• No results found

Promises made to be broken : an analysis of excuses and the selfish motivations behind cooperation

N/A
N/A
Protected

Academic year: 2021

Share "Promises made to be broken : an analysis of excuses and the selfish motivations behind cooperation"

Copied!
34
0
0

Bezig met laden.... (Bekijk nu de volledige tekst)

Hele tekst

(1)

University of Amsterdam

Promises Made to be Broken: An

Analysis of Excuses and The Selfish

Motivations Behind Cooperation

Jack Carey

11373393

2 August 2017

MSc Business Economics

Managerial Economics & Strategy

Master’s Thesis

15 ECTS

dr. J. van de Ven

(2)

Acknowledgements

Foremost, I extend my gratitude to dr. Jeroen van de Ven, my academic supervisor, for his feedback and commentary throughout the process of developing this thesis paper. I also wish to recognise the assistance of dr. Casper Troost in allowing me to conduct my experiments in those

first-year finance classes he coordinates. Lastly, I appreciate the advice and guidance from numerous other staff in the faculty of Economics and Business at the UvA. I am grateful for the

instrumental role the above-mentioned have played in contributing towards this research.

Statement of Originality

This document is written by Student Jack Carey who declares to take full responsibility for the contents of this document. I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been

used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

Abstract

It is well-known that promises abet coordination in social and strategic dilemmas. Insofar however, experimental literature has largely maintained the assumption of perfect information in games to derive this understanding. We diverge from this, employing a modified trust game that incorporates a mid-game shock, affecting the information available to second-mover play and her subsequent decision making. Our research finds that people leverage excuses (or shocks) to renege on earlier promises and in turn act self-regarding. We do not find support to conclude however that those self-regarding actions are any more exploitative given the same excuse. Although we cannot precisely comment on social preferences in our game, we propose that social norms become tenuous in the presence of a shock – acting as a catalyst for non-compliance.

(4)

1

Introduction

Cooperation underpins much of the value in social and economic interactions. Within the context of a firm, many dealings involving agents are characterised by collaboration and teamwork. From allocation of work flows through to deadline agreements – cooperation serves a functional role in achieving efficient outcomes. Important questions addressing the tension between selfish urges and mutual interests offer insight into the dynamics of prosocial action. It is not surprising that extensive research in experimental economics has been devoted to discovering these psycho-behavioural drivers. Fundamental to this body of literature is the persistence of cooperation in situations where non-cooperation is a dominant strategy. Even in the absence of any enforcing mechanisms such as reputation concerns, repeated interactions or punishment threats, do these environments still support such cooperative equilibria (Coricelli et al., 2006). Closely associated, contract theory continues to explore partnerships and their resulting consequences. Here, considerable attention has been given to settings with hidden action (non-contractible future choice) and hidden information (non-conditional private information). Such contractual incompleteness may facilitate opportunistic behaviour and as such jeopardise any potential cooperative benefits to be had in social interactions (Charness & Dufwenberg, 2011).

Corporate policy and social agenda need integrate measures to curb opportunism and reduce its harmful effects on society. Opportunistic activities like exploitation and shirking are unproductive and typically erode others’ incentive to contribute to enterprise. To mitigate this, earlier findings have documented that communication and non-binding promises to cooperate in social dilemmas are often credible vehicles to deter these types of behaviour (Ellingsen & Johannesson, 2004). It has been shown that the exchange of promises leads to heightened cooperation. Naturally, this prompts us to ask, under which settings do these promises continue to manifest? By extension, we also question, how readily are these promises undermined after being established? To comment on this, we need to consider the underlying support structures of promises. To that end, a backdrop of experimental games has suggested that it may not so much be a preference for fair and altruistic outcomes per se that sustains prosocial efforts (Van der Weele et al., 2014), but rather self-perception and social appearances.

We do not explicitly test for this in our study, however, we do discuss the interacting roles of social image concerns and shame aversion as the bedrock for upholding promises. These forces remain checked against a set standard of compliance with social norms. We argue that as transparency regarding actions and information decline, or behavioural standards become ambiguous, individuals become increasingly self-interested and more readily exploit the goodwill of others. That is, if it appears seemingly justifiable, individuals may be seen to leverage an excuse

(5)

to undermine a previous commitment. Specifically, this research concerns whether a shock (in the form of new asymmetric information) will be utilised as an ‘excuse’ to fail to execute a promise. Introducing this feature is designed to more accurately reflect uncertainty in a context outside the sterility of an experimental setting, and moreover sheds light into the robustness of promises. To that extent, we formulate our research question as follows: “Do people leverage excuses (or shocks) to renege on their promises and in turn act self-regarding?”

To operationalise our experiment, we employ a one-shot trust game – allowing for pre-commitment to a strategy – with a novel treatment variable to proxy for a shock. The unique Sub-game Perfect Nash Equilibrium (SPNE) involves no exchange. We first compare the distributions of promises between treatment conditions, we find statistical evidence to reject the notion that our results are driven by ex-ante differences in promises pledged. Further, our findings indicate that a statistically significantly higher proportion of people defect on promises given an excuse, relative to the circumstance where that same excuse is absent. Specifically, we observe an increase in the frequency of broken promises from 22.58% in our control condition (without game shock) to 43.94% in our treatment intervention condition (with game shock). Contingent on this result, we test whether those broken promises, with the game shock, attract greater levels of exploitative behaviour. We find no statistical support for this conclusion. Bolstered by our statistical findings, we discuss the capacity within our experimental design for the ‘false consensus’ effect and the role of ‘shade’ behind uncertainty. We assert there is a strong likelihood that these ideologies serve as the justifications for the anti-social action we observe.

We now start by discussing the related literature which precedes our research. Section 3 describes our methodology, treatment design, procedures and develops the experimental hypotheses we seek to examine. Section 4 analyses the results and poses potential limitations to our findings. Finally, section 5 concludes with discussion and final remarks.

(6)

2

Related Literature

The standard building blocks of conventional economic theory relies on the concept of “homo-economics”. This can be described under a set of assumptions characterising a ‘rational’, purely self-regarding individual whose utility is a function of only her material wellbeing. This derives the conclusion that people act selfishly and do not care about the well-being of others. However, a wealth of evidence contradicts this notion of pure selfishness. Experiments have proven to incite other-regarding preferences where subjects can be seen to give and return substantial sums of money – indicating non-pecuniary utility. Examples have been found in a wide variety of environments and in many institutional settings including positive provision of public goods, behaviour in ultimatum and dictator games, and, reciprocity in gift exchange games (Charness, 2000; Dufwenberg & Kirchsteiger, 2004; Van der Heijden et al., 2001).

Similarly, in the classic investment game (and subsequent replications) developed by Berg et al. (1995), despite misaligned incentives between first and second mover – we still observe positive giving and sequential reciprocity. Nonetheless, we determine our theoretic expectation for this game via a process of backward induction – which essentially results in a stalemate – the second-mover returns no money, and thus the first-mover never engages. This inconsistency between ‘rational’ economics and experimental outcomes has attracted much attention of theorists and prompted the development of alternative models of distributional preferences. Such approaches can accommodate for fairness considerations, tastes for social efficiency, disutility borne from inequities etc. (Charness & Dufwenberg, 2011).

