Inequality Aversion,Efficiency,and Maximin Preferences in Simple Distribution Experiments:Comment
By E RNST F EHR,M ICHAEL N AEF,AND K LAUS M.S CHMIDT*
A major puzzle in the literature on social preferences is that a large majority of subjects behaves as if completely lf-interested in some circumstances,such as in competitive experi-mental markets with standardized goods or in thefinal rounds of public goods experiments; while in other circumstances a large majority behaves as if strongly motivated by fairness ,in competitive markets with in-complete contracts or in public goods experi-ments with punishment opportunities.Recently developed models of inequity aversion(Fehr and Schmidt,1999;Gary E Bolton and Axel Ockenfels,2000)can explain this puzzle by assuming a heterogeneous population of lfish and inequity-aver subjects.Dirk Engelmann and Martin Strobel(2004,henceforth E&S) question the relevance of inequity aversion in simple dictator game experiments,claiming that a combination of a preference for efficiency1 and a Rawlsian motive for helping the least well-off is more important than inequity aversion.
In this paper,we show that E&S overstate the relevance of efficiency motives and understate the relevance of inequity aversion.The par-ticipants of the E&S experiments were under-graduate students
of economics and business administration.The subjects lf-lected into theirfield of study(economics)and learned in thefirst mester that efficiency is desirable. Noneconomists,however,may value efficiency much less than economists do.We replicated the most relevant E&S experiments with vari-ous subject pools and are able to show that the dominance of the efficiency motive over the equity motive is restricted to students of eco-nomics and business administration.Students from various other disciplines and a sample of nonacademic employees value equality much more highly than efficiency.
This rais the question whether there are other subject characteristics such as gender or political attitudes that affect the preferences for efficiency versus equality.Wefind a nonnegli-gible gender effect,indicating that women are more egalitarian than men.Perhaps surpris-ingly,however,the dominance of equality over efficiency is unrelated to political , subjects who vote for right-wing parties and favor right-wing political attitudes are as likely to favor equality as are subjects with left-wing attitudes.
I.Economists versus Noneconomists
The most interesting results of E&S concern their treatments Ey and P that are designed to discrimin
ate between preferences for efficiency, the Rawlsian maximin motive,and inequity aversion(e Table1below).In both treat-ments,person2is the decision maker(dictator) who can choo between allocations A,B,and C.Note that own choices never affect person 2’s payoff.Person2can redistribute income from a richer person1to a poorer person3in
*Fehr:Institute for Empirical Rearch in Economics, University of Zurich,Bluemlisalpstras10,CH-8006 Zurich,Switzerland(e-mail:efehr@iew.unizh.ch);Naef: Institute for Empirical Rearch in Economics,Univer-
sity of Zurich,Bluemlisalpstras10,CH-8006Zurich, Switzerland(e-mail:naef@iew.unizh.ch);Schmidt:De-partment of Economics,University of Munich,Ludwig-stras28,D-80539Muenchen,Germany(e-mail: klaus.schmidt@Lrz.uni-muenchen.de).A previous ver-sion of this paper was circulated under the title“The Role
of Equality,Efficiency,and Rawlsian Motives in Social Preferences.”We would like to thank Susanne Krem-helmer for excellent rearch assistance and Mathias Erlei for very helpful comments on an earlier version of
this paper.This paper is part of the rearch priority program at the University of Zurich on the“Found
ations
of Human Social Behavior:Altruism and Egoism.”Part
of this rearch was conducted while the third author visited Yale University;he would like to thank the Eco-nomics Department for its great hospitality.Financial support by the EU-Rearch Network ENABLE(MRTM-
CT-2003-505223)is gratefully acknowledged.Schmidt also gratefully acknowledgesfinancial support by the German Science Foundation through SFB-TR15.
1“Efficiency”in Engelmann and Strobel(2004)is not
defined as Pareto efficiency but as surplus maximization.
