Wise feedback and Feedforward, Hanoi, Sep

2026-09-21

2026-12-18

Test results for all tests (will be 10 in total);

1. Test scores: The student's score (on a 0–10 scale) on the weekly in-class mathematics tests.

2. Optional question attempt: A dummy variable equal to 1 if the student attempted the bonus question (which did not count toward the official grade) on the quiz, and 0 if they left it blank. Both measures were recorded weekly throughout the semester.

Student autonomy and effort in study

It will be measured at the beginning and end of the course through two short surveys and tests administered throughout the semester.

This study is a three-arm randomized controlled trial conducted among undergraduate students in an introductory mathematics course. Students are assigned at the individual level (1:1:1 ratio) to receive different forms of qualitative written feedback on their regular quizzes: (1) an evaluative control group, (2) an instructional feedforward group, or (3) an integrated feedback group combining wise feedback with instructional guidance. The experiment tracks weekly assessment scores, voluntary problem attempts, and survey-based measures of student study habits across the semester.

Not available

Randomization done by a computer.

Randomization is conducted at the individual level using stratified block randomization within each classroom section. Specifically, students within each of the two course sections are stratified into mutually exclusive blocks based on baseline performance and geographic origin (urban vs. rural). Within each stratum, individual students are randomly allocated in an equal 1:1:1 ratio to one of the three conditions. There is no cluster-level randomization

No

0

Up to 1,200 student-week panel observations (120 students observed weekly across 10 in-class quizzes), alongside 240 baseline and endline survey responses.

40 students for each treatment arm.

Power calculations assume a standard significance level of (alpha = 0.05) and 80% statistical power (beta = 0.20), evaluating pairwise comparisons between any treatment arm (n = 40) and the control arm (n = 40). 1. Primary Continuous Outcome: Academic Quiz Score, Cross-sectional baseline benchmark: In a simple two-arm comparison without covariates (n_1 = n_2 = 40), the minimum detectable effect size (MDES) is approximately 0.63 standard deviations (SD).Panel estimation with baseline controls & fixed effects: Accounting for the longitudinal repeated-measures design (10 weekly quizzes; 6 active intervention rounds and 3 washout rounds) and controlling for baseline score (Quiz 1), the effective residual variance is substantially reduced. Assuming a moderate-to-high quiz score autocorrelation (rho is in [0.50, 0.70]), the MDES shrinks to approximately 0.28 to 0.36 SD for average treatment effects across the intervention period. 2. Primary Binary Outcome: Optional Task Attempt, Assuming a baseline control uptake rate of 20% to 25% on the voluntary non-credit challenge problem, the design can detect a minimum difference of 18 to 22 percentage points in single-period tests, and an average shift of 10 to 13 percentage points across the pooled panel of weekly repeated rounds with clustered standard errors at the individual level. 3. Secondary Survey Outcome: Self-Reported Academic Autonomy, Estimated via ANCOVA controlling for baseline survey scores (R^2 approx 0.30 - 0.40), the design provides 80% power to detect a change of 0.48 to 0.52 SD on the 3-item Likert autonomy index at endline.