The enforceability of exchanges in social dilemma games, especially the investment game, relies fundamentally on the presence of trust (Van der Heijden et al., 2001). Cooperation is rooted here in the ability to signal and communicate trustworthy intentions. As sub-components of experimental design – promises, pre-play communication and visual cues (stemming from face-to-face interaction) help overcome coordination problems and comprehensively enable higher degrees of trust. Vanberg (2008) explores how the exchange of promises is greatly influential in fostering pro-social behaviour and enhancing cooperation. He formulates two explanations for promise keeping – the ‘commitment-based’ and ‘expectation-based’ – both inducing a form of emotional commitment to fulfil contractual obligations. Following a proposed model of social preferences by Ellingsen and Johannesson (2004), the former explanation includes a “taste (...) [or preference] for keeping one’s word.” The defining feature here is that subjects are directly concerned about their consistency with personal obligations. The latter ‘expectation-based’ explanation is founded by a basic disposition to experience guilt when others are disappointed relative to expectations. Using an adaptation of the Charness and Dufwenberg (2006) design,

(7)

Vanberg achieves independent variation in the respective explanations by a process of random matching in his dictator game. He concludes that an increased frequency of cooperation appears to be incompatible with the ‘expectation-based’ explanation for promise keeping. In his ‘switching’ condition, to determine the validity of the ‘expectation-based’ explanation, he finds the receipt of a promise by partners only increases the frequency of the cooperative outcome by 2% (i.e. increased to 54% from 52%). Rather, his results indicate that the promise of a subject, has an immutable effect on the behaviour (only) towards the other subject to whom the promise was directed. Specifically, he observes an increase of 21% in the frequency of the cooperative outcome in his ‘no-switching’ condition (i.e. increased to 73% from 51%).

In essence, this suggests that the behaviour of a promise-maker (trustee) is motivated by self-manifested concerns regarding consistency with their own commitments. This is opposed to the alternative explanation that says the trustee has indirect concerns regarding the second-order beliefs the trustor has formed about their payoff expectations. Notwithstanding this, it should be qualified that the ‘commitment-based’ explanation is, at its core, self-propelling in the avoidance of guilt. Simply, in assuming a promise-maker is guilt-averse, it is costly for her to default on her promise and thus she prefers to fulfil her commitment. Often associated with social unease, guilt is said to arise from interpersonal transactions that involve the infliction of harm or loss on a relationship partner (Baumeister et al., 1994; Battigalli & Dufwenberg, 2007). Thus, to avoid potential contradiction to the earlier discussion of self-manifested concerns, we believe this sentiment of distress could more appropriately be defined as shame. This refers to the internal consciousness of, rather than the external feeling of responsibility for, some mishap or social malaise. Shame, alongside guilt, manifests itself most pronounced in contexts of communal relationship whereby norms are unambiguously defined.

Consistent with the outcomes of Vanberg (2008) relating self-manifested concerns to the likelihood of promise-keeping, Dana et al. (2007) posit that people intrinsically dislike appearing unfair – either perceived by themselves or others. Social appearances and reputation form the basis of image motivation. Referring to an individual’s tendency to be partly motivated by others’ perceptions, image motivation depends crucially on the visibility of ones’ actions. That is, people like to create a positive image of themselves when their prosocial act is publically visible (Ariely et al., 2009). Yet, it is important to acknowledge the one-directional nature of this process. The fact that people seek to maintain positive identifications if and when they can be awarded credit, does not necessarily affirm the opposite – that anonymity or hidden action results in antisocial behaviour. Dana et al. (2007) in their paper exploring moral ‘wiggle room’ offer insights into this converse relation. They postulate that in some situations, giving is motivated simply by a desire to

(8)

avoid appearing selfish. In stark contrast to the motive of altruism, behavior may simply be piloted by self-interest coupled with a desire to maintain an illusory preference for fairness. They construct different treatments whereby senders are able to leave the relationship between their actions and resulting outcomes uncertain. They construct two sources of ambiguity – the first allows for self-ignorance and the second, the withholding of information to receivers affected by their actions. This gives subjects the moral ‘wiggle room’ to behave self-interestedly.

In the absence of pre-play strategic commitments, Dana et al. (2007) find significantly less generous behaviour in these ‘uncertain’ manipulations relative to the ‘transparent’ baseline. To highlight this, in their plausible deniability treatment 59% (17/29) of dictator subjects chose the non-socially optimal outcome (i.e. chose to be self-regarding). Of those dictators whom were not cut-off by the software package, 55% (12/22) actively chose the selfish outcome. Overall, this reveals that senders capitalise on receiver’s incomplete information, an outcome framed as ‘other-deception’ (Dana et al., 2007). Dissimilarly to this study however, Van der Weele et al. (2014) find that ‘an excuse’ in a trust game does not encourage subjects to partake in more evasive behaviour. They draw a basis for comparison against Dana et al. (2007) and argue that earlier findings are due to an overwhelming lack of moral context in dictator games. They further that image concerns are not a key driver of reciprocal behaviour (in contextually richer settings) and excuses that simulate time-pressure have no effect on the incidence of cooperation. It should be noted, Van der Weele et al. (2014) employ a strategy method whereby subjects do not receive feedback about their partners’ behaviour. In light of this contradiction, we can assert that the extent of giving strongly depends on the environmental cues of a game (Camerer, 2003).

Especially pertinent to promise-keeping, norm-abiding behaviour is a common thread that ties both the analysis of Vanberg (2008) and Dana et al. (2007). Social norms can be interpreted as moral expectations, which people feel inclined to live up to. If promise keeping is considered a ‘social norm’ this naturally explains a ‘taste for keeping one’s word’, in line with the ‘commitment-based’ explanation of Vanberg (2008). Likewise, we can understand shame (or guilt) as a complex triggered by inconsistencies with what ‘should’ be done. Hence provided people have some inner compass dictating normative behaviour, a promise has intrinsic value. That is, independent of their effect on the trustor, promises are self-guiding in their commitment by the trustee. A departure from these implicit standards is inevitably damaging. The implications can potentially lead to impaired social status or even social alienation in the extreme. However, Dana et al. (2007) examine how a lack of transparency allows senders to remain insulated from such moral judgment of receivers, to prevent any deterioration of social image. Despite heterogeneous preference and

(9)

utility functions, it is evident that a desire to conform to social pressures underlies much of the rationale for promise-keeping.

Related literature has comprehensively examined the types of mechanisms that improve efficiency in social dilemmas, with an admittedly distinct focus on communication and promises. However, there is a marked void concerning how immune these mechanisms are to external factors. People submit to situational pressures that dictate giving in certain contexts, although little research has studied the various situational justifications that people exploit for behaving selfishly. Perhaps there is limited scope for generalizability of the fact that promises are a resilient instrument to improve coordination. Our study intends to determine whether the standard of promise-keeping is an actual goal that people seek to implement or merely a constraint that people position themselves to circumvent (Dana et al., 2007). By relaxing experiment parameters, we can stress-test the role of promises in cooperation, and determine when their functionality begins to falter. In mandating nonbinding pre-play communication in the setting of a trust game, we build on previous literature to offer insights into whether the availability of an excuse counteracts the cooperative benefits of promises. This lends to us the ability to discuss potential behavioural motives for promise-keeping, and conversely, promise-breaking.

