Discrimination

One of the most hotly debated and widely discussed sources of inequality is discrimination.  There is no question that there are substantial differences in the earnings of males and females and Whites and Blacks.  There is also no doubt that many take this as evidence of gender and racial discrimination, just as some wanted to take average earnings differences between male and female faculty as evidence of gender discrimination.  Unfortunately, proof of discrimination is quite a bit more difficult than a simple comparison of average earnings.  In fact, if one is to make any sense of the research on discrimination, it is necessary to begin with defining discrimination.  

According to Yinger, one of the authors in a collection of articles in the Spring 1998 volume of The Journal of Economic Perspectives, the "civil rights laws define discrimination as the unfavorable treatment of a person based solely on the basis of that person's membership in a 'protected class.'" This is reflected in the 1968 Fair Housing Act that states "it shall be unlawful . . . to discriminate against any person . . . because of race, color, religion, sex, or national origin."  The discrimination can take two forms - the '"disparate treatment' of  customers on the basis of their membership in a protected class.    . . .  the use of practices that do not explicitly consider a person's group membership but instead have an adverse impact on a protected class without any 'business necessity.'"  If an action is shown to adversely affect a given class, then the burden is on the business that must support its decision.

Some economists, the most notable being Becker, take a narrower view of discrimination.  Becker (1971) sees discrimination as the product of "prejudice or a 'taste for discrimination' and it requires the discriminator pay or forfeit income for the privilege of exercising prejudicial tastes." (Ladd, 1998)  To help move the discussion forward, we will distinguish four separate "types" of discrimination and examine their life expectancies. Will these forms of discrimination persist and their elimination depend upon active government intervention, or will the market system tend to eliminate them over time of these types. The four discrimination types are employer, statistical, employee, and customer discrimination.

Employer discrimination exists when employers restrict their hiring to certain groups.  An example would be a decision to restrict hiring to males or to people over 6 feet tall despite the fact that none of these decisions is based on productivity differences.  In terms of the supply and demand model of the labor market, demand would be lower for those individuals discriminated against so we would expect wages to be lower for females and those under 6 feet.  

But will this discrimination persist or will it eventually disappear?  According to Becker, this discrimination will disappear because firms that discriminate will pay more for their labor and this will raise their production costs to a point where non discriminating firms will be able to undercut them.  In the end the high-cost, discriminating firms will be driven from the market and employer discrimination will be eliminated. 

As appealing as this view may be for economists, it is hard to reconcile this with the fact that discrimination clearly existed as late as the mid 1960s.  Darrity and Mason (1998) provided some rather startling examples of racial preferences in hiring that appeared in Help Wanted advertisements of some of the nation's largest papers.  It was not until passage of the federal Civil Rights Act of 1964 that the explicit discrimination that these ads reflect began to disappear.  Either Becker's view is wrong or the adjustment process is so long as to make it irrelevant.

A second form of discrimination would be statistical discrimination, an idea developed by Arrow and Phelps.  Statistical discrimination exists when individuals attach different estimates of productivity based on the group to which the individual belongs. For example, decision makers often need to make choices with incomplete information and as a result they tend to make assumptions to fill in the gaps. For example, an older worker may lose out to a comparably trained and skilled younger worker because the employer believes the older worker is likely to retire. The younger person may get the offer even though the older individual has no plans to retire and the younger person is actively pursuing opportunities in other fields.  As one of my students pointed out in class, the same situation would exist when males were favored over females because of a belief that the 'odds' are that the female is likely to take time out to raise her family.  While this may be true of the averages, there is no reason to believe that the average provides much insight into the individual situation.  

Will statistical discrimination be competed away?  Statistical discrimination will be competed away if the statistical profiles were based on incomplete or inaccurate information.  It is true that females have born the primary responsibilities for raising children and they are most likely to be taking time off after the birth of the child, but this is only part of the story since there are other aspects of females' behavior which would raise their productivity.  To the extent decisions based on statistical averages were wrong, the firms that practiced it would be less competitive and they would eventually lose out to non discriminating firms.

Two other types of discrimination that will not be eliminated by discrimination are employee discrimination and customer discrimination. Employee discrimination exists when employees treat different groups differently. If, for example, men do not like working for a female supervisor, then you are likely to find a situation where men are less productive working for women.  As a result women will be less productive as managers and demand for female managers would be lower.  This would push their earnings lower and it would limit their opportunities for advancement to the upper reaches of management - a possible explanation for the "glass ceiling" that you are likely to have heard about.

