The effect of the Free High School Tuition law on upper-secondary school choice in Japan

1 The effect of the Free High School Tuition law on upper-secondary school choice in Japan Petr Mariel 1 Department of Quantitative Methods University of the Basque Country (UPV/EHU) Avda. Lehendakari Aguirre, 83 E48015 Bilbao, Spain E-mail: petr[email protected] Tel.: +34.94.601.3848 Nobuhiro Sanko Graduate School of Business Administration Kobe University 2-1 Rokkodai-cho, Nada-ku, Kobe-shi 657-8501 Japan E-mail: [email protected] Tel.: +81.78.803.6987 Ainhoa Vega-Bayo Departament of Economic Analysis University of the Basque Country (UPV/EHU) Avda. Lehendakari Aguirre, 83 E48015 Bilbao, Spain, E-mail: ainhoa.veg[email protected]s Tel.: +34.94.601.7076 Funding: The authors acknowledge financial support from the FEDER/Ministry of Science, Innovation and Universities (ECO2017-82111-R, PID2020-113650RB-I00) and the Basque Government through grants IT1359-19 (UPV/EHU Econometrics Research Group) and IT1336-19 (Bilbao Research Team in Economics). 1 Corresponding author This is the accept manuscript of the following article that appeared in final form in Studies in Educational Evaluation 70 : (2021) ID 101065 which has been published in final form https://doi.org/10.1016/j.stueduc.2021.101065. Copyright © 2021 The Author(s) Published by Elsevier under CC BY-NC-ND licence ( https://creativecommons.org/ licenses/by-nc-nd/4.0/) 2 The effect of the Free High School Tuition law on upper-secondary school choice in Japan Japan is one of the most homogeneous countries in the world. Japanese is the uniform, universally used language in the country, and the education system is based on high-quality public and private schools. The private education sector has some curiosities. It is generally used by higher classes looking for prestige, as is usual in other countries, even though quality and selectivity indices typically indicate that high-quality education is also offered in public schools (James et al., 1988). Prestigious private upper-secondary schools are generally located in urban areas. In prefectures in urban areas, the highest-ranked public and private uppersecondary schools are competitive, whereas, in prefectures in suburban areas, there are generally many public upper-secondary schools that are ranked higher than the highest-ranked private upper-secondary schools. The upper-secondary school choice in Japan seems to be a decisive point affecting students’ future job opportunities and social status. The quality of the chosen upper-secondary school to a large extent determines students’ prospective opportunities, and that is why this decision can be seen as the start of a long-term competition for a professional career. According to rational choice theory, individuals are conscious decision makers, and their choices are influenced by a cost–benefit analysis. This analysis of the upper-secondary school choice decision in Japan was based on students’ socio-economic background, educational aspirations, and previous academic achievement. These aspects can cause class-specific decisions that may contribute to educational as well as social inequalities. The aim of this study is to conduct a detailed analysis of the first of these three aspects of the upper-secondary school choice decision, that is, of students’ socio-economic background. The first specific goal of this study is to analyse the factors that lead families to choose a specific upper-secondary school in Japan. The second specific goal is to determine 3 the impact of the new law called “Free High School Tuition”, which was introduced in 2010. We use data on the school and family characteristics included in the PISA (OECD, 2009, 2012) questionnaires and apply a discrete choice model to analyse school choices before and after the implementation of this law and the way in which these choices vary depending on students’ socio-economic background and an urban or a non-urban location. These PISA data categorise Japanese schools according to two dimensions: their orientation (academic/vocational) and their funding (public/private). We analyse the relationship between those two school dimensions and the socio-economic characteristics of the students’ family. PISA data have been widely used in recent years to study similar and related topics. For example, school choice and school efficiency were investigated by Montes and Rubalcaba (2014), who assessed the extent to which choice and choice equity mechanisms influenced school efficiency. Another recent analysis based on PISA data was performed by Givord (2019), who showed that the share of students whose schools admitted students based on their home address reduced by approximately 20% in Japan between 2000 and 2015. A typical study based on students’ achievement was that of Sakellariou (2017), who analysed the private versus public school performance in mathematics for 40 countries and showed, in the case of Japan, a public school advantage. The paper is organised as follows. Section 2 describes the education system in Japan, Section 3 presents the data and methods, and Section 4 discusses the empirical results. Finally, Section 5 presents the main conclusions. The Education System in Japan Japan has a single-track comprehensive school system that is similar to the US model (Entrich, 2015). The highly competitive and relatively rigid education system in Japan is compulsory at the first two stages, which are elementary school (shōgakkō, for 6to 12-yearolds) and lower-secondary school (chūgakkō, for 12to 15-year-olds). After this stage, 4 students can continue in upper-secondary school (kōtōgakkō, for 15to 18-year-olds) or attend a college of technology (kōtō-senmon-gakkō). The upper-secondary school can be either vocational (senmongakka) or academic (futsūka). Afterwards, upper-secondary school graduates advance to universities or specialised training colleges, undertake vocational training, or enter the labour market directly. The percentage of those who enrol in universities differs significantly between academic and vocational upper-secondary school students. Colleges of technology were established in 