DRAFT FOR DISCUSSION- DO NOT CITE International and Time Comparisons of Inequality in Tertiary Education Béatrice d Hombres and Phuong Nguyen-Hoang February 13, 211 1
Abstract This paper proposes to provide a regional comparison of the level of intergenerational educational mobility in tertiary education. We examine the relationship between participation in tertiary education and family background across countries and over time. Empirical results are drawn from datasets covering 68 countries. The general conclusion is that participation in tertiary education is strongly associated with fathers education. The link between socio-economic background and participation in tertiary education is the highest in Latin America and lowest in Asian countries. We find substantial gender differences in the level of intergenerational educational mobility. Time comparisons show a decline in intergenerational educational mobility in Asia and Latin American while the opposite is found for the group of industrial countries. 2
Table of contents Tables and Figures Introduction... 6 International studies of the role of social origin in determining educational attainment... 7 Survey data for measuring intergenerational mobility... 9 Family background related information... 16 Overview of the method... 18 Summary statistics... 21 Inequality in tertiary education across socioeconomic groups... 25 Relationship between educational outcomes of father and children...25 Macro regional results....25 Country estimates...3 Conclusion... 56 Bibliography... 57 3
Tables and Figures Table 1:European and Central Asia: data source, samples size, 28 countries... 11 Table 2: Latin America countries: data source, samples size, 8 countries.... 13 Table 3: Industrial countries: data source, samples size, 21 countries... 14 Table 4: East and South Asia and Pacific countries: data source samples size, 9 countries.. 16 Table 5: Summary statistics, Industrial countries... 22 Table 6: Summary statistics, Latin America and Caribbean... 23 Table 7: Summary statistics, Europe and Central Asia... 24 Table 8: Summary statistics, East and South Asia... 25 Table 9:Intergenerational correlations in accessing tertiary education... 26 Table 1: Transmission of the educational status from father to child... 27 Table 11: Time comparisons in the relationship between father education and participation in higher education, regional comparisons... 29 Table 12: Time comparisons in the relationship between father education and participation in higher education, regional comparisons,... 29 Table 13: Time comparisons in the relationship between father education and participation in higher education, regional comparisons, and... 28 Table 14: Time comparisons in the relationship between father education and participation in higher education, regional comparisons,... 28 Table 15: Transmission of the educational status from father to child, time comparisons, ID countries... 34 Table 16: Intergenerational correlations in accessing tertiary education, LAC countries... 39 Table 17: Transmission of the educational status from father to child, time comparisons, LAC... 4 Table 18: Intergenerational correlations in accessing tertiary education, ECA countries... 45 Table 19: Transmission of the educational status from father to child, time comparisons, ECA countries... 47 Table 2 : Intergenerational correlations in accessing tertiary education East and South Asia... 52 Table 21: Transmission of the educational status from father to child, time comparisons, ESA countries... 54 Table 22 : Access to post secondary studies, country ranking based on... 57 Table 23: Participation in post secondary education for a set of EU countries... 63 Table 24: Participation in post secondary education, conditional on the famiy background. Panama and Guatemala... 63 4
Figure 1: Time comparisons in the relationship between father education and participation in higher education,... 29 Figure 2: Time comparisons in the relationship between father education and participation in higher education, and... 3 Figure 3: Gender differences in the intergenerational transmission of education ID countries... 33 Figure 4: Time comparisons in the relationship between father education and participation in tertiary education, ID countries... 36 Figure 5: Time comparisons in the relationship between father education and participation in higher education, gender differences, ID countries... 37 Figure 6: Gender differences in the intergenerational transmission of education, LAC... 39 Figure 7: Time comparisons in the relationship between father education and participation in higher education, LAC countries... 41 Figure 8: Time comparisons in the relationship between father education and participation in higher education, gender differences, LAC countries... 42 Figure 9: Gender differences in the intergenerational transmission of education, ECA countries... 44 Figure 1: Time comparisons in the relationship between father education and participation in tertiary education,eca countries... 48 Figure 11: Time comparisons in the relationship between father education and participation in higher education, gender differences, ECA countries... 49 Figure 12: Gender differences in the intergenerational transmission of education, ESA... 51 Figure 13: Time comparisons in the relationship between father education and participation in higher education, ESA countries... 53 Figure 14: Time comparisons in the relationship between father education and participation in tertiary education, gender differences, ESA..56 Figure 15: Relationship between father education and participation in tertiary education, full sample... 65 5
