Predicting Persistence and Fadeout Across Multi-Site RCTs of an Early Childhood Mathematics Curriculum Intervention
Watts TW, Botvin CM, Bailey DH, Hart ER, Mattera S, Clements DH, Sarama J, Farran DC, Lipsey MW · 2026
grade Bmeta-analysisdeveloper-involvedreplicated
Sample
Multiple cluster randomized trials of the same preschool mathematics curriculum, pooled meta-analytically
Population
US preschoolers across several independent evaluation sites.
Design
Meta-analytic synthesis across multi-site cluster RCTs of one preschool mathematics curriculum, modelling what predicts the size of the impact at 1-year follow-up given the impact at post-test.
Key findings
Fadeout is predictable from the size of the immediate effect, and it happens FASTER where the counterfactual environment is better — the opposite of what the sustaining-environments story predicts. Practical implication: forecast follow-up impact from immediate impact, and do not expect follow-through instruction alone to preserve an early gain.
Genetic confound
The catch-up finding cuts against a purely hereditarian reading of fadeout: if fadeout were only stable differences reasserting, the rate should not depend on how much the CONTROL group later learns. Fadeout is at least partly control-group catch-up, so it does not cleanly adjudicate heredity versus environment.
Replication notes
The strongest current test of the 'sustaining environments' hypothesis, and it goes the wrong way for that hypothesis.
Effects
| Outcome | Metric | Value | Measure | Timing | Vs | Horizon | Class |
|---|---|---|---|---|---|---|---|
| Post-test impact as a predictor of 1-year follow-up impact | prediction | strong — immediate impact magnitude is the best available forecast of follow-up impact | mixed | post-test to +1 year | none | under-1yr | domain-skill |
| Site-level later learning and fadeout speed | pattern | contrary to the sustaining-environments theory, impacts faded FASTER at sites where students learned more in the year after the intervention ended | mixed | +1 year | none | under-1yr | domain-skill |
| Fidelity and dosage | prediction | did not strongly predict fadeout once post-test effect size was accounted for | mixed | +1 year | none | under-1yr | domain-skill |
Cited by
- Math curriculum choice — how much does the textbook matter?mixedconf: highgc: low