American Economic Association
Assessing Affirmative Action
Author(s): Harry Holzer and David Neumark
Source: Journal of Economic Literature, Vol. 38, No. 3 (Sep., 2000), pp. 483-568 Published by: American Economic Association Stable URL: http://www..jstor.org/stable/2565419 Accessed: 18-03-2015 10:04 UTC
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at http://www.jstor.org/page/info/about/policies/terms.jsp
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.
American Economic Association is collaborating with JSTOR to digitize, preserve and extend access to Journal of Economic Literature.
STOR
http://www.jstor.org
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
Journal of Economic Literature
Vol. XXXVIII (September 2000) pp. 483-568
Assessing Affirmative Action
Harry Holzer and David Neumark1
1. Introduction
THE FUTURE OF affirmative action in the United States is uncertain. Proposition 209 in California, passed in 1996, prohibits all government institutions from "discriminating against or giving preferential treatment to any individual or group in public employment, public education, or public contracting on the basis of race, sex, color, ethnicity, or national origin."2 A similar initiative (Initiative 200) passed in 1998 in Washington. Recent court cases (such as Adarand v. Pena) set up strict standards for race-conscious programs to pass constitutional muster. Legislation or referenda pending in many states may further circumscribe affirmative action programs. On the other hand, public opinion polls still indicate public support for some forms of affirmative action (The Gallup   Organization   1997),3  and the
1 Holzer, Georgetown University. Neumark, Michigan State University and NBER. We are grateful to Jess Reaser for outstanding research assistance, and to Scott Adams, Heather Bednarek, Richard Attiyeh, Timothy Bates, Elchanan Cohn, Sandy Darity, Kevin Lang, John McMillan, Karen Roberts, William Rodgers, Dick Startz, Stefanie Wilk, and two referees for helpful comments and suggestions.
2 An earlier 1995 decision by the University of California Board of Regents eliminated the use of race as a factor in undergraduate and professional school admissions.
3 This needs to be qualified. Support for affirmative action appears to drop substantially, especially among whites, when questions are couched
Clinton Administration is committed to "support affirmative action measures that promote opportunities in employment, education and government contracting for Americans subject to discrimination or its continuing effects" (White House Memorandum, July 19, 1995).
Although the debate over affirmative action is both high profile and high intensity, neither side's position is based on a well-established set of research findings. Economics provides an extensive, well-known literature on which to draw regarding the existence and extent of labor market discrimination against women and minorities, although views may conflict (see, e.g., William Darity and Patrick Mason 1998, and James Heckman 1998), and a less extensive but also well-known literature on the effects of affirmative action on the employment of women and minorities (see,
in terms of "preferential treatment" or quotas (Seymour Lipset and Martin Schneider 1978; James Kluegel and Eliot Smith 1986; Donald Kinder and Lynn Sanders 1990; Civil Rights Monitor Leadership Conference on Civil Rights Online Monitor). This was illustrated dramatically in developments surrounding a Houston ballot measure (Proposition A) to ban affirmative action in city contracting and hiring, which was voted down in 1997. Proponents of the ban originally collected signatures for a proposition to ban "discrimination" and "preferential treatment," but the City Council reworded the proposition that went on the ballot to ask whether the city should ban affirmative action.
483
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
484       Journal of Economic Literature, Vol. XXXVIII (September 2000)
e.g., Jonathan Leonard 1989, 1990). However, research by economists provides much less evidence and even less of a consensus on the question of whether affirmative action improves or impedes efficiency or performance, which is perhaps the key economic issue in the debate. This review focuses on all of these questions regarding affirmative action, but the major focus is on the efficiency/performance question. Aside from being a central economic policy question, the research findings that potentially bear on this question are less well-known, appear (often recently) in a wide-ranging set of disciplines and journals, and have not been reviewed, synthesized, and evaluated so as to provide researchers and policymakers with a thorough understanding of our current state of knowledge.
Illustrating the absence of a research base with which to assess affirmative action, the Clinton Administration's extensive Review of Federal Affirmative Action Programs (George Stephanopou-los and Christopher Edley 1995) draws on empirical evidence limited nearly exclusively to discrimination and to the effects of affirmative action on boosting employment, university admissions, and contracting among women and minorities.4 Critics also typically fail to base their views on empirical evidence. Shelby Steele (1990), for example, levels numerous charges against affirmative action, including that it "offers entitlements, rather than development, to blacks" (p. 89), that "blacks . . . stand to lose more from it than they gain"
4 The only evidence it presents on efficiency/performance questions is based on examining whether in open-ended survey responses contractors happened to suggest that affirmative action operated like a quota system (Section 6 of the Revieiv). Since this question was not directly posed, the fact that only a small number of respondents made this suggestion is not very informative.
from it (p. 113), that affirmative action results in the "lowering of normal standards to increase black representation" (p. 117), and does "nothing whatever to stop the very real discrimination that blacks may encounter" (p. 121). Each of these claims is inherently empirical, yet no evidence is offered to support them.
Our review in this paper aims to delineate the key questions economists should be asking about affirmative action to adequately assess the set of policies it represents, and to point out the shortfalls between what we do know and what we would like to know.5 At the same time, there is a growing literature that, in our view, begins to ask and answer some of the right questions.
2. An Overview of Affirmative Action
To begin, we require a definition of affirmative action. In principle, at least, affirmative action can be distinguished from other antidiscrimination measures by requiring pro-active steps (hence the phrase "affirmative") to erase differences between women and men, minorities and nonminorities, etc., in contrast to laws that only prevent employers from taking steps that disadvantage minorities in the labor market, such as refusing to hire them. However, it is more difficult to construct a working definition of affirmative action policies, for a number of reasons. First, the definition of affirmative action as a specific "policy" is fuzzy, since it is more an amalgam of components of other legislation and of court rulings than a single coherent policy. Second, affirmative
5 Our focus on questions of direct interest to economists dictates that we give short shrift to political or philosophical issues, such as the "procedural fairness" of affirmative action (e.g., Robert Folger and Mary Konovsky 1989). However, later we briefly note that these considerations may have economic ramifications, if workers' behavior depends on perceived fairness.
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
Holzer and Neumark: Assessing Affirmative Action
485
TABLE 1
Key Executive Orders, Regulations, and Court Decisions Regarding Affirmative Action in the Lahor Market
1961: Kennedy Executive Order 10925
1965: Johnson Executive Order 11246
1967: Johnson Executive Order 11375
1968: Department of Labor Regulations governing Executive Orders 11246 and 11375
1970: Department of Labor Philadelphia Plan
1979: United Steelworkers of America v. Weber
1984: Firefighters Local Union No. 1784 v. Stotts
Required government contractors not to discriminate against employees or job applicants, and mandated that contractors "take affirmative action to ensure that applicants are employed and employees are treated during employment without regard to their race, creed, color, or national origin."
Reiterated Executive Order 10925.
Amended Executive Order 11246 to cover women.
Requires federal contractors with 50 or more employees or contracts of at least $50,000 to identify underutilization of women or minorities and establish corrective goals and timetables.
New regulations under Orders 11246 and 11375 establishing goals and timetables for employment of minorities in construction.
With regard to an in-house training program reserving 50 percent of spaces for blacks, U.S. Supreme Court ruled that Title VII "does not prohibit such race-conscious affirmative action plans." Defined "permissible" plans as those that break down existing patterns of racial segregation, do not "unnecessarily trammel" on the interests of white employees nor create an absolute bar to their advancement, and are temporary, intended to "eliminate a manifest racial imbalance," rather than "to maintain racial balance."
U.S. Supreme Court stated that court-authorized affirmative action plans were authorized by Title VII to provide relief "only to those who have been actual victims of illegal discrimination."
Sources: Bloch 1994; Edley 1996; Radford 1997.
action is commonly used to refer to policies or behavior in different spheres, including employment, education, and government contracting. Third, affirmative action may operate at a number of different levels and in a number of different ways, including public vs. private, federal vs. state vs. local, and involuntary vs. voluntary. Fourth, affirmative action may cover many different activities, including recruitment, training, hiring, promotion, etc. Finally, the status of affirmative action is undergoing change contemporaneously, as a result of both policy initiatives and court rulings. Our approach in this survey is to cover the relevant literature without restricting attention to
any particular subset of policies, strategies, etc., that fall under the rubric of affirmative action in the civilian sector. In the remainder of this section we provide an overview of the different dimensions, definitions, and domains of affirmative action.6
2.1 Affirmative Action in the Labor Market vs. Equal Employment Opportunity Enforcement
To better understand the potpourri of policies, court rulings, etc., that comprise affirmative action, Table 1 provides a summary of executive orders
6 For fuller discussions, see Edley (1996), and the Review of Federal Affirmative Action Programs (Stephanopoulos and Edley 1995).
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
486       Journal of Economic Literature, Vol. XXXVIII (September 2000)
and court rulings that might be regarded as encompassed by affirmative action in the labor market.7 Executive Order 11246 (which restated an earlier Kennedy executive order) is probably the single "policy" most commonly interpreted as establishing affirmative action (Leonard 1989), because this order is most strongly linked with the interpretation of affirmative action as emphasizing numerical yardsticks with respect to the hiring of minorities and women. In particular, in the initial implementation of Executive Order 11246, the Department of Labor developed what is referred to as the "Philadelphia Plan," which aimed to increase minority representation in construction and is viewed as the "precursor of the numerical 'goals and timetables' obligations of federal contractors" (Bloch 1994, p. 70). Currently, employers with federal contracts and fifty or more employees, or with contracts worth $50,000 or more, are required to file reports indicating "underutilization" of women or minorities in any job group in which minorities or women are underrepre-sented. Contractors are then obliged to address this underutilization by making corrective efforts including the use of written "goals and timetables." Contractors may be sued and barred from federal contracts if they are judged to be discriminating or not pursuing affirmative action, although apparently this latter outcome is rare (Stephanopoulos and Edley 1995).
Describing the current status of affirmative action in the labor market, and its likely status in the near future, is made difficult because of two factors. First, regarding its future status, chal-
7 For thorough discussions of the legislation, rulings, etc., listed in Table 1 see, e.g., Richard Epstein (1992), Farrell Bloch (1994), Stephanopoulos and Edley (1995), Edley (1996), and Mary Radford (1997).
lenges to affirmative action are being mounted at state, local, and federal levels. Although many of these challenges focus on public employment (such as the "Houston Civil Rights Initiative," which failed in 1997), some seek to rewrite federal law in such a way as to undermine affirmative action.8 While more public attention has focused on challenges to affirmative action in education (discussed below), there appears to be a perception that the legal status of affirmative action in the labor market may also undergo serious challenges; this may come not only from public initiatives but also from an expansion of private lawsuits alleging reverse discrimination.
Second, regarding its current status, other legislation that has ostensibly targeted discrimination in the workplace has also led to affirmative action in practice.9 Title VII of the Civil Rights Act of 1964, which established Equal Employment Opportunity (EEO) as law, allows for affirmative action as a means of remediation for past discrimination. While the main focus of this legislation is the prohibition of discrimination in employment, the act also allows the courts, when finding that an employer is engaging in an unlawful employment practice, to "order such
8 See, for example, the list of bills presented to the 105th U.S. Congress, on the web site of the American Association for Affirmative Action (www.affirmativeaction.org). This website also provides a thorough list of state initiatives regarding affirmative action.
9 Conversely, Executive Order 11246, usually associated with affirmative action, has an explicit antidiscrimination component. The Review of Federal Affirmative Action Programs (Stephanopoulos and Edley 1996) reports that because the Office of Federal Contract Compliance Programs (OFCCP) is responsible for enforcement of both the affirmative action and nondiscrimination components of this order, contractors are often confused and incorrectly attribute requirements to hire a woman or minority worker to remedy past discrimination to affirmative action, rather than to antidiscrimination efforts.
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
Holzer and Neumark: Assessing Affirmative Action
487
affirmative action as may be appropriate, which may include reinstatement or hiring of employees. . . . " As indicated in some of the key court rulings summarized in Table 1, this provision of Title VII has served as the basis for court-ordered affirmative action plans, although the courts have gone back and forth on what is allowed or encouraged under Title VII.
In addition to the absence of a clear legal distinction between affirmative action and EEO legislation, in practice the difference is muddied further. Many employment discrimination cases concern hiring, and are based on evidence of "disparate impact," according to which underrepresentation of women or minorities—relative to some suitably-defined pool of job candidates—is sufficiently large to support an inference of discrimination. Much of the argument in such cases concerns the definition of the appropriate candidate pool (see, e.g., Epstein 1992, ch. 18). But regardless of how this issue is settled, it is obvious that employers concerned with a possible disparate impact discrimination claim will seek to ensure that women and minorities are adequately represented among their hires. Indeed, EEOC guidelines for defining disparate impact essentially establish a system of numerical yardsticks, embodied in the "80 percent" or "four-fifths" rule, which states that "a selection rate for any race, sex, or ethnic group which is less than four-fifths ... of the rate for the group with the highest rate will generally be regarded by the Federal enforcement agencies as evidence of adverse impact ..." (Code of Federal Regulations, Section 1607.4, 1998). This is easily monitored for many companies, as employers with 100 or more employees are required to file EEO-1 reports indicating the percentages of female and minority workers in broad occupa-
tional categories.10 Indeed, employment discrimination lawsuits can stem not only from complaints brought by private plaintiffs, but also from "Commissioner's Charges" brought by the EEOC following review of EEO-1 reports, without a single complainant (Bloch 1994). Although such complaints are relatively rare (e.g., United States Equal Employment Opportunity Commission 1994), there is indeed some evidence suggesting that EEO-1 reporting acts to deter discrimination. First, Bloch (1994) reports that while minority representation is generally highest among federal contractors, it is also generally higher among firms filing EEO-1 reports than in the labor force as a whole (Table 6.4, p. 102). Second, in an important case, EEOC v. O&G Spring and Wire Forms Specialty Company—which when reheard was in some ways a test case for the Wards Cove v. Atonio ruling that required plaintiffs not only to present evidence of statistical disparities but also to identify specific discriminatory employer practices— evidence based on EEO-1 reports figured prominently in the plaintiff's case (Bloch 1994, ch. 4).
The difficulty of separating equal opportunity from numerical yardsticks is also reflected in federal civilian hiring. In 1969, President Nixon issued an executive order requiring federal agencies to pursue equal employment opportunity for minorities and women, by the establishment of "Affirmative Employment Programs" (AEP's). Beginning in 1978, the EEOC had advisory authority for these plans, including reviewing and approving annual equal opportunity plans submitted by each federal agency,
10 Bloch (1994, p. 105) suggests that "noncon-tractors required to, file EEO-1 reports that are, monitored by the EEOC . . . would not be acting rationally if they were to avoid hiring minorities and women."
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
488       Journal of Economic Literature, Vol. XXXVIII (September 2000)
although the EEOC has no broad enforcement authority over these agencies. Initially, the EEOC focused on underrepresentation, requiring each agency to determine whether minorities or women were underrepresented in the agency and, if so, to set annual goals for rectifying these imbalances. Beginning in the late 1980's, the requirements to set goals in agencies' AEP's were dropped in place of greater emphasis on removing barriers to recruitment, hiring, and promotion of women and minorities. Since Adarand the requirements for such plans must be, among other things, more narrowly tailored, so that Justice Department guidelines for federal agencies suggest that an analysis of particular occupations must establish that representation is two standard deviations below that for comparable occupations in the civilian labor force in order for an agency to use an AEP; in contrast, prior to Adarand the requirements for establishing underrepresentation appear to have been neither as stringent nor as narrowly defined.11
Thus, in our view an analysis of affirmative action in the labor market should not be limited solely to the effects associated with contractor status, but ought to focus as well on policies or actions that might encourage anything other than race- or sex-blind behavior in the labor market.12 Using a broad
11 This information is based on the Clinton Administration's Review of Federal Affirmative Action Programs (Stephanopoulos and Edley 1996), and private communications with the staff of the Office of the Chief Economist of the Department of Labor.
12 Note that this broad working definition of affirmative action is quite similar to that adopted in the Clinton Administration's Review of Federal Affirmative Action Programs (Stephanopoulos and Edley 1996). The authors of that review defined affirmative action as "any effort taken to expand opportunity for women or racial, ethnic and national origin minorities by using membership in those groups that have been subject to discrimina-
working definition provides a more thorough analysis of the tapestry of policies that might be regarded as "affirmative action," and that might be affected by policies barring any form of preferential treatment based on race, sex, or other criteria. In addition, this broad definition clarifies our view that the position of some critics of affirmative action—that we can do away with affirmative action but maintain vigorous enforcement of antidiscrimination laws (i.e., "color blindness")—is to some extent based on an artificial distinction.13 Laws barring race- or sex-conscious behavior in hiring, promotions, etc., are likely to undermine not only explicit forms of affirmative action, but also any prohibitions of discrimination that rely on disparate impact analyses for their enforcement.14
2.2 Affirmative Action in Education and Contracting
There are no explicit federal policies regarding affirmative action in university admissions. However, universities have implemented affirmative action admissions policies that are widely regarded as giving preferential treatment to women and minority candidates. As outlined in Table 2, such admissions policies initially came under attack in the Bakke case, in which the Supreme Court declared that policies that set aside a specific number of places for minority students violated the Fourteenth
tion as a consideration ..." (footnote 1). Finis Welch (1981) takes a similarly broad (if not broader) view, using affirmative action to refer to the full apparatus of antidiscrimination policies, rulings, etc.
13 For example, Steele (1990) writes, "I would . . . like to see affirmative action go back to its original purpose of enforcing equal opportunity— a purpose that in itself disallows racial preferences" (p. 123). See also Stephen Carter (1991).
14 They would presumably still leave open "disparate treatment" cases, which rest on explicit instances of discriminatory behavior.
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
Holzer and Neumark: Assessing Affirmative Action
489
TABLE 2
Key Court Decisions and Referenda Regarding Affirmative Action in Universities
1978: Regents of the University of California v. Bakke
1995: Podberesky v. Kirwan
1996: California Proposition 209
1996: Hopwood v. State of Texas
Court agreed that special admissions program reserving spaces for minority students violated Title VII and 14th Amendment.
U.S. Supreme Court let stand a ruling of the U.S. 4th Circuit Court of Appeals that the Banneker scholarship program at the University of Maryland violated the 14th Amendment.
Prohibited discrimination or preferential treatment in public employment, public education, or public contracting on the basis of race, sex, color, ethnicity, or national origin.
U.S. 5th Circuit Court of Appeals ruled that an affirmative action plan at the University of Texas Law School that admitted some minority students with lower grade-point averages and test scores than white applicants who were not admitted violated the 14th Amendment.
Sources: Bloch 1994; Edley 1996; Radford 1997; Civil Rights Monitor Leadership Conference on Civil Rights Online Center.
Amendment of the U.S. Constitution, which bars states from depriving citizens of equal protection of the laws. As a consequence, the courts have ruled that the Fourteenth Amendment applies to affirmative action programs at state and local levels of government. However, while this decision is viewed as declaring strict quotas illegal, it is also interpreted as ruling that race can be used as a "flexible factor" in university admissions (Edley 1996, p. 58).15 Affir-mative action in university admissions most recently suffered serious setbacks as a result of Proposition 209 in California, and a court ruling against the University of Texas Law School in Hopwood v. State of Texas. The latter addresses an admissions program that granted preferential treatment to minorities in the evaluation of grade point averages and test scores. Proposition 209 addressed preferential treatment in any program, although most of the at-
15 Specifically, the decision allowed admissions officers to "take race into account" as a means to secure the educational benefits of a student body with diverse backgrounds and experience (William Bowen and Derek Bok 1998, p. 8).
tention has focused on its impact on university admissions.
In addition to admissions procedures, financial assistance programs may give preferential treatment to particular minority groups. A well-known example was the Banneker program for black students at the University of Maryland, which ran parallel to another colorblind program, but with lower scoring thresholds (Edley 1996). In Podberesky v. Kirwan, the Supreme Court let stand a Circuit Court ruling that this program violated the Fourteenth Amendment. However, the Circuit Court ruling clarified that such programs were not necessarily unconstitutional. To pass muster, two criteria had to be met. First, a program must be based on a "compelling government interest," in this case a remedy for past discrimination. Second, it must be narrowly tailored to the specific problem of past discrimination it addresses. These two criteria were established as "strict scrutiny" for race-based state and local affirmative action measures, in a 1989 case regarding a contracting program (City of Richmond v. J. A. Croson Co.).
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
490       Journal of Economic Literature, Vol. XXXVIII (September 2000)
Aside from scholarships, other programs also seek to increase representation of women or minorities through incentives for higher education. A partial list of these includes Department of Education programs to encourage minority students to become teachers, Health and Human Services programs to encourage minorities to enter the health professions, National Science Foundation programs targeting both minorities and women, and federal aid to historically black colleges and universities (Stephanopoulos and Edley 1995). These programs may come under scrutiny in the future. As federal programs they are not subject to the Fourteenth Amendment, but they could be reviewed in light of the Fifth Amendment, which guarantees that citizens shall not "be deprived of life, liberty, or property, without due process of law." As explained below, the courts have relatively recently applied the Fifth Amendment to federal contracting and procurement programs. Even more speculative is the question of whether affirmative action programs at private universities will come under review. A case based on the fact that most universities have federal contracts is conceivable, while a more extreme possibility is revocation of tax-exempt status by the IRS, as occurred in the face of racially discriminatory policies at Bob Jones University and Goldsboro Christian Schools (Bob Jones University v. United States).
The third major component of affirmative action is contracting and procurement programs. At the federal level, these have principally taken the form of preferential treatment in bidding (such as sole-source contracting for small projects (the Section 8(a) program), "bid price preferences," and "rule-of-two set-asides") for Small/Disadvantaged Businesses (SDBs), and Small Business Administration (SBA) programs of tech-
nical assistance.16 These contracting and procurement programs focus more on minorities than women.17 Some have been dropped or modified in response to court decisions discussed below. In addition to the federal level, numerous states and localities have used programs aimed at increasing the share of contracts awarded to minority-owned businesses.
As summarized in Table 3, court rulings in the last decade or so have challenged the legal standing of such programs. City of Richmond v. J. A. Croson Co. established strict criteria ("strict scrutiny") that must be met for state programs to be legal. However, because the Fourteenth Amendment applies to state and local government policies, court rulings restricting federal programs (Fullilove v. Klutznick and Metro Broadcasting, Inc. v. FCC) initially applied less strict standards ("intermediate scrutiny"). However, in Adarand Constructors, Inc. v. Pena, the Supreme Court ruled that strict scrutiny could also apply to federal programs. The Court ruled that federal race-conscious programs will be evaluated on the basis
16 Bid price preferences allow contracting officers to add a specified amount to non-SDB bids (typically 10 percent at the federal level) and then to award the contract based on the adjusted bids. Rule-of-two set-asides allow contracting officers to "limit bidding on a particular contract to . . . SDB's if two or more such firms are potential bidders and the officer determines the prevailing bid will likely be within 10 percent of the fair market price" (Stephanopoulos and Edley 1995, Section 9).
17 The Review of Federal Affirmative Action Programs (Stephanopoulos and Edley 1995, Section 9) states that eligibility for these programs "is targeted to minority-owned businesses (and in some cases nonminority women-owned businesses), but by statute available more broadly to "socially and economically disadvantaged" individuals." The targeting toward minorities is implemented via definitions of these terms, in particular because "By statute, persons from certain racial and ethnic groups—but not women—are presumed to be socially disadvantaged." On the other hand, the Review suggests that for some agencies (notably, the Department of Transportation), SDB programs make all women eligible.
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
Holzer and Neumark: Assessing Affirmative Action
491
TABLE 3
Key Court Decisions Regarding Affirmative Action in Contracting
1980: Fullilove v. Klutznick
1986: City of Richmond v. J. A. Croson Co.
1990: Metro Broadcasting, Inc. v. FCC
1995: Adarand Constructors, Inc. v. Pena
U.S. Supreme Court upheld provision in federal Surface Transportation Act setting goal of 10 percent of contract dollars for "disadvantaged business enterprises" (principally minority-owned). Applied "intermediate scrutiny" to federal race-based affirmative action programs, requiring that they serve "an important governmental interest" and be "substantially related" to that interest.
U.S. Supreme Court applied "strict scrutiny" in striking down a city ordinance establishing a 30 percent target for the proportion of city contracts awarded to minority businesses. The strict scrutiny criteria established were a "compelling government interest" (remedial response to past discrimination and its lasting effects) and that a program be "narrowly tailored" to achieve that interest.
U.S. Supreme Court upheld congressional measures to increase minority ownership of broadcast licenses, on the grounds of advantages of diversity in viewpoints expressed, rather than remediation for discrimination.
U.S. Supreme Court overturned Fullilove and Metro Broadcasting in ruling that strict scrutiny apply to congressionally-authorized race-based programs, in this case a Department of Transportation program encouraging contractors to sub-contract with firms owned by "socially and economically disadvantaged" individuals.
Sources: Bloch 1994; Edley 1996; Radford 1997; Civil Rights Monitor Leadership Conference on Civil Rights Online Center.
of the Fifth Amendment, with much the same interpretation with which the Court had applied the Fourteenth Amendment to non-federal programs. These various rulings do not prohibit race-based programs at any level of government, although they do raise the standards for their legal justification. It is too early to be able to characterize how local, state, and federal programs will ultimately be shaped in response to these rulings.
2.3 Variation in Affirmative Action
Aside from the distinction between federal vs. state and local programs, there are other important sources of variation in affirmative action. First, affirmative action can arise not just in the public sector, but in the private sector. Second, affirmative action may be used voluntarily or involuntarily. As noted
above, involuntary affirmative action plans may be implemented in the private sector as a court-ordered remedy for past discrimination. Such involuntary plans are authorized by Section 706(g)(1) of Title VII of the Civil Rights Act, which allows a court to "order such affirmative action as may be appropriate, which may include reinstatement or hiring of employees, . . . " (italics added).18 Voluntary affirmative action plans may be initiated by employers in order to accomplish some other goal, such as avoiding EEOC violations or becoming eligible for federal
18 However, such plans are limited by language in Section 703(j) indicating that Title VII shall not be interpreted as requiring any employer "to grant preferential treatment to any individual or to any group" on the basis of their race, sex, etc. Clearly these sections of Title VII leave some ambiguities as to precisely what type of affirmative action is permissible.
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
492       Journal of Economic Literature, Vol. XXXVIII (September 2000)
contracts. Alternatively, an employer may perceive direct benefits from increased hiring of underrepresented groups, perhaps because it broadens a company's appeal to its customers (Radford 1997). As indicated in the description of United Steelworkers of America v. Weber, in Table 1, the Supreme Court has attempted to establish criteria under which voluntary affirmative action plans are permissible under Title VII; note that Title VII regulates private affirmative action, in contrast to the application of the Fourteenth Amendment to public affirmative action.
Variation also exists because affirmative action goals may be pursued through a number of different channels. For example, in the labor market affirmative action may influence the "pre-hiring" stages of recruitment and screening, hiring itself, or "post-hiring" behavior such as training or promotion. Similarly, in a university setting affirmative action may affect recruiting, admissions, or remedial education. As an example, drawing on some of our own research (Holzer and Neumark 1999, 2000), we find that employers using affirmative action tend to hire minorities who are less qualified in terms of some readily observable measure of qualifications such as education; this suggests that, as critics of affirmative action contend, it leads to hiring of less-qualified minority workers. Yet we also find that employers using affirmative action recruit more extensively and use more intensive screening of job candidates, which raises the possibility that employers are uncovering other information about minority workers that offsets their lower educational qualifications. Thus, if we do not try to look at the gamut of behaviors or outcomes that affirmative action might influence, we may be led to incorrect assessments about its effects.
2.4 An Uncertain Future
The final difficulty that arises in defining and assessing affirmative action is that it is currently in flux. In addition to uncertainty over the effects that policies and rulings such as Proposition 209, Initiative 200, and Adarand will have on the shape of affirmative action programs, additional referenda, legislation, and court cases loom in many states.19 These all raise questions about the ways in which affirmative action might be circumscribed in the future. There is also uncertainty about how existing policies might be modified to achieve some of the goals of affirmative action while remaining within the parameters of the law as interpreted by the Supreme Court. For example, in response to the Hopwood decision, in 1997 the Texas state legislature passed a bill to admit all students in the top ten percent of their high school class, regardless of their test scores. One critical question is the extent to which such policies will still target groups that may have received preferential treatment in the past. As Bowen and Bok (1998) point out, such a policy will do less for blacks, in particular, because they are only "half as likely as whites to finish in the top 10 percent of . . . [their] high school class" (p. 272). On the other hand, minority students at highly segregated schools would likely see their
19 Two leading pending court cases are Gratz and Hammacher/Grutter v. The Regents of the University of Michigan, filed in 1997, which challenges undergraduate admissions procedures at that university, alleging unlawful preference to minorities, and Smith v. University of Washington Law School, also filed in 1997, which challenges affirmative action in admissions. As another example, a bill was recently introduced in the Michigan House to amend the state constitution to ban any preferential treatment based on race, sex, religion, etc., in employment, public education, or public contracting; similar bills are being considered in other states (see the Civil Rights Monitor Leadership Conference on Civil Rights Online Center).
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
Holzer and Neumark: Assessing Affirmative Action
493
chances of admission increased under such a proposal.
3. Does Affirmative Action "Level the Playing Field?"
In this section, before trying to assess the efficiency and performance effects of affirmative action, we consider the prior question of whether affirmative action is likely to "level the playing field." Two issues arise in attempting to answer this question. The first is whether significant discrimination against minorities and females persists, in which case it is more likely (although not a given) that affirmative action can help to level the playing field by countering discrimination, rather than generating reverse discrimination. We therefore begin by providing our assessment of the evidence on discrimination. We also discuss more general issues of unequal economic access, which might be thought of as a form of "societal discrimination," and specifically unequal opportunities in education and en-trepreneurship. One could further argue that even if the case for discrimination against minorities and women is not compelling, the absence of evidence of reverse discrimination undermines the argument that affirmative action (and the current set of EEO policies that do not fall under this heading) generate unfair preferential treatment of minorities and women. Rather, it may be more plausible to conclude that these policies have produced a "level" playing field, or at worst do no harm.
The second issue—assuming that discrimination persists (which is our assessment)—is whether affirmative action is likely to be a helpful policy. In theory, the "redistribution" effects are clear, and in the latter part of this section we review evidence on shifts in employment, admissions, etc., generated by affirmative action.
3.1 Is the Playing Field Level? Evidence on Labor Market Discrimination
The primary approaches to testing for race or sex discrimination in the labor market include: 1) wage regression tests; 2) audit studies; 3) direct evidence on the relative wages and productivities of different demographic groups; and 4) direct evidence on employer characteristics and behavior.20 Below, we review evidence from each of these approaches, and offer our reasons for concluding that the overall evidence points to some continuing discrimination against women and blacks.
Before doing so, though, one issue that must be addressed in an economic analysis of discrimination is whether, as originally suggested by Gary Becker (1971), market competition precludes discrimination. Although some researchers (e.g., Victor Fuchs 1988; June O'Neill 1994) have used Becker's framework to dismiss the possibility that race or sex differences in labor markets could reflect discrimination, Becker (in Chapter 3) was careful to specify the conditions under which his "market competition" result holds, and conversely those under which it does not hold. In particular, if product markets are perfectly competitive, and there are sufficient potential employers with nondiscriminatory tastes, then discrimination will disappear over time through competition in product markets. Alternatively, if product markets are not perfectly competitive but some entry is possible, and there are non-decreasing returns to scale and at least one potential employer with nondiscriminatory tastes, then discrimination
20 Evidence on labor market discrimination against minorities and women has recently been reviewed quite extensively (Francine Blau 1998; Darity ancf Mason 1998; Heckman 1998; Joseph Altonji and Rebecca Blank 1999). Here we provide a shorter summary of this evidence than is provided in these reviews.
