Hard work versus luck : a study on competitiveness and spiteful behaviour

(1)

HARD WORK VERSUS LUCK: A STUDY ON

COMPETITIVENESS AND SPITEFUL BEHAVIOUR

AMSTERDAM, JANUARY 15

th

₂₀₁₈

Hanaa Mekdad (10000529)

Thesis Msc. Economics (15 ECTS)

Track: Behavioural Economics and Game Theory University of Amsterdam

(2)

Statement of Originality

This document is written by Student Hanaa Mekdad who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in

creating it.

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

(3)

Abstract

Research shows that competition leads to higher levels of spiteful behaviour. There also seems to be a difference in spiteful behaviour towards losers and winners of a competition. An experiment was conducted to test whether higher levels of

competition lead to higher levels of spiteful behaviour towards others. In the first part subjects participated in a lottery or an ability-based competitive task. In the second part subjects were randomly matched and participated in a Joy-of-Destruction game. The destruction levels in a Joy-of-Destruction game were used as a measure of

spiteful behaviour. The results indicate that higher levels of competition (ability-based competitive task) do not lead to higher levels of spiteful behaviour towards others than in lower levels of competition (lottery). In addition, winners do not choose higher levels of destruction towards winners than losers. If losers choose to destroy, they choose higher levels of destruction towards winners than towards losers.

(4)

People are not solely motivated by their own self-interest; people do not adhere to the concept of the “Homo Economicus”. People do not only care about their own well-being, but they also care about the well-being of others. However, research shows that individuals also have spiteful preferences (Zizzo and Oswald (2001; Abbink and Sadrieh (2009). The literature defines spiteful preferences as the desire to hurt other individuals even it does not have a strategic or monetary benefit (or even costs money)(Hermann and Orzen, 2008). Economic models such as the ERC model of Bolton and Ockenfels (2000) demonstrate that individuals have altruistic and selfish preferences. However, these models fail to include spiteful preferences. Hermann and Orzen (2008) show that excluding spiteful preferences results in models that do not represent the real word.

Spiteful behaviour is present in many situations: vandalism, physical violence, bullying. According to several experiments, spiteful behaviour is more prevalent in situations that involve competition (Grosch and Rau (2017) and Jauernig et al. (2016). The relation between competition and antisocial punishment is relevant for many areas, such as: management, public policy and societal conflicts. An example is work place hostility. Imagine a workplace that is competitive and rewards those who perform better with a promotion. The individual who is promoted can be considered the “winner” and the rejected candidates can be considered as the “losers”. If

competition positively affects spiteful behaviour, it could lead to a hostile environment in the workplace (Hedges, 2015).

However, the literature does not explain fully what happens to the range of spiteful behaviour at the different levels of competitiveness. Grosch et al. (2017) and Grosch and Rau (2017) found that different levels of competitiveness do indeed affect spiteful behaviour, but they fail to study how spitefulness differs between the parties’ winner or loser status. Jauernig et al. (2016) study whether spitefulness towards others differs depending on their own and others’ status following the competitive task. They show that winners are more likely to display spiteful behaviour towards losers. They also show that losers are less spiteful towards other losers. Jauernig et al. (2016) do not assess the effect of different levels of competitiveness. The aim of this thesis is to extend the existing literature by combining the two strands of research in conducting tests in an experimental setting. Combining the different findings resulted

(6)

in the following research question: What effect does a higher level of competitiveness have on the range of spiteful behaviour across winner and loser status?

To provide an answer to the research question an experiment is conducted. The experiment consisted of two treatments: one that involved a higher level of competiveness and a second treatment that involved a lower level of competiveness. The higher-level competitive treatment was labelled as the Ability Treatment and the lower-level competitive treatment was labelled as the Lottery Treatment. The names of treatments are given based on the task of that treatment. In the Ability Treatment subjects participated in an ability-based competitive task and in the Lottery Treatment subjects participated in a Lottery. Similar to the experiment of Jauernig et al. (2016) and Grosch Rau (2017), this experiment conducts a Joy-of-Destruction game to measure the level of spiteful behaviour. The Joy-of-Destruction game is a game in which subjects can destroy the earnings of their partner. In this experiment, subjects were randomly matched after the ability-based task or the lottery to participate in a Joy-of-Destruction game. First, this thesis focuses on the question whether

participating in the Ability Treatments results in higher destruction levels than in the Lottery Treatment. Second, this thesis assesses whether spiteful behaviour in a Joy-of-Destruction game is dependent on the competition treatment outcome of the subject with whom one is matched. Specifically, it is tested whether winners engage in higher levels of spite towards losers than towards winners and whether losers engage in higher levels of spite towards winners than towards losers. This is similar to the tests that Jauernig et al. (2016) conducted.

The results of the experiment show that there is no significant difference in destruction levels in the Ability and Lottery Treatment. This indicates that levels of competitiveness do not influence spiteful behaviour, which contradicts the findings of Grosch et al. (2017) and Grosch and Rau (2017). Furthermore, the results do not show that Winners choose higher destruction levels towards Losers than Winners. Losers do not choose higher destruction levels towards Winners than Losers at the aggregate level. At the conditional level (only those who destroy), Losers choose higher

destruction levels towards Winners than Losers. This also contradicts the results of Jauernig et al. (2016).

The following chapters provide an overview of the related literature, the method, the results, the discussion of the results and a conclusion. Chapter two

(7)

provides an in depth analysis of the related literature. Chapter three discusses the experimental design, the procedure, statistical methods that are used and the

hypotheses. This chapter also includes a general summary of the sample. The fourth chapter includes the results. The fifth chapter discusses the findings, the limitations of the experiment and provides some suggestions for future research. The sixth chapter provides a conclusion of the findings of this thesis.

2. Literature Review

2.1. Other-Regarding Preferences

Traditional economics assumes that individuals make decisions based on rationality and only consider their own-wellbeing. There is a vast amount of literature available that demonstrates that individuals do not solely base their decisions on rationality. Individuals actually care about fairness, reciprocity and the wellbeing of others. The following paragraph discusses these other-regarding preferences.

