SIGNIFICANCE TESTS AND CONFIDENCE INTERVALS FOR THE ADVERSE IMPACT RATIO

Abstract

The two most common methods for assessing adverse impact, the four‐fifths rule and the z‐test for independent proportions, often produce discrepant results. These discrepancies are due to the focus on practical versus statistical significance, and on differing operational definitions of adverse impact. In order to provide a more consistent framework for evaluating adverse impact, a new significance test is proposed, which is based on the same effect size as the four‐fifths rule. Although this new test was found to have slightly better statistical power under some conditions, both tests have low power under the typical conditions where adverse impact is assessed. An alternative to significance testing would be to report an estimate of the adverse impact ratio along with a confidence interval indicating the degree of precision in the estimate.