09*** 0,42*** 0,16*** 0,04*** 0,01*** 0,13*** 0,21*** 0,82*** 10,20*** 10,69*** 10,95*** 30,13*** 0,43*** 0,001*** 0,59*** 10,05*** 0,71*** 10,41*** 10,56*** 20,45*** 0,035*** 0,001*** 0,54*** 10,53*** 10,80*** 10,71*** 20,91*** 20,41*** 20,31*** 20,16*** 20,94*** 10,14*** 10,38*** 10,39*** 20,09*** 0,15*** 0,09*** 0,00 0,49*** 0,32*** 0,38*** 0,33*** 0,52*** 2003 1990 2003 1990 2003 1990 0,08*** 0,09*** 0,35*** 0,55*** 0,42*** 10,46*** 10,72*** 10,23*** 20,10*** 0,107*** 0,001*** 0,81*** Cuenta Propia Dependiente Mujer y Hombre Pareja Cuenta Propia 2003 0,09*** 0,27*** 0,33*** 0,48*** 0,49*** 0,08** 0,86*** 0,76*** 10,43*** 0,061*** 0,001*** 0,42***
Mujer Jefa de Hogar Cuenta Propia 2003 0,21*** 0,14*** 0,22*** 0,08*** 0,61*** 10,48*** 0,20*** 0,02 0,24*** 0,138*** 0,002*** 0,17*** 0,16*** 0,20*** 0,68*** 0,11*** 0,07*** 0,08*** 0,06*** 10,79*** 0,132
Dependiente
1990
2003
1990
0,17*** 0,04* 0,47*** 0,23***
0,25*** 0,42*** 0,35*** 0,35***
0,31*** 0,36*** 0,25*** 0,50***
Primaria Incomp. Primaria Completa Secundaria Incomp. Secundaria Humanista Comp. Secundaria TP Comp. TP/IP Incomp. TP/IP Completa Universidad Incomp. Universidad Completa Exp Exp 2 rural Casado Convive Joven N de nios N de jvenes N de adultos N de mayores Trabaja el Jefe? Mujer Constante 0,32*** 0,39*** 0,35*** 0,14*** 0,02*** 0,11** 0,19*** 20,23*** 0,60*** 0,106
0,10*** 0,02 0,06** 0,04 0,81*** 0,034*** 0,002*** 0,52*** 0,90*** 0,46*** 0,67*** 0,23*** 0,05*** 0,09*** 0,18***
0,18*** 0,46*** 0,00 0,32*** 0,91*** 0,094*** 0,002*** 0,70*** 0,32*** 0,19*** 0,80*** 0,29*** 0,20*** 0,07*** 0,03***
0,13*** 10,12*** 10,22*** 0,51*** 0,00 0,057*** 0,001*** 0,08*** 0,50*** 0,14*** 10,22*** 0,07*** 0,05*** 0,11*** 0,15***
20,12***
10,16***
0,36***
0,34*** 0,68*** 0,10*** 0,12*** 0,03*** 0,02 0,25*** 20,79*** 10,38*** 0,119
0,34*** 0,54*** 0,08*** 0,02*** 0,06*** 0,09*** 0,29*** 20,20*** 0,48*** 0,106
Pseudo-r2
0,190
0,132
0,190
323
En todas las estimaciones se incluyeron dummies regionales y, dependiendo del rol en el hogar, variables sobre ingreso y participacin laboral de otros miembros del hogar, as como caractersticas educacionales y etarias de otros adultos. ***, **, * representa que el coeficiente es estadsticamente significativo a < 0,01, < 0,05 y < 0,1, respectivamente.
CUADrO A-6
324
ESTIMACIN MULTILOGIT SOBrE DECISIONES DE PArTICIPACIN LABOrAL, 1990 y 2003

