Studying Affect Dynamics Using Epistemic Networks

Abstract

The study of how affect develops and manifests over time during learning is a popular area of research called affect dynamics. Students’ affective states are recorded in authentic settings like classrooms using direct observations by culturally sensitive, trained, and certified coders. A popular approach to studying affect dynamics in the last decade involved a transition metric called the L statistic. However, recent studies have reported statistical errors and other discrepancies with L statistic leading to questions about its reliability. Thus, we turn to epistemic network analysis (ENA), an emerging technique that is gaining popularity in studying the structure of temporal interconnections between codes. In this paper, we present an alternative approach to study affect dynamics by extending ENA to include directionality in the network edges to capture transitions. We also propose a new approach to running significance tests on network edges to identify significantly likely transitions. Then, we apply the two techniques – L statistic and ENA - to a previously collected affect dataset from a middle school math class, in order to better understand the trade-offs between these methods. Our analysis revealed that ENA could be a promising new approach to conduct affect dynamics analysis. In addition to avoiding statistical errors seen in L statistic, ENA offers better visualization which better emphasizes the magnitude of a transition’s strength. We discuss the assumptions in ENA that need to be vetted further and the possibility for new kinds of analysis in the future for affect dynamics research using ENA.