Peer Effects, Teacher Incentives, and the Impact of Tracking (Kenya RCT)
Duflo, E., Dupas, P., & Kremer, M. · 2011
grade Brctindependentreplicatednumbers spot-checked
Sample
121 schools (60 tracked / 61 random assignment); ~10,000 grade-1 pupils
Population
Kenyan grade 1; baseline classes ~80-100 split in two.
Design
The decisive tracking experiment. Context caveats: extreme baseline heterogeneity, contract teachers with renewal incentives, curriculum far above the median student — do not export magnitudes to developed countries.
Key findings
Tracking raised achievement across the WHOLE distribution — including the bottom half — via teach-to-level, not peers. Tracking works iff instruction adapts; reshuffling bodies without curricular differentiation does nothing.
Genetic confound
School-level randomization; sorting by prior achievement IS the treatment.
Replication notes
Verified. US corroboration of the mechanism: Collins & Gan (Dallas, IV) find homogeneous sorting benefits both high and low performers.
DOI / URL
10.1257/aer.101.5.1739
Effects
| Outcome | Metric | Value | Measure | Timing | Vs | Horizon | Class |
|---|---|---|---|---|---|---|---|
| Tracking vs random assignment, total score | SD | +0.138 (+0.175 w/controls) at 18 months | researcher-designed | 18 months | active-alternative | 1-2yr | domain-skill |
| Bottom half / top half of initial distribution | SD | +0.155 (p=.04) / +0.19 — ALL quantiles gained | researcher-designed | 18 months | active-alternative | 1-2yr | domain-skill |
| Persistence | SD | +0.163 one year post-program | researcher-designed | 1yr post | active-alternative | 1-2yr | domain-skill |
| Mechanism | evidence | median students gained equally in either section (rejects pure peer effects); teachers +9.6pp more likely found teaching — teach-to-level + effort | researcher-designed | during | active-alternative | end-of-treatment | behaviour |
Cited by
- Ability grouping and tracking — four practices, four verdictsmoderate supportconf: highgc: low