Edited by Gerrit De Geest
Chapter 10: Employment discrimination
[In: Volume 2, Kenneth G. Dau-Schmidt, Seth D. Harris and Orly Lobel (eds) Labor and Employment Law and Economics]
1 Scope of law and economics analysis of employment discrimination
The law and economics scholarship on employment discrimination examines the welfare consequences of the legal rules regulating employment discrimination. While substantial overlap exists, the inquiry is distinct from the efforts of labor economists to determine the extent of discrimination in labor markets. For surveys in this vein, see Darity and Mason (1998) and Blau (1998). It is also distinct from the efforts of academic lawyers to describe and criticize employment discrimination laws. The major task of the law and economics work surveyed here is to link the two inquiries by studying the effects of law on the labor market. This review limits itself to the literature on race, sex, and disability discrimination, ignoring the important area of age discrimination. See Posner (1989) for an excellent overview of the issues.
2 Definitions of discrimination
The term ‘discrimination’ is elusive. A wine expert is admired for having discriminating taste, but an employer with a taste for racial discrimination is despised (Cooter (1994, p. 137). For an extensive discussion of various definitions of discrimination, see Kelman 1991). An overly broad definition would say that an employer discriminates whenever it distinguishes between two workers because they belong to different groups. To see that the definition is too broad, consider an employer who pays ‘highly productive’ workers more than ‘less productive’ workers. Although distinctions based on productivity (or, more broadly, merit) violate a policy of egalitarianism, they do not implicate traditional concerns of discrimination. One must be careful with the meaning of ‘productivity’ as well. Aigner and Cain (1977) define productivity in terms of physical output or actual job performance. They recognize that discrimination against, say, blacks, can always be explained away if their productivity includes others’ antipathy towards them unrelated to their output or performance.
As a second attempt at a definition, perhaps discrimination occurs whenever an employer treats equally productive workers differently because of any group characteristic other than, but perhaps related to, productivity. As Cooter (1994, p. 137) suggests, ‘discrimination in economic life usually consists in sorting people according to traits rather than productivity’. This definition, at least for the law, is both too broad and too narrow.
p. 297The definition is overbroad because every distinction can be said to be based on a group characteristic, but only some group distinctions are discriminatory. The problem comes in separating invidious discrimination from appropriate (or benign, or merit-based, or non-problematic) distinctions. For example, an employer might distinguish between equally productive college and high school graduates, or between equally productive workers scoring 95 and 85 on a test, or between equally productive black and white workers. Only the last distinction is generally labeled discrimination because it is based on a suspect or protected group. Antidiscrimination laws have prohibited employers from making distinctions based on a number of characteristics, including race, color, sex, age, religion, national origin, age, sexual orientation, marital status and disability. In general, but not invariably, these are immutable characteristics over which the person has no control. Often they are irrelevant characteristics for proper or profit-maximizing employment decisions, but models of statistical discrimination wrestle with situations when these characteristics are correlated with productivity. Epstein (1992, p. 413) has attacked as inherently arbitrary the labeling of certain distinctions in US law as invidious discrimination. As he puts it, a ‘central part of [my] argument against Title VII was that there is no independent conception of which characteristics count as “merit” characteristics and which are invidious’. While difficult lines must be drawn (do distinctions based on beauty, obesity, or veterans’ status count as discriminatory?), American law has singled out a few distinctions as legally discriminatory, the most important being distinctions based on race and sex.
Still, it is too narrow to define discrimination as occurring only when an employer treats two equally productive workers differently because of a protected characteristic. This definition matches the legal concept of ‘disparate treatment’ fairly closely. But the definition is narrower than one that focuses on effects. A worker in a protected group could be said to be discriminated against whenever he or she is treated differently than someone in another group on grounds other than productivity. This definition includes the legal concept of ‘disparate impact’. For example, employers often distinguish between workers on the basis of non-protected characteristics, such as education, test scores, or criminal convictions. While US antidiscrimination laws do not protect criminals or those with low education or test scores, the laws do forbid employers from distinguishing on such a trait if it has a disparate impact on racial or gender lines and the employer cannot demonstrate that the practice is job related and a business necessity.
In addition to disparate treatment or disparate impact, a third legal frame for discrimination is reasonable accommodation. The legal command for p. 298employers to reasonably accommodate occurs most prevalently in disability discrimination and, to a lesser degree, religious discrimination. Seth Harris and Michael Ashley Stein address the economics of workplace disability accommodations in greater detail in this volume (Chapter 12). Nonetheless, while some scholars, including Jolls (2001), have argued that reasonable accommodation is a form of disparate treatment or disparate impact, the first-order economic effects of banning these forms of discrimination are quite different (Issacharoff and Nelson 2001; Schwab and Willborn 2003). In enacting a law prohibiting disparate treatment, a legislature could believe it is not imposing costs on employers (except for second-order administrative or litigation costs fending offinvalid claims). So long as the employer hires the best person for the job without regard for disability, race, or gender, a valid disparate treatment claim cannot be made against it. Similarly, the legislature could believe it is not imposing costs by a disparate impact standard. Consistently with the disparate impact standard, an employer may use any business-related criteria or test to sort workers. Tests that are not business related cannot improve a firm’s ability to hire qualified workers. True, proof issues and errors in the legal system make disparate treatment and disparate impact laws costly, but these are second-order costs. By contrast, the reasonable accommodation command is for employers to spend money to accommodate the disabled worker (so long as such accommodations do not cause the employer undue hardship). Some argue that in practice most reasonable accommodations can be made with little cost (Blanck et al. 2007). Others argue that reasonable accommodation mandates, even if costly, can enhance the efficiency of internal labor markets (Harris 2003). Nevertheless, the command to reasonably accommodate contemplates costs not contemplated by the negative command to not discriminate in treatment or impact.
