Do borrowing constraints matter for intergenerational educational mobility? Evidence from Japan

Niimi, Yoko Working Paper Do borrowing constraints matter for intergenerational educational mobility? Evidence from Japan ADBI Working Paper, No. 830 Provided in Cooperation with: Asian Development Bank Institute (ADBI), Tokyo Suggested Citation: Niimi, Yoko (2018) : Do borrowing constraints matter for intergenerational educational mobility? Evidence from Japan, ADBI Working Paper, No. 830, Asian Development Bank Institute (ADBI), Tokyo This Version is available at: https://hdl.handle.net/10419/190251 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-nc-nd/3.0/igo/ ADBI Working Paper Series DO BORROWING CONSTRAINTS MATTER FOR INTERGENERATIONAL EDUCATIONAL MOBILITY? EVIDENCE FROM JAPAN Yoko Niimi No. 830 April 2018 Asian Development Bank Institute The Working Paper series is a continuation of the formerly named Discussion Paper series; the numbering of the papers continued without interruption or change. ADBI’s working papers reflect initial ideas on a topic and are posted online for discussion. Some working papers may develop into other forms of publication. In this report, “$” refers to US dollars. Suggested citation: Niimi, Y. 2018. Do Borrowing Constraints Matter for Intergenerational Educational Mobility? Evidence from Japan. ADBI Working Paper 830. Tokyo: Asian Development Bank Institute. Available: https://www.adb.org/publications/do-borrowing-constraints-matter-intergenerational-educationalmobility-japan Please contact the authors for information about this paper. Email: [email protected] The empirical work undertaken in this paper utilizes microdata from the Preference Parameters Study of Osaka University’s 21st Century Center of Excellence (COE) Program, “Behavioral Macrodynamics Based on Surveys and Experiments,” and its Global COE Project, “Human Behavior and Socioeconomic Dynamics.” I acknowledge the program/project’s contributors: Yoshiro Tsutsui, Fumio Ohtake, and Shinsuke Ikeda. I am also grateful to Erbiao Dai, Isaac Ehrlich, Tomoki Fujii, Tatsuo Hatta, Joel Hellier, Jean Hindriks, Charles Yuji Horioka, Jong-Wha Lee, Xiaonan Sun, Keiko Tamada, Ryuichi Tanaka, Anh Tran, Tien Manh Vu, and other participants of the Asian Development Bank Institute–Asian Growth Research Institute (AGI) Workshop on Public and Private Investment in Human Capital and Intergenerational Transfers in Asia, the Macro Seminar at the Department of Economics, Korea University, the Workshop on “Equity in Education” at the Catholic University of Leuven, the Rokko Forum at the Graduate School of Economics, Kobe University, and the AGI Staff Seminar for their valuable comments. This work was financially supported by the Japan Society for the Promotion of Science (JSPS) KAKENHI (Grants-in-Aid for Scientific Research) Grant Number 15H01950, a project grant from AGI, and a grant from the Ministry of Education, Culture, Sports, and Science and Technology (MEXT) Joint Usage/Research Center at the Institute of Social and Economic Research, Osaka University. Yoko Niimi is an associate research professor at the Asian Growth Research Institute and Kyushu University. The views expressed in this paper are the views of the author and do not necessarily reflect the views or policies of ADBI, ADB, its Board of Directors, or the governments they represent. ADBI does not guarantee the accuracy of the data included in this paper and accepts no responsibility for any consequences of their use. Terminology used may not necessarily be consistent with ADB official terms. Working papers are subject to formal revision and correction before they are finalized and considered published. Asian Development Bank Institute Kasumigaseki Building, 8th Floor 3-2-5 Kasumigaseki, Chiyoda-ku Tokyo 100-6008, Japan Tel: +81-3-3593-5500 Fax: +81-3-3593-5571 URL: www.adbi.org E-mail: [email protected]rg © 2018 Asian Development Bank Institute ADBI Working Paper 2018 Y. Niimi Abstract This paper examines the intergenerational transmission of educational attainment using data on Japan. By exploiting unique information on whether children have ever given up schooling for financial reasons and, if they have, which level of schooling they have forgone, it attempts to assess the role of borrowing constraints in determining intergenerational educational mobility in a more direct manner than previous attempts made in the literature. We find that there has been a steady increase in the extent of the intergenerational transmission of educational attainment, resulting in lower intergenerational mobility, during the postwar period in Japan. We also find that while the importance of borrowing constraints for determining intergenerational educational mobility declined at one time, it seems to have become significant enough once again to lower intergenerational educational mobility for the youngest cohort we examined in this paper. However, our analysis also shows that the relative importance of adolescent academic ability for children’s educational attainment has increased in recent years, thereby underlining the increasing importance of early investments in children’s human capital for their subsequent academic advancement. Keywords: borrowing constraints, education, intergenerational mobility, Japan JEL Classification: I24, J62 ADBI Working Paper 2018 Y. Niimi Contents 1. INTRODUCTION ......................................................................................................... 1 2. CONCEPTUAL FRAMEWORK ................................................................................... 3 3. RELATED LITERATURE ON BORROWING CONSTRAINTS ................................... 4 4. EDUCATION SYSTEM IN JAPAN .............................................................................. 8 5. DATA AND ESTIMATION METHODS ...................................................................... 11 5.1 Data ............................................................................................................... 11 5.2 Estimation Methods ....................................................................................... 12 6. EMPIRICAL RESULTS.............................................................................................. 14 6.1 Descriptive Statistics ..................................................................................... 14 6.2 Mobility Indices .............................................................................................. 16 6.3 Regression Results ....................................................................................... 18 6.4 Discussion ..................................................................................................... 