Logistic Regression and Model Based Recursive Partitioning for Item Evaluation

Abstract

Complex sampling plans are common in large datasets that are developed in national and international contexts. The multilevel structure of such data is handled most frequently by multilevel models. However, analysis becomes complex under such conditions, particularly if estimates (e.g., regression slopes) of individual units is desired. Such cases include validity studies where differential item function analyses (DIF) are conducted. This simulation study evaluated the accuracy of model based recursive partitioning (MBRP) using logistic regression for DIF detection under various conditions. Results suggest that MBRP is a promising technique for linear and logistic models, including those used for DIF detection. Implications and future directions for research are discussed.