Some scholars, including Aigner and Cain (1977), have separated individual from group discrimination. They argue that individual or withingroup discrimination is inevitable. As an example, they note that when college graduates are paid their average productivity and high school graduates paid their (lower) average productivity, some high school graduates will be paid less than their individual productivity (and others paid more, although rarely will an individual complain about being overpaid). But no group discrimination exists. Similarly, they note that racial group discrimination might not occur even when black and white individuals of the same underlying ability are not paid equally, because the overpayment of some might cancel the underpayment of others. In their definition, then, discrimination only occurs when groups with the same average productivity receive different average pay. Antidiscrimination law, however, does not separate individual and group discrimination. An individual black who p. 299is treated worse than a white with the same productivity has a claim under the antidiscrimination laws regardless of whether other blacks might be treated better than whites of corresponding ability.
A final definitional complication is that some writers separate discrimination from segregation. For example, Becker (1971, p. 57) declared that ‘[m]any serious errors have been committed because of a failure to recognize that market segregation and market discrimination are separate concepts’. Racial discrimination occurs when blacks earn less than whites of comparable productivity; racial segregation occurs when race affects job assignments by particular firms. Segregation without discrimination can occur when all-black firms pay workers the same as all-white firms of equal productivity. Discrimination without segregation occurs when blacks and whites of equal productivity work in the same firms, but blacks are paid less. Segregation and discrimination can go together. Strauss (1991, p. 1635) warned of the long-term dangers of ‘separate but equal’ segregation. Antidiscrimination laws prohibit employers both from discriminating in wages and from segregating their workforces.
3 Economic models of discrimination
Economists have developed two prominent models of why employers discriminate – a taste model and a statistical discrimination model. Three other models – a sorting model, a status-production model, and a cartel model – have appeared more recently.
A Becker’s taste for discrimination model
In the 1950s, Gary Becker (1957, 2nd edition 1971) introduced the ‘taste for discrimination’ model that captures the intuitive notion of invidious discrimination. An employer has a taste for discrimination, according to Becker, when he acts ‘as if he were willing to pay something … to be associated with some persons instead of others’ (Becker 1971, p. 14). This definition is general enough to finesse the issue discussed above of which group distinctions are discriminatory, but Becker focused on race discrimination. Assume that group-W and group-B workers (to update Becker’s original notation of W and N) are equally productive and that employers are willing to pay ‘d’ not to associate with B workers. In equilibrium, the market wage for W workers is w, equal to the marginal revenue product of W workers. The wage of B workers, however, is w – d. Under this model, in the short run, black workers will receive a lower wage than white workers of equal productivity and discriminatory firms will earn lower monetary profits than nondiscriminatory firms.
Becker also introduced variants of his model in which customers or employees had a taste for discrimination, meaning that they would demand p. 300lower prices or higher wages when associating with black employees. Profit-maximizing firms will respond by segregating their workforces. If enough nondiscriminatory customers or employees exist, all-black or integrated firms can pay the same wages as all-white firms. If the taste for discrimination is pervasive, black workers will be paid less as well.
Many scholars have asserted that, under Becker’s model, in the long run competitive markets will eliminate firms with a taste for discrimination. Posner (1987), for example, said that discrimination can persist only with some kind of market failure.
The issue is more complex, however. The basic point, not fully realized in the literature, is that markets cater to tastes rather than drive them out. People willing to pay for the costs of apples or safety or discrimination will have it provided to them. True, well-functioning markets confront actors with the full costs of their actions, so only actors who value their taste more than it costs to produce will indulge. In this sense, markets discipline tastes. One cannot indulge the taste for discrimination or apples on a whim, but will have to pay for it. Further, the taste for discrimination in labor markets is tied to another product – profits for employers, wages and other working conditions for workers, and the physical product or service for customers. Markets will confront actors with the costs of tying. Discrimination will be tied to the product only for actors willing to pay for the joint good. But it seems wrong to say, in general, that competitive markets will drive out a taste. Rather, markets drive out tastes that people are not willing to pay for and sustain tastes where value exceeds cost.
The taste for discrimination differs from the taste for apples in that it directly affects other market actors. A customer’s preference for red over green apples lowers the price of green apples, but the green apples don’t care. A customer’s preference for white over black workers lowers the wage of black workers and the black workers suffer lower utility. It may seem that a discriminatory taste imposes an externality on black workers. But this is not so. Externalities occur when the full costs of a market transaction are not borne by the decision makers. Here, however, the discriminatory customer faces higher prices passed on by the employer who hires a more expensive white workforce.
Becker himself was more cautious, and claimed only that ‘under certain conditions’ would competitive markets eliminate discriminatory employers (Becker 1971, p. 45). A number of points – some pointing to market failure but others not – can be made that suggest that a taste for discrimination is consistent with competitive markets in the long run.
First, production technology matters. If constant-returns-to-scale technology exists, so that one firm can expand indefinitely without increasing its average costs, then a single nondiscriminating firm can undersell all p. 301others and produce the entire industry output (Becker (1971, p. 44). But if costs rise with output, more than one firm is required to eliminate discrimination. If employer discrimination is pervasive, there may not be enough nondiscriminating firms to hire all the minority workers and eliminate the wage differential. Becker 1968, p. 210) suggested that a shortage of entrepreneurial skill may prevent firms from fully eliminating discrimination, and thereby declared that ‘discrimination exists, and at times even flourishes, in competitive economies’. Donohue (2003, p. 205) is unconvinced by the shortage-of-entrepreneurs explanation. It did not take special skill to know that hiring low-wage black workers in the textile industry would be profitable – unless, as Donohue puts it, ‘the scarce skill was knowing how to do this without having one’s mill burned down by the Ku Klux Klan’.
