Is there a slippery slope in the dictator game?

(1)

UNIVERSITEIT VAN AMSTERDAM – FACULTY OF BUSINESS AND ECONOMICS

Is there a slippery slope in

the dictator game?

MSc thesis – Game theory and behavioural economics

Petra Cristiana Dinescu - 10636536

7/15/2016

The dictator game is one of the most widely used frameworks for investigating social preferences. This paper explores more closely the effect of a particular kind of framing on decision making in a dictator game. Specifically, we look at whether the sequential farming of choices creates a slippery slope effect that can catalyze unethical behavior amongst dictators. This research question is relevant for understanding whether psychological mechanisms can shift the pattern of decisions undertaken in an ethical context and if so, to what extent.

(2)

1

1. Introduction and research question

Unethical behavior is subject to a lot of debate, with arguments ranging from the fields of philosophy, sociology, psychology and economics. The argument of the slippery slope is often invoked when analyzing gradual ethical erosion and its (sometimes disastrous) consequences. While psychological research on the subject has been undertaken, there is almost no empirical research looking at the evolution and determinants of unethical behavior in an economic

(3)

2 context of a ‘slippery slope’. Mentions of the effect are plentiful in the mainstream press (from politics to corporate scandals) or business ethics textbooks (Welsh et al., 2015), yet the

accounts are rather anecdotal than empirics-based. For example, the slippery slope argument has been evoked in a number of business scandals, such as in the investment fraud case of Bernie Madoff. He even described his case himself as a slippery slope – “It starts out with you taking a little bit, maybe a few hundred, a few thousand. You get comfortable with that, and before you know it, it snowballs into something big.”

A motivating example in which a slippery slope might have played a role is Stanley Milgram’s ‘Obedience to Authority’ 1960s experiments, which were never centered around gradualism or slippery slopes. This experiments portrays subjects in their role of ‘teachers’, another form of dictators, having the power to administer incremental electric shocks to their ‘students’, every time they did not correctly answered a task. While the main theme of the study lies in its title – obedience to authority – the design of the experiment holds some of the defining

characteristics of a slippery slope – incremental shocks administered while engaged in a ‘teaching’ exercise. The published first results – 65% of subjects reached the final 450 volts shock – generated a lot of debate and criticism, and so the psychological mechanisms that shaped these findings could not be investigated in a similar manner. So the question was born whether the sequential, incremental ‘framing’ of the shock punishment contributed to the surprising results. This question has generated the present research, which investigates the effect of the sequential framing of decisions in a standard dictator-game.

Unethicality can be represented by a number of actions, amongst which theft, cheating, unfairness, bending the rules, intentional damage. The present study uses taking money from another as a measure for unethical behavior. The research aims to find out whether there is a ‘slippery slope’ effect in the context of a variant of a dictator game and what might the

underlying causes of it be. The slippery slope is represented here by a sequential framing of the decision, and takes place in the absence of external costs, social exposure or monitoring. The research uses a between subjects experimental design and is conducted through a survey distributed online.

(4)

3 One of the hypothesized causes of the effect is the constant change in the reference point of the agent, translated into an update of the decision value at each step. In other words, we would like to test for the so called myopic decision making – the failure to respond to minor changes in the environment, as each step can be perceived by the subject as relative to the one before. A secondary research question is concerned with the willingness of people to reverse or update their decision when presented with the possibility to do so. The motivations for them to do this depend on the initial decision taken as well as on the beliefs formed during the

experiment and will further be investigated in the Results section.

The findings of this paper can provide valuable insight into how people make unethical choices and whether gradualism can accelerate or hinder such a decision. Repeated decision making in unethical circumstances appears to not encourage people to cheat or steal, possibly even more so when the ‘stakes’ are insignificant. However, further research is needed to give some

conclusions on the topic, as this paper has several limitations discussed at the end. There is a broad scope for future investigations, since the topic is not yet thoroughly covered by the scientific community. Real-life applications include the design of policies or analysis of unethical decision making among powerful individuals.

As follows, the literature section contains a brief overview of the existing research on adjacent topics and a summary of the findings used to guide this study. The methodology part presents the elements of the experiment and the motivation for each choice, together with the

theoretical predictions / hypotheses driven by arguments based on the literature presented at the beginning. The results and the discussion sections are concerned with the actual findings and interpretation of this research, as well as limitations and further avenues for developing this area of study further.

2. Literature review

The dictator game, along with other social preferences games such as the ultimatum and the trust games, has been used extensively in experimental economics research ever since the

(5)

take-4 off of the field. The dictator game consists of two players, out of which only one of them – the dictator – has decision power. He chooses how to divide a given amount of money/units between himself and his recipient and his decision is incontestable. In contrast to what the theory of the rational economic agent predicts, a majority of dictators give a positive amount to their recipient.

A number of social preferences models have been proposed, for which the inclusion of social preferences in utility theories has been a central element. The models of degree altruism, reciprocal altruism and inequity aversion are probably the most widely used (Clavien & Chapuisat, 2015) and hold through repeated experimental trials. These models are relevant when analyzing the data of our own study as well.

List (2007) shows that subtle manipulations of the set of actions or framing of possible

decisions have a large influence on the outcome in dictator game settings. As the author points out, this has great implications for understanding generosity and social preferences outside of the lab and how we can use that in institutional settings. Specifically, he explored the existing literature on dictator games and introduced the new variation of the dictator that can take rather than give, strengthening the fact that behavior is not only determined by the

preferences of subjects, but also by the characteristics of the situation in which they make decisions.

We use this design implemented by List (2007), as well as Bardsley (2007), where dictators have the option to take from another’s endowment rather than to give to the other. In the taking game implemented by Bardsley (2007), dictators have the possibility to either give or to take in various amounts. One of the treatments only allows for giving (just as in a standard dictator game), while the other one allows for both taking and giving. The proportion of subjects who give (in the first treatment) is significantly higher than the proportion of subjects who do not take (in the second treatment). Moreover, whenever participants have the taking option along the standard giving one, the percentage of givers decreases, showing once again how framing can affect aggregate decision making in such settings.

(6)

5 Unlike the study of Bardsley (2007), which incorporates taking choices in the choice set and looks at how the distribution of givers changes, we only use taking as a given option. We instead look at how the framing of the decision to take from one’s recipient affects the final distribution of amounts between the dictator and the recipient. Moreover, we try to exclude the social normative aspect that can partly shape results of a standard dictator game - the generous behavior of dictators can be enhanced by the norm to not act entirely selfish with an endowment that was offered to them.

The slippery slope argument has been used as a form of reasoning in multiple domains,

biomedical ethics, legal argumentation and business ethics being just a few. To give a definition of the phenomenon, Walton (2015) gathers several characteristics that a given argument needs to have in order to fall in this category of the slippery slope, amongst which:

 The agent faces a sequence of actions, each being the result of the carrying out of the previous action; in the beginning, each sequence step appears to be insignificant, but consequences become larger as the sequence proceeds.

