Effective instruction conditions for educational robotics to develop programming ability of K-12 students: A meta-analysis

36 studies, 4,180 K-12 students, 85 effect sizes

K-12 including kindergarten, international, with heavy Asian representation.

The authors invented the "essential versus generative ability" outcome split themselves for this meta-analysis, so the outcome construct is not independently defined. The moderator profile is a textbook inflation signature: the largest effects appear at kindergarten, for interventions of one to five weeks, for samples under 50, and depend on the choice of measurement scale. Every one of those points the same direction — small, short, self-measured studies drive the estimate. Read at abstract depth.

Educational robotics moved "essential" programming ability g = 0.539 (95% CI 0.327-0.752) and "generative" ability g = 0.535 (0.426-0.643). Individual programming favoured essential ability and cooperative programming favoured generative ability. Effects were largest in kindergarten, largest for one-to-five-week interventions, and largest for samples under 50.

Medium. Mixed experimental and quasi-experimental base.

Robotics meta-analyses cluster around g = 0.5-0.9 but with wildly different inclusion rules and outcome constructs, so the apparent agreement is not replication — the construct being pooled is not the same across them.