Abstract

In a linear mixed-effects model for psychological experimental data analysis, the effect of the model selection procedure on detecting the experimental condition effect was investigated through a Monte Carlo simulation study. Specifically, while changing the complexity of the random effect structure of the data-generating model, the type I error rate and power were compared between the model selection strategies. As a result, the maximal model approach (or the random slope model) showed relatively low statistical power under the condition that the structure of the random component of the data was simple. On the other hand, when the model comparison approach was used, the Type I error rate approached the significance level, and the power was superior to or equivalent to that of the maximal model approach in all simulation conditions of this study. Finally, we discussed the points that experimental researchers should consider when using the linear mixed-effects model as an analysis tool.