. 10
( 13)


Worries White Male
-0.4 about
Bills (-0.42*)
Oppor- Household
tunity Income >
in USA $150,000

Figure 10.2
Determinants of the support for redistribution.
Note: Bars represent ordinary least squares coef¬cients (value of the estimated coef¬cient
is in parentheses) predicting support for redistribution. The dependent variable is stand-
ardized so that the estimated coef¬cient represents the effect of the variable indicated on
concern about poverty measured in standard deviation units. The equation also includes:
seven additional income dummies, age, a dummy for attended college, and dummies for
˜˜worries about bills most of the time,™™ and ˜˜worries about bills some of the time.™™ The
omitted category for household income is less than $10,000 per year. The omitted catego-
ries for causes of poverty and wealth are ˜˜lack of effort™™ and ˜˜strong effort™™ respectively.
To simplify the presentation of race effects, we use the sample of white and black
respondents only. Omitted category for ˜˜worries about bills™™ is ˜˜all of the time.™™ There
are 3417 observations. R 2 ¼ :260. This regression uses sample weights, although the
results are not sensitive to them. We use robust standard errors. All coef¬cients are sig-
ni¬cant at the 1 percent level.
288 Fong, Bowles, and Gintis

(White) Educa-
0.10 (0.089)*** Connec-
(0.084)*** tion Parents
Matters Matter Dishon-
Race Inheri-
Matter Luck
(0.059)** (0.057)** esty
(0.055)*** Matters Matters tance Gender
(0.042)*** (Black) Matters Matters
(0.033) (0.035)** (Female)

Regression Coefficient


-0.10 Matters

Figure 10.3
Effects on the support for redistribution of beliefs about the importance of various factors
in getting ahead in life.
Notes: Bars represent ordinary least squares coef¬cients (value of the estimated coef¬-
cient is in parentheses) predicting support for redistribution. The dependent variable is
standardized. Independent variables are the respondent™s belief in the importance of the
factor shown to getting ahead in life. The coef¬cients are the estimated effects of a one-
point increase in the response scale for a given belief on standard deviations of support
for redistribution. Regressions also include all of the self-interest measures included in
¬gure 10.2. The number of observations was 3,437. This regression uses sample weights,
although the results were not sensitive to them. R 2 ¼ :184. *** Signi¬cant at the 1 percent
level. ** Signi¬cant at the 5 percent level.

education, age, and the frequency with which respondents worry
about meeting family expenses in the regressions.6
In ¬gure 10.2, we present results from an ordinary least squares re-
gression that predicts support for redistribution using two sets of vari-
ables: (1) beliefs about the causes of wealth and poverty, and (2) the
measures of self-interest. To facilitate interpreting the coef¬cients, we
have standardized the dependent variable to have a mean of zero and
a standard deviation of one. The interpretation is as follows: those who
say that bad luck alone causes poverty are a 0.50 standard deviation
higher in their support for redistribution than those who think lack
Reciprocity and the Welfare State 289

of effort alone causes poverty. Those who think that good luck alone
causes wealth are a 0.39 standard deviation higher on the support for
redistribution scale than those who think effort alone causes wealth,
and people who respond that there is plenty of opportunity in the
United States to get ahead scored a 0.42 standard deviation lower in
support for redistribution than people who do not think there is plenty
of opportunity.
Measures of self-interest also have signi¬cant effects in the expected
direction on support for redistribution. Those who are in the highest
income category (annual household income greater than $150,000)
scored a 0.47 standard deviation lower on support for redistribution
than those in the lowest income category (income less than $10,000).
Those who almost never worry about bills are signi¬cantly less sup-
portive of redistribution than those who worry all of the time. The
self-interest variables are jointly signi¬cant at the 1 percent level.
The effect of being white is large and highly signi¬cant, and the
effect of being male is even larger. At ¬rst glance, this may appear to
contradict an empirical regularity that among the socioeconomic vari-
ables, race has one of the largest and most reliable effects while sex
does not. However, if we omit the beliefs variables, the magnitude of
the effects of race and sex increase and become roughly equivalent in
size. This is consistent with the argument, put forth by Gilens (1999),
that the effect of race is mediated by beliefs about the characteristics of
the poor, especially poor blacks.
If we take the view that all of the socioeconomic variables together
capture self-interest, then the effect of self-interest appears consider-
ably larger than if we simply consider the size of the coef¬cient on in-
come. Using ordered probit to estimate similar equations, Fong (2001)
has estimated the sizes of the effects of the independent variables on
the probabilities of scoring in each of the six categories of the support
for redistribution scale. In an equation that controls for both beliefs
about the causes of wealth and poverty and a large number of objec-
tive and subjective measures of and proxies for self-interest, the effects
of being in the least privileged category (non-white, female, single,
union member, part-time worker, no college education, in lowest in-
come category, household size greater than four, and almost always
worries about bills) as opposed to the most privileged are similar
in size to the effects of believing that luck alone causes wealth and
poverty, as opposed to believing that effort alone causes wealth and
290 Fong, Bowles, and Gintis

Could our results be driven by missing self-interest variables? Peo-
ple who believe that poverty is caused by bad luck or circumstances
beyond individual control may be those who have low-mean, high-
variance incomes. Such individuals may have higher expectations of
needing government assistance in the future, and therefore demand
more redistribution purely out of self-interest. For similar reasons,
those who believe that the poor are lazy may simply be people who
have higher-mean, lower-variance incomes and therefore less self-
interest in redistribution. If this is true, then the effect of these beliefs
on redistributive policy preferences may have nothing to do with the
psychology of holding the poor accountable and blaming them for
their outcomes. It would simply be the case that beliefs about the
causes of income are correlated with a person™s ¬nancial position,
which in turn determines his or her demand for redistribution.
If the beliefs about the causes of poverty and wealth operate through
self-interest, then they should have no effect among people at the top
and bottom of the distribution of income who expect to remain there.
Those who do not expect to bene¬t should demand no redistribution
at all, regardless of their beliefs about the causes of income, while those
who expect to bene¬t should register the highest degree of support for
redistribution regardless of their beliefs about the causes of income. To
test whether this is the case, we use sub-samples of (1) individuals with
incomes over $75,000 per year who expect to be better off in ¬ve years
than they are today and who worry about bills less often than ˜˜all of
the time™™; (2) individuals with incomes under $10,000 per year; and (3)
individuals with incomes under $30,000 per year who do not expect to
be better off in ¬ve years than they are today and who worry about
bills more often than ˜˜almost never.™™
In all of these sub-samples, a quite inclusive set of measures cap-
turing self-interest is jointly insigni¬cant. That is, we cannot reject the
hypothesis that every single socioeconomic variable has a coef¬cient
of zero. Yet, beliefs about roles of luck, effort, and opportunity in gen-
erating life outcomes were jointly signi¬cant for all three sub-samples
and in most cases were individually signi¬cant in the expected direc-
tions as well.7 Therefore, among those individuals who are poor
and do not expect their lives to improve and those individuals who
believe that lack of effort causes poverty oppose redistribution. Anal-
ogously, support for redistribution is high among those securely
well off respondents who believe that poverty is the result of bad
Reciprocity and the Welfare State 291

In another test of self-interest, we use questions on the respondents™
views on the importance of various factors, including a person™s race
and sex, to getting ahead in life. Figure 10.3 presents an ordinary least
squares regression of support for redistribution on the importance of
various determinants of success, controlling for the same socioeco-
nomic variables included in the regression presented in ¬gure 10.2.
Beliefs that ˜˜willingness to take risks™™ and ˜˜hard work and initiative™™
explain ˜˜why some people get ahead and succeed in life and others do
not™™ have highly signi¬cant negative effects on support for redistribu-
tion. Beliefs that education, people™s parents, connections, good luck,
dishonesty, and inherited money explain why some people get ahead
have signi¬cant positive effects on support for redistribution. In addi-
tion, beliefs that a person™s sex is important to getting ahead have sig-
ni¬cant positive effects on support for redistribution for men, while
the effect of this belief for women is also positive but smaller and sta-
tistically insigni¬cant. Beliefs that a person™s race is important to get-
ting ahead in life have signi¬cant positive effects for whites, while the
effect of these beliefs for blacks is positive but smaller and statistically
If people think that a person™s race and sex are important to get-
ting ahead in life, then the effects of these beliefs on self-interested
demand for redistribution should operate in opposite directions for
those who expect to bene¬t and those who expect to lose from racial
or gender discrimination.8 In other words, whites who think race is im-
portant to getting ahead will expect to be economically advantaged
and would have fewer self-interested reasons to support redistribution
than whites who think that race does not matter. Similar reasoning
holds for men who think a person™s sex is important to getting ahead
in life.
However, using an alternative form of the same regression pre-
sented in ¬gure 10.3, we ¬nd that the effect of believing that a person™s
sex is important to getting ahead in life is signi¬cantly more positive
for men than it is for women. This interaction effect is signi¬cant at the
1 percent level (unreported). As we have seen, this is inconsistent with
self-interest, because men and whites with these beliefs would expect
to bene¬t from discrimination and hence have less likelihood of bene-
¬ting from redistributive programs.
Concerns about the incentive effects of taxation are a ¬nal mecha-
nism through which self-interest might cause beliefs that the poor are
lazy and the rich industrious to decrease the demand for redistribution.
292 Fong, Bowles, and Gintis

