The judgment of honest behavior in a dice-game

(1)

The judgment of honest behavior in a dice-game

An experimental study on norm violation

Master thesis written by:

Anish Bagga a.bagga@student.rug.nl (Student Role Number: S3199762)

Subject of studies:

Finance (M.Sc.)

Faculty of Economics and Business University of Groningen

Supervisor:

Dr. Chiara Lisciandra

(2)

II

Abstract

The present study conducts a dice-game experiment including a treatment where subjects are asked to observe other (confederate) participants playing the dice-game. With this procedure, the study intends to examine how (dis)honest subjects judge other observed participants’ behavior. The results show that subjects do not cheat in the dice-game experiment. Moreover, they judge observed behavior clearly as fair and unfair for the respective depicted behavior of participants. Furthermore, this study comprises a survey that examines moral values of subjects. Thus, fairness judgments on observed participants are analyzed with respect to moral values that represent norm compliance.

(3)

III

The degree to which an individual is able to act freely in a society is mostly constrained by a construct of norms that exist within that society. In social science, the term “norm” can be interpreted in many ways to explain the rules of individuals’ behavior. According to Cialdini, Reno and Kallgren (1990), for instance, descriptive norms are rules of behavior that are widely accepted by other individuals. In other words, they describe the “is-situation” of behavior, which explains what most people would do in a given situation. With their experiment, Cialdini et al. (1990) show that individuals litter more when they are exposed to a littered environment and see other individuals littering as well. Here “littering” is depicted as the descriptive norm. Moreover, injunctive norms differ from this perspective by accurately describing an “ought-situation” (i.e. how one should behave in a situation). Furthermore, injunctive norms relate more to the perception of other individuals on one self’s action or behavior. Related to the previous example, Cialdini et al. (1990) added another condition to the littering experiment where subjects see a (confederate) subject picking up litter. Consequently, the “ought-situation” here is depicted as “pick up your litter”, which results in subjects littering less, than in the previous example. As clearly specified before, there are many characteristics on how norms can be described. Besides descriptive and injunctive norms, Rimal and Real (2005) point out other studies’ findings in terms of different types of norms, such as subjective norms (Fishbein and Ajzen, 1975) or social norms (Perkins and Berkowitz, 1986). In a further work, Cialdini and Trost (1998) define social norms as follows: “Social norms are rules and standards that are understood by members of a group, and that guide and/or constrain social behavior without the force of laws. These norms emerge out of interaction with others; they may or may not be stated explicitly, and any sanctions from deviating from them come from social networks, not the legal system” (Cialdini and Trost, 1998, p.152). It is necessary to mention that norms are not continuously activated processes which permanently guide individuals to their executed behavior (Biel and Thøgersen, 2007), but are rather made salient to receive the focus of an individual (Cialdini et al., 1990). Also, Fehr and Gaechter (2000) confirm the concept of social norms and define it as a tool of regulating behavior within a society or social group that is based on shared beliefs and informal social sanctions.

(5)

2

composition is given, then one can speak of a well-performing society. Thus, individuals and all other entities or actors (ideally) have to follow the existing norms, which results in good efficiency. In contrast, bad efficiency would occur, in the case of norm violation. Now, extending this deliberation, one can argue that a society’s performance is successful and efficient when individuals are able to comply with norms in a longer run and also adapt to new norms. This suggests the existence of tensions between the maintenance of norm compliance and the occurrence of norm violation. In what follows, explanations of norm violation are presented. Van Kleef, Homan, Finkenauer, Gündemir and Stamkou (2011) define norm violation as a social concept or process that is omnipresent and versatile in its occurrence. Moreover, there has to be a solid opportunity that enables an individual to derive from present norms and to violate them (Shannon, 2000). Furthermore, van Kleef et al. (2011) define norm violation as a “behavior that infringes one or more principles of proper and acceptable behavior” (van Kleef et al., 2011, p.1). Norm violation is perceived when a displayed behavior does not fall into the personal view of acceptance and is therefore seen as inappropriate (Levine, Anders, Banas, Baum et al., 2000). The source of norm violating behavior lies within the individual itself or can be derived from other societal causes (van Kleef, Wanders, Stamkou, and Homan, 2015). In their study, van Kleef et al. (2015) argue that individuals are inclined to violate norms more often when they have a higher power status or social status. That is, individuals who think of themselves as more powerful tend to violate norms more often than individuals who seem to have less power. Van Kleef et al. (2015) also argue that the role of others is the main force that leads to a violating behavior.

The present research analyzes subjects’ judgments on the norm of honesty when subjects are exposed to a norm-violating situation. Moreover, the main interest of this study is to compare the tolerance of norm violation between honest and dishonest subjects within the context of a dice-game experiment. In addition, the experimental results are linked to the collected survey data to find more insights on subjects’ judgments on honesty. The reason why the survey is included in the experimental project is to observe whether certain moral values are good indicators of norm compliance or violation. These values and dimensions are selected from Schwartz (1992) and Hofstede and Minkov (2010), because they can be related to socio-normative behavior and can be applied to the above mentioned definition of social norms in the most suitable way.

(6)

3

type to determine honesty and dishonesty of subjects. Other studies also used the dice-game concept: Arbel, Bar-El, Siniver and Tobol (2014) investigated religiosity effects among students in Israel; Gaechter and Schulz (2016) analyzed intrinsic honesty and norm violations in various countries. The dice-game, in its most simple form, asks subjects to roll a die anonymously and report a number. Each side of the die is matched with a corresponding reward (for example: 1€ for reporting a “1”, 2€ for reporting a “2” […], and 6€ for reporting a “6”). Cheating occurs when subjects report other numbers than their actual die number. Based on a large sample size one can assume that each number has the probability of 16.67%. So, if subjects report higher numbers more often than the implied probability of 16.67%, then one can assume that subjects cheated by reporting a higher number other than the true die roll.

This study extends the dice-game by asking the subjects to watch two videos that contain confederate participants playing the dice-game. In the first set-up, participants play the dice-game honestly, whereas in the second set-up, participants play it dishonestly. This procedure allows the study to implement the observing condition and to find results regarding subjects’ judgments on norm violations. To be more precise, subjects are asked to indicate how socially appropriate they consider non-cheating and cheating behavior. As mentioned above, this study’s experiment links its results with a conducted survey1, which captures data of subjects regarding moral values with respect to the “parameters” formulated by Schwartz (1992) and Hofstede et al. (2010). The purpose of the survey is to find moral values that explain subjects’ (un)fairness judgments on observed participants with respect to norm compliant behavior.

