Pro-social or anti-social, or both? A within- and between-subjects study of social preferences

Accepted Manuscript

Pro-social or Anti-social, or Both? A Within- and Between-subjects Study of Social Preferences Le Zhang , Andreas Ortmann PII: DOI: Reference:

S2214-8043(16)30009-X 10.1016/j.socec.2016.03.001 JBEE 181

To appear in:

Journal of Behavioral and Experimental Economics

Received date: Revised date: Accepted date:

7 December 2014 2 March 2016 4 March 2016

Please cite this article as: Le Zhang , Andreas Ortmann , Pro-social or Anti-social, or Both? A Withinand Between-subjects Study of Social Preferences, Journal of Behavioral and Experimental Economics (2016), doi: 10.1016/j.socec.2016.03.001

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

ACCEPTED MANUSCRIPT

Pro-social or Anti-social, or Both? A Within- and Between-subjects Study of Social Preferences Le Zhang and Andreas Ortmann Abstract: The literature on dictator [D] and joy-of-destruction [JoD] games demonstrates that some people can be nice and some people can be nasty. We study, by way of an

CR IP T

experiment with between-subjects and within-subjects features, to what extent social preferences are consistent or context dependent. We find that participants’ giving amount in D games, and the amount they destroy in JoD games, depend on the choice set. While the choice set strongly affects participants’ giving decisions, its effect on participants’ destruction

AN US

decisions is much weaker. We observe inconsistent choices (giving in D games and destroying in JoD games) for about one in five subjects but also find this mixed-motive preference dramatically reduces when the choice sets of standard D and JoD games are enlarged. Most of our participants are selfish although they also tend to make choices that

M

increase social welfare when given the opportunity. The Machiavellian attitudes we elicited

ED

predict the giving amount in D games but not the destruction amount in JoD games.

PT

Keywords: Dictator game  Joy-of-destruction game  Altruism  Nastiness  Mach-IV test

CE

JEL Classification A13  C91 D03 D64

AC

Le Zhang (corresponding author): Department of Economics, School of Economics and Finance, Curtin Business School, Curtin University, Australia. Email: [email protected] Andreas Ortmann: School of Economics, UNSW Business School, University of New South Wales, Australia. Email: [email protected] Acknowledgements: We thank participants at the 7th ANZ Workshop in Experimental Economics (Perth, Australia), the 2013 Asia-Pacific ESA Conference (Tokyo, Japan) and at seminars at CERGE-EI (Prague, Czech Republic), the University of New South Wales (Sydney, Australia), the University of Vienna, the University of Nijmegen, and the Max-Planck Institute for Human Development (Berlin, Germany) for probing questions and helpful comments and suggestions. We specifically would like to thank Michal Bauer, Mirta Galesic, Lata Gangadharan, Philip Grossman, Ralph Hertwig, Kai Konrad, Jianying Qiu, Leonidas Spiliopoulos, James Tremewan, Jean-Robert Tyran, Jana Vyrastekova, Jan Woike, Daniel Zizzo, and two anonymous referees for this journal for their comments. We appreciate Jeanette Deetlefs’ assistance in running the experiments. We thank the UNSW Business School for partial support of this project via a small project grant to LZ and research funds to AO. All mistakes are ours. AO thanks CERGE-EI, Prague, Czech Republic, and the Max-PlanckInstitute for Tax Law and Public Finance, Munich, Germany, for their hospitality.

1

ACCEPTED MANUSCRIPT I.

INTRODUCTION

Initial results from dictator (D), ultimatum, and trust games were widely interpreted as people being more altruistic than economic theory had traditionally assumed (see Camerer, 2003). Since then, numerous new studies have demonstrated strikingly the dependence of experimental outcomes in D games on various design and implementation characteristics (see,

CR IP T

for example, Dana et al., 2006; Dana et al., 2007; Grossman and Eckel, 2012) including the choice set (see, for example, Andreoni and Bernheim, 2009; Bardsley, 2008; Cappelen et al., 2013; Krupka and Weber, 2013; Lazear et al., 2012; List, 2007; Zhang and Ortmann, 2014). In addition, Andreoni and Miller (2002) have demonstrated that pro-social preferences such

AN US

as altruism are price sensitive, a finding unlikely to surprise economists but one that contradicts the concept of a “primitive” (Berg et al., 1995; Ortmann et al., 2000). The results of List (2007) confirm the price-sensitivity of altruism in that his participants jettison moral

M

scruples when they become too costly.

More recently, an emerging literature on joy-of-destruction (JoD) games demonstrates

ED

that some people can be outright nasty in the sense that they are willing to reduce other

PT

participants' endowment although they do not benefit directly or might even have to pay for it (Abbink and Herrmann, 2011; Abbink and Sadrieh, 2009; Sadrieh and Schröder, 2012; Zizzo

CE

and Fleming, 2011).1 Not surprisingly, such anti-social preferences, and the JoD outcomes they bring about, are also dependent on various design and implementation characteristics.

AC

In this study, using variations of both D and JoD games, we examine whether and to

what extent social preferences (i.e., pro-social, anti-social, and preferences that are mixed) are consistent or context dependent. The studies closest to ours are Zizzo and Fleming (2011) and Sadrieh and Schröder (2012). Zizzo and Fleming (2011) find that social pressure, while its effects on public good contributions are mixed, predicts giving in their D games and 1

As we will show below, the JoD games are in interesting ways related to the not-so-nice take-options in D games that authors such as List (2007) have recently studied.

2

ACCEPTED MANUSCRIPT destruction in their costly JoD games (which they call money-burning games).2 Sadrieh and Schröder (2012) also study giving and destruction choices in their within-subject give-ordestroy games, in which subjects can increase or decrease recipients’ endowment by giving up parts of their own endowment. One of their interesting findings is that one third of their population exhibits both pro-social and anti-social preferences which they interpret as “desire

CR IP T

to influence others”. Our study differs from these earlier papers in many design and implementation details, although all three papers provide within-subject examinations of prosocial, anti-social, and other preferences. Among the important differences are: First, the destruction choices in the other two papers are costly, with the cost-effectiveness ratio being

AN US

1:3 in Zizzo and Fleming (2011) and 1:1 in Sadrieh and Schröder (2012). In contrast, the destruction choices in our JoD games are costless. We chose to examine costless nasty behavior in JoD games because we believed it to be a common real-life feature. Second, we

M

designed zero-sum D games and non-zero-sum JoD games to study a wider range of social preferences, such as completely pro-social, completely anti-social, completely self-interested,

ED

inconsistent, social-welfare maximizing, and so on. The examination of social preferences

PT

under different choice sets, for both D and JoD games, can reveal not only the price sensitivity of pro-social or anti-social preferences in the sense of List (2007) separately, but

CE

also whether social preferences in both games are consistent or context dependent. Third, and following well-established precedents (e.g., Gunnthorsdottir et al., 2002), we also investigate

AC

whether personality traits (i.e., the Mach-IV test) predict choices in D and JoD games. In line with previous results (namely, List, 2007), we find considerable context effects of

choice sets on giving decisions in D games. When take-options are included in D games, fewer participants choose to give money and they give less money on average. We also find 2

There are numerous differences between our studies, from the specifics of the game, to the action spaces, and the recruitment and selection of participants. Sadrieh and Schröder (2012), for example, recruit students leaving a cafeteria which we believe to be a potentially problematic approach; Zizzo and Fleming (2011) screen their subjects for English-language proficiency.

