One issue that has not been disputed to date in the litigation on race-based affirmative action has been whether schools that were sued to stop the preferential admission of minorities in fact engaged in affirmative action. In every case the schools sued had policies that espoused affirmative action, and they admitted they were following these policies. But suppose they had not admitted to affirmative action admissions. Could the plaintiffs have proved that the defendant schools were engaging in race-based affirmative action?
The question is not purely academic, or at least it might not remain so. A number of states, my home state of Michigan among them, now preclude race-based affirmative action, and the word is that those opposed to affirmative action plan ballot proposals to ban affirmative action in numbers of other states. Judging from recent referenda, if such proposals make the ballot, they are likely to pass. If they do, or where they have, should it appear that there are more blacks, Hispanics, Native Americans, or members of other minority groups on campus than race-blind admissions would bring, law suits might well ensue. Unlike earlier litigation, however, plaintiffs would be attempting to prove facts – the existence of forbidden affirmative action – rather than seeking to argue the constitutionality of an admittedly ongoing process. How hard will it be to do this?
To aid in thinking about this question, consider Grutter v. Bollinger. In Grutter, the plaintiff, through its statistical expert, presented evidence designed to show that a huge advantage was given to black applicants in the Law School’s admissions process. To the casual observer, the plaintiff had a slam dunk case, and the advantage blacks enjoyed from affirmative action was huge. There were cells in a grid formed by the intersection of an LSAT score range (say 160 – 165) and an undergraduate grade point average range (say 3.25 – 3.5) where, say, 15% of white applicants were admitted and 80% of black applicants were admitted. (I am making these numbers up, for I don’t have the data in front of me or remember admission percentages by cells, but the general point holds.) It looked from these data as if blacks not only benefited from affirmative action, but, as the plaintiff claimed, they benefited greatly. Moreover the average LSAT score and the average GPA of Michigan’s white admits were substantially greater than the black student averages, and differences were highly significant in the statistical sense of the term. (Again I don’t have the data in front of me, but I am sure the significance level of the differences between the mean scores of admitted whites and blacks on both the LSAT and UGPA were significant well beyond .001.) Yet these numbers alone, though they may appear overwhelmingly probative, do not necessarily demonstrate that the University of Michigan Law School gave race great weight in its admissions process.
First, there is the semantic question of what it means to give a factor great weight. Suppose there are two applicants competing for a single slot, who are identical in every way except one is a legacy – his mother was a law school graduate. If the legacy student is admitted for this reason, his legacy status has been decisive, but does this mean it has been given great weight. Arguably it need only to have had the weight of a feather to tip the balance. So a study that might show that when two students with more or less similar credentials are competing for admission, the legacy student always wins shows that legacy status can be hugely important in the admissions decision, but it does not necessarily mean it has been given great weight.
Now suppose that legacy students are admitted over non-legacy students even when non-legacy students look stronger on certain dimensions, LSAT scores for example. Can we now say that legacy status has been given great weight? Not necessarily, for this depends on the weight that LSAT scores receive in the admissions process. If they are barely considered, legacy status might not count for much, yet still often prove decisive, for it doesn’t take much to outweigh a minor factor.
Suppose then that we could show that other things being equal schools always chose the person with the higher LSAT score except when a student was a legacy. Would this mean that legacy status was given great weight? Again, not necessarily. It could be that LSAT score was a tie breaker when other things were equal, meaning that LSAT scores were often decisive (if students in the admissions pool were similar) but were not given great weight. Thus the fact that legacy status trumped LSAT scores might mean that legacy status counted as a feather and a half while LSAT scores were a mere feather. A credential has great weight only if the credentials it outweighs are themsleves weighty.
Finally suppose we could show through multiple regression that moving from non-legacy status to legacy status had the same effect on one’s chance of admissions as scoring a full standard deviation higher on the LSAT. Could we now say that legacy status was given great weight. Again we couldn’t on these statistics alone because to say this we would have to know how much a standard deviation improvement meant in the admissions process. The point is we are always importing general knowledge about what schools value and how they treat different information when we use statistical equivalence to assess the weight given to a factor like legacy status (or race.) It is because we sense that a standard deviation difference in LSAT scores is highly influential in the admissions process, the difference in some schools between who will be placed in the pool for closer examination and who will be rejected out of hand, that we argue that when factors like legacy status or race seem as important as large LSAT score differences they are being given great weight.