When earned income is more sensitive to work effort, taxation may
cause greater effort disincentives and reduce aggregate income. If so,
then beliefs about the roles of effort, luck, and opportunity in generat-
ing income may affect the level of support for redistribution through
concerns about incentive costs of redistribution (Piketty 1995). This
type of incentive concern should not apply only to redistribution, but
to any tax-funded expenditure, including expenditures such as na-
tional defense.
According to this tax-cost hypothesis, if beliefs that income is caused
by factors under individual control decrease demand for redistribu-
tion, then they should decrease demand for other kinds of tax-funded
expenditures (including defense spending) as well. But there is no evi-
dence that tax cost concerns adversely affect the demand for public
expenditures. Using the 1990 General Social Survey, we estimate or-
dered probit regressions predicting support for spending on welfare,
national defense, halting the rising crime rate, and dealing with drug
addiction, respectively.9 The independent variables are beliefs that the
poor are poor because of lack of effort, and ¬ve demographic variables
(income, education, race, sex, and age). In the samples reported above,
the belief that the lack of effort causes poverty has a highly signi¬cant
negative effect on support for redistribution. However, these same
beliefs have no effect on support for spending on crime or drug addic-
tion, and they have a signi¬cant positive effect on support for spend-
ing on defense. If these beliefs simply measure tax cost concerns, then
their effect on support for all of these expenditure items should have
been negative.
However, even more convincing evidence on this point comes from
the experiment including actual welfare recipients described earlier in
this section. There were no disincentive costs in this experiment, yet
student subjects gave more to the welfare recipients with the stronger
work commitments. These results lend support to previously made
hypotheses about well-known patterns in survey data. Heclo (1986)
reports that 81 percent of survey respondents favor public funding for
child care if the mother is a widow who is trying to support three chil-
dren, while only 15 percent favor such funding when the mother has
never married and is not interested in working. Heclo also reports the
results of a survey in which the wording of a question about support
for public redistribution was manipulated so that some subjects were
asked about spending on ˜˜welfare™™ while others were asked about
spending on ˜˜assistance for the poor,™™ or ˜˜caring for the poor.™™ In that
Reciprocity and the Welfare State 293

experiment, 41 percent of respondents stated that there is too much
spending on welfare and 25 percent stated that there is too little. By
contrast, only 11 percent and 7 percent of the respondents said that
there is too much spending on assistance for and caring for the poor,
respectively, and 64 percent and 69 percent said that there is too little
spending on assistance for and caring for the poor, respectively. In a
similar vein, Page and Shapiro (1992) report that support for social se-
curity spending has been very high and stable over time, while sup-
port for spending on welfare has been consistently low. The
interpretation commonly given for ¬ndings such as these is that people
are less generous to recipients who they think are not working when
they could and should be, or who are otherwise considered to be in
questionable moral standing (Heclo 1986; Gilens 1999). We have
shown that these ¬ndings cannot be explained away by a fuller and
more rigorous account of self-interest.

10.5 Strong Reciprocity and the Welfare State: Unhappy Marriage?

The following generalizations sum up the relevance of the experimen-
tal, survey, and other data to the problem of designing and sustaining
programs to promote economic security and eliminate poverty. First,
people exhibit signi¬cant levels of generosity, even towards strangers.
Second, beliefs about the causes of high and low incomes matter.
Third, people contribute to public goods and cooperate to collective
endeavors and consider it unfair to free-ride on the contributions and
efforts of others. Fourth, people punish free riders at substantial costs
to themselves, even when they cannot reasonably expect future per-
sonal gain from these actions of punishment (chapters 1, 5“8, this
It would not be dif¬cult to design a system of income security and
economic opportunity that would tap rather than offend the motiva-
tions expressed in these four generalizations. Such a system would
be generous towards the poor, rewarding those individuals who per-
form socially valued work and who seek to improve their chances of
engaging in such work, as well as to those individuals who are poor
through accidents not of their own making, such as illness and job
While strong reciprocity may support egalitarianism, it may also
help explain opposition to welfare state policies in some of the ad-
vanced market economies during the past few decades. Speci¬cally,
294 Fong, Bowles, and Gintis

in light of the empirical regularities outlined above, we suspect the
following to be true as well: Egalitarian policies that reward people in-
dependent of whether and how much they contribute to society are
considered unfair and are not supported, even if the intended recipi-
ents are otherwise worthy of support and even if the incidence of non-
contribution in the target population is rather low. This would explain
the opposition to many welfare measures for the poor, particularly
since such measures are thought to have promoted various social
pathologies. At the same time it explains the continuing support for
Social Security and Medicare in the United States, since the public per-
ception is that the recipients are ˜˜deserving™™ and the policies are
thought not to support what are considered antisocial behaviors. The
public goods experiments reported in chapter 5 of this volume are also
consistent with the notion that tax resistance by the nonwealthy may
stem from their perception that the well-to-do are not paying their fair
A striking fact about the decline in the support for the former Aid to
Families with Dependent Children, food stamps, and other means-
tested social support programs in the United States, however, is that
overwhelming majorities oppose the status quo, whatever their income,
race, or personal history with such programs. This pattern of public
sentiment, we think, can be accounted for in terms of the principle of
strong reciprocity.
We rely mainly on two studies. The ¬rst study, Farkas and Robinson
(1996), analyze data collected in late 1995 by Public Agenda, a non-
pro¬t, nonpartisan research organization. The authors conducted eight
focus groups around the country, then did a national survey involving
half-hour interviews of 1,000 randomly selected Americans plus a na-
tional oversample of 200 African-Americans. The second study, politi-
cal scientist Martin Gilens™ Why Americans Hate Welfare, is an analysis
and review of several polls executed during the 1990s and earlier by
various news organizations.10
In the Public Agenda survey, 63 percent of respondents thought the
welfare system should be eliminated or ˜˜fundamentally overhauled™™
while another 34 percent thought it should be ˜˜adjusted somewhat.™™
Only 3 percent approved of the system as is (Farkas and Robinson
1996, 9). Even among respondents from households receiving welfare,
only 9 percent expressed basic approval of the system, while 42 per-
cent wanted a fundamental overhaul and an additional 46 percent
wanted some adjustments.
Reciprocity and the Welfare State 295

The cost of welfare programs cannot explain this opposition. While
people generally overstate the share of the Federal budget devoted to
welfare (Farkas and Robinson 1996, 9), this cannot account for the
observed opposition.11 Farkas and Robinson note that

By more than four to one (65 percent to 14 percent), Americans say the most
upsetting thing about welfare is that ˜˜it encourages people to adopt the wrong
lifestyle and values,™™ not that ˜˜it costs too much tax money.™™ . . . Of nine possi-
ble reforms presented to respondents”ranging from requiring job training to
paying surprise visits to make sure recipients deserve bene¬ts”reducing bene-
¬ts ranked last in popularity.

