Assessing the impact of increased selection of minority applicants on performance and monetary utility

There are usually two main purposes for increased minority hiring: (1) reparation for prior discrimination, and (2) present and future selection fairness. Previous discrimination refers to intentional or unintentional bias that results in disparate impact in hiring, while present and future selection fairness refers to avoiding bias in selection as a result of bias in measurement or as a result of how fairness is defined. Minority individuals have, on average, sometimes performed lower on both predictor and criterion measures. As a result, if such differences exist, are not due to bias in measurement, and minority hiring is increased in order to make up for prior discrimination or for possible bias in measurement, then losses in performance and monetary utility will accrue. A Monte Carlo simulation was used to quantify these losses both in terms of overall performance and in terms of the performance level of the minority hire expected to perform worst. This was done for various combinations of predictor validity, selection ratio, minority applicant base rate, and criterion difference between majority and minority groups. As minority hiring was increased above the valid minority hiring rate, performance and monetary utility losses began to appear. These losses accelerated as minority hiring increased. Additionally, the performance percentile of the minority hire expected to perform worst, can drop rapidly as minority hiring increases. This should be a prime concern for jobs where poor performance can lead to loss of life or other similar consequence. Uses for the utility loss calculations were suggested. For the purpose of reparation for prior discrimination, this utility loss information could provide a means to evaluate the impact of affirmative action programs. For the purpose of present and future selection fairness, this information could be used to replace present selection fairness models since the measurement error assumptions made by those models are not likely to exist to the degree that they postulate. Rather, a compromise is necessary, and utility loss information provides such a means. Limitations of the present study were presented and suggestions for future research were made.