1962, and, unlike universities or junior colleges, they accept those who have completed lower-secondary schooling. Students of these institutions are granted the title of associate (jun-gakushi) and may apply for admission to the upper division of university. Generally, these institutions focus on teaching specialised subjects, with the aim of helping students to develop the abilities that are required for vocational life (Ministry of Education, Culture, Sports, Science and Technology (MEXT), 2021a). Secondary schools in Japan do not usually include both stages of secondary education (lower secondary and upper secondary), so the vast majority of students have to change schools between these two stages. According to the Ministry of Education, Culture, Sports, Science and Technology (MEXT, 2012), only 3% of students in 2012 did not change schools between these stages. Nearly 1.2 million students in that year started their education at the upper-secondary level, but fewer than 5,000 students continued to study at a secondary school that had both lowerand upper-secondary divisions, and fewer than 30,000 students enrolled in an upper-secondary school that was operated by the same educational body as their lowersecondary school or that had a close relationship with their lower-secondary school. Admission to the upper-secondary school system plays an important role. After having finished their compulsory education, students are admitted to highor low-prestige uppersecondary schools depending on their scores in compulsory entrance examinations. Schools at 5 the upper-secondary level are valued by their reputation, that is, according to the percentage of their students who pass the difficult entrance examinations to the most prestigious universities in the country (Stevenson & Baker, 1992). Reputable upper-secondary schools receive many applications; therefore, students applying to these schools need higher exam scores. Thus, schools are ranked by the minimum exam scores to be admitted, and students apply for schools that fit their academic achievements. The system therefore represents a type of hierarchical academic ranking system (Kariya & Rosenbaum, 1999). However, in recent years there have been certain changes to the admission process for upper secondary schools. For example, Entrich (2019, p 275) recently wrote: "From 1997 onwards, students’ individual motivation and aptitudes were given more weight for high school admission through the evaluation of practical or technical examinations, essays, and interviews, and by stronger considering extra-curricular activities and recommendation letters. Additionally, the catchment areas for high school admission were expanded, wherefore students can choose from a larger range of high schools now and take more than one entrance examination in the same year." We investigate how the upper-secondary school choice is related to the socioeconomic characteristics of the family and how this relationship changed in 2010 with the implementation of the new “Free High School Tuition” law. This direct approach would not be valid for the elementary school choice because elementary schools and lower-secondary schools are compulsory and most students are allocated to public schools within their school district. Therefore, some parents are willing to pay more for housing in specific school districts. Kuroda (2018), for example, showed that parents exhibited higher willingness to pay housing rent in a better public elementary school district. However, this has been shown only for married couples who have children attending or expecting to attend elementary schools. Similarly, Ushijima and Yoshida (2009) and Yoshida et al. (2008) analysed the effect of 6 school quality on land prices. They concluded that the quality of elementary schools influences land prices in the school district but that this effect varies depending on the time period analysed and the type of district. A possible relationship between the type of uppersecondary school and students’ socio-economic characteristics has been studied indirectly in the following two streams of research. The first stream is based on students’ test scores. In this literature, the variances of scores are decomposed into between-school and in-school variances. If students were allocated to schools based on their achievement level, then a high between-school variance in scores would be expected. This was shown, for example, by Knipprath (2010), in an analysis of the mathematics and science scores of PISA 2000, 2003, and 2006, and by Taki (2011), in a study focusing on the mathematics scores of PISA 2003. Although these studies showed that the students in the same upper-secondary schools have similar scores, they did not directly model school choice. Nevertheless, Knipprath (2010) found correlations between students’ performance in mathematics and science and their socio-economic status at the high school level and concluded that “The PISA studies also showed that Japanese students are allocated to high schools according to their achievement level and their economic, social, and cultural background” (Knipprath, 2010, p. 403). The second stream of research focuses on the quality of schools. Defining the quality of a school is not a straightforward task, which is why the definitions differ among studies. Yamamoto and Brinton (2010) analysed the 1995 Social Stratification and Mobility Survey and defined the quality of upper-secondary schools using respondents’ reports on the proportion of classmates who proceeded to junior college or university. They concluded that the top-ranked secondary schools were chosen by families with a higher socio-economic status. Nakanishi (2011) applied a different approach and defined the quality of an uppersecondary school using students’ self-reported achievement at lower-secondary schools, 7 concluding that the achievement, and therefore the quality of the school, is highly related to the father’s socio-economic characteristics. Recently, Entrich (2019), using the Hyōgo High School Students (HHSS) survey (Ojima & Aramaki, 2018), not only related the school ranking to classical school choice drivers, like