Ben-Ner, Culture, identity

8/7/2019 Ben-Ner, Culture, identity

1/18

Journal of Economic Behavior & Organization 72 (2009) 153170

Contents lists available at ScienceDirect

Journal of Economic Behavior & Organization

j o u r n a l h o m e p a g e : w w w . e l s e v i e r . c o m / l o c a t e / j e b o

Identity and in-group/out-group differentiation in work and

giving behaviors: Experimental evidence

Avner Ben-Ner a,, Brian P. McCall b, Massoud Stephanec, Hua Wang d

a Carlson School of Management, University of Minnesota, 321-19th Avenue South, Minneapolis, MN 55455, United Statesb School of Education, 610 East University Avenue, Ann Arbor, MI 48109, United Statesc Brain Sciences Center, VA Medical Center (11B), One Veterans Drive, Minneapolis, MN 5541, United Statesd Schulich School of Business, York University, 4700 Keele Street, Toronto, Ontario M3J 1P3, Canada

a r t i c l e i n f o

Article history:

Received 4 September 2006

Received in revised form 30 April 2009

Accepted 1 May 2009

Available online 18 May 2009

JEL classification:

C91

D8

J16

Keywords:

Identity

Experiments

Self-other differentiation

Cooperation

In-group and out-group

a b s t r a c t

We investigate the existence and relative strength of favoritism for in-group versus out-

group along multiple identity categories (body type, political views, nationality, religion,

and more) in four alternative contexts: (1) giving money in a dictator game, (2) sharing an

office, (3) commuting, and (4) work. We carried out two studies. The first study entailed

hypothetical situations and imaginary people; the second study was similar to the first,

but the dictator game component was incentivized (actual money) and involved actual

receivers. Our subjects behavior towards others is significantly affected by their respective

identities. (1) Those that belong to the in-group are treated more favorably than those who

belong to the out-group in nearly all identity categories and in all contexts. (2) Family and

kinship are the mostpowerfulsource of differentiation,followed by politicalviews, religion,

sports-team loyalty, and music preferences, with gender being basically insignificant. (3)Thehierarchy of identity categories is fairly stableacross thefour contexts. (4) Subjects give

similar amounts and discriminate between in-group and out-group to similar degrees in

the hypothetical and incentivized dictator games.

2009 Elsevier B.V. All rights reserved.

1. Introduction

Identityisapersonssenseofself(Akerlof andKranton,2000); itis theconceptthatindividuals come torealizewhen they

answer the elemental question of whoam I? The answer,typically, includes multiplecategories or attributes such as gender,

facial features, and height, as well as religion, ethnicity, social-group affiliation, sports-team loyalty, family, profession, artistic

preferences, culinary preferences, and place of origin. These attributes represent how a person views himself or herself, and

are likely to have different weights to the sense of self.

Identity is often the source of positive and desirable outcomes, such as the warm feeling of amity and affiliation, con-

structive and cooperative behavior in the context of social, ethnic, and religious organizations, as well as desirable diversity

and variety (e.g., Eckel and Grossman, 2005; Page, 2007) However, identity is also the basis for discrimination and hatred,

exclusion, enmity, sports riots, national and religious wars, ethnic cleansing and extermination, distrust and conflict (e.g.,

Costa and Kahn, 2003; Putnam, 2007).

Corresponding author. Tel.: +1 612 624 0867.

E-mail address: [email protected] (A. Ben-Ner).

0167-2681/$ see front matter 2009 Elsevier B.V. All rights reserved.doi:10.1016/j.jebo.2009.05.007
http://www.sciencedirect.com/science/journal/01672681http://www.elsevier.com/locate/jebomailto:[email protected]://dx.doi.org/10.1016/j.jebo.2009.05.007http://dx.doi.org/10.1016/j.jebo.2009.05.007mailto:[email protected]://www.elsevier.com/locate/jebohttp://www.sciencedirect.com/science/journal/01672681


2/18

154 A. Ben-Ner et al. / Journal of Economic Behavior & Organization 72 (2009) 153170

Generally, people act more favorably towards persons who share with them an important attribute of their identity

compared to persons who differ significantly on that attribute. For example, fans of the same sports team give each other

high-fives but jeer fans of a rival team; enthusiasts of certain musical groups may work more readily with those who share

their preferences than with others; and members of some religious groups sacrifice their own lives but take the lives of

members of other groups to advance their groups cause. Even arbitrary assignment of identity in the context of a psychology

experiment can elicit partisan behavior (Tajfel and Turner, 1979; see also examples in Akerlof and Kranton, 2000, 2005).

The difference in how someone treats a person of the same identity as compared to a person of a different identity is

likely to depend on several factors: the identity attribute in question, the circumstances of the interaction between subject

and object, as well as the subjects individual characteristics. There is a large and expanding body of literature on identity

in several disciplines.1 However, many questions with regard to how different identity attributes affect behavior towards

others remain unaddressed in the literature. For example, does religion evoke more passion than ethnicity or than sports?

Are all differences in identity fertile grounds for discrimination? Do differences affect equally various social and economic

behaviors? We address these questions in this paper.

An understanding of the role identity plays in the context of various interactions is important for both economic theory

and policy, as Akerlof and Kranton (2000, 2002, 2005) illustrate. In markets where the identity of the transacting parties

is known, the same good or service may have different prices, depending on the degree of similarity in the identities of

the parties. Employees that identify with their organization or team require fewer and different incentives to exercise high

levels of effort than other employees (Akerlof and Kranton, 2005; Eckel and Grossman, 2005). But are identity effects of

significant magnitude in an economic sense? Are identity effects widespread or are they restricted to a few identity cat-

egories only? Are identity effects relevant to diverse contexts? These questions have received only scattered theoretical

and empirical attention, and our paper is the first to address them collectively within a consistent and comprehensive

framework.

In this paper, we outline three complementary theoretical perspectives on the role of identity in interactions between

individuals: inclusive fitness theory, evolutionary theory, and social identity theory. The threeperspectives suggest that iden-

tity and the distinction between in-group and out-group are important, although they have somewhat different implications

regarding the relative importance of different categories.

The central contribution of the paper consists of an empirical examination of the extent to which attitudes and behaviors

of individuals towards in-group members differ from those towards out-group members. We study this question relative

to multiple identity categories, from gender, body type and culinary preferences to religion, nationality and political views.

We evaluate the relative importance of these categories in the context of giving in a dictator game, willingness to work

with another person on a project that is critical to ones career advancement and other situations. We do so by carrying out

two studies. In Study I, we asked 222 subjects how willing they were to give money (out of a $10 endowment) to another

person, work with another person on a critical project, commute with another person, and share an office with another

person. Subjects were asked to consider separately dozens of other persons, each described by a single attribute. In this

study subjects were surveyed about imaginary persons and in the dictator game we used hypothetical money. We were thus

able to present 91 alternative persons, most of whom could not be found in a commonly available subject pool, as well as

keep the cost of the study at reasonable levels.

In order to validate the differences in the giving behavior towards in-group versus out-group despite reliance on hypo-

thetical money and hypothetical persons, we carried out a second study. In Study II, we asked 37 subjects to participate as

senders in eight dictator games; the subjects were also asked to indicate their willingness to work on a project critical to

their career advancement, commute and share an office with each of eight individuals.

The two studies suggest that attitudes and behaviors individuals exhibit towards others are affected strongly by the

similarity of the identity of the two parties. (1) Those that belong to the in-group are treated more favorably than those

who belong to the out-group in nearly all identity categories and in all contexts.2 (2) Family and kinship are the most

powerful source of differentiation identity in our sample, followed by political views, religion, sports-team loyalty, and

music preference, with gender being basically insignificant. (3) The hierarchy of identity categories is fairly stable across the

four contexts, although some identity categories are substantially more important in some contexts than in others (notably,

family is most importantin theworkcontext). (4)Subjectsfavor anddiscriminate othersto similar degrees in thehypothetical

and incentivized dictator games.

Our subjects represent a fairly homogenous sample of young men and women who have very little experience with strife

associated with religious, national, or ethnic identities, the kind of conflicts that fuel much of the most visible identity-

based behaviors. Such a sample is likely to inform about the presence or absence of deep-seated, perhaps hard-wired,

sentiments about the differentiation between in-group and out-group people, and behaviors driven by such sentiments,

possibly mixed with culturally-transmitted values regarding such differentiation, but with only limited contribution from

direct life experiences.

