Endowments, perceived similarity, and dictator giving.

Goerg, Sebastian J. ; Johnson, David B. ; Rogers, Jonathan D. 等

I. INTRODUCTION

Imagine a college student receives a prepaid credit card as a gift from her grandmother for the start of the school year. The same day when checking out at a grocery store, she is asked if she would like to make a donation to a charity. Now imagine that the same student received the same prepaid card but this time as part of a propaganda package distributed by a racist organization to lure students. Would this change her donation behavior? Given that a fundamental assumption in economics is that income is fungible regardless of its source, the answer should be no.

However, behavioral literature is full of examples in which income is not treated as fungible. Usually, this phenomenon is attributed to mental accounting (e.g., Thaler 1980, 1985, 1999). Mental accounting is the routine of organizing income streams into mental accounts and the subsequent spending of the income within the categories of these accounts. (1) As Abeler and Marklein (2016) demonstrate (with field and lab experiments) this can lead to distorted consumption decisions. In their experiments, a plurality of subjects treat income from different sources as nonfungible. That is, in-kind grants lead to higher consumption of the target good compared to cash grants allowing for the same consumption bundle. (2) More recently, Clingingsmith (2015) tests for fungibility by comparing amounts sent by dictators when the endowment is earned versus when the endowment is windfall income (i.e., money provided rather than earned). In line with a mental accounting paradigm, subjects treat income from both sources as only partially fungible. (3)

In this article, we propose an additional channel through which the source of income influences decision making: the decision maker's similarity to the income source. The underlying proposition is that individuals feel more attached to income if they are similar to the source of the income. We base this proposition on the fact that studies in social-psychology and behavioral economics use group identity to highlight the similarity between agents. Generally, a strong sense of group identity leads to stronger prosocial behavior within the in-group and discrimination against individuals who are not members of the in-group (Allport 1954; Chen and Li 2009; Tajfel 1970). Group biases can be observed in both naturally occurring groups and laboratory-induced groups. Group biases have been demonstrated to occur in natural groups, based, for example, on race or occupation (e.g., Chen et al. 2014; Fershtman and Gneezy 2001: Goette and Meier 2006). (4) Moreover, these types of biases are seen in groups (created in the laboratory) based on trivial characteristics and/or random assignment (e.g., Chen and Chen 2011; Chen and Li 2009; Eckel and Grossman 2005; Tajfel and Turner 1979).

It has been previously shown that group affiliation influences perceived similarity and results in higher cooperation rates (e.g., Fischer 2009. 2012; Glockner, Goerg, and Fischer 2014) and altruistic punishment of norm violations (Muss-weiler and Ockenfels 2013). Evolutionary strategies based on similarity can even guarantee the surviving in relatively hostile and uncooperative environments (Fischer et al. 2013). While this body of literature demonstrates the influence of similarity on behavior towards other individuals, similarity with an otherwise irrelevant income source has not been investigated. To investigate the influence of similarity to the source of one's endowment on individual economic behavior, we use dictator games in which we manipulate the similarity between the endowment sources and the dictators by varying the source. (5)

While we are not the first to investigate income source effects in the dictator game, much of the previous work focused on the ownership of the income and whether it was earned or generated by windfall gains. While an in-depth discussion of these results is beyond the scope of the present work, windfall income usually leads to higher generosity than earned income (e.g., Cherry 2001; Clingingsmith 2015; Engel 2011; Hoffman et al. 1994). Likewise, when recipients earn the amounts to be distributed, dictators often assign larger amounts to the recipients (Engel 2011 : Oxoby and Spraggon 2008).

In our experiment, income is generated by members of three different groups: another university student, an online worker, or a user of a racist website. Income is generated prior to the laboratory experiments. This design is unique, because by randomly assigning dictators to different endowment sources, we exogenously manipulate subjects' similarity with their endowment sources, while also keeping the population of the recipients and the size of the endowment fixed. Thus, any differences in generosity are the result of the subject's similarity to the endowment sources.

Using a self-assessed perceived similarity measure, we demonstrate a significant relationship between the dictator's perceived similarity to the endowment provider and dictators' generosity. Specifically, dictators with greater perceived similarity to the endowment source pass off a smaller portion than those reporting lower similarity levels. However, due to heterogeneity in dictators' perceived similarity with endowment sources, we find no significant differences in dictator transfers across the three groups. The observed negative correlation between perceived similarity and generosity is in line with out-conjecture that individuals value money more if it comes from a relatively more similar source. Thus, consumer decisions may be susceptible to changes in income sources not only as a result of mental accounting but also through attitudes towards income sources.

II. EXPERIMENTAL DESIGN

The experiment has two stages. In the first stage, we use nonlaboratory subjects from three distinct groups (Stormfront.org [STORM] users, Amazon Mechanical Turk [AMT| Workers, and Florida State University [FSU] students) to generate the endowments of the dictators in the second stage. The amount of the generated endowment is always the same. This first stage is only implemented to manipulate the subjects' similarity to the endowment source without the use of deception. Details for this first stage can be found in the Appendix.

In the second stage, we use a dictator game in the laboratory with a standard student subject pool. (6) Subjects are informed that the endowment was generated by a member of one of three different groups. As such, the randomly assigned sources (Florida State University students, Amazon Mechanical Turk Workers, and Stormfront.org users) generate our three treatments: FSU, AMT, and STORM.