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
494       Journal of Economic Literature, Vol. XXXVIII (September 2000)
will be competed away. In the case of total barriers to entry in product markets, even with decreasing returns to scale, if businesses are transferable and there is a potential nondiscriminating employer, discrimination may disappear through competition in the market for firms. Finally, if employers' tastes are nepotistic rather than discriminatory, then the "discriminatory" wage gap will not be eliminated by competition in the market for firms, although product market competition may still suffice.
Thus, the Becker model does not unambiguously indicate that discrimination cannot persist. This conclusion is reinforced by the development of other theoretical models with taste discrimination on the part of some agents in which discrimination can persist. Dan Black (1995) offers a search model with wage differentials stemming from discriminatory tastes in equilibrium. Lawrence Kahn (1991) shows that customer discrimination can persist and create a discriminatory wage differential. Aside from these theoretical considerations, there is very little empirical evidence on whether market competition roots out discrimination (see Judith Hellerstein, Neumark and Kenneth Troske 1997). Thus, while this argument has to be taken seriously, there are neither theoretical nor empirical grounds for dismissing evidence of discrimination out of hand.
3.1.1 Earnings/Employment Regressions
The traditional approach to the analysis of race/sex discrimination has been the estimation of regressions of the log of wages or earnings on observable proxies for productivity that are not themselves, at least in principle, attributable to race or sex.21 "Residual" dis-
21 As an example of this latter issue, the "feedback" hypothesis argues that some productivity-related characteristics are on average lower for
crimination is then estimated either as the coefficients on dummy variables for race or sex, or from decompositions of overall race or sex differences into those based on differences in observable characteristics versus those based on differences in coefficients, with the latter interpreted as measuring discrimination (e.g., Ronald Oaxaca 1973; Neumark 1988).
The literature on race and sex differences based on this residual method is huge, but a sense of the magnitudes of the estimates can be obtained from a few recent, high-quality papers in this area; without providing a more extensive review, we would suggest that earlier research is by and large consistent with the evidence reported in these papers. Using data from the National Longitudinal Survey of Youth (NLSY) for 1987, covering men aged 22-29, O'Neill (1990) reports that the black/white wage ratio is 0.83. Estimating separate regressions for blacks and whites, and controlling for region, schooling, and potential work experience, the ratio rises to a range of 0.852-0.877 (depending on whether the wage regression for blacks or whites is used to weight characteristics), suggesting that most of the wage gap, a residual differential of about 12-15 percent, is unaccounted for by human capital differences. However, this is hardly a complete set of human capital controls available in micro-data sets. Adding industry controls and a skill index defined at the level of 3-digit occupations, the wage ratios rise to 0.887-0.912. This evidence points to race gaps in wages that are on the order of perhaps 15-20 percent to begin with, and fall to about ten percent once human capital controls are added.
women than for men because of past or anticipated sex discrimination (Blau and Marianne Fer-ber 1991; Reuben Gronau 1988; Neumark and Michele McLennan 1995).
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
Holzer and Neumark: Assessing Affirmative Action
495
Turning to residual sex gaps in wages, a survey of estimates of this residual is presented as part of a meta-analytic study by T. D. Stanley and Stephen Jar-rell (1998). In the sample of studies they consider, the average residual log wage gap ranges from 0.276 to 0.29, depending on the precise sample of studies used. Using PSID data from 1979-88, Blau and Kahn (1997) report raw sex wage gaps, and residual gaps first controlling only for human capital variables, and then also for industry, occupation, and collective bargaining status. In these data, the mean log wage gap fell from 0.475 to 0.323 between 1979 and 1988. In 1979, this gap falls to 0.335 adding the human capital variables, and 0.254 adding the remaining variables. In 1988, the gap declines to 0.217 adding the human capital variables, and 0.126 adding the other variables. In terms of wage ratios, these numbers imply an unadjusted ratio of 0.622 in 1979, and a final adjusted ratio of 0.776. In contrast, the corresponding figures for 1988 are 0.724 and 0.882. Thus, the raw wage gaps between men and women are much larger than those between whites and blacks, and the remaining residual wage gaps are on the order of ten percent or more.
Of course, any residual race/sex differences can be interpreted as discrimination or as unobserved productivity that is correlated with race/sex. Most labor economists presume that certain kinds of unobserved skills are lower among women and especially minorities, and that their omission leads us to overestimate race and sex differences in the market (e.g., Becker 1985). This motivates the inclusion of variables that better control for these skills, to see whether residual race/sex differences are eliminated; if so, we would be inclined to conclude that labor market discrimination   against   the relevant
groups does not exist. In fact, evidence from several recent studies indicates that residual wage differentials nearly disappear for some groups of minorities and women when we include controls for previously unmeasured skills. For instance, inclusion of Armed Forces Qualifications Test (AFQT) scores in log wage equations can account for much of the residual black-white difference in the National Longitudinal Survey of Youth (O'Neill 1990; Nan Maxwell 1994; Derek Neal and William Johnson 1996), while controls for language ability and education largely account for residual wage differences between whites and Mexican-Americans (Stephen Trejo 1997).
On the other hand, if the direction of the bias due to other omitted variables is indeterminate, or some of these additional control variables are invalid, then the inference of little remaining discrimination from these results is not necessarily warranted (Darity and Mason 1998).22 Also, although some of the studies listed above indicate that education and cognitive skills account for large percentages of racial/ethnic differences in wages, it would be premature to infer that discrimination against these groups (especially blacks) no longer exists. Significant racial differences can still be found for earnings in other data sets (e.g., Richard Murnane, John Wil-lett, and Frank Levy 1995), and for employment even controlling for test scores (Neal and Johnson 1998). Even in the NLSY data, there are significant
22 For instance, Darity and Mason (1998) claim that when Arthur Goldsmith, Jonathan Veum, and Darity (1997) include some self-reported psychological variables in an earnings equation that also includes AFQT, the evidence "restores a negative effect on wages of being African-American" (p. 75). However, the authors do not present estimates for their sample excluding these psychological variables, so it is not clear what the basis is for claiming that including the psychological variables results in a sharper race difference in wages.
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
496       Journal of Economic Literature, Vol. XXXVIII (September 2000)
differences across race-sex groups in the returns to components of the AFQT (William Rodgers and William Spriggs 1996a; John Cawley et al. 1996) or other measures of skill, suggesting that the labor market still rewards comparably-skilled individuals differently.23 Yet another issue that arises is whether there is racial bias in these test scores; we are not convinced that there is decisive evidence of bias, but do not necessarily believe that test score differentials reflect only productivity.24
Regarding male-female differences in the labor market, the "gender gap" in wages clearly narrowed in the 1980s but remained significant even after controlling for experience and job tenure (Blau 1998). While significant sex differences in educational attainment no longer exist, there is evidence of persistent sex differences in math test scores (Altonji and Blank 1999) and in high school/college curricula (Charles Brown and Mary Corcoran 1997), although these do not seem to account for remaining sex differences in earnings, especially among the less-educated. Also, Jane Waldfogel (1998) has recently shown that the most significant sex gap in pay now exists be-
23Bernt Bratsberg and Dek Terrell (1998) report higher returns to experience for whites, but comparable or higher returns to tenure for blacks. The extent to which racial differences in returns to experience are consistent with models of statistical discrimination is considered in Altonji and Charles Pierret (1997) and Gerald Oettinger (1996).
24 For instance, Rodgers and Spriggs (1996a) argue that the AFQT is racially biased, based on differences in regression equations between whites and blacks that are used to account for these scores. But the independent variables used in their equations are quite sparse, and the "racially-neutral" predicted AFQT score that they generate contains only a small fraction of the variance in the original variable. See Christopher Jencks and Meredith Phillips (1998) and John Campbell (1996) for arguments against the notion that test scores are racially biased as predictors of job performance or earnings; in general, test scores have roughly comparable predictive power across racial groups.
tween men and women with children. The extent to which this gap reflects discrimination as opposed to differences in unobserved skills, preferences, or job characteristics between mothers and fathers remains unclear.
3.1.2 Audit Studies
The "audit" methodology has become a popular means of testing for discrimination in housing and labor markets. In audit studies, researchers send matched pairs of individuals with similar education and experience but of different races/sexes to employers, landlords, realtors, etc. Assuming that the auditor pairs are well-matched, any resulting differences in treatment between race/sex groups should be attributable to discrimination rather than heterogeneity in skills or behavior that is correlated with race or sex. The labor market audit studies (e.g., Marc Bendick, Charles Jackson, and Victor Reinoso 1994; Michael Fix and Raymond Struyk 1994; Genevieve Kenney and Douglas Wissoker 1994; Neumark 1996; Claudia Goldin and Cecilia Rouse, forthcoming) generally indicate that significantly fewer minorities or females obtain job offers than white males. The magnitudes of the net differences favoring whites or males in the probabilities of receiving job offers generally range from 5-20 percentage points. Because in virtually all of these studies most job candidates do not get offered a job, the estimated net differences are generally small relative to the pool of all applicants, but large as a percentage of those who are actually getting offers.
Heckman (1998) has criticized many of these audit studies on a wide range of empirical and conceptual grounds. The empirically-based criticisms include the fact that the job openings are not randomly generated (e.g., they are often based on newspaper ads, which
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
Holzer and Neumark: Assessing Affirmative Action
497
account for small fractions of overall hiring) and that the empirical magnitudes of the biases generated are, according to Heckman, quite small. The conceptual criticism of these studies is that whites and minorities may continue to differ, on average, with respect to characteristics that remain unobserved but that employers anticipate.
This raises the question of just what we mean by discrimination. If we think in terms of the empirical literature on discrimination, or its legal treatment, discrimination occurs when, for example, men and women are treated differently without any basis for this in the productivity-related characteristics that ■wje can observe. However, we can think of characteristics that an employer might "know" to distinguish between women and men, and therefore provide a basis for differential treatment, even though this cannot be documented for a particular set of applicants or workers. A prime example might be expected turnover, which might be higher on average for women even though it cannot, of course, be documented to be true for a set of women in, say, a particular applicant pool (because it is something that happens in the future). Such a factor could conceivably influence employer hiring decisions in audit studies for nondiscriminatory reasons. Conceptually, if the employer is right about such characteristics, and differences in these characteristics can explain the differential treatment, then as economists we might want to rethink labeling the employer's behavior as discriminatory. On the other hand, from a legal perspective the answer seems relatively clear; such statistical discrimination is illegal. This is made most explicit with respect to sex, where EEOC guidelines define the following as illegal discrimination: "The refusal to hire an individual based on assumptions of the com-
parative employment characteristics of women in general. For example, the assumption that the turnover rate among women is higher than among men" (Code of Federal Regulations, Section 1604.2, 1998). With respect to race, it is widely accepted that the EEOC's regulations concerning employee selection largely rule out the use of statistical discrimination (Epstein 1992, ch. 2).
Heckman also argues that, even if valid, audit studies tell us only about individual rather than market discrimination, by average rather than marginal employers. This argument relies heavily on Becker's model of a labor market in which employers have varying degrees of discriminatory tastes. The point is that minorities, for example, may avoid the most discriminatory employers where they would earn less (or not be hired), but still be able to find employment at enough nondiscriminatory employers such that at the margin they can avoid discrimination. This argument is clearly stronger the more likely it is that discriminatory employers are driven from the market, an issue discussed above; otherwise the marginal employer may well be a discriminator. Also, in other models, such as Black's (1995) search model with discriminatory tastes, the presence of any discriminatory employers creates a market wage gap. Finally, Heckman's critique also hinges on whether any other barriers or disadvantages prevent minorities/women from gaining jobs with nondiscriminatory employers. We discuss some of these barriers or disadvantages below.
3.1.3 Direct Evidence on Relative Productivities and Wages
To date, the paucity of data on worker productivity by race or sex has made direct tests of discrimination with large data sets very rare, except for a limited number of studies of athletes
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
498       Journal of Economic Literature, Vol. XXXVIII (September 2000)
(e.g., Kahn and Derek Sherer 1988).25 But large data sets that match workers and their establishments have recently become available that enable researchers to directly compare relative worker productivities and earnings across race and sex groups. To date, Hel-lerstein, Neumark, and Troske (1999) have provided evidence of lower relative wages among female workers in manufacturing, but comparable productivities between them and male workers, in data for the U.S. In contrast, they find little evidence of significant differences in earnings or productivity by race in manufacturing. The absence of a race differential in earnings contrasts with individual-level wage regression estimates, and stems in large part from the segregation of blacks in higher-pay plants (in which they earn lower wages than whites, but higher wages than they would earn elsewhere); see also William Carrington and Troske (1998). For this study, though, the important parameter is the difference between the pay gap and the productivity gap, the estimate of which is less likely to be biased because of this segregation, and hence the results still suggest no discrimination against blacks.26
3.1.4 Evidence on Employer
Characteristics and Behavior
Evidence on employer characteristics and behavior is likely to yield additional
25 Kahn and Sherer document a salary gap (of 20 percent) between white and black NBA players, controlling for performance, which could reflect employer, co-worker, or customer discrimination. Clark Nardinelli and Curtis Simon (1990) test for customer discrimination by examining the prices of baseball cards of whites and non-whites, conditional on performance, finding a 10-13 percent shortfall for non-whites.
26 Similar estimations for manufacturing establishments in Israel (Hellerstein and Neumark 1999) do not reflect wage discrimination against women. Leonard (1984a) also estimated productivity differentials by race and sex, although using
information of two varieties. First, data on hiring out of applicant pools potentially tell us more about the demand side of the market from which discrimination emanates. Second, variation in behavior toward different race/sex groups based on employer characteristics can provide evidence on specific hypotheses regarding discrimination. However, few studies have presented evidence on employers, due to data limitations.
Holzer (1996) uses data from a recent survey of 3000 establishments in four large metropolitan areas to provide evidence on employer willingness to hire applicants from different race/sex groups into jobs that differ by skill needs, location, compensation, etc. Unfortunately, this survey does not include data on the quality of the individual applicants, although there is information on relative skills of the various population groups by geographic area. The evidence in this study indicates that Hispanic applicants are generally hired at higher rates than blacks, while among blacks, female applicants are hired at higher rates than males; these differences do not appear to be accounted for by differences in relative skills or in the characteristics of jobs for which each demographic group applies.27
Other research looking at employer characteristics uncovers evidence potentially consistent with discrimination.
state-by-industry data rather than plant-level data; he found lower productivity associated with non-white male employees.
27 These hiring patterns are quite consistent with more qualitative evidence on employer behavior from interviews or ethnographic studies (e.g., Joleen Kirschenman 1991; Kirschenman and Katheryn Neckerman 1991; Phillip Moss and Chris Tilly 1995). At the same time, these studies point to at least perceived differences among some of these groups; for example, employers seem to prefer other ethnic groups, especially immigrants, over native-born blacks because they perceive a stronger work ethic and better attitudes among the former.
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
Holzer and Neumark: Assessing Affirmative Action
499
Holzer (1998) finds that small establishments hire blacks at lower rates than do larger establishments; greater discrimination in hiring among small employers is consistent with the lower likelihood that small employers are covered by EEO law or monitored for affirmative action compliance. Kenneth Chay (1998) reports further evidence consistent with this hypothesis, finding that the relative employment and pay of blacks appear to have been boosted at small establishments that came under EEO law with the Equal Employment Opportunity Act of 1972. While consistent with discrimination on the part of small employers, such evidence is also consistent with reverse discrimination on the part of large employers covered (earlier, in the case of the Chay paper) by EEO or affirmative action. Thus, large employers are not necessarily an appropriate standard of "neutrality" with which to compare the behavior of small employers. Also, Holzer and Keith Ihlanfeldt (1998) find that the race of the customers at an establishment has significant effects on the race of workers hired into jobs that involve direct contact with these customers, and on the wages that they are paid (see also Ihlanfeldt and Madelyn Young 1994), consistent with customer discrimination. Finally, Stephen Raphael, Michael Stoll, and Holzer (1998) show that black owners/managers hire more black employees than white owners/managers in similar locations.
3.1.5 Summary
Taken together, the various studies summarized above suggest that, while differences in educational attainment and cognitive skills account for large fractions of racial differences in wages, employer discrimination continues to play a role in generating different labor market outcomes by race and sex. While
some studies of race differences can account for most or all of the entire race gap in wages, this is not true of all studies, and is certainly not true of studies of residual sex gaps in wages.
It is true that one can construct arguments explaining away specific types of evidence as discrimination, and as new types of evidence have been brought to bear in response to criticisms of early evidence (e.g., the introduction of audit studies in studying labor market discrimination), such arguments have arisen in response (e.g., Heckman's critique of audit studies). However, many of the empirical assumptions underlying these arguments are themselves untested, and it is our view that increasingly subtle arguments are needed to explain away evidence consistent with discrimination as newer, more reliable evidence is obtained in response to earlier criticisms. In contrast, a uniform, relatively simple behavior—discrimination—can explain much of both the older and newer evidence. While this does not necessarily imply that discrimination is the explanation of the evidence accumulated thus far, these considerations make us far more inclined to reach this conclusion.
3.2 Other Sources of Disadvantage in Labor Markets: "Societal Discrimination"
Aside from the potential effects of affirmative action in countering discrimination, some other arguments in favor of affirmative action—in particular, perhaps, with respect to university admissions—may also be strengthened by the presence of other disadvantages that limit the abilities of minorities to develop their human capital and hence to compete on an equal footing in the labor market. For instance, Janet Currie and Duncan Thomas (1995), Neal and Johnson   (1996),   and   Rodgers and
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
500
Journal of Economic Literature, Vol. XXXVIII (September 2000)
Spriggs (1996a) document that family background and neighborhood characteristics adversely affect the test scores obtained by young blacks, while George Borjas (1995), Katherine O'Regan and John Quigley (1996), and Ingrid Ellen and Margery Turner (1997) report similar effects on employment and earnings of young urban blacks. Other evidence documents race differences in school resources. For example, John Kain and Kenneth Singleton (1996) show that per-pupil expenditures and other measures of school quality remain lower among black students than white students in Texas, although the effects of school resources on educational outcomes are still heavily debated. The role of residential racial segregation in limiting the educational and employment opportunities of blacks has been clearly demonstrated by David Cutler and Edward Glaeser (199 7).28 This segregation appears to reflect the effects of past labor market discrimination against blacks and other minorities, as well as past and current discrimination in housing and credit markets.29
Thus, a case can be made that the inferior labor market outcomes of minorities and women reflect labor market discrimination, while minorities also face a variety of other disadvantages and barriers. Of course, women and men grow up on average in similar socioeconomic conditions, so a case for other "societal" disadvantages faced by
28 One effect of this segregation is "spatial mismatch" between suburban employers and inner-city workers (see reviews in Jencks and Susan Mayer 1990; Holzer 1991; Kain 1991; Ihlanfeldt and David Sjoquist 1998).
29 Evidence of continuing discrimination in housing and mortgage markets is reviewed in John Yinger (1998) and Helen Ladd (1998). For evidence that existing racial segregation cannot be fully accounted for by differences in group incomes or by the preferences of black residents themselves see Reynolds Farley, Elaine Fielding, and Maria Krysan (1997).
women would have to be based on other factors (such as role models, responsibility for children, etc.). We are not aware of evidence on role-model effects on female labor market outcomes, but there is ample literature at least consistent with child-care responsibilities adversely affecting women's labor market outcomes (e.g., Waldfogel 1998; Sanders Korenman and Neumark 1992). Based on these considerations, it seems a reasonable conclusion from all of the evidence that the "playing field" in the labor market is not level across the various groups. This does not necessarily imply that affirmative action is the best policy response, but in our view it provides a prima facie case for a serious consideration of affirmative action as a policy tool to address race and sex differences in labor market outcomes.
3.3 Discrimination in Education and Entrepreneurship
Prior to the civil rights movement there was clear discrimination against blacks in Southern schools, and at a minimum very low representation in Northern schools (Bowen and Bok 1998; Stephan Thernstrom and Abigail Thernstrom 199 7).30 Historically-black colleges and universities were created as a part of a system of segregated universities, or to provide educational opportunities to black students given their exclusion from other universities, especially in the South. Per student expenditures at
30 One might argue that preferential treatment of "legacy" students to some extent perpetuates a history of preferential treatment of whites; see John Larew (1991). Bowen and Bok (1998), for a set of three selective schools, find that admissions rates for legacy students are about the same as those for black students, although the legacy students have stronger academic records. Bernard Lentz and David Laband (1989) find that sons of doctors get accepted to medical schools at higher rates, controlling for a wide variety of qualifications and traits, which they suggest may reflect favoritism toward legacies (p. 408).
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
Holzer and Neumark: Assessing Affirmative Action
501
these colleges and universities were historically lower than at other schools, and some spending gaps persist.31 On the other hand, evidence on returns to college education suggests that black students enrolling in the 1970s were not disadvantaged—if anything, the opposite—by attending historically-black schools (Jill Constantine 1995; Ronald Ehrenberg and Donna Rothstein 1994), although this may reflect poor education of blacks at mixed colleges. In addition, it was relatively common for undergraduate or professional schools to exclude women in the earlier part of the century, but very uncommon by the end of World War II (e.g., Stephen Cole 1986; Ohechukwu Oko 1996). Thus, taking a very long view, one could argue that women and minorities did not face a level playing field in higher education. This view is reinforced by the evidence cited in the previous subsection on lower quality of elementary and secondary education for minorities, which can disadvantage minorities in the competition for admission to the better universities.32
The rapid rise in enrollments of minorities and women in professional schools suggests that something must have changed, whether a reduction in discrimination or implementation of affirmative action. However, it is not entirely clear what to make of changes in enrollments in professional schools once formal barriers to admission of women or minorities were dropped. In particular, Cole (1986) finds no evi-
31 The National Center for Education Statistics reports that as of 1994 the per student gap was 12 percent at public institutions, and 14 percent at private institutions.
32 A salient example is the University of California's addition of extra grade point credits to high school students who complete advanced placement courses, which are apparently much more likely to be offered in affluent school districts with fewer minorities (Dan Walters 1999).
dence that acceptance rates into medical school were lower for women than for men from the 1940s to the 1980s. In general, qualifications (MCAT scores) and acceptance rates have been very similar for men and women. The rapid rise in the 1970s of the percentage of medical students who were female was accompanied by a parallel rapid increase in application rates. As the author points out, though, nothing in his analysis rules out the hypothesis that lower application rates of women stemmed from discrimination either prior to or after medical school.
Turning to the present and the more recent past, there is relatively little work that investigates or finds evidence of current discrimination against women or minorities in university admissions. One example is Kathy Cannings, Claude Montmarquette, and Sophie Mahsered-jian (1996), who find no evidence that an applicant's sex affected the probability of medical school admission at the University of Montreal medical school in data from 1987. In a sample from 1979, Lentz and Laband (1989) find evidence consistent with slight discrimination against women and in favor of blacks and Hispanics relative to whites and Asians in medical school admissions. Moreover, there is rather convincing evidence that universities now treat minorities preferentially in admissions decisions, as discussed in Section 3.4.2. Thus, in contrast to the labor market—where in our view discrimination persists even in the face of affirmative action, and would likely worsen in its absence—we cannot conclude that in the absence of affirmative action in education women and minorities would be treated unfairly (although we also cannot rule this out).
Like research on labor market discrimination, research on discrimination affecting women- and minority-owned
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
502       Journal of Economic Literature, Vol. XXXVIII (September 2000)
business formation and success is driven by large disparities in the incidence and size of such businesses, with the share of black and Hispanic business ownership and receipts much below their representation in the population.33 In Fullilove, Chief Justice Burger cited numerous factors holding down minority entrepreneurship generally, and government business in particular, including "deficiencies in working capital, inability to meet bonding requirements, disabilities caused by an inadequate "track record," lack of awareness of bidding opportunities, unfamiliarity with bidding procedures, preselection before the formal advertising process, and the exercise of discretion by government procurement officers to disfavor minority businesses" (cited in Stephanopoulos and Edley 1995). Of course, only the last of these sounds like outright discrimination, and direct discrimination by the government at that.
Burger did not discuss discrimination in lending to blacks. There is less research on business lending than mortgage lending, but most of it points in the same direction, suggesting that capital market discrimination is partly responsible for some of the other disadvantages that Burger cited. In a study of the determinants of approval of loans to established small businesses in all fifty states, Faith Ando (1988) finds that many factors (business experience, size, credit rating, previous bankruptcies, etc.) influence approval rates in the ex-
33 For example, even as late as 1990, "African Americans accounted for 12.1 percent of the population but they owned only 3.1 percent of the total business and 1.0 percent of receipts of all U.S. firms. That same year, Hispanic Americans accounted for 9 percent of the population, but only 3.1 percent of U.S. businesses and 1.2 percent of all receipts" (United States Commission on Minority Business Development, Final Report 1992, p. 6, cited in Stephanopoulos and Edley 1995).
pected directions, but that after controlling for these factors loans to blacks are still less likely to be approved.34 Using data from the 1993 National Survey of Small Business Finances, David Blanchflower, Phillip Levine, and David Zimmerman (1998) find that "after controlling for a large number of characteristics of the firms, . . . black-owned firms are substantially more likely to be denied credit than other groups" (p. 1), although they report little or no evidence that women are discriminated against in this market.35 Conditional on getting approved for a loan, and controlling for other characteristics of owners and firms, it also appears that black-or minority-owned firms get less (or more expensive) credit. Using the 1982 Characteristics of Business Owners survey, Timothy Bates finds that white owners typically command much more financial capital than blacks at the point of business start-up (1991, p. 66). Moreover, he finds that blacks get smaller bank loans than whites with otherwise identical characteristics.36
34 However, her sample is based on a low (24.1 percent response rate), and the fact that it displays some unusual characteristics relative to other samples—in particular, blacks and Hispanics having approximately the same human and financial capital as whites—raises questions about representativeness.
35 Ken Cavalluzzo and Linda Cavalluzzo (1998) report similar evidence from an earlier version of the same survey, and Cavalluzzo, Cavalluzzo, and John Wolken (1999) report such evidence, plus additional analyses, using the 1993 survey. One problem with drawing inferences from differences in loan approval rates is selection in terms of who applies for a loan. As an example, Blanchflower, Levine, and Zimmerman (1998) report qualitative evidence indicating that black-owned firms are more likely to report being seriously concerned with credit market problems and hence are less likely to apply for credit out of fear that the loan would be denied (p. 1).
36 Bates reports that "while the white business borrower gets $2.09 and $1.91 in debt capital per dollar of equity capital, other things equal, according to the equations for all bank loan recipients and exclusive bank loan recipients, the black business borrower generates only $0.69 and $0.74 in
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
Holzer and Neumark: Assessing Affirmative Action
503
Of course, as in the mortgage literature (and as in the labor market literature), one can suggest unobservables that could explain these differences in access to credit. As in the case of labor market discrimination, in interpreting such behavior it may be important to think about to whom the information is unobservable. It is conceivable that lenders know something that researchers do not know from the available data, and act rationally on the basis of this, in which case we might not want to consider the evidence as indicating discrimination. But to suggest such unobservables is not to prove that they explain the findings. Presumably, the unobservables of interest are those related to business failures (or defaults on mortgages). Indeed, Bates (1991) reports that black business owners have higher discontinuance rates measured over the sample period, when he does not take account of the size of the loan, and suggests that "these patterns may reinforce banker attitudes that black business startups are risky relative to whites" (p. 78). However, when he adjusts for the loan size differential associated with race (net of other factors), the evidence suggests that black-white differences "in business discontinuance rates would have been trivial among bank loan recipients" (p. 79); the implication is that had blacks gotten loans as large as those of whites, discontinuance rates would have been equal. This evidence makes it more difficult to argue that unobservables related to loan risk are driving the race differences.37
debt capital for a dollar of equity capital input" (1991, p. 72). Caren Grown and Bates (1992) provide similar evidence in a study focusing on the construction industry, using the same data, while Bates and William Bradford (1992) report similar results with regard to venture capital.
37 A largely unexplored historical explanation of lower business ownership rates among women is that, prior to the enactment of the Equal Credit
We are not aware of research that directly addresses the issue of discrimination in contracting and procurement. Although the small shares of government contracts that went to minority-and women-owned businesses prior to the implementation of affirmative action programs are consistent with such discrimination (Stephanopoulos and Edley 1995, Ch. 9), it is conceivable that the problem was one of lower minority and female business ownership, rather than discrimination per se. Of course, since business formation likely depends on expected future revenues, discrimination in contracting and procurement would be expected to hold down the formation of businesses owned by women and minorities, making it difficult to establish cause and effect.
Overall, in assessing whether the "playing field" is level, when it comes to university admissions and government contracting and procurement the evidence is weaker. We think it is fair to say, though, that there is clearly a history of past discrimination against some groups along both dimensions, and that the evidence is most consistent with continuing discrimination against blacks in business lending.
3.4 The Distributional Effects of Affirmative Action
Whether discrimination exists is clearly important in assessing whether affirmative action levels the playing field. In our view, the preceding discussion of employment, enrollment, and business ownership and success suggests that in many respects the playing
Opportunity Act of 1974, women faced unique problems, as single women were not considered credit-worthy, married women had difficulties establishing credit histories, and alimony and child support were not counted as income (Stephanopoulos and Edley 1995, ch. 9).
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
504       Journal of Economic Literature, Vol. XXXVIII (September 2000)
field is not level. This then raises the question of whether affirmative action can, in practice, help to level the playing field. In this section we review the existing evidence on the effects of affirmative action in shifting patterns of employment, enrollment, etc. To summarize briefly, there appears to be compelling evidence that affirmative action does increase employment, enrollments, and contracting for minorities and women in the ways we might expect. This leaves the critical questions of the efficiency and performance effects of affirmative action, which we discuss at length in Section 4.
3.4.1 The Effects of Affirmative Action on Employment
The effects of affirmative action on the employment of white males, white females, and minorities have received considerable attention. Early papers include those by Orley Ashenfelter and Heckman (1976), Morris Goldstein and Robert Smith (1976), and Heckman and Kenneth Wolpin (1976). Leonard's work (1984b,c) is perhaps best known in this regard, though James Smith and Welch (1984) analyzed many of these same issues concurrently with Leonard, and the earlier papers just mentioned also analyzed them. This work has also been updated to some extent by Rod-gers and Spriggs (1996b). All of these studies use federal EEO-1 data on contractors and non-contractors to analyze the employment effects of affirmative action; in contrast, Holzer and Neu-mark (1999, 2000) use micro-level employer data from other sources. These earlier studies generally involve some comparison of the shares of employment (or employment growth) accounted for by various demographic groups between establishments that practice affirmative action and those that do not. The key independent vari-
able is usually a measure of whether or not the firm is a federal contractor, and therefore subject to affirmative action requirements. Establishment-level control variables are included for size, location, recent employment growth, etc.
Leonard's studies compare changes in employment shares of different demographic groups over the period 1974-80 between contractor and non-contractor establishments; his studies are reviewed in Leonard (1989, 1990). He finds that the shares of employment accounted for by women and minorities rose at contractor establishments between 1974 and 1980, while those accounted for by white males declined. In particular, his estimates indicate that black male employment relative to white male employment grew 0.82 percent faster per year in contractor establishments than in non-contractor establishments, with the differential growth rates for black males (0.62 percent faster for contractors) and white males (0.2 percent slower for contractors) significant at the one-percent level (Leonard 1990). These estimates are from comparisons of means, but regression results with detailed sets of controls are similar (e.g., Leonard 1984c). Leonard's evidence on the effects of affirmative action on employment of women reveal modest positive effects for white women, and stronger positive effects for black women. For example, he reports regression estimates of the elasticity of employment in particular demographic groups with respect to total employment growth, finding that for white women this elasticity is 1.01 among contractors, compared with 0.97 among non-contractors. For black women the evidence is more striking, with elasticities of 1.18 for contractors and 1.07 for non-contractors.