The cost, apparently, is not the problem. In focus groups:
Participants invariably dismissed arguments about the limited ¬nancial costs
of welfare in almost derisive terms as irrelevant and beside the point. (Farkas
and Robinson 1996, 9“10)

Nor can the perception of fraud account for this opposition. It is true
that 64 percent of respondents (and 66 percent of respondents on wel-
fare) believe welfare fraud is a serious problem. However most do not
consider it more serious than in other government programs, and only
35 percent of survey respondents would be more ˜˜comfortable with
welfare™™ if fraud were eliminated (Farkas and Robinson 1996, 11“12).
In commenting on this fact Martin Gilens (1999, 1, 2) observes that
˜˜Politics is often viewed, by elites at least, as a process centered on the
question ˜who gets what.™ For ordinary Americans, however, politics
is more often about ˜who deserves what™ and the welfare state is no
exception.™™ In the Public Agenda study, respondents overwhelmingly
consider welfare to be unfair to working people and addictive to
recipients. By a more than ¬ve-to-one margin (69 percent to 13 percent
overall, and 64 percent to 11 percent for people receiving welfare),
respondents say that recipients abuse the system (for instance, by not
looking for work) rather than actually cheating the system (for ex-
ample, by collecting multiple bene¬ts) (Farkas and Robinson 1999,
12). Moreover, 68 percent of respondents and (59 percent of welfare-
receiving respondents) think that welfare is ˜˜passed on from genera-
tion to generation, creating a permanent underclass.™™ In the same
vein, 70 percent of respondents (and 71 percent of welfare-receiving
respondents) say welfare makes it ˜˜¬nancially better for people to stay
on welfare than to get a job,™™ 57 percent (62 percent of welfare re-
cipients) think welfare encourages ˜˜people to be lazy™™ and 60 percent
(64 percent of welfare recipients) say the welfare system ˜˜encourages
296 Fong, Bowles, and Gintis

people to have kids out of wedlock.™™ (Farkas and Robinson 1999, 14“
15) Note that the welfare recipients and other citizens hold similar
views in this respect.
That the respondents are correct in thinking that the welfare state
cause these behaviors is beside the point. Whether or not welfare causes
(for example) out of wedlock births or fosters an unwillingness to
work, citizens object that the system provides ¬nancial support for
those who undertake these socially disapproved behaviors. Their
desire is to bear witness against the behavior and to disassociate them-
selves from it, whether or not their actions can change it.
Racial stereotyping and opposition to welfare are closely associated.
The public agenda survey shows that whites are much more likely
than African-Americans to attribute negative attributes to welfare
recipients and much more likely to blame an individual™s poverty on
lack of effort. Gilens (1999) writes that the survey data show,

For most white Americans, race-based opposition to welfare is not fed by ill-
will toward blacks, nor is it based on whites™ desire to maintain their economic
advantages over African Americans. Instead race-based opposition to welfare
stems from the speci¬c perception that, as a group, African Americans are not
committed to the work ethic.

There is some evidence that people are more tolerant of redistribu-
tions to their own ethnic and racial categories than they are of redistri-
bution to other ethnic or racial categories. Erzo Luttmer (2001) found
for a U.S. sample that individuals are more opposed to welfare if they
live in neighborhoods where a higher percentage of welfare recipients
are of a different race. Luttmer™s ¬ndings are consistent with our reci-
procity interpretation of redistributive politics, in light of the evidence
that when people identify with a social group, they are more likely to
blame outgroup members (holding these members individually re-
sponsible) for their bad outcomes and behaviors and to give outgroup
members little credit (holding factors other than the outgroup mem-
ber™s voluntary control responsible) for their good outcomes and
behaviors (Brewer and Miller 1996). However, the salience of race in
Luttmer™s U.S. data may not be as pronounced in other cultural con-
texts, since the characteristics that determine who are ˜˜insiders™™ and
who are ˜˜outsiders™™ is culturally speci¬c. Taking account of the fact
that many Americans see the current welfare system as a violation of
deeply held reciprocity norms does not require that policy makers
adopt punitive measures and stingy budgets for the poor. Indeed, the
Reciprocity and the Welfare State 297

public strongly supports income support measures when asked in
ways that make the deserving nature of the poor clear: a 1995 New
York Times/CBS poll, for instance, found that twice as many agreed
as disagreed that ˜˜it is the responsibility of the government to take
care of people who can™t take care of themselves.™™

10.6 Conclusion

Like Petr Kropotkin (1989[1903]) a century ago, we ¬nd compelling
evidence for the force of human behavioral predispositions to act both
generously and reciprocally rather than self-interestedly in many social
situations. While many economists have failed to appreciate the practi-
cal importance of these predispositions in policy matters, their salience
was not missed by Frederick Hayek (1978, 18, 20):
[The] demand for a just distribution . . . is . . . an atavism, based on primordial
emotions. And it is these widely prevalent feelings to which prophets (and)
moral philosophers . . . appeal by their plans for the deliberate creation of a
new type of society.

If we are right, economists have misunderstood both the support for
the welfare state and the revolt against welfare (where it has occurred),
attributing the latter to sel¬shness by the electorate rather than the fail-
ure of many programs to tap powerful commitments to fairness and
generosity and the fact that some programs appear to violate deeply
held reciprocity norms. Egalitarians have been successful in appealing
to the more elevated human motives precisely when they have shown
that dominant institutions violate norms of reciprocity and may be
replaced by institutions more consistent with these norms.
To mobilize rather than offend reciprocal values, public policies
should recognize that there is substantial support for generosity to-
wards the less well off as long as they have tried to make an effort to
improve their situation and are in good moral standing. The task of
politically viable egalitarian policy design might thus begin by iden-
tifying those behaviors that entitle an individual to reciprocation.
Among these behaviors in the United States today would be saving
when one™s income allows, working hard, and taking risks in both
productive endeavors and schooling. Persistent poverty is often the
result of low returns to socially admired behaviors: low wages for
hard work, a low rate of return on savings, costly access to credit for
those wishing to engage in uncertain entrepreneurial activities, and
298 Fong, Bowles, and Gintis

educational environments so adverse as to frustrate even the most dili-
gent student. Policies designed to raise the returns of these activities
when undertaken by the less well off would garner widespread sup-
port. A second principle of reciprocity-based policy design should be
to insure individuals against the vagaries of bad luck without insuring
them against the consequences of their own actions, particularly when
these actions violate widely held social norms against such things as il-
licit drug use or child bearing in the absence of reasonable guarantees
of adequate parenting.
Many traditional egalitarian projects, such as land reform and em-
ployee ownership, are strongly consistent with reciprocity norms, since
they make people the owners not only of the fruits of their labors, but
more broadly of the consequences of their actions (Bowles and Gintis
[1998] and Bowles and Gintis [1999] provide overviews based on
contemporary principal-agent models). The same may be said of more
conventional initiatives such as improved educational opportunities
and policies to support home ownership. There is good evidence, for
example, that home ownership promotes active participation in local
politics and a willingness to discipline personally those engaging in
antisocial behaviors in the neighborhood (Sampson, Raudenbush, and
Earls 1997). An expansion of subsidies designed to promote employ-
ment and increase earnings among the poor, suggested by Edmund
Phelps (1997), would tap powerful reciprocity motives. Similarly, so-
cial insurance programs might be reformulated along lines suggested
by John Roemer (1993) to protect individuals from risks over which
they have no control, while not indemnifying people against the results
of their own choices (other than providing a minimal ¬‚oor to living
standards). In this manner, for example, families could be protected
against regional ¬‚uctuations in home values”the main form of wealth
for most people”as Robert Shiller (1993) has shown. Other forms of
insurance could partially protect workers from shifts in demand for
their services induced by global economic changes.
An egalitarian society can be built on the basis of these and other
policies consistent with strong reciprocity, along with a guarantee of
an acceptable minimal living standard consistent with the widely
documented motives of basic needs generosity. But if we are correct,
economic analysis will be an inadequate guide to policymaking in the
area unless it revises its foundational assumptions concerning human
Reciprocity and the Welfare State 299


1. The numbers of observations for these questions were 78 and 79 for the poor group
and 294 and 281 for the rich group. Gilens (1999) makes similar observations using earlier
2. See Mof¬tt (1983) for an early model of welfare stigma. See also Lindbeck, Nyberg,
and Weibull (1995) for related work that addresses the role of work norms in redis-
tributive politics and treats such norms as endogenous to the provision of government
3. The exact wording of this questions is: ˜˜Why, in your opinion, are there people who
live in need? Here are four opinions; which is the closest to yours? 1. Because they have
been unlucky; 2. Because of laziness and lack of willpower; 3. Because there is much
injustice in our society; 4. It is an inevitable part of modern progress; 5. None of
these.™™ Our dummy variable is one for respondents who answered ˜˜Because of laziness
and lack of willpower,™™ and zero for respondents who gave one of the other four
4. These results do not depend on the particular sample and speci¬cation that we pres-
ent. In all speci¬cations, the effect of moving up to the next income quartile is an order
of magnitude smaller than the effect of believing that poverty exists because the poor are
lazy. When the question about whether or not the public authorities are doing enough for
the poor was omitted from our composite measure of concern about poverty, the effect of
income was not even signi¬cant, regardless of whether other demographic variables
were included in the regression, while the effect of beliefs that the poor are lazy remained
large and highly signi¬cant.

