Narrow Tailoring

by Ian Ayres*

Table of Contents

Introduction

I. When Does Narrow Tailoring Mandate Race-Neutral Means

II. Quotas vs. Credits

A. Comparing Quotas and With Simple Credits

B. Declining Credit Schedules

C. Quasi-quotas

III. Tailoring the Scope of Racial Preferences In The Real World

Conclusion

Appendix

B. Declining Credit Schedules

C. Quasi-quotas

Introduction

Since the Supreme Court announced in Adarand v. Pena that federal affirmative action programs will be subject to "strict scrutiny," a debate has reemerged over what constitutes a compelling government interest for classifications that favor traditionally disadvantaged races. This essay, however, does not address this interesting and vital issue. It instead focuses on the second prong of strict scrutiny analysis: the requirement that racial classifications "must be narrowly tailored" to further the compelling government interest at stake.

While many people believe that the government never has a compelling interest to use racial classifications, for the purposes of this essay I ask the reader to accept the holding of both Adarand and City of Richmond v. Croson that remedying past discrimination can constitute such a compelling interest. Specifically, imagine that the Department of Transportation could show that past and present disparate treatment by private and public actors has caused the market share of qualified minority guardrail contractors to be substantially below their non-minority counterparts. Assuming that a federal court found sufficient evidence to demonstrate a compelling government interest to increase minority participation, this essay considers what type of affirmative action programs would satisfy the narrow tailoring requirement.

Two criticisms of the Supreme Court's narrow tailoring approach in Adarand and Croson form the central thesis:

1. The Court's preference for "the use of race-neutral means to increase minority participation" is inconsistent with narrow tailoring and may not be a less restrictive alternative than explicit racial classifications. Extending affirmative action subsidies to non-victim whites produces less-tailored, overinclusive programs. And because both race-neutral and explicitly racial means share the same race-conscious motivation of remedying past discrimination, it is not clear that race-neutral means represent a less restrictive alternative.

2. The Court's (and others') antipathy for quotas is overstated. A quota remedy might be more narrowly tailored to achieve the government's remedial interest than many types of simple subsidies. While quotas may be poorly tailored because they mandate an inflexible level of minority participation, bidding credits (and other preferences) may be poorly tailored because they induce too much uncertainty and volatility in minority participation. More narrowly tailored programs will exhibit a "sliding scale" of racial preferences in which the size of the preference will vary inversely with the degree of successful minority participation in the program. Under a narrowly tailored program, the farther minority participation falls below what it would be in the absence of discrimination, the larger the racial preference government might legitimately confer.

Together these two criticisms suggest that when the government has a compelling interest to remedy past discrimination, the narrow tailoring principle should not impede racial classifications that tailor the size of the racial preference to the remedial need.

Sliding-scale preferences of this kind may come close to setting aside a minimum quota of contracts for minority bidders, but this essay will show that such quasi-quotas (for fractions of the legitimate remedial goal) are consistent with narrow tailoring when dramatic shortfalls in minority participation would undermine the government's remedial effort. For example, in an industry where the government has a legitimate interest in increasing minority participation to 30% (what it would be in the absence of discrimination), the government might independently find that allowing the minority participation to fall below 5% would affect the long term viability of all minority business. Under such circumstances, the government might be justified under the narrow tailoring principle in granting substantial bidding credits for 5% of the government contracts -- which effectively guarantee that at least 5% will go to minorities.

Quasi-quotas might satisfy the narrow tailoring requirement for four reasons:

(1) Unlike an absolute quota, a quasi-quota by assuring a minimum level of minority cost-effectiveness would at least to a small degree tailor the level of participation to the strength of minority applicants.

(2) A quasi-quota would only be appropriate if the government found that shortfalls in participation below some minimum level would seriously undermine the government's remedial effort. In such circumstances, a quasi-quota would be better tailored because it would cause decisionmakers to better internalize the true social costs of dramatic shortfalls in minority participation.

(3) Because a quasi-quota would only set aside a fraction of the government's legitimate remedial goal, it would impose a smaller burden on the interests of non-beneficiaries.

(4) Granting minority enterprises effective guarantees of minimum participation can increase the quality of minority participants -- so as to reduce the disparity between minority and non-minority recipients.

In sum, the Court's disinclination to authorize remedial programs with explicit racial classifications and particularly those programs that display quota-like attributes cannot be adequately justified by resort to the narrow tailoring principle.

The first two sections of the essay explore dichotomous choices that policymakers face in designing an affirmative action program: whether the affirmative action preference should be explicitly racial; and (if explicitly racial) whether the preference should take the form of a quota or a credit. In contrast to the top-down reasoning of these sections, the last section then briefly surveys current affirmative action programs and current evidence of past discrimination to consider what types of programs should satisfy the narrow tailoring requirement.

I. When Does Narrow Tailoring Mandate Race-Neutral Means

The idea that a remedy needs to be tailored to further the government's legitimate interest is unexceptional -- and is captured by the idea that remedial classifications should not be too over- or underinclusive. In considering the validity of affirmative action remedies for past discrimination, the Supreme Court's primary concern has been one of overinclusion: that is, giving affirmative action preferences to people who were not affected by past discrimination. For example, Justice O'Connor in Croson challenged the inclusion of non-black minorities in Richmond's affirmative action plan:

There is absolutely no evidence of past discrimination against Spanishspeaking, Oriental, Indian, Eskimo or Aleut persons in any aspect of the Richmond construction industry. . . . It may well be that Richmond has never had an Aleut or Eskimo citizen. . . . If a 30% set-aside as "narrowly tailored" to compensate black contractors for past discrimination, one may legitimately ask why they are forced to share this "remedial relief with an Aleut citizen who moves to Richmond tomorrow? The gross overinclusiveness of Richmond's racial preference strongly impugns the city's claim of remedial motivation.

Croson's discussion of Aleuts thus represents a straightforward application of the narrow tailoring principal. The Richmond preferences were not narrowly tailored to remedy past discrimination, because some of the beneficiaries were not the victims of past discrimination.

After this plausible beginning, however, the Croson opinion uses narrow tailoring as a principle militating in favor of "the use of race-neutral means to increase minority participation." Just two paragraphs after the quoted discussion of Aleuts, Justice O'Connor argues that the racial preferences were not narrowly tailored because the Richmond City Council did not consider a variety of race-neutral preferences -- including small business preferences, reduced bond requirements and "training and financial aid for disadvantaged entrepreneurs of all races." In essence, Croson requires policymakers to find that race-neutral means could not achieve the government's compelling interest prior to implementing a racial classification.

The problem with this requirement is that the "overinclusion" version of narrow tailoring, if anything, points away from race-neutral subsidies. If preferring the minuscule number of Aleuts in Richmond is "grossly overinclusive," then extending preferences to a much larger class of whites a fortiori would fail the narrow tailoring requirement. Whites, like Aleuts, were not the victims of race discrimination. Narrowly tailoring the beneficiary class for remedial subsidies so that it will not be overinclusive almost necessitates non-race neutrality.

Justice O'Connor's inclination toward race-neutral means, however, might be squared with another principle. In Regents of the University of California v. Bakke, Justice Brennan noted that:

[A] government practice or statute which restricts "fundamental rights" or which contains "suspect classifications" is to be subjected to "strict scrutiny" and can be justified only if it furthers a compelling government purpose and, even then, only if no less restrictive alternative is available.

When applied to racial classifications, this principle seems to require the government to achieve its compelling remedial interest in the way that least restricts or burdens the fundamental rights of the program's non-beneficiaries. Again, Powell in Bakke noted:

We have never approved a classification that aids person perceived as members of relatively victimized groups at the expense of other innocent individuals in the absence of judicial, legislative, or administrative findings of constitutional or statutory violations. After such findings have been made, the governmental interest in preferring members of the injured groups at the expense of others is substantial, since the legal rights of the victims must be vindicated. . . . [T]he remedial action usually remains subject to continuing oversight to assure that it work the least harm possible to other innocent persons competing for the benefit.

In Wygant v. Jackson Board of Education, the Supreme Court in a footnote described the "less restrictive" alternative requirement as being a subpart of narrow tailoring:

The term "narrowly tailored," so frequently used in our cases, has acquired a secondary meaning. More specifically, as commentators have indicated, the term may be used to require consideration of whether lawful alternative and less restrictive means could have been used.

Whether viewed as an independent requirement or as a subpart of narrow tailoring, this "less restrictive alternative" principle conflicts with the principle that affirmative action preferences be narrowly tailored to victims of discrimination: Narrow tailoring of subsidies to the victim class seems to require racial classifications which exclude non-victim whites from the remedial subsidy, while the Supreme Court -- via something like the "least restrictive alternative" principle -- disfavors such classification.