Andreoni and Miller (2002) have conducted an experiment in which they show that individuals exhibit rational altruistic preferences. The results show that approximately 98% of the subjects’ preferences can be presented with a quasi-concave utility function. Andreoni and Miller (2002) also show that there is heterogeneity in preferences across the subjects. Subjects can be divided into three groups: Selfish (47,2%), Leontief (30,4%) or Rawlsian (22,4%). Leontief or Rawlsian preferences indicate that an individual receives utility when the (monetary) difference between them and another individual is minimal. Having utilitarian preferences indicates that an individual receives utility when they maximize the (weighted) sum of monetary payoffs.

Social preference models can be categorized into two approaches: the distributive approach and intentions/reciprocity approach (Falk and Fischbacher, 2001). The most important models include the equity/distributive approach, which are the models of Fehr and Schmidt (1999) and Bolton and Ockenfels (2000). Both Fehr and Schmidt (1999) and Bolton and Ockenfels (2000) argue that fair behaviour is driven solely by the outcome of the monetary distributions. According to Fehr and Schmidt (1999) individuals are driven by inequity aversion. This means that individuals do not like to be worse off than others, but they also do not like to be

(8)

better off than others. They postulate that individuals receive negative utility from having disadvantageous inequality and advantageous inequality.

Bolton and Ockenfels (2000) also incorporate distributive preferences in their model. They show that motives drive the behaviour that individuals show in social decision-making. These motives differ across individuals and similar to Fehr and Schmidt (1999) they include heterogeneous preferences. Ultimately, behaviour depends on the weight that individuals place on their “motivations” or preferences. Bolton and Ockenfels (2000) postulate that individuals receive more utility if their monetary payoff is close to the average monetary payoff of their reference group. This means that when there are individuals that have very little or individuals that have a lot in a group, the ERC model predicts that individuals will receive the highest utility if their monetary payoff is close to the average payoff. This is different from the model of Fehr and Schmidt (1999) because the model of Fehr and Schmidt (1999) predicts that individuals maximize utility if inequality is minimal between each individual.

In addition to distributive preferences, there are models that incorporate intentions and reciprocity. The following models that have included reciprocity are the models of Rabin (1993), Charness and Rabin (2002), Dufwenberg and

Kirchsteiger (2004) and Falk and Fischbacher (2006). Rabin (1993) has introduced the notion of intentions in a model. The main idea of the models is that individuals hurt individuals who hurt them and reward individuals who are good to them. According to Rabin there are three components that constitute fair decisions: (i) the strategy of player i (ii) player i’s beliefs about player j’s available strategies (iii) the beliefs player i has about player j’s beliefs about the strategies available to player i. The model eventually leads to fairness equilibrium that can be either positive or negative. A positive equilibrium is a situation in which players mutually maximize each other’s payoffs and a negative equilibrium is a situation in which both players mutually minimize each other’s payoffs.

The model of Charness and Rabin (2002) combines both distributional and reciprocal preferences. Charness and Rabin (2002) show that individuals’ behaviour is not driven by difference aversion, but rather by a preference to maximize total social welfare. The formal model of Charness and Rabin (2002) also incorporate factors

(9)

such as preferences for competitiveness, difference aversion and social welfare. The inclusion of these factors allows for heterogeneity across individuals.

Dufwenberg and Kirchsteiger (2004) developed a model that can be applied in extensive games and focuses solely on reciprocal preferences. Dufwenberg and Kirchsteiger (2004) apply the formal model of Rabin (1993) and generalize it to N-person extensive games and introduce the concept of “Sequential Reciprocity Equilibrium”. The model incorporates Psychological Game Theory. Psychological Game Theory is different from standard game theory. It does not only take into account the strategy profile in a game, but also the beliefs an individual has about the other players’ beliefs/strategic choices. The main difference from the model of Rabin (1993) and this model is the update of beliefs. The main assumption in the model of Dufwenberg and Kirchsteiger (2004) is that in each sub-game of their extensive game, individuals update their beliefs about the other individual at each sub-game and is based on the previous behaviour of the other player in the previous sub-game. So, the strategic choice of a player at a sub-game is motivated by the beliefs he has about the other player at that sub-game.

Falk and Fischbacher (2006) also introduced a model that includes reciprocal preferences. The model incorporates two factors that form the central idea (“The evaluation of kindness of an action”): the consequences of an action and what the underlying intentions are of the player. However, unlike Rabin (1993) and

Dufwenberg and Kirchsteiger (2004), the model of Falk and Fischbacher (2006) is not only driven by intentions, it also takes into account relative outcome comparisons. The model of Falk and Fischbacher (2004) incorporates the notion of kindness. Individuals reciprocate behaviour based on whether they perceive this behaviour as kind or unkind.

2.2 Spiteful Preferences

In addition to altruistic or selfish preferences, people also exhibit antisocial

preferences. Yamagishi et al. (2017) indicate that individuals who engage in spiteful behaviour have a higher activation in the part of the brain called the caudate nucleus. The following section provides an overview of a theoretical model and additional experiments that demonstrate that people have spiteful preferences.

(10)

2.2.1 Theoretical Model

Social preference models do not always include antisocial preferences in their analysis. Levine (1998) introduced a social preference model that provides a theoretical frame that incorporates altruistic, selfish and spiteful preferences.

Individuals do not only care about their own monetary payoff, but they also care about the monetary payoff of their opponent. The model of Levine (1998) does not only incorporate altruistic or selfish preferences, the model also provides the opportunity to include spiteful preferences. The main idea behind the model is that individuals receive utility based on their preferences and what they believe the preferences are of their opponents. First, the utility of an individual depends on his or her utility. Second, the utility of an individual depends on their adjusted utility. The adjusted utility incorporates the player’s own utility, their social preference

(spiteful/altruistic/selfish) and what they believe the preferences are of their opponent. The model explains the behaviour in Ultimatum Games, Market Games and Public Goods. The model cannot explain why individuals give positive amounts in a Dictator Game.

