EMULATION, INEQUALITY, AND WORK HOURS: WAS
THORSTEN VEBLEN RIGHT?*
Samuel Bowles and Yongjin Park
We investigate the manner in which a desire to emulate the rich inﬂuences individualsÕ allocation
of time between labour and leisure, greater inequality inducing longer work hours as a
result. Data on work hours in ten countries over the period 1963–98 show that greater inequality
is indeed associated longer work hours. These ÔVeblen effectsÕ are large and the estimates are
robust using country ﬁxed effects and other speciﬁcations. Because consumption inequality is a
public bad, a social welfare optimum cannot be implemented by a ﬂat tax on consumption but
may be accomplished by more complicated (progressive) consumption taxes.
At the close of the nineteenth century, Thorsten Veblen proposed what he termed
pecuniary emulation as the foundation of a theory of consumption. Spending, he
maintained, is driven by relative status considerations, that is by the desire to be a
particular type of person as much as by the desire to enjoy the consumer goods per se.
The Joneses, with whom one had to keep up, were not the neighbours but the rich;
their level of living became the never-attainable objective in a consumption arms race
among the less well-to-do. In The Theory of the Leisure Class, he wrote:
The motive is emulation – the stimulus of an invidious comparison ...
especially in any community in which class distinctions are quite vague, all
canons and reputability and decency and all standards of consumption are
traced back by insensible gradations to the usages and thoughts of the
highest social and pecuniary class, the wealthy leisure class. (1934, p. 81).
While valued by some economists as capturing common-sense aspects of consumption
as a form of status seeking, Veblen’s view of social preferences was soon
eclipsed by the simpler and more tractable neoclassical theory of the consumer.
Relegated to the underworld of economics, Veblen’s ideas have nonetheless resonated
over the ensuing years in the writing of Duesenberry (1949), Leibenstein
(1950) and Galbraith (1958) at the middle of the past century and Schor (1998)
and Frank (1997) at the century’s close.
We investigate the importance of Veblen effects in the determination of work hours,
namely the manner in which a desire to emulate the consumption standards of the
rich may inﬂuence an individualÕs allocation of time between labour and leisure.
Veblen effects are derived from a class of social-comparison-based utility functions
on which there is a growing literature and some empirical evidence.1
Clark and
* Thanks to Michael Ash, Giacomo Corneo, James Rebitzer, Juliet Schor, Alois Stutzer and Elisabeth
Wood and two anonymous referees for comments, Bridget Longridge for research assistance, and the MacArthur
Foundation and the Behavioral Sciences Program of the Santa Fe Institute for ﬁnancial support.
1
See Bagwell and Bernheim (1996), Layard (1980), Frey and Stutzer (2002), van Praag (1993), Sen
(1983), Hirsch (1976), Scitovsky (1976) and Easterlin (1974). Frank (1997), Cole et al. (1995) and Clark
and Oswald (1996) provide extensive additional references to the empirical literature. By comparison to
the economic literature, the relevant sociological and social psychological literature is extensive and
venerable: Homans (1961) and Festinger (1957) are inﬂuential contributions.
The Economic Journal, 115 (November), F397–F412. Ó Royal Economic Society 2005. Publishedby Blackwell
Publishing, 9600 Garsington Road, Oxford OX4 2DQ, UK and 350 Main Street, Malden, MA 02148, USA.
[ F397 ]
Oswald (1996) for example found that the satisfaction levels reported by British
workers (in the British Household Panel Survey) vary inversely with the wage levels
of peers. Neumark and Postlewaite (1998), using data from the US NLSY, studied
the labour supply decisions of relatives, ﬁnding some evidence that women whose
sister’s husband had a higher income than their own husband were more likely to
be employed.
These studies provide some support for comparison-based utility functions, but
do not test Veblen effects directly. An explicitly Veblen-inspired study by Schor
(1998) using a US sample asked respondents how their Ôﬁnancial statusÕ compared
to that of those in their reference group (primarily co-workers and friends). While
a majority of her sample responded that they personally did not feel pressure to
Ôkeep up with the JonesesÕ, Schor found that, independently of the effects of
annual and permanent income and other standard regressors, those whose
ﬁnancial status was below their reference group saved signiﬁcantly less than those
who were better off than their reference group. Interestingly, those who watched
TV more saved less, conditional on the other regressors.
Our model of the choice of work hours, presented in the next section, captures
Veblen effects by taking account of the inﬂuence of the consumption of the wellto-do
on the marginal utility of own consumption of the less well-off. The main
result is that work hours are increasing in the degree of income inequality. We
then use data on average annual work hours in ten countries over the period 1963–
98, along with data on inequality of income to explore this hypothesis. Inequality is
a predictor of work hours in both OLS and ﬁxed-effects estimates; its effects are
large, and estimates are robust across a variety of speciﬁcations. This result is
consistent with the hypothesis that social comparisons are upwards to a richer
reference group and is inconsistent with the alternative hypothesis that social
comparisons are downward-looking, people’s consumption and work choices
reﬂecting a desire to distance themselves from a poorer reference group. We then
address an alternative interpretation in which a positive relationship between work
hours and inequality is due to the incentive effects of the latter (Bell and Freeman,
2001). In the penultimate Section we consider some of the normative implications
of Veblen effects, identifying a class of policies which can implement a social
welfare optimum: included are subsidies for the leisure of the rich and a graduated
consumption tax (but not a ﬂat consumption tax).2
1. Veblen Effects on Work Hours
Veblen held that consumption is motivated by a desire for social standing as well as
for the enjoyment of the goods and services per se
the proximate ground for expenditure in excess of what is required for
physical comfort is . . . a desire to live up to the conventional standard of
decency. . .(1934, p. 81)
2
Corneo and Olivier (1997) analyse optimal taxation in Veblen-inspired model of an indivisible
conspicuous consumption good with both snobbish and conformist consumers. As in the model below,
the tax implications of the Veblen effects they model depend on the number of consumers.
