What Has Economics to Say About Racial Discrimination?

Kenneth J. Arrow

Racial discrimination pervades every aspect of a society in which it is found. It is

found above all in attitudes of both races, but also in social relations, in intermarriage, in
residential location, and, frequently, in legal barriers. It is also found in levels of
economic accomplishment; that is, income, wages, prices paid, and credit extended. This
economic dimension hardly appears in general treatments of economics, outside of the
specialized literature devoted to it. Nevertheless, it is important not only in itself but as a
test of standard theories.
There is no way of separating completely the study of racial discrimination (or
indeed many other aspects of economics) from moral feelings. There are many modern
varieties of liberalism, which draw the boundaries between social and individual action
in different places, but all agree in rejecting racial discrimination, by which is meant
allowing racial identification to have a place in an individual's life chances. It is, of
course, important to be analytic; moral feelings without analysis can easily lead to
unconstructive policies.

It is natural to suppose that economic analysis can cast light on the economic effects
of racial discrimination. But its pervasiveness must give us pause. Can a phenomenon
whose manifestations are everywhere in the social world really be understood, even in
only one aspect, by the tools of a single discipline? I want to explore here the scope and
limits of ordinary economic analysis for understanding racial discrimination even in
markets.

Some Empirical Constraints on Theory

We must start with a simple observation. Before any legal steps were taken to address
economic discrimination, it existed in perfectly open form, with no need for subtle
economic analysis. Darity and Mason, in their paper in this volume, do well to remind us
that help-wanted advertisements stated racial preferences plainly, without even the
fig-leaf of a code. I can speak as a witness here. It was simply well known that most good
jobs were not available to blacks. Not only employers but also labor unions, particularly
craft unions, were explicit on maintaining the color bar. During the U.S. participation in
World War II, the no-strike pledge by labor unions was well-honored, with one glaring
exception: when the Philadelphia rapid transit system, caught in the wartime labor
shortage, tried to hire blacks, the workers went on a successful strike to prevent the
attempt.

Residential discrimination was of course also overt, enforced primarily through

voluntary choice by sellers, but also by covenants attached to the land. About 1950, 1
looked into joining a cooperative housing development, the members of which were
primarily Stanford faculty, known liberals. I expressed my dismay on finding a clause
limiting non-white participation to 10 percent of the whole. I was assured that it was
considered a radical and courageous act to set the proportion above zero and that there
could be no mortgage financing if they went further. Intermarriage was rare in the north
and illegal in many states.

Cafeterias in at least some U.S. government departments would not serve blacks as
late as World War II. Strict segregation in the military during World War II, including
exclusion from combat, was the norm. In virtually all southern states, there were explicit
* Kenneth J. Arrow is Joan Kenney Professor of Economics Emeritus, Stanford University, Stanford,
California.
 -2-

"Jim Crow" laws requiring segregation in public facilities, transportation, and education,
and their existence reflected social attitudes. But the phenomena just discussed occurred
also in the parts of the economy not directly governed by legislation.

The presence of racial discrimination throughout American society was, to use the
words of Samuel Johnson, a fact "too evident for detection and too gross for
aggravation." To establish the existence of discrimination, estimating wage equations
would have been beside the point. Of course, society and scholars would want to know
the quantitative implications of discrimination for income as well as other indices of
well-being. But the fact of discrimination would not have needed testing.

One point of this reminder of the past is to remind us that any theory of racial
discrimination, including any theory of its economic implications, has to be consistent
with these patent facts. A second point is to raise the possibility that such a widespread
set of values is likely to change only slowly. This is not to deny "rat that value changes
have occurred and are occurring. It is to suggest that there is likely to be a large residue
of discriminatory values and the values that arise among those discriminated against even
after the slightly more than 30 years since the passage of the Civil Rights Act. A third
point is that the passage of legislation means that the gross evidence available before
1964 is no longer at present. We are forced to resort to indirect inferences, such as those
in the other papers in this symposium.

