Fairness in employment testing.
Fairness in employment testing.
复制标题
就业测试的公平性。
DOI：
10.1126/science.2740906
复制
发表时间：
1989
期刊：
影响因子：
56.9
通讯作者：
A. Wigdor
中科院分区：
文献类型：
--
作者：
J. Hartigan;A. Wigdor
康斯坦斯霍尔顿（新闻与评论，6月2日，第1036页）写了一个有用的总结，在国家研究理事会的一般能力倾向测试电池（GATB）的最新报告中解决的问题。然而，她严重歪曲了委员会关于调整黑人和西班牙裔考生分数的立场，以便有能力的非白色工人与有能力的白色工人有同样的机会被推荐工作。正如霍尔顿文章中的数字所显示的那样，直接使用测试分数，而不进行调整，将导致错误地拒绝更多有能力的黑人和西班牙裔工人，而不是有能力的白色工人（尽管所有群体中的一些有能力的工人都会被错误地拒绝）。这并不像霍尔顿说的那样是因为测试有偏见。”假阴性”效应不是种族或族裔的函数。相反，有能力的少数民族工人被不成比例地拒绝是由于两个因素的相互作用：测试的适度预测准确性和这些申请人的平均测试分数较低。两者的结合意味着更多的黑人和西班牙裔考生谁可以表现良好的工作将被错误地预测为不满意。此外，委员会并不相信，正如文章的读者可能会怀疑的那样，目前由劳工部就业服务部门使用的组内百分位数评分系统是确保有能力的少数民族和白色申请人获得平等推荐机会的唯一方法，或者在任何情况下都是适当的方法。我们对组内百分位数方法的认可显然与GATB目前的预测能力有关。只要GATB预测工作绩效的准确性不高（相关系数为0.3），那么基于群体规范的得分将使有能力的白色、黑人和西班牙裔员工的推荐率大致相等。关键的一点是，影响推荐结果所需的调整大小必然取决于测试预测工作绩效的准确性。报告中特别认可的第二种评分策略（一种所谓的基于绩效的方法，即按组调整考试分数，以便在给定的工作绩效水平下，所有组的考试分数分布相同）的吸引力在于，它能对考试预测准确性的变化做出反应。高度准确的预测意味着较小的分数调整;在当前的准确度水平下，调整将与组内百分位数系统产生的调整大致相同。最后，霍尔顿引用了一位人事管理办公室官员的话，他说，大量的研究表明，纯粹的分数排序”是唯一科学合理的立场。“我们不同意。的确，”纯等级排序”的选择将产生具有最高预期生产率的劳动力。但同样真实的是，有能力的黑人和西班牙裔工人在同样的工作表现水平上被这种推荐政策拒绝的频率要比白人高得多。这是一个科学事实，在报告中从理论和经验上得到了证明。约翰·哈迪根
Constance Holden (News & Comment, 2 June, p. 1036) has written a useful summary of the issues addressed in the National Research Council's recent report on the General Aptitude Test Battery (GATB). However, she has seriously misrepresented the com-mittee's position on adjusting the scores of black and Hispanic test-takers so that able nonwhite workers have the same chance of referral to jobs as able white workers. As the figure reproduced in Holden's article illustrates, the direct use of test scores, without adjustments, will resultin the false rejection of a larger proportion of able black and Hispanic workers than ofable white workers (although some able workers in all groups will be erroneously rejected). This is not because the test is biased, as Holden says the report implies. The" false negative" effect is not a function of race or ethnicity. Rather, the disproportionate re-jection of able minority workers is due to the interplay of two factors: the modest predictive accuracy of the test and the lower average test scores of these applicants. The combination of the two means that propor-tionately more black and Hispanic test-takers who could perform wellon the job will be falsely predicted to be unsatisfactory. Moreover, thecommittee does not be-lieve, as a reader ofthe article might surmise, that the within-group percentile scoring sys-tem currently being used by the Labor De-partment's EmploymentService is the only way-or in all circumstances an appropriate way-to ensure equal referral chances for able minority and white applicants. Our endorsement of the within-group percentile method is clearly linked to the current predictive power of the GATB. As long as the GATB predicts job performance with only modest accuracy (correlation, 0.3), scores based on group norms will achieve approximately equal referral rates for able white, black, and Hispanic workers. A crucial point is that the size of the adjustment needed to effect the recommended outcome will necessarily depend on the accuracy with which job performance is predicted by the test. The attraction of the second scoring strategy specifically endorsed in the report (a so-called performance-based method in which testscores are adjusted by group so that the distribution oftest scores at a given level of job performance is the same for all groups) is that it is responsive to changes in the predictive accuracy of a test. Highly accurate prediction would mean small score adjustments; at current levels of accuracy the adjustments would be just about the same as those produced by the within-group percentile system. Finally, Holden quotes an official from the Office of Personnel Management as saying that the vast bulk of research shows that pure rank-ordering of scores" is the only scientifically justified position." We disagree. It is indeed true that selection on" pure rank-ordering" will generate a work force with the highest expected productivity. But it is also true that able black and Hispanic workers will be rejected far more frequently by such a referral policy than whites at the same level ofjob performance. This is a scientific fact, demonstrated theo-retically and empirically in the report. JOHN HARTIGAN