As previously stated, the treatments are designed to manipulate the similarity between the dictator and the endowment source. In the following, we will distinguish between objective and perceived similarity. (7) Objective similarity is based on the comparison of the most prevalent attribute of the assigned endowment sources with the average dictator in the lab. This means that the dictators in the lab have the highest objective similarity to other FSU students, second highest to AMT workers, and the lowest to STORM users. (8)

While useful in many respects, objective similarity does not take into account individual differences in dictators' self-perception and the perception of their endowment provider. (9) To capture these individual differences, we elicit, from each subject, their subjective perceived similarity to their assigned endowment source. To asses perceived similarity, subjects are asked two questions on a scale ranging from 1 (not at all) to 11 (very much). The first one elicits how similar subjects believe they are to an average person from their assigned endowment source provider (SIMILAR) and the second asks how much the subject shares opinions and beliefs (OPINIONS) with the provider. (10)

In the lab, after being seated and having the instructions read to them, subjects complete a short quiz to ensure understanding of the dictator game. After completing the quiz, the experiment begins. Each experiment takes approximately 45 minutes. In addition to a $5 show-up fee, each subject makes, on average, $5. Dictators on average earn $1.86 more than responders. Half of the subjects are assigned the role of the dictator (called "proposer" in the instructions) and are given $10. The other half are assigned the role of the recipient and receive no initial endowment. A summary of the treatments and earnings can be seen in Table 1.

All participants learn that subjects from three different groups generated initial endowments and that the computer randomly matched each dictator with one of the three sources. Once the game starts, the dictator learns about their matched source by viewing screen shots from the sources' websites and the groups' motto/mission statement (presented in Figures Al-A3). Only the matched endowment sources are revealed to the dictator, the other two possible sources remain unknown to the dictator. The dictator then decides the amount to be transferred to the recipient (in U.S. Cents). The recipient earns the amount the dictator transfers to him, while the dictator earns what is left of the initial endowment. Figure 1 gives the structure and timeline of the experiment. After the dictator game, all subjects answer a questionnaire which includes a short demographic survey and the perceived similarity statements. (11)

To prevent signaling concerns, responders only observe their earnings and neither learn the actual source of the dictators' endowment nor the set of possible sources. Subjects are truthfully told that their decisions will not be linked to their name and that their decisions will remain private. In addition, we minimize the possibility of session and experimenter demand effects (Zizzo 2010). In each individual session, dictators are randomly matched with all three sources. Students learn the source of their endowment only on the screen (i.e., it is not read aloud) and only they know the source of their own endowment. Additionally, we pay subjects using a double blind mechanism which subjects are aware of from the start of the experiment.

Double blind payment is accomplished by recruiting an additional subject who left the laboratory for the duration of the experiment. This subject's only role is to deliver the payments to subjects by calling a number that corresponds to a given computer terminal and handing each participant an envelope with their terminal number written on it. The envelope contains a check with the payment for the experiment. The check is signed and has the correct payment amount written on it, but does not include the name of the subject, who is instructed to add it herself later. The researcher preparing the checks is not in the same room as the participants and has no way to link individuals to terminal numbers and ipso facto their allocation decisions or survey responses. (12)

III. HYPOTHESES

We posit that the source of an endowment plays an important role in determining how money is valued and thus spent. In effect, this means that money provided by a source that is very similar to the subject is valued more than the same amount if received from someone with a different background or set of beliefs. In a dictator game, this will affect the marginal rate of substitution between the subject's own earnings and the earnings of the other via the dictators' generosity. To illustrate this point, we provide a simple extension to the linear model of inequity aversion by Fehr and Schmidt (1999). (13)

According to the classical model of inequity aversion, a dictator would derive positive utils from own earnings and negative utils from either advantageous inequity or disadvantageous inequity. In the dictator game, the dictator's earnings are the initial endowment [E.sub.d] reduced by the transfer x to the recipient. The degree of aversion against advantageous inequity [beta] [greater than or equal to] 0 represents the regret the dictator feels if their earnings are larger than the earnings of the recipient (i.e., [E.sub.d] - x > x.). Conversely, the degree of aversion against disadvantageous inequity [alpha] [greater than or equal to] 0 represents the envy the dictator feels if earnings are smaller than the earnings of the recipient (i.e., [E.sub.d] - x < x).

We incorporate the effect of similarity to the endowment source by weighting the dictator's earnings with the term (1 - [gamma]D), where D measures the (dis-)similarity and [gamma] [greater than or equal to] 0 the weight the dictator puts on similarity. The similarity to the endowment source is measured as the distance between an attribute [a.sub.d] of the dictator and the attribute [a.sub.s] of the source or D = [absolute value of [a.sub.d] - [a.sub.s]].

For example, D could measure the difference in social status, education, or political orientation. If D = 0 the source and the dictator share the same attribute and are extremely similar; as D increases the similarity between the dictator and the source decreases. (14) This small addition suggests dictators will value money more if the source is very similar. Hence, the dictator's utility function has the following form:

(1) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

Based on this utility function, the dictator's transfers will depend on the similarity with the endowment source D, the weight put on this similarity y, as well as. the inequity aversion parameters a and p. (15) Therefore, the amount the dictator transfers to the recipient is determined by the conditions given in Equations (2) and (3).

(2) [beta] [gamma] [gamma]D/2 [greater than or equal to] [1/2]

(3) [gamma]D [greater than or equal to] 1 2[alpha].

If condition 2 is not satisfied, the dictator will transfer nothing to the recipient. If the condition is satisfied, the dictator will equalize the earnings by transferring half of the initial endowment to the recipient. In addition, if condition 3 is satisfied, the dictator is willing to transfer the whole initial endowment to the recipient. In this extreme case, the dictator considers the source of the money to be so different that the disutility from keeping the money is larger than the disutility from the resulting inequity due to the higher earnings of the receiver. (16) Put differently, the individual would rather give away the entire endowment than be associated with its source. The following implications of the model can be summarized as follows:

* A money-maximizing dictator, who has no inequity aversion and no similarity concerns ([alpha] = 0, [beta] = 0, and [gamma] = 0), will transfer nothing to the recipient

* A dictator with inequity aversion but no similarity concerns ([alpha] > 0, [beta] > 0, and [gamma] = 0) will either transfer nothing or the earnings equalizing amount to the recipient. The dictator will transfer half of the initial endowment if [beta] > 1/2.

