The rate of return to early childhood education in Japan: estimates from the nationwide expansion

ABSTRACT

We present new estimates of the internal rate of return to early childhood education. Utilizing the nationwide expansion of preschool education in Japan between 1960 and 1980, we initially assess the impact of preschool attendance on high school graduation and college enrollment for men. Subsequently, we compute the social rate of return to preschool attendance for men by drawing upon national wage statistics to project expected wage trajectories and referencing government documents to account for the social cost of preschools. Our findings indicate a social rate of return ranging from 5.7% to 8.1%, consistent with previous estimates in the literature.

KEYWORDS:

• Internal rate of return
• universal preschool expansion
• kindergarten
• nursery school

JEL CLASSIFICATION:

• I22
• I28
• H75

Acknowledgments

We are grateful to the editor and anonymous referees for useful and encouraging comments. We benefited from the useful comments from the participants of various seminars and conferences. All remaining errors are solely ours.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 Garces, Thomas, and Currie (2002) find that children participating in the Head Start program tend to exhibit higher wages, an increased likelihood of high school graduation, an elevated probability of college enrollment, and a reduced propensity for engaging in criminal activities during their early twenties. Barnett (2004) delves into the long-term effects of the Perry Preschool Project, examining outcomes at age 27, while Belfield et al. (2006) provides further updates on these effects at age 40. Meanwhile, J. Heckman et al. (2010) and Akabayashi and Tanaka (2013) have conducted an in-depth examination of both short-term and long-term program effects, employing rigorous statistical analysis. For comprehensive overviews of early interventions' effects, refer to Currie (2001) and Blau and Currie (2006).

2 In this paper, we focus only on men because it is difficult to estimate the long-term monetary benefits of preschool education for women whose labor market attachment was much lower than men during the period we studied.

3 This policy was followed by the Second Plan in 1971.

4 In fact, the Japanese government is currently in the process of amalgamating two preschool education institutions into a single entity. The MEXT is actively promoting the adoption of a system established in Fiscal Year 2006 known as the ‘Center for Early Childhood Education and Care.’ This system offers integrated education and childcare services, along with an Action Program for the Promotion of Pre-School Education.

5 The nursery school data are unavailable for the years 1963–1965. To address this gap in our analysis, we applied linear interpolation to estimate the missing data. In a robustness check, we repeated the same estimations using data from 1964 onward, and found that these results closely mirrored our initial findings.

6 The Annual Report on Religion is published by the Agency of Cultural Affairs. The data in this report is derived from the results of the Survey of Religion, which is an annual census survey conducted by the MEXT. The primary objective of this survey is to gather fundamental information, including the number of religious groups, the workforce associated with these groups, and the count of believers.

7 The maximum number exceeds 100% partly since this is the gross graduation rate that ignores the grade repetition, and there were some high schools which had 4 year programs.

8 The BSWS is an annual survey that targets regular workers employed by businesses with more than nine employees. The age categories in the published tables are ‘under 17,’ ‘18–19,’ ‘20–24,’ ‘25–29,’ …‘Over 65.’

9 One may also think that the short-run benefits of childcare use should include its impact on parental labor supply. While there is a large empirical literature on the relationship between childcare availability and female labor force participation, the evidence is mixed. In Japan, for example, Asai, Kambayashi, and Yamaguchi (2015) found, using data from 1990 to 2010 and employing a prefecture fixed effect model, the variation in the availability of full-time childcare, which corresponds to nursery school in our study, was not significantly associated with the labor supply of mothers with children under 6. Their findings suggest the presence of crowding out of informal childcare by other family members due to formal care. During the 1960s-70s, informal care was more prevalent than today, suggesting that the impact on employment would have been even smaller. Additionally, as kindergarten programs were typically half-day, the effect on employment was likely to be further minimized. Considering these factors, we have chosen not to incorporate the potential effects on parental labor supply into the benefits of childcare.

10 In OLS, standard errors are robust to heteroskedasticity. In GMM, standard errors are heteroskedasticity- and autocorrelation-consistent (HAC: Newey and West 1987) using Bartlett kernel with band-width 3=T1/3 (Li and Racine 2006, 405).

11 Here, we assume that using the 4-year college advancement rate for its graduation rate does not significantly bias our estimates, although it is challenging to assess the potential bias due to the lack of official data on college dropout rates during the 1970–1980s.

12 We are grateful to the suggestion from Shun-ichiro Bessho on the recommended value of DWL.

13 For a more comprehensive robustness check, we estimated Model 1, allowing for the incorporation of sample weights based on preschool enrollment at age 5, as well as clustering of errors at the prefecture-level. When introducing the weights, the estimated IRRs ranged from 7.11% to 8.17%. Furthermore, with the additional introduction of clustering, the point estimates ranged from 7.09% to 8.15%, with statistical significance for kindergarten. We also considered the potential presence of severe aggregation bias due to the nature of our prefecture-level dataset. According to Hanushek, Rivkin, and Taylor (1996) and Fertig and Wright (2005), aggregation bias in the estimation of the education production function is likely to overestimate the effect of schooling. We assumed that a substantial positive aggregation bias existed in the first stage estimation and that the true coefficients based on individual data would be half of our estimated values. Even under this assumption, the calculated IRRs based on the point estimates (as the standard error would also change with individual data) still ranged between 5.71% and 7.88%.

Additional information

Funding

This research was financially supported by the JSPS Grant-in-Aid for Scientific Research (Grant Numbers 24330090, 21H04982 for Akabayashi, 20H05629 for Tanaka).