5. We drop non-responses and ˜˜don™t know™™ responses. Another option would be to in-
clude ˜˜don™t know™™ as a valid response. However, how and why people develop well-
de¬ned preferences and beliefs is beyond the scope of this chapter. We focus on why
people oppose or support income redistribution given that their beliefs and preferences
are well-de¬ned.
6. There are several additional questions that might capture self-interest that are
excluded from the model presented here. See Fong (2001) for a discussion and analysis
of these variables.
7. Space limitations prevent us from presenting these results here. However, the ¬nding
using ordered probit are presented in Fong (2001).
8. We assume that people agree on which group bene¬ts and which loses when they be-
lieve that a person™s race or sex is important to getting ahead.
9. The sample size in these regressions ranges from 584 to 594.
10. A third study by Weaver, Shapiro, and Jacobs (1995), drawing in addition on Na-
tional Opinion Research Council and General Social Survey data, comes to broadly simi-
lar conclusions.
11. As a general rule, non-experts vastly overstate the share of the tax revenues devoted
to things of which they disapprove, whether it be foreign aid, welfare, AIDS research, or
military expenditure”the opposition is generally the cause of the exaggeration, not vice
300 Fong, Bowles, and Gintis


Alesina, Alberto, Edward Glaeser, and Bruce Sacerdote. ˜˜Why Doesn™t the United States
Have a European-Style Welfare State?™™ Brookings Papers on Economic Activity 2 (2001):
Atkinson, A. B. The Economic Consequences of Rolling Back the Welfare State. Cambridge,
MA: MIT Press, 1999.
Besley, Timothy, and Stephen Coate. ˜˜Understanding Taxpayer Resentment and Statisti-
cal Discrimination,™™ Journal of Public Economics 48 (1992): 165“183.
Bowles, Samuel, and Herbert Gintis. ˜˜Schooling, Skills and Earnings: A Principal-Agent
Approach,™™ in Kenneth Arrow, Samuel Bowles, and Steven Durlauf (eds.), Meritocracy
and Economic Inequality. Princeton, NJ: Princeton University Press, 1998.
”””. Recasting Egalitarianism: New Rules for Markets, States, and Communities. Erik Olin
Wright (ed.). London: Verso, 1999.
Brewer, Marilynn B., and Norman Miller. Intergroup Relations. Paci¬c Grove, CA:
Brooks/Cole Publishing Company, 1996.
Deutsch, Morton. Distributive Justice. New Haven: Yale University Press, 1985.
Eckel, Catherine, and Philip Grossman. ˜˜Altruism in Anonymous Dictator Games,™™
Games and Economic Behavior 16 (1996): 181“191.
Edda. ˜˜Havamal.™™ In D. E. Martin Clarke (ed.), The Havamal, with Selections from Other
Poems in the Edda. 1923.
Farkas, Steve, and Jean Robinson. The Values We Live By: What Americans Want from Wel-
fare Reform. New York: Public Agenda, 1996.

Fong, Christina M. ˜˜Social Preferences, Self-Interest, and the Demand for Redistribution,™™
Journal of Public Economics 82,2 (2001): 225“246.
”””. ˜˜Empathic Responsiveness: Evidence from a Randomized Experiment on Giving
to Welfare Recipients.™™ Carnegie-Mellon University, 2004.
Forsythe, Robert, Joel Horowitz, N. E. Savin, and Martin Sefton. ˜˜Replicability, Fairness
and Pay in Experiments with Simple Bargaining Games,™™ Games and Economic Behavior
6,3 (May 1994): 347“369.
Gallup Organization, Haves and Have-Nots: Perceptions of Fairness and Opportunity. Gallup
Press, 1998.
Gilens, Martin. Why Americans Hate Welfare. University of Chicago Press, 1999.
Hayek, Frederick. The Three Sources of Human Values. London: London School of Econom-
ics, 1978.
Heclo, Hugh. ˜˜The Political Foundations of Antipoverty Policy.™™ In Sheldon H. Danziger
and Daniel H. Weinberg (eds.), Fighting Poverty: What Works and What Doesn™t. Cam-
bridge: Harvard University Press, 1986, 312“341.

Kollock, Peter. ˜˜Transforming Social Dilemmas: Group Identity and Cooperation.™™ In
Peter Danielson (ed.), Modeling Rational and Moral Agents. Oxford: Oxford University
Press, 1997.
Reciprocity and the Welfare State 301

´ ´ ´´
Kolm, Serge-Christope. La Bonne Economie: La Reciprocite Generale. Presses Universitaires
de France: Paris, 1984.
Kropotkin, Petr. Mutual Aid: A Factor in Evolution. New York: Black Rose Books,
Lindbeck, Assar, Sten Nyberg, and Jorgen Weibull. ˜˜Social Norms and Incentives in the
Welfare State,™™ Quarterly Journal of Economics 114 (1995): 1“35.
Loewenstein, George F., Leigh Thompson, and Max H. Bazerman. ˜˜Social Utility and De-
cision Making in Interpersonal Contexts,™™ Journal of Personality and Social Psychology 57, 3
(1989): 426“441.
Luttmer, Erzo F. P. ˜˜Group Loyalty and the Taste for Redistribution,™™ Journal of Political
Economy 109, 3 ( June 2001): 500“528.
Miller, David. Principles of Social Justice. Cambridge, MA: Harvard University Press, 1999.
Mof¬tt, Robert. ˜˜An Economic Model of Welfare Stigma,™™ American Economic Review 73
(1983): 1023“1035.
”””, David Ribar, and Mark Wilhelm. ˜˜Decline of Welfare Bene¬ts in the U.S.: The
Role of Wage Inequality,™™ Journal of Public Economics 68, 3 ( June 1998): 421“452.
Moore, Barrington Jr. Injustice: The Social Bases of Obedience and Revolt. White Plains: M. E.
Sharpe, 1978.

Page, Benjamin, and Robert Shapiro. The Rational Public: Fifty Years of Trends in American™s
Policy Preferences. Chicago: University of Chicago Press, 1992.

Phelps, Edmund S. Rewarding Work: How to Restore Participation and Self-Support to Free
Enterprise. Cambridge, MA: Harvard University Press, 1997.
Piketty, Thomas. ˜˜Social Mobility and Redistributive Politics,™™ Quarterly Journal of Eco-
nomics 110, 3 (August 1995): 551“584.
Reif, Karlheinz, and Anna Melich. Euro-Barometer 31A: European Elections, 1989: Post-
Election Survey, June“July 1989 [Computer File]. Conducted by Faits et Opinions, Paris.
ICPSR ed. Ann Arbor, MI: Inter-university Consortium for Political and Social Research,
Roberts, Kevin. ˜˜Voting over Income Tax Schedules,™™ Journal of Public Economics 8 (De-
cember 1977): 329“340.

Roemer, John. ˜˜A Pragmatic Theory of Responsibility for the Egalitarian Planner,™™ Philos-
ophy and Public Affairs 22 (1993): 146“166.
Sampson, Robert J., Stephen W. Raudenbush, and Felton Earls. ˜˜Neighborhoods and Vi-
olent Crime: A Multilevel Study of Collective Ef¬cacy,™™ Science 277 (August 15, 1997):
Scott, James C. The Moral Economy of the Peasant: Rebellion and Subsistence in Southeast Asia.
New Haven, CT: Yale University Press, 1976.
Shiller, Robert J. Macro Markets: Creating Institutions for Managing Society™s Largest Eco-
nomic Risks. Oxford: Clarendon Press, 1993.
Skitka, Linda, and Philip Tetlock. ˜˜Providing Public Assistance: Cognitive and Motiva-
tional Processes Underlying Liberal and Conservative Policy Preferences,™™ Journal of Per-
sonality and Social Psychology 65, 6 (1993): 1205“1223.
302 Fong, Bowles, and Gintis

Walster, Elaine, G. William Walster, and Ellen Berscheid. Equity, Theory and Research. Bos-
ton: Allyn and Bacon, 1978.
Weaver, R. Kent, Robert Y. Shapiro, and Lawrence R. Jacobs. ˜˜Poll Trends: Welfare,™™ Pub-
lic Opinion Quarterly 39 (1995): 606“627.
Weiner, Bernard. Judgments of Responsibility: A Foundation for a Theory of Social Conduct.
New York: The Guilford Press, 1995.
Williamson, John B. ˜˜Beliefs about the Motivation of the Poor and Attitudes toward Pov-
erty Policy,™™ Social Problems 21, 5 ( June 1974): 734“747.
11 Fairness, Reciprocity, and
Wage Rigidity

Truman Bewley

11.1 Introduction

Most empirical tests of the many competing theories of wage rigidity
use publicly available data on pay rates and employment that reveal
little about the institutions and motivations that explain wage behav-
ior. In order to learn more, some economists have analyzed unusual
sources of data or have conducted surveys and experiments. Manage-
ment scientists and organizational psychologists have for years been
collecting data relevant to wage rigidity. I here report on what I know
of these sources of information about the origins of wage rigidity.

