The impact of a block-based visual programming curriculum: Untangling coding skills and computational thinking

331 second-graders across 21 classrooms in 11 public schools and 10 districts; 313 analysed (about 95% retention)

US second graders, mean age 8.4 (range 7.1-9.9); 21% free or reduced lunch, 10% with an individualised education programme, 11% with limited English proficiency.

A school-level cluster randomised trial of the 24-lesson Coding as Another Language ScratchJr curriculum, funded by the US Department of Education: 5 schools and 9 classrooms treatment, 6 schools and 12 classrooms control (business as usual). Baseline equivalence met, assessors blinded to assignment, multilevel models with school random intercepts (ICC 0.31). The critical fact for grading is that Bers developed both the CAL curriculum AND the TechCheck instrument on which it returned a null — this is a DEVELOPER-LED NULL, which is the strongest form of null evidence, because the efficacy-to-effectiveness decay curve runs the other way. The authors flag underpowering for the computational-thinking outcome, noting roughly 30 schools would be needed for a medium effect at two levels.

Coding skill moved a lot: post-test Coding Stages Assessment coefficient B = 8.05, Cohen d = 0.68, p < .001, with the treatment mean rising 4.62 to 15.68 against control 4.50 to 7.23. Computational thinking did not move at all: TechCheck coefficient B = -0.1, p = 0.81, Cohen d = -0.02. The authors conclude that coding skills and computational thinking may not be causally related among second-grade students, and caution against teaching abstract computational-thinking practices. They also observe that a contrary positive result elsewhere plausibly reflects a testing effect, because the unplugged activities used closely resemble the computational-thinking test items.

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

The authors state that three separate trials using the same instrument across three different curricula consistently reported no significant effect on students' computational thinking, and the English Code Club randomised trial reproduces the null with a different validated instrument. This is a replicated null.