1) Introduction Intergenerational mobility is defined as the ability of an individual to access and participate in higher education regardless of his or her economic or social background. By contrast, intergenerational immobility occurs when people from less advantaged backgrounds are unable to access high skilled and high paid jobs because of lack of access to education. Equal access to education based on individuals effort and not on factors unrelated to productivity (family background, gender, minority groups, etc) is thus largely recognized as socially and economically desirable. The expansion of tertiary education experienced over the last decades in many countries (Schofer and Meyer 25) does not necessarily mean that tertiary education systems are equally accessible to all social groups. Increasing tertiary education rates can indeed result from the participation of a greater proportion of students from families with a relatively high socio-economic status. This paper aims at shedding some light on this equity aspect by examining how family background shapes access to tertiary education. There have been few cross-country studies on the level of intergenerational transmission of education. Most of the papers on intergenerational mobility in education are based on a single country. Notable exceptions include Behrman et al. (21) and Hertz et al. (27) although none of these studies focuses on access to tertiary education. Chevalier et al (23) as well as a recent study by Koucky et al. (21) examines the relationship between access to tertiary education and family background but only covers industrial countries. This present paper is intended to fill this gap in the literature by presenting a comparative measurement of the extent of intergenerational mobility in tertiary education. More specifically, based on the compilation of large datasets providing information on individuals family background, we examine the variation in the association between the likelihood of participating in tertiary education and parents schooling attainment in 68 countries. We also explore the evolution of intergenerational mobility through education across time and by gender. We use regression analysis to make inferences about the importance of family socioeconomic status for access to tertiary education. Our results show a strong and significant relationship between socio economic background and participation in tertiary education. Furthermore, we observe unambiguously that intergenerational mobility in tertiary education is the highest in Latin America and the lowest in Asian 6
countries. We also find substantial gender differences in the level of intergenerational educational mobility. In South and East Asia as well as in Latin America, the probability to participate in tertiary education is more strongly associated with the family background for males than for females. The opposite is observed in Europe and Central Asia whereas industrial countries do not display gender differences in terms of relationship to socio economic background and participation in tertiary education. Time comparisons show a decline in intergenerational educational mobility in Asia and Latin American; the opposite is found for the group of industrial countries. Note that in this study, we look at the correlation and not the causality - between family origin and access to tertiary education studies. Our work does not allow us to specify the reasons underlying the observed association. The paper proceeds as follows: Section 2 briefly reviews cross country studies on intergenerational mobility in education. Section 3 describes the data required for cross country comparisons of intergenerational mobility in attaining tertiary education while Section 4 discusses the proxy used for measuring the family background across a large number of countries and surveys. Section 5 presents the empirical strategy employed for analyzing inequalities in access to tertiary education. Section 6 presents some summary statistics and section 7 discusses the empirical results. 2) The role of social origin in determining educational attainment (global perspective) International comparisons of educational mobility are scarce. Only few recent papers (Koucky et al. (21), Chevalier et al. (23), Hertz et al., (27), Behrman et al. (21)) compare across countries the level of intergenerational educational mobility by estimating the extent to which family background determines schooling attainment. Estimates of intergenerational educational mobility in Europe and for industrial countries can be found in Chevalier et al. (23) and Koucky et al. (21). Chevalier et al. (23) examine the relationship between access to tertiary studies and paternal education in 2 countries, among them 15 are European countries. 1 The analysis is based on the International Adult Literacy Survey (IALS), a survey in which countries participated in different collection cycles 1 The countries covered are Finland, Northern Ireland, New Zealand, Denmark, Great Britain, United States, Canada, Czech Republic, Sweden, Poland, Chile, Ireland, Italy, Norway, Hungary, Switzerland, Belgium, Germany, and Slovenia. 7
over the period 1994-1998. 2 The results suggest wide differences across nations in the degree of educational mobility with Czech Republic and Canada displaying respectively the highest and lowest levels of educational mobility. [Time variations in the effect of paternal education on the probability of attending tertiary studies are very country specific.] NOT SURE WHAT THIS MEANS However, in Central and Easter European countries, on average educational mobilityhas decreased while the opposite happens in European Nordic countries. In this case, educational mobility was measured as the probability of attaining tertiary education based on parental education. Recently, Koucky et al. (21) analysed the level of inequality present in accessing tertiary education in 23 European countries 3 using the European social survey (waves 22-24, 26 and 28). Large variations across countries in the level of inequality in access to tertiary education are also reported, with, for instance, Hungary displaying very high levels of inequality and Denmark and Sweden low levels with respect to the EU average. As in Chevalier et al. (23), access to higher education has been increasingly linked to social origins in Eastern Europe while South and North Western Europe show a decreasing trend over the last 5 years. Comparisons of educational mobility in Latin America and Caribbean countries can be found in Behrman et al. (21). In this paper, the authors study the intergenerational transmission of education and occupation in four Latin American countries - Brazil, Colombia, Peru, and Mexico - and in the United States. They observe that the association between parental education and the educational attainment of the children is the highest in Brazil and Colombia. The educational mobility is far higher in the United States than in Latin American and Caribbean countries. In addition, women are found to exhibit higher educational mobility in Brazil and Colombia while the opposite is observed in Mexico and Peru. A comparative assessment of educational mobility in Africa can be found in Bossuroy et al. (27). In this study, the authors estimate the level of occupational and educational intergenerational mobility in five African countries: Cote d Ivoire, Ghana, Guinea, Madagascar and Uganda. Their findings suggest that Ghana and Uganda display lower levels of intergenerational persistence in education than Cote d Ivoire, Guinea and 2 See also de Broucker and Underwood (1998). 