Hombre Otro Parentesco Cuenta Propia 1990 0,38*** 0,71*** 0,88*** 10,02*** 10,27*** 10,37*** 10,21*** 10,66*** 10,20*** 0,03*** 0,001*** 0,57*** 0,35*** 0,34*** 10,32*** 10,45*** 0,090 0,073 0,53*** 0,45*** 10,83*** 10,15*** 0,65*** 0,72*** 0,11*** 0,002*** 0,47*** 0,29*** 0,60*** 10,12*** 0,09*** 0,11*** 0,38*** 0,66*** 10,75*** 20,08*** 20,04*** 20,91*** 20,05*** 20,47*** 20,27*** 30,37*** 10,86*** 20,24*** 10,91*** 20,66*** 2003 1990 2003 1990 Dependiente Cuenta Propia 2003 10,72*** 20,18*** 20,02*** 20,76***
Hombre Jefe de Hogar
Dependiente
1990
2003
0,62*** 0,87*** 10,07*** 0,95***
0,23*** 0,23*** 0,52*** 0,74***
Primaria Incomp. Primaria Completa Secundaria Incomp. Secundaria Humanista Comp. Secundaria TP Comp. TP/IP Incomp. TP/IP Completa Universidad Incomp. Universidad Completa Exp Exp2 rural Casado Convive Joven Trabaja el Jefe? Constante 30,15*** 20,37*** 30,38*** 20,26*** 30,67*** 0,304*** 0,006*** 0,87*** 10,28*** 0,86*** 0,28*** 0,33*** 40,17*** 0,205 0,220 30,87*** 30,11*** 30,67*** 20,95*** 30,94*** 0,28*** 0,005*** 0,51*** 0,70*** 10,09*** 0,19*** 0,50*** 40,17***
0,96*** 10,12*** 10,02*** 10,30*** 0,55*** 0,003*** 0,001*** 0,59*** 0,63*** 0,38*** 10,16***
0,52*** 10,09*** 0,88*** 0,62*** 0,05** 0,06*** 0,002*** 0,24*** 0,47*** 0,63*** 0,90***
30,61***
20,37***
20,92*** 10,88*** 20,89*** 20,09*** 30,18*** 0,29*** 0,005*** 0,86*** 10,25*** 10,36*** 0,37*** 0,37*** 50,71*** 0,205
20,86*** 20,07*** 20,91*** 20,51*** 30,00*** 0,29*** 0,004*** 0,73*** 0,55*** 10,32*** 0,22*** 0,26*** 50,60*** 0,220
Pseudo-r2
0,090
0,073
En todas las estimaciones se incluyeron dummies regionales, variables sobre ingreso y participacin laboral de otros miembros del hogar, as como caractersticas educacionales y etarias de otros adultos. ***, **, * representa que el coeficiente es estadsticamente significativo a < 0,01, < 0,05 y < 0,1, respectivamente.
325
CUADrO A-7
ESTrUCTUrA DE PENSIONES, SEGN SUBGrUPOS DE POBLACIN
Subgrupo Hombres < 65, 0-3 esc 65-74, 0-3 esc > 74, 0-3 esc < 65, 4-6 esc 65-74, 4-6 esc > 74, 4-6 esc < 65, 7-11 esc 65-74, 7-11 esc > 74, 7-11 esc < 65, >11 esc 65-74, > 11 esc > 74, >11 esc Mujeres < 65, 0-3 esc 65-74, 0-3 esc > 74, 0-3 esc < 65, 4-6 esc 65-74, 4-6 esc > 74, 4-6 esc < 65, 7-11 esc 65-74, 7-11 esc > 74, 7-11 esc < 65, >11 esc 65-74, >11 esc > 74, >11 esc Total
% con pensin 1990 2003 Cambio
Pensin promedio (M$ 2003) 1990 2003 Cambio %
10,5 76,3 94,1 8,2 75,6 93,4 3,4 74,3 88,1 3,8 73,8 86,9
8,8 61,5 69,8 8,2 70,8 80 3,7 72,6 79 3,1 73,3 75,6
1,7 14,8 24,3 0,0 4,8 13,4 0,3 1,7 9,1 0,7 0,5 11,3
77,6 73,1 69,7 105,6 98,3 97,1 157,6 149,2 129,6 211,2 246,5 255,2
88,6 95,8 94,4 130,3 119,9 129,2 164,5 180,2 166,3 271,9 289,5 253,2
14,3 31,0 35,5 23,3 22,0 33,0 4,4 20,8 28,3 28,7 17,5 0,8
14,9 64 79,3 8,1 62,3 77,6 3,4 56,6 72 3,4 60,5 74,1 11,5
11,7 44,7 64,1 8,7 47,4 68,6 3,3 50,9 68,3 2,9 59,4 66,1 10,5
3,2 19,3 15,2 0,6 14,9 9,0 0,1 5,7 3,7 0,5 1,1 8,0 1,0
50,3 54,7 57,7 64,7 67,3 66,5 74,7 86,1 89,0 137,2 155,7 143,6 103,4
79,6 83,5 88,8 84,5 92,0 101,9 105,1 116,5 132,5 176,3 210,4 183,0 143,5
58,2 52,5 53,9 30,6 36,6 53,3 40,6 35,4 48,8 28,5 35,1 27,5 38,8
Fuente: En base a encuestas Casen 1990 y 2003.
326
CUADrO A-8
GrUPOS DEMOGrFICOS
Participacin % en total 1990 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 7,67 16,15 25,43 4,88 1,42 4,41 5,41 0,55 0,47 1,38 4,26 1,92 0,34 0,63 0,93 1,53 14,42 6,35 0,24 0,29 1,31 2003 7,61 14,58 22,68 3,82 2,54 6,5 6,82 0,68 0,55 1,52 3,87 1,63 0,83 0,98 1,07 1,69 13,2 5,62 0,76 0,6 2,43 Diferencia 0,06 1,57 2,75 1,06 1,12 2,09 1,41 0,13 0,08 0,14 0,39 0,29 0,49 0,35 0,14 0,16 1,22 0,73 0,52 0,31 1,12
