Prof. Bryan Caplan

Econ 321


Weeks 12-13: Discrimination

I.             Wage Differences versus Wage Discrimination

A.           People don't earn the same income, and neither do groups.  There are, on average, large wage differences.

B.           From the NLSY (1992 data): Average annual labor income was $17,100.  Compared to white males, what did members of other groups earn on average?


Labor Income Gap



Other Non-White




C.           No one disputes that there are large wage differences.  The debate, rather, is about why.  Are these gaps partly or wholly explained by the fact that groups differ in average characteristics relevant to marginal productivity?

D.           Two kinds of characteristics: the ones we measure (or "observe") like education and IQ, and the ones we don't, like culture and creativity.  Can wage differences be explained by differences in observable characteristics? 

1.            If so, we don't even need to worry about unobservable characteristics.

E.           Let's start with an easy one.  What if we control for marital status and number of children?  A lot of women have no labor income because they don't work and/or don't work as much because they take care of kids. 

F.            Suppose we compare never-married, childless males and females?  The -$12,000 gap shrinks to a mere -$1,100 gap!  It's not even "statistically significant" as econometricians say.

G.           Now let's move to something harder.  Is there any way to account for racial income differences?  Let's start by controlling only for education and experience.  What then?

H.           The "other non-white" gap essentially disappears, but the white-black gap only mildly shrinks.


Labor Income Gap



Other Non-White


I.             What if, following up on earlier discussions, we also control for measured intelligence?  (The NLSY administered extensive intelligence tests to people surveyed). 

J.            Other non-whites actually earn more than observably identical whites; the white-black gap drastically shrinks.


Labor Income Gap



Other Non-White


K.           Many scholars who have studied black poverty have put some blame on differences in family structure.  On average, blacks are much less likely to marry and remain married; yet blacks on average have more children.  What if we add in controls for family variables?

L.            Remaining black-white gap shrinks still further, becoming statistically insignificant.  Other non-whites look even better off than before.


Labor Income Gap



Other Non-White


M.          There are definitely large differences in labor earnings, and they match the popular stereotypes about which groups the market treats "unfairly."

N.           But it is wrong to infer discrimination from inequality.  You must control for real group differences first.

O.           Once you do so, there is little evidence of discrimination.  (And some of it cuts the wrong way!)  Labor income differs between groups because - on average - groups differ in education, intelligence, family structure, etc.

II.            Compensating Differentials and Apparent Discrimination

A.           Suppose some differences did persist controlling for observable characteristics?  In the interests of full disclosure, adding controls for education, experience, and IQ actually increases the male-female gap to -$2000.  (Still not statistically significant, though).

B.           Should we immediately infer discrimination?  Another explanation: Different fun/money trade-offs.

C.           One especially plausible instance: Women seem much more likely than men to enjoy jobs that involve nurturing and caring.  Teaching and nursing are the classic examples.

1.            In contrast, on average, men seem to focus more on money.

D.           If these claims are right, then without discrimination of any kind, women will earn less.  Why?  Because more of them trade-off money for satisfaction.

E.           More generally, if on average some groups focus more on money, we should expect to see wage gaps.  If we had good measures of "focus on money" we could even statistically control for it. 

III.          Discrimination as a Preference

A.           We have seen that the empirical case for discrimination is weak. 

B.           Interestingly, many economists doubted - on theoretical grounds - that discrimination had much effect long before much data was available.

C.           Why?  Let us begin by defining "discrimination" more precisely.  In economic terms, we can think of pure dislike or hatred for others as a taste for discrimination, a willingness to pay to avoid people you don't like.

D.           For example, suppose a Serbian employer hates Croatians.  But how much is he willing to pay for this?  Would he give up $1,000,000 to avoid hiring a Croatian?  Probably not.  There is some amount of money sufficient to make the Serbian hire the Croatian in spite of his discriminatory taste.

E.           Similarly, how much in wages would an Israeli worker be willing to give up to work at a firm with no Palestinians?

F.            Or, how much extra would a Romanian consumer pay to shop at a Romanian-owned store rather than a Turkish-owned store?

IV.          Discrimination by Employers

A.           Once we understand this notion of the "taste for discrimination," we can use it to analyze a variety of cases.  Let us begin with employer-on-worker discrimination.

B.           Assumptions:

1.            Most employers have a taste for discrimination against Asians.  Their willingness to pay to satisfy this taste ranges from $2/hour/worker to $0/hour/worker, with an average of $1/hour/worker.

2.            No one else has discriminatory tastes.

3.            Asian and non-Asian workers are equally productive.

4.            Labor markets are competitive and there are no anti-discrimination laws.

C.           What happens?  Labor demand for Asians is lower and they earn lower wages - at first.

D.           Who hires them?  The least-discriminatory employers!  If the wage gap is $1.00, then employers who value discrimination by less than $1.00 hire only Asians.

E.           More racism thus means lower profits.  Less racist employers hire cheaper Asian labor, while more racist employers higher more expensive non-Asian labor. 