To resolve this doctrinal conflict, it is useful to articulate what larger purposes the narrow tailoring and least restrictive alternative principles are meant to serve. Although stated in a variety of somewhat amorphous ways, several commentators have suggested that the purpose of strict scrutiny is to limit the racial divisiveness caused by remedial efforts. Aspects of both the narrow tailoring and the less restrictive alternative requirements can be squared with this meta-purpose: Granting government benefits to minorities that were not the victims of past discrimination in a particular jurisdiction smacks of the illegitimate interest group politics that strict scrutiny was meant to "smoke out." And superficially at least race-neutral classifications seem to be less likely to provoke the kinds of racial enmity that would in part undermine the remedial purpose of the legislative action itself. This possibility that explicit racial classifications would be more likely to fan the flames of racial divisiveness might provide a rationale for the least restrictive alternative requirement to take priority over the requirement that the beneficiary class not be grossly overinclusive. The Constitution might tolerate or even require a more poorly tailored class of beneficiaries if using a race-neutral proxy for being a victim of discrimination were less likely to inflame further prejudice and divisiveness.

This constitutional preference for race-neutral means is strongest when considering simple mandates for race-blind decisionmaking. If merely enjoining disparate treatment (by government and private actors) could adequately remedy past and current discrimination, this type of injunction would clearly be less restrictive -- especially judged by the meta-purpose of minimizing racial division -- than explicit racial classifications.

But in Croson, Justice O'Connor implicitly acknowledges that remedying past and current discrimination may require more than merely prohibiting prospective disparate treatment. When Croson declares a preference for race-neutral means, it is contemplating affirmative subsidies -- including training, bond and financial subsidies for small or disadvantaged entrepreneurs -- that would not obtain absent the desire to remedy past (and possibly ongoing) private disparate treatment. Unlike the simpler mandates merely to end disparate treatment, Croson takes on the harder task of comparing two different kinds of "affirmative" remedies neither of which would have obtain in the absence of the government's race conscious efforts to remedy past discrimination.

The point here is not merely that the preference for race-neutral subsidies contradicts the over-inclusiveness prohibition, but to take head on the idea that race neutral subsidies are less restrictive alternatives -- or in some sense less of a burden on constitutional norms concerning racial divisiveness.

Properly understood, it is not clear that the Court's idea of "race-neutral means" is not less restrictive than affirmative action programs with explicit racial classifications. The central problem with the Court's conception is that the race-neutral means are still racially motivated. The key phrase from Croson, which is quoted again in Adarand, is the admonition that policy makers must consider "the use of race-neutral means to increase minority business participation." The court is still counseling legislatures to engage in race-conscious decisionmaking -- that is, to enact certain subsidies because of the race of the beneficiaries. And of course the court cannot avoid this causal connection: any race-neutral program attempting to remedy past racial discrimination would necessarily have a motive to benefit the victimized race.

Thus understood, the Court's preference for "facially neutral, but racially motivated" programs cannot sidestep strict scrutiny analysis. Affirmative subsidies for small or economically disadvantaged entrepreneurs would easily pass constitutional muster under the lenient rational relation test -- if the legislature granted these subsidies merely to redistribute wealth to the poor or to spur economic growth. Under these circumstances, there would be no tension between prohibiting overinclusion and a preference for race-neutral subsidies, because the narrow tailoring requirements would apply.

But the type of subsidy preferred by Justice O'Connor in Croson and Adarand is intended to be a "means to increase minority business participation." While there are still nice questions about what degree of legislative race-consciousness rises to the level of "intentional discrimination" for purposes of equal protection analysis, the hypothesized genesis of O'Connor's race-neutral subsidies is sufficient to trigger strict scrutiny. For example, last term in Miller v. Johnson, the Supreme Court, in assessing whether a facially race-neutral districting plan violated the Equal Protection Clause, formulated the following "predominant factor" standard:

Redistricting legislatures will, for example, almost always be aware of racial demographics; but it does not follow that race predominates in the redistricting process. See Personnel Administrator of Mass. v. Feeney, 442 U.S. 256, 279, 60 L. Ed. 2d 870, 99 S. Ct. 2282 (1979) ("'Discriminatory' purpose' . . . implies more than intent as volition or intent as awareness of consequences. It implies that the decisionmaker . . . selected or reaffirmed a particular course of action at least in part 'because of,' not merely 'in spite of,' its adverse effects") (footnotes and citation omitted). . . . The plaintiff's burden is to show, either through circumstantial evidence . . . or more direct evidence going to legislative purpose, that race was the predominant factor motivating the legislature's decision . . . .

The Adarand/Croson preference for "race-neutral means to increase minority participation" clearly contemplates legislative action "because of" its effects on minority entrepreneurs. And while it is difficult to clearly specify the minimum necessary requirement for establishing what constitutes a "predominant" motivating factor, it should not be difficult to conclude that subsidies fashioned to increase minority participation had race as a predominant motivating factor. If we intend to subject racially motivated legislation to strict scrutiny, at the end of the day we must still answer which racially motivated means is the least restrictive alternative.

In conducting this analysis, the Supreme Court cannot use legislative motive to distinguish between "explicitly racial" and "race neutral" means -- because by hypothesis they both share the same motive of remedying past discrimination by subsidizing increased minority participation. The choice between programs that on their face use race classifications and those that are race neutral is not a question of whether the legislature intended to benefit people because of their race. After Washington v. Davis, it would seem that citizens' core equal protection interest is in being free from the effects of racially motivated legislative action. But both the race-neutral and explicitly racial classifications considered in Croson/Adarand equally burden this right, because both share the same motive.

If one cannot distinguish on the basis of motive or degree of the legislature's race-consciousness, then in what other sense might race-neutral means be less of a burden on constitutional interests? Here I would like to consider two senses in which "racially motivated, but facially neutral" might be "less restrictive" -- especially in the sense of inducing less racial divisiveness.

A. The Argument From Opaqueness

First, race-neutral means might induce less racial divisiveness because the citizenry is less likely to become aware of the legislature's underlying motivation and race-consciousness. Even if empirically true (or maybe I should say, especially if empirically true), this justification seems an extremely weak if not embarrassing basis for constitutional decisionmaking. To begin rather grandly, preferring race-neutral subsidies because its racial motivation is less visible not only violates Kant's publicity principle that "[a]ll actions relating to the right of other human beings are wrong if their maxim is incompatible with publicity." While the Supreme Court at times has countenanced legislative or executive secrecy to further compelling government interests, this would be the only time that the Constitution preferred or even required legislative opaqueness.

Indeed, a particularly cynical extension of this argument would infer that the Supreme Court in Croson/Adarand was counseling legislatures not to explicitly discuss remedying past discrimination. Under this cynical reading, the Court would in effect be telling legislatures: if you stop talking about remedying past discrimination, we will allow you to pass race-neutral subsidies which disproportionately favor minorities.

Whether stated cynically or not, the argument from opaqueness flatly contradicts Justice O'Connor's own classic statement of the process justification for strict scrutiny:

Absent searching judicial inquiry into the justification for such race-based measures, there is simply no way of determining what classifications are "benign" or "remedial" and what classifications are in fact motivated by illegitimate notions of racial inferiority or simple racial politics. Indeed, the purpose of strict scrutiny is to "smoke out" illegitimate uses of race by assuring that the legislative body is pursuing a goal important enough to warrant use of a highly suspect tool.

Given our freedom of the press attempts to shield the citizenry from legislators' true degree of race consciousness are bound to also shield the Court as well. Whether intended or not, Adarand and Croson are bound to induce legislatures to conceal their race conscious decision making behind race-neutral classifications. The net effect of these cases, perversely may be to subject racially motivated legislation to less scrutiny than under the more lenient Metro-Broadcasting standard -- because the Court will have more difficulty determining for which statutes race played a predominant motivating factor. It would be less burdensome to the equal protection rights of non-beneficiaries to force racially motivated policies into the open, where they can be more readily scrutinized by the courts and by the democratic process. Racially motivated legislation is inherently suspect, but unacknowledged racial motivation by legislatures is all the more worrisome.

B. Avoiding Individual Racial Determinations

The second potential justification deserves more serious consideration. It is possible that race-neutral means induce less racial divisiveness (and are therefore "less restrictive alternatives") because they do not involve the government in individual determinations of race. Justice Stevens first gave voice to this concern in his dissent in Fullilove v. Klutznick:

[T]he very attempt to define with precision a beneficiary's qualifying racial characteristics is repugnant to our constitutional ideals. . . . If the National Government is to make a serious effort to define racial classes by criteria that can be administered objectively, it must study precedents such as the First Regulation to the Reich's Citizenship Law of November 14, 1935.