Tamao promedio hogar 1990 3,43 4,32 4,16 2,86 2,68 4,00 4,15 2,36 2,99 3,32 2,85 1,97 2,68 2,62 5,56 6,39 6,26 5,42 3,93 4,62 5,46 4,06 2003 3,28 4,01 3,77 2,53 2,4 3,59 4 2,18 2,79 3,35 2,7 1,91 2,56 2,79 5,78 5,97 5,92 5,37 3,37 4,83 5,36 3,79 Diferencia 0,15 0,31 0,39 0,33 0,28 0,41 0,15 0,18 0,20 0,03 0,15 0,06 0,12 0,17 0,22 0,42 0,34 0,05 0,56 0,21 0,10 0,27
327
CUADrO A-9
SUBGrUPOS DE POBLACIN SEGN CArACTErSTICAS JEFE y NMErO DE NCLEOS
Edad 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 15-29 30-45 46-64 65 y ms 15-29 30-45 46-64 65 y ms 15-29 30-45 46-64 65 y ms 30-45 46-64 15-29 30-45 46-64 65 y ms 15-29 30-45 46-64
Escolaridad 0-12 0-12 0-12 0-12 13 y ms 13 y ms 13 y ms 13 y ms 0-12 0-12 0-12 0-12 13 y ms 13 y ms 0-12 0-12 0-12 todos 13 y ms 13 y ms 13 y ms
Sexo Hombre Hombre Hombre Hombre Ambos Hombre Hombre Ambos Mujer Mujer Mujer Mujer Mujer Mujer Ambos Ambos Ambos Ambos Ambos Ambos Ambos
Nmero ncleos 1 1 1 1 1 1 1 1 1 1 1 1 1 1 2 y ms 2 y ms 2 y ms 2 y ms 2 y ms 2 y ms 2 y ms
CUADrO A-10
PENSIONES INDIVIDUALES ($ 2003) ANEXO
Media 1990 efectiva 2003 efectiva Simulacin 90 Precio 03 Participacin 03 Ambos 03 132,5 107,8 136,8 103,4 142,1
Gini 43,4 39,8 37,8 42,9 37,9
90/10 7,7 4,6 4,2 7,6 4,3
90/50 3,1 3,5 3,1 3,4 3,3
10/50 0,40 0,76 0,74 0,45 0,77
75/25 2,2 2,0 1,7 2,3 1,8
328
GrFICO A-1
PArTICIPACIN EN SUBSIDIOS, CENTILES DE HOGArES (INGrESO PEr CPITA) 2003 VS 1990
.8
Participacin en subsidios
.6
.4
.2
20
40 Percentil Promedio 1990
60
80
100
Promedio 2003
GrFICO A-2
MONTO PrOMEDIO DE SUBSIDIO, CENTILES DE HOGArES SEGN INGrESO P/C, 1990 VS 2003 (Considera solo hogares con subsidios) 30.000
Monto promedio de subsidios
25.000 20.000 15.000 10.000 5.000 0 20 40 Percentil Promedio 1990 Promedio 2003 60 80 100
329
GrFICO A-3
TASA DE OCUPACIN MUJErES 15-65, POr PCTILE INGrESO PEr CPITA, 1990 y 2003 .6 .5 .4 .3 .2 .1 0 20 40 Percentil Promedio 1990 Promedio 2003 60 80 100
Tasa de ocupacin mujeres
GrFICO A-4
SIMULACIN DE PENSIONES 1.5
Tasa de cambio
.5
0 0 20 40 60 80 100
Centiles de la distribucin Acceso Montos
SUSCRIPCIONES AO 2011 Suscripcin anual CHILE: Instituciones Individuo Estudiantes Nmeros sueltos (Incluido transporte areo) $ 20.000 $ 15.000 $ 10.000* $ 8.000 c/u Subscriptions rates All other countries: Institutions Individuals Single copy rate (Includes postage)
US$ 40 US$ 35 US$ 15
*Se debe presentar certificado de matrcula. Toda correspondencia puede hacerse al Editor; en cuanto a pedido de publicaciones, puede hacerse mediante cheque o solicitarlas mediante factura proforma a: Facultad de Economa y Negocios, Universidad de Chile, Campus Andrs Bello, Diagonal Paraguay 257, torre 26, casilla 3861, Santiago, Chile.
Correspondence should be addressed to Editor, Estudios de Economa. Make all remittances payable to: Facultad de Economa y Negocios, U. de Chile. P.O. Box 3861, Santiago, Chile.
Suscripcin y venta: Campus Andrs Bello: Diagonal Paraguay 257, Oficina 1606 Piso 16 - Torre 26, Santiago, Chile. Telfonos: 978 3411 - 978 3410
Vol. 48 No. 1 | MAY, 2011
CONTENTS
Fighting informality in segmented labor markets A general equilibrium analysis applied to Uruguay Carmen Estrades | Mara Ins Terra ........................................... 1 On the properties of general equilibrium with default in economies with incomplete markets Eduardo A. Rodrguez .............................................................. 39 The impact of financial transactions taxes on money demand in Colombia Marcela Giraldo | Brian W. Buckles .......................................... 65 Revenue elasticity of the main federal taxes in Mexico Felipe J. Fonseca | Daniel Ventosa-Santaulria ............................ 89
SUBSCRIPTIONS
Prices for annual subscription (2 issues, May and November)

