Affect Dynamics Analysis

D’Mello and Graesser’s (2012) highly-cited model of affect dynamics proposes a sequence of theoretically-grounded transitions between affective states during learning. However, empirical studies in a range of contexts have not produced the predicted results. In this work, we provide a detailed analysis of the prior affect dynamics studies elaborating on the contextual and methodological differences in them. We describe the steps involved in affect dynamics analysis using L with clarifications on the edge cases that have mostly been omitted from write-ups on how the prior affect dynamics studies were conducted. We present a mathematical evidence that several past studies used the L statistic incorrectly, leading to invalid conclusions of statisticial significance and provide a correction to the interpretation of L statistic. Using a corrected analysis method, we re-analyze ten past affect datasets collected in diverse contexts and synthesize the results to find if there is an empirical evidence for the D’Mello and Graesser’s widely accepted model.

[git] [presentation]

Data - Affect observations
Method - Meta analysis, transition analysis

• Karumbaiah, S., Andres, J.M.A.L., Botelho, A.F., Baker, R.S., Ocumpaugh, J. (2018) The Implications of a Subtle Difference in the Calculation of Affect Dynamics. Proceedings of the 26th International Conference on Computers in Education (ICCE). [pdf] Nominated for Best Paper Award
• Karumbaiah, S., Baker, R.S., Ocumpaugh, J. (2019) The Case of Self-Transitions in Affective Dynamics. Proceedings of the 20th International Conference on Artificial Intelligence in Education (AIED). [pdf] [git]