The Adverse Impact Toolkit software program combines "multiple events" (e.g., passing rates between two groups on a test over several years) or multiple data sets (e.g., divisions or job groups) to analyze adverse impact. The procedure involves the same type of statistical comparison as applied in the Selection Rate Comparison program, but allows for multiple data sets.
- Homogeneity of Odds/Rates Test (two-tail): Because combining multiple events or data sets can sometimes create statistical anomalies, two tests are included in the program to assess whether the multiple events or data sets can be combined for a valid overall analysis. These tests are the Breslow-Day (with Tarone’s Correction) and the Treatment by Strata Interaction Test proposed by Mehrotra. Both tests typically output similar results, and statistically significant findings indicate possible complications with combining all events into a single analysis.
- Statistical Significance Tests (two-tail): These analyses include two versions of the Cochran-Mantel-Haenszel procedure and a test called the Minimum Risk Weights Test to estimate the statistical probability of the observed passing rate difference between groups for the overall, combined data set. All three test typically output similar values, and one version of the Cochran-Mantel-Haenszel procedure typically produces values that very closely mirror the exact probability value.
|