11.2 Are Wages and Salaries Downwardly Rigid?

It is sensible to check whether wages really are downwardly rigid be-
fore considering why they are. This question is surprisingly hard to an-
swer, because appropriate data are lacking. It is not even clear what
the appropriate de¬nition of the wage should be. A ¬rm™s marginal
costs depend on the average hourly nominal labor cost per job. Em-
ployee welfare depends on total nominal compensation per worker.
A third possibility is nominal compensation for an employee with a
given job tenure and continuing in the same position with the same
employer under ¬xed working conditions. If the employee is paid by
the hour, it is the hourly rate and the bene¬ts that count. Total com-
pensation is the relevant pay rate for salaried employees. This third
de¬nition is the one most closely associated with employees™ and man-
agers™ notions of fairness and hence is most pertinent to the managerial
concerns that explain downward wage rigidity.
In order to adhere even more closely to the sense of fairness prevail-
ing in business, it might be advisable to include only base pay and
304 Bewley

exclude variable components, such as bonuses. The three pay rates can
change independently. For instance, the average hourly labor costs of a
job can increase with no change in any worker™s pay, if the seniority of
workers assigned to the job increases. Similarly, changes in hours
worked or in job assignments can change an individual™s total pay
without changing hourly pay rates or labor costs per job. There are
conceptual ambiguities associated with bene¬ts. For instance, if an in-
crease in the costs of a given medical insurance policy were shared be-
tween the ¬rm and its work force, the ¬rm™s nominal labor cost per job
would increase, but workers would probably feel that the total value of
their medical bene¬ts had decreased.
A wage cut should be de¬ned as a reduction in the wage of the third
de¬nition of wage”the pay of an employee continuing to work under
unchanged conditions. Unfortunately, this pay rate is the most dif¬cult
to measure, because it requires knowledge of much more than just
total pay.
Lebow, Saks, and Wilson (1999) is the only study I know of that
measures the ¬rst de¬nition of wage”the ¬rm™s average labor costs.
The authors use U.S. Bureau of Labor Statistics data and ¬nd that
wage costs are somewhat rigid downward, although there is a consid-
erable amount of wage reduction.
There is a large literature that uses surveys of the pay of individual
workers to study variation in the third kind of wage. The studies in-
clude McLaughlin (1994, 1999), Lebow, Stockton, and Wascher (1995),
Card and Hyslop (1997), Kahn (1997), Fehr and Goette (1999), and
Smith (2000, 2002). Some of these authors had to struggle with possible
errors in the reporting of wage rates. All of the studies suffer from ig-
norance of changes in hours worked, job assignments, bonuses, or
working conditions, so that it is not clear that the data reveal the wage
of the third de¬nition. All the studies report large amounts of wage
Surveys of ¬rms on wage rigidity reach con¬‚icting conclusions.
Roger Kaufman (1984), Alan Blinder and Don Choi (1990), Jonas Agell
and Per Lundborg (1999), and myself, Bewley (1999), simply asked
employers whether they had reduced pay. The responses probably ap-
ply to the third de¬nition of wage, but there is no way to be sure. None
of the ¬rms in Kaufman™s sample of 26 British ¬rms had considered
nominal wage cuts during the recession occurring at the time of his
study. Blinder and Choi found a high incidence of pay reduction, in 5
of the 19 American ¬rms they studied. Agell and Lundborg, on the
Fairness, Reciprocity, and Wage Rigidity 305

other hand, found almost no wage cutting; two out of 153 responding
Swedish ¬rms had experienced nominal wage cuts during the previous
seven years, a period of high unemployment and low in¬‚ation. The
wage cuts that did occur were for just a few employees. The near ab-
sence of wage cutting may be explained by institutional factors speci¬c
to Sweden. Although I conducted my survey during a recession and
actively sought out ¬rms that had cut pay, I found a low incidence of
pay cuts; of 235 businesses studied, 24 had reduced the base pay of
some or all employees during the recession of the early 1990s.
Similarly con¬‚icting results appear in surveys of union wage agree-
ments. In Current Wage Developments and the Monthly Labor Review, the
Bureau of Labor Statistics reports on general wage changes for both
union and nonunion manufacturing production workers for the years
1959 through 1978. These data show a negligible number or wage
reductions; cuts for less than a half a percent of the workers in every
year.1 (The corresponding percentage for my sample was 0.14 per-
cent.) Con¬‚icting evidence has been found by Mitchell (1985), who
uses Bureau of Labor Statistics data to calculate that 13 percent of all
workers covered by major new contracts suffered wage cuts in 1983.
Similarly, Fortin (1996) ¬nds that 6 percent of 1,149 large non“cost-of-
living-adjusted union wage settlements in Canada from 1992 to 1994
involved wage cuts.
Much less ambiguous evidence of downward rigidity in the third
kind of wage is contained in the few studies that use company records
to learn the histories of job assignments, hours worked, and pay of in-
dividual employees. The studies include Baker, Gibbs, and Holmstrom
(1994); Wilson (1996); and Altonji and Devereux (2000). Unfortunately,
these authors study only three ¬rms; Baker, Gibbs, and Holmstrom
study one ¬rm, Wilson studies two, one of which is the ¬rm studied
by Baker, Gibbs, and Holmstrom, and Altonji and Devereux study the
third. Only Altonji and Devereux report data on hourly workers. The
other two studies have information only on salaried employees.
All three studies ¬nd a negligible number of pay reductions. Altonji
and Devereux ¬nd that 2.5 percent of hourly workers experienced
wage cuts, but almost all of these were ˜˜associated with changes
between full and part time status, or with changes in whether perfor-
mance incentives are part of compensation.™™ These ¬ndings are rein-
forced by a telephone survey Akerlof, Dickens, and Perry (1996) made
of 596 people in the Washington, DC, area. The key question was
˜˜Excluding overtime, commissions, and bonuses, has your base rate of
306 Bewley

pay changed since a year ago today?™™ A negligible number reported
pay reductions. Given the form of the question, this evidence probably
pertains to the third de¬nition of wages. Contradicting this evidence
are two similar surveys conducted in New Zealand in 1992 and 1993,
where 8 percent and 5 percent, respectively, of the respondents re-
ported hourly wage reductions (Chapple, 1996, tables 11.2 and 11.3).
More work should be done. No one has yet conducted a large survey
that accurately measures the incidence of cuts in pay according to the
third de¬nition of wages.

11.3 Evidence from Surveys by Economists

There are six surveys by economists of business managers responsible
for compensation policy. The goal of ¬ve of these was to learn the
reasons for downward wage rigidity”the studies of Roger Kaufman
(1984), Alan Blinder and Don Choi (1990), Jonas Agell and Per Lund-
borg (1995, 1999), Carl Campbell and Kunal Kamlani (1997), and Bew-
ley (1999). The sixth study, that of David Levine (1993), also contains
relevant information. Although the ¬ndings of the studies differ to
some extent, they give a consistent picture of the sources of wage rigid-
ity. I also discuss a paper by Jennifer Smith (2002), who analyzes a sur-
vey of British workers.
I ¬rst summarize my own ¬ndings, based on interviews with 246
company managers and 19 labor leaders in the northeastern United
States during the early 1990s when unemployment was high because
of a recession. I present my ¬ndings as re¬‚ecting the views of manag-
ers, although labor leaders had almost exactly the same opinions on
the matters discussed. The primary resistance to wage reduction comes
from upper management, not from employees. The main reason for
avoiding pay cuts is that they damage morale. Morale has three com-
ponents. One is identi¬cation with the ¬rm and an internalization of
its objectives. Another is trust in an implicit exchange with the ¬rm
and with other employees; employees know that aid given to the ¬rm
or to co-workers will eventually be reciprocated, even if it goes unno-
ticed. The third component is a mood that is conducive to good work.
The mood need not be a happy one, though happiness is important for
the performance of some jobs, such as those that involve dealing with
customers. The mood could be dislike of an unpleasant job combined
with grim focus on achievement or pride in accomplishment. Good
morale is not equivalent to happiness or job satisfaction. Workers may
Fairness, Reciprocity, and Wage Rigidity 307