General findings of the current study suggest that subjects neither cheat in this particular dice-game set-up nor do they stretch norms by cheating marginally. Additionally, subjects perceive other participants’ behavior clearly as fair or unfair when they are exposed to cheating and non-cheating observing conditions. A more detailed analysis tries to explain the influences of values on the subjects’ fairness judgments in respective treatments. However, due to data limitation problems the regression models cannot deliver any statistical significant clues on the relation between fairness judgments and moral values. The bottom line is that subjects in the underlying dice-game only exhibit honesty and socially correct judgments.

The remainder of this work is organized as follows: section 2 provides a literature review and presents the scrutinized hypotheses. Section 3 describes the methodology

1

(7)

4

regarding the survey data and dice-game experiment including the observing treatments. In section 4, the empirical findings are illustrated, whereas section 5 provides information on the limitations of the study. In addition to the limitations, section 5 also gives supplementary improvements for future research. Finally, section 6 presents the conclusion of this study.

2. Literature review

The current work employs two influencing information streams: first, the output of the survey data collection and second, the dice-game experiment. In the present study, the moral values and beliefs of subjects from mainly Dutch2 individuals are explored. Hereafter, moral survey parameters and the main implementation of experimental games are reviewed. This section concludes with formulating four working hypotheses based on the dice-game and previous studies.

It can be assumed that individuals from two different countries perceive norms and rules in a different manner. For instance, respecting punctuality is a stronger perceived norm in USA than in Morocco (White, Valk, and Dialmy, 2011). However, there are basic cultural and moral values that can be used to explain the variation of different behavior and perceived norms. The parameters are extensively described in previous renowned studies and researches by Hofstede et al. (2010) and Schwartz (1992, 2012) and can be employed to describe cultural groups, societies or individuals because these values are universally recognized and structured in a same way across groups (Schwartz, 2012). The present study also refers to their findings and formulates an independent survey so that the moral values of individuals are derived for further purposes such as the motivational behavior of being (dis)honest in the subsequent game experiment.

As the studies by Hofstede et al. (2010) and Schwartz (1992) are widely presented in research, a brief representation of the respective cultural dimensions and moral values are given in the following. Hofstede et.al (2010) divides culture into six major dimensions to illustrate the respective national culture. These dimensions are: power distance index, individualism versus collectivism, masculinity versus femininity, uncertainty avoidance index, long term orientation versus short term orientation, and indulgence versus restraint. Similarly, Schwartz (1992) analyzes eleven cultural values, which are described as:

2

(8)

5

direction, stimulation, hedonism, achievement, power, security, conformity, tradition, spirituality, benevolence, and universalism. Schwartz’s elaborated values suggest that there is “an universal organization of human motivations” (Schwartz, 2012, p.3). Nevertheless, social entities (cultures, societies, and individuals) give more or less importance to certain values. With regard to the underlying objective of this work it can be said that there are values that can be preferably used as indicators for either norm compliance or norm violation. Since norms prescribe a behavior for a given situation and tell us what is expected to do, the compliance to these norms depends on the values the social entities follow (Schwartz, 2012). Therefore the present study only employs seven particular values and dimensions from the aforementioned list.

Thus, the seven values are briefly described as follows: The first value used in the survey is “conservation and tradition” and embodies the behavior of respecting and following religious and cultural customs as well as committing oneself to its beliefs (Schwartz, 1992). According to Schwartz (2012), a shared belief is constructed with the help of common practices, symbols, and ideas, which are followed and supported by others as well. Secondly, the value “conformity” represents denying extensive impulses or actions, which might contrast with other people’s beliefs, values, and expectations towards prevailing norms and rules (Schwartz, 1992). This value supports the idea of a smooth interaction with (other) group members (Schwartz, 2012). With regard to norm compliance and the survey content of the present study, it can be assumed that the previous two values indicate the basic concepts of norm compliance, because they provide (behavioral) rules for social entities in general. Furthermore, they also implicate social sanctions in case of disregarding norms. For simple illustration: if an individual does not follow common practices in a ritual process, the individual can be excluded from the ritual process as result of non-compliance. The main driver of the third value called “security” can be described as the stability within a society derived from harmony or protection from dangerous “enemies” and the safety of kin (Schwartz, 2012). This value is also implemented in the survey, because it can be used as an indicator for norm compliance. The reason for inclusion is the content of the value. “Security” implies that one way of achieving society’s stability can be earned through safety and harmony which in turn strongly relies on individuals’ consensus of following or supporting similar norms.

(9)

6

risk seeking attitude. It also includes the stimulation of oneself towards exciting experiences (Schwartz, 1992). From that it can be derived that the risk seeking nature is stimulated when individuals break norms or rules. Moreover, the value “stimulation” relates with the value “self-direction”, since both stand for the comprehensive dimension called “openness to change” (Schwartz, 2012). Thus, the fifth value “stimulation”, expresses that subjects choose own life goals and achievements supported by individuality in terms of being creative and curious (Schwartz, 1992). Moreover, this value is determined by independence and freedom of own choices (Schwartz, 1992). “Stimulation” and “self-direction” together oppose the concept of “conservation” illustrated by the first three values (Schwartz, 2012). Since “stimulation” and “self-direction” depict the general dimension of “openness to change” (Schwartz, 2012), this study uses this concept for the purpose of indicating norm violation.

The sixth value is “individualism versus collectivism” and explains individuals’ integration within a society and if individuals put collective welfare or wellbeing before personal success (Hofstede, 2011). This value describes the interdependence of individuals in a society (Tsakumis, Curatola, and Porcano, 2007). The higher the individuality within a society the more the individuals are concerned about themselves as in contrast to collectivistic societies. With regard to tax evasion for instance, Tsakumis et al. (2007) suggest that a high individualistic society is less tolerant concerning tax evasion, because there are stricter regulatory systems discouraging to do so. The seventh value “uncertainty avoidance” depicts the degree to which an individual is able to bear the unknown future with given instructions derived from behavioral codes, rules, and other norms (Hofstede, 2011). According to Hofstede et al. (2010), weaker uncertainty avoiding societies do not feel uncomfortable with unknown situations. As a result, they do not try to minimize the anxious feeling through codes, laws, or rules. In the context of the present survey it can be said that high uncertainty avoiding indicates norm compliance. This is, because a higher importance given to uncertainty avoidance speaks for following rules, codes, and rules.

(10)

7

conformity, security, (high) uncertainty avoidance, and individualism (which is low collectivism) will provide a good indication for norm compliance and fair judgments. The more important individuals perceive the values to be, the less they are prone to cheat. In contrast, values such as self-direction and stimulation relate more with a judgment that displays higher tolerance for cheating behavior. This is because individuals who exhibit higher importance to these two values are prone to violate norms more “unresistingly”, because they are inclined to be stimulated by independence and risk-taking attitude which in turn might lead to violate prevalent norms.

