Early reaction time variability predicts implicit statistical learning: a comparison of four variability indices

High intra-individual reaction time variability (RTV) is traditionally viewed through a deficit perspective and interpreted as a maladaptive signature of attentional lapses, cognitive inefficiency, and systemic noise. However, theories from motor learning and the competitive neurocognitive networks framework suggest that behavioral variability and reduced top-down control might actually facilitate certain forms of implicit skill acquisition. The present study addresses the apparent conflict between these perspectives by investigating whether elevated RTV serves as an adaptive, functional precursor to implicit statistical learning. Across two independent studies, participants completed the Alternating Serial Reaction Time (ASRT) task. We quantified early RTV during the initial task phase using multiple metrics – coefficient of variation, inter-trial RTV, and ex-Gaussian parameters Sigma and Tau – to predict subsequent statistical learning. Analyses controlled for baseline response speed and early learning artifacts, and test-retest reliability measures were also evaluated. Our results show that early RTV predicted later statistical learning measured via reaction times. This predictive relationship was most consistent for metrics capturing dynamic, moment-to-moment fluctuations (inter-trial RTV and Sigma) rather than extreme attentional lapses (Tau). While the effect size was relatively small, the association remained significant after controlling for potential statistical confounds. Furthermore, early RTV demonstrated strong test-retest stability. These findings challenge the exclusively deficit-oriented perspective on behavioral noise. Instead, we propose that elevated RTV may reflect an adaptive, exploratory processing tendency, analogous to kinematic exploration in motor learning, that could support the brain’s ability to implicitly extract and model probabilistic environmental regularities.