The Evidence on Teaching

Similarity Matters: A Meta-Analysis of Interleaved Learning and Its Moderators

Brunmair, M., & Richter, T. · 2019

grade Cmeta-analysisindependentunreplicatednumbers spot-checked
Sample
59 studies / 238 effect sizes
Population
Mostly university/lab; inductive category learning + math; artificial/aligned materials.
Design
Multilevel meta of interleaving vs blocking on inductive learning, with material-type moderators, publication-bias (trim-and-fill), and p-curve analyses.
Key findings
Interleaving is real but NARROW: strong for perceptual/inductive category learning (paintings 0.67), small for math (0.34), NULL for expository text, NEGATIVE for word learning (-0.39, blocking wins). Mechanism is discriminative contrast (learning WHICH category applies), not general deep processing — so it helps most when categories are confusable. Trim-and-fill drops the overall effect to 0.29.
Genetic confound
Domain discrimination-skill outcome; no g. Larger for younger/pure-student samples.
Replication notes
Numbers verified. The definitive statement that interleaving is narrow and material-dependent.
DOI / URL
10.1037/bul0000209

Effects

OutcomeMetricValueMeasureTimingVsHorizonClass
Overall interleaving vs blockingg0.42 (trim-and-fill 0.29)researcher-designedmixed; lab-dominatedactive-alternativeuncleardomain-skill
Paintings / visual category learning (best case)g0.67researcher-designedlabactive-alternativeend-of-treatmentdomain-skill
Mathematicsg0.34 (small)researcher-designedmixed lab/classroomactive-alternativeuncleardomain-skill
Word learning — BLOCKING wins (reversal)g-0.39researcher-designedlab; shortactive-alternativeend-of-treatmentdomain-skill
Expository text and tastesgnon-significantresearcher-designedlabactive-alternativeend-of-treatmentdomain-skill

Cited by