23 7. CONCLUSIONS ........................................................................................................ 24 REFERENCES ..................................................................................................................... 26 ADBI Working Paper 2018 Y. Niimi 1 1. INTRODUCTION Intergenerational mobility measures the degree to which socioeconomic status is transmitted from one generation to the next, and as such, it can be considered a fundamental indicator of the equality of economic opportunities in a society. A strong association between the socioeconomic status of parents and that of children indicates low intergenerational mobility whereby children from a disadvantaged family are likely to remain disadvantaged throughout their lives and may not be able to achieve their economic potential regardless of their abilities or efforts (Blanden, Gregg, and Macmillan 2007). While various aspects of intergenerational mobility have been measured in the past, including earnings, occupation, and education, this paper pays particular attention to the intergenerational transmission of educational attainment. According to the intergenerational human capital investment model proposed by Becker and Tomes (1986), education, or human capital, is an important channel through which earnings ability is transmitted from parents to children. High-earning parents have a greater financial capacity to invest in their children’s education. Such parents may also pass on certain attributes (genetic and cultural endowments) to their children, which make it easier for children to acquire education. More education then enables children to obtain higher earnings in adulthood. Education therefore plays a central role in passing on socioeconomic advantage (or disadvantage) from generation to generation. Besides the importance of examining the intergenerational transmission of educational attainment per se, there are a number of advantages to using education, rather than earnings, to measure intergenerational mobility. Education is less sensitive to well-known problems with measurement error and life cycle bias than earnings (Solon 1992; Mazumder 2005; Zimmerman 1992). Information on the educational attainment of both parents and children is more readily available in household surveys and recall-based information on the educational attainment of parents is likely to be of better quality than that on their earnings. Since education is a more or less permanent characteristic and most individuals complete their education by their mid-20s, education is also less prone to life cycle bias than earnings. As a result, in the absence of accurate information on permanent income, the use of education is arguably the best way to measure the intergenerational transmission of socioeconomic status. Moreover, Blanden (2013) shows that intergenerational mobility in earnings and education tend to be highly correlated and suggests that educational mobility could be considered a good proxy for earnings mobility. There is a growing literature that examines the intergenerational transmission of educational attainment, and the empirical evidence consistently shows a strong association between parents’ and children’s educational attainment. 1 On the other hand, the evidence on the importance of borrowing constraints for children’s educational attainment as well as for intergenerational earnings/educational mobility remains inconclusive. For instance, several studies find that parents’ financial resources play only a limited role in children’s educational attainment, suggesting that borrowing constraints are relatively unimportant (e.g., Cameron and Heckman 1998, 2001; Carneiro and Heckman 2002) while other studies find stronger evidence of borrowing constraints, particularly in more recent years (e.g., Belley and Lochner 2007; Lochner and Monge-Naranjo 2012). 1 See Black and Devereux (2011) for a comprehensive survey of the literature on intergenerational mobility. ADBI Working Paper 2018 Y. Niimi 2 Note that when discussing the equality of opportunity, the determinants of any outcome can be separated into two components: “circumstances,” such as family background, and “efforts” (e.g., Roemer 1998). Inequality arising from circumstances that are beyond one’s control certainly calls for policy interventions. Given that borrowing constraints for children’s education largely stem from circumstances, analyzing the relative importance of borrowing constraints for intergenerational mobility has important policy implications. If borrowing constraints are responsible for the intergenerational persistence of educational attainment, measures for relaxing such constraints should be adopted. However, if the formation of children’s human capital is determined mainly by families’ underlying characteristics (e.g., genetic and cultural environments including aspiration for education) that are passed on from generation to generation, policies that merely address financial constraints would not be effective in addressing the intergenerational persistence of educational attainment and there is room for discussion about whether policy interventions are needed (and if so, what interventions are needed) to address the lack of intergenerational mobility. Nevertheless, examining empirically the importance of borrowing constraints for children’s educational attainment and intergenerational mobility is not a straightforward exercise given the difficulty of identifying which families are credit constrained. Whether or not borrowing constraints bind is determined not only by parents’ financial resources but also by the endowed ability of children and whether well-functioning credit markets (or alternatives to credit market financing of human capital investment) exist in a society (Grawe and Mulligan 2002). Fortunately, the Preference Parameters Study of Osaka University, which we use for our analysis, contains unique information on whether children have ever given up schooling for financial reasons, and if they have, which level of schooling they have forgone. By exploiting such information, this paper aims to contribute to the existing literature on the implications of borrowing constraints for the intergenerational transmission of educational attainment in two ways. First, we assess the importance of borrowing constraints for intergenerational educational mobility by comparing the extent of the intergenerational transmission of educational attainment estimated using original data on children’s years of schooling with that estimated using the hypothetical number of years of schooling for children who had to give up schooling for