Experimentally elicited productions: Differences and similarities between mixed effects and ANOVA analyses

Currently, many experimental studies of speech production use fully counterbalanced designs to examine variation in categorical (e.g., correct/incorrect) or relatively continuous measures (e.g., reaction times, voice onset times). These data present several challenges to ANOVA analyses. Some of these issues are well known to be the speech community; for example, the non-normality of dependent variables such as proportion correct. Others have been less extensively addressed; for example, many speech studies account for participant- but not item-specific contributions to variance. I’ll discuss the opportunities and challenges in using linear mixed effects models to address these issues. I’ll review some of the common issues that arise in using such models and discuss how to interpret and report their results.