RaschFit.sas: A SAS Macro for Generating Rasch Model Expected Values, Residuals, and Fit Statistics created by Tian Song, Edward W. Wolfe

The SAS macro RaschFit.sas produces data-to-Rasch model person and item fit indices as well as expected values and residuals, given parameter estimates and scored data. Specifically, the macro takes as input three files: (a) a file containing person Rasch model parameter estimates, (b) a file containing item Rasch model parameter estimates, and (c) a file containing a scored data matrix. RaschFit.sas uses these input files to compute four person and item fit statistics (unweighted and weighted mean square and the standardized unweighted and weighted mean square) using the equations specified in Chapter 5 of Wright and Masters (1982). In addition, the macro outputs files containing (a) a listing of the expected and residual scores for each person-by-item combination and (b) a matrix of data-to-model residuals. The macro can accommodate any combination of dichotomous, partial credit, and/or rating scale item structures. The output has been validated by replicating the fit values output by WINSTEPS (Linacre, 2012). The macro should be useful for researchers or analysts interested in computing fit indices or outputting expected scores and residual matrices from parameter estimation programs that do not provide that information. The macro requires the user to provide separate Excel files containing item and person parameter estimates for the dichotomous rating scale or partial credit Rasch model as well as a data set provided by the user. The person parameter estimates file must include separate columns for a person identifier and the ability estimate. The item parameter file must include separate columns for an item identifier and the item difficulty estimate. In addition, the item parameter file should include a separate column containing item thresholds when more than two score categories exist (i.e., for the rating scale or partial credit models) in addition to a final column indicating the number of categories for the item. Scored data are read from a flat file that assumes a single column for each item response.