1 See the review article by Ellemers et al. (2002), and literature reviews in Akerlof and Kranton (2000) and Leonard and Levine (2006).2

Our results arelikelyto representan underestimateof the degree of differentiationbetweenin-groupand out-groupin our sample. Although confidentialand anonymous, there is still the possibility that some subjects did not express fully their discriminatory attitudes.


3/18

A. Ben-Ner et al. / Journal of Economic Behavior & Organization 72 (2009) 153170 155

The paper is organized as follows. Section 2 explores the concept of identity and its key categories, outlines theories

that predict differential behaviors towards ones in-group versus out-group members, and generates key hypothe-

ses. Section 3 describes Study I and presents its results; Section 4 focuses on Study II. Section 5 concludes the

paper.

2. Identity and behavior towards in-group and out-group members: conceptual background

2.1. Why identity?

Identity, or a persons sense of self, is the outcome of a developmental process whereby differentiation between self

(in-group) and other (out-group) occurs. It is a process that starts in early childhood from the undifferentiated unit of

mother and child (Klein, 1984). In adulthood, identity is associated with identification with groups or categories such as

gender, ethnicity, religion, musical preferences, and dressing style. A sense of self and group belonging is also observed

among animals, who display the ability to recognize their kin ( Fletcher and Michener, 1987; Hepper, 1991). This is the basis

for differentiation between in-group and out-group members.

Identity is the answer to who am I? characteristically given with reference to multiple groups or categories (Hamachek,

1992; Akerlof and Kranton, 2000). For example, one may identify oneself as tall, music lover, who loves to eat health

food, Protestant, fan of certain sports teams, and so on. Identity is thus a composite of multiple attributes ( Sen, 2006).

The relative composition and weight of each of these attributes may vary over a persons life cycle, across people, and

with the circumstances of their lives (Hamachek, 1992). For example, musical preferences may be very important andreligion only marginal in some persons concept or sense of identity; the weight of ethnicity may be enhanced by the

presence of multiple ethnic groups or ethnic confrontation at the expense of other attributes such as cultural or musical

preferences.

Individuals perceive others identities in comparison to their own identity, and evaluate the similarity and dissimilarity

between them. Others identities can be inferred from various signals. For example, surface-level attributes such as gender,

race, dress style and other attributes are readily observable in face-to-face interactions, whereas deeper-level attributes

such as nationality, cultural values, religion, political views and other attitudes and beliefs generally become known through

extended interactions (Harrison et al., 1998).

Identity and the assessment of others identities have genetic, cultural and neural bases grounded in an evolutionary pro-

cess (Cavalli-Sforza and Feldman, 1981). Genetic relatedness, whether observed and known or only inferred and assumed,

offers a strong basis for answering the question of who am I? and the related question of who is the other person?

If we are our genes, then the people with whom we share a greater proportion of our genes are an immediate instance

of in-group as compared to less related out-group people. Going from identical twins, who are genetically identical, tomembers of an extended family, who are closely related, to members of a tribe, who share only a small proportion of

common genes, and so on, the declining proportion of shared genes provides an instant basis for increasing differenti-

ation between in-group and out-group; this is the key insight of Hamilton (1964) theory of inclusive fitness. Genes that

incline their bearers to be caring toward those who carry similar genes would have been selected in the process of human

evolution.

Evolutionary theorists note the value of steady affiliation with a group, and claim that the desire to belong to a group may

be hard-wired in some species, including humans. Group affiliation provides physical protection (Shaw and Wong, 1989),

facilitates the ability to read facial, behavioral, or linguistic clues regarding feelings such as guilt and the detection of lying,

which confers an obvious advantage (Wilson, 1978), and facilitates reciprocity, a key element of sustained cooperation and

trust.3

Other sources of identity may have little to do with genetic relationship and more with the psychological needs for

association with and distinctiveness from others.4 Group affiliation may be based on demographic characteristics such as

age and generation, or on functional association, such as a work group, neighborhood, common interest, culture, or hobby;therefore, the range of possible identities is very large. One theory that advances this view, social identity theory (Tajfel and

Turner, 1979), is widely accepted among sociologists and social psychologists.5

Many identity attributes have been recognized in the literature, and those have been aggregated into a set of broad, partly

overlapping categories. Most of these categories can be derived without much stretch from all three theories. The panel

below lists the most important categories that appear in the literature, and when available, cites references that elaborate

on each category from diverse theoretical perspectives.

3 See, for example, Ben-Ner et al. (2004a,b) and Ben-Ner and Putterman (2009). Ethnically homogenous middlemen groups confer benefits on their

members and reduce transaction costs associated with exchange uncertainty (e.g., Greif, 1993; Landa, 1997).4 It is quite possible thatthe psychological needsfor association and distinctiveness can be explained in evolutionary theoretic terms. Such an exploration

goes beyond the scope of this paper.5

Foran expansivediscussion of this andrelated theories,see Ellemers etal. (2002). Darityet al.(2006) advance thistheory by developingan evolutionarygame model to show how racial identity may evolve in a society in which individuals are easily identified by racial criteria.


4/18


Broad identity categories.

Identity category Literature

Family and kinship Shaw and Wong (1989); Skefeld (1999); Alderfer (1997); Van den Berghe (1999)

Gender Davis (2000); Akerlof and Kranton (2000); Barkow (1989); Dickson and Pollack (2000); Wade (2001)

Occupation Spreitzer et al. (1974); Cartwright et al. (1978); Becker and Carper (1956); Savickas (1999)

Ethnicity Barkow (1989); Dien (2000); Alderfer (1997); Davis (2000); Devos (1972); Van den Berghe (1999)

Culture Skefeld (1999); Dien (2000); Davis (2000); Devos (1972)

Nationality Dien (2000); Wade (2001)

Race Abdullah (1998); Alderfer (1997); Davis (2000); Hirschfeld (1995); Wade (2001)Religion Barkow (1989); Miller et al. (2001); Skefeld (1999)

Political philosophy Miller et al. (2001)

Dress style Miller et al. (2001); Dickson and Pollack (2000); Hayes (1999)

Community type Hummon (1986); Davis (2000)

Interests Hummon (1986); Pitts (2002)

Hobbies and leisure Spreitzer et al. (1974); Anderson and Farris (2001); Baughman (2000); Dickson and Pollack (2000)

Knowledge Hummon (1986)

Sentiment Hummon (1986)

Generation and age Alderfer (1997); Dickson and Pollack (2000)

Socio-economic status Cartwright et al. (1978); Akerlof and Kranton (2000); Devos (1972)

Musical preference Brown (2000); Pitts (2002); Tarrant et al. (2001); Wade (2001)

Sexual preference Brown (2000); Wade (2001)

The literature suggests that individuals tend to assign people with whom they interact to a class of in-group or out-

group according to these categories. The in-group/out-group differentiation may go beyond a stark dichotomy; for instance,

individuals distinguish among immediate relations such as parents and siblings, more distant relatives, such as cousins, and

even more distant members of an extended family, and likewise, some religions or denominations within broad religions

may be considered closer to each other than to others. However, there is also a strong tendency to make a simple division

between in-group and out-group, us and them; we will follow such a dichotomy in the remainder of this paper. 6

2.2. Behaviors towards in-group and out-group

Humans seem to have a deep-rooted propensity to respond emotionally to symbolic representations of members of their

in-group by exhibiting spontaneous joy, pride, and so on (Isaacs, 1975; Tnnesmann, 1987), and these emotions are aroused

and reinforced through the language of kinship and the use of rituals, flags, anthems, marches, and so on (Johnson, 1995).

It has been widely noted that individuals engage in more favorable behaviors towards people who share with them some

salient identity attributes than towards people who are different from them. Behaviors and relationships affected in this

fashion by the in-group/out-group differentiation have been discerned in many contexts, such as conflict (Shaw and Wong,1989), teacherstudent relations (Akerlof and Kranton, 2002; Hamachek, 1992), managersubordinate interactions (Boone

et al., 1999; Akerlof and Kranton, 2000), job performance (LePine and Van Dyne, 2001), and occupational choice (Cartwright

et al., 1978).

Similarity in identity may entail more trust, reciprocity, fewer concerns about being taken advantage of, efficiency due

to shared language, norms, understandings or concerns, as well as psychological comfort, a sense of warmth and so on.

