The Detection of Discrimination in Hiring: Toward Better Measurement

Paul Muter and John J. Furedy

Dept. of Psychology
University of Toronto
E-mail: muter@psych.utoronto.ca or furedy@psych.utoronto.ca

(c) Copyright 1996 Paul Muter and John J. Furedy

There is a clear need for a means to detect discrimination in hiring. Discussions of affirmative action often rest on an assumption of past discrimination. The recent U.S. Supreme Court decision stated that affirmative action is justified only if discrimination has occurred in the past. Contrary to a widespread current belief, as embodied, for example, in the Ontario Government's Employment Equity Act of 1994, the presence of discrimination cannot be deduced merely from nonproportional hiring. For example, in the National Basketball Association, white people are under-represented, but it hardly follows that the explanation is discrimination against white people in hiring. A more likely explanation is that, for a number of complex and unknown reasons, probably including different preferences of the two groups, non-whites on the average perform better than whites in basketball.

In many other contexts it is more difficult to judge whether discrimination in hiring has occurred. We propose an alternative measure of discrimination in hiring: post-hiring performance. It is assumed that qualifications are distributed normally for each group and that only people to the right of a criterion are hired. Consider a case in which it is assumed that qualifications are distributed equally for men and women, and in which it is assumed that there is discrimination against women in hiring: The hiring criterion for women is higher than the criterion for men. Under these assumptions, it can easily be demonstrated that the qualification scores of hired women have a higher mean and lower variance (a measure of variability) than the qualification scores of men. If there has been discrimination in hiring against women in a certain period, it is reasonable to predict that the post-hiring performance of women hired in that period should have a higher mean and lower variance than the post-hiring performance of men. (It is easy to demonstrate that this prediction remains the same if it is assumed that the means of the two underlying distributions are unequal.) If there were no discrimination in hiring, then performance means and variances of men and women should tend to be the same, under these assumptions. Finally, if there were discrimination in favour of women, their scores should have a lower mean and higher variance. (If there were no discrimination in hiring and women on the average had superior qualifications, it is easy to demonstrate that the prediction would be that post-hiring performance of women would have a higher mean and higher variance, and that women would be hired more than proportionally.)

Two limitations of the post-hiring performance measure must be recognised. The first limitation is that it reflects discrimination only at the point of hiring, and not in the total life history of each group. Accordingly, a demonstration of reverse discrimination in hiring does not automatically imply that the hiring policy should be changed. Those who argue for the importance of compensating for past hiring discrimination, or that, for other reasons, a group should have more representation, could still suggest that a reverse discrimination hiring policy is appropriate.

A second limitation of the proposed measure is that there are many possibilities for post-hiring discrimination that might diminish the mean performance of one group. (A measure that would avoid this problem would be qualifications at the time of hiring, if this information were available; the same reasoning and analysis would apply for this measure as for post-hiring performance.) However, the variance of the post-hiring performance of the group discriminated against should still be lower than the variance of the other group, under the present assumptions. In addition, assuming that post-hiring discrimination is in the same direction as any hiring discrimination, if post-hiring performance of one group is superior, the indication of discrimination against that group is quite clear, though a finding of no difference or a difference in the opposite direction may be ambiguous because of this problem of post-hiring discrimination.

A corollary to the principle that nonproportional hiring does not necessarily indicate discrimination in hiring is that proportional hiring does not necessarily indicate fairness in hiring. As an illustrative example, in major league baseball in 1993 (the most recent complete season before 1996), five of the twenty-eight teams were managed for the entire season by non-white men. This 5:23 ratio corresponds roughly to the proportion of non-whites to whites in North America. However, these five teams had a record of 443 wins and 367 losses, substantially better than the combined record of the other 23 teams. This post-hiring performance measure confirms the expectation that there was hiring discrimination against non-whites for these positions. The interpretation in terms of the present analysis would be that the mean qualifications score for non-whites was higher, but the criterion was also (unfairly) higher, yielding approximately proportional hiring. (A similar pattern for N.F.L. head coaches is reported in Harper's, Sept. 1996, p. 15, though the sample size is not given).

The win/lose performance index in sports is, of course, much more objective than the performance indices used in most fields of endeavour, but in many cases performance can be assessed along a number of meaningful dimensions with sufficient accuracy to allow us to go beyond the simplistic proportionality measure for assessing whether discrimination in hiring has occurred.

Author Note: Paul Muter is a member of the Society for Academic Freedom and Scholarship (SAFS); John J. Furedy is president of SAFS.