These estimates can be contrasted with those for earlier periods and later
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
Holzer and Neumark: Assessing Affirmative Action
505
periods, both characterized by weaker enforcement. The earlier studies focused on minority hiring, because affirmative action guidelines referring to sex were not implemented until 1972 (Leonard 1989). As summarized by Brown (1982), papers by Goldstein and Smith (1976), covering 1970-72, and Ashenfelter and Heckman (1976), covering 1966-70, use similar estimation strategies to Leonard. While the first paper reports smaller effects than Leonard, the second reports similar magnitudes. In particular, the estimated one-year percentage change in black male relative to white male employment is 0.8, comparable to the 0.82 figure cited above from Leonard (1990). Brown suggests that a partial explanation for the weaker evidence in Goldstein and Smith is the classification of all establishments in contractor firms as contractors, whereas Ashenfelter and Heckman distinguished by establishment. A study by Heckman and Wolpin (1976) also covers the early 1970s, using data for the Chicago SMSA, and paying more careful attention to issues of serial correlation in what are essentially stock-adjustment models. They obtain considerably stronger estimates of the effects of affirmative action on minority hiring—an 8.8 percent change in relative employment of black males per year. Unfortunately, as Brown points out, they never analyze whether this is attributable to the econometric methods they use (including the serial correlation correction, weighting, and the inclusion of industry control variables), because they never report estimates of the earlier specifications with their data. Finally, Brown suggests that based on descriptive statistics reported in the paper, the program effects may have been particularly strong for the 1972-73 period. If we interpret the estimates from the late 1960s and early 1970s as
comparable to Leonard's estimates for the late 1970s, we are left with somewhat of a puzzle, since Leonard argues that enforcement of affirmative action before 1974 was relatively weak.38 Smith and Welch (1984) present evidence suggesting that the employment effects of affirmative action might have actually been greater in the earlier period than in the period Leonard studied, despite its weaker enforcement, which would suggest that the cumulative effects of affirmative action on the distribution of employment are likely larger than those that appear in Leonard's work.39
Turning to the more-recent period, Leonard (1990) reports that estimates of the effects of affirmative action on employment weakened in the early 1980s, as a result of lax enforcement of affirmative action regulations in the early years of the Reagan administration. He summarizes the evidence as indicating that before 1980, the elasticity of black male employment growth to total employment growth was 1.7 among contractors, versus 1.2 among non-contractors. In contrast, after 1980 the corresponding numbers were 1.0
38 To support this contention, Leonard cites reports by the United States Civil Rights Commission and United States General Accounting Office. The current enforcement structure, based in the OFCCP in the Department of Labor, was established in 1978.
39 Leonard argues that while the overall employment effects are comparable between the earlier and later periods, affirmative action's effect on minority/female representation in highly skilled jobs was greater in the latter period. But Smith and Welch present data showing that black representation in professional and managerial occupations, as well as in employment more generally, grew strongly in the contractor sector relative to the non-contractor sector in the early 1970s, and in the sectors covered by EEO-1 in the late 1960s as well. Unfortunately, Smith and Welch cannot distinguish contractors from non-contractors within the EEO-1 data during this earliest period, making it impossible to calculate an exact cumulative effect of affirmative action on employment shares.
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
506       Journal of Economic Literature, Vol. XXXVIII (September 2000)
among contractors, versus 1.1 among non-contractors. Rodgers and Spriggs (1996b) compare EEO-1 data extending from 1979 to 1992, estimating the cross-sectional relationship between contractor status and the share of employment of various demographic groups, controlling for other factors. They find that from 1982 to 1992, the positive relationship strengthened for the black employment share, from a 0.83 percentage point effect in 1982 to a 1.36 percentage points effect in 1992, effects that they appear to regard as roughly comparable. Because these are cross-sectional estimates, it is not clear how to compare them with Leonard's estimates. Also, as Leonard pointed out in his earlier research, because contractor and non-contractor establishments may differ in other ways, changes in rates of employment growth may be more informative.
Whether or not the contractor program contributed to rising inequality within the black community also remains somewhat unclear. In other work, Leonard (1984c) shows that contractor effects on employment are found across the entire occupational spectrum and did not increase inequality among blacks between 1974 and 1980. But Smith and Welch's results indicating rising black representation in professional and managerial occupations raise some questions about these findings.
The data and estimation strategies used in the Holzer-Neumark studies are quite different from those used by Leonard and the others, coming from surveys of establishments that (weighted by employee size) are more fully representative of the labor force than are the EEO-1 data (Holzer 1996). Unfortunately, these data are drawn from just four large metropolitan areas (Atlanta, Boston, Detroit, and Los Angeles), and
provide only cross-sectional estimates of employment differentials across establishments. The estimated effects for white males and females are therefore more likely to be biased due to unobserved heterogeneity across establishments, though the direction of this bias is not clear a priori.40 The studies using EEO-1 data also limit the effects of affirmative action to federal contractors, while the Holzer-Neumark studies use a broader, self-reported measure of affirmative action practice that includes establishments that might use affirmative action for other reasons.41
Despite these differences, results from the two data sources are quite consistent in most regards. Holzer and Neumark's data suggest that the employment of white males in the affirmative action establishments is lower by roughly 10-15 percent, which is redistributed mostly to white females and black males.42 Although Holzer and Neumark's estimates might suggest somewhat greater redistribution of employment from white males to white females
40 Interestingly, white and black women appear to be underrepresented among contractors in a simple cross-section of EEO-1 establishments; positive effects of such contractor status on their employment only appear in the difference-indifference estimates reported by Leonard or in regression equations reported by Rodgers and Spriggs that include a variety of controls for location and establishment characteristics.
41 The questions on which these analyses are based asked whether "EEO or affirmative action played any role" in the recruiting or hiring of the last employee at the establishment. Non-contractor employers might engage in affirmative action either as a court-imposed remedy for previous discrimination or to avoid "disparate impact" charges against them. Along many dimensions, such as the percentage of establishments reporting affirmative action and correlations with size and industry, the Holzer-Neumark data look similar to those used by Leonard.
42 Other studies using micro-level employer data that find positive effects of affirmative action on the hiring of blacks include Howard Block and Robert Pennington (1981), and Thomas Hyclak, Larry Taylor, and James Stewart (1992).
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
Holzer and Neumark: Assessing Affirmative Action
507
in firms practicing affirmative action, comparisons of the estimates to those reported by Leonard have to be done with caution; Leonard's estimates refer to growth rates of employment, whereas Holzer and Neumark's estimates capture a cross-section differential that may reflect growth over many years, although it need not be a "long-run" effect. For example, Leonard's estimate of a 0.82 faster growth rate of black male relative to white male employment per year would translate into similar magnitudes to those reported by Holzer and Neumark over a period of about fifteen years. The Heckman and Wolpin study is the only one to explicitly report long-run effects based on a partial adjustment model. The estimate for black male relative to white male employment is on the order of 11 percent, again paralleling the other estimates, although as noted above, the short-run estimates in these data, at least, appear unusually high in this study. We think there are two points to keep in mind in interpreting how "big" these magnitudes are. First, the findings from all of the studies imply that increases in the shares of employment in the affirmative action sector accounted for by minorities, who may have somewhat more limited skills and qualifications than the white males they replace, are quite modest in the aggregate.43 Second, the estimates for white females tend to be larger, but one might speculate that any impact of substitution of white females for white males on efficiency is mitigated because these groups are more
43 While Smith and Welch argue that the share of all black employment accounted for by the EEO-1 covered sector grew from under 50 percent to almost 70 percent between 1966 and 1980, with most of this increase likely concentrated among contractors, their data still suggest that the shares of contractor employment accounted for by blacks rose by just a few percentage points in that period.
similar in terms of skills, and any impact on incomes is mitigated because these individuals often reside in the same households.
What are the gains enjoyed by minorities and females and the costs borne by white males from redistribution of employment across sectors? Even if we assume high labor supply elasticities across the two sectors and little net increase in overall labor demand for blacks (e.g., Brown 1982), the wages of minorities and females are likely to be somewhat higher (and those of white males lower) than they would have otherwise been, because wage levels generally differ between the two sectors. For example, in the data set used in Holzer and Neumark (1999), the estimated wage differential between establishments using affirmative action and those not using affirmative action is ten percent, after controlling for worker characteristics (so this could partially reflect size effects). Further, the Holzer-Neumark data also indicate that race and sex differences in wages are smaller in establishments using affirmative action, suggesting further relative wage gains of women and minorities stemming from affirmative action.
One final point worth emphasizing here is that the establishments that are most likely to engage in affirmative action are not necessarily those that are most likely to discriminate in the first place. For instance, contractors and other practitioners of affirmative action tend to be large establishments that appear to engage in much less hiring discrimination against blacks than do smaller establishments (Holzer 1998; William Carrington, Kristin McCue, and Brooks Pierce 1995; Chay 1998). Leonard (1985) also notes that compliance reviews, which have particularly large effects on establishment hiring, are   targeted   towards establishments
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
508       Journal of Economic Literature, Vol. XXXVIII (September 2000)
that already have relatively strong minority employment rates. This has two implications. First, it suggests that to some extent affirmative action may not be a remedy for present discrimination at the establishment level, but rather may work to counteract broader disadvantages in the economy facing blacks by improving their opportunities elsewhere in the labor market. Second, it suggests that the estimates we obtain from the types of evidence described in this section may not necessarily uncover the causal effects of affirmative action—i.e., the effects that would occur if the policy were truly exogenous. To the extent that affirmative action poses less of a constraint for nondiscriminating establishments, the causal effects would be understated. On the other hand, if the "spurious" differences associated with affirmative action in minority hiring, for example, were sufficiently sharp as to outweigh the effects of affirmative action when the policy represents a true constraint, the causal effects could be overstated. This problem plagues the Holzer and Neumark studies most severely (since their measure of affirmative action is not based on exogenous contractor status), but it also plagues the EEO-1 studies to the extent that enforcement is targeted to select types of firms. This problem is endemic to research on affirmative action, as there is an absence of the types of evaluation studies that have become prominent in other areas of labor economics research. Although it is difficult to conceive of a more compelling research design to study the effects of affirmative action,44 this is one of the most prominent needs in research on the redistributive effects of affirmative action.
44 Holzer and Neumark (1999) report on some attempts to do this based on firm-size cutoffs at which different affirmative action requirements become effective.
3.4.2 The Effects of Affirmative Action on Enrollment
The effects of affirmative action on college enrollments have not, until recently, received as much attention as its effects on employment. In addition, the existing research is somewhat less convincing in its ability to isolate the effects of affirmative action, because there is no explicit distinction (such as contractor status) with which to classify colleges or universities as bound or not bound by affirmative action. Indeed, we are not aware of any systematic information regarding colleges or universities that do not use affirmative action in admissions.45 We suspect that less attention has been paid to redistributive effects of affirmative action in education because there is little doubt among researchers that it played a prominent role in increasing admissions of minorities (e.g., Bowen and Bok 1998). This may be a reasonable position. But since the period of rapidly rising enrollments (roughly the 1960s and 1970s) was one of rapid declines in poverty among minorities,46 following closely upon the desegregation of public schools, and was accompanied by antidiscrimination forces in the labor market that likely affected the returns to higher education for minorities, simple time-series trends in minority enrollments may overstate the independent effects of affirmative action. Finally, we have to be cautious in concluding that rather compelling anecdotal evidence, at least, of affirmative   action  in   admissions implies
45 There are, however, a handful of colleges that apparently resist affirmative action in admissions (as well as hiring, by refusing federal funds). One example is Hillsdale College in Michigan. In principle, a study of such colleges (assuming they are nondiscriminatory) with other matched schools might yield useful information on the effects of affirmative action on enrollments.
46Thernstrom and Thernstrom (1997) provide tables summarizing these statistics.
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
Holzer and Neumark: Assessing Affirmative Action
509
strong effects of affirmative action on enrollments. The "partial equilibrium" effects of affirmative action on a single college's admissions policies may coexist with much smaller "general equilibrium" effects on overall enrollments, since part of what may occur is simply more schools chasing the same minority students, each of whom can ultimately enroll at only one college or university.
Nevertheless, the increase in minority college enrollments over this period is striking. Summarizing data from numerous sources, Thernstrom and Thernstrom (1997) report that black enrollments as a percentage of all enrollments in schools other than black colleges rose from 1.8 percent in 1960 to 4.2 percent in 1970, 8.2 percent in 1980, and 9.0 percent in 1994. Similarly, Bowen and Bok (1998) report that from 1960 to 1995 the percentage of blacks aged 25 to 29 who had graduated from college rose from 5.4 percent to 15.4 percent. Even sharper changes are evident for professional schools in this period, with the percentage of blacks growing from one percent to 7.5 percent in law schools, and from 2.2 percent (in 1964) to 8.1 percent in medical schools. Data for Hispanics do not go back as far, but also indicate sharp gains since 1970. Even if these increases overstate the gains due to affirmative action, the lawsuits over race-based preferences (e.g., Bakke) and evidence of stepped up recruiting of minorities (Bowen and Bok 1998) strongly suggest that affirmative action played a major role.47
47 Naturally, such gains accrue disproportionately to those from better-off families, as minority students who meet criteria for admission under affirmative action programs tend to come from such families. For example, Bowen and Bok (1998) report that in their sample for 1989, 15 percent of black matriculants come from families they classify as "high socioeconomic status," compared with three percent for the national black population (p. 48).
Finally, Cecilia Conrad and Rhonda Sharpe (1996) provide evidence at a point in time (1994) for the University of California system that more clearly indicates the role of affirmative action. In both undergraduate and professional graduate programs, a large fraction of students are admitted solely on academic credentials, while another group is admitted based on other criteria— including diversity—which prior to Proposition 209 included race and ethnicity. The figures they cite show that minority representation is considerably higher in the latter group of admits than in the former, or that non-academic admission criteria are generally more important for minority students.
Another way to approach the issue of the effects of affirmative action in current admissions decisions to undergraduate and graduate institutions is by examining relative qualifications of minority admits. Linda Datcher Loury and David Garman (1993) find shortfalls in SAT scores among blacks even in schools with average quality below the median.48 In contrast, Thomas Kane (1998) argues that college admission rates for blacks, controlling for high school grades, SAT scores, and personal characteristics, are only higher for the top quintile of schools. These results are not necessarily inconsistent, though the exact reason for the racial gap in below-median   schools   is unclear.49
48 For instance, their data show that almost 75 percent of whites, but only about 10 percent of blacks, have SAT scores above 850 in schools with average scores below 1000.
49 Kane's data are based on self-reported applications to different colleges, while Datcher Loury and Garman look only at schools attended. The latters' results might reflect racial differences in decisions about which school to attend, conditional on acceptance, or racial differences in average test scores in the same schools, conditional on being above a common cutoff. Given the wide range of schools within the lower category of schools considered by Datcher Loury and Garman, it is also quite possible that blacks are
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
510       Journal of Economic Literature, Vol. XXXVIII (September 2000)
There are also some related studies for graduate school admissions. Regarding professional schools, studies show that blacks admitted to medical schools (e.g., Steven Keith, Robert Bell, and Albert Williams 1987; Robert Davidson and Ernest Lewis 1997) and graduate business programs (Mary Dugan et al. 1996) have lower grades and/or higher probabilities of being admitted, conditional on their grades, consistent with preferential treatment. In the only study of Ph.D. programs of which we are aware, Gregory Attiyeh and Richard Attiyeh (1997) study admissions decisions in five disciplines (economics, biochemistry, English, mathematics, and mechanical engineering) at 48 leading graduate schools that are members of the Association of Graduate Schools of the Association of American Universities. They estimate probit models for admissions decisions including demographic variables, information on GRE scores, college grades, selectivity of undergraduate college, other graduate degrees, major, and institutional characteristics (including institution dummy variables). Their results indicate modest preferential treatment of women in three out of five disciplines, and far more substantial preferential treatment of minorities relative to other U.S. citizens in all five disciplines. Thus, across these professional and academic fields, there is fairly straightforward evidence of preferential treatment of women and especially minorities.
Bowen and Bok (1998) provide a broader look at the evidence regarding undergraduate admissions. They first note the well-known statistics on differences in SAT scores between minority and nonminority students at selective
concentrated within a different set of schools (with lower average test scores) than whites in this category.
undergraduate institutions. However, they provide some important counterpoints to the typical claim that such differences unambiguously or solely reflect lower standards for minorities (e.g., John Bunzel 1996). First, they present a simple and clear exposition of a point that is obvious to statisticians, but not to everyone else—that lower test scores in the minority population relative to the nonminority population imply that even if an identical cutoff is used in deciding whom to admit to a particular school, the average SAT score of nonminority admits will exceed that of minority admits (p. 16).50 Thus, evidence of differences in SAT scores does not prove and almost certainly overstates the role of preferential treatment in admissions. In addition, it is sometimes argued that test scores are poorer predictors of college success for blacks than for whites, which could lead admissions officers to rely less on test scores for blacks, and would likely lower the test scores of black relative to white admits at comparable institutions (see William Dickens and Kane 1999). Looking at data for selective institutions, Frederick Vars and Bowen (1998) and Bowen and Bok (1998) find some evidence that test scores are worse predictors for blacks.
The role of overall differences in the distribution of test scores in generating test score gaps for admits has received some recent attention from researchers. Linda Wightman (1997) studied 30 of the most competitive law schools, comparing the test scores and grades of black students with those of white students in the bottom decile of white admits. The notion is that these white
50 Think of estimating average height in two populations, one consisting of all men over six feet tall in the overall population, and another of all men over six feet tall in the NBA. Clearly, the second average would exceed the first.
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
Holzer and Neumark: Assessing Affirmative Action
511
students have characteristics close to those of the white students who were displaced by affirmative action admissions. While the overall difference in LSAT scores between white and black students was 24 percent, the difference between these least-qualified whites and the black students fell to 10 percent. Given that the black students score lower on the LSAT, this is consistent with a smaller differential between the marginal black students admitted and the marginal white students denied admission; this indicates that there is still preferential treatment, but of a smaller magnitude than suggested by the overall test score differential. However, the evidence is incomplete, because it does not compare admits and those denied admission, and it is not necessarily generalizable to other settings.
A recent paper by Dickens and Kane (1999) discusses the same issue regarding test score differentials as Bowen and Bok, but makes an additional important point—namely, that among the many indicators of college success, test scores are the one on which blacks fare worst in relative terms, which accentuates test score differences among admits even when equal standards are being used for blacks and whites. Even more importantly, Dickens and Kane report some simulations/calculations that address the question of whether the observed test score differentials at elite schools are consistent with strong preferential treatment, or instead a race-blind process, taking account of both differences in test score distributions as well as other factors that may be used in judging qualifications, in terms of outcomes that matter; in the case of college admissions, they use predicted college grade point averages. As a basis for comparison, they compute the test score differences that would
emerge from a race-blind admissions process in various narrow ranges of predicted college grade point averages. Depending on the precise simulation, these differences (in terms of population standard deviations) range from 0.87 to 1.37, although most are between 0.89 and 1.05. These estimates contrast with a population difference of 1.08 standard deviations, implying that a race-blind admissions process can generate black-white test score shortfalls among admits that are nearly as large as the population shortfalls, precisely because it is on test scores that blacks fare worst in relative terms. Dickens and Kane also present some comparisons of the implications of their simulations with test score differentials at universities reported by Richard Herrnstein and Charles Murray (1994). They suggest that some of the extreme cases (such as the University of California, Berkeley) are larger than what could be expected from a race-blind process, but that test score differentials for other elite schools are in the range consistent with a race-blind process.
We view this type of research as critical to the debate over affirmative action in university admissions. We also suspect that this paper will spark additional analyses that may yield a wider range of estimates. Provisionally, though, this research suggests that, on average, the types of test score differences we observe may not reflect strong preferential treatment for minorities. More specifically, this work emphasizes that there is not a compelling case that existing test score differentials reflect strong preferences at more than some isolated universities.
Bowen and Bok also present two other striking statistics regarding SAT scores, which put recent debates regarding test score differentials at selective schools into perspective. First, they
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
512       Journal of Economic Literature, Vol. XXXVIII (September 2000)
report that the black-white gap in SAT scores at a subset of four selective institutions with the requisite data narrowed between 1976 and 1989 (although it may have widened prior to 1976). Second, these same figures show that the average SAT score of black matriculants at these schools in 1989 was higher than the average SAT score of all matriculants in 1951 (although the average for white matriculants in 1989 was considerably higher). Again, these figures do not dispute the claim that there is preferential treatment in admissions for blacks and minorities generally, but they do suggest a more sober evaluation of the casual use of statistics on SAT scores in this debate.
Bowen and Bok then move on to an analysis of the impact of affirmative action on admissions, using data on five universities for which they have the needed data on applicants. Although future academic research will undoubtedly probe deeper into the details of their analysis, they attempt to carry out what is the correct approach. As they note:
The best way to measure the degree of preference given is by comparing the credentials of those black students who presumably would not have been enrolled under a race-neutral standard . . . with the credentials of an equivalent number of rejected applicants (mostly white) who would have been admitted under a race-blind procedure, (p. 18)
The estimation and calculation needed to answer this question includes the simulation of a race-neutral admissions policy, although in a less sophisticated fashion than Dickens and Kane. Bowen and Bok conclude that under a race-neutral policy the probability of admission for black applicants would fall to 0.13 (as compared with 1989 figures of 0.42 for black applicants, and 0.25 for white applicants). As confirmation that their estimation and simulation yield
reasonable estimates, they also report on figures for the University of California, Berkeley before and after the adoption of race-neutral admissions because of Proposition 209. In 1997, when race-sensitive admissions were used, the admission rates were 0.485 for blacks, and 0.299 for whites. For the following year the rate for blacks fell to 0.156, and the rate for whites rose to 0.303 (pp. 32-33).51 Thus, their calculations indicate that preferential treatment of minorities in admissions is responsible for a large share of minority enrollment at these schools, and that large swings in enrollments could result from the elimination of what may be rather mild preferential treatment. However, as the Dickens and Kane study suggests, a race-neutral policy that more properly combines multiple predictors of success might lead to smaller shifts than those implied either by Bowen and Bok's estimates, or actual experiences following Proposition 209.
3.4.3 The Effects of Affirmative Action on Contracting with Minority-and Women-Owned Businesses
Evidence on the effects of affirmative action on contracting with minority- and female-owned businesses is more sparse, although there is some evidence consistent with strong positive effects of these programs. In their Review of Federal
51 Of course applicant behavior might change in response to changes in admissions procedures; for example, blacks might begin to apply to more schools. As a result, these figures probably overstate the response to changes in admissions policies. Nonetheless, it is unlikely that for the relatively small subset of selective schools this could lead to very different conclusions with regard to student "yields," and Bowen and Bok present some simulations that confirm this (p. 35). At the same time, as hinted at by the small increase in the white admission rate at Berkeley, and indicated by other calculations Bowen and Bok present, the admissions taken from blacks under a race-neutral policy would result in only very marginally higher admission rates for whites.
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
Holzer and Neumark: Assessing Affirmative Action
513
Affirmative Action Programs, using information obtained directly from the federal agencies involved, Stepha-nopoulos and Edley report that between 1982 and 1991 there was a 24 percent increase in the dollar volume of all federal procurement contracts over $25,000. Over this same period, contracts awarded to firms owned by women increased by more than 200 percent, and those awarded to minority-owned firms increased by more than 125 percent (Stephanopoulos and Edley 1995, ch. 9, p. 5).52 They also report strong gains in Department of Defense (DoD) contracting with small, disadvantaged businesses, with such contracting growing from 2.1 percent of DoD procurement in 1985 to 5.5 percent in 1994. More generally, they report that by 1993 affirmative action in contracting and procurement had achieved near-proportional representation for minority-owned businesses, reporting that as of this year "prime contracts for minority-owned businesses were 6.4 percent of the total dollar volume. This approaches the proportion of minority-owned businesses among all U.S. firms" (ch. 9, p. 5).53 One problem with this evidence regarding federal programs is that it does not identify the independent effects of affirmative action; it is possible that other factors partly account for
52 They do not explicitly state whether the figures for minority- and women-owned businesses also refer to contracts over $25,000.
53 However, they are also quick to point out that minorities are underrepresented among business owners, so there is less evidence (indeed, none is discussed in their review) that affirmative action has spurred minority and female entrepreneur-ship. On the other hand, although Stephanopoulos and Edley do not emphasize this point, they note that the receipts of women- and minority-owned businesses are typically much smaller than those of white-owned firms (in 1990, by a factor of more than two; ch. 9, pp. 2-3). Thus, one could argue that as a fraction of receipts of minority-owned businesses, federal programs have gone beyond proportionality.
the growth in contracting and procurement with minority- and female-owned businesses.
Aside from the federal programs, state and local governments frequently favor minority-owned businesses. Bates and Darrell Williams (1995a), Bates (1998), Samuel Myers and Tsze Chan (1996), and Mitchell Rice (1995) provide some summary information suggesting large impacts of local procurement programs on the minority share of contracting and procurement, in Atlanta and other cities. For example, in 1973, although the majority of Atlanta's population was black, black-owned firms received only one-tenth of one percent of the city's procurement business. But after implementing a program of promoting minority business ownership, this share rose to 19.9 percent by 1976, and 38.5 percent by 1978 (Bates 1998, p. 11). Similarly, Bates (1998) reports that controlling for firm characteristics including size, industry, etc., minority-owned businesses are more likely to sell to government than are other businesses.
Overall, then, there is a prima facie case for concluding that affirmative action was responsible for growth in government contracts with minorities and women. Of course the focus on government contracting to some extent misses the point, which is to help foster minority and female business ownership. Because this latter question pertains more to the efficiency/performance effects of affirmative action, we discuss related evidence in the next section.
4. Beyond Redistribution: The Efficiency/Performance Effects of Affirmative Action
Although there are some remaining questions, the evidence reviewed in the previous section is most consistent with
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
514       Journal of Economic Literature, Vol. XXXVIII (September 2000)
the conclusion that affirmative action succeeds in boosting employment of women and minorities, minority enrollment in universities, and government contracts for minority- and women-owned businesses. In addition, our view is that there is sufficient evidence of discrimination in at least some of these spheres to rule out a facile conclusion that affirmative action must reduce efficiency or performance. Consequently, an assessment of theoretical and empirical research on the efficiency/performance effects of affirmative action is essential. Indeed our review has a heavy emphasis on this latter issue, simply because it has been studied less, review of research on this topic is lacking in the literature, and it is a central economic issue in the debate over affirmative action.
We begin this section by reviewing the theoretical literature that asks whether policies that explicitly or implicitly make race or sex a consideration are likely to improve the workings of labor markets that are influenced by discrimination. We also consider the potential efficiency effects of affirmative action in university admissions and government contracting. One might think that, if discrimination exists, then a theoretical conclusion that affirmative action can improve efficiency or performance would be straightforward. Thinking simplistically, affirmative action policies are likely to reduce efficiency in a world with no discrimination by leading to the hiring and promotion of less-qualified women and minorities. On the other hand, in the presence of discrimination affirmative action should increase efficiency, by discouraging the hiring and promotion of less-qualified white males over more-qualified women and minority workers. It turns out, however, that this question is more complicated than this simple dichotomy suggests.
After the theoretical discussion, we turn to evidence on efficiency/performance effects. As our theoretical review explains, economic theory does not make unambiguous predictions about the efficiency effects of affirmative action in employment. Nonetheless, in looking at labor market effects, economic theory at least offers some guidance as to the types of effects we should examine—e.g., productivity, profitability, unit labor costs, etc. With respect to education, theory provides even less guidance; as there is not a compelling case that university admissions policies in the absence of affirmative action are efficient, it is not clear what measures we should look at to try to infer whether affirmative action leads to deviations from efficiency. Rather, when we turn to education we examine evidence on specific hypotheses regarding the effects of affirmative action that might bear on efficiency (e.g., incentives, "fit," and externalities). Finally, we review the little evidence that exists addressing the efficiency/performance effects of affirmative action in contracting and procurement.
4.1 The Efficiency Effects of Affirmative Action in the Labor Market: Theory
A theoretical assessment of the efficiency effects of affirmative action requires both a model of discrimination, and a "form" of the policy. The discussion in this section is based on papers that have introduced discrimination and affirmative action in alternative ways— some leading to sharp predictions, and others less so. One implication of this variety of predictions is that it is important, as an empirical matter, to assess both the nature of discrimination and the workings of affirmative action. Throughout we assume that when we work with models of taste discrimination the utility or disutility generated by
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
Holzer and Neumark: Assessing Affirmative Action
515
discriminatory tastes does not enter the social welfare function.
4.1.1 Becker-Style Models of Discrimination
The link between discrimination and efficiency is not as straightforward as it might seem. In a simple Becker-style model of racial wage discrimination stemming from employers' discriminatory tastes, discrimination need not be inefficient; instead it may leave total output and employment unchanged, while simply resulting in redistribution among employers and workers of different races (Kenneth Arrow 1972). In such a model, if production functions are identical, efficiency prevails as long as total employment is equally distributed among all firms. A sufficient condition for this is that Becker's "discrimination coefficients" be constant and identical across all employers.54 Inefficiencies can be introduced rather simply, however. For example, if we suppose that employers get no utility from hiring whites but suffer disutility from hiring blacks, and if some employers have more discriminatory tastes than others, the distribution of employment across plants is likely to be unequal, which will impair efficiency as long as there are not constant returns to scale.
The simplest Becker-style model that Arrow develops is not particularly well-suited to thinking about inefficiency from the supply side. Since it has a single labor input and inelastic supplies of white and black labor, all labor is employed despite the discrimination. However, if we introduce elastic labor supply functions, assumed to be the same for whites and blacks, then the lower market wage offered to blacks because of discrimination will cause them to un-
54 Arrow illustrates this for the case in which black and white workers are perfect substitutes in production, but it holds generally.
dersupply their labor to the market; their marginal product exceeds the marginal value of their leisure. Such models are also less than ideal for exploring the relationships between discrimination, efficiency, and affirmative action. Typically in such models labor inputs from different groups are combined into a single labor input, often as perfect substitutes, with issues like promotion or hiring into more-skilled positions swept under the rug. Thus, it is more useful to sketch a model in which there is hiring or promotion into different positions, and explore the role of discrimination in generating inefficiencies in such a model. Models in which worker productivity is linked to the quality of the job match are more natural for this purpose.
4.1.2 Simple Job Matching Models
Michael Rothschild and Joseph Stiglitz (1982) develop a model in which the output (Q) of an employee depends positively on his own ability (A), but negatively on the deviation between his ability and skill requirements (S), so that both overqualification and underqualification are costly:
Q{A,S) = a + (3A - y(A - S)2.
We can modify this model slightly, normalizing the output price to one, and assuming that the firm's production function is simply the aggregation of individual outputs over a continuum of possible values of S:
Q = J {a + pA - y(A - S)2}dS.
When the firm hires a worker, its problem is to assign this worker to the optimal S, subject to paying that worker the wage he would expect in the competitive market. It is assumed that firms have some information (Y) about the worker which is informative about A. Risk-neutral, nondiscriminatory employers maximize expected output, so
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
516       Journal of Economic Literature, Vol. XXXVIII (September 2000)
that workers are assigned to jobs with S = £(A|Y). In this case output of workers with characteristics Y is given by
Q(A,S\Y) = a + PA - y(A - E(A|Y))2,
and the wage equals expected output and is given by
W(Y) = E[Q(A,S\Y)]
= a + $E(A\Y)-yVar(A\Y).
Discrimination in hiring and promotion (e.g., stemming from tastes) could be manifested as a refusal or reluctance to hire blacks (B) into high S jobs. Assume that this discrimination takes the form of employers assigning black workers to jobs with S = E(A\Y) - (b/y)1^,55 and that race is uninformative about A. Then output is
Q(A,S\Y) = a + PA - y(A - E(A|Y))2
- iXb/y) + 2A{b/y)^ - 2E(A\Y)(b/y)^}
and expected output equals
E[Q(A,S\Y)] = a + pE(A|Y)
-yVar(A\Y)-b.
If all employers behave the same, expected output is lower by b everywhere, and the market wage is lower by the same amount, so expected profits are unaffected. If discriminatory behavior varies, then more discriminatory employers earn lower profits. Either way, output is lower than it would be in the absence of the discriminatory behavior, since blacks are assigned to jobs for which they are generally over-qualified, entailing inefficiency. In this model, we can regard an "ideal" affirmative action policy as requiring employers to assign blacks to jobs with S = E(A\Y). This would raise black wages, increase the assignment of blacks to higher-skilled jobs, and increase output.
