Effects of Activity-based Elementary Science on Student Outcomes: A Quantitative Synthesis

57 controlled studies, ~400 comparisons, ~13,000 students, ~1,000 classrooms

US elementary school science (roughly ages 5-12); the three NSF activity-based programmes ESS, SAPA and SCIS versus conventional textbook science.

Meta-analysis of controlled evaluations of the three big NSF hands-on elementary science curricula. Designs pooled range from randomised to matched-comparison; the meta-analysis inherits that mix, hence grade C. The treatment is a WHOLE CURRICULUM (activity-based, materials-centred science), not an isolated practical-work component, so the estimate cannot be attributed to hands-on work specifically — it bundles curriculum content, teacher training and pedagogy. Crucially, Bredderman coded every outcome measure for BIAS TOWARD THE PROGRAMME (does the test cover what only the activity-based curriculum taught?) and reported effect sizes separately by bias category. That coding is what makes this file valuable. The paper also reports gains in 'intelligence' and 'creativity', which this archive treats as far-transfer claims requiring extraordinary evidence and which almost certainly reflect programme-aligned researcher measures.

THIS IS THE PAIRED-MEASURE FILE. Bredderman coded each outcome test for whether it favoured the activity-based programme and reported: d = 0.54 on measures biased toward the hands-on programme, d = 0.24 on unbiased measures, d = 0.13 on measures biased toward the conventional comparison. That is a 2.25x inflation from measure alignment alone, inside a single meta-analysis, on the same studies — an unusually clean estimate of the alignment tax in science education and closely matching METHODOLOGY's nominal 2x. The second structural finding is the process/content split: d = 0.52 on science-process tests versus d = 0.16 on science content. Activity-based science reliably teaches students to do the activity; it barely moves what they know. Bredderman also reports gains in creativity (0.42), language, mathematics (0.12) and 'intelligence' — far- transfer claims from 1983 on programme-aligned instruments that this archive treats as presumptively absent and that no modern evidence supports. Caveat that matters: the treatment is an entire curriculum, so nothing here isolates the hands-on component.

Medium. Studies are controlled but not uniformly randomised; the disadvantaged-vs-advantaged moderator in particular could reflect regression and selection rather than differential responsiveness.

No modern replication. Shymansky et al. reanalysed the same NSF-curriculum literature in the 1980s with broadly similar conclusions, but nothing since has re-run this synthesis with current standards (no funnel-plot or PET-PEESE correction here; Bredderman notes published reports gave higher effects than unpublished ones, i.e. undiagnosed publication bias).