However, identity may also be a clue to possession of instrumental skills (such as occupational and educations status), a

special need (such as low socio-economic status), or reproductive capacity (being of opposite gender); in such cases the

effects on behavior of similarity or difference in identity may be emphasized, or attenuated and even reversed, so that a

wealthy person may give more money to a needy poor person than to a rich one, or a man may prefer to engage with a

woman rather than a man in an interaction that holds the potential for procreation.

The three theories predict that individuals will treat others whom they consider to be in-group more favorably than

those whom they regard as out-group. Indeed, there is ample direct evidence that identity matters for behavior: ethnic,

national, and religious wars dot history, discrimination on the basis of almost any conceivable grounds is commonplace,and a visit to a schoolyard during recess shows how children divide into random teams to play a ball game and develop

instantly strong feelings towards members of their own team and their temporary adversaries.7 Studies have demonstrated

that people generally favor in-group over out-group members in distribution of rewards (for example, Brewer, 1979; Tajfel

and Turner, 1986; Brewer and Brown, 1998), and that they attribute more positive views to in-group members than to out-

6 For an argument that human beings process information with the aid of categories rather than more detailed attributes, see Fryer and Jackson (2003).7 Theimportance of thedistinction between in-group andout-group forbehaviorcan be perhaps best grasped from studies that usearbitrary distinctions

between artificially-createdgroups,but whichgenerate significantdifferences in behaviortowards in-group andout-groupmembers.In the famous Robbers

Cave Experiment, Sherif et al. (1961) showed how deep antagonism can arise among two groups 12-year old boys of similar backgrounds just on the basis

of random assignation to groups. In a series of experiments, Tajfel and Turner (1986) divided subjects arbitrarily into groups according to preferences for

painting styles, andthen asked members of different groups to share money with members of their own preference group or other groups. Those who were

assigned to a particular preference favored persons who were assigned the same preference. These dictator-game like experiments showedhow importantare in-group and out-group identities, irrespective of their arbitrariness.


5/18


group members (Allen, 1996; Rustemli et al., 2000).8 Trust game experiments showed differences in trusting on ethnic and

national lines (see Glaeser et al., 2000; Fershtman and Gneezy, 2001; Bornhorst et al., 2006).

Inclusive fitness theory suggests that family and kin relations constitute the most important identity category. Other

categories willbe ranked according to the known, perceived or inferred genetic relatedness; categories that indicate potential

genetic relatedness such as ethnicity and nationality will rank higher than other categories. Since genetic similarity can only

rarely be observed directly, individuals may use clues correlated with genetic identity: ethnicity, skin complexion, religion,

culinary preferences, place of origin, physical similarity, etc. (Van den Berghe, 1999). In contrast, in the case of gender out-

group (opposite sex)members are requiredfor procreation, so inclusive fitness considerationssuggest favoritism to go exactlythe other way from all other categories. Evolutionary theory predicts that long-term affiliation is valuable, pointing to the

same categories as inclusive fitness theory, and to groups with which individuals tend to be attached for long periods of

time. Categories that signal deep-level affinity, such as political and cultural views, will also be important for differentiation

between in-group and out-group. Particularly importantare categories that are the intersection or overlapof other categories.

The foregoing discussion implies that the strength of differentiation between in-group versus out-group relative to a

particular identity category is likely to depend on context. For example, trust and reciprocity are likely to be more important

in work than in social situations, and cultural compatibility is more important in leisure contexts than at work.

2.3. Hypotheses

We offer three broad hypotheses that capture the principal ideas represented by the literature and the three theories

discussed above.

H1. In general, individuals favor in-group over out-groupmembers. This applies to a broad range of identities and to diverseactivities.

In the specific empirical context of this paper, we expect that study subjects will exhibit preference for in-group versus

out-group for most of the identity categories and in the contexts of giving money, work and other activities.

H2. The degree to which individuals favor in-group members over out-group members varies across categories of identity.

In the empirical context of this paper, the strongest differentiation between in-group versus out-group is for the family

and kinship category, and weakest for gender. The ranking of other categories remains an empirical question.

H3. The effects of identity vary across contexts and behaviors.

Theremainder of the paper tests these hypothesesand provides empiricalevidence on the relative importance of different

identity categories and contexts.

3. Study I

3.1. Subjects and procedure

All freshmen at the University of Minnesota were invited by email to participate in economic-psychological experiments;

nearly 10 percent responded, with 222 actually showing up at the experiment in September 2002. The average age of the

sample was 18.8 years with 92.8 percent of individuals being between 18 and 21, with female and Caucasian majorities (64.0

percent and 71.4 percent, respectively).

Subjects were assigned to one of two identical sessions. After taking a seat and receiving a random identification number,

subjects completedthe followingsteps:(1) a timed12-mincognitive-ability test, (2) a personality inventory,(3) a willingness-

to-give survey-experiment, (4) a willingness-to-commute survey-experiment, (5) a willingness-to-work survey-experiment,

(6) a willingness-to-share an office survey-experiment, (7) a split-the-pie survey-experiment, and (8) a detailed personal

background questionnaire. Each step was contained in a separate envelope that was sealed after its completion. All seven

envelopes were then placed into a larger envelope that carried a random identification number and which was dropped

into a collection box near the exit door; subjects were then paid the promised $15 participation fee. The entire session took

approximately 45 min. In the remainder of this section we describe the four survey-experiments completed in steps 36 and

the background questionnaire, which are used in this paper.

3.2. Design

We designed four survey-experiments that capture various behaviors in social and economic contexts aiming to: (1)

examine in-group/out-group differentiation, (2) explore differences in the relative strength of identity categories, and (3)

8

Allen (1996) found an in-group bias effect for individuals of European and African descent, such that both groups attributed more positive traits tomembers of their respective in-group.


6/18


investigate differences across giving, work and social contexts relative to identity categories.9 In the four experiment-surveys

subjects were asked to express their willingness to give money to, work with, share an office with, and commute with

91 different persons characterized by various identity attributes.10 Information about subjects was gathered through a

background questionnaire that permitted the creation of in-group/out-group dummy variables indicating whether a subject

was similar to or different from each of the various imaginary persons with whom they were paired. In this study we used

imaginary rather than real persons in order to be able to pair our subjects with a much large number of attributes of other

persons than what would have been available in ourpoolof subjects,or forthat matter, inmost pools accessible toresearchers.

We used imaginary situations (work, commute and share office) in order to investigate work and social interactions thatcannot be ordinarily simulated in the laboratory.

The first survey-experiment was designed as a zero-sum, one-shot game, where the subjects were asked to . . . imagine

yourself in a situationin which you aregiven $10, which you cankeepto yourselfor give to another person, allor any portion of

it. Subjects were asked to consider sharing their hypothetical (imaginary) $10 endowment with another (imaginary) person.

This mimics the familiar dictator game that is carried out with actual money. The dictator game is a one-person decision

process: one player, the dictator, divides a fixed amount of money between himself or herself and another person, the

recipient, who is entirely passive and has no say in the decision. In this situation, giving any amount to the other person costs

the subject exactly that amount, dollar for dollar. Because a selfish subject who understands the extremely simple structure

of the game would give nothing, the common interpretation is that any giving implies caring, altruism and unconditional

cooperation towards the other person. This is thus especially relevant to the question concerning differential caring for

in-group versus out-group members.

The hypothetical nature of the money may introduce two types of bias. First, study subjects may exaggerate how much

they would give because it costs them nothing to be generous. Such exaggeration would be problematic if it meant thatsubjects gave the same amount (e.g., the maximum $10) to all persons. But if the exaggeration takes the form of a factor

that multiplies any amount they would give in an actual-money incentivized experiment, then the difference between the

giving to in-group and to out-group would be multiplied by that factor, and the test of our theory would be whether the

measured difference in giving to in-group and out-group is greater than zero. There is no reason to expect that this factor

is related to identity categories, so the test for the existence of a hierarchy among identity categories would be revealed

by the test of the differences in the differences in giving to in-group versus out-group across identity categories.11 Sec-

ond, subjects may understate the degree to which they favor in-group over out-group, again because it costs nothing to

be politically correct; this social-acceptability effect would entail giving more of hypothetical money to out-group than

actual money. This effect would reduce observed differences in behavior towards in-group versus out-group and militate

against our hypotheses.12 A related issue with a potentially similar effect arises from the fact study subjects may not distin-

guish among other persons as carefully as they would if these were actual rather than imaginary individuals. An argument

can be made against the expectation of these two biases on the basis that subjects have nothing to gain from misstating

their preferences in an anonymous study, and that the effort of stating their actual preferences is so small that only theextremely effort-averse subjects will agree to participate in the study and be paid for their participation but respond ran-

domly (or give not-considered responses, such as the same response) to questions asked in the study. Hence the alternative

hypothesis against the existence of these biases is that the dictator game with hypothetical money and imaginary recipients

will yield essentially the same results as the dictator game with actual money and actual recipients (see Ben-Ner et al.,

2008).13

In addition to the explicitly economic situation of giving money, we examined hypothetical behaviors in work and social

situations. In three separate survey-experiments subjects were asked to answer yes or no to three questions: do you

want, or not want, to commute daily to school with a particular person, do you want, or not want to work with a particular

person on a project critical to your career advancement, and whether you like or dislike sharing an office with this person.