be content simply because they do nothing. Good morale has to do
with a willingness voluntarily to make sacri¬ces for the company and
for coworkers.
A general sense of fairness is conducive to good morale; it contrib-
utes to an atmosphere of mutual trust. The sense of fairness is created
by having supervisors treat workers decently, by having impartial
rules for settling disputes and determining promotions and job assign-
ments, and by using reasonable standards for setting the relative pay
of different employees. These standards are often elaborate systems
and are termed ˜˜internal pay structures.™™ They clearly determine pay
differentials on the basis of such factors as training, experience, tenure
at the ¬rm, and productivity. The structures are extremely important,
because any perceived pay inequity within a ¬rm may cause indigna-
tion and disrupt work. The standards of internal equity are somewhat
arbitrary, can depend strongly on company tradition, and may not
specify that pay be proportional to productivity. Many employers be-
lieve that productivity of the work force as a whole is maximized
when pay increases less than productivity, although some individuals
might produce more if given stronger ¬nancial incentives.
There is a division of opinion within business about how sensitive
pay should be to productivity. Big income differentials due to differen-
ces in productivity can cause resentment, especially if productivity is
dif¬cult to measure (which it often is). Nevertheless many ¬rms use
piece rates when productivity can be measured unambiguously, and
even when piece rates are impractical ordinary notions of equity re-
quire that differences in people™s contributions be rewarded ¬nancially
to some extent. The sensitivity of pay to productivity may be blunted
by the in¬‚uence of other factors on pay, such as longevity with the
¬rm. No matter how sensitive the pay of individuals is to their produc-
tivity, ¬rms automatically keep the average pay of broad categories of
workers roughly equal to the value of their average marginal product
by adjusting the number of workers in each category to the pro¬t max-
imizing level.
Managers are concerned about morale because of its impact on labor
turnover, recruitment of new employees, and productivity. Disgrun-
tled employees are likely to quit as soon as they ¬nd another job. A
company™s best recruiters are its employees, so it is important not to
have them go around complaining about their company. Morale has
little impact on productivity in the sense of speed in carrying out rou-
tine tasks. Habit and working conditions largely determine this sort of
308 Bewley

productivity. Managers have in mind the impact of morale on workers™
willingness to do the extra thing, to encourage and help each other, to
make suggestions, and to work well even when not supervised. Also,
workers with bad morale waste time complaining to each other. In
considering the impact of morale on productivity, it is important to re-
alize that supervision is so expensive that many employees are not
closely supervised and have a signi¬cant amount of freedom on the
job. Except in some low-level jobs, employers rely on workers™ volun-
tary cooperation and do not simply give orders.
When considering why wage cuts hurt morale, it is necessary to dis-
tinguish new from existing employees. The morale of existing employ-
ees is hurt by pay cuts because of an insult effect and a standard of
living effect. Workers are used to receiving regular pay increases as a
reward for good work and loyalty and so interpret a pay cut as an af-
front and a breach of implicit reciprocity, even if the pay of all employ-
ees is reduced. Individual workers may take a pay cut less personally if
everyone™s pay falls, but when everyone in a company suffers, they all
complain to each other and stimulate each other™s discontent. The stan-
dard of living effect is the resentment caused by the fall in income.
Workers blame their employer when they ¬nd their lifestyles curtailed.
This effect is closely related to what experimental economists call ˜˜loss
The arguments just given do not apply to newly hired workers. They
probably would hardly care if their ¬rm had a general pay cut just
before they were hired. It is possible, however, to reduce the pay of
newly hired workers while continuing to give normal pay increases to
existing employees; new workers hired after a certain date would sim-
ply be paid according to a reduced pay scale. Some ¬rms have ex-
perimented with such two-tier pay structures. Managers say that new
workers hired in the lower tier might be glad at ¬rst to have their jobs,
but that their attitude would change after they learned that their pay
violated the traditional internal pay structure. They would believe
they were being treated unfairly, their resentment would hurt their
morale, and their discontent could spread to others.
Resistance to wage reduction and the need for internal pay equity
stem from ideas of fairness that usually refer to some reference wage.
The reference wage for pay cuts is the previous wage. The reference
wage for internal equity is that of other workers within the ¬rm with
similar quali¬cations and similar jobs. The fairness of wages has little
to do with pro¬ts or productivity, although both workers and man-
Fairness, Reciprocity, and Wage Rigidity 309

agers ¬nd it appropriate that employees share to some extent in the
success of their company. While managers attempt to use reasonable
criteria when establishing an internal pay structure, once a structure is
established, tradition by itself makes it a standard of fairness.
The explanation of downward pay rigidity just given is closely re-
lated to the morale theory proposed by Solow (1979), Akerlof (1982),
and Akerlof and Yellen (1988, 1990). They assert that morale and hence
productivity increase with the wage and that the trade-off between la-
bor costs and productivity determines a wage that is independent of
the unemployment rate. Akerlof (1982) uses his gift exchange model
to explain the link between the wage and morale. According to this
model, workers offer more effort than is demanded by the employer in
exchange for pay rates in excess of market clearing levels, so that effort
increases with the wage level.
I do not believe that this theory is fully accurate, however, because
employers say they do not see much connection between effort or mo-
rale and wage levels; productivity and morale do not increase with pay
levels, although they can be hurt by pay reductions or disappointingly
small raises. Even generous pay increases do not increase morale or
productivity, because workers quickly get used to increases and grow
to believe they have a right to them. They soon lose track of any idea
that they should offer extra effort in exchange for higher pay. Employ-
ers do not think about a trade-off between labor costs and the produc-
tivity of existing employees when determining pay levels, though
managers do consider the trade-off between labor costs and the quality
of labor that a ¬rm can attract and retain.
In the theory of Akerlof, Solow, and Yellen, morale depends on the
level of the wage, whereas in the explanation I have described, wages
affect morale only when reduced. What is accurate in the Akerlof-
Solow-Yellen theory is the idea that employers avoid cutting pay be-
cause doing so would hurt morale. What the theory misses is that
employees usually have little notion of a fair or market value for their
services and quickly come to believe they are entitled to their existing
pay, no matter how high it may be. Workers do not use pay rates at
other ¬rms as reference wages because they know too little about
them. Exceptions to this statement may occur when workers are repre-
sented by an active labor union that keeps them informed about what
other ¬rms are paying.
Although pay cuts are unusual, they do occur and usually do not
have the harmful effects described by managers when arguing that
310 Bewley

pay should not be cut. The explanation for this inconsistency is that
pay cuts are accepted by the work force if they prevent a ¬rm from
closing or if they save a large number of jobs. Managers are con¬dent
that they can convince employees that a pay cut is necessary, if it is in
fact the case.
One of the puzzles discussed in the literature on wage rigidity has
been why ¬rms lay off workers rather than reduce their pay. I found
that most managers believe that the elasticity of their company™s de-
mand for labor is so low that pay cuts would not reduce an excess sup-
ply of labor within the ¬rm. The elasticity is small, because direct labor
is a small fraction of marginal costs and the price elasticity of product
demand is far from in¬nite. Only in ¬rms with a high elasticity of
product demand, such as construction companies, is it believed that
pay cuts can signi¬cantly increase the demand for labor. Many of the
pay cuts that occurred in the companies I researched were made in
these types of ¬rms or in ones that were in danger of closing. Other
¬rms where pay reduction was an alternative to layoffs were those
that laid off workers simply to save money, not to get rid of excess la-
bor (and there were many such companies). The main argument for
preferring layoffs to pay cuts is that layoffs do less damage to morale.
Laid off workers suffer, but they are no longer in the ¬rm. In the words
of one manager, ˜˜Layoffs get the misery out the door.™™ Good manage-
ment practice is to delay potential layoffs until the employer can make
a large number all at once, and then to assure those who remain that
there will be no more layoffs for some time.
Any damage to morale from layoffs is temporary, whereas that of a
pay reduction is long-term. Other arguments are that layoffs increase
productivity (whereas pay cuts hurt it) and that layoffs give manage-
ment some control over who leaves (whereas the best workers are like-
ly to quit when pay is reduced). The tendency for the best workers to
quit is a concern in many ¬rms, because the leveling effects of internal
equity on pay mean that pay for workers within a given job category
increases less than their contribution to pro¬ts as this contribution in-
creases. Another consideration is that feasible layoffs often save much
more money than feasible pay cuts, which usually cannot be more
than about 20 percent of base pay. Layoffs save the ¬xed costs of em-
ployment, which are substantial, whereas cuts reduce only the variable
part of pay.
Another puzzle appearing in economics literature is why unem-
ployed workers do not try to take jobs away from employed people by
Fairness, Reciprocity, and Wage Rigidity 311