2.2.2 Spiteful Behaviour in Experiments

Experiments show that individuals engage in spiteful behaviour. Spiteful behaviour is present regardless whether individuals receive (monetary) benefit from spiteful behaviour and is even present if it entails a cost.

Zizzo and Oswald (2001) demonstrate that individuals are willing to lose money in order to destroy income of others. They argue that people do not only exhibit positive interdependent preferences but they also exhibit negative

interdependent preferences. In the experiment of Zizzo and Oswald (2001) subjects first participated in a group of four subjects in a betting game in which they could win or lose money. In the experiment, half of the players in the group received a

discriminatory advantage in terms of money (500 extra of a currency called Doblon). All subjects knew the scores of all group members. After the betting stage the

participants could burn money of the other three group-members. Zizzo and Oswald (2001) found that 62.5% is willing to burn money of others even if this entails that it will cost them. They provide two explanations for their findings. First, the results might be driven by rank egalitarianism (punishing those who earned the most (highest

(11)

rate compared to those who did not have the advantage in the betting stage. The second reason is reciprocity. Disadvantaged individuals might think that the richest subjects undeservingly earned the money and therefore retaliate against those who have a higher income. It is difficult to find to which extent reciprocity and rank egalitarianism influence money burning. Another problem that arises is the fear of being burnt by other group members. The expectation of money burning by a partner positively affects the money burning decision of the subject.

Zizzo (2003) replicated the study of 2001 with a different set-up in the money burning stage. Everyone participated in the money burning stage, however only one choice would randomly be implemented. This set-up avoids that money-burning decisions were based on the pre-emptive fear of retaliation that was present in the experiment of 2001. In this experiment money burning occurred less and money burning was more sensitive to price compared to the experiment in 2001. Subjects are less willing to burn money when the cost of burning money is high. Individuals also seem rank egalitarian; they are more likely to burn money of the richest group members.

Abbink and Sadrieh (2009) developed a method called the Joy-of-Destruction game to show that individuals have antisocial preferences. In an experiment they show that individuals are willing to destroy the income of another individual even if there are no strategic benefits. In this experiment subjects were anonymously and randomly matched. Subjects had to participate in a real effort task and for each completed task a subject received 80c or 120c depending on the treatment. The experiment consisted of two treatments: a hidden treatment and an open treatment. In the open treatment all subjects had the opportunity to destroy the income of their partner (choice between 0 and 80c and multiples of 10c). In the open treatment the destruction level of the subject is communicated to their partner. The destruction of money did not result in any monetary gain for the subject. In the hidden treatment there was also a chance that Nature (based on chance) would destroy a part of their partners’ income (each value between 0 and 160 and each value equally likely). In this treatment subjects could additionally chose how much they were willing to destroy. In the hidden treatment the partner would only be informed about the total destruction of their income and not the source of destruction. Individuals chose a lower destruction rate in the open treatment (8.5%) compared to the hidden treatment

(12)

(39.4%). Abbink and Sadrieh (2009) report that transparency in the source of destruction influences the destruction levels that individuals choose. If destruction levels are non-transparent to an individual, a destroyer could increase their

comparative advantage by decreasing the level of income of their partner. The results indicate that destruction is present in both treatments and Abbink and Sadrieh (2009) claim that this is due to pre-emptive retaliation. This means that individuals are driven by their expectation that their partner will destroy. As a response to this expectation they choose to destroy the earnings of their partner as a pre-emptive measure. Abbink and Sadrieh (2009) claim that in that case individuals do not make their decision based on the “pleasure of being nasty”, but based on the expectation that other individuals receive a “pleasure from being nasty”.

Kessler et al. (2012) expand on the results of Abbink and Sadrieh (2009) and introduce the Destructor Game. In the Destructor Game there is only one person in each pair that can choose to destroy. Kessler et al. (2012) conducted an experiment that shows that in the absence of bilateral destruction, individuals still choose to destroy. The authors claim that this stems from “an intrinsic pleasure to hurt others”. During the experiment subjects are randomly matched and each pair consists of a randomly assigned destructor and a passive player. The names of the roles are not communicated to the subjects. The destructor was called player A and the passive partner was called player B. The destructor has the opportunity to destroy the earnings of the passive partner and the passive player cannot retaliate. In addition, there is also the possibility that a random event destroyed the earnings of the passive subjects, but subjects were only informed about their final earnings. The maximum amount that could be destroyed (both random event and destructor) was 40%. Every subject knew the rules of the game. The results show that 15.5% of the subjects choose a

destruction rate larger than zero. Kessler et al. (2012) removed any motives that might influence spiteful behaviour such as: pre-emptive retaliation, inequity aversion,

monetary benefits or pre-emptive retaliation. Kessler et al. (2012) argue that the 15.5% of subjects receive an “intrinsic pleasure to hurt others”.

Abbink and Herrmann (2011) report that in an adjusted joy-of-destruction game (called mini-JOD) subjects are willing to destroy the income of another

individual even if it involves losing money. The experiment had the similar set-up as the Joy-of-Destruction game described in the previous paragraphs. The subjects were

(13)

endowed with 10 money units (or MU). Again, all subjects can destroy income of their partner without any monetary gain. Participants were randomly divided in the open or the hidden treatment. In the open treatment the subject could choose between destroying 5MU of their partners income at a cost of 1MU and destroying nothing. In the hidden treatment the subject could again choose between destroying 5MU of their partners’ income and not destroying at all. In the hidden treatment there was a chance of 1/3 that Nature would destroy 5 MU. Thus, choosing to destroy 5MU of their partners’ income would be ineffective, because 5MU is already destroyed by Nature and it would also cost them 1MU. The findings indicate that there is a significant difference in the percentage of choosing to destroy between the open treatment (10.8%) and the hidden treatment (25.8%). There is a significant correlation between those who destroyed money and those who expected their partner to destroy their income (pre-emptive fear of retaliation). Abbink and Herrmann (2011) report that including the chance that Nature destroys income of their partner decreases moral cost of destruction, because subjects think that there is a chance that Nature will destroy it anyway. In addition to random destruction in the hidden treatment, the source of destruction remains non-transparent. Individuals can destroy their partners’ income without the partner knowing who it was. According to Abbink and Herrmann (2011) these two factors (non-transparency and chance) lower the moral cost of income destruction.

