Intergenerational educational mobility within Chile

Muñoz S., Ercio Working Paper Intergenerational educational mobility within Chile IDB Working Paper Series, No. IDB-WP-01707 Provided in Cooperation with: Inter-American Development Bank (IDB), Washington, DC Suggested Citation: Muñoz S., Ercio (2025) : Intergenerational educational mobility within Chile, IDB Working Paper Series, No. IDB-WP-01707, Inter-American Development Bank (IDB), Washington, DC, https://doi.org/10.18235/0013513 This Version is available at: https://hdl.handle.net/10419/324783 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by/3.0/igo/ Intergenerational Educational Mobility within Chile Ercio Munoz WORKING PAPER No IDB-WP-01707 Inter-American Development Bank Gender and Diversity Division April 2025 Intergenerational Educational Mobility within Chile Ercio Munoz Inter-American Development Bank Gender and Diversity Division April 2025. Cataloging-in-Publication data provided by the Inter-American Development Bank Felipe Herrera Library Muñoz, Ercio. Intergenerational educational mobility within Chile / Ercio Muñoz. p. cm. — (IDB Working Paper Series ; 1707) Includes bibliographical references. 1. Educational mobility-Chile. 2. Educational equalization-Chile. 3. Social mobility-Chile. I. Inter-American Development Bank. Gender and Diversity Division. II. Title. III. Series. IDB-WP-1707 http://www.iadb.org Copyright © 2025 Inter-American Development Bank ("IDB"). This work is subject to a Creative Commons license CC BY 3.0 IGO (https://creativecommons.org/licenses/by/3.0/igo/legalcode ). The terms and conditions indicated in the URL link must be met and the respective recognition must be granted to the IDB. Further to section 8 of the above license, any mediation relating to disputes arising under such license shall be conducted in accordance with the WIPO Mediation Rules. Any dispute related to the use of the works of the IDB that cannot be settled amicably shall be submitted to arbitration pursuant to the United Nations Commission on International Trade Law (UNCITRAL) rules. The use of the IDB's name for any purpose other than for attribution, and the use of IDB's logo shall be subject to a separate written license agreement between the IDB and the user and is not authorized as part of this license. Note that the URL link includes terms and conditions that are an integral part of this license. The opinions expressed in this work are those of the authors and do not necessarily reflect the views of the Inter-American Development Bank, its Board of Directors, or the countries they represent. Intergenerational Educational Mobility within Chile∗ Ercio Mu˜noz† Inter-American Development Bank April 2025 Abstract I provide estimates of intergenerational mobility (IGM) in education at a disaggregated geographic level for Chile, a country with high school-level stratification by socioeconomic status and a decentralized administration of public schools. I document wide variation across communes. Relative mobility is correlated to the number of doctors, the number of students per teacher, and earnings inequality. Using a LASSO, I find that the share of students enrolled in public schools, the number of students per teacher, population density, and municipal budget are the strongest predictors of IGM. I also document within-country variability in how parental education is associated with other children’s outcomes. JEL-Codes: D63, I24, J62. Keywords:Socioeconomic mobility, Geography, Chile, Education. ∗I am grateful to Jonathan Conning, Christos Giannikos, Roy van der Weide, N´uria Rodr´ıguez-Planas, Wim Vijverberg, and the participants of the micro brown bag, the dissertation seminar at CUNY and the virtual WEAI 2021 Conference for useful comments. Pablo Vargas provided outstanding research assistance. Any views expressed in this paper are those of the authors and should not be attributed to the Inter-American Development Bank. †Email: [email protected] 1 I Introduction How much of an individual’s educational achievement is due to his or her parents’ educational achievements? High persistence in educational outcomes across generations can lead to unrealized human capital potential and inefficient allocation of resources and talents that result in lower economic growth. Moreover, it can be a mechanism by which economic advantage is inherited, as education is linked to the capacity to generate income and wealth. Economists have made important progress in documenting the level of intergenerational mobility (IGM) in education (i.e., the relationship between educational outcomes of parents and children) for many countries (see Van der Weide et al. 2024).1However, the evidence at the country level can hide important variation within countries, as it has been shown by a growing literature (for example, Alesina et al. 2021,2023,Asher et al. 2024,Card et al. 2022,Derenoncourt 2022,Hilger 2016,Feigenbaum 2018,Munoz 2024) that estimates IGM for small geographical units, extending the literature on IGM in income initiated by Chetty et al. (2014). In this paper, I contribute to this literature in three ways. First, I estimate intergenerational mobility in education in Chile at the country, region, and commune level using census data for a cohort born in the 1990s. I offer eight indicators that describe the association between children’s and parents’ years of schooling,2capturing different policy-relevant concerns and aspects of this association (e.g., how much more educated children with parents with an extra year of schooling tend to be, how likely are children from low-educated parents to be low-educated, how likely are children to surpass the education of their parents, etc.), therefore providing a broad view of IGM.3I provide these estimates in an online data appendix for future research. Second, I show how other children’s outcomes, such as teenage 1See Torche (2021) for a survey focused on developing countries. 