parents’ socio-economic status, students’ academic performance, and institutional constraints, but also showed that a significant share in the school choice decision is made by individual students themselves. This was achieved by showing that students’ concrete future plans significantly affected their decision making at the transition to high school. There is also a vast literature focusing on the importance of tracking, that is, the early determination of whether students will follow an academic track or a vocational track, in the Japanese system. Taki (2010, p. 247) concluded that “Japan is a country having the distinct characteristic wherein almost all the relevance between SES [socio-economic status] and the academic performance is being converted into inter-school disparity by the high school entrance exam”. Similarly, Hallinan (1994) and Oakes (1994) observed an impact of tracking on achievement, attitudes, and educational attainment. They also concluded that tracking is related to students’ economic, social, and cultural background; the school environment; and the classroom climate, among other variables. This is an important fact because, subsequently, students from low-ranked schools are unlikely to enter competitive universities (Ono, 2001). Moreover, students with a higher socio-economic status usually have easier access to shadow education lessons that help them to improve their academic performance, which in turn leads to admission to competitive universities. Matsuoka (2015), for example, indicated that, in 2007, the percentage of students attending cram schools (juku) to learn academic subjects increased heavily in the period when students took high school entrance examinations, reaching 50.9% for eighth-graders and 65.4% for ninth-graders. Furthermore, more recent information has shown that the importance of shadow education is not 8 decreasing. According to e-Stat (2018), 79.8% of ninth-graders in public schools attend cram schools that teach academic subjects, and their estimated average annual expenditure is JPY 393,000. One of the most important and comprehensive studies on this topic is Entrich (2018), who conducted an analysis of the impact of shadow education on social inequality formation in Japan based on several detailed empirical analyses. This work examined the reasons for the high Japanese enrolment rates in cram schools and private tutoring, together with their causes and their implications for social inequality. There is also an extensive literature focusing specifically on the choice of high school. Fujihara (2012), for example, developed two hypotheses related to this topic. The first is the relative risk aversion hypothesis (Breen & Goldthorpe, 1997), and the second is the downward educational mobility aversion hypothesis (Kikkawa, 2006). These hypotheses were tested with data on second-year students at Japanese senior high schools. As expected, the results showed that fathers’ occupation and parental education had direct effects on the rank of high school attended. An important conclusion with respect to the aim of our work was drawn by Kariya (2016), who investigated the operation of mechanisms that have produced social inequality in education over recent decades. Specifically, his work analysed three data sets collected in three different decades focusing on the relationship between the hierarchy of Japanese senior high schools and inequality. The results provided empirical evidence of an increasing impact of students’ family background on their academic grades and the rank positions of the high schools in which the students are enrolled. Moreover, the results showed a decreasing and indirect effect of the family background on the students’ allocation to high schools over recent decades. This indirect effect has occurred primarily through its influences on students’ academic achievement. Specifically, the “selection of students into different future SES strata 9 takes place visibly through educational differentiation at the upper secondary educational level” (Kariya, 2016, p. 151). Finally, Sakai (2010) analysed the career consciousness of students at an urban commercial high school and their possible motivations to attend highor middle-level universities. His results helped to provide an understanding of how the economic recession of the 1990s influenced schools’ policy on career guidance. Specifically, he concluded that the students’ consciousness lacked a concrete future perspective and that helping them to overcome this issue was consistent with the management strategy of the school. Upper-secondary school is not compulsory, but the vast majority of all lowersecondary school graduates continue their studies at either public or private upper-secondary schools. According to MEXT (2021b), the percentage of private upper secondary students in 2010 was 29.8%. Neither public nor private schools are free, but the cost of public uppersecondary schools is lower than that of private upper-secondary schools. In April 2010, the Japanese Government made public upper-secondary schools tuition free. At the same time, students at private upper-secondary schools started to receive an amount equivalent to the tuition fee at public upper-secondary schools as a subsidy. However, the household expenditure on education per student (including private school tuition after the subsidy and other expenses for school education and extracurricular school activities) for private school students is two to three times the expenditure for public school students. Since April 2014, a household income limit has been applied to determine eligibility to receive the subsidy. Modifications are continuously being made to this system. The goal of the tuition-free high school programme was stated as follows: “Minimizing the financial burden on households to ensure that all motivated high school students can feel secure about receiving education, irrespective of the financial situations of their families, is an issue that needed to be tackled urgently” (MEXT, 2009). The justification 16 Table 2 Summary Statistics of the Explanatory Variables Variable 2009 2012 City and large city Town