offering to replace them at lower pay. Robert M. Solow (1990) has
proposed that the unemployed do not engage in such undercutting be-
cause of a social convention against it. I found that explicit undercut-
ting is impossible for most people, because they do not know exactly
what job they are applying for or what its pay is. However, it is not un-
common during periods of high unemployment for job applicants to
offer to work for extremely low pay. These offers are not frowned
upon but are almost never accepted, except to reduce pay during the
initial probationary period of employment, because accepting the offers
would violate the internal pay structure and could demoralize the new
A similar puzzle is why ¬rms do not replace employees during
recessions with cheaper unemployed workers. In reply to this ques-
tion, Assar Lindbeck and Dennis J. Snower (1988) proposed, with their
insider-outsider theory”that ¬rms seldom replace workers because
old employees who remained would harass and refuse to cooperate
with and train the replacements, thereby reducing their productiv-
ity. I found that the main reasons employers do not replace employ-
ees are that the new ones would lack the skills of the existing ones
and replacement would demoralize the work force. The skills would
be lost in part because many of them are speci¬c to the ¬rm. Man-
agers agreed that after replacement, the unreplaced workers would
probably boycott the new ones, but asserted that other factors took
precedence as an explanation of why employees were not replaced
during recessions.
John Maynard Keynes (1936) proposed that downward wage rigid-
ity is explained by employees™ preoccupation with pay differentials
among workers in similar jobs at different ¬rms. I found, however,
that such external pay differentials are not an issue, except in highly
unionized industries. In most companies, employees know so little
about pay rates at other ¬rms that they do not know whether or not
they are underpaid. Although labor unions do try to keep their mem-
bers informed of pay rates at other companies, unions are weak in the
United States.
A popular explanation of wage rigidity is the ˜˜No Shirking Theory™™
of Shapiro and Stiglitz (1984). According to their model, managers in-
duce workers to perform well by ¬ring them if their productivity falls
below a prescribed level. Being ¬red is more costly to the worker
the higher is the wage, so that higher wages make it possible to insist
on greater productivity. According to Shapiro and Stiglitz™s theory,
312 Bewley

managers set wages so as to optimize the trade-off between wage costs
and productivity. This theory does not really explain downward wage
rigidity, because it implies that wages should decline when unemploy-
ment increases. As unemployment rises, however, it becomes harder to
¬nd a new job, so that losing a job is more costly to the worker. The
theory also implies that ¬rms can then obtain the same productivity at
lower wages, which is not necessarily the case.
Despite these drawbacks, the No Shirking Theory is popular among
economists. However, when I asked managers and labor leaders about
it, they almost always told me that it did not apply. As was explained
in connection with the Akerlof™s gift exchange model, employers do
not see much connection between pay and morale. Nor do employers
obtain cooperation by threatening to ¬re shirkers. To do so would cre-
ate a negative atmosphere that could damage morale and encourage
rebelliousness. Workers may malinger on the job, but are seldom dis-
missed for doing so, except during the short probationary period after
hiring. Shirking is usually dealt with through discussions and repri-
mands, and workers are normally ¬red only because of a pattern
of egregious behavior. Managers elicit effort by clearly explaining to
employees what is expected of them, identifying their shortcomings in
a constructive manner, pointing out the importance of the tasks they
perform, showing interest in and appreciation of their work, and mak-
ing them feel they are valued members of the organization. Most
employees like to work, and cooperate, and please their boss.
Despite the inapplicability of the no shirking theory, the incentive
mechanism it posits can be effective. For instance, employees do work
harder during economic slowdowns when new jobs are dif¬cult to
¬nd and layoffs are imminent, especially if layoffs are done on the
basis of performance”that is, if the least productive workers in a job
category are laid off ¬rst. The increase in effort occurs both because
job loss becomes more dangerous during an economic slowdown and
because workers try to avoid layoffs by being cooperative and produc-
tive. Because layoffs stem from circumstances not controlled by man-
agement, they do not generate the hostility that might be generated by
systematically ¬ring slackers.
Although ¬ring is not used to stimulate work effort, ¬nancial incen-
tives are thought to be very effective in doing so and are believed not to
impair morale. Incentives can even improve morale, because workers
¬nd it fair that they be rewarded for their contributions to the com-
pany. Provided incentives are not exaggerated, they contribute to inter-
Fairness, Reciprocity, and Wage Rigidity 313

nal equity. Discipline and even ¬ring can contribute to internal equity
as well, because workers who make the effort to do their job well and
obey company rules can be outraged if they see others get away with
¬‚agrant misbehavior. The main purposes of ¬ring are to protect the
company from malefactors and incompetents and to maintain internal
equity. Dismissals that are managed correctly earn managers respect.
What needs to be avoided is an atmosphere of retribution that menaces
everyone. This assertion appears not to apply, however, to low-level
jobs. There was evidence that employers do sometimes use coercion to
motivate workers in low-paying jobs that require little training and
where employees are easily supervised.
Another popular explanation of wage rigidity is the adverse selec-
tion model of Andrew Weiss (1980, 1990). There are two versions of
this model, dealing with quits and hiring, respectively. In the quits ver-
sion, managers prefer layoffs to pay cuts because the best workers
leave if pay is reduced, whereas if managers lay off workers, they can
select those who leave. According to the hiring version, managers be-
lieve that the higher is the level of pay that a job applicant is willing to
accept, the higher his or her unobservable quality will be, and pay
offers to new hires are determined by the trade-off between worker
quality and pay. Weiss asserts that the relation between pay and job
candidate quality is determined by alternative employment in the sec-
ondary sector, where quality is perfectly observable. The secondary
sector consists of home production or jobs that have high turnover and
are usually part-time. The hiring version of Weiss™s adverse selection
theory applies to the primary sector, where jobs are long-term and usu-
ally full-time. He assumes that real wages in the secondary sector are
downwardly rigid because of constant returns to labor in production
in this sector. According to the theory, this downward rigidity is then
transferred to the primary sector through the impact of adverse selec-
tion on hiring pay.
I found strong support for Weiss™s theory as it applies to quits, but
none as it applies to hiring. Although managers believe that a pay cut
would cause their best employees to quit, I found no evidence that
recruiters use pay aspirations as an indicator of job candidate quality.
Job recruiters treated the trade-off between pay and worker quality as
a basic fact of life, but they did not learn more about candidate quality
from pay demands. Recruiters used the trade-off as a reason for not
reducing pay only for skills that were in short supply despite the eco-
nomic slowdown. For most skills, they believed they could hire all the
314 Bewley

workers they needed during the recession at lower rates of pay. The
secondary sector does not sustain candidates™ reservation wages. Hir-
ing pay is more ¬‚exible in the secondary than the primary sector”
the opposite of the effect predicted by Weiss™s theory. Two-tier or
multiple-tier wage structures are commonplace in the secondary sec-
tor, because the part-time and casual nature of the jobs keeps workers
from getting to know each other well and so reduces the need to avoid
internal pay inequities.
Kaufman™s (1984) results support my main ¬ndings. He conducted
interviews in twenty-six British ¬rms in 1982 during a period of high
unemployment. He too found that employers ˜˜believed they could
¬nd quali¬ed workers at lower wages.™™ He found that employers
avoid replacing workers with cheaper ones because of the value of
skills and of long-term employment relationships. Employers avoid
pay cuts because of concerns about productivity. Because supervision
is costly, employers rely ˜˜heavily on the goodwill of their employees.™™
Workers view wages as ˜˜a reward for performing competently™™ and
would regard a wage cut as an ˜˜affront.™™ Employers avoid hiring
new employees at lower pay rates than existing ones because doing
so would create ˜˜intolerable frictions,™™ especially with ˜˜the newer
workers who would eventually become disgruntled about the two tier
wage structure.™™ Managers feel they can cut nominal pay if ˜˜severe cut-
backs or closure will be necessary unless the nominal wage cuts are
Blinder and Choi (1990) interviewed managers at nineteen ¬rms,
and their ¬ndings largely agree with my own. They found little evi-
dence to support Andrew Weiss™s idea that job candidates™ wage
demands are useful indicators of productivity. Few of Blinder and
Choi™s nineteen respondents thought that a higher wage would induce
greater work effort, although a majority thought that a wage cut
would diminish effort. The majority said that effort would decrease
after a wage cut because of reduced morale. None mentioned the
decreased penalty for being ¬red. A majority of their respondents
believed that higher unemployment would bring greater work effort.
All respondents answering the question felt that a wage cut would
increase labor turnover, although only one of the ¬ve ¬rms that had
recently reduced pay had experienced a signi¬cant increase in quits.
˜˜The reason for the wage cut seemed to matter. . . . Generally, wage
reductions made to save the ¬rm from failure or to align wages with
those of competitors are viewed as justi¬able and fair while those
Fairness, Reciprocity, and Wage Rigidity 315