My common sense tells me that usually this is true, but imagine a school many of whose applicants have LSAT scores 3 or 4 standard deviations above the mean (Yale, maybe). A school like this would probably be smart if it paid little attention to LSAT score differences in choosing which applicants to admit – most of its applicants would have strong predicted aptitudes and could easily handle the challenges professors threw at them. Thus a legacy or race effect equivalent to a standard deviation difference in LSAT scores might mean little in such a school even if these factors were decisive in the sense that a non-legacy student or a white would have little chance of admission with an LSAT score that was, let us say, “only” two standard deviations above the mean while a minority student would almost always be admitted. If race-based affirmative action were outlawed, these niceties would, of course, mean little, for any boost based solely on race would be illegal. But in the policy debate which often raises questions about the degree of advantage minorities enjoy, such technicalities are important and are usually overlooked because it is assumed that large test score differences always count hugely in a school’s admissions process, no matter what its applicant pool looks like.
A second issue that makes it conceptually more difficult than it might appear to conclude that a factor like race is hugely important in an admissions process stems from the fact that the influence of race on decisions is not the same for every applicant, though talk about affirmative action suggests that minorities, whatever their other qualifications, always enjoy – by virtue of their race – a tremendous advantage. There were, for example, many cells in Michigan Law School’s admission grid where the chance of admission was vanishingly small, whether one was white or minority, and other cells where admission was highly probable regardless of race, so race mattered little for these applicants. Thus one must, to be precise, speak somewhat differently than plaintiffs are wont to speak; namely the claim must be that over a limited range race matters and/or that for students who have certain credentials race matters greatly in determining whether they are admitted.
But even this is not unequivocally proven by the statistical data. Consider the following thought experiment. Suppose the University of Michigan Law School selected students without any attention to race, but on the basis of some characteristic, let us say demonstrated leadership, that was orthogonal to race. If this were done, there would still be a vast difference between the mean LSAT scores and UGPAs of Michigan’s black and white admittees because there is a huge difference in the average scores of black and white applicants in the applicant pool. Hence differences in average LSAT scores or UGPAs would not necessarily tell us anything about whether race played a role, much less a large role, in the admissions decision. (Note there are large differences between the admissions test scores and grades of all of Michigan’s affirmative action eligible minorities and whites. I shall, however, use blacks as the example throughout because much of the empirical research on affirmative action focuses only on blacks and whites and popular debates on affirmative action often focus on a black-white dichotomy.)
It might be thought that if we looked within cells, the inference problem would disappear. But immediately we would confront the issue of how to define cells. If they contained too few people of either race, statistically significant relationships would seldom appear. For example, if the only 2 blacks in a cell and 10 of 20 whites in the same cell were admitted, 100% of the blacks would have gained admission and only 50% of the whites would have, but we could not safely conclude that race played a role in the admissions process. On the other hand, we have already seen that it would not be surprising to have a highly significant disparity in a process where race played no role if we had only one cell; that is we compared the scores of all white admittees to all black ones. Thus given Michigan’s applicant pool, the larger the range covered by defined cells, the less we could say about the size of the advantage that race played in admissions despite large disparities in mean admissions probabilities within those cells. We could not even be certain that race played any role in deciding who within a cell was admitted. A corollary is that cell sizes might be manipulated to show race effects or non-effects.
Of course, there would be some cells where the numbers might be at the right level to reveal statistically differences in admissions probabilities by race for those falling in the cell without being so large that disparities of the size seen simply reflected disparities in the average credentials of those black and white applicants who fell into the cell. But the statistical tests would be invalid if the cells were specifically drawn to contain people in a particular range knowing about their distribution. The grid, in other words, would have to be drawn a priori without knowing the characteristics of those within them. Non-parametric techniques might, however, help with these statistical problems. If “too many” cells had higher proportions of blacks than whites accepted, this would be some evidence that race mattered even if the nonparametric nature of the test made it be difficult to say how much weight race was given.
To further show the inferential complexities that can exist here, let us change the hypothetical in a not unreasonable way. Suppose rather than being orthogonal to LSAT scores and UGPAs, the characteristic on which we were admitting was positively correlated with these scores within racial groups. One might, for example, expect leaders to be more able intellectually than followers, and since people’s opportunities for leadership are likely to more often be within racial groups than across them, black students with demonstrated leadership might disproportionately congregate in the grid cells occupied by, say, the top 15% of black applicants while white students with demonstrated leadership might be concentrated in the cells occupied by the top 15% of white applicants. Given the distribution of white and black test scores and undergraduate grade point averages, there would be little overlap. Thus even in a pure race blind admission system, one would expect to find cells where most black applicants were admitted (because they were leaders) and few whites were admitted (because they were followers.) It would also appear that whites had to have far higher scores than blacks to be admitted when, in reality, they simply had to have demonstrated the same degree of leadership.