3 The countries covered are Austria, Belgium, Czech Republic, Denmark, Estonia, Finland, France, Germany, Greece, Hungary, Ireland, Netherlands, Norway, Poland, Portugal, Romania, Russia, Slovak republic, Slovenia, Spain, Sweden, Switzerland, Turkey, Ukraine, and United Kingdom. 8
Madagascar. Note that Bossuroy et al. s study is based on surveys undertaken between the mid-eighties and the beginning of the nineties. Finally, Hertz et al. (27) examine for a sample of 42 economies the intergenerational persistence of education. The sample includes Latin American, African, Asian and Western European nations.4 Hertz et al. (27) show that the intergenerational educational mobility over the past 5 years has been particularly low in Latin America and high in the Nordic countries while Asian and transition countries are situated in between. Our contribution to this literature is threefold. First, we cover many more countries than the studies mentioned above. While Hertz et al. (27) has constituted so far the most exhaustive study on the issue, the work presented in this paper covers 26 more countries than Hertz et al. (27). Second, when possible, for each of the countries included in this study, we rely on several household surveys in order to boost country-cohort sample sizes (see section 3 for additional information). As a result, the number of observations per country and cohort is on average much higher than in the papers mentioned before. The greater the number of observations the more precise the country-cohort estimates of the correlation between family background variables and the chance of acceding in higher education. Third, while Hertz et al. (27) have undertaken cross-country comparisons of the association between educational attainment and parental education, we focus more specifically on access to tertiary education. 3) Survey data for measuring intergenerational mobility Studying intergenerational mobility requires datasets with information on the educational attainment of both parents and children. Most existing household surveys (such as Multiple Indicator Cluster Surveys, Demographic and Health Surveys, Household Budget and Nutrition Surveys, Literacy surveys, Population Census, demography surveys, Household Expenditure and Consumption Surveys, Household Budget Surveys, Population census, etc) gather information on the educational histories of household members that are living together at the time of interview. 4 More precisely, the countries covered are Peru, Ecuador, Panama, Chile, Brazil, Colombia, Nicaragua, Indonesia, Italy, Slovenia, Egypt, Hungary, Sri Lanka, Pakistan, USA, Switzerland, Ireland, South Africa, Poland, Vietnam, Philippines, Belgium, Estonia, Sweden, Ghana, Ukraine, East Timor, Bangladesh, Slovakia, Czech Republic, Netherlands, Norway, Nepal, New Zealand, Finland, Northern Ireland, Great Britain, Malaysia, Denmark, Kyrgyzstan, China, Ethiopia. 9
However, reliance on such household surveys for the study of intergenerational transmission of education limits the scope of the study to those students who are still dependent on their parents for housing In other words, young adults who have left their parental household will not be covered in the sample. While this problem is a non-issue for studies on intergenerational mobility in primary or secondary education, it may be more problematic for studies focusing on the intergenerational transmission of tertiary education. There is potentially a sample selection bias given that young adults are likely to leave their parental home in ways that are related to their social origins. Furthermore, this potential sample selection problem is probably not identical across countries (see BOX 1 in appendix). As a result, cross-country comparisons could be flawed. To compare the distribution of access to tertiary education across countries, we need to rely on surveys providing socio economic information for two or more generations of the same family. After a systematic search, we were able to find comparable data on family background characteristics, and hence comparable information on intergenerational mobility, for 68 countries. More precisely, we cover eight Latin American and Caribbean countries (Brazil, Chile, Colombia, Ecuador, Guatemala, Mexico, Panama and Peru), 28 countries from European and Central Asia (Albania, Armenia, Azerbaijan, Belarus, Bosnia, Bulgaria, Croatia, Czech Republic, Estonia, Georgia, Hungary, Kosovo, Kyrgyzstan, Latvia, Lithuania, Moldova, Montenegro, Poland, Romania, Russia, Serbia, Slovakia, Slovenia, Tajikistan, Turkey, Ukraine, Uzbekistan), 9 countries from East and South Asia (China, Indonesia, Japan, Mongolia, Nepal, Philippine, South Korea, Taiwan and Timor Lest), 23 Industrial countries (Austria, Belgium, Canada, Cyprus, Denmark, Finland, France, Germany, Greece Iceland, Ireland, Italy, Japan, Luxembourg, Netherlands, New Zealand, Norway, Portugal, Spain, Sweden, Switzerland, United Kingdom and United States). Unfortunately we do not include African and Middle Eastern countries in our sample. This is because we were unable to find enough recent micro data that was representative of the whole country. Tables 3-6 present the data sources and country sample sizes for each of the countries included in this study. Table 1: European and Central Asia : data source, samples size, 28 countries Country Survey Year Albania Life in Transition Survey 26 Albania Living Standards Mesurement Surveys 25 European Values Study 28 Sample size after sample selection 829 1
