Title:A Doubly Latent Space Joint Model for Local Item and Person Dependence: An Application for Analyzing Deductive Reasoning Test Data

Abstract:We present a new statistical approach to analyze binary item response data from the Competence Profile Test of Deductive Reasoning - Verbal (DRV). The analysis of the DRV test data is challenging due to the data's unknown local item and person dependence structure. Existing methodologies cannot properly handle such data. Our innovation is to expand the current latent space joint modeling framework to deal with dependencies among items and among respondents. Specifically, our approach estimates relative distances between pairs of items and relative distances between pairs of persons to describe item and person dependence structures. The proposed approach was successfully applied to analyze DRV test data. Our analysis results revealed that there were three person clusters and four item clusters in the DRV data. We also evaluated the performance of the proposed method in terms of person clustering and compared the results against mixture Rasch analysis in a simulation study.