Justice Steven's concern was later echoed and expanded by Justice Kennedy in Metro Broadcasting, Inc. v. FCC:

The Court fails to address the difficulties, both practical and constitutional, with the task of defining members of racial groups that its decision will require. The Commission, for example, has found it necessary to trace an applicant's family history to 1492 to conclude that the applicant was "Hispanic" for purposes of a minority tax certificate policy.

Race-neutral means to increase minority participation still force the government to determine the racial characteristics of groups, but do not expose society to the intrusive painful process of defining the race of individual citizens: determining that a particular neighborhood is predominantly Latino or predominantly Anglo may be much easier than determining whether any particular person is Anglo or Latino.

Determinations of race that turn on how individuals label themselves have led in the past to at least a few instances where individuals seemed to have claimed minority status merely to qualify for an affirmative action benefit. While determining the racial characteristics of larger classes might also turn in part on self-labeling (such as contained in Census data), groups would have much greater difficulty coordinating false claims of minority status in order to qualify for benefits.

Our visceral reaction against laws that define race undergirds a coherent justification for the Supreme Court's preference for race-neutral subsidies. Far from requiring a narrowly tailored class of beneficiaries, the Constitution might require using imprecise (poorly tailored) proxies for racial discrimination. Even if the general populace knows the remedial motivation for the subsidies, race neutral means might avoid what Justices Stevens and Scalia consider to be the abhorrent process of determining the race of individuals.

While this provides a coherent justification, it is not ultimately persuasive. Racial classification need not ultimately turn on the arcane measures of consanguinity used in the Nuremberg laws or in South Africa's Apartheid or our own Jim Crow regimes. Instead, section 8d of the Small Business Act wisely defines "social disadvantage" as being ""subjected to racial or ethnic prejudice or cultural bias because of their identity as a member of a group without regard to individual qualities." Defining an individual's race from a viewer's perspective at once tailors the definition to the remedial goal of helping those that have been subject to disparate treatment without denying that race is "socially constructed." As Henry Louis Gates has observed: "To declare that race is a trope, is to deny its palpable force in the life of every African American who tries to function every day in a still very racist America." Defining an individual's race by making inferences (or possibly requiring evidence) about whether she was exposed to disparate treatment should not raise the same constitutional concerns.

As an empirical matter, the vast majority of people claiming affirmative action subsidies on the basis of self-reported minority status are bone fide under this standard in that they have been subjected to systematic disparate treatment because of their perceived minority status. While hard cases of fraudulent or undeserving claims of minority status exist, they are by comparison statistical aberrations. While the current stability of racial identity is empirically contingent, the Supreme Court difficulty posed by making determinations in hypothetical cases that might be presented is not a sufficient "case in controversy" to justify the Court's strong preference for race neutral means.

In sum, avoiding individual determinations of race represents the strongest justification for preferring race-neutral subsidies, but such a gain comes at the certain expense of poorer tailoring and the very probable expense of less candid lawmaking and therefore less exacting scrutiny. Richard Fallon notes in his contribution to this symposium:

It is at least oddly disparate to maintain, on the one hand, that explicitly race-conscious reasoning is permissible in justifying an affirmative action program, but to insist, on the other, that race-consciousness is an evil that may not be reflected in an affirmative action program's distributive criteria.

This section has attempted to unpack this odd disparity. When the government's compelling interest is to remedy racial discrimination, the overinclusion prohibition of Croson unescapably points toward explicit racial classifications. Although the Supreme Court seems to have a "least restrictive alternative" principle behind its preference for race-neutral means, this section has shown that properly understood explicit racial classifications do not necessarily burden fundamental rights more than "race-neutral means to increase minority participation."

II. Quotas vs. Credits

The United States insists that Japan commit itself to a clearly stated increase in the number of dealers handling American cars and that the two sides measure the progress annually. Japan has rejected that approach as a use of "numerical targets" and managed trade.

The front page of the June 28 New York Times was remarkable because in adjacent articles the Clinton administration advanced seemingly contradictory arguments on the equity of quotas. In discussing the looming trade war with Japan, the administration argued that Japan had systematically discriminated against United States automobile imports and demanded that the Japanese government agree to import explicit numeric quotas. But in another article discussing the administration's review of federal affirmative action, Clinton reiterated that he was against quotas. When the United States had been the victim of long-standing, systematic discrimination, quotas seem to be the only credible remedy: Assurances from the Japanese that they would stop discriminating and even start implementing U.S. import preferences rang hollow. But like the Japanese, when the United States was placed in the role of dismantling a history of past discrimination, exposing its constituents to quota obligations seemed unduly burdensome. The moral of these newspaper stories is that from a victim's perspective quotas do not seem so inequitable.

The purpose of this section is to show, using an extremely reductive economic model, that quotas or something close to them might be consistent with -- and possibly even required by -- the narrow tailoring principle. This section builds upon the last in that the reader is now asked to accept that the government has established a compelling interest and that explicit racial preferences are permissible. Given this premise, this section asks what form the preferences should take.

The central question I seek to answer is whether quotas or credits are more consistent with narrow tailoring. By quotas, I mean a preference that only allows minorities to compete for set aside government benefits -- such as guardrail contracts or broadcast licenses or medical school slots. By credits, I mean a preference that forces minorities to compete with whites for government benefits, but gives minorities an advantage over similarly-situated whites. A quota guarantees the highest-ranking minority a benefit, while a credit does not.

But before considering directly the choice between quotas and credits, it is useful to point out that either quotas or credits can be implemented by rules or standards. Beginning with Bakke, analysts have often conflated these two dimensions by comparing a rule-like quota with a standard-like credit. Thus, Justice Powell compares Davis' quota/rule with Harvard's fuzzy plus-factor system -- a classic credit/standard. But as illustrated in Figure 1, the other two permutations are also prevalent: a rule-like credit was at issue in Adarand, and Executive Order 11,246 might create a quota standard.

While Powell tried to justified his preference for credits over quotas, he did not explain his seeming preference for standards -- exemplified by his praise for Harvard's fuzzy individualized admissions process where race is used as merely one among many ambiguously weighted factors. It is not self-evident that using standards to implement racial preferences is worthy of praise. Courts will have more difficulty scrutinizing whether the scope of fuzzy racial preferences is narrowly tailored to achieve its goals. Unlike rules, standards by definitions submerge the degree of preference. Although a rules vs. standards debate is not the focus of this section, explicit rules might be more consistent with narrow tailoring.

A. Comparing Quotas and Simple Credits

This section compares rule-like quotas and simple credits (which create constant marginal subsidies to increase minority participation). For example, affirmative action programs that granted minority applicants to law school a constant number of points to their LSAT scores (or that granted employers an invariant subsidy for each additional minority hired) would constitute such a simple credit. The Appendix presents an economic model which illustrates for very particular assumptions that quotas can be better tailored to achieve the government's legitimate goals than simple credits. But because economic models are so off-putting -- especially in the civil rights context -- this section describes the results of the model without graphs or equations.

1. Trading-off Quantity and Quality Burdens

To decide whether a quota or credit is the "less restrictive alternative," it is important to see that the constitutional burden of a racial classification is not solely captured by calculating the number of cases in which the racial preference was dispositive; the racial disparity in quality is also relevant. Allocating a given number of contracts (or admission spots) to minorities is more burdensome when the preference countenances substantial disparities in quality. Reverting back to the meta-goal of minimizing racial divisiveness, non-minorities have a more legitimate claim to being treated unfairly, if they lose out to minorities who are substantially less qualified than to minorities who are only slightly less qualified. In order to capture how much affirmative action deviates from race-blind decision making, it is accordingly necessary to measure not only the number of cases in which race was dispositive, but the average disparity in quality when race was dispositive. In sum, both the quantity (of minority participation induced by affirmative action) and the quality (disparity between minorities and non-minorities) are constitutionally relevant.

A natural way to tailor affirmative action to minimize the divisive burden of racial classification is to acknowledge a tradeoff between the quantity-burden and quality-burden of affirmative action subsidies. A program that granted minorities a 40% bidding credit on 5% of government contracts might be no more burdensome than a program that granted minorities a 5% bidding credit on 40% of government contracts. Even though the 40% bidding credit might operate in effect as a set aside (and impose a substantial quality burden), it only forecloses non-minorities from 5% of the market: the quantity burden -- that is, the number of contracts where race is dispositive -- is much less burdensome.