Second, a firm with a taste for nepotism, as Becker called it, will thrive in a competitive market. A nepotistic employer gains utility from hiring whites as distinct from a discriminatory firm that loses utility from hiring blacks. An all-white nepotistic employer, then, has a (subjective) cost advantage over both nondiscriminatory and discriminatory employers and thus could produce the entire output, assuming constant returns to scale (Becker (1971, p. 44 n. 4; Goldberg 1982). Donohue 1986, p. 1422) finds the nepotism theory unimportant, arguing that the basic assumption that favoritism rather than animus drives discrimination is ‘arbitrary and unrealistic’.
Third, as mentioned above, employers willing to pay for their taste in discrimination with lower profits can remain competitive indefinitely. Two scenarios are imaginable here, the closely held firm and the jointstock company. An entrepreneur who works alongside his employees may dislike working with blacks, and thus prefer an 8 per cent monetary return with an all-white workforce to a 10 per cent monetary return with an integrated or all-black workforce. He will stay in business if he is willing to reject offers from nondiscriminatory entrepreneurs willing to pay the full monetary value (10 per cent return) for the firm. If only a few entrepreneurs with his discriminatory tastes exist, these few firms will be segregated with no market discrimination. If enough discriminatory entrepreneurs exist, discrimination against blacks as well as segregation will occur. The capital market will pressure discriminatory entrepreneurs to sell out, thereby capturing the high monetary return, and retire or go into another business that does not require association with blacks to make high profits. But this repeats the point that if entrepreneurs are willing to pay for their taste in discrimination, the market allows them to do so. The alternate scenario involves dispersed shareholders who do not physically associate with the firm. Some investors might willingly accept p. 302lower monetary returns in order to indulge in their taste for investing in discriminatory firms, increasing their overall return as they see it. This is the flip side to the South Africa divestiture movement of the 1980s where investors willingly accepted lower returns by refusing to invest in firms involved in South Africa apartheid. If only a small number of investors act this way, segregation may occur without discrimination. If enough discriminatory investors are willing to accept lower monetary returns, they can permanently lower black wages.
Perhaps a more important explanation for long-run discrimination is that profit-maximizing employers in competitive markets will cater to the discriminatory tastes of employees or customers. If some customers will pay less for a product or service from a black employee, or some employees demand higher wages to work with blacks, black employees become less valuable. As Strauss (1991) ably recounted, the result could be segregation or discrimination. The issue is analogous to the debate over whether domestic labor standards are ineffective in the face of foreign competition. There, the assertion is that workers in developing countries willing to work in unsafe conditions or forego other benefits will undercut domestic workers with a taste for safety. While it is far from clear that cheap foreign labor even sets overall levels of compensation domestically (Freeman 1995), profit-maximizing domestic employers will provide benefits that workers are willing to pay for with reduced wages, even in the face of foreign workers with no taste for benefits. Analogously, workers with no taste for discrimination will not undercut workers willing to pay for their taste for discrimination.
Other scholars point to clear market imperfections to explain the persistence of discrimination. Epstein (1992) blames governmental Jim Crow laws for maintaining discrimination. Like other governmental regulation that interferes with labor markets, these laws mandating segregation perpetuated discrimination. Critics of Epstein have noted that Jim Crow laws, while mandating segregated schools, buses, and marriages, almost never regulated labor markets (Verkerke 1992). The impediments were far broader than governmental restrictions. Indeed, a debate in the literature is the degree to which discriminatory social norms can persist without government involvement.
Becker pointed to monopolies as a market imperfection (Becker 1971, pp. 46–7). He emphasized that one of the fruits of a product monopoly is the ability to indulge in a taste for discrimination in the labor market without suffering competitive harm. The firm will earn lower profits than if it hired the cheapest possible labor of a given quality, but can still earn above-average profits. Still, if the monopoly is transferable, a discriminating monopolist has pressure to sell out to those without a taste for p. 303discrimination. In other words, competitive capital markets as well as competitive product markets put pressure on employers.
In a somewhat different vein, Akerlof (1985) shows that discrimination can persist in competitive markets with transaction costs. Suppose some traders have a taste for discrimination against blacks while others do not and that it takes time or money to distinguish discriminators from others. Firms with black employees will miss out on some trades and be under a perpetual cost disadvantage.
B Statistical discrimination
In Becker’s employer-taste-for-discrimination model, employers lose profits by discriminating, even if they gain in utility. The deviation from the profit-maximizing assumption was troubling. This deviation meant that competitive pressures might reduce or eliminate discrimination, although as discussed above, more or less ad hoc or secondary assumptions can explain the persistence of discrimination within the taste-for-discrimination model. As an empirical fact, the seeming persistence of discrimination casts doubt on Becker’s model. As Kenneth Arrow (1972, p. 192) observed, only a poor model ‘predicts the absence of the phenomenon it was designed to explain’. Still, as Paul Samuelson (1951, p. 323) put it in discussing labor market models, ‘In economics it takes a theory to kill a theory; facts can only dent the theorist’s hide’.
In response to the shortcomings of the taste model, economists developed models of statistical discrimination as an application of the growing insights from limited-information theories (Arrow 1973; Phelps 1972; Aigner and Cain 1977). These models assume no prejudice or invidious motive by employers. Rather, employers use group characteristics as a cost-effective way of predicting individual worker attributes in a world of limited information. A profit-maximizing employer wants to hire the worker with the highest expected productivity for the going wage. But gathering information to predict individual productivity is costly. Rather than undertake an expensive inquiry into the quality of the high schools attended by job applicants, the employer might assume that on average blacks attend lower-quality high schools than do whites. Or rather than undertake a lengthy interview, the employer might assume that a woman job applicant is more likely to quit than an otherwise similar man or is likely to put a greater drain on the pension fund by living longer in retirement. These stereotypes are merely statistical correlations, hence the term statistical discrimination. Employers that rely on false stereotypes will face a competitive disadvantage similar to employers acting on a taste for discrimination because they are not hiring the most productive workers. But some stereotypes are true in the sense that the average differs by group even p. 304though the generalization does not apply to many members of the group. Profit-maximizing employers will statistically discriminate whenever the net gains from using the cheap but often-inaccurate proxy outweigh the net gains of more accurate but more costly individualized information.
