Improving Higher-Order Thinking Skills in Mathematics in Bhutanese Primary Schools

This study introduces a classroom instruction improvement program, targetng Grade 5 mathematics specifically. The program is designed to shift classroom instruction from teacher-centered transmission learning toward student-centered active learning, where students actively engage in problem-solving, construct explanations, and discuss ideas with peers. In each treatment school, the intervention is provided to one Grade 5 classroom's students and their teacher.

2026-07-28

2026-12-31

Student learning measured in test score points: total scores and sub-scores by knowing, applying, and reasoning cognitive domains in mathematics.

The research team will design a mathematics test at endline, allowing measurement of math skills by cognitive domain. Data will be processed using Item Response Theory. In addition, school-administered exam scores and grades will be drawn from official school records for treatment and control students.

Student motivation and attitudes toward mathematics; classroom instructional practices; principals' and teachers' practices and attitudes in teaching HOTS; student reading proficiency in English and the national language.

Student motivation and enjoyment will be measured through a student self-report questionnaire administered at endline.

Student reading comprehension in English and the national language will also be measured at endline. Since the intervention targets HOTS-oriented mathematics pedagogy specifically, reading outcomes serve two purposes: as a specificity check, since a treatment effect concentrated in reading rather than mathematics would suggest a general attention effect rather than a mathematics-pedagogy-specific mechanism; and as a test of cross-subject transfer, since Grade 5 teachers in Bhutan typically teach multiple subjects to the same students, and HOTS-oriented instructional practices learned in training may carry over into other subjects.

Teacher content and pedagogical knowledge will be measured through a teacher assessment administered at endline, covering understanding of HOTS-oriented task design and facilitation techniques covered in training.

Teacher classroom practices will be measured via structured classroom observation using the study's observation rubric (covering task launch, facilitation of student exploration and reasoning, and orchestration of discussion), scored by independent observers at endline.

Principals' efficacy and understanding will be assessed using questionnaires as relevant to the intervention.

Heterogeneity in treatment effects by students' baseline reading proficiency, drawn from mid-term exam records, will be examined in a forthcoming pre-analysis plan.

Stratified cluster RCT. Unit of randomization is the school. Stratification by zone (West, Central, and East) and school size (large/small), yielding 6 strata. 100 schools total, randomly assigned to treatment (49) or control (51), with balanced allocation within each stratum.

Not available

Computer-generated random assignment within strata Within each stratum, schools were assigned a random uniform draw and ranked; the bottom half by rank was assigned to treatment. The seed was fixed in advance following a pre-specified, documented selection procedure (see hidden design field) and is recorded in the project's randomization do-file for full reproducibility.

School

Yes

100 schools

Approximately 4,890 students (primary sample: ~3,000 students in 100 classrooms; spillover sample: ~1,890 students in 63 classrooms in schools with more than one Grade 5 class)

49 schools treatment; 51 schools control

We calculate the minimum detectable effect size (MDES) for a two-level cluster randomized trial with schools as the unit of randomization. The study randomizes 100 schools into 49 treatment and 51 control arms using stratified randomization across six strata (three geographic zones × two school size categories). The primary outcome is a custom IRT-scaled Grade 5 mathematics test, standardized to mean zero and standard deviation one within the sample. Power is set at 80% with a two-tailed significance level of 0.05, and 30 students are sampled per school. The ICC is estimated at 0.029 from our Phase I's school survey data with 50 schools in Western districts, which is corroborated by a district-level ICC of 0.031, constructed based on the Bhutan National Education Assessment 2024 covering all districts nationally. Under this estimate, the design effect is 1.84, yielding an MDES of 0.139 standard deviations. The design remains adequately powered (MDES ≤ 0.20 SD) for ICC values up to 0.097, approximately three times the observed estimate, providing reasonable robustness against higher clustering in the Central and Eastern zones not covered by Phase I data.