3

ACCEPTED MANUSCRIPT that fewer participants choose to destroy money in JoD games when the choice sets include the options of adding money to recipients’ endowment, although the treatment effects for JoD games are statistically insignificant. We also find that 21 percent of our participants in the Baseline treatment make choices that could be identified as inconsistent. We find, however, statistically significant reductions of the percentage of inconsistent choices when choice sets

CR IP T

are enlarged to be more symmetric around the origin of the endowment allocation. Our finding of a negative relationship between percentage of inconsistency of behavior and size of the action set is novel and warrants attention in future research.

The remainder of our study is organized as follows. In Section II and III we provide

AN US

details of the experimental design and implementation. In Section IV we report the effects of choice sets on giving decisions in D games and destroying decisions in JoD games, and investigate the extent and intensity of inconsistent choices. In Section V we briefly discuss

EXPERIMENTAL DESIGN

ED

II.

M

relevant studies. Section VI contains our conclusion.

We implemented three treatments, “Baseline”, “Treatment$1”, and “Treatment$5” to study

PT

whether social preferences are context dependent. Within each treatment, our participants

CE

made decisions in both D and JoD "game" scenarios specific to the treatment.3 Participants had the same information across treatments. The only difference across treatments was the

AC

different choice sets (see Fig. 1 below for the pictorial description). For the D decisions, our treatments replicate the key treatments in List (2007).

3

Both games do not really deserve the label, as they are not truly interactive; specifically, the second mover (who is in effect a recipient rather than a responder) is always at the mercy of the first mover.

4

ACCEPTED MANUSCRIPT

Treatment$1

Treatment$5

15

15

10

10

10

5 0

Recipient

15

Recipient

5 0

0

5

10

15

D

0 0

Decision maker

5

10

Decision maker

JoD

5

D

JoD

15

CR IP T

Recipient

Baseline

0

5

10

15

Decision maker D

JoD

AN US

Fig. 1 Pictorial description of payoffs for the three treatments4

Baseline Treatment. The D and JoD decisions within each treatment have, as point of departure, the same initial endowment allocation of $10 for dictators and $5 for recipients as

M

previously used in List (2007). The participant who is endowed with $10 is the decisionmaker. In the D game, the decision maker could give, in integer steps, up to $5 of their

ED

endowment to an anonymous recipient, hence the final outcome is one of the integer

PT

outcomes that lie on the line connecting [10, 5] to [5, 10], which below we call the giving segment of the choice set. In the JoD game, the decision maker could destroy, in integer

CE

steps, up to all of the recipient’s $5 initial endowment. The final outcome is one of the integer outcomes that lie on the vertical line from [10, 5] to [10, 0], which below we call the

AC

destruction segment of the choice set. Treatment$1. This treatment is similar to the Baseline treatment; the only difference is

that the choice set is larger. In the D game, the decision maker can give up to $5 or take $1 from the recipient (“take-option”). Thus the choice set of this D treatment includes a $1 takeoption that extends the giving segment. In the JoD game, the decision maker can decrease the 4

Since participants knew that the $5 show-up fee would be independent from decisions taken by themselves or others, Fig 1 excluded the show-up fee.

5

ACCEPTED MANUSCRIPT endowment of the recipient by up to $5 or add $1 to the recipient’s endowment (“addingoption”). Thus, the choice set of this JoD decision includes a $1 adding-option that extends the destruction segment. Hence, the final outcome could be one of the integer outcomes between [11, 4] to [5, 10] in the D game and from [10, 6] to [10, 0] in the JoD game. Treatment$5. This treatment is similar to Treatment$1, the only difference is that the

CR IP T

choice set is even larger. In the D game, the decision maker can now give up to $5 to or take up to $5 from the recipient. The choice set of this D treatment thus includes both giving and taking segments, which are symmetric to the origin of endowment. In the JoD game, the decision maker can decrease the endowment of the recipient by up to $5 or add up to $5 to

AN US

that recipient’s endowment. The choice set of this JoD treatment thus includes both destruction and adding segments, which are also symmetric to the origin of endowment. Hence, the final outcome could be one of the integer outcomes between [15, 0] to [5, 10] in

M

the D game and from [10, 10] to [10, 0] in the JoD game.

Note that decisions in D and JoD games, notwithstanding the symmetry suggested by

ED

Fig. 1, have different social welfare consequences. In the D game scenarios, the give- and

PT

take-options imply a transfer from one participant to the other, hence there are no welfare consequences (the total aggregate amount available to participants remains unchanged). They

CE

are zero-sum games. In contrast, the destruction options as well as the options to have money added in the JoD game scenarios do have welfare consequences. They are non-zero-sum

AC

games. Specifically, the options of having money added increase the welfare of the participants.5 And the destruction options entail a reduction in social welfare. We are interested in treatment effects (differences across treatments, the between-

subjects component of our design), as well as participants’ behaviors within each treatment (the within-subjects component of our design) – i.e., whether participants’ preferences are 5

However, it invites another form of destruction – that of the endowment of the experimenter – although it seems unlikely that subjects think about it in this way (Krupka and Weber, 2013).

6

ACCEPTED MANUSCRIPT pro-social or anti-social, or both. Our study is guided by three hypotheses that are partially inspired by List (2007): Hypothesis 1: The percentage of givers in D games does not change across treatments. The rationale for this hypothesis is the assumption of stable preferences which predicts that, in the aggregate, the proportion of participants who choose to give money to recipients

CR IP T

(the percentage of tangency points of indifference curves on the giving segment) ought not to be affected by the addition of take-options. Under the assumption of stable preferences, the take-options are expected to differentiate only between those that are constrained not to give money in the Baseline treatment.

AN US

Hypothesis 2: The percentage of destroyers in JoD games does not change across treatments.

The rationale is similar for the destruction choices in JoD games. The proportion of

M

participants who choose to destroy recipients’ endowment (the percentage of tangency points of indifference curves on the destruction segment) ought not to be affected by the addition of

ED

adding-options that allow decision makers to increase recipients’ endowment.

treatments.

PT

Hypothesis 3: The percentage of (in)consistent choices does not change across

CE

Since our participants make decisions in both the D and the JoD games, we could theoretically categorize participants according to their choices (see a detailed typology in

AC

Table 3), under the assumption that our participants have stable preferences across both decision situations and that choices would directly reveal types. The common choice sets across treatments are the giving choices in the D games and the destruction choices in the JoD games. Appealing again to the assumption of stable preferences, we expected that the percentage of (in)consistent choices (i.e., where inconsistency is defined as giving in D games and destroying in JoD games) would not change across treatments, although

7

ACCEPTED MANUSCRIPT inconsistent choices were predicted by the sensitivity to social pressure (Zizzo and Fleming, 2011) and what Sadrieh and Schröder (2012) called the “desire to influence others”. III.

EXPERIMENTAL IMPLEMENTATION

All experimental sessions took place in the BizLab of the UNSW Business School. We

CR IP T

recruited 48 participants for each treatment (24 participants in each session and two sessions for each treatment, with the counter-balanced order in the second session, to control for order effects that are concerns in designs with within-subjects components).6 In total, we had 167 participants in 7 sessions which included one session for the double-blind of Treatment$5.