* A dictator with inequity aversion and similarity concerns ([alpha] > 0, [beta] > 0, and [gamma] > 0) will transfer nothing, half of the initial endowment, or the entire endowment to the recipient. (17) Transfers increase in D and [gamma]. Dictators will transfer positive amounts even when they have lower [beta]'s than dictators who care only about inequity aversion. If either D or [gamma] become sufficiently high, they will outweigh the disadvantageous inequity aversion a and full transfers will occur.

Based on these implications, we can derive behavioral hypotheses for our experiment, which will be supported if a considerable number of participants care about the similarity to the endowment source. While we assume that inequity aversion ([alpha], [beta]) and the weight put on similarity ([gamma]) are fixed characteristics of the subjects, it is possible to influence the similarity between a dictator and the source of an endowment (D) by varying the source. As such, our treatments are designed to exogenously manipulate similarity and thus allow us to formulate the following hypotheses:

HYPOTHESIS 1: Dictators' transfers are negatively correlated with the similarity to the endowment source. Based on objective similarity, we expect the following ordering of average transfers FSU < AMT < STORM.

Our treatments manipulate objective and perceived similarity. We expect both similarity measures to be correlated. However, due to individual differences in dictators' self-perception and the perception of the sources, we expect a stronger relationship between perceived similarity and transfers.

HYPOTHESIS 2: Dictators' transfers are negatively correlated with the perceived similarity to the endowment source. The higher the perceived similarity to the endowment source the lower the transfers will be.

IV. RESULTS

Subjects in the experiment are typical of the standard university population. The average age of subjects in our experiment is 21.02 years and 48.2% are female. Because of perceived links between social issues and political affiliation and religion, we also ask subjects their political affiliation and religion. (18)

On average, dictators give 31.3% of their endowment to the recipients, which is very close to the mean of 28.35% reported in the meta-study by Engel (2011). (19) However, the distribution of amounts sent looks quite different from what is reported in Engel (2011). Out of 20,813 dictators in Engel (2011), 36.11% give nothing to the recipient, and 16.74% choose the equal split. In our study, a lower percentage of dictators give nothing (20.37%), and more dictators give half of their endowment (33.33%). According to our model, this could be the case if some of our endowment sources are perceived as less similar than the anonymous source (or experimenter) that is common in dictator games.

A. Objective Similarity and Dictators' Transfers

Table 2 provides summary statistics of transfers in our three treatments. As expected, we observe the lowest average transfer in FSU. Mean transfers for STORM and AMT are higher and roughly at the same level. However, mean transfers do not differ significantly between the three endowment sources (Kruskal-Wallis equality-of-populations test p=.19). (20)

The group of dictators being matched with AMT has the lowest standard deviation of transfers, the one matched with FSU the second highest, and the one matched with STORM the highest. The difference of standard deviations is significant for the comparison between AMT and STORM (two-sided SD-test p=.034), but not for AMT and FSU (p=.13) or FSU and STORM (P=. 52).

Objective similarity did not influence average transfers as much as expected. If there exists a relationship between similarity and transfers, there are two possible reasons why this might occur: (1) the experimental manipulation failed and did not influence similarity, or (2) it did influence perceived similarity, but in a heterogeneous way. In the next step, we will focus on the relation between objective and subjectively perceived similarity.

B. Objective Similarity and Subjectively Perceived Similarity

To assess the subjectively perceived similarity we used two questionnaire items (SIMILAR and OPINIONS) on a scale from 1 to 11. These two items are highly correlated (r = .86, with p <.001). We use these two items to construct a similarity score with an excellent reliability coefficient (Cronbach's [alpha]=.927). (21)

Figure 2 gives box plots for the similarity score across treatments. (22) It demonstrates that the pairing with different endowment sources successfully manipulated lhe perceived similarity and that the average ranking of perceived similarity is as expected. Thus, objective and perceived similarity ratings are significantly correlated (Spearman rank correlation using an ordinal objective similarity ranking p=.51 and p<.001). Comparing the similarity scores confirms the differences to be highly significant between the different endowment sources (Kruskal-Wallis equality-of-populations rank test p <.01). (23)

At the same time, Figure 2 reveals that the perceived similarity ratings have very high variances for the endowments sources of STORM and FSU and a rather small one for AMT. While the standard deviations of STORM and FSU do not differ significantly (two-sided SD-test p=.46), both are significantly larger than the one in AMT (p <.01 and p=.016).

Our experimental manipulation successfully manipulates the perceived similarity and the average effect is in line with the objective similarity. However, dictators' similarity perceptions with the endowment sources are very heterogeneous for STORM and FSU. This suggests that the nonsignificant effect of objective similarity on transfers is likely caused by objective similarity being an imprecise measure of similarity. In a next step, we investigate whether the more accurate measure of subjective similarity with the endowment source does a better job in predicting dictators' giving.

C. Subjective Similarity and Dictators' Transfers

Figure 3 shows the distribution of amounts given by dictators. The left panel of Figure 3 provides the distribution of offers for dictators reporting a perceived similarity to the endowment source that is below or equal to the median similarity level in our sample. The right panel provides the distribution of offers for dictators with reported similarity above the median level.