The present study aims at explaining the relation between moral values and judgments of (dis)honest behavior. In the following, a review of studies that investigate this particular relation is provided. An interesting source, to which the present study can be connected to, is the study by Gaechter and Schulz (2016). In their study, they analyze norm compliance at the country level in relation to the intrinsic honesty of individuals. First, they derive a so-called “Prevalence of Rules Violation” (PRV3) index, which is developed by analyzing 23 countries in terms of their norm compliance. After building this PRV index - a proxy for norm compliance - they conduct a time-delayed dice-game experiment with young subjects (in the age of approximately 20 years). The time delayed dice-game guarantees that participants of the dice-game have not been involved in the previous PRV index calculations.

The dice-game in their study is strongly related to Fischbacher and Heusi (2013) and works in a similar way, where subjects are asked to self-report a number from a manual die roll. Hence, they are able to map the results of the PRV index with the results of the time delayed dice-game experiment and are therefore able to infer intrinsic honesty in a society. The outcome of the study shows that societies are more honest when they have a low prevalence for rules violation index relative to those with a high index. They conclude that rules violation on a societal level can affect the behavior towards violation attitudes on the individual level. However, a question stays unanswered in their paper, which can be connected to one of the objectives that is examined in this paper: How does (dis)honest behavior predict judgments of (dis)honest behavior? And which moral values do relate with (un)fairness judgments?

The idea of nationality as explanatory means for norm violation can be extended to further realms. A recent study by Andrighetto, Zhang, and Ottone (2016) examines

3_{It is the index for rule violation and is derived by evaluating information on corruption, tax evasion and}

(11)

8

cultural dishonesty found on the disposition of tax payments to local tax authorities. The extremes between two countries, Italy and Sweden, are analyzed in a tax compliance experiment. The results show that the average level of tax evasion does not differ significantly, but the amount and magnitude of tax evasion does. It can be said that Swedes are either very tax compliant or underreport taxes heavily, whereas Italians prefer to be partially dishonest in their tax declaration process. A reason for this can be found in the moral parameters which these two countries exhibit.

Other studies examine the corruption level of subjects from a particular country in relation to (dis)honest behavior in experimental games. Innes and Mitra (2013) give evidence that dishonest behavior (here: corruption) spills over from a peer-group. The effect, that dishonest subjects accept dishonest behavior from other participants more generously, is also analyzed in one of the scrutinized hypotheses. The results of Innes and Mitra (2013) also find support by the previous findings of Barr and Serra (2010), who analyze the prevalence of corruptive behavior of individuals4 from societies where a serious level of corruption is present.

So far some literature insights on the cultural and moral dimensions with respect to norm violating behavior on a country level are presented. Given that this work includes the dice-game to detect honesty, some related information regarding (dis)honesty in other experimental set-ups is presented in the following. The meta-analysis by Rosenbaum, Billinger, and Stieglitz (2014) investigate 63 experiments concerning honesty and truth-telling. The review strategy divides these experiments into six subcategories of game types and draws the conclusion, that every experimental game type reveals different levels of honesty that can be explained by the different understanding of (dis)honesty and dissimilar experimental set-up and task. Besides that, their research exhibits that there are ambiguous subjects which are honest, cheat unconditionally or are economically biased in the sense, that they do not choose the very best payoff, although there is no or small risk of detection.

Interestingly, the (pioneering) study by Gneezy (2005), tackles the issue of people’s reluctance to lie (i.e. behaving honestly). His findings conclude that people are sensitive to lying when it affects others independently from the gains they earn from lying. Similar results are also verified by other studies by Hurkens and Kartik (2006) and Gneezy, Rockenbach, and Serra-Garcia (2013). With regard to the present study one can argue, that subjects are not supposed to exhibit strong lying aversion, because the dice-game is a one sided game and

4

(12)

9

therefore does not affect any other players. The only obstacle is in the subject’s mindset or conscience itself. Mazar, Amir, and Ariely (2008) call this mindset the concept of moral beliefs. The idea of the self-concept maintenance is to find moral constraints in which dishonesty or norm violation is not regarded as (personal) norm violation as such and that executed behavior fits to one’s beliefs.

The idea of the present research is to find out whether (dis)honest subjects judge (dis)honest behavior more (un)fairly in the given observation treatments and which moral values could be predictors for the respective judgments. For the judgments, subjects are exposed to observe other participant playing the same dice-game. With that setting in mind the present research is able to derive four conditions or scenarios that can be further hypothesized. The dice-game is also implemented in the studies of Fischbacher and Heusi (2013) and Gaechter and Schulz (2016) where participants displayed honest and dishonest behavior. For the purpose of the investigated hypotheses, the present study also relies on these two behaviors and derives the hypotheses with regard to the observing conditions. Since the outcomes of the dice-game cannot be predicted, many concepts are applied to formulate the hypotheses. First, it can be expected that subjects are honest by playing the dice-game truthfully because some subjects in general do not prefer to lie per se (Vanberg, 2008) or when lying affects others (Gneezy, 2005). Also, Charness and Dufwenberg (2006) show that people do not lie because it creates a guilty feeling. From that, it can be assumed that the social norm of honesty is present and therefore subjects in the present dice experiment are averse to cheating behavior. With regard to the fairness judgement, it can be said that honest subjects appreciate honest depicted behavior in the observing treatments, because subjects generally follow the norm of reciprocity (Gouldner, 1960). The norm of reciprocity can be explained by Gouldner (1960) who suggests, that “people will usually help those who helped them” (Gouldner, 1960, p.173). With respect to the set-up in the present experiment, it can be said that subjects who played the game by following the norm of honesty expect others to play honestly as well and therefore appreciate this behavior in return. The first hypothesis can be formulated as follows:

Hypothesis 1: Honest subjects find it socially appropriate if other observed participants behave honestly in the dice-game experiment.

(13)

10

of honesty is prevalent and therefore they do not cheat. Any deviating behavior leads to a violation of honesty and can be considered as an act of norm violation as defined in Levine et al. (2000). There it is stated that a behavior which is not in the personal realms of acceptance is considered as norm violation (Levine et al., 2000). Thus, subjects who are honest might sanction dishonest behavior (Cialdini and Trost, 1978) or will not tolerate it. Based on this and on the experimental set-up, the second hypothesis can be formulated accordingly:

Hypothesis 2: Honest subjects find it socially inappropriate if other observed participants behave dishonestly in the dice-game experiment.