Armenia Life in Transition Survey 26 164 European Values Study 28 Azerbaijan Life in Transition Survey 26 1816 European Values Study 28 Belarus Life in Transition Survey 26 1599 European Values Study 28 Bulgaria Life in Transition Survey 26 European Values Study 28 3946 European Social Survey 2-28 Bosnia Life in Transition Survey 26 1763 European Values Study 28 Croatia Life in Transition Survey 26 3493 Adult Education Survey 28 Czech Republic Life in Transition Survey 26 European Values Study 28 8829 European Social Survey 2-28 European Survey on Living Conditions 25 Estonia European Survey on Living Conditions 25 European Social Survey 2-28 European Values Study 28 11644 Life in Transition Survey 26 Living conditions survey of the Baltic region 1999 Georgia European Values Study 28 1665 Life in Transition Survey 26 Hungary European Survey on Living Conditions 25 European Social Survey 2-28 13881 European Values Study 28 Life in Transition Survey 26 Kazakhstan Life in Transition Survey 26 74 Kazakhstan Living Standards Mesurement Surveys 1996 Kosovo European Values Study 28 17 Kyrgyzstan Life in Transition Survey 26 474 Kyrgyzstan Poverty Monitoring Surveys 1997 Table 1 continuation Latvia European Survey on Living Conditions 25 Living conditions survey of the Baltic region 1999 European Values Study 28 6473 Life in Transition Survey 26 Lithuania European Survey on Living Conditions 25 7926 11
Living conditions survey of the Baltic region 1999 European Values Study 28 Life in Transition Survey 26 Moldovia European Values Study 28 Life in Transition Survey 26 Montenegro European Values Study 28 Life in Transition Survey 26 Poland European Survey on Living Conditions 25 European Social Survey 2-28 European Values Study 28 Life in Transition Survey 26 Romania European Values Study 28 Life in Transition Survey 26 Russia European Values Study 28 Life in Transition Survey 26 European Social Survey 2-28 Slovakia European Survey on Living Conditions 25 European Social Survey 2-28 European Values Study 28 Life in Transition Survey 26 Serbia European Values Study 28 Life in Transition Survey 26 Slovenia European Values Study 28 Life in Transition Survey 26 European Social Survey 2-28 European Survey on Living Conditions 25 Ukraine European Values Study 28 Life in Transition Survey 26 European Social Survey 2-28 Turkey Life in Transition Survey 26 European Social Survey 2-28 Uzbekistan Life in Transition Survey 26 787 Tajikistan Life in Transition Survey 26 TajikistanLiving Standards Mesurement Surveys 27 We were able to pool at least two datasets or two waves of a given dataset for the vast majority of countries. 5 In addition, for all countries but Brazil, Chile and New Zealand, the survey year of at least one of the datasets employed in the analysis falls between 2 and 28. Data for Europe and Central Asia (henceforth ECA) are primarily drawn from the 26 Life in transition Survey, 28 European Values Study, 2-1554 1812 2845 151 4631 12519 169 9627 422 1998 1235 5 All LAC countries but Chile are covered by a single household survey. However, these surveys are very large, as shown in Table 2. 12
28 European Social Survey, 25 European Survey on Living Conditions and 1999 Living conditions survey of the Baltic region. For Albania, Tajikistan and Kyrgyzstan, we also employ Living standards Measurement Surveys undertaken in the end of the nineties. After sample selection, the mean country sample size for countries of this region numbers to 5314 individuals. Note that many countries in the ECA region, such as Moldavia, Uzbekistan, Tajikistan, have never so far been included in crosscountry comparisons of educational mobility. Table 2: Latin America countries: data source, samples size, 8 countries Country Survey Year Sample size after sample selection Colombia Colombia Quality of Life Survey 23 2799 Ecuador Ecuador Living Standards Measurement Surveys 1998, 1999, 26 3343 Guatemela Guatemala Survey on Living Conditions 2 1399 Mexico Mexican Family Life Survey 22, 25 18323 Panama Panama Standards Living Surveys 1997, 23, 28 26123 Chile International Social Survey Programme 1999 294 International Adult Literacy Survey 1999 Peru Peru Living Standards Measurement Survey 21 11361 Brazil Pesquisa Nacional por Amostra de Domicilios 1996 118679 All countries but Chile and Brazil in Latin America and Caribbean area (henceforth LAC) are covered by large household surveys whose at least one wave has taken place after 2. In comparison, the analyses in Behrman et al. (21) relied on 4 LAC countries with surveys implemented between 1985 and 1997. Our analysis for Colombia is based on a Quality of Life survey conducted in 23 while we use the 22 and 25 Family Life Surveys for Mexico. Data for Guatemala, Panama, Ecuador and Peru are drawn from various Living Standards Measurement Surveys (LSMS) between 1997 and 28. The information on Brazil comes from a very large household survey carried out in 1996 while Chilean figures are derived from two cross country surveys (International Social Survey Program and International Adult Literacy Survey) in 1998-1999. In this region, the mean country sample size amounts to 3192 observation. 6 Table 3: Industrial countries: data source, samples size, 21 countries 6 This large figure is largely driven by Brazil. The mean country sample size with Brazil excluded, is equal to 1858 observations. 13
Country Survey Year Austria European Survey on Living Conditions 25 European Social Survey 2-28 European Values Study 28 Belgium European Survey on Living Conditions 25 European Social Survey 2-28 European Values Study 28 Sample size after sample selection 12 11139 Canada Canada General Social Survey 26 12899 Cyprus European Survey on Living Conditions 25 European Social Survey 2-28 European Values Study 28 Denmark European Survey on Living Conditions 25 European Social Survey 2-28 European Values Study 28 Finland European Survey on Living Conditions 25 European Social Survey 2-28 European Values Study 28 France European Survey on Living Conditions 25 European Social Survey 2-28 European Values Study 28 Germany European Survey on Living Conditions 25 European Social Survey 2-28 European Values Study 28 Greece European Survey on Living Conditions 25 European Social Survey 2-28 Table 3 continuation European Values Study 28 Ireland European Survey on Living Conditions 25 European Social Survey 2-28 European Values Study 28 Italy European Survey on Living Conditions 25 European Social Survey 2-28 Luxembourg European Survey on Living Conditions 25 European Social Survey 2-28 European Values Study 28 Netherlands European Survey on Living Conditions 25 European Social Survey 2-28 European Values Study 28 6973 8927 12 584 15129 21824 1187 9467 2927 761 11569 New Zealand International Social Survey Programme 1999 331 14