Chile Latin America Rest of the world $9,000 US$30 US$40
Please fill out the form available at:
ww.laje-ce.org/subscriptions
INSTITUTO DE ECONOMIA PONTIFICIA UNIVERSIDAD CATOLICA DE CHILE
Sumario
N 121 Verano 2011
ESTUDIOS PBLICOS
www.cepchile.cl
Harald Beyer
Qu veinte aos no es nada...? Una mirada a la desigualdad de ingresos a partir de las encuestas Casen Alcances jurdicos del Convenio 169 Implementacin y evolucin de los derechos contenidos en el Convenio OIT 169: Aporte de la jurisprudencia y pronstico de desarrollo futuro de sus implicancias ambientales y regulatorias Una visin panormica al Convenio OIT 169 y su implementacin en Chile El Convenio 169 de la OIT sobre pueblos indgenas y tribales: Derecho internacional y experiencias comparadas Conferencia Dilema energtico: La energa nuclear Libros La independencia de Chile y el republicanismo (La Creacin de la Repblica, de Vasco Castillo) Mercaderes, Empresarios y Capitalistas (Chile, Siglo XIX), de Gabriel Salazar Comentario al libro de Gabriel Salazar Mercaderes, Empresarios y Capitalistas (Chile, Siglo XIX)
Jos Miguel Valdivia Andrs Fernndez y Cristhian de la Piedra
Santiago Montt y Manuel Matta Elina Mereminskaya
Jorge Zanelli
Renato Cristi
Alfredo Jocelyn-Holt Rolf Lders
CENTRO DE ESTUDIOS PBLICOS