Average amounts shared by dictators, with above median similarity to the endowment source, are significantly smaller than their counterparts with less similarity to the endowment provider (239 Cents vs. 361 Cents, two-sided Fisher-Pitman permutation test p=.064). Furthermore, dictators with above median similarity are more likely to keep the whole amount (two-sided Fisher's exact test p =.085). (24) Overall, transferred amounts are negatively and significantly correlated with the perceived similarity levels (r =-.2628, with p <.055). (25)

In a last step, we test the robustness of the above results with additional regression analyses. Table 3 displays the results of several estimates based on Tobit regressions. (26) Models 1-4 replicate the previous results while taking the censored structure of the data into account. Model 1 confirms that being matched with STORM or FSU does not lead to significantly different transfers compared to being matched with AMT. (27) However, perceived similarity has a significant influence on transfers and this influence can be demonstrated for both similarity questions (SIMILAR, OPINION) and the combined perceived similarity score (Models 2, 3, and 4). Model 5 controls for additional characteristics of the subjects, which are taken from the sociodemographic questionnaire. (28) In addition, our design allows to control for single session effects as in each session dictators were paired with three endowment sources. Controls for session effects are included in Model 6. In all models, the perceived similarity is a significant driver of transfers. Thus, our additional analysis demonstrates that the effect of the perceived similarity with the endowment source on dictators' generosity is remarkably robust.

V. DISCUSSION

In this article, we explore how the impact of similarity to an endowment source influences the value individuals place on their endowment and hence the dictators' generosity. To manipulate the similarity with the source, members of three different groups generated the initial endowments: Stormfront.org users, AMT workers, or FSU students. In line with our hypotheses, similarity with the source leads to less generosity, while less similarity leads to more. However, the heterogeneity in perceived similarity is so large that the effect can only be observed for the individually perceived similarity and not the objective similarity.

Based on the intuition behind our model, we can explain why someone would try to avoid having money from a source with which they have little in common. In fact, looking at the maximal transfers in our dictator game reveals that some subjects transferred all of their initial endowment. Moreover, all of these subjects received their endowment from the STORM source. These subjects felt so strongly that they did not want to be linked with the source of the endowment, that they preferred to give away the entire sum.

Our work demonstrates perceived similarity between the decision maker and endowment provider affects economic behavior. However, it remains unknown if the perceptions of others affect decision making. For instance, if the recipient of an endowment knows that all members of the population are aware of a lack similarity between the source of the endowment and the population as a whole, would this lead to changes in recipient's decision? There are numerous examples of public figures donating the proceeds of funds received from sources deemed as dubious by the population, but it is not known if their donation was due to possible economic/social punishments in the future or from the public figures' education of the sources' dubious nature. (29) Such an exploration is particularly well-suited for experimentalists because it is not always possible to know if the decision maker is aware of the nature of the source of their endowment. Alternatively, the decision maker may only learn critical information about the source after the decision to accept the endowment. In either case, there may be economic advantages to feigning ignorance, so it is difficult to assess whether an individual is or is not fully aware of from where the money comes.

To close, our results suggest money from loved ones is spent differently than money from strangers. The fact that the source of income can influence monetary decisions beyond mental accounting is a novel extension to the literature and opens up interesting questions for future research that go beyond the topic of generosity. For example, does it matter for the spending of welfare recipients whether they receive the subsidies from members of their own communities or some anonymous pool of money? If we better understand the way that endowments are perceived, then we can improve our predictions of how they will be spent. Methodologically speaking, it also suggests that researchers should be aware of how they are perceived by subjects and it is another reason not to use a researcher's own students in the lab. If a student likes (or dislikes) their professor, then they may behave differently in an experiment than they would with a stranger. A dollar buys a dollar's worth of goods, no matter its source, but it can carry psychological costs that must also be considered. ABBREVIATIONS AMT: Amazon Mechanical Turk FSU: Florida State University STORM: Stormfront.org

APPENDIX A

FIRST STAGE AND THE ENDOWMENT GENERATION

The first stage is only used to generate the endowments for dictators in the second stage. This is done by having subjects take a short survey and subsequently indicate their preference over a pair of allocations. Stage one subjects are paid solely based on this choice, which is the end of their involvement. The allocation preference task is presented in Table Al. The first allocation pays 10 dollars to the decision maker and 8 dollars times a constant to an unspecified "other" party. The second allocation pays 12 dollars to the subject making the decision and 10 dollars times the same constant to the unspecified other party, the other party in the experiment being participants in the second stage of the experiment. Participants in the first stage are not told anything about the participants in the second stage.

The specific numbers here are irrelevant as Allocation B dominates Allocation A for both the source and the unspecified other. For any constant, z > 0. It is easy to defend strong priors that B will be chosen in nearly every instance, and indeed all stage one subjects make this selection. The purpose of this first stage is not to observe differences in behavior, but so that subjects in stage two can be truthfully told the source of their endowment.

Three subjects, each front a different group, participated in the first stage of the experiment. The first subject is an undergraduate student who participated in a pen and paper version of the task. The remaining two subjects participated through the online labor market Amazon Mechanical Turk (AMT). One was recruited through the open AMT worker pool while the other was recruited directly from Stormfront.org.30 To ensure the group membership of the Stormfront subject, this individual was recruited first and participated in a password protected version of the task. The password could only be obtained by forum users. All three participants were aware that they were participating in an experiment. These subjects are only used for endowment generation and as such we do not analyze any of the data provided by them.

APPENDIX B

SCREEN SHOTS OF ENDOWMENT PROVIDERS

FIGURE B1

Stormfront.org: "We are a community of racial realists and idealist. We are White Nationalists who support true diversity and a homeland for all peoples. Thousand of organizations promote the interest, values and heritage of non-white minorities. We promote ours. We are the voice of the new embattled White minority."