International Adult Literacy Survey 1999 Norway European Survey on Living Conditions 25 European Social Survey 2-28 Portugal European Survey on Living Conditions 25 European Social Survey 2-28 European Values Study 28 Spain European Survey on Living Conditions 25 European Social Survey 2-28 European Values Study 28 Sweeden European Survey on Living Conditions 25 European Social Survey 2-28 Switzerland European Social Survey 2-28 European Values Study 28 International Adult Literacy Survey 1994, 1998 UK European Survey on Living Conditions 25 European Social Survey 2-28 USA General Social Survey 1996-28 8699 191 23671 8672 8935 11857 15259 Data for industrial countries (henceforth ID countries) come from the 25 European Survey on Living conditions, 28 Adult Education Survey, 2-28 European Social Surveys, 28 European Values Study, 1999 International Social Survey programme and 1999 International Adult Literacy Survey. In addition, we rely on General Social Surveys for USA and Canada. In comparison Kouck et al (21) rely on the 2-28 Europe Social Survey in their analysis of the transmission of higher education across generations in Europe. The sample size of each industrial country is large enough to obtain precise country-cohort estimates of the association between the probability of having accessed tertiary studies and the family background. The mean country sample size is 12377 observations. 7 Table 4: East and South Asia and Pacific countries: data source, samples size, 9 countries Country Survey Year Sample size after sample selection China East Asia Social Survey 26, 28 51 Phillipines International Social Survey Programme 1999 813 Mongolia Life in Transition Survey 26 823 7 We note, however, that we should be careful in the interpretation of the results for New Zealand, because this country is only covered by surveys implemented before 2. 15
Timor-Leste Timor-Leste Living Standards Measurement Survey 21 313 Nepal Nepal Living Standards Survey 24 685 Taiwan East Asia Social Survey 26, 28 5573 South Korea East Asia Social Survey 26, 28 51 Indonesia Indonesia Family Life Survey 27 15545 Japan East Asia Social Survey 26, 28 2879 Information on both educational attainment and family background-related variables for East and South Asia countries (henceforth EAS) were found for most of the countries in the 26 and 28 East Asia Social Surveys, 1999 International Social Survey Programme, 26 Life in Transition Survey. Timor- Leste, Nepal and Indonesia are covered by Living standards Measurement Surveys and Household surveys. Except for the 3 countries just mentioned above, the country sample sizes are much lower than for the countries of the other regions as we usually rely on only one survey per country. In this region, the mean country sample size amounts to 5,87 observations. 4) Family background related information The education level and occupation of one s parents are core components of an individual s family socioeconomic status as explained hereafter. Ideally, we would combine these three measures into a single measure of family background. 8 This is not possible because of underlying data problems with the latter two components. First, while parental income could be a possible measure of the wealth of teenage respondents, it is not usually available unless we focus our analysis on individuals who are still living with their parents. 9 Furthermore, information on income is usually not accurate and income categories are not comparable across countries. Income also fluctuates over time so much so that it constitutes a poor measure of the respondent s long term socioeconomic status (Hertz et al., 27). 8 Note that, of course those 3 dimensions of the family socio economic status do not capture all aspects of the family background in determining access to tertiary studies. For instance, the number of siblings can have significant effects on an individual s educational attainment (Blake 1989). 9 Previous studies (e.g., Sandefur and Wells 1999) (Sandefur and Wells 1999) find that the absence of a parent or parents may have significant effects on educational attainment. 16
Second, parental occupation, is also problematic for cross national comparative purposes. 1 Indeed the measurement of the parental occupational status differs so substantially between surveys that it is impossible to make this measurement comparable across surveys. In some surveys, the variables related to the parental status allow for several occupational categories, in other surveys the number of categories is reduced to a few occupational categories. It is thus impossible to compare across countries/surveys the parental occupational status. Finally, the last available alternative, which we adopt in this study, is to rely on parental education as a measure of family socioeconomic status. Studies have found a strong correlation between education and income (see Psacharopoulos and Patrinos, 24 for a review of studies on this topic). Income is in turn highly correlated with the type of occupation a person does. In other words, parental education constitutes a principal characteristic of the family background. Also, measurements of education are often available and more readily comparable across surveys. In this study, we thus use as a proxy for a respondent s family background the educational level of his/her father. Other proxy candidates are the educational level of the respondent s mother, the highest level of education attained by his/her parents (such as in Behrman et al., 21) or the average of both parents education (Hertz et al., 27). We adopt