2Throughout the paper, I will refer to the cohort of interest as “children,” and I will refer to their parents or older relatives living in the same household as “parents”. I will precisely define who will be considered as a parent in Section II. 3Deutscher & Mazumder (2023) provide a framework that highlights the key concepts and properties of different indicators of IGM. 2 pregnancy and tertiary education attendance, are also associated with parental education at the country level and display wide variation within the country. Finally, I explore how the estimates of educational IGM are correlated with a rich set of variables related to income, geography, education, municipal budget, and other characteristics of the communes. Furthermore, I investigate by means of a lasso (least absolute shrinkage and selection operator), what correlates have the most predictive power over IGM at the level of commune. IGM literature for Chile. Previous studies have used different household and opinion surveys (see for example, Torche 2005,Hertz et al. 2007,Nunez & Miranda 2010,Narayan et al. 2018,Celhay et al. 2010,Celhay & Gallegos 2015,Sapelli 2016,Neidh¨ofer et al. 2018, Van der Weide et al. 2024) to document IGM in income, education, and other socioeconomic measures. However, they all have in common that the samples are not representative at the commune level, so they focus on country-level estimates. Two exceptions are Celhay & Gallegos (2015) which also explores mobility at the regional level (the coarser administrative unit in which the country is divided), and Cort´es Orihuela et al. (2023), which uses labor earnings in the formal sector from administrative records to estimate income mobility at the regional level and between communes (the smallest administrative unit) in the Metropolitan Region. Institutional background. Chile is an interesting case study to analyze IGM at the subnational level. On the one hand, the country is one of the richest economies in the Latin American region and has shown significant progress in poverty reduction and income per capita growth in the last three decades. On the other hand, income inequality is relatively high for OECD standards, and previous research has documented high school-level stratification by socioeconomic status (Mizala & Torche 2012,Guti´errez & Carrasco 2021), which has fueled some educational reforms in the last decade. In addition, the country is marked by the free-market reforms inherited from the military dictatorship (1973-1990). This includes a universal voucher system and decentralization of the administration of public schools, which 3 are managed by municipalities.4 In terms of IGM at the country level, the best evidence available at a global scale (Narayan et al. 2018,Van der Weide et al. 2024) shows some interesting findings for Chile. Among the 148 countries for which there are estimates of educational mobility for the cohort born in the 1980s, the country ranks relatively low when a summary statistic of relative mobility, such as one minus the Pearson correlation coefficient between years of schooling of children and parents is used but somewhat more mobile according to one minus the regression coefficient between these two variables.5In contrast, the country seems much more mobile when we look at a measure of absolute mobility like the share of students with higher education than parents. However, when a measure that aims to capture directional mobility from the bottom to the top is considered (i.e., “rags to riches” or poverty to privilege rate as named in Narayan et al. 2018), then the country appears among the least mobile ones (see Figure A1 in the Appendix).6 The evolution across different cohorts for these indicators also show some interesting patterns when compared to simple averages by region as classified in Narayan et al. (2018).7Chile does not show much progress in most of the indicators relative to regional averages except for absolute mobility (share of students with higher education than parents) and relative mobility measured as 1 −β. In contrast, relative mobility measured as 1 −ρ(independent of the marginal distributions of education) has remained at lower levels than all the regional averages for all the cohorts in the same way as the poverty to privilege ratio (or rags to riches). 4A recent reform started a process of centralization in 2018. 5The correlation coefficient can be transformed into the regression coefficient by multiplying it by the ratio of the standard deviation of child schooling to parent schooling. Therefore, differences between them are explained by changes in inequality across generations. 6Ranked 138 among the 148 available estimates. 7Figure A2 in the Appendix plots all these indicators across cohorts. 4 II Data and Methods Data. I use individual-level data from the 2017 census of housing and population obtained from the National Institute of Statistics.8This statistical operation, which aimed to capture the total population of Chile, includes demographic details such as age, sex, education, household composition, as well as detailed geographical information. Sample definition. The full-count census database contains information about 17,574,003 individuals. I keep people born in Chile aged between 21 and 25 years and drop those considered domestic service, living in collective housing, or in transit, which reduces the sample to 1,155,207 individuals, 568,231 men and 586,976 women.9 Education. The census data contains a variable reporting schooling, regardless of the track or kind of study. When I study how the educational attainment of children relates to the attainment of parents, I take the highest attainment among the individuals in the older generation.10 Given the typical educational path in Chile, where students start first grade at the age of 6, the average student would be able to attain at most 15 years of schooling at the age of 21. To accommodate for this, the indicators are computed using years of schooling censored at 15 for both children and parents.11 Geography. Chile is divided into 16 regions, 56 provinces, and 346 communes.12 The data set contains information on where the interview was conducted and the place of birth in terms of these three administrative divisions. I use the latter to assign people to places and estimate IGM for the country, by region, and by commune. 8The data can be accessed at https://www.ine.gob.cl/estadisticas/sociales/censos-de-poblacion-yvivienda/censo-de-poblacion-y-vivienda. 9Van der Weide et al. (2024) use the same age range to estimate IGM in education using survey data for 39 countries. 10The results are qualitatively similar if I use the average rounded to the nearest integer instead of the maximum. 11Similar censoring of years of schooling is used in Neidh¨ofer et al. (2018) with survey data to compute IGM at the country level for 18 countries in Latin America. Figure A3 displays the distribution of educational attainment of parents and children. 