and small town City and large city Town and small town Number of students 3,394 1,495 3,608 1,368 Dummy variables Proportion Proportion Father not working 0.07 0.08 0.07 0.07 Mother not working 0.66 0.59 0.65 0.60 Two-parent family 0.91 0.91 0.92 0.91 Siblings 0.88 0.91 0.87 0.89 Grandparents living with the family 0.27 0.40 0.25 0.34 Values Discrete variables 1 2 3 4 5 6 2009 Educ. level mother City and large city 0% 2% 12% 33% 25% 28% Town and small town 0% 4% 16% 38% 22% 19% Educ. level father City and large city 0% 4% 13% 26% 7% 50% Town and small town 0% 7% 18% 36% 7% 32% 2012 Educ. level mother City and large city 0% 2% 10% 33% 26% 29% Town and small town 0% 4% 16% 40% 22% 18% Educ. level father City and large city 0% 5% 12% 27% 8% 48% Town and small town 0% 7% 19% 35% 8% 31% City and large city Town and small town Continuous variables Mean S.D. Min. Max. Mean S.D. Min. Max. 2009 Cultural possessions -0.25 0.93 -1.39 1.24 -0.30 0.90 -1.39 1.23 Family wealth -0.43 0.70 -2.88 2.81 -0.43 0.71 -2.56 2.06 Home educ. resources -0.32 1.02 -4.53 1.60 -0.51 1.07 -4.53 1.60 Highest parental occupational status 52.78 14.27 23.00 80.00 50.2 15.20 23.00 80.00 2012 Cultural possessions -0.38 0.96 -1.51 1.27 -0.57 0.94 -1.51 1.27 Family wealth -0.24 0.63 -3.30 2.92 -0.16 0.62 -2.50 1.80 Home educ. resources -0.44 0.81 -3.93 1.12 -0.60 0.81 -3.93 1.12 Highest parental occupational status 52.30 20.07 11.56 88.70 46.32 19.95 11.56 88.70 17 Another index variable coded by PISA 2012 is Home educational resources, which is constructed using answers to questions concerning, for instance, whether students have a desk and a quiet place to study, a computer for schoolwork, educational software, books to help with their schoolwork, technical reference books, and dictionaries. As can be seen in Table 2, for our study, this index ranged between -4.53, 1.60 and -3.93, 1.12 for 2009 and 2012, respectively. Finally, the highest parental occupational status shows the higher International Socio-Economic Index of occupational status score of either parent or, in the case of singleparent households, of the only available parent. As stated above, Bukodi and Goldthorpe (2013) decomposed the social origins into parental class, parental status, and parental education. In our case, the parental class can be represented by Family wealth, the parental status by Highest parental occupational status, and the parental education by Educ. level mother and Educ. level father. The estimation of a multinomial logit model for a school choice using PISA data could seem to be a non-standard approach, as PISA studies usually take into account the complex sample structure through the use of multilevel models and related techniques. Nevertheless, these studies are usually focused on the collected students’ achievements, which may depend on the schools’ and students’ characteristics, and the complex sample structure must be taken into account. In our case, the explained variable (school choice) was determined prior to the data collection; therefore, the methodology does not need this multilevel approach. Empirical Results We estimate multinomial logit models by maximum likelihood using the 2009 and 2012 samples for the two city/town areas to analyse the reasons that drive Japanese parents’ and students’ decision to choose a certain type of upper-secondary school. The dependent variable in our model is a categorical variable, Type of school, with four different values, as defined in Table 1. The explanatory variables are those included in Table 2. We checked for 18 possible multicollinearity among these variables but found very low interdependency. Tables A1 and A2 in the Appendix present the estimation outcome of the model with 36 parameters obtained for the two areas and two years. To summarise this outcome, we present the likelihood ratio tests for all the explanatory variables in Table 3. With four dependent categories in our dependent variable, there are three sets of parameters associated with each explanatory variable. The joint hypothesis that an explanatory variable does not affect the dependent variable (school type) therefore involves the set of three coefficients corresponding to each specific variable being equal to zero. Table 3 Likelihood Ratio Tests of the Explanatory Variables 2009 2012 City and large city Town and small town City and large city Town and small town 𝜒V statistic pvalue 𝜒V statistic pvalue 𝜒V statistic pvalue 𝜒V statistic pvalue Father not working 10.43 0.02 2.34 0.51 0.19 0.98 10.04 0.02 Mother not working 12.91 0.01 0.56 0.91 10.20 0.02 3.25 0.35 Two-parent family 2.42 0.49 8.06 0.05 1.47 0.69 6.74 0.08 Siblings 15.30 0.00 1.74 0.63 13.01 0.01 6.63 0.09 Grandparents living with the family 7.12 0.07 5.39 0.15 3.04 0.39 4.85 0.18 Cultural possessions 24.00 <0.01 6.42 0.09 24.52 0.00 18.98 0.00 Educ. level mother 58.51 <0.01 20.67 <0.01 46.51 0.00 12.09 0.01 Educ. level father 119.12 <0.01 34.51 <0.01 78.29 0.00 19.64 0.00 Family wealth 11.72 0.01 11.57 0.01 3.26 0.35 1.57 0.67 Home educ. resources 0.46 0.93 7.40 0.06 3.26 0.35 5.67 0.13 Highest parental occupational status 11.45 0.01 4.30 0.23 24.82 0.00 3.51 0.32 The specific effects of each variable on the school choice are discussed below, but probably the most important result in Table 3, related directly to our goal to analyse the impact of the new law, can be observed for Family wealth. Although its effect is significant at the 5% level in both city/town areas in 2009, prior to the passing of the law, it is no longer significant in 2012. This is in line with the conclusion reached by Hori and Shimizutani (2018) that the tuition-free programme enhanced the high school enrolment rate for lower-income households; 19 that is, it gave them incentives to send their children to high school thanks to the exemption from the tuition payment. Another important part of the interpretation of our results is based on changes in the probability defined in equation (1) of choosing a specific type of school depending on different values of our explanatory variables ( 𝑥$ ). Figure 1 shows the change in the probability of choosing a certain type of school in the two