FIGURE B2

Amazon Mechanical Turk: "We give business and developers access to an on- demand, scalable workforce. Workers select from thousands of tasks and work whenever it's convenient."

APPENDIX C QUESTIONNAIRE ITEMS

Perceived Similarity

To assess subjects' perceived similarity, we used two questionnaire items to be answered on an 11 -point scale ranging from "Not at All" to "Very Much." Subjects were asked only for their matched endowment source, with source being www.stormfront.org user, www.mturk.com user, or Florida State University student. The exact wording was:

(SIMILAR) What do you think, how similar are you to an average source?

(OPINIONS) What do you think, how much do you share opinions and beliefs with the average source1

In addition, to draw attention from the similarity questions, we asked in between the two similarity questions how much they would trust the endowment source:

(TRUST) What do you think, how much can you trust an average source?

The first two similarity items are both significantly correlated with dictators' transfers (/?=--.2386 with p=.082 and r =--.2684 with p=.0497). The trust item is not significantly correlated with dictators' transfers (r =--.0918, and p=.51) and turns significant only after controlling for perceived similarity.

APPENDIX D

ADDITIONAL FIGURES AND TABLES

APPENDIX E

INSTRUCTIONS AND CONSENT

Experimental Instructions

Thank you for participating in today's experiment. During this experiment, you can earn money. At the end of the experiment your personal earnings will be added to your $5 participation fee. These earnings will be paid to you with a check as soon as the experiment is over. This payment will be anonymous, which means that no other participant of this experiment will be informed about the size of your payment.

Procedure for Your Payment. We invited one additional subject, who will not participate in the experiment. She/he will be randomly selected and leave the laboratory for the duration of the experiment. Therefore, she/he will not know the topic of today's experiment. She/he will receive the participation fee of $5 as well as an additional compensation.

After the experiment, this person is going to do the payments. She/he will call the numbers of the terminals and give an envelope to each participant. The envelope contains a check with the payment for this experiment. The check is signed and has the earned amount on it, but you still need to add your name. In addition, the envelope contains a receipt. Please add all necessary information to the receipt and put it back into the envelope and place the envelope in a box. This box will go directly to the financial accountants of the economics department and will be used only for accounting reasons. Neither the researcher nor the students supporting the conduction of this experiment will open these envelopes.

The participant organizing the payments will neither know the amount on the check nor the amount on the receipt! Please remember to take the check with you!

Why Do We Use This Procedure for Your Payment? This procedure will ensure that the researchers conducting this experiment are not able to connect a single decision in this experiment with an actual person or name. The researchers know only the cabin number, but they will never know your private information (e.g., name, gender, age). Furthermore, as the researchers are in a different room, they will not know your faces.

Thank you for participating in today's experiment. We will read through the instructions so that all participants of this experiment receive the same information. It is very important that you do not talk to other participants at any point during the entire experiment. In case you do not understand some parts of the experiment, please read through these instructions again. If you have further questions after this, please give us a sign by raising your hand. We will then approach you in order to answer your questions personally.

During this experiment, you can earn money. At the end of the experiment, your personal earnings will be added to your $5 participation fee. These earnings will be paid to you by check and/or in cash as soon as the experiment is over. This payment will be anonymous, which means that no other participant will be informed about the size of your payment.

During the course of the experiment, you will be randomly assigned to participate with another participant. You will receive information about the matched participant's decision, but you will get no information on who those person actually is, neither during the experiment nor at some point after the experiment. Similarly, the other participants will not be given any information about your identity.

Please remain quiet and do not communicate with other participants during the entire experiment. Raise your hand if you have any questions. One of us will come to you to answer them.

The experiment consists of one decision game and several questionnaires. The instructions for each part will be given separately at the beginning of that part.

In this game, half of all the participants in this room will be assigned to the role of the proposers, the other half will be assigned to the role of the responder. The computer randomly assigns each responder to one proposer. At the beginning of the experiment, the computer will inform you about your assigned role.

In this game, the proposer decides about the allocation of 1,000 U.S. Cents. People having different group affiliations generated this initial endowment. There are three different group affiliations. One third of today's proposers will be matched with a person from the first group, one third with a person from the second group, and one third with a person from the third group. Before the proposer chooses his allocation, the computer will reveal the affiliation of the person who generated the initial endowment. All people involved in the payments at the end of the experiment have no information on the group affiliation of the person that generated your initial endowment. The proposer can decide about the exact allocation of the 1,000 U.S. Cents.

This means the proposer decides about:

* the amount he/she wants to keep

* the amount he/she wants to give the responder

The allocation will be in U.S. Cents and the total sum of the allocated amounts must add up to 1.000 U.S. Cents. This game will be played once--no repetition will follow. After the proposer fixed the allocation, both responder and proposer will be informed about the chosen allocation as well as the final distribution of amounts. Responders will make no decisions. Thereafter, we ask you to answer some questionnaires.

Please raise your hand now if you have some questions about this game. We will come to your terminal and try to help you. Before proposers learn the group affiliation of the individual that generated your initial endowment, we will have you take a short quiz. Once everyone has completed the quiz, proposers will learn the group affiliation of the individual that provided the initial endowment.

APPENDIX F

OBJECTIVE SIMILARITY ASSESSMENT

Thank you for participating in this short questionnaire. In the following, we will ask you to assess how similar a random FSU student is to an average member of a given group. We will ask you this question for three different groups:

1. other students at Florida State University (FSU)

a. FSU is a large state university in Florida.

b. Their mission statement is: Strength. Skill, Character

2. users of Amazon Mechanical Turk (AMT)

a. AMT is an online labor market where workers complete short jobs to earn money.

b. Their mission statement is: We give business and developers access to an on-demand, scalable workforce. Workers select from thousands of tasks and work whenever it is convenient.

