Learning Historical Thinking With Oral History Interviews: A Cluster Randomized Controlled Intervention Study of Oral History Interviews in History Lessons

900 students in 35 ninth-grade classes at 10 academic-track schools (30 intervention classes, 5 control classes); intervention subsample N = 769

Ninth graders (mean age 14.5) in German Gymnasium history classes, topic "Peaceful Revolution in the GDR"

The best-executed randomised design in this cluster and, tellingly, the one that produces the least flattering picture of source work. Within-school cluster randomisation: schools with three history classes had classes randomly assigned across the three intervention conditions (live eyewitness, video of the same interview, transcript of the same interview); schools with four classes also had one class assigned to a non-treated control. Pre-test, post-test and a follow-up 2-3 months later. Models used robust SEs accounting for clustering, multiple imputation, and an unusually long covariate list (pre-test achievement, cognitive and reading ability, SES, family interest in history, grades, self-concept, interest, test motivation). Two design decisions cap the interpretation. First, the CONTROL GROUP WAS NEVER TAUGHT THE TOPIC - it continued with other units - so the enormous factual-knowledge effect is a coverage artefact, and the authors say so themselves. Second, the researcher personally taught all 30 intervention classes, so teacher and developer are the same person and the contrast to "usual teaching" includes an expert-outsider effect. Five outcome tests were all built for the study (factual knowledge; understanding reconstruction; understanding deconstruction; understanding oral history; concepts of sources and accounts).

Adjusted effects versus control at post-test / follow-up: factual knowledge b = 1.46 / 1.14 SD (an artefact of the control never studying the topic); understanding reconstruction 0.31 / 0.35; understanding ORAL HISTORY 0.30 / 0.42; concepts of sources and accounts 0.19 at post-test; and understanding DECONSTRUCTION b = 0.01, p = .879 at post-test and 0.03, p = .772 at follow-up. Deconstruction - recognising that a source is a constructed narrative to be interrogated rather than absorbed - is the core critical-source-analysis skill, and three weeks of intensive work with a primary witness moved it exactly nothing. The head-to-head comparison is sharper still and gives the paper its working title, "More Fun - Less Learned": students who met a LIVE eyewitness scored significantly LOWER than students who got the identical interview on video or in transcript on understanding oral history and on understanding deconstruction, while being more convinced they had learned a lot. The most engaging encounter with a primary source produced the weakest critical stance toward it. Those live-vs-mediated differences had faded by the 2-3 month follow-up.

Low. Randomisation was within school at class level with a rich covariate set and pre-test control, so genetic differences between arms are equated in expectation. The remaining threat is not genetic but structural - a non-treated control that never covered the content.

No direct replication. Its value here is convergent rather than confirmatory - it independently reproduces the Wilke 2022 and Reisman 2012 pattern in which the taught, proximal, researcher-scored skill moves and the deeper critical-source outcome does not, and it adds the counter-intuitive result that the most vivid primary-source encounter is the least critically instructive.