2.3 Competition and Spite

Spitefulness is more frequent in situations that involve competitiveness. Situations such as status, resource scarcity, envy and entitlement are possible predictors for spiteful behaviour in an experiment. The next section provides an overview of experimental results that involve competition and spiteful behaviour.

Behaviour of individuals is not always consistent, individuals who behave pro-social in some situations might engage in antipro-social behaviour in other situations (Sadrieh and Schröder, 2017). Prediger et al. (2013) show that resource scarcity is a reason for spiteful behaviour. In an experiment conducted in Namibia they show that in fields, where resources are scarce, subjects (farmers) are twice as likely to destroy another subjects income. Resource scarcity leads to higher levels of competition, which in turn affects spiteful behaviour. Hermann and Orzen (2008) and Bolle et al. (2014) show that in games that involve rent-seeking or vendettas result in higher

(14)

levels of spite. In games in which subjects fight for a prize, they are more likely to sabotage each other. Even if sabotage ultimately entails a cost. In the experiment of Hermann and Orzen (2008) individuals are more likely to overinvest when another subject is present, resulting in higher costs. Bolle et al. (2014) show that in a

(sequential) Vendetta Game that individuals engage in retaliatory behaviour towards another individual in order to sabotage the individual from winning, even if this leads to a cost. Bolle et al. (2016) argue that emotions and inequality in the probability of winning have a small but significant effect on this retaliatory behaviour. However, the results show that even if the experiment controlled for these effects (emotions and inequality), subjects still choose to sabotage the other individual. This leads to high efficiency loss; in which the subjects spends money to sabotage the other individual but simultaneously lowers their own probability to win.

Status and competitive rivalry have a positive effect on spiteful behaviour. Fehr et al. (2008) show in an experiment conducted in India that high-caste subjects (higher status individuals) are more likely to engage in spiteful behaviour towards other subjects, especially when inequality is disadvantageous. The results of Fehr et al. (2008) are in line with theories about status. Individuals are willing to sabotage the status of other individuals even if this behaviour entails a cost (Charness et al., 2013). This shows that individuals are willing to improve their status even if it does not have a monetary benefit. Second, Goette et al. (2011) show in an experiment in the Swiss army that in a competitive setting, individuals are more likely to engage in antisocial behaviour towards out-group subjects than in-group subjects, regardless whether these subjects were defecting or cooperating. In addition, the results show that competition improves cooperation with in-group members. The findings of Fehr et al. (2008), Charness et al. (2010) and Goette et al. (2011) show that a competitive environment contributes to higher levels of spite towards others.

Winners behave more spiteful towards “losers” than winners, according to the experiment of Muller et al. (2012) and Jauernig et al. (2016). In addition, Jauernig et al. (2016) show that losers are less spiteful towards losers than winners. The paper of Muller et al. (2012) shows that competition affects spiteful behaviour in three

experiments. In all three experiments, the subjects participated in a real-effort task. All the subjects were matched with an imaginary partner. In the second part the subject had the opportunity to engage in spiteful behaviour (study 1: send a noise

(15)

blast, study 2: same as study 1, study 3: Hot Sauce task). In the first study subjects could send a noise blast to their partner that ranged from 60dB to 105dB (level of a smoke alarm). The second experiment was similar to the first experiment; there was only a slight difference in the instructions. The third experiment consisted of a Hot Sauce task in which the subjects could administer salt and Tabasco to the drink of their partner. The results in all three experiments show that winners of the first task were more likely to aggress towards losers than winners. The findings are in line with the Power Theory of Aggression. Muller et al. (2012) suggest that winning leads to higher levels of entitlement and empowerment, which in turn leads to higher levels of aggression. The experiment of Jauernig et al. (2016) expands on the findings of Muller et al. (2012) and tries to find a link between spiteful behaviour towards out-group members and in-out-group members. They test whether spiteful behaviour in a competitive setting stems from the Power Theory of Aggression (Muller et al., 2012) or the Frustration Aggression Hypothesis. The Frustration Aggression Hypothesis assumes that losers are more likely to engage in aggressive behaviour, because losers feel inferior from losing a competition. Losers use aggression as a venting mechanism to compensate for their loss. The experiment consists of two steps: a competitive task and the Joy-of-Destruction game. Subjects first competed in a competitive task that consisted of a speed-based calculation task. Each subject was paired with another subject in the first task. After finishing this task there was a winner and a loser in each pair. The scores were communicated to their partner for social comparison. After completing the task the subjects were re-matched with their old partner or re-matched with a new partner. The matching and re-matching of partners resulted in three treatments (In-group Stranger Treatment, group Stranger Treatment and Out-group Partner Treatment and two conditions (winners or losers)). After the re-match each subject received 100 ECU (10ECU=€0.70) and participated in a

Joy-of-Destruction game. After completing the Joy-of-Joy-of-Destruction game (but before receiving the results) the subject completed a short questionnaire about their happiness. They find that there is no difference in aggression levels towards a new partner or their old partner. The effect of winning or losing does not lead to a personal feud between old partners. Second, they find that losers are more severely punished by winners who had a large score gap between them and their partner in the first task. However, losers that lost with a large score gap did not show any difference between

(16)

the chosen destruction levels towards winners or losers. Similar to the findings of Muller et al. (2012) losers are more severely punished by those who win. Winners are more likely to aggress towards those who scored worse in the competitive task. The third result they find is that losers are punished less by others losers. Winners are punished the same by both winners and losers. This stems from the possibility that losers feel empathy towards other losers. The fourth finding shows that a loser with a very low score was more likely to choose high punishment levels towards another loser. They did not find the same significant outcome for winners. This can be influenced by the humiliation that losers with a low score feel. Humiliation can trigger higher destruction levels towards a subject who had a lower score gap.