F398 [ N O V E M B E RT H E E C O N O M I C J O U R N A L
Ó Royal Economic Society 2005
His key idea (quoted at the outset) was that the best-off members of a community
– Ôthe leisure classÕ – establish the standards for the rest.
But why is it the consumption of the leisure class that is emulated rather than
their leisure? Veblen’s response was that under modern conditions consumption is
a more visible form of display.
The exigencies of the modern industrial system frequently place individuals
and households in juxtaposition between whom there is little contact
in any other sense than juxtaposition. One’s neighbours, mechanically
speaking, often are socially not one’s neighbours, or even acquaintances;
and still their transient good opinion has a high degree of utility. The only
practicable means of impressing one’s pecuniary ability on these unsympathetic
observers of one’s everyday life is an unremitting demonstration
of the ability to pay. (1934, p. 71)
Veblen’s ideas are thus a precursor to the contemporary theory of costly signalling
of otherwise unobservable qualities initiated in economics by Spence
(1973) and in biology by Zahavi (1975).3
The following model embodies the two propositions underlying Veblen’s
account, namely that people compare consumption (or wealth) but not leisure,
and that they refer upwards, choosing their work and spending activities in order
to be more like a higher income group, rather than seeking social distance from
lower income groups. Suppose individuals differ in some trait that inﬂuences
hourly wages and that they choose their hours of work (h) to maximise a utility
function, the arguments of which are leisure (which we normalise as 1 À h) and
what we term effective consumption, cÃ
deﬁned as their own consumption level (c)
minus a constant v (for Veblen) times the consumption level of some higher
income reference group (c$
). The individual’s reference group might be the very
rich, or it might be an intermediate group. The reference group’s rank in the
income distribution is taken as exogenous, as is the Veblen constant v. It may be
convenient to think of each individual as belonging to a homogeneous income
class, each member of which takes the next highest income class as its reference
group (the richest class have no reference group). Together, the reference group
and v measure the nature and intensity of the relevant social comparisons. Individuals
do not save, so c ¼ wh, where w is the wage rate. Thus for some individual
not in the richest group we have
u ¼ uðcÃ
; hÞ
u ¼ u½ðwh À vc$
Þ; h
ð1Þ
where u is increasing and concave in its ﬁrst argument and decreasing and convex
in the second. Leisure and consumption are complements so ucÃ
h < 0. The effect
of increased consumption by members of the reference group thus is both to lower
the utility of the individual and to raise the marginal utility of effective
consumption. The individual will choose hours to be hÃ
, namely that which
3
See the works cited in Gintis et al. (2002).
2005] F399E M U L A T I O N , I N S E R T I O N , A N D W O R K H O U R S
Ó Royal Economic Society 2005
equates the marginal rate of substitution between leisure and effective consumption
to the wage rate.4
We can now consider the effects of an exogenous increase in the wages of the
richest group (raising c$
relative to wh for every income class except the richest).
Differentiating the individual’s ﬁrst order condition for the choice of work hours
(and using the second order condition) we ﬁnd that dhÃ
/dc$
has the sign of
À(ucÃ
cÃ þ ucÃ
h), which is positive. The effect of the larger gap between the consumption
levels of the individual and the reference group is to reduce effective
consumption wh À vc$
, and thus raise the marginal utility of consumption relative
to the marginal utility of leisure, inducing an increase in the hours of work.
The effect on work hours of variations in the Veblen constant have the same sign
(dhÃ
/dv > 0), reﬂecting an increase in the intensity of social comparison and
perhaps capturing the negative effect of TV watching on saving in Schor’s study.
If, contra Veblen, the reference group were the poor (others seeking to distance
themselves from the reference group), then an increase in inequality associated
with a decline in the wages of the poorest group would induce a reduction in work
hours of other higher income groups. Thus are able to test empirically whether the
comparisons are upwards, to a rich reference group as in the Veblen hypothesis, or
downwards to a poorer group from which others are seeking social distance.
For individuals less well off than the reference group, a simple labour supply
function may be derived that is additive in its conventional and Veblen effects. To
see this, normalise the wage of the less well-off to unity and suppose that all (the
rich and the not-so-rich) share the following utility function (an example of (1)
above).
u ¼ lncÃ
À dh: ð2Þ
In the absence of Veblen effects (v ¼ 0, so cÃ
¼ wh) each utility maximising
individual would select h so as to equate the disutility of labour (d) to the marginal
beneﬁt of labour via its contribution to consumption (1/h), thus setting h ¼ 1/d.
With v > 0 the work hours of the rich are unaffected but those not in the
reference group (with superscript n) will now set their work hours at
hn
¼ l=d þ vwr
hr
ð3Þ
where the superscript r refers to the rich reference group. As can be seen from (3)
those not in the reference group work more hours, as we would expect, the second
additional term representing the Veblen effect. An extension of this model with
many income groups each of which (except the richest) takes the next richest
group as its reference group (footnotes 8 and 10) shows that a increase in
consumption by the rich generates a downward cascade of Veblen effects,
increasing work hours throughout the income distribution.
4
If the utility function is Cobb-Douglas in leisure and effective consumption (with a the coefﬁcient of
cÃ
and 1 À a the coefﬁcient of (1 À h)) then the choice of hours is such that
hÃ
=ð1 À hÃ
Þ ¼ a=ð1 À aÞ þ vc$
=wð1 À hÞ
with the increased hours indicated by the second term on the right hand side representing the Veblen
effect (if v ¼ 0, h ¼ a).
