Analysis of event-related potential difference waves can benefit from linear mixed effects modeling: Recommendations for analyses and general model fitting

Abstract

Linear mixed effects models (LMEs) have advantages for analyzing mean amplitude event-related potential (ERP) data. Compared to ANOVA and linear regression, LMEs retain more subjects and yield unbiased parameter estimates by accounting for trial-level sources of variability. However, LME analysis of ERP mean amplitude difference waves may be problematic due to the need to pair single trial data to create trial-level difference waves. In both simulated and real pediatric ERP data, the present study compares ERP difference wave results across conventional ANOVA/regression analyses and six trial-level LME approaches in different low trial-count scenarios. We evaluate each approach based on accuracy of estimates and statistical power in simulated data, and magnitude of effect detected in real ERP data from 3- to 5-year-old neurotypical children (N = 64). Two analysis approaches were unbiased: creating trial-level difference waves by pairing trials on all study design features (the ‘exact match’ approach) and fitting an interaction term; and the interaction term had greater power to detect a significant effect in simulated data. Both simulations and analysis of real preschooler ERP data support using LMEs to analyze difference waves. We also include recommendations for researchers for picking a difference wave approach appropriate for their research question.