2.4 Competitive Context and In-group/Out-group Competition

The previous paragraphs show that competition can trigger spiteful behaviour towards others. However, these studies do not include what happens to spiteful behaviour when the level of competiveness varies. The following sub-section provides an overview of the articles that provide a motivation for conducting this thesis.

Van Anders and Watson (2007) focus on the difference in testosterone levels between the participation in a lottery versus a competitive task. They find that men who lost the competitive task do appear to have a significantly larger decrease in testosterone levels compared to men who won the competition. They do not find the same result in the lottery task. The results for women are inconclusive; they do not find a significant decrease/increase in both the lottery and the ability treatment. This indicates that testosterone levels of men are affected by external recognition for their effort. There is evidence that higher levels of testosterone are linked with higher levels of antisocial behaviour (Carré et al., 2013). There is a lack in economic literature that focuses on competitive context and the levels of spite. However, the article of Van Anders and Watson (2007) shows that there is a reason to believe that competitive context does influence spiteful behaviour. This provides a motivation to conduct an experiment in an economic setting that focuses on the link between lottery and a competitive task and the effect on spiteful behaviour.

Grosch et al. (2017) conducted a lab-in-the-field experiment at a Ghanaian farm that focuses on the levels of competitiveness and pro-sociality in a Public Goods Game (hereafter, PGG). The experiment consisted of three treatments. The

(17)

competitive treatment consisted of a real effort task (assembling ballpoints) and winners were chosen based on relative performance. The random treatment consisted of the same real-effort task, but winners and losers were randomly chosen. Subjects in the threshold treatment participated in the similar task as the other treatments, but winners were chosen if they assembled 40 ballpoints. Grosch et al. (2017) also included another treatment: dispersion. They looked at whether a bigger difference in endowments between losers and winners does lead to lower levels of pro-sociality than in the situation when the difference in endowments between losers and winners is small. In the high dispersion treatment, winners would receive 15 Ghanaian Cedi (GHS) and losers would receive 5 GHS. In the low dispersion, winners would receive 12 GHS and losers would receive 8 GHS. First, the findings indicate that participating in a competitive real effort task leads to significant lower cooperation in a subsequent PGG than participating in a task that assigned winners and losers randomly. This result shows that higher levels of competitiveness lead to lower levels of pro-social behaviour. The authors claim that this is driven by inequality aversion, because this result is only significant at the high dispersion treatment. When the dispersion between earnings is low, there is no difference in pro-social behaviour between the competitive treatment and the lottery treatment. This indicates that pro-social behaviour only decreases in the competitive task when the stakes are high. Second, the decrease in pro-sociality is stronger for subjects who usually work together and now have to compete against each other. This indicates that the decrease in pro-sociality might be driven by the change from in-group to out-group competition. The results also show that subjects who were not aware of the current bonus-system at the firm were also less pro-social. This indicates that subjects are perhaps driven by their frustrations towards the firm or other co-workers, because they did not know about the current bonus-system.

Grosch and Rau (2017) also conducted an experiment in the lab in which they show that individuals do engage in spiteful behaviour towards others. The experiment also includes discrimination, competition and randomness as treatments. They show that when subjects are in the discriminatory treatment destroy more than in a

competitive environment. They show that discrimination acts like a “frustrator” and leads to higher levels of destruction in a Joy-of-Destruction Game. They do not find any difference between the destruction rates between losers and winners in the

(18)

competition and random treatment. They do show that in the competitive treatment and the random treatment, individuals with high aversion to inequality destroy more than subjects with lower aversion to disadvantageous inequality. They show that in the discrimination treatment, subjects have higher destruction levels. The authors claim that this might be driven by the frustration that they feel, because they were treated unequally. Second, they also find that destruction levels are higher when people are paid not according to their effort. These subjects behave more spiteful than those who are paid according to their ‘believed effort”. This is true for all treatments, but especially in the Discrimination treatment. The experiment shows that spiteful behaviour is present in all treatments, however destruction levels are the highest in the discrimination treatment. However, similar to the experiment of Grosch et al. (2017) they fail to include in-group and out-group competition. Jauernig et al. (2016) show that spiteful behaviour towards losers is lower than towards winners.

This thesis tries to find a link between the level of competitiveness and the effect on the range of spiteful behaviour. There is evidence that competitiveness increases levels of spite (see section 2.3). However, there are not many experiments that incorporate varying levels of competition and the effect on spiteful behaviour. Grosch et al. (2017) and Grosch and Rau (2017) show the level of competitiveness affect spiteful behaviour or decrease pro-sociality, but the experiments fail to include in- and out-group behaviour. First, the experiment of Grosch et al. (2017) does show that subjects who usually work together were less pro-social in the experiment. This might be driven by the change from being an in-group member to an out-group member. Second, the experiment of Grosch et al. (2017) focuses only on the level of pro-social behaviour, but including anti-social behaviour results provides a more inclusive view of the motivations behind the decrease in pro-sociality. The article of Grosch and Rau (2017) focuses solely on the difference between discrimination and competitive environment. They only focus at the difference between the

discrimination treatment versus the other competitive and random treatment. They did not include the difference between the competitive and the random treatment. Similar to Grosch et al. (2017) they do not test for differences between in- and out-group members. Jauernig et al. (2016) do find a significant result that destruction levels are higher towards out-group members than in-group members. However, Jauernig et al. (2016) do not vary the levels of competition (lottery vs. a competitive environment).

(19)

This thesis combines, in particular, the findings of Jauernig et al. (2016), Grosch et al. (2017) and Grosch (2017) and provides a better understanding of the dynamics

between levels of competitiveness and their effect on spiteful behaviour.

3. Methodology

The following sections discuss the methodology that is used to test the hypotheses. The sections are structured as follows: the first section discusses the experimental design, the second section focuses on the procedure of the experiment, the third section includes the hypotheses, the fourth section discusses the method that is used to test the hypotheses and the fifth section provides information about the data collection and a summary of the sample characteristics.