Monseor Stero Sanz 162. Santiago, Chile. Fono (56-2) 328-2400. Fax (56-2) 328-2440.
SUSCRIPCIONES: Anual $ 9.000 Bianual $ 13.500 Estudiantes $ 5.000
CONTENIDO DEL VOLUMEN 25 N 2 DICIEMBRE DE 2010

Nmero Especial / Special Issue Inequidad y movilidad de ingresos en Latinoamrica / Inequality and income mobility in Latinamerica Editores Invitados / Invited Editors Marcela Perticar, Hugo opo ARTICULOS / ARTICLES Marcela Perticar, Hugo opo Introduccin Introduction Viviane M. R. Azevedo, Csar P. Bouillon Intergenerational social mobility in Latin America: A review of existing evidence Movilidad social intergeneracional en Amrica Latina: Una revisin de la evidencia actual Pablo Celhay, Claudia Sanhueza, Jos R. Zubizarreta Intergenerational mobility of income and schooling: Chile 1996-2006 Movilidad intergeneracional del ingreso y la educacin: Chile 1996-2006 Ana Ins Navarro Estimating long term earnings mobility in Argentina with pseudo-panel data Medicin de la movilidad de ingresos de largo plazo en Argentina usando pseudo-paneles Catalina Franco, Johanna Ramos Diferenciales salariales en Colombia: Un anlisis para trabajadores rurales y jvenes, 2002-2009 Earnings differentials in Colombia: A study of young and rural workers, 2002-2009 Marcelo Brgolo, Fedora Carbajal Exploring the urban-rural labor income gap in Uruguay: A quantile regression decomposition Explorando la brecha de ingresos laborales urbano-rural en Uruguay: Una descomposicin de regresin por cuantiles Alejandro Badel, Ximena Pea Decomposing the gender wage gap with sample selection adjustment: Evidence from Colombia Descomponiendo la brecha salarial de gnero con ajuste de sesgo de seleccin: El caso colombiano Guillermo Paraje, Melvyn Weeks Income nonresponse and inequality measurement Falta de respuesta en ingresos y medicin de la desigualdad
SUSCRIPCIONES Los valores de suscripcin para los dos nmeros de 2010, incluyendo envo areo, son los siguientes: Chile Resto del Mundo (pesos chilenos) (dlares) Individuos Instituciones Nmeros sueltos 12.000 16.000 6.000 35.00 40.00 20.00
El pago del resto del mundo debe hacerse en cheque en dlares de Estados Unidos o por medio de una money order internacional, pagadero a Universidad Alberto Hurtado. Enve toda correspondencia relativa a esta materia a: Editor, Revista de Anlisis Econmico, ILADES/Georgetown University, Erasmo Escala 1835, Santiago, Chile. SUBSCRIPTIONS The subscription rates for 2010 (Vol. 25, two issues) including air mail, are: Individuals Institutions Single copy rate 12.000 16.000 6.000 35.00 40.00 20.00
Make all remittances payable to Universidad Alberto Hurtado, in United States Currency. Correspondence should be addressed to Editor, Revista de Anlisis Econmico, ILADES/Georgetown University, Erasmo Escala 1835, Santiago, Chile.
ECONOMIC ANALYSIS
R E V I E W
ANDROS IMPRESORES www.androsimpresores.cl