Statistical discrimination models, while differing in their precise formulations, show how individuals from disfavored groups can receive less pay than individuals with identical ability but from favored groups. The stereotypes can arise from ‘innate’ differences between the groups (differences in longevity being one of the less controversial examples, although even here some argue that women live longer than men because they have been excluded from high-stress jobs), from invidious discrimination outside labor markets (for example, in education), and from the effects of prior labor market discrimination. In Aigner and Cain’s prominent early model, the group differences arose because an individual productivity test predicted white performance more accurately than black performance. Statistical discrimination can persist over time, for it is engaged in by profit-maximizing employers. Nevertheless, pressures exist to reduce statistical discrimination. Money can be made through developing low-cost tests and other methods of predicting individual worker ability that reduce the gains from blunt stereotyping.
It might seem that statistical discrimination increases overall efficiency by allowing individual employers to maximize their individual profits. It is true that in a world of imperfect information, statistical discrimination helps place workers in jobs where their expected productivity is most valued. But statistical discrimination, by tying individuals to group averages beyond their control, creates an externality that can thwart overall efficiency. Statistical discrimination uses average valuations rather than marginal valuations, which are necessary for efficient resource allocation. Individual incentives can be dulled. For example, in the Aigner–Cain model, workers are paid a weighted average of their individual predicted productivity and their group average productivity. Workers in such a model invest too little in training, because they are not completely compensated for individual increases in productivity. As Lundberg and Startz (1983) show, statistical discrimination can exacerbate the distortions. While white workers with statistical discrimination will be encouraged to receive more training, this can be more than offset by the discouragement of blacks to receive training. One empirical problem with the Lundberg–Startz model is that the returns to education are higher for blacks than whites, at least in recent years (Donohue and Heckman 1991b; Sunstein 1991, pp. 29–30), contrary to the prediction that statistical discrimination will discourage blacks from acquiring human capital.
Statistical discrimination might also exacerbate labor-supply distortions, p. 305as shown by Schwab (1986). In this model, workers can work in an individualized market that recognizes individual productivity (self-employment being an obvious example) or in a ‘factory’ that recognizes only average worker ability. Suppose the more productive workers in the individualized market are also more productive in the factory market. If so, some workers will inefficiently choose the individualized market even though they would produce more in the factory market because they would receive only the average factory wage. Now suppose factory employers learn one more thing about their workers, which is their group status, and productivity varies by group. Favored-group workers are encouraged to work more in the productive factory market, but disfavored-group workers are discouraged. Under plausible scenarios, the discouraged workers can outweigh the encouraged workers, reducing overall efficiency. For example, suppose employers statistically discriminate against women because of their greater average quit rate. This may inefficiently divert women toward self- or home-employment. If men are likely to be employees in the outside labor market regardless of statistical discrimination (that is, men have a more inelastic labor supply), the discouragement of women will outweigh the encouragement of men.
C Sorting model
In his important book advocating a repeal of employment discrimination laws, Epstein (1992) introduces a sorting and searching model that blends aspects of the statistical and taste models. For a sympathetic review, see Crespi (1992). Epstein argues first that decentralized markets provide substantial protection against discrimination. Epstein contrasts the victims of violence and the victims of discrimination. Without strong laws against violence, people have to be on guard against violent people. That is why we lock our house even when 99 per cent of our neighbors are friendly. Even if one pays off one violent person, others will arise. Once laws against violence are enforced, people can concentrate on finding people who make good offers. Even if 90 per cent of the people refuse to deal with someone because of their race, he can concentrate on doing business with the remaining 10 per cent. The bigots are powerless to block their mutually beneficial deals.
Second, Epstein recognizes that competitive forces will not totally eliminate discrimination. Much of remaining discrimination is an efficient response by firms to sorting problems. To have smooth workplace arrangements, firms must solve many collective goods problems. What type of firm atmosphere will exist, from choice of piped-in-music to work intensity to degree of office talk? When workers have very different tastes, a firm has difficulty making a decision that will not anger many. ‘To the extent that p. 306individual tastes are grouped by sex, by age, by national origin – and to some extent they are’, says Epstein, smooth operation of the firm may call for sorting of these groups. Epstein recognizes that diversity, say in a sales staff, has benefits as well. Some profit-maximizing firms will opt for diversity, but others for homogeneity. For a related argument, see Cooter (1994, pp. 141–4).
D Status-production and other cartel models of discrimination
Becker’s taste model assumed rather simplistically that individual discriminators wanted to avoid associating with people with certain characteristics and were willing to pay to indulge in this taste. The model ignored the fact that discriminatory whites were often willing to closely associate with blacks, so long as the relationship maintained hierarchy. A prime example was the common practice of whites employing black domestic workers in their homes. More importantly, the taste model gives no importance to group status. Richard McAdams (1995) developed a quite different model which he labels a status-production model. The key element is that whites form a socially connected group that invests in elevating their self-esteem by subordinating blacks. The importance of group solidarity explains why whites will not deviate from the social norm – say by hiring blacks into important positions. Building on Akerlof’s (1985) model, McAdams posits that the deviator loses intragroup status and that the group develops a secondary norm of shunning or otherwise punishing the deviator. Thus, the profit-maximizing entrepreneur calculates that he loses more than he gains by hiring the shunned but cheap blacks.
The status-production model captures the virulent aspects of racism. Its vision of discrimination is more brutal than the prissier taste model which seems to apply more ‘to a kind of tea party discrimination than to the blood and steel of the southern racial scene’ (Higgs 1977, p. 9). As McAdams emphasizes, the model predicts that discrimination ‘will persist in the face of market competition’ (McAdams 1995, p. 1063). It also explains why lower-class whites are most likely to discriminate because they are least able to produce status in other ways.