(10)

3

Methodology

This research generated a novel dataset using student subjects, predominantly from the University of Amsterdam. The students were subscribed to a first-year introductory finance class belonging to a bachelor’s degree in the faculty of Business and Economics. We supplemented our participant numbers using additional students at a local student housing facility in Amsterdam. We conducted a classroom style experiment (during class time) to elicit our responses regarding promise-keeping behaviour. We compiled our data by means of both hand-written responses and via use of online survey software, Qualtrics. This afforded us an efficient process to achieve experimental scale. In the following section – we outline the experimental setup, describe our treatment groups, list the experimental procedures and conclude by framing our hypothesis.

3.1 Game Design and Treatment Setup

The basic format of our game involves the interaction of randomly paired subjects in a modified trust (investment) game. This is a one-sided sequential dilemma, prompting players to experience tension between private motives and potential gains from cooperation. We use standard game parameters of the Berg et al. (1995) experiment where subjects play one round, the first-mover (trustor) has an initial endowment and the second-mover (trustee) can return any proportion of the endowment shared with her. It is common knowledge between players that all money passed from the trustor to the trustee is augmented by a factor of three. We structured the game using a neutral framing – neither encouraging nor deterring cooperation of subjects.

We operationalise the exchange of ‘promises’ between subjects by a closed, structured form of numerical communication. This mandated that trustees pledge or announce to trustors the (non-binding) amount they were committed to return, for every given amount they were to receive in the future. This deduced a significantly large dataset of trustee promises that could be tested using the direct-response method. We avoided conventional free-form communication as this threatened to invalidate the observations of promises. That is, if the trustor does not behave consistently with the earlier free-form message announced by the trustee, that promise is made

redundant before the trustee has the opportunity to execute.Further, with limited experimental

time and resources available, narrow communication ensured the messages sent were relevant and

controlled.1 As argued in Bicchieri (2002, 2006), ‘relevant communication’ is critical in producing

cooperation and if discussion lacks explicit promising, the game may be ineffective in supporting trust.

(11)

We introduced the treatment intervention of a shock to test for the subjects’ willingness to leverage an excuse to defect on commitments. This shock can be perceived as new information and is exogenous to any of the earlier decisions and interactions of players. The shock constructed is essentially a lottery, affecting only the payoff of the trustor. Both players are aware that the trustor has a positive likelihood of winning the lottery, which will increase her payoff with absolute certainty. However, the probability and payoff characteristics of the lottery remain unknown to prevent the trustee from conditioning her strategy on the outcome. The likelihood of success in the lottery is set artificially high to induce the greatest number of informative observations. That is, only the roll of a number six on a six-sided die will the trustor not be successful in the lottery. This lottery is drawn after the trustee has made her promise, although before she has had the opportunity to execute the promise.

The lottery conceptually justifies an excuse to bypass a promise as it can be considered grounds for less generous behaviour. Moreover, as the lottery outcome is only revealed to the trustee, there is an element of shade whereby the trustee’s actions are not entirely transparent to the trustor. As neither player explicitly learns the payoff of their partner, the trustor can only imperfectly deduce whether their partner upheld their promise. This offers sufficient scope for the trustee to behave self-interestedly without stirring any legitimate concerns for equality or fairness.

3.2 Experimental Procedures

Subjects were informed that they were participating in an economic decision-making experiment. To begin, subjects were randomly assigned to either one of two roles: contestant A (trustor) and contestant B (trustee). The distribution of face-down identification cards to all subjects determined these roles and pairings. Within each sub-session, one participant would be randomly selected for payment and remunerated according to the decisions made within subject pairings. It was announced that subjects were permitted to communicate, via means of hand-written message, if

and only when indicated to do so by the experimenter. We proceeded to read the instructions2

aloud, including a summary of important information, before inviting questions.

Herein, the experiment was formally underway. The procedures were divided into a messages stage and two decision stages, with the treatment intervention also featuring a lottery

stage. Firstly, contestant B was requested to send a message3 to contestant A, indicating the amount

she promised to return for every amount she could receive (messages stage). Using this information, contestant A decided how much of her initial five-euro endowment to share with

2 Refer to Appendix – section 6.2 for treatment experiment instructions.

(12)

contestant B (first decision stage). Only lottery condition sub-sessions then had an intermittent stage whereby a lottery was drawn for all subject pairs, and only contestant B learned the outcome (lottery stage). Following this, contestant B could then decide how much to return to contestant A – considering all information provided to her (second decision stage). The possible strategy set available to contestant B here is an exact function of the amount sent by contestant A in the preceding decision stage.

At the conclusion of the experiment, subjects privately receive their payoff. We also surveyed general characteristics of participants including gender, economic background and relationship status with their randomly matched partner. Overall, the experimental standards were satisfactory. The commitment and decision-making tasks required of participants was not time-intensive nor cognitively demanding. We maintained a high degree of experimental control across the majority of sub-session and the incentives sufficiently dominated any subjective costs of subjects.

3.3 Research Hypotheses

Our main hypotheses and examination relates to the behaviour of the trustee (contestant B). We test our results between treatment samples to determine whether the lottery shock has any significant impact on promise-keeping behaviour of trustees. The SPNE of the game suggests that in either condition, giving should be null and thus we do not observe any variation between sample groups. However, as experimental literature predicates, we should expect positive giving. We formulate our main hypotheses below, with contingent sub-hypotheses that follow.

Formally, our experiment will test the difference in proportions of broken promises (or equally promises upheld) resulting from the lottery shock. Supported from a growing body of literature qualifying positive giving in experiments, we expect that given the experimental context, trustees will break their promise and return lower amounts coinciding with a winning lottery outcome, acting selfishly to benefit themselves. Further, we posit that the winning lottery outcome will induce greater exploitative behaviour ex-post as subjects feel seemingly more removed (less accountable) for their actions and thus greater exploit their partners.

Null Hypothesis 1: The incidences of trustees breaking promises are not statistically different between the lottery and no-lottery conditions.

Null (Sub-)Hypothesis 2: Considering only those promises broken, the deviation, i.e. the amount shirked from that promised (as a proportion of the amount promised) is not statistically different between the lottery and no-lottery conditions.