 There are factors that catalyze the movement along the sequence, making it increasingly difficult for the agent to stop moving ahead; loss of control and impossibility of reversal are often drawn upon when defining slippery slopes. Rizzo & Whitman (2003) also identify the three characteristics common to all slippery slope arguments as being: ‘an initial, seemingly acceptable, decision, a ‘danger case’ that is clearly unacceptable and a process or mechanism by which the initial decision will raise the likelihood of the danger case’. We are not as much interested in the danger case or catastrophic

consequences of slippery-slopes, as we are in the causal mechanism that leads to these end-points. We therefore focus in this study on the first characteristic of slippery slopes, that of the sequential series of actions, each being the result of the previous. Needless to say, a slippery slope leads, as stated previously, to a bad, usually unforeseen and undesired, outcome. Hence, a certain negative aspect of the actions undertaken by agents is needed when studying this phenomenon.

(7)

6 The empirical studies that do exist on the subject are placing the slippery slope in the domain of incentives. The subject is thus exempted of agency and the main effect tested is the reaction or change in behavior following different incentives or payoff schemes that are modeled as slippery slopes (Offerman & van der Veen, 2015; Welsh et al., 2015). The general finding is that gradual changes, as compared to abrupt ones, lead to a higher degree of unethical behavior. Gino & Bazerman (2009) investigate ethical judgment in the context of a slippery slope and find that people are more inclined to criticize the misconduct of others when it occurs abruptly, rather than in small, gradual steps. Through their experiments, they show that subjects having the role of ‘watchdogs’ are less-likely to reject inflated estimations of coins in jars (their measure of unethicality) when the estimations increase gradually, rather that abruptly once. Part of this effect is due to the failure to observe the ethical erosion when it occurs gradually, according to the authors. However, the measure used in this study is not the most reliable and the differences in amounts monitored are perhaps too small to be representative of the findings.

What might be the underlying psychological mechanism for the slippery slope effect? Bandura (1991) views moral reasoning as an accumulation of interacting factors that influence each other, thus a ‘self-regulatory mechanism’ is required to guide moral conduct. Welsh et al. (2015) extend this view by attributing the increased unethical behavior in a slippery slope context to the impairment of one’s capacity to maintain appropriate behavior and an eased moral disengagement. This suggests that self-control plays a role when it comes to ethical behavior, and it surely plays a role in a slippery slope context as well. The authors prove this by using gradual and abrupt changes in monetary incentives following the completion of a task in which participants could cheat by over-reporting their correct answers. Their results show a clear correlation between the gradual change condition and over-reporting. Moreover, they show that the relationship between gradualism and cheating is mediated by increased moral disengagement and reduced self-regulatory ability, measured through questions similar to the ones used in our scale.

(8)

7 Mazar et al. (2008) suggest that dishonesty follows a constant pattern up until a certain level of external (monetary) benefits, after which it becomes limited. In other words, they (also) used false reporting of correct answers as a proxy for unethical behavior and found that lower payment per correct answer lead to slightly more unethicality. These findings provide insights about the psychological mechanism behind the assessment of internal and external rewards in economic settings. Small unethical acts seem to be easily rationalized when the external reward is kept under a threshold. When that threshold is exceeded, the internal cost of ‘misbehaving’ becomes too big to be covered by the (larger) external reward. For this reason, we choose a small amount to be taken at each step and believe that such a negligible amount, relative to the entire endowment of the recipient, can be easily internalized by the dictators.

Similarly, Peer et al. (2014) find partial confessions of unethical behavior to be prevalent and more common than null or full confessions. However, the emotional cost associated with only partially confessing are the highest, indicating that the external reduced negative consequences of a full-confession are transferred to the internal domain as emotional costs and lower levels of guilt relief.

Other studies focus on self-justification in an unethical behavior context. One research shows that individuals prefer justified dishonesty when available (Shalvi et al., 2015). By using pre-violation (of rules) justifications, people redefine unethical actions as excusable, thus

maintaining a moral self-concept. In the same time, using post-violation justifications alleviate the threat to the moral self-concept by compensating for the unethical act. In a larger context, honesty, measured through self-reporting of die-roll outcomes, is significantly influenced by the level of corruption attributed to the country where the individual resides (Shalvi, 2016).

In another experiment conducted by Rilke et al. (2016), self-reporting of a company’s employees is analyzed for the same purpose of determining dishonesty and its underlying psychological mechanism, yet in a real-world organizational environment. The two treatments they use are denoted ‘one-by-one’ (a series of smaller and frequent reports) and ‘all-at-once’ (one large end report) indicating the procedure of self-reporting performance of employees in a company. They find that participants inflated their performance more in the one-by-one policy

(9)

8 treatment (with a 7 percentage points difference). However, the exact extent to which they did so is hard to be determined given the experimental design and measures used by the authors, which do not allow knowing whether and how much participants lied (there is not possibility to check actual performance).

One particular feature of our research design is the anonymous and private decision space, supporting the individual nature of the dictator game (the social interaction is very limited). As Shalvi et al. (2015) point out, people strive to keep a positive image of their selves not only publicly, but also in private. Individuals maintain their moral conduct through various control mechanisms. The absence of a social context and the guaranteed anonymity rule out social sanctions as the main control mechanism in our experiment, leaving self-reactive control as the main one (Bandura, 1991) . Control mechanisms are not however immutable. Self-regulatory influences only operate once activated, and disengagement can lead to detrimental social consequences (Bandura, 1991). One way of sponsoring moral disengagement is achieved not by changing moral standards or the characters of agents, but rather through restructuring moral values and providing moral justification. Self-serving justification play a major role in this process, by diminishing the psychological costs that are linked to unethical behavior, and act as a license for assimilating two opposing desires – that of maximizing the external reward and that of minimizing the internal costs (Shalvi et al., 2015). As Barkan et al. (2015) point out, ambiguity is used to create a ‘gray zone’ in which people can blur the discrepancy between what is considered right and wrong, thus protecting their moral-self-perception. By keeping the design of the experiment abstract, potential fuels of self-serving justifications (such as feelings of entitlement after a task) are kept minimal. Ambiguity is thus present only in the form of the uncertain initial endowment of the partner, which will hopefully act as a self-justification for initiating the unethical behavior of taking. Moreover, no authority/control is present. This can be another catalyzing factor in the process of moral disengagement, as shown by Milgram (1978).

While psychological self-justification mechanism seem to play an important role in a slippery-slope setting, another element that can influence the way people act in such circumstances

(10)

9 refers to the change in reference points, especially when the actions are taken over a larger time frame. Prospect theory (Kahneman & Tversky, 1979) including reference point theory has been prominent in the last decades’ economic research. Gains and losses are, according to this paradigm, not regarded in absolute terms of wealth, but rather in the difference between the value of the gain or loss and a pre-existent reference point.

There is a lot of experimental literature indicating that prospect theory represents a better theoretical framework for real-life decision making of individuals. Lee et al.(2008) combine adaptation-level theory (Helson, 1964) with prospect theory, looking at the shift in reference point in an investments setting. They are particularly interested in the adaption mechanism of investors to losses and their decision to capitulate or hold on to the investment. Adaption level theory originates from studies on sensory systems, but is apparently applicable to the cognitive system as well. The adaptation level comprises all the previous stimuli and is thus updated at each step. Lee et al.(2008) show that there is a positive relation between adaption and the total loss accumulated, as well as that adaption is negatively correlated with one’s positive

expectations and feelings about a bad investment. In our experiment, the stimuli are

represented by the change in the external reward accumulated. An increase in adaption to the acquired gains could thus lessen the cost associated with taking an additional small amount.