Working on a project critical to ones promotion requires a choice of partner who can be trusted to cooperate, reciprocate

andgenerally actfavorably to ones interests, andwho is likelyto be a good worker. Sharing an office is an ongoing activitythat

has milder instrumental implications and stronger social-compatibility requirements. Commuting together is an activity of

short duration that entails social interactions without any instrumental elements. In each survey-experiment subjects were

9 We use the compound term survey-experiment because there was no random manipulation between individuals in an independent variable, but

there were within-individual manipulations of the hypothetical situations.10 Of the 222 subjects, 20 participated in a single-attribute dictator game to test for possible bias associated with the inclusion of multiple attributes in

the same experiment. For the attribute in question (being female) tests show that there was no difference between what was given by the 202 subjects in

the multiple-item survey-experiment and the 20 subjects in the single-item study-experiment.11 A similar argument concerns exaggeration that takes the form of a constant add-on to what subjects would have given with actual money, as long as

the amounts of hypothetical money given to in-group and out-group do not hit the upper bound of $10. This not the case in our sample. We define the

nature of these potential biases precisely in the context of the estimating Eqs. (1) and (2) in Section 3.4.12 The question why respondents may misstate their attitudes in anonymous and confidential surveys and experiments is discussed in Ben-Ner et al.

(2008).13 Basic statistics of giving in Study I indicate that subjects seem to have given consideration to their responses. Over all imaginary persons, the average

within person standard deviation in giving is 2.146. For imaginary in-group persons the standard deviation equals 2.354 and for imaginary out-group

persons the standard deviation equals 1.672. The percent of respondents who gave the same amount to all imaginary people is 2.99%. 4.98% of respondentsgave the same amount to all imaginary in-group people and 10.45% of respondents gave the same amount to all imaginary out-group people.


7/18


Fig. 1. In-group and out-group comparisons, by context and identity category.

paired separately and sequentially with 91 different persons characterized in ways that are directly associated with an

identity category. Instructions are included in Appendix A.

3.3. Empirical specification of in-group and out-group and of identity categories

In this paper, of the 91 imaginary persons we used only the 53 persons that fit into one of the following categories of

identity: family,political views, sports-team loyalty, music preferences, nationality, religion,socio-economicstatus,television

viewing habits, food preferences, birth order, body type, dress type, and gender. These 13 categories correspond to most of

the categories presented in Section 2. Table 1 illustrates the bases for creating the in-group/out-group dummy variables.

This was done by matching the persons listed in the survey-experiments with corresponding characteristics reported by

subjects in thebackgroundquestionnaire. Forexample, if a participant indicated that s/he belongs to oneof several Protestant

denominations, then the in-group/out-group dummy variable was coded as 1 (in-group) when the person on the list was

described as Protestant, and was coded as 0 (out-group) for a person described as Buddhist, Muslim, or Jewish.14 For a study

participant shorter than 66 (for males), the variable was coded as in-group for a person described in the list as short, and

out-group for tall.

An identity categorygenerally consists of multipleattributes or items; for example, there aremultiplemusical preferences,

several religions, two ways of characterizing body type (height and weight), and so on. We created the in-group/out-group

dummy variables by taking the average over the items in each category. In the sports-team loyalty category, we use only

one item, fan of ones team versus fan of a rival team. In the family and kinship category, in-group includes family relations

of varying degrees, as well as persons described as looks like you and resembles you. The last two items were included

because clues to genetic closenessare associated with looks.Out-group forthis categoryis theperson described as a stranger,

the obvious non-kin.15 See Table 1 for details.

3.4. Results

Fig. 1 displays the sample averages and proportions broken down by in-group and out-group for each identity category,

by context (type of behavior). The upper left panel shows that for all identity categories, with the exception of gender,

mean levels of giving are larger for in-group than for out-group. The differences are particularly large for the family, reli-

gion, political views, sports-team loyalty and music preferences categories. Similar results are seen in the remaining panels

14 For various reasons, we did not include race and ethnicity in our experiments (Jewish was included in the religion category).15

Out-group characterizations such as someone youve seen crossing the street and someone youve seen at the checkout counter at the supermarketare less loaded that the term stranger but produce similar results.
http://-/?-http://-/?-http://-/?-http://-/?-


8/18


Table 1

In-group and out-group definitions, by identity category.

Identity category Participants self-characterization

in the background questionnaire

The other person is considered in-group

relative to the study participant when the

person is characterized as

The other person is considered out-group

relative to the study participant when the

person is characterized as

Family and kinship

Your brother Stranger

Your close relative Stranger

Your father Stranger

Your brother-in-law StrangerYour stepfather Stranger

Your cousin Stranger

Resembles you Stranger

Looks like you Stranger

Political viewsPolitically liberal (13 on a 6

point scale)

Politically liberal Politically conservative

Politically conservative (46 on a

6 point scale)

Politically conservative Politically liberal

Sports-team loyalty Fan of your favorite sports team Fan of your rival sports team

Music preferences

Bluegrass is a favorite type Listens to bluegrass music

Alternative is a favorite type Listens to alternative music

Contemporary pop/rock is a

favorite type

Listens to contemporary pop/rock

New age is a favorite type Listens to new age music

Rap/hip-hop is a favorite type Listens to rap/hip-hop music

Opera is a favorite type Listens to opera music

Bluegrass is not listed as a

favorite type

Listens to bluegrass music

Alternative is not listed as a

favorite type

Listens to alternative music

Contemporary pop/rock is not a

favorite type

Listens to contemporary pop/rock

New age is not a favorite type Listens to new age music

Rap/hip-hop is not a favorite type Listens to rap/hip-hop music

Opera is not a favorite type Listens to opera music

NationalityAmerican American Chinese, and from France, Iraq, Argentina,

Russia and Poland

Out-group nationalities: parallel

treatment

Religion Belongs to a Protestantdenomination

Protestant, Lutheran Muslim, Buddhist, or Jewish

Out-group religions: parallel

treatment

Socio-economic status

Family experienced financial

difficulties while growing up

Poor Financially well-off

Family was financially well-off Financially well-off Poor

Had to work while in high school Had to work while in high school Did not have to work in high school

Father is pro fessional worker Fath er is a physician Father is a factory worker

Father is unskilled or semiskilled

worker

Fath er is a factory wo rker Father is a physic ian

T V viewing Watches T V f or at least 3 h ours a

day

Watches a lot of TV Hardly ever watches TV

Watches T V at most 1 ho ur a day Hardly ever watch es T V Watch es a lot of T V

Food preferences

Convenience foods such as chips

are favorite

Eats chips often Eats salad often

Vegetarian meal is favorite Vegetarian Eats hamburger often

Birth order Youngest child Youngest child Oldest child

Oldest child Oldest child Youngest child

Body type

Taller than 73 if male, 68 if

female

Tall Short

Shorter than 58 if female, 66 if

male

Short Tall

Body mass index (BMI) 27 if female, 27.3 if male Overweight Skinny

Dress type Dresses like you Dresses differently from you

GenderFemale Female Male

Male Male Female


9/18


of Fig. 1 for the proportions of subjects liking to share an office, wanting to commute, or wanting to work with another

person.

The identity categories in Fig. 1 are listed from the smallest in-group/out-group difference in giving to the largest. The

largestaverage differencefor allfour behaviorsis for the family category: for giving,the in-group/out-groupdifference is $2.93,

while for the share office, commute and work behaviors the differences in proportions are 0.28, 0.52 and 0.53, respectively.

The smallestaverage in-group/out-group difference for giving is for the gender category ($0.14), for share office is the body

type category (0.016), for work is the body type category (0.005), and for commute is the gender category (0.013). A

slightly higher proportion of subjects favor out-group than in-group in the gender category for the giving and commute

behaviors; out-group in the body type category is also shown a slightly more favorable attitude, on average, than in-group

in the share office, work and commute behaviors but not in giving.

