The statistics used in education research are based on central trends such as the mean or standard deviation, discarding outliers. This paper adopts another viewpoint that has emerged in statistics, called extreme value theory (EVT). EVT claims that the bulk of normal distribution is comprised mainly of uninteresting variations while the most extreme values convey more information. We apply EVT to eye-tracking data collected during online collaborative problem solving with the aim of predicting the quality of collaboration. We compare our previous approach, based on central trends, with an EVT approach focused on extreme episodes of collaboration. The latter provided a better prediction of the quality of collaboration.