Effects of Part-Time Subject-Specialist Teachers on Student Achievement: A Cluster-Randomized Trial in Japanese Elementary Schools

The intervention assigned part-time subject-specialist teachers to public elementary schools in Chiba Prefecture, Japan, during the 2023 academic year (April 2023 to March 2024). The specialists were retired former full-time teachers holding valid teaching licenses, recruited by the prefecture for their expertise in mathematics or science. Each specialist taught mathematics or science classes to fourth-grade students in place of the homeroom teacher, while the homeroom teacher continued to teach all other subjects and remained responsible for student guidance, grading, and evaluation. The homeroom-based system was thus preserved; the intervention changed only who taught the designated subject. Specialists taught an average of about 10 classes per week, and a single specialist could serve multiple schools, grades, or classes. Schools assigned to the mathematics treatment received a mathematics specialist only, and schools assigned to the science treatment received a science specialist only; no school received specialists in both subjects. Control schools continued with the conventional homeroom-based system, in which the homeroom teacher taught all subjects. The total number of instructional hours was unchanged in all schools, as instructional hours are fixed by the national curriculum; the intervention reallocated who delivered instruction, not how much instruction students received.

This study was conducted in cooperation with the Chiba Prefectural Board of Education. Using data from the 2022 National Assessment of Academic Ability, we identified relatively low-performing public elementary schools in each of Chiba Prefecture's five school districts. Within each district, 12 schools were selected and randomly assigned in equal numbers to three arms: part-time mathematics specialist, part-time science specialist, or control, yielding 20 schools per arm (stratified randomization by district).

The specialists were retired former full-time teachers holding valid teaching licenses (average age 55.8 years), recruited explicitly for subject expertise in mathematics or science. They taught an average of 9.68 classes per week, and a single specialist could be assigned to multiple schools, grades, and classes; 38 specialists in total accepted posts for fourth-grade classes. Specialists taught only the designated subject, while homeroom teachers retained responsibility for all other subjects, student guidance, grading, and evaluation. Total instructional hours were unchanged, as they are fixed by the national curriculum guidelines.

Deployment formally began in April 2023 but was not immediate in all schools: specialists were in place in 13 of 20 mathematics schools and 15 of 20 science schools by the baseline assessment (May 2023), rising to 93 percent and 98 percent of schools, respectively, by the end of the academic year. We use random assignment as an instrument for actual deployment to address partial compliance.

Baseline assessments (standardized achievement tests in mathematics, science, and reading, plus student questionnaires) were administered in May 2023, and endline assessments in May 2024. Fifth-grade students in the same schools took identical tests and questionnaires for placebo and benchmarking purposes. Homeroom teacher outcomes were collected through a survey in February 2024. Data from two schools were excluded due to administrative errors in student identifier assignment that prevented longitudinal matching. Semi-structured qualitative interviews with three pairs of homeroom and specialist teachers were conducted in June, September, and October 2023.

2023-04-01

2024-05-31

Student academic achievement in mathematics, science, and reading, measured by standardized achievement tests administered at baseline (May 2023) and endline (May 2024). Test scores are standardized to have a mean of zero and a standard deviation of one within the analytic sample. Standardized achievement tests in mathematics, reading, and science were administered to all fourth-grade students in participating schools during regular class hours. Science tests were conducted by a separate testing provider and, in science treatment schools, administered on a different day from the mathematics and reading tests. Scores in each subject are standardized (mean 0, SD 1). Treatment effects are estimated in a value-added framework, regressing endline scores on baseline scores, treatment assignment indicators, student gender, class size, total enrollment, and randomization strata fixed effects, with standard errors clustered at the school level.

Achievement scores in mathematics, science, and reading are constructed from raw scores on standardized achievement tests. Scores in each subject and wave are standardized to have a mean of zero and a standard deviation of one within the analytic sample. In addition to overall science scores, science achievement is decomposed into five domains for supplementary analysis: two difficulty-based categories (basic and applied items) and three competency-based categories corresponding to the evaluation criteria of Japan's national Course of Study (knowledge and skills; inquiry, reasoning, and expression; and proactive attitude toward learning). Each domain score is standardized in the same manner.

(1) Student learning strategies: flexible, planning, task-oriented, cognitive, and effort-regulation strategies. (2) Student non-cognitive skills: self-control and self-efficacy. (3) Science achievement by domain: basic and applied items; knowledge and skills; inquiry, reasoning, and expression; and proactive attitude toward learning. (4) Homeroom teacher outcomes: working hours, lesson preparation time, perceived workload, instructional practices (lecture-based, repetitive practice, inquiry-based), mental well-being (WHO-5), and growth mindset. (5) Teacher-student relationships and classroom behavior: a student-reported teacher-student relationship index, a behavioral problems index, and twelve classroom discipline indicators.