3. usersofStormfront.org.

a. Stormfront is a White supremacist forum.

b. Their mission statement is: We are a community of racial realists and idealists. We are White Nationalists who support true diversity and a homeland for all peoples. Thousands of organizations promote the interest, values, and heritage of non-White minorities. We promote ours. We are the voice of the new embattled White minority.

Please give your similarity ratings on a scale from 0 to 10 (0 meaning Not at all and 10 meaning Very much).

1. On a scale from 0 to 10, how similar do you think the average FSU student is to other FSU students? Answer:

2. On a scale from 0 to 10, how similar do you think the average FSU student is to users of Amazon Mechanical Turk? Answer:

3. On a scale from 0 to 10, how similar do you think the average FSU student is to users of users of Stormfront? Answer:

Thank you!

doi: 10.1111/ecin.12408

Caption: FIGURE 1: Structure of the Experiment

Caption: FIGURE 2: Box Plot of the Subjectively Perceived Similarity Score

Caption: FIGURE 3: Histograms of Dictator Giving for Low (Left) and High (Right) Similarity

Caption: FIGURE B3: Florida State University: "Strength, Skill. Character."

Caption: FIGURE D1: Mean Transfers and Standard Errors

Caption: FIGURE D2: Box Plot of Subjectively Perceived Similarity Measures

REFERENCES

Abeler, J., and F. Marklein. Forthcoming. "Fungibility, Labels, and Consumption." Journal of the European Economic Association, 2016.

Allport, G. W. The Nature of Prejudice. Reading, MA: Addison-Wesley, 1954.

Andreoni, J., and B. D. Bernheim. "Social Image and the 50-50 Norm: A Theoretical and Experimental Analysis of Audience Effects." Econometrica, 77(5), 2009. 1607-36.

Bolton, G. E., and A. Ockenfels. "ERC: A Theory of Equity, Reciprocity, and Competition." American Economic Review, 90(1), 2000, 166-193.

Chen. R., and Y. Chen. "The Potential of Social Identity for Equilibrium Selection." American Economic Review, 101(6), 2011,2562-89.

Chen, Y., and S. X. Li. "Group Identity and Social Preferences." American Economic Review, 99(1), 2009, 431-57.

Chen. Y., S. Lin. T. X. Liu, and M. Shih. "Which Hat to Wear? Impact of Natural Identities on Coordination and Cooperation." Games and Economic Behavior, 84, 2014,58-86.

Cherry, T. L. "Mental Accounting and Other-Regarding Behavior: Evidence from the Lab." Journal of Economic Psychology. 22(5), 2001. 605-15.

Clingingsmith. D. "Mental Accounts and the Mutability of Altruism: An Experiment with Online Workers." Working Paper, 2015.

Cooper. D. J., and D. B. Johnson. "Ambiguity in Performance Pay: An Online Experiment." Available at SSRN 2268633, 2013.

Currarini. S., and F. Mengel. Forthcoming. "Identity, Homophily and In-Group Bias." European Economic Review, 2016.

Eckel, C. C., and P. J. Grossman. "Managing Diversity by Creating Team Identity." Journal of Economic Behavior and Organization, 58(3), 2005, 371 -92.

Engel. C. "Dictator Games: A Meta Study." Experimental Economics, 14(4), 2011, 583-610.

Fehr, E., and K. M. Schmidt. "A Theory of Fairness, Competition, and Cooperation." Quarterly Journal of Economics, 114(3), 1999, 817-68.

Fershtman, C., and U. Gneezy. "Discrimination in a Segmented Society: An Experimental Approach." Quarterly Journal of Economics, 116(1), 2001, 351-77.

Fischbacher, U. "z-Tree: Zurich Toolbox for Ready-Made Economic Experiments." Experimental Economics, 10(2), 2007, 171-78.

Fischer. I. "Friend or Foe: Subjective Expected Relative Similarity as a Determinant of Cooperation." Journal of Experimental Psychology: General, 138(3), 2009, 341.

--. "Similarity or Reciprocity? On the Determinants of Cooperation in Similarity-Sensitive Games." Psychological Inquiry, 23, 2012, 48-54.

Fischer, I., A. Frid. S. J. Goerg. S. A. Levin. D. I. Rubenstein. and R. Selten. "Fusing Enacted and Expected Mimicry Generates a Winning Strategy that Promotes the Evolution of Cooperation." Proceedings of the National Academy of Sciences, 110(25), 2013. 10229-33.

Fosco, C., and F. Mengel. "Cooperation through Imitation and Exclusion in Networks." Journal of Economic Dynamics and Control, 35, 2011, 641-58.

Glockner. A., S. J. Goerg, and I. Fischer. "Similarity and Cooperation." Mimeo. 2014.

Goette. D. H., and S. Meier. "The Impact of Group Membership on Cooperation and Norm Enforcement: Evidence Using Random Assignment to Real Social Groups." American Economic Review, 96(2), 2006, 212-16.

Greiner, B. "An Online Recruitment System for Economic Experiments." MPRA Working Paper No. 13513, 2004.

Hilgers, D., and M. Wibral. "How Malleable Are Choice Brackets? The Case of Myopic Loss Aversion." Working Paper. 2014.

Hoffman, E., K. McCabe, K. Shachat, and V. Smith. "Preferences, Property Rights, and Anonymity in Bargaining Games." Games and Economic Behavior, 7(3), 1994, 346-80.

Kahneman, D., and A. Tversky. "Choices, Values and Frames." American Psychologist, 39, 1984. 341-50.

Koszegi, B., and M. Rabin. "Reference-Dependent Consumption Plans." American Economic Review, 99, 2009, 909-36.

