Teaching high school students to use heuristics while reading historical texts.

246 students (128 male, 118 female), ages 16-17, in 8 history classrooms at 2 high schools in the western United States

US 11th-grade history students

The one clean 2x2 factorial experiment in this literature, and the only causal test that separates the two things document-based history bundles together: the TEXTS (multiple primary sources vs a single textbook) and the STRATEGY INSTRUCTION (explicit teaching of sourcing, corroboration and contextualization vs content instruction). Eight classrooms were randomly assigned to one of four conditions over a three-week unit. Pre- and post-tests on content knowledge and on heuristic use. Graded C rather than B for a cluster-randomisation error that the design cannot escape: eight classrooms across four cells is two clusters per condition, so treatment is fully confounded with teacher/classroom in every cell, and student-level analysis over-states precision. Outcomes were built by the researchers, including a count of how often students deployed sourcing and corroboration in a final essay - the archive's standard ~2x discount for researcher-designed, treatment-aligned measures applies. Independence is the paper's chief virtue: University of Utah authors with no stake in any commercial or branded curriculum, five years before Reading Like a Historian was evaluated by its own designer.

The factorial separates the ingredients and the answer is that the TEXTS did more work than the STRATEGY LESSONS. Students who read multiple documents scored higher on history CONTENT than students who read the textbook, and used sourcing and corroboration more often - i.e. simply putting several conflicting sources in front of adolescents changed both what they knew and how they read, without anyone teaching a heuristic. Explicit heuristic instruction increased use of sourcing and corroboration but did NOT produce a corresponding content advantage, and contextualization proved resistant across conditions. Reisman 2012 cites this study as having already established that multiple-document instruction increases content retention, and her own factual-knowledge result (d ~ 0.29) is best read as a partial replication of it. The paper reports no standardised effect sizes in the record obtained here, which is why the numeric rows below are qualitative.

Low. Classrooms were randomly assigned to condition, so allele distributions are equated in expectation; the residual concern is the small number of clusters, which is a precision problem rather than a confounding one.

No direct replication. Its central claim - that multiple-document instruction raises content retention - is corroborated in the same direction by Reisman 2012 (d ~ 0.29 on factual knowledge) and by the PACT trial series (0.19-0.45 on content), though neither replicates the factorial that isolates texts from strategy instruction. The failure of contextualization to move is corroborated by Reisman 2012 (no effect on contextualization or corroboration sub-scales) and by Bertram et al. 2017 (understanding deconstruction b = 0.01).