analysed years and areas if one of the explanatory variables changes its value. In this comparison, the remaining explanatory variables are set to a representative value, taking into account the descriptive statistics presented in Table 2, by setting them to a “benchmark family”. Representative values for the dummy variables are determined based on the majority rule, that is, the father is working, the mother is not working, it is a two-parent family with one or more siblings, and there are no grandparents living with the family. Moreover, the discrete variable for fathers’ and mothers’ education level is set to a median value (5), while the values for the continuous explanatory variables (Family wealth, Cultural possessions, Home educ. resources, and Highest parental occupational status) are set to their corresponding mean value. All the changes presented in Figure 1 are calculated as discrete changes (i.e. 0/1) for the dummy variables, as the unit change for the median value of fathers’ and mothers’ educational level (i.e. from 5 to 6), and as a standard deviation change for the continuous variables (i.e. from (mean - st. dev./2) to (mean + st. dev./2)). More details can be found in the study by Long and Freese (2005). Figure 1 allows for several comparisons. Given our goal to analyse the impact of the new law, the most relevant results for us are the differences between 2009 and 2012. We can observe that, for example, a change from zero to one in the Father not working variable in the city area in 2009, that is, a change from a family with a working father to a family with a nonworking father, decreases the probability of families choosing the private and academic (outcome 3) type of school by about 0.08, but in 2012 the effect of this variable is close to 20 zero. On the other hand, this variable increases the probability of choosing a public and academic (outcome 1) type of school in the city area in 2009 by about 0.06, but this change in 2012 is negative and close to zero. Figure 1 Change in Probability with Respect to the Benchmark Family 2009 2012 City and large city Town and small town 1 – Public and Academic 3 – Private and Academic 2 – Public and Vocational 4 – Private and Vocational To interpret the most relevant results in Figure 1, we will focus first on the city area. Comparing all of the changes and taking into account that there are zero/one changes, unit changes, and standard deviation changes in the mix, we can see that the biggest changes in probability come from the variable Siblings, for both 2009 and 2012. That is, having more than one child increases the probability of choosing the public and academic type of school (outcome 1) by about 0.08 but decreases the probability of choosing the private and academic (outcome 3) type of school by about 0.10 in both years. Change in Predicted Probability for SCTYPEREC -.11 -.07 -.04 0 .04 .07 .11 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 Father not working-0/1 Mother not working-0/1 Two-parent family-0/1 Siblings-0/1 Grandparents-0/1 Educ. level mother Educ. level father Family wealth-std Cultural possessions-std Home educ. resources-std Parental occup. status-std Change in Predicted Probability for SCTYPEREC -.11 -.07 -.04 0 .04 .07 .11 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 Father not working-0/1 Mother not working-0/1 Two-parent family-0/1 Siblings-0/1 Grandparents-0/1 Educ. level mother Educ. level father Family wealth-std Cultural Possessions-std Home educ. resources-std Parental occup. status-std Change in Predicted Probability for SCTYPEREC -.11 -.07 -.03 .01 .05 .09 .13 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 Father not working-0/1 Mother not working-0/1 Two-parent family-0/1 Siblings-0/1 Grandparents-0/1 Educ. level mother Educ. level father Family wealth-std Cultural possessions-std Home educ. resources-std Parental occup. status-std Change in Predicted Probability for SCTYPEREC -.11 -.07 -.04 0 .04 .07 .11 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 Father not working-0/1 Mother not working-0/1 Two-parent family-0/1 Siblings-0/1 Grandparents-0/1 Educ. level mother Educ. level father Family wealth-std Cultural Possessions-std Home educ. resources-std Parental occup. status-std 21 Moreover, the part of Figure 1 devoted to the city area indicates that, approximately, the group of variables with the second most important effect in 2009 (behind the largest Siblings effect) can be considered to be Father not working and, as expected from the literature review, Educ. level mother and Educ. level father. It is also important to note that the effects of the variables Father not working and Mother not working are different. According to Table 2, in the city area, the mother is not working in 66% and 65% of the families in 2009 and 2012, respectively, and the effect of this variable in the city area in Figure 1 remains relatively stable between 2009 and 2012. On the other hand, the father is working in 93% of the families in the two analysed years (Table 2), and, as mentioned earlier, its effect in Figure 1 changes drastically between 2009 and 2012. This shows close dependence of the school choice on the family budget, which, according to this result, seems to be linked to fathers’ but not mothers’ employment status in the city area. Therefore, Father not working seems to be an indicator of a low-income household. On the other hand, Mother not working seems to represent a broader family characteristic than the work status per se – for example, a sufficiently high family income that allows this mother’s employment status. A non-working mother, therefore, is likely to indicate a specific family class. It also highlights the fact that the change in Mother not working is similar to a standard deviation increase in Cultural possessions, both implying a higher probability of choosing an academic but a lower probability of choosing a vocational type of school. As expected from the literature, the education level of the father and mother also have an important effect on the type of school chosen. A unit change in these variables (from 5 to 6) increases the probability of choosing the private and academic (outcome 3) type of school in the city area and in the two years by approximately 0.04–0.07, but it decreases the probability of choosing the public and vocational (outcome 2) type of school by about 0.03– 0.04. 