Rasch models in SAS

Introduction

Rasch models are used when a set of items are used to measure a uni-dimensional latent variable. Originally used in educational testing models, they are also used for measurement in psychology and physical rehabilitation. This paper describes a collation of SAS macros that fits Rasch models. Implementation of item parameter estimation for polytomous Rasch models using MML, CML and PCML in SAS has been discussed earlier (Christensen 2006, Nandakumar 2012), and SAS macros have been made available (Hardouin 2007, Christensen 2013, Christensen 2013a).

The SAS macros presented here provide a unified framework where any method of item parameter estimation can be applied to items with varying number of response options. Item parameter estimation can be done using marginal maximum likelihood (MML), conditional maximum likelihood (CML) or pairwise conditional maximum likelihood (PCML) estimation. Person locations can be estimated using ML or weighted MLE (Warm 1989). A number of item fit statistics and graphical presentation of item characteristic curves (ICC’s) and item fit plots are included.

Simulation is implemented.