INVESTIGATING THE FIT OF THE GENERALIZED GRADED UNFOLDING MODEL (GGUM) WHEN CALIBRATED TO IRT GENERATED DATA FROM DOMINANCE AND IDEAL POINT MODELS

The assessment of model fit in latent trait modelling, better known as item response theory (IRT), is an integral part of model testing if one is to make valid inferences about the estimated parameters and their properties based on the selected IRT model. Though important, the assessment of model fit has been less utilized in IRT research than it should. For example, there have been less research investigating fit for polytomous dominance models such the Graded Response Model (GRM), and to a lesser extent ideal point models such as the Generalized Graded Unfolding Models (GGUM), both in its dichotomous and polytomous forms. For such reasons, examining fit for the GGUM is paramount and should be investigated thoroughly. The current study tests for different fit indices when calibrating the GGUM model to generated data from different IRT models. The tested outcomes consist of type I error and power rates across 100 replications for selected number of items and sample sizes with respect to different model fit indices utilized in previous IRT literature. Results from the simulation study show that relative fit indices performed well in identifying the correct dichotomous data model (i.e., GGUM) when the delta ranges are extended beyond the specified distribution ranges for the dominance models. Also, polytomous GGUM data were identified as the best fitting model in almost all the cases, irrespective of the number of items and sample size. On the other hand, the majority of absolute fit indices did not perform well in identifying fit/misfit. Advisor: Ralph De Ayal