financial reasons assuming that they had not faced borrowing constraints. We believe that this is a more direct way of assessing the importance of borrowing constraints for intergenerational educational mobility than previous attempts made in the literature. Second, by analyzing the case of Japan, we try to broaden our understanding of the intergenerational transmission of educational attainment and the role of borrowing constraints in determining it in Japan, for which empirical evidence hardly exists. Japan is an interesting case to study given that Japanese parents tend to bear the relatively heavy financial burden of tuition fees for tertiary education and that they also tend to bear the high cost of sending their children to cram schools to make sure that their children are well prepared to succeed in the fierce competition for entrance to upper secondary schools and universities (or even to lower levels of schooling).2 2 For instance, according to the National Transfer Accounts data collected by the Center for the Economics and Demography of Aging, University of California at Berkeley (available at http://www.ntaccounts.org/web/nta/show/Country%20Summaries), about 94% of the private consumption of those aged 18–21 is financed by private transfers (predominantly from parents) in the case of Japan, whereas the corresponding figure for the United States is about 58% (Niimi and Horioka 2017). ADBI Working Paper 2018 Y. Niimi 3 The rest of the paper is structured as follows. The next section describes the conceptual framework. Section 3 reviews the related literature with a particular focus on the empirical evidence on the role of borrowing constraints in intergenerational mobility. Section 4 describes the education system in Japan. Section 5 describes the data and estimation methods. Section 6 presents the estimation results. Section 7 ends with some concluding remarks. 2. CONCEPTUAL FRAMEWORK The empirical analysis of the intergenerational transmission of educational attainment in this paper is based on the intergenerational human capital investment model developed by Becker and Tomes (1986) for analyzing the intergenerational transmission of earnings, assets, and consumption across generations. The model posits education, or human capital, as the central mechanism through which socioeconomic advantage (or disadvantage) is passed on across generations. In the model, a child’s earnings or economic status in adulthood depend on his/her endowments as well as on his/her parents’ investments in the child’s human capital. Children are assumed to inherit genetic and cultural endowments (e.g., genetic traits, cognitive and noncognitive abilities, and family environment) from parents while utilitymaximizing parents are assumed to be altruistic and make optimal investments in their children’s human capital. If parents can readily borrow to finance optimal investments in their children, the degree of intergenerational earnings mobility will simply be equal to the inheritability of endowments given that there will be no direct relationship between parents’ financial resources and investments in their children’s human capital. In this case, an autoregressive process in earnings across generations is expected. By contrast, if access to capital markets is limited and parents cannot borrow against their children’s future earnings to finance investments in children’s human capital, this is likely to result in the intergenerational persistence of human capital and earnings. Financial market imperfections are likely to occur because children cannot credibly commit to paying back the loans parents take out on their behalf. In such a case, given that high-income families can more readily self-finance a given amount of investment in their children than lowand middle-income families, the inability of parents to borrow depresses the earnings of poor children vis-à-vis rich children with the same ability. This, in turn, strengthens the correlation between the earnings of parents and children in families that do not have enough funds to invest optimally in their children’s human capital. In other words, in the presence of imperfect credit markets, the degree of intergenerational earnings mobility will also depend on the earnings of parents and their willingness to self-finance investments in their children. In sum, the intergenerational transmission of human capital and earnings rests upon the intergenerational transmission of endowments, either genetically or through the environment, and the presence of borrowing constraints. In the presence of borrowing constraints, the model predicts that intergenerational earnings/educational mobility would be lower among constrained groups than among unconstrained groups given that the former would make a suboptimal investment in their children’s human capital. Note, however, that the model proposed by Becker and Tomes (1986) predicts that the presence of borrowing constraints merely slows down the process of convergence of successive generations toward the mean and does not prevent it in the long term. Nevertheless, given that this seems at odds with the observed intergenerational ADBI Working Paper 2018 Y. Niimi 4 persistence of educational attainment, a number of alternative models have subsequently been developed to explain this phenomenon.3 Becker et al. (2015), for example, consider how the persistence of economic status depends on the distribution of income, and in doing so, they emphasize the importance of complementarities between parents’ human capital and investments in children in the production of children’s human capital (i.e., highly educated parents are more productive at teaching their children). They show that when returns to investments in children increase in parents’ human capital, the equilibrium relationship between parents’ and children’s human capital tends to be convex, resulting in greater intergenerational persistence among high-income families even in a world with perfect capital markets and without differences in children’s innate ability. On the other hand, borrowing constraints may produce high persistence among low-income families. As a consequence, their theory predicts that intergenerational mobility will be low at both ends of the income distribution and that successive generations of the same family may cease to regress toward the mean if complementarities in the production of children’s human capital are strong enough (Becker et al. 2015).4 Moreover, Hellier (2017) provides a useful synthetic and encompassing framework for modeling several factors that have been considered in the literature as determinants of the slowdown in the pace of human capital convergence or of the emergence of human capital stratification/low-education traps. These include credit market imperfections, fixed costs of education, “S-shaped” production of children’s human capital, local externalities and neighborhood effects, and the structure of education systems. As shown in Hellier (2017), many of these factors, as well as various combinations thereof, can divide the population between several education groups, causing education-based social stratification and low-education traps. While all of these factors are equally important, we focus our analysis on the role of borrowing constraints in determining intergenerational educational mobility. 