Finite mixture modeling of eye-tracking data: A tutorial using an example of L2 metaphor reading

Eye movement measures provide detailed information about the time-course of linguistic processing. The current standard for statistical analyzes of eye movement data is the linear mixed model, which assumes that the observations (i.e., eye fixation times) originate from a single process. However, many eye fixation measures typically reported in reading studies may reflect a mixture of underlying processes. For example, gaze duration, which is calculated as the sum of fixations made on a word during its first-pass reading, may contain a composite of observations originating from two types of processes: a fast and efficient process characterized by short gaze durations, and a slower and more effortful process characterized by longer gaze durations. Finite mixture models offer a statistical tool for examining whether gaze durations are driven by multiple underlying processes. Furthermore, these models allow testing hypotheses on whether linguistic manipulations influence the means of these distributions or their proportion. In the present tutorial, we briefly explain the background of this analytic approach and demonstrate how finite mixture models can be used to analyze an eye movement dataset. We suggest that finite mixture models are a valuable tool for examining the underlying processes reflected in eye fixation times or other similar measures, and hope that this tutorial will help researchers to make use of this method.