The Distribution of Between-Dimension Correlation in Misspecified Multidimensional Rasch Models in Unidimensional Data

• In Event: Issues of Rasch Dimensionality, Scaling, and Fit

Abstract

The purpose of this project is to elicit the distribution of between-dimension correlation for misspecified multidimensional Rasch models in unidimensional data. Specifically, we sought to answer the following research questions: What is the distribution of between-dimension correlation for misspecified two-dimensional models for unidimensional data? How do test length, sample size, and misfit affect the distributions? How do these affect the accuracy of decisions using the convention of treating correlations of 0.90 as representing a unidimensional instrument? Based on the results, the rule of thumb seems to be a reasonable basis for a decision most of the time. The rule could lead to an erroneous decision, however, in cases of short exams, especially when the data contain more model-data misfit.

Author

• Leigh M. Harrell-Williams, Georgia State University