All Other Things Being Equal: Acquisition and Transfer of the Control of Variables Strategy

87 children aged 7-10 (grades 2-4) in the hands-on study; 55 third and fourth graders (24 previously trained, 31 untrained classmates) in the 7-month posttest

Two US elementary schools; grades 2-4, ages 7-10.

Individual-interview lab study, not a classroom trial. Children were randomly assigned across three conditions — Training + probe questions, probes only (no training), and neither — and worked with physical apparatus (springs, ramps/slopes, sinking objects) over two days. Phases: Exploration (baseline), Assessment, Transfer-1 and Transfer-2 (new mechanics domains, one week later). Part II was a paper-and- pencil posttest 7 months later, in a different format (evaluating printed designs rather than building them), with different experimenters, in five non-mechanics domains: plant growth, cookie baking, model aeroplanes, drink sales, running speed. NOTE the comparison: the "control" for the delayed posttest is classmates who never took part at all, so it is a business-as-usual contrast, not an active alternative. The authors call the posttest "remote transfer"; by this archive's taxonomy it is still the same strategy applied to a structurally identical task with a new cover story, so it is near transfer. Nothing in this study measures performance in another school subject, on a general reasoning test, or on any independent standardised instrument.

CVS is teachable and it generalises modestly — within the strategy. Direct instruction plus probe questions raised CVS use from 34% of trials at baseline to 65% after training and 61-64% in the two new-domain transfer phases; probe questions ALONE did not help, and neither did unguided exploration. The delayed test showed a sharp developmental ceiling: at 7 months, 79% of trained fourth graders were "good reasoners" (13+ of 15 correct) vs 15% of untrained fourth graders (χ² = 10.78, p < .001), but for third graders 40% vs 22% was not significant, and second graders transferred only to very near situations within the training domain. So: reliable near transfer for the oldest children only, no evidence beyond CVS-shaped tasks, and no measure at all of anything a school would call an outcome.

Low for the immediate contrasts (random assignment within schools). Medium for the 7-month posttest, where the comparison group is self-selected non-participating classmates rather than a randomised arm.

Directly extended by Klahr & Nigam (2004), already in this archive, which replicated the superiority of direct instruction over discovery for CVS in a classroom-adjacent setting and likewise found transfer only to a CVS-shaped judgement task. The finding that CVS is trainable is robust; the claim that training it generalises to scientific thinking broadly is not tested here and was directly attacked by Kuhn & Dean (2005).