A related analysis in which discrimination affects the allocation of minority
55 This particular form of the constant leads to a simple form for the constant in the solution.
and nonminority labor to particular jobs is the customer discrimination model considered by Kahn (1991). He shows that in such a model affirmative action—in the form of an equal pay constraint plus a requirement of proportional representation—can improve efficiency, replicating the same allocation of resources that is produced in the absence of discriminatory tastes. As Kahn points out, a relatively small share of employment in the federal contractor sector is in industries in which there is customer contact and in which customer discrimination is likely to be very important. But given the broader definition of affirmative action that we adopt in this review, customer discrimination should perhaps be given greater attention.
Nonetheless, perhaps driven in part by the view that employer taste discrimination is unlikely to persist in labor markets (e.g., Shelly Lundberg and Richard Startz 1983), and by the view that customer discrimination is of limited importance, most of the existing work on the efficiency effects of affirmative action has focused on models of statistical discrimination. Before describing this work, however, we consider some earlier work on models of affirmative action in which discrimination (taste or statistical) is not central.
4.1.3 Affirmative Action as Quotas
Welch (1976) considers a stark case of affirmative action, namely employment quotas for minorities.56 He allows for the possibility of taste discrimination that lowers wages of minority workers, but because in his model taste discrimination results only in redistribution, this discrimination can be ignored when looking at allocative efficiency. More significantly,   though,  Welch's model
56 George Johnson and Welch (1976) explore similar issues in a closely related analytical framework.
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
Holzer and Neumark: Assessing Affirmative Action
517
does not allow discrimination of the type outlined above that might prevent the allocation of qualified minority workers to higher-skilled jobs.
In the model, employers need workers in different skill categories (call them unskilled and skilled), and average skill levels are lower for the minority than for the majority population. Skilled and unskilled workers each have a comparative advantage in the corresponding type of work, so the skilled vs. unskilled distinction is more than just a label. The quota requires that for every majority skilled worker hired, r (< 1) minority workers must be hired in the same position. The minority proportion of the population is n, and the minority proportion with qualifications needed for the skilled position is nm, with 7t > nm and r > nm. The quota is assumed to be accompanied by an equal pay constraint for the skilled category. In such a model, employment quotas could result in too few skilled workers being hired, because the quota for minority workers cannot be met in the skilled category. This obviously increases production costs and hence entails inefficiencies. However, Welch points out that the costs may be mitigated (but not eliminated) by "skill bumping," which occurs when employers hire unskilled minority workers into the skilled category in order to be able to hire more skilled majority workers. This latter strategy may correspond to the assertion that affirmative action results in the hiring and promotion of unqualified minority (or female) workers.57
Welch's model leads to empirical implications that can guide the search for and interpretation of evidence on af-
57 Welch considers a one-sector model, as well as a two-sector model in which affirmative action is imposed in only one sector. The productivity implications are similar, although the implications for income redistribution are not.
firmative action. Affirmative action results in some less-skilled minority workers being hired into skilled jobs. Also, to the extent that firms' production costs are increased, performance measures related to cost (inversely) or profitability should decline. Hence, micro-level evidence regarding costs or profitability of firms, and the qualifications and performance of workers hired into particular jobs, comes to the fore. If empirical research fails to detect lower qualifications, lower profitability, or higher costs at the micro level, we would be more skeptical regarding aggregate efficiency effects. We assume that one cannot actually estimate the aggregate welfare losses associated with such a policy, although Welch does go through a simulation exercise to try to gauge these losses. This simulation makes the important point that when skill bumping occurs (which depends on parameters of the model) the social inefficiency associated with quotas is driven primarily by the extent to which misallocated labor is less productive; in the specific model he uses, these costs are much greater if the lost productivity from placing an unskilled worker in a skilled job is high, and these costs can be quite low if this lost productivity is low.
Welch also anticipates an issue that later became more prominent, namely whether affirmative action might affect skill acquisition. In particular, because a quota raises the relative earnings of the minority group, the relative returns to investment in skill for minorities may rise, leading to skill convergence. Welch shows, however, that this result may not be so simple. First, in a two-sector model (with quotas applied in one sector, such as the contractor sector), the earnings of skilled workers relative to unskilled workers may fall in both sectors, so that although skill convergence may result, the average skill level
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
518       Journal of Economic Literature, Vol. XXXVIII (September 2000)
declines.58 Second, the returns to skill in a world with quotas depend on how majority and minority workers are matched, because with the equal pay constraint wages paid to minority workers are an average of skills of minority and majority workers with whom they are associated. Nonetheless, Welch suggests that quotas are indeed more likely to increase the wages of skilled minority workers relative to unskilled minority workers, and hence to increase the returns to skill and relative investment by minority workers.
In Welch's model, quotas and equal pay constraints induce inefficiencies. In a sense, it is no surprise that beginning with a model with perfect information and competition, such policies induce distortions. A more recent line of inquiry into the efficiency effects of antidiscrimination measures proceeds from the perspective of imperfect information in the form of statistical discrimination. In our view, the major contribution of this research is to suggest how affirmative action policies may enhance efficiency.
4.1.4 Statistical Discrimination
In early models of statistical discrimination (Dennis Aigner and Glen Cain
58 For example, in the case without skill bumping, we can think about the first effect of the quota as leading to a shift of skilled nonminority workers to the uncovered sector, which raises skilled wages in the covered sector, lowers skilled wages in the uncovered sector, and raises unskilled wages in the uncovered sector. (Unskilled wages in the covered sector could go either way, as marginal productivity falls but output price rises.) Equilibrium is then restored by the migration of labor. In particular, if unskilled wages have risen by more in the uncovered sector, unskilled labor flows to that sector, which reduces the marginal productivity of skilled labor in the covered sector. Welch shows that depending on the magnitudes of the elasticity of substitution between skilled and unskilled labor in the covered sector, and the elasticity of demand for the product of the covered sector, it is possible for the net result to be declines in relative earnings of skilled labor in both sectors.
1977) racial wage differences emerge because the reliability of information about different groups varies. If, for example, information about blacks available to those making hiring decisions is less reliable, then more-qualified blacks are paid less than comparable whites, less-qualified blacks are paid more than comparable whites, but average pay equals average productivity. Thus, there is no "group" discrimination, in that for each group average wages equal average productivity. However, Lundberg and Startz (1983) point out that if we take into account the response of human capital investment, then the lower returns to these investments in these qualifications for blacks can lead to underinvestment by blacks.59 The result is group discrimination, because "groups with equal average initial endowments of productive ability do not receive equal average compensation in equilibrium" (Lundberg and Startz 1983, p. 342). Furthermore, they show that as long as the marginal cost of training or investment is increasing for each individual, this equilibrium is socially inefficient, because the cost of training the marginal white exceeds the cost of training the marginal black. A policy that forbids unequal wage schedules to the two groups shifts some investment to blacks for whom marginal costs are lower, and hence improves efficiency. However, such a policy corresponds more closely to equal pay laws than to affirmative action.
A natural question to ask of models like these is whether there is any basis
59 If we think of investment solely in terms of easily observable characteristics like schooling, Lundberg and Startz note that it is not obvious why information should be less reliable for one group than the other. However, citing earlier work by Arrow (1973), they suggest that human capital investment be interpreted more broadly as a set of behaviors that lead to better performance on jobs, tests, etc., and hence is imperfectly observable.
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
Holzer and Neumark: Assessing Affirmative Action
519
for believing that employers have less reliable information about minorities or women. Kevin Lang (1986) suggests that differences in methods of speaking and listening (as well as nonverbal communication) between blacks and whites and between men and women, coupled with white men being disproportionately represented among those doing the hiring, can create a situation of less reliable information about minority and female job applicants. Bradford Cornell and Ivo Welch (1996) formally demonstrate that in this type of situation majority group employers are likely to treat majority group workers favorably, even when they have no discriminatory preferences. Neumark (1999) provides the only empirical analysis of the reliability of labor market information about various groups. OLS and IV estimates of a regression of starting wages on a proxy for actual productivity permit the estimation of the variance of expected to actual productivity, or the reliability of information about new workers. These estimates indicate that employers have worse information about female new hires than male new hires, although not about minority relative to nonminority new hires. Of course, over time employers might be expected to learn about their workers, so one might argue that any effects of differential reliability should fade away. But once we allow for agents to respond to the decisions employers make based on initial information, such as through human capital investment, it is possible that persistent effects can arise.
Lundberg (1991) extends the earlier theoretical work in two important ways. First, paralleling Rothschild and Stiglitz (1982), she considers the efficiency effects associated with matching workers correctly to jobs, using a framework with heterogeneous jobs in which higher-skilled jobs pay higher wages. Given the
imperfect information, employers make optimal decisions by allocating relatively more higher-qualified (in terms of unobserved productivity) whites than blacks to higher-skilled jobs. The antidiscrimination constraint in this model— which corresponds more closely to affirmative action—forces employers to place more blacks in such jobs, even though employers have difficulty in assessing their qualifications, which generates some bad matches. This creates a tradeoff between increased efficiency in human capital investment and decreased productive efficiency; the antidiscrimination policy is more likely to increase efficiency when the cost of mismatches is low, and the costs of distorted human capital investment are high.60 Of course, though not related to efficiency per se, it is likely to be precisely when the costs of mismatches are low—because skill differences between the jobs are slight—that distributional gains are likely to be smaller since wage differences are also slight. On the other hand, cases in which skill differences
60 Lundberg also considers different types of antidiscrimination policies, in particular what she characterizes as affirmative action vs. a prohibition of disparate treatment. Because we do not draw as sharp a distinction between these types of policies in this review, we leave it to the reader to examine these differences further. Paul Milgrom and Sharon Oster (1987) develop a model which similarly considers job assignment and investment decisions, but in which discrimination arises because promotion conveys information to the outside market and therefore leads to higher wages for employers. Those groups about whom outside information is the worst—which they presume to be minority and female workers—suffer the most promotion discrimination in this context, because there is an externality posed by promotion that the worker but not the employer captures, which is strongest for minorities and women. In this case, investment decisions and job assignment decisions are distorted, and a government policy that mandates the optimal promotion rate for each group can unambiguously increase efficiency. This differs from Lundberg's model in which employers make optimal job assignment decisions given the information constraint.
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
520       Journal of Economic Literature, Vol. XXXVIII (September 2000)
are larger seem more likely to result in larger gains in efficiency in human capital investment, and larger distributional gains. So there is no obvious tradeoff between efficiency gains overall and distributional consequences.
When it comes to assessing empirical evidence on the efficiency effects of affirmative action, the results of this paper and the Lundberg and Startz paper are somewhat discouraging. We can perhaps hope to determine whether affirmative action leads to the hiring of less-qualified minority or female employees. What the models in these papers show, though, is that such evidence does not necessarily imply that affirmative action is on net inefficient. In the framework of these models, assessing efficiency also requires an empirical analysis of the allocation of human capital investment, a challenging goal that has not been seriously attempted.61
In the Lundberg and Startz model, the statistical discrimination takes the form of less accurate information about minority workers. In contrast to earlier work on statistical discrimination, an appealing feature of this model is that by incorporating human capital investment, this seemingly benign form of statistical discrimination leads to a discriminatory outcome. Stephen Coate and Glenn Loury (1993) consider a less benign form of statistical discrimination, specifically negative stereotypes about minority workers. This model extends Arrow's (1973) earlier work, with the most important difference being that employers' lower initial evaluations of minority workers' qualifications result in less-frequent assignments to more-demanding jobs, rather than wage discrimination. As such, the Coate and
61 However, Holzer and Neumark (2000) present some evidence on the relationship between affirmative action and training of different groups of workers.
Loury set-up is more amenable to thinking about the effects of affirmative action.
The Coate and Loury model is one in which employers have to optimally assign workers to jobs. Mismatches between workers and jobs—in particular the assignment of unqualified workers to the more-highly rewarded jobs—are costly to the firm. Conversely, successful matches of qualified workers to these jobs entail a positive return to employers. As in the Lundberg and Startz model, workers invest in the qualifications needed to perform the more-demanding job prior to entering the labor market. Imperfect information is still important, because with perfect information (and no discriminatory tastes) employers would do this assignment based on individual productivity. As a result of the imperfect information, employers base expected productivity in each job in part on group membership. In this framework, it is straightforward to demonstrate the effects of negative stereotypes, by which Coate and Loury mean an overly negative evaluation of the likelihood that minority workers are qualified for the more-demanding job. This negative stereotype results in minority group members facing lower returns from human capital investment, which leads them, in fact, to have weaker qualifications than the majority group.62
In equilibrium, then, employers' negative stereotypes are confirmed, generating a "self-fulfilling prophecy." Coate and Loury also show that equilibria with self-fulfilling negative stereotypes   are   inefficient,   entailing both
62 This contrasts with the earlier literature on statistical discrimination, where it was argued that misperceptions of relative productivity of groups of workers would be unlikely to persist in competitive markets, because employers with these mis-perceptions would be at a competitive disadvantage (Aigner and Cain 1977).
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
Holzer and Neumark: Assessing Affirmative Action
521
lower rewards for workers and a smaller pool of qualified workers for employers. Note that, although the information structure is different than in Lundberg and Startz, the outcome is essentially the same. Minority workers have the same innate distribution of abilities (or costs of investment), but as an outcome of imperfect information coupled with human capital investment decisions, end up with different qualifications on average, and hence different pay as well.63
Coate and Loury also use this model to ask whether affirmative action is likely to break down employers' negative stereotypes (in which case an affirmative action policy would no longer be needed), by leading to an equilibrium in which employers' beliefs that group membership is not associated with qualifications are confirmed. Government intervention may be needed because no single employer in the "negative stereotype equilibrium" has an incentive to behave differently; a single employer deciding to assign more blacks to the more-demanding job would not change the incentives to invest, and would only lower profits by ending up with worse matches. But the equilibrium can potentially be changed by employers acting together, perhaps because of government.64
Coate and Loury conceptualize af-
63 An interesting question that remains unexplored in the existing models is how the conclusions about efficiency change if part of the acquisition of skills occurs on the jobs to which workers are assigned, via training.
64 This is closely related to another potentially important type of intergenerational effect of affirmative action (pointea out by Kevin Lang), specifically the reduction of "social distances" or "transactions costs" between racial and ethnic groups. As employers and others are forced to learn how to evaluate individuals from these different groups in order to find qualified candidates, the "costs" of inter-group interactions (stemming, for example, from miscommunication or stereotypes) may fall. Such changes may have a public good component, making it unprofitable for a sin-
firmative action as requiring that the rate of assignment to the more-demanding job be the same for the minority and majority groups. This becomes a binding constraint on the employers' job assignment problem if employers initially hold different beliefs about the productivity of workers in different groups. They then ask whether under affirmative action there is an equilibrium in which employers hold homogeneous beliefs about the productivity of majority and minority workers, and whether all equilibria under affirmative action satisfy this condition. If this is true, then affirmative action moves the economy to an equilibrium with homogeneous beliefs, and upon its removal the economy stays there. Since the initial negative stereotypes are inefficient, affirmative action would increase efficiency by eliminating negative stereotypes.
The predictions of the theory are ambiguous. There are equilibria under affirmative action satisfying these criteria, but there are also equilibria in which negative stereotypes will not be eliminated, and may be worsened (resulting in a larger skill gap). Intuitively, affirmative action in this model affects the standard that is used to allocate people to different jobs. Given that the two groups may test differently, the standard need not be the same for the two groups, since it is the results (i.e., job assignments) that are constrained. When affirmative action lowers the standard for the group about whom the negative stereotype is initially held, that group may respond by investing less, thus widening the ex post difference in productivity. Coate and Loury refer to this as a "patronizing equilibrium," because it is caused by employers, believing a group
le employer to change his evaluation methods on is own, but leading to social gains if many employers do so as a result of government policy.
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
522
Journal of Economic Literature, Vol. XXXVIII (September 2000)
TABLE 4
Main Theoretical Results on the Efficiency Effects of Affirmative Action in the Labor Market
Nature of discrimination Effects of affirmative action
Welch (1976)	None, or taste discrimination	Reduce efficiency in production
Lundberg and Startz (1983)	Statistical discrimination (worse information about minorities)	Increase efficiency in human capital investment, decrease efficiency in production, net effect ambiguous
Milgrom and Oster (1987)	Promotion discrimination, to maintain "Invisibility" of minorities and women	Increase efficiency in production and human capital investment
Lundberg (1991)	Statistical discrimination (worse information about minorities)	Increase efficiency in human capital investment, decrease efficiency in production, net effect ambiguous
Coate and Loury (1993)	Negative stereotype about skills of minorities	Ambiguous
to be less productive, responding to the affirmative action constraint by lowering standards. Whether this patronizing equilibrium emerges depends on a number of things, including: the benefit the worker receives from being placed in the higher-paying job vs. the cost of acquiring skills; the change in the probability of being placed in the higher-paying job conditional on the test (i.e., the change in rationing); and the change in the optimal level of investment as a result of the change in the standard.65
4.1.5 Summary of Results from
Models of Discrimination and Affirmative Action
This concludes our review of theoretical papers on the efficiency effects of affirmative action in the labor market. As Coate and Loury state, the results of their study (and of the Lundberg and Lundberg/Startz studies) is to "give credence to both the hopes of advocates of
65 See also Dean Foster and Rakesh Vohra (1992) for a similar model developed concurrently that illustrates the ambiguity regarding whether affirmative action reduces or increases incentives for investment.
preferential policies and the concerns of critics" (p. 1239). That is, theoretical models of affirmative action types of policies yield ambiguous predictions regarding the efficiency effects of these policies. This conclusion is documented in Table 4, which indicates the fairly wide class of models that lead to ambiguous predictions about the effects of affirmative action. Ultimately, then, it is a challenge for empiricists to assemble evidence that measures efficiency/performance effects of affirmative action, or tests some of the assumptions and conditions that these models suggest are important in determining whether affirmative action increases or decreases efficiency. However, although the theoretical models do not specifically address this point, they seem to suggest that the potential efficiency gains (costs) of affirmative action are likely to be higher (lower) in jobs that are not at the extreme ends of the skill distribution. Affirmative action policies that place less-skilled workers in jobs with very high skill demands are likely to entail relatively stronger losses of productive efficiency, while acquisition of the requisite human capital may not
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
Holzer and Neumark: Assessing Affirmative Action
523
come easily. On the other hand, policies that simply open up more low-skilled jobs to minorities are unlikely to yield efficiency gains on the human capital side, because they are unlikely to have strong effects on human capital investment. While this conclusion is intuitively appealing, it remains to be addressed in the context of a specific theoretical model; at the same time, it may provide some guidance in future empirical work that tries to assess where affirmative action is relatively more helpful or harmful.
4.1.6 Externality Arguments
Aside from models that begin with discrimination in the labor market, there are some papers that suggest the existence of externalities that place us in a world of second-best choices in which a distortionary policy like affirmative action could increase efficiency. One example relates to the earlier discussion of unequal economic resources in black and white communities. Loury (1977, 1981) argues that, when coupled with racial segregation, initially unequal resources can lead to perpetual differences in economic outcomes for blacks and whites under a regime of equal opportunity, even when the distribution of innate abilities is the same for the two races. This occurs because, given racial segregation, low-income black families impose a negative externality on higher-income families in the provision of community public goods such as schools; this externality reduces investments that affect individual productivity, essentially impeding the ability of the higher-income black families to catch up to higher-income white families. In such a case, it is possible that an affirmative action policy that boosts the incomes of blacks could, at least in principle, offset this externality.
The  potential intergenerational ef-
fects of affirmative action to which this argument points may be central. To the best of our knowledge, there is no empirical work that directly addresses the intergenerational effects of affirmative action. However, there is a limited amount of potentially related evidence. First, a long-standing issue in economics and sociology is intergenerational income mobility. Recent research that takes account of measurement error (Gary Solon 1992; David Zimmerman 1992) suggests that intergenerational correlations of father's and son's earnings are on the order of 0.4, and on the order of 0.55 when family income is used. High estimates like these (in contrast to estimates in the 0.2 range from earlier studies) are taken as indicating low mobility. However, they are not usually used to think about the consequences of policy interventions that raise income. With respect to affirmative action, high correlations might be interpreted as a good thing, since they might suggest that income gains from the parents are "passed on" to the children.66 However, it is not clear that individual-level correlations such as these should be used to draw inferences about programs such as affirmative action that presumably raise the incomes of many minority families simultaneously, especially in light of Loury's hypothesis that there are important positive externalities. Positive externalities pose two complications: first, children in families without increases in parents' income also gain, muting any effect of parents' income on their own children; and second, one could easily imagine that the effects of simultaneously raising
66 If the correlations were higher for minority groups, the evidence would correspondingly be more promising. However, the existing research does not disaggregate results by demographic group, because of the small sample sizes involved (in the 300's in the work by Solon and Zimmerman).
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
524       Journal of Economic Literature, Vol. XXXVIII (September 2000)
many incomes in a community are multiplicative, so that the effect of higher parents' incomes in a single family is not the right experiment. In addition, it is not obvious that one would want to use estimates from representative samples to infer the intergenerational effects of income gains that come from affirmative action; the latter income gains may have different characteristics that imply different intergenerational transmission.
There is also some research that might be used to evaluate Loury's argument based on evidence of effects of community characteristics, rather than intergenerational correlations. Corcoran et al. (1992) evaluate the effects of family and community effects on economic status. They find relatively little evidence of effects of community characteristics from men's youth on current economic status (average income, log earnings, log hours, etc.), with the only persistent negative effect attributed to the participation rate in welfare in the respondent's zip code, whereas race, per se, still plays a large role. (See also the papers cited in Section 3.2.) Of course, poor measurement of community characteristics, including omitted variables and insufficient disaggregation, can lead to overstatement of race effects (and understatement of the effects of community characteristics), insofar as blacks tend to be located in areas with more adverse "true" community characteristics, as seems likely.
An alternative type of model with externalities concerns the effects of mentoring. Susan Athey, Christopher Avery, and Peter Zemsky (1998) develop a model in which within-firm mentoring is more productive when mentors are of the same "type" as the mentees; specifically, lower-level workers learn more from mentors of the same race, sex, etc. One implication ■ of this model is that employers will consider a worker's race,
sex, etc., in making decisions regarding whom to promote to positions in which they will serve as mentors. A firm may use a promotion rule biased in favor of minorities, although this need not occur. They also explore the dynamic implications of this model; they find that when history or initial conditions matter and there are multiple equilibria, firms may stay in an equilibrium with little diversity even though per-period profits are higher with more diversity. In this case, a short-run policy of encouraging diversity— such as affirmative action—can move the firm to an equilibrium with higher per-period profits. Once this occurs, the policy can be dropped and the firm will stay in the "high-diversity" equilibrium.
Although this is not a general equilibrium model, and hence overall efficiency effects cannot be evaluated, the model does have two implications which are rather unfortunate with respect to assessing evidence on affirmative action. First, affirmative action may improve firm performance, although this may not occur in the short run while the firm is moving to the new equilibrium. Second, firms may promote minorities or women who are less qualified on some dimensions; the problem is that we may not pick up the increased productivity of mentoring associated with such promotions. Thus, in both cases evidence on firm performance or worker qualifications may be less than fully informative about efficiency effects of affirmative action. This also suggests that research on the nature of mentoring relationships may prove valuable in helping us interpret empirical tests of the effects of affirmative action.
A related possibility that we think merits some exploration is the potential externality entailed in role-model effects. If minority or female role models help to open up new occupations or sectors  to  minorities  or women (by
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
Holzer and Neumark: Assessing Affirmative Action
525
influencing, for example, the possibility that they would consider entering such an occupation), then there is a positive benefit derived from early minority and female entrants that the individual cannot capture.67 This may provide another motivation for affirmative action, which may reduce the expected costs that these early entrants will have to pay.68
4.2 The Efficiency Effects of Affirmative Action in Education: Theory
When we turn to the analysis of the efficiency effects of affirmative action in education, we can adopt a similar perspective on alternative policies to that which we adopted in looking at labor markets. As before, we continue to view affirmative action broadly, as any policy that makes race or sex a factor in admissions; some of these were discussed earlier. Counterposed to affirmative action might be a strict antidiscrimination policy, as reflected in Title IX and Title VI of the Civil Rights Act.69 Whereas in the labor market the application of a disparate impact standard to equal employment opportunity blurred the lines between antidiscrimination laws and affirmative action, the absence of such a standard with respect to educational institutions leaves the distinction sharper.70
67 For an example of this argument with respect to the legal profession, see Oko (1996).
68 In the context of hiring by educational institutions, this could also be considered as a potential argument for affirmative action in graduate school admissions, since such admissions serve as the gateway to being trained as a professor. We therefore treat this issue of academic hiring in our discussion of affirmative action in education, rather than the labor market.
69 Title IX bars sex discrimination in any educational programs or activities receiving federal funding. Title VI more generally bars discrimination on the grounds of race, color, or national origin in any program or activity that receives federal funding.
70 Bowen and Bok (1998) state that beginning in the 1970s "federal officials had incorporated re-
Reflecting the policy debate, this section is couched in terms of affirmative action for minorities, and not for women. There are, indeed, some affirmative action programs for women in education, such as the National Science Foundation programs referred to in Section 2.2, extra funding opportunities for female graduate students pursuing Ph.D.'s, and mentoring programs (see the references in Neumark and Rosella Gardecki 1998). But our sense is that most of the attention paid to affirmative action in education refers to minorities; the lawsuits, state referenda, etc., which have addressed affirmative action in education seem geared exclusively to the minority/nonminority distinction, presumably because most affirmative action programs are targeted toward minorities.
In addition, the discussion in this section and when we turn to the empirical evidence centers on affirmative action in admissions. Our sense is that the debate over affirmative action in education focuses on admissions, not financial aid. For example, The Shape of the River (Bowen and Bok 1998) makes no mention of affirmative action influencing financial aid decisions, even though it tries to confront head-on all of the issues that arise in the affirmative action debate. We believe that this is because, at the undergraduate level at least, explicit programs directing financial aid toward minority students are not widespread, although there are some isolated programs like those discussed in Section 2.2. However, this is to some extent an under-investigated issue, and our first-hand experience at
ports on student enrollment into the affirmative action plans of universities" (p. 8). However, it is unclear if any actions were taken based on these reports; indeed, we have been unable to find any other documentation of the existence of such reports.
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
526       Journal of Economic Literature, Vol. XXXVIII (September 2000)
the graduate level—where financial aid decisions are much less closely related to financial need—suggests that decisions regarding financial support based on demographic characteristics may be more pervasive.
4.2.1 Thinking about "Preferential Admissions"
In order to analyze the potential efficiency implications of affirmative action in college or university admissions— whether at the undergraduate or graduate level—we must consider the processes by which students might be admitted under a variety of assumptions, and decide whether or not such a process would be more or less efficient in the absence of affirmative action considerations. This is not an easy or clear-cut task; we have found little if any discussion in the literature of socially optimal admissions policies of universities, and a full-blown treatment of this issue is beyond the scope of this paper.71 However, consideration of the issues involved leads to two conclusions:
a) Even in the absence of affirmative action, few universities would simply admit students above certain cutoffs of grades and test scores, which appears to be the "alternative" preferred by critics of affirmative action.
b) If universities did so, it would be unlikely to be socially efficient.
It is useful to begin by considering the outlines of what a "socially optimal" matching of students to universities would look like in a fairly simple and abstract setting, and whether or not affirmative action in admissions would necessarily  result  in  a  less optimal
71 The only paper we have found that even attempts to look at various admissions policies from the perspective of economic efficiency is Conrad and Sharpe (1996). Gordon Winston (1999) discusses some of the unique features of higher education that make standard welfare or efficiency analysis extremely difficult.
matching. In the abstract, we might assume that optimal matching would result from strict price-rationing of university positions, rather than rationing of slots based on grades and test scores. Under a price-rationing scheme, those who most value attending any particular university would be willing to pay the most for admission. The resulting equilibrium across a differentiated set of universities would maximize social welfare. This assumes, of course, the existence of no other major market failures, such as capital market constraints, externalities, etc.; we will return to these issues below.
However, even aside from sorting on other institutional characteristics, it is not clear that returns from education across students are strongly positively correlated with previous grades and test scores, so it is not apparent that this simplistic social optimum would resemble what results from simple cutoffs based on grades and test scores. If there is a positive monotonic relationship between private returns and qualifications, and a negative monotonic relationship between qualifications and the costs of education (by which we mean the costs of "learning," rather than financial costs), then university admissions procedures based on admitting everyone above some threshold of qualifications may mimic the socially efficient outcome. But these are strong assumptions that need not hold.72
The discussion to this point assumes
72 For example, Zvi Griliches (1977) develops a model in which higher ability individuals invest less in schooling than lower-ability individuals. Rod Garratt and John Marshall (1994) present a model in which a cutoff on achievement is optimal, but this rests on the assumption that the returns to college are increasing in achievement. Other empirical work (e.g., McKin-ley Blackburn and Neumark 1993; Cawley, Heck-man, and Edward Vytlacil 1998) explore whether returns to schooling are higher for higher-ability individuals.
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
Holzer and Neumark: Assessing Affirmative Action
527
that universities have no preferences over which students to accept, and that students' preferences are independent of other students' attendance decisions. In reality, neither of these conditions holds. Both students and universities care about university quality, which depends to some extent on the academic quality of students admitted (and on the quality of faculty, physical facilities, etc.). Students may care about the quality of the institution because its reputation enters into their utility functions, or because the quality of education that they receive depends on the abilities of their fellow students.73 The latter implies the existence of educational externalities in the classroom, a point to which we return below. Therefore, it makes sense that universities would ration their slots at least partly on the basis of student grades and test scores, as well as on the basis of price. In addition, of course, it is apparent that universities are for institutional or other reasons constrained from using price rationing.74
But even if quality of students is the central concern of universities and other students, socially optimal matching would not necessarily entail universities relying exclusively on grades and test  scores  for  admissions.  For one
73 Rothschild and Lawrence White (1995) analyze the education market when students are consumers as well as inputs in the production process. This and other issues related to the market for higher education are reviewed in Caroline Hoxby (1997).
74 They do not, for example, appear to charge higher prices to lower-quality students who might impose negative externalities, although perhaps one could argue that legacy students, who are probably on average less qualified but also wealthier and less likely to receive scholarships or financial aid, are paying a higher price for this reason. They also do not vary tuition across classes based on variation in educational costs, with the possible exception of higher tuition for upperclassmen— who are generally in smaller classes—at some universities.
thing, these are imperfect measures of student quality. Thus, even highly competitive graduate and professional schools supplement the use of grades and test scores in the admissions process with consideration of personal essays, interviews, and the like, by which they presumably gain more information about student motivation, creativity, or intellectual depth. The result is likely that these admissions programs may improve matching, by admitting some students with lower grades and scores over those with higher ones, but with other positive characteristics.75
Aside from multiple characteristics providing information on student quality, university administrators behave as if diversity in student attributes—especially among undergraduates—is itself important to the mission of the university (e.g., Bowen and Bok 1998), although in our view this remains to be established by the evidence. One could think of this in terms of the educational value of a diverse student body—discussed in the next subsection—or alternatively as universities optimizing over a multi-dimensional objective function. For example, the most prestigious universities appear to value having the political leaders of poorer or more rural states among their alumni, along with professional athletes, actors, musicians, etc. They therefore seem to value a wide range of characteristics and talents among their undergraduates, including some vaguely defined sense of future leadership potential. Affirmative action
75 Although not central to the argument here, if the predictive power of grades and test scores with respect to expected quality is lower for minorities, it might be optimal to allow admissions procedures to vary across these racial groups in ways consistent with affirmative action. There is a large set of questions concerning the definition of merit, the reliability of tests, etc., into which we do not delve in this section. Rather, we consider the economic issues that arise conditional on having a measure of merit or qualifications.
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
528       Journal of Economic Literature, Vol. XXXVIII (September 2000)
may be used out of a similar motivation, as those alumni who achieve positions of leadership and visibility within particular ethnic communities may be similarly valued. In this environment grades and test scores receive more or less weight in the admissions procedures of different schools relative to many other student attributes, resulting in a complex matching process between schools and students.
