The transfer effect of computational thinking (CT)-STEM: a systematic literature review and meta-analysis

37 studies, 7,832 students, 96 effect sizes

Kindergarten through university, with computational thinking integrated into STEM subjects.

A random-effects meta-analysis of computational-thinking-STEM integration studies, with a fatal gap for this topic: no active-versus-passive control moderator was coded or tested, which is the one variable that collapses the estimate in the best meta in the field. Six other moderators were tested. The outcome taxonomy is also not the archive's — the near-transfer bucket includes STEM achievement and the far-transfer bucket is creativity, critical thinking and problem solving.

Overall transfer g = 0.601 (95% CI 0.510-0.691), I-squared = 88.9%. Near transfer g = 0.645 (0.536-0.753); far transfer g = 0.444 (0.312-0.576). The far-transfer funnel plot is asymmetric and Egger's test is t(24) = 3.90, p < .001 — clear publication bias on exactly the estimate the policy claim depends on, which the authors report and then headline the uncorrected number anyway. The sample-size moderator is the other tell: studies with n under 50 give g = 0.826 while studies with n over 150 give g = 0.233 (Q_B = 48.0, p < .001), a textbook small-study effect the authors interpret as a substantive class-size finding. Effects also shrink with longer interventions, which no dose-response account of transfer predicts.

Medium. A mixed-design pool with no control-type moderation.

Its overall estimate is consistent with the uncorrected headline of Scherer et al. (2019), but it lacks the control-type moderator that collapses that estimate, so it does not independently confirm anything about far transfer.