The efficacy of a computer science curriculum for early childhood: evidence from a randomized controlled trial in K-2 classrooms

13 schools cluster-randomised (6 treatment, 7 control); 464 consented treatment students across 28 classrooms plus control students across 36 classrooms; about 1,000-1,057 K-2 students in analytic samples

Kindergarten to grade 2 in public schools in a north-eastern US state, with Title I schools encouraged to apply.

A genuine cluster randomised trial with randomisation before any outcome testing and baseline equivalence established — the best design in the programming-instruction literature. Three weaknesses that bound it. (1) Developer-led throughout: Bers and the DevTech group built ScratchJr, the Coding as Another Language curriculum, the Coding Stages Assessment that measured the positive result, and the TechCheck instrument that measured the null. (2) High individual-level non-response, which is why the What Works Clearinghouse rated it "meets standards with reservations". (3) Only 13 clusters, so cluster-level power is thin. Note what the WWC reviewed: the literacy and TechCheck domains — not the coding outcome that carries the entire positive result. Reported in a technical report (ERIC ED626865) and a journal article; treated here as one study.

Twelve weeks of the CAL-ScratchJr curriculum moved coding performance by Hedges g = 0.47 against a business-as-usual control. It moved computational thinking (TechCheck) by g = 0.04 and standardized literacy by g = -0.01. A companion analysis of 747 students found no curriculum effect on mathematics or literacy, although within-student growth in coding skill predicted both. The journal version reports that students with disabilities and female students did not gain equally with peers. The one outcome that moved is the one measured by an instrument the curriculum's own developers built.

None material. School-level randomisation with baseline equivalence established.

No independent replication of the CAL-ScratchJr curriculum exists. The pattern it produces — coding skill up around 0.5, computational thinking at zero — is reproduced by the same group's later second-grade trial and, independently and with a different validated instrument, by the English Code Club randomised trial.