3.1 Experimental Design

The experimental design consists of two tasks to determine how a higher level of competition affects the range of spiteful behaviour. In the first part the subjects participate in either a competitive task or in a lottery in which they can earn experimental currency units (hereafter, ECU). The second part consists of participating in a money-burning task.

The experiment is conducted to test how higher levels of competition affect the range of spiteful behaviour towards other individuals. First, the difference in competition level is established by conducting two different tasks: a group that participates in an ability-based competitive task and a group that participates in a chance-determined competitive task. The experiment incorporates a between-group design to avoid experimenter demand effect that might arise in a within-group design

(Charness et al., 2012).The difference in level of competition is adopted based on the experiments of Van Anders and Watson (2007) and Grosch and Rau (2017). These experiments use a speed-based calculation task as a competitive task. The other task consists of a lottery in which winners and losers were chosen based on chance. The competitive task in this experiment also consists of a speed-based task in which subjects have to find two unique numbers in a matrix that add up to ten. The second type of task involves in participating in a lottery. More details about this task are given in sub-section 3.1.1 and sub-section 3.1.2.

The second task in the experiment involves participating in a Joyof

(20)

to show that individuals engage in spiteful behaviour. This is also used in the experiments of Jauernig et al. (2016) and Grosch and Rau (2017). They use the Joy-of-Destruction game to test whether competition has an effect on spiteful behaviour. The chosen destruction levels are used as a way to measure spiteful behaviour towards others. This experiment builds further on the results of Jauernig et al. (2016) and Grosch and Rau (2017) and therefore a Joy-of-destruction game is used as a measure of spiteful behaviour. More details about how the Joy-of-Destruction is conducted in this experiment can be found in sub-section 3.1.2.

3.1.1 Treatments and Sub-Treatments

This research consists of two treatments that are the main focus of this thesis: the Ability Treatment and the Lottery Treatment. In addition, this experiment also includes sub-treatments: In-group and out-group sub-treatments.

In the Ability treatment, subjects participate in a speed-based calculation task that involves addition. The task is similar to the task that is used in an article of Jauernig et al. (2016). In this task subjects have to identify two unique numbers in a matrix that add up to 10 in a matrix and add them together. Solving matrices is an easy task that does not require any preliminary knowledge. The task is also practical during an experiment, due to the low effort it takes to check the answers. Before subjects start with the first task, they have to answer a control question to ensure that they understand the task. If every subjects understands the ask, the subjects receive five minutes to solve as many matrices from a list of 32 matrices. Half of the subjects that have highest number of correct answers are labelled as Winners. The other half of subjects that have the lowest number of correct answers are labelled as Losers. Losers receive a note that informs them that they either earned 100 ECU (“You earned 100 ECU”). Winners receive a note that informs them that they have earned 200 ECU (“You earned 200 ECU”). All subjects receive their information about the outcome anonymously in a folded note. Note that in the experiment, subjects are not told that they are Winners or Losers. Informing subjects that they belong to a winning team or a losing team could lead to out-group and in-group framing. Furthermore, subjects are only informed about their own earnings and not the earnings of other subjects. This eliminates any effects of social comparison. Jauernig et al. (2016) show that a large score gap influences spiteful behaviour. This is less relevant for this thesis,

(21)

because this thesis focuses solely whether different levels of competition affect the range of spiteful behaviour.

In the Lottery treatment, subjects participate in a lottery in which each subject has an equal chance to win. Each subject draws a ticket from a box that contains a number. Subjects who draw the lowest numbers (1 or 2 for example) are labelled as Winners and subjects who draw the highest numbers are labelled as Losers. Similar to the first treatment, half of the participants are Winners and the other half are labelled as Losers. Winners receive a note that informs them that have earned 200 ECU (“You earned 200 ECU”) and Losers receive a note that informs them that they have earned 100 ECU (“You earned 100 ECU”). Similar to the Ability Treatment, all subjects receive their information about the outcome anonymously in a folded note. The payoffs in the Lottery treatment are the same as in the Ability Treatment. This ensures that any difference between the treatments is not due to differences in payoff between the two treatments.

In addition, this thesis focuses on spiteful behaviour towards Winners and Losers based on whether an individual is a Winner or a Loser. Jauernig et al. (2016) show that Winners choose higher levels of destruction towards Losers than towards Winners. The results also show that Losers choose lower levels of destruction towards Losers than towards Winners. Jauernig et al. (2016) call Losers facing Losers and Winners facing Winners the Ingroup Stranger Treatment Winners facing a Loser and a Loser facing a Winner was called the Outgroup Stranger/Partner Treatment.

Stranger refers to when a subject was re-matched with a stranger and Partner refers to when a subject was re-matched with their previous partner in the first task (see section 2.3). In this experiment subjects are only matched in the second part of the

experiment, so there is no re-matching. However, the experiment does include an In-group and Out-In-group treatment, which will be referred to from now on as the sub-treatments. Subjects are not told that they belong to the winning team or losing team to avoid in- and out-group framing. Note that, this thesis does not focus on in- and out-group competition based on the team subjects belong to. Similar to Jauernig et al. (2016), this thesis focuses on whether Losers are more likely to engage in spiteful behaviour due to their frustration of losing and whether Winners are more likely to engage in spiteful behaviour due to higher levels of entitlement that result from winning. The subjects only receive information about the earnings of their opponent

(22)

(“You have earned ... ECU in part 1. You are matched with a participant that has earned ... ECU in part 1”).

3.1.2 Joy-of-Destruction Game

In the second part of the experiment the subjects are randomly and anonymously matched with a partner. In this part the subjects participate in a Joy-of-Destruction game. A Winner in the first stage might face a Loser (out-group) or a Winner (in-group). A Loser might face another Loser (in-group) or a Winner (out-(in-group).

