Japan's R&D capabilities have been decimated by reduced class hours for science and math subjects

Japan centrally sets class hours and content through national curriculum guidelines (MEXT), resulting in uniform instructional time across schools. The study investigates whether government-driven reductions in science and mathematics class hours in junior high school have contributed to declines in academic performance and, ultimately, to stagnation in Japan’s R&D capabilities, as reflected in fewer scientific papers and patent applications since the 2010s. Although overall R&D expenditure has trended upward over decades, output indicators have lagged, suggesting factors related to human capital. Declines in enrollment in science and engineering, especially at graduate levels, and a documented shift away from science among students coincide with reductions in class hours and content since the 1980s. The research question is whether reduced instructional time in junior high school science and mathematics adversely affected later R&D outputs at the individual level.

Prior work links education investment and human capital to innovation and R&D performance. Studies report positive relationships between education indicators (e.g., share of GDP for education, literacy, junior/high school enrollment) and researchers’ productivity and national innovative capacity (Furman et al., 2002; Akhmat et al., 2014; De Rassenfosse & van Pottelsberghe, 2009). Doctoral education is associated with higher patenting and patent quality in Europe (Mariani & Romanelli, 2007; Schettino et al., 2013). International assessments indicate instructional time in science and mathematics correlates positively with achievement (Baker et al., 2004), and cognitive skills drive economic development (Hanushek & Kimko, 2000; Hanushek & Woessmann, 2008; Breton, 2011). The Japanese context adds a policy dimension: major reductions in class hours and content due to curriculum revisions since the 1980s (Hino, 2016) and a broader shift away from science among students (Science Council of Japan, 2016) are hypothesized to affect long-run R&D capacity. Earlier Japanese policy reports note declines in the number and quality of young R&D personnel in the 2000s due to multiple factors (NISTEP, 2010).

Data: Two online anonymous surveys with identical questionnaires of R&D personnel in Japan. The 2020 Survey (RIETI; fielded by Rakuten Insight, March 2020) targeted individuals working in engineering or research roles; screening continued until 5000 valid responses. The 2016 Survey (NTTCom Online Marketing Solutions, March 2016) yielded 4129 valid responses. Participants registered with survey monitors and consented to privacy policies. Sample (2020): 5000 valid responses; 92.9% male, 7.1% female; mean age 48.5 (range 23–69); mean working years 26.2 (0–51). Education: 73.3% university/graduate school, 2.2% PhD. Employment: 4.9% at universities/research institutes; majority in private companies. Main responsibility: basic/applied research 6.7%, with development and other areas >90%. Dependent variables: Cumulative counts over respondents’ careers of (a) patent applications, (b) patent renewals, and (c) presentations and papers at academic venues. Distributions were heavily zero-inflated and followed a power-law with long right tails. To control for experience, outputs per working year were computed by dividing totals by working years (four respondents with zero working years were excluded; N=4996 for per-year analyses). For patent applications per working year: mean 0.1255, SD 0.4619, min 0, max 10.0. In raw counts, 73.4% had zero patent applications, 85.0% zero patent renewals, and 64.1% zero presentations/papers; among those with at least one output, the 1–9 range dominated across indices. Key independent variables: Class hours in junior high school for science, mathematics, and their total. Hours were assigned by mapping each respondent’s junior high cohort to the applicable MEXT curriculum guidelines version (via age): total science+math hours by cohort were 805 (age ≥61; 420 science, 385 math), 840 (52–60; 420, 420), 735 (40–51; 350, 385), 700 (31–39; 315, 385), and 605 (≤30; 290, 315). For the 1993–2001 guidelines (age 31–39), science hours varied 315–350; the analysis conservatively used 315. Controls: Sex (female dummy), education dummies (university, master’s, PhD; reference: high school/junior college/technical college), type of business, initial research/technical area at company entry, and company size. Estimation: Type I Tobit model with lower bound at zero, appropriate for the censored distributions. Analyses conducted with STATA MP v13 and SPSS v26. Models estimated separately for science hours, math hours, and total hours. Likelihood-ratio chi-square tests indicated all models were significant at the 1% level. Additional cohort comparisons used the 2016 and 2020 datasets to analyze age trends and cohort effects (3-year moving averages) and to compare total class hours with average patent applications per working year by age group.

• Increased junior high school class hours in science and mathematics are significantly associated with higher patent applications per working year among R&D personnel.

• Science hours: Coef. 0.0046, Std. err. 0.0005, P<0.01 (Model 1).
• Mathematics hours: Coef. 0.0047, Std. err. 0.0009, P<0.01 (Model 2).
• Total science+math hours: Coef. 0.0029, Std. err. 0.0003, P<0.01 (Model 3). • Control variables: Female dummy showed a statistically significant negative association with patent applications per working year; higher educational attainment (university, master’s, PhD) showed progressively larger positive associations compared with the reference group. • Distributional facts: 73.4% of respondents had zero patent applications; 85.0% had zero patent renewals; 64.1% had zero presentations/papers. Among those with outputs, counts followed a power-law with most in the 1–9 range. • Cohort versus age effects: Comparing age-profile curves from 2016 and 2020 (3-year moving averages), shapes are nearly identical after shifting the 2016 curve by 4 years, indicating strong cohort effects beyond pure aging. Correlation coefficients after shift: 0.923 for patent applications; 0.895 for patent renewals. • Age-group alignment with curriculum hours: Average patent applications per working year by age group moved consistently with total junior high science+math hours (840 hours in ages 52–60 > 805 hours in ages 61–69; lower outputs for younger age groups aligned with 735, 700, and 605 hour regimes).

The findings support the hypothesis that reductions in junior high school science and mathematics class hours—accompanied by reduced curricular content—contributed to diminished R&D outputs later in careers, as proxied by patent applications per working year. This relationship persists after controlling for sex, education, industry, initial technical field, and firm size, and is robust across separate models for science, mathematics, and their total. Cohort analyses suggest the decline in outputs among younger generations cannot be explained by age alone. The study posits several mediating pathways: fewer class hours reduce content exposure and achievement, weaken interest and attitudes toward science (with observed positive albeit weak correlations between junior high science hours and liking science in college), reduce the likelihood of choosing physics in high school (r=0.209, p=0.000), and may diminish the quality of future science/math teachers, collectively affecting the pipeline into university STEM fields and advanced degrees. These mechanisms align with broader literature linking cognitive skills and science/math achievement to economic development. Importantly, the stagnation in Japan’s R&D outputs occurred despite long-run increases in overall R&D expenditure, underscoring human capital formation—shaped by education policy—as a critical constraint. Policy significance: Education policy shifts, particularly the “relaxed” curricula from the 1980s onward that reduced both time and content, can have long-lag negative impacts on national innovation capacity and economic growth. Careful, evidence-based curriculum revisions with attention to long-term human capital outcomes are essential.

Reductions in junior high school science and mathematics class hours in Japan since the 1980s are associated with lower subsequent R&D outputs at the individual level, contributing to national stagnation in publications and patenting. Because education shapes the long-term stock and quality of R&D human capital, curriculum revisions should be implemented only after thorough, long-horizon assessments. University reforms alone, without strengthening primary and secondary education, are unlikely to raise R&D capacity; improvements must consider the entire pipeline of human capital accumulation from early schooling through higher education. Future work should further quantify the long-run effects of specific curricular components and pathways (e.g., subject choices in high school, teacher quality) on R&D outcomes and examine policy designs that bolster STEM engagement and achievement.