Fischbacher and Heusi (2013) and Gaechter and Schulz (2016) show in their studies that subjects also tend to cheat in the dice-game. Based on their findings, this research also expects their subjects to behave untruthfully. With respect to the judgments on observed behavior, the present study is able to assess also two further hypotheses strongly related to the previous two in terms of their structure. As the previous two hypotheses focus on honest behavior, the next two hypotheses analyze dishonest subjects and their judgments. The degree to which a subject is willing to behave unethically, while maintaining a positive self-image, is dependent on the flexibility of rationalizing unethical behavior (Mazar et al. 2008). In the present experiment, subjects have the opportunity to cheat marginally (i.e. reporting a “4” instead of the obvious income-maximizing die roll “5”) to maintain their self-image. According to aforementioned social norm definitions (Cialdini and Trost, 1978 and Cialdini et al., 1990) and the context malleability within the experiment, subjects might have recognized the opportunity to behave dishonestly and personally see this unethical behavior as morally acceptable. From this it can be assumed that they perceive this (marginal) unethical behavior as a justifiable behavior which should be followed by others as well. If this behavior is not followed by others, then cheating subjects might not appreciate observed honest behavior. Thus, the third hypothesis is composed as follows:

Hypothesis 3: Dishonest subjects find it socially inappropriate if other observed participants behave honestly in the dice-game experiment.

(14)

11

unethical behavior is higher among similar group-members. Thus, when cheating subjects in the present study compare themselves with the depicted confederate participants, they find self-affirmation and therefore appreciate dishonest behavior. In a similar vein, Gino and Galinsky (2012) investigate the psychological closeness among subjects. They find that (un)ethical behavior is contagious and influences the perception of (dis)honesty, when subjects are (fictively) connected to (dis)honest participants. For instance, they indicate that subjects, who take the perspective of selfish participants in an experimental setting, would see particular selfish behavior as less morally incorrect, than subjects who are not asked to take the perspective of a selfish participant. The fourth hypothesis can be concluded as such:

Hypothesis 4: Dishonest subjects find it socially appropriate if other observed participants behave dishonestly in the dice-game experiment.

3. Methodology

The present study operates with two major data sources of which the first one extracts information via an online survey on the individuals’ moral values. The second data input comes from an experimental dice-game derived from the study of Fischbacher and Heusi (2013). The following subsection describes the survey design. Then, subsection 3.2, explains the original dice-game set-up implemented by Fischbacher and Heusi (2013). After that, subsection 3.3 gives a description on the particular design and method of the dice-game used in the present work. Finally, the subsection 3.4 illustrates the method of data analysis.

3.1 Survey design

One data source of this present study is an online survey that investigates judgments on moral values which might give a first indication on (dis)honest behavior. The selected values5 are chosen from the work of Hofstede et al. (2010) and Schwartz (1992). The online survey6 collects data concerning values - answered on a five-point Likert scale - and also sociodemographic information. For example, one instrument in the survey asks subjects to give their perception on “having respect for tradition (preserving time-honored customs)”. Here subjects have the option to state their opinion by placing a mark on one of the following

5_{See section 2 for further details.}

6_{See: https://rug.eu.qualtrics.com/jfe/form/SV_0Bpc3rdmHSjIC0t ; last online access checked on 9}th_{of May}

(15)

12

five points: “Opposed to my principles” (1); “Somewhat opposed to my principles” (2); “Neither opposed to nor important” (3); “Somewhat important” (4) or “Of supreme importance” (5). Subjects for whom certain statements have a higher importance in their life or environment are inclined to give higher ratings (four or five points on the scale). Other instruments, such as “collectivism” or “norm violation” follow the same structure but only differ in the description of the Likert scales. For a more accurate overview, the complete online survey can be found in appendix A.

3.2 Original dice-game

The second data input comes from an experimental dice-game derived from the study of Fischbacher and Heusi (2013). The purpose of the dice-game experiment is to find under which conditions subjects are willing to behave (dis)honestly while playing the dice-game. In the simple experimental dice-game individuals are exposed to a single decision making situation where they are asked to roll a die privately and note down the figure that determines the payoff (reward). Thus, lying occurs when participants report different figures than the actual die showed. However, the experimenter is not observing subjects during the dice roll procedure and only receives the reported number of each subject at the end of the experiment. The following payoff matrix in table 1 depicts the die figures [1, 2, 3, 4, 5, 6] and their matched payoff structure of [1, 2, 3, 4, 5, 0] MU (monetary units) respectively. The payoff of 0 MU for roll number “6” is implemented to incentivize subjects to report other numbers than “6”.

Table 1: Payoff matrix of the dice-game (Fischbacher and Heusi, 2013)

Die number: ₁ ₂ ₃ ₄ ₅ ₆

Matched payoff in Monetary Units: ₁ ₂ ₃ ₄ ₅ ₀

(16)

13

probability of 16.67%. By aggregating subjects’ self- reported numbers Fischbacher and Heusi (2013) show that subjects report the number “6” less frequently than the expected 16.67% and report other higher numbers such as “4” and “5” more often than 16.67%. Therefore it can be concluded that subjects show traces of dishonesty. This game design allows subjects to act anonymously which leads us to assume that subjects are induced to cheat fully or marginally. Moreover, subjects are not observed by any other subjects, participants or the experimenter, thus there is no possibility of detection which consequently enhances the chance of behaving untruthfully.

3.3 Experimental design

The present study employs a dice-game for two reasons. Firstly, it is a simple game that can be conducted with any type of subjects to explore (dis)honest behavior. Secondly, the underlying true distribution of the die number is known, thus no other control experiments are needed to verify (dis)honesty (Fischbacher and Heusi, 2013). However, the original experiment by Fischbacher and Heusi (2013) is not able to detect lying on the individual level, whereas the present study is able to track back (dis)honest reports, due to its implementation in a laboratory lab and by using Microsoft Excel. 7

In the following, the experiment is described in more detail. The basic procedure of the present experiment can be divided in two steps. First subjects play the dice-game by themselves and report their personal outcome of the die roll. The die rolling process consists of tossing a virtual die that is programmed in the Excel file and practically does what a real manual die would do. The only difference between a real die and the virtual die is that the rolling process is set into motion by clicking on a visual button. The reason why the experiment uses a virtual die is to ensure complete anonymity to all subjects during the conducted experiment because otherwise subjects might feel observed and therefore would not be inclined to cheat.

Second, in order to observe the judgment on norm violations, subjects are asked to watch a video in which other participants play the dice-game by themselves. The observing process is constructed in four stages and includes watching prepared videos of confederates playing the dice-game honestly and dishonestly. First subjects are asked to watch a video of a (confederate) “participant A” playing the dice-game honestly. Then in the second stage, subjects are requested to give a fairness judgment on the observed honest behavior. In the

7

(17)

14

third stage, the same subjects are asked to watch a video that shows a (confederate) “participant B” (different participant than “participant A”) playing the dice-game dishonestly. Lastly, they give a fairness judgment on dishonest behavior here as well. Judgments are given on five-point Likert scale indicating unfairness and fairness as the extreme scale points.