1912
both treatments by choosing allocations B or C instead of allocation A.Choices B and C in-volve a relatively high efficiency loss,however.In treatment Ey,every additional money unit that is given to the poor person reduces the rich person’s income by four money units,while the rich person suffers an in
come reduction of three units if the poor person’s income is incread by one unit in treatment P.In both treatments,a choice of the efficient allocation A constitutes evidence not only against specific functional form assumptions (like piece-wi linearity or positional asymmetry,meaning that subjects prefer advantageous inequity to disadvanta-geous inequity)in the Fehr-Schmidt approach,but also against general nonlinear versions of inequity aversion.2Treatment P is particularly important becau the decision maker (person 2)always earns the lowest income,independent
of which allocation is implemented.Thus,the decision cannot be affected by Rawlsian pref-erences.Therefore,treatment P constitutes a clean test of the relevance of inequity aversion in comparison to the efficiency motive.
The E&S evidence for treatment P also illus-trates an important point regarding the interpre-tation of their results.E&S conclude that “inequality aversion does not em to be a ma-jor part in a complete explanation”of their data.This conclusion is bad on their logit regres-sions,which neglect any individual heterogene-ity.3One-third of their subjects choo the most
2
The inequity aversion approach by Fehr and Schmidt (1999)is compatible with the evidence in ver
al other games conducted by E&S (the “envy”games)if one allows for nonlinear forms of inequity aversion and gives up the assumption of positional asymmetry.
3
The conditional logit analysis in E&S is problematic for two reasons.First,E&S can estimate the preferences only of an “average subject,”which neglects any individual heterogeneity.Second,the independent variables Eff (which captures the sum of the subjects payoff),Self (which cap-tures the decision maker’s own payoff),FS ␣(which cap-tures the disadvantageous inequality),and FS (which captures advantageous inequality)are linearly dependent.Thus,they have to exclude one or more of the variables from the analysis.The decision which variable one should exclude is,however,completely arbitrary,but has strong effects on the regression outcomes.If,as E&S do,Self is
T ABLE 1—S UBJECT P OOL E FFECTS :E CONOMISTS
VERSUS
N ONECONOMISTS
Allocation
Treatment Ey
Treatment P
A B C A B C Person 1payoff 21171314118Person 2payoff 999444Person 3payoff 345567Total payoff
333027232119Average payoff of 1and 31210.5
99.58.57.5Efficiency prediction
A
A Inequity aversion prediction C C Rawlsian maximin prediction
C
A or
B or
C (A)Economists:Berlin (E&S,2004)
Choices (absolute)1271118210Choices (percent)40.023.336.760.0 6.733.3(B)Economists:Munich
Choices (absolute)721225631630Choices (percent)
66.111.022.957.814.727.5(C)Noneconomists:Munich
Choices (absolute)221348211745Choices (percent)
26.5
15.7
57.8
25.320.554.2(D)Noneconomists:Zurich
Choices (absolute)8820Choices (percent)22.2
22.255.6(E)Economists:Zurich
Choices (absolute)3191831918Choices (percent)
53.515.531.053.515.531.0(F)Noneconomists:Zurich
Choices (absolute)612378532584Choices (percent)
37.7
14.2
48.1
32.7
15.4
51.9
1913
VOL.96NO.5FEHR ET AL.:SIMPLE DISTRIBUTION EXPERIMENTS:COMMENT
inefficient and most egalitarian allocation in treatment P,however,suggesting that they are motivated by inequality aversion(e Table1). Perhaps the most important message that comes from social preferences models is that in strate-gic interactions the heterogeneity of social pref-erences is extremely important.As shown in Fehr and Schmidt(1999),even a minority of inequality-aver subjects may have powerful effects on the outcome of strategic interactions. Thus,even if it were true that in general only one-third of the population is motivated by in-equality aversion,this motive can have impor-tant effects.As we will e below,however, among noneconomists,much more than one-third of the subjects em to be motivated by inequality aversion.
In afirst wave of paid experiment,we exam-ined the subject pool hypothesis by conducting the Ey and P treatments of E&S with two dif-ferent subject pools from the University of Mu-nich.4Thefirst subject pool consisted of109first-year undergraduate students in economics and business administration(henceforth called “economists”),while the cond subject pool was made up of83first-and cond-year un-dergraduates from other disciplines,mostly the social sciences(henceforth called“nonecono-mists”).Each subject had to make a decision in both of the distribution games discusd above. The results of the experiments are reported in Table1.For convenience,we also show the results from the E&S experiments in panel A of this table.
