What One Hundred Years of Research Says About Ability Grouping and Acceleration (second-order meta-analysis)
Steenbergen-Hu, S., Makel, M. C., & Olszewski-Kubilius, P. · 2016
grade Cmeta-analysisindependentreplicatednumbers spot-checked
Sample
13 first-order grouping metas + 6 acceleration metas
Population
K-12 across a century of studies.
Design
The taxonomy anchor — four DISTINCT practices, never pool them. Older corpus with researcher-designed-measure inflation (~2x discount).
Key findings
Four practices, four verdicts: within-class and cross-grade grouping are modest free wins; undifferentiated between-class streaming does nothing; gifted-class effects are selection; and acceleration's honest reading is 'no harm + a saved year.' The organizing principle: grouping works exactly insofar as INSTRUCTION ADAPTS to the group.
Genetic confound
Grouped/gifted/accelerated students are selected on substantially heritable prior ability; matching cannot remove it. The vs-older-classmates cell is the only selection-resistant acceleration comparison.
Replication notes
All component g's verified. Kulik & Kulik's dose-response signature folded in: gains scale with degree of CURRICULAR adjustment, not grouping per se.
DOI / URL
10.3102/0034654316675417
Effects
| Outcome | Metric | Value | Measure | Timing | Vs | Horizon | Class |
|---|---|---|---|---|---|---|---|
| Within-class ability grouping | g | 0.19-0.30 (modest, ~free) | mixed | end of treatment | business-as-usual | end-of-treatment | domain-skill |
| Cross-grade subject grouping (Joplin) | g | 0.26 | mixed | end of treatment | business-as-usual | end-of-treatment | domain-skill |
| Between-class streaming/tracking | g | 0.04-0.06 (~nil — mostly pre-1990 UNdifferentiated streaming) | mixed | end of treatment | business-as-usual | end-of-treatment | domain-skill |
| Special grouped gifted classes | g | 0.37 (selection-inflated; see the RD nulls) | mixed | end of treatment | business-as-usual | end-of-treatment | domain-skill |
| Acceleration vs same-age peers / vs older classmates | g | 0.70 (SELECTION, not causal) / 0.09 (the honest cell: accelerants keep pace a year ahead) | mixed | end of treatment | unclear | end-of-treatment | domain-skill |
| Ability-level moderation | pattern | effects did NOT differ for high/medium/low ability — no meta-analytic support for grouping harming low-ability students | mixed | end of treatment | business-as-usual | end-of-treatment | domain-skill |
Cited by
- Ability grouping and tracking — four practices, four verdictsmoderate supportconf: highgc: low
- Acceleration and gifted programs — the label vs the contentmoderate supportconf: highgc: medium