AN US

Average earnings were over $20 per person, including a $5 show-up fee. The experiment lasted less than one hour.

To repeat, in each treatment decision makers had two independent decisions to make,

M

one based on a version of the D game and the other based on the corresponding version of the JoD game. Participants were not informed of the second task until the first task was

ED

completed. The D and JoD game decisions were denoted as Decision 1 and Decision 2 in the instructions (see Appendix A).

PT

The three treatments, for the D as well as the JoD games, constitute a between-subjects

CE

design. Following List (2007), we did not use asset legitimacy (but note the robustness check in his Earnings treatment).7 In each of the two decision scenarios (D and JoD) across three

AC

treatments, all participants were told that their initial endowment was $5. Furthermore, they 6

Because directly relevant precedent studies did not exist, we computed the optimal sample size for our experiment by using the results in Frank (1998) as the alternative hypothesis: approximately 24 participants per treatment were required for aspiration levels of 95% confidence level and 80% statistical power. There were only 47 subjects in the Baseline treatment as some students did not show up in time; overall we had 143 (1x47, 2x48) participants for the three treatments plus 24 more participants for a double-blind version of the third treatment (Treatment$5 – Double-blind). 7 Asset legitimacy means that endowment and earnings are not “manna from heaven” provided by the experimenter, as is typically the case in economics experiments. For there to be asset legitimacy, at least part of the endowment has to be earned. Cherry et al. (2002) confirm that asset legitimacy makes a difference for the level of giving. As we were interested in the differences between treatments, as well as those within treatments, and implementing asset legitimacy in our context would have added another layer of complexity, we decided to eschew it.

8

ACCEPTED MANUSCRIPT were told that, as the decision makers, they would be endowed with yet another $5. They were also told that one of the two independent decisions made during a treatment would be randomly selected as their payoff-relevant decision and that each participant, in addition, would be at the receiving end of another participant’s randomly selected payoff-relevant decision.8 Once all decisions were made, each participant flipped a coin to identify the pay-

CR IP T

off relevant game for them; the participant at the receiving end was anonymously and randomly selected (under a random-permutation protocol) among all participants in the same session.9 After all participants (without knowing other participants’ decisions) randomly selected their pay-off relevant decisions, we asked them to answer a post-experiment survey,

AN US

which was composed of the Mach-IV test and several general demographic questions (like gender, age, major etc.).10 The standard Mach-IV test (see Appendix B for details) included 20 statements which participants were asked to rate on a 5 – point Likert scale ranging from

M

strongly disagree to strongly agree. Following the standard procedure for the administration of the Mach-IV test, participants were not incentivized to answer it. We also added a “catch

ED

question” to gauge whether participants paid attention to the questions. The Mach-IV test was

PT

implemented through Qualtrics. Each page featured only one of the questions and participants

8

CE

could not go back to previous questions. A time delay (12 seconds), about which participants

AC

We were aware that this might induce indirect-reciprocity confounds. A cleaner design would have been to let half of the participants in a treatment always act out the role of dictator and the other half act out the role of recipient; we did not use such design as it would have made some recipients, especially those whose $5 were destroyed by others, quite unhappy. An even cleaner design would have been to match each of the three D scenarios with each of the JoD scenarios. Obviously, some such design (apart from various other problems) would have been a more expensive undertaking. We consider it unlikely that it would have led to significantly different insights. The indirect-reciprocity argument is applicable to all three treatments, and hence is unlikely to affect our interpretation of the treatment effects. 9 They were randomly selected but without replacement in the sense that two people would not be matched twice and a participant could never be a recipient of her- or himself. 10 Since there is considerable evidence on types being a function of demographic characteristics as well as personality traits (see, for example, Paulhus and Williams, 2002), we implemented a well-known and validated assessment instrument, the Mach-IV test, to shed light on the determinants of our participants’ decisions through an assessment of their Machiavellian instincts. Wilson et al. (1996) designed the first Mach test as an assessment instrument for personality traits. The Mach-IV test – see Appendix B for the battery of questions used – has been widely applied and cited both in economics (e.g., prominently Gunnthorsdottir et al., 2002) and in psychology (Christie and Geis, 1970; Harrell and Hartnagel, 1976; Jones and Kavanagh, 1996; Mudrack and Mason, 1995).

9

ACCEPTED MANUSCRIPT were informed, was implemented for each question to discourage participants from rushing through the questions. IV.

RESULTS

i. Giving Decisions Table 1 Aggregate giving behavior

CR IP T

a. Summary statistics and non-parametric tests

Observations

Proportion of givers

Mean

Median

Standard Deviation

Baseline

47

0.64

1.28

1.00

1.30

Treatment$1

48

0.35

0.33

0.00

1.48

Treatment$5

48

AN US

Giving

0.15

-2.48

-4.00

2.92

Table 1 summarizes the results of giving decisions across three treatments. The results from the Baseline treatment were similar to those from other D game experiments (see Camerer,

M

2003): 64 percent of the participants (30 out of 47) gave money to others, and the average

ED

giving was around 25%. When the $1 take-option was added to the choice set, 17 out of 48 participants (35%) gave money to others, with the average giving being close to zero. When

PT

decision-makers could take up to $5, only 7 out of 48 decision makers (15%) gave money to recipients, with the average giving dramatically decreasing to -$2.48.11 The effect on the

CE

median offer was even more dramatic. Hence, giving amount in D games were context

AC

dependent: decision makers gave less money when take-options are included in the choice set. The shift of the distribution of giving amount is illustrated in Fig 2. Our results replicate the results in List (2007).12

11

Recall, that negative giving is taking. And we find that the double-blind treatment does not have an effect on choices in D games. 13% of the participants gave money, with the average giving being -$2.83. The difference between Treatment$5 and Treatment$5 – Double-blind is statistically insignificant. 12 Wilcoxon Mann-Whitney tests are implemented to test the distribution of giving amount across treatments. The results demonstrate statistically significant differences across treatments (at 5% significance level).

10

CR IP T

ACCEPTED MANUSCRIPT

Fig. 2 Histogram of giving decisions across treatments

We use Fisher’s exact tests to investigate more formally the proportion of givers across

AN US

treatments.13 The differences in the percentages of givers across the three treatments are statistically significant, thus we reject Hypothesis 1 that the proportion of people giving positive amounts is the same.14 Fewer decision makers gave positive amounts when take-

M

options were included in the choice set. ii. Destruction Decisions

ED

Table 2 Aggregate destruction behavior

Observations

Proportion of destroyers

Mean

Median

Standard Deviation

Baseline

47

0.28

0.60

0.00

1.26

Treatment$1

48

0.15

-0.29

-1.00

1.09

Treatment$5

48

0.17

-2.15

-4.00

3.46

CE

PT

Destroy

Table 2 summarizes the results of destruction decisions across the three treatments.15 Similar

AC

to their giving decisions, decision makers’ destruction decisions differ across treatments: 13 out of 47 participants (28%) destroyed recipients’ endowment, which was similar to the results in Abbink and Herrmann (2011). However, when decision makers could add money to

13

All the results from Fisher’s exact tests are consistent with those from unconditional Fisher-Boschloo tests which arguably are more appropriate. We report the Fisher’s exact tests since they are widely used and understood. 14 P-values are 0.005, 0.016 and 0.000 for differences between Baseline and Treatment$1, Treatment$1and Treatment$5, Baseline and Treatment$5 respectively. 15 Negative destruction amounts mean that decision makers increase recipients’ endowment (adding-option).

