Poster #31 - The Different Effects of Collapsing Categories on Graded Response Model and Generalized Partial Credit Model

• In Event: 79.052 - MTCC Poster Session 19
  In Poster Session: 79.052-2 - Educational Measurement, Psychometrics, and Assessment Poster Session

Abstract

In ordinal data analysis, when an item has some category with low endorsement rate, it is common practice to collapse the adjacent categories. However, until recently the effect of collapsing on polytomous item response theory (IRT) modeling has not been thoroughly investigated. Harel (2014) provided an in-depth analysis on the effect of collapsing categories on partial credit model (PCM) and generalized partial credit model (GPCM), claiming that collapsing would lead to model misspecification and hence should be avoided in most circumstances. In this paper, we provide theoretical and simulation results to show that graded response model (GRM), another commonly used polytomous IRT model, is a safer option when collapsing categories is required.

Authors

• Shengyu Jiang, University of Minnesota - Twin Cities

• Chun Wang, University of Washington