Affirmative Action Policies in Admissions Decisions
During the last year, I have been working to finish a paper that analyzes affirmative action policy from a Brunswikian perspective. The paper is jointly authored with Tom Stewart and Radhika Nath, and I presented the work at the July Brunswik meeting for those of you who were there.
The research on affirmative action policy follows from some closely related work on psychiatric emergency room decision making, breast cancer screening policy, and global warming.
Only true Brunswikians are likely to understand the link between such disparate topics! The link is that for each of these problems we are applying the key ideas that Ken Hammond explored in his 1996 book-Human Judgment and Social Policy: Irreducible Uncertainty, Inevitable Error, Unavoidable Injustice.
We are studying affirmative action in the context of admissions to highly selective undergraduate institutions. Although it would seem that little new could be left to be said on this topic, which has received extensive attention in recent years, the 60-year old Taylor-Russell approach sheds new light on the subject, just as Ken told us it would.
Admissions decisions can be divided into four exhaustive and mutually exclusive categories: false positives (admit unqualified applicants), false negatives (reject qualified applicants), true positives (admit qualified applicants) and true negatives (reject unqualified applicants).
To understand the full implications of different affirmative action policies, four factors must be considered simultaneously:
o selection rate, or percentage admitted;
o base rate, or percentage of those applying who could do the work if admitted;
o predictive accuracy, or degree of correspondence between predictions of performance and actual performance; and
o the costs associated with false positive and false negative errors, as well as the
o benefits associated with true positive and true negative diagnoses.
The paper reports a series of nine analyses, which were conducted under varying assumptions concerning the presence or absence of an affirmative action plan, whether observed racial differences in college admissions test scores reflect systematic bias, and whether there are racial differences in the level of random error associated with predictors of future performance.
To give a flavor of the results, the analysis illustrated the following: Assume there is comparatively greater random error in predictors of future performance for African-American applicants (a proposition for which there is reasonably good circumstantial evidence). Then under a program of affirmative action, proportionately more false positives (matriculated students who do not succeed) would be found among minority group students than among majority group students. In other words, proportionately more admitted students would fail to graduate.
Further, the number of false negatives (rejected applicants who could have succeeded) would also be proportionately higher among minority students. In other words, proportionately more deserving applicants would be rejected.
Moreover, if we reduce the number of admitted minority applicants in order to address their higher failure rates, we simply aggravate the false negative problem, rejecting an even higher percentage of deserving minority applicants.
In sum, Brunswikian-inspired judgment analysis sheds new light on the notion of fairness in affirmative action, demonstrating that, in this context, what is "fair" for majority group applicants is unlikely to be "fair" for minority group applicants, and vice versa.