Reciprocal Teaching: A Review of the Research (with the 1996 question-generation review)
Rosenshine, B., & Meister, C. (1994); Rosenshine, Meister & Chapman (1996) · 1994
grade Cmeta-analysisindependentreplicatednumbers spot-checked
Sample
16 reciprocal-teaching studies
Population
Grades 4+ (many grade 7-8); general and low-comprehension readers; US.
Design
The canonical demonstration of measure-dependent effect inflation. Reports median effect sizes separately by outcome-measure type. NOTE: the NRP 2000 comprehension chapter reused these medians (0.88/0.32) - it was a narrative review with no independent pooled ES, so do not count NRP as a second source.
Key findings
Reciprocal teaching works, but its effect is ~2.75x larger on experimenter-made tests (median d=0.88) than on standardized tests (0.32), and dose-response is flat (~6 sessions ≈ ~50). Empirical basis for the 'strategies are a one-time trick, not a trainable skill' view.
Genetic confound
Experimental contrasts; the standardized-measure d=0.32 is the defensible transfer estimate.
Replication notes
Measure-inflation and flat-dose patterns corroborated in NRP 2000 and Willingham; RT itself replicates as a real but modest effect.
DOI / URL
10.3102/00346543064004479
Effects
| Outcome | Metric | Value | Measure | Timing | Vs | Horizon | Class |
|---|---|---|---|---|---|---|---|
| Reading comprehension (reciprocal teaching) | median d | 0.88 (researcher-designed) vs 0.32 (standardized) | mixed | end-of-treatment | business-as-usual | end-of-treatment | domain-skill |
| Dose-response (~6 vs ~50 sessions) | d | no difference (flat) | mixed | end-of-treatment | none | end-of-treatment | domain-skill |
Cited by
- Reading-comprehension strategy instruction (reciprocal teaching, summarizing, questioning)mixedconf: mediumgc: low