There was another more specific aspect of labor market discrimination that is
well-known to economic historians but seems to have played little role in macroeconomic
explanations.' As Higgs (1977) and Whatley and Wright (1994) have shown, black and
white wages for the same job very frequently differed but little. Discrimination mainly
took the form of limiting the range of jobs in which blacks were hired at all. The form
which racial discrimination took was the same as in residential segregation. It was not
that blacks were charged higher rents for the same residence but that they were excluded
from certain (most) areas.

Is there evidence of racial discrimination in the economy today? I have to take the
evidence given in the three accompanying papers as decisive. Especially striking are the
audit studies on differential treatment in the housing and automobile markets, reported on
by Yinger. While one can always invent hypotheses to explain away these results, there is
really no reason not to draw the obvious conclusions. There is also convincing evidence
of discrimination in the mortgage market. In addition, we have the strong evidence
presented in Massey and Denton (1993), not cited in the three papers published in this
issue, that residential segregation by race is extremely high. Any conceivable
explanation, whether by discrimination in the housing market or by voluntary choice
based on racial preferences for neighbors, is based on racial discrimination.

To summarize, we have clear evidence that blacks were in the past excluded from a
significant range of good jobs and from the purchase of housing and restaurant services.
We have very strong evidence that these practices persist in some important measure. I
am going to suggest in this paper that market-based explanations will tend to predict that
racial discrimination will be eliminated. Since they are not, we must seek elsewhere for
non-market factors influencing economic behavior. The concepts of direct social
interaction and networks seem to be good places to start.

Economic Theory and Racial Discrimination-Some Generalities

What light can standard economic analysis cast on answering this question and on
analyzing the causal factors? Can the broad facts, inadequately summarized above, fit
into a mold to which economic theory can apply?
 -3-

The answer depends in part as to what we mean by economic theory. Certainly,

“rational choice theory” is broader than “economic theory.” Rational choice theory means
that the individual actors act rationally (that is, by maximizing according to a complete
ordering) within the constraints imposed by preferences, technology, and beliefs, and by
the institutions which determine how individual actions interact to determine outcomes.
Further, the beliefs are themselves formed by some kind of rational process. By economic
theory, we mean that in some sense, markets are the central institution in which
individual actions interact and that other institutions are of negligible importance.

The theoretical picture of a market is one of impersonal exchange. I confine myself

to the competitive case. At a given price (or, more precisely, given all prices), individual
agents choose how much to supply and how much to demand. These supplies and
demands are simply added up; when the prices are such that total supply equals total
demand in each market, equilibrium prevails. There is no particular relation between a
supplier and a demander; that is, a supplier is indifferent about supplying one demander
or another, or vice versa. This is not a bad description of highly organized exchanges,
such as securities and futures markets, but hardly complete for even most commodity
markets, let alone the labor and credit markets. Suppliers and demanders have direct
personal relations, even or perhaps especially among sophisticated agents, as in
interindustry trade.' Certainly, employment of labor involves direct personal relations
between employee and employer (or the latter's agents) as well as among employees.
Similarly, credit relations other than those represented by marketable securities have
typically required direct personal interaction between debtor and financial institution.
Nevertheless, most of economic analysis, within the range in which it is applicable,
presupposes that the market idealization gives at least a reasonable approximation for the
purposes of predicting prices and total quantities. Let us ask whether a market-based
model can broadly satisfy the empirical constraints suggested in the preceding section.
On the usual interpretation, it cannot. If the members of the two races, after adjusting
for observable differences in human capital and the like, received different wages or were
charged different prices in commodity or credit markets, an arbitrage possibility would be
created which would be wiped out by competition. Most analysts, following Becker
(1957), add to the usual list of commodities some special disutility which whites attach to
contact with blacks, taste-based discrimination. Many variations are possible; dislike by
employers, dislike by white workers or by foremen, or whatever, as in Welch (1967) or
Arrow (1972a, 1972b, 1973).
The trouble with these explanations is that they contradict in a direct way the usual
view of employers as simple profit-maximizers. While they do not contradict rational
choice theory, they undermine it by introducing an additional variable. First, consider the
simple hypothesis of employer discrimination. If employers have one variable other than
profits in their maximands, why not others? Indeed, other such variables have been
hypothesized from time to time-for example, effort (Hicks, 1935; Scitovsky, 1943) ,
growth or size (Marris, 1964; Baumol, 1959) , or retention of control by withholding
information from employees (Marglin, 1974)and it would be easy to state many others
that may or may not have appeared in the literature.
1
What has been studied in the literature is the effect of segregated employment on the wages of the group
discriminated against. The first paper is that of Millicent Fawcett (1892!), with reference to gender
discrimination. Her work was cited and elaborated upon by Edgeworth (1922) and the idea rediscovered
and developed empirically in a noteworthy article of Bergmann (1971).
2
White (1995) has argued persuasively that product differentiation as an active strategy presupposes that
the relations among the economic agents are not anonymous.
 -4-