3. RELATED LITERATURE ON BORROWING CONSTRAINTS According to the conceptual framework discussed above, the intergenerational transmission of socioeconomic status would be greater in a world with borrowing constraints. However, the evidence on the role of borrowing constraints in determining intergenerational earnings mobility has so far been mixed. 5 Moreover, it is not a straightforward exercise to test this hypothesis empirically given the difficulty of identifying which families are credit constrained. Whether or not borrowing constraints bind is determined not only by family income but also by the endowed ability of children and whether well-functioning credit markets (or alternatives to credit market financing of human capital investment) exist in a society (Grawe and Mulligan 2002). Han and Mulligan (2001) indeed suggest that 3 See Chusseau, Hellier, and Ben-Halima (2013) and Hellier (2017) for a comprehensive review of the theoretical literature. 4 Heckman and Mosso (2014) similarly note that even in the absence of imperfect credit markets, the intergenerational correlation between human capital and earnings may be observed due, for example, to the accident of birth because returns to parental investments depend on parents’ own human capital by affecting the productivity of investments. 5 See Black and Devereux (2011) for a comprehensive survey of the literature on intergenerational mobility. ADBI Working Paper 2018 Y. Niimi 11 Figure 3: Share of University Tuition Fees in Average Annual Household Income (%) Source: Data on household income are from the Comprehensive Survey of Living Conditions conducted by MHLW (available at http://www.mhlw.go.jp/toukei/list/20-21.html); data on tuition fees are from MEXT (available at http://www.mext.go.jp/a_menu/koutou/shinkou/07021403/__icsFiles/afieldfile/2017/09/26/1396452_03.pdf). 5. DATA AND ESTIMATION METHODS 5.1 Data The data used for the empirical analysis come from the “Preference Parameters Study” of Osaka University. This survey was conducted annually in Japan during the 2003–13 period by the 21st Century Center of Excellence (COE) Program “Behavioral Macrodynamics Based on Surveys and Experiments” and the Global COE Project “Human Behavior and Socioeconomic Dynamics” of Osaka University. A sample of individuals aged 20–69 was drawn to be nationally representative using two-stage stratified random sampling. The sample has a panel component, although fresh observations were added in 2004, 2006, and 2009 to overcome the problem of attrition. Given that data from the Preference Parameters Study contain information not only on respondents’ educational attainment but also on that of their parents, the data are well suited for analyzing intergenerational educational mobility. Data from the 2012 wave (4,588 observations in total) are mainly used for the present analysis as they contain unique information on whether respondents have ever given up schooling for financial reasons. In other words, the data allow us to identify exactly which families faced borrowing constraints for their children’s education. The 2012 wave also includes other useful questions, such as those on respondents’ mothers’ presence in the household when respondents were 3, 7, and 15 years old, respectively. We also make use of information on respondents and their families when respondents were 15 years old from the 2009 wave, namely the number of respondents’ siblings, the standard of living of their families, the prefecture of their residence, and their academic performance. The age of respondents in the 2012 wave ranges from 22 to 78. We restrict our estimation sample to respondents who were born after World War II. In addition, we drop respondents who said that they were still students in 2012. We then divide the remaining sample into three cohorts, more specifically respondents born between 1946 ADBI Working Paper 2018 Y. Niimi 12 and 1955 (Cohort I), between 1956 and 1965 (Cohort II), and between 1966 and 1989 (Cohort III), in order to see whether there have been any changes in intergenerational educational mobility over time.16 After excluding observations with missing information on the variables used in our analysis, we are left with 3,012 observations. 5.2 Estimation Methods To examine the intergenerational transmission of educational attainment, we estimate the following standard equation using ordinary least squares (OLS): 𝑌 𝑖𝑐 =𝛼+𝛽𝑌 𝑖𝑝 +𝛾𝑋𝑖+𝜀𝑖 (1) where Yic and Yip represent the years of schooling completed by child c and parent p of family i, respectively. Xi is a vector of variables representing the characteristics of the child and his/her family, and εi is an error term. β is the parameter of interest and indicates the extent of the intergenerational persistence of educational attainment. A greater absolute value of β implies that children’s schooling is more heavily influenced by their parents’ schooling while a value close to 0 implies that children’s schooling tends to be independent of their parents’ schooling. Note that the estimated 𝛽 󰆹 is given by: 𝛽 󰆹=𝜌𝑐𝑝 𝜎𝑐 𝜎𝑝 (2) where σc and σp are the standard deviations of children’s and parents’ years of schooling, respectively, and ρcp is the correlation between children’s and parents’ years of schooling. Equation (2) implies that an increase (decrease) in the estimated intergenerational persistence of educational attainment 𝛽 󰆹 may simply be the result of an increase (decrease) in the dispersion of children’s schooling relative to the dispersion of parents’ schooling (Checchi, Fiorio, and Leonardi 2013; Hertz et al. 2007). Hence, we also estimate the following equation whereby we normalize children’s and parents’ years of schooling by the corresponding standard deviations: 𝑌𝑖𝑐 𝜎𝑐 =δ+ρ𝑌𝑖𝑝 𝜎𝑝 +𝜏𝑋𝑖+𝜀𝑖 (3) The coefficient β therefore takes into account changes in the dispersion of educational outcomes in children’s and parents’ generations, providing a relative measure of intergenerational persistence. On the other hand, the coefficient ρ provides an absolute measure