Recent empirical research in the social sciences has emphasized the importance of causal inference. However, causal inference is challenging when using observational data, primarily cross-sectional, even if it includes relevant variable information, as in international and large-scale educational surveys. The difficulty is more pronounced when the variable of interest, such as a national-level policy, is systemic. This paper demonstrates that by using pseudo-panel data derived from repeated cross-sectional data, we can obtain findings relevant to policy-making, thereby mitigating some of the challenges in causal inference, particularly biases from unobserved confounding factors.

   The specific topic addressed in this paper is the assessment of policy factors related to the youth's choice to teach. In general, improving the availability and quality of teacher personnel is a universal and important issue for public education policy (OECD 2018). These research areas concerning the choice of teaching career and teacher supply have been interdisciplinary in education (educational policy studies, sociology of education, educational psychology, etc.) and economics (economics of education, labor economics). In particular, empirical research on the basic issues of "who chooses to teach" and "what factors increase the number of people who want to teach" has been conducted in many countries. While educational and psychological research have pointed out the importance of psychological factors, work environment factors have not been recognized as the main factors influencing career choice (Watt et al. 2017). On the other hand, empirical studies in the economics of education and labor economics have focused exclusively on the impact of salary levels as a policy variable on entry and exit from the workforce and have partially argued for its contribution (Corcoran et al. 2004; Dolton 1990; Manski 1987).

   Moreover, Japan, where the presenter is from, has historically excelled in maintaining high-quality teachers, as evidenced by their high competency (Hanushek et al. 2019) and low turnover rates, compared to other countries. However, recent years have seen a growing trend among young people to avoid teaching careers. Japan now faces challenges similar to many countries experiencing a structural teacher shortage. Public debates often cite the relatively inferior work environment of teaching compared to other white-collar jobs as a factor in this avoidance. Yet, substantial evidence is lacking to inform policy priorities in this area.

   In this study, we position and extend the groundbreaking recent studies that have used PISA student-level data to analyze the youth’s choice of teaching profession (Park & Byun 2015; Han 2018) as important prior work. We differ from that study in terms of methodology, using pseudo-panel data composed of subpopulations of countries as units; we apply a cross-classified hierarchical model to ask "Which policy factors" promote "whose" entry into the teaching profession among young people? We specifically focus on policy factors related to the working environment, namely, the relative salary level of teachers compared to other professions and the workload of teachers (working hours, number of students per teacher, and time spent on non-teaching tasks).

   Applying a cross-classified hierarchical model to the pseudo-panel data, we respond to the question of "which policy factors" encourage "whom" of young people to enter the teaching profession, addressing both causal inference (controlling for time-invariant confounders) and policy relevance (heterogeneity of policy effects). The cross-classified model, which sets up the random effects/coefficients in two types of units, country, and subpopulation, has a major advantage in that it allows for different policy implications for each country. To further increase the robustness of our model, we are expanding it into a semiparametric model (infinite mixture model) that does not rely on a multivariate normal distribution for random effects and coefficients.

One problem with existing quantitative empirical studies of the choice of teaching profession and teacher supply is their weak consideration of causal inferences (especially in addressing unobserved confounding factors). This paper attempts to address these problems through an analysis using pseudo-panel data. Pioneering studies based on pseudo-panel data in education (but different from the topic of this paper) include Gustafsson (2008, 2013), who applied them to data from large-scale international surveys, and the ideas in this paper also rely on them. In this paper, we use student-level data from OECD member countries in the Programme for International Student Assessment (PISA) as data related to teacher choice. PISA survey data are usually used in empirical analyses with academic achievement as the outcome variable, but they have already been used in several studies of career choices because they include questions on items related to occupations in which students expect to be employed at age 30 (Park & Byun 2015; Han 2018; Han et al. 2018, 2020). Existing studies often rely on cross-section data from a specific time period. In contrast, our analysis uses pseudo-panel data compiled from multiple time points. As each PISA survey targets different respondents (15-year-old students from each country at each time point), it does not constitute individual-level panel data. However, by reorganizing this data into a subpopulation-based panel format, incorporating multiple attribute information, we can exploit the benefits of panel data, such as controlling for time-invariant confounding factors. In creating the pseudo-panel data, subpopulations were defined based on information about gender, parental occupation (whether the parent's occupation was in teaching or not), and cognitive ability (subdivided into 10 groups based on PISA scores). The aspiration rate of primary and secondary education teachers within each subpopulation is used as the dependent variable to clarify which policy factors related to the working environment each youth group strongly responds to, influencing their choice or rejection of the teaching profession. Policy factors concerning the working environment include 1) salary level, 2) teacher-student ratio, 3) working hours, and 4) the amount of non-teaching tasks, focusing on the national and temporal levels. The data on policy factors are based on country and time units. These data are analyzed using Bayesian cross-classified parametric/semi-parametric hierarchical models. By employing a cross-classified hierarchical model, we can assume that the effects of policy factors vary between countries and subpopulations, allowing us to obtain policy-relevant insights.

