The Sensitivity of Rasch Tree DIF Detection to Model Misspecification

The Rasch Tree model for differential item functioning (DIF) detection is a useful tool that uses exploratory yet interpretable methods to investigate intersectional DIF. Recent research has explored Rasch Tree performance under specific data conditions. The present research expands upon this by examining the effect of Rasch model misspecifications (i.e., misfit) on Rasch Tree estimation performance. Using a Monte Carlo simulation, we test the effects of including discrimination and pseudo-guessing parameters in the data generation process on DIF detection accuracy, DIF magnitude estimates, and resulting Rasch Tree structure. Results indicate that imposing the Rasch model on data that contain varied item discriminations and guessing behavior leads to a decrease in DIF detection accuracy and stability. In addition, we also assess these effects under different sample sizes and test lengths and examine the Rasch Tree’s ability to distinguish between person ability differences and item level DIF under various conditions. We discuss implications and future directions from this research.