The reliability of six item bias indices

Abstract

The reliabilities of six item bias indices were investigated for each of the eleven tests of the Iowa Tests of Basic Skills, using random samples of fifth-grade students. The reliability of an index was defined as its stability from one randomly equivalent group to another. Both racial and sexual bias were considered. In addition, correlations among bias indices were investigated. The results indicate that the item bias indices investigated were fairly unreliable when based on sample sizes of 200 minority and 200 majority examinees. Consequently, this study suggests that, with sample sizes of about 200, the use of item bias indices to screen achievement test items cannot be expected to lead to consistent decisions about which items are biased.