F400 [ N O V E M B E RT H E E C O N O M I C J O U R N A L
Ó Royal Economic Society 2005
One aspect of the model deserves comment, namely the assumption that individuals
choose their hours of work. In a collective bargaining framework or an
efﬁciency wage model, employers play a major role in setting work hours, and the
relationship between individual preferences and observed hours may be considerably
attenuated. Not surprisingly, a signiﬁcant fraction of employees in the
advanced economies would prefer hours different from what they have (Bell and
Freeman, 2001). However in the studies reported, a majority preferred current pay
with current hours (rather than more hours and more pay, or less hours and less
pay) and Bell and Freeman report evidence that most European Community
workers would prefer increases in pay (at the current hours) to decreases in hours
(at the current total earnings) suggesting that they are close to the hours they
would have chosen, even if the institutional setting allows no direct relationship
between individual hours choices and outcomes. Bo¨heim and Taylor (2004) report
similar results using the British Household Panel Survey.
This evidence that work hours respond to employee preferences may reﬂect the
fact that employers and unions alike have an interest in taking account of employee
preferences concerning hours of work (to maximise job rents and improve labour
discipline, for example), even if this interest competes with tax and beneﬁts
arrangements which sometimes produce signiﬁcant differences between actual and
desired hours. As a result, individual preferences will affect observed work hours even
in environments in which employees do not literally choose their work hours.
A second comment on the model concerns its behavioural foundations. We do
not suppose that people engage in a conscious optimising process in selecting
their work hours. A more plausible view is that individuals have norms concerning
the appropriate division of their time between family, friends, work, and other
activities, and that these norms differ from group to group and evolve over time.
Suppose this is the case, and that people simply seek to implement their Ôwork
hour normÕ, occasionally updating this norm in response to two kinds of information:
their perceptions of the subjective well-being of others and the hours of
work of others. A plausible model of this learning process would combine payoffbased
updating with conformism: that is, individuals adopt the norms of those in
their social group perceived to be happier, but with a conformist bias towards
adopting norms held by large numbers of their associates, independently of the
associated utility levels (Bowles, 2004). Then the model just presented gives the
payoff-based aspect of the updating of the work hour norm. The main result is that
the work hour norm typical of a given group (other than the richest) is increasing
in the level of inequality but that the short-run Veblen effect might be attenuated
by conformist effects.
2. Work Hours and Inequality
The importance of both social norms and labour market institutions in the
determination of work hours suggests that it may be illuminating to study work
hours averaged over individuals. We use data on average annual hours of work for
ten advanced economies. The annual data for the ten countries presented in
Figure 1 indicate substantial and growing differences between economies. The
2005] F401E M U L A T I O N , I N S E R T I O N , A N D W O R K H O U R S
Ó Royal Economic Society 2005
work year in Germany exceeded that in the US by 231 hours in 1960, and had
fallen to 365 hours less than the US by 1998. Many countries show a decline in
hours prior to the early 1980s followed by a levelling off or increase (in Sweden the
work year fell by 388 hours over the ﬁrst two decades and then increased by
128 hours over the next two decades).
Because the reference group for Veblen effects is the rich, we chose a measure of
income inequality that is sensitive to upper incomes, namely the ratio of the highest
earnings in 90th percentile (that dividing the 90th from the 91st percentile) to the
highest earnings in the 50th percentile. We also present estimates using two alternative
measures of inequality, the Gini coefﬁcient of after-tax incomes from the
Luxemburg Income Study and a Theil index of inter-industry wage differences.
Figure 2 presents the percentile data along with the annual hours, as well as the
country means for these variables. The simple correlation (r ¼ 0.66) is substantial,
but as we will see, it arises in part from covarying inﬂuences on hours and inequality.
We therefore estimate a more complete model.
1300
1400
1500
1600
1700
1800
1900
2000
2100
2200
1960 1965 1970 1975 1980 1985 1990 1995 2000
Averageannualworkhours
Belgium Canada France Italy Netherlands
Norway Sweden United Kingdom United States W Germany
Sweden
Norway
France
Italy
Canada
US
Netherlands
UK
Belgium
Germany
Fig. 1. Work Hours Over Time
Source: OECD Labour Market Statistics Data Set (http://www1.oecd.org/scripts/cde/
members/lfsdataauthenticate.asp)
Average annual hours of work are the number of hours worked on average by persons
for total employment (average employment over the year) Refer to Data Appendix for
detailed deﬁnition.
F402 [ N O V E M B E RT H E E C O N O M I C J O U R N A L
Ó Royal Economic Society 2005
hit
¼ a þ bgit
þ cxit
þ ki
þ dt
þ lit
ð4Þ
where hit
is the natural logarithm of work hours in country i in time t, g is the
measure of inequality, xit
is a vector of other possible exogenous inﬂuences on
hours (with c its vector of estimated coefﬁcients), ki
is a country ﬁxed effect, dt
is a
year ﬁxed effect, and lit
is an error term. The country ﬁxed effects will take
account of cultural and institutional differences and other country-speciﬁc
unobserved inﬂuences on hours. Among the x-variables we considered union
density (to capture possible time-varying institutional differences), real gross
domestic product per capita (to measure possible inﬂuences of income levels on
consumption and leisure preferences) and real manufacturing wages (to capture
conventional labour supply effects). The latter two were expressed in common
units using purchasing power parity conversions. Because hours vary cyclically in
response to labour demand rather than to individual labour supply decisions, we
also include a measure of aggregate unemployment. To account for changes in the
gender composition of the workforce we include the women as a fraction of
employment. We included year ﬁxed effects to capture the possible inﬂuences of
1300
1400
1500
1600
1700
1800
1900
2000
2100
1.30 1.50 1.70 1.90 2.10 2.30 2.50
Earnings inequality (P90/P50 percentile ratio)
Averageannualworkhours
US
Sweden
Netherlands
UK
Belgium
Germany
Canada
France
Norway
Italy
Fig. 2. Earnings Inequality and Average Annual Work Hours
Source: OECD Labour Market Statistics Data Set (http://www1.oecd.org/scripts/cde/
members/lfsdataauthenticate.asp)
Earnings Inequality (percentile ratio) is based on gross earnings of full-time workers and
is the ratio of earnings at 90th percentile to the median earnings (as described in the
text). Refer to Data Appendix for detailed deﬁnition.