The status-discrimination model may well explain the Jim Crow American South. More open to question is whether it captures the central features of discrimination today, particularly discrimination based on sex, age or disability. Epstein (1995) has termed the status-production model a sideshow applicable to government-sponsored discrimination of the Jim Crow South but little else. He suggests that the enforcement of discriminatory norms is impossible without violence. As he puts it (Epstein, (1995, p. 1092), ‘[c]oercion is always the main event; status production is the side show’. The question is whether a cartel involving thousands or millions of people p. 307can be maintained without government regulation that punishes deviants. Epstein emphasizes the legal regulation of Jim Crow; McAdams the social norms. For an excellent review of the debate, see Donohue 2005).
4 The economic role of antidiscrimination laws
One’s views on the need for and effectiveness of law depend greatly on one’s model of discrimination. The taste-for-discrimination model sees little need for law. Starting with the premise that markets would root out taste-based discrimination, early law-and-economics writers urged civil rights advocates to rely on markets rather than laws (Milton Friedman (1962; Demsetz 1965). An exception to this early writing is Landes 1968). He posits that fair employment laws raise the cost of discrimination and thus, at the margin, will discourage employers from discriminating and will raise the black–white wage ratio.
Posner (1987) complains that the antidiscrimination laws are often counterproductive. Profit-maximizing employers break down discriminatory barriers by hiring qualified blacks at their lower market wage rate, thereby obtaining a cost advantage over discriminatory employers. But the anti-discrimination laws prohibit an employer from paying unequal wages for equal work (indeed, the US Equal Pay Act specifically prohibits wage discrimination against women, in addition to the general antidiscrimination prohibition of Title VII, which covers wage discrimination based on race or sex). Antidiscrimination laws are less effective at finding employers liable when they refuse to hire minority workers at all. Thus, the laws penalize profit-maximizing employers who pay low wages to blacks, but in practice condone employers who refuse to hire blacks. Posner’s argument assumes that wage discrimination occurs within a firm, but this rarely happens. As Wright (1996) has shown, throughout the 20th century employers rarely paid blacks less than whites for identical work within a firm, even when it was legal to do so. Rather, blacks were excluded from many high-paying jobs and industries.
Even law-and-economics scholars more sympathetic to the basic thrust of antidiscrimination laws have worried about some of its perverse incentives. For example, Ayres and Siegelman (1996) warn that disparate impact litigation might induce employers to discriminate against minorities in hiring. Their conclusion runs counter to the usual assertion that disparate impact litigation (which focuses on the relative numbers of minority and majority workers) leads to hiring quotas by employers. But Ayres and Siegelman show that, in recent years, most disparate impact opinions involve firing cases rather than hiring cases. A firing disparate impact case is easier to show because the relevant pools exist in the firm (for example, a greater fraction of black than white employees were terminated), whereas in p. 308a hiring case great debate concerns the appropriate pool to compare hiring patterns with (all workers? all skilled workers? all suburban workers?). The danger to employers, then, is that hiring large numbers of minorities opens them up to a disparate impact firing suit. For a related argument that antidiscrimination laws do not lead to quota hiring, see Issacharoff (1992, pp. 1238–9).
Other scholars looked with more sympathy on the efficiency justifications for antidiscrimination laws, even within the taste-based model. Donohue (1986) argued that the laws speed up the market’s push towards nondiscrimination. Employers with a taste for discrimination have higher costs than nondiscriminatory employers. Antidiscrimination laws, to the extent they are effective, impose additional costs on employers who discriminate and thus drive them from the market more rapidly. Posner (1987) criticized Donohue for ignoring the costs of government enforcement, arguing that government intervention could make the transition to nondiscriminatory markets inefficiently quick, just as a government mandate that shippers adopt a new, fuel-efficient technology would wastefully cause old ships to be scrapped too quickly. The market is best, asserted Posner, at both eradicating discrimination and determining the optimal timing with which to eradicate discrimination. Posner’s basic objection to Donohue’s argument was that Donohue argued that government intervention was efficient without pointing to some market failure, such as externalities, monopoly or high information costs, that casts doubt on the market solution. Donohue, in response, rejected Posner’s analogy to shippers because it implicitly assumed that shipping was in equilibrium when a new technology was discovered, while the existence of discriminatory employers shows that labor markets were not in equilibrium under the Becker model. Companies with outmoded machines cannot be suddenly made efficient by changing ownership, but companies with a taste for discrimination can be made more efficient by switching ownership, because the psychic costs of discrimination immediately disappear.
Later, taking up more directly Posner’s challenge to point to the market failure, Donohue (1992) pointed to third-party moralists who abhor discrimination directed against others. These outside preferences are not considered when employers negotiate contracts with either their white or their black employees. Even if discriminating employers are willing to pay for their taste for discrimination in reduced profits, the outcome is not efficient because employers are ignoring the external harm their discrimination causes these moralists. Conducting a ‘thought experiment’, Donohue argued that if the average adult American was willing to pay $100 per year to keep Title VII, the resulting $17.5 billion benefits would outweigh a ‘middle-case’ estimate of its administrative and incentive costs. Donohue p. 309(2003, pp. 29–30) has acknowledged that economists generally try not to rely on altruism or the preferences of moralists in making efficiency arguments.
The status-discrimination model makes the sharpest efficiency justification for the antidiscrimination laws. In that model, employers are trapped by the fear of white stigma from integrating their workforces. As Lessig (1995) put it, hiring blacks would mark an employer as having either a special greed for money or affection for blacks. The antidiscrimination laws put an important ambiguity in the decision; perhaps the employer hiring a black merely wants to obey the law. By altering the social meaning of discrimination, the antidiscrimination laws can break the cartel-like solidarity among whites (McAdams 1995).