F.            Thus, over time the most racially tolerant employers become a larger and larger part of the market, and racist employers are driven out of business. 

G.           This shifts employers' distribution of discriminatory tastes in the direction of tolerance - raising the demand for Asian labor and reducing the demand for non-Asian labor.  So the wage gap falls.

H.           As long as there are enough employers who care solely about money, not race, the ultimate effect is that racist employers are driven from the market, and equally-productive labor earns the same wage.

I.             Even if most people are racist, selective pressure favors non-racist employers.  Businesspeople are competing to make money; any goals other than making money - good or bad - hold them back.

J.            In other words, more greedy, less racist employers tend to drive less greedy, more racist employers out of business.

K.           Corollary 1: Government regulation is necessary to sustain discrimination by profit-seeking employers.

L.            Corollary 2: Discrimination is much more likely to appear in the non-profit sector.

V.           Discrimination by Workers

A.           We now turn to worker-on-worker discrimination.

B.           Assumptions:

1.            All non-Asian workers have a taste for discrimination against Asians. 

2.            No one else - including employers - has discriminatory tastes.

3.            Asian and non-Asian workers are equally productive.

4.            Labor markets are competitive and there are no anti-discrimination laws.

C.           Employers who make non-Asians work with Asians will have to pay the non-Asians a compensating differential.  This reduces demand for Asian labor.

D.           Simple solution: segregated workplaces.  If non-Asian workers don't like Asians, employers can save money by setting up all-Asian plants. 

E.           Given the assumptions, this leads to full segregation and equal wages for both types of employees.  Racism doesn't disappear, but it doesn't have any impact on wages.

VI.          Discrimination by Consumers

A.           Last case - suppose consumers don't like Asians.  What then?

B.           Profit-maximizing solution: move Asian workers out of the public eye - essentially, another form of segregation.

C.           This does mean lower demand for Asian labor, and lower Asian wages, but the effect is probably small.  People rarely know anything about 95% of the people who worked to produce their groceries.

D.           Still, markets are less likely to weed out discrimination by consumers than any other form of discrimination.

E.           But how common is it?  Consumers today are probably more inclined to boycott firms for racism than tolerance.  (Note further that anti-discrimination laws provide little protection against consumer-on-worker discrimination).

VII.        Occupational Discrimination and Economies of Scale

A.           The effects of worker-on-worker discrimination become more severe in industries with large economies of scale.

B.           Why?  If there are few economies of scale, then any disliked group of workers can get a "firm of their own" to avoid hostile co-workers.

C.           As economies of scale rise, this becomes less feasible.  You can't have an all-Albanian auto plant in the U.S.

D.           Similarly, if there are very few people of a disliked group in an industry, it will be hard for them to have a "firm of their own."

E.           This can conceivably be a self-reinforcing situation.  Auto firms won't hire blacks; there aren't enough black autoworkers to set up their own firm; and since auto firms won't hire blacks, blacks don't learn how to become autoworkers.

F.            In practice, though, people worked through cracks in the system.  Some firms' workers are less racist than others.  Minority workers who wanted to enter a non-traditional occupation sought them out and got their start there.  Once you reach a "critical mass" of workers in an occupation, the separate firms solution becomes viable.

G.           In a number of interesting cases, occupations started out as hobbies, creating the necessary "critical mass" indirectly.  Minorities in athletics and entertainment are a good example.  (Incidentally confirming that consumers don't care much about race).

VIII.       Stereotypes and Information Economics

A.           Gathering more information takes time, and time is foregone income.  Thus, people inevitably - and sensibly - quit gathering information once they think their understanding is "good enough."

B.           Of course, "mistakes will be made."  People are choosing between two evils - wrong judgments and lost time.

C.           This is the essence of stereotyping: Generalizing in a useful but fallible way based on limited information.

D.           People use stereotypes all of the time.  You may have wondered if I was the professor on the first day of class.  Why?  Because I don't fit the stereotypical age of a professor.  Were you irrational to use this stereotype?  Hardly.  Most professors are older - I am still the youngest faculty member at Mason.

E.           What would your day be like if you used no stereotypes?  You use stereotypes about traffic patterns to choose your route to school.  You use stereotypes about campus police to decide whether to illegally park.  You use stereotypes about couples to guess whether two people are married.

F.            Many people think stereotypes are plainly false.  But it's an empirical question.  This is a huge topic, but there is a lot of evidence that most stereotypes are right on average most of the time.

G.           Moreover, people who don't like stereotypes still use them.  "Police are bigots" is a stereotype.  "White people make more money than black people" is a stereotype.  Both may be true on average, but they are stereotypes nevertheless.

H.           Not sure?  Test your own stereotypes against objective statistics.

I.             The basic stereotype fallacy: Confusing averages and universals.  But does anyone actually do this?

IX.          Statistical Discrimination

A.           Suppose employers rely on a stereotype to make employment decisions, and that stereotype is true on average.