22 A typically observed effect in the literature is that higher wealth leads to a higher probability of choosing a private and academic school. This result, represented by the effect of the variable Family wealth, is observed only in 2009. A standard deviation increase raises the probability of choosing the private and academic (outcome 3) type of school by approximately 0.03 in the city area, but it decreases the probability of choosing the public and academic (outcome 1) type of school by approximately the same amount. The comparison of the changes in probabilities between 2009 and 2012 is expected to shed light on the effect of the “Free High School Tuition” law implemented in 2010. The differences in the effects of the explanatory variables between 2009 and 2012 stay the same for some variables and vary for others. Continuing to focus on this in the city area, the largest Siblings effect stays very similar in 2009 and in 2012. This is a very important conclusion because the corresponding part of the family budget devoted to tuition costs obviously increases with the number of children and the fact that this effect remains constant in the two years seems to suggest that the new law has a more mitigated effect on families with more than one child. However, this stable effect of Siblings is not observed for other variables related to family budget. The negative effect on the probability of choosing the private and academic type of school (outcome 3) in the city area of Father not working (a low-income household indicator) observed in 2009 diminishes in 2012. This is confirmed by the fact that its effect is significant at the 5% level in 2009 but not significant at the 5% level in 2012 (Table 3). Interestingly enough, the same result is obtained in Table 3 for Family wealth, a fact that is also reflected graphically in Figure 1 as its effect diminishes in 2012 in the city area. Therefore, according to Table 3 and Figure 1, the effect of the two variables directly or indirectly related to the family budget (Father not working and Family wealth) decreased significantly in 2012. This seems to be a direct effect of the new law’s implementation in 2010. 23 The results are slightly different for the town area. The difference in the school offer as well as in the social class composition with respect to the city area becomes apparent in the descriptive statistics in Tables 1 and 2. Regarding the school offer, the largest differences are represented by a greater proportion of public and vocational and a lower proportion of private and academic types of schools in the town area. When it comes to the socio-economic characteristics, the largest differences are represented by a greater share of grandparents living with the family and a lower education level of both parents in the town area. Focusing on Figure 1, the biggest changes in probability in the town area come from the variables Father not working and Two-parent family. The effect of Father not working in the town area, with a lower offer of private schools, seems to be different from that in the city area, with a much larger share of private schools (Table 1). In this case, its effect increases the probability of choosing the vocational type and decreases the probability of choosing the academic type of school in 2012. This effect is not significant at the 5% level in 2009 (Table 3). This kind of family does not seem to be affected by the new law as the decision seems to be more between academic and vocational schools than between public and private schools. This is probably because of a lower offer of private schools locally; considering private schools in cities nearby would be linked to higher time and travel costs. Moreover, the situation of a non-working father in this more rural area could be associated with a desire for faster incorporation of the offspring into the labour market. In spite of the large effect in Figure 1, the effect of the Two-parent family is only marginally significant in 2009 and not significant at the 5% level in 2012 (Table 3). However, similar to the city area, the effects of Educ. level mother and Educ. level father are both significant at the 5% level (Table 3). Again, it appears that, given the lower offer of private schools, the preference of parents with higher education is more for the academic than for the 24 vocational type of schools, unlike the preference for private over public schools observed in the city area. On the other hand, an important result is represented by the same effect of Family wealth as obtained in the city area. Its effect on the probability of choosing a private and academic school in 2009 is positive, whereas it is negative for the public academic type of school. This effect practically disappears (Figure 1) and becomes not significant at the 5% level in 2012 (Table 3). These are the same results that were obtained for the city area. Therefore, the direct effect of the new law can be observed both in the city and in the town area. Figure 2 shows the probability of choosing a specific type of school for specific values of the variables Educ. level father (as an example of a variable with similar effects in the two years considered) and Family wealth (as an example of a variable with different effects in the two years). The remaining variables in Figure 2 are set at the benchmark family’s values, as previously specified in Figure 1. Focusing on the graphs showing the effect of the father’s education, the biggest changes in the probability of choosing a specific type of school belong, in the city area, to the private–academic and public–vocational types of schools. For the lowest levels of the father’s education, the probability of choosing the public and vocational type of school is higher than the probability of choosing the private and academic type of school. This situation is reversed for the higher levels of education. These effects are almost the same in 2009 and 2012. The same results for the town area are represented by the downward trend of the probability of choosing the public and vocational type of school. The expected upward trend for the private and academic types of school is not observed due to their limited offer in the town area. The most important result is that the trends of all four probabilities are very similar in the two areas and in the two analysed years. 