12Chile does not have a commonly used designation of commuting zones, such as the one used in Chetty et al. (2014) for the United States. Many communes are within commuting distance in the country, particularly within the Metropolitan area. However, the estimates at the regional level for this specific case are close to what could be considered a commuting zone. 5 case of women.22 These outcomes are potentially very consequential in life trajectories, as the first one is positively associated with earnings and other indicators of well-being (Oreopoulos & Petronijevic 2013), while the second is negatively associated with income and some indicators of well-being (Fletcher & Wolfe 2009). Furthermore, these outcomes can be measured at earlier ages than our main outcome (i.e., years of schooling), reducing the magnitude of any potential coresidence bias, as coresidence rates decrease with age. First, I estimate the probability of attending at least one year of tertiary education using a sample of individuals between the ages of 19 and 21. Figure A4c shows this likelihood for each parental educational attainment, finding a positive slope approximately equal to 0.046 with a somewhat prominent discontinuity at 12 years of schooling and a somewhat nonlinear relationship for low values of parents’ years of schooling. This contrasts with the virtually linear relationship between parental income rank and college attendance documented for the US in Chetty et al. (2014). Despite these differences and other differences in terms of measurement and concepts, I find similar gaps. The gap in the likelihood of attending tertiary education for individuals with low-educated vs. highly-educated parents is around 60 percentage points while Chetty et al. (2014) documented a gap of 67.5 percentage points in the US for individuals with lowest-income vs. highest-income parents. Second, I estimate the probability of becoming a mother as a teenager, defined as having a child for females between the ages of 15 and 19. Figure 2b shows this likelihood for each parental educational attainment, finding a negative relationship close to linear with a slope of -0.017. The gap between highly-educated and low-educated parents is around 20-25 percentage points (Chetty et al. 2014, documents a gap of 29.8 percentage points for highest-lowest parents’ incomes). 22I use the same econometric specification as in Equation 1with a different dependent variable. 12 Figure 2: Other child’s outcomes 0 .2 .4 .6 .8 1 Offspring's likelihood of attending tertiary education 0 5 10 15 20 Parents' years of schooling (a) Tertiary education attendance 0 .1 .2 .3 Likelihood of being mother as teenager 0 5 10 15 20 Parents' years of schooling (b) Teenage birth (females only) Notes: The first plot displays the likelihood of completing at least one year of tertiary education for each level of education of the generation above (highest years of schooling among parents and older relatives living in the same household). The second plot displays the likelihood of having a child as teenager for each level of education of the generation above. The samples include individuals with age between 19 and 21 (left) and 15 and 19 (right). The size of the bubble varies according to the number of individuals. III.2 Intergenerational mobility within Chile Region-level estimates. Before presenting the most disaggregated estimates, Table 4summarizes the eight measures of interest estimated for the 16 regions of Chile. Non-negligible differences can be found across regions in most of these dimensions. For example, the chances of reaching the top quintile of the educational distribution for children with parents at the bottom quintile (i.e., P1,5) is more than 200% higher in the northern Arica y Parinacota region relative to Ays´en region. Similarly, in terms of absolute mobility (i.e., α), there are regions with more than one year of difference, and relative mobility (i.e., 1−β) is 17% higher in Arica y Parinacota than in Metropolitana de Santiago or Los Rios. When I consider relative mobility measured with the correlation coefficient (1−ρ), the level in Arica y Parinacota is approximately 30% higher than in the region with the lowest value (Araucan´ıa).23 23Table A4 in the Appendix reports the last three indicators of IGM using the distribution of educational attainment at the country level versus at the region level. There is heterogeneity in the direction of change, but for the three indicators, the range of variation decreases using the local distribution. 13 Table 4: Region-level estimates of IGM Statistics Region α1−β¯ Y¯y≥1−ρ P1,5P1,1P5,5 Tarapac´a 9.66 0.74 11.53 0.60 0.70 0.10 0.37 0.32 Antofagasta 9.25 0.71 11.61 0.57 0.68 0.08 0.40 0.31 Atacama 9.54 0.73 10.99 0.62 0.68 0.07 0.40 0.29 Coquimbo 9.44 0.72 10.53 0.65 0.66 0.07 0.38 0.32 Valpara´ıso 9.61 0.72 11.23 0.65 0.68 0.09 0.35 0.34 Libertador General Bernardo O’Higgins 10.06 0.76 9.95 0.71 0.71 0.10 0.34 0.31 Maule 9.75 0.73 9.84 0.73 0.67 0.09 0.36 0.32 Biob´ıo 10.12 0.74 10.65 0.74 0.66 0.11 0.33 0.37 Araucan´ıa 9.58 0.71 9.88 0.75 0.61 0.07 0.38 0.37 Los Lagos 9.35 0.71 9.77 0.71 0.65 0.07 0.41 0.31 Ays´en del General Carlos Ib´a˜nez del Campo 9.38 0.76 9.59 0.65 0.73 0.05 0.44 0.23 Magallanes y de la Ant´artica Chilena 10.36 0.77 11.33 0.66 0.72 0.10 0.30 0.30 Metropolitana de Santiago 9.38 0.70 11.33 0.64 0.65 0.09 0.37 0.36 Los R´ıos 9.46 0.70 10.18 0.72 0.62 0.06 0.38 0.34 Arica y Parinacota 10.76 0.82 11.49 0.61 0.79 0.14 0.28 0.31 ˜ Nuble 10.02 0.74 9.86 0.76 0.68 0.11 0.33 0.36 Notes: The table reports region-level estimates of absolute mobility, relative mobility (1−β), average parents’ education, the share of children with higher education than parents, relative mobility (1−ρ), rags to riches, intergenerational low, and intergenerational high, respectively. A description of the measures can be found in Table 2. Rows are sorted by the official designated number that each region used to have until 2018. Commune-level estimates. I document wide within-country variation. Relative mobility measured as 1 −β, excluding places with less than 50 individuals24, ranges between 0.54 in Quemchi, a commune located in the south of the country, and 0.97 in San Pedro de Atacama, a commune located in the north. Non-negligible variation is found in all the indicators studied. Figure A10 in the Appendix shows the distributions of the communelevel estimates for the eight measures and Table A3 of the Appendix similarly reports some descriptive statistics of these estimates. For all the indicators I can find communes with levels at least 100% greater than others, in some cases several times greater. The measures of mobility based on conditional probabilities derived from quintiles of educational attainment are constructed using the distribution of attainment at the country level for children and similarly for parents. Similar measures could be constructed using the distribution of attainment by commune. In this case, moving from the bottom to the top 24Figure A9 in the Appendix displays the CDF of the sample size by commune, showing that less than 5% of the communes have less than 50 observations. 