Fostering historical thinking and democratic citizenship? A cluster randomized controlled intervention study

668 students (402 intervention, 266 control) in 47 classes and 17 schools, down from 40 teachers in 32 schools recruited

12th-grade students (age 17-18) in Flemish (Belgian) general secondary history education

Labelled a cluster RCT and randomised at school level, but the analysed sample is not the randomised sample and the paper is graded down accordingly. Of 32 schools recruited, only 17 (6 control, 11 intervention) finished; dropout was HIGHER in the control arm, and the authors additionally applied "strict exclusion criteria ... throughout the research to ensure treatment fidelity", i.e. removed teachers after randomisation. Both moves break intention-to-treat and leave an arm-size imbalance (402 vs 266) that randomisation of 32 schools should not have produced. Covid disruption is the stated cause of much of the attrition. The intervention is a six-lesson (roughly 12 x 50 min) inquiry unit built by the authors on cognitive-apprenticeship principles: modelling of historical inquiry, multiple-document tasks around authentic historical questions, explicit teaching of procedural knowledge and of the constructed nature of historical knowledge, plus whole-class discussion (the ingredient Reisman 2012 hypothesised was missing from her own trial). Control teachers taught their own post-1945 topics with no restrictions; materials were collected to monitor contamination. Both pre-test and post-test outcomes are researcher-built: a historical essay task scored 0-24 on a research rubric, an open-ended procedural-knowledge test, Likert scales for epistemological beliefs, and a parallel "topical" essay task on a contemporary controversy as the transfer measure. Groups were NOT equivalent at pre-test on procedural knowledge (p < .001) or democratic skills (p = .03), so pre-tests are covariates rather than a balance check. This is the cleanest test in the cluster of the far-transfer claim, because the lesson series deliberately taught NO civic content, leaving transfer to happen spontaneously or not at all.

Disciplinary outcomes moved and everything else did not. Controlling for pre-test and covariates, intervention students scored B = 4.79 points higher out of 24 on the historical inquiry essay task (p = .007; f-squared = 0.15, "medium"), which is roughly a standardised difference near 1 SD on the post-test metric - a large effect on a researcher-built rubric applied to the exact skill taught. Procedural knowledge about inquiry rose from 4.40 to 6.42 in the intervention arm while the control arm was flat (3.85 to 3.94). NO effect on either subscale of epistemological beliefs about history, and the authors report that procedural knowledge was not even correlated with inquiry skill at pre-test - students who can state the rules do not thereby apply them. Crucially, NO effect on democratic skills or democratic dispositions: the multilevel models found no intervention effect on the contemporary-topic essay task. Within-condition, students who gained more on the historical essay also scored higher on the topical essay, but the authors themselves note both are writing tasks and general writing ability was not controlled, so the correlation may be a writing artefact rather than transfer. Their conclusion is explicit: the data "do not provide clear evidence for a transfer" and "caution for an excessive optimism regarding the idea that historical thinking automatically contributes to democratic citizenship".

Low-to-medium. Initial assignment was random at school level, which equates allele distributions in expectation; the residual threat is post-randomisation attrition and fidelity-based exclusion, which were differential by arm and can reintroduce selection. Pre-test imbalance on two of five measures is consistent with that concern.

Unreplicated as a trial. It does, however, independently corroborate the pattern seen in Reisman 2012 and Bertram et al. 2017 - large gains on the taught disciplinary task, nothing on the distal claim - and it directly tests the transfer assumption that Reisman 2012 could only gesture at with a researcher-built 20-item multiple-choice transfer test.