There are at least two objections to this line of analysis. One is that introducing new
variables easily risks turning the "explanation" into a tautology. Molière had one of his
characters explain sleep as the result of an accumulation of the "dormitive principle";
opium's effects were due to the fact that opium possessed a large quantity of the
dormitive principle, and so forth. This was clearly intended as a parody of scientific
discourse, and it certainly would be a parody of economics to multiply entities in this
anti-Occamian fashion. Perhaps more serious is the neglect of Darwinian principles.
Presumably the population of employers is not uniform in its discriminatory tastes.
Then, under the usual assumption of constant (or increasing) returns to scale,
competition would imply the elimination of all but the least discriminatory employers.
If there are any non-discriminatory employers, they would drive out the others.

A further objection to the hypothesis that racial wage differentials arise from
employer discrimination is that large corporations hire a major fraction of the labor
force. Attributing taste to impersonal entities is a hypothesis of dubious usefulness. It is
hardly in the stockholders' interests to discriminate under the postulated condition, and
competition in the capital market should be effective in eliminating discrimination.
Finally, the hypothesis of employer discrimination does not at all explain segregation
by occupation.

An alternative hypothesis is that labor market discrimination is due to discrim-

inatory tastes of other employees. In the case of large corporations, for example, it
would be those of the executives, although other scenarios have been advanced. But
then it is easy to see that in simple cases, the natural equilibrium would be segregation
within an industry-that is, firms with either all black or all white labor forces. (Some of
these arguments appeared in my papers cited above.) The matter is a little more
complicated when there are complementary labor inputs, like foremen and floor
workers or skilled and unskilled labor. If for some historical reason, the foremen, for
example, are all white and require compensation for working with blacks, then racial
differences in wages will appear. But if foremen differ in their discriminatory tastes,
then the less discriminatory will receive higher wages for working in plants with higher
proportions of black floor workers. This implication has not been tested, but certainly
seems dubious. Again, in any case, the model of worker-based discriminatory tastes
may explain segregation within industries but not segregation by occupation.

Finally, what can market-based theories make of discrimination against black

consumers? Sellers of houses and mortgages have refused to sell to black customers and
still refuse to some extent. It is hard to think of any market-based explanation for refusal
to sell. Sellers of automobiles sell to black customers only at higher prices. Why does not
competition prevent this discrimination, according to well-known arguments?

Statistical Discrimination
Modern economic theory for the last 30 years has emphasized how information, more
properly, beliefs and expectations influence economic behavior. These beliefs may in
turn be based on some kind of evidence; the rational choice theory implies that beliefs
contradicted by experience will not survive. In the present context, this has given rise to
the theory of statistical discrimination. Suppose blacks and whites do in fact differ in
productivity, at least on the average. This is in turn due to some cause, perhaps quality of
education, perhaps cultural differences; but the cause is not itself observable. Then the
experience of employers over time will cause them to use the observable characteristic,
race, as a surrogate for the unobservable characteristics, which in fact cause the
productivity differences (Phelps, 1972; Arrow, 1972a, b, 1973; for a more recent version,
see Lundberg and Startz, 1997). This is a market-based explanation, which does not
require tastes for discrimination.
 -5-

If there are a number of observable variables, such as quantity of education, then the
hypothesis of statistical discrimination implies that an estimate of wages based on those
observables will be significantly improved by adding race as a predictor. But this is the
same conclusion as arrived at by hypothesis of market-based discrimination based on
taste.