14 may require a higher number of years of schooling in some places compared to others and capture a different aspect of mobility. As an additional exercise, I compute those measures and find that P1,5measures constructed in both ways are highly correlated while P1,1is to a lesser degree while in contrast, P5,5is not correlated (see Figure A11 in the Appendix.). Figure 3maps relative mobility (1 −β) across the country. There are some regions with clusters of communes showing relatively similar levels of IGM, such as the northern regions and more heterogeneity in the center of the country. Figure A15 in the Appendix plots relative mobility dividing the map of the country into three parts, a northern region less the metropolitan region, the metropolitan region, and a southern region. These three regions have communes with relatively low and high levels of intergenerational educational mobility. However, in this map the variety in IGM levels of the metropolitan region (where the highest share of the population lives) can be appreciated in more detail. Correlations among different measures of IGM. Table 5presents the Pearson correlation coefficients between the eight mobility statistics computed at the commune level. I find the strongest positive correlation to be between absolute and relative mobility, both measured with 1 −βand 1 −ρ. These three measures are at the same time positively correlated to above parents and rags to riches, especially absolute mobility. Intergenerational low is negatively correlated with the other six indicators. Table 5: Correlation among IGM statistics α1−β¯ Y¯y≥1−ρ P1,5P1,1P5,5 Absolute mobility (α) 1 Relative mobility (1 −β) 0.912∗∗∗ 1 Average education (¯ Y) -0.0175 -0.146∗∗ 1 Above parents (¯y≥) 0.268∗∗∗ 0.139∗-0.716∗∗∗ 1 Relative mobility (1 −ρ) 0.713∗∗∗ 0.874∗∗∗ -0.0917 -0.0128 1 Rags to riches (P1,5) 0.478∗∗∗ 0.259∗∗∗ 0.296∗∗∗ 0.0604 0.228∗∗∗ 1 Intergenerational low (P1,1) -0.730∗∗∗ -0.517∗∗∗ -0.207∗∗∗ -0.166∗∗ -0.380∗∗∗ -0.537∗∗∗ 1 Intergenerational high (P5,5) -0.00472 -0.141∗0.0369 0.233∗∗∗ -0.140∗0.236∗∗∗ -0.0715 1 ∗p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001 Other outcomes within Chile. I estimate the relationship between parental education and 15 Figure 3: Intergenerational educational (1 −β) mobility within Chile Notes: The map plots relative IGM measured as one minus the regression coefficient (by commune) between children’s years of schooling (using individuals aged between 21 and 25) against parents’ years of schooling. Educational attainment is censored at 15. Communes with less than 50 individuals are left as missing (Figure A9 in the Appendix displays the CDF of the sample size by commune). Figure A15 in the Appendix displays a version of the map dividing Chile into three areas. the two alternative outcomes described in the previous section: the likelihood of attending at least one year of tertiary education, and the likelihood of being a mother as a teenager 16 for females. Table 6reports these estimates at the regional-level. There is significant variation across regions in the effect of an additional year of parental schooling on the chances of attending tertiary education. Araucania shows the strongest effect (0.044), which suggest that the gap between individuals with uneducated parents and those with highly educated ones (21 years) in the chances of attending tertiary education is approximately 92 percentage points (21 ×0.044). A caveat to note is that this calculation may overestimate the effect in light of the non-linearity observed at the national level in Figure A4c for lower levels of parental education. If I assume that the effect is null in the first 5 years of education, then the gap is approximately 70 percentage points. On the other extreme, Ays´en region shows the smallest average effect (0.019). Similarly, the effect of an extra year of parents’ schooling on teenage birth rates varies significantly across regions. The effect of one year goes from a fall in the likelihood of a teenage birth equal to 0.8 percentage points in ˜ Nuble to 1.6 percentage in Antofagasta or Coquimbo. This last effect implies a gap between uneducated and highly educated parents of approximately 33.6 percentage points, which again is meaningful but may be an overestimation due to non-linearities. IV Correlates of IGM within Chile In this section, I examine whether intergenerational educational mobility at the commune level is correlated with a broad set of variables, including income distribution, educational characteristics, municipal budgets, geographic factors, and other local attributes. Understanding these relationships is crucial to identifying the main factors that contribute to the persistence of educational outcomes across generations. Given the large number of potential predictors, the analysis follows a two-step approach. First, I investigate the correlations between IGM and selected key variables that the literature and theoretical frameworks suggest 17 Table 6: Parental education effect on other outcomes Region Tertiary education Teenage birth Tarapac´a 0.038 -0.013 Antofagasta 0.038 -0.016 Atacama 0.042 -0.012 Coquimbo 0.040 -0.016 Valpara´ıso 0.042 -0.014 Libertador General Bernardo O’Higgins 0.028 -0.010 Maule 0.034 -0.012 Biob´ıo 0.039 -0.013 Araucan´ıa 0.044 -0.013 Los Lagos 0.035 -0.014 Ays´en del General Carlos Ib´a˜nez del Campo 0.019 -0.015 Magallanes y de la Ant´artica Chilena 0.033 -0.009 Metropolitana de Santiago 0.043 -0.015 Los R´ıos 0.039 -0.013 Arica y Parinacota 0.026 -0.012 ˜ Nuble 0.037 -0.008 Notes: The table reports the association between an extra year of parents’ schooling and the likelihood of completing at least one year of tertiary education, as well as the likelihood of having a child as teenager for females (computed using an OLS regression). The samples include individuals aged between 19 and 21 (left) and 