Using Hidden Markov Models to Investigate Learning Behavior and Its Correlates: An Application to Process Data from a Morphological Training App

Esther Ulitzsch

Abstract

Learning apps generate rich, high-dimensional data that offer substantial potential for understanding how students approach learning tasks. Hidden Markov models (HMMs) provide a principled framework for summarizing sequential multivariate data in terms of interpretable latent states while explicitly modeling transitions among them. In this talk, I illustrate and critically examine the utility of HMMs using data from Captain Morph, a morphological vocabulary-training app for primary school children, across three task types to evaluate replicability and generalizability. I conclude by discussing implications for the design of learning apps and reflecting on the broader potential of HMMs to support adaptive learning.