The subject pool effects displayed in Table1 are striking.The Munich economics and busi-ness administration students corroborate the main E&S result,confirming that preferences for efficiency play a major role among econo-mists(compare panels A and B of Table1);
66.1%of the economists opted for the efficient but most inegalitarian allocation A in treatment Ey,even exceeding the40percent in the E&S study.The efficiency advantage of allocation A is somewhat lower in game P and the fraction of economists opting for the efficient allocation A decreas slightly to57.8percent,very similar to the E&S results.The behavior of nonecono-mists from the University of Munich contrasts sharply with the results,however(e panel C of Table1).In games Ey and P,the nonecono-mists cho the inefficient but most egalitarian allocation C at the rate of57.8percent and54.2 percent,respectively,while only25to27per-cent opted for the efficient allocation A.The differences between the noneconomists and the economists from the University of Munich are statistically highly significant(pϽ0.001in each treatment,Fisher exact test).
An additional paid experiment with non-economists(college students)in Zurich,Swit-zerland,who participated only in treatment P, further confirms the robustness of the subject pool effect.A comparison of panels C and D of Table1shows that the students’choices were almost identical to tho of the noneconomists from the University of Munich(pϾ0.93, Fisher exact test).Only22percent
choo the efficient allocation A,while55.6percent choo the allocation predicted by inequity aversion.The choices of the noneconomists in Munich and Zu-rich also differ significantly from the choices of the E&S subjects(pϽ0.001,Fisher exact test), further supporting a strong subject pool effect. II.The Impact of Political Attitudes on Social
Preferences
The strong difference in behavior between economists and noneconomists rais the ques-tion of whether there are other subject pool characteristics that might affect the results.Sub-jects’social preferences could be related to gen-der,age,or their political attitudes.One might conjecture that more right-wing subjects are less in favor of equality and prefer more efficient allocations than left-wing subjects.If this were the ca,or if other subject pool characteristics had a significant impact,we would have to worry about whether our subject pool is repre-ntative with respect to the characteristics. To address this concern,we conducted addi-tional experiments where—after subjects had made their choices in treatments Ey and P—we
excluded,FS␣and FSare not significant.If,however,FS
is excluded,Self and Eff are not significant.Thus,if one
takes their regression approach riously,one could equally
well conclude that concerns for efficiency are irrelevant.To
make things wor,if one arbitrarily excludes Eff from the
analysis,Self and FS␣suddenly become highly significant.
We owe this argument to Mathias Erlei who respecified and
reestimated the E&S regressions.
4In all experiments reported in this paper,subjects made
their decisions anonymously,they were paid in private,and
there was no role ,the decision makers knew
that they were in the role of person2.
1914THE AMERICAN ECONOMIC REVIEW DECEMBER2006
collected information about their political atti-tudes and how they voted in the last general election.We also collected information about their age,gender,and their membership in or-ganizations (such as sports clubs or local char-ities).We recruited 58third-mester students of economics or business administration from the University of Zurich and 100third-mester students from other disciplines (law and medi-cine)and an additional 62noneconomists from outside the university.The last subjects were nonmanagement employees of banks and other financial institutions.They had no college edu-cation,but all of them had completed an ap-prenticeship for their current job.
The subjects’political preferences were elic-ited in two ways.First,subjects ranked them-lves on a scale from 0to 10,where 0indicates the most extreme left-wing position and 10in-dicates the most extreme right-wing position.This lf-report measure of political attitudes is now widely ud in reprentative surveys in
Switzerland.In addition,we asked subjects how they voted in the last national election.Both economists and noneconomists participated in the elections at a rate of approximately 70percent.For the subjects,we computed a Spearman rank correlation between their lf-reported political attitudes on the 0–10scale and their votes for left,center,and right par-ties.5This correlation i
s 0.78(p Ͻ0.001),in-dicating that subjects’ranking on the left-right scale and their actual voting behavior are consistent.