B.           Is that "discrimination"?  In a sense, yes - you are being judged for your group, not yourself.  But in another sense, no - the group differences are real, and people don't dislike your group as such.  Economists call this statistical discrimination.

C.           A good example: gender and auto insurance premiums.

D.           Another example: who cabbies will pick up late at night.

E.           Unlike taste-based discrimination, statistical discrimination can survive and thrive in markets.  If group differences are real, and it is costly to judge case-by-case, then people who don't discriminate lose money.

F.            Important point: Statistical discrimination does not reduce mean group income.  It just narrows the distribution.  People who exceed their group stereotype's performance level are under-paid; people who fall short of their group stereotype's performance level are over-paid.

G.           Once they understand the idea of statistical discrimination, many people become concerned about "self-fulfilling prophesies."

1.            Ex: People think teen-age males are criminally inclined (and they are), this angers the teen-age males, leading them to commit more crimes.

2.            Ex:  People think men aren't good with children.  So no one lets men work with children, and as a result their skills do not develop.

H.           This is possible, but hardly the only possibility.  Perhaps members of stigmatized groups respond by trying harder to distinguish themselves from their group average.

I.             Interesting psychological research exists along these lines: When individuals clearly violate stereotypes, people over-react.  This means that the marginal payoff of demonstrating ability is actually greater if people assume you're less able because of your group.

X.           The Effect of Discrimination Laws

A.           Suppose, once again, that discrimination is a pure taste.  What do anti-discrimination laws accomplish?

B.           If they correctly identify discrimination, then very little.  Markets already severely punish employers who pay more for workers than necessary.

1.            They might however exacerbate worker-on-worker discrimination by forbidding segregation.

C.           However, if "discrimination" laws blur the line between "difference" and "discrimination," effects can be severe.  The law then effectively requires employers to pay workers of different ability levels the same; employers respond by preferring the more productive group, making life even harder for the less productive group.

D.           In other words, discrimination laws act as a price control, requiring equal wages in two labor markets where the market clears at different wage levels.

E.           To some extent, though, discrimination laws might be seen as quantity restrictions (hire x workers of group y or else!).  The short-run effect of this on group y can be positive; but in the longer-run employers figure out ways to avoid this burden.

1.            E.g. Relocate the firm to states with small "protected" populations.

F.            For statistical discrimination, discrimination laws have the same negative effects.  Groups are really different on average, but the law says employers must treat them the same.  Firms then do their best to avoid paying people more than they're worth.

G.           Ex: How might unregulated markets induce cab-drivers to pick up late at night in dangerous areas?

H.           Similarly, able members of low-productivity groups might - in an unregulated market - agree to work for free on a temporary basis to prove themselves.  This would probably be illegal under current law.

XI.          Discrimination Laws In Practice

A.           Under the discrimination laws, aggrieved individuals can sue employers for discriminating against them.

B.           Employers can defend themselves by showing that the worker was judged on the basis of individual performance.

C.           Still, the defense always labors under the equivocation between difference and discrimination.

D.           Interestingly, most discrimination suits come from workers who say their current employer mistreated them, not from workers who say they were not hired in the first place. 

1.            The irony is that an employer who was actually racist, or simply wanted to avoid legal headaches, is probably less likely to be sued than someone who gives individuals a chance.

E.           If employers practice statistical discrimination, why would they want to fire a worker after hiring him?  Only if he is below his group mean!

F.            Discrimination laws have also severely curtailed the use of IQ tests, even though these are probably the best predictors of job performance available.

G.           Interestingly, early developers of IQ tests often saw them as a way to judge people on their merits as individuals.  But now they have fallen out of favor.

H.           Question: If you really wanted to stop discrimination, which would make more sense to ban: IQ tests or face-to-face interviews?

XII.        Why the Standard History of Discrimination Is Wrong

A.           The standard story: White males arbitrarily discriminated against everyone else out of pure malice.  Then activists "raised awareness" and discrimination laws were passed to open up opportunities for people other than white males.  While a strong legacy of racism and sexism persists, these laws have created the progress that disadvantaged groups have enjoyed since 1965.

B.           Why it's wrong:

1.            Even if average levels of malice were high, employers are among the least racist people around.  They are selected to care about profits, not skin color.

2.            White males have earned more money on average, but most or all of that difference disappears controlling for characteristics.

3.            Blacks and other groups were enjoying rapid economic progress long before any civil rights acts were passed.  Asians already equaled or exceeded white income - even Japanese-Americans, who lost most of their wealth during WWII internment.

4.            Lower-earning groups enjoyed progress before the civil rights laws in large part because their average characteristics were changing.  Blacks were acquiring more education and skills, immigrants were acquiring language fluency, women were changing their family plans, and so on.

5.            Most of the progress that non-white-males have enjoyed has been inevitable.  On net, civil rights laws may have impeded their progress by making employers reluctant to hire people who might potentially sue them.  There may have been some small effect; but as in other cases, there are probably negative long-run effects as well as positive short-run effects.