Square blocks represent country average of each country and diamond shaped points are
annual data.
2005] F403E M U L A T I O N , I N S E R T I O N , A N D W O R K H O U R S
Ó Royal Economic Society 2005
changes in preferences (or other determinants of work hours) possibly reﬂecting
the diffusion of what Inglehart (1977) terms Ôpost materialist valuesÕ. However,
extensive experimentation with the available measures of Ôpost materialist valuesÕ
did not reveal any systematic results. Finally, we used a measure of government
expenditures relative to gross domestic product. The variable proved insigniﬁcant
while having no appreciable effect on the results reported in table 1.
We treat g as exogenous. A more adequate approach would take g and h to be
jointly determined.5
A plausible exogenous instrument for g proved impossible to
ﬁnd. Thus our results could capture the effect of exogenous shifts in labour supply
on the degree of inequality. To test this possibility we use both contemporaneous
and one year lagged inequality measures on the right-hand side of (4). The results
with lagged inequality measures (presented in Table 2) are virtually identical to
those with contemporaneous regressions. Our view that the endogeneity problem
is not accounting for our results is supported by a companion study of the US
(Park, 2004) in which the labour force participation of wives of full time full year
working men covaries with measures of income inequality among men of similar
age and of the same locality. Because it is unlikely that the labour force decisions of
wives affected income inequality among men (especially during the period
studied, 1969–79) it appears inequality was the cause of increased work hours
rather than the converse.
Our estimates appear in Table 1. Our preferred estimate (I) as well as alternative
estimates using other measures of inequality (II) and (III) indicate signiﬁcant
positive effects of inequality on work hours. Moreover, these effects are large. A
standard deviation change in 90/50 percentile ratio, Gini, and Theil, is associated
with a predicted increase in annual hours of 3.4%, 2.2% and 1.8% respectively.
Taken literally this means that the difference in the US and Swedish percentile
ratio in 1992 accounts for 59% of the difference between the hours of work in the
two countries.
The estimates also suggest a small (and in the preferred estimate, not signiﬁcant)
negative labour supply elasticity consistent with other estimates of labour
supply functions and with the derived labour supply function above (3). The
unemployment rate has the predicted coefﬁcient, as does the female proportion in
employment. In OLS estimates (not shown) Union Density had a large and statistically
signiﬁcant negative coefﬁcient; but in these country ﬁxed-effects equations
its coefﬁcient is small and positive, suggesting that our country ﬁxed effects
may be capturing some of the institutional differences associated with the degree
of unionisation. The speciﬁc country effects across all of the equations indicate
major differences among the countries due to idiosyncratic effects of time
invariant cultural, institutional and other country differences uncorrelated with
the regressors. Sweden and Norway are similar in their short work year while the
5
We have (in reduced forms): g ¼ g(h; z) and h ¼ h(g; k) where z and k are exogenous inﬂuences.
We would like to estimate the partial effect on h of an exogenous shift in g that is hg. What we observe,
however, are intersections of these two functions, inferences from which (unless gh ¼ 0) will over or
underestimate the true effect. If g varies inversely with h – increases in work hours of those in the
middle of the earnings distribution attenuating inequality – we underestimate the true effect, and
conversely.
F404 [ N O V E M B E RT H E E C O N O M I C J O U R N A L
Ó Royal Economic Society 2005
English-speaking countries are distinct and not signiﬁcantly different from one
another in their long work hours; the remainder of the continental countries
occupy a middle ground with Belgium closest to the Nordic pattern. The countryeffect
difference between the English speaking and the Nordic group is about
295 hours per year, indicating large idiosyncratic effects presumably due to
cultural, political, and other differences.
We estimated the same ﬁxed-effects equations as in Table 1, but using as our
dependent variable the natural logarithm of the US Bureau of Labour Statistics
series on average annual hours of manufacturing workers. This series may provide
a more accurate measure of hours (but for a more limited portion of the population.)
The results in Table 2, which cover the same countries and time period,
show that the coefﬁcients of our three inequality measures are highly signiﬁcant,
and of approximately the same magnitudes as those using the OECD labour hours
series. Table 2 also presents the estimated coefﬁcients of one-year lagged
inequality measures. The results with lagged inequality measures are very similar to
or slightly stronger than those with contemporaneous regressions, suggesting that
our results are not driven by the endogenous relation between work hours and
inequality. Lastly, we show the coefﬁcients of inequality measures for a speciﬁcation
without the country ﬁxed effects (but with the year ﬁxed effects). As expected,
the estimates of the Veblen effect are considerably larger but these are likely to be
upward biased because of the co-variation of both hours and inequality with timeinvariant
country-speciﬁc differences, the effects of which are captured in our
ﬁxed-effects estimates.
