Using Diagnostics for Identification of Biased Test Items

Abstract

This paper demonstrates how recent developments in the analysis of regression models may prove useful in the identification of atypical and potentially biased test items. Regression diagnostics studied are based on analysis of the sensitivity of leverage points, studentized residuals, and ratios of covariances due to the sequential deletion of each test item from the analysis. These procedures appear to offer a substantial refinement over existing approaches.