The dice-game experiment of this study was conducted in the experimental laboratories at the University of Groningen, Netherlands. Subjects were directly recruited from different classes in Business Ethics and were gathered in a room to receive a short verbal introduction about the experimenter and experimental steps. Subjects were told that they could receive payoffs according to their performances in the dice-game experiment. There were clearly no verbal hints given about dishonesty, honesty, cheating, and non-cheating behavior. Before the experimental game could start, each participant received a manual sheet8 for explanation and introductory purposes. The manual sheet also informed the subjects when to open the Excel file including the virtual dice-game. The present experiment uses also a similar payoff matrix (see table 2) as in the study of Fischbacher and Heusi (2013).

Table 2: Payoff matrix of the present dice-game experiment

Die number: 1 2 3 4 5 6

Matched payoff in Payoff Units: ₂ ₄ ₆ ₈ ₁₀ ₀

To stimulate dishonest behavior, the payoffs were doubled for each number on the die roll. Nevertheless, the zero payoff for a rolled “6” was retained so that subjects would have the chance to correct it by reporting other numbers than “6”. Due to budgetary constraints it was only possible to provide candies as payoff-units. Subjects were told to take out the amount of candies equal to the number they have reported in the dice-game. If any further cheating happened at the payoff collection, it was not recorded as part of the experiment.

The present study thus extends the basic concept of the dice-game with a second phase that asks participants to observe and to give a fairness judgment on others playing the game honestly or dishonestly. The overall design of the experiment allows us to combine extracted information from the survey and the dice-game experiment. Thus, individual attitudes towards moral parameters, which are derived from the survey’s results could be matched or tested

8

(18)

15

against the results of the experimental game. This procedure should enable us to gain further insights on reaction to other individuals’ violation or compliance of norms.

3.4 Data analysis

First, a survey validation is conducted to determine the degree of validity, reliability, and sensitivity of the survey instruments. For this, an Exploratory Factor Analysis (EFA) is conducted followed by Cronbach’s alpha and sensitivity analysis. The factor analysis is conducted to determine the underlying factorial structure (Bentler and Weeks, 1980) of the survey. The present study uses the subscale averages to present and analyze the descriptive statistics and correlation matrix. A correlation analysis is undertaken to evaluate the relation between variables.

With regard to the experimental dice-game results, descriptive statistics of the subscale averages are presented to indicate (dis)honest behavior and to infer some conclusions on the formulated hypotheses. This also includes the statistical analysis of subject’s first (actual) die roll and reported number in the dice-game. Here, a t-test is conducted to check whether there is a difference in the means of the two samples (actual die roll and reported number). Then, correlation matrices are examined to find significant survey values that are in relation to both (un)fairness judgments in the dice-games. This also includes a presentation of multicollinearity diagnostics. Finally, multiple regression analyses9 are conducted with each judgment as the dependent variable and the relevant moral values for norm compliance as the explanatory variables. The equation looks as follows, and depicts as the constant and as the factors for each moral value taken from the survey.

= + + + . . . + + , ℎ = 1 10

4. Empirical findings

4.1 Description of pool

The online survey was conducted during the first two weeks of February 2017 at the University of Groningen, Netherlands. Subjects from the bachelor of science of an economics and business program were informed about the online survey and the link was shared through

9_{Analyses are performed in the statistical software IBM SPSS and Microsoft Excel.}

10

(19)

16

the university platform. A total of 153 participants took part in the survey of which around 80% grew up in the Netherlands. There were four German and three Turkish subjects, who took part in the survey. Other nations such as Thailand, Brazil and South Korea were also marginally represented with one to approximately three participants each. The gender distribution is balanced with 74 males and 79 females. The average age of participants is 21.7 years.

Roughly 6 weeks after the online survey, subjects from the same pool were recruited to participate in the dice-game experiment. In total 51 subjects participated in the dice-game experiment within two days11. The same underlying distribution in gender (26 female and 25 male participants) and age as in the survey data can be observed. However, there is less variance in the origin country which means that close to 85% of the subjects have a Dutch background.

4.2 Survey validation and analyses

In the first stage, an Exploratory Factor Analysis (EFA)12 was conducted to determine the underlying factorial structure and to evaluate validity (Bentler and Weeks, 1990; Hair, 2009; and Matsunaga, 2010). Then, Cronbach’s Alphas were computed for the reduced scales to determine reliability. Finally, sensitivity was evaluated by analyzing skewness, kurtosis, minimums and maximums of the survey items.

Validity is determined by using the methods of an EFA which indicates whether or not the survey items represent the underlying construct adequately (Hair, 2009). The EFA in the context of this study applies the method of a factorial validity of the survey. Thus, the EFA was conducted using all 48 items while performing the Principal Axis Factoring method. This method was chosen as it is the most commonly used, and is also robust to potential deviations from normality (Yong and Pearce, 2013). Oblique rotation was additionally used for the underlying factor analysis to increase the interpretability of the results (Matsunaga, 2010). To be more precise, a direct oblimin rotation method was applied. Oblique rotation was chosen over orthogonal rotation because some degree of correlation is expected between the factors (Hair, 2009 and Pallant, 2011). This analysis yielded a Kaiser-Meyer-Olkin (KMO) value of 0.699 which suggests that a factor analysis is almost suitable when respecting the acceptable threshold of 0.7 (Hair, 2009). Furthermore, the value of the Bartlett’s test for sphericity is significant (p=0.000), therefore the factor analysis is appropriate (Pallant, 2011).