3. Methodology

1. Experimental design

In the context of the present research question, an experimental design is advantageous given the rather small number of experimental studies on the subject. Moreover, it is the only type of research design that allows for a high degree of control. A field experiment cannot easily isolate a specific framing such as the slippery slope we use, even more so when unethicality is studied, while keeping everything else constant and not having other factors (such as uncertainty) shape the context.

(11)

10 The experiment comprises a decision-making survey conducted online using Qualtrics, which is designed following the model of a dictator game. It involves two players, one represented by the sole decision maker (called the dictator) and the other, called the recipient, represented by the charity GiveDirectly. All participating subjects take the role of the dictator in this

experiment and each is paired with this charity as its recipient. GiveDirectly most effectively mirrors an individual recipient, given that it transfers donations (in proportion of 91%) to eligible individuals in poor African communities. Subjects were recruited online, mainly from the two general student groups of Business and Economics at the University of Amsterdam and the University of Economic Studies in Bucharest.

Following the design of a standard dictator game, each player is initially endowed with a certain amount, denoted in units, which remains constant for both the control and the treatment group. The dictator has to decide whether to take from his recipient’s endowment and how much to take. The amount of units that can be taken is incremental in steps of 12 units; hence, one can take an amount of the form n*12, with n being a multiple in the range of [0, 20]. This sort of framing of the decision, in which dictators can take instead of give, was used before in dictator game experiments (Bardsley, 2007). The initial endowment of the dictator is fixed at 238 units. The initial endowment of the recipient is randomly generated in Microsoft Excel from the uniform distribution [268, 296] and is equal to 288. This random determination is known to the dictator, but the drawn value of the endowment of the recipient remains unknown to the dictator, and thus is denoted E2 throughout the experiment. The starting point of the interval is chosen as to provide the dictator with an incentive to start taking. The endpoint is just below the option of the dictator to receive 298 units, which indicates a total take of 60 units from the recipient (which is realized in 5 steps in the treatment group). This provides a context to check whether subjects form self-serving beliefs about their recipient’s endowment and whether they adapt their choices to those beliefs. The uncertain endowment of the recipient was chosen as a design element in order to reduce the artificiality of the situation and create a more realistic context, as well as to not encourage beliefs regarding the over-control of the experimenter over the endowments. The interval, overpassing the fixed endowment of the dictator, was chosen

(12)

11 as such to provide a clear incentive for the dictators to start taking, as well as to make it

possible to differentiate between altruistic and inequity-averse preferences.

The experiment manipulates the way in which the choice over the final distribution of units is presented. Thus, the two conditions are as follows: for the control group, the choice set is presented at once in a list, from which the participants can choose the one distribution scheme that they want to implement; the treatment group has the same final distributions as choices, as well as the same information, only presented one step at a time with a constant amount of 12 units being added to their own pot with each step. The decision making for the treatment group is structured sequentially, with the two available options at each step being ‘do not take’ and ‘take’ the additional 12 units, an amount equal to 4.4% of the minimal total endowment that a recipient can have (268). We choose this amount for each step of the slope because it is a small amount that can be easily rationalized. For example, the second question in the

treatment group, if reached, looks like this:

‘You have chosen to take 12 units from your recipient. Your endowment is now equal to 250 units and your recipient's endowment is equal to E2-12. You are now given the possibility to either take 12 units again or not. Please choose one of the following.

o Yes, take 12 units from the endowment of my recipient and add them to mine. o No, do not take 12 units.’

Participants are randomized between the two groups by the survey software, which randomly presents one of the two question blocks upon opening the survey. Instructions and control questions are equivalent for both groups, as well as the supporting questionnaire. The two groups are thus identical in terms of experimental design, with the only difference being the framing of the decision to take from the recipient. This design enables a between-treatments analysis of results and determining whether there is any effect of the framing of decisions on outcome.

An additional experimental feature involves the given possibility to reassess one’s own decision outcome in both groups. That is, after any subject settled on the amount that he wishes to take

(13)

12 and following the supporting general questionnaire, he is faced with the whole display of

distributions and asked whether he wishes to choose a different distribution and implement it as final. In this question, which is not preannounced, the subjects are reminded about their own starting endowment value, the endowment interval of their recipient, as well as their previous decision (the amount taken from their recipient in the main part of the experiment). This design element is chosen in order to provide additional insight into the motivations for taking or not taking, as well as to test whether preferences are indeed constant as traditional theory predicts. If a high rate of (decreasing) change is recorded, this might indicate either that people are carried away through the slippery slope which leads to wanting to compensate for the bad outcome reached or that they motivate their unethical decisions through self-serving, artificial beliefs about a future possibility to give something back or make up for their behavior. The extent to which each participant took from his recipient was used as a proxy for unethical behaviour. Given that both subjects are endowed with a certain amount to start with, taking from one’s recipient to one’s own benefit can be considered analogous to stealing. As the taken amount increased, so does the unethicality of a subject. However, in this particular context, taking can also be explained by other factors such as an aversion to charity giving, hence we include some control variables.

Following Welsh et al. (2015), the final questionnaire includes measures of moral

disengagement. These consisted of several questions that the subjects were asked to fill-in at the end of the survey, along with the demographic questions such as gender, study track, age and nationality. Moral disengagement was measured using four items from the scale developed by Moore et al. (2012), out of which one item was adapted to this particular situation (‘It is okay if somebody steals as long as they have a good reason for it’). Each of the items was assessed on a scale of 1 to 7 (1 representing a ‘strongly disagree’ attitude), hence the scale value for any participant lies between [4,28], with the upper limit indicating the highest moral

disengagement. The four questions in the moral disengagement scale each touch upon one of these dimensions – stealing, authority, means of achieving noble ends and taking credit; the higher the score for each of these factors, the more morally disengaged the subject is

(14)

13 considered. Additional measures include a risk-taking self-assessment and charity giving habits. While Welsh et al. (2015) also include a measure of the depletion of self-regulatory resources, our design does not contain an activity that should deplete the subjects of self-regulation (such as solving math puzzles). Given that measuring depletion of resources should be done both before and after the experimental task, which in our research is concise and takes a short time, we choose to not to include this measure in this study. All the measures can be found under the Instructions summary in the Appendix.

An abstract, context-free design such as the one proposed has indeed the disadvantage of decreased external validity. However, in the case of our research question, this sort of design is optimal because it restricts possible belief formation and framing effects. Thus, the possible psychological motivations of subjects are reduced to those that we particularly look for. Still, in case no effect is found in the experiment, it might be the case that the slippery slope effect is context-dependent and real-life elements such as uncertainty are necessary for it to manifest. Subjects are incentivized through a lottery draw of one participant to be paid, once all

responses are collected. All subjects are asked for an email contact prior to the start of the decision-making part. The email serves only for contact purposes in the case of a draw and will not be used for individual analysis of data. All emails are deleted after the draw is made and this is made explicit in the instructions. Earnings for each subject are determined as the sum of their initial endowment of 238 units and the total amount taken from their recipient. The conversion rate is equal to 30 units = 1 euro. The maximum amount that any subject can make is thus equal to 478 units/30 = 15.9 euros.