11

ACCEPTED MANUSCRIPT recipients’ initial endowment (and hence increase overall welfare), fewer decision makers behaved in the destructive manner (7 out of 48 and 8 out 48 participants in Treatment$1 and Treatment$5, respectively). In fact, the majority of decision makers increased recipients’ endowment in Treatment$5, and more than 40 percent of them added the full amount, with mean destruction amount being -$2.15.16 We concluded that destruction amount, like giving

CR IP T

amount, were context dependent: decision makers destroyed less money when efficiency gains could be captured, increasingly so as efficiency increased from $1 to $5. The shift of

ED

M

AN US

the distribution of destruction amount is illustrated in Fig 3.17

Fig. 3 Histogram of destruction decisions across treatments

PT

We use Fisher’s exact tests to investigate more formally the proportion of destroyers

CE

across JoD treatments. We find that the differences in destruction decisions are statistically weakly significant between Baseline, Treatment$1 and Treatment$5 (but not between the

AC

latter two treatments). 18 Although fewer decision makers destroyed the recipients’ endowment when adding-options were included in the choice set, the reduced percentage of 16

Recall, that negative destruction means that money is added to recipients’ endowment. The double-blind does not have an effect on choices in JoD games as well. 13% of participants destroyed money, with the average destruction being -$2.5. The difference between Treatment$5 and Treatment$5 –Double-blind is statistically insignificant. 17 Wilcoxon Mann-Whitney tests are implemented to test the distribution of destruction amount across treatments. The results demonstrate statistically significant differences across treatments (at 5% significance level). 18 P-values of one-tailed Fisher’s exact tests are 0.095, 0.50 and 0.148 for differences between Baseline and Treatment$1, Treatment$1and Treatment$5, Baseline and Treatment$5 respectively. The treatment effects are statistically insignificant if we use two-sided tests.

12

ACCEPTED MANUSCRIPT destroyer – from 28% in the Baseline treatment to 15% and 17% in Treatment$1 and Treatment$5 – is statistically insignificant. We thus cannot reject Hypothesis 2 that the proportion of destroyers in JoD games is the same across treatments. b. Inconsistency of pro-social and anti-social behavior

CR IP T

By implementing D and JoD games with enlarged choice sets, we can study a wider range of choices and categorize them into different types (see Table 3).19 Across the three treatments, decision makers choose whether to give money, and how much, in the D games, and whether to destroy, and how much of, recipients’ endowment in the JoD games. We do not concern

AN US

ourselves with status-quo behavior (participants not making changes to initial endowment in both D and JoD games) as it is difficult to interpret the underlying preferences: Participants may prefer the initial endowment allocation or they may be constrained by the choice set,

M

hence choose the corner solutions. We observe dramatic shifts of types of choices, in particular for inconsistent choices (see Table 4). In the Baseline treatment, 21 percent of

ED

decision makers’ choices were inconsistent as they gave money to recipients but also destroyed recipients’ endowment.20 The proportion of participants belonging to this type

PT

(T3) decreases when decisions in the “negative domain” are included (i.e., take-options as

CE

negative giving and adding-options as negative destruction). The proportion of inconsistent

AC

choices dramatically reduces to 10% in Treatment$1 and 4% in Treatment$5, when choices

19

Not all of the types can be fully observed in the Baseline treatment, in which D game includes only the giving segment and JoD game includes only the destruction segment. Therefore, we focus on the common type (T3) across treatments. 20 The Spearman product-moment correlation coefficient of the Baseline treatment is 0.13, with a p-value of 0.39. This result differs from the statistically significant correlation of 0.341 in Zizzo and Fleming (2011). The different result might be due to differences in experimental design and implementation. For instance, the destruction choice is costless in our setting, while it is costly in Zizzo and Fleming (2011) where the transfer price is 1/3 for both D and JoD (‘money-burning’) games. The Spearman correlation coefficients of the other two treatments in our experiment are not comparable as a positive Spearman correlation coefficient can reflect social-welfare maximizing behavior as well (i.e., take money in the D game and increase recipients’ endowment in the JoD game). We therefore prefer to discuss the types of choices rather than the Spearman product-moment correlation coefficient.

13

ACCEPTED MANUSCRIPT sets expand to the negative domain.21 These decreases of inconsistent choices are statistically significant and we hence reject our Hypothesis 3 that the percentage of inconsistent choices does not change across treatments.

AC

PT

CE

Treatment$1 & Treatment$5

ED

M

AN US

Baseline

Table 3 Categorization of types according to participants’ choices Types Give Destroy Description Status quo. Decision makers do not change the T0 0 0 default endowment allocation. Pro-social. Decision makers give money to recipients and they do not destroy recipients’ endowment. (Included are types – T1-2 and T1-3 – that are here constrained by the choice set. See explanations under T1 + 0 Treatment$1 & $5.) Anti-social. Decision makers do not give money to recipients and they destroy recipients’ endowment. (Included are types – T2-2 and T2-3 – that are here constrained by this choice set. See explanations T2 0 + under Treatment$1 & $5.) Inconsistent. Decision makers make inconsistent choices – give money to recipients but destroy T3 + + recipients’ endowment. Status quo. Decision makers do not change the T0 0 0 default endowment allocation. Warm-glow pro-social. Decision makers give money to recipients, and they neither destroy nor add T1 + 0 money to recipients’ endowment. Completely pro-social. Decision makers give money to recipients, and they also add money to recipients’ T1-2 + endowment. Partially pro-social. Although decision makers neither give money to nor take money from recipients, they add money to recipients’ endowment T1-3 0 when it is costless for them to do so. Cold-prickle anti-social. Decision makers do not take money from recipients, but they reduce T2 0 + recipients’ endowment. Completely anti-social. Decision makers not only take money from recipients but also reduce T2-2 + recipients’ endowment. Self-interested. Decision makers take money from recipients to maximize their own payoff, but they do not reduce recipients’ endowment if they cannot T2-3 0 benefit from it. Inconsistent. Decision makers make inconsistent choice – give money to recipients but reduce T3 + + recipients’ endowment. Social-welfare maximizing. Decision makers take money from recipients and they also add money to T4 recipients’ endowment.

CR IP T

Treatment

21

The 4% is on the level of the noise one often finds in experiments (Moon and Martin, 1996).

14

ACCEPTED MANUSCRIPT The intensity of options in the negative domain also matters. While the options in the negative domain (i.e., take $1 in D games and add $1 in JoD games) number only one fifth of what they are in the positive domain (i.e., give up to $5 in the D game and destroy up to $5 in the JoD game) in Treatment$1, they are the same in the negative and the positive domain for Treatment$5. Namely, in Treatment$5, participants could give or take in dollars up to $5 in

CR IP T

the D game and destroy or add up to $5 in the JoD game. The differences of outcomes for the D game as well as for the JoD game reflect the effects of the strength (“intensity”) of options in the negative domain. In Treatment$1, 25% participants chose options that maximized social welfare (i.e., take money in D games and add money in JoD games). The proportion of

AN US

this type doubles to 50% in Treatment$5. In line with the arguments proposed in List (2007) for the D game, this dramatic increase can also be explained by moral scruples. The proportion of inconsistent choices also decreases further from 10% in the Treatment$1 to 4%

M

in Treatment$5. The differences in distribution of types between Treatment$1 and Treatment$5 are also statistically significant.

