Comparing parametric item response theory and nonparametric item response theory: Application in psychological research using polytomous items
The current study focused on a comparison between two widely used representative Item Response Theory (IRT) models of their respective families, the Modified Graded Response Model (MGRM) from the parametric Item Response Theory (PIRT) family and the Nonparametric Kernel Smoothing (NKS) model from the nonparametric Item Response Theory (NIRT) family, using data (N = 8,780) collected on the Positive and Negative Syndrome Scale (PANSS). The models were compared in aspects of item calibration, estimation stability using various sample sizes, and cross validation performance. Results showed that the MGRM and the associated widely used software package PARSCALE demonstrated the advantage in providing summarized model output in the form of estimated parameters, higher estimation stability in equal-sized samples, and the possibility of post-hoc analyses that involve using model parameters and ability estimates. The NKS model and its associated software package TESTGRAF, showed strengths in accommodating options with low endorsement frequency, fully capturing information in the original data, generating item category response functions (ICRFs) that could be used for qualitative item analyses, producing better cross validation results for scales with adequate frequency in all response categories, as well as providing a user-friendly interface. The conclusion was that for psychological measurements that require high stability of model parameter estimates and ability estimation, PIRT models such as the MGRM, could be more desirable. For psychological measurements that require more flexibility in item calibration, when it is costly to increase sample size, and/or qualitative evaluation of each item is desirable, NIRT models such as the NKS model could provide more benefits to researchers. It was highly recommended that future research should leverage more of the modeling advantages provided by the NIRT models.
Experimental psychology|Quantitative psychology
Zhou, Ying, "Comparing parametric item response theory and nonparametric item response theory: Application in psychological research using polytomous items" (2011). ETD Collection for Fordham University. AAI3512338.