The raw averages presented in Fig. 1 suggest that (a) in-group is treated more favorably than out-group, with very minor

exceptions, (b) there are marked differences in the way in-group and out-group are treated across identity categories, and

(c) there are differences across behaviors. The remainder of this section explores these points in more detail and relative to

the hypotheses put forth in Section 2.

In order to further investigate in-group/out-group differences by identity category and behavior type, we estimated

fixed-effects regression and fixed-effects logit models. For the level of giving, we assumed that

giq = 0 + is + io +c C

scI(q = {c, s}) +c C

ocI(q = {c, o}) + iq (1)

where i denotes the individual, q denotes the particular imaginary person that subject i is paired with, c denotes the identity

category under consideration, c{1, . . ., C}, s denotes whether the imaginary person q is of thein-grouptype ando the denotes

whether the imaginary person q is of the out-grouptype. The parameters is and io are individual fixed effects for imaginarypeople who fall into the in-group and out-group types, respectively. Thus, these parameters measure the average giving to

in-group and out-group across all identity categories for a particular individual. By allowing for individual fixed effects for

in-group and out-group in our hypothetical dictator giving game, our in-group/out-group difference estimates by identity

group will be valid estimates for an actual dictator giving game even if the average levels of giving and the within-individual

differences in these average levels of giving between in-group and out-group differ between actual and hypothetical dictator

games; all that is required is that the difference in differences across identity categories are the same.16 The parameters scand oc measure the category deviation from the person-specific mean for in-group and out-group types. For simplicity, we

have assumed that these deviations themselves are notperson specific. Finally, iq is an individual-imaginary person-specific

error term. For the commute, work and share office behaviors the fixed-effects logit model

ln

piq

1 piq

= 0 + is + io +

c C

scI(q = {c, s}) +c C

ocI(q = {c, o}) (2)

is estimated, where piq represents the probability that individual i says yes to the question posed that pertains to imaginary

person q.

Table 2 presents estimates of sc oc , the in-group/out-group difference, by identity category, relative to the in-

group/out-group difference in the excluded category, body type, in which little favoritism is expressed (see Fig. 1).17 These

estimates are based on the estimates of the fixed-effects models described by Eqs. (1) and (2). The full set of fixed-effect

estimates is presented in Table A1. Column (1) presents estimates based on the fixed-effects regression estimates for the

giving survey-experiment, whereas columns (2)(4) present estimates based on the fixed-effects logit estimatesfor the share

office, work and commute survey-experiments, respectively.

(1) Table 2 conveys a strong message: in-group members are significantly favored over out-group in nearly all identity

categories in all four survey-experiments. Exceptions are glaringly few: body type, birth order and dress type, where

there is no favoritism, socio-economic status, where in-group is preferred in work and social contexts by out-group is

favored in giving, and gender, where out-groupis favored in all four contexts. We conclude that Hypothesis 1 is supportedby our empirical evidence.

(2) The results in Table 2 show large differences in the treatment of in-group versus out-group across identity cat-

egories, in all four contexts. For giving, we can reject the null hypothesis that the in-group/out-group difference

is independent of identity category (F= 21.99, p-value = 0.000). For sharing an office, and commuting and working

with another individual, we also soundly reject (p-value = 0.000) the null hypothesis that the in-group/out-group dif-

ference is independent of identity category (2(12)= 115.83, 2(12) = 208.93, 2(12)= 201.56, respectively). Thus, theextent of in-group/out-group favoritism varies substantially across identity categories, providing empirical support for

Hypothesis 2.

16 Thismeans thatthe parametersis andio maydiffer between the hypotheticaland actual dictatorgames, because for our estimates to be valid estimates

of in-group/out-group differences in actual giving game we only require that sc and oc are the same.17

The numbers underlying the body type category in the four panels of Fig. 1 are $0.15 (giving), 0.016 (sharing office), 0.005 (work) and 0.032(commute).


10/18


Table 2

Estimated differences in behaviors towards in-group and out-group, by identity category, Study I (hypothetical money and imaginary receivers).

Variable Regression Logit

Giving (1) Share office (2) Work (3) Commute (4)

Family 3.325*** 4.853*** 7.771*** 7.326***

Political views 1.585*** 4.170*** 3.106*** 4.120***

Sports-team loyalty 1.648*** 3.818*** 2.217*** 3.112***

Religion 1.068*** 2.844*** 2.243*** 2.453***

Music preferences 1.057*** 5.301*** 2.968*** 4.962***

Television viewing 0.610 2.419*** 2.776*** 1.876***

Food preferences 0.477*** 1.262** 1.000** 1.274**

Birth order 0.263 0.767 0.281 0.138

Dress type 0.107 0.398 0.671 0.825

Nationality 0.053 2.570*** 3.176*** 2.180***

Socio-economic status 1.048** 2.228*** 2.631*** 1.048**

Gender 0.938* 2.176** 2.179** 1.863**

N 202 222 222 222

Notes: The table reports estimated differences in behaviors towards in-group and out-group based on the fixed-effect estimates reported in Table A1.

Significance tests are based on two-sided asymptotic z-tests of differences in the in-group/out-group estimatedcoefficients for each identity category. One,

two and three asterisks indicate statistical significance at the 10%, 5% and 1% levels, respectively. The excluded category is body type, so these differences

are relative to the in-group/out-group difference for body type, the average of which (0.1429) can be computed by looking at differences in the average

estimated individual effect between in-group (0.0745) and out-group (0.0684) across individuals.

In order to explore Hypothesis 2 in more detail and gain insight into the relative importance of various identity cate-

gories, we analyzed the relative ranks of the in-group/out-group differences by identity category for the four contexts.

Our point estimates for the giving survey-experiment show that the family category has the largest in-group/out-group

difference, followed by sports-team loyalty, political views, and religion and music preferences. What is the likelihood

that this ordering is due to chance? We used bootstrapping techniques using 1000 replications to examine the rank-order

distribution of our estimates. Bootstrapping treats the sample as a population and then re-samples with replacement a

number of times and computes relevant statistics for each replacement sample. The empirical distribution of the boot-

strapped sample statistics is then used to address questions of statistical significance (see Efron and Tibshirani, 1993 for

details). Here we analyze the bootstrapped samples empirical distribution of relative ranks since we were interested in

addressing the question of whether, for example, the in-group/out-group difference for family was statistically signifi-

cantly more important in terms of rank than political views. We would judge the difference in rankings as statistically

significant if in more than 95 percent of the replications the ranking of family in-group/out-group differences was higher

than the ranking of political views in-group/out-group differences. Because of the computational complexity of estimat-ing the fixed-effects logit model, the rank order of in-group/out-group differences was bootstrapped only for the giving

experiment, which was based on a fixed-effects regression model. The results are presented in Table 3. For each identity

category, the table reports the mean rank and the 5th and 95th percentiles of the rank distribution based on the 1000

replications. For example, the in-group/out-group difference for the religion category was ranked on average 4.5 across

all identity groups while in the upper 5 percent of the replications the difference was ranked third or higher and in the

lower 95 percent of the replications the difference ranked sixth or lower.

The in-group/out-group difference was largest for the family category in all 1000 replications. The next two highest

mean ranks were for the sports-team loyalty and political views categories. However, since sports-team loyalty was

Table 3

Ranks of in-group/out-group differences for giving, Study I summary statistics from bootstrap replications.

Variable Mean rank 5th percentile 95th percentile

Family 1 1 1

Political views 2.763 2 4

Sports-team loyalty 2.493 2 3

Religion 4.463 3 6

Music preferences 4.546 3 6

Television viewing 6.561 4 9

Food preferences 7.050 6 9

Birth order 8.354 6 11

Dress type 9.255 7 11

Nationality 9.593 8 11

Body type 9.937 8 11

Socio-economic status 12.601 12 13

Gender 12.384 12 13

Notes: Ranks are based on 1000 bootstrap replications of the model described by Eq. (1) in the text. For each replication, the rankings

are based on the differences between the estimated coefficients for in-group and out-group by identity category relative to the excludedcategory Body type, which is normalized to zero.