15 and 19 (right). Rows are sorted by the official designated number that each region used to have until 2018. as particularly relevant. Second, I apply a LASSO (Least Absolute Shrinkage and Selection Operator) regression to identify the strongest predictors of mobility. This method performs a selection of correlates based on their predictive strength. The selection of correlates follows considerations of data availability and is based on previous research, particularly Alesina et al. (2021), Chetty et al. (2014) and Van der Weide et al. (2024). An important caveat is that this analysis should not be interpreted as causal. The sole purpose is to document stylized facts that can later be used to theoretically model or estimate empirically the mechanisms behind local differences in intergenerational mobility. IV.1 Bivariate associations This section explores the bivariate relationship between indicators of IGM and a broad set of variables, including income inequality, governance, public investment, education, and 18 health services, among others. The focus is on relative mobility measured as one minus the regression coefficient for simplicity and because it is arguably the most widely used indicator. However, a summary table with results for the remaining indicators can be found in the Appendix (Table A5). The definition of the correlates and their data sources are listed in the Appendix (see Table A1). To allow for a lag between the contextual environment and the observed level of mobility, all variables are measured in the year 2010.25 Figure 4reports the coefficients from regressions with relative mobility (standardized to have mean 0 and variance 1) as the dependent variable and a given correlate (standardized to have mean 0 and variance 1) as the independent variable, which are labeled as unconditional, and then the same regression controlling for average education of the older cohorts.26 This analysis provides an initial understanding of how these factors relate to IGM before applying a selection method based on their predictive ability. The unconditional estimates provide a descriptive perspective on the raw correlations between mobility and each variable, while the conditional estimates help isolate the effect above and beyond the average educational attainment of parents. Income distribution. Higher levels of inequality are often associated with lower mobility, as economic advantages and disadvantages persist across generations (Corak 2013). I include indicators such as the Gini index, specific income quantiles (10th, 50th, 90th, 95th), and income ratios (90/10, 90/50, 50/10) to capture different dimensions of inequality. The hypothesis is that greater income disparity, particularly between the richest and poorest households, limits access to high-quality education and economic opportunities, thereby reducing mobility. I find that the Gini index, 90th quantile, 95th quantile, 90/10 ratio, and 90/50 ratio are all negatively and significantly correlated with relative mobility at the 5% level. These results suggest that in Chile, higher intergenerational mobility in education is more strongly associ25The exception in terms of year of measurement is population, which is computed using Census 2017. 26Following the approach used in Alesina et al. (2021). 19 Figure 4: Correlates of relative mobility (1 −β) at the commune-level Gini index Avg. labor earnings 10th quantile 50th quantile 90th quantile 95th quantile 90th over 10th 90th over 50th 50th over 10th Area Dist. to reg. capital Density Population Municipal quality Personnel female share Felonies Budget Total expenditure Social expenditure Educational expenditure Students in public schools Students per teacher Secondary test scores Primary test scores Nurses Doctors Infant mortality Water network -0.6 -0.4 -0.2 0 0.2 0.4 0.6 Unconditional Conditional on avg. education of older cohorts Notes: The figure reports the coefficients from regressions with relative mobility at the commune level measured as one minus the regression coefficient (standardized to have mean 0 and variance 1) as the dependent variable and a given correlate (standardized to have mean 0 and variance 1) as the independent variable (in green, labeled as unconditional), and the coefficients from the same regression controlling for average education of the older cohorts (in orange). 95% confidence intervals are included. ated with lower levels of income inequality in the upper half of the income distribution. This contrasts with findings from Corak (2020), which show that in Canada, mobility in income is more associated with inequality in the lower half of the income distribution. However, these results align with country-level evidence reported by Narayan et al. (2018), showing that income inequality is positively associated with intergenerational persistence in education, meaning it is negatively associated with relative mobility. These findings suggest that the well-documented relationship between inequality and mobility at the international level may also hold within countries (see Figure A13 in the Appendix).27 27It is worth noting that the administrative dataset used to construct measures of income inequality, which comes from the unemployment insurance system, only considers the formal sector. This could lead to 20 To complement this analysis, I also examine the relationship between relative mobility and inequality in education among parents (individuals aged 40–60 at the time of the Census). This exercise is equivalent to constructing a “Great Gatsby curve” for education within Chile, and I find a negative relationship between educational inequality and mobility. This suggests that the relationship documented across countries by the Narayan et al. (2018) also holds within countries (see Figure A14 in the Appendix). Demographic and geographic characteristics. Spatial and demographic factors can influence mobility by affecting access to education, labor markets, and public services. I consider variables such as population size, density, commune area, and distance to the regional capital. The hypothesis is that larger and more urbanized communes, or those closer to economic centers, may offer better educational resources and job opportunities, leading to higher mobility. However, none of these demographic variables exhibit a statistically significant relationship with mobility. This suggests that within-country variations in population distribution and geography do not play as strong