of intergenerational persistence, which is adjusted for changes in the distribution of educational attainment from one generation to the next. We estimate both equations (1) and (3) and report both measures of the intergenerational persistence of educational attainment. Our dependent variable is children’s (i.e., respondents’ in this case) years of schooling and the main explanatory variable of interest is parents’ years of schooling. Given that information on educational attainment is reported as categorical variables based on the completion/incompletion of various academic qualifications in the Preference 16 If we had restricted Cohort III to those born between 1966 and 1975 (a 10-year interval) to be consistent with Cohorts I and II, the sample size would have been relatively small (768 observations). We therefore decided to include all observations of those born after 1966 in this age group. However, even if we restrict the sample to those born between 1966 and 1975 for Cohort III, the regression results are similar and the findings presented in the paper remain the same. ADBI Working Paper 2018 Y. Niimi 13 Parameters Study, we calculate years of schooling as the minimum length of time required to obtain a particular qualification for both children and parents. For parents’ years of schooling, we use the average of fathers’ and mothers’ years of schooling, but we also try including both fathers’ and mothers’ years of schooling simultaneously. As noted earlier, the 2012 wave of the Preference Parameters Study includes questions asking respondents whether they have ever given up schooling for financial reasons, and if they have, which level of schooling (upper secondary school, college of technology, two-year junior college, university, or graduate school) they have forgone. It is therefore possible to identify which respondents were credit constrained for their education and how much schooling they could have gotten if they had not faced borrowing constraints. Using this unique information, we recalculate respondents’ years of schooling assuming that nobody faced borrowing constraints. 17 We then re-estimate equation (1) based on this version of our dependent variable and compare the estimates of β to assess the importance of borrowing constraints for the intergenerational transmission of educational attainment. Theory predicts that intergenerational educational mobility would be lower in a society with borrowing constraints given that some parents would not be able to invest in children’s human capital at the optimal level. We therefore expect our estimate of β to be greater when we use the original data on children’s years of schooling than when we use the hypothetical number of years of schooling for children who had to give up schooling for financial reasons on the assumption that nobody had faced borrowing constraints. This is a more direct assessment of the importance of borrowing constraints for intergenerational educational mobility than previous attempts made in the literature. As illustrated above, respondents are considered to have faced borrowing constraints in the present analysis if they have ever given up their schooling for financial reasons. As a result, “borrowing constraints” are defined in a relatively broad sense here, and are not restricted to a situation in which respondents or their parents were not able to borrow to finance respondents’ education. Moreover, the self-reporting nature of the question poses some limitations to accurately identifying families facing borrowing constraints for children’s education. Nevertheless, when we analyze the determinants of the likelihood of facing borrowing constraints, we obtain reasonable results with most coefficients having the expected signs, as shown below. While we acknowledge the limitation of using responses to the self-reported questions in this analysis, we take some comfort from these findings. As far as the other explanatory variables are concerned, we include respondents’ characteristics, namely their gender and their academic performance when they were 15 years old. Respondents are asked to indicate how high/low their grades (i.e., low, relatively low, middle, relatively high, or high) were relative to others in their grade for all subjects, particularly Japanese, and mathematics. We construct a categorical variable that indicates the relative rank of respondents’ grades for all subjects. This variable can be interpreted as the academic ability of respondents when they were adolescents. We also include four measures of family background using information on respondents’ families when respondents were 15 years old: (i) a variable that indicates whether the respondent was the eldest son in the family; (ii) the number of siblings the respondent had; (iii) a variable that indicates whether the mother was absent from the household 17 We calculate the lower bound of the hypothetical number of years of schooling assuming, for example, that respondents went only to upper secondary school if they said that they gave up going to upper secondary school for financial reasons (i.e., we assume that they did not go on to university in this case). ADBI Working Paper 2018 Y. Niimi 14 (i.e., divorced or died); and (iv) the standard of living of the respondent’s family. The latter is based on respondents’ answers to a question asking them to indicate their relative standard of living during their childhood on a scale of 0–10, with 10 being “wealthiest” and 0 being “poorest.” We treat this variable as being cardinal. Given that the family’s standard of living is likely to reflect permanent income more than short-term liquidity constraints, including this variable in the estimation model allows us to examine the effect of permanent income on children’s educational attainment, at least to some extent. We additionally include respondents’ cohort fixed effects (5-year intervals) to control for cohort trends in education as well as regional fixed effects based on the prefecture of respondents’ residence when they were 15 years old to allow for differences in geographical characteristics. We also include cohort fixed effects (10-year intervals) for respondents’ fathers and mothers. Finally, we include a variable that indicates the annual average ratio of active job openings to applicants in the year in which respondents were 18 years old to reflect the economic situation around the time respondents considered going to university.18 Note that the estimated intergenerational transmission of educational attainment in this analysis should be regarded as a descriptive measure of intergenerational association in educational attainment rather than as a measure of the causal effect of parents’ education on children’s education. The behavior and decisions of children may be affected by the unobserved characteristics of parents, such as genes, preferences, and/or family environment. As a result, OLS estimates are potentially biased upwards. Unfortunately, we lack the data needed to implement appropriate identification strategies that would allow us to disentangle these effects and identify a causal mechanism that underlies the relationship between parents’ and children’s educational attainment. The coefficients we estimate should therefore be interpreted as a combination of all of these effects. We leave the examination of the causal relationship underlying the intergenerational transmission of educational attainment as an agenda for future research. 