Rethinking the Use of Tests: A Meta-Analysis of Practice Testing
Adesope, O. O., Trevisan, D. A., & Sundararajan, N. · 2017
grade Cmeta-analysisindependentreplicatednumbers spot-checked
Sample
~217 studies / 272 effect sizes
Population
Majority undergraduates + some K-12; mostly verbal/factual materials; ~89% lab, ~11% classroom.
Design
Anchor testing-effect meta. Random-effects. Predominantly lab with researcher-designed, response-congruent tests.
Key findings
Retrieval practice beats restudy (g=0.51) and no-activity (0.93); effect grows with delay and effortful formats. But ~89% lab with researcher-designed response-congruent tests — magnitude inflated by measure alignment.
Genetic confound
Retention outcome only; no g.
Replication notes
Numbers verified. Consistent with Rowland 2014, Yang 2021; magnitude sensitive to measure alignment.
DOI / URL
10.3102/0034654316689306
Effects
| Outcome | Metric | Value | Measure | Timing | Vs | Horizon | Class |
|---|---|---|---|---|---|---|---|
| Retention vs ALL comparisons | g | 0.61 [0.58,0.65] | researcher-designed | mixed; predominantly lab | unclear | end-of-treatment | domain-skill |
| Retrieval vs RESTUDY (rereading) | g | 0.51 | researcher-designed | mixed lab | active-alternative | end-of-treatment | domain-skill |
| Retrieval vs no-activity/filler | g | 0.93 | researcher-designed | mixed lab | business-as-usual | end-of-treatment | domain-skill |
| Moderator: retention interval | direction | larger when study-to-test gap >1 day | researcher-designed | delayed > immediate | unclear | unclear | domain-skill |
Cited by
- Retrieval practice (the testing effect)strong supportconf: highgc: low