Performance Observability and Feedback Avoidance among Pupils

Students participating in academic quiz tournaments on the Maidaan digital learning platform are invited to take part in an additional voluntary “practice” round. The invitation message and feedback display vary in how performance information is presented (publicly or anonymously) and in how the difficulty of the practice round is labeled (easy or hard). The study examines how different formats of feedback and task framing affect students’ willingness to engage in further learning activities.

The field experiment is conducted on the Maidaan educational platform in India with students in grades 3–9 (ages 8–14). It follows a 2×2 factorial design that varies feedback observability (Public vs. Anonymous) and task difficulty label (Easy vs. Hard).

After each regular Maidaan tournament, students receive a standardized WhatsApp message inviting them to participate in a short, optional “bonus practice round.”

Public feedback condition: Students’ leaderboard entries show their name, grade, and school.

Anonymous feedback condition: Students appear as “Player N, Grade X,” with school information hidden.

Easy/Hard labels: The message and practice round content indicate the stated difficulty relative to the prior tournament, drawn from Maidaan’s existing question bank.

Randomization occurs at the individual level within each tournament wave (Nov 1, Nov 15, Jan 10, Jan 31 – 2025/26), using stratified block randomization (block size = 4) across school, grade, gender, performance tier, school type (contract / non-contract), and pool.

The primary outcome is opt-in to the bonus round (binary: 1 = participates, 0 = does not). Secondary outcomes include bonus-round performance (score and accuracy), engagement with leaderboards (“Leaderboard” and “Your Stats” tabs), and the timing of opt-in (minutes/hours after message).

The study tests whether anonymizing feedback increases willingness to re-engage after competitive play and whether gender and performance differences moderate this effect.

2025-11-01

2026-01-31

opt-in to the bonus round (categorical: 3=finished, 2=started 1=clicked to play, 0=did not). Main interest is binary (3 vs 0).

(conditional on opt-in): bonus-round performance (score and accuracy), engagement with leaderboards (Leaderboard’ and ‘Your Stats’ tabs), and the timing of opt-in (minutes/hours post-message). Engagement data are derived from Maidaan backend logs where available.

The study is a 2×2 randomized field experiment implemented on the Maidaan educational platform in India with students in grades 3–9. After completing a regular academic quiz tournament, students receive a standardized WhatsApp message inviting them to play a short, optional “bonus practice round.” The experiment varies two features of this invitation:
(1) Feedback observability — whether performance on the bonus round is displayed publicly on the leaderboard or anonymously, and
(2) Task difficulty label — whether the bonus round is described as easy or hard.
Participation in the bonus round is entirely voluntary. The main outcome is whether students choose to play. The study examines how different ways of presenting feedback and difficulty influence students’ willingness to re-engage with learning tasks.

The experiment is conducted across four tournament waves (Nov 1, Nov 15, Jan 10, Jan 31 – 2025/26) on Maidaan, a digital quizzing platform used by school students in India. Participants (grades 3–9, ages 8–14) are regular platform users drawn from 67 schools (~5,400 unique users).

Design:
A 2×2 factorial structure crossing Feedback Observability (Public vs Anonymous) and Task Difficulty (Easy vs Hard). Each student is randomly assigned to one of four treatment arms:

Public × Easy 2. Public × Hard 3. Anonymous × Easy 4. Anonymous × Hard.
In the Public arms, leaderboards show the student’s name, grade, and school. In the Anonymous arms, entries appear as “Player N, Grade X” with school name hidden. The “easy/hard” labels correspond to Maidaan’s standard difficulty levels.

Randomization:
Individual-level stratified block randomization (block size = 4) is implemented within each wave. Stratification variables: school, grade, gender, performance tier (Final Rank, DNQ narrow, DNQ large, DNP), school type (contract vs non-contract), and pool. Returning students are re-randomized independently in each wave. Randomization code with a fixed seed will be included in the replication package.

Procedures:
Invitations are sent via WhatsApp at 19:00 IST on Mondays after tournaments; messages contain no incentives and are tagged as utility notifications. The bonus-round remains open 7 days. Backend logs capture message delivery, read rates, clicks, and bonus-round performance. Amendment: From wave 2, we send a reminder message on WhatsApp at 19:00 IST on Saturdays after tournaments.

Randomization occurs at the individual level within each wave, using stratified block randomization (block size = 4, one student per arm). Stratification variables: school, grade, gender, performance tier {Final Rank, DNQ narrow, DNQ large, DNP}, school_type (contract/non-contract), and pool. Returning students are re-randomized independently in each wave; prior assignments are not carried forward. We will record the treatment sequence participants were exposed to and control for previous treatments.

Individual student.
Randomization is conducted at the individual level within each tournament wave. Stratified block randomization (block size = 4; one student per treatment arm) is implemented within strata defined by school, grade, gender, performance tier (Final Rank, DNQ narrow, DNQ large, DNP), school type (contract/non-contract), and pool.

Returning students are re-randomized independently in each wave (Nov 1, Nov 15, Jan 10, Jan 31 – 2025/26). Analyses account for repeated participation by clustering standard errors at the individual level and using school fixed effects.

Yes

67 schools.

The experiment includes approximately 67 distinct schools (7 contract schools and 60 non-contract schools) participating across four tournament waves, with an average of 80–120 students per school per wave.

Approximately 20,000 student-wave observations (≈ 5,000 unique students × 4 waves). Each student is randomized in up to four tournament waves between November 2025 and January 2026. The expected total includes both contract and non-contract school participants from 67 schools. Analyses will pool all wave-level observations and cluster standard errors at the school and individual levels.

Public × Easy: ≈ 5,000 student-wave observations (≈1,250 per wave; students from all 67 schools)
Public × Hard: ≈ 5,000 student-wave observations (≈1,250 per wave; students from all 67 schools)
Anonymous × Easy: ≈ 5,000 student-wave observations (≈1,250 per wave; students from all 67 schools)
Anonymous × Hard: ≈ 5,000 student-wave observations (≈1,250 per wave; students from all 67 schools)

Main outcome: Opt-in to bonus round (binary). Baseline rate (p₀): 0.25 (Pilot-based, delivered subsample). Standard deviation (Bernoulli): √[p₀(1−p₀)] = 0.433. Unit for effect size: percentage points (pp) difference in opt-in rates. Sample design assumptions (used for MDES): 2×2 factorial, balanced arms; ≈ 5,000 student-wave obs per arm (≈20,000 total). 67 school clusters; individual randomization, but SEs clustered by school. Average cluster size across waves ≈ 300 students/school (≈75 per school per wave × 4 waves). α = 0.025 (Bonferroni across 2 primary tests), Power = 0.80. ICC (school) assumed 0.01–0.02 → design effect ≈ 1.8–2.5. Minimum Detectable Effect Size (MDES): ≈ 6.5 pp (if ICC = 0.01; DE ≈ 1.8) ≈ 8.0 pp (if ICC = 0.02; DE ≈ 2.5) In SD units: 6.5 pp / 0.433 ≈ 0.15 SD; 8.0 pp / 0.433 ≈ 0.18 SD. Relative change vs baseline: 6.5 pp on 25% = +26%; 8.0 pp on 25% = +32%.