2. Theoretical predictions / hypotheses

One feature that is particularly different in this experimental design is that subjects are

attributed full responsibility for their decisions, which have a direct influence on the amount of their potential earnings. Hence, they are not passive reactors to a certain gradual increase in incentives or arguments. The actual findings of this study can thus differ from what is predicted based on the existing literature, on account of this difference in agency.

(15)

14 Also, the current experiment does not provide much uncertainty or ‘moral wiggle room’ (Dana et al., 2007) and the external rewards of the participants are directly proportional to the individual choices made. Ambiguity is only induced through the uncertainty that the dictator has regarding his recipient’s payoff, as well as through the slippery slope experimental design in which each step is not pre-announced (for the treatment group). According to standard

microeconomic theory, if subjects have a certain utility function (accounting for both their payoff and their internal cost of taking), the uncertainty over the exact sequence of decisions they are facing should not affect their decision. Hence, the entire sequence is not presented upfront, having each decision be taken anew. This supports the update of decision values at each steps and prevents premature decision making about the desired final distribution and thus reinforcement of self-control mechanisms (such as deciding upfront over the desired distribution and using it as a guide when deciding at each step).

For selfish, rational agents, the dominant strategy in any of the groups is taking the maximum possible amount (240 in this experiment). Similarly, subjects are predicted to be consistent in their choices and not change their decision once they are given this possibility at the end of the survey. However, based on previous empirical studies, other-regarding preferences can be incorporated into one’s utility function, as well as inequity aversion and the cost of morally misbehaving. The utility function of any dictator might take the simplified form: Ui = πi + απ-i - c(ti), where πi is the payoff to self, π-i is the payoff of the recipient, α is the degree of altruism and belongs to the interval [-1, 1], and c(ti) is the cost inflicted by taking the amount ti.

Following this rationale, a series of small unethical choices should facilitate unethical behavior, given that the emotional / interior cost is relatively low at each step and can be easily absorbed once the decision is made. Taking in small, insignificant amounts relative to the entire pot, rather than in one go, could also be analogous to the self-justification tendency of cash substitutions, such as taking home office supplies, or taking more tokens instead of cash in a payment self-collection context (Barkan et al., 2015). Hence, taking a small amount at a time could be seen as more justifiable and less pervasive than taking a large amount at once. We thus have the following hypotheses:

(16)

15

Hypothesis 1: The sequential framing of the decision to take will result in higher transfer values

in the treatment group, compared to the corresponding control group, proving a slippery slope effect.

As Welsh et al. (2015) point out, a gradual increase in unethical behavior is facilitated by moral disengagement. Taking an additional unit in a situation in which one has already been taking is more easily justified than when the possibility of taking is new and suddenly presented.

Moreover, it is more likely that the self-control mechanisms of subjects are worn out over a long series of unethical decisions, and hence it can be assumed that disengaging morally occurs more often in the treatment design.

Hypothesis 2: Over this series of decisions, people will become more morally disengaged when

taking gradually than at once. In other words, the respondents in the treatment group will elicit a higher moral disengagement score than their control group counterparts.

Moral self-assessment makes use of a balance sheet moral tool – a mechanism in which when people exceed their planned or ideal moral selves, they feel entitled to act less morally in another situation, in a compensatory manner (De Cremer, 2009; Shalvi et al., 2015). This mechanism might explain a potential high reversal rate in the additional reassessment decision that subjects are presented with after each treatment, meaning that a high transfer is reversed and replaced with a low or moderate one. However, given that they do not profit from the first moral/immoral choice, this assumption might be somewhat ambitious.

On the other hand, taking in several steps as opposed to at once can feel less decisive. Subjects might lose track of their decision making process and thus be inclined to re-evaluate it and change it once given this chance.

Hypothesis 3: People are more likely to reassess their decision when they take in a gradual manner. A larger reassessment rate in the treatment group, as compared to the control, would confirm this hypothesis.

(17)

16

4. Results

1. Descriptive statistics

Firstly, some descriptive statistics are needed in order to get an overview of the data. A total of 109 respondents completed the entire survey and thus could be used as participants. From these, three were excluded because they indicated in the self-filling explanatory question (which asked them to motivate their choices) that they did not understand the task at all. Another six have been dropped because they stated estimations of their recipient’s endowment below the threshold of 200, which we chose as to mark an understanding of the instructions. The remaining 100 participants were divided almost equally over the two treatments, with 49 respondents being randomly assigned to the control group and 51 respondents being randomly assigned to the treatment. Each person could fill in the survey only once and thus a between-subjects design was used (this was double checked by looking for doubles in the mail addresses provided). The sample comprised 68% females and the mean age of the respondents is 24.8, with a mode of 23. Most of the participants are either of Dutch or Romanian nationality, as these are the characterizing majorities of the student samples asked for participation. 28% of all respondents are of other nationalities.

Before the actual experiment questions were displayed to the participants, they had to answer correctly three control questions. All subjects who completed the entire survey answered these questions correctly, as otherwise they could have not proceeded to following questions, yet a few still indicated that they did not understand the instructions and that they randomly chose their answers. Those were dropped out from the data, as indicated above.

Means of amounts taken are displayed in Table 1 with standard errors in parentheses. Since the endowment of all participants is the same, 238 units, we can look at amounts in absolute values. As seen below, the means between the two groups are not very different, with a slightly higher average amount taken in the control group. A variance ratio test shows that the

(18)

17 we used a paired t-test for comparing means, which confirms that the difference between them is not significant (t=0.48, p=0.62).

Table 1 – Descriptive statistics across treatments

control treatment

variable mean (s.d) min / max mean (s.d) min / max take 38.69 (73.65) 0 / 240 31.76 (68.76) 0 / 240 retake 41.63 (76.17) 0 / 240 30.57 (58.13) 0 / 240 changea _{0.20 (0.40)} _{0 / 1} _{0.35 (0.48)} _{0 / 1} estimation 268.75 (29.05) 200 / 365 271.15 (18.64) 220 / 300 riskb _{5.28 (2.31)} _{1 / 10} _{5.52 (2.08)} _{2 / 10} donationc _{2.24 (0.77)} _{1 / 4} _{2.52 (0.90)} _{1 / 4} nobleends 4.06 (1.70) 1 / 7 3.58 (1.57) 1 / 7 authority 3.53 (1.78) 1 / 7 2.64 (1.50) 1 / 7 stealing 2.46 (1.56) 1 / 7 2.60 (1.55) 1 / 6 credit 1.87 (1.31) 1 /7 1.66 (1.19) 1 / 6 moraldisd _{11.93 (3.39)} _{4 /20} _{10.50 (3.31)} _{4 / 17} female 0.71 (0.45) 0 / 1 0.64 (0.48) 0 / 1 age 23.36 (2.91) 17 / 32 26.13 (7.98) 16 / 52 RO 0.71 (0.45) 0 / 1 0.50 (0.50) 0 / 1 NL 0.06 (0.24) 0 / 1 0.15 (0.36) 0 / 1 FEB 0.69 (0.46) 0 / 1 0.62 (0.48) 0 / 1

a _{coded as 1 for a change in amount taken between the take and the retake step and 0 for no change} b _{measured on a scale from 1 to 10, with 1 representing the maximum risk-aversion}

c _{measured through the question ‘How often do you donate money to charity?’, on a scale from 1 to 4, with 1=Never,} 2=Rarely, 3=Occasionally, 4=Regularly

d_{measured as the summation of the scores of all 4 scale factors, thus giving an increasing range of [4,28], with 4=least} morally disengaged