Even Epstein has recognized the usefulness of the antidiscrimination laws in ending rigidities in labor markets in the American South. As he puts it, ‘Title VII was heaven-sent’ (Epstein 1992, p. 251). Epstein emphasizes that Title VII destroyed the web of indirect legal sanctions against firms trying to exploit the profit potential in discrimination rather than merely private behavior. As he explains, in the pre-Title VII South it was all too easy for an all-white town board to lose a building application or postpone a meeting to approve the sewage connection of a firm that violated the segregation norms of the era. Epstein’s major point, however, is that much contemporary discrimination is efficient, as indicated in his sorting model and in some statistical discrimination models. He therefore concludes that continuance of the antidiscrimination laws imposes heavy costs that are no longer worth the gains.
Other law-and-economics scholars have questioned the means rather than ends of antidiscrimination law. Cooter (1994), for example, argued that the absolute regulatory prohibition against discrimination is an inefficient method of controlling discrimination. Cooter’s premise is that some discrimination is efficient and thus should not be prohibited, but that market imperfections may allow too much discrimination and perhaps justify intervention. Analogizing to the use of taxes or tradable permits to control pollution, Cooter suggests that antidiscrimination laws should tax employers who employ too few blacks or women or give employers tradable rights, rather than prohibit discrimination altogether. A tax or tradable rights regime can reach equivalent levels of discrimination differing only in whether government officials adjust prices or quantities. Either method can enforce discrimination laws at lower cost than a flat prohibition, in part by allowing firms with especially high costs of complying (for example, few qualified minorities are in the pool) to hire fewer minorities. Strauss (1991) makes a similar argument, phrasing it as a preference for disparate impact analysis than disparate treatment analysis. Current p. 310disparate treatment law prohibits firms from ever considering race in a decision. Strauss argues that this is costly to enforce because it requires extensive case-specific findings. Many cases of invidious discrimination go unchecked, while some cases of efficient discrimination (particularly efficient statistical discrimination) are banned. Far better, Strauss says, would be to require employers to pay a fine if they do not hire proportionately to the nationwide labor force. If a firm has correct bottom-line numbers, it can then make individual decisions to hire or fire without scrutiny by the antidiscrimination laws. It should be pointed out that Strauss uses an unconventional conception of disparate impact because he would not require the government to point to a specific employment practice, such as a test, that caused the firm to hire fewer blacks. In a similar vein, Mashaw (1991) argued for racial quotas, with firms with a small percentage of blacks in their workforce being allowed to buy permits from firms with large black workforces. One problem with such tax, fine, or tradable rights schemes is that the government needs detailed information to set the optimal tax rate. Enormous wasteful lobbying would occur by rent-seeking advocates pursuing higher or lower rates or special exemptions. A greater problem is the symbolism of allowing trade in rights to discriminate, although this symbolism has been overcome in allowing tradable pollution rights. Bell (1992, pp. 47–64) savagely satirized the concept of tradable rights. Bell calls this hypothetical Racial Preferencing Licensing Act a ‘legalized reincarnation of Jim Crow’.
McCaffery (1993) likewise advocated tax reform as a solution to sex discrimination. McCaffery sees extensive market failures that create disparities between men and women, including search costs and incomplete markets, but he emphasizes tax policy. The result of these market failures, he suggests, is that women have been given a stark choice: act like men and become highly committed to the paid labor force, or get out. The market has been dictating choices rather than accommodating them, as Epstein would see it. McCaffery advocates higher taxes on primary workers, typically married men, and lower taxes on secondary workers, typically married women. These changes could unravel the dynamic that leads to polarized options for women.
5 Empirical studies of employment discrimination law
Unlike law-and-economics scholarship in some other areas, the scholarship in employment discrimination has gone beyond model building and taken a serious empirical look at discrimination litigation and the effects of antidiscrimination law. The amount of employment discrimination litigation in the United States has exploded over the last 40 years. Between 1970 and 1989, employment discrimination case filings started from a tiny p. 311base to rise by 2166 per cent, compared to a rise of only 125 per cent in the overall civil docket (Donohue and Siegelman 1991). By 2000, employment discrimination cases comprised nearly 10 per cent of the federal civil docket (Clermont and Schwab 2004; Clermont et al. 2003; Eisenberg and Clermont 1995). Employment discrimination cases are less likely than other cases to privately settle out of court. The percentage of employment discrimination cases going to trial has fallen dramatically over the last generation from 18.2 per cent in 1979 to 3.7 per cent in 2001. Still, a substantially higher percentage of employment discrimination cases go to trial. While the ratio of other trials to discrimination trials was 10:1 in 1979, it was only 4.66:1 in 2000 (Clermont and Schwab 2004). Employees win at trial less frequently than other plaintiffs and their judgments are reversed on appeal more often than other plaintiffs (Clermont and Schwab 2004).
Most employment discrimination cases protest firings, not refusals to hire. The common image of discrimination is that employers reject black or women job applicants or pay black or women workers less for the same work. Indeed, in the early years of Title VII litigation, most cases were hiring cases. By the 1980s, however, firing cases were six times more common than hiring cases (Donohue and Siegelman 1991, p. 1015). An employer is thus far more likely to be sued when it terminates an individual worker than when it refuses to hire job applicants. This makes employers more reluctant to hire black or women workers in the first place.
Paradoxically, societal success in overcoming discriminatory barriers may explain much of the rise in discrimination litigation (Donohue and Siegelman 1991, pp. 1006–15). A worker is more likely to sue when rejected from a high-paying job than a low-paying job because the payoff of a successful lawsuit increases while the costs of litigation vary little with pay. Further, lawsuits protesting discrimination are more likely as the workforce becomes more integrated because it becomes easier to show that the rejected black or woman was treated less favorably than a white male. Thus, as women and blacks enter high-paying, integrated jobs, the number of discrimination lawsuits rises.
In addition to long-term trends, discrimination lawsuits respond to business cycles. The number of lawsuits and amount of damages rise during recessions, but win rates fall (Donohue and Siegelman 1993). The key link is that victims of discrimination receive higher damages the longer they are out of work.