All subjects (Winners and Losers) participate in a Joy-of-Destruction game designed by Abbink and Sadrieh (2009). The Joy-of-Destruction game is used to test the level of spiteful behaviour of the subjects. In this part of the experiment, subjects are randomly matched with another subject. First subjects have to read the

instructions and they also receive a decision sheet. In this decision sheet subjects are informed about their earnings of their opponent. The identity of their opponent is not disclosed, subjects are only informed about the earnings of their opponent. After they have read the instructions, the subjects participate in a Joy-of-Destruction game (hereafter, JOD). Similar to Jauernig et al. (2016) and Grosch and Rau (2017), iin this task subjects are asked how much of their partner’s earnings they are willing to destroy. The instructions do not include the word “destroy” to avoid framing or any influence from experimenter-demand effect. They are shown the following sentence: “I choose to reduce my counterpart’s earnings by”. Similar to the experiment of Abbink and Sadrieh (2009), subjects can choose destruction rates between 0 and 100 (discrete numbers only and multiples of 20)(see Appendix B for the JOD game presented in the experiment). Subjects are also told that choosing a positive destruction rate will have no effect on the subject’s own monetary payoff.

At the end of this part each subject circles their choice on a decision sheet and is collected by the experimenter. In each pair only one decision will be chosen to determine the final outcome. This ensures that participants will not deliberately

choose a too low or too high level of destruction because of pre-emptive retaliation. In Jauernig et al. (2016) winners faced a probability of 40% of their income being

destroyed and losers faced a probability of 60%. They argue that the subjects that lose the competition are more aggressive towards winners than vice versa. However, the empirical results of Jauernig et al. (2016) show that winners are more aggressive

(23)

towards losers than the other way around. Hence, in this experiment both Winners and Losers face a probability of 50% of their destruction being chosen.

3.2 Procedure

The following section provides all the steps of the experiment. This section also includes information about how the data was collected.

Each session of the experiment consisted of a group of an even number of participants (four or eight subject in each session). The experiment consisted of eight sessions: four sessions of eight subjects and four sessions of four subjects. The total number of subjects that participated was 48. Subjects were all friends or family (or friends of the family) of the experimenter, so many subjects knew each other. Subjects were recruited via e-mail and knew that this experiment is for a thesis. The e-mail included a link to a website, in which they could choose a time and location that suited them best. The e-mail did not include any information about the subject or objective of the experiment, but subjects did know that it involved a behavioural economics experiment. The sessions took place in the dining room of the

experimenter or at the University of Amsterdam (D3.00). The experimenter made sure that there was enough space between subjects to ensure privacy (one chair between each subject). More information about the subjects can be found in sub-section 3.5.1.

The two treatments Lottery and Ability are randomly assigned beforehand. Half of the participants were assigned to the Lottery Treatment and the other half were assigned to the Ability Treatment. The In-group and Out-group sub-treatment were also randomly assigned beforehand. The experiment consisted of three parts and took about 30 minutes to 45 minutes. Before the experiment started all subjects were seated with enough space between them. This was important to ensure that nobody could see what the other was writing on his or her answering sheet. After they were seated, they picked a folded ticket out of box to ensure that each seat number was randomly assigned.

Before the experiment started, subjects received a list with general instructions about the experiment. First, the experiment was anonymous. Subjects could not

communicate with each other and they were also told to refrain from any other means of communication (mobile phone). They were also told that they could not write their name on any of their sheets (instructions or decision sheet). They were told that their identity was not important for the experiment. This minimizes the

(24)

experimenter-demand effect (Charness et al., 2012). Note that this experiment was not double-blind, because the experimenter had to provide information to each in private subject about their results of the first task, second task and their final earnings. Second, if they had questions they could raise their hand and the experimenter would come to answer their question in private. The general instructions also included information about the duration of the experiment and the incentives. The experiment included a monetary incentive to provide an extrinsic motivation to participate in this experiment. The maximum amount a subject could win was 200 ECU (=10 euro). The subjects were informed that at the end of the experiment a subject was randomly chosen that would receive their earnings of the experiment in real money.

In the Ability Treatment subjects participated in a competitive speed-based task. First, they received instructions about the rules of the task. The instructions included information about the task, earnings, how the results were communicated and a control questions. In the speed-based task, subjects had to find two unique numbers that add up to 10 in a 4x3 matrix. The goal is to solve as many matrices as possible in five minutes. They were also shown an example of a matrix (see figure 3.2). The instructions also included information about earnings. Subjects were informed that the experiment would rank all the subjects based on their correct answers. Half of the subjects that had the highest number of correct answers in each session would receive 200 EC. The other half of the subjects that had the lowest number of correct answers would receive 100 ECU. Furthermore, they were informed that they would not be informed about their realized ranks and the ranks of other subjects. They would also not receive any information about the earnings of other subjects. The results would remain private and are communicated through a folded note. After the instructions, subjects had to answer a control question to make sure that they understood the task. Figure 3.2 shows the control question. Subjects had to raise their hand if they had finished with the control question. When every subject raised their hand, the experimenter would then hand out the exercises and answer sheet of the speed-based task. The subjects could only start with the task if the experimenter gave them a signal to do so. After five minutes, the experimenter collected the exercises and the answer sheets of the subjects. The experimenter then checked the answers for each subject. Subjects were not allowed to talk or

(25)

answers of each subject from highest to lowest. After the experimenter checked the answers, the subjects were given a note with information about their earnings. Half of the subjects in each session that had the highest number of correct answers received 200 ECU. They received a note with the following message: “You earned 200 ECU”. The other half of subjects that had the lowest number of correct answer received a note with the message: “You earned 100 ECU”.

In the Lottery Treatment subjects participated in a lottery. First they received instructions about the lottery, earnings and how the results would be communicated. The lottery involved drawing a folded ticket from a box. Each ticket contained a number and subjects were told that they had to keep their ticket number private. Subjects were also informed that half of the subjects that draw the lowest number would receive 200 ECU and the other half with the highest numbers would receive 100 ECU. Similar, to the Ability Treatment the subjects were informed that they would receive the results on a folded note. Their realized rank and earnings would remain private. When every subject finished reading the instructions and there were no questions, the experimenter would then walk around with a box of tickets. Each ticket was folded four times to ensure that subjects could not see the number. The experimenter also walked around with a notebook to write down seat numbers and corresponding ticket numbers. This ensured that subjects would no try to deceive the experimenter and lie about their ticket number. After the lottery, the experimenter ranked the ticket number of each subject from lowest to highest. Half of the subjects with the lowest number would receive a note with the message: “You earned 200 ECU”. The other half would receive a note with the message: “You earned 100 ECU”. Similar to the Ability Treatment, the subjects received their note in private. They were not informed about their realized rank or the rank of other subjects. Subjects were also not informed about the earnings of others.