E Economia 38 1 2011

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

E Economia 38 1 2011

Uploaded by

Copyright:

Available Formats

ESTUDIOS DE ECONOMIA

VOLUMEN 38 N 1 / JUNIO 2011

Introduction to the issue David Bravo

UNIVERSIDAD DE CHILE FACULTAD DE ECONOMIA Y NEGOCIOS DEPARTAMENTO DE ECONOMIA

129 169 195 223 243

Introduction to the issue

Centro de Microdatos, Departamento de Economa, Universidad de Chile. E-mail: dbravo@ econ.uchile.cl

Estudios de Economa, Vol. 38 - N 1

Introduction to the issue / David Bravo

Estudios de Economa, Vol. 38 - N 1

How much might human capital policies / Jere R. Behrman

Group mean (unweighted)

Estudios de Economa, Vol. 38 - N 1

How much might human capital policies / Jere R. Behrman

Discount Rate Lag in years 10 40 60 100

3.00% $ 717 $ 281 $ 134 1

5.00% $ 592 $ 130 $ 42 0

Survival Prob 0.964 0.915 0.787 0.007

Survival probabilities based on WHO Life Tables for Brazil at http://www.who.int/countries.

Estudios de Economa, Vol. 38 - N 1

Based on Behrman, Cheng and Todd (2004).

How much might human capital policies / Jere R. Behrman

Estudios de Economa, Vol. 38 - N 1

Based on Angrist et al. (2002) as summarized in Knowles and Behrman (2004).

How much might human capital policies / Jere R. Behrman

Estudios de Economa, Vol. 38 - N 1

Lottery Winners 336

? ? ? 227 236 186 109 86 0

? ? ? 227 236 186 195

How much might human capital policies / Jere R. Behrman

Discounted to age 13a

Estudios de Economa, Vol. 38 - N 1

Source: See text. a Using a discount rate of 5%.

How much might human capital policies / Jere R. Behrman

Estudios de Economa, Vol. 38 - N 1

How much might human capital policies / Jere R. Behrman

Estudios de Economa, Vol. 38 - N 1

How much might human capital policies / Jere R. Behrman

1b. Treatment for women with presumptive STD

1c. Drugs for pregnant women with poor obstretic history

$ 1,000 $ 5,000 $ 1,000 $ 5,000 $ 1,000 $ 5,000

$ 13,135 $ 45,062 $ 13,135 $ 45,062 $ 13,135 $ 45,062

$ 1,100 $ 1,100 $ 276 $ 276 $ 154 $ 154

$ 2,766 $ 7,460 $ 2,766 $ 7,460 $ 2,766 $ 7,460

$ 1,100 $ 1,100 $ 276 $ 276 $ 154 $ 154

$ 26,542 $ 89,484 $ 1,024 $ 1,024

3b Vitamin A (pre child under six years)

$ 1,000 $ 5,000 $ 1,000 $ 5,000 $ 1,000 $ 5,000

$ 12,244 $ 40,607 $ 520 $ 2,377 $ 14,831 $ 46,758

$ 2,304 $ 5,147 $ 257 $ 1,221 $ 3,261 $ 7,955

Estudios de Economa, Vol. 38 - N 1

1a. Treatments for women with asymptomatic bacterial infections

1b. Treatment for women with presumptive STD

1c. Drugs for pregnant women with poor obstretic history

How much might human capital policies / Jere R. Behrman

3a. Iodine (per woman of child bearing age)

3b Vitamin A (pre child under six years)

Estudios de Economa, Vol. 38 - N 1

How much might human capital policies / Jere R. Behrman

Gini Coefficient SE 0.23 0.001 0.53 0.003 0.49 0.004

9.5 4.3 2149 3195 1193 1751

Estudios de Economa, Vol. 38 - N 1