Treatment group – A total of 30 (58.8%) subjects chose to not take the first 12 units offered. Out of those, nine participants indicated that they believed the endowment of their recipient is

(19)

18 lower than the minimum stated amount (the lower end of the range, 268), thus suggesting that they formed beliefs that could enhance their altruistic self-image. Four out of the total of 51 (7.8%) took the entire possible amount of 240 in 20 sequential steps.

Control group – 29 (59.1%) of the control group respondents chose the option to keep the given distribution of endowments, and thus not to take anything from their recipient. Eight out of those believed that the starting endowment of their recipient is below the minimal

communicated value of 268. Five out of the 49 (10%) took the maximum amount in the control group.

In standard dictator games, usually more than 60% of participants give a positive amount to their recipient, with a mean of around 20% of the endowment (List, 2007). Similarly, more than half of the subjects in our study maintained the status quo distribution across treatments, showing some kind of altruistic behavior given that the starting endowment of the dictator was smaller than the one of the recipient. Figure 1 shows the proportions across treatments of each distribution of amounts chosen by the respondents. Still, a larger proportion of people (25/100) chose a potentially egalitarian distribution (up to 60 units taken) than an advantageously unequal one (16/100). These are split rather evenly across the treatments, with no significant differences. Twenty, and twenty-one people respectively, chose distributions different than the starting one, hence the treatment had no influence over the decision of people to change the given distribution, or to start taking. A Mann-Whitney test confirms that there are no significant differences in the distribution of the taken amounts (z=0.33, p=0.73).

(20)

19 Figure 1 – Distribution of amounts taken across conditions

2. Hypotheses support

Table 2 provides means, standard deviations and correlations between the study variables, generated in Stata. Opposite to our expectations, there was no positive correlation between the treatment and the amount taken by subjects. Hypothesis 1 thus receives no experimental evidence in this case. The correlation has the opposite sign, yet it is close to 0.

Similarly, there was a negative significant correlation between the treatment and the moral disengagement factor, which comes against the predicted Hypothesis 2. This may again be explained by the potential opposite effect of the treatment in the present experimental setting.

Hypothesis 3 is confirmed by the data. Only four people from the treatment group decreased

the amount taken in the reassessment question. Five participants did so in the control group. This indicates that the reassessment option did not have the expected effect, in line with the absence of a slippery slope effect. Most people kept their initial chosen distribution, which

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0 12 24 36 60 72 96 192 240 % of total subjects in group amount taken treatment control

(21)

20 indicates constant preferences, unaltered by timing or framing circumstances, supported by classic economic theory. As can be seen in Table 2, there is no significant correlation between the treatment assigned and the decision to change one’s decision (in whichever direction). However, 12.2% of the respondents in the control group decided to increase their share

following the possibility to change their decision. In comparison, 23.2% of the treatment group respondents increased their share at the reassessment question, with the difference being significant at the 10% level (z=-1.39, p=0.08). This can perhaps be explained by the fact that the decision making in the treatment group appears much less definitive than the one in the

control group. In other words, subjects in the treatment sample do not know explicitly that when they choose to not take the next 12 units, the distribution up until that point becomes final. The control group question offers a better overview over all possible choices and perhaps this is the reason why a larger number of subjects chose to change their decision (in the final reassessment question) in the treatment than in the control.

3. Correlations between variables and regression

Additionally, we looked at other correlations as well. Dummies for being Romanian and Dutch were also created and looked at, resulting in: Romanian nationality is linked, -0.18 (p=0.01), with a slight decrease in the amount taken, as well as a decrease in the frequency of charity donations (-0.18, p=0.06). In the opposite direction, Dutch nationality is weakly correlated, 0.27 (p=0.006), with the amount taken.

Separating the data for each of the two treatments, in the treatment group we find a rather high, significant positive correlation between risk attitude and amount taken (0.28, p<0.05), which is not present for the control group sample. This fact shows that the uncertainty specific to the treatment framing of decision making acts as a hindrance for the most risk-averse. The same correlation can be traced between risk attitude and moral disengagement (0.27, p=0.05), which indicates that as risk-aversion decreases, moral disengagement increases. As for the control group, the only significant correlation found is between moral disengagement and estimation of the recipient’s endowment (-0.28, p<0.05), which suggests that estimations

(22)

21 become lower as people become more disengaged. As there is no relation between amounts taken and estimations, it is difficult to assess to what degree these serve as self-serving beliefs. The two treatment-corresponding correlations tables can be found in the Appendix.

The moral disengagement variable represents an aggregate of the four dimensions included in the end questionnaire – stealing, authority obedience, noble ends and taking credit. The reliability of the scale, however, is rather low (α=0.27), with an almost null item correlation (0.08). In their study, Welsh et al. (2015) use two scales based on the same model, one with two statements and a reliability of 0.7 and another with eight statements and a reliability of 0.84.

We also looked at the scale factors separately, in relation to the other study variables. However, the only significant correlations appears to be between the ‘taking credit’ variable and estimation (-0.30, p=0.0021) and between treatment and ‘authority’ (-0.26, p=0.0086). This suggests that people may feel more in control when they refuse (or accept) the proposition to take a small isolated amount. Moreover, both the authority and the noble ends measures are significantly correlated with the taking credit factor. The female dummy is again negatively correlated with three of the scale dimensions, reaching some significance threshold with each. This fact strengthens the finding that females are inclined to be more morally-engaged and not easily endorse the self-serving motivations of questionable behaviors.

