Selecting a Measurement Model for the Analysis of the National Institutes of Health Stroke Scale

Abstract

To select the most appropriate model for the analysis of data from the National Institutes of Health Stroke Scale (NIHSS), the graded-response, Rasch partial credit, and generalized partial credit models were used to analyze NIH stroke data of 1,191 acute ischemic stroke patients. Based on Akaike's Information Criterion (AIC) and Bayesian Information Criterion (BIC), the generalized partial credit model has the most generalizable parameters. Items on the NIHSS have different discriminating powers. The generalized partial credit model, which allows varying slopes of item response functions, is the most appropriate model for the analysis of the NIHSS.

Keywords:

• AIC
• BIC
• item response theory
• measurement models
• NIHSS
• stroke