(13)

4

Results

We collected data from a participant pool of 224 subjects and conducted 16 experiment sub-sessions. After trimming our observations for incomplete and irreconcilable responses, we achieved a total sample size of 208 observations. Within this, 62 students (5 sub-sessions) participated in the no-lottery (control) condition and the remaining 146 students (11 sub-sessions) participated in the lottery (treatment intervention) condition. We employ different sample sizes to account for the inherently greater variance in the lottery (LT) condition relative to the no-lottery

(NL) condition.4 The average session times was approximately 15 minutes depending on the

treatment variation and class size. The average earnings in all sub-sessions, split by trustee (trustor) payoffs in the no-lottery and lottery conditions respectively were €6.39 and €6.61 (€6.48 and

€6.82).5, 6

Concerning our hypothesis, we are only interested in the behaviour of the trustee (contestant B in our experimental design). Thus, the results presented in this section only relate to the 104 observations of those participants assigned the trustee role. We analyse their promise making decisions, from both ex-ante and ex-post perspectives, to build comprehensive discussion related to our hypothesis. The majority of our results employed nonparametric hypotheses testing,

however, we do conduct parametric tests for additional robustness where it permits. To further

qualify our findings, we also fitted a linear regression model and conducted other tests for differences. For supplemental information regarding the coding of string variables (relevant for categorical variables) in our statistical tests, please refer to appendix – section 6.1; Figure A2. We further explain our findings, and the justifications behind our choice of tests in the body below.

4.1 Introduction and Descriptive Statistics

We first present summary results describing our participant pool, sorted by treatment groups. Figure 1: Summary statistics: Observed trustee characteristics

Gender Economics experience Relationship status

Female Male Yes No Friends Acquaintances Strangers

NL 13 15 20 8 10 14 4

LT 30 36 46 20 15 32 19

28 (66) observations of characteristics in NL (LT) treatment due incomplete survey responses

4 This yields us additional information (explanatory power) in the lottery data set, offsetting any statistical

impairment due to differences in variance.

5 The payoff outcomes reasonably approximate a normal distribution and as such we conduct a t-test to determine if

any statistically significant difference persists. We find no statistically significant difference in payoffs (at the 5% level of significance) between treatment conditions, for both trustees and trustors.

(14)

Our participant pool, and experiment sub-sessions were randomly assigned to either treatment group. By visual inspection, Figure 1 results show no large discrepancies in any of the surveyed characteristics, between treatment groups. This loosely confirms our process of random assignment was successful. However, we also ratify this conclusion by testing for equality of proportions, especially for the ‘Relationship status’ category. Using a fisher’ exact test for equality of proportions we find that there is no statistically significant difference (i.e. P = 0.250 for a non-directional test) in the proportion of ‘Friends’, ‘Strangers’ and ‘Acquaintances’ between treatment

groups.7 This allows us reject, with a high degree of certainty, that the following results are driven

by differences in treatment group composition.

We continue to present results, sorted by treatment group, concerning the giving and receiving of trustees in Figure 2. These indicate that on average, the ‘Returned amount’ by trustees to trustors, are lower in the lottery condition (€4.363) as opposed to the no-lottery condition (€5.323). Such an outcome is consistent with our overarching hypothesis. However, it should be recognised that the corresponding ‘Received amount’ by trustees are on average lower in the lottery condition, €10.973 relative to €11.710. Further, the standard deviation of the lottery condition exceeds that of the no-lottery condition across all variables. Together, these findings perhaps signal additional uncertainty (for trustors) in the lottery condition. Nonetheless, this inference of subdued giving in the no-lottery condition relative to the lottery condition, is further supported by lower proportion of ‘Ret./Rec.’ in the lottery condition.

Figure 2: Descriptive statistics: Treatment outcomes concerning giving (and receiving) of trustees

Trustee exchange of money (€)

Received amount Returned amount Ret./Rec.*

NL Obs. 31 31 31 Mean 11.710 5.323 0.403 Std. Dev 4.406 2.957 0.207 Min (Max) 0 (15) 0 (10) 0 (0.667) LT Obs. 73 73 73 Mean 10.973 4.363 0.345 Std. Dev 4.723 3.349 0.220 Min (Max) 0 (15) 0 (15) 0 (1)

Ret./Rec. is the amount returned to the trustor as a proportion of the amount received by the trustee

We proceed to statistically test this claim. Our sufficiently large sample size (i.e. n > 30) allows us to assume normality in the distribution of returned amounts and thus we conduct a mean-comparison t-test to determine if a significant difference persists between conditions. We

(15)

find weak support (at the 10% level of significance) to conclude that trustees return a higher

amount under the no-lottery condition relative to the lottery condition (i.e. Pr (T > t) = 0.076).8

However, we should be sceptical that this result is not merely explained by differences in the ‘Received amount’ between treatment conditions. We construct a linear regression to control for this effect (included as ‘Amt.Rec’ variable in specification below), and also introduce the appropriate interaction term – the model is presented below.

ReturnedAmount = α + βAmt.Rec. + γTreatment + δAmt.Rec. · Treatment + ε

We use robust standard errors to account for the likely uneven variance of the regression’s residual error term across observations (i.e. heteroscedasticity). After controlling for trustee’s ‘Received amount’ we find the ‘β’ coefficient on the ‘Treatment’ condition dummy variable (and the ‘γ’ coefficient on the interaction term) is not statistically significant (i.e. P >

t = 0.799 and P > t = 0.646 respectively). Despite the directional (negative) effect of these coefficients being consistent with our previous t-test, the lack of significance in our regression

output suggests the treatment condition does not explain the variation in ‘Returned amount’.9

Next, we provide descriptive statistics of the amounts promised to be returned by trustees, for every given amount of endowment received by trustors. Figure 3 details five categories of ‘Amount received’, sorted by treatment groups. Upon inspection, we can deduce that the no-lottery condition attracts, on average, higher (mean) pledged promise amounts, relative to the lottery condition. This holds true for all categories of amounts received, with the exception of ‘Fifteen (15)’. We also observe a marginally smaller standard deviation of promises pledged in the lottery treatment, relative to the no-lottery treatment. Results are tabulated below in Figure 3. Figure 3: Descriptive statistics: Promised return amounts of trustee for amounts received (trustee messages) 10

Amount received

Three (3) Six (6) Nine (9) Twelve (12) Fifteen (15)

NL Obs. 31 31 31 31 31 Mean 1.226 2.677 4.306 5.855 7.048 Std. Dev 0.902 1.301 1.786 2.188 2.687 Min (Max) 0 (3) 0 (6) 0 (9) 0 (11) 0 (15) LT Obs. 73 73 73 73 73 Mean 1.096 2.466 3.925 5.630 7.116 Std. Dev 0.878 1.281 1.739 2.245 2.787 Min (Max) 0 (3) 0 (6) 0 (9) 0 (11) 0 (15)

Promised amounts for entire sample set of 104 trustee observations

8 Refer to appendix – section 6.1; Figure A4 for results of t-test difference of means for trustee amounts returned. 9 Refer to appendix – section 6.1; Figure A5 for full linear regression results of returned amounts.

(16)

We complete our descriptive statistics by presenting data on promise-keeping behaviour. Figure 4 details the average return amount against each received endowment amount by the trustee. We group our results according to treatment condition and compare the actual amounts returned by the trustee against their own average promises pledged. To further clarify, the pledged promises are only averaged for those subjects who actually receive the corresponding endowment amount (i.e. 7 73 trustee subjects received €6.00 in the lottery condition, the average promised return of those 7 participants was €2.00). The bar-chart below provides a visualisation for this comparative.