11/18


ranked higher than political views only in 55 percent of the replications, the difference in mean ranks is not statistically

significant. The fourth and fifth highest mean ranks were the religion and music preferences categories, respectively, but

since the in-group/out-group difference for religion was larger than that for music preferences in only 52 percent of the

replications, the rank differences are not statistically significant. When comparing sports-team loyalty to the religion

and music preferences categories, the in-group/out-group differences for the sports-team loyalty category are larger

than both religion and music preferences categories in over 95 percent of the replications. Thus, the rank differences

are statistically significant. The political views in-group/out-group difference was larger than the religion and music

preferences differences for giving in 90 percent and 91 percent of the replications, respectively. The evidence is therefore

not as strong as for sports-team loyalty. At the other end of the ranking we have gender and socio-economic status; the

difference in their ranks is not statistically significant, but the difference between each of these and those ranked higher

is statistically significant.

The rank ordering of identity categories obtained from bootstrapping replications is, not surprisingly, essentially the

same as that implied by the relative magnitude of differences in giving across identity categories in the fixed-effects

regression reported in column (1) of Table 2. The rank ordering of different identity categories for the other contexts

presented in columns (2)(4) is similarly implied by the relative magnitude of the estimated differences for each behav-

ior. The order of importance of identity categories varies across the four columns, but the preeminent role of family

persists across behaviors. Family is far ahead of other categories in terms of favoritism for in-group versus out-group

not only with respect to giving (estimated difference of 3.325 as compared to 1.648 for sports-team loyalty, the next

largest difference), but also for work (estimated difference of 7.771 as compared to 3.176 for nationality preferences, the

next largest difference), and commute (estimated difference of 7.326 as compared to 4.962 for music, the next largest

difference); only in the share office the estimated difference between in-group and out-group for family is second, being

slightly smaller than the difference for music preferences (4.853 versus 5.301). The findings concerning the rank of the

family category lend support to our secondary hypothesis regarding the centrality of family and kin suggested by the

inclusive fitness and evolutionary theories.

We also hypothesized that gender is nota sourceof identity like other categories because the role of sexin reproduction

and the functional role of opposites. Indeed, our empirical findings suggest that in the gender category a small advantage

is given to out-group over in-group in the gender category.18 There is a also a small difference in preference for giving to

out-groupthan in-group in the socio-economic status category, probably explained by the fact that most people, whether

well-off or not, prefer to give money to the poor rather than the well-off, so the charitable motivation trumps identity in

this particular context.

The relative importance of other identity categories linked to long-term affiliation cannot be tested without classifying

identity categories accordingto the duration of affiliationor the potential contribution of variousgroups to the procreative

success and survival of those affiliated with them. Such a classification is not available in the literature, and is a task that

is well beyond the scope of this paper.

(3) Hypothesis 3 suggests that theimportance of similarity in identity varies acrosscontexts. While weare unable to compare

directly parameter estimates from regression and logit analyses, we can do so across the logit analyses concerning the

share office, work, and commute contexts. In order to evaluate the importance ofsimilarity for a given identity category

we tested the equality of the (in-group) (identity category) coefficients across the share office, work and commute

contexts; the chi-square tests reject the null of equality at the 1 percent level for the identity categories of family,

music preferences, and sports-team loyalty (and at the 10 percent level for dress type and birth order). As the parameter

estimates on (in-group) (identity category) in Table A1 suggest, our study subjects value more commuting and working

with their kin than sharing an office with them, and they prefer commuting with someone who shares their musical

preferences and sports-team loyalty, but this similarity does not seem to be very important for sharing an office and

certainly not for working on a critical project. In other categories similarity (rather than difference) in identity does not

seem to play a role.

To explore this hypothesis in more detail we compared behaviors towards in-group and out-group across the three

contexts by carrying out pair-wise tests of equality between the logit estimates in columns (2)(4) of Table 2 for each

identity category. The chi-square tests and direction of the difference in estimates are presented in Table 4. The difference

in the extent to which in-group is favored over out-group is statistically significant and substantial across the three contexts

in the case of only two identity categories: family and music preferences. The differentiation between in-group and out-

group in the family category is greater in the work and commute contexts than in the share office behavior, whereas in

the case of music preferences the difference is larger for the share office and commute behaviors than for work. The music

preferences category is likely to bear more on compatibility in social situations such as commuting and sharing an office

than on trust and cooperation and therefore the order we just discussed makes sense. The family category probably bears

more on trust, reciprocity and cooperation than on compatibility in social settings and therefore should be more important

for work than sharing an office or commuting; the former relationship is found in our data, but not the latter. Less consistent

18

There might be asymmetries in ways that men and women treat each other, which we did not explore here intentionally (but see the dictator gameexperiments on this issue in Ben-Ner et al., 2004a,b).


12/18


Table 4

Test of equality of coefficients across share office, work and commute contexts, by identity category, Study I.

Share office versus work Share office versus commute Commute versus work

Family 21.36***() 12.67()*** 0.36(+)

Political views 2.45(+) 0.01() 1.94(+)

Sports-team loyalty 7.74(+)*** 1.32(+) 2.08(+)

Religion 1.55(+) 0.48() 0.14(+)

Music preferences 20.78(+)*** 0.30() 9.32(+)***

Television viewing 0.26() 0.50(+) 1.20()Food preferences 0.31(+) 0.00() 0.25(+)

Birth order 0.34(+) 0.51(+) 0.02(+)

Dress type 0.16() 2.43() 4.32(+)**

Nationality 0.91() 0.33() 1.95()

Socio-economic status 0.62() 4.30(+)** 6.52()**

Gender 0.00(+) 0.06() 0.09(+)

Notes: Each cellshows thechi-squareteststatistic.() indicatesthatthe estimateon thefirst-listed context issmaller than theestimate on thesecond-listed

context; (+) indicates the opposite. The comparison is between estimates presented in Table 2. One, two and three asterisks indicate statistical significance

at the 10%, 5% and 1% levels, respectively.

and significant differences concern sports-team loyalty, which is more important for sharing an office and commuting than

for work, similar to music preferences, and dress type, which is more important for commuting than for work, again similar

to music preferences.

Hypothesis 3 is thus generally supported by these findings, which suggest that some identity categories are more impor-tant for activities in which trust and cooperation is central (work), and others are more important for contexts that entail

a large element of social interaction. However, for the several remaining identity categories there is no strong differential

impact by identity on behavior.

4. Study II

In order to verify that the results derived from the giving survey-experiment with imaginary other persons and without

real money is not an artifact of its hypothetical nature, we conducted an incentivized dictator game experiment where the

other persons were real people and there was an actual $10 endowment. We also conducted three survey-experiments,

similar to the work, share office and commute survey-experiments in Study I.

4.1. Subjects and procedure

All undergraduate and graduate students in the Management and Education schools at the University of Minnesota were

invited to participate in a study on decision-making with a promise of $10 in show-up fee, and up to additional $10 in

paper-and-pencil experiment earnings. Advance on-line registration, which included completion of a personal background

questionnaire, was required. Of the 89 registrants, 54 individuals showed up at the experiment in September 2007. The

average age of the sample was 25.6, with 44.4 percent of individuals being between 18 and 21, with female and Caucasian

majorities (66.7 percent and 75.9 percent, respectively). The on-line registration took approximately 15 min, whereas the

experiments took about 30 min.

Subjects were assigned to three rooms, two of which were referred to as Room A and one as Room B. The 37 subjects in

the A rooms were senders and the 17 B room subjects were passive receivers. 19

4.2. Design

Like Study I, Study II had four components: a dictator game, and expression of the degree of desirability to work,share an office, and commute with each of eight other persons. The eight persons-attributes we have chosen are sim-

ilar but not identical to those in Study I because of the nature of our subject pool and the limited scope of the

study (validation). The eight persons were characterized as follows: born in the USA, below average height, does not

believe in God, politically liberal, politically conservative, born in a foreign country, taller than average, and believes

in God. These attributes belong to four identity categories: political views, nationality, belief in God, and body type. 20

The nationality category is summarized by born in the U.S. versus born in another country, religion is replaced with

belief in God, and body type is summarized by height (omitting weight); the political views identity category is iden-

19 Because there were fewer Room B subjects than subjects in the two A rooms, they received money from more than one sender. Senders were not told

that their counterpart in Room B may receive money from other senders.20 The category of belief in God was introduced in Study II in lieu of the religion variable. In the background questionnaire there was an item: On a scale

from 1 to6 where1 isDo not believe/Agnostic and 6 isVery strong belief, how would you describe your faith in God? Subjectswho checked 1, 2 or 3 wereclassified as not believers, and those who checked 4, 5 or 6 as believers.