4.2.2 Diversity
In addition to these motivations for adopting affirmative action, many educators believe that diversity enhances the quality of the education that undergraduates receive, as they learn from each others' experiences and perspectives. This argument is summarized in the following statement by the presidents of the American Association of University Presidents:
We believe that our students benefit significantly from education that takes place within a diverse setting. In the course of their university education, our students encounter and learn from others who have backgrounds and characteristics very different from their own. As we seek to prepare students for life in the twenty-first century, the educational value of such encounters will become more important, not less, than in the past. (Cited in Bowen and Bok 1998, p. 252.)
This argument—like the earlier argument about why student quality is important—implicitly recognizes that students are not merely consumers of educational services, but are themselves inputs into the educational process. The gist of this argument is then that positive externalities from adding a minority student (especially when such students are underrepresented) may be larger than those from adding a nonminority student. Of course, a similar motivation may drive universities to seek out diversity along other dimensions, including
geography, talents, interests, etc. Thus, many colleges and universities appear to choose some degree of affirmative action in their admissions processes, even absent any pressure from the Federal government (which, as we noted above, may not be a major consideration in admissions) or from other interest groups.76 Whether they choose the socially optimal degree is difficult to assess, without a much clearer sense of what an optimal set of admissions procedures would be. Moreover, whether this diversity does improve educational quality is, in our view, a critical question to examine empirically.
4.2.3 Imperfections in the "Market" for College Admissions
Imperfections in the "market" for college admissions may also be generating departures from efficient behavior that could be reduced by affirmative action practices. These include capital market imperfections and externalities for various communities, as described below:
a. Capital Market Imperfections. One argument offered in support of possible efficiency-enhancing effects of affirmative action in education is that capital market imperfections prevent lower-income individuals from taking advantage of a college education (Loury 1977).77 In terms of the framework outlined above, although the social benefit of education of some of these lower-income individuals exceeds that of the education of higher-income individuals who receive a college education, the private costs for the former are higher, and hence sub-optimal education decisions
76 Pressure to engage in affirmative action in admissions may come from heavily politicized faculty members or from other organizations, if not from the government per se.
77 One possible reflection of this is the generally higher returns to education among blacks than whites (e.g., John Bound and Richard Freeman 1992).
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
Holzer and Neumark: Assessing Affirmative Action 529
are made.78 While it is true that minorities are more likely to come from low-income families, if access to capital is solely a function of socioeconomic background, this argument really pertains to income rather than race. As such, this argument might be viewed as providing a basis for granting preferential treatment to low-income students, rather than minorities per se (i.e., class-based rather than race-based preferences). On the other hand, access to capital may be limited by race discrimination. In addition, one might argue that the capital market problem should be addressed via the capital market itself—by either rooting out imperfections or targeting discrimination—rather than via preference in admissions, which seems like a blunt tool, especially to the extent that it is race-based and may benefit higher-income minorities who do not face capital market barriers. For example, student loans, scholarships, etc., are forms of intervention in capital markets, and no doubt disproportionately favor lower-income students.
b. Community Externalities. The second externality argument concerns the positive externalities that educated women, minorities, etc., may offer to some communities. An example is the argument that black doctors will go back to their community to practice, and that this might be worth encouraging even if the cost is the denial of a place in medical school to a more-qualified white candidate. This argument has two requirements (putting aside capital market issues). The first is that the social—but not the private—return to practicing medicine in the black (i.e., poorer) community is higher. Given possible imperfections in the health care market, this is conceivable. The
78 Of course, there may be valid reasons for credit to be more expensive for children from lower-income families.
second is that blacks are actually more likely to do this. Aside from the empirical evidence on either of these questions, one could again argue that this is primarily an issue of income, not race. Also, one might wonder whether there are other mechanisms to increase specific types of professional care for poorer communities.79 Externalities may be conveyed in other ways, such as mentoring, role-model effects, etc., for which the social returns cannot be captured privately.
4.2.4 Arguments against Affirmative
Action: The Fit, Underperformance, and Stigma Hypotheses
Of course, other criticisms of affirmative action in education have been advanced that for the most part do not explicitly (but perhaps implicitly) address the issue of efficiency. Dinesh D'Souza (1991) has articulated what Bowen and Bok label the "fit" hypothesis, arguing that the educational experience of minorities has suffered as university administrators have recruited and admitted such students into environments where they cannot compete. Instead, he suggests, they might be better off in universities where they can "compete against evenly-matched peers" (p. 43). This parallels another prominent argument against affirmative action, namely the "stigma" hypothesis proffered by Charles Murray (1994), who suggested that the "evil of preferential treatment ... [is that it] perpetuates the impression  of inferiority"   (p.   207). Steele
79 For example, the National Health Services Corps Scholarship Program offers tuition or loan repayments for medical school in return for primary care service in underserved areas ("health professional shortage areas"), which include both rural and urban settings (Fitzhugh Mullan 1997). However, although Congress has mandated preferential consideration for minority applicants to the program, rural placements predominate (Donald Pathman and Thomas Konrad 1996).
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
530       Journal of Economic Literature, Vol. XXXVIII (September 2000)
(1990) subscribes to a similar view. He suggests that the "first-order" effect is that preferential treatment for blacks lowers the incentives for high efforts that might lead to some of the achievements otherwise attained because of preferences—or as he puts it, "the hidden incentive not to do what we believe preferences will do" (p. 119). Moreover, he argues that this is reinforced by psychological repercussions, as "the effect of preferential treatment—the lowering of normal standards to increase black representation—puts blacks at war with an expanding realm of debilitating doubt, so that the doubt itself . . . undermines their ability to perform, especially in integrated situations" (pp. 117-18).
Roughly speaking, there is a common implication of all of these hypotheses. Specifically, minority students who apparently benefit from preferential treatment will have worse achievements and performance than otherwise comparable nonminority students. It is worth pointing out, though, that arguments that suggest lower effort on the part of beneficiaries in response to lower standards are not necessarily correct. Like in the labor market, there may be some reasons for thinking that university admissions are best characterized as rank-order tournaments.80 Andrew Schotter and Keith Weigelt (1992) have shown that in such tournaments, if a group (such as minorities) faces a cost disadvantage to attaining a particular performance level, then affirmative action programs can, under some conditions, increase effort of the minority group. This is more likely to occur when the
80 While labor markets may mimic the efficient allocations that would occur in rank-order tournaments (e.g., Edward Lazear and Sherwin Rosen 1981), universities may not be able to do this, since they do not really control the rewards to college admission for the "winners."
cost disadvantage is sizable, as the affirmative action program causes some of its members to "participate" in the tournament rather than to drop out. To some extent, this hinges on the long-running dispute over whether lower test scores, for example, of some minority groups reflect cost disadvantages (such as inferior schooling) or ability disadvantages. We do not mean to suggest that university admissions should necessarily be regarded as rank-order tournaments. But this work does point out that, to the extent that cost disadvantages do play a role in generating differences in the academic achievements used in university admissions, preferential admissions may not lower and could raise incentives for achievement.
4.3 The Efficiency Effects of Affirmative Action in Government Contracting/Procurement: Theory
As with the empirical evidence on discrimination, and later the evidence on efficiency/performance effects of affirmative action, there is far less work on government contracting/procurement programs than on employment. Indeed, there are virtually no theoretical treatments of affirmative action in contracting/procurement and its possible effects, although some issues have been raised in the literature.
If, as suggested by Chief Justice Burger in Fullilove, government agencies were discriminating in not giving business to minority- or women-owned firms, then affirmative action could increase efficiency by awarding contracts to lower-cost or higher-quality women-or minority-owned firms. In addition, the encouragement of minority ownership may spur minority employment, as there is some evidence suggesting that minority owners are more likely to employ minority workers (although this evidence is not directly linked to set-asides;
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
Holzer and Neumark: Assessing Affirmative Action 531
Bates 1994).81 This is not an efficiency argument per se, but given severe employment and crime problems in some minority neighborhoods, higher employment may entail positive externalities by discouraging non-market activities that have negative effects on others. Aside from any externality argument, Bates (1993, ch. 1) also points out that antidiscrimination laws and affirmative action in hiring are targeted toward large businesses partly by law and partly because the costs and benefits of enforcement encourage a focus on large employers. Thus, Bates argues, encouraging minority ownership may be a complementary policy that promotes minority employment in small businesses.82
On the other hand, researchers have identified a few problems that may be created by using affirmative action in contracting and procurement to try to help develop women- and minority-owned businesses. One potential problem is that overall goals for shares of contracts going to minority- or women-owned businesses may lead government agencies to concentrate their contracting in fields where there are already a large number of such firms. This is not necessarily problematic if for some reason (such as geographic concentration or marketing skills) minorities or women have comparative advantages in these fields. But the relative ease of achieving overall contracting goals by concentrating efforts in fields that already have large shares of minority- or
81 Bates (1993) reports that this pattern persists for white-owned businesses in minority neighborhoods and minority-owned businesses in nonmi-nority neighborhoods, so these differences in hiring patterns are not solely a reflection of the racial composition of the local labor market.
82 Although Bates recognizes that jobs with large employers are preferable, he notes that small businesses have been and will continue to be a potentially important source of minority employment in urban areas.
women-owned businesses could be disadvantageous if it fails to encourage the development of businesses in other fields, especially if this reinforces historical—possibly discriminatory—forces that initially led to the existing distribution of minority- or women-owned businesses. Second, Bates and Williams (1996) have suggested that in some environments, set-aside programs for government contracts may lead business owners to overextend themselves, leading to a higher rate of failure. Third, these same authors have raised the possibility that programs for minority-owned businesses may have encouraged the formation of front companies for white firms that provide nothing but fees and perhaps a little income for those who represent the front company.83 Such fraudulent behavior is likely to introduce inefficiencies because it results in contracts being awarded for reasons unrelated to either the cost or quality of the goods and services provided, or the goals of affirmative action in contracting and procurement. Fourth, government preferences in awarding contracts may prop up weak companies, or breed dependence on government contracts, rather than furthering the overall goal of creating more independent women- and minority-owned businesses. Finally, by restricting competition and in some cases using explicit bid-price preferences, set-asides may raise the cost of government purchases of goods and services.84
One might argue that the true goal of affirmative action in contracting/procurement is not to raise the share of
83 Thomas Denes (1997) suggests a similar problem for general small business set-asides under the 1953 Small Business Act.
84 Allan Corns and Schotter (1999) show that bid-price preference programs may under some conditions reduce rather than increase costs, because of their effects on the equilibrium behavior of bidders.
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
532       Journal of Economic Literature, Vol. XXXVIII (September 2000)
minority business with the government, but rather to boost minority (and female) entrepreneurship. An important barrier to this may be discrimination in credit markets, in which case increased contracting with the government may not be the best remedy. Instead, it may be preferable to combat the credit market discrimination directly. It is likely that standard discrimination in credit markets generates inefficiencies, although, as has turned out to be the case in theoretical analyses of labor markets, the question is probably more complex. We have uncovered only one study that looks at these questions. Loury and Andrew Weiss (1998) develop a model in which an original perceived (or believed) difference in investment behavior of blacks and whites can lead to an equilibrium in which whites receive what amounts to preferential treatment in credit markets, even in the absence of differences in ability or wealth or outright discriminatory tastes. The essence of this model is that there may be historical reasons for the perceived difference in investment behavior, but even if that difference no longer exists ex ante, it exists ex post in equilibrium; in other words, blacks get stuck in a "bad equilibrium." In this case, it is possible that government intervention requiring equal treatment of blacks and whites in credit markets can move blacks to the "good equilibrium" in which investment behavior is the same as whites. However, other outcomes are also possible—such as pushing whites to the bad equilibrium—so that the implications for overall output (and efficiency) are ambiguous.
4.4 Efficiency/Performance Effects in the Labor Market: Evidence
Theoretical work on the effects of affirmative action in the labor market identifies a number of potential chan-
nels for efficiency effects, including influences on the qualifications of hires and on production costs, impacts on human capital investment, changes in information or beliefs about minority and female applicants, and externalities. Because the potential effects of affirmative action on the productive efficiency of firms depends in large part on how affirmative action affects employee or establishment productivity and performance per unit of labor cost, most research has focused on this question in one way or another. However, because there are many settings in which it is not feasible to estimate directly the effects on productivity or costs, many empirical strategies aside from production or cost function estimation have been used, including: examination of company financial data; drawing inferences from employee performance ratings or attitudes; empirical or qualitative studies of specific sectors or firms; and studies of employee selection procedures. This evidence is described in the following subsections, and summarized in Table 5.
4.4.1 Production/Cost Function Estimates
For economists, estimates of production or cost functions should be the most straightforward way to infer the effects of affirmative action on worker or establishment productivity or costs, when the appropriate data are available. Leonard (1984a) estimated production functions at the level of state-by-two-digit industry cells, using data from the Census of Manufactures and the Annual Survey of Manufactures in the 1970s. He augments the labor input in the production function to include information on the fraction of establishments that are federal contractors, in order to estimate whether productivity of labor is lower in such establishments (actually,
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
Holzer and Neumark: Assessing Affirmative Action 533
TABLE 5
Summary of Studies on Effects of Affirmative Action in the Labor Market on Efficiency/Performance
Studies
Data
Results/Conclusions
Comments
1. Production/Cost Function Estimates
Leonard 1984 State x industry cells, Cen-
sus of Manufactures, Annual Survey of Manufactures; EEO-1 averages
Griffin 1992 EEO-1 merged with Com-
Griffin et al. 1996 pustat; imputed wages
from Population Census
2. Company Financial Data
Wright et al. 1995 Daily stock valuations for
34 companies
3. Employee Performance Ratings
Holzer and Neumark     Survey of employers in 4
1999 and 2000 metropolitan areas
Lewis 1997
1% sample of federal personnel records
4. Analyses of Specific Sectors
A. Police Steel & Lovrich 1987 Lovrich & Steel 1983
Lott 1997
Carter and Sapp 1991
Gottfredson 1996
254 cities:
Crime rates, arrests, demographics of police
Homicide rates and police demographics for cities
Survey of 700 police depts., large cities, and counties, and site visits to selected depts.
Nassau County Police Dept. test results, studies, and description
No negative effects of contractor status.
Contractors have 6.5 percent higher costs.
Announcement of award for exemplary affirmative action practices raises company value.
Minorities have lower educational attainment but generally not lower performance ratings under affirmative action. Lower performance evident only for Hispanic males, and in establishments using affirmative action in hiring rather than recruiting.
Women receive higher ratings than men; minorities' ratings are a bit lower (within grade).
No relationship between police demographics and crime/police outcomes.
Affirmative action for blacks raises homicide rates.
Black and female police officers more likely to have BAs, education in bona fide occupational qualification.
Attempts to reduce disparate impact reduced importance of cognitive test, reducing test validity.
Highly aggregated data.
Questionable wage imputations, implausible own labor demand elasticities.
Small and self-selected sample, questionable interpretations
Self-reported EEO/ affirmative action status, subjective outcomes.
Subjective outcomes not necessarily linked to affirmative action.
Affirmative action defined on basis of police demographics.
Unusual sample of cities: questionable specification and use of instrumental variables.
Descriptive study, little description of actual evidence.
No info, on performance of recruits under different testing regimes.
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
534
Journal of Economic Literature, Vol. XXXVIII (September 2000)
TABLE 5 (Cont.)
Studies Data Results/Conclusions Comments
B. Academics Kolpin and Singell 1996
Barbezat 1989
Elmore and Blackburn 1983
Espinosa 1992
Economic Departments: Guide to Graduate Study in Economics; Economists: AEA Directory
Carnegie Council Surveys of American Professoriate
Surveys at Big Ten universities
Demographic information and affirmative action reports for California city
Women had better publication records than men in top departments in 1970s.
Women paid less than men, controlling for publications; blacks paid more than whites and have fewer publications.
Whites and blacks have comparable publication records.
Recruitment and training efforts lead to affirmative action success.
Recruitment training, leadership commitment, incentives for managers lead to affirmative action success.
Recruitment is important for affirmative action success.
Leadership commitment, monitoring of performance, participation of women, and more extensive recruitment lead to affirmative action success.
Goals and timetables not met; exaggerated progress.
Data for 1970s. Quality of publications?
Data for 1970s. Quality of publications?
Quality of publications?
Qualitative study.
Qualitative study.
Qualitative study.
Qualitative study. Measures of "success"?
Descriptive study.
5. Case Studies/Institutional Evidence
Badgett 1995 One large nonunion firm
Vernon-Gerstenfeld 9 large companies
and Burke 1985
Stoops 1982 Houston Police Dept.
Hyer 1985 3 universities
in the state-by-industry cells). Leonard found no negative effects on productivity of the presence of contractors in a cell; indeed the point estimate was positive. While the aggregate nature of Leonard's data likely decreases their ability to draw strong inferences, Leonard points out (1984a, footnote 27) that in a similar type of study using his
data, but with state-by-industry information on OSHA regulations, evidence of significant negative productivity effects of such regulations was found, indicating that his data and experiment appear capable of detecting negative productivity effects of regulation.
More recently, Peter Griffin (1992) estimated translog cost functions at the
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
Holzer and Neumark: Assessing Affirmative Action
535
firm level, using EEO-1 data merged with Compustat data. His results suggest that the constraints imposed on the labor demand choices of contractors raised their labor costs by roughly 6.5 percent relative to those of non-contractors. Moreover, in accordance with the LeChatelier principle, he finds that firms in the contracting sector have less elastic labor demands. However, there are two potential reservations regarding this conclusion. First, data on wages had to be imputed, based on race/sex/occupational cells within industries from the 1980 Census of Population. As Griffin notes, the use of his wage index as a firm-level labor cost measure requires the assumption that input markets are perfectly competitive (p. 254). The problem this poses is that wage differences by industry, union status, etc., are attributed causally to contractor status. Given that contractors are likely to be in high-wage industries, such as construction, this imputation procedure may well lead to upward-biased estimates of cost differences attributable to affirmative action.85 Second, the very large elasticities of demand with respect to own wages estimated for these establishments (generally 1-2 for contractors and 2-5 for non-contractors) create some doubt about these results (see Daniel Hamermesh 1993). Finally, a potential problem with both the Griffin and Leonard studies, but one that is impossible to resolve without a true experiment, is that contractor and non-contractor firms could produce different mixes of goods, which could give rise to production function differences. However, it is not obvious that this would bias the results in a particular direction, especially when—as in the Griffin paper—parameters such as the
85 This point is acknowledged explicitly in Griffin, Arthur Getis, and Ernst Griffin (1996, footnote 1).
elasticity of demand for labor and elasticities of substitution are studied. Also, the Leonard paper estimates differenced forms of production functions, so this would matter only if the mix of goods changed as a result of the establishment of affirmative action for federal contractors.
Overall, while the approach is promising and well-motivated by economic theory, the use of production and cost function estimates to infer affirmative action effects on productivity has so far generated inconclusive results.
4.4.2 Company Financial Data
Another method which labor economists use to infer the effects of workplace practices on productivity is the analysis of company-level financial data. Applying this method to the study of affirmative action, Peter Wright et al. (1995) carried out an event study that analyzes the responsiveness of stock prices (relative to the overall market) to announcements that companies had been cited by the Department of Labor for "Exemplary Voluntary Efforts" on affirmative action. Looking at 90-day periods that precede and follow such announcements, they find that stock prices react positively and significantly to such announcements, and conclude that the market views affirmative action as having positive effects on companies' profits. However, questions remain about the extent to which this result would generalize to a larger sample of establishments and to involuntary efforts based on contractor status.86 There are also questions about the nature of the "event." In particular, the authors also find negative effects of announcements
86 The sample analyzed contained only 34 companies. Voluntary self-selection of affirmative action efforts might imply much lower efficiency costs than those imposed on an establishment by their "contractor status."
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
536       Journal of Economic Literature, Vol. XXXVIII (September 2000)
of discrimination settlements on stock prices (also see Joni Hersch 1991), suggesting that the positive effects of announcements of exemplary affirmative action efforts may simply provide information on "immunity" from discrimination suits. On the other hand, although the awards presumably confirm practices already in place, they can still convey information about these practices by reducing the uncertainty associated with this assessment. If discrimination is inefficient, reflecting non-profit-maximizing behavior, then the "event" that reveals nondiscriminatory behavior to investors might be expected to result in higher rather than lower stock prices.
4.4.3 Employee Performance Ratings and Attitudes
An alternative approach examines performance ratings of individual employees in affirmative action vs. non-affirmative action settings. Because these ratings are subjective and person-specific, they may contain classical measurement error across establishments and employees, and possibly systematic biases in favor of or against specific race/sex groups. Still, there is reason to believe that these performance ratings contain useful information on performance differences across individual employees.87
87 In a National Research Council report, George Milkovich and Alexandra Wigdor (1991) argued that performance ratings by supervisors generally do not appear to be seriously biased by race and sex, based on evidence of positive correlations between ratings and both ability and objective performance measures, and on evidence of only moderate effects of race and sex on performance ratings (Gregory Lewis 1997). However, Lewis contends that this is too strong a statement, as there is some evidence of bias in performance ratings against blacks and women. This suggests that cross-sectional studies may be biased toward finding worse performance for these groups. Kath-erine Abraham and James Medoff (1981) were among the first economists to use these ratings as proxies for individual productivity.
This individual-level empirical approach is of particular value for another reason. The evidence reviewed in the previous two subsections looks for effects of affirmative action at the aggregate (establishment or firm) level. It is entirely possible, though, that affirmative action affects a relatively small fraction of the workforce, and that differences in qualifications or performance are sufficiently small that it is very difficult to detect any effects at the firm or establishment level. In itself, this may be valuable information from a policy perspective, from which effects on establishments or firms may be of paramount concern. But perhaps more from the point of view of understanding behavioral responses to affirmative action, establishment- and firm-level analyses may obscure important results. In contrast, data on individual-level qualifications and performance can be more revealing about these responses. Finally, if the individual-level results tend to point in the same direction as the establishment- or firm-level results in failing to detect deleterious efficiency or performance effects of affirmative action, the latter results would be strengthened.
In one set of papers, Holzer and Neumark (1999, 2000) use a difference-in-differences framework to analyze supervisory performance ratings of white male and minority/female employees in establishments that do and do not practice affirmative action.88 The difference in the gap between white males and other groups between the two sectors constitutes the difference attributable to affirmative action. They find that, on the one hand, the qualifications of minorities in affirmative action establishments (as measured by educational
88 The data focus on the last worker hired in the establishment, and performance is measured on a 0-100 scale.
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
Holzer and Neumark: Assessing Affirmative Action
537
attainment) are lower than in non-affirmative action establishments. But, on the other hand, there was little evidence of lower performance ratings of women/minorities in establishments using affirmative action (with the possible exception of Hispanic males). In the study that looked at affirmative action in recruiting and in hiring, to the extent that performance was lower in establishments engaging in affirmative action, these findings were limited to establishments that used affirmative action in hiring as opposed to recruitment only. In contrast, among the broader set of establishments using affirmative action in recruiting, the results indicate that minorities and women in these establishments perform, if anything, better than white males; the ratings differentials are positive for all groups, and significant for black males (at the ten-percent level) and black females.
The findings also indicate that these establishments engage in a variety of human resources activities that apparently help them find minority/female employees for whom some easily observable qualifications (such as education) may be somewhat limited but whose performance is up to standards. For instance, establishments engaging in affirmative action recruit more extensively, attract more black applicants, pay less attention to various negative personal attributes during the screening process (perhaps because they uncover other, more pertinent or compensating information), rely more heavily on formal performance evaluations, and provide more hours of training to their new employees, than do non-affirmative action establishments.
These results are consistent with affirmative action establishments either finding minority applicants with above-average unobservable characteristics, or
managing to offset the observable deficiencies through extra feedback and training. Note that this extra training is consistent with the prediction of the Lundberg (1991) model of affirmative action. On the other hand, if more intensive recruiting is largely responsible for the absence of lower performance among minorities hired under affirmative action, we have to be careful in drawing conclusions for the labor market as a whole, as opposed to a subset of firms. In particular, if the supplies of skills or abilities available in the market are fixed, and if affirmative action became more widespread (with more firms recruiting more extensively), it might become more difficult to find minority workers with characteristics that offset some of their lower qualifications. If, instead, the supplies of skills or abilities respond positively to affirmative action (by no means a necessity, as shown by Coate and Loury 1993), then the results may be more applicable to the labor market as a whole. However, the latter inference cannot be drawn from the available studies.
These results are plagued by some statistical questions, including uncertainty over how the affirmative action status of establishments is defined and potential unmeasured differences in ratings across these establishments.89 But the results of the studies are largely borne out by those of Lewis (1997), who analyzed differences by race/sex in performance ratings of workers in a one-percent sample of all federal personnel
89 The difference-in-differences estimation enabled us to eliminate average differences in ratings by race/sex in the non-affirmative action sector (which should remove any influence of bias in ratings, unless this bias differs in the two sectors), while fixed effects associated with individual supervisors across establishments are eliminated by using the deviation in performance ratings between the last person hired and the "typical employee in this position."
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
538
Journal of Economic Literature, Vol. XXXVIII (September 2000)
records. Given that the federal government uses affirmative action in hiring, any observed differences in personnel ratings by race/sex might well be attributable to these practices, and the focus on one such very large employer eliminates concerns about unobserved heterogeneity across employers with different race/sex preferences and hiring practices.90 His results show that, within occupational grade and controlling for differences in education and experience, white women are significantly more likely to receive "outstanding" performance ratings than white men (Table 2, column 4). Black and Hispanic women are no more likely to receive such ratings than white men, indicating, in turn, that they are less likely to receive them than white women. The point estimates also indicate that black and Hispanic men are less likely to receive outstanding ratings than white men, although these differences are not statistically significant. Lewis obtains parallel results for the probability of receiving low ratings. Thus, Lewis's results correspond with those found by Holzer and Neumark in their analysis of mostly private-sector establishments. There is no evidence that white women hired under affirmative action perform worse, and weak evidence at best that minorities do so.91
90 Even without affirmative action, there are likely to be differences in average performance across race groups that reflect differences in educational quality and other measures of background. Thus, any estimated differences in performance ratings by race are likely to be upper bounds for the effects associated with affirmative action. The difference-in-differences estimation in the Holzer-Neumark papers overcomes this problem.
91 Studies in psychology point to some potential problems with subjective assessments of hires if they are explicitly identified as "affirmative action hires." This literature, which is summarized in Campbell (1996), shows that the attitudes expressed towards affirmative action hiring and the hires themselves can vary widely depending upon
4.4.4 Analyses of Specific Sectors
The effects of affirmative action on outcomes have been studied most extensively for two groups of workers: police and academics.92
a. Police. The representation of blacks on the police forces of many large cities has grown dramatically since the 1970s, particularly as minorities have become much larger fractions of
how affirmative action is implemented. This research has the further implication that productivity or performance of workers hired under affirmative action may be influenced by how it is implemented. For example, Madeleine Heilman (1997) and Heilman, Caryn Block, and Jonathan Lucas (1992) argue that labeling individuals as "affirmative action hires," with no qualifications or caveats, almost certainly stigmatizes minority/female applicants, and perhaps even reduces their own perceptions of their qualifications and productivity. On the other hand, these negative perceptions can be avoided when respondents are provided more information about the possible presence of discrimination in the absence of affirmative action and about the actual selection process (e.g., Laura Graves and Gary Powell 1994; Alison Konrad and Frank Linnehan 1995; Francine Tougas et al. 1996).
Research in management comes at essentially the same issue from the perspective of "procedural justice." This research suggests that employee perceptions of employer "fairness" have positive implications for firms by influencing workers' behavior (J. Greenberg 1990; Joanne Leek, David Saunders, and Micheline Charbon-neau 1996; Daniel Skarlicki and Folger 1997). An important component of this perceived fairness is procedural justice, which refers to the fairness of the means used to determine what employees receive (as opposed to distributive justice, which refers to the fairness of the distribution of what is actually received (Folger and Konovsky 1989; Skarlicki and Folger 1997)). However, while some research using hypothetical examples presented to test subjects suggests that how managers explain affirmative action-type decisions may influence perceived procedural justice (Ramona Bobocel and Aaron Farrell 1996), we have not come across research that tests how perceptions of procedural justice or, more importantly, performance change with actual variation in implementation of affirmative action.
92 Medicine has also been studied, but because the issue with respect to this field is couched in terms of medical school admissions, it is discussed in the following section on affirmative action in education.
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
Holzer and Neumark: Assessing Affirmative Action
539
these urban populations, but also as the local political landscapes have changed (William Lewis 1989). Police departments are of considerable interest in the affirmative action debate because they have frequently been subject to lawsuits and consent decrees that have, in many cases, contributed to the growing minority and female representation in many of these departments. Tim Sass and Jennifer Troyer (1999) claim that "the volume of antidiscrimination litigation involving police departments has been substantial, perhaps greater than any other occupation group on a per capita basis" (p. 572). As a specific example, John Lott (1998) reports data from the U.S. Department of Justice's Civil Rights Division indicating that 19 of 189 cities with complete data in the Law Enforcement Management and Administrative Statistics (LEMAS) survey were covered by consent decrees regarding race or sex during the 1987-93 period, and presents evidence that these consent decrees were associated with relative declines in the representation of white males, and relative increases in the representation of minorities and women.
Nicholas Lovrich and Brent Steel (1983) and Steel and Lovrich (1987) analyze differences in crime outcomes across cities with varying representation of women and minorities on their police forces. Defining "high affirmative action" and "low affirmative action" cities either by current representation of minorities/females or growth over time, they find little evidence of differences across these cities in changes in crime rates, successful arrest rates per crime committed, and per capita expenditures on crime control. In our view the latter two are more direct and relevant measures of police efficiency. One might wonder whether the failure to find evidence of effects of affirmative action is
a result of imprecise and therefore rather uninformative estimates, or instead a result of a small confidence interval centered on zero. The evidence presented in these papers suggests that it is the latter. For example, Table 4 in the paper by Steel and Lovrich reports crime clearance rates for each year from 1970-80. These never differ by more than one percentage point between their treatment and control cities (here defined with respect to affirmative action for women), the standard errors are well under one percentage point, and the crime clearance rate is lower in the treatment group in some years and higher in others. On the other hand, the papers do not report standard errors for the figures on changes in crime rates; for most of the years the differences in the annual percentage change in crime rates are less than seven, with switching from year to year in whether the change is higher in the treatment or control group.93
Of course, growing minority representation on a city's police force may reflect local political/demographic factors rather than the effects of affirmative action programs per se. For example, a growing proportion of young blacks may be associated with both increased crime rates and increased minority representation. Alternatively, minority representation may be increased in response to increased crime, or blacks may be more heavily represented on the police forces of cities with larger
93 John Donohue and Steven Levitt (1998) focus more on the relationship between the racial composition of police forces and the racial patterns of arrests and crime. They find that own-race policing (by which they mean a better match between the demographics of a city's population and its police force) appears to reduce property crime and overall arrests, without influencing violent crime. They interpret the combined evidence as providing an "efficiency rationale" for affirmative action in policing, although their paper does not analyze affirmative action directly.