Figure 4: Average pledged promises and actual returned amounts of trustees

sd

Number of actual zeros received were 1 and 6 for the NL and LT treatments respectively

We observe for the no-lottery and lottery treatments alike; the pledged promises consistently exceed (or equal) the actual amount returned. The magnitude of this difference varies across x-axis categories, however, the greatest difference is observed for ‘Fifteen (15)’. Figure 5

below offers further commentary on this behaviour.11 Figure 5 demonstrates a greater mean of

‘Promises broken’ in the lottery condition, relative to the no-lottery condition. Moreover, as similarly indicated in Figure 4, the mean amount the trustee deviated from the promised amount (‘Abs. Deviation’) is larger in the lottery condition relative to the no-lottery condition (–€1.164 compared to –€0.774). Although this indicates the lottery treatment suffers more damage to absolute levels of social efficiency, a better measure to examine subject behaviour is the ‘Dev./Promise’ variable. This accounts for the size of the promised amount, and creates a variable to determine the proportional deviation from that promise. Contrasting this between treatment

11 Refer to appendix – section 6.1; Figure A2 for coding of negative values for ‘Abs. Deviation’ and ‘Dev./Promise’.

0.0 1.5 3.0 4.5 6.0 7.5 9.0

Three (3) Six (6) Nine (9) Twelve (12) Fifteen (15)

Av er ag e t rus tee re tur n am ount s

Trustee received amounts

(17)

conditions, we see no substantive change (difference of 0.003) in the proportional deviation. We defer presenting statistical tests for these comparisons until sections 4.3 and 4.4 which offer commentary in greater detail. The remaining descriptive statistics are presented below in Figure 5.

Figure 5: Descriptive statistics: Trustee broken promises and deviation from promised amount

Trustee broken promises and deviations

Promises broken Abs. deviation Dev./Promise* NL Obs. 31 31 31 Mean 0.226 -0.774 0.150 Std. Dev 0.425 2.077 0.322 Min (Max) 0 (1) -10 (0) 0 (1) LT Obs. 73 73 73 Mean 0.438 -1.164 0.153 Std. Dev 0.500 2.321 0.495 Min (Max) 0 (1) -12 (3) -3 (1)

Abs. deviation is the amount the trustee deviated from the promised amount (conditional on a broken promise)

* Dev./Promise is the amount the trustee deviated from the promise as a proportion of the promise made (conditional on a broken promise)

4.2 Treatment Effects on Promise Distribution

To comment on our main hypothesis regarding the incidences of promise-breaking between treatment conditions, we first conduct a preliminary test. A confounding factor that could be driving the difference in proportions of broken promises is the difference in ex-ante promises sent from trustees between treatments. That is, perhaps the (potential) promise-breaking we observe is due to differences in the distribution of promises pledged as a result of the introduction of the lottery. If true, this could perhaps indicate that promise-breaking is driven by proactive behaviour (prior to the lottery draw) rather than reactive behaviour (after the lottery draw).

In Figure 3 above, we observe that on average, the no-lottery condition attracts higher promises than the lottery condition, across all categories but ‘Fifteen (15)’. To legitimise our conclusions and disqualify this explanation of alternative promise distributions, we regress potential amounts received from trustors against corresponding promises pledged by trustees. Our regression is comprised of five pledged promise amounts for every subject, totalling 520 data points between both the lottery and no-lottery condition. As we observe each participant on multiple instances, we cluster at the subject (observation) level to allow for within-subject variation in the error term. The linear regression specification and our results are presented below.

(18)

Figure 6: Linear regression of trustee pledged promises (dependent variable) with subject-level clustering

Regression results

Coefficient Robust Std. Error t P > |t|

Amt.Rec. 0.494*** 0.033 14.76 0.000

Treatment -0.302 0.222 -1.36 0.177

Amt.Rec.*Treatment 0.014 0.040 0.34 0.734

Constant -0.224 0.198 -1.13 0.261

*** Significant at the 1% level.

Model R-squared is 0.563. Treatment dummy variable coded 0 (1) as NL (LT). Amt.Rec. is amount the amount received by the trustee from the trustor.

As we cannot be certain our model is correctly specified, we ran our regression using different specifications including a default linear regression and heteroscedastic-robust standard

errors linear regression.12 The ‘β’ coefficient on ‘Amount Received’ is highly significant (P >

t = 0.000) leading us to conclude a positive relation between amounts received and promises pledged (consistent with intuitive expectations). However, we are primarily interested in the treatment variable, and the interaction effect variable between the treatment and the amount received. It is important to acknowledge that we coded our treatment dummy variable using the no-lottery (control) condition as the default (zero). Neither of our coefficients of interest are statistically significant and thus we can conclude that the distributions of ex-ante promises are unaffected by the presence of the lottery (at the 5% level of significance). That is, the ‘Treatment’ condition explains no statistical variation in the level of ‘Promises Pledged’. However, we should caution against the interpretability of these results. We attempt to fit a model to describe ‘Promises Pledged’ using only three explanatory variables (one interaction term) presented above. As such, we should recognise the model is highly stylised and only imperfectly describes the dependent variable. In turn, the model likely suffers endogeneity, rendering imprecise coefficient estimates – i.e. the correlation of explanatory variables with residual error term inflates coefficient estimates.

4.3 Defected Promises

We proceed to examine the main hypothesis relating to our research: are the incidences of trustees breaking promises statistically different between the lottery and no-lottery conditions? Complemented by the linear regression above, our findings to this hypothesis enable us to comment on whether the presence of an excuse renders people more likely to break their promises. We compare the proportions of those subjects in the lottery and no-lottery condition who break (or inversely keep) their promises. However, to ensure our observations in the lottery condition only relate to those

12 Note that our results deliver the same conclusion using these additional specifications. Moreover, we found our

results are robust to (within-estimation) entity fixed effects regression. It should be noted however that this model is not well-suited to our data – i.e. we have no concern of unobserved time-invariant heterogeneity in our sample.

(19)

subjects who exploit a successful lottery outcome to renege on their promise, we need further manipulate our data set. We assume any unsuccessful lottery outcomes conceptually results in the lottery (control) condition reverting back to the no-lottery (intervention) condition. This is due to the fact that the information and the payoffs relevant to both trustor and trustee are ex-ante (at the time of the trustee return decision), unchanged. To that end, we trim any lottery observations that coincide with unsuccessful lottery outcomes – a total of 7 observations where the trustor does not win the lottery. Within this subset of observations there are 3 incidences of promise-breaking (without matched trustor being successful in lottery). Thus, our lottery condition observations are reduced from 73 to 66, and by construction our observations of broken promises from 39 to 36. In either treatment condition, we do not assume a valid parametric distribution for the proportions of defected promises and hence we conduct a fisher’s exact test. We do not reject our null hypothesis for (P > 0.05) and reject our null, in favour of the alternative for all (P ≤ 0.05). The results of this test are presented below in Figure 7.

