Fixation duration and the learning process: an eye tracking study with subtitled videos

Learning is a complex phenomenon and education researchers are increasingly focussing on processes that go into it. Eye tracking has become an important tool in such research. In this paper, we focus on one of the most commonly used metrics in eye tracking, namely, fixation duration. Fixation duration has been used to study cognition and attention. However, fixation duration distributions are characteristically non-normal and heavily skewed to the right. Therefore, the use of a single average value, such as the mean fixation duration, to predict cognition and/or attention could be problematic. This is especially true in studies of complex constructs, such as learning, which are governed by both cognitive and affective processes. We collected eye tracking data from 51 students watching a 12 min long educational video with and without subtitles. The learning gain after watching the video was calculated with pre- and post-test scores. Several multiple linear regression models revealed a) fixation duration can explain a substantial fraction of variation in the pre-post data, which indicates its usefulness in the study of learning processes; b) the arithmetic mean of fixation durations, which is the most commonly reported eye tracking metric, may not be the optimal choice; and c) a phenomenological model of fixation durations where the number of fixations over different temporal ranges are used as inputs seemed to perform the best. The results and their implications for learning process research are discussed.