a role in educational mobility. Institutional and governance quality. The quality of local governance and institutional strength may affect mobility by influencing the efficiency of public service delivery. I include indicators such as municipal governance quality, female representation in the public sector, and felonies. The hypothesis is that better governance and lower crime levels may create an environment more conducive to higher intergenerational mobility. I do not find a significant relationship between these governance indicators and mobility. This may indicate that variations in local government performance across Chile are not substantial enough to drive differences in mobility outcomes. Public spending and social infrastructure. Public investment in education and social services can mitigate economic disadvantages and enhance mobility. I analyze municipal budgets, total public expenditure, social expenditure, and educational expenditure. The hypothesis is that higher public spending, particularly on education, should be positively an underestimation of income inequality, as informal workers are excluded. 21 Hertz, T., Jayasundera, T., Piraino, P., Selcuk, S., Smith, N. & Verashchagina, A. (2007), ‘The Inheritance of Educational Inequality: International Comparisons and Fifty-Year Trends’, The B.E. Journal of Economic Analysis Policy 7(2). Hilger, N. (2016), ‘The Great Escape: Intergenerational Mobility in the United States Since 1940’, NBER Working Paper (21217). Mazumder, B. (2016), Estimating the intergenerational elasticity and rank association in the united states: Overcoming the current limitations of tax data, in ‘Inequality: Causes and consequences’, Vol. 43, Emerald group publishing limited, pp. 83–129. Mizala, A. & Torche, F. (2012), ‘Bringing the schools back in: the stratification of educational achievement in the chilean voucher system’, International Journal of Educational Development 32(1), 132–144. Munoz, E. (2024), ‘The Geography of Intergenerational Mobility in Latin America and the Caribbean’, Econom´ıa LACEA Journal 23, 333–354. Munoz, E. & Siravegna, M. (2023), ‘When Measure Matters: Coresidence Bias and Intergenerational Mobility Revisited’, IDB Working Paper 01469. Narayan, A., Van der Weide, R., Cojocaru, A., Lakner, C., Redaelli, S., Gerszon Mahler, D., Ramasubbaiah, R. & Thewissen, S. (2018), Fair Progress?: Economic Mobility Across Generations Around the World, The World Bank. Neidh¨ofer, G., Serrano, J. & Gasparini, L. (2018), ‘Educational Inequality and Intergenerational Mobility in Latin America: A New Database’, Journal of Development Economics 134, 329–349. Nunez, J. I. & Miranda, L. (2010), ‘Intergenerational Income Mobility in a Less-Developed, High-Inequality Context: The Case of Chile’, The B.E. Journal of Economic Analysis Policy 10(1). 28 Oreopoulos, P. & Petronijevic, U. (2013), ‘Making college worth it: A review of research on the returns to higher education’. Sapelli, C. (2016), Cap 3: La Movilidad Intergeneracional de la Educacion en Chile, in ‘Chile : ¿M´as Equitativo? Una mirada a la dinamica social del Chile de ayer, hoy y ma˜nana’. Torche, F. (2005), ‘Unequal But Fluid: Social Mobility in Chile in Comparative Perspective’, American Sociological Review 70, 422–450. Torche, F. (2021), Educational Mobility in the Developing World, in V. Iversen, A. Krishna & K. Sen, eds, ‘Social Mobility in Developing Countries: Concepts, Methods, and Determinants’, Oxford University Press. Van der Weide, R., Lakner, C., Gerszon Mahler, D., Narayan, A. & Ramasubbaiah, R. (2024), ‘Intergenerational Mobility around the World: A New Database’, Journal of Development Economics 166, 103167. 29 Appendices The appendix provides additional tables and figures, and other relevant information. Table A1 lists the set of correlates that I use together with a short description and data sources. Table A2 reports all the indicators computed by sub-populations (male vs. female, indigenous vs. non-indigenous, and urban vs. rural). Table A3 reports some descriptive statistics of the estimates of IGM at the level of commune. Figure A1 plots different measures of intergenerational mobility in education at country-level highlighting where Chile falls relative to Latin America and the Caribbean and the world. Figure A2 plots different measures of intergenerational mobility in education for Chile compared to simple averages of regions for five different cohorts. Figure A3 displays an histogram with the distributions of educational attainment of parents and children. Figure A4 displays the evolution of mobility across birth cohorts in recent literature versus my estimate. Figure A5 plots the average coresidence rate against IGM indicators at the commune level. Figure A6 plots the average coresidence rate by level of education. Figure A7 plots the average coresidence rate by age. Figure A8 displays the transition probabilities between educational attainment of parents and children (classified into three categories). Figure A9 shows the cumulative distribution of the sample size by commune. Figure A10 displays the distribution of all the measures at commune-level. Figure A11 displays scatter plots comparing indicators of mobility (that use quintiles) using country level distribution of educational attainment vs. local distribution. Figure A15 maps the level of educational intergenerational mobility at the commune level separating the country into north, metropolitan region, and south. 30 Figure A12 reports the results of the correlations with a set of variables using all the measures of IGM. Figure A13 shows a binscatter plot between relative mobility and income inequality measured with the Gini coefficient at the commune level. Figure A14 shows a binscatter plot between relative mobility in education and educational inequality measured with the standard deviation of years of schooling at the commune level. 