Figure 7: Difference in proportions of broken promises between treatment groups

Promise outcomes

Promises kept Promises broken Total

Treatment Group No-Lottery 24 7 31 (77.42%) (22.58%) (100.00%) Lottery 37 29 66 (56.06%) (43.94%) (100.00%) Total 61 36 97 (24.47%) (26.60%) (100.00%)

Fisher’s exact test for equality of proportions = 0.0406 Fisher’s 1-sided exact test for equality of proportions = 0.0304

Two sample t-test for equality of proportions (diff = mean(0) – mean(1)) Pr (diff<0) = 0.016**

Percentages in parentheses are row proportions of total NL and LT observations 66 observations for LT treatment due to trimming of non-winning lottery outcome

Our output yields significant results at the 5% level of significance. We observe p-values of 0.041 and 0.030 for the fisher’s exact, and one-sided exact test for equality of proportions respectively. This prompts us to reject our null hypothesis, in favour of the alternative and conclude that, all else remaining equal, the lottery yields a significantly higher proportion of broken promises than the no-lottery condition. We also perform additional tests to determine the degree

of robustness in our findings. We find our results are also robust to a parametric t-test13 (i.e.

(20)

conducting a t-test on the mean of ‘Promises broken’ in Figure 5) and are unchanged when using the larger dataset of 73 trustee observations in the lottery condition.

4.4 Magnitude of Deviations for Defected Promises

Our last set of statistical tests are an extension to the results we observe above. The following concerns the degree of exploitative behaviour of those trustees observed to break promises. In addition to the increased self-regarding behaviour we can conclude from Section 4.3 – here, we set out to determine whether an excuse affords subjects more reason to greater capitalise on a self-regarding act, than they otherwise would without an excuse. As such, we are chiefly interested in the average level of deviations between treatments. However, we further extend our analysis to ascertain potential differences in variance that may also offer commentary to this opportunistic behaviour. As a forward disclaimer, we concede there are significant limitations that impede our ability to comment on the magnitudes of these broken promises. Firstly, an obvious caveat is that the deviation magnitude of any given promise is contingent on the amount received by the trustee. Secondly, as these tests are conditional on a broken promise, we need trim our dataset to a total of 7 and 29 observations for the lottery and no-lottery condition respectively. Due to the evident difference in sample sizes, we need exercise caution when deriving conclusions regarding the variance of either sample set. Notwithstanding these considerations, we conduct a series statistical tests in attempt to deliver insight to our second (sub-)hypothesis.

We begin by plotting the absolute and proportional deviations14 of trustee promised

amounts against actual amounts returned. With albeit limited data points, these visuals may suggest negligible differences in the mean level and variance of deviations between treatment conditions – although at large they offer little insights. Consequently, we seek to ratify any potential difference of means using statistical methods. We find no statistically significant results from a mean-comparison two-sample t-test with unequal variance – testing both the ‘Abs. Deviation’ and

‘Dev./Promise’ variables.15 This indicates that the mean deviations of amounts returned, from

broken promises, do not exhibit any significant differences between treatment conditions. We proceed to test for potential differences in variance. We conduct a non-parametric test for equality of distribution functions (to proxy for equality of variance) between treatment conditions using a

K-Smirnov test.16 It remains that there is a lack of statistical support to claim any significant

differences in the variance of the samples between treatment conditions. This outcome holds true

14 Refer to appendix – section 6.1; Figure A7 and A8 for respective absolute and proportional plots of deviations. 15 Refer to appendix – section 6.1; Figure A9 for reported results of two sample t-test.

16 We deem that a parametric test for comparison of variance (assuming an F-distribution) is non-valid here due to

(21)

for both the ‘Abs. Deviation’ and ‘Dev./Promise’ variables – we do not report these results as they

are not central to our underlying (sub-)hypothesis.17 Again, we should acknowledge the diminished

statistical power of these tests. This analysis regarding the magnitude of deviations for defected promises is taken from an ex-post stand point – analysing only those observations which are consistent with promise breaking. We do not find any evidence that the presence of the lottery shock gives rise to any further exploitative (self-regarding) behaviour of those trustees observed to break promises. This suggests that the socially damaging implications of broken promises, conditioned on the pre-requisite of a promise already being broken, is the same under both treatment conditions.

(22)

5

Discussion and Conclusion

5.1 Results Summary and Research Discussion

Our results section lends our research question three key findings. Firstly, we reject the notion that our lottery shock results in ex-ante changes to the distribution of promises made by trustees. We model a linear regression fitting ‘Promises Pledged’ against ‘Amount Received’ and find no significant effect of the lottery treatment condition on promises pledged. Secondly, addressing the underlying hypothesis of our research we find that given an exogenous shock that functions as a credible excuse, subjects shirk away from previously committed promises. We conduct a fisher’s exact test and t-test for equality of proportions and learn that the incidences of trustees breaking promises are statistically greater (at the 5% level of significance) in the lottery condition relative to the no-lottery conditions. Lastly, concerning our other (sub-)hypothesis we find trustees who engage in promise breaking do not retain more money for themselves in the lottery condition relative to the no-lottery condition. That is, given our small sample size restriction after trimming, we do not find a statistically significant difference in the deviation levels of returned amounts from pledged promises (for those promises broken) between treatments.

5.2 Social Preferences and Psycho-Behavioural Drivers

There are several interacting factors that may explain the tendency to renege on promises when presented with an excuse. However, the design we adopted does not allow us to exactly discern what kind of social preferences and behavioural drivers are at work (including those that are violated) in our research. Rather we can only deduce likely candidates via a process of deduction. We begin now by discussing the environment established within our setup. Subjects are subscribed to the same academic course and belong to the same class cohort, experiencing routine encounters with others on a bi-weekly basis. Due to both frequent interaction and community values of the university – there is likely a strong prescription for fair and pro-social behaviour imposed here (Ortmann et al., 2000). We argue that given this, subjects do not want to behave out of self-interest at the expense of others. However, by introducing a shock that delivers new information and uncertainty – it seems intuitive that these social pressures will dissipate. We discuss two separate impetus for the dampening of these pressures below.

Firstly, similar to the study of Dana et al. (2007), we consider the relevance of ‘shade’ prompting the rationale for promise-breaking. Shade manifests itself when there is an ambiguous link or non-causal link between personal actions and outcomes experienced by others. In our setup, the trustor can only imperfectly determine if their paired trustee upheld their promise. That is, the composition of the trustors gross payoff is unknown to the trustor herself – she only

(23)

observes whether it more or less than that consistent with the trustee’s pledged promise. Aligned with the notion of social image concerns, such shade offers motive to act self-interestedly and renege on earlier commitments. However, this is not entirely convincing. The risk to the trustee maintains that, because she is unaware of the lottery payoff, depending on the amount she decides to withhold, the trustor may be able to deduce that a promise was broken. Evidently, this demands keen understanding of the experimental situation.