Table 1
Estimates of the Relationship Between Work Hours and Inequality
I II III IV
Constant 9.635 7.833 10.279 9.878
(16.95) (12.16) (30.18) (16.27)
Percentile earnings ratio 0.177 0.126
(4.81) (2.95)
GINI coefﬁcient (after-tax income) 0.030
(2.22)
Inter-industry earnings inequality 0.023 0.020
(5.74) (2.81)
Ln(real wage) À0.021 À0.041 À0.055 À0.017
(À0.69) (À2.56) (À7.47) (À0.51)
Ln(real GDP per capita) À0.234 À0.065 À0.256 À0.243
(À3.70) (À0.98) (À7.30) (À3.57)
Union density 0.023 0.002 0.002 0.018
(3.60) (0.30) (0.64) (2.65)
Unemployment rate À0.005 À0.008 À0.005 À0.005
(À5.17) (À5.67) (À6.25) (À4.34)
Female proportion in employment À0.094 0.038 À0.070 À0.106
(À3.82) (1.17) (À4.35) (À4.18)
Country and year ﬁxed effects Yes Yes Yes Yes
Observations 155 89 240 143
Adjusted R-squared 0.958 0.979 0.967 0.961
Note: The dependent variable is ln(Average annual work hours). The number of observations
is limited by the fact that our inequality measures are not available for all years.
2005] F405E M U L A T I O N , I N S E R T I O N , A N D W O R K H O U R S
Ó Royal Economic Society 2005
3. Other Explanations
The fact that inequality predicts work hours is consistent with the Veblen effects
proposed at the outset but there are other consistent explanations. Bell and
Freeman (2001) have suggested that inequality induces longer work hours because
those who work longer hours attain a higher percentile rank in the wage distribution
at the workplace and an increase in rank implies greater wage gains the
more unequal is the wage distribution. They provide convincing evidence for this
effect: in the US and Germany wage inequality within detailed occupation/
industry cells is positively correlated with work hours for those working more
35 hours per week and longer.
Discriminating empirically between this incentive-based account and the social
comparisons interpretation offered here may be impossible, and it is very likely
that both incentive and Veblen effects are at work. However, we are not persuaded
that the Bell and Freeman model accounts for the relationship apparent in
Figure 2 and Table 1. First, Bell and Freeman treat long hours as an effective
signal of a difﬁcult to observe quality likely to result in promotion. While this is
true for young lawyers as in the account by Landers et al. (1996), we think it more
likely that hard work when on the job (that is, effort, not hours) is a more common
way to move up. Second, the fact that their inequality-hours relationship is much
weaker (in both the US and Germany) for all workers (rather than just those
working full time or more) is not easy to reconcile with their model. Finally Bell
(1998) found that black workers in the US in 1990 are more responsive to measures
of earnings inequality among blacks only. Bell suggests that this may be
because the black-only distribution is a better indicator of the gains to working
longer hours (but points out that it is not easy to explain why this would be so). A
more parsimonious explanation might be that the relevant reference group for
black workers is other black workers and their response to measures of black-only
inequality is picking up a Veblen effect.
These caveats about the Bell-Freeman interpretation are far from decisive,
however. It would be valuable to see if the evidence for Veblen effects is robust
when using a measure of inequality that could not plausibly be related to the
incentive effects they stress. Two measures accomplish this. First, the previouslymentioned
study (Park, 2004) showing that wivesÕ labour force participation
covaries with male income inequality cannot plausibly be capturing the incentive
Table 2
Alternative Estimates of the Veblen Effect
P90/50 GINI THEIL
Using BLS (manufacturing) hours 0.090 0.042 0.033
(2.47) (2.66) (7.60)
Using one-year lagged inequality 0.170 0.049 0.027
(4.45) (3.50) (5.90)
Without country ﬁxed effect 0.528 1.015 0.066
(9.30) (7.20) (11.36)
Note: The dependent variable is ln(Average annual work hours).
F406 [ N O V E M B E RT H E E C O N O M I C J O U R N A L
Ó Royal Economic Society 2005
effects, unless we have a reason to believe that having a wife with a job has positive
implication on a husband’s promotion.
Second, the most plausible measure of inequality for the incentive effects view
would be within ﬁrm or within industry inequality, of the type Bell and Freeman
used. The reason is that if workers are putting in extra hours to impress their
employer, it is the ﬁrm’s wage structure that is providing the incentive, not the level
of inequality within other ﬁrms, and less still, the difference in average wages between
ﬁrms. (Employers in other ﬁrms have no way of knowing how many hours a worker
puts in.) Thus the Theil index of inter-industry average wage inequality provides
such a test. The fact that this measure of inequality is a signiﬁcant predictor of work
hours (equation III in Table 1) suggests that the Veblen effects model captures
some of the causal mechanisms at work, for this measure could not possibly be
capturing the Bell-Freeman incentive effects. Notice (equation IV) that the estimate
of its coefﬁcient is reduced only marginally by the addition of the percentile ratio to
the equation, suggesting that the estimated effect on work hours in equation III is
not primarily due to the correlation of the Theil index with other measures of
inequality that may be picking up incentive effects modelled by Bell and Freeman.
A second alternative interpretation of the inequality-hours relationship is that
the acceleration of skill-intensive technical change over the last two decades may
have increased inequality and at the same time increased hours of work. Freeman
(2002) for example, found that in the US those using computers or the Internet at
work put in longer hours, and we know from Krueger (1993) that computer use
has raised the economic returns to schooling. Taken together, these two facts
suggest that an exogenous increase in computer use may account for a positive
correlation between hours and earnings inequality. We do not think this accounts
for our results, however, because when we split our time period (at 1983) using the
Theil index (the only measure on which we have sufﬁciently long time series to do
this) we ﬁnd that its estimated coefﬁcient in the early period is almost twice that in
the later period.6
4. Consumption Inequality as a Public Bad
If Veblen effects of the type modelled here are important, there may be a case for
public policies to limit consumption on the conventional grounds that it generates
social costs not accounted in the private calculations of the consumer. Frank
(1997) and others have proposed a tax exemption for savings on just this grounds.7
Veblen effects are an example of this class of consumption externalities, but with
two special characteristics.
