Project Follow Through: National Planned-Variation Evaluation
Stebbins, L. B., et al. (Abt Associates); syntheses by Watkins 1997 · 1977
grade Cquasi-experimentindependentmixed
Sample
~70,000+ children across ~180 communities
Population
K-3 economically disadvantaged children; reading/math/language + cognitive + affective outcomes.
Design
Largest US education experiment, but PLANNED VARIATION not randomized: communities self-selected models; each site vs a local comparison. Direct Instruction (Engelmann) at ~19 sites. Norm-referenced tests (MAT).
Key findings
Field-scale evidence that Direct Instruction beat 8 competing models (open education, cognitive, self-esteem) on basic skills AND cognitive AND affective outcomes — including on models that explicitly targeted those outcomes. But it is a huge quasi-experiment with self-selected sites and contested analysis; the '#1' ranking is suggestive, not a clean causal result.
Genetic confound
Raised achievement toward norms; earlier Bereiter-Engelmann IQ gains did not durably hold — no durable g movement.
Replication notes
Contested: House, Glass, McLean & Walker 1978 ('No Simple Answer') found WITHIN-model site variance exceeded BETWEEN-model variance, effects small; self-selected sites, non-random, ~50 years old. Suggestive, not causal-clean.
Effects
| Outcome | Metric | Value | Measure | Timing | Vs | Horizon | Class |
|---|---|---|---|---|---|---|---|
| Basic skills (reading/math/spelling/language) | rank on standardized tests | DI ranked #1; raised disadvantaged children toward ~50th percentile | standardized | end of grade 3 | business-as-usual | end-of-treatment | domain-skill |
| Cognitive & affective outcomes | index vs comparison | DI positive on both; several affect/cognition-focused models scored NEGATIVE | mixed | end of grade 3 | business-as-usual | end-of-treatment | g |
Cited by
- Explicit instruction vs discovery/inquiry — how much guidance?moderate supportconf: mediumgc: low