father s education rather than the other two proxies for two reasons. First, the non-response rate on mother education (and in consequence the number of missing data) is almost always higher in the available surveys than for father s education. Second, several country-specific studies have shown that father s education is a better predictor of a child s educational attainment than the mother s education. However, note, as indicated earlier, that by focusing only on father s education, we necessarily neglect other dimensions of the family background but we gain the ability to compare the transmission of education across a large number of countries. A potential source of bias lies in the fact that parents education has been measured with differing degrees of reliability. Country or/and survey-specific variations in the amount of measurement error might create spurious variations in the coefficients associated with parental education. It is important to bear this in mind while looking at the empirical results. 1 Parental occupation is used as a family background variable in Bossuroy et al. (27) and Kouck et al. (21). 17
5) Overview of the method Country specific estimates of the level of inequality in access to tertiary education Empirical specification We estimate equation (1) intended to quantify the country-cohort-specific level of inequality in access to tertiary education: (1) where, the outcome variable, is a dichotomous variable taking on the value one if the individual i reports having participated in tertiary education and zero otherwise. For each country we have divided the population into 4 cohorts: 1969-1978, 1959-1968, 1949-1958, 1939-1948. These 4 cohorts cover individuals being in age of participating in higher education in 4 different periods. Equation (1) is estimated separately for each country j=1 68 and cohort c=1...4. As mentioned earlier, the measurement of education differs from one survey to another. In most surveys, respondents are asked what is the highest level of education they have participated in. However, in some of the surveys used in this study, the education variable is about degree achievement and not participation. If degree failure varies across socioeconomic groups, this might affect the country estimates of intergenerational correlation of education and we will underestimate the extent of the relationship between social class and higher education achievement. Note that we also include survey specific effects for all country-cohort estimates. As explained earlier, the explaining variables related to the family background of the respondents, and, refer to the educational attainment of the father. is equal to one if the father of the respondent has at maximum completed a primary education and zero otherwise while is equal to one if the father has attained a tertiary education and zero otherwise. The excluded category will thus be composed of all individuals with a lower or upper secondary education. is the disturbance term of equation (1). Equation (1) is estimated using a probit model. We restrict the analysis to adults aged between 25 and 65 at the time of the survey and cohorts born between 1938 and 1978. 18
The marginal effects to be estimated, and, describe the influence of the family background on the respondent s educational achievement for country j and cohort c. is an estimate of the increased probability for individuals whose father has a tertiary education to enrol in tertiary studies compared to individuals whose father has only completed a secondary education (henceforth the reference category). Similarly, is the penalty, with respect to the reference group, in terms of the likelihood of having access to tertiary studies for individuals with a low family background. BEA PLEASE EXPLAIN Country and macro regional estimates The estimated value of for a given country j is given by, i.e., it is the average value of the estimated cohort-coefficients. Similarly,. We follow Hertz et al. (27) approach in that we give the same weight to each cohort so that and are not sensitive to the size of the cohorts.11 This makes country comparisons more reliable as country results will not be affected by differences in the age structure in the various surveys. Based on these country cohort specific intergenerational education regressions, we also compute for the 4 sets of countries considered in the study Industrial Countries, Europe and Central Asia, South and East Asia, Latin America and Caribbean - the macro regional estimate of and which are given by and with the number of countries belonging to the macro-region. A global measure of inequality of opportunity for each country is given by the sum of the absolute values of two estimated coefficients: macro regional measure of inequality of opportunity is equal to High estimates of.. The are interpreted as limited intergenerational mobility, while low estimates of suggest that schooling outcomes are not closely related across generations. 12 11 An alternative solution would be to estimate equation (1) for each country on the full country sample. In that case, 68 coefficients would be estimated (and similarly for ) but larger cohorts would be given higher weight. 12 Kouck et al. (21) also estimate a discrete model to quantify the level of inequality in tertiary education. Parental education and parental occupation related variables are the explanatory variables used to quantify by how much the probability of achieving a tertiary education is related to family background. The authors of the study use the ROC curve (Receiver Operating Characteristics) in order to come up with a single estimated measure of country inequity. 19