Welfare Maximiser T4 0.25 0.50

AC

CE

PT

ED

Table 4 Distribution of types (proportion of each type) across treatments Types Status Pro-social Anti-social Inconsistent Quo T1T1T2T2Treatment T0 T1 T2 T3 2 3 2 3 Baseline 0.30 0.43 0.06 0.21 Treatment$1 0.13 0.10 0.15 0.15 0.02 0.02 0.08 0.10 Treatment$5 0.06 0.02 0.08 0.08 0.02 0.10 0.08 0.04

c. Regression Analysis22 We use the OLS model to examine the effects of demographics on personality traits measured by Mach-IV scores. Departing from much of the literature (and, prominently, Gunnthorsdottir et al., 2002), we do not find any effects of the standard demographic factors on participants’

22

Since the choice set varies across treatments in which different lower limits were censored, we also use a Tobit model separately for the three treatments. The results are robust under Tobit models.

15

ACCEPTED MANUSCRIPT Mach-IV scores: neither gender, age nor culture is a significant moderator (see the Table 5). This may be due to the subject pool being quite homogeneous: All participants are students from the University of New South Wales and their ages cluster around 22. Table 5 Giving/destruction decisions OLS regressions Mach-IV Score Mach-IV Score

Years Allowance Careless Double-blind Developed Country

Treatment$5

Treatment$1*JoD first

CE

Treatment$5*JoD first

AC

N

(Pseudo) R-sq

0.00

-0.03

0.11*

(0.02)

(0.03)

(0.01)

(0.00)

(0.03)

(0.05)

1.50

0.18

-0.54

0.08

-0.05

-0.28

-2.08**

(1.12)

(0.36)

(0.39)

(0.09)

(0.07)

(0.36)

(0.80)

0.05

-0.03

-0.05

-0.00

-0.01

0.01

0.19

(0.21)

(0.08)

(0.06)

(0.02)

(0.01)

(0.09)

(0.16)

-0.14

0.36

-0.41

-0.03

-0.09

-0.04

-1.91

(1.96)

(0.54)

(0.50)

(0.14)

(0.12)

(0.85)

(1.68)

-0.07

0.02

-0.01

0.01

0.01

0.00

0.02

(0.12)

(0.05)

(0.03)

(0.01)

(0.01)

(0.05)

(0.11)

0.01

0.00

0.00

0.00

0.00

0.00

-0.00

(0.00)

(0.00)

(0.00)

(0.00)

(0.00)

(0.00)

(0.00)

-1.34

0.80+

0.25

0.27*

0.08

0.11

0.86

(1.43)

(0.42)

(0.50)

(0.11)

(0.10)

(0.42)

(0.72)

-0.81

-1.37

-1.94+

-0.17

-0.14+

-0.69

-0.11

(1.81)

(0.83)

(1.05)

(0.13)

(0.08)

(0.68)

(0.67)

-0.38

-0.32

-0.10

-0.15

-0.12

0.88

1.88*

(1.26)

(0.40)

(0.41)

(0.10)

(0.09)

(0.63)

(0.76)

-1.50

-1.53***

-0.91+

-0.38***

-0.20

-1.23+

3.84**

(1.80)

(0.42)

(0.47)

(0.09)

(0.12)

(0.67)

(1.25)

-0.01

-3.30***

-0.96

-0.51***

0.10

-0.19

-0.08

(1.88)

(0.70)

(0.82)

(0.11)

(0.12)

(0.66)

(1.13)

-0.23

-1.11**

0.39

-0.32*

0.20

-1.26*

-1.44+

(1.87)

(0.41)

(0.42)

(0.13)

(0.13)

(0.51)

(0.84)

1.62

1.23*

-0.28

0.41*

0.08

2.19*

-6.08***

(2.67)

(0.58)

(0.57)

(0.20)

(0.22)

(1.06)

(1.60)

0.93

-0.57

-3.84***

-0.01

3.52***

(2.65)

(0.93)

(0.92)

(0.24)

(1.05)

56.59***

5.95*

1.93

(4.71)

(2.30)

(1.90)

166

166

166

166

142

0.065

0.493

0.372

0.254

0.129

PT

JoD first

_cons

Destruction

-0.01+

ED

Treatment$1

Giving

0.00

CR IP T

Australia

Destruction

-0.07**

AN US

Age

Truncated

Giving

M

Male

Probit Whether to Destroy

Whether to Give

57

31

Notes: Marginal effects are reported for Probit model; *** p< 0.001; ** p< 0.01; * p< 0.05; + p<0.1

We also examine participants’ Mach-IV scores, treatment dummies and other demographic characteristics on giving decisions by the OLS model: Participants who have higher Mach-IV scores gave less money at the 1% significance level, although the effect size

16

ACCEPTED MANUSCRIPT (-0.07 in the OLS regression) is small. 23 Our result is consistent with the result of Gunnthorsdottir et al. (2002) where high-Machs are observed to be more apt to behave unethically than low-Machs. There is neither a gender effect nor a culture effect on giving amount in our experiment. Participants who incorrectly answered our catch-question also gave more money, although the difference was weakly statistically significant. However,

CR IP T

except for treatment effects, none of these factors (specifically, being high-Machs) have a statistically significant effect on destruction decisions. The treatment effects for giving amount and destruction amount remain statistically significant after controlling for the demographic factors. We also find the order effects on decisions in D games: Participants

AN US

gave more money and took less money when D game is played first for Baseline treatment and Treatment$5.24 When the order of decisions is controlled, the treatment effects on giving amount become more significant when D decisions are elicited first.

M

An interesting result is that, when D decisions in Treatment$5 were elicited first, participants continued to destroy at a comparatively high rate even after they took money

ED

from others. We can think of two possible explanations: Firstly, randomization was not

PT

perfect and those participants were unusually nasty on average. Given that we do not find a difference in Mach-IV scores, that explanation seems unlikely. Secondly, participants

CE

enjoyed the power of destroying others’ endowment although they could not benefit from it. Why did participants not destroy as much when JoD was elicited first? Although we stressed

AC

in the instructions that their two decisions were independent, we conjecture that the uncertain nature of the second decision – recall that they only knew that there was a second decision but no specifics – might have played a role and made them more hesitant. Specifically, we conjecture that some of our participants might have thought that the second decision could

23

However, the coefficient only means the difference in giving amount when the Mach-IV scores are increased by one unit. Hence, the giving decisions might be very different if participants’ Mach-IV scores differ a lot. 24 The linear combination of the JoD first and Treatment$5*JoD first parameters is statistically significant.

