Investigating the Performance of Person-Fit Measures Under Rasch Multidimensional Models

• In Event: Comparing, Decomposing, and Exploring Models

Abstract

Person-fit measures aim to diagnose aberrant responses. Most of the existing person-fit measures were proposed and evaluated under unidimensional item response theory model. The current study examined how the selected person-fit measures (HT, D( ), MCI, U3, lz, and M) performed when data followed multidimensional Rasch model. We varied the following factors: model for data generation (noncompensatory, compensatory and between item Rasch model), test length (20, 40 and 80), percentage of aberrant examinees (15% and 25%), percentage of aberrant response items (10% and 20%) and the condition with no aberrant examinees. Cutoff values based on unidimensional Rasch models were derived first, based on which, Type I error rates and detection rates were calculated for each condition.

Authors

• Yan Xia, Florida State University

• Insu Paek, Florida State University