First, note that the usual consumption externalities are symmetrical (my consumption
reduces the well being of the JonesÕ I am trying to keep up with, just as
theirs reduces mine). But Veblen effects are asymmetrical: if the JonesÕ are richer
6
Both estimates are smaller than the estimate in Table 1 and are only marginally signiﬁcant, suggesting
that inequality may explain much of the distinct nature of the two periods evident in Figure 1,
while providing a weaker account of the within-period movements.
7
Among others, Boskin and Sheshinski (1978), Ireland (1994) and Oswald (1983) have made similar
proposals.
2005] F407E M U L A T I O N , I N S E R T I O N , A N D W O R K H O U R S
Ó Royal Economic Society 2005
than me, they do not care about my consumption but instead are trying to keep up
with some even richer reference group. Thus Veblen effects cascade downward
through the income distribution, with the richest group inﬂicting subjective costs
on the next group, whose emulation of the consumption of the rich then augments
its own consumption level, thus passing additional subjective costs to the
groups further down.
A second difference is that the inﬂuence of a reference group in the Vebleninspired
model may be substantially independent of its size, so a relatively small
number of well-off but visible consumers may constitute the reference consumption
standard for a much larger number of less well-off individuals. In this case
their consumption decisions may inﬂict subjective costs on large numbers of less
well-off individuals. For both reasons – the asymmetry of the effects and the differing
sizes of various ranks in the income distribution – an appropriate policy
response to Veblen effects may be a progressive consumption tax rather than the
ﬂat consumption tax implied by symmetrical consumption externalities.
To see why this is true take a simple two-class society in which there are a
number (normalised to unity) of well-off individuals indicated (as above) by the
superscript r, and a larger number, n, of less well-off people. As our point is to
clarify the logic of policies to correct Veblen effects rather than to advocate particular
policies, we will retain our simplifying assumptions (including that there is
no saving). We also set the wage of the less well-off at unity. Using the utility
function (2) and the resulting labour supply function (3), suppose that a social
planner wished to know what level of work hours of both groups would maximise
the sum of utilities in this society, x, where
x ¼ lnðhr
wr
Þ À dhr
þ n½lnðhn
À vwr
hr
Þ À dhn
: ð5Þ
The planner would know that in the social optimum the consumption of the
well-off will be less than under private optimisation, and because there are no
savings, the only way to accomplish this is to reduce the work hours of the well-off.
As the work hours of the lower group generate no externalities (they are the
reference group for no one) the planner would simply vary hr
to maximise x, using
(3) to take account of the endogenous response of hn
to the planner’s chosen level
of hr
. While private optimisation induces the rich to equate the marginal
contribution of work to (private) consumption utility (1/hr
) to the (private)
disutility of labour (d), social welfare optimisation requires
1=hrÃ
¼ d þ dnvwr
ð6Þ
where the ﬁrst term on the right is the private cost (disutility of labour)
experienced by the rich and the second is the sum of the marginal social cost
imposed on those attempting to emulate the well-to-do. The aggregate-welfare
maximising level of work hours of the rich is thus given by
hrÃ
¼ 1=dð1 þ nvwr
Þ ð7Þ
which shows that the welfare optimum requires the rich to work less than 1/d by a
proportional amount nvwr
which is equal to the sum of the loss in effective
consumption imposed on the lower income group. The required change in the
F408 [ N O V E M B E RT H E E C O N O M I C J O U R N A L
Ó Royal Economic Society 2005
work hours of the rich is proportional to both the relative size of the two income
groups and to their wage rates.8
As the social optimum requires a change in the
labour-leisure allocations of the higher income reference group but not of the
lower income group, the social planner will not introduce an across the board
consumption tax (applying to both groups). A well-designed policy will target the
consumption of the rich speciﬁcally, as it is this which generates the negative
externalities.
From (6) we see that the implied reduction in the work hours of the rich could
be implemented by policies that enhance their marginal utility of leisure (or what
is equivalent, increasing their marginal disutility of labour) by a proportional
amount dnvwr
.9
Suppose the social planner’s only instrument is a linear tax on the
consumption of the well-off. The particular utility function used here implies that
the tax will not affect the labour hours they perform, so a tax at rate s will reduce
the consumption of the reference group by the same rate. Assuming that the tax
revenues, when spent, yield a per dollar contribution to aggregate welfare of b, the
planner will vary s to maximise:
x ¼ ln½hr
wr
ð1 À sÞ À dhr
þ nfln½hm
À vwr
hr
ð1 À sÞ À dhn
g þ bswr
hr
: ð8Þ
The resulting optimal tax rate sÃ
is given (using (3) and hr
¼ 1/d) by
nvwr
þ bwr
hr
¼ 1=ð1 À sÃ
Þ: ð9Þ
This requires that the tax rate be selected to equate the marginal beneﬁts of
additional taxes (that is the reduced Veblen effects for the less well-off plus the
expenditure beneﬁts, shown on the left-hand side of (9)) to the marginal costs (in
reduced consumption) to the well-off (the right-hand side).
Setting b ¼ 0 so as to abstract from the expenditure related beneﬁts of the tax
policy), and assuming that nvwr
< 1,
sÃ
¼ 1 À 1=nvwr
ð10Þ
As expected, the optimal tax is increasing in the relative size of the less well-off
group, the size of the Veblen effect and the relative wages of the better-off group.10
8
Were there m members of a third (poorer) class for whom the next highest income class is the
reference group, with a wage rate wo
and hours of work ho
, a tedious calculation shows that
hrÃ
½1 þ vwr
ðn þ mv=wo
Þ ¼ 1=d:
A comparison with (7) shows that the optimal work hours of the rich are further reduced by the
consideration of additional poorer classes.
9
This could be accomplished, for example by subsidising the leisure activities of the rich. Under
these conditions the rich would maximise
ur
¼ lnðhr
wr
Þ À dhð1 þ nvwr
Þ
and their private optimisation would give the ﬁrst order condition (6) thus implementing (7).