17

ACCEPTED MANUSCRIPT entail some form of retaliation (Jones and Kavanagh, 1996). As a matter of fact, none of the 24 participants engaged in destruction activity when JoD decisions were elicited first, indicating that – if our conjecture is true – that fear must have been considerable. We have implemented non-parametric Fisher’s exact tests for the hypotheses about the proportion of participants who give money in D games and those who destroy money in JoD

CR IP T

games. The results from Probit regressions confirm our findings from the Fisher’s exact tests.25 While the take-options in D games dramatically reduce the percentage of givers in D games, the effects of the adding-options in JoD games are much weaker.26

Among those participants who reduced the recipients’ endowment in JoD games, males

AN US

and those with higher Mach-IV scores destroyed even more money, with the effect being statistically significant (see the Truncated regression on destruction in Table 5). Participants destroyed more money in Treatment$1 compared to Baseline treatment, but they destroyed

M

dramatically less when JoD decisions were elicited first.

DISCUSSION

ED

V.

The literature on asset legitimacy (see, for example, Cherry et al., 2002, and also the Earnings

PT

treatment in List, 2007) suggests that our results would be affected by it; essentially the

CE

effects would be reduced and we would see less giving (and destruction). Asset legitimacy could, for example, induce a feeling of entitlement towards the endowment that could be

AC

given or taken. Interestingly, how exactly entitlements are implemented is important. Cherry et al. (2002) find a powerful effect on giving when dictators earn their endowment. Cappelen et al. (2013) essentially replicate List (2007), but add asset legitimacy for the initial

25

24 observations are missing because Treatment$5*JoD first perfectly predicts the lack of destruction behavior (nobody destroyed others’ endowment when JoD is played first). Hence the treatment effects on decisions of whether to destroy cannot be fully reflected in this regression. 26 Recall that the treatment effects on the percentage of destroyers in JoD games are statistically insignificant by two-sided tests.

18

ACCEPTED MANUSCRIPT endowment through a production phase. They find that two-sided asset-legitimacy does not make a difference. Contradicting the assumption of stable preferences but confirming earlier results on giving reported in List (2007), we demonstrate that the distributions of giving and destruction amount dramatically shift to the left, with the proportion of participants that give positive

CR IP T

amounts decreasing significantly when negative choices are allowed and the proportion of participants who destroy money likewise decreasing although the shift is statistically insignificant. In line with List (2007), we interpret these changes of participants’ choices as cost-sensitive moral scruples, i.e., a form of self-signaling based on internalized social norms.

AN US

Alternatively, if we accept the arguments in Zizzo and Fleming (2011), participants are susceptible to experimental demand effects, or to social pressure, both of which can be interpreted as forms of signaling. While our data do not allow us to tease apart to what extent

M

internalized social norms or social pressure drive the behavior we observe between treatments (see also Levitt and List, 2007; List, 2007), they do show that both self-signaling and/or

ED

signaling are weakened if participants can deviate from defaults (i.e., the initial endowment) in two opposite directions because it is ambiguous what the relevant (internalized) social

PT

norm is. In Treatment$1, it is too “costly” for participants to choose the options in the

CE

negative domain as deviation in one direction is one fifth of what it is in the other direction. In other words, the pressure of choosing to give money to recipients and to destroy recipients’

AC

endowment disappears to a greater extent in Treatment$5, in which participants could deviate from the initial endowment in both directions by the same distance. Since a considerable fraction of our participants is swayed by efficiency

considerations,27 it is natural to ask to what extent inequality aversion might have affected our results. We believe this concern to be moot. We have two sessions with decisions in

27

This result is well established in the literature (Engelmann and Strobel, 2004; Kimbrough and Reiss, 2012).

19

ACCEPTED MANUSCRIPT different orders for each treatment. Participants might add money because of inequality aversion, but inequality aversion could not explain the fact that they took money from others in the D games, which increased the inequality as the decision makers received a higher amount. Thus inequality aversion is not a likely rationale for the addition of money that we observe, while social welfare (efficiency) is.

CR IP T

We found order effects: When JoD games were played first, participants were less altruistic in D game scenarios in Baseline treatment and Treatment$5, and less nasty in JoD game scenario in Treatment$5. It seems that being prompted to answer the JoD decision first makes participants aware of the fact that (if they are at the receiving end) they might have

AN US

some of their endowment destroyed by other decision makers in the first decision scenario, or destruction activity in the first decision scenario might entail some form of retaliation in the second decision scenario. The order effects on giving decisions could also be due to a social

M

norm: participants usually ask for (or expect) return when they are nice to others. Since we only have one session for treatments with decisions in each order, the order effects could also

ED

be confounded with session effects. If there are indeed session effects (which we think are

PT

unlikely in general), they seem more likely for destruction decisions (see the different results in Treatment$5 when JoD is played first). Additional sessions could be run to check for the

CE

session effects (for example, we could run both orders in one session but that would cause other problems); we doubt that this is what drives our results.

AC

Decision makers’ adding behavior weakly indicates that they care more about the

welfare of their peers (i.e., other participants in the same session) than the experimenters’, this result being in line with the results reported in Fréchette (2012) and also Frank (1998). We did not find any gender or age effects on Mach-IV scores, but the Mach-IV score had a statistically significant negative relationship with giving decisions, albeit not for destruction decisions. Participants who obtained high Mach-IV scores were more selfish:

20

ACCEPTED MANUSCRIPT more likely to take money from others, which was beneficial for themselves. However, participants who had higher Mach-IV scores did not destroy more on average. It could be interpreted that participants with high Mach-IV scorers are very cunning, but they are not nasty if it is not beneficial for them to do so. CONCLUSION

CR IP T

VI.

Much of the literature on D game experiments shows that people are quite altruistic (but see Bekkers, 2007; Cherry et al., 2002; Zhang and Ortmann, 2014), while the literature on Joyof-Destruction game experiments implies that some people might be quite nasty in some

AN US

situations. We also know, especially for D game experiments, that altruistic behavior is likely to be affected by institutional changes that – in a well-defined sense – should not make a difference (List 2007). We test such institutional effects for both D and JoD games. By asking participants to make decisions in D and JoD games, we explore the

M

consistency of participants’ social preferences and the robustness across different contexts;

ED

we find considerable context dependence (see the dramatic shifts of distribution for both decisions). For the giving decisions in the D game, the strong and statistically significant

PT

effect in Treatment$5 reflects that moral scruples – whether triggered by self-signaling or

CE

signaling -- get thrown overboard when they become too costly. Our study replicates the results of List (2007) which demonstrate that giving decisions (specifically, whether to give)

AC

are context dependent.28 We also demonstrate that destruction decisions (whether to destroy) are context dependent when destruction decisions (JoD game) are elicited first, even though the destruction decisions are not as sensitive as giving decisions. There are several explanations for the context effects that we and others (Levitt and List, 2007; List, 2007; Sadrieh and Schröder, 2012; Zizzo and Fleming, 2011) have observed.

28

It also seems to suggest that indirect reciprocity, whether negative or positive, that our design makes possible in principle, seems not something we have to worry about much.

21

ACCEPTED MANUSCRIPT Confusion is one of them but it is not a persuasive explanation, as the shifts in behavior seem consistent across treatments (as well as even across studies; see our replication of the key results in List, 2007). In our view, the evidence in favor of this explanation in our experiment is not very persuasive and does not explain well what looks like systematic directional changes in distributions of giving and destruction across treatments. Zizzo and Fleming

CR IP T

(2011) have linked the inconsistent choices they observed with the sensitivity to social pressure, an explanation also mentioned in List (2007) and Levitt and List (2007); Sadrieh and Schröder (2012) have appealed to a “desire to influence others”. As mentioned, our data do not allow us to tease apart to what extent internalized social norms or social pressure drive

AN US

the behavior we observe between treatments although we believe this question to be an interesting issue that warrants further research.