(23)

22 Table 2 – Means, standard deviations and correlations among study variables

Variable Mean s.d. 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 1.treatmenta _.51 _.50 _- 2.take 35.16 70.92 -.04 - 3.retake 36.52 68.30 -.08 .80*** - 4.changeb _.27 _.44 _.17 _.12 _.09 _- 5.riskc _5.41 _2.19 _.05 _.18* _.16 _.002 _- 6. donationd_2.39 _0.85 _.16* _.009 _.07 _-.009 _{.27*** -} 7.moradise _11.21 _3.41 _-.21** _.13 _.14 _-.05 _.03 _-.13 _- 8.estimation 269.98 24.21 .04 .06 .07 -.09 .21** .05 -.11 - 9.age 24.77 6.18 .22** -.08 -.08 .02 -.02 .13 -.22** -.10 - 10.female 0.68 0.46 -.07 -.22** -.06 .07 -.18* .13 -.17* -.05 -.09 -

a_{coded as 1 for treatment group and 0 for control group}

b _{coded as 1 for a change in amount taken between the take and the retake step and 0 for no change} c _{measured on a scale from 1 to 10, with 1 representing the maximum risk-aversion}

d _{measured through the question ‘How often do you donate money to charity?’, on a scale from 1 to 4, with 1=Never, 2=Rarely,} 3=Occasionally, 4=Regularly

e_{measured as the summation of the scores of all 4 scale factors, thus giving an increasing range of [4,28], with 4=least morally disengaged} ***p<0.01, **p<0.05, *p<0.1

(24)

23 Additionally, we ran some simple regressions to confirm the above findings. It appears indeed that there is no effect of the treatment on the amount taken (dependent variable), nor on the decision to maintain the status quo as opposed to taking. Table 3 summarizes this data.

Table 3 – Regression output

variable coefficient (s.e.) p-value treatment -12.52 (14.21) 0.381 risk 4.68 (3.27) 0.157 donation -1.02 (8.65) 0.906 moraldis 1.31 (2.08) 0.530 female -26.27* (15.28) 0.089 age -1.30 (1.16) 0.266 FEB 16.19 (14.71) 0.274 NL 58.02** (24.25) 0.019 RO -19.51 (15.98) 0.225 constant 48.91 (55.09) 0.377 R-squared 0.2042 N 100

As can be seen above, the only significant effects are of the gender and Dutch nationality variables. As expected and as also indicated by the correlations analysis, being a female

negatively influences the amount taken, decreasing it on average by 26 units. Coming from the Netherlands positively impacts the amount taken, increasing it by 58 units. The other variables besides treatment have the expected sign, yet lack significance. Treatment has a negative coefficient, strengthening the fact that there is no real slippery slope effect at best. The effect, if any, seems to be in the opposite direction than expected, for which several explanations are provided further.

(25)

24

5. Discussion

1. Interpretation of results

A clear majority of the subjects preferred to keep the status quo distribution, even if it was disadvantageous to them and they could have altered it to their own benefit or just to more egalitarian levels. Looking at the self-provided motivations that respondents filled in, a common reason for this is attributed to the fact that their recipient is a poor individual / represented by a charity. Out of the entire sample, 28% stated that they would not want to take money from the charity or that their recipient can benefit more from the extra units, regardless of the treatment. The results indicate nonetheless that theories of dictator game giving including other-regarding motivations cannot be easily dismissed. However, given that we used a taking dictator game, alternative interpretations of the subjects’ behavior should be considered. Taking in this case might weight more or bear a larger internal cost than not giving when there is a giving opportunity. Hence, it is difficult to assess to what extent the recorded results can be attributed to pure altruism, as subjects did not have an opportunity to act upon that in this current experiment (for example a treatment in which they could also give).

Another researched social preference that our experiment confirms to some extent is inequity aversion. Approximately a quarter of all participants in both treatments take between 12 and 60 units, thus suggesting some sort of egalitarian preference over the distribution of units. The self-provided motivations strengthen this fact, with 16 of them (16%) explicitly stating that they wished a somewhat equal amount of units as their recipient.

The estimation of one’s recipient’s endowment serves as a potential indication of belief

formation. As the estimation drops, it may be expected that the amount taken in the first place also decreases, suggesting that subjects form self-serving beliefs that could help motivate their decision to take a certain amount from their recipient. Moreover, the moral disengagement scale serves the same purpose of measuring whether moral beliefs are in line with the extent of unethical behavior undertaken by participants. Hypothesis 1 predicted that participants would more likely be taking from their recipient in the treatment than in the control, yet this effect

(26)

25 was not proven. A curious finding is that the amount taken or retaken by subjects (which acts as our measure for unethicality) is not significantly correlated with the degree of self-reported moral disengagement. This suggests that people in this study did not form self-serving beliefs about moral attitudes in line with their own moral behavior, which comes in contradiction to several other studies.

Given that the frequency of donation is also positively correlated with the treatment condition, it might be the case that people in the treatment group are in general more altruistic and have stronger ethical values, hence are less morally disengaged as well. This can help explain the negative significant relation between moral disengagement and being assigned to the treatment group.

Furthermore, it appears that as the self-reported risk tolerance of participants increases, so does the amount they take from their recipient, as well as their estimation of the endowment and their donation frequency. Following a commitment to an unethical act, self-measured risk aversion decreases, suggesting that people balance their questionable behaviors by taking a more risk confronting attitude; this might be a natural adaption given that punishment sometimes follows wrongdoings.

Being a female and the taken amount, as well as moral disengagement, are negatively

correlated. This is consistent with the seemingly stronger moral engagement amongst women that has been found by other studies as well (Dollar et al., 2001; Glover et al., 2002). Heinz et al. (2012) show that female dictators exhibit significantly more reciprocity in a real-effort dictator game. However, they also find that there is no difference between taking-rates in a windfall money treatment, where the money is not earned through a real-effort task, which does not stand in line with the findings of this paper. Gender studies in behavioral games are plenty and there is no clear consensus on a general conclusion; it also is beyond the scope of this paper to assess it.

The fact that the data suggests a slight opposite effect of the slope can be attributed to the artificial setting and the very short time between decisions. Another explanation for this may be that taking in several steps leads to worse feelings than just taking once, regardless of the

(27)

26 amount (in this context). A repeated questionable action feels worse to the agent than a single one. The repeated decisions to take, even in small amounts, may create a kind of moral

reminder that increases fairness. It is shown that signing the oath on forms at the beginning of reporting significantly reduces dishonesty (Shu et al., 2012), for which the main explanation given is that signing triggers an increase in the salience of moral standards and thus instills a fairer behavior amongst subjects. Also, it is possible, as some participants suggested, that the amount was too small to make the cost of stealing worthwhile.

Given the low amount of people taking the first 12 units, it is possible that the amount chosen for each of the sequential steps to have had the opposite effect than expected initially.

Combined with the rest of the design elements, it might have influenced people to think that it is not worth it to incur the cost of taking for such a negligible amount, hence the external reward was not enough of an incentive for taking the first (and following) steps. Another supposition is that taking at once, in a singular decision step, feels less bad than pursuing the action of taking over several steps, thus choosing to behave unethically more than once. This may be supported by the higher number of people who choose to stop after taking the first 12 units, relative to the control group and to the other distributions in the treatment. In this sense, it is possible that the amount of decisions that one makes for doing something questionable also plays a role. Yet the most well-grounded explanation probably concerns the particular choices of this study, namely having a charity / poor individual as a recipient. As most self-provided motivations tell, the circumstances of their recipient played a decisive role in the decision taken by many participants.

2. Limitations

There are a few limitations that are prone to occur due to narrow access to research resources. First of all, given the impossibility to financially reward all subjects, findings cannot be rendered final or indisputable. A replication of the experiment with real monetary incentives would be needed in order to correct for this. However, subjects are incentivized by a random draw of one

(28)

27 of the respondents (and their recipient) to be paid, which should increase the accuracy of the results.