The results of our new experiments are displayed in panels E and F of Table 1and in Table 2.We first tested for differences between
5
In the calculations,we treat the social democrats and the green party as left-wing parties,the liberal party (FDP),the Christian democrats (CVP),and the Evangelical peo-ple’s party (EVP)as center parties,and the Swiss people’s party (SVP)as a right-wing party.This classification is also ud by political scientists.
T ABLE 2—I MPACT
OF
S UBJECT P OOL ,P OLITICAL A TTITUDE ,
AND
G ENDER
Independent variables Dependent variable:Choices in the distribution game (0ϭallocation A,1ϭallocation B,2ϭallocation C)
New data
(panel E and F in Table 1)
All data
(panels B to F in Table 1)
(1)(2)(3)(4)(5)(6)Ey P Ey&P Ey P Ey&P Dummy for economist (1ϭeconomist)Ϫ0.159**Ϫ0.207***Ϫ0.183***Ϫ0.258***Ϫ0.242***Ϫ0.250***(0.070)(0.069)(0.060)(0.043)(0.043)(0.037)Gender dummy (1ϭwomen)0.0670.125*0.095*0.091*0.086*0.088**(0.066)(0.066)(0.058)(0.048)
(0.045)
(0.039)
Political attitude
(0ϭleft,10ϭright)Ϫ0.015Ϫ0.003Ϫ0.009(0.017)(0.017)(0.015)Age
Ϫ0.002Ϫ0.002Ϫ0.002(0.006)(0.006)(0.006)Dummy for membership in organizations
0.0730.1050.089(0.068)(0.069)(0.059)Dummy for Ey-treatment Ϫ0.034Ϫ0.040(0.034)(0.025)Pr (C)if a noneconomist 0.4820.5210.5010.5130.5300.521Number of obrvations 216216432407443850Cluster per subjects no no yes no no yes Prob Ͼchi20.1030.0080.0060.0000.0000.000Pudo R 2
0.021
0.036
0.028
0.0470.039
0.044
Notes:The table reports the marginal effects of the different variables on choosing the egalitarian allocation C.Numbers in parenthes denote the standard error of the marginal effects.The marginal effects are evaluated at the point where the dummy for economists is t at zero.The estimated baline probability of choosing allocation C for noneconomists is given in the row “Pr(C)if a noneconomist.”
*Indicates the coefficient is statistically different from zero above the 0.1significance level.**Indicates the coefficient is statistically different from zero above the 0.05significance level.***Indicates the coefficient is statistically different from zero above the 0.01significance level.
1915
VOL.96NO.5FEHR ET AL.:SIMPLE DISTRIBUTION EXPERIMENTS:COMMENT
students from noneconomic disciplines and em-ployees.It turns out that the behavior of the two groups is very similar,which is confirmed by statistical tests(pϭ0.775in the Ey treat-ment,pϭ0.739in the P treatment,Fisher exact test).Therefore,we pooled their data in panel F of Table1.The table shows that both in treat-ment Ey and P,53.5percent of the economists prefer the efficient allocation A,whereas only 31percent preferred the egalitarian allocation C (e panel E of the table).This patter
n is re-verd,however,among the noneconomists (panel F of Table1).A majority of nonecono-mists(51.9percent)prefers the most egalitarian allocation C and only32.7percent choo the efficient allocation A in treatment P.Similarly, a relative majority of the noneconomists(48.1 percent)prefers allocation C in treatment Ey and only37.7percent choo the efficient allo-cation A.
The differences between economists and noneconomists in lf-reported political atti-tudes are small.On average,economists score 5.28on the left-right scale,whereas the non-economists are somewhat more left wing and score4.95.This difference is statistically insig-nificant(pϭ0.41,Mann Whitney test).There is,however,strong individual variation in both groups.It is therefore interesting to examine whether individual differences in political atti-tude can explain the individual variation in so-cial preferences across subjects.