10
If there exists a third, poorer class, as deﬁned in the previous footnote, and the intermediate class
is taxed at the rate sm
< 1, the optimal tax on the consumption of the rich increases to
sÃ
¼ 1 À fnvwr
½1 þ mvð1 À sm
Þ=nwo
gÀ1
ð100
Þ
to take account of the indirect Veblen effects (via increased work and consumption by the middle
group) on the well-being of the poorest group (the increase in sÃ
varying positively with the relative size
of the poorer class and inversely with its wage).
2005] F409E M U L A T I O N , I N S E R T I O N , A N D W O R K H O U R S
Ó Royal Economic Society 2005
5. Conclusion
We have shown that increased inequality induces people to work longer hours and
have also provided evidence that the underlying cause is the Veblen effect of the
consumption of the rich on the behaviour of those less well off. The effects are
large enough to invite attention from policy makers.
The design of policies to attenuate possible market failures arising from Veblen
effects requires attention to considerations wholly absent above, including
their effects on savings, distributional impacts and political viability (the public
might not favour subsidising wilderness retreats for the well-off, even if, as the
leisure subsidy example requires, they were inconspicuous!) We will not address
these issues here. It is clear, however, that policies designed to discourage consumption
per se (such as the ﬂat consumption tax advocated by many) are not
optimally designed to address Veblen effects. The reason is that where Veblen
effects are important, the social cost imposed by consumption depends on who is
doing it, on the structure of reference groups (who cares about whom) and the
size of the hierarchically ordered reference groups. The consumption of those
who, like the well-to-do, are directly or indirectly reference models for many
would ideally be taxed at a higher rate than the consumption of those who are
models to none or to few.11
Such a policy would be doubly attractive as it as it
would enhance the welfare of the less well-off by limiting the downward cascade
of welfare-reducing Veblen effects while funding valued social projects or allowing
the reduction of other incentive-distorting forms of taxation. As (10) shows,
the richer and smaller is the reference group, the higher is the tax rate that
maximises total social utility.
For well-known reasons, policies that raise average living standards while
favouring the less well-off should be attractive to vote-maximising political parties
and candidates. Speciﬁc taxes on high-end consumption items have occasionally
been advocated and the village of Mamaroneck, New York, even placed a limit
on house size speciﬁcally to curb Veblen effects (Foderano, 2001). But Vebleninspired
policies are a rarity in both academic and policy circles.12
Trends in work
hours have responded to other inﬂuences.
Over a century ago Veblen thought that conspicuous consumption would
increase in importance
11
A government that sought to increase output (rather than maximising the sum of utilities) could
mobilise Veblen effects by shifting the tax burden from the rich to the less well-off, thereby
inducing higher levels of work hours among the latter.
12
This may be due to their seemingly punitive stance towards the well-off. It is easy to see, however,
that one could design a Pareto-improving policy to attenuate negative consumption spillovers. Let the
rich have a utility function that values some form of Ôinconspicuous consumptionÕ that does not stimulate
emulation, in addition to both conspicuous consumption and leisure. Suppose that instead of a
tax on the consumption of the rich, a restriction on their work hours was introduced. Because in the
pre-restriction allocation they selected their work hours to maximise their utility (choosing hr
so that
dur
/dhr
¼ 0), a sufﬁciently small reduction in their work hours would have only second-order effects on
their utility while conferring a cascade of ﬁrst-order beneﬁts on those below them in the social comparison
ranking. Thus there exists some Pareto improving hours restriction on the rich, accompanied
by a transfer of income to the rich from the rest, with the proviso that it must be spent on inconspicuous
consumption.
F410 [ N O V E M B E RT H E E C O N O M I C J O U R N A L
Ó Royal Economic Society 2005
The means of communication and the mobility of the population now
expose the individual to the observation of many persons who have no
other means of judging his reputability than the display of goods... the
present trend of the development is in the direction of heightening the
utility of conspicuous consumption as compared with leisure. (1934,
pp. 71–2)
The description seems almost contemporary, and could well apply across nations,
as large cosmopolitan elements in many populations now take their consumption
standards from the well to do in New York, Milan or Tokyo rather than
their domestic exemplars of style and respectability.
But since he wrote, leisure has not been crowded out by consumption, conspicuous
or otherwise. Indeed in the nations on which data are available, since
1870 work hours have declined substantially, by roughly 50% in continental Europe
and by about a third in the English-speaking nations (Huberman, 2004a,b). It
seems plausible, and consistent with our estimates for a much shorter period, that
among the causes included the sustained increase in per capita income, the
increase in women’s labour force participation, and the very long term decline in
the income share of the top income earners, in many countries extending from
the early part of the twentieth century until well into its ﬁnal quarter.13
Santa Fe Institute
University of Siena
Connecticut College
References
Bagwell, L. S. and Bernheim, D. (1996). ÔVeblen effects in a theory of conspicuous consumptionÕ,
American Economic Review, vol. 86, pp. 349–73.
Bell, L. (1998). ÔDifferences in work hours and hours preferences by race in the U.S.Õ, Review of Social
Economy, vol. 56(4), pp. 481–501.
Bell, L. and Freeman, R. (2001). ÔThe incentive for working hard: explaining hours worked differences
in the US and GermanyÕ, Labour Economics, vol. 8(2), pp. 181–202.
Bo¨heim, R. and Taylor, M. (2004). ÔActual and preferred working hoursÕ, British Journal of Industrial
Relations, vol. 42(1), pp. 149–66.
Boskin, M. and Sheshinski, E. (1978). ÔOptimal redistributive taxation and when individual welfare
depends on relative incomeÕ, Quarterly Journal of Economics, vol. 92, pp. 589–601.