6. EMPIRICAL RESULTS 6.1 Descriptive Statistics Table 1 shows summary statistics for the dependent and explanatory variables used in the empirical analysis, separately for each cohort. The table shows that there has been an increase in the years of schooling completed by parents and children (i.e., respondents) over time. While our variable that indicates the standard of living of children’s families when children were 15 years old is essentially a relative term, the number of children who feel that their standard of living was relatively high seems to have increased over time as well. At the same time, the proportion of children who gave up schooling for financial reasons declined, from about 16% for Cohort I to about 9% for Cohort III. This may be partly due to the increase in the standard of living of children’s families, but it may also be partly due to the steady decline in the number of siblings children have, as shown in the table. 18 Data on the active job openings to applicants ratio are from statistics on Employment Referrals for General Workers collected by MHLW (available at http://www.mhlw.go.jp/toukei/list/114-1.html). ADBI Working Paper 2018 Y. Niimi 15 Table 1: Descriptive Statistics Cohort I (1946–1955) Cohort II (1956–1965) Cohort III (1966–1989) mean s.d. mean s.d. mean s.d. Children’s characteristics Years of schooling 12.98 2.10 13.65 1.91 13.76 1.95 Female 0.52 0.55 0.58 Academic performance at age 15 Low rank 0.04 0.06 0.07 Relatively low rank 0.09 0.10 0.15 Middle 0.36 0.36 0.38 Relatively high rank 0.32 0.31 0.26 High rank 0.19 0.17 0.14 Gave up schooling for financial reasons 0.16 0.11 0.09 Hypothetical number of years of schooling 13.52 2.07 13.98 1.90 14.02 1.95 Parents’ characteristics Average years of schooling 10.32 1.69 10.97 1.87 11.89 1.91 Father’s years of schooling 10.43 2.18 11.20 2.45 12.06 2.46 Mother’s years of schooling 10.21 1.59 10.75 1.72 11.72 1.82 Family background at age 15 Eldest son 0.16 0.21 0.21 Number of siblings 2.35 1.51 1.61 1.05 1.44 0.74 Standard of living 4.55 1.75 4.89 1.71 5.34 1.81 No mother (divorced or died) 0.02 0.01 0.01 Job openings to applicants ratio 1.10 0.32 0.69 0.17 0.88 0.28 No. of observations 905 941 1,166 s.d. = standard deviation. Source: Calculations based on data from the 2009 and 2012 Preference Parameters Study. Carneiro and Heckman (2002) estimate that at most about 8% of youth in the US are subject to short-term liquidity constraints that affect their post-secondary schooling using data from the 1979 NLSY. This cohort corresponds to Cohort II in our analysis,19 and the percentage of children who gave up going to a four-year university or a twoyear junior college is estimated to be about 10.0% for this cohort. Although this figure is for Japan, the percentage of children who gave up tertiary education for financial reasons seems relatively similar in Japan and the US. 19 Those who were born between the years 1957 and 1964 were surveyed in the 1979 NLSY. ADBI Working Paper 2018 Y. Niimi 16 Figure 4 shows the level of schooling children gave up for financial reasons for each cohort. Although more than half of the children who were credit constrained gave up going to university in all three cohorts, there are some differences across cohorts in the level of schooling children gave up. The percentage of children who did not go to upper secondary school for financial reasons was relatively high (about 26%) for the oldest cohort compared with younger cohorts (about 8% and 5%, respectively). By contrast, there was an increase in the percentage of children who did not go to graduate school for financial reasons in younger cohorts. Figure 4 thus suggests that borrowing constraints used to bind more at lower educational levels for older cohorts than for younger cohorts. Nevertheless, Table 1 shows that the hypothetical number of years of schooling, assuming that nobody faced borrowing constraints, is consistently greater than the years of schooling children actually completed for all three cohorts. Figure 4: Proportion of Children Who Gave Up Schooling for Financial Reasons by Level of Schooling (%) Source: Calculations based on data from the 2012 Preference Parameters Study. 6.2 Mobility Indices Before moving on to our regression analysis, we first examine intergenerational educational mobility based on transition matrices. Table 2 presents the matrices of transition among four education categories defined based on years of schooling: (i) equal to or less than 9 years (up to lower secondary education); (ii) more than 9 but equal to or less than 12 years (up to upper secondary education); (iii) more than 12 but less than 16 years (some tertiary education); and (iv) equal to or more than 16 years (at least four-year university degree). ADBI Working Paper 2018 Y. Niimi 17 Table 2: Transition Probability Matrices and Mobility Indices by Birth Cohort Children Parents <=9 >9 and <=12 >12 and <16 >=16 Mobility Index (Shorrocks) Average Jump Index Cohort I (1946–1955) <=9 0.13 0.64 0.11 0.13 0.770 0.978 >9 and <=12 0.04 0.52 0.19 0.25 >12 and <16 0.03 0.16 0.28 0.54 >=16 0.00 0.15 0.08 0.77 Cohort II (1956–1965) <=9 0.03 0.63 0.17 0.16 0.838 1.026 >9 and <=12 0.01 0.43 0.26 0.30 >12 and <16 0.00 0.15 0.29 0.56 >=16 0.00 0.04 0.22 0.74 Cohort III (1966–1989) <=9 0.03 0.73 0.16 0.08 0.822 0.842 >9 and <=12 0.01 0.45 0.27 0.27 >12 and <16 0.00 0.23 0.29 0.48 >=16 0.00 0.04 0.19 0.77 Source: Calculations based on data from the 2012 Preference Parameters Study. A 4 x 4 transition matrix P is computed based on the four categories of educational outcomes, as described above, for parents and children of each cohort. The elements of the matrix are pij, which represent the probabilities that educational outcomes move from category i in parents’ generation to category j in children’s generation. To assess how mobility has changed over time, we calculate two types of mobility indices: (i) the mobility index proposed by Shorrocks (1978) and (ii) the average jump index proposed by Bartholomew (1973). Shorrocks’ (1978) mobility index is defined as: 𝑀𝑆=𝑘−𝑡𝑟𝑎𝑐𝑒(𝑃) 𝑘−1 (4) where P is a transition matrix with k educational categories and trace(P) is the sum of the elements of the main diagonal of P. This index ranges from 0 (zero mobility) to 1 (perfect mobility). The computed indices are shown in Table 2. While we find that educational mobility increased from 0.770 to 0.838 between Cohort I and Cohort II, it declined slightly to 0.822 for Cohort III. One of the drawbacks of Shorrocks’ mobility index is that it is insensitive to any moves other than those on the diagonal. As a complementary measure, we also calculate the “average jump” index proposed by Bartholomew (1973), which addresses this issue by taking into account movements off the diagonal. It is defined as: 𝑀𝐵=∑∑𝑝𝑖.