In addition to considering the population as a whole, we also examine how the intergenerational correlations in education vary by sex. For this, equation (1) is simply estimated separately by country, cohort and sex. Time comparisons We compute linear time trends in intergenerational mobility by regressing for each country respectively and on a variable taking on the values 1 to 4, with the value 1 for the oldest cohort to attend tertiary and 4 for the youngest one.. Regional linear trends are obtained by relying on similar estimates but while pooling together all countries belonging to the region R and adding country fixed effects in the equations. The coefficients associated with the variable, and, show the absolute change in and occurring from one cohort to the next. If and are positive (negative), the correlation between the probability to participate in postsecondary studies and paternal education has increased (decreased) over time. We also examine how mobility in tertiary education has evolved in relative terms by measuring the rate of change across each cohort as follows:,. In addition, in order to detect non linear changes across cohorts in the estimated values of and, we also plot the estimated coefficients and against. Ranking Countries are ranked in three groups according to the degree of educational persistence across generations as measured by. The best performing countries in terms of educational mobility lie between 1th percentile and 75th percentile. The worst performing countries lie in the bottom 25th percentile. The rest of the countries are assigned to the middle group. 1 th percentile Best performers in educational mobility 75 th percentile Middle group 25 th percentile Worst performers in educational mobility Before turning to the empirical results, remember that many other factors, such as genetic endowments, affect the value of. Our objective is to estimate the extent of intergenerational mobility in education across countries regardless 2
of its underlying causes. The regression coefficients discussed in the result section constitute thus descriptive measures of the intergenerational correlations in accessing tertiary education. 6) Summary statistics Access to tertiary education: synopsis Tables 7-1 provide the summary statistics of the surveys we use in this study. The proportion of individuals having reached tertiary education amounts to around 37% for both the group of ID and ECA countries. Much lower participation rates are observed in the two other regions where the percentage of respondents having reached tertiary education numbers of less than 2%. Table 5: Summary Statistics, Industrial Countries Proportion having participated in tertiary studies Full Sample Respondents Fathers Female Male Oldest Cohort Youngest Cohort Austria.237.73.216.268.2.28 Belgium.44.25.416.389.295.532 Canada.768.334.758.741.651.844 Cyrpus.34.85.269.297.182.425 Denmark.386.236.433.351.36.462 Finland.383.18.437.343.296.473 France.396.134.367.35.263.516 Germany.433.323.392.496.383.461 Greece.251.92.211.245.125.322 Ireland.397.12.368.376.234.526 Italy.185.34.175.168.92.256 Luxembourg.35.166.247.332.215.46 Netherlands.397.182.34.426.37.439 NewZealand.357.189.337.383.314.358 Norway.448.346.43.44.33.537 Portugal.152.38.143.19.78.197 Spain.273.89.248.26.14.384 Sweeden.428.184.456.363.335.512 Switzerland.37.23.234.43.255.357 UK.459.288.421.456.326.527 USA.377.466.374.386.358.393 Canada (76%) and Portugal (15%) have the highest and lowest proportions of people with a tertiary degree among the ID countries. The highest proportion of people who participated in post-secondary studies are observed in Croatia (76.8%) Colombia (25.%), the Philippines (32.2%) in ECA, LAC, and EAS 21
while Turkey (7.9%), Guatemala (5.6%), Timor-Leste (2.%) post the lowest proportions of college participants in these four regions, respectively. Access to tertiary studies: variations over time Not surprisingly, we observe that children from ID countries are more likely than their father to participate in postsecondary education. Similar figures are observed in the other regions. Respondents from ECA, ID, LAC and EAS countries are on average 1.7, 2.3, 3.6 and 6.2 times more likely than their father to participate in tertiary education. Results in ESA are largely driven by Timor- Leste. If this country is excluded, the corresponding figure for this region is around 3. The relatively limited progress in tertiary education rates in the ECA region might be in part explained by historical factors. Bosnia, Croatia, Montenegro, were Serbia were involved in their wars to fight for their independence of Yugoslavia between 1991 and 1995. Most of the countries of this regions also experienced the cost short terms effects of the transition to a market economy. Table 6: Summary Statistics, Latin America and Caribbean Proportion having participated in post secondary studies Full sample Respondent Father Female Male Oldest Cohort Youngest Cohort Brazil.16.2.19.11.89.99 Chile.182.61.179.186.13.259 Colombia.25.81.238.265.192.27 Ecuador.183.41.162.193.92.25 Guatemala.56.15.4.73.33.57 Mexico.95.3.66.119.5.14 Panama.179.48.178.152.97.24 Peru.184.38.173.19.121.197 Access to tertiary studies: gender gap In the LAC and ESA areas as well in the group of ID countries, men are, on average, more likely to reach tertiary education than their female counterparts. The gender gap amounts to 1.4 percentage points for the set of ID countries and to 1.5 and 4 percentage points respectively in LAC and ESA regions. Those figures hide substantial differences across countries within each region: in Germany, 49.6% of males have participated in postsecondary studies while this is the case for only 39.2 % of females. High gender disparities are also found in Switzerland where male respondents are 72.2% more likely to reach tertiary studies than their female counterparts. On the other hand, in Spain, UK or Italy, males and females have nearly the same chance of accessingtertiary 22
