Analyzing Processing Time Data in Applied Linguistics and Second Language Research: A Multivariate Mixed-effects Approach

In this paper, I introduce the use of an extension of the popular linear mixed-effects models: multivariate mixed models which can handle multiple outcome variables. This technique is especially useful for applied linguists and second language researchers who use processing time data (e.g., accuracy and reaction times) obtained from timed decision tasks, text-based eye tracking, and self-paced reading. It can address the long-standing issue of multiple comparisons as a result of having multiple outcomes (e.g., first fixation durations and total reading times in text-based eye tracking, and time spent in different regions in self-paced reading). This technique also provides exciting opportunities for researchers to ask new questions that could not be addressed in a straightforward manner with traditional statistics. With this technique, researchers are able to investigate differential effects of a predictor on different outcomes. Through a demonstration in R using published, open eye-tracking data, I contextualize my discussion of the technique, offering also practical, step-by-step, and annotated guidelines for interested researchers.