Balancing Type One error and power in linear mixed models

Balancing Type One error and power in linear mixed models

Article history: Received thirteen November twenty fifteen revision received twenty-one December twenty sixteen Available online ten February twenty seventeen

Introduction

During the last ten years, linear mixed-effects models, LMMs, have increasingly replaced mixed-model analyses of variance, ANOVAs, for statistical inference in factorial psycholinguistic experiments. The main reason for this development is that LMMs have a number of advantages over ANOVAs. From a pragmatic perspective, the most prominent one is that a single LMM can replace two separate ANOVAs with subjects, F one ANOVA, and items, F two ANOVA, as random factors, which does away with ambiguities of interpretation when effects are significant in only one of the two default ANOVAs. Other advantages are, for example, better preservation of statistical power in the presence of missing data and options for simultaneous analyses of experi-