If the size of the quality disparity did not play an important role in assessing the burden of a particular affirmative action program, then one might easily argue that quotas were perfectly tailored. As long as the quota was set to equal what minority participation would be in the absence of discrimination, then non-minorities would not be foreclosed from competing for any contracts that it would have won in the absence of past discrimination. But the Court's antipathy for quotas stems at least in part from the fact that quotas, by "insulating" minorities from non-minority competition, can countenance arbitrarily large differences in quality -- and that racial differences in quality as well as the amount of participation induced by a racial preferences are relevant. In her Metro Broadcasting dissent, Justice O'Connor has come closest to acknowledging the relationship between the quantity and quality burdens:

The Court's emphasis on the multifactor process should not be confused with the claim that the preference is in some sense a minor one. It is not. . . . [R]ace is clearly the dispositive factor in a substantial percentage of comparative proceedings.

Justice O'Connor sees that the size of the racial preferences as well as the number of cases in which it is dispositive are relevant to determining the burden imposed by the racial classification. What is missing, however, is the crucial notion that the Constitution can tolerate more of one burden if the affirmative action program exhibits less of the other.

Recognizing this quantity-quality tradeoff provides a constitutional underpinning for the rather intuitive notion that in a narrowly tailored program minority participation should be reduced when minority applicants are relatively weak. If the quality disparity between minorities and non-minorities is unexpectedly large, then to reduce the effective burden of the program the maximum quantity of contracts (or admission slots) where race is dispositive should be reduced. For example, even if the government could establish that the minority share of procurement contracts would be 20% in the absence of past discrimination, the Constitution might prohibit the use of 25% bidding credits to induce this participation rate, but it might countenance the use of 25% bidding credits if the credits only were decisive in 5% of the bidding. In short, the "less restrictive alternative" requirement suggests that the level minority participation should be at least in part a function of the strength of minority quality.

2. The Constitutional Burdens of Quotas and Credits

Credits seem to be "less restrictive alternatives" than quotas because with a quota the level of minority participation is not sensitive to the relative quality of minorities competing for the benefit in question. Credits seem to be more narrowly tailored -- in the sense of minimizing the quantity and quality burdens on non-minorities -- because they allow flexibility on just this dimension. For example in a law school admissions program, if minority applicants are relatively strong, a fixed LSAT credit will lead to increased minority acceptances. Or for a construction program in which a general contractor was given a fixed bonus for each additional minority contractor, unexpectedly high minority bids will lead to a reduced participation. In each case, credits induce fluctuations in the level of minority participation that seem better tailored to the government's interest.

But commentators have not realized that credits can induce too much fluctuation: inducing too large a reaction to changes in relative minority quality. While credits laudably harness the general contractors' self-interest to reduce minority participation when minority bids are unexpectedly high, the general contractor is not a perfect agent in pursuing governmental objectives.

The possible divergence of the government's interest in adjusting minority participation to respond to the relative strength of minority bidders and the general contractor's interest under a credit regime represents a classic externality. If minority bids are unexpectedly strong, an invariant credit might induce general contractors to overshoot -- by increasing minority participation above the level it would be in the absence of discrimination. But such a result would clearly be inconsistent with narrow tailoring: the government has no compelling remedial interest in causing minority participation to exceed what it would be in the absence of discrimination.

Analogously, if minority bids are unexpectedly weak, the general contractor -- in reducing minority participation -- internalizes the foregone credit subsidy, but has no reason to consider the impact on the government's remedial purpose. The government might have a particularly strong interest in avoiding drastic shortfalls in minority participation: For example, one might imagine markets in which failing to maintain a minimum market share would undermine the long-term viability of all existing minority businesses. Simple credits would give private decisionmakers no reason to consider how dramatic, short-term shortfalls in minority participation would affect the government's long-term remedial interest.

In short, the government's remedial interest in inducing marginal increases of minority participation is not constant, (but simple credits give decisionmakers constant marginal subsidies). The government's interest of inducing "token" participation may be quite small, but assuring some critical mass of participation may be particularly important to achieving the government's long term remedial goal. Moreover, there are strong reasons to think that the benefits of marginal increases in minority participation must at some point decline toward zero. Consider two markets where in the absence of discrimination minorities would have a 40% market share. Most civil rights advocates would find that it would be more valuable to raise minority participation from 15 to 20 percent than it would be to increase participation from 35 to 40 percent. But simple credits which grant a constant subsidy for marginal increases in minority participation do not capture these differences. And as discussed in the last section, the "less restrictive alternative" analysis might countenance larger quality disparity when smaller percentages of the market are affected.

The crucial failure then of simple credits (which provide a constant marginal credit for increased minority participation) is that they might overshoot the legitimate response to relative strength of minority bidders -- countenancing large fluctuations in minority participation. When a credit might create the possibility of large fluctuations in minority participation and when there are significant social harms from large or small levels of participation, the fluctuations in participation caused by the credits might be more poorly tailored than the invariant participation caused by quotas. The Appendix shows this possibility for very specific assumptions, but the intuition is much broader: If credits do not give private decisionmakers the correct incentives to choose the correct level of minority participation, then delegating this decisionmaking via a credit program may be more poorly tailored to achieve the government's interest then using a quota to mandate an invariant level of participation.

B. Declining Credit Schedules

Let me be quick to emphasize that quotas are not perfectly tailored. A more enlightened choice than simply choosing between quotas or simple credits is to establish a declining credit schedule which conditions the size of the credit on the quantity of minority representation. The goal would be to set the credit schedule to approximate the government's actual interest in marginal increases in participation. Instead of granting minorities a 10% bidding credit for all procurement contracts, a credit schedule might create the incentives depicted in Table 1:

Table 1: Arbitrary Example of Declining Credit Schedule

Cumulative Effect

Bidding Disparity Between Minority and Non-Minority Bidders

Maximum Minority Market Share

a 15% credit for 5% of the contracts

10 - 15%

5%

a 10% credit for 20% of the contracts

5% - 10%

25%

a 5% credit for 5% of the contracts

0% - 5%

30%

a 0% credit for 70% of the contracts

< 0%

100%

The next section will say more about how these percentages might be calculated, but for now it is only important to see that the declining credit schedule still allows for the strength of minority bids to affect the degree of participation: if minority bids are less than 5% above non-minority bids, the minority market share can be as high as 30%; but if minority bids are more than 10 percent above non-minority bids, minority participation falls to at least 5%. The key to a declining-credit schedule is that the larger the disparity in quality (measured here by the size of the bid), the smaller the minority market share.

Creating a credit schedule that more closely tracks the government's remedial interest would mitigate the externality problem of simple credits: The declining credit schedule would cause the general contractor to internalize the welfare costs associated with foregone minority participation. Regardless of whether the minority bidders placed unexpectedly high or low bids, the general contractor facing a credit schedule would have a better incentive to adjust participation without over- or under-shooting.

To implement a declining-credit schedule, it would not be necessary that all the procurement contracts be bid simultaneously or that the contracts with the highest minority credit be bid first. For example, to implement the declining credit schedule, a government procurement agency could announce that it would award all contracts to low bidders, but that at the year's end, minority bidders (who had won contracts) would be eligible for a bonus. The size of this year-end bonus would be a function of the minority market share of total procurement dollars. For example, the declining credit schedule depicted in Table 1 would pay prevailing minorities 15% more than their nominal bids if minorities won less then 5% of the procurement contracts, but only a 10% bonus if the minority market share was between 5% and 25% and no bonus if the minority market share was more than 30%.

Indeed, the simplest way to create a declining-credit effect would be for a procurement agency to announce that winning minority bidders in a given fiscal period would share pro-rata in a fixed bonus. If several minority contractors prevailed, a fixed bonus split several ways would amount to a small bidding credit; but if relatively few minority contractors prevailed, the fixed bonus would be substantially higher. This fixed-bonus method naturally trades-off the quantity- and quality-burdens described above: dividing a fixed bonus pro-rata among minorities could ensure that there would only be large disparities in quality (measured as before by price) when there was a relatively small quantity of the market being allocated on the basis of race (and conversely race would only be decisive for a larger market share if the average quality difference was relatively small). Fixed bonuses also have affinities with cumulative voting. Just as cumulative voting allows minority shareholders to cumulate their votes on a few candidates to ensure some minimum level of representation, fixed bonuses allow the government procurement office to cummulate (or spread) its remedial effort to assure the maximum amount of minority participation, while assuring that the total quantity and quality burdens to non-minorities would in no event be greater than a predetermined number of dollars. And as Peter Cramton and I have shown elsewhere, the cost to government of distributing such fixed bonuses is almost certainly less than the face value of the bonus -- as these ex post bidding credits can induce more competition and induce the non-preferred firms to bid more competitively.