11

On the 23rd and 24th of March 2017

12

(20)

17

The Kaiser’s criterion is used to extract possible factors at the threshold of Eigenvalues greater than 1 (Hair, 2009). The underlying output of the EFA suggests 15 factors that explain 68.58% of the variance. The high number of the factors cannot be seen as an ideal solution with respect to the present survey structure used in this study. As a consequence, other indicators for the correct factor extraction were taken into account (Hair, 2009). The scree plot13 serves as another indicator and suggested possible seven or eight factor solutions due to the location of the kink. Furthermore, cumulative explained variance suggested that at least 8 factors should be considered to exceed at least 50% of explained variance (this solution explains 50.852% of variance). Thus, initial factor analysis models were restricted at the 8 factor threshold. For interpretation purposes, a pattern matrix was used because it shows the clear underlying structure of the possible factor and loadings. Moreover, cutoffs for item removal were considered at 0.30 as suggested by the literature (Hair, 2009 and Pallant, 2011). Items which did not reached the cutoff were removed and the model was subsequently re-estimated as long until no further items were candidates for exclusion. Items with cross-loadings into other factors were not removed at this stage because there was some degree of overlap between the concepts. The reason for this is that otherwise too many cross-loading items would have been removed during the extraction and exclusion process. During the process, the number of factors was reduced to seven as one of the factors was reduced to two items only. Also, the seven factor solution increased in explained variance, suggesting the exclusion of the eighth factor. The removed items were: Tradition1, Tradition2, Tradition4, Tradition6, Security5, Violation1, Violation5. The final model has seven factors14 and explains 53.086% of variance. The factor loadings are summarized in the following table 3:

13_{Please see Appendix D.}

14_{The seven factors are not (re)named according to their describing concept. For simplicity each factors are}

(21)

18

Table 3: Exploratory Factor Analysis with oblique rotation

For the survey analysis an exploratory factor analysis is conducted using the principal axis factoring method which results in the depicted seven factors with the respective items. Standardized loadings from direct oblimin rotation are reported. Cross-loadings are not removed because otherwise too many items be removed. Bolded values indicate loadings above 0.30.

Item Factor Loading

(22)

19

Potential issues remain with items Security4, SelfDirection1, Violation10, Violation12, and Conformity3 because they exhibit cross-loadings. Future research using these instruments should be aware of this. Nevertheless, the extracted structure based on the factors is quite similar as compared to the original survey. Factor 1 contains the items: “conformity“, “security“, “self-direction“, and “tradition“. It can be interpreted as a representation of stability or tendency to conservative attitudes. Also in Schwartz (2012), these values have an opposed or harmonious relation to each other. The second factor is comprised of items from the “stimulation“, “self-direction“, and “security dimensions“.15 Overall, it can be interpreted as a representation of more self-oriented attitudes. The following factors show a better construct than the previous two. Factor 3 contains only items of the “violation“ scale. Factor 4 contains the “collectivism“ scale, whereas factor 5 represents “uncertainty avoidance“. Factor 6 is a subset of “violation“ items. Factor 7 is interesting because its content is related to order-seeking and authority obedience. When taking a closer look to factors 1, 2, and 7 one can find a variation in the underlying items which does not explain a clear construct. Here, one can assume that many (moral) concepts are compromised into one factor. The merging of multiple dimensions or items into a single factor suggests the possibility of a second-order factor (Bentler and Weeks, 1980). With respect to factor 1 one can assume that “tradition“,“conformity“ and “security“ items might belong to one factor, whereas “self-direction“ items could produce a second-order factor. For further studies using a similar survey, factor 1, 2, and 7 should be revised.

Next, reliability was computed on the basis of Cronbach’s Alpha (Cronbach, 1951), using the items obtained from the factors identified in the EFA. Cronbach’s Alpha is an indication of internal consistency (Hair, 2009 and Cramer and Howitt, 2004). These are reported in the following table 4:

Table 4: Cronbach’s Alphas

The factors from the underlying EFA of this study are analyzed with regard to reliability. Cronbach’s Alphas show an acceptable level of reliability. A threshold of 0.7 is considered to be a good indicator for reliability. Factors 1, 2, 4, and 5 are above the threshold value of 0.7 and therefore indicate reliability. Factor 1 2 3 4 5 6 7 Cronbach’s Alpha 0.806 0.839 0.678 0.793 0.770 0.675 0.593

15_{Usually, negative signs imply a reverse interpretation of the items. However, the present study uses the}

(23)

20

The results indicate an acceptable level of reliability. Factors 1, 2, 4, and 5 are above the threshold of 0.7 (Bland and Altman, 1997). Factor 3 is slightly under this threshold. However, by removing the item Violation216 the Cronbach’s Alpha for this factor can be further improved to 0.765. For Factor 6 there are no possible improvements when one wants to remove items, but the value of 0.675 is close enough to the threshold of 0.7. It can be concluded that it should not create any greater issues for the analysis. Finally, Factor 7 is the most problematic, suggesting that this factor may be a statistical error, although it has the common construct of orderliness, security and obedience.

Sensitivity is considered as the ability for an instrument to distinguish between two individuals, and is achieved by a reasonable approximation to normality (Ferketich, 1991). Based on the literature, a variable is considered to have a sufficiently normal distribution, and thus sensitivity, when its absolute skewness is under 3 and its absolute kurtosis under 10 (Kline, 2005). Based on this criterion, all variables except Violation1 and Violation5 were considered sufficiently normal distributed (please see Appendix E). These variables indicated an exclusion in the EFA, thus confirming that future studies must be aware of these possible issues.

In the following, the survey is analyzed regarding the descriptive statistics (see table 5) of the subscale averages. Survey subscale averages are defined as one representative variable for each value. For instance, the value “tradition” is composed of six items, which are aggregated to calculate one average value for the specific subject. This process is done with each observation and each value. This procedure seemed the most practical approach because these subscale values are used for the dice-game analyses and subsequent regression analyses. Table 5 illustrates the subscale averages of each survey value. In general, the outcome of means in table 5 shows that subjects generally give above average importance to each survey value. Especially values such as “conformity”, “self-direction”, and “norm violation” exhibit above average outcomes. It can be found, that subjects tend to give more importance to values such as “tradition”, “conformity” and “security” (subscale averages vary from 3.67 to 4.24), thus indicating a favor for norm compliance. Also, with regard to “norm violation” it can be said that subjects slightly disapprove of norm violation because there is a mean outcome of 4.035. Moreover, “stimulation” and “self-direction” exhibit subscale averages higher than 3.80. Thus, subjects in the survey pool also tend to be independent, risk seeking and adventurous, therefore indicating a tendency to be norm breaking with respect to

16

(24)

21

aforementioned theory. Interestingly, “conformity” and “self-direction” have mean outcomes greater than 4.00 which expresses that norm compliance and violation might cancel each other out depending on the context these values occur. Moreover, Schwartz (2012) states that values are related to each other (for e.g. tradition and conformity) and have the same underlying motivations to an individual. In addition to that, Schwartz (2012) argues that values stand in a dynamic relationship to each other and therefore create tensions or harmony when pursuing certain values. With respect to the present values “conformity” and ”self-direction”, it can be argued that they have a dynamic relationship among each other because they oppose themselves in the context of Schwartz’s (2012) proposed structure. This means that “conformity” represents the overall concept “conservation”, whereas “self-direction” stands for the “openness to change”. Now, depending on the context these values might occur and the motivation an individual attributes to these opposing values, further implications can be derived. Identical results occur for the values “tradition” and “stimulation” (tradition mean = 3.670 and stimulation mean =3.837), supporting the fact that these values oppose each other and are context dependent. “Collectivism” and “uncertainty avoidance” do not exhibit any exceptional results. The subscale averages are close to 3 and therefore any interpretations of these results should be undertaken very carefully.