Customer discrimination exists when customers have discriminatory preferences.  An example would be fans of professional sports that were willing to pay more when watching teams that had more White players - a result observed in a 1988 study of professional basketball games.  In this situation you would expect the marginal revenue product of these players would be higher and this would translate into higher earnings.

The common denominator in employee and customer discrimination is that they are unlikely to be eliminated by the workings of the market.  If female managers tend to be less productive because of their male employees, then the firms are likely to be less competitive and competition will therefore not work to eliminate this discrimination.  Similarly, if customers are willing to pay more to see White basketball players, then the market will continue to generate earnings premiums for the these basketball players. 

It is clear from the above discussion that discrimination can take many forms and that it's contribution to earnings differentials can be significant. The same is true in output and capital markets.  Yinger (1998) reviewed the research on the housing and auto markets and concluded that "discriminatory barriers to consumption show no signs of diminishing over the last 20 years. . . . Black and Hispanic households still face a significant chance of encountering discrimination when they inquire about housing or visit a car dealer."  Ladd (1998), in her review of the research on discrimination in mortgage lending by the Federal Reserve Bank of Boston, concludes "it is clear that mortgage lenders discriminate against minorities."  Finally, Darrity and Mason (1998) conclude their review of discrimination in labor markets by noting that "the strong evidence of the persistence of discrimination in labor markets calls into question any theoretical apparatus that implies discrimination must inevitably diminish or disappear." 

But what were these claims based on - how does one go about proving the existence of discrimination?  Most economists would begin with the premise that there are an array of factors that influence earnings and discrimination would exist only if after we have explained away differences due to the ability and skills of the individual, characteristics of the jobs, and differences in market power, we still find some differences in earnings.  The problem is it is very difficult in practice to identify what part of the differential is due to discrimination and what part is due to other 'economic' factors. 

To get some idea of the nature of the problem consider the data that indicates in 1995 the average earnings of year-round, full-time females was two-thirds the level of males.  Some would argue that this was evidence of discrimination, but economists would tend to look beyond these simple summary numbers For example, based on the earlier analysis, we would expect differences in earnings based on a list of factors that would include age, experience, occupation, education, industry, and location and that any measure of average earnings would depend upon the mix across these various dimensions.  The fact that in 1996 the share of workers under age twenty-five, workers who we would expect to have lower average earnings, was ten percent higher for females than males would suggest that some of the differences in average earnings is due to difference in the age distribution of workers.

There are also differences in average earnings that could be attributable to occupational mix.  In the health care industry, approximately one-quarter of physicians are females while more than ninety percent of all dieticians and registered nurses are females.  Given the differences between the earnings of physicians and registered nurses, we can expect the average earnings figures for males to be higher than that for females.  It is also true that males tend to have more work experience which should on average translate into higher earnings.

The bottom line is that existing earnings differentials are not to be considered as evidence of discrimination and that measurement of discrimination is quite difficult.  Two approaches that go beyond this simple comparison are regression analysis and audits.  Regression analysis is designed to determine the extent to which race or gender explain the value of some market performance variable.  In the housing industry the variable could be home prices, in the mortgage market it could be rejection rates, and in the labor market it could be earnings.  In each instance there would be an equation estimated that would include as explanatory variables characteristics of the variable (neighborhood, type of car . . .) and characteristics of the individual (income, gender, race ...).  Discrimination would be proved if, after the impact of all other factors had been accounted for, race or gender still were shown to be related to the variable in question, that race or gender helped explain differentials in home prices, earnings, car prices, or mortgage rejection rates.  [If you have some experience with regression analysis, you may want to look at an overview of the regression approach to identifying discrimination].

In each instance the attempt is made to uncover discriminatory behavior without ever directly observing it.  The audit approach is a matched-pair survey technique that allows for direct observation of discriminatory behavior.  People from two different groups, one from a protected class, are selected and trained to work as pairs.  Every effort is made to eliminate all differences between the pairs with the exception of their class status.  For example, if this were a test of the housing market, two individuals would be primed to dress similarly and would be given similar socio-economic profiles - similar jobs and earnings.  These individuals would then apply to a variety of loan providers and their treatment would be recorded.  If any differential in treatment is observed, it would be attributed to discrimination. 

Each of these techniques has its strengths and weakness, but the weight of evidence seems to point rather convincingly toward discrimination as a significant factor in explaining earnings differential - a first step toward understanding inequality and poverty.