Conditional and Marginal Strengths of Affect Transitions During Computer-Based Learning

Understanding the transitions among affective states during computer-based learning may guide the design of affect-responsive learning environments. Current studies have focused on the marginal strength of an affect transition, which is the average transition tendency over possible affective states preceding the transition. However, marginal strength ignores the potential influence of the preceding state on the transition. In contrast, a conditional strength, which is the transition tendency given a particular state preceding the transition, accounts for this influence and may contribute to a more comprehensive understanding of students’ learning processes. This paper presents a methodological framework that utilizes the logistic mixed model to compute the conditional strengths of affect transitions and examines whether conditional and marginal strengths are equal. In three real-world datasets, we found that the conditional and marginal strengths of a transition were not identical in most cases. Prediction analysis indicated that accounting for the state preceding the transitions resulted in better affect prediction performance. In addition, empirical data analyses showed that the framework had higher power in detecting the impact of students’ factors on affect and affect transitions. The framework also allows researchers to specify the reference transition when computing a transition strength and handle self-transitions, a critical issue in affect transitions. Empirical data analyses showed that the strength of a transition varied substantially when the reference transition changed, highlighting the careful selection of reference transitions in transition analyses.