Secondly, the somewhat limited sample size and the homogenous profile of the subjects (most being fellow students in the field of economics) might add biases to the results. Distributing the survey online is a way of limiting this effect and reaching a larger, more heterogeneous pool of subjects. Nonetheless, the study should be regarded as a pilot and results treated as merely indicative.

Besides these administrative limitations, the external validity question comes to mind. While the design of the study is kept as abstract and simple as possible in order to isolate effects, one can always say that the real world decision making is never as clear-cut. A slippery slope is rarely ever only about the sequential framing of choices, yet it is important to determine how strong of an influence this particular factor has. In relation to this, one disadvantage of the present design in terms of external validity is that it isolates an ethical choice. It is usually the case that people are facing several ethical decisions in a certain time span and their sequence or the history of decision making is highly influential for the next choice to be taken (De Cremer, 2009).

3. Further research

Given the results of this study, a replication of the experiment at a larger scale and in proper laboratory conditions (for increased control) should provide stronger conclusions. The insurance of anonymity and immediate payment are important elements that could not be secured in this study. While we only focused on the sequential display of amounts as defining the slippery slope in the context of the dictator game, a more complex design might increase external validity and provide a more authentic context for subjects. Dimensions such as costs of withdrawing from the slope, more ambiguity or risk of getting caught/punished can create a more externally valid environment. A recipient with a similar background, rather than a charity, might lead to different responses from dictators. Hence, a re-design of the current research should take care of the above elements. Moreover, in terms of choices of the particular design

(29)

28 elements in this paper, some changes can be suggested. Firstly, the amount to take at each step, as well as the total number of steps, should be reconsidered in order to provide a clearer incentive to start taking. The suggested amount should be higher and provide additional monetary payoffs, while the number of steps can be reduced as 20 almost identical questions can discourage the subjects to continue taking. Finally, the supporting questionnaire could better assess preferences and behaviors, such as disposition towards altruistic actions or motivations for the choices made.

Various other games, such as the common pool game, might provide useful settings for such research. An increased time-frame and a larger dispersion of decision points should also be considered, as they better provide the setting for a slippery slope. In the same time, for an extended empirical research on slippery slopes, other drivers of the effect should be

considered. Definitions of slippery slopes often contain a mechanism through which an initial decision or set of decisions increases the likelihood that the subsequent decisions will be taken, thus reinforcing the ‘slope’ of the argument (Rizzo & Whitman, 2003). Such a mechanism has not been artificially created in in any empirical research study, to our knowledge. A way to achieve this sort of motion is by introducing costs of withdrawal for the decision makers, after they reach a certain point down the slope. This also resembles many real-world situations in which agents are ‘caught up’ in a slippery slope and withdrawal comes with certain risks or consequences. The appropriate control condition for this design involving costs of withdrawal has to account for those costs and integrate them into the outcomes presented to the control group subjects. Further research could also look into whether known consequences to the agents lead to a weakened effect of the slippery slope or the extent to which uncertainty plays a catalyzing role.

A slightly different research area could constitute moving the slippery slope to the positive domain, to see if achieving optimal outcomes is facilitated by a gradual change in the agent’s choice set. There is already evidence that gradualism can be a strong catalyst for socially optimal results. Ye et al.(2013) showed that coordination in groups is more successful when stakes are gradually increased over time, as opposed to starting or jumping to a high stake.

(30)

29 Similarly, groups coordinate better when the size of the group is increased gradually (Weber, 2006). An interesting next step would be to see whether altruistic behavior is also enhanced by gradual demands or by a sequential framing of sharing decisions. A change of framing in the present design – a giving dictator instead of a taking one- can provide a starting point for such research. It would be interesting to observe how results shift in response to such minor framing changes, when compared to the design used in this paper.

6. Conclusion

There are a number of studies that suggest that social-preferences interpretations of dictator game studies need revision. This paper aimed to provide another step towards this

reinterpretation of motivations and behaviors documented by experiments. Choice sets and framing have been proven to be subtle influences on decision making outcomes, through changing expectations and beliefs. In the same manner, we employ a between-subjects design that alters the way choices are structured and presented to the decision-makers.

In general, dictator-game findings are confirmed and no slippery slope effect is found in the present experimental research. Distributions over the amounts taken are similar across the two conditions, with a majority of around 60% of each group keeping the status-quo distribution. Most of the remaining subjects are inclined towards a more egalitarian split. The effect of the treatment framing appears to have an opposite effect than expected on the decision to take, thus suggesting that the slippery slope created here fails because of the highly abstract context and certain design elements or that there is indeed no such effect of framing in a dictator game setting measuring social preferences. Other findings are that females have stronger altruistic preferences, or a higher cost of unethical behavior, and that people of Dutch nationality take relatively more than other nationalities.

Moreover, a larger amount of people from the treatment group chooses to increase their share in the reassessment opportunity, suggesting that the uncertainty that characterized their decision framing affects belief formation and their consequent choices. It is possible that the

(31)

30 first step that the control group was presented with discouraged some subjects from taking and thus, when the option to reassess one’s decision was clearly presented at the end as definitive, part of these people decided upon their real preferences. This hypothesis is supported by the positive significant correlation between the risk factor (risk taking) and the amount taken. The correlation is particularly strong in the treatment group.

To conclude, there is still plenty of room for further investigations of the slippery slope effect in a variety of experimental settings. While the findings of this study have confirmed to a large extent existing literature on the dictator game, there remain unanswered questions about the role of gradualism in an unethical decision making context.

7. References

Websites http://www.businessknowhow.com/growth/ethics-survey.htm ,

http://www.bbc.com/capital/story/20140806-the-slippery-slope and

http://www.baylor.edu/business/ethics/index.php?id=87726 , retrieved May 15th 2016.

Bandura, A. (1991). Volume 1: Theory. Handbook of Moral Behavior and Development.

Bardsley, N. (2007). Dictator game giving: Altruism or artefact? Experimental Economics, 11(2), 122–133. http://doi.org/10.1007/s10683-007-9172-2

Barkan, R., Ayal, S., & Ariely, D. (2015). Ethical dissonance, justifications, and moral behavior. Current Opinion in Psychology, 6, 157–161. http://doi.org/10.1016/j.copsyc.2015.08.001 Clavien, C., & Chapuisat, M. (2015). The evolution of utility functions and psychological altruism.