𝑝𝑖𝑗|𝑖 − 𝑗| 𝑘 𝑗 𝑘 𝑖 (5) ADBI Working Paper 2018 Y. Niimi 18 where pij is the value of the element in row i and column j and pi. is the marginal distribution of educational category i for parents’ generation. Their product is multiplied by the distance between the two educational categories. The average jump index shows a trend similar to the one shown by Shorrocks’ mobility index: educational mobility increased, then decreased, during the postwar period. 6.3 Regression Results Intergenerational Transmission of Educational Attainment To further analyze the intergenerational transmission of educational attainment, we turn to a regression analysis of the determinants of children’s years of schooling. Table 3 reports the OLS regression results of equation (1). We first regress children’s years of schooling on parents’ years of schooling along with basic variables, namely a female dummy, the job openings to applicants ratio, regional fixed effects, and cohort fixed effects for children as well as for their fathers and mothers. We then add variables that reflect family background when children were 15 years old, namely whether the child is the eldest son, the number of siblings, the relative standard of living, and mother’s presence. We then further add variables that indicate how well children performed academically at school when they were 15 years old. We first compare the adjusted R2 of different regression models for each cohort to assess the importance of family background and of children’s ability when they were adolescents as determinants of children’s years of schooling. The adjusted R2 of the basic variant (with parents’ educational attainment, a female dummy, the job openings to applicants ratio, and cohort and regional fixed effects included) is relatively similar for all cohorts. While the adjusted R2 increases for all cohorts when we add variables relating to family background, the size of the increase becomes smaller as we move to younger cohorts. While adding a set of variables relating to family background increases the adjusted R2 by about 30% for Cohort I, it only increases the adjusted R2 by about 21% and about 5% for Cohorts II and III, respectively. This suggests that the relative importance of family background for children’s educational attainment has declined over time. If we look at individual coefficients, the penalty for having a larger number of siblings and for growing up in a household with a relatively low level of standard of living is found to have declined over the years. If we consider the standard of living variable as a proxy for permanent income, the regression results suggest that the family’s permanent income is still an important determinant of the child’s educational attainment, but its relative importance seems to have declined over time to some extent. The cost of growing up in a broken family (the mother is absent) is also observed only among older cohorts. Unfortunately, we do not have information on whether the father was present in the household when the child was growing up. Given the increasing number of single mothers in recent years in Japan, examining the implication of growing up in a single-mother household for children’s educational attainment as well as for intergenerational educational mobility is left as an important agenda for future research. ADBI Working Paper 2018 Y. Niimi 19 Table 3 : Regression Results for the Determinants of Children’s Years of Schooling Cohort III (1966–1989) 0.309*** [0.027] –0.055 [0.155] –0.110* [0.066] 0.056** [0.027] 0.053 [0.424] –0.321*** [0.122] – 0.521*** [0.196] – 0.518*** [0.144] 0.838*** [0.121] 1.821*** [0.149] –0.047 [0.303] 9.678*** [0.622] 0.331 1,166 Note: ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels. Standard errors are in parentheses. Regional and cohort dummies are included in all regressions. S ource: Estimation based on data from the 2009 and 2012 Preference Parameters Study. 0.379*** [0.029] 0.107 [0.170] –0.164** [0.072] 0.082*** [0.030] –0.181 [0.466] –0.282** [0.134] –0.047 [0.334] 9.212*** [0.678] 0.185 1,166 0.400*** [0.029] –0.335*** [0.106] –0.045 [0.335] 9.195*** [0.662] 0.176 1,166 Cohort II (1956–1965) 0.266*** [0.029] 0.351** [0.160] –0.175*** [0.054] 0.109*** [0.031] –1.319** [0.559] –0.721*** [0.126] –0.606*** [0.233] –0.351* [0.182] 0.902*** [0.126] 1.475*** [0.152] –0.437 [0.311] 11.768*** [1.678] 0.342 941 0.318*** [0.031] 0.371** [0.174] –0.208*** [0.059] 0.141*** [0.034] –1.681*** [0.605] –0.763*** [0.136] –0.297 [0.336] 10.624*** [1.816] 0.227 941 0.362*** [0.031] –0.883*** [0.113] –0.330 [0.344] 10.884*** [1.840] 0.188 941 Cohort I (1946–1955) 0.279*** [0.037] –0.130 [0.188] –0.207*** [0.044] 0.146*** [0.035] –1.184*** [0.402] –0.943*** [0.134] –0.455 [0.302] –0.848*** [0.215] 0.760*** [0.140] 1.696*** [0.163] 0.283 [0.277] 10.272*** [0.762] 0.354 905 0.366*** [0.039] –0.163 [0.204] –0.257*** [0.048] 0.189*** [0.038] –1.159*** [0.438] –0.947*** [0.146] 0.187 [0.301] 9.971*** [0.828] 0.234 905 0.446*** [0.039] –0.750*** [0.128] 0.145 [0.311] 8.650*** [0.798] 0.180 905 Parents’ years of schooling Family background at age 15 E ldest son Number of siblings Standard of living No mother (divorced or died) Children’s characteristics Female Academic performance at age 15 (base category: middle) Low rank Relatively low rank Relatively high rank High rank Job openings to applicants ratio C onstant Adjusted R2 No. of observations ADBI Working Paper 2018 Y. Niimi 20 As far as adolescent academic ability is concerned, it is found to be an important determinant of children’s educational attainment for all cohorts, which is consistent with previous studies (e.g., Cameron and Heckman 1998, 2001; Carneiro and Heckman 2002). Moreover, unlike family background, the importance of adolescent