Several affirmative action programs currently implement a type of declining credit schedules. For example, in a recent auction of 30 (narrowband) paging licenses, the FCC granted what amounted to a 50% bidding credit for 12.5% of the licenses; a 16% bidding credit for 25%; and 0% for the remaining 62.5%. Moreover, the affirmative action program at issue in Adarand itself exhibited a kind of varying credit schedule: A maximum subsidy equaling 1.5 percent of the general contract could be earned for hiring one minority subcontractor, but only .5 percent more could be earned for hiring a second minority subcontractor, and no additional subsidy was offered for hiring more than two minority subcontractors.

Moreover, implicit credit schedules lie behind many of the admission programs that grant minority applicants preferences. I submit that few programs would give minorities an invariant credit. Imagine, for example, a college wishing to admit 1000 applicants almost exclusively on the basis of SAT scores -- but with a preference for racial minorities. Even if a 10 percentile credit had been sufficient to admit 100 minority students in recent years, admissions directors would almost inevitably make this credit sensitive to the relative strength of minority applications in a given year. For example, in a particular year, if the 10 percentile credit would only succeed in admitting 8 minorities instead of the usual 100, I predict that most admissions offices would in effect increase the credit to admit a few more minority applicants. Conversely, if a 10 percentile credit succeeded too well by admitting 400 minorities, I predict that most admissions offices would decrease the minority preference. Thus, if we pierced the veil of standard-like affirmative action programs to discover the underlying rule-like formula, we are likely to find that the de facto credit is implicitly a function of the degree of minority participation. Far from being bad, the foregoing analysis suggests that conditioning the size of a minority preference on the quantity of minority participation is generically the most tailored means. Countenancing larger preferences when minority participation is low better tailors the means to the government's legitimate interests without increasing the burden on non-minorities (because the larger disparity in quality is offset by the smaller quantity of minority beneficiaries).

C. Quasi-quotas

Credit schedules might, however, be criticized on the grounds that they create quasi-quotas. For example, granting minorities a 50% bidding credit for a small number of FCC paging licenses may have effectively set aside licenses -- insulating preferred bidders from non-minority competition. Granting substantial bidding credits to ensure at least small amounts of minority participation can, however, still be defended on narrow tailoring (and least restrictive alternative) grounds for four reasons:

First, unlike an absolute quota, a quasi-quota, by assuring a minimum level of minority quality, would at least to a small degree tailor the level of participation to the strength of minority applicants. A 50% bidding credit unlike a set aside, does not absolutely guarantee that the benefit will be allocated to a minority. While it is unlikely that minority bidders would be so weak that they could not prevail, the substantial credit -- as opposed to an absolute set aside -- provides some assurance that prevailing minorities must have some minimum qualifications. Moreover, a substantial bidding credit is more easily subject to strict scrutiny than a quota. A quasi-quota states directly how much of a subsidy the government would be willing to pay (how much of a cost difference) in order to assure a minimum level of minority participation. The absolute quota does not as directly "price" the government's interest or if it does, it implausibly implies that ensuring minority participation would be worth any burden on non-minorities. Many racial preferences, which seem to be absolute set-asides, have built into them limited amounts of interrace competition. For example, both 8a and the dreaded "Rule of Two" only allow minority set asides if the procurement officer certifies that minority bids will not be more than 10% above non-minority bids.

Second, a quasi-quota would only be appropriate if the government found that shortfalls in participation below some minimum level would seriously undermine the government's remedial effort. To justify a quasi-quota, the government would need to do more than merely establish its overall remedial goal: the level of minority participation that would have developed in the absence of discrimination. The secondary showing would likely focus on the long term impact of having a dramatic shortfall in minority participation far below the ultimate goal. For example, evidence about the "minimum efficient scale" of production and technological "learning curves" might suggest that such a shortfall would threaten the long-term success of the remedy -- by threatening the viability of minority contractors who were in the process of establishing themselves as self-sufficient competitors. In such circumstances, a quasi-quota would be better tailored because it would cause decisionmakers to better internalize the true social costs of dramatic shortfalls in minority participation.

Third, because a quasi-quota would only set aside a fraction of the government's legitimate remedial goal, it would impose a smaller burden on the interests of non-beneficiaries. As discussed above, in estimating the burden of an affirmative action plan on non-minorities there is a tradeoff between the quantity of slots allocated because of race and the quality of the disparity between minorities and non-minorities when race is dispositive. A quasi-quota almost by definition imposes a larger quality-burden, but because it applies to fraction of the government's overall goal, the quantity burden on non-minorities is relatively small. Non-minorities are practically foreclosed from a portion of the market, but the non-minority share is still much larger than it would be in the absence of discrimination. Creating quasi-quotas to meet the government's entire remedial goal would not be narrowly tailored. But substantial bidding credits which created what I have called quasi-quotas would constitute the least restrictive or burdensome way of forestalling against the negative remedial consequences of a dramatic shortfall in minority participation. For example, if the government found that in the absence of (past or ongoing) discrimination minorities would provide 30% of procurement contracts, it might be appropriate to create quasi-quotas for 5% or 10% of the procurement dollars -- but it would not be narrowly tailored to effectively set aside 25 or 30% of the procurement dollars.

Finally, granting minority enterprises effective guarantees of minimum participation can increase the quality of minority participants -- so as to reduce the disparity between minority and non-minority recipients. Offering minorities a "safety-net" can enhance the quality of minority participation and increase the quality of minority applicants/bidders. For example, a law school that guarantees that it would be willing to accept up to a 50 percentile difference in LSAT scores if it is necessary to admit 5 percent minorities might induce a much stronger pool of minority applicants (if minorities want to minimize the risk of going to a school where they will represent a token presence). Effectively guaranteeing at least 5 percent participation may allow the school to increase minority participation with a much smaller LSAT credit. The reverse of this "herding" effect is occurring now in University of California schools. The Regents' directive that race not be used as an admissions criterion -- even though it does not become effective until 1998 -- has already caused a substantial decline in minority admissions.

The usefulness of quasi-quotas in strengthening minority quality is not limited to educational contexts where non-economic motives abound and where minority applicants see their participation as complementary. The strongest evidence can be found in recent FCC paging auctions. As described above in the regional narrowband auction of 30 licenses, the FCC granted designated entities (i.e. minority- and female-controlled firms) what amounted to a 50 percent bidding credit for 10 of the licenses and a 16% bidding credit on the remaining 20 licenses. The 50% bidding credit -- by virtually assuring that at least 10 licenses would go to "designated" firms -- very likely induced a number of minority and female businesses to form. Knowing that there were at least 10 licenses effectively set aside may have induced many designated firms to undertake the fixed costs to organize, investigate and finance. In this narrowband auction, this guarantee was so effective in inducing minority- (and female-) controlled firms to form that the designated entities bid away 40% of the bidding credit on the set-aside licenses (meaning the nominal bids on the 10 safety-net licenses were 40% higher than those paid by non-preferred firms for similar licenses) and a designated entity even ended up winning one of the licenses where only the 16% bidding credit obtained.

There is an important lesson here. The government can offer very substantial bidding credits to ensure minimal minority participation often without bearing the constitutional burden of large disparities in quality (or price). Guaranteeing a minimum amount of participation may induce stronger designated firms to form so that in the end the government need not payoff on the guarantee. Perversely, guaranteeing your willingness to grant a very substantial subsidy for a minimum amount of participation may actually reduce the racial credit that is needed to achieve the government's remedial goal.

Even though quasi-quotas might seem to violate the spirit of Bakke's prohibition against quotas, this subsection has shown that substantial racial preferences which virtually insulate minority bidders or applicants from competition are consistent with -- and might be required by -- the underlying narrow tailoring and least restrictive alternative principles. Upon a showing that dramatic shortfalls in minority participation (from what it would be in the absence of discrimination) would substantially impede the government's remedial effort, and especially upon a showing that a quasi-quota would likely enhance minority participation, the government would be justified in establishing substantial bidding credits to assure a small proportion of its overall remedial goal.

* * *

The take home lessons of Part II have been three-fold:

(1) Invariant credits may be less narrowly tailored than a quota to further the government's remedial objective because they may induce too much fluctuation in minority participation.