13/18


Table 5

Estimateddifferences in behaviors towardsin-group and out-group, by identity category, Study II (actual money and actual receivers in the dictator game

experiment).


Giving (1) Share office (2) Work (3) Commute (4)

Belief in god 1.486*** 5.366*** 4.516*** 6.621***

Nationality 0.124 2.887 1.534 0.669

Political views 1.959*** 5.660*** 3.933*** 6.901***

N 37 37 37 37

Notes: The table reports estimated differences in behaviors towards in-group and out-group based on fixed-effect estimates (akin to those presented in

Table A1). Significance tests are based on two-sided asymptotic z-tests of differences in the in-group/out-group estimated coefficients for each identity

category. One, two and three asterisks indicate statistical significance at the 10%, 5% and 1% levels, respectively. The excluded category is body type (tall

versus short).

tical in the two studies. The coding of the in-group/out-group dummy variables was done in the same fashion as in

Study I.

In the dictator game, Room A subjects were asked to make a decision to keep $10 or send any portion of it in increments

of one dollar to eight different individuals who were located in Room B. They were told that they will be paid only for one

decision: However, you will not know until the experiment is over which of the decisions is the one related to this participant,

so your best bet is to treat each decision as if it is the real one. The experiments followed the protocol employed by Bardsley

(2000) and others, by offering Room A subjects payment for one choice out of several. The full instructions and decisionsheet are in Appendix B. The dictator game components in Study I and Study II were otherwise structured similarly, except

that in Study II subjects were asked to indicate how much they want to keep of their $10 endowment, and how much they

want to send to the other person, whereas in Study I they were asked how much they want to give to the other person, and

how much they want to keep.

After they completed the dictator game experiment, subjects were asked to complete three additional survey-

experiments, asking to rate each of the eight persons described above in terms of their desirability for sharing a daily

ride to school, as office mates, and as members of a work team critical to the respondents career advancement and who, on

the basis of tests, were deemed to be equally competentto carry out the tasks associated with the project. Study II survey-

experiments differ from those in Study I in that the questions were on a 14 scale rather than no or yes, and the explicit

statement regarding the equal competency of potential work-team members. Study II survey-experiments are reproduced

in Appendix B.

Room B subjects had a passive role as receivers of As sending. In order to justify the statement in the invitation referring

to additional earnings of up to $10 from participation in a pencil-and-paper experiment, Room B subjects were asked toanswer a multiple-choice question (about the murder rate in Minneapolis), on the basis of which they were compensated

$1, in addition to receiving whatever they were sent from subjects in Room A.

4.3. Results

We estimated the model in Eq. (1) for the dictator game decisions, using data from Study II. For the commute with, share

office with, and work with survey-experiments, after reversing the scores we grouped the 1 and 2 responses together and

the 3 and 4 responses together to make them compatible with Study I, we estimated the model in Eq. (2).21 As in the previous

analyses, the excluded category is body type. The results are presented in Table 5, which parallels Table 2. In the identity

category of belief in God, the estimated difference for giving (sending) in the dictator game in column 1 is 1.486 and highly

significant, comparable to the 1.068 in the religion category in Table 2. The nationality category difference is 0.124, small

and statistically insignificant, comparable to the 0.053 difference in Table 2. The estimated difference for the political views

category is 1.959, statistically significant and comparable to 1.585 in Table 2.22 The analysis of behavior in the contexts of

share office, work and commute is presented in columns (2)(4) ofTable 5. The estimates reveal that the differences in the

treatment of in-group versus out-group are greater in Study II than in Study I for the belief in God/religion and the political

views categories, but are more muted in the nationality category. (This was true in the work context, despite the explicit

statement of equal competency; if anything, one might have expected that the nationality category, which may correlate

with problems in communication, would be a basis for favoring in-group, which for most subjects in the study means US

bornbut this was not the case.) We thus conclude that the analysis of behaviors in Study II reveals very similar patterns to

those uncovered in Study I.

21 We also estimated a fixed effects regression model using the (reversed) 14 scores as the dependent variables. The results were similar.22 When themodel in (1) is estimatedusingStudyI data andonly those identity categories used in StudyII, there are no statisticallysignificant differences

between thetwo studies in the estimatedin-group/out-group differences in averagegiving by identity category when both person-specific fixed effects forin-group and out-group giving are included.
http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-


14/18


Table 6

Means of giving and differences in giving to in-group versus out-group, Study I versus Study II, by identity category.

Category Study I Mean giving ($) S.E. Study II Mean giving ($) S.E. Difference ($)

(1) (2) (3) (4) (5) = (1)(3)

a. Mean giving

All 2.25 0.05 2.03 0.14 0.22

Political views 2.27 0.16 2.04 0.28 0.23

Nationality 2.32 0.07 2.07 0.38 0.25

Religion/belief in God 2.24 0.08 1.85 0.27 0.39Body type 2.14 0.09 2.01 0.27 0.13

b. Estimated differences in giving to in-group and out-group

Category Study I Estimated

in-group/out-group

differences (1)

S.E. (2) St udy II Est imated

in-group/out-group

differences (3)

S.E. ( 4) D if ference (5) = (1) (3)

All 0.70 0.07 0.27 0.16 0.43**

Political views 1.28 0.19 1.08 0.32 0.19

Nationality 0.38 0.14 0.03 0.35 0.35

Religion/belief in God 0.86 0.20 0.36 0.32 0.50

Body type 0.23 0.16 -0.30 0.30 0.53

Notes: One, two and three asterisks indicate statistical significance at the 10%, 5% and 1% levels, respectively. Differences in panel (b) are derived from the

coefficient estimates for fixed-effects regression models. See text for differences in the composition of the identity categories between Study I and Study II.

We now turn to investigate whether use of hypothetical money and imaginary individuals instead of actual money and

actual individuals is associated with the two biases discussed in Section 3: sending more hypothetical money than actual

money, and not being as discerning as when actual money and actual individuals are considered. We test for the existence of

the first bias by comparing the average amount of giving overall and in each of the four identity categories included in both

studies. We test for the second bias by comparing the difference in giving to in-group versus out-group overall and in each

of the four identity categories between Study I and Study II.

The results of the analysis of the firstbiasare presented in panel (a) ofTable 6. Over all four identity categories, the average

amount of giving is higher in Study I than in Study II ($2.25 versus $2.03), but the difference is not statistically significant

at conventional levels (p-value= 0.136). When broken down by identity category, the differences in average giving, while all

positive, are never statistically significant. The small and insignificant difference may be indeed due to a slight bias (to mildly

exaggerate giving when it does not cost anything to do so), or to minor idiosyncratic events that have an influence in smallsamples23 or minor difference in protocol between the two studies.24 Our conclusion is that the first bias is insignificant:

subjects seem to give very similar amounts of hypothetical and actual money.25

The results of the investigation of the second bias are presented in panel (b) of Table 6, based on person fixed-effects

estimation. The overall difference between Study I and Study II in estimated in-group/out-group differences is positive and

statistically significant (p-value = 0.014). However, when brokendown by identity category, while all positive, none the Study

I-Study II differences in the estimated in-group/out-group differences are statistically significant, and we conclude that there

is no significant bias of the second type.

To determine whether the small and insignificant observed differences in giving where due to unobserved differences

in the sample of subjects across Study I and Study II, we also analyzed whether there were any were any differences

between these studies in the proportion of subjects who were willing (in Study I) or finding it desirable (in Study II)

to share an office with another person. Since in both studies this question was hypothetical we would expect no sig-

nificant differences between the two studies. The results for sharing office are reported in Table A2.26 The estimated

overall difference in the fraction willing to share office in Study I (0.79) and Study II (0.72) is positive and statisticallysignificant, but when broken down by identity category the differences are mixed in sign and are statistically insignif-

icant. The overall Study IStudy II difference in the in-group/out-group estimated difference in the fraction willing to

share an office, on the other hand, is negative and statistically significant, but again when broken down by identity cat-

egory the differences are mixed in sign, and are never statistically insignificant.27 These results suggest that there is no

23 In one of thetwo A rooms,a participant commented outloud (despite the explicitprohibition) that it was obvious that one should notsend any money.