ability is found to have increased for the youngest cohort. Simply adding the adolescent ability variable increases the adjusted R2 further by about 52% and about 51% for Cohorts I and II, respectively, but it increases the adjusted R2 even more (about 79%) for Cohort III. In other words, how well children performed when they were 15 years old seems to play a greater role in determining their subsequent educational attainment today than in the past, thereby suggesting the increasing importance of investments in children’s human capital at early ages.20 Even after controlling for family background as well as for children’s ability when they were adolescents, we still find a significant association between parents’ and children’s schooling, as found in the literature. It is, however, more disturbing to find that the extent of the intergenerational transmission of educational attainment increased for the youngest cohort in comparison with older cohorts. A one-year increase in parents’ average years of schooling is associated with a 0.28 and a 0.27 increase in children’s years of schooling for Cohort I and Cohort II, respectively, but it is associated with a 0.31 increase in children’s years of schooling for the youngest cohort. Table 4: Relative and Absolute Measures of Intergenerational Persistence in Educational Attainment Cohort I (1946–1955) Cohort II (1956–1965) Cohort III (1966–1989) β (relative) 0.279*** [0.037] 0.266*** [0.029] 0.309*** [0.027] ρ (absolute) 0.224*** [0.029] 0.261*** [0.029] 0.303*** [0.026] No. of observations 905 941 1,166 Note: *** denotes statistical significance at the 1% level. Standard errors are in parentheses. Source: Estimation based on data from the 2009 and 2012 Preference Parameters Study. To examine whether this increase in the intergenerational transmission of educational attainment during the postwar period is due to an increase in the dispersion of children’s schooling relative to the dispersion of parents’ schooling, we also estimate equation (3), in which we normalize children’s and parents’ years of schooling by the corresponding standard deviations. We include a full set of explanatory variables in the equation, and the relevant results are shown in Table 4.21 The estimated ρ is much smaller than the estimated β for Cohort I, whereas they are relatively similar for younger cohorts. The relatively large difference between the estimates of β and ρ for Cohort I is due to the fact that the dispersion of children’s years of schooling is significantly greater than that of parents’ years of schooling (see Table 1). By contrast, the dispersion of children’s and parents’ years of schooling 20 Note, however, that the observed increase in the importance of adolescent academic ability for children’s subsequent educational attainment may also be due to an increase in the importance of “residual” ability not explained by other explanatory variables. This may be caused by, for example, increased heterogeneity in school quality. Due to the lack of data on the quality of schools that children attended, a further investigation of this point is left for future research. However, I would like to thank Tomoki Fujii for drawing my attention to this very important point. 21 The full regression results are available from the author upon request. ADBI Working Paper 2018 Y. Niimi 27 Grawe, N. D. and C. B. Mulligan (2002), “Economic Interpretations of Intergenerational Correlations,” Journal of Economic Perspectives, 16(3), pp. 45–58. Han, S. and C. B. Mulligan (2001), “Human Capital, Heterogeneity and Estimated Degrees of Intergenerational Mobility,” Economic Journal, 111(470), pp. 207–243. Heckman, J. J. and S. Mosso (2014), “The Economics of Human Development and Social Mobility,” Annual Review of Economics, 6, pp. 689–733. Hellier, J. (2017), “Education, Intergenerational Mobility and Social Stratification in Economic Theory: An Encompassing Approach,” mimeo, University of Nantes. Hertz, T., T. Jayasundera, P. Piraino, S. Selcuk, N. Smith, and A. Verashchagina (2007), “The Inheritance of Educational Inequality: International Comparisons and Fifty-Year Trends,” The B. E. Journal of Economic Analysis & Policy, 7(2), Article 10. Higuchi, Y. (1992), “Kyoiku wo Tsujita Sedaikan Shotokuiten” (Intergenerational Income Transfers through Education) (in Japanese), Nihon Keizai Kenkyu (JCER Economic Journal), Vol. 22, pp. 137–165. Hojo, M. (2008), “Kyouikuwo Tsuujita Sedaikan Shotokuiten,” (Intergenerational Income Transfers through Education) (in Japanese), in Horioka, C. Y. and the Institute for Research on Household Economics (eds) Setainai Bunpai to Sedaikan Iten no Keizai Bunseki (Economic Analysis of Intra-household Distribution and Intergenerational Transfers), Kyoto: Minerva Shobo. Karagiannaki, E. (2017), “The Effect of Parental Wealth on Children’s Outcomes in Early Adulthood,” Journal of Economic Inequality, 15(3), pp. 217–243. Keane, M. P. and K. I. Wolpin (2001), “The Effect of Parental Transfers and Borrowing Constraints on Educational Attainment,” International Economic Review, 42(4), pp. 1051–1103. Kubota, K. (2016), “Credit Constraints, Education, and Intergenerational Mobility: Evidence from U.S. Survey Data,” paper presented at Tokyo Labor Economics Workshop on May 13, 2016. Lochner, L. and A. Monge-Naranjo (2012), “Credit Constraints in Education,” Annual Review of Economics, 4, pp. 225–256. Mazumder, B. (2005), “Fortunate Sons: New Estimates of Intergenerational Mobility in the United States Using Social Security Earnings Data,” Review of Economics and Statistics, 87(2), pp. 235–255. Mulligan, C. B. (1997), Parental Priorities and Economic Inequality, Chicago: University of Chicago Press. Niimi, Y. and C. Y. Horioka (2017), “The Impact of Intergenerational Transfers on Wealth Inequality in Japan and the United States,” World Economy, forthcoming, DOI: 10.1111/twec.12544. Oshio, T. and W. Senoh (2005), “The Economics of Education in Japan: A Survey on Empirical Studies and Unresolved Issues” (in Japanese), Keizai Bunseki (Economic Analysis), 175, pp. 105–139. Restuccia, D. and C. Urrutia (2004), “Intergenerational Persistence of Earnings: The Role of Early and College Education,” American Economic Review, 94(5), pp. 1354–1378. ADBI Working Paper 2018 Y. Niimi 28 Roemer, J. E. (1998), Equality of Opportunity, Cambridge: Harvard University Press. Shorrocks, A. F. (1978), “The Measurement of Mobility,” Econometrica, 46(5), pp. 1013–1024. Solon, G. (1992), “Intergenerational Income Mobility in the United States,” American Economic Review, 82(3), pp. 393–408. Zimmerman, D. J. (1992), “Regression Toward Mediocrity in Economic Stature,” American Economic Review, 82(3), pp. 409–429.