To add a further ingredient, suppose that LSAT scores and UGPAs were not completely irrelevant to admissions decisions but were used both to exclude students whom a law school thought had too great a risk of failure and as a tie breaker to decide between students of equally evident leadership. This would further strengthen the observed pattern. Students with very low test and grade credentials, predominantly black, would not be admitted to the school even if some were every bit as strong leaders as the whites and blacks who were admitted, and there would be a further tendency to admit students who had stronger test and grade credentials within races and across them because of the tie breaking rule. The resulting distribution of admittees would make it appear that LSAT scores, UGPA and race all figured importantly in determining who was admitted because white admittees would be clustered at the top of the white applicant range, black admittees would be clustered at the top of the black applicant range, and there would be little overlap between them due to the distribution of grades and test scores in the white and black applicant populations. This is exactly the pattern one sees at Michigan and at many other law schools, and from this pattern the plaintiffs in Grutter argued that race had to be hugely important in determining who was admitted to Michigan. The plaintiffs would have been right had they merely suggested that this pattern is a possible result of admissions schemes that look at little other than race, test scores and grades in admitting students, but as our thought experiment reveals, the pattern could also result from an admissions scheme that gave no weight to race and used LSAT scores and UGPA only to exclude applicants regarded as too risky or to break ties when applicants were regarded as very close on other important factors. The pattern is also, of course, compatible with many admissions models in between the race is almost everything and the race counts for nothing model, including models that, as the Michigan Law School claimed in Grutter, treat race as only one among a number of factors that determine admissions. A phenotype formed of information on LSAT scores, undergraduate grades, race and the admissions decision is, in other words, consistent with many different genotypes.
We see from these examples that it might be difficult if not impossible to prove from statistical data that reported only race, admissions decisions, undergraduate grades and admissions test scores that a law school was not engaged in race blind admissions or, if it considered race, that minority students were hugely advantaged by the process. The easy acceptance of this claim in litigation to date has come about because most people have a model of the admissions process in which LSAT scores and UGPAs are most of the story, and because litigation to date has involved constitutionally based attacks on affirmative action at schools that admitted to considering race in admissions. For the latter reason defendant schools have not vigorously contested the claim that race greatly matters, though it is notable that in the Grutter case the Michigan Law School Dean of Admissions testified that she looked far beyond test scores and race in deciding whom to admit. Indeed, I heard a report from the trial that when the Michigan admissions dean was trying to describe to the court what she looked for in reading a file, she became so enmeshed in the details of the file she was using as an example that she almost lost sight of the fact that she was testifying and not puzzling over an applicant. Characteristics other than race and scores can be that important and engrossing.
Turning to the post affirmative action world as it exists in Michigan, California and other states, I wish to make a few further points. The first is that for those who simply look at the race of admittees and their test scores and begin with the idea that the latter largely determine who is admitted, it is very likely to appear that schools that engage in color blind admissions are “cheating,” for their admitted classes will show differences, perhaps large differences, in the test scores of whites and minorities because of score differences in the applicant population. Moreover, the more attention a school pays to factors other than test scores in deciding whom to admit, the larger these differences are likely to be. They are, however, not likely to be as big as differences found in schools that are able to practice affirmative action even if identical admissions models exist. The reason is that when a school is known to have an affirmative action program, self selection is likely to broaden test score discrepancies between the white and black applicant pools. Whites with scores that are below average for a school are less likely to apply because they feel they have little chance of admission while blacks, because they know race counts somewhat, will apply with scores that would have discouraged them had they not known their race might help. For this reason, it is particularly difficult to draw from score discrepancies in schools actively practicing affirmative action conclusions about how much race counts.
In a post affirmative action world, score discrepancies by race due to self selection should not exist, but we cannot say that they won’t. People may still evaluate their credentials differently by race and make application decisions accordingly. For example, a black student with an LSAT score a standard deviation and a half above the mean may have achieved the highest LSAT score in her circle of friends and be encouraged by the sense she stands out to apply to a highly selective law school while a white scoring similarly may be motivated to aim lower. Nevertheless, one would expect self selection by applicants to contribute less to differences in average applicant credentials by race in a post affirmative action world than in one where affirmative action is allowed.