Bowles, S. (2004). Microeconomics: Behavior, Institutions, and Evolution, Princeton: Princeton University
Press.
Clark, A.E. and Oswald, A. (1996). ÔSatisfaction and comparison incomeÕ, Journal of Public Economics,
vol. 61(3), pp. 359–81.
Cole, H., Mailath, G. and Postlewaite, A. (1995). ÔIncorporating concern for relative wealth into
economic modelsÕ, Federal Reserve Bank of Minneapolis Quarterly Review, vol. 19(3), pp. 12–21.
Corneo, G. and Olivier, J. (1997). ÔConspicuous consumption, snobbism, and conformismÕ, Journal of
Public Economics, vol. 66, pp. 55–71.
Dell, F. (2003). ÔTop incomes in Germany over the twentieth century: 1891–1995Õ, CEPREMAP: Paris.
Duesenberry, J. (1949). Income, Saving, and the Theory of Consumer Behavior, Cambridge, MA: Harvard
University Press.
13
See Dell (2003) on Germany and the data on France, UK and the US presented in Piketty and Saez
(2003).
2005] F411E M U L A T I O N , I N S E R T I O N , A N D W O R K H O U R S
Ó Royal Economic Society 2005
Easterlin, R. (1974). ÔDoes economic growth improve the human lot? Some empirical evidenceÕ, in
Nations and Households in Economic Growth: Essays in Honor of Moses Abramovitz, New York: Academic
Press.
Festinger, L. (1957). A Theory of Cognitive Dissonance, Stanford: Stanford University Press.
Foderano, L. (2001). ÔIn suburbs, they’re cracking down on the JonesesÕ, New York Times, March 19.
Frank, R. (1997). ÔThe frame of referance as a public goodÕ, Economic Journal, vol. 107(445),
pp. 1832–47.
Freeman, R. (2002). ÔThe labour market in the new information economyÕ, Oxford Review of Economic
Policy, vol. 18(3), pp. 288–305.
Frey, B. and Stutzer, A. (2002). Happiness & Economics: How the Economy and Institutions Affect Human-WellBeing,
Princeton: Princeton University Press.
Galbraith, J. K. (1958). The Afﬂuent Society, Boston: Houghton Mifﬂin Company.
Gintis, H., Eric, S. and Bowles, S. (2002). ÔCostly signaling and cooperationÕ, Journal of Theoretical Biology,
vol. 213(1), pp. 103–19.
Hirsch, F. (1976). Social Limits to Growth, Cambridge, MA: Harvard University Press.
Homans, G. (1961). Social Behavior: Its Elementary Forms, New York: Harcourt Brace.
Huberman, M. (2004a). ÔWorkdays, workweeks, and annual hours, new international estimates 1870–
2000Õ, mimeo, Universite de Montreal.
Huberman, M. (2004b). ÔWorking hours of the world unite: new international evidence of worktime,
1870–1913Õ, Journal of Economic History, vol. 64, pp. 964–1001.
Inglehart, R. (1977). The Silent Revolution: Changing Values and Political Styles Among Western Publics,
Princeton: Princeton University Press.
Ireland, N. (1994). ÔOn limiting the market for status signalsÕ, Journal of Public Economics, vol. 53(1),
pp. 91–110.
Krueger, A. (1993). ÔHow computers have changed the wage structureÕ, Quarterly Journal of Economics,
vol. 108(1), pp. 33–60.
Landers, R., Rebitzer, J. and Taylor, L. (1996). ÔRat race redux: adverse selection in the determination of
work hours in law ﬁrmsÕ, American Economic Review, vol. 86(3), pp. 329–48.
Layard, R. (1980). ÔHuman satisfactions and public policyÕ, Economic Journal, vol. 90(360), pp. 737–
50.
Leibenstein, H. (1950). ÔBandwagon, snob, and Veblen effects in the theory of consumersÕ demandÕ,
Quarterly Journal of Economics, vol. 64(2), pp. 183–207.
Neumark, D and Postlewaite, A. (1998). ÔRelative income concerns and the rise in married women’s
employmentÕ, Journal of Public Economics, vol. 70, pp. 157–83.
Oswald, A. (1983). ÔAltruism, jealousy, and the theory of optimal nonlinear income taxationÕ, Journal of
Public Economics, vol. 20, pp. 77–87.
Park, Y. (2004). ÔVeblen effects on labour supply: male earnings inequality increases women’s labour
force participationÕ, Department of Economics, Connecticut College.
Piketty, T. and Saez, E. (2003). ÔIncome inequality in the United States: 1913–1998Õ, Quarterly Journal of
Economics, vol. 118, pp. 1–39.
Schor, J. (1998). The Overspent American: Upscaling, Downshifting, and the New Consumer, New York: Basic
Books.
Scitovsky, T. (1976). The Joyless Economy: An Inquiry into Human Satisfaction and Consumer Dissatisfaction,
New York: Oxford University Press.
Sen, A. (1983). ÔPoor, relatively speakingÕ, Oxford Economic Papers, vol. 35, pp. 153–67.
Spence, M. (1973). ÔJob market signalingÕ, Quarterly Journal of Economics, vol. 90, pp. 225–43.
van Praag, B. (1993). ÔThe relativity of the welfare conceptÕ, in (M. Nussbaum and A. Sen, eds.), The
Quality of Life, pp. 363–92, Oxford: Clarendon Press.
Veblen, T. (1934). The Theory of the Leisure Class, New York: Modern Library.
Zahavi,A. (1975). ÔMateselection–a selectionfor handicapÕ, Journalof TheoreticalBiology, vol. 53, pp. 205–14.
Ó Royal Economic Society 2005
F412 [ N O V E M B E R 2005]T H E E C O N O M I C J O U R N A L