31 Table A1: Covariates Label Source Description Gini Index UID Gini Index Average earnings UID Average earnings in the formal sector 10th quantile UID 10th percentile of earnings in the formal sector 50th quantile UID 50th percentile of earnings in the formal sector 90th quantile UID 90th percentile of earnings in the formal sector 95th quantile UID 95th percentile of earnings in the formal sector Ratio 90-10 UID Ratio 90th to 10th percentile of earnings in the formal sector Ratio 90-50 UID Ratio 90th to 50th percentile of earnings in the formal sector Ratio 50-10 UID Ratio 50th to 10th percentile of earnings in the formal sector Area SINIM Log of the total surface of commune Distance to regional capital SINIM Log of the distance between the commune and the regional capital Population density per km2 SINIM Log of population density per km2 by commune Population SINIM Log of commune’s estimated population in June 2012 Municipal professionalization SINIM Share of college educated workers in the municipality Female Share in Municipality SINIM Share of female workers over the total workers in personnel of the municipality Crimes CEAD Log of the number of crimes with greater social connotation Budget availability SINIM Log of commune’s budget availability per capita Total expenditure SINIM Log of commune’s total expenditure per capita Social expenditure SINIM Log of the commune’s total expenditure in the social programs area per capita Education expenditure SINIM Log of the commune’e total expenditure education programs Students in public schools ACE Number of students enrolled in public schools over total enrollment Students per teacher SINIM Log of students per teacher ratio in the municipal education system Standarized test - secondary ACE Average score between math and language in SIMCE taken in high school Standarized test - primary ACE Average score between math and language in SIMCE taken in 4th grade Nurses by 100K inhabitants SINIM Log of number of nurses by 100.000 inhabitants within the commune Doctors by 100K inhabitants SINIM Log of number of doctors by 100.000 inhabitants within the commune Infant mortality rate SINIM Number of children under 1 year of age who die for every 1.000 live births Water network SINIM Percentage of homes connected to drinking water network in the commune Parental education Census Average education of individual older than 24 but younger than 66 Unemployment insurance database (UID) can be accessed at: https://www.spensiones.cl/apps/bdp/index.php. National system of municipal information (SINIM) can be accessed at: http://datos.sinim.gov.cl/datos municipales.php. Center for crime studies and analysis (CEAD) can be accessed at: http://cead.spd.gov.cl/estadisticas-delictuales/. Research unit, education quality agency data (ACE) can be accessed at: https://informacionestadistica.agenciaeducacion.cl/#/bases. Census 2017 data can be requested from the National Institute of Statistics at: https://www.ine.cl. 32 Table A2: IGM at country-level for subgroups Male Female Non-indigenous Indigenous Urban Rural α9.049 10.129 9.535 9.881 9.622 9.476 β0.688 0.742 0.710 0.748 0.717 0.718 ¯ Y11.126 11.123 11.260 10.247 11.336 9.203 ¯y≥0.628 0.707 0.663 0.690 0.657 0.754 ρ0.624 0.658 0.640 0.676 0.652 0.649 P15 0.068 0.108 0.089 0.082 0.092 0.071 P11 0.419 0.310 0.365 0.367 0.359 0.390 P55 0.331 0.378 0.357 0.306 0.353 0.358 The table reports country-level estimates of absolute mobility, relative mobility (1 −β), average parents’ education, share of children with higher education than parents, relative mobility (1 −ρ), rags to riches, intergenerational low, and intergenerational high, respectively, all computed by subgroup. A description of the measures can be found in Table 2. Table A3: Descriptive statistics of IGM at commune-level Mean SD Min Max N α9.79 0.66 7.16 11.73 330 1−β0.74 0.05 0.54 0.97 330 ¯ Y10.00 1.19 6.13 14.50 330 ¯y≥0.71 0.07 0.48 0.90 330 1−ρ0.68 0.06 0.50 0.96 330 P15 0.09 0.03 0.02 0.22 312 P11 0.36 0.06 0.09 0.57 313 P55 0.33 0.05 0.10 0.46 190 The table reports descriptive statistics of estimates of absolute mobility, relative mobility (1 −β), average parents’ education, share of children with higher education than parents, relative mobility (1−ρ), rags to riches, intergenerational low, and intergenerational high, respectively, all of them at the commune-level. I omit estimates with less than 50 observations. A description of the measures can be found in Table 2. 33 Figure A1: Chile relative to the world in terms of educational IGM .2 .4 .6 .8 1 1-beta Chile Latin America & Caribbean Other countries (a) Relative mobility (1 −β) .2 .4 .6 .8 1 1-rho Chile Latin America & Caribbean Other countries (b) Relative mobility (1 −ρ) 0 .2 .4 .6 .8 1 Share (weakly) Chile Latin America & Caribbean Other countries (c) Share of students with higher education than parents (share) .05 .1 .15 .2 .25 BHQ4 Chile Latin America & Caribbean Other countries (d) Directional IGM - Probability child from bottom half ends up in Q4 (BHQ4) .2 .3 .4 .5 .6 Q4Q4 Chile Latin America & Caribbean Other countries (e) Directional IGM - Probability child from Q4 ends up in Q4 (Q4Q4) .25 .3 .35 .4 .45 BHQ1 Chile Latin America & Caribbean Other countries (f) Directional IGM - Probability child from bottom half ends up in Q1 (BHQ1) Source: Elaboration by the author with data from Narayan et al. (2018). 34 Figure A2: Mobility in Chile versus average by region for five cohorts (a) Relative mobility (1 −β)(b) Relative mobility (1 −ρ) (c) Share with higher education than parents (d) Prob. child from bottom half ends up in Q4 (e) Probability child from Q4 ends up in Q4 (f) Prob. child from bottom half ends up in Q1 Source: Elaboration by the author with data from Narayan et al. (2018). Regional averages are unweighted. Regions are EAP: East Asia & Pacific; ECA: Europe and Central Asia; HI: High income; LAC: Latin America and the Caribbean; MENA: Middle East and North Africa; SA: South Asia; SSA: Sub-saharian Africa. 35 Figure A3: Histogram of education 0 .1 .2 .3 .4 Density 0 5 10 15 Year of schooling Children Parents 36 Figure A4: Own estimates versus recent literature at the country level (a) Relative mobility (1 −β) from Narayan et al. (2018) and own estimate. (b) Relative mobility (1 −β) from Neidh¨ofer et al. (2018) and own estimate. (c) Relative mobility (1 −ρ) from Narayan et al. (2018) and own estimate. (d) Relative mobility (1 −ρ) from Neidh¨ofer et al. (2018) and own estimate. The figure shows estimates of intergenerational educational mobility obtained from regressing children years of schooling against parents’ years of schooling, and the Pearson correlation coefficient between the same two variables. Narayan et al. (2018) uses CASEN survey while Neidh¨ofer et al. (2018) also uses Latinobarometro survey (LBM). The former uses 10-year cohorts, the latter uses 4-year cohorts (the most recent one is 1992-1995), and my estimate uses individuals approximately born between years 1991-1995. The last four cohorts using LBM survey contain smaller samples (831, 413, 179, and 24 observations), and hence are somewhat unreliable. 