Examining the Impact of Nonnormality on Parameter Estimation in Bifactor Graded Response Model

• In Event: Applications in Psychometric Models

Abstract

This study investigates the impact of nonnormality on parameter estimation in a bifactor graded response model (Bifactor-GRM) using Monte Carlo simulations. The Bifactor-GRM accommodates a general factor and multiple specific factors, allowing for a more flexible representation of latent constructs. We examine how the skewness and kurtosis of the general and specific factors, along with other design factors, including sample size and number of items per factor, affect parameter recovery. Item parameters (discrimination and thresholds) and person ability on the general and specific factors estimates are assessed using bias, root mean squared error (RMSE), and the correlation. The results indicate that nonnormality significantly impacts parameter estimation and that the choice of estimation method plays a crucial role in accurate parameter recovery.

Authors

• Jujia Li, University of Alabama

• Chunhua Cao, University of Alabama

• Wenchao Ma, University of Minnesota