The Effects of Contextual Interference Learning on the Acquisition and Relatively Permanent Gains in Skilled Performance: A Critical Systematic Review with Multilevel Meta-Analysis

The paradoxical effects of contextual interference (CI) assume that high CI practices hinder performances during the acquisition phase of learning, while providing more permanent enhancement during the retention phase. This meta-analysis evaluates the possible generalizability of the CI phenomenon in physical education (PE) and sports contexts, with regard to the acute and relatively permanent gains in performance outcomes. A total of 933 records from five electronic databases were screened using the PICOS criteria, of which 36 studies were selected. Outcomes evaluating the performance changes (Δ) from pre-post, post-retention, and pre-retention tests were included. Out of 183 overall pooled outcomes, Δ in only 37 performance outcomes (20%) agreed with the paradoxical CI effects on the acquisition or the relatively permanent gains. No statistically significant overall difference was detected for “Δ pre-post” between low (blocked) (28.9 ± 59.5%) and high (random/serial) (27.9 ± 52.8%) CI (effect size (ES) = 0.1, p = 0.35). An overall significant difference (p = 0.001) in favor of high CI practice was detected in “Δ post-retention.” However, this difference was not large enough (ES = − 0.35) to produce an overall greater long-term gain following high (24.56 ± 4.4%), compared to low (21.9 ± 9.8%) CI (ES = − 0.13, p = 0.18). Out of 10 tested variables, only the age significantly moderated both CI effects (p < 0.0001 for both Δ pre-post and Δ pre-retention) and the female proportion significantly moderated only the first CI effect (p = 0.009 for Δ pre-post). These findings found very limited evidence supporting the recommendation to employ high CI practices to gain a longer-term performance advantage, calling into question the generalization of the CI model to PE and sports practices. High-quality follow-up research evaluating alternative motor-learning models are therefore needed.