By utilizing Bayesian cross-classified hierarchical models on pseudo-panel data regarding the youth’s choice of teaching profession, we could analyze the impact of various policy factors related to the working environment. This approach allowed us to control for time-invariant confounding factors and clarify heterogeneity in the effects of each policy factor across different subpopulations and countries. Regarding overall trends, enhancing the working environment appears to motivate female students to choose teaching as a profession more than male students. Specifically, improvements in relative salary, student-teacher ratios, and reduced working hours significantly encourage highly qualified individuals to enter the teaching field. Concerning the effect's magnitude, we observed that a one standard deviation improvement in these factors increases the proportion of students aspiring to teach by 0 to 2 percentage points. However, for high-ability male students whose parents are not teachers, we found no significant incentive to pursue a career in teaching. While it is difficult to summarize the differences in policy effects across countries, focusing on Japan, which is the primary concern of the presenter, we find the relative salary level and relative working hours compared to other occupations have a stronger impact. Similarly, the analysis results can point to specific characteristics in other countries. These findings contrast with previous research in education and psychology on the choice of the teaching profession, which often underestimates the role of extrinsic factors due to the analogical application of motivational theories of learning. Our findings reveal that the working environment plays a crucial role in influencing young people's decisions to enter the teaching profession and in determining the overall supply of teachers. Moreover, they identify which policy factors will affect the quality of teacher supply.

Bryk, A. S., and S. W. Raudenbush (2002) Hierarchical Linear Models: Applications and Data Analysis Methods, Sage Publications. Condon, P. D. (2020) Bayesian Hierarchical Models with Applications Using R, 2nd edition, CRC Press. Corcoran, S. P., W. N. Evans, and R. M. Schwab (2004) “Women, the Labor Market, and the Declining Relative Quality of Teachers,” Journal of Policy Analysis and Management, 23(3): 449-470. Dolton, P. J. (1990), “The Economics of UK Teacher Supply: The Graduate's Decision,” The Economic Journal, 100: 91–104. Gustafsson, J. (2008) “Effects of International Comparative Studies on Educational Quality on the Quality of Educational Research,” European Educational Research Journal, 7(1):1-17. Gustafsson, J. (2013) “Causal Inference in Educational Effectiveness Research: A Comparison of Three Methods to Investigate Effects of Homework on Student Achievement,” School Effectiveness and School Improvement, 24(3): 275-295. Han, S. W. (2018) “Who Expects to Become a Teacher? The Role of Educational Accountability Policies in International Perspective,” Teaching and Teacher Education, 75:141–152. Han, S. W., F. Borgonovi, and S. Guerriero, (2018) “What Motivates High School Students to Want to Be Teachers? The Role of Salary, Working Conditions, and Societal Evaluations About Occupations in a Comparative Perspective,” American Educational Research Journal, 55(1): 3–39. Han, S. W., F. Borgonovi, and S. Guerriero (2020) "Why Don’t More Boys Want to Become Teachers? The Effect of a Gendered Profession on Students’ Career Expectations," International Journal of Educational Research, 103:101645. Hanushek, E. A., J. F. Kain, and S. G. Rivkin (2004) “Why Public Schools Lose Teachers”, Journal of Human Resources, 39(2): 326–354. Hanushek, E. A., M. Piopiunik, and S. Wiederhold (2019) “The Value of Smarter Teachers: International Evidence on Teacher Cognitive Skills and Student Performance,” Journal of Human Resources, 54(4), 857-899 Kleinman, K. P. and J. G. Ibrahim (1998) “A Semiparametric Bayesian Approach to the Random Effects Model," Biometrics, 54:921-938. Manski, C. F. (1987) “Teachers Ability, Earnings, and the Decision to Become a Teacher: Evidence from the National Longitudinal Study of the High School Class of 1972,'' in D. A. Wise ed., Public Sector Payrolls, University of Chicago Press. OECD(2018) Effective Teacher Policies: Insight from PISA, OECD Publishing. Park, H., and S. Y. Byun (2015) “Why Some Countries Attract More High-Ability Young Students to Teaching: Cross-National Comparisons of Students’ Expectation of Becoming a Teacher,” Comparative Education Review, 59(3): 523–549. Watt, H. M. G., P. W. Richardson, and K. Smith eds. (2017) Global Perspectives on Teacher Motivation, Cambridge University Press.