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
540       Journal of Economic Literature, Vol. XXXVIII (September 2000)
minority populations and higher crime rates. All of these scenarios predict a positive association between crime rates and minority representation, and hence positive bias in the types of analyses Lovrich and Steel conduct. Lott (1998) uses presence of a consent decree and the length of time it has been in effect, as well as whether the city's mayor is black, as instruments for the race and sex composition of the police force. Upon instrumenting, Lott finds a large positive effect of minority police representation on the rates of incidence of some crimes (homicides and assaults). However, this study is plagued by a variety of problems concerning sample definition and specification of equations, making the conclusions unreliable. Most importantly, Lott makes a statistical error in the main results reported in his tables. His dependent variables in the second-stage regressions are crime rates, yet he includes these in the first-stage regression. In a footnote he reports results when he (correctly) does not do this, finding much weaker and generally insignificant effects of sex or race composition on violent crime rates. These problems are reflected in the likelihood (for reasons explained in the text) that endoge-neity of minority representation with respect to. crime rates should lead to upward, rather than downward, biases in OLS estimates of the effects of minority representation on crime, while Lott finds the opposite.94
The issue of whether affirmative ac-
94 In addition, the exogeneity of a consent decree is questionable in this context, as it may reflect underlying racial tensions and demographic/political forces that could be endogenous to the local crime rate. We are also concerned with how to interpret the very different results for different categories of crime, suggesting some spurious relationships in the data; for example, his results indicate that a higher black or minority presence is associated with more murders and assaults, and fewer rapes and car thefts.
tion leads to lower standards has been quite prominent with respect to police work. First, concurrent with increased efforts at raising minority representation have been increased educational requirements for police officers, which are more likely to disqualify minority than nonminority candidates (David Carter and Allan Sapp 1991). However, as Carter and Sapp document, the courts have generally ruled that higher education is a bona fide occupational requirement. Interestingly, Carter and Sapp find that female and minority police are more likely than their male counterparts to have college degrees, though this could simply reflect their relatively younger ages.95
A second prominent issue has been that of "race-norming" of test scores, which has the effect of reducing standards for minority candidates. Carter and Sapp argue—apparently based on impressions gained from site visits, although the evidence is not presented— that race-norming is unnecessary, and that "qualified minorities can be identified and hired by law enforcement agencies without changing employment standards" (p. 20). However, they suggest that police departments pursued the latter strategy because they were skeptical that more aggressive recruitment would meet their hiring goals (p. 20). Linda Gottfredson (1996) provides a detailed critique of the Nassau County Police Department's development—under court order—of a test designed to reduce disparate negative impact of testing on minorities. She shows that the overall test procedure downplayed cognitive tests relative to other tests. She also argues strongly that the tests  that received increased weight
95 They present only univariate analyses, arguing—for reasons inexplicable to us—that this is appropriate because "the findings represent the population studied, not a sample" (p. 7).
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
Holzer and Neumark: Assessing Affirmative Action
541
were less job related, which reduced the overall validity of the test. However, although Gottfredson predicts dire consequences of this test for performance of the department, she presents no direct evaluation of evidence before and after the test's implementation.
b. Academics. As virtually all universities are federal contractors, most are bound by affirmative action requirements. Do the women and minorities who have been hired in greater numbers in academia over the past several years have lower productivity than their white male counterparts? Van Kolpin and Larry Singell (1996) focused on economics departments in the late 1970s and early 1980s, finding that highly-ranked departments hired relatively few women in the 1970s, but those that did gained ground in rankings relative to those that did not. Furthermore, women had higher publication rates than did their male counterparts at departments with comparable ranks in the 1970s.
Other studies focus on salary differentials between different demographic groups, and therefore do not directly relate to affirmative action, although these studies sometimes present evidence on publication differences between white males and others. Focusing on academic departments more broadly defined than in the Kolpin and Singell study, Debra Barbezat (1989) reports considerably lower publication of articles for women than men, and smaller differences in the same direction for books. However, these estimates do not control for field (and Barbezat claims that the average level of publication varies tremendously across fields). They also do not control for age or experience, which are lower for women in her sample. On the other hand, in salary regressions that control for age and experience (but not field), adding publica-
tions reduces the positive male-female differential, consistent with lower publications for women conditional on the included variables. Barbezat does not report overall publication differences by race, but she does find that blacks were earning positive premia that grew after controlling for publications, consistent with lower publications for blacks (although again with no controls for field). However, there is contrary evidence regarding lower performance among black academics, as Charles Elmore and Robert Blackburn (1983) find comparable rates of publication between whites and blacks in Big Ten universities in the late 1970s.
The disadvantage of these studies relative to the Kolpin and Singell study is that they do not incorporate information on the quality of publications. In principle this could be done, although such information is likely to be valid only within fields. The Kolpin and Singell study implicitly incorporates information on the output of faculty members via department rankings. In general, research on academics is promising for two reasons. First, the debate about affirmative action in hiring in academia is particularly fierce (see, e.g., Amitai Etzioni 1971, and the letters in response to this article). Second, much of individual output is observable in academia in publicly-available forms; although quality is harder to gauge, this can be attempted using either citations or rankings of journals.
4.4.5 Case Study/Institutional Evidence
Aside from quantitative evidence on the efficiency effects of affirmative action, there are some useful case studies that present qualitative or descriptive evidence, although these studies focus more on how affirmative action was implemented rather than on performance explicitly.  For instance,  Lee Badgett
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
542       Journal of Economic Literature, Vol. XXXVIII (September 2000)
(1995) reports on a very large nonunion manufacturer that implemented affirmative action for blacks as part of a settlement of a race discrimination lawsuit. The affirmative action plan entailed hiring and promotion goals, and technician training. Badgett reports that the hiring and promotion goals were met by the company expanding its recruiting efforts (as evidenced by large increases in the relative numbers of black applicants, as well as increased hiring).96 This company was increasing skill requirements of workers, via both increased pre-employment testing and increased training, at the same time it was implementing affirmative action. Badgett argues that rather than these two goals competing with one another, they may have been at least in part complementary, because achieving them required similar tools (e.g., increased recruiting and training), although it is difficult to point to concrete evidence.97 Susan Vernon-Gerstenfeld and Edmund Burke (1985) studied nine companies during the 1980s, interviewing and surveying personnel and affirmative action directors. Although they were provided no data with which to assess the assertions these directors made, they report that affirmative action plans were viewed as more likely to be effective when they emphasized recruitment and training, and used ongoing performance assessment. Although the nature of "effective" is not specified, and there appears to be no comparison across
96 The hiring comparison is based not only on changes over time, but also a comparison to hiring at similar employers in the same period.
97 One finding illustrates how the company at least reduced the tradeoff between these goals (without demonstrating complementarity per se). Specifically, in order to reduce disparate impact from increased pre-employment testing, managers designed recruiting efforts to "oversample" blacks, so that "the actual numbers of minority applicants who pass the test remain high even though the group's pass rate is lower" (p. 503).
companies based on the methods used to make affirmative action more effective, it is nonetheless of interest that these conclusions parallel the findings in Holzer and Neumark (2000) that firms using affirmative action report intensified recruitment, training, and performance evaluation. Similarly, Rick Stoops (1982) stresses the successful recruitment campaign for minority police in Houston in the late 1970s and early 1980s, although the study contains no information on the qualifications of recruits. Patricia Hyer (1985) concludes that factors such as more extensive recruitment and strong leadership contributed to the success of affirmative action at the three academic institutions she studies, although in her case success is measured only in terms of recruitment of female faculty, and there is again no comparison group.98
Not all case studies describe positive programs. Dula Espinosa (1992) studies the implementation of affirmative action in a California city, finding that goals were not met and that progress toward meeting those goals was exaggerated. Among other conclusions, this paper points to the importance of obtaining objective measures of the success of affirmative action, and suggests some caution in relying on subjective (or even ostensibly quantitative) assessments provided by those involved in the program.
Not surprisingly, these studies indicate that an emphasis on generating
98 More quantitative evidence on the important role of recruitment, training, and overall management strategy in affirmative action programs appears in a study of 141 manufacturing firms in Tennessee by Horace Johns and Ronald Moser (1989), while many of the same factors are stressed by John Butler and Charles Moskos (1998) in their discussion of the success of affirmative action in the U.S. Armed Forces in leading to promotions of minorities into higher ranks (although these authors do not assess the performance effects).
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
Holzer and Neumark: Assessing Affirmative Action
543
employees of high quality through a variety of means seems to contribute to "successful" affirmative action programs. In general, although we think the case studies provide useful descriptive information that might guide more systematic research, they do not lead to conclusive evidence on their own. Supplementing these studies with comparison groups of companies with less successful affirmative action programs, and looking at success in terms of worker or organizational performance would greatly increase the value of case studies; but they will always suffer from a lack of generalizability relative to more quantitative studies of larger, more representative samples.
4.4.6 The Employee Selection Literature
A very different and much more indirect approach to the study of affirmative action appears in the literature in industrial psychology and human resources management on selection procedures in employment. This literature, summarized in Campbell (1996), generally shows that cognitive test scores are correlated with a variety of measures of job performance at the level of 0.3-0.6. Given racial test score gaps of up to a standard deviation between whites and blacks, comparable hiring standards across the two groups will lead to lower employment levels of blacks and lower average performance among black employees on jobs where such performance matters. Not surprisingly, simulations (with little economic content) suggest that lowering standards for minorities under affirmative action will lead to lower average performance among those minorities hired, and higher average performance among nonminorities, although effects on overall firm performance are likely to be modest unless goals for minority hiring far exceed the actual proportion in the
relevant population (e.g., Jay Silva and Rick Jacobs 1993)."
On the other hand, a broader range of tests (including psychomotor skills, personality tests, or other job-specific skills) indicate much smaller or no racial differences, and predicted racial differences in job performance will vary according to the extent to which these different measures of skills are relevant to any particular job. Indeed, multiple predictors generally generate more "valid" outcomes with less adverse impact against minority candidates (see the discussion in Gottfredson 1996). Thus, successful affirmative action programs may require not a lowering of standards, but a careful evaluation of what appropriate standards are, how they are best gauged in minority candidates for employment, and how extensive recruiting and training can be used to generate such employees.
4.4.7 Administrative Costs
Aside from these influences on worker performance, there are also potential direct costs of administering affirmative action. Some estimates of these direct costs, based on a variety of surveys of limited numbers of establishments, appear in Leonard (1985) and Conrad
99 Generally, cognitive tests are as strongly related to performance among blacks as among whites. But Robert Thorndike (1971) points out that even with fair tests (i.e., the regression line relating performance to test scores is identical for blacks and whites), so long as the tests themselves are not perfectly correlated with the true "criterion" for which tests are administered, if blacks have lower mean scores on a predictor variable the race shortfall in the rate of hiring (based on whether the conditional expectation of performance exceeds a cutoff 17) will exceed the race shortfall in the proportion exceeding q on the test. This has generated some calls for within-group race-norming of predictors, though not necessarily norming between groups. Within-group norming in this context means that results are adjusted so that the group differences based on the test are no larger than they would be based on the true criterion. For more discussion see Campbell (1996).
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
544
Journal of Economic Literature, Vol. XXXVIII (September 2000)
(1995). They cite surveys of very large companies from the mid-1970s, in which average annual compliance costs were $78 per employee in 1976-77 dollars, or roughly three times as much in current dollars. The variance in these estimates across companies was extremely high, and one could argue that average costs per firm today are either higher or lower than those estimated over twenty years ago.100 The additional costs imposed on an establishment by a compliance review per employee are, in current dollars, roughly $2-15. Despite this wide range of estimates, it seems fair to say that the direct administrative costs to contractors of administering affirmative action programs raise their compensation costs by roughly one percent on average.101 Of course, additional resource costs arise in the public sector from administering affirmative action.
4.4.8 Summary of Employment Findings
The studies summarized above, based on very different methodological approaches and frequently flawed in one way or another, do not generate a definitive conclusion about the efficiency/ performance effects of affirmative action on employees and establishments. Still, a number of findings appear with some frequency, and the following generalizations are suggested by the data:
100 On the one hand, Conrad argues that costs today may well be lower than those of twenty years ago, due to company learning over time. On the other hand, since these estimates focus only on very large companies that no doubt enjoy major economies of scale in human resource administration costs, administrative costs for a more representative sample of companies might be higher.
1°1 Since median wages for these large companies will generally be above the economy-wide average of about $13 per hour, median annual earnings for most will oe above $20,000, and mean earnings will be even higher. In current dollars, enforcement costs will be roughly $240-250 per year per employee.
• There is virtually no evidence of significantly weaker qualifications or performance among white women in establishments that practice affirmative action, especially within grade or at a given wage level.
• There is some evidence of lower qualifications for minorities hired under affirmative action programs, especially when such qualifications are measured using test scores or educational attainment. Evidence of lower performance among these minorities appears much less consistently or convincingly, and to the extent that it appears at all, it is associated more with affirmative action in hiring than with affirmative action in recruiting.
• Extensive recruitment and training are important ways for individual firms to generate pools of qualified minority applicants and employees when using affirmative action.
• Careful evaluation of selection procedures for validity and fairness is important. It may be possible to broaden selection "standards" without necessarily lowering them.
4.5 Efficiency/Performance Effects in Education: Evidence
There is evidence on the effects of affirmative action in education that focuses on many of the specific hypotheses raised in the theoretical discussion in Section 4.2, although there is, of course, no "omnibus" measure on the basis of which to assess the efficiency effects of affirmative action in education. The central issues in thinking about the efficiency consequences of preferential admissions, which include the "fit" and "underperformance" hypotheses, are whether blacks gain as much as whites from admission to selective colleges, and the relative performance of women or minorities admitted under preferential programs. Research
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
Holzer and Neumark: Assessing Affirmative Action
545
has also addressed the questions of positive externalities in the form of service to underserved communities or role-model/ mentoring effects, as well as what might be considered an externality from diversity in the classroom and in colleges in general.102 The available evidence is reviewed in the following subsections, and summarized in Table 6.
4.5.1 Relative Performance of Preferential Admits
A couple of studies have focused on medicine and the relative performance of those who were likely initially admitted to medical schools under affirmative action, based on surveys of physicians who attended medical school at some point since the mid-1970s. Davidson and Lewis (1997) look at special admits—which include but are not limited to minorities—to a particular medical school. They find that these special admits have significantly lower grades in core basic and clinical science courses (although no higher failure rates) than the control group of regular admits, slightly lower graduation rates (94 percent vs. 98 percent, significantly different at the five-percent level), and significantly lower scores and pass rates on Board certification exams (for example, taking the exams an average of 1.3 times, vs. 1.1 times for the control group, to achieve a passing grade). However, there was no reported difference in the completion of residency training (identical percentages), the evaluation of performance by residency directors (slightly favoring the control group, but not significantly), or the likelihood of choosing
102 One argument we do not address here is that raised by Loury (1977) regarding the potential efficiency gains from redistributing educational resources towards minorities in or from communities with underinvestment in public goods. Conrad and Sharpe (1996) offer this as an economic rationale for affirmative action as "reparations" for past discrimination (p. 19).
primary care (25 percent of the special admits, vs. 24 percent of the regular admits, an insignificant difference). In contrast, Keith, Bell, and Williams (1987) look at minority vs. nonminority students, finding that minority physicians are significantly more likely to choose primary care specialities (55.9 percent vs. 41.5 percent) and less likely to have Board certification in their specialties (48 percent vs. 80 percent),103 with the latter difference becoming smaller (but remaining) after accounting for the generally lower pre-medical school performance of minorities.
There is also some general evidence of lower performance of preferential admits in undergraduate institutions. Bowen and Bok (1998, ch. 3) document the lower overall performance of black students relative to white students in the selective institutions they study. For example, graduation rates from the first school attended are 81 percent for blacks and 86 percent for whites, while cumulative GPAs are lower by 0.52 point on a four-point scale. This difference does not appear to be biased in any way by differences in the distribution of students across majors. In this sample, at least, the distributions of majors of black and white students do not reflect lower representation of blacks in majors sometimes conjectured to use more stringent grading guidelines, such as the sciences. The percentages of black and white students pursuing science, math, and engineering degrees are nearly identical, while blacks are overrepresented in some of the social sciences, and whites overrepresented in some of the humanities. They also document (as in Vars and Bowen 1998)
103 Board certification is available in a number of specialized fields. A doctor is eligible to take a certification exam after completing all of the graduate training requirements of a given board, which range from three to seven years.
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
546       Journal of Economic Literature, Vol. XXXVIII (September 2000)
TABLE 6
Summary of Studies on Effects of Affirmative Action in Education on Efficiency/Performance
Studies
Data
Results/Conclusions
Comments
1. Gains Experienced by Preferential Admits Datcher Loury NLS Class of 1972
and Garman 1993 and 1995
Kane 1998
High School and Beyond
Bowen and Bok 1998
College and Beyond Surveys
2. Relative Performance of Preferential Admits Davidson and Survey of Graduates from
Lewis 1997 UC Davis, 1968-87
College and Beyond Surveys
Vars and Bowen 1998
3. Externalities A. Service to Minority Communities
Keith et al. 1987 Survey of all minority physicians in U.S., from class of 1975 and a sample of non-minority physicians
Penn et al. 1986
Cantor et al. 1996
Survey of 113 graduates from UCSD
1987, 1991 Surveys of Young Physicians
B. Mentoring/Role Model Effects Solnick 1995 Data from 15 women's and
coed colleges including anticipated and final majors of individual students
Constantine 1995
National Longitudinal Survey of Class of 1972
Blacks have lower GPAs and graduation rates, especially those with low SAT scores at better institutions. But blacks still benefit from selective institutions.
Similar results, but worse performance of blacks at selective institutions disappears when account is taken of attendance at historically-black college.
Conditional on test scores, etc., black students at most selective institutions graduate at higher rates.
Lower scores and grades for minority doctors; small differences in outcomes.
Lower GPAs and graduation rates of blacks conditional on test scores.
More minority/poor patients; more primary care specialities; fewer with board certification.
More minority patients, more primary care, more rural/inner-city practices.
Minority doctors serve more minority/poor.
Women at women's colleges more likely to switch into traditionally male majors.
There is a large positive labor market return to black students attending historically black colleges and universities.
Careful attempt to address relative gains of black and white admits at selective institutions.
Undermines claim that black preferential admits gain less.
Inconclusive evidence on "fit" hypothesis.
Special Admits: only half are minority.
Unclear applicability to "fit" or "underperformance" hypothesis.
Large sample.
Small sample.
Large sample.
Does not isolate effect of students, or necessarily generalize to effects of affirmative action.
Does not isolate effect of students, or necessarily generalize to effects of affirmative action, and does not address other educational outcomes.
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
Holzer and Neumark: Assessing Affirmative Action
547
TABLE 6 (Cont.)
Studies Data Results/Conclusions Comments
Dynan and Rouse 1997	Data on Harvard undergraduates	Having female faculty in introductory economics courses did not affect the probability that males or females majored in economics.	Limited sample.
Canes and Rosen 1995	Panel data on science and engineering departments at three institutions	An increased share of women on the faculty did not lead to an increase in the share of female majors.	Limited number of schools.
Rothstein 1995	National Longitudinal Survey of Class of 1972	Positive association between female faculty and probability that female students attain graduate degree.	Large, representative sample.
Neumark and Gardecki 1998	Data on female hiring and female graduate students in economics over 18 years	More female faculty and female dissertation chairs do not improve female students' job placements, but more female faculty results in shorter completion times.	Limited to one academic field.
C. The Benefits of Diversity Bowen and Bok        College and Beyond 1998                     Surveys of 1976 and 1989 matriculants		Greater percentage black among college students is positively associated with perceived effect of college on racial understanding, and interracial interactions during and after schooling.	First systematic evidence on this issue.
that this performance shortfall exists even conditional on test scores; regression estimates indicate a significant black-white shortfall in class rank of 16 percentage points, compared with a raw differential of 30 percentage points. Finally, Vars and Bowen also report that conditional on test scores, there is a performance shortfall of blacks relative to whites—measured in terms of grade point average or graduation—that is larger for those with the highest SAT scores. Bowen and Bok suggest that this might be taken as evidence in favor of the "fit" or "underperformance" hypotheses, but then dismiss this possibility based on the additional evidence
that this shortfall is greater at higher test scores. They argue that these hypotheses are least likely to apply to the most academically talented black students. But we do not see why this follows, since the question concerns black-white differences conditional on test scores. The "fit" or "underperformance" hypotheses do not make any predictions regarding where in the test score distribution such effects should be strongest.
We have to be a bit careful in interpreting this evidence, however, since statistical issues could underlie some of it. An overall performance shortfall in the data can arise from the test scores
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
548       Journal of Economic Literature, Vol. XXXVIII (September 2000)
available to researchers being error-ridden proxies for the "true ability" on the basis of which admissions decisions are made. One way to see this is to note that in this case, under the usual assumptions that true ability and the error are joint-normally distributed and uncorrelated, the expected grade point average conditional on race and test scores is a weighted average of test scores and the race-specific mean. Alternatively, in a regression of grade point average on test scores and race, the test score coefficient is biased downward toward zero, and given that blacks have lower test scores, the race coefficient is also biased downward. A larger shortfall at the upper end of the test score distribution is expected if test scores predict white performance better than black performance; as mentioned earlier, Vars and Bowen (1998) and Bowen and Bok (1998) present some evidence of this. This follows because, letting P denote performance and T denote test scores:
E(P\T, black) = XT + (1 - X)E(P\black)
E(P\T, white) = X'T + (1 - X')E(P\white). With   test   scores   more   accurate for whites, X' > X, so that the performance gap conditional on test scores (E(P\T, black) - E(P\T, white)) increases with T.
Furthermore, it is not clear that looking at achievement conditional on test scores is the right way to test these hypotheses, since it is not obvious why blacks should feel less "qualified" or "deserving" than whites with comparable test scores and therefore perform worse, although stereotyping could cause this. Rather, the important question is whether black students at selective institutions—who do as a group have lower test scores—underperform relative to how they would have done in the absence of affirmative action, in which case some of them would have gone to
less selective institutions. Bowen and Bok struggle with answering this difficult question. In particular, they show that conditional on SAT scores, high school grades, and socioeconomic characteristics, graduation rates of blacks were higher at more selective institutions (pp. 61-63), and thus reject the "fit" hypothesis. However, it seems likely that there are unobservables positively related to academic and economic success among those blacks who were admitted to the selective schools; that is, conditional on observable factors such as SAT scores, grades, etc., the selective schools manage to select and enroll the most qualified students who are therefore the most likely to graduate. Thus, while this exercise is a useful first step— certainly a lower graduation rate of minorities at selective schools, conditional on these variables, would provide evidence in favor of the "fit" hypothesis— more research is needed to better approximate the "true" experiment of how those minorities admitted to the selective schools would have fared elsewhere.
4.5.2 Gains Experienced by Beneficiaries of Preferential Admissions
Papers by Datcher Loury and Gar-man (1993, 1995) and by Kane (1998) analyze the SAT scores, college grades, graduation rates, and earnings of young whites and blacks, focusing especially on how these vary by college quality (measured by average SAT scores) for each racial group. Because affirmative action arguably only boosts minority enrollments at the most selective colleges (see also Bowen and Bok 1998), differences in outcomes by race and selectivity of college are used to infer the relative gains to minorities and nonmi-norities from affirmative action. This is a slightly different question than that addressed in the previous subsection; even if preferential admits perform less
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
Holzer and Neumark: Assessing Affirmative Action
549
well, the gains from preferential admissions may still be high, and could be higher than those for other admits. The former authors primarily use data from the NLS Class of 1972, while Kane uses the High School and Beyond data.
Superficially, the results of the two papers appear to differ somewhat. While both show that blacks have lower GPAs and graduation rates, the Datcher Loury and Garman results imply that these findings are much more likely to hold among blacks with low SAT scores who are attending schools of above-average quality. Still, both papers imply that, all else equal, whites as well as blacks benefit from attending more selective schools, in terms of GPA, graduation rates, and subsequent earnings. Thus, blacks with lower SAT scores are not necessarily worse off when they are admitted to a more selective college than they would have otherwise been; instead, their average gain from having been admitted is not as high as that observed among whites. Furthermore, Kane argues that the negative interaction between race and college selectivity on graduation disappears in equations that also control for whether the person attended a historically-black college, where average SAT scores are relatively low but graduation rates are high.
4.5.3 Externalities
a. "Service" to Minority Communities by Preferential Admits. One potential positive externality from affirmative action in admissions is the creation of expanded service by professionals or others to traditionally underserved (typically minority) communities. This issue has also been studied with respect to the medical profession. The results show quite uniformly that "special admit" and even more so minority physicians are more likely to treat patients who are mi-
norities, poor (especially those paying through Medicaid), non-English speakers, and/or those located in rural/inner-city (or "physician shortage") areas (Keith, Bell, and Williams 1987; Nolan Penn et al. 1986; Joel Cantor et al. 1996; and M. Komaromy et al. 1996). These studies are consistent with the conclusion that the special admissions programs are generating social benefits to disadvantaged groups that go beyond the physician in question. This evidence bears on the externality arguments for affirmative action, which may compensate for the effects of affirmative action in admitting/producing less-qualified doctors, although it does not prove the existence of a market failure that is remedied by affirmative action's apparent redistribution of medical services from majority to minority communities.
In addition, although it would not necessarily offset the benefits of service to underserved populations, it is possible that the tendency of minority physicians to serve minority patients reflects a lower ability of these physicians to serve in the hospitals or practices that cater to relatively more nonminority patients. This could stem from discrimination, or from lower qualifications of minority physicians (consistent with some of the research findings). While there is no definitive research distinguishing among these alternative possibilities, there is some pertinent evidence. First, Cantor et al. (1996) provide evidence that physicians who report that they found limited opportunities in their training or career due to discrimination also report serving a higher proportion of black, Hispanic, and poor patients; unfortunately, these results are not broken out by the race or ethnicity of the physician. On the other hand, these authors also report that physicians who state strong preferences for serving minority patients, or for serving patients who
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
550       Journal of Economic Literature, Vol. XXXVIII (September 2000)
have trouble getting care, work with a higher percentage of minority patients. Thus, while it is at least possible that discrimination plays a role in directing minority physicians toward minority patients, it appears that physicians' preferences also play a role. Most importantly, perhaps, Keith, Bell, and Williams (1987) report some descriptive information relating performance on Board exams to practicing in medically un-derserved areas. For nonminority physicians, those with higher performance are less likely to practice in such areas. But for minority physicians, performance is positively related to practicing in underserved areas, which appears to contradict the argument that the higher incidence of minority practice in an underserved area stems from lower qualifications; taken literally, this evidence would imply that if minority physicians were more qualified, even more of them would work with underserved populations.
b. Mentoring/Role-Model Effects. Another externality argument is that increased numbers of highly-educated women and minorities generate mentoring relationships or role-model effects that lead to better educational and professional outcomes for other women and minorities. This question has been addressed on two levels.
First, some research attempts to address whether the presence of female or minority students has beneficial contemporaneous effects for other female or minority students. Sara Solnick (1995) studies whether women who attend women's colleges are more likely to major in traditionally male fields. Because there may be selection into these colleges on the basis of preferred major, she studies changes from originally-declared major to final major, and how these transitions differ between women at  women's   colleges  and  women at
coeducational colleges. She finds that at women's colleges, women are indeed more likely to switch to traditionally male fields of study, but are no less likely to switch out of these fields. Con-stantine (1995) estimates the effects of attending an historically black college or university (HBCU) on the future wages of black students, carefully trying to control for selection on unobservables, finding evidence of sizable positive returns. Both of these studies are consistent with positive effects on female or minority students of the presence of other women or minority students.104 However, these studies have some limitations with respect to drawing conclusions about affirmative action. First, they do not necessarily generalize to the effects of increasing female or minority enrollment at traditional coeducational institutions. Second, they do not separate out the effects of the students attending women's colleges or HBCUs from other differences in the environment of these schools, including the sex and race composition of the faculty, which is discussed below.
The second level on which the mentoring/role-model hypothesis has been addressed is with respect to faculty-student relationships. That is, does the training and subsequent hiring of female or minority faculty improve outcomes for subsequent female or minority students?105 Looking at impacts of faculty on choice of major of undergraduates, Karen Dynan and Rouse (1997) find that adding female faculty
104 One could interpret such a conclusion a couple of ways. It could indicate the potentially positive impact of higher percentages of minority or female students, and hence provide an argument for diversity. On the other hand, it could potentially be used to argue for the value of segregated educational institutions, an argument that is frequently made with respect to women's colleges.
1°5 As noted earlier, one might also consider this hypothesis as an argument for affirmative action in faculty hiring in colleges and universities.
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
Holzer and Neumark: Assessing Affirmative Action
551
to economics departments has no effect on the likelihood that either men or women continue to study economics, while Brandice Canes and Harvey Rosen (1995) find no evidence that increases in the proportions of female faculty in science or engineering departments lead to increases in the share of female majors. Turning to choice of graduate study, Rothstein (1995) reports that having female faculty as an undergraduate has a positive association with the probability that female students attain a graduate degree (although no effect on earnings). Finally, looking at graduate student success, Neumark and Gardecki (1998) examine whether increased hiring of female faculty results in more successful outcomes for female graduate students in economics. They find no evidence of beneficial effects on job placements from either additional female faculty or having a female dissertation chair, although they do find that adding female faculty appears to shorten completion times for female students.106
c. The Benefits of Diversity. The third type of externality argument invoked in favor of affirmative action in education is the positive benefits of diversity in the classroom and in college life.107 As
106 Other papers address issues regarding graduate student-faculty role-model relationships, without explicitly considering the effects of changing the number of female faculty. Marsha Shelburn and Patsy Lewellyn (1995) describe some differences in the relationships of male and female graduate students to predominantly male faculty members. Lucia Gilbert (1985) provides descriptive information from an academic department on factors that were important (to students) determinants of same-sex student-faculty role-model relationships. Ehrenberg, Daniel Goldhaber, and Dominic Brewer (1995) examine the effects of teachers' race, sex, and ethnicity among students between 8th and 10th grade.
107As noted by Conrad and Sharpe (1996), counter posed to this argument is the criticism that affirmative action reduces educational quality by "watering down the curriculum." Sandy Darity has pointed out that one instance in which affir-
Bowen and Bok note, however, there is little concrete evidence on the effects of diversity, "in part because definition, measurement, and analysis are very difficult in this area" (1998, p. 219). There is, indeed, little if any prior research that they cite.108 However, they use their survey data to conduct some rather extensive new analyses of the role of diversity in colleges and universities, although in our view this evidence still falls short of establishing benefits of diversity.
Bowen and Bok report that in their survey of 1976 college matriculants collected between 1995 and 1997, sizable fractions of both whites (42 percent) and blacks (74 percent) described the "ability to work effectively and get along well with people from different races/cultures" as very important (the highest rating on a five-point scale). These percentages were higher (mainly for whites) in the survey of 1989 matriculants. Of course, as the authors note, this only establishes that students regard this ability as important, and perhaps of increasing importance. Even if we accept this as evidence that this ability has objectively become more important, this evidence does not imply that diversity in education produces this outcome.
To assess the contribution of colleges and universities to this goal, they report evidence on respondents' perceptions of the contribution of colleges and universities to their ability to work effectively and get along well with people from different races/cultures. Here, they find that among 1976 matriculants, sizable percentages of blacks and whites
mative action/diversity clearly has affected the curriculum is in literature. There is, of course, a raging debate on whether this improves or dilutes the curriculum.
1°8 Conrad and Sharpe (1996) cite a few studies—some older and unpublished—that address the diversity issue.
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
552       Journal of Economic Literature, Vol. XXXVIII (September 2000)
responded that college made a "great deal" of difference, with these percentages rising among 1989 matriculants. This evidence has two obvious weaknesses, however. First, it is not clear what the "control" group or "counter-factual" is. Since the survey respondents all attended college, we have no way of assessing whether they would have developed this ability, or regarded something else (such as work experience) as valuable in developing this ability, had they not attended college. Second, even if there is a causal effect of college, this positive effect of college need not be attributed to diversity in college, per se, rather than other aspects of college education. However, they also report a positive relationship between the relative size of the black student population and white students' perceptions that college improved their ability to understand and get along with people of other races/cultures (Appendix Table D.8.1), which is at least consistent with diversity having a causal effect. One could argue that students more interested in better understanding of other races/cultures are more likely to attend schools with more minority representation, but recall that the students at these schools are reporting a positive effect of college on this understanding, so the argument would have to be one about selection on the propensity to be influenced by diversity. Moreover, although the authors do not make this point, the same table reports no relationship between the relative size of the black student population and the proportion of white students who report that getting along with people of different races/cultures is very important, suggesting that this selection is not at work.109
109 To attempt to study the link between diversity and ability to get along with people of other races or cultures, Bowen and Bok study interac-
Bowen and Bok's evidence on diversity is significant because our perception is that diversity in colleges and universities is a central argument for affirmative action in college admissions, and there is little other evidence on this question. However, the evidence is far from decisive. It does not establish a causal link from diversity to other outcomes, although it presents some evidence consistent with a causal link. More importantly, perhaps, it takes as the outcome of interest greater understanding of and interaction with people of different races or cultures. It is perhaps not surprising (nor uninteresting) that more racial/cultural interaction generated by diversity in college admissions breeds better racial/cultural understanding and further racial/cultural interaction, but using this finding to argue the benefits of diversity could be regarded as assuming the answer to the question.110 Given the multi-dimensional objectives of education, we do not want to claim that improved racial/ cultural understanding or interaction is
tions between students of different races, among the 1989 matriculants. They report high degrees of interaction for black students and white students. They also report that this interaction is positively associated with the relative size of the black student population, consistent with diversity having some positive influence on interactions; again, this evidence is more compelling because those who value racial/cultural understanding are apparently no more likely to attend schools with relatively more black students, and there is no relationship between the racial mix of students' high schools and their reported racial interactions in college (p. 238). Finally, they report that white students who had interactions with students of opposite races were more likely to report that the ability to understand or get along with people from other races/cultures was very important.