Learning strategies are measured using scales developed by Sato and Arai (1998), covering five domains: flexible, planning, task-oriented, cognitive, and effort-regulation strategies. Self-control is measured using the Japanese translation of the domain-specific impulsivity scale of Tsukayama, Duckworth, and Kim (2013), and self-efficacy using the Self-Efficacy for Learning and Performance subscale of the MSLQ (Pintrich et al., 1991); each consists of eight items. All indices are constructed by averaging item responses on a 5-point Likert scale and standardizing to mean zero and standard deviation one in the analytic sample. Science domain scores are constructed from test items classified into difficulty-based categories (basic, applied) and competency-based categories corresponding to Japan's national Course of Study (knowledge and skills; inquiry, reasoning, and expression; proactive attitude toward learning), each standardized in the same manner. Teacher outcomes are drawn from a survey administered by the Chiba Prefectural Board of Education in February 2024: perceived workload is an eight-item index (4-point Likert scale) aligned with the 2019 TIMSS teacher questionnaire; instructional practices follow the TIMSS 2019 classification (lecture-based, repetitive practice, inquiry-based); mental well-being is measured by the WHO-5 Well-Being Index; and growth mindset uses two items adapted from Yeager et al. (2022), with higher values indicating stronger endorsement of a growth mindset. The teacher-student relationship index is constructed from three student-reported items (whether teachers recognized students' strengths, listened to their concerns, and explained until understanding was achieved). The behavioral problems index is the sum of twelve student-reported classroom discipline items (punctuality, attendance, proper shoe arrangement, organization and orderliness, proper greetings, responsiveness, clarity of speech, polite language use, preparedness of learning materials, and others), each on a four-point Likert scale.

This study is a cluster-randomized controlled trial with schools as the unit of randomization. From public elementary schools identified as relatively low-performing based on the 2022 National Assessment of Academic Ability, 12 schools were selected in each of Chiba Prefecture's five school districts. Within each district, schools were randomly assigned in equal numbers to one of three arms: (1) deployment of a part-time mathematics specialist teacher, (2) deployment of a part-time science specialist teacher, or (3) control (no specialist deployment), yielding 20 schools per arm and 60 schools in total. Randomization was stratified by school district. No school received specialists in both subjects. Outcomes are measured for fourth-grade students using standardized achievement tests and questionnaires administered at baseline (May 2023), before specialist deployment was completed, and at endline (May 2024). Because deployment was not complete in all treatment schools at baseline, we estimate both intention-to-treat effects and local average treatment effects, using random assignment as an instrument for actual specialist deployment. Fifth-grade students in the same schools, who were not part of the experiment, took identical tests and questionnaires to allow placebo tests at the school level.

Randomization done in office by a computer, using statistical software (Stata), stratified by school district.

School. Randomization was conducted at the school level (cluster randomization), stratified by school district. There is only one level of randomization; students and teachers within a school share the school's treatment status.

Yes

60 schools

Approximately 2,780 fourth-grade students in 60 schools (1,021 in mathematics treatment schools, 852 in science treatment schools, and 913 in control schools, based on enrollment as of April 2, 2023)

20 schools assigned to the part-time mathematics specialist arm, 20 schools assigned to the part-time science specialist arm, and 20 schools assigned to the control arm.

No formal power calculation was conducted prior to the trial.

We provide experimental evidence on when subject specialization improves learning in elementary schools. In a cluster-randomized experiment across 60 Japanese public elementary schools, experienced part-time specialists in mathematics or science were assigned while preserving the homeroom structure. Science specialists raise science achievement by 0.18 SD and mathematics by 0.13 SD; mathematics specialists yield no gains. This asymmetry reflects comparative advantage: student achievement is strongly linked to homeroom teachers\textquotesingle{} subject expertise, and specialists generate gains only where such expertise is scarce. In mathematics, specialists crowd out homeroom teachers who are themselves math-specialized; in science, where fewer than 8 percent of homeroom teachers report subject expertise, specialists fill a genuine skill gap. We find no increase in coordination costs and no deterioration in teacher-student relationships. The intervention delivers gains at roughly one-twelfth the cost of class size reduction. Specialization improves learning when its design aligns with teacher comparative advantage and subject-specific skill constraints.

RIETI Working Paper Series, 25-E-111

https://www.rieti.go.jp/jp/publications/summary/25120001.html