(2) Declining credit schedules which grant larger racial preferences the larger the shortfall in minority participation (from what it would be in the absence of discrimination) are most consistent with the narrow tailoring and less restrictive alternative principles. Declining credit schedules are better tailored to the government's remedial interest, and countenancing a larger quality disparity for a smaller quantity of government benefits can be a less restrictive alternative.

(3) A narrowly tailored credit schedule may attempt to achieve a small fraction of the government's legitimate remedial goal with bidding credits that are so substantially that they effectively set aside some of the government benefits at issue.

While the foregoing analysis is theoretically coherent, the next section explores the difficulty of creating a declining credit schedule in practice.

III. Tailoring the Scope of Racial Preferences in the Real World

In contrast to the top-down doctrinal and economic analysis of the first two sections, this section works from the bottom up -- looking at the kinds of data that are currently being used to justify remedial affirmative action programs -- to see how a declining credit schedule might be implemented. The declining credit curve is meant to approximate the government's remedial benefit from marginal increases in minority participation. Calculating such a curve is a daunting task. There is little social consensus whether there are any remedial benefits to race-conscious remedies. It blinks reality to think that government could calculate numeric quotas that precisely capture the government's remedial interest.

In the absence of precision, how might the government proceed? As a first step, one might retain the current method of assessing what minority participation would have been in the absence of discrimination. Calculating this market share would at least pin down one end of the credit schedule -- by establishing the participation level at which the bidding credits should end. A number of states have conducted post-Croson studies which estimate what the participation of minorities would be in particular markets in the absence of discrimination. While this benchmark provides some evidence of the minority participation level where the credit schedule should phase out, it does not help numerically estimate how steeply a narrowly tailored credit schedule should slope. This section first discusses the "state of the art" in calculating the overall remedial goal -- what the minority market share would be in the absence of discrimination -- and then speculates what types of evidence might be brought to bear in setting the size of bidding credits to encourage lower levels of participation.

A. Estimating the Overall Remedial Goal

The current "state of the art" is extremely crude. In recent post-Croson disparity studies, the estimate of what the minority business market share would be in the absence of discrimination is calculated in two stages. In the first stage, the number of minimally qualified minority firms is divided by the total number of minimally qualified firms. Much turns on how one defines "minimally qualified." In a New York study, firms were selected as being minimally qualified if they appeared on "agency procurement 'bidders lists' which contain firms that have either made themselves available to do business, or have actually done business, with one or more agency." Using the raw number of minority firms, this first-stage calculation yields the minority percentage of firms who have attempted to bid. The second stage then adjusts this percentage to account for the possibility that more minority firms would have been available to bid in the absence of discrimination. To make this "but-for discrimination" adjustment, the studies calculate a regression to determine how much more likely non-minorities with similar educational and work experience are likely to be entrepreneurs.

In the New York disparity study, minority business enterprises were found to constitute 11.6% of the bidders list. The "but-for discrimination" regression suggested that there would have been 20% more minority firms in the absence of discrimination -- so that after adjustment, the percentage of minority bidders who would have existed in the absence of discrimination would be 13.9. This number serves as a crucial benchmark for establishing the government's overall remedial goal. In New York, the minority market share of state procurement dollars was only 5.8 percent. The disparity between the government's estimate of what the minority market share would be in the absence of discrimination (availability: 13.9%) and what the minority market share currently was (utilization: 5.8%) establishes not only the government's compelling interest but illuminates whether a particular affirmative action program is narrowly tailored to achieve the 13.9% remedial goal.

However, this whole methodology is based on a fairly extreme assumption that if 13.9% of the firms are minority-owned, these minority firms would, in the absence of discrimination control, 13.9% of the procurement revenues. Courts have rejected head-counting -- in that they have rejected using the minority percentage in the general population as a benchmark of what market share minorities would control in the absence of discrimination -- but in its stead, they have accepted a fairly crude method of firm-counting. The assumption that existing minority firms would control a proportionate market share in the absence of discrimination heroically compares firms with radically different capacities. In this calculus, a Fortune 500 firm is given the same weight as a firm one-thousandth its size. Even though the disparity studies label the first stage estimate as a measure of "availability," there is little or no attempt to control for the capacity of existing firms. The methodology is especially crude in a post-Croson, post-Adarand world where one byproduct of two decades of affirmative action might be the nominal existence of a large number of small, low-capacity minority firms. Indeed, there is nothing in this methodology to limit the market share estimate to the minority population in the general population.

At the end of the day, I believe it is wishful thinking to believe that firm-counting provides a more accurate benchmark than head-counting (for example, the minority percentage among college graduates). Indeed, it might be appropriate simply to do away with the first stage altogether and simply use the second-stage regression analysis not merely as an adjustment, but to accomplish a more sophisticated type of head counting. The regression method used in the New York disparity study could be used -- without ever adverting to existing number of firms -- to calculate the number of minority entrepreneurs that should exist given the number of existing minorities' various educational and work experiences. The regression is consistent with the Croson requirement that a state may not use procurement affirmative action to remedy discrimination in prior educational opportunity, because it takes the current number of minority MBAs as given. The regression in a sense calculates how many of these MBAs should be entrepreneurs -- assuming that in the absence of discrimination minority MBAs would become entrepreneurs at the same rate as non-minority MBAs. The multivariate regression methodology of course controls not just for a person's MBA status, but for a variety of human capital variables. The point is that this sophisticated head-counting is likely to be more reliable than firm-counting because the assumption that human capacities would be similar in the absence of discrimination, while controversial, is more reliable than that firm capacity would be similar in the absence of discrimination.

B. Setting the Size of the Credits

The prior discussion only attempted to calculate the level of minority participation at which racial credits should equal zero. It becomes all the more Herculean to estimate what size of credit is appropriate to try to encourage lower levels of participation. To annunciate any particular bidding credit sounds arbitrary because there is no empirical method for balancing the remedial benefit of increasing minority participation against the various potential costs (including inefficient production and reduced opportunities for non-beneficiaries). And I can offer here no magical formula for calculating the optimal size of bidding credits within a bidding schedule. However, it still may be possible to make progress by recognizing that a declining credit schedule might pass strict scrutiny even if the government cannot deduce that this program uniquely maximizes some metric of social welfare.

Instead, the government might be able to establish that there are strong reasons for encouraging some minimum level of minority participation. Even though a credit schedule relates a level of participation to a particular bidding credit size, producing evidence that maintaining a minimum participation level is essential to the government's remedial goal would go a long way toward justifying a more substantial bidding credit -- including the type of quasi-quotas described above -- even though it still would not provide very commensurable evidence about just how large this more substantial credit should be.

In essence then, I recommend that governments estimate multiple market shares. As before, it would estimate the share which would achieve the overall remedial goal, but it might also identify lower levels of minority participation that were particularly important to effectuate the remedy. Even though an idealized credit schedule (as depicted in the Appendix) could comprise an infinite number of bidding credits, given the limits of administrative feasibility, it is much more likely to have two or three bidding credit categories -- as arbitrarily exemplified earlier in Table 1. The government would adduce evidence that a minimum participation level was particularly important and then, probably without more, argue that the specific bidding credit was sufficiently tailored to meet this heightened remedial need.

Evidence that a minimum market share is particularly useful or necessary -- even though not currently used in post-Croson litigation -- might be adduced. Industrial organization economists could estimate the minimum efficient scale of existing minority enterprises and predict the effect of particular size credits on the ability of these firms to exist and grow. Evidence that minority enterprises were establishing credit ratings and acquiring market-specific know how might be particularly relevant in justifying a short-term safetynet.

Moreover, in reviewing the size of bidding credits the Court should be clear to distinguish between facial and applied reviews of affirmative action programs. As discussed above, programs which on their face grant substantial bidding credits may in practice not countenance substantial disparities. Thus, the 50% bidding credits granted to minorities in FCC auctions induced so many minority (and female) enterprises to form that the vast majority of the credit was bid away. In reviewing the reasonableness of a large credit, the Court needs to think about the likelihood that it will be used -- that is, the expected difference between minority and non-minority bidders. The recent experience at the FCC auctions suggests that even a 50% bidding credit might actually reduce the expected difference between minority and non-minority bids, and thus provides an especially strong rationale for what on its face seems to be a rather extreme racial subsidy.