Studies in History and Philosophy of Science Part C :Studies in History and Philosophy of Biological and Biomedical Sciences, 1–16. http://doi.org/10.1016/j.shpsc.2015.10.008 Dana, J., Weber, R. A., & Kuang, J. X. (2007). Exploiting moral wiggle room: Experiments

demonstrating an illusory preference for fairness. Economic Theory, 33(1), 67–80. http://doi.org/10.1007/s00199-006-0153-z

(32)

31 De Cremer, D. (2009). Psychological perspectives on ethical behavior and decision making. Dollar, D., Fisman, R., & Gatti, R. (2001). Are women really the “fairer” sex? Corruption and

women in government. Journal of Economic Behavior and Organization, 46(4), 423–429. http://doi.org/10.1016/S0167-2681(01)00169-X

Gino, F., & Bazerman, M. H. (2009). When misconduct goes unnoticed: The acceptability of gradual erosion in others’ unethical behavior. Journal of Experimental Social Psychology, 45(4), 708–719. http://doi.org/10.1016/j.jesp.2009.03.013

Glover, S. H., Bumpus, M. A., Sharp, G. F., & Munchus, G. A. (2002). Gender differences in ethical decision making. Women in Management Review, 17(5), 217–227.

http://doi.org/10.1108/09649420210433175

Heinz, M., Juranek, S., & Rau, H. A. (2012). Do women behave more reciprocally than men? Gender differences in real effort dictator games. Journal of Economic Behavior and Organization, 83(1), 105–110. http://doi.org/10.1016/j.jebo.2011.06.015

Helson, H. (1964). Adaptation-level theory.

Kahneman, B. Y. D., & Tversky, A. (1979). Prospect Theory : An Analysis of Decision under Risk Author ( s ): Daniel Kahneman and Amos Tversky Published by : Econometric Society Stable URL : http://www.jstor.org/stable/1914185 REFERENCES Linked references are available on JSTOR for this article :, 47(2), 263–292.

Lee, C., Kraeussl, R., Lucas, A., & Paas, L. J. (2008). A Dynamic Model of Investment Decision Making: How Adaptation to Losses Affects Future Selling Decisions. Tinbergen Institute Discussion Paper.

List, J. A. (2007). On the Interpretation of Giving in Dictator Games. Journal of Political Economy, 115(3), 482–493.

Mazar, N., Amir, O., & Ariely, D. (2008). The Dishonesty of Honest People: A Theory of Self-Concept Maintenance. Journal of Marketing Research, 45(6), 633–644.

(33)

32 http://doi.org/10.1509/jmkr.45.6.633

Milgram, S. (1978). Obedience to authority.

Offerman, T., & van der Veen, A. (2015). How to subsidize contributions to public goods: Does the frog jump out of the boiling water? European Economic Review, 74(October 2008), 96– 108. http://doi.org/10.1016/j.euroecorev.2014.11.007

Peer, E., Acquisti, A., & Shalvi, S. (2014). “I cheated, but only a little”: partial confessions to unethical behavior. Journal of Personality and Social Psychology, 106(2), 202–17. http://doi.org/10.1037/a0035392

Rilke, R. M., Schurr, A., Barkan, R., & Shalvi, S. (2016). One-by-one or all-at-once? Self-reporting policies and dishonesty. Frontiers in Psychology, 7(February), 1–7.

http://doi.org/10.3389/fpsyg.2016.00113

Rizzo, M., & Whitman, G. (2003). The Camel’s Nose is in the Tent: Rules, Theories and Slippery Slopes. UCLA Law Review. Retrieved from http://works.bepress.com/mario_rizzo/4/ Shalvi, S. (2016). Corruption corrupts. Nature, 531, 456–457.

Shalvi, S., Gino, F., Barkan, R., & Ayal, S. (2015). Self-Serving Justifications: Doing Wrong and Feeling Moral. Current Directions in Psychological Science, 24(2), 125–130.

http://doi.org/10.1177/0963721414553264

Shu, L. L., Mazar, N., Gino, F., Ariely, D., & Bazerman, M. H. (2012). Signing at the beginning makes ethics salient and decreases dishonest self-reports in comparison to signing at the end. Proc. Natl. Acad. Sci. U. S. A., 109(38), 15197–15200.

http://doi.org/10.1073/pnas.1209746109

Walton, D. (2015). The basic slippery slope argument. Informal Logic, 35(3), 273–311.

Weber, R. (2006). to Achieve Efficient Coordination Growth Managing in Large Groups. Growth (Lakeland), 96(1), 114–126. http://doi.org/10.1257/000282806776157588

(34)

33 Welsh, D. T., Ordóñez, L. D., Snyder, D. G., & Christian, M. S. (2015). The Slippery Slope: A

Self-Regulatory Examination of the Cumulative Effect of Minor Ethical Transgressions. Journal of Applied Psychology, 100(1), 114–127.

Ye, M., Asher, S., Casaburi, L., & Nikolov, P. (2013). One Step at a Time: Does Gradualism Build Coordination? Working Papers, (February).

(35)

34

8. Appendix

1. Aggregate summary statistics

variable mean std. deviation min max

take 35.16 70.92 0 240 retake 36.52 68.30 0 240 changea _0.27 _0.44 ₀ ₁ estimation 269.98 24.21 200 365 riskb _5.41 _2.19 ₁ ₁₀ nobleendsc _3.82 _1.64 ₁ ₇ authorityc _3.08 _1.69 ₁ ₇ stealingc _2.54 _1.55 ₁ ₇ creditc _1.77 _1.25 ₁ ₇ moraldis 11.21 3.41 4 20 donationd _2.39 _0.85 ₁ ₄ female 0.68 0.46 0 1 age 24.77 6.18 16 52 RO 0.61 0.49 0 1 NL 0.11 0.31 0 1 FEB 0.66 0.47 0 1

a _{coded as 1 for a change in amount taken between the take and the retake step and 0 for no change} b _{measured on a scale from 1 to 10, with 1 representing the maximum risk-aversion}

c _{measured on a scale from 1 to 7, with 1 representing the least agreement}

d _{measured through the question ‘How often do you donate money to charity?’, on a scale from 1 to 4,} with 1=Never, 2=Rarely, 3=Occasionally, 4=Regularly

(36)

35 2. Correlations among variables in the control group

0. 8361 0. 9421 0. 2124 0. 9057 0. 1067 0. 6453 0. 3924 age - 0. 0303 - 0. 0106 0. 1813 - 0. 0174 - 0. 2333 0. 0674 0. 1249 0. 1252 0. 2273 0. 5114 0. 5095 0. 1419 0. 8932 0. 0258 f emal e - 0. 2220 - 0. 1756 0. 0961 - 0. 0965 - 0. 2129 - 0. 0197 0. 3184 0. 9125 0. 7436 0. 4856 0. 4886 0. 2980 0. 1184 donat i on - 0. 0161 0. 0479 0. 1020 0. 1013 - 0. 1517 0. 2260 1. 0000 0. 4596 0. 3754 0. 4375 0. 1052 0. 2594 r i sk 0. 1081 0. 1294 0. 1135 0. 2342 - 0. 1643 1. 0000 0. 3828 0. 3099 0. 1580 0. 0436 mor al di s 0. 1275 0. 1481 0. 2048 - 0. 2895 1. 0000 0. 9139 0. 5054 0. 4019 est i mat i on 0. 0158 0. 0974 - 0. 1225 1. 0000 0. 8093 0. 7705 r eassessdu~y - 0. 0354 0. 0428 1. 0000 0. 0000 r et ake 0. 9752 1. 0000 t ake 1. 0000 t ake r et ake r easse~y est i ma~n mor al di s r i sk donat i on

Is there a slippery slope in the dictator game?