Can one distinguish between statistical and taste-based discrimination? Clearly, to do

so in the case of the labor market depends on the ability to observe a measure of the
individual's marginal productivity. Unfortunately, such data do not in general exist.
Parallel evidence may be better found in the mortgage market. The evidence, as
summarized by Ladd in this symposium, is not very extensive, but it suggests that, given
the observed variables, blacks do default somewhat more. If discrimination were
taste-based, we would expect the opposite.

Of course, it is not very satisfactory to postulate that the unobserved determinants of

performance just happen to be correlated with race. The hypothesis that statistical
perceptions change behavior as well as reflect it is alluded to by Darity and Mason
toward the end of their paper and was earlier formulated by Arrow (1972a; b; 1973) and
Coate and Loury (1993) . To prepare to work requires investment by the worker. Not all
of this investment is observable; it may require changes of habits and attitudes towards
work, and diligence in school and home tasks, to give some examples.3 If the employer is
going to judge by race, then there is no reward for these investments. They will not be
acquired, and then the statistical judgments will be confirmed.

The discussion of statistical discrimination so far assumes that the employers or

creditors use all the information available throughout the economy. In Bayesian terms,
the posterior information is sufficiently rich to make the contribution of the prior
minimal. But of course this is not so. Each employer has a very limited range of
experience, and so prior beliefs can remain relatively undisturbed. Indeed, to the extent
that discrimination takes the form of segregation, then there will in fact be little
experimentation to find out abilities. As Whatley and Wright (1994) point out, the very
fact of segregation will reinforce beliefs in racial differences.

Social Interactions and Networks

Enough has been said to suggest that market-based theories give an inadequate
account of the effects of racial discrimination on economic magnitudes and the effects of
the economic system on racial discrimination. It is increasingly recognized that many
social interactions with economic implications are not mediated through a
depersonalized market, but rather through the cumulative effect of individual choices.
An early example is Schelling's (1971) analysis of residential segregation. He started
with preferences towards the races of neighbors but pointed out that even mild
discriminatory attitudes, if widespread, might lead to a very segregated equilibrium.
Implicitly, he assumed that it was not possible to have discriminatory prices in a given
location, for example, lower rents for whites in predominantly black neighborhoods.

The hypothesis that prices do not reflect every kind of social interaction, even those
of economic importance, is used in many contexts. Every now and then, economists
studying the labor market have found it important to postulate some kind of rigidity of
relative
3 In evidence surveyed a long time ago by Bowles and Gintis (1976, ch. 13), it was observed that job
performance was better forecast by habits and behavior of students than by their grades or achievement test
scores.
 -6-