studies.with regard to LAC countries, men are 8% Mexico) and 82% (Guatemala) more likely to have accessed tertiary studies than women. Gender disparities are low in Colombia and Chile while in Panama and Brazil, it is more probable for females to participate in higher education than it is for males. The participation in tertiary education in ESA countries (e.g., Nepal, Timor- Leste, Pakistan, China, and even South Korea) is really low both for men and women. However, in relative terms, sizeable gender differences are found in Nepal, Timor-Leste, Japan and China. Government policies or gender stereotypes in these countries could explain gender disparities. For instance, in China where son preference is still prevalent, females are discriminated against and not provided with equal educational opportunities by their parents (Croll 2; Wang 25). Table 7: Summary Statistics, Europe and Central Asia Proportion having participated in post secondary studies Full sample Respondent Father Female Male Oldest Cohort Youngest Cohort Albania.126.49.82.139.111.19 Armenia.4.236.374.423.44.365 Azerbaijan.448.427.444.438.444.467 Belarus.525.318.426.575.366.595 Bosnia.324.212.19.419.293.315 Bulgaria.288.131.292.28.229.35 Croatia.768.156.661.798.624.85 Czech_Republic.2.143.15.22.141.21 Estonia.428.212.493.372.396.42 Georgia.54.368.491.55.62.524 Hungary.23.129.183.189.167.21 Kazakhstan.592.441.65.57.625.612 Kosovo.211.9.131.245.14.236 Kyrgyzstan.415.236.366.371.347.366 Latvia.398.178.441.333.355.398 Lithuania.533.23.576.465.45.497 Moldova.391.161.286.448.439.355 Montenegro.412.289.293.491.45.382 Poland.24.68.229.167.151.295 Romania.261.116.162.316.257.25 Russia.6.326.595.551.512.623 Serbia.49.273.334.483.432.397 Slovakia.193.99.17.188.148.22 Slovenia.229.114.237.22.168.28 Tajikistan.276.3.25.325.373.248 Turkey.79.18.45.79.5.91 Ukraine.637.355.637.64.536.669 23
Gender stereotypes are also common in East Asian countries. People have preconceived ideas or beliefs about women s physical characteristics, personality traits, abilities and roles in society. There is still a negative connotation associated with tertiary education as Asian parents fear that too much education will prevent their daughters from marriage (Mak 27). Gender differences in the ECA area contrast with what was found in ESA. Indeed, in 1 out of the 27 ECA countries, we find that women are more likely than men to participate in tertiary education. The largest gender gap is observed in Estonia where 49.3% of females have participated in tertiary studies while this figure only amounts to 37.2% for males. Noticeable differences in favor of females are also observed in Lithuania and Latvia. Table 8: Summary Statistics, East and South Asia Proportion having participated in post secondary studies Full sample Respondents Father Female Male Oldest Cohort Youngest Cohort China.53.1.21.38.15.59 Indonesia.138.28.111.15.95.149 Japan.31.184.24.388.216.36 Mongolia.51.335.486.57.551.455 Nepal.21.9.8.5.16.23 Philippine.32.133.319.321.271.385 SouthKorea.53.1.21.38.15.59 Taiwan.251.81.214.246.342.212 TimorLeste.2.1.9.32.5.51 24
7) Inequality in tertiary education across socioeconomic groups The results presented in this section discuss the estimated coefficients and., our measure of overall inequality in accessing post-secondary studies, combines both the penalty in terms of participation in higher education for having a father with low education and the premium for having a father with a high educational level. The reference category is composed of individuals whose father has participated in secondary education (lower or upper, completed or not). We use interchangeably terms such as respondents with a low (high) socioeconomic family background, respondents whose father has a low (high) educational level, or disadvantaged (privileged) respondents. These terms refer to individuals whose father has at best completed primary education (whose father has at least participated in some form of post-secondary education). We also employ the terms participate, enroll and access to tertiary education interchangeably. The macro regional comparisons (Industrial countries, Europe and Central Asia, East and South Asia, Latin America and Caribbean) are based on the regional averages of country specific estimates of and. As a result, (i) macro regional estimates are sensitive to the number of countries in each region covered by our study, and (ii) within each region, an equal weight is assigned to each country independently of its population size. A) Relationship between educational outcomes of father and children A.1) Macro regional results. Transmission of the educational status from father to child, snapshot Table 9 reports the average relationship between father s education and participation in tertiary education for the four macro regional groups under analysis. Table 9: Intergenerational correlations in accessing tertiary education ID -,167,285,452 LAC -,276,228,54 ECA -,118,246,378 25
ESA -,14,214,318 ID: Industrial country (23 countries), ECA: Europe and Central Asia (28 countries) ESA: East and South Asia (9 countries), LAC: Latin America and Caribbean (8 countries) The results suggest that an individual s probability of accessing tertiary education depends significantly on the educational status of his/her father. In addition, the association between family origin and access to tertiary studies varies significantly across regions. We observe that LAC countries display the highest level of inequity while ESA countries, followed by ECA countries, show the lowest figures. In the LAC region, individuals whose father has reached tertiary studies are 22.8 percentage points more likely to participate in higher education than their peers whose father has a secondary education. Similarly, individuals with the lowest family background are 27.6 percent less likely to reach tertiary studies than the reference group. In ESA, children coming from a family with a low social status have 11.8 percentage points less chances to enroll in tertiary studies as compared to the reference group and the comparative advantage for coming from a family with a high social status family takes the form of an additional 21.4 percentage points chance of enrolling in tertiary studies. Table 1: Transmission of the educational status from father to child, gender differences Females ID -,157,281,443 LAC -,231,241,473 ECA -,129,27,46 ESA -,15,168,282 Males ID -,176,293,469 LAC -,315,259,574 ECA -,11,221,322 ESA -,126,247,373 ID: Industrial country (23 countries), ECA: Europe and Central Asia (28 countries) ESA: East and South Asia (9 countries), LAC: Latin America and Caribbean (8 countries) For the set of ID countries, having a father with a high level of education increases more than in any other region the probability of participating in tertiary education (point estimate:.285). However, the disadvantage of having a father with a low educational status is much lower than in the LAC region (point estimates in absolute values:.167 for ID countries versus.276 for LAC countries). 26