Moving beyond empirical studies of the litigation process, several studies have examined whether the antidiscrimination laws have improved blacks’ earnings. Such studies are notoriously tricky. Essentially, the researcher is trying to examine changes after a nationwide law goes into effect when many other changes in society occur as well. In the case of p. 312US antidiscrimination laws, Title VII became effective in 1965, coinciding with important executive orders requiring affirmation action by government contractors, a new Voting Rights Act, continuation of the civil rights movement, and general social unrest. Smith and Welch (1989) argued that federal antidiscrimination laws had little effect on black economic progress because black/white earnings ratios had been steadily increasing since the 1940s. Better black education and migration from the South to higherpaying jobs in the North are the most important explanations for black economic progress, they suggest, neither of which is directly attributable to law. Donohue and Heckman (1991a) contested this conclusion. As the title of their review article indicates, they find black economic progress to be episodic rather than continuous. The first period of progress occurred around World War II. The second critical episode was the period from 1964 to 1975, when the black/white earnings ratio increased from 0.62 to 0.72, with most of the progress coming in the American South (Donohue and Heckman 1991, p. 1607). Black outflow from the South had slowed dramatically by 1965 and improvements in amount and quality of blacks’ schooling explain only part of the post-1965 jump in blacks’ earnings. Donohue and Heckman conclude that Title VII, along with other civil rights legislation of the period, had a positive impact on black/white earnings by shaking the economic and social taboos of the South that had prevented employers from hiring black workers. A particularly important example comes from the textile industry in South Carolina, as examined by Heckman and Payner (1989). Prior to 1965, few blacks were employed in the South Carolina textile industry when black employment levels and wages suddenly rose. Shifts from agriculture and improved black education cannot explain the timing. The mid-1960s saw a tight labor market making the underemployment of blacks particularly costly. Heckman and Payner find quite plausible the story that in 1965 entrepreneurs ‘seized on the new federal legislation and decrees’ and began hiring black workers as they had wanted to do. Epstein (1992, p. 245) attributes the surge of black employment to the federal government’s attack on Jim Crow regulation, pointing out that South Carolina had a 1915 statute, not formally repealed until 1972, mandating segregation in textile factories. Verkerke (1992, p. 2091) responds that nearby Southern states also had highly segregated textile factories but no express segregation law.
Some intriguing studies have exploited changes in coverage of Title VII to measure its effectiveness. In 1972 Congress expanded the coverage of Title VII from employers with 25 employees to employers with 15. Chay (1998) found that relative black employment and earnings rose in industries in the South with a high percentage of small employers, suggesting that the statute increased the relative demand for blacks. Bloch p. 313(1994) also emphasized differences in coverage, noting that the most intensive enforcement of antidiscrimination policy is for federal contractors, followed by employers with 100 or more employees required to file detailed EEO-1 reports on their hiring practices, followed by employers with 15–100 employees. While significant data problems exist, he shows that minorities appear to have greater representation in more regulated firms. Welch (1989), as reported in Donohue (1989), likewise found that employment of blacks and women increased between 1966 and 1980 for EEO-1 reporting firms. Bloch warns, however, that this pattern does not necessarily show any aggregate increase in black employment or wages and may represent only a shift in black employment from the uncovered to covered sector.
Most analysts agree that, after the 1965–75 surge, blacks have made far less economic progress and that blacks’ earnings remain substantially less than for whites, even after adjusting for education and other productivity factors. The antidiscrimination laws seem to have had little effect in the last 30 years (although it is always hard to disprove the negative assertion that the laws prevented a decline). Whether more vigorous enforcement of the antidiscrimination laws would help, or whether they have run their course and further progress must come in other directions, is unclear. Acemoglu and Angrist (2001) and Blanck et al. (2007) join this debate in the context of the Americans with Disabilities Act’s (ADA) effects on the employment rate of workers with disabilities. While the differing story between the efficacy of the ADA and Title VII is certainly not without controversy, it is consistent with the notion that the ADA, by design, imposes costs on employers, whereas Title VII does not.
6 Special issues in sex discrimination laws
Much of the general law-and-economics approach to employment discrimination applies equally to sex, disability (but for the reasonable accommodation requirement) and age discrimination, as well as race discrimination. Sex discrimination raises a number of special issues, however, and as Epstein (1992) emphasized, the parallels between race and sex discrimination are imperfect at best. As Wasserstrom (1977) ably explained, color-blindness is the eventual goal of the antidiscrimination laws, but sexblindness is not. No employer could provide separate but equal bathrooms for blacks, but all large employers provide separate bathrooms for women. Pregnancy and childbirth uniquely affect women workers.
The economic models fit differently for race and sex discrimination. Becker’s taste-for-discrimination model is formulated generally enough to accommodate a preference against associating with women or blacks. It seems to fit well enough the fact that many bosses, co-workers, or p. 314customers prefer dealing with men (and, in other situations, dealing with women). Perhaps men gain status by maintaining group solidarity against women, but McAdams’s status-production model seems somewhat hollow when applied to sex discrimination. Statistical discrimination models, by contrast, often apply well to sex stereotypes. Epstein (1992, p. 65) applied his sorting model to sex discrimination, emphasizing the divisions within the firm that can arise when women want part-time status. In Posner’s (1989) view, the economics of sex discrimination is distinctive because of the interdependence in utilities of men and women. Married couples have joint consumption and often act altruistically towards each other. Thus, for example, married women indirectly benefit if labor market discrimination increases husbands’ pay. Because racial intermarriage remains rare, blacks do not indirectly benefit when whites are favored.
Antidiscrimination law has recognized differences between race and sex discrimination. Most important is the bona-fide occupational qualification (bfoq) defense under Title VII which applies to sex but not race distinctions. Title VII prohibits employers in any context from using race to sort workers, but allows employers to sort by sex if sex is a bfoq. Thus, employers can have separate but equal bathrooms or grooming standards for men and women, but not for race.