wages. For example, Dunlop's (1957) study showed that the wages of the same
occupation, truck drivers, varied with the general wage levels of the different industries,
which employed them. Similarly, a frequently-maintained hypothesis about
unemployment is that there is some fair level of wages, which must be maintained (for
example, Hahn and Solow, 1995, ch. 5).
I intend these points as an illustration of a more general principle-that beliefs and
preferences may themselves be the product of social interactions unmediated by prices
and markets. This concept has been the object of significant theoretical research recently
(for example, Blume, 1997; Durlauf, 1997a, b) and empirical application to the frequency
of criminal activity (Glaeser, Sacerdote, and Scheinkman, 1996).
Another variation of the theme that social linkages alter resource allocation processes
is found in the concept of social capital that was introduced by Loury (1977), developed
by Coleman (1990, ch. 12) and used empirically, among others, by Putnam (1993) and
Borjas (1992). These scholars have hypothesized that a dense network of social
connections, even though developed for noneconomic purposes, will enhance both
political and economic efficiency. Admittedly, the concept of social capital is very hard
to pin down in an explicit model, but enough has been done to show its importance.
I want to conclude by concentrating on one particular type of social structure, which
has already shown its applicability to the labor market, the network of acquaintances and
friends. Sociologists and some economists who have worked in this area have shown by
careful empirical work that a very large fraction of the jobs are filled by referrals by
current employees. There are many such studies; for especially careful and definitive
ones, see Rees and Shultz (1970, ch. 13) and Granovetter 1974. The network concept of
labor allocation differs considerably from a market. It is indeed very easy to say how
social segregation can give rise to labor market segregation through network referrals.
Discrimination no longer has any cost to the discriminator; indeed, it has social rewards.
Profit maximization is overcome by the values inherent in the maintenance of the
network or other social interaction. The methodological demands, which are satisfied by
a network approach have been outlined by Granovetter (1988) and White (1995). More
definite modeling of networks in the labor market still needs to be done. Clearly, the
anonymous market, in which in effect every seller is connected with every buyer, is one
extreme of a network. Intuitively, it is clear that a sufficiently dense network will mimic a
market (Kranton and Minehart, 1997). But the empirical accounts of employment suggest
instead a network with relatively few links compared with all those possible.
The main point is that personal interactions occur throughout this process, and
therefore there is plenty of room for discriminatory beliefs and preferences to play a role,
which would be much less likely in a market subject to competitive pressures. The
network model seems most appropriate for the labor market, and perhaps less so for the
housing, automobile, and credit markets. But in all of these, each transaction is a social
event. The transactors bring to it a whole set of social attitudes which would be irrelevant
in the market model..Models of racial discrimination in which all racial attitudes are
expressed. Models of racial discrimination in which all racial attitudes are expressed
through the market will get at only part of the story. At each stage, direct social
transactions unmediated by a market play a role. Even the market manifestations will be
altered by these direct social influences.
 -7-