made just to raise pro¬ts are not.™™ Managers felt strongly that having a
wage policy that was viewed as unfair ˜˜would affect work effort, quits,
and the quality of future applicants. . . . Attitudes like this must be
strong deterrents to implementing an ˜unfair™ wage policy though . . .
that does not necessarily rule out wage reductions under the right
circumstances™™ (Blinder and Choi 1990, 1008“1009). Blinder and Choi
found strong support for the idea that worker concern about relative
wages is a reason for downward wage rigidity. The question they
asked, however, did not distinguish between internal and external pay
comparisons, so the support their ¬ndings give to Keynes™ relative
wage theory is ambiguous.
Campbell and Kamlani (1997) surveyed 184 ¬rms, sending question-
naires to managers who were asked to rate the importance of various
statements on a scale from one to four. Most of Campbell and Kamla-
ni™s ¬ndings agree with my own and those of the other surveys. Their
respondents attached the greatest importance to the idea that wage
cuts would induce the best workers to quit, which is Weiss™s adverse
selection idea as it applies to quits (Weiss 1990). Campbell and Kam-
lani found that the best workers are valued because pay does not in-
crease in proportion to productivity and employees™ skills are often
¬rm-speci¬c. Other important management concerns were that a wage
cut would increase turnover (and hence hiring and training costs) and
would generate bad feeling that would lead to less work effort. Camp-
bell and Kamlani found less support for the idea that pay cuts would
make recruitment more dif¬cult and found no support for the no shirk-
ing model. Managers did not agree that cutting pay would decrease
effort because of a reduced fear of job loss, but did agree that effort
would decline because of decreased gratitude and loyalty.
Furthermore, good management-worker relations were thought to
have a much greater impact on effort than high wages, close super-
vision, or high unemployment. There was also no support for the
insider-outsider theory. Most managers did not believe that if the ¬rm
discharged some of its current workers and replaced them with new
ones at a lower wage that the old workers who remained would harass
and refuse to cooperate with the newly hired ones. The reasons for a
pay cut matter”its negative impact on effort would be greater if the
¬rm were pro¬table than if it were losing money. There is an asymme-
try between the impact of wage increases and decreases; the deleteri-
ous effect a wage decrease would have on effort would greatly exceed
the positive effect of a wage increase. Similarly, a wage decrease would
316 Bewley

have a worse impact on effort and morale than having paid the lower
wage for a long time.
Agell and Lundborg (1995, 1999, 2003) did questionnaire surveys of
managers in Swedish manufacturing ¬rms, obtaining responses from
179 ¬rms in 1991 and from 157 of those ¬rms in a follow-up survey in
1998. A strong majority of the respondents felt that a nominal wage cut
would be strongly resisted by employees and that at least 50 percent of
the ¬rm™s jobs would have to be threatened to make a cut acceptable.
This ¬nding may be in¬‚uenced by the fact that Swedish laws make
it dif¬cult to reduce pay. The respondents gave strong support to
Keynes™s theory that the desire to preserve external wage relativities
explains downward wage rigidity. The inconsistency between this
¬nding and my own is probably explained by the much greater impor-
tance of labor unions in Sweden than in the United States. Agell and
Lundborg found little or no support for the no shirking model. Manag-
ers did not regard shirking as very common, and ˜˜employees who
were repeatedly caught shirking were punished by a simple verbal re-
buke™™ (Agell and Lundborg 1999, 11). Like Campbell and Kamlani,
Agell and Lundborg found that good management-worker relations
were much more important to work effort than high wages, supervi-
sion, or unemployment. When managers were asked to list the fac-
tors most important to worker motivation, ˜˜they answered that their
employees ought to be given stimulating work assignments, and to
feel involved in decision-making. Some stressed that it was important
that all employees felt noticed and trusted, and provided with con-
tinuous feedback and appreciation™™ (Agell and Lundborg 2003, 25,
16). As the authors note, these answers were very similar to the ones
I heard from U.S. managers.
Managers reported that higher unemployment increased worker
effort, and workers seemed to be providing more effort in 1998 (when
there was high unemployment) than in 1991 (when there was little).
These ¬ndings on the effect of unemployment con¬rm those of myself
and of Kaufman. Like Blinder and Choi, Agell and Lundborg found
little support for Weiss™s idea that job candidates™ reservation wages
are a useful signal of productivity (Agell and Lundborg 1999, table
11.6). Agell and Lundborg also found little support for Solow™s theory
about undercutting. They found (as I also found in my research) that
offers to work for little pay were not uncommon, although fewer such
offers occurred in 1998 than in 1991, perhaps because the much higher
unemployment rate in 1998 discouraged job searching. Managers usu-
Fairness, Reciprocity, and Wage Rigidity 317

ally rejected low offers, because accepting them would create pay
inequities within the ¬rm and low bidders were thought to have poor
skills (Agell and Lundborg 1995, 299). In my survey, I often heard the
¬rst explanation, but seldom the second.
Levine (1993) obtained responses to questionnaires on pay policy
from 139 compensation managers of large American corporations. The
questions focused on the determinants of wages and salaries rather
than on the reasons for downward wage rigidity. Nevertheless, he
found that the unemployment rate and other measures of excess de-
mand for labor had almost no impact on pay. Also, internal equity con-
siderations took precedence over changes in market pay rates in the
determination of relative pay rates for closely related jobs and skills.
In summary, the six surveys”Kaufman (1984), Blinder and Choi
(1990), Campbell and Kamlani (1997), Agell and Lundborg (1995,
1999, 2003), and Levine (1993)”are largely consistent and point to an
explanation of wage rigidity based on morale rather than on the kind
of incentives that play a role in the no shirking model or in Weiss™s
model of adverse selection in hiring. Adverse selection in quits does
seem to be part of the explanation of wage rigidity, however.
I turn next to the analysis by Jennifer Smith (2002) of nine years
of data from the British Household Panel Study of 6,000 employed
workers from 1991 to 1999. She used data on the 70 percent of workers
who did not change employers or job grades over the nine-year period.
The data include monthly income and responses to questions about job
and pay satisfaction. She found that in a typical month, about 28 per-
cent of workers suffered nominal pay cuts (in the sense that their
monthly income declined) and the pay of about 6 percent of workers
was frozen (in that their monthly nominal income did not change).
Smith studied the association between changes in satisfaction and
monthly income and found that workers who suffered cuts were on
average less satis¬ed than those who enjoyed pay increases, although
the difference in satisfaction was not striking. Of those workers whose
income fell, nearly 40 percent were satis¬ed with their pay and nearly
60 percent were satis¬ed with their job.
Smith also found that those workers whose pay was frozen were just
as satis¬ed as those whose income declined. She interpreted this ¬nd-
ing as evidence against the morale theory of wage rigidity outlined
earlier in this chapter, because according to that theory pay cuts should
cause greater unhappiness than do pay freezes. The theory, of course,
may be wrong, but it is not clear what conclusions should be drawn
318 Bewley

from Smith™s analysis, because she probably does not have data on pay
cuts and freezes in the sense of the third de¬nition given in
the previous section, and this is the de¬nition that is relevant to
downward wage rigidity. Monthly incomes can ¬‚uctuate for a great
many reasons”such as changes in overtime, shifts, job assignments,
bonuses, or hours, and Smith has information on none of these vari-
ables except for hours, and she is not sure the data on hours are accu-
rate. Pay raises, freezes, and cuts have to do with the rules by which
pay is calculated.
A great deal more information is required than total monthly income
in order to detect changes in these rules. I ¬nd it extremely unlikely
that an average of 28 percent of the work force suffered pay cuts from
one month to the next according to the proper de¬nition of pay cut.
Another issue is that actual pay cuts often turn out to do little harm to
morale, because they are done for a good reason and are accepted by
workers as fair. When managers say that pay cuts would hurt morale,
they refer to unjusti¬ed cuts. Also, job and pay satisfaction are proba-
bly not good measures of morale. I imagine, nevertheless, that workers
who suffered true pay cuts would be a great deal less satis¬ed than
workers who had received raises.

11.4 Evidence from Experimental Economics

Experimental evidence is accumulating that primarily agrees with
what managers say about their own choices and about worker motiva-
tion. The most important ¬nding is the prevalence of reciprocity. Many
people, when placed experimentally in the role of worker or employer,
give extra effort when offered extra pay or offer extra pay after receiv-
ing extra effort, even when no quid pro quo is required. People also re-
ciprocate bad for bad. In experiments, subjects incur a cost in order to
harm others who have hurt them. The general willingness to recipro-
cate good for good is the essence of good morale. Negative reciprocity
is what underlies the insult effect of pay cuts, which is resentment
caused by the ¬rm™s perceived breach of positive reciprocity; workers
expect employers to offer pay increases, not cuts, in exchange for loy-
alty and effort. The pervasiveness of negative reciprocity probably
explains managers™ belief that the systematic use of ¬ring would not
motivate employees to work well. Another ¬nding is that ¬nancial
incentives do inspire effort, provided they are framed in a way that
avoids any impression of menace. Surveys of the experimental litera-
Fairness, Reciprocity, and Wage Rigidity 319

ture are Fehr and Gachter (1998b, 2000), Fehr and Falk (2002), and
chapter 5 of this volume.
A series of laboratory experiments demonstrate the importance of
reciprocity in mock employment relationships (Fehr, Kirchsteiger, and
Riedl [1993, 1998]; Kirchler, Fehr, and Evans [1996]; Fehr et al. [1998];


. 10
( 13)