Alternatively, we suggest the ‘false consensus’ effect as a potential catalyst for broken promises. A false consensus usually entails that a person believes others would act similarly, rather than a person considering their own second-order beliefs about their choice set (Charness, & Dufwenberg, 2006). Essentially it can be seen as an egotistic bias to overestimate the degree to which others are like us. It is a possibility that through this, or other self-serving type reasoning, subjects manage to put at bay emotions of shame. Social norms that normally dictate the fulfilment of promises are now distorted by the self-serving conviction that others would act similarly. In turn, what is considered socially appropriate is no longer well defined. This gives rise to a reduced sense of accountability and subjects can comfortably justify shirking from their commitments.

Typical of much of the experimental literature testing strategic dilemma games, we believe it would be remiss to avoid mention of equity concerns. We discard that inequity aversion of the trustee could influence promise-breaking behaviour in our setup for a similar contingency reason as explained above. As the trustee is unaware of the lottery amount, they cannot accurately rebalance the relative payoffs to trustor and trustee by defecting on their promise. That is, an exogenous lottery shock with unknown outcomes regarding payoff cannot viably be used to break a promise on grounds of equity concerns. Notwithstanding this, we do observe subjects deviate from what is considered to be normative behaviour for self-regarding intentions. The preference for socially conforming behaviour (and consistency with commitments) is tenuous in the presence of an excuse or shock.

5.3 Limitations of Experimental Design

In this section, we discuss limitations to the interpretability of our results and the major flaws we retrospectively identified in our experimental design. These weaknesses in our design affect the external validity of our findings and in turn our ability to extrapolate conclusions to more extensive settings. Foremost, it is essential we acknowledge that our findings are highly sensitive to our choice of implemented shock. The lottery introduced affects the payoff to the trustor, a successful outcome resulting in her receiving more than anticipated by the trustee. Alongside other experimental choice parameters including payoffs, the possibility for monitoring and coordination, known

(24)

likelihood of outcomes – these variables influence (sometimes albeit subtly) economic play in games. In its entirety, our game successfully created an environment for informed decision making and enabled the observed reneging of promises.

We now continue to critique design elements of our research – specifically, the construction of promises in our game setting. Firstly, we did not structure the instructions to explicitly make use of the word ‘promise’. This was a conscious decision to avoid the subtle (and unanticipated) framing effects the word ‘promise’ could set in motion. However, there is an obvious risk that our alternative description of a ‘non-binding message’ could have been interpreted as an immaterial commitment, thus increasing the number of observed defected promises in both treatments alike. Secondly, when comparing free-form messages, to more restricted messages (only containing intended choices), literature suggests that these may be conducive to weaker levels of commitment. Rich communication allows for people to communicate their social motivations and personalises the interaction (Brosig et al., 2003; Cason & Mui, 2015; Charness & Dufwenberg, 2010). To incorporate this, a potential improvement to our game could be to reduce the choice set of players. For example, if the trustee were responsible for only a binary decision, we could implement free-form messages – all the while maintaining the relevance and validity of those observed promises. Lastly, our decision to mandate promises could perhaps have jeopardised useful self-selection elements in our design. That is, forcing trustees to make a promise does not allow for type detection – perhaps those we observed to renege on their promises, would ex-ante not have offered any commitment. Thus, an alternative explanation could simply be that all we observed is a tendency to be more selfish, which may not be appropriately mapped to a decision to break a promise that was voluntarily sent. Overall, the above choices in game design may well have inflated the number of observed defections in our experiment.

5.4 Concluding Remarks

In this paper, we posit the research question: “Do people leverage excuses (or shocks) to renege on their promises and in turn act self-regarding?” We employ a modified trust game that incorporates a mid-game shock, affecting the information available to trustee play and her subsequent decision making. Our findings indicate that, contrary to a wealth of experimental literature, promises may not be quite as fool proof a mechanism to enhance cooperation in settings of strategic dilemmas. If indeed people are rational and selfish – excuses, shocks and hidden action may be a guise under which efficient contracting may founder (Charness and Dufwenberg, 2006). We definitively draw the conclusion that there is an overwhelming effect of situational cues on pro-social behaviour and human generosity. This indicates that perhaps promises are, all else equal,

(25)

merely a mechanism that people feel obliged to act consistently with, yet capable of exploitation under certain conditions.

A key objective of this research is to improve economic theory, and the understanding of the determinants of pro-social and cooperative behaviour. However, we also seek to apply our findings to a practical context – informing the types of economic activity that entail cooperation, incentives and contracting, social norms, market failure etc. In many environments, there is a credible threat of parties reneging on their commitments. Preventative measures should be introduced in the early stages, and throughout the process of contract development. In turn, this should act to reduce the cost and frequency of disputes, and furthermore create efficiencies in bargaining processes. Insights from this paper speak to the real-world inadequacy of promises as a mechanism for complete coordination in such economic dealings. The broader message this communicates is that evasive and opportunistic behaviour has damaging socio-economic implications that need be curtailed.

Our findings indicate that the functionality of promises in enhancing cooperation is perhaps not as robust as assumed in existing literature. The behavioural heuristics regarding coordination in social dilemmas are thus not as informative as once presumed. We encourage future research to explore the role of promises in contexts that are more reflective of real-world environments. That is, in environments outside the confines of a controlled, experimental setting where there is legitimate susceptibility to shocks or inflows of new information. Specifically, we urge research to conduct further testing into the cognitive processes behind promise breaking – the tension in social preferences, and the psycho-behavioural drivers that dominate actions. Finally, we welcome research to investigate the potential premeditated (or impulsive) nature of defected promises, and, by extension, to determine whether people engage in deliberate strategic misinformation to further a self-regarding agenda.

(26)

6

Appendices

6.1 Additional Results Outputs

Figure A1: T-test difference in proportions of average payoffs (€)

Average Payoffs

Trustees Trustors Combined Average

Treatment Group No-Lottery 6.387 6.484 6.435 (0.494) (0.358) Lottery 6.610 6.815 6.712 (0.396) (0.257) Combined Average 6.543 6.716 6.630

Diff = mean(NL) – mean(LT) H0 : diff = 0 HA : diff < 0 Pr (T < t) 0.363 0.240 HA : diff > 0 Pr (T > t) 0.637 0.760

Standard errors in parentheses

31 (73) in the NL (LT) conditions for both the trustee and trustor participant pool

Figure A2: Coding of String Variables (Excel; Stata)

Number coding

Zero (0) One (1) Two (2)

Promise Upheld Keep Break

Treatment Group Control (NL) Intervention (LT)

Lottery Success Lose Win Control (NL)

Gender Male Female

Economics Experience

Yes No

Relationship Status Strangers Friends Acquaintances

Deviation levels coded negative (positive) if returned less (more) than promised.

Figure A3: Difference in proportions of relationship status between treatment groups

Relationship status

Strangers Friends Acquaintances Total

Treatment Group No-Lottery 4 10 14 28 (4.26%) (10.64%) (14.89%) (29.79%) Lottery 19 15 32 56 (20.21%) (15.96%) (34.04%) (70.21%) Total 23 25 46 94 (24.47%) (26.60%) (48.94) (100.00%)

Fisher’s exact test for equality of proportions = 0.25028

Percentages in parentheses are cell proportions of total 94 observations.