References
Arrow, Kenneth J., "Models of Job Discrimination." In A.H. Pascal, ed. Racial
Discrimination in Economic Life. Lexington, Mass.: D.C. Heath, 1972a, 83-102.
Arrow, Kenneth J., "Some Mathematical Models of Race Discrimination in the Labor
Market." In A.H. Pascal, ed. Racial Discrimination in Economic Life. Lexington,
Mass.: D.C. Heath, 1972b, 187-204.
Arrow, Kenneth J., "The Theory of Discrimination." In Orley Ashenfelter and Albert
Rees, eds. Discrimination in Labor Markets. Princeton, N.J.: Princeton University
Press, 1973. 3-33.
Baumol, William, Business Behavior, Value, and Growth. New York: Macmillan, 1959.
Becker, Gary, The Economics of Discrimination. Chicago: University of Chicago Press,
1957.
Bergmann, Barbara, "The Effects on White Income of Discrimination in
Employment,"Journal of Political Economy, 1971, 792, 294-313.
Blume, Lawrence, "Population Games." In W.B. Arthur, S. Durlauf, and D. Lane, eds.
The Economy as an Evolving Complex System. Menlo Park, Calif.: Addison Wesley
Longmans, 1997.
Borjas, George, "Ethnic Capital and Intergenerational Income Mobility," Quarterly
Journal of Economics, 1992, 107, 123-50.
Bowles, Samuel, and Herbert Gintis, Schooling in Capitalist America. New York: Basic
Books, 1976.
Coate, Stephen, and Glenn Loury, "Will Affirmative Action Policies Eliminate Negative
Stereotypes?" American Economic Review, 1993, 83, 1220-42.
Coleman, James S., Foundations of Social Theory. Cambridge, MA: Harvard University
Press, 1990.
Dunlop, John T., "The Task of Contemporary Wage Theory." In John T. Dunlop, ed. The
Theory of Wage Determination. London: Macmillan, and New York: St. Martin's,
1957, 3-27.
Durlauf, Stephen, Statistical Mechanics Approaches to Socioeconomic Behavior." In
W.B Arthur, S. Durlauf, and D. Lane, eds. The Economy to an Evolving Complex
System. Menlo Park, Calif.: Addison Wesley Longmans, 1997a.
Durlauf, Stephen, "The Memberships Theory of Inequality: Ideas and Implications,"
Working Paper, Department of Economics, University of Wisconsin, 1997b.
Edgeworth, Francis Y., "Equal Pay to Men and Women for Equal Work," Economic
Journal,1922, 32, 431-57.
Fawcett, Millicent, "Mr. Sidney Webb's Article on Women's Wages," Economic Journal,
1892, 2, 173-76.
Glaeser, E., B. Sacerdote, and J. Scheinkman, "Crime and Social Interactions," Quarterly
Journal of Economics, 1996, 11:2, 507-48.
Granovetter, Mark, Getting a job: A Study of Contacts and Careers. Cambridge, Mass.:
Harvard University Press, 1974.
Granovetter, Mark, "Economic Action and Social Structure." In Barry Wellman and S.D.
Berkowitz, eds. Social Structures: A Network Approach. Cambridge, U.K., and New
York: Cambridge University Press, 1988.
Hahn, Frank, and Robert Solow, A Critical Essay on Modern Macroeconomic Theory.
Cambridge, Mass., and London: The MIT Press, 1995.
Hicks, John R., "Annual Survey of Economic Theory: the Theory of Monopoly,"
Econometrica, 1935, 3, 1-20.
Higgs, Robert, Competition and Coercion: Blacks in the American Economy 1865-1914.
New York: Columbia University Press, 1977.
Kranton, Rachel E., and Deborah F. Minehart, "Link Patterns in Buyer-Seller Networks:
Incentives and Allocations in Graphs," Working Paper, University of Maryland and
Boston University, June 1997.
 -8-

Loury, Glen, "A Dynamic Theory of Racial Income Differences." In P. A. Wallace and
A. Le Mund, eds. Women, Minorities, and Employment Discrimination. Lexington,
MA.: Lexington Books, 1977.
Lundberg, Shelley, and Richard Startz, "Race, Information, and Segregation." Working
Paper, Department of Economics, University of Washington, 1997.
Marglin, Stephen, "What Do Bosses Do? The Origins and Functions of Hierarchy in
Capitalist Production," Review of Radical Political Economy, 1974, 6:2, 60-112.
Marris, Robin, The Economic Theory of 'Managerial' Capitalism. New York: The Free
Press of Glencoe, 1964.
Massey, D., and N. Denton, American Apartheid. Cambridge, Mass.: Harvard University
Press, 1993.
Phelps, Edmund S., "The Statistical Theory of Racism and Sexism," American Economic
Review, 1972, 62, 659-61.
Putnam, Robert D., Making Democracy Work. Princeton, NJ.: Princeton University Press,
1993.
Rees, Albert, and George P. Shultz, Workers and Wages in an Urban Labor Market.
Chicago: University of Chicago Press, 1970.
Schelling, Thomas, "Dynamic Models of Segregation," Journal of Mathematical
Sociology, 1971, 1, 143-86.
Scitovsky, Tibor, "A Note on Profit Maximization and its Implications," Review of
Economic Studies, 1943, 11:1, 57-60.
Welch, Fmis, "Labor Market Discrimination: An Interpretation of Income Differences in
the Rural South," Journal of Political Economy,1967, 75,225-40.
Whatley, Warren, and Gavin Wright, "Race, Human Capital, and Labour Markets in
American History." In George Grantham and Mary MacKinnon, eds. Labour Market
Evolution. London and New York: Routledge, 1994.
White, Harrison C., "Social Networks Can Resolve Actor Paradoxes in Economics and in
Psychology," Journal of Institutional and Theoretical Economics, 1995, 151:1,
58-74.