Table 5: Descriptive statistics of survey data

Table 5 shows the descriptive statistics of all survey values with N=153 observations. All Subscale averages are calculated and analyzed. Values such as “conformity”, “security”, “stimulation”, and “self-direction” exhibit high mean values. Furthermore, this table depicts the absolute values for skewness and kurtosis of each subscale. The standard errors are presented in parentheses. Thresholds of 3 for skewness and 10 for kurtosis are considered to indicate normality in the data.

(25)

22

Subsequently, sensitivity was also analyzed for the averages using the original intended allocation of each item and method described above. As it can be seen, all dimensions fall within the proposed threshold of absolute skewness under 3, and absolute kurtosis under 10 (Kline, 2005), with the exception of “conformity” which exhibits a slight kurtosis because the value is above 10. From that it can be concluded that potential deviation from normality must be considered in this dimension.

The correlation analysis in table 6 with a subject pool size of N=153 shows that there is a moderate relation between security and conformity. The correlation coefficient is

; ! = 0.552** and is significant at p < 0.01. With regard to the subjects, it can

(26)

Table 6: Pearson correlation analysis between survey values with N=153 observations

Correlation analysis of the survey values with 153 observations. Each survey value is measured on a five-point scale, where a higher number indicates a higher

importance/stronger agreement (or stronger disapproval in case of the last value) for the respective value type. Survey items are aggregated for each value and observation. For "norm violation" there are no statically significant correlations with other values. *. Correlation is significant at the 0.05 level (two-tailed). **. Correlation is

significant at the 0.01 level (two-tailed).

TRADITION CONFORMITY SECURITY STIMULATION SELFDIRECTION COLLECTIVISM

(27)

24 4.3 Findings from dice-game experiment

The experimental design allows the current study to derive the true reported numbers, due to its implementation in Excel that enables using a Visual Basic for Application (VBA) code. This step was required as the present study does not have a very large sample size to apply an equal distribution of die numbers. In fact, evaluating the possibility of each computational die number of this experiment shows only a possibility 16.67% for die numbers “1”, “5”, and “6”. In six out of 51 observations, a reporting error could be observed. However, in these cases, it can be assumed that subjects misinterpreted the term “dice number” and therefore reported the number of “throws” taken, which was lower than the actual number. Also, some persons did not give any information regarding the judgments. Thus, incorrect observations were excluded from the total sample, which results in an actual sample size of 45 observations. Surprisingly, only two subjects reported numbers that increased their payoffs. The remainder of subjects always reported the number resulting from the first die roll. In other words, two subjects showed dishonest behavior that is cheating, whereas the rest of the subjects played truthfully. Table 7 shows the mean and variance of the first (actual) number captured by Excel and the reported number. The average actual number of the die roll of 3.378 (and a variance of 3.286) is very similar to the average reported number of 3.333 (and a variance of 3.272). A two-sided t-test on the zero difference in those means does not reject the null hypothesis (t = 0.340; p = 0.736). Hence, it can be concluded that there is no difference in these observations, which statistically supports the fact that subjects exhibit honest behavior.

Table 7: Mean and variance of the first actual number and reported number in the dice-game experiment and t-test (two-tailed) outputs

Output of a t-test with 45 observations and alpha = 5%. Testing the difference between the means of both samples. The first sample represents the actual numbers from the first die roll (Mean = 3.378). The second sample represents the reported numbers (mean = 3.333). With t-statistics = 0.340 (p-value of 0.736) the null hypothesis cannot be rejected, implying that there is no significant difference between actual number and reported number. It can be concluded that subjects are honest in the present observation pool.

First number Reported number

(28)

25

The results concerning the fairness judgments support norm compliance as well. Table 8 represents the outcome of fairness judgments. Subjects perceive non-cheating participants also as fair and therefore indicate 4.600 points on average on the five-point scale. In a similar fashion, subjects perceive cheating participants as unfair and therefore give lower fairness judgments, which is 1.444 on average. The higher the number the fairer (or socially appropriate) the observed behavior is perceived by subjects. Nevertheless, there is only one subject, who plays the dice-game untruthfully and indicates fair judgments on both behaviors. However, one single observation is not sufficient to derive any conclusion regarding norm violations.

Table 8: Descriptive statistics of judgments

Descriptive statistics of both judgments with total 45 observations. Subjects are asked to indicate the social appropriateness on a scale from 1 to 5. The higher the number the fairer (or socially appropriate) the observed behavior is perceived by subjects. Table 8 shows that subjects give non-cheating behavior 4.600 points on average, indicating strong social appropriateness. In contrast to this, cheating behavior is perceived as inappropriate with 1.444 points on average. It can be concluded, that subjects judge non-cheating behavior clearly as fair and cheating behavior as unfair.

Judgment on non-cheating participant Judgment on cheating participant Mean 4.600 1.444 Median 5.000 1.000 Standard deviation 0.809 1.139 Variance 0.655 1.298 Number of observations 45

(29)

26 4.4 Prior note on limitations

The individual codes from the survey and experiment are matched and therefore it is possible to assign survey information to the experimental results. A drawback that has to be mentioned is that there is a very small overlap between survey data and experimental data. In total, this study can only provide 21 overlapping observations derived from the online survey and experimental data. With a sample size of N=21 it is clear that a normal distribution cannot be assumed and parametric applications might not provide unbiased results.

4.5 Analysis of judgments

Although the previous results cannot give any illustrations on the dishonesty hypotheses, this subsection tries to find some explanation for the fair judgments, which could be observed in the experimental result. The correlation matrix depicted in table 9 shows the correlation coefficients between the first judgment, second judgment, and all moral values. The results in table 9 indicate that there are no relevant significant relations between values:

Table 9: Pearson correlation analysis between judgments and survey with N=21 observations

The scrutinized variables are the two (un)fairness judgments put into relation with the eight survey values. Tradition, conformity, security, collectivism, and uncertainty avoidance exhibit norm compliance features, whereas stimulation and self-direction elicit norm violence. The correlation analysis builds the fundamental approach for the subsequent regression analysis. *. Correlation is significant at the 0.05 level (two-tailed). **. Correlation is significant at the 0.01 level (two-tailed). Judgment on non-cheating behavior Judgment on cheating behavior Judgment on non-cheating behavior 1 -0.135 Judgment on cheating behavior -0.135 1 Tradition -0.201 0.016 Conformity -0.281 -0.068 Security -0.238 -0.406 Stimulation -0.067 -0.317 Self-Direction 0.171 -0.04 Collectivism -0.266 -0.144 Uncertainty Avoidance -0.215 -0.101 Norm Violation -0.107 0.141

(30)

27

relation between the variables. From the low coefficients it can be concluded that there should not be any multicollinearity issues (Brooks, 2008).

