A comparison of Andrich's rating scale model and Rost's succesive intervals model

This study compared and contrasted two IRT models for measuring attitudes: Andrich’s rating scale model (ARSM) and Rost’s successive intervals model (SIM). While these IRT models require that the attitude scale be unidimensional, they make different assumptions in the development of their item parameters. The ARSM and SIM were compared in the context of a computer adaptive test (CAT). Two data sets were used. The Audit of Administrator Communication (ADCOM) data set is archival and allowed for comparison of the models in an actual testing environment. A second data set was simulated using the linear factor analytic approach. The models were compared using Pearson product-moment correlations, standard errors and number of items administered during a CAT. In addition, RMSE and bias estimates were calculated. Results indicated that each model has advantages within a CAT context. The ARSM provided a better estimate of theta and the SIM required fewer items to estimate theta. Suggestions are provided as to the choice of model to use in different research settings.