(27)

Figure A4: T-test difference in means of actual returned amounts by trustees Returned Amounts Trustees Treatment Group No-Lottery 5.323 (0.531) Lottery 4.363 (0.392) Diff = mean(NL) – mean(LT)

H0 : diff = 0 HA : diff < 0 Pr (T < t) 0.923 HA : diff > 0 Pr (T > t) 0.076 Standard errors in parentheses

31 (73) in the NL (LT) conditions for trustee returned amounts

Figure A5: Linear regression of trustee returned amounts (dependent variable) with robust std. errors.

Regression results

Coefficient Robust Std. Error t P > |t|

Amt.Rec. 0.530*** 0.060 8.82 0.000

Treatment -0.172 0.674 -0.25 0.799

Amt.Rec.*Treatment 0.036 0.079 -0.46 0.646

Constant -0.888 0.153 -1.67 0.099

*** Significant at the 1% level.

Model R-squared is 0.529. Treatment dummy variable coded 0 (1) as NL (LT). Amt.Rec. is amount the amount received by the trustee from the trustor.

Figure A6: Average pledged promises of trustees by treatment conditions

sd

Average of entire set of pledged promises from 31 (73) observations in NL (LT) conditions

0.0 1.5 3.0 4.5 6.0 7.5 9.0

Three (3) Six (6) Nine (9) Twelve (12) Fifteen (15)

Av er ag e t rus tee p ro m ise d am ount s

Trustee received amounts

(28)

Figure A7: Absolute deviation from promised amounts for observed broken promises

We observed two incidences of positive giving which are not included in this scatter plot; for trustees promising to return 0 (2) they returned 3 (3) – i.e. abs. deviation of 3 (1)

Figure A8: Proportional deviation of promised amounts for observed broken promises

We observed two incidences of positive giving which are not included in this scatter plot; for trustees promising to return 0 (2) they returned 3 (3) – i.e. prop. deviation of 3 (0.5)

0 3 6 9 12 15 0 2 4 6 8 10 12 14 16 Ab so lut e d ev ia tio n fro m p ro m ise

Trustee promised amounts

Lottery No Lottery 0.000 0.200 0.400 0.600 0.800 1.000 0 2 4 6 8 10 12 14 16 Pr op or tio na l d ev ia tio n fro m p ro m ise

Trustee promised amounts

(29)

Figure A9: T-test difference of means for deviations from broken promises

Deviations variable

Abs. deviation Dev./Promise* NL Obs. 7 7 Mean -3.429 0.663 Std. Err 1.251 0.132 LT Obs. 29 29 Mean -2.897 0.375 Std. Err 0.544 0.0.134

Diff = mean(NL) – mean(LT) H0 : diff = 0 HA : diff < 0 Pr (T < t) 0.353 0.929 HA : diff > 0 Pr (T > t) 0.647 0.071

Abs. deviation is the amount the trustee deviated from the promised amount (conditional on a broken promise)

* Dev./Promise is the amount the trustee deviated from the promise as a proportion of the promise made (conditional on a broken promise)

(30)
(31)
(32)

6.3 Sample Information Exchange Sheets (Lottery Condition)

Page 1/1

Information Exchange

Class Number: ………

Participant Number – Contestant A: ……… Participant Number – Contestant B: ………

Stage One – Contestant B Message

Contestant B please complete the final column:-

Contestant A sends… Contestant B receives… Contestant B is willing to return… €0 €1 €3 €……… €2 €6 €……… €3 €9 €……… €4 €12 €……… €5 €15 €………

Stage Two – Contestant A Decision

Contestant A please complete the below:- Contestant A has decided to send €………

Contestant B receives triple the above amount equalling a total of €………

**Mystery lottery drawn**

Stage Three – Contestant B Decision

Contestant B please complete the below:-

Was your paired contestant A successful in the lottery? ( Yes / No ) Contestant B has decided to return €………

(33)

7

References

Ackert, L. F., Church, B. K., & Davis, S. (2006). Social Distance and Reciprocity (Working Paper). Georgia State University: Experimental Economics Center.

Ariely, D., Bracha, A., & Meier, S. (2009). Doing Good or Doing Well? Image Motivation and Monetary Incentives in Behaving Prosocially. American Economic Review, 99(1), 544-555. Battigalli, P., & Dufwenberg, M. (2007). Guilt in Games. American Economic Review, 97(2),

170-176.

Baumeister, R. F., Stillwell, A. M., & Heatherton, T. F. (1994). Guilt: An interpersonal approach. Psychological Bulletin, 115(2), 243-267.

Berg, J., Dickhaut, J., & McCabe, K. (1995). Trust, Reciprocity, and Social History. Games and Economic Behavior, 10(1), 122-142.

Bicchieri, C., Lev-On, A., & Chavez, A. (2009). The medium or the message? Communication relevance and richness in trust games. Synthese, 176(1), 125-147.

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

Brosig, J., Weimann, J., & Ockenfels, A. (2003). The Effect of Communication Media on Cooperation. German Economic Review, 4(2), 217-241.

Camerer, C. F. (2011). Behavioral Game Theory: Experiments in Strategic Interaction. Princeton: Princeton University Press.

Cason, T. N., & Mui, V. (2015). Rich communication, social motivations, and coordinated resistance against divide-and-conquer: A laboratory investigation. European Journal of Political Economy, 37, 146-159.

Charness, G. (2000). Bargaining efficiency and screening: an experimental investigation. Journal of Economic Behavior & Organization, 42(3), 285-304.

Charness, G., & Dufwenberg, M. (2006). Promises and Partnership. Econometrica, 74(6), 1579-1601. Charness, G., & Dufwenberg, M. (2010). Bare promises: An experiment. Economics Letters, 107(2),

281-283.

Charness, G., & Dufwenberg, M. (2011). Participation. American Economic Review, 101(4), 1211-1237.

Charness, G., & Garoupa, N. (2000). Reputation, Honesty, and Efficiency with Insider Information: An Experiment. Journal of Economics & Management Strategy, 9(3), 425-451. Coricelli, G., Morales, L. G., & Mahlstedt, A. (2006). The Investment Game with Asymmetric

Referenties

GERELATEERDE DOCUMENTEN

This is also backed up by archival photometry from Sonneburg archive, which does not reveal dimmings at the predicted times, radial velocity observations from TRES, which sees no

With a strong focus on three case studies, this thesis studies what has constructed the concept of national identity in the party positions of right wing Western-European

This research examines how the extent of cooperation in international strategic alliances influences the level and content of coordination between the

Although it is true that one well-powered study is better than two, each with half the sample size (see also our section in the target article on the dangers of multiple underpowered

The experiments show that people are willing to forego a material gain to prevent future regrets and that the reluctance to exchange lottery tickets is (partly) caused by

The pressure drop in the window section of the heat exchanger is split into two parts: that of convergent-divergent flow due to the area reduction through the window zone and that

week period of treatment. Aloe ferox gel material showed a 1.1% increase in skin hydration after 1 week of treatment; but thereafter also exhibited a dehydrating effect

Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners and it is a condition of