Sensitivity Analysis of Fit Indices for Multilevel-Multidimensional Item Response Models

• In Event: Measurement Models

Abstract

Multilevel-multidimensional item response theory (IRT) models have been developed to address the nested structure of item response data from the complex sampling or repeated measures designs (e.g., Fox, 2007). While the high-dimensional models have been rapidly developed along with efficient computation algorithms (e.g., Metropolis-Hastings Robbins-Monroe (MH-RM) algorithm; Cai, 2009), the model fit indices that provide information about the overall goodness of data-model fit have been less explored within the framework of multilevel-multidimensional IRT models. The purpose of this study is to evaluate the performance of currently available relative model fit indices (e.g., log-likelihood, AIC, and BIC) through simulation study and to suggest an alternative approach in the framework of multilevel-multidimensional IRT models.

Authors

• Dandan Liao, University of Maryland

• Ji Seung Yang, University of Maryland - College Park