Because of the proof difficulties I have pointed to, a crucial issue in litigation against schools accused of cheating may be who is responsible for keeping records of the factors that admissions officers weigh in reaching decisions, if anyone is responsible at all. If as is now the case, there is no obligation on a decision maker to record or assemble all factors that enter into a decision, it will be difficult for plaintiffs without substantial discovery of raw records and expensive coding to show that race rather than some other factor has allowed one person or group of people rather than another to be admitted. Moreover some factors thought legitimate, like the impression an applicant makes in an interview, may not be recorded at all or, if recorded appear only as a simple impression (“outstanding interview”) that will allow for no validation. Those who fought to prohibit affirmative action are likely to regard such circumstances as creating loopholes big enough to drive a truck through.
Thus in lawsuits arising from complaints of cheating, a crucial question will be what information is needed to make out a prima facie case. If it is simply the difference between the proportion of black and white applicants admitted, challenges will be very difficult if Michigan’s experience is a guide, since least historically at Michigan black and white applicants are admitted more or less in proportion to their presence in the applicant pool. If differences between the proportion of blacks and whites admitted controlling for LSAT scores and UGPAs can make out a prima facie case, then it should be easy for plaintiffs to establish such cases where cheating is occurring. But it will also be easy for them to establish such cases where admissions are in fact race blind. In either case, the justification for making such controlled discrepancies sufficient to shift burdens must rest on a court’s assumption that a school is following, or on the judicial imposition of, an admissions model which highly privileges the role of test scores and prior grades in academic admissions processes.
In the perhaps analogous area of employment discrimination and hiring, it makes sense to assume that all firms are looking for more or less the same characteristics and that these include education, experience, ability to get along with others as demonstrated in prior jobs and other job-related skills. But institutions of higher education have many goals, including even efforts to remediate prior educational deficiencies. It is questionable whether there is a principled basis for imposing one model, which emphasizes test and grade credentials, on all schools, and even if one could find a justifying principle (for those who believe affirmative action is discrimination, erring on the side of overinclusiveness to promote non-discrimination might be such a principle), it is not, I would argue, educationally wise to enforce this one size fits all principle on the world of higher education.
The reason why the standard for establishing a burden shifting prima facie case in litigation is so important is that it is likely to influence what records are kept and how. If a case for going forward can be established based on aggregate racial discrepancies in admissions conditioned on test scores, schools that wish to consider factors other than test scores will have to keep records of what they considered in each case lest they be sued, and the information must be readily accessible. If the burden is not shifted by showing controlled group discrepancies, it will be up to plaintiffs to find the information needed to show that factors other than race don’t explain apparent differences. Schools in this world will have less of an obligation to keep and organize information on the factors that influence their decisions and will have more room for subjective judgments in their admissions decision processes. A danger if burdens are placed on schools is that bans on affirmative action will not lead to a world of non-discrimination as their backers argue, but will rather reintroduce discrimination against blacks and other minorities. If the only numerical portrait that ensures a school safety from litigation or popular condemnation is one which shows white and black admittees to be have very similar credentials then it is likely that schools will either be forced to further elevate the role that test scores and grades play in their admissions process or subtly discriminate against blacks in order to create a litigation-proof appearance of non-discrimination. Indeed, admissions officers may come to see in credential similarity a test of their own legal compliance, and this could stimulate subtle, perhaps unconscious, discrimination to erase the credential differences expected naturally in a race blind admissions world.
Finally, there is the issue of who if anyone has standing to claim a school has failed to admit him and what he would have to prove to show discrimination. Here there is an interesting precedent in the death penalty area. In McClesky v. Kemp, the Supreme Court was presented with a grid much like the grids that have been developed in some of the affirmative action litigation. David Baldus and his colleagues who did the statistical analysis offered the courts found that there were numerous homicides whose characteristics were such that even if murder was charged, almost no one received the death penalty regardless of whether the victim was white or black, and there were a much smaller numbers of murders that were so heinous that virtually every defendant involved received the death penalty no matter what the race of the victim. These cases are analogous to the cases of those applicants whose scores and grades are so low or so high that admission almost never happens or is virtually guaranteed. But there were a number of murders with middle range (in terms of heinousness) characteristics, and in these cases killers of whites were far more likely to receive the death penalty than killers of blacks. The Court held, however, that these data were of no avail to McClesky who had killed a white in circumstances where a death sentence had he been black was unlikely. To prevail, the Court held, McClesky had to present not numbers, but specific evidence that in his case it was the race of the victim that determined the death sentence. That showing was for all practical purposes impossible to make. If the Court took a similar attitude toward the relevance of statistical evidence in affirmative action cases, then no matter what schools did, individual plaintiffs would almost always be unable to make out cases. I do not expect this to happen. Affirmative action and its jurisprudence is, unfortunately, almost as politically determined as the death penalty.