Statistical Analysis Software (SAS), is widely-used and available in research and industry settings. Item response theory (IRT) is an important technique to calibrate and evaluate items and to score examinees from item responses on tests, surveys, and similar instruments. The PROC IRT procedure in SAS/STAT®, released December 2013, is a procedure to perform unidimensional and multidimensional item response theory with the one-, two-, three-, and four-parameter logistic models when data are dichotomously scored, and the graded response model when data are polytomously scored.

The purpose of this study is to provide a comprehensive review of the procedure and to assess its ability to recover item and latent score parameters.

A simulation study is used to evaluate the efficiency and accuracy of PROC IRT. Unidimensional and multidimensional 2PL, 3PL, and graded response models are applied to datasets of N=200, 250, 500, and 1,000 and 20 items. Evaluation criteria includes run time, percentage of non-convergence datasets, model fit statistics, item fit statistics, and measures of item parameter mean square error and bias.

In a preliminary study of 50 replications, all 2PL model fits reached convergence. The convergence rates of the 3PL models at N=250, 500, and 1000 were 22%, 72%, and 70%, respectively. The MSE of item parameter estimates decreased as N increased, and was better for the 2PL model over the 3PL.

The full paper will discuss in more details the input, options, interface, and output of PROC IRT. The procedure will also be compared to other procedures in SAS that are capable of performing IRT.

If excellent performance is displayed, this software may be preferred for conventional IRT applications over others by practitioners in the large-scale assessment or testing industry due to the essential utilization of SAS for data management and other statistical and psychometric analyses.