Rosholm et al. tested three variants of a 12-week, four-session-per-week small-group math intervention for the 20% lowest achievers in grades 2 and 8, using two two-stage randomized trials with 2:1 pupil-level randomization within schools .
The grade 2 advantage was not a designed comparison (P = 0.26 for equality), and the authors appropriately note winner's-curse risk in arm-specific estimates such as 0.80 SD. Medium-term measurement also differs by grade: national tests in grade 3 (~16β18 months) versus school-leaving exams (~14 months), complicating cross-grade fade-out inference.
Strengths include preregistration (AEARCTR-0002361), stratified randomization, register-based outcomes, and honest framing via Kraft (2020) effect-size benchmarks. Weaknesses: only ~50% of intended 48 sessions were delivered; schools were recruited by phone with possible self-selection; the intervention-specific post-test was developed by the research team and administered on the intervention's last day, which inflates immediate effects relative to distal standardized measures; and spillover analysis to non-participating pupils was underpowered. Context matters: baseline standardization within the low-achieving target population mechanically enlarges SD units (the authors report 0.26/0.17 SD when standardized on the full population). Compared with meta-analytic benchmarks for small-group instruction (~0.30β0.36 SD), the pooled effects are plausible, but arm-level differences (e.g., the null 1-group+coaching result in grade 2) likely reflect noise, not mechanism .
Bottom line: credible causal evidence that cheap small-group instruction (1 teacher : 3β4 pupils, ~$392/pupil) helps grade 2 low achievers; weaker, fading evidence for grade 8. Falsification: a replication with independent standardized outcomes and high fidelity showing null pooled effects would overturn the claim.
Know what changed, what holds up, and what remains uncertain. Every Friday. No ads.