Analysis of reaction times using mixed-effects models

Abstract

The aim of this article is to illustrate the modeling of response times using linear mixed-effects model technique, offered as an alternative to standard analyses from the general linear models family. Characteristics of response times, used as the main dependent variable in many psychological experiments, are analyzed in detail. This measure often does not meet the statistical prerequisites of various parametric analyses in which it is used, such as the normality of distribution, independence of observations and the homogeneity of variance. Various transformations and complex statistical procedures, linear mixed modeling being one, are used in those situations. A specific advantage of the analysis used in this research is its ability to model random effects, for example, of participants and stimuli. This way, additional sources of response times variability are accounted for and the findings become generalizable to other objects of research. The same is made difficult for analyses in the general linear models family because of the use of values averaged over participants and/or stimuli and F1 × F2 testing procedure. Models with different degrees of complexity are compared in this research using fit indices, to test whether the introduction of additional parameters in model is warranted. Finally, a model criticism is offered, confirming its suitability for the data used.