To further analyze possible issues or relations within the variables multicollinearity diagnostics are conducted. To identify potential multicollinearity, all variables were regressed on each other iteratively to compute the Variance Inflation Factor (VIF) scores.17 Similar to the previous correlation analysis the subscale average values of the matched pool are taken into consideration.18 The maximum threshold for this score is 10 (Marquiardt, 1970). In this case, all VIF scores are significantly within an acceptable range and thus it is possible to say that no multicollinearity issues are present. The following table 10 summarizes the VIF scores:

Table 10: Multicollinearity diagnostics

All independent variables are regressed on each other and the respective VIF (Variance Inflation Factor) are presented in the table. All of them are below 10, thus multicollinearity is not an issue in this data set. The independent variables are the subscale averages of the matched data with N=21.

Dependent Variable Independent

Variable Tradition Conformity Security

Stimulat-ion Self-Direction Collectiv-ism Uncertainty Avoidance Norm-Violation Tradition - 2.602 2.554 2.651 2.551 1.615 2.741 2.412 Conformity 4.466 - 4.648 4.446 3.647 4.301 4.112 3.658 Security 2.092 2.218 - 2.203 2.246 2.250 1.643 2.237 Stimulation 2.624 2.564 2.662 - 2.361 2.174 2.073 2.114 Self-Direction 2.669 2.223 2.870 2.496 - 2.497 2.762 2.735 Collectivism 2.137 3.316 3.634 2.906 3.156 - 3.233 3.583 Uncertainty Avoidance 3.890 3.401 2.847 2.972 3.747 3.469 - 3.824 Norm Violation 2.754 2.434 3.118 2.439 2.984 3.092 3.076 -

As a result to the descriptive statistics, experimental result and the respective correlation analysis and nonexistence of variance in the data, the present study is only analyzing values that are related to norm compliance. Since values such as “conformity” and “uncertainty avoidance” do not exhibit any large impact in the descriptive analysis, they are still considered in the following analyses. To further derive indications on fairness judgments, two multiple regression analyses are conducted with each judgment as the dependent variable and norm compliant values as explanatory variables. Table 11 illustrates the outputs of the regression analyses. As the correlation analysis already suggests, there are no statistical

17_{The author found that this approach is the most appropriate while using IBM SPSS.} 18

(31)

28

significant relations between the first judgment, and values favoring norm compliance. All coefficients except the constant are not significant. Ultimately, conclusions on values that might give any explanation on subjects’ fairness judgments are statistically redundant for this specific data set.

Investigating the second judgment on confederate participants’ dishonest behavior yields identical results. Interestingly, the value “security” with a coefficient of -1.922 is almost significant. The p-value here is p=0.052 which is marginally above the significant threshold of p<0.05. Still, the null hypothesis cannot be rejected, implying that the security coefficient is different from “0”. The previous correlation analysis exhibits a moderate coefficient for security and judgment. However, one can very cautiously assume that “security” could have some explanatory power, if the sample size had been larger. With respect to this scenario, it could be interpreted, that on average subjects would perceive observed dishonest participants less fairly, while subjects increase the importance given to the moral value by one unit.

Table 11: Multiple regression analyses of the (un)fairness judgments on observed (non)-cheating behavior as the dependent variables

Dependent variables: (un)fairness judgments. For the analyses only norm compliant values are taken as the explanatory variables, because subjects do no exhibit dishonest behavior when participating in the dice-game by themselves. The estimation method is OLS. Coefficients and standard errors are presented. P-values smaller than 0.01 and 0.05 are indicated by **, * respectively. Regression outputs for each explanatory variable do not exhibit any statistically significant coefficients. Moreover, the adjusted R-squared is negative which means that this regression model is not suitable.

Dependent variable: First judgment on non-cheating behavior

Dependent variable: Second judgment on cheating behavior Variable Coefficient Standard error Coefficient Standard error

(32)

29

Further interpretations of these results are difficult as there are not enough data points available. Moreover, the applied regression models are not suitable for this particular data set and therefore drawing any conclusion with regard to values is not feasible. Also, it has to be considered that the regression models exhibit a low R-squared for both calculations. Apart from the low values of R-squared, the values for the adjusted R-squared are negative (adjusted R-squared for the first regression analysis is equal to -0.244; adjusted R-squared for the second regression analysis is equal to -0.003). This means, that the proposed regression model for the judgment analysis fits very poorly and therefore any conclusion that are derived are not representative. The reason why the R-squared are positive and the adjusted R-squared are negative could be a mathematical one. R-squared can only assume positive values as it is the square of Pearson’s R, while the adjusted version penalizes this score based on the number of data points and regressors, thus being able to assume negative values. In this case, the lack of significant predictors lowered the adjusted R-squared enough to become negative (Auer and Rottmann, 2011). For further studies, one should increase the observation points to find stable results.

5. Limitations and improvements

The results presented throughout this paper might be biased in the way that experiment specifications and participant selection can be accounted for subjects’ pure honesty and fair judgments. In the following, the main limitations and possible improvements are presented.

(33)

30

The result section already implies the difficulties this work is exposed to. Basically, one can divide limitations in two different contexts. While the first set of limitations address the subjects’ pure honest behavior, the second set of limitations concerns the small sample size of matched survey and experimental data.

In particular, the first limitation is that the experiment does not elicit dishonest behavior and variance in the fairness judgments. In other words, subjects are consistently honest and find other participants’ observed behavior clearly appropriate for the non-cheating condition and inappropriate for the cheating condition respectively. To support the occurrence of honesty, one can argue that Pascual-Ezama, Fosgaar, Cardenas et al. (2015) find overall honesty in student subjects in their experiment, which also examines (dis)honesty in a self-reporting game set-up. With respect to the current experiment, it can be said that a certain priming effect could have occurred during the recruitment and experiment phase. Students were primed to the effect that they were recruited by their own professor just after a business ethics class and were guided to the laboratory to participate in the experiment. Because of this unnatural recruiting process, students might have felt the need to behave correctly (i.e. not cheating). Moreover, the extensive information input during the experimental procedure might have unconsciously influenced subjects to behave compliant to rules and therefore only honest behavior was elicited (Bargh, Chen, and Burrows, 1996).

The judgment of honest behavior in a dice-game