A Comparison of the Partial Credit and Graded Response Models in Computerized Adaptive Testing

Abstract

Computerized adaptive testing (CAT) is a procedure for administering tests which are tailored individually for each examinee. Although the majority of CATs are based on dichotomous item response theory (IRT) models, some researchers have explored the use of polytomous IRT models, such as the graded response (GR) model and the partial credit (PC) model, in CAT. This study investigated (a) the robustness of PC and GR model-based CATs' ability estimation to the use of items which did not fit these models, and (b) how accurate the PC and GR model-based CATs' ability estimates were compared to one another. Results showed that for the PC CAT reasonably accurate ability estimation (r% r 0.921) may be obtained despite adaptive tests which, on average, contained up to 45% misfitting items. Furthermore, the inclusion of misfitting items did not appear to increase the PC CAT test lengths. The GR CAT administered tests comparable in length to the PC CAT and provided ability estimates which were slightly more accurate...