An exploratory meta-analytic review on the empirical evidence of differential learning as an enhanced motor learning method

897 (DL 453 / control 446) across 27 studies

University sports classes, club athletes, rehab patients; soccer, tennis, basketball, shot put, high jump, balance, fine motor tasks

27 studies / 31 experiments, 897 participants (DL 453, control 446). Acquisition pool = 23 studies / 27 experiments; retention pool = 12 studies / 12 experiments; transfer reported too rarely to meta-analyse. Differential learning = deliberately high, non-repeating movement variability, the strongest form of variability-of-practice. Risk of bias rated high for nearly all studies (only 15/31 low risk on randomisation; allocation concealment unclear in all but six; selective reporting high risk in 9/31); funnel plots show moderate rightward asymmetry; I²=77-78% acquisition, 79% retention. Retention intervals ranged 1 h to 1 year, most 1-2 weeks. The paper is internally inconsistent on the headline acquisition estimate: abstract says d=0.26 [0.10, 0.42], results section and part of the discussion say d=0.27 [0.12, 0.42]. THE EVIDENCE BASE IS DEVELOPER-CONCENTRATED even though the meta-analysts are not: the retention technical-skills subgroup "mainly encompassed studies from one research group" (Schöllhorn's, DL's originator), Schöllhorn-lab experiments make up a large share of the pool, and the authors cite the developer's own institutional publication list of unpublished abstracts as evidence that publication bias "is present and affected the results of the meta-analysis". Three of seven retention subgroups rest on a single study each.

The only meta-analytic evidence on high-variability practice, and it is weak. Differential learning shows a small acquisition effect (d=0.26) growing to medium at retention (0.61) — directionally consistent with the story that variability pays off later, not sooner. But the authors themselves say 'inferences about the effectiveness of DL would be premature': high risk of bias in nearly all studies, funnel asymmetry indicating publication bias, I² ~78-79%, small underpowered samples, and no transfer analysis at all. DECISIVE AND PREVIOUSLY UNRECORDED: the whole DL advantage is against TRADITIONAL REPETITIVE practice. Head-to-head against other variability-based methods it disappears — vs structural learning d=-0.19 at acquisition and -0.18 at retention, vs contextual interference d=-0.04 (technical skills) and +0.13 at retention. The authors' own abstract concedes: "For robust comparisons on the relative effectiveness of DL to different variability-based motor learning methods, scarce and inconclusive evidence was found." Two whole acquisition subgroups are null or negative: fine motor skills d=-0.12 and tactical behaviour d=0.20 (ns). Related primary work (Caballero et al. 2024, PeerJ 12:e17575, 'more is not better') finds variability is not monotonically beneficial.

Not addressed. Baseline skill and intrinsic movement variability differ substantially between learners and are plausibly heritable; DL studies do not model who benefits.

No preregistered replication. The classic schema-theory variability-of-practice prediction remains inconsistently supported; no modern meta-analysis isolates variable vs constant practice on delayed transfer.