Reduction of Bias in Rasch Estimate * Due to Aberrant Patterns. Rapport 87-5.

The purpose of this study is to determine an efficient way to reduce the bias in estimates of the Rasch model parameters due to aberrant response patterns. First, the benefits of using one- or two-sided goodness-of-fit tests of patterns witn the model are discussed. Then, the consequences of removing non-fitting patterns from Rasch model data arc considered. Finally, an iterative procedure to reduce the bias is presented. This procedure replaces non-fitting patterns by certain patterns sampled according to the model. The effectiveness of this procedure is investigated in a simulation study using Rasch model data mixed with aberrant response data. It is also demonstrated that, for aberrant response behavior ttlat too often results in ideal patterns, another strategy for detecting aberrant patterns is n'eded. Such a strategy would provide a possibility to detect more types of aberrant behavior and to reduce bias in the estimates of tha model parameters. Four data tables and seven graphs are included. (Author/TJH) * Reproductions supplied by EDRS are the best tnat can be made * * from the original document. * *** * ********* * 4*** * ********* * x **** * **********t********** * *********** * *