This pronouncement may have had a norm-setting effect as the average amount sent was significantly lower in this room compared to the other, $1.68

versus $2.39. (In the estimation in Table 5, any effect of the comment on sending is absorbed by the individual in-group and out-group fixed effects.)24 As noted earlier, in Study I the options were listed first give and second keep whereas in Study II the options were reversed; the first option listed

may have had a minor advantage.25 Ben-Ner et al. (2008) could not find an average difference in giving hypothetical versus actual money, although they found that there are certain

personality differences between those who give more or less hypothetical than actual money.26

Comparisons regarding work with and commute with another person yield similar results and are available upon request.27 Estimates are based on fixed-effects logit models hence the coefficient estimates need not be between 0 and 1.


15/18


consistent pattern of differences between the two studies with respect to the fraction of subjects willing to share an

office.

We conclude that the results of Study II reinforce our conclusions from Study I.

5. Conclusions

The assumption that behavior is independent of the identity of those who participate in an economic interaction is

central to economists understanding of how markets operate, how firms work internally, how nations trade with each

other, and much else. In this paper we show that the distinction between in-group and out-group affects significantly

economic and social behavior. We found that our student subjects favor those who are similar to them on any one of a wide

range of categories of identity over those who are not like them. Whereas family and kinship (including persons described

as looks like you and resembles you in addition to various relatives) are the most powerful source of identity in our

sample, it appears that there is hardly any identity category that is considered inconsequential when it comes to favoring

in-group over out-group. Subjects favor in-group over out-group in giving in dictator game (whether with hypothetical or

with actual money endowments), in choosing members of a work team (from among equally competent candidates who

differ only, for example, in their political views or belief in God), in sharing an office or in commuting. In addition to family

and kinship, political, religious and cultural beliefs and affiliations are important bases for discrimination in various contexts

(we did not test for race and ethnicity, whereas nationality ranks relatively low); body type, dress style and especially gender

are not.

Although other interpretations are possible, the important identity categories may be viewed as modern-day equivalents

of tribal or hunting-band affiliation of yore when belonging to groups was particularly important for survival and, thus, may

have evolutionary roots.28 That gender is not found to be a source of oppositional identity is also predicted by inclusive

fitness theory.

Our findings are derived from two similarly structured studies. The studies differed in a major methodological respect:

usage of hypothetical money and imaginary recipients in a survey-experiment that mimics the dictator game in one study,

and an incentivized dictator game experiment with actual money and actual recipients in the other study. The differences in

the amounts that dictators sent in the two studies are not statistically significant; furthermore, the difference across studies

in the difference given to in-group versus out-group in each of the identity categories that were included in both studies

is also statistically not significant. This finding invites renewed consideration of the contexts in which financial incentives

must be meted in experiments in order to elicit truthful responses, and when actual people are needed even in passive roles

(e.g., recipients in dictator game experiments).

Not everyone favors in-group over out-group to the same degree, but identity-based behavior seems to drive behavior

in many contexts and with respect to many identity categories. As economists strive to incorporate other-regarding and

social preferences such as trust and reciprocity in their models of economic behavior, it is apparent that preferences for

in-group versus out-group also need to be integrated in our analyses. Further research in this area is much needed to

understand economic andsocial interactionsin firms,cities and countries amongan increasinglydiversepopulation.Whereas

in the present paper we considered one identity attribute at a time, it would be important to investigate how multiple

identity categories, which is the way individuals often present themselves in reality, affect behavior. Do certain dimensions

trump others? For example, does information about ones political views make other data about that person unimportant or

irrelevant in the eyes of many?

The costs of identity-based favoritism and discrimination include inefficient job assignments, incorrect promotion prac-

tices, and unfair treatment of people, as well as conflict and war, to name a few examples. Can these costs be reduced? Can

favoritism be reduced or its scope altered? We argued that there might be evolutionary bases for the preference of in-group

over out-group, but we did not claim that evolution has prescribed specific identity categories. Further research may inves-

tigate what our intuition suggests, that the importance of some identity categories is learned through experience, which can

be affected by policy (Ben-Ner and Hill, 2008). Furthermore, our findings show that several categories are equally important,

some of which have been historically the bases for harsher discrimination than others. Is there a possibility for elevating the

cultural importance of some identity categories, such as sports-team fanship, over other categories, such as religion, in order

to channel preferences for in-group over out-group to more tolerable venues? Much more research is needed in this critical

area.

Acknowledgements

We would like to thank the Carlson School of Management at the University of Minnesota for financial support. We are

grateful for the helpful comments of the co-editor, referees, many readers, discussants and audiences at seminars at the

University of Minnesota (2003), Monash University (2003), and the Hebrew University (2008), and at the Conference on

28

In a blog entitled Which Side Are You On? in the New York Times of December 3, 2008, the singer Suzanne Vega writes that [s]ongs brand us a partof a tribe.


16/18


Embodied Cognition (2003), Society for the Advancement of Socio-Economics (2003), Behavioral Economics and Eco-

nomic Psychology IAREP-SABE Congress (2006), Economic Science Association European Meeting (2006), Identity, Equity

and Opportunity conference at Oxford University (2006), and the Identity and Organization OB Camp at Yale School of

Management (2007).

Appendix A.

Table A1

Fixed-effects regression and fixed-effects logit estimates, Study I.


Giving (1) Share Office (2) Work (3) Commute (4)

Family 0.532 (0.173) 2.802 (0.213) 4.085 (0.242) 3.633 (0.230)

Nationality 0.158 (0.104) 0.420 (0.156) 0.816 (0.156) 0.359 (0.169)

Political views 0.429 (0.186) 2.373 (0.226) 1.518 (0.244) 2.443 (0.241)

Television viewing 0.014 (0.145) 0.805 (0.201) 1.207 ( 0.200) 0.960 (0.213)

Religion 0.168 (0.112) 1.280 (0.159) 1.150 (0.163) 1.296 (0.170)

Music preferences 0.293 (0.109) 2.090 (0.154) 1.349 (0.159) 1.961 (0.165)

Food preferences 0.040 (0.127) 0.343 (0.189) 0.391 (0.191) 0.416 (0.201)

Sports-team loyalty 0.641 (0.173) 1.223 (0.224) 1.180 (0.232) 1.227 (0.239)

Socio-economic status 0.996 (0.117) 0.559 (0.171) 0.861 (0.171) 0.350 (0.187)

Gender 0.643 (0.173) 2.282 (0.465) 1.900 (0.391) 1.441 (0.404)

Dress type 0.158 (0.173) 0.550 (0.295) 0.594 (0.295) 0.162 (0.276)

Birth order 0.293 (0.162) 0.189 (0.311) 0.026 (0.440) 0.642(0.296)

In-group family 2.793 (0.249) 2.050 (0.338) 3.686 (0.361) 3.693 (0.364)

In-group nationality 0.211 (0.256) 2.150 (0.460) 2.359 (0.449) 1.821 (0.506)

In-group political views 1.157 (0.302) 1.797 (0.396) 1.589 (0.430) 1.677 (0.427)

In-group television viewing 0.625 (0.339) 1.614 (0.535) 1.570 (0.503) 0.916 (0.518)

In-group religion 0.902 (0.237) 1.565 (0.344) 1.093 (0.346) 1.157 (0.371)

In-group music preferences 0.763 (0.246) 3.211 (0.391) 1.620 (0.367) 3.001 (0.435)

In-group food preferences 0.437 (0.248) 0.919 (0.369) 0.609 (0.366) 0.858 (0.399)

In-group sports-team loyalty 1.006 (0.286) 2.595 (0.453) 1.037 (0.402) 1.884 (0.458)

In-group socio-economic status 0.051 (0.229) 1.672 (0.354) 1.770 (0.354) 0.698 (0.370)

In-group gender 0.295 (0.286) 0.106 (0.686) 0.279 (0.573) 0.422 (0.579)

In-group dress type 0.265 (0.286) 0.152 (0.459) 0.077 (0.459) 0.633 (0.471)

In-group birth order 0.556 (0.302) 0.578 (0.591) 0.026 (0.440) 0.780 (0.648)

Constant 2.167 (0.085)

Number of observations 10660 8784 8695 8484

Person-in-group/out-group groups 402 345 335 312

R2 = 0 .0812 Log likelihood = 2965.41 Log likelihood= 2946.19 Log likelihood= 2675.82

Notes: Each observation corresponds to a particular person-identity category in-group/out-group value. For the fixed-effects logit estimates, all observa-

tions in which in-group/out-group-identity category groups haveno variation in the dependent variable are dropped from the estimations. Standard errors

are in parentheses.

Table A2

Proportion of participantswilling to share an office, an

Documents

Ben-Ner, Culture, identity