25 Figure 2 Effect of the Father’s Education and the Family Wealth on the Probability of School Choice Father’s education 2009 2012 City and large city Town and small town Family wealth 2009 2012 City and large city Town and small town The effect of family wealth, however, changes in 2012 compared with 2009 in the two areas (lower part of Figure 2). In 2009, having greater family wealth increases the probability of choosing private schools but decreases the probability of choosing public schools both in 0.1 .2 .3 .4 .5 1 2 3 4 5 6 Educ. level father Pub Academic Pub Practical Pri Academic Pri Practical Probability 0.1 .2 .3 .4 .5 1 2 3 4 5 6 Educ. level father Pub Academic Pub Practical Pri Academic Pri Practical Probability 0.2 .4 .6 1 2 3 4 5 6 Educ. level father Pub Academic Pub Practical Pri Academic Pri Practical Probability 0.2 .4 .6 .8 1 2 3 4 5 6 Educ. level father Pub Academic Pub Practical Pri Academic Pri Practical Probability 0.2 .4 .6 -2 0 2 Family wealth Pub Academic Pub Practical Pri Academic Pri Practical Probability 0.1 .2 .3 .4 .5 -2 0 2 Family wealth Pub Academic Pub Practical Pri Academic Pri Practical Probability 0.2 .4 .6 .8 -2 0 2 Family wealth Pub Academic Pub Practical Pri Academic Pri Practical Probability 0.2 .4 .6 .8 -2 0 2 Family wealth Pub Academic Pub Practical Pri Academic Pri Practical Probability 32 Ishida, K., Nakamuro, M., & Takenaka, A. (2016). The academic achievement of immigrant children in Japan: An empirical analysis of the assimilation hypothesis. Educational Studies in Japan, 10, 93–107. doi:10.7571/esjkyoiku.10.93. James, E., Benjamin, G. R., & Mendras, M. (1988). Public policy and private education in Japan. Springer. Kariya, T. (2016). 10 understanding structural changes in inequality in Japanese education. In D. Chiavacci & C. Hommerich (Eds.), Social inequality in post-growth Japan: Transformation during economic and demographic stagnation (pp. 149–165). Routledge. Kariya, T., & Rosenbaum, J. E. (1999). Bright flight: Unintended consequences of detracking policy in Japan. American Journal of Education, 107(3), 210–230. doi:10.1086/444216. Kikkawa, T. (2006). Education and social inequality: Contemporary educational credentialism in Japan [Gakureki to Kakusa Fubyodo: Seijuku Suru Nihon-Gata Gakureki Shakai]. Tokyo Daigaku Shuppan. Knipprath, H. (2010). What PISA tells us about the quality and inequality of Japanese education in mathematics and science. International Journal of Science and Mathematics Education, 8(3), 389–408. doi:10.1007/s10763-010-9196-5. Kuroda, Y. (2018). The effect of school quality on housing rents: Evidence from Matsue City in Japan. Journal of the Japanese and International Economies, 50, 16–25. doi:10.1016/j.jjie.2018.05.002. Long, J. S., & Freese, J. (2005). Regression models for categorical dependent variables using Stata (2nd ed.). Stata Press. 33 Matsuoka, R. (2015). School socioeconomic compositional effect on shadow education participation: Evidence from Japan. British Journal of Sociology of Education, 36(2), 270–290. doi:10.1080/01425692.2013.820125. McFadden, D. (1974). Conditional logit analysis of qualitative choice behaviour. In P. Zarembka (Ed.), Frontiers in econometrics (pp. 105–142). Academic Press. Ministry of Education, Culture, Sports, Science and Technology (MEXT). (2009). Making public high schools tuition-free and high school enrollment subsidies. Retrieved from https://warp.da.ndl.go.jp/info:ndljp/pid/11402417/www.mext.go.jp/b_menu/hakusho/h tml/hpab200901/detail/1305888.htm. Accessed April 27, 2021. Ministry of Education, Culture, Sports, Science and Technology (MEXT). (2012). School basic survey. Retrieved from https://www.e-stat.go.jp/statsearch/files?page=1&toukei=00400001&tstat=000001011528&year=20120. (In Japanese). Accessed April 27, 2021. Ministry of Education, Culture, Sports, Science and Technology (MEXT). (2020). Survey on issues related to student advisory including problematic behaviours and truancy. Retrieved from https://www.e-stat.go.jp/statsearch/files?page=1&layout=normal&toukei=00400304&tstat=000001112655&result _page=1 (in Japanese). Accessed April 27, 2021. Ministry of Education, Culture, Sports, Science and Technology (MEXT). (2021a). Overview. Retrieved from https://www.mext.go.jp/en/policy/education/overview/index.htm. Accessed April 27, 2021. Ministry of Education, Culture, Sports, Science and Technology (MEXT). (2021b). Private school in Japan. National Institute for Educational Policy research. Retrieved from https://www.nier.go.jp/English/educationjapan/. Accessed July 11, 2021. 34 Montes, O., & Rubalcaba, L. (2014). School choice, equity and efficiency: International evidence from PISA-2012. Investigaciones de Economía de la Educación. In A. García & I. Neira (Eds.), Investigaciones deeconomía de la educación 9 (Volume 9, Chapter 31, pp. 585–614. Asociación de Economía de la Educación. Nakanishi, H. (2011). Reexamination of Japanese educational selection system: Focusing on stratified contest mobility. Journal of Child Study, 17, 69–81. Oakes, J. (1994). Ability grouping and tracking in schools. In P. Husen (Ed.), The international encyclopedia of education, 1, 6–12. OECD. (2009). PISA 2009 technical report. Retrieved from https://www.oecd.org/pisa/data/pisa2009technicalreport.htm. Accessed April 27, 2021. OECD. (2012). PISA 2012 technical report. Retrieved from http://www.oecd.org/pisa/data/pisa2012technicalreport.htm. Accessed April 27, 2021. Ojima, F., & Aramaki, S. (2018). Kōkōsei-tachi no yukue – gakkō paneru chōsa kara mita shinro to seikatsu no 30nen [High school students’ pathways – 30 years of life course and school life trajectories as seen from a school panel survey]. Ono, H. (2001). Who goes to college? Features of institutional tracking in Japanese higher education. American Journal of Education, 109(2), 161–195. doi:10.1086/444265. Sakai, A. (2010). Educational selection, career decisions, and school support: The case of an urban commercial high school. In Challenges to Japanese education: Economics, reform, and human rights (pp. 87–106). Teachers College (New York). Sakellariou, C. (2017). Private or public school advantage? Evidence from 40 countries using PISA 2012-Mathematics. Applied Economics, 49(29), 2875–2892. doi:10.1080/00036846.2016.1248361. 