Measuring the Stability of Rasch Fit Statistics Using Nonparametric Bootstrapping

• In Event: Methodological Considerations for Rasch Measurement Research

Abstract

In Rasch analyses, residual fit statistics such as infit and outfit are commonly used to evaluate model-data fit. However, these statistics are typically reported as point estimates, lacking measures of uncertainty. While previous studies have used bootstrap methods to derive empirical confidence intervals for fit statistics, these resampling techniques could also be used to give measures of instability for fit statistics. This study evaluates the stability of Rasch fit statistics using the coefficient of variation across nonparametric bootstrapped resamples. We assessed stability across different simulation conditions and found that small samples increased item fit instability, and short tests increased person fit instability. These results highlight the need for caution when interpreting fit statistics derived from small samples or short tests.

Authors

• Andrew Krist, University of Alabama

• Stefanie A. Wind, University of Alabama