110 This same limited focus on racial/cultural issues is reflected in Alexander Astin (1993), who reports evidence of a positive association among college students between socializing with students from different ethnic or racial groups, on the one hand, and commitment to promoting racial understanding, on the other.
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
Holzer and Neumark: Assessing Affirmative Action
553
unimportant. But additional evidence that attempts to link diversity in education to economic outcomes—such as reduced discrimination, higher wages, different occupational choices, etc.—would be of great value and in our view more decisive.
What this discussion makes most clear, perhaps, is the difficulty of thinking about how to obtain direct, objective measures of the contribution of diversity to educational quality. However, while the! absence of concrete evidence regarding the gains from diversity might give some solace to critics of affirmative action in education, we should also recognize the absence of evidence indicating adverse effects from the pursuit of diversity.
4.5.4 Summary of Education Findings
The available research has begun to provide evidence regarding specific arguments for or against affirmative action in education. Even more so than for affirmative action in employment, the jury is still very much out—not only because researchers have generally not yet succeeded in garnering decisive evidence regarding specific hypotheses, but also because there is generally less research, and because theory provides less guidance regarding the outcomes we should or can study to assess the efficiency/performance question. Nonetheless, there are a few areas in which some general conclusions can be drawn from the existing studies:
• Blacks admitted to college have on average lower GPAs and graduation rates. However, in the universe of colleges and universities excluding the historically-black colleges and universities, the race difference in graduation rates is no larger at the most selective institutions. In addition, blacks benefit from attending selective schools as do whites. Together, this evidence under-
mines the argument that affirmative action admission programs at selective schools result in poor "fits" or "mismatches" for black students, placing them in challenging academic environments where on average they are not able to compete and do worse as a result. However, it is difficult to establish a definitive conclusion regarding this question.
• Research on medical education finds evidence that minority students and in some cases "special admits" (a broader category) perform less well in school, and are less likely to achieve high levels of expertise. At the same time, this research also suggests that these students are ultimately more likely to serve minority patients, which may provide a positive externality that helps offset the lower qualifications or skill levels that these students attain.
• Evidence on whether female or minority faculty or students spur achievement or otherwise positively affect other female or minority students is mixed. There is some evidence that women's colleges and historically black colleges have positive impacts on women and blacks, respectively. But evidence that female faculty at coeducational institutions either serve as role models or mentors encouraging women to study traditionally male fields, or helping women to perform better, is mixed, with relatively more evidence indicating no positive effects.
• A case can probably be made that a diverse student body positively impacts interracial and intercultural relations. However, in our view the claims for the positive effects of diversity go beyond this in suggesting that diversity results in better education overall. This is a difficult question to assess, but perhaps no more difficult than in other areas of research on educational quality. There is as yet no evidence on
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
554       Journal of Economic Literature, Vol. XXXVIII (September 2000)
TABLE 7
Summary of Studies on Effects of Affirmative Action in Contracting/Procurement
on Efficiency/Performance
Studií
Data
Result/Conclusions
Comments
1. Concentration in Industries with Strong Minority Stephanopoulos GAO report
and Edley 1995
2. Cost Differentials
Denes 1997 500 dredging contracts sup-
plied by US Army Corps of Engineers
3. Propping Up Weak Companies
Bates and Williams Characteristics of Business 1995a and 1996 Owners data base, profiles
of local procurement programs
Myers and Chan 1996
Survey of contractors in New Jersey
4. Spurring Entrepreneurship
Chay and Fairlie        CPS, 1979-89
1998
Presence
Agencies concentrate their minority contracting in fields such as construction where there are a significant number of existing minority firms.
Costs no higher for bids restricted to small businesses.
MBEs with large share of revenues from local government sales more likely to go out of business, although some of this may be a spurious result of front companies. Programs that impose penalties for fraud, and that provide assistance, promote success and survival.
Award/bid ratio was unchanged when set-asides ended, and average number of contracts reported by black relative to white contractors rose slightly after set-asides ended.
Black self-employment appears to rise in cities as a result of establishment of minority set-aside programs.
Not peer-reviewed study.
May not generalize to preferential programs for minorities or women.
Useful introduction of institutional evidence. Evidence of fraudulent front companies is indirect.
Data do not indicate greater success of minority-owned firms under set-asides.
Addresses key question. Difficulties in establishing timing of effect and hence causality.
this broader hypothesis regarding the potential benefits of affirmative action in college and university admissions.
4.6 Efficiency/Performance Effects in Contracting/Procurement: Evidence
As is the case for evaluating the efficiency/performance effects of affirmative action in education, there is no
overall test for the efficiency effects of affirmative action in contracting and procurement. Rather, there is only sketchy evidence on some of the specific hypotheses that have been advanced regarding the potential negative side effects of this form of affirmative action. The available evidence is discussed below, and summarized in Table 7.
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to 1STOR Terms and Conditions
Holzer and Neumark: Assessing Affirmative Action
555
4.6.1 Concentration in Industries with Strong Minority Presence
According to a General Accounting Office study (see Stephanopoulos and Edley 1995, ch. 9), agencies tend to concentrate their minority contracting in certain fields—such as construction— where there are a significant number of existing minority firms. As noted earlier, this is potentially problematic as such programs may not achieve their potential in encouraging business development in other fields.
4.6.2 Cost Differentials
We have not found any work that examines relative costs of contracts awarded to minority- or women-owned businesses. However, Denes (1997) looks at bid price relative to government cost estimates comparing solicitations restricted to small businesses to unrestricted solicitations.111 He finds that costs are no higher in the restricted bids, and that more bids are submitted for these solicitations, perhaps because some companies specialize in them. This study need not generalize to set-asides for minority- and women-owned businesses, and almost surely does not generalize to bid-price preference programs; but the study suggests that programs that ostensibly restrict competition do not necessarily result in higher prices. Moreover, the research design might be fruitfully applied to affirmative action programs in government contracting and procurement.
4.6.3 Propping Up Weak Companies
Another potential problem is that affirmative action in contracting and procurement may prop up weak companies. To determine whether this occurs, we
111 In part to avoid very different types of goods and services provided, he focuses solely on dredging contracts.
would ideally like to know whether companies that receive preferences in contracting would tend to fail more than representative companies, absent this treatment. Of course, we cannot observe this directly. Furthermore, if discrimination (in capital markets, for example) disadvantages women- or minority-owned businesses, then this test is not necessarily the right one. The view that has been adopted in the little work that exists on this issue focuses on the question of whether companies that benefit from affirmative action in contracting or procurement continue to succeed after "weaning" from the affirmative action program, relative to businesses generally. Indeed, there are provisions built into some programs that aim to wean companies from preferential treatment, and hence in a sense "set up" this and related tests. Most prominently, 1987 amendments to the statute establishing the section 8(a) program set a "graduation" period of nine years and require that, over time, firms achieve an "increasing mix of business from outside the section 8(a) program and outside federal contracting" (Stephanopoulos and Edley, ch. 9, p. 3).
Stephanopoulos and Edley (1995) report evidence that they claim shows that these amendments are effective. In particular, they cite Small Business Administration statistics for 1993 indicating that "of the 710 firms that were graduates in that or previous years, 56 percent were still fully operational, 6 percent had curtailed operations, 3 percent had been acquired by other companies, and 35 percent had ceased operations . . . ," and that "comparisons with Census data suggest that the failure rates of graduated [section] 8(a) firms are no worse than, and in fact may be better than, those seen in small businesses generally" (ch. 9, p. 5). However, this comparison is based on the failure
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
556
Journal of Economic Literature, Vol. XXXVIII (September 2000)
rate of section 8(a) firms that graduated from the program—and hence have survived at least nine years prior to evaluating their continued success—with the failure rate for new firms, an obviously unfair comparison, and one that would no doubt look less favorable were failure rates of section 8(a) firms contrasted with failure rates for a general sample of firms conditional on surviving nine years.112 These authors also report that firms are quite successful at meeting the non-section 8(a) business requirement, noting that in 1995 nearly two-thirds of firms in the fifth through ninth year of section 8(a) participation met or exceeded the minimum requirements for non-8(a) business (ch. 9, p. 6).113
Additional evidence on whether affirmative action props up weak companies comes from a study of procurement programs in New Jersey (Myers and Chan 1996), which reports that the introduction of preferences for minority-owned businesses led to sharp involvement of such businesses in government procurement. But as Bates (1998) points out, their numbers also reveal that when these preferences were sharply curtailed, the share of procurement going to minority vendors was not affected (p. 24).114 This evidence might be interpreted as consistent with preferential procurement programs provid-
112 The numbers and comparison they cite are taken from reports of the Small Business Administration (SBA). In private communications, a staff member of the SBA agreed that their figures do not indicate that section 8(a) graduates are more successful than other firms.
113 we fully recognize that much of this evidence comes from a political document, and it would be preferable to rely on evidence reported by independent researchers. '
114 In particular, the award/bid ratio was essentially unchanged when set-asides ended (Table 4 of Myers and Chan), and the average number of contracts reported by black contractors relative to the average number reported by white contractors rose slightly after set-asides were ended (Table 5).
ing a jump start to minority-owned businesses, but ultimately leading to businesses that could stand on their own. However, one unexpected finding casts some doubt on the data used in this study. Specifically, the estimates indicate that the advent of the set-aside program reduced the average number of awards for black contractors relative to white contractors, despite state figures indicating that the share of awards going to minority-owned firms rose sharply under set-asides (Myers and Chan 1996). Interesting complementary evidence would be whether the firms that took a large slice of government contracts also experienced growing private-sector business, and continued to prosper after the preferential programs were curtailed.
Similarly, Bates and Williams (1996) report evidence indicating that minority business enterprises (MBEs) that derive a high percentage of their revenue from local government sales were more rather than less likely to go out of business. In particular, using the Characteristics of Business Owners data base of the U.S. Bureau of the Census and looking at survival over the 1987-91 period, they find that among all MBEs, and among the subset of MBEs with any government sales, those MBEs deriving at least 25 percent of revenue from local government sales were more likely to go out of business. One could read this evidence as suggesting that local government contracting with MBEs certainly does not prop firms up. However, Bates and Williams suggest that some of this may be spurious, especially because in their data government procurement seems to be targeted at MBEs in operation for one year or less. They suggest that "some of the young MBEs may be front companies. One variant of fronting entails setting up an MBE to participate—along with large, nonminority
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
Holzer and Neumark: Assessing Affirmative Action
557
business partners—in a specific government contract or project; the MBE closes when the project ends" (p. 297).115 Bates and Williams also examine factors associated with successful MBE programs, based on evidence that comes from merging their data with profiles of MBE programs operated in more than thirty large cities. They point out that some MBEs operate in "an environment in which MBE certification is comprehensive, bonding and working capital assistance are available, and assistance is delivered by a staff dedicated to aiding potential and actual MBE vendors" (p. 3). Their evidence suggests that these types of assistance promote success and survival among MBE vendors, for MBEs deriving less than 25 percent of sales from government. They also note that these more beneficial program characteristics are associated with local procurement and contracting programs in cities with black mayors. On the other hand, they suggest that the absence of beneficial effects of these assistance programs for companies deriving at least 25 percent of their sales from government is attributable to the fact that more of these MBEs are front companies. The evidence for this claim is that among these latter companies, controlling for other characteristics, sales are lower under programs that use financial penalties and possible jail terms to punish fraudulent behavior; they interpret this finding as evidence that "[t]hese MBEs behave collectively as though front companies are operating in their midst" (p. 14). In contrast, among the MBEs with lower government sales, this relationship between penalties and sales does not ap-
115 Bates and Williams (1995a) provide more details, and responses by other researchers in the same journal issue and a rejoinder by Bates and Williams (1995b) debate the issue of front companies.
pear. Although this is not direct evidence, it is a clever inference based on the available data.
4.6.4 Spurring Entrepreneurship
As pointed out earlier, the overriding goal of affirmative action in government contracting and procurement may be not to increase the share of business for minority- or female-owned firms, but rather to spur minority and female entrepreneurship. This question has received virtually no attention in the literature. However, a recent study by Chay and Robert Fairlie (1998) asks whether the establishment of minority set-aside programs in large cities appears to cause increases in black self-employment, using CPS data for 1979-89 (prior to the Supreme Court decision in City of Richmond v. J. A. Croson Co.). The advantage of the experimental design is that it allows for growth over time in black self-employment, only attributing the causal effect of set-asides to differential growth in cities establishing set-aside programs. The authors report that city-level set-aside programs appear to have generated growth in black self-employment, although they are somewhat cautious in their conclusions because of difficulties of pinning down the timing of the establishment of these city-level programs, and hence in determining whether the growth in self-employment came after the set-aside programs.
4.6.5 Summary of Contracting/ Procurement Findings
Evidence on the efficiency/performance effects of affirmative action in contracting and procurement is limited. This area is ripe for further research, especially as recent court decisions generate variation in both the scope and mode of operation of these programs. The evidence to date suggests a mixed
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
558       Journal of Economic Literature, Vol. XXXVIII (September 2000)
bag of conclusions, with some of the criticisms of these programs (such as those regarding fraud) borne out, but others (regarding higher costs and propping up weak companies) not supported by the existing evidence.
5. Conclusions
We have tried to assess what we know about affirmative action from the perspective of economic theory and empirical evidence, with a particular but not exclusive emphasis on the efficiency effects of affirmative action programs. We have considered affirmative action defined broadly to include an array of special efforts to improve the status of minorities and women in the labor market, educational institutions, or business procurement—whether these efforts result from requirements on federal contractors, court-imposed remedies, voluntary efforts, or other policies.
Not surprisingly, the theoretical literature from labor economics generates ambiguous results on whether or not affirmative action programs result in efficiency gains or losses; much depends on whether or not discrimination exists in the absence of these programs, the presence of other market failures (such as externalities, and capital market and information imperfections), and on how they affect human capital formation and job assignments among protected and unprotected groups. The theoretical literature regarding affirmative action effects in university admissions and business procurement is virtually nonexistent, although hypotheses have been advanced that are related to efficiency/performance effects, which we try to evaluate.
The empirical literature—both in economics and other disciplines—on the presence of discrimination and the effects of affirmative action is much more extensive. While it is impossible
to assess the overall efficiency or welfare effects of affirmative action from this evidence, the following inferences pertinent to these effects seem justified:
• Significant labor market discrimination against minorities and women persists, as do other forms of disadvantage for minorities in the attainment of human capital (which some refer to as "societal discrimination").
• Affirmative action programs redistribute employment, university admissions, and government business from white males to minorities and women, though the extent of the redistribution may not be large.
• There is virtually no evidence of weaker educational qualifications or job performance among females who benefit from affirmative action relative to males, especially within occupational grade.
• The educational performance and labor market credentials of minority beneficiaries are weaker than those of their white counterparts. But evidence of weaker performance in the labor market among these groups is much less frequently observed or is less credible. Evidence on the performance of the minority businesses who benefit from special procurement programs is also mixed.
• The potential effects of affirmative action on performance, at least in the labor market, appear to depend on how it is implemented. Employers that practice affirmative action can (and often do) mitigate its potentially negative effects on performance by extensive recruitment and screening before workers are hired, as well as special training and evaluation efforts afterwards.
• Although minority students admitted to colleges and universities perform less well, on average, than nonminority students, this evidence is generally no
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
Holzer and Neumark: Assessing Affirmative Action
559
stronger at the most selective schools that have been the focus of the affirmative action debate. Both black and white students benefit from attending selective colleges and universities.
• There is some evidence consistent with positive externalities from affirmative action, but not for each type of externality that has been posited by its advocates. For example, minority doctors are more likely to treat minority and/or low-income patients than are other physicians. Evidence on role-model/mentoring effects in universities is weaker and more mixed, especially with respect to coeducational institutions. There is no evidence of positive (or negative) effects of a diverse student body on educational quality.
• There is mixed evidence regarding whether affirmative action in contracting and procurement props up weak companies. In some studies, firms that initially benefit from these programs but then move into an environment without set-asides do not appear to fail at higher rates than comparable firms. On the other hand, there is some evidence that minority business enterprises deriving a large percentage of their revenue from local government are relatively more likely to go out of business. Some evidence suggests, however, that this phenomenon is attributable to the fraudulent formation of front companies for the sole purpose of qualifying for these programs. Local government programs with genuine assistance to small enterprises, and penalties for fraudulent behavior, appear able to promote success of minority business enterprises.
All in all, the evidence suggests to us that it may be possible to generate affirmative action programs that entail relatively little sacrifice of efficiency. Most importantly, there is at this junc-
ture very little compelling evidence of deleterious efficiency effects of affirmative action. This does not imply that such costs do not exist, nor that the studies we review have captured the overall welfare effects of affirmative action. It does imply, though, that the empirical case against affirmative action on the grounds of efficiency is weak at best.
On the other hand, advocates of affirmative action might draw more encouragement from the existing evidence. Affirmative action seems to have major redistributive effects that operate in markets in which discrimination still exists, and it may create some positive externalities; it might therefore lead to increased efficiency. This set of findings on efficiency effects, which displays some variation but tends to be centered on little or no effect, as well as specific evidence pointing toward some benefits of affirmative action (such as externalities), might be interpreted as somewhat favorable to this set of programs.
However, because there are resource costs associated with enforcement of affirmative action, some evidence of overall efficiency gains is probably needed to make a case for affirmative action on efficiency grounds, rather than redistributive grounds. Although we can by no means fully quantify these, it seems to us that there is not yet sufficient evidence to conclude that there are overall efficiency gains, although based on the evidence we think this is more rather than less likely. Thus, we regard the current state of the evidence as most consistent with the view that affirmative action offers significant redistribution toward women and minorities, with relatively small efficiency consequences. A major outstanding question that could tip the scales more in the direction of efficiency gains is the extent to which this redistribution increases
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
560       Journal of Economic Literature, Vol. XXXVIII (September 2000)
efficiency by countering discrimination in the labor market. We have argued that there is evidence of continuing discrimination against women and minorities. In this case it is possible that affirmative action generates additional efficiency gains, although theory does not necessarily imply this.
In any event, it is also clear that the research evidence to date on the effects of affirmative action also remains quite incomplete, especially regarding effects on efficiency and performance. To accurately evaluate the efficiency effects of affirmative action programs on establishments or other appropriate economic units, we would likely need evaluations approximating random assignment, as have been used in research on employment and training programs. As a political matter, it seems very unlikely that such policy experiments will ever be explicitly allowed or implemented, although policy changes may provide informative "quasi-experiments."
Perhaps the best we can hope for is the continued development of research that addresses more limited questions, but over time provides a collage of evidence that gives us a more thorough understanding of the effects of affirmative action. Included in this list would be additional sector-specific studies, like those that we have seen to date in the fields of medicine, crime, and academia, but in a wider variety of sectors, and with better measures of performance and cleaner comparisons across groups than we have seen to date. In addition, studies that test specific conjectures about affirmative action, like those we have seen to date regarding mentoring, externalities, job performance, and business survival, can generate a lot more added value even in the absence of "ideal" experiments.
Of course, it is certainly possible that other remedies for the disadvantages
experienced by women and minorities are equally or more desirable than those provided by affirmative action. For instance, Steele (1990) writes, instead of affirmative action, "Give my children fairness; give disadvantaged children a better shot at development—better elementary and secondary schools, job training, safer neighborhoods, better financial assistance for college, and so on" (p. 124). Virtually no one would argue against the notion that efforts to enhance human capital formation among minorities are extremely important as a means of improving their relative economic status, and are probably more desirable over the long run. But these efforts are quite costly and generate few returns in the short run, and there are many disagreements about exactly how to achieve the desired skill enhancement.116 Race-neutral employment and training programs for the disadvantaged may generate more immediate gains, though the magnitudes of the improvements generated are frequently low and the costs of effective remedies quite high (e.g., Robert Lalonde 1995). In any event, none of these are necessarily incompatible with ongoing affirmative action programs.
Some also argue against the notion that affirmative action should be based on race or sex, and instead argue for income- or class-based remedies. While one might be better able to support these arguments on equity grounds, it should be recognized that because there are many non-minorities in lower-income or lower-class groups, such programs would likely result in significant reductions in the proportions of minorities that benefit  from  affirmative  action programs
116 For recent research on the consequences of school-finance equalization, see David Card and A. Payne (1998), Sheila Murray, William Evans, and Robert Schwab (1998), and Raquel Fernandez and Richard Rogerson (1998).
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to 1STOR Terms and Conditions
Holzer and Neumark: Assessing Affirmative Action
561
(Maria Cancian 1998; Kane 1998).117 Of course, this fact, in and of itself, is not a criticism of class- or income-based preferences, since the whole point would be to open up some preferences for disadvantaged nonminorities. But in assessing class- or income-based policies, we should not be under the illusion that we would get approximately the same outcome under a more "acceptable" ostensible set of criteria for receiving preferential treatment.118 Furthermore, income- or class-based policies appear to ignore the potential role of affirmative action in countering discrimination and other disadvantages faced by women and minorities in the labor market and elsewhere, even controlling for family background. And, to return to a point made earlier, it seems difficult to construct race- and sex-blind antidiscrimination policies that might instead be used to address these latter issues. Finally, any negative effects of affirmative action on efficiency, employee performance, or student quality that we currently experience might become even more serious under such a scheme, as low-income white males replace middle- (or upper-) income women and minorities in universities and workplaces.119 In addition, income-
117 Cancian considers classes defined based on poverty status, parents' education, and family structure. She also documents that these alternative methods of classifying individuals and their families are likely to result in very different sets of individuals eligible for affirmative action, highlighting the political difficulty of agreeing on a class or income criterion to use. Conrad and Sharpe (1996) cite similar evidence based on simulations carried out by the University of California system.
118 An alternative policy for university admissions is a version of Texas' law guaranteeing admission to the University of Texas to the top ten percent of students in each high school. This, of course, is not based explicitly on race, but would likely favor minorities when high schools are segregated by race.
119 Simulations of the effects of Proposition 209 in the University of California system, as reported
based affirmative action schemes may create disincentives to increase income, either directly via labor market decisions, or indirectly via decisions regarding family structure.120 On the other hand, disagreements over exactly which ethnic groups should have "protected" status, and for how long, are certainly legitimate;121 these are likely to continue to provide a rationale for considering income- or class-based remedies. We expect research on such remedies to become quite prominent.
Our overall goal in this review is not to offer a set of policy recommendations regarding affirmative action as it is currently practiced, or alternatives to it. Rather, our goal is to convey to the research and policy community what we know about affirmative action, and to point to the important unanswered questions. The fulfillment of both of these goals would better inform the debate over affirmative action.
References
Abraham, Katherine and James Medoff. 1981. "Are Those Paid More Really More Productive? The Case of Experience," /. Human Res. 16:2, pp. 186-216.
Aigner, Dennis J. and Glen C. Cain. 1977. "Statistical Theories of Discrimination in Labor Markets," Ind. Labor Bel. Rev. 30:2, pp. 175-87.
Altonji, Joseph and Rebecca Blank. 1999. "Race and Gender in the Labor Market," in Handbook of Labor Economics, Vol. 3C. O. Ashenfelter and D. Card, eds. Amsterdam: North Holland, pp. 3143-259.
Altonji, Joseph and Charles Pierret. 1997. "Employer Learning and Statistical Discrimination," NBER working paper 6279.
Ando, Faith. 1988. "Capital Issues and Minority-
by Conrad and Sharpe (1996) suggest that substituting income-based for race-based admissions policies will lower median SAT scores among admitted students.
120 We find it ironic that many conservative critics of affirmative action fail to note these potential incentive effects of income-based policies, despite emphasizing such incentive effects with respect to other social programs.
121 Furthermore, the original notion that affirmative action was meant to be transitional rather than permanent deserves some consideration.
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
562       Journal of Economic Literature, Vol. XXXVIII (September 2000)
Owned Business," Rev. Black Polit. Econ. 16:4, pp. 77-109.
Arrow, Kenneth J. 1972. "Some Mathematical Models of Race Discrimination in the Labor Market," in Racial Discrimination in Economic Life. A. Pascal, ed. Lexington, MA: D.C. Heath, pp. 187-203.
---- 1973. "The Theory of Discrimination," in
Discrimination in Labor Markets. O. Ashenfel-ter and A. Rees, eds. Princeton: Princeton U. Press, pp. 3-33.
Ashenfelter, Orley and James Heckman. 1976. "Measuring the Effect of an Antidiscrimination Program," in Evaluating the Labor Market Effects of Social Programs. O. Ashenfelter and J. Blum, eds. Princeton: Princeton U. Press, pp. 46-84.
Astin, Alexander. 1993. Four Critical Years: Effects of College on Beliefs, Attitudes and Knowledge. San Francisco: Jossey-Bass.
Athey, Susan; Christopher Avery, and Peter Zem-sky. 1998. "Mentoring and Diversity," NBER working paper 6496.
Attiyeh, Gregory and Richard Attiyeh. 1997. "Testing for Bias in Graduate School Admissions," J. Human Res. 32:3, pp. 524-48.
Badgett, Lee. 1995. "Affirmative Action in a Changing Legal and Economic Environment," Ind. Rel. 34:4, pp. 489-506.
Barbezat, Debra. 1989. "Affirmative Action in Higher Education," Research Labor Econ. 10, pp. 107-56.
Bates, Timothy. 1991. "Commercial Bank Financing of White- and Black-Owned Small Business Start-ups," Quart. Rev. Econ. Business, 31:1, pp. 64-80.
---- 1993. Banking on Black Enterprise. Washington, DC: Joint Center Political Economic Studies.
--- 1994. "Utilization of Minority Employees
in Small Business: A Comparison of Nonminor-ity and Black-Owned Urban Enterprises," Rev. Black Polit. Econ. 23:1, pp. 113-21.
---- 1998. "Minority Business Access to Government Procurement," mimeo, Wayne State U.
Bates, Timothy and William D. Bradford. 1992. "Factors Affecting New Firm Success and Their Use in Venture Capital Financing," J. Small Bus. Fin. 2:1, pp. 23-38.
Bates, Timothy and Darrell Williams. 1995a. "Preferential Procurement Programs and Minority-Owned Businesses," J. Urban Affairs, 17:1, pp. 1-17.
---- 1995b. "Preferential Procurement Programs, A Rejoinder," J. Urban Affairs, 17:1, pp. 29-32.
--- 1996. "Do Preferential Procurement Programs Benefit Minority Business?" Amer. Econ. Rev. 86:2, pp. 294-97.
Becker, Gary S. 1971. The Economics of Discrimination: 2nd Ed. Chicago: U. Chicago Press.
___ 1985. "Human Capital, Effort, and the
Sexual Division of Labor," /. Labor Econ. 3:1, pt. 2, pp. S33-S58.
Bendick, Marc; Charles Jackson, and Victor Rei-noso. 1994. "Measuring Employment Discrimination Through Controlled Experiments," Rev. Black Polit. Econ. 23:1, pp. 25-48.
Black, Dan A. 1995. "Discrimination in an Equilibrium Search Model,"/. Labor Econ. 13:2, pp. 309-34.
Blackburn, McKinley L. and David Neumark. 1993. "Omitted-Ability Bias and the Increase in the Return to Schooling," /. Labor Econ. 11:3, pp. 521-44.
Blanchflower, David G; Phillip B. Levine, and David J. Zimmerman. 1998. "Discrimination in the Small Business Credit Market," mimeo, Dartmouth College.
Blau, Francine. 1998. "Trends in the Well-Being of American Women, 1970-95," J. Econ. Lit. 36:1, pp. 112-65.
Blau, Francine D. and Marianne A. Ferber. 1991. "Career Plans and Expectations of Young Men and Women,"/. Human Res. 26:4, pp. 581-607.
Blau, Francine D. and Lawrence M. Kahn. 1997. "Swimming Upstream: Trends in the Gender Wage Differential in the 1980s," /. Labor Econ. 15:1, pt. 1, pp. 1-42.
Bloch, FarrelL 1994. Antidiscrimination Law and Minority Employment. Chicago: U. Chicago Press.
Block, Howard and Robert Pennington. 1981. "An Econometric Analysis of Affirmative Action," Rev. Black Polit. Econ. 11:2, pp. 267-76.
Bobocel, Ramona D. and Aaron C. Farrell. 1996. "Sex-Based Promotion Decisions and Interactional Fairness: Investigating the Influence of Managerial Accounts," /. Applied. Psych. 81:1, pp. 22-35.
Borjas, George. 1995. "Ethnicity, Neighborhoods, and Human Capital Externalities," Amer. Econ. Rev. 85:3, pp. 365-90.
Bound, John and Richard B. Freeman. 1992. "What Went Wrong? The Erosion of Relative Earnings and Employment among Young Black Men in the 1980s," Quart. J. Econ. 107:1, pp. 201-32.
Bowen, William G. and Derek Bok. 1998. The Shape of the River. Princeton: Princeton U. Press.
Bratsberg, Bernt and Dek Terrell. 1998. "Experience, Tenure, and Wage Growth of Young Black and White Men,"/. Human Res. 33:3, pp. 658-82.
Brown, Charles. 1982. "The Federal Attack on Labor Market Discrimination: The Mouse That Roared?" Research Labor Econ. 5, pp. 33-68.
Brown, Charles and Mary Corcoran. 1997. "Sex-Based Differences in School Content and the Male/Female Wage Gap," /. Labor Econ. 15:3, pt. 1, pp. 431-65.
Bunzel, John H. 1996. "Race and College Admissions," Public Interest, 122, pp. 49-58.
Butler, John Sibley and Charles Moskos. 1998. "Labor Force Trends: The Military as Data," paper presented at NRC Conference on Racial Trends in America.
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
Holzer and Neumark: Assessing Affirmative Action
563
Campbell, John. 1996. "Group Differences and Personnel Decisions: Validity, Fairness, and Affirmative Action," J. Vocation. Behav. 49:2, pp. 122-58.
Cancian, Maria. 1998. "Race-Based versus Class-Based Affirmative Action in College Admissions," J. Policy Analysis Manage. 17:1, pp. 94-105.
Canes, Brandice J. and Harvey S. Rosen. 1995. "Following in Her Footsteps? Faculty Gender Composition and Women's Choices of College Majors," Ind. Labor Kel. Rev. 48:3, pp. 486-504.
Cannings, Kathy; Claude Montmarquette, and Sophie Mahseredjian. 1996. "Entrance Quotas and Admission to Medical Schools: A Sequential Probit Model," Econ. Ed. Rev. 15:2, pp. 163-74.
Cantor, Joel; Erika Miles, Laurence Baker, and Dianne Baker. 1996. "Physician Service to the Underserved: Implications for Affirmative Action in Medical Education," Inquiry, 33:2, pp. 167-80.
Card, David and A. Abigail Payne. 1998. "School Finance Reform, the Distribution of School Spending, and the Distribution of SAT Scores," NBER working paper 6766.
Carrington, William; Kristin McCue, and Brooks Pierce. 1995. "Using Establishment Size to Measure the Impact of Title VII and Affirmative Action," mimeo, Johns Hopkins U.