C. Overview of the "160" Federal Affirmative Action Programs

To get a handle on how much Adarand is likely to change federal affirmative action programs, it is useful to begin with the "comprehensive list" of race- and gender-conscious programs that the Congressional Research Service compiled at Senator Bob Dole's request. Although it has been reported repeatedly that there are 160 programs, quantifying the pervasiveness of affirmative action by this type of program counting is pure folly. For example, 19 of these programs are mere exhortations that the recipients of various government grants are "encouraged" to deposit their funds in minority- and women-owned banks. Since so little (read: nothing) turns on these exhortations, these provisions could be consolidated or eliminated without affecting the viability of minority- or women-owned banks. Thus, if Clinton wanted to, he could easily halve the number of programs by merely eliminating the toothless exhortations -- without affecting the core preferences in federal subsidies where for example $4.8 billion in contracts were set aside for minority and female contractors under the Small Business Administration's 8(a) program.

The Clinton administration might also defend dozens of grants for "historically Black Colleges and Universities" (often referred to as "HBCUs") by arguing that they are "race-neutral means to increase minority participation." Unlike women's colleges which need to discriminate against male applicants in order to preserve a female identity, many HBCUs could maintain a minority majority with a color-blind admissions process. However, to avoid explicit racial classifications, some of the preferences would need to be more narrowly drawn to exclude the additional preference for "institutions having at least 50% minority enrollment." While these programs might be strongly defended on remedial grounds, it should be noted that these very substantial preferences are clearly inconsistent with Amar and Katyal's diversity theory. Far from their image of bringing diverse racial groups together, subsidizing the continued existence of what will continue to be virtually all-black colleges seems to produce just the balkanization that Amar and Katyal claim describes private contracting.

Stepping back, one sees in the comprehensive list all four permutations of rules/standards and quotas/credits. It may be, contrary to my earlier proposal, that the Supreme Court will scrutinize fuzzy (standard-like) racial preferences more leniently, but at least with respect to rule-like quotas and credits which are especially prevalent in federal procurement, the administration will need to undertake disparity studies to establish not only its compelling remedial interest, but to justify the size of the racial preferences. Before such studies are completed, it is impossible to assess precisely what types of programs can survive strict scrutiny. However, the essay has tried to show that stark shortfalls in minority participation (compared to what it would have been in the absence of discrimination) might -- consistent with the narrow tailoring principle -- justify a rule-like, explicitly racial preference.

Conclusion

In an article reporting that the Clinton administration planned to impose a three-year moratorium on procurement set-asides, an anonymous administration official suggested that the administration was moving toward an approach that is consonant with declining credit schedules:

The Administration has decided to allow Federal agencies, if they can justify it, to use other kinds of preferences, like giving price breaks and extra points in evaluating contract bids by minority and woman-headed companies. . . . Agencies determine a benchmark for the percentage of minority or female contracts it ought to have. That benchmark will determine what kind of affirmative action steps the agency may take. The further away from the benchmark an agency finds itself, the more blatant[ ] preference . . . in contracting it may employ.

The article makes clear that the agency benchmark will be derived from the type of post-Croson "disparity studies" described above that will attempt to determine what the minority market share would be in the absence of discrimination. The last quoted sentence in particular seems to indicate that the preferred racial preferences will have the essential attributes of a declining credit schedule, in that when minority participation is "further away from the benchmark," the agency can justify a larger racial credit.

I have argued that facially neutral, but racially motivated classifications are not more narrowly tailored (and may not provide a less restrictive alternative) than explicitly racial classifications. And that in choosing among different types of racial classifications, policy-makers should consider declining credit schedules that reduce the marginal racial subsidy as racial participation increases toward the overall remedial goal. Most provocatively, this essay has argued that very substantial bidding credits, which come close to guaranteeing a minimum level of minority participation, may be consistent with narrow tailoring -- both because they may better reflect the substantial remedial benefits of maintaining a minimum, critical mass of participation, and because guaranteeing a virtual safetynet can enhance the quality of minority applicants and reduce the racial disparity among participants.

From a narrow doctrinal perspective "something must give" in the Croson and Adarand opinions. The Supreme Court cannot consistently prohibit non-victim races from receiving remedial subsidies and at the time encourage race-neutral subsidies (which include preferences for certain non-victim whites). And if the court chooses to privilege race-neutral means to increase minority participation, it will likely need to abandon the "consistency" principle so that it can uphold racially-motivated statutes that benefit the victims of racial discrimination

While this essay has considered quotas vs. credits, rules vs. standards, and race-neutral vs. non-neutral means, there are several other dimensions on which the narrow tailoring principle might apply. For example, almost no consideration has been given to what might be called "Coasean" tailoring: Whether it is more tailored to grant benefits directly to minorities or to parties who contract with minorities. When the stringent assumptions of the Coase theorem do not hold, the remedial incidence of direct and indirect subsidies may diverge. After all, Adarand itself involved a subsidy to a non-minority general contractor and the Viacom controversy this past summer involved an FCC tax subsidy for a non-minority licensee. This Coasean choice has practical import. It is often overlooked that most of our affirmative action efforts are directed toward increasing the participation of minority-owned firms and not in increasing minority employment. While these capitalist-centered programs have much to say for them, the vestiges of past discrimination are at least as equally prevalent in many job categories, and increasing the power of minority capitalism may prove to be a poor method (as seen with other "trickle down" theories) to benefit minority labor.

This essay has also not discussed the crucial issue of temporal tailoring. There are increasing indications that a majority of the Supreme Court might require sunset provisions for affirmative action programs. A sunset requirement makes most sense if discrimination is viewed strictly as a thing of the past. If this were true, the causal nexus between the initial victims of discrimination and the subsequent beneficiary class would become increasingly attenuated. And it would become all the more difficult to estimate with confidence what market share minorities would control in the absence of discrimination. Under such conditions, a temporally tailored program might gradually decrease the market share at which bidding credits (or other plus factors) are phased out. Notice this temporal tailoring seeks to achieve diacronically what the declining credit schedule seeks to achieve syncronically. Indeed, declining credit schedules can help provide evidence of whether the racial preferences are still necessary to induce the desired level of minority participation. If credits diminish toward zero as participation rises toward the remedial goal, then the goal will only be reached if the weakest minority bidder can participate in the market with a very small subsidy. As long as the vestiges of discrimination increased the costs of minority bidders, it would be unlikely that a declining credit schedule would overshoot the overall remedial goal -- so there would be a smaller likelihood that courts would suspend a program mistakenly thinking that the credits were no longer necessary.

Let me quickly add, however, the assumption that discrimination is a thing of the past is repeatedly and strongly contradicted by a wide variety of empiricism. We would not consider imposing a sunset requirement on traditional remedies for tortious assault, because we have evidence that assault continues to occur in the present day. A sufficient showing that the government is -- to use the generative Croson phrase -- "a passive participant" in present acts of disparate racial treatment strongly rebuts the unqualified argument that remedial affirmative action must soon end. For this reason, agencies seeking to respond to temporal tailoring arguments should consider testing for both private and public disparate treatment as part of on-going "disparity studies" to satisfy strict scrutiny.

While this essay has used some economics to illuminate the narrow tailoring principle, let me end by emphasizing that there are many issues where economics is not particularly helpful. For example, I doubt whether economic analysis could help illuminate whether the remedial or diversity bases for affirmative action are "compelling state interests." But tailoring a remedy to further a particular goal may constitute a type of low-level "tinkering" in which economics -- especially with its focus on marginal costs and benefits -- may aid constitutional policymaking.

Appendix

In the following discussion, whether a program is narrowly tailored is assessed using marginal cost-benefit analysis. Marginalism is particularly appropriate for the task because it focuses on whether incremental increases in the scope of one program or another is worth the candle. However, to proceed we need to make some assumptions about the general shape of the marginal costs and marginal benefits. The goal here is merely to describe the slope that the marginal cost and benefit curves might take and to show the possibility that an invariant quota might be more effective (read: more narrowly tailored) in inducing the desired minority participation than an invariant credit.

To undertake this comparison, I begin by making several highly reductive assumptions about the effects of different affirmative action programs in an Adarand-like procurement context. These assumptions allow me to apply the analysis from Martin Weitzman's classic article, "Prices vs. Quantities." Just as Weitzman's article questioned the "vague preference" that economists have toward price subsidies, my purpose here will be to show that the preference for bidding or other price-like credits (and against quotas) is overstated.

We first assume that the general contractor's marginal costs of increasing minority participation are increasing. This assumption is depicted in Figure 2 by the upward sloping marginal cost curve -- where the horizontal axis measures some quantum of minority participation such as the dollar revenue going to minority subcontractors (or the number of minority employees) and the vertical axis measures the marginal cost in dollars. This increasing marginal cost curve also crucially describes how the contractor will react to various minority subcontracting credits. The first credits that we consider are constant marginal subsidies for increased minority participation. Rational decisionmakers will respond to a credit by increasing minority participation to the point where the marginal costs equals this constant marginal subsidy. As shown in the figure, if a general contractor were offered C* dollars for each unit increase in minority participation, then the general contractor would increase minority participation by Q*. And intuitively increasing the size of the credit (say in the Figure 2, up to C') induces the general contractor to increase minority participation even more (to Q').