It is well known that the way in which an experiment is conducted is eminently

M

important (Ortmann, 2010; Zizzo, 2010). In addition to controlling for social distance (as we did in a double-blind treatment) or order effects, a number of other robustness tests suggest

ED

themselves: increasing the stakes and studying the effect of asset legitimacy (especially the

PT

entitlement for the money that could be taken) come to mind immediately. We anticipate that

AC

CE

such robustness tests would go in predictable directions.

22

ACCEPTED MANUSCRIPT References: Abbink, K., Herrmann, B., 2011. The Moral Costs of Nastiness. Economic Inquiry 49, 631-633. Abbink, K., Sadrieh, A., 2009. The pleasure of being nasty. Economics Letters 105, 306-308. Andreoni, J., Bernheim, B.D., 2009. Social Image and the 50-50 Norm: A Theoretical and Experimental Analysis of Audience Effects. Econometrica 77, 1607-1636. Andreoni, J., Miller, J., 2002. Giving According to GARP: An Experimental Test of the Consistency of Preferences for Altruism. Econometrica 70, 737-753.

CR IP T

Bardsley, N., 2008. Dictator game giving: altruism or artefact? Experimental Economics 11, 122-133.

Bekkers, R., 2007. Measuring altruistic behavior in surveys: The all-or-nothing dictator game, Survey Research Methods. University of Groningen, pp. 139-144. Berg, J., Dickhaut, J., McCabe, K., 1995. Trust, Reciprocity, and Social History. Games and Economic Behavior 10, 122-142.

AN US

Camerer, C.F., 2003. Behavioral Game Theory: Experiments in Strategic Interaction. Princeton University Press, Princeton. Cappelen, A.W., Nielsen, U.H., Sørensen, E.Ø., Tungodden, B., Tyran, J.-R., 2013. Give and take in dictator games. Economics Letters 118, 280-283. Cherry, T.L., Frykblom, P., Shogren, J.F., 2002. Hardnose the Dictator. American Economic Review 92, 12181221.

M

Christie, R., Geis, F.L., 1970. Studies in Machiavellianism. Academic Press, New York.

ED

Dana, J., Cain, D.M., Dawes, R.M., 2006. What you don’t know won’t hurt me: Costly (but quiet) exit in dictator games. Organizational Behavior and Human Decision Processes 100, 193-201. Dana, J., Weber, R., Kuang, J., 2007. Exploiting moral wiggle room: experiments demonstrating an illusory preference for fairness. Economic Theory 33, 67-80.

PT

Engelmann, D., Strobel, M., 2004. Inequality Aversion, Efficiency, and Maximin Preferences in Simple Distribution Experiments. American Economic Review 94, 857-869.

CE

Frank, B., 1998. Good news for experimenters: subjects do not care about your welfare. Economics Letters 61, 171-174. Fréchette, G.R., 2012. Session-effects in the laboratory. Experimental Economics 15, 485-498.

AC

Grossman, P., Eckel, C., 2012. Giving versus taking: a “real donation” comparison of warm glow and cold prickle in a context-rich environment. Monash University, Department of Economics. Gunnthorsdottir, A., McCabe, K., Smith, V., 2002. Using the Machiavellianism instrument to predict trustworthiness in a bargaining game. Journal of Economic Psychology 23, 49-66. Harrell, W.A., Hartnagel, T., 1976. The Impact of Machiavellianism and the Trustfulness of the Victim on Laboratory Theft. Sociometry 39, 157-165. Jones, G.E., Kavanagh, M.J., 1996. An Experimental Examination of the Effects of Individual and Situational Factors on Unethical Behavioral Intentions in the Workplace. Journal of Business Ethics 15, 511-523. Kimbrough, E.O., Reiss, J.P., 2012. Measuring the Distribution of Spitefulness. PLoS ONE 7, e41812.

23

ACCEPTED MANUSCRIPT Krupka, E.L., Weber, R.A., 2013. Identifying social norms using coordination games:Why does dictator-game sharing vary? Journal of the European Economic Association 11, 495-524. Lazear, E.P., Malmendier, U., Weber, R.A., 2012. Sorting in Experiments with Application to Social Preferences. American Economic Journal: Applied Economics 4, 136-163. Levitt, S.D., List, J.A., 2007. What Do Laboratory Experiments Measuring Social Preferences Reveal about the Real World? The Journal of Economic Perspectives 21, 153-174. List, J.A., 2007. On the Interpretation of Giving in Dictator Games. Journal of Political Economy 115, 482-493.

CR IP T

Moon, P., Martin, A., 1996. The search for consistency in economic search. Journal of Economic Behavior & Organization 29, 311-321. Mudrack, P.E., Mason, E.S., 1995. Extending the Machiavellianism construct: A brief measure and some unexplored relationships. Journal of Social Behavior and Personality 10, 187-200.

Ortmann, A., 2010. The Way in Which an Experiment is Conducted is Unbelievably Important': On the Experimentation Practices of Economists and Psychologists, in: E.H.Witte, Gollan, T. (Eds.), Tagungsband zum 25. Hamburger Symposion -Schwerpunktthema: Sozialpsychologie und Ökonomie, Pabst Verlag, pp. 38-55.

AN US

Ortmann, A., Fitzgerald, J., Boeing, C., 2000. Trust, reciprocity, and social history: A re-examination. Experimental Economics 3, 81-100.

Paulhus, D.L., Williams, K.M., 2002. The Dark Triad of personality: Narcissism, Machiavellianism, and psychopathy. Journal of Research in Personality 36, 556-563. Sadrieh, A., Schröder, M., 2012. The desire to influence others. FEMM Working Paper No. 12027.

M

Wilson, D.S., Near, D., Miller, R.R., 1996. Machiavellianism: A Synthesis of the Evolutionary and Psychological Literatures. Psychological Bulletin 119, 285-299.

ED

Zhang, L., Ortmann, A., 2014. The effects of the take-option in dictator-game experiments: a comment on Engel’s (2011) meta-study. Experimental Economics 17, 414-420. Zizzo, D.J., 2010. Experimenter demand effects in economic experiments. Experimental Economics 13, 75-98.

AC

CE

PT

Zizzo, D.J., Fleming, P., 2011. Can experimental measures of sensitivity to social pressure predict public good contribution? Economics Letters 111, 239-242.

24

ACCEPTED MANUSCRIPT Appendix A: Instructions Welcome to this experiment on individual decision-making. Please do not talk during the experiment. If you have a question, please raise your hand and an experimenter will come to your carrel and answer your question. Please also turn off your

CR IP T

mobile phone, or other electronic gadgets, now.

Anonymity

Your decisions in this experiment will not be revealed to anyone.

AN US

Total payment:

In this experiment, each of you will be paid $5 for having shown up on time. This show-up fee is independent of any decision that you make in the experiment that follows. You will

M

be able to earn additional dollars through the decisions we ask you to make in Part I. [Next Page]

ED

PART I:

PT

You will be asked to make two independent decisions. In each of the two decision situations, you are the decision maker (DM). After you have made your decisions, we will

CE

ask you to flip a coin. The result of the toss will determine which decision is payoffrelevant (if it is head, your payoff-relevant decision is decision 1; otherwise your payoff-

AC

relevant decision is decision 2). Hence you can affect your earnings directly through your decisions.