In Table2,we report the marginal effects of ordered probit regressions where the choice of the most unequal allocation A is reprented with0,allocation B with1,and the egalitarian allocation C with2.As explan-atory variables,we included a dummy vari-able for economists(1for economist),a gender dummy(1for women),the political attitude on the left-right scale,age,and a dummy for whether the subject is a member in an organization or club.Regression(1)is bad on data from the Ey treatment and regression(2)us the data from the P treat-ment.We pool the data from both treatments in regre
ssion(3)and control for the potential dependence of subjects’decisions across treatments by clustering on subjects.We also control for a treatment effect in regression(3) by including a dummy for the Ey treatment. The most important fact stemming from the first three regressions reported in Table2is that the marginal effect for the economists’dummy is negative and highly significant,even after controlling for political attitudes.In fact,while political attitude has virtually no effect on social preferences,regression(3)shows economists have an18-percentage-point lower probability of choosing the egalitarian allocation C.Age, membership in organizations,and the Ey-dummy have no significant effects.The gender variable is weakly significant,however,and indi-cates that women are somewhat more egalitarian. If we average over both ,take re-gression3),women are roughly10percent more likely to choo the egalitarian allocation C.
We pooled all data from our experiments in regressions4to6to estimate the overall impact of the economists’dummy and a gender dummy on social preferences.6The results of the regressions indicate that economists are25 percentage points less likely to choo the egalitarian allocation C,while women are more likely to choo C by9percentage points.Note, also,that if we pool the data from both treat-ments(regression6)the gender dummy is sig-nificant at the5-percent level.
III.Conclusions
Our experiments indicate that there is a strong subject pool effect in the simple distri-bution games of Engelmann and Strobel(2004). While a majority of economists prefer effi-ciency over equity,various groups of nonecono-mists,ranging from students of various other disciplines to low-level employees of banks and financial institutions,show the opposite pattern. On average,more than50percent of the non-economists prefer the most egalitarian(and least efficient)distribution,while the probability of an economist choosing this allocation is25per-centage points lower.Thus,we conclude that preferences for equity are of major importance in simple distribution experiments,even though there is a significant minority of subjects who em to be concerned about efficiency.We have also shown that subjects’political preferences do not affect their social preferences for effi-6We took all data reported in panels B to F in Table1. Since we do not have values for age,political attitude,and membership in organizations in the experiments reported in panels B to D,we cannot u the variables in the regres-sions4to6.
1916THE AMERICAN ECONOMIC REVIEW DECEMBER2006
ciency and equity.Subjects with a right-wing political attitude are as likely to choo the egalitarian allocation as left-wing subjects. Women,however,favor the egalitarian alloca-tion more often than men do.
The relative importance of the different moti-vational forces ems to differ not only across subject pools but also across games.The E&S experiments all involve dictator games without strategic interaction.Efficiency concerns are an important motivational force for a significant frac-tion of the subjects in the nonstrategic distribu-tion games,but they em to be far less important in strategic games such as bargaining,trust,or public good games.The ultimatum game is a striking ca in point.Efficiency requires that the responder accept any positive offer.There is a huge amount of experimental evidence,however, showing that low offers are frequently rejected. Inequity aversion does a much better job of pre-dicting behavior in the latter games,and it is consistent with the choices of a large fraction of the subjects in the E&S distribution games.
A better understanding of the functioning of different motivational forces in different en-vironments is an important question for future rearch.
REFERENCES
Bolton,Gary E,and Axel Ockenfels.2000.“ERC: A Theory of Equity,Reciprocity,and Com-petition.”American Economic Review,90(1): 166–93.
Dirk,and Martin Strobel.2004.“In-equality Aversion,Efficiency,and Maximin Preferences in Simple Dist
ribution Experi-ments.”American Economic Review,94(4): 857–69.
Ernst,and Klaus M.Schmidt.1999.“A Theory of Fairness,Competition,and Coop-eration.”Quarterly Journal of Economics, 114(3):817–68.
1917
VOL.96NO.5FEHR ET AL.:SIMPLE DISTRIBUTION EXPERIMENTS:COMMENT