Likert, R. "A Technique for the Measurement of Attitudes." Archives of Psychology, 22(140), 1932. 55.

Michaels. S. "Nelly Furtado to Donate Sim Gaddafi Concert Fee to Charity." 2011. Accessed May 2013. https:// www.theguardian.com/music/2011/mar/01/nellyfurtado-gadaffi-concert-fee

Mussweiler, T., and A. Ockenfels. "Similarity Increases Altruistic Punishment in Humans." Proceedings of the National Academy of Sciences, 110(48), 2013, 19318-23.

Oxoby, R. J., and J. Spraggon. "Mine and Yours: Property Rights in Dictator Games." Journal of Economic Behavior & Organization, 65(3), 2008. 703-13.

Rabin, M., and G. Weizsacker. "Narrow Bracketing and Dominated Choices." American Economic Review, 4, 1999, 1508-43.

Tajfel, FI. "Experiments in Intergroup Discrimination." Scientific American, 223(5), 1970, 96-102.

Tajfel, H., and J. Turner. An Integrative Theory of Intergroup Conflict. The Social Psychology of Intergroup Relations. Monterey, CA: Brooks/Cole. 1979.

Thaler, R. "Toward a Positive Theory of Consumer Choice." Journal of Economic Behavior & Organization. 1, 1980. 39-60.

--. "Mental Accounting and Consumer Choice." Marketing Science, 12, 1985. 183-206.

--. "Mental Accounting Matters." Journal of Behavioral Decision Making, 12, 1999, 183-206.

Zizzo, D. J. "Experimenter Demand Effects in Economic Experiments." Experimental Economics, 13. 2010. 75-98.

SEBASTIAN J. GOERG, DAVID B. JOHNSON and JONATHAN D. ROGERS

* We are grateful for valuable comments provided by the referees and participants at the University of Calgary seminar series. All mistakes are our own.

Goerg: Assistant Professor, Department of Economics, Florida State University, Tallahassee. FL 32306-2180. Phone 850-544-7950. Fax 850-644-4535, E-mail sgoerg@fsu.edu

Johnson: Assistant Professor. Department of Economics. Finance and Marketing, University of Central Missouri, Warrensburg, MO 64093. Phone 336-639-2190, Fax 660-543-8465, E-mail djohnson@ucmo.edu

Rogers: Visiting Assistant Research Professor, Social Science Experimental Laboratory, New York University Abu Dhabi (NYUAD). Abu Dhabi. UAE. Phone +26285445, E-mailjonathan.rogers@nyu.edu

(1.) Related is the topic of narrow bracketing in which individuals view each investment decision in isolation and not in the context of other choices (e.g., Kahneman and Tversky 1984; Rabin and Weizsacker 1999: Thaler 1985). Resulting decisions seem to be mistakes (Hilgers and Wibral 2014) rather than outcomes due to preferences as suggested by Koszegi and Rabin (2009).

(2.) In-kind grants are vouchers that can only be spent on one good, while cash grants are lump-sum vouchers that can be spent without this restriction.

(3.) In addition, dictators are less generous if they receive money from both sources compared to receiving the same endowment from only one source (Clingingsmith 2015).

(4.) Furthermore, due to homophily. members of endogenously formed groups tend to be very similar (see Currarini and Mengel 2016 and Fosco and Mengel 2011 for work by economists on this topic).

(5.) We use the dictator game because it is a well-established workhorse (Engel 2011). nonstrategic riskless game. and. therefore, has a simple decision/payoff structure.

(6.) The experiment is conducted in the Experimental Social Sciences (xs/fs) Laboratory at Florida State University. The experiment is run using zTree Fischbacher (2007). Subjects are all students at Florida State University and are recruited with the ORSEE recruitment system (Greiner 2004).

(7.) We thank one referee for suggesting the terms perceived and objective similarity.

(8.) All participants in the experiments are FSU students, thus, they all share at least one attribute with the FSU group. Most participants are employed in part-time jobs, thus most of them share at least one attribute with the AMT group. Finally, most students do not support the views of the white supremacists of the STORM group. This ranking was confirmed by a poll conducted with students at a different university in which responders assigned an average similarity rating of 6.04 to the comparison with the FSU group, of 4.72 for AMT group, and of 3.12 for STORM group. For more details, refer to the Appendix.

(9.) For example, an individual student may not feel similar to his or her fellow students because of relative differences in their age. nationality, religiosity, et cetera.

(10.) Exact question wording can be found in the Appendix.

(11.) As additional controls, we elicited subjects' age. gender. religion, political affiliation, and how much they would trust an average person at the source. More detailed analysis regarding these variables is available upon request.

(12.) See the Appendix for a detailed description of the payment process.

(13.) However, it is important to mention that the influence of source similarity could analogously be incorporated into most models of other-regarding preferences or norm compliance (e.g., Andreoni and Bernheim 2009: Bolton and Ockenfels 2000).

(14.) A natural extension to settings with several attributes would be to think of D as the length of a vector connecting the point that describes the multidimensional attributes of the dictator and with the point describing the source.

(15.) The linear model of inequity aversion results in "extreme predictions" for the dictator game. It either yields no transfers or payoff equalizing transfers, but no transfers in-between. However, as Fehr and Schmidt (1999, 848) note, replacing the linear function of inequity aversion with a concave one would allow for interior solutions. For simplicity, we stick to the linear version of the model.

(16.) Usually, disadvantageous inequity is ignored in dictator games, as inequity aversion provides no reason for dictators to transfer amounts resulting in lower earnings than the ones of the recipient. However, our extension adds an additional trade-off between similarity and disadvantageous inequity aversion. Therefore, it needs to be considered.