To complete the model we need to consider the marginal costs and benefits of increased minority participation on others in society. The goal here is to think about the shape of a "Marginal Net Benefits" (MNB) curve which aggregates the marginal impact on social welfare of increasing minority participation (excluding only the costs to the general contractor which is already captured in the marginal cost curve). Under our assumption that the total benefits of increasing minority participation are larger than the social cost, it must be true that the marginal net benefits must be positive for at least some increases in minority participation.

Moreover, there are strong theoretical reasons to think that the MNB curve must at some point become downward sloping. While it is true that the net benefits of creating "token" minority participation may be small, the marginal net benefits must at some point start to decline toward zero. Once a minority participation reaches what it would be in the absence of discrimination, there is no reason to think that the marginal net benefit of increasing participation beyond this point would be positive: the benefit of remedying past discrimination would no longer be present and the cost to the non-beneficiary race, if anything, would be exacerbated.

The assumption that marginal (net) benefits of affirmative action are downward sloping can be easily captured in an example. Consider a market where, in the absence of discrimination, minorities would have a 40% market share. The declining marginal benefit assumption suggests that more social benefits would be created from increasing minority participation from 15 to 20 percent than there would be from increasing minority participation from 35 to 40 percent.

This analysis is depicted graphically in Figure 3. Because the marginal benefit of inducing token participation is not as large, the NMB curve rises at (from Q' to Q''). But after overcoming this "token" effect and reaching some critical mass, the curve declines toward Point Q''', which represents the participation that minorities would have attained in the absence of discrimination. The NMB curve might hit the horizontal axis before this "proportional representation" point depending on how one accounts for the burdens to non-beneficiaries. For the purposes of this analysis, what is important is that the marginal cost curve generally declines after minority participation reaches some critical mass, and that the marginal benefits for government inducing increased participation are extinguished at least by the time when minority participation is what it would have been in the absence of discrimination.

Even though there may well be factors -- such as the "token" effect -- that cause the marginal benefit curve to slope upwards, the remainder of this section (for simplicity) considers only downward sloping MNB curves. Again, I only need to convince you that the MNB curve might plausibly be downward sloping and that the general contractor's marginal cost curve might be upward sloping. Accepting that a particular market might exhibit such marginal costs and benefits is enough to assess whether quotas or credits are more narrowly tailored.

To make the question interesting, we need to introduce some uncertainty about how the general contractor will react to a credit scheme. If the marginal cost and net benefit curves are precisely known, then either a quota or a credit might implement the same equilibrium. By identifying the intersection of the two curves, a social planner could choose a quota or a credit which tailors the increased minority participation to the government compelling interest. This is done in Figure 4 by a credit set at C* or quota set at Q*.

To make quotas and credits non-equivalent policy instruments, imagine that the government is uncertain about how much higher minority bids will be than white subcontracting bids. There are many types of uncertainty that the government might face, but (just to create a simple example) assume that the slope of the marginal cost curve is known, but that the intercept of the curve might be shifted either up or down by some amount, .

With these assumptions we can explore graphically whether a quota or a credit is more successful. A perfectly tailored rule would make the degree of minority participation depend on the relative strength of the minority bidders. If minority bids are unexpectedly high, increasing minority participation is more expensive -- so the optimal amount of participation is lower. Analogously, if the minority bids are unexpectedly low, increasing minority participation is less expensive and the optimal participation rate is higher. These optimum participation rates for the two possible states of the world are depicted in Figure 5 by q*L and q*H -- which represent the quantities where the MNB curve intersects the two possible MC curves.

Figure 5 can help us assess how well quotas or credits succeed in tailoring actual participation rates to these optimal benchmarks. The quota induces minority participation of qQ regardless of whether the minority bidders enter relatively high or low bids. When the minority bids are unexpectedly high, the quota induces too much minority participation (qQ > q*H). The inefficiency caused by this oversupply is represented by a triangle in Figure 5 drawn between the optimal and actual quantity: when the minority bids are unexpectedly high, the marginal cost of minority participation at qQ is larger than the marginal net benefit. The marginal cost to the general contractor remains above the marginal net benefit to the rest of society at all points between qQ and q*H and the triangle area represents the total loss in welfare from the quota's failure to tailor when minority bids are unexpectedly high.

Quotas create an analogous inefficiency by inducing too little minority participation when minority bids are unexpectedly low. When the cost of increasing minority participation is unexpectedly low, the optimum minority participation is greater than the quota (q*L > qQ) -- but the quota restrains participation to a level where the marginal benefit is greater than the marginal cost. The tailoring inefficiency associated with this shortfall is depicted by an analogous triangle.

Figure 5 also shows, however, that a simple credit fails to induce the optimum participation rates. Quotas are inefficient because they do not vary with the strength of minority bids; credits are inefficient because they vary too much. When minority bids are unexpectedly high, the quota mandated too much minority participation, but Figure 5 shows that a credit induces too little participation. General contractors' responding to a credit and high minority bids will increase minority hiring only to the point where the marginal subsidy of the credit equals the marginal cost of increasing minority participation. In Figure 5, this point occurs at qCH. But at this level of participation, the marginal net benefits to society are larger than the marginal costs. The tailoring inefficiency associated with this shortfall is represented by the smaller, shaded triangle between qCH and qQ. An analogous tailoring inefficiency is created when minority bids are unexpectedly low. The credit induces too much minority participation (qCL > qQ), which is depicted by an analogous small triangle.

Comparing the tailoring inefficiencies, Figure 5 reveals the conventional result. The quota produces a less tailored outcome than the credit. The credit is better tailored because it produces participation rates that are closer to optima. Credits induce more variation in minority participation. And for the assumed marginal cost and benefit curves, making minority participation sensitive to the strength of minority bids is more tailored than making minority participation completely invariant. The residual inefficiency of the credit is caused by general contractors ignoring the marginal benefits of enhanced minority participation -- they only internalize the constant marginal subsidy of the credit. But as drawn in Figure 5 the externality caused by the credit is small, because the marginal benefit curve is relatively flat -- meaning that under a credit there are only slight differences between the marginal benefits to society and the marginal subsidy created by the simple, invariant credit.

If we stopped here, there would be little added-value for this economic modeling. Consistent with Powell's discussion in Bakke, Figure 5 shows that a quota is less well tailored to the government's compelling interest than a credit. However, it is easy to construct an example in which the quota is better tailored than the credit. Indeed, if we merely increase the steepness of the marginal net benefit curve, the quota can become the better tailored means. Figure 6 shows just this possibility. As before, a credit makes the minority participation rate more sensitive to the strength of minority bidding. But in this case, the swings in minority participation are excessive compared to what an optimally tailored participation would be: The mandated quota participation (qQ) is closer to the optimal participation (q*H and q*L) than the level induced by the credit (qCH and qCL). The deviations from the optimal participation level create analogous inefficiency triangles, but in Figure 6 the tailoring inefficiency of the credit is more than that for the quota.

The geometry suggests that quotas become better tailored (relative to credits ) as (1) the marginal benefit curve becomes steeper or (2) the marginal cost curves becomes flatter. But what is the intuition behind this counterintuitive result? The flatness of the marginal cost curve determines how sensitive minority participation will be to size of the credit. If the marginal cost curve is very flat, then unexpectedly high minority bids will cause a large decline in minority participation, and unexpectedly low minority bids will cause a large increase in minority participation. In other words, if a number of minority contractors are likely to place similar bids (but the government is uncertain about how much higher these bids will be than the white bids), it will be very difficult for the government to assess how a particular credit will affect minority participation. If the government sets the credit too low, the general contractor will not be willing to hire any minorities, but if the government sets the credit too high, the general contractor may higher only minorities. The slope of the marginal cost curve thus shows how much variation in minority participation will result from the use of simple credit.

The slope of the marginal net benefits curve then determines the size of the externality associated with this variation. As argued above, when a general contractor decides to reduce minority participation because of unexpectedly high minority bids, it only internalizes the foregone credit subsidy. The general contractor does not consider the foregone opportunity to remedy discrimination. When the marginal net benefit curve is steep, this externality problem becomes larger, and credits accordingly do a poorer job of tailoring minority participation to the government's compelling interest.

Return to top of the Page