Not New, but Nearly Forgotten: the Testing Effect Decreases or even Disappears as the Complexity of Learning Materials Increases

Narrative review; Table 1 vote-counts ~50 published experiments, and the argument rests on 8 experiments with worked examples plus 4 by Tran et al. and 2 by De Jonge et al. No pooled effect size is computed.

Undergraduates, vocational-education students and elementary schoolchildren, across the reviewed experiments; materials range from word pairs to worked examples on electrical circuits.

Guest editors introduction/review article for a special issue of Educational Psychology Review (27:247-264), not an empirical study and not a meta-analysis. Method is a nonexhaustive vote-count table in which the authors assign their own subjective element-interactivity ratings (many marked "M-H?" or "H?") to each study, plus narrative synthesis. Grade D is correct: expert argument with researcher-assigned exposure coding and no quantitative synthesis. Allegiance caveat the record must carry: Sweller is the originator of cognitive load theory and of element interactivity, the construct the paper uses to explain the boundary; both authors guest-edited the special issue and co-authored the studies that supply its central nulls (van Gog et al. 2015; Leahy, Hanham & Sweller 2015). Independent of the retrieval-practice literature it critiques, but not disinterested in the explanation it offers.

States the central boundary at the strength the paper actually claims, which is weaker than the way it is usually cited. The claim is that material complexity "might constitute another boundary condition" and that further research "would be worthwhile" - a hypothesis backed by converging nulls, offered as a research agenda, not a settled result. The strongest evidence in the paper is the within-material manipulation: scrambling a coherent text, which lowers element interactivity without changing the words, restores a (weak) testing benefit. The forgotten-history half is well sourced: Kuhn (1914) and Gates (1917) both found the recitation advantage shrinking as material became more connected and meaningful. Note the direction of the caution - element interactivity is defined relative to the learner, so the same material is high-EI for a novice and low-EI for an expert.

Concerns working-memory load x material complexity, not genetics.

Mixed, and the two directions must both be recorded. Supporting: the boundary is not one lab - nulls on high-EI material come from van Gog/Kester, Leahy/Hanham/Sweller, De Jonge/Tabbers/Rikers and Tran/Rohrer/Pashler; the paper also reports that Gates (1917) directly replicated Kuhn (1914) on the graded material effect; and a later classroom study (see cvs-retrieval-2024-classroom-reversal) found an actual reversal on complex procedural content. Against: the paper itself concedes "caution is always warranted when interpreting null effects" and that in some of the studies other known boundary conditions - test format, initial retrieval success, feedback, spacing - "cannot be ruled out entirely as an alternative explanation", and it admits results "may have been different if feedback or restudy opportunities after testing had been provided". A commentary by Karpicke and Aue on the special issue contributions is acknowledged in this paper but is a separate publication and was NOT read for this record; its content is therefore not asserted here.