35 Stevenson, D. L., & Baker, D. P. (1992). Shadow education and allocation in formal schooling: Transition to university in Japan. American Journal of Sociology, 97(6), 1639–1657. Taki, H. (2010). The relationship between socio-economic status and academic achievement: Exploring specific characteristics of education and inequality in Japan. Sociological Theory and Methods, 25(2), 229–248. doi:doi.org/10.11218/ojjams.25.229. Taki, H. (2011). School track and inequality in Japanese upper secondary schools: An analysis using PISA data. Soshioroji, 55(3), 37–52. doi:doi.org/10.14959/soshioroji.55.3_37. Tsuneyoshi, R. (2013). Junior high school entrance examinations in metropolitan Tokyo: The advantages and costs of privilege. In G. DeCoker & C. Bjork (Eds.), Japanese education in an era of globalization: Culture, politics, and equity (pp. 164–182). New York: Teachers College Press, Columbia University. Ushijima, K., & Yoshida, A. (2009). Shougakkou Ni Okeru Kyouiku No Shitsu Ha Chika Ni Eikyou Suruka? – Tokyoto Tokubetsuku No Chika Data Wo Mochiita Kenshou. Journal of Applied Regional Science, 14, 37–47. Yamamoto, Y., & Brinton, M. C. (2010). Cultural capital in East Asian educational systems: The case of Japan. Sociology of Education, 83(1), 67–83. Yoshida, A., Zhang, L., & Ushijima, K. (2008). Gakkou No Shitu To Chika – Adachiku No Chika Data Wo Mochiita Kenshou. Jutaku Toshi Keizai, 68, 10–18. 36 Appendix Table A1 Multinomial Logit Model Estimation: 2009 City and large city Town and small town Coef. Std error Coef. Std error Category: 1 – Public and academic Father not working -0.491 0.247 ** 0.094 0.555 Mother not working 0.359 0.166 ** 0.005 0.288 Two-parent family 0.375 0.246 1.018 0.436 ** Siblings 0.174 0.250 0.454 0.412 Grandparents living with the family -0.104 0.181 0.257 0.304 Cultural possessions 0.181 0.098 * 0.160 0.170 Educ. level mother 0.176 0.083 ** 0.117 0.145 Educ. level father 0.222 0.067 *** -0.036 0.124 Family wealth -0.262 0.125 ** -0.462 0.220 ** Home educ. resources 0.001 0.088 -0.249 0.148 * Highest parental occupational status 0.009 0.006 0.012 0.010 Constant -0.710 0.559 0.145 0.930 Category: 2 – Public and vocational Father not working -0.631 0.279 ** 0.405 0.569 Mother not working 0.044 0.180 0.080 0.302 Two-parent family 0.256 0.266 0.754 0.454 * Siblings 0.178 0.275 0.438 0.437 Grandparents living with the family 0.010 0.196 0.166 0.317 Cultural possessions -0.104 0.108 -0.003 0.179 Educ. level mother 0.024 0.091 -0.178 0.152 Educ. level father -0.095 0.073 -0.367 0.130 *** Family wealth -0.174 0.136 -0.361 0.231 Home educ. resources 0.030 0.096 -0.375 0.155 ** Highest parental occupational status 0.002 0.007 0.012 0.011 Constant 0.836 0.606 2.095 0.969 ** Category: 3 – Private and academic Father not working -0.876 0.280 *** 0.024 0.593 Mother not working 0.393 0.175 ** 0.099 0.309 Two-parent family 0.357 0.269 0.444 0.468 Siblings -0.308 0.258 0.623 0.460 Grandparents living with the family -0.312 0.191 -0.104 0.326 Cultural possessions 0.192 0.103 * -0.012 0.183 Educ. level mother 0.431 0.088 *** 0.134 0.156 Educ. level father 0.429 0.072 *** -0.099 0.133 Family wealth -0.062 0.130 -0.114 0.236 Home educ. resources 0.026 0.092 -0.324 0.158 ** Highest parental occupational status 0.015 0.007 ** 0.003 0.011 Constant -3.289 0.599 *** 0.067 1.005 37 Category: 4 – Private and vocational Base category Base category Log-likelihood -3670.0 -1542.7 Number of parameters 36 36 Observations 3394 1495 ***, **, and *: significance at the 1%, 5%, and 10% levels. 38 Table A2 Multinomial Logit Model Estimation: 2012 City and large city Town and small town Coef. Std error Coef. Std error Category: 1 – Public and academic Father not working -0.089 0.311 -0.673 0.520 Mother not working 0.260 0.171 -0.453 0.369 Two-parent family 0.058 0.305 0.966 0.489 ** Siblings 0.335 0.228 0.771 0.418 * Grandparents living with the family 0.193 0.203 0.130 0.400 Cultural possessions 0.318 0.100 *** 0.092 0.203 Educ. level mother 0.092 0.087 0.150 0.171 Educ. level father 0.192 0.071 *** -0.092 0.146 Family wealth -0.079 0.146 0.326 0.307 Home educ. resources -0.199 0.111 * -0.249 0.230 Highest parental occupational status 0.006 0.004 -0.005 0.009 Constant 0.105 0.555 1.938 1.088 * Category: 2 – Public and vocational Father not working -0.032 0.328 -0.053 0.528 Mother not working -0.017 0.182 -0.597 0.378 Two-parent family -0.035 0.319 0.550 0.502 Siblings 0.389 0.247 0.835 0.440 * Grandparents living with the family 0.338 0.214 0.333 0.408 Cultural possessions 0.181 0.108 * -0.205 0.210 Educ. level mother -0.131 0.093 -0.082 0.176 Educ. level father -0.110 0.076 -0.295 0.150 ** Family wealth -0.172 0.156 0.305 0.316 Home educ. resources -0.171 0.119 -0.054 0.236 Highest parental occupational status -0.001 0.005 -0.007 0.010 Constant 1.949 0.587 *** 3.250 1.118 *** Category: 3 – Private and academic Father not working -0.036 0.320 -1.411 0.879 Mother not working 0.292 0.175 * -0.373 0.435 Two-parent family 0.200 0.317 1.351 0.773 * Siblings -0.027 0.231 0.144 0.506 Grandparents living with the family 0.230 0.208 -0.284 0.484 Cultural possessions 0.409 0.102 *** 0.322 0.239 Educ. level mother 0.248 0.089 *** 0.126 0.207 Educ. level father 0.281 0.073 *** 0.121 0.180 Family wealth -0.011 0.149 0.457 0.372 Home educ. resources -0.192 0.114 * -0.114 0.277 Highest parental occupational status 0.013 0.005 *** 0.005 0.011 Constant -1.546 0.576 *** -1.484 1.403 39 Category: 4 – Private and vocational Base category Base category Log-likelihood -4022.1 -1159.7 Number of parameters 36 36 Observations 3608 1368 ***, **, and *: significance at the 1%, 5%, and 10% levels.