A Latent Markov Model for Multimodal Time-Series Data in Computer-Interactive Assessments (Poster 5)

• In Event: Excellence in Education Research: Early Career Scholars and Their Work

Abstract

The study presents a latent Markov model (LMM) that describes student’s interaction patterns in computer-interactive assessments. Current development in LMMs assumes strong measurement invariance and has limited utility for modeling assessment interaction data. In this study, we propose a flexible and efficient LMM framework that supports analysis of time-intensive multimodal computer-interaction data. The study constructs element models of LMM accommodating various interaction variables (e.g., item response, time stamp, action counts) and population variables (e.g., demographic profiles, cluster covariates). The inferential procedures for the new framework are developed based on the Baum-Welch algorithm and are validated through Monte Carlo simulation and empirical data application. The current investigation suggests that the new framework adequately accounts for item and subject effects and provides more robust inference. Compared to the existing model, the new framework achieves greater model fit and is more robust and accurate in eliciting the latent profiles and retrieving the generating parameter values. The proposed framework provides a synthetic analytical tool for analyzing multivariate cross-sectional time series and well accommodates educational assessment data that show distinct item and individual effects. In particular, the new framework promises great potential for describing test-takers’ problem-solving modalities, such as distinct interaction patterns, evolution of latent state profiles, and manifestation in the observable indicator variables.

Author

• Hyeon-Ah Kang, University of Texas at Austin