37 Figure A11: Comparison of indicators using country level distribution of educational attainment vs. local distribution (a) Rags to riches (P1,5)(b) Intergenerational low (P1,1) (c) Intergenerational high (P5,5) The figure compares estimates of rags to riches, intergenerational low, and intergenerational high measures computed using quintiles based on country-level educational attainment versus communelevel attainment (denoted local). Each uses a sample of individuals of age 21-25 omitting communes with less than 50 individuals. For details about the indicators see Table 2. 44 Figure A12: Correlates of the IGM at the commune-level (all the indicators) (a) Unconditional (b) Conditional on education of old cohorts 45 Figure A13: Intergenerational mobility in education vs. income inequality .7 .72 .74 .76 .78 Relative IGM in education .25 .3 .35 .4 .45 Gini coef. for individuals ages 18-60 (labor earnings) Notes: The figure shows a binscatter plot between relative IGM (measured as one minus the regression coefficient between child’s years of schooling against parents’ years of schooling) and the Gini coefficient computed using labor earnings in 2010 of individuals ages 18-60. Educational attainment is censored at 15 and the sample includes individuals with age between 21 and 25. Communes with less than 50 observations are not included. 46 Figure A14: Intergenerational mobility in education vs. inequality in education Pearson's correlation coefficient: -.123 .64 .66 .68 .7 .72 Relative IGM in education 3.2 3.4 3.6 3.8 4 4.2 Standard deviation of years of schooling (age 40-60) Notes: The figure shows a binscatter plot between relative IGM (measured as one minus the Pearson correlation coefficient between child’s years of schooling against parents’ years of schooling) and the standard deviation of years of schooling computed using individuals ages 40-60 that are used as parents. Educational attainment is censored at 15. Communes with less than 50 observations are not included. 47 Table A4: Region-level estimates of IGM Statistics Region P1,5Plocal 1,5P1,1Plocal 1,1P5,5Plocal 5,5 Tarapac´a 0.10 0.11 0.37 0.33 0.32 0.31 Antofagasta 0.08 0.10 0.40 0.36 0.31 0.33 Atacama 0.07 0.10 0.40 0.34 0.29 0.36 Coquimbo 0.07 0.08 0.38 0.34 0.32 0.31 Valpara´ıso 0.09 0.10 0.35 0.36 0.34 0.33 Libertador General Bernardo O’Higgins 0.10 0.10 0.34 0.34 0.31 0.31 Maule 0.09 0.09 0.36 0.37 0.32 0.31 Biob´ıo 0.11 0.09 0.33 0.36 0.37 0.32 Araucan´ıa 0.07 0.07 0.38 0.39 0.37 0.36 Los Lagos 0.07 0.08 0.41 0.38 0.31 0.32 Ays´en del General Carlos Ib´a˜nez del Campo 0.05 0.09 0.44 0.35 0.23 0.32 Magallanes y de la Ant´artica Chilena 0.10 0.10 0.30 0.31 0.30 0.33 Metropolitana de Santiago 0.09 0.09 0.37 0.36 0.36 0.36 Los R´ıos 0.06 0.06 0.38 0.37 0.34 0.35 Arica y Parinacota 0.14 0.14 0.28 0.30 0.31 0.32 ˜ Nuble 0.11 0.09 0.33 0.35 0.36 0.27 The table reports region-level estimates of rags to riches, intergenerational low, and intergenerational high (a description of the measures can be found in Table 2). It compares measures that assign individuals into quintiles using the distribution of educational attainment at the country level with measures that use the distribution of each region (those with the superscript “local”). 48 Table A5: Correlates of intergenerational mobility. All the indicators. Relative mobility (1 −β) Absolute mobility (α) Average education (¯ Y) Relative mobility (1 −ρ) Above parents (¯y≥) Rags to riches P(1,5) Intergen. low P(1,1) Intergen. high P(5,5) Gini index ⊗ ⊗ × ⊗ ⊗ ◦ Avg. labor earnings × × × × × 10th quantile ⊗ ⊗ × ◦ 50th quantile × × × × × 90th quantile ⊗ ⊗ × ⊗ ⊗ ⊗ × 95th quantile ⊗ ⊗ × ⊗ ⊗ ⊗ × 90th over 10th ⊗ ⊗ × ⊗ × ◦ × 90th over 50th ⊗ ⊗ × ⊗ ⊗ 50th over 10th × ⊗ ⊗ × ◦ Area × ⊗ ⊗ Dist. to reg. capital × ◦ ⊗ ⊗ Density × ⊗ × Population × ⊗ × × ◦ Municipal quality (prof) Personnel female share ⊗ Felonies × ⊗ ◦ Budget × ⊗ ⊗ × × Total expenditure × ⊗ × Social expenditure ⊗ ⊗ Educational expenditure × ⊗ Students in public schools × ⊗ × × ⊗ ◦ Students per teacher ⊗ ⊗ ⊗ ⊗ × × Secondary test scores × × × × ⊗ ⊗ Primary test scores ⊗ ◦ × Nurses × × × Doctors ⊗ ⊗ ⊗ × Infant mortality Water network ⊗ ⊗ × Notes: The table reports the statistical significance (at the 5% level) of the coefficients from regressions with an indicator of intergenerational mobility at the commune level (standardized to have mean 0 and variance 1) as the dependent variable and a given correlate (standardized to have mean 0 and variance 1) as the independent variable (labeled as unconditional model), and the coefficients from the same regression controlling for average education of the older cohorts (labeled as conditional model). ×if statistically significant in unconditional model. ◦if statistically significant in conditional model. ⊗if statistically significant in both models. 49 Table A6: LASSO variable selection across IGM indicators Relative mobility (1 −β) Absolute mobility (α) Average education (¯ Y) Relative mobility (1 −ρ) Above parents (¯y≥) Rags to riches P(1,5) Intergen. low P(1,1) Intergen. high P(5,5) Gini index × Avg. labor earnings 10th quantile 50th quantile × × × 90th quantile 95th quantile 90th over 10th × × × × 90th over 50th × 50th over 10th × Area × × × × × × Dist. to reg. capital × × × × Density × × × × Population × Municipal quality (prof) × × × × × Personnel female share × × × × × Felonies × × × Budget × × × × Total expenditure Social expenditure × × × Educational expenditure × Students in public schools × × × × × Students per teacher × × × Secondary test scores × × × Primary test scores × × × Nurses × × × × × Doctors × × × × Infant mortality × × Water network × × × × × Notes: The table presents the variables selected with a LASSO estimation using the optimal value of λ, highlighting the set of correlates that remain nonzero after regularization for each indicator of intergenerational mobility. 50 Figure A15: Intergenerational educational mobility within Chile North Metropolitan Region South Relative mobility 0.54 to 0.69 0.69 to 0.70 0.70 to 0.71 0.71 to 0.72 0.72 to 0.73 0.73 to 0.75 0.75 to 0.76 0.76 to 0.78 0.78 to 0.80 0.80 to 0.97 Missing (a) Relative mobility by commune - Chile, 2017 Notes: The map plots relative IGM measured as one minus the regression coefficient (by commune) between child’s years of schooling (using age between 21 and 25) against parents’ years of schooling. Educational attainment is censored at 15. Communes with less than 50 observations are left as missing. 51