Power and errors in adverse impact indices
研究了八种不利影响指标在不同条件下的统计功效和错误率，发现Fisher精确检验和连续性校正的Z检验在多数情况下表现较好，但某些条件下实用测试更优。
Adverse impact indices are intended to detect potentially discriminatory employment practices, and although the number of such indices continues to increase, with a few exceptions, little is known about the comparative performance of those indices under various conditions. We examined the performance of eight such indices, including statistical significance tests, ad hoc “practical” tests proposed by the courts, and effect size measures. The power of these indices was examined under various levels of adverse impact, applicant pool sizes, and selection ratios. Although no index performed well in all circumstances, typically, Fisher’s exact test and the Z test for proportions (with correction for continuity) performed adequately most often. However, in some conditions, these indices were outperformed by some of the “practical” tests.