(17.) We think it is reasonable to assume that a dictator with similarity concerns also has other-regarding preferences. However, one could also imagine a dictator with similarity concerns but not inequity aversion. In this case, the dictator would either keep the whole endowment or transfer the whole endowment to the recipient. The dictator would transfer the whole endowment if [gamma]D3 > I.

(18.) The breakdown of subjects' political affiliations is as follows: 29.6% of subjects in our experiment indicate that they are republicans, 35.2% indicate they are democrats, while the remaining 35.2% indicate an "other" political affiliation. The summary of subjects' religion (or lack-thereof) is as follows: 59% indicate they are Christian. 5.6% Jewish, and 20.4% Atheist with the remaining 14.8% indicating an "other" religion.

(19.) Our mean contributions do not differ significantly from the average reported in the meta-study (two-sided two-sided Fisher-Pitman permutation test p=.356).

(20.) All pairwise comparisons with a two-sided Fisher Pitman permutation test yield p >.45.

(21.) The perceived similarity variable is calculated by taking the average of OPINION and SIMILAR survey questions. This technique is common and has been in use for over 80 years (Likert 1932).

(22.) See Figure A5 for additional box plots with the two questionnaire items (SIMILAR and OPINIONS).

(23.) Pairwise comparisons between endowment sources confirm this finding (two-sided Fisher-Pitman permutation test, Stormfront vs. Mechanical Turk p <.01, Stormfront vs. Florida State p <.01, and Florida State vs. Mechanical Turk p=.029). The same holds for similarity items SIMILAR and OPINIONS (all comparisons withp<.05).

(24.) As previously mentioned, while the overall distribution differs from the distribution reported in the meta-study by Engel (2011), the distribution of amounts given by dictators with above median similarity to the endowment source looks quite similar to the distribution reported in Engel (2011).

(25.) This correlation can also be confirmed separately for the two questionnaire items on which the similarity score is based: SIMILAR r=-.239 with p=.08 and OPINION r =-.268 with p=.0497. Furthermore, we can already see this relationship within treatments FSU (r=-.523) and STORM (r =-.269).

(26.) We apply Tobit regressions as dictators' transfers are left- and right-censored. Seventeen percent of the data is left-censored and 4% is right-censored.

(27.) Regressions with the objective similarity score rather than source dummies lead to no significant results. However. the sign of the coefficient is in the right direction with increases in objective similarity being negatively correlated with amounts sent by dictators. Results are not reported here but are available upon request.

(28.) Table D1 gives the coefficients of these additional controls.

(29.) Forexample, in 2007 Nelly Furtado donated I million dollars that she had earned by performing a show for Muammar Gaddafi's family (Michaels 2011). TABLE 1 Earnings by Treatment and Role Subjects Dictators Recipients (Dictators) STORM 36(18) 674.72 325.28 AMT 36(18) 666.66 333.33 FSU 36(18) 718.89 281.11 TABLE 2 Summary Statistics for Dictators' Transfers Mean SD SE Min Max STORM 325.3 310.4 73.2 0 1000 AMT 333.3 182.3 42.9 0 500 FSU 281.1 213.5 50.3 0 500 TABLE 3 Tobit Regressions on Dictators' Transfers Transfer (1) (2) (3) STORM -5.71 (102.09) FSU -71.45 (82.77) SIMILAR -27.80 ** (13.70) OPINION -28.19 ** (13.23) Perceived similarity Constant 311.6 *** 450.3 *** 454.8 *** (68.95) (86.46) (84.81) Additional controls -- -- -- Session dummies -- -- -- Observations 54 54 54 F-test p-value p = .678 p = .043 p = .035 Transfer (4) (5) (6) STORM FSU SIMILAR OPINION Perceived similarity -30.08 ** -77.08 *** -92.97 *** (13.93) (22.78) (22.48) Constant 464.5 *** 585.5 528.3 (87.95) (536.5) (471.5) Additional controls -- Yes Yes Session dummies -- -- Yes Observations 54 54 54 F-test p-value .032 p = .0028 p = .0009 Notes: Robust standard errors in parentheses. Given the censored structure of the data at zero and ten dollars, coefficients are estimated with Tobit regressions. Additional controls are gender, age, religion, party affiliation, and trust towards the source. *** p < .01, ** p < .05, * p < . 1. TABLE A1 Source Player Choice Option A Option B ME: 10.00 Me: 12.00 OTHER: 8.00*z 10.00*z TABLE D1 Dictators' Transfers Transfer (1) (2) (3) Similarity score -30.08 ** -77.08 *** -92 97 *** (13.93) (22.78) (22.48) Female 25.52 2.46 (84.22) (79.22) Age -15.80 -19.41 (24.03) (20.16) Trust 58.19 ** 77.53 *** (24.15) (23.87) GOP -86.02 -135.70 (96.12) (104.8) DEM 89.47 71.91 (82.92) (73.2) Libertarian -81.97 -53.38 (161.5) (138.4) Jewish 219.40 * 279.10 ** (115.80) (104.60) Christian 139.40 163.1 (120.60) (96.88) Other religion 179.50 212.70 * (111.50) (105.50) Session 2 160.40 * (88.40) Session 3 292.40 ** (109.80) Session 4 153.30 * (80.67) Session 5 174.50 (110.10) Constant 464.50 *** 585.50 528.30 (87.95) (536.50) (471.50) Observations 54 54 54 F-test p=.032 p=.0028 p=.0009 Notes: Robust standard errors in parentheses. Given the censored structure of the data at zero and ten dollars, coeffi- cients are estimated with Tobit regressions. Dropped category for religion is atheist and for political affiliation other. GOP. Republican; DEM, Democrat. *** p < 0.01, ** p < .05, * p < .1.