Carrington, William and Kenneth Troske. 1998. "Interfirm Racial Segregation," /. Labor Econ. 16:2, pp. 231-60.
Carter, David and Allan Sapp. 1991. "Police Education and Minority Recruitment: The Impact of a College Requirement," discussion paper, Police Executive Research Forum.
Carter, Stephen. 1991. Reflections of an Affirmative-Action Baby. NY: Basic Books.
Cavalluzzo, Ken S. and Linda C. Cavalluzzo. 1998. "Market Structure and Discrimination: The Case of Small Business," /. Money, Credit, Banking, 30:4, pp. 771-92.
Cavalluzzo, Ken S.; Linda C. Cavalluzzo, and John Wolken. 1999. "Competition, Small. Business Financing, and Discrimination: Evidence from a New Survey," mimeo, Georgetown U.
Cawley, John; James Heckman, and Edward Vytlačil. 1998. "Cognitive Ability and the Rising Return to Education," NBER working paper 6388.
Cawley, John; Kathleen Bonneely, James Heckman, and Edward Vytlačil. 1996. "Measuring the Effects of Cognitive Ability," NBER working paper 5645.
Chay, Kenneth. 1998. "The Impact of Federal Civil Rights Policy on Black Economic Progress: Evidence from the Equal Opportunity Act of 1972," Ind. Labor Rel. Rev. 51:4, pp. 608-32.
Chay, Kenneth and Robert Fairlie. 1998. "Minority Business Set-Asides and Black Self-Employment," mimeo, U. California, Berkeley.
Coate, Stephen and Glenn C. Loury. 1993. "Will Affirmative-Action Policies Eliminate Negative
Stereotypes?" Amer. Econ. Rev. 83:5, pp. 1220-40.
Cole, Stephen. 1986. "Sex Discrimination and Admission to Medical School, 1929-1984," Amer. J. Soc. 92:3, pp. 549-67.
Conrad, Cecilia A. 1995. "The Economic Cost of Affirmative Action," in Economic Perspectives on Affirmative Action. M. Simms, ed. Washington, DC: Joint Center Political Economic Studies, pp. 31-54.
Conrad, Cecilia A. and Rhonda V. Sharpe. 1996. "The Impact of the California Civil Rights Initiative (CCRI) on University and Professional School Admissions and the Implications for the California Economy," Rev. Black Polit. Econ. 25:1, pp. 13-59.
Constantine, Jill M. 1995. "The Effect of Attending Historically Black Colleges and Universities on Future Wages of Black Students," Ind. Labor Rel. Rev. 48:3, pp. 531-46.
Corcoran, Mary; Roger Gordon, Deborah Laren, and Gary Solon. 1992. "The Association between Men's Economic Status and Their Family and Community Origins," /. Human Res. 27:4, pp. 575-601.
Cornell, Bradford and Ivo Welch. 1996. "Culture, Information, and Screening Discrimination," /. Polit. Econ. 104:3, pp. 542-71.
Corns, Allan and Andrew Schotter. 1999. "Can Affirmative Action Be Cost Effective? An Experimental Examination of Price-Preference Auctions," Amer. Econ. Rev. 89:1, pp. 291-305.
Currie, Janet and Duncan Thomas. 1995. "Race, Children's Cognitive Achievement and the Bell Curve," NBER working paper 5240.
Cutler, David and Edward Glaeser. 1997. "Are Ghettoes Good or Bad?" Quart. J. Econ. 107:3, pp. 827-72.
Darity, William A. and Patrick L. Mason. 1998. "Evidence on Discrimination in Employment: Codes of Color, Codes of Gender," /. Econ. Perspect. 12:2, pp. 63-90.
Datcher Loury, Linda and David Garman. 1993. "Affirmative Action in Higher Education," Amer. Econ. Rev. 83:2, pp. 99-103.
—:---        1995. "College Selectivity and Earnings,"
/. Labor Econ. 13:2, pp. 289-308.
Davidson, Robert and Ernest Lewis. 1997. "Affirmative Action and Other Special Consideration Admissions at the University of California, Davis School of Medicine," /. Amer. Med. Assoc. 278:14, pp. 1153-58.
Denes, Thomas A. 1997. "Do Small Business Set-Asides Increase the Cost of Government Contracting?" Public Admin. Rev. 57:5, pp. 441-44.
Dickens, William T. and Thomas J. Kane. 1999. "Racial Test Score Differences as Evidence of Reverse Discrimination: Less than Meets the Eye," Indust. Rel. 38:3, pp. 331-63.
Donohue, John J. and Steven D. Levitt. 1998. "The Impact of Race on Policing, Arrest Patterns, and Crime," mimeo, Stanford U.
D'Souza, Dinesh. 1991. Illiberal Education. NY: Free Press.
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
564       Journal of Economic Literature, Vol. XXXVIII (September 2000)
Dugan, Mary Kay; Nazli Baydar, William Grady, and Terry Johnson. 1996. "Affirmative Action: Does It Exist in Graduate Business Schools?" Selections, 12:2, pp. 11-18.
Dynan, Karen E. and Cecilia Elana Rouse. 1997. "The Underrepresentation of Women in Economics: A Study of Undergraduate Economics Students," J. Econ. Ed. 28:4, pp. 350-68.
Edley, Jr., Christopher. 1996. Not All Black and White. NY: Hill and Wang.
Ehrenberg, Ronald G.; Daniel D. Goldhaber, and Dominic J. Brewer. 1995. "Do Teachers' Race, Gender, and Ethnicity Matter? Evidence from the National Educational Longitudinal Study of 1988," Ind. Labor Rel. Rev. 48:3, pp. 547-61.
Ehrenberg, Ronald G. and Donna S. Rothstein. 1994. "Do Historically Black Institutions of Higher Education Confer Unique Advantages on Black Students? An Initial Analysis," in Choices and Consequences: Contemporary Policy Issues in Education. R. Ehrenberg, ed. Ithaca: ILR Press.
Ellen, Ingrid and Margery Austin Turner. 1997. "Does Neighborhood Matter? Assessing Recent Evidence," Housing Policy Debate, 8:4, pp. 833-66.
Elmore, Charles and Robert Blackburn. 1983. "Black and White Faculty in White Research Universities,"/. Higher Ed. 54:1, pp. 1-15.
Epstein, Richard. 1992. Forbidden Grounds: The Case Against Employment Discrimination Laws. Cambridge, MA: Harvard U. Press.
Espinosa, Dula. 1992. "Affirmative Action: A Case Study of an Organizational Effort," Socio. Per-spect. 35:1, pp. 119-36.
Etzioni, Amitai. 1971. "On Academic Blood Tests," Science, 2, p. 1087.
Farley, Reynolds; Elaine Fielding, and Maria Krysan. 1997. "The Residential Preferences of Blacks and Whites: A Four Metropolis Analysis," Housing Policy Debate, 8:4, pp. 763-800.
Fernandez, Raquel and Richard Rogerson. 1998. "Public Education and Income Distribution: A Dynamic Quantitative Evaluation of Education-Finance Reform," Amer. Econ. Rev. 88:4, pp. 813-33.
Fix, Michael and Raymond Struyk. 1994. Clear and Convincing Evidence. Washington, DC: Urban Institute Press.
Folger, Robert and Mary A. Konovsky. 1989. "Effects of Procedural and Distributive Justice on Reactions to Pay Raise Decisions," Academy Manage. J. 32:1, pp. 115-30.
Foster, Dean P. and Vohra, Rakesh V. 1992. "An Economic Argument for Affirmative Action," Rationality and Society, 4:2, pp. 176-88.
Fuchs, Victor R. 1988. Women's Quest for Economic Equality. Cambridge, MA: Harvard U. Press.
Gallup Organization. 1997. "Special Reports: Black/White Relations in the U.S.," June 10.
Garratt, Rod and John M. Marshall. 1994. "Public Finance of Private Goods: The Case of College Education,"/. Polit. Econ. 102:3, pp. 566-82.
Gilbert, Lucia A. 1985. "Dimensions of Same-Gender Student-Faculty Role-Model Relationships," Sex Roles, 12(1/2), pp. 111-23.
Goldin, Claudia and Cecilia Rouse. "Orchestrating Impartiality: The Impact of "Blind" Auditions on Female Musicians," forthcoming in Amer. Econ. Rev.
Goldsmith, Arthur H.; Jonathan R. Veum, and William Darity, Jr. 1997. "The Impact of Psychological and Human Capital Variables on Wages," Econ. Inquiry, 35:4, pp. 815-29.
Goldstein, Morris and Robert Smith. 1976. "The Estimated Impact of the Antidiscrimination Program Aimed at Federal Contractors," Ind. Labor Rel. Rev. 29:3, pp. 524-43.
Gottfredson, Linda. 1996. "Racially Gerrymandering the Content of Police Tests to Satisfy the U.S. Justice Department: A Case Study," Psych. Public Policy and Law, 2, pp. 418-46.
Graves, Laura and Gary Powell. 1994. "Effects of Sex-Based Preferential Selection and Discrimination on Job Attitudes," Human Rel. 47:2, pp. 133-57.
Greenberg, J. 1990. "Employee Theft as a Reaction to Underpayment Inequity: The Hidden Cost of Pay Cuts," /. Applied Psych. 75, pp. 561-68.
Griffin, Peter. 1992. "The Impact of Affirmative Action on Labor Demand: A Test of Some Implications of the LeChatelier Principle," Rev. Econ. Statist. 74:2, pp. 251-60.
Griffin, Peter; Arthur Getis, and Ernst Griffin. 1996. "Regional Patterns of Affirmative Action Compliance Costs," Annals Reg. Sci. 30:3, pp. 321-40.
Griliches, Zvi. 1977. "Estimating the Returns to Schooling: Some Econometric Problems," Econometrica, 45:1, pp. 1-22.
Gronau, Reuben. 1988. "Sex-related Wage Differentials and Women's Interrupted Labor Ca-reers-the Chicken or the Egg," /. Labor Econ. 6:3, pp. 277-301.
Grown, Caren and Timothy Bates. 1992. "Commercial Bank Lending Practices and the Development of Black Owned Construction Companies,"/. Urban Affairs, 14:1, pp. 25-41.
Hamermesh, Daniel S. 1993. Labor Demand. Princeton: Princeton U. Press.
Heckman, James J. 1998. "Detecting Discrimination,"/. Econ. Perspect. 12:2, pp. 101-16.
Heckman, James and Kenneth Wolpin. 1976. "Does the Contract Compliance Program Work? An Analysis of Chicago Data," Ind. Labor Rel. Rev. 29:3, pp. 544-64.
Heilman, Madeleine. 1997. "Sex Discrimination and the Affirmative Action Remedy: The Role of Sex Stereotypes," /. Bus. Ethics, 16:9, pp. 877-89.
Heilman, Madeleine; Caryn Block, and Jonathan Lucas. 1992. "Presumed Incompetent? Stigma-tization and Affirmative Action Efforts," /. Applied Psych. 77:4, pp. 536-44.
Hellerstein, Judith and David Neumark. 1999. "Sex, Wages, and Productivity: An Empirical
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
Holzer and Neumark: Assessing Affirmative Action
565
Analysis of Israeli Firm-Level Data," Int. Econ. Rev. 40:1, pp. 95-123.
Hellerstein, Judith; David Neumark, and Kenneth R. Troske. 1997. "Market Forces and Sex Discrimination," NBER working paper 6321.
--- 1999. "Wages, Productivity, and Worker
Characteristics: Evidence from Plant-Level Production Functions and Wage Equations," /. Labor Econ. 17:3, pp. 409-446.
Herrnstein, Richard J. and Charles Murray. 1994. The Bell Curve: Intelligence and Class Structure in American Life. NY: Free Press.
Hersch, Joni. 1991. "Equal Employment Opportunity Law and Firm Profitability," /, Human Res, 26:1, pp. 139-53.
Holzer, Harry. 1991. "The Spatial Mismatch Hypothesis: What Does the Evidence Show?" Urban Stud. 28:1, pp. 105-22.
---- 1996. What Employers Want: Job Prospects for Less-Educated Workers. NY: Russell Sage Foundation.
--- 1998. "Why Do Small Establishments
Hire Fewer Blacks than Larger Ones?" /. Human Res. 33:4, pp. 896-914.
Holzer, Harry and Keith Ihlanfeldt. 1998. "Customer Discrimination and the Employment Outcomes of Minorities," Quart. ]. Econ, 113:3, pp. 833-65.
Holzer, Harry and David Neumark. 1999. "Are Affirmative Action Hires Less Qualified? Evidence from Employer-Employee Data," /. Labor Econ. 17:3, pp. 534-69.
-- 2000. "What Does Affirmative Action
Do?" Ind. Labor Rel. Rev. 53:2, pp. 240-71.
Hoxby, Caroline M. 1997. "How the Changing Market Structure of U.S. Higher Education Explains College Tuition," Ind. Labor Rel. Rev. 53:2, pp. 240-71.
Hyclak, Thomas; Larry Taylor, and James Stewart. 1992. "Some New Historical Evidence on the Impact of Affirmative Action: Detroit, 1972," Rev. Black Polit. Econ. Fall, pp. 81-98.
Hyer, Patricia. 1985. "Affirmative Action for Women Faculty: Case Studies of Three Successful Institutions," /. Higher Ed. 56:3, pp. 282-99.
Ihlanfeldt, Keith and David Sjoquist. 1998. "The Spatial Mismatch Hypothesis: A Review of Recent Studies," mimeo, Georgia State U.
Ihlanfeldt, Keith and Madelyn Young. 1994. "In-trametropolitan Variation in Wage Rates: The Case of Atlanta Fast-Food Workers," Rev. Econ. Statist. 76:3, pp. 425-33.
Jencks, Christopher and Susan Mayer. 1990. "Residential Segregation, Job Proximity, and Black Job Opportunities: The Empirical Status of the Spatial Mismatch Hypothesis," in Inner City Poverty in the United States. L. Lynn and M. McGeary, eds. Washington, DC: National Academy Press, pp. 187-222.
Jencks, Christopher and Meredith Phillips eds. 1998. The Black-White Test Score Gap. Washington, DC: Brookings Institution.
Johns, Horace and Ronald Moser. 1989. "Where
Has EEO Taken Personnel Policies?" Personnel, Sept., pp. 63-66.
Johnson, George E. and Finis Welch. 1976. "The Labor Market Implications of an Economywide Affirmative Action Program," Ind. Labor Rel. Rev. 24:4, pp. 508-22.
Kahn, Lawrence M. 1991. "Customer Discrimination and Affirmative Action," Econ. Inquiry, 29:3, pp. 555-71.
Kahn, Lawrence and Derek Shearer. 1988. "Racial Differences in Professional Basketball Players' Compensation,"/. Labor Econ. 6:1, pp. 40-61.
Kain, John. 1991. "The Spatial Mismatch Hypothesis Three Decades Later," Housing Policy Debate, 3:2, pp. 371-462.
Kain, John and Kenneth Singleton. 1996. "Equality of Educational Opportunity Revisited," New England Econ. Rev. May/June, pp. 87-110.
Kane, Thomas. 1998. "Racial Preferences and Higher Education," In C. Jencks and M. Phillips eds. The Black-White Test Score Gap. Washington, DC: Brookings Institution, pp. 431-56.
Keith, Steven; Robert Bell, and Albert Williams. 1987. Assessing the Outcome of Affirmative Action in Medical Schools. LA: Rand Corp.
Kenney, Genevieve and Douglas Wissoker. 1994. "An Analysis of the Correlates of Discrimination Facing Young Hispanic Jobseekers," Amer. Econ. Rev. 84:3, pp. 674-83.
Kinder, Donald R. and Lynn M. Sanders. 1990. "Mimicking Political Debate with Survey Questions: The Case of White Opinion on Affirmative Action for Blacks," Social Cog. 8:1, pp. 73-103.
Kirschenman, Joleen. 1991. "Gender within Race," mimeo, U. Chicago.
Kirschenman, Joleen and Katheryn Neckerman. 1991. "We'd Love to Hire Them But," in The Urban Underclass. C. Jencks and P. Peterson, eds. Washington, DC: Brookings Institution, pp. 203-34.
Kluegel, James R. and Eliot R. Smith. 1986. Beliefs About Inequality: Americans' Views of What Is and What Ought To Be. NY: Aldine De Gruyter.
Kolpin, Van and Larry Singell. 1996. "The Gender Composition and Scholarly Performance of Economics Departments: A Test for Employment Discrimination," Ind. Labor Rel. Rev. 49:3, pp. 408-23.
Komaromy, M.; K. Grumbach, M. Drake, K. Vra-nizan, N. Lurie, D. Keane, and A. B. Bindman. 1996. "The Role of Black and Hispanic Physicians in Providing Health Care for Underserved Populations," New England J. Med. 334:20, pp. 1305-10.
Konrad, Alison and Frank Linnehan. 1995. "Race and Sex Differences in Line Managers' Reactions to Equal Employment Opportunity and Affirmative Action Interventions," Group Org. Manage. 20:4, pp. 409-39.
Korenman, Sanders and David Neumark. 1992. "Marriage, Motherhood, and Wages," /. Human Res. 27:2, pp. 233-55.
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
566       Journal of Economic Literature, Vol. XXXVIII (September 2000)
Ladd, Helen F. 1998. "Evidence on Discrimination in Mortgage Lending," J. Econ. Perspect. 12:2, pp. 41-62.
Lalonde, Robert. 1995. "The Promise of Public Sector-Sponsored Training Programs," /. Econ. Perspect. 9:2, pp. 149-68.
Lang, Kevin. 1986. "A Language Theory of Discrimination," Quart. J. Econ. 101:2, pp. 363-82.
Larew, John. 1991. "Why Are Droves of Unqualified, Unprepared Kids Getting into Our Top Colleges?" Washington Monthly, June, pp. 10-14. Reprinted in Race and Gender in the American Economy. Susan F. Feiner, ed. Englewood Cliffs, NT: Prentice Hall, pp. 145-50.
Lazear, Edward P. and Sherwin Rosen. 1981. "Rank-Order Tournaments as Optimum Labor Contracts,"/. Polit. Econ. 89:5, pp. 841-64.
Leek, Joanne D.; David M. Saunders, and Micheline Charbonneau. 1996. "Affirmative Action Programs: An Organizational Justice Perspective,"/. Org. Behavior, 17:1, pp. 79-89.
Lentz, Bernard F. and David N. Laband. 1989. "Why So Many Children of Doctors Become Doctors: Nepotism vs. Human Capital Transfers,"/. Human Res. 24:3, pp. 396-413.
Leonard, Jonathan. 1984a. "Anti-Discrimination or Reverse Discrimination? The Impact of Changing Demographics, Title VII, and Affirmative Action on Productivity," /. Human Res. 19:2, pp. 145-74.
--- 1984b. "Impact of Affirmative Action on
Employment,"/. Labor Econ. 2:4, pp. 439-63. 1984c. "Employment and Occupational
Advance under Affirmative Action," Rev. Econ, Statist. 66:3, pp. 377-85.
1985. "Affirmative Action as Earnings Re
distribution: The Targeting of Compliance Reviews,"/. Labor Econ. 3:3, pp. 363-84.
1989. "Women and Affirmative Action," /.
Econ. Perspect. 3.T, pp. 61-75.
1990. "The Impact of Affirmative Action
Regulation and Equal Employment Law on Black Employment," /. Econ. Perspect. 4:4, pp. 47-63.
Lewis, Gregory. 1997. "Race, Sex, and Performance Ratings in the Federal Service," Public Admin. Rev. 57:6, pp. 479-89.
Lewis, William. 1989. "Toward Representative Bureaucracy: Blacks in the City Police Organizations, 1975-1985," Public Admin. Rev. 49:3, pp. 257-67.
Lipset, Seymour Martin and William Schneider. 1978. "The Bakke Case: How Would It Be Decided at the Bar of Public Opinion?" Public Opinion, 1:1, March/April, pp. 38-48.
Lott, John. 1998. "Does a Helping Hand Put Others At Risk?: Affirmative Action, Police Departments, and Crime," mimeo, U. Chicago.
Loury, Glenn C. 1977. "A Dynamic Theory of Racial Income Differences," In P. Wallace and A. LaMond eds. Women, Minorities, and Employment Discrimination. Lexington, MA: D.C. Heath, pp. 153-88.
---1981. "Is Equal Opportunity Enough?"
Amer. Econ. Rev. 71:2, pp. 122-26.
Loury, Glenn and Andrew Weiss. 1998. "Discrimination without Racism," mimeo, Boston U.
Lovrich, Nicholas and Brent Steel. 1983. "Affirmative Action and Productivity in Law Enforcement Agencies," Rev. Public Personnel Admin. 4:1, pp. 55-66.
Lundberg, Shelly J. 1991. "The Enforcement of Equal Opportunity Laws under Imperfect Information: Affirmative Action and Alternatives," Quart. J. Econ. 106:1, pp. 309-26.
Lundberg, Shelly J. and Richard Startz. 1983. "Private Discrimination and Social Intervention in Competitive Labor Markets," Amer. Econ. Rev. 73:3, pp. 340-47.
Maxwell, Nan. 1994. "The Effect on'Black-White Wage Differences of Differences in the Quantity and Quality of Education," Ind. Labor Re-lat. Rev. 47:2, pp. 249-64.
Milgrom, Paul and Sharon Oster. 1987. "Job Discrimination, Market Forces, and the Invisibility Hypothesis," Quart. J. Econ. 102:3, pp. 453-76.
Milkovich, George and Alexandra Wigdor, eds. 1991. Pay for Performance: Evaluating Performance Appraisal and Merit Pay. Washington, DC: National Academy Press.
Moss, Philip and Chris Tilly. 1995. "Soft Skills and Wages," working paper, Russell Sage Foundation.
Mullan, Fitzhugh. 1997. "The National Health Service Corps and Inner-City Hospitals," New England J. Med. 336:22, pp. 1601-604.
Murnane, Richard; John Willett, and Frank Levy. 1995. "The Growing Importance of Cognitive Skills in Wage Determination," Rev. Econ. Statist. 77:2, pp. 251-66.
Murray, Charles. 1994. "Affirmative Racism," in Debating Affirmative Action: Race, Gender, Ethnicity, and the Politics of Inclusion. N. Mills, ed. NY: Delta, pp. 191-208.
Murray, Sheila E.; William N. Evans, and Robert M. Schwab. 1998. "Education-Finance Reform and the Distribution of Education Resources," Amer. Econ. Rev. 88:4, pp. 789-812.
Myers, Samuel and Tsze Chan. 1996. "Who Benefits from Minority Set-Asides? The Case of New Jersey," /. Policy Analysis Manage. 15:2, pp. 202-26.
Nardinelli, Clark and Curtis Simon. 1990. "Customer Racial Discrimination in the Market for Memorabilia: The Case of Baseball," Quart. J. Econ. 105:3 , pp. 575-96.
Neal, Derek and William Johnson. 1996. "The Role of Premarket Factors in Black-White Wage Differences," /. Polit. Econ. 104:5, pp. 869-95.
___ 1998. "Basic Skills and the Black-White
Earnings Gap," in The Black-White Test Score Gap. C. Jencks and M. Phillips, eds. Washington, DC: Brookings Institution, pp. 480-97.
Neumark, David. 1988. "Employers' Discriminatory Behavior and the Estimation of Wage Discrimination,"/. Human Res. 23:3, pp. 279-95.
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to 1STOR Terms and Conditions
Holzer and Neumark: Assessing Affirmative Action
567
--- 1996. "Sex Discrimination in Restaurant
Hiring: An Audit Study," Quart. J. Econ. 111:446, pp. 915-942.
--- 1999. "Wage Differentials by Race and
Sex: The Roles of Taste Discrimination and Labor Market Information," Ind. Relat. 38:3, pp. 414-45.
Neumark, David and Rosella Gardecki. 1998. "Women Helping Women? Role-Model and Mentoring Effects on Female Ph.D. Students in Economics,"/. Human Res. 33:1, pp. 220-46.
Neumark, David and Michele McLennan. 1995. "Sex Discrimination and Women's Labor Market Outcomes,"/. Human Res. 30:4, pp. 713-40.
Oaxaca, Ronald. 1973. "Male-Female Wage Differentials in Urban Labor Markets," Int. Econ. Rev. 14:3, pp. 693-709.
Oettinger, Gerald. 1996. "Statistical Discrimination and the Early Career Evolution of the Black-White Wage Gap," /. Labor Econ. 14:1, pp. 52-78.
Oko, Ohechukwu. 1996. "Laboring in the Vineyards of Equality: Promoting Diversity in Legal Education through Affirmative Action," Southern U. Law Rev. 23:2, pp. 189-212.
O'Neill, June. 1994. "Discrimination and Income Differences," in Race and Gender in the American Economy. S. Feiner, ed. Englewood Cliffs, NJ: Prentice-Hall, pp. 13-17.
---- 1990. "The Role of Human Capital in
Earnings Differences between Black and White Men,"/. Econ. Perspect. 4:4, pp. 25-45.
O'Regan, Katherine and John Quigley. 1996. "Teenage Employment and the Spatial Isolation of Minority and Poverty Households," /. Human Res. 31:3, pp. 692-702.
Pathman, Donald E. and Thomas R. Konrad. 1996. "Minority Physicians Serving in Rural National Health Service Corps Sites," Med. Care, 34:5, pp. 439-54.
Penn, Nolan; Percy Russell, Harold Simon, Teresa Jacob, Claire Stafford, Eddie Castro, Jeannette Cisneros, and Mary Bush. 1986. "Affirmative Action at Work: A Survey of Graduates of UCSD Medical School," Amer. J. Public Health, 76:9, pp. 1144-46.
Radford, Mary F. 1997. "The Affirmative Action Debate" in Government Regulation of the Employment Relationship. B. Kaufman, ed. Madison, WI: Industrial Relations Research Assoc., pp. 343-67.
Raphael, Stephen; Michael Stoll, and Harry Holzer. 1998. "Are Suburban Firms More Likely to Discriminate against African Americans?" discussion paper, Institute for Research on Poverty.
Rice, Mitchell F. 1995. "Minority Procurement Policy Design, a Reply to Timothy Bates and Darrell Williams," /. Urban Affairs, 17:1, pp. 19-22.
Rodgers, William M. and William E. Spriggs. 1996a. "What Does the AFQT Really Measure: Race, Wages, Schooling and the AFQT Score," Rev. Black Polit. Econ. 24:4, pp. 13-46.
---- 1996b. "The Effect of Federal Contractor
Status on Racial Differences in Establishment-Level Employment Shares: 1979-1992," Amer. Econ. Rev. 86:2, pp. 290-93.
Rothschild, Michael and Joseph E. Stiglitz. 1982. "A Model of Employment Outcomes Illustrating the Effect of the Structure of Information on the Level and Distribution of Income," Econ. Letters, 10, pp. 231-36.
Rothschild, Michael and Lawrence J. White. 1995. "The Analytics of the Pricing of Higher Education and Other Services in which the Customers are Inputs,"/. Polit. Econ. 103:3, pp. 573-86.
Rothstein, Donna S. 1995. "Do Female Faculty Influence Female Students' Educational and Labor Market Attainments?" Ind. Labor Rel. Rev. 48:3, pp. 515-30.
Sass, Tim R. and Jennifer L. Troyer. 1999. "Affirmative Action, Political Representation, Unions, and Female Police Employment," /. Labor Res., 20:4, pp. 571-87.
Schotter, Andrew and Keith Weigelt. 1992. "Asymmetric Tournaments, Equal Opportunity Laws, and Affirmative Action: Some Experimental Evidence," Quart. J. Econ. 107:2, pp. 511-39.
Shelburn, Marsha R. and Patsy G. Lewellyn. 1995. "Gender Bias in Doctoral Programs in Economics,"/. Econ. Ed. 26:4, pp. 373-82.
Silva, Jay and Rick Jacobs. 1993. "Performance as a Function of Increased Minority Hiring," /. Applied Psych. 789:4, pp. 591-601.
Skarlicki, Daniel P. and Robert Folger. 1997. "Retaliation in the Workplace: The Roles of Distributive, Procedural, and Interactional Justice,"/. Applied Psych. 82:3, pp. 434-43.
Smith, James and Finis Welch. 1984. "Affirmative Action and Labor Markets," /. Labor Econ. 2:2, pp. 269-301.
Solnick, Sara J. 1995. "Changes in Women's Majors from Entrance to Graduation at Women's and Coeducational Colleges," Ind. Labor Rel. Rev. 48:3, pp. 505-14.
Solon, Gary. 1992. "Intergenerational Income Mobility in the United States," Amer. Econ. Rev. 82:3, pp. 393-408.
Stanley, T. D. and Stephen B. Jarrell. 1998. "Gender Wage Discrimination Bias?" /. Human Res. 33:4, pp. 947-73.
Steel, Brent and Nicholas Lovrich. 1987. "Equality and Efficiency Tradeoffs in Affirmative Action—Real or Imagined? The Case of Women in Policing," Social Science J. 24:1, pp. 53-70.
Steele, Shelby. 1990. The Content of Our Character. NY: St. Martin's Press.
Stephanopoulos, George and Christopher Edley, Jr. 1995. "Review of Federal Affirmative Action Programs," unpublished White House document.
Stoops, Rick. 1982. "Minority Recruitment," Personnel /. 61:4, pp. 254-55.
Thernstrom, Stephan and Abigail Thernstrom. 1997. America in Black and White: One Nation, Indivisible. NY: Simon & Schuster.
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions
568       Journal of Economic Literature, Vol. XXXVIII (September 2000)
Thorndike, Robert L. 1971. "Concepts of Culture-Fairness,"/. Ed. Measure. 8:2, pp. 63-70.
Tougas, Francine; Stephane Joly, Ann Beaton, and Line St.-Pierre. 1996. "Reactions of Beneficiaries to Preferential Treatment: A Reality Check," Human Rel. 49:4, pp. 453-64.
Trejo, Stephen J. 1997. "Why Do Mexican Americans Earn Low Wages?" /. Polit. Econ. 105:6, pp. 1235-68.
United States Equal Employment Opportunity Commission. 1994. Annual Report.
Vars, Frederick E. and William G. Bowen. 1998. "Scholastic Aptitude, Test Scores, Race, and Academic Performance in Selective Colleges and Universities," in The Black-White Test Score Gap. C. Jencks and M. Phillips, eds. Washington, DC: Brookings Institution, pp. 457-79.
Vernon-Gerstenfeld, Susan and Edmund Burke. 1985. "Affirmative Action in Nine Large Companies: A Field Study," Personnel, 62:4, pp. 54-60.
Waldfogel, Jane. 1998. "Understanding the 'Family Gap' in Pay for Women with Children," /. Econ. Perspect. 12:1, pp. 137-56.
Walters, Dan. 1999. "Basic Fairness Still Elusive," Sacramento Bee, August 8.
Welch, Finis. 1976. "Employment Quotas for Minorities,"/. Polit. Econ. 84:4, Pt. 2, pp. S105-39.
----1981. "Affirmative Action and Its Enforcement," Amer. Econ. Rev. 71:2, pp. 127-33.
Wightman, Linda F. 1997. "The Threat to Diversity in Legal Education: An Empirical Analysis of the Consequences of Abandoning Race as a Factor in Law School Admission Decisions," NYU Law Rev. 72:1, pp. 1-53.
Winston, Gordon C. 1999. "Subsidies, Hierarchy and Peers: The Awkward Economics of Higher Education,"/. Econ. Perspect. 13:1, pp. 13-36.
Wright, Peter; Stephen Ferris, Janine Hiller, and Mark Kroll. 1995. "Competitiveness through Management of Diversity: Effects on Stock Price Valuation," Academy Manage. J. 38:1, pp. 272-87.
Yinger, John. 1998. "Evidence on Discrimination in Consumer Markets," /. Econ. Perspect. 12:2, pp. 23-40.
Zimmerman, David J. 1992. "Regression toward Mediocrity in Economic Stature," Amer. Econ. Rev. 82:3, pp. 409-29.
This content downloaded from 147.251.185.127 on Wed, 18 Mar 2015 10:04:22 UTC All use subject to JSTOR Terms and Conditions