Women tend to live longer than men. Many employers acted on this ‘true stereotype’ in constructing pension plans, using sex-based annuity tables that required women to make higher contributions for equivalent monthly pensions or receive lower monthly pensions for equivalent contributions. The practice is a paradigmatic example of statistical discrimination. No malicious discrimination was intended, but employers were explicitly using sex as a proxy to predict longevity. Nevertheless, it is illegal in the US. In two cases, Los Angeles Department of Water and Power v. Manhart and Arizona Governing Committee v. Norris, the Supreme Court prohibited employers from using sex-based tables in calculating individual contributions or benefits (although the Court allowed employers to consider the overall percentage of women in calculating whether overall contributions match benefits). The Court reasoned that the antidiscrimination laws demand fairness toward individuals before fairness to groups and the sex-based classification involved group generalizations rather than ‘thoughtful scrutiny of individuals’. In short, the Court thus objected to statistical discrimination even when ‘true’.
An extensive law-and-economics commentary has criticized the pension decisions, arguing that employers should be allowed to consider sex in calculating pension contributions or benefits (Epstein 1992; Benston 1982; Freed and Polsby 1981; Kimball 1979). They see employers or their insurance companies using sex classifications to solve an adverse selection p. 315problem. With unisex tables, women covered by a pension plan will receive 6 to 12 per cent more than they contribute to the plan (Epstein 1992, p. 323, citing Benston 1982, p. 515). These distributional consequences reduce overall welfare as plan participants begin ‘gaming’ the system. As Epstein points out, strategic considerations become largest when a husband and wife are each entitled to pension benefits. With unisex tables, the couple has an incentive to shift their benefits towards the wife by having the husband take a self-and-survivor annuity and the wife take an annuity for her life alone. Employers have an incentive against hiring women or toward dropping life-time pension benefits.
Similar controversy arose over the issue of whether employers violated antidiscrimination laws by excluding pregnancy from health insurance benefits or pregnancy leave from disability benefits. Epstein (1992, p. 32) emphasizes that pregnancy is a poor candidate for insurance because it is largely a controlled event and thus a moral hazard problem is created by insuring its costs – women are more likely to become pregnant. Many European countries, concerned with low birth rates, regard this as a favorable aspect of insurance and provide for pregnancy health costs (generally as part of national health insurance) and paid maternity leave with government programs. Indeed, whether childbirth benefits should be provided through government programs or employer mandates is a good example of the tradeoffs articulated by Summers (1989). In a thoughtful article, Issacharoff and Rosenblum (1994) contrast the extensive European government benefits toward pregnancy with the limited employer mandates in the United States – essentially a nondiscrimination command and an employee’s right to 12 weeks’ unpaid leave for childbirth. They propose a more extensive set of benefits modeled after unemployment insurance.
In the early US pregnancy cases, the Supreme Court upheld employer benefit plans that excluded pregnancy. The Court reasoned that a classification between pregnant persons and other persons was not explicitly sex based and therefore was not disparate treatment; and because the costs of providing health insurance were as great for women (even without pregnancy coverage) as for men, no disparate impact claim was made. Congress promptly reversed this holding, with the Pregnancy Discrimination Act amending Title VII. Employers must treat pregnancy like any other disability. Posner (1989) notes that this compels employers to ignore the higher average cost of employing women who are more likely to leave the workforce to have children. The consequence of ignoring these differences, Posner argues, is inefficient and may not benefit women as a whole. Women directly lose because employers have an incentive not to hire them. Applying his interdependence analysis, Posner argues that women as wives indirectly lose as their husbands’ wages fall. The biggest losers are married p. 316but childless working women, because the wife and husband have lower wages than if employers could consider pregnancy and she will not gain from the higher child-related fringe benefits.
Donohue (1989) challenges Posner’s argument that statistical discrimination against women is socially efficient. Following the approach of Lundberg and Startz (1983) and Schwab (1986) discussed above, Donohue reiterates that while acting on statistical averages may maximize firm profits, it creates dynamic distortions. For example, if employers are reluctant to train women for top corporate positions because they tend to quit to have children, even women who plan to stay in the labor market have reduced incentives to invest in human capital. Donohue also more broadly challenges Posner’s interdependency thesis. Even if consumption is joint and women are altruistic, Donohue says, women may have different preferences and may want altruistically to spend more on their children than their husbands would wish. If sex discrimination lowers the economic power of women, their bargaining within a marriage is probably weakened.
Sexual harassment continues to be a major issue in employment discrimination law and the law-and-economics literature helped frame the issues. Hadfield (1995) used an avowedly economic approach to define sexual harassment. Harassment occurs, under her test, whenever a woman is subjected to actions that a man is not subjected to that would cause the woman to alter her behavior to avoid the actions if she could do so without cost. Hadfield emphasizes that many women are trapped (not at the margin) and thus must endure the harassment, but if possible might refuse night shifts, overtime or travel assignments where harassment is likely. Whatever the definition, Posner (1989) notes that sexual harassment is not in the employer’s self-interest, but often the costs of prevention are high. Just as other antisocial workplace behavior, such as embezzlement, is subject to public enforcement, so too should sexual harassment be – depending on the relative costs of public and private enforcement. Epstein (1992) agrees that prohibitions on sexual harassment are perhaps the easiest antidiscrimination prohibitions to justify, at least if the claims stay close to their tort roots. Verkerke (1995) explored the standards upon which employers should be liable for sexual harassment by its employees. He argues first that employer liability for creating a sexually hostile environment should not differ from employer liability for other forms of discrimination, but rather should depend on creating the proper incentives for employers to acquire information about harassment. Conditional notice liability, whereby employers are not liable if they have ‘experimented’ in creating notice procedures and have not been notified of the harassment, may be most appropriate. But Verkerke emphasizes that the choice of the optimal liability standard requires data currently not available.