Your payoff-relevant decision might also affect another person (from participants in this experiment) that you will be matched with, and you will also receive a payment which is determined through another participant’s decision.

25

ACCEPTED MANUSCRIPT

So your earnings for this part is your payment as a decision maker resulting from your payoff-relevant decision, plus the payment as a recipient resulting from another participant’s payoff-relevant decision. All matches are anonymous and will be done by random permutation. Hence, two If participant A is the recipient of participant B, then

CR IP T

participants will not be matched twice.

participant B will not be the recipient of participant A and each participant would not be the participant of her-or himself.

AN US

[Any questions?]

[We now discuss in more detail the two decisions that you have to make.] [Next Page]

M

In Decision 1, you will be randomly and anonymously matched with another participant. You

ED

are the decision maker (DM) and the other participant is the recipient.

PT

Now imagine that each of you will be given $5. As DM, you will be given an additional $5,

CE

for a total of $10. The other participant, as a recipient, does not have a decision to make.

You can either leave payments unchanged, or give part or all of the additional $5 to the other

AC

participant [Treatment$1: or increase the other participant’s payment by $1; Treatment$5: or take up to $5 from the other participant.] All possible final payments are given in the table below:

[Baseline: Options

How much you Your payment give

26

The other participant’s Payment

ACCEPTED MANUSCRIPT

Treatment$5: Options

-1 0 1 2 3 4 5

11 10 9 8 7 6 5

How much you Your payment give -5 -4 -3 -2 -1 0 1 2 3 4 5

The other participant’s payment 4 5 6 7 8 9 10

CR IP T

How much you Your payment give

CE

PT

Option 1 Option 2 Option 3 Option 4 Option 5 Option 6 Option 7 Option 8 Option 9 Option 10 Option 11

5 6 7 8 9 10

AN US

Option 1 Option 2 Option 3 Option 4 Option 5 Option 6 Option 7

10 9 8 7 6 5

M

Treatment$1: Options

0 1 2 3 4 5

ED

Option 1 Option 2 Option 3 Option 4 Option 5 Option 6

15 14 13 12 11 10 9 8 7 6 5

The other participant’s payment 0 1 2 3 4 5 6 7 8 9 10

[Please study the table now. We will ask you in a couple of minutes to choose your preferred

AC

option online. ]

[Does anyone need more time? Any questions?]

[Now, it is time to make your decision.] [Next Page]

27

ACCEPTED MANUSCRIPT In Decision 2, you will be randomly and anonymously matched with another participant. You are the decision maker (DM) and the other participant is the recipient.

Now imagine that each of you will be given $5. As DM, you will be given an additional $5,

CR IP T

for a total of $10. The other participant, as a recipient, does not have a decision to make.

Your payment remains unchanged; however, you need to decide whether to leave the other participant’s payment unchanged, or decrease part or all of her/his $5 [Treatment$1: or increase the other participant’s payment by $1; Treatment$5: or increase the other player’s

AN US

payment by up to $5]. All possible final payments are given in the table below:

[Baseline:

How much you Your payment decrease

Option 1 Option 2 Option 3 Option 4 Option 5 Option 6

0 1 2 3 4 5

PT

ED

M

Options

CE

Treatment$1: Options

AC

Option 1 Option 2 Option 3 Option 4 Option 5 Option 6 Option 7

10 10 10 10 10 10

How much you Your payment decrease -1 0 1 2 3 4 5

10 10 10 10 10 10 10

28

The other participant’s Payment 5 4 3 2 1 0

The other participant’s payment 6 5 4 3 2 1 0

ACCEPTED MANUSCRIPT Treatment$5: Options

Option 1 Option 2 Option 3 Option 4 Option 5 Option 6 Option 7 Option 8 Option 9 Option 10 Option 11

-5 -4 -3 -2 -1 0 1 2 3 4 5

10 10 10 10 10 10 10 10 10 10 10

The other participant’s payment 10 9 8 7 6 5 4 3 2 1 0

CR IP T

How much you Your payment decrease

AN US

[Please study the table now. We will ask you in a couple of minutes to choose your preferred option online. ]

M

[Does anyone need more time? Any questions?]

ED

[Now, it is time to make your decision.]

CE

[Next Page]

PT

[We are now going around and determine which of your decisions is payoff-relevant.]

[Part II: Online questionnaire

AC

Experimenter B: while experimenter A is preparing your payment, now we ask that you answer the brief questionnaire that you find online. There are no right or wrong answers. We appreciate you reading the questions carefully – please note that we have put in a time delay for each of the questions, so rushing through is not possible. We appreciate you answering the questions honestly and completely. Thank you for your cooperation.]

29

ACCEPTED MANUSCRIPT Appendix B: The Mach-IV test The Mach-IV test includes 20 statements which participants were asked to rate on a 5 – point Likert scale ranging from strongly disagree to strongly agree. The answers are aggregated and mapped into a range that allows classification of types as high- and low- Machs.29 We added a “catch question”30

to gauge whether participants paid attention to the questions.

AC

CE

PT

ED

M

AN US

CR IP T

Q1. Never tell anyone the real reason you did something unless it is useful to do so. Q2. The best way to handle people is to tell them what they want to hear. Q3. One should take action only when sure it is morally right. Q4. Most people are basically good and kind. Q5. It is safest to assume that all people have a vicious streak and it will come out when they are given a chance. Q6. Honesty is the best policy in all cases. Q7. There is no excuse for lying to someone else. Q8. Generally speaking, people won't work hard unless they're forced to do so. Q9. All in all, it is better to be humble and honest than to be important and dishonest. Q10. When you ask someone to do something for you, it is best to give the real reasons for wanting it rather than giving reasons which carry more weight. Q11. Most people who get ahead in the world lead clean, moral lives. Q12. Anyone who completely trusts anyone else is asking for trouble. Q13. The biggest difference between most criminals and other people is that the criminals are stupid enough to get caught. Q14. Most people are brave. Q15. It is wise to flatter important people. Q16. It is possible to be good in all respects. Q17. P.T. Barnum was wrong when he said that there's a sucker born every minute. Q18. It is hard to get ahead without cutting corners here and there. Q19. People suffering from incurable diseases should have the choice of being put painlessly to death. Q20. Most people forget more easily the death of their parents than the loss of their property.

29

Here is the scoring rule: Score = $_POST['Q1']+ $_POST['Q2'] + (6-$_POST['Q3']) + (6-$_POST['Q4']) + $_POST['Q5'] + (6-$_POST['Q6']) + (6-$_POST['Q7']) + $_POST['Q8'] + (6-$_POST['Q9']) + (6$_POST['Q10']) + (6-$_POST['Q11']) + $_POST['Q12'] + $_POST['Q13'] + (6-$_POST['Q14']) + $_POST['Q15'] + (6-$_POST['Q16']) + $_POST['Q17'] + $_POST['Q18'] + $_POST['Q19'] + $_POST['Q20']; 30 The “catch question” is “A person asks you whether you are studying in the University of Sydney. Please choose “strongly disagree” as you are studying in the University of New South Wales now”.

30