The second task was the same for all subjects in each treatment. First, the experimenter randomly matched each subject with another subject, because in the second part subjects participate in a JOD game. Subjects were matched based on the sub-treatment: In-group or Out-group. The sub-treatments were randomly determined beforehand. After the random matching, the subjects received the instructions and decision sheet for the second task of the experiment. The second task involved participating in a JOD game. Subjects were informed that they were randomly

(26)

matched with another subject. Second, they were informed that they would have to make a decision that involved reducing the earnings of their opponent and that they could reduce their opponent’s earnings by multiples of 20 ECU with a maximum of 100 ECU (0, 20... 100 ECU). They were also informed that their opponent would also be given the simultaneous opportunity to reduce their earnings. The instructions also included information about the implementation of the decisions. They were told that only one decision in each pair would be randomly implemented. Finally they were informed that their identity would not be revealed to their opponent. The only thing that would be communicated is their earnings. The subjects were also given a decision sheet that included the JOD game. Subjects were instructed to keep their decision private. After reading the instructions, subjects could immediately start making a decision.

After each subject was finished with making a decision they had to raise their hand. The experimenter would then come and collect their decision sheet. The subjects are then given a short questionnaire that included questions about their age, gender and educational background. These variables are included as a control variable for the regression model. Some experiments show that factors such as age, gender influence spiteful behaviour Cremer et al. (2007) show that males are more likely to engage in spiteful behaviour than females. Second, Fehr et al. (2013) show that younger people are more likely to engage in spiteful behaviour than older people. Educational background was included to control for any influence that might arise from a certain field of study. For example, Frank et al. (1993) show that economic students are more selfish than non-economics students.

Similar to Jauernig et al. (2016) in each pair only one destruction choice is randomly implemented, to remove any effects of pre-emptive retaliation. In this experiment each choice of both Losers and Winners had an equal chance (50%) of being chosen. While the students were filling out the questionnaire, the experimenter was choosing which random choice was implemented for each pair. After subjects were finished with the questionnaire they were given a note. This note contained information about their final earnings and which choice was implemented. The choice that was not implemented was not communicated to the subjects, because it was not relevant. The note only included information about their own earnings. Subjects were not informed about the earnings of the other subjects.

(27)

At the end, the experimenter randomly assigned a subject who received money in euro, based on the earnings of their experiment. Each subject received a thank you note in an envelope from the experimenter. The subject that was randomly assigned to receive money also received money in the envelope according to his or her earnings (200 ECU=10 euro).

Figure 3.1 This figure shows all the steps that were conducted during the experiment.

Figure 3.2 This figure shows the control question that was included in the Ability Treatment.

(28)

3.3 Hypotheses

The literature lacks studies that focus on the relation between competition and the range of spiteful behaviour. There is evidence in the literature that shows that competition does influence spite. However, this is not elaborate enough to provide a clear-cut answer. The following paragraphs describe the hypotheses that provide an answer to the research question.

3.3.1 Ability Versus Lottery

There is a difference between participating in a competitive task and participating in a lottery. According to Van Anders and Watson (2007) ability-based competition results in different outcomes compared to chance-determined competition. According to Van Anders and Watson (2007) individuals who win a competition based on ability are driven by their desire to win or lose (external recognition) but also whether the outcome corresponds to their ability. Grosch et al. (2017) show that a higher level of competitiveness does result in lower levels of pro-sociality. This generated into the following hypotheses:

H1.1: Subjects who participate in the Ability Treatment are more likely to destroy in a

Joy-of-Destruction Game than subjects who participate in the Lottery Treatment

H1.2: Subjects who participate in the Ability Treatment choose higher levels of

destructions in a Joy-of-Destruction Game than subjects who participate in the Lottery Treatment.

3.3.2 In-group Versus Out-group

According to Muller et al. (2012) individuals who won in an ability-based competitive task are more aggressive towards those who lost. This is because of the status that is related to winning. Muller et al. (2012) argue that winning a competition leads to feelings of entitlement and influences aggressive behaviour. Jauernig et al. (2016) provide further proof that winners are more likely to aggress towards losers and not winners of a competitive task. They also show that subjects who lost in the

competitive task choose lower levels of destruction towards losers than towards winners. The conclusions of Muller et al. (2012) and Jauernig et al. (2016) established the following hypotheses:

(29)

H2.2: Losers choose lower levels of destruction towards Losers than towards

Winners.

3.4 Method of Analysis

The experiment resulted in the following variables: results of part 1 (Winners or Losers), earnings of part 1, matched pairs (whether they are matched with an in-group member or an out-group member), destruction rate, gender, age and field of study, final earnings and whether their choice was implemented. Ultimately, the following variables are the most important for testing the hypotheses: results of part 1 (whether they were Winners or Losers), the person they were matched with (an in-group

member or an out-group member), whether or not they destroyed and their destruction rate. The destruction rate is a way to measure spiteful behaviour in this experiment.

Since the sample size is small (N=48)(N=24 for each treatment and N=12 for each sub-treatment (in-group or out-group sub treatment)), this thesis relies on non-parametric tests to make any statistical inferences. The Exact Fischer test (one-sided) is used to test whether participating in the Ability Treatment will result in a higher frequency of destroyers than participation in the Lottery Treatment (H1.1). The variable that is used to test this is the dummy variable Destroy (1 if you destroyed and 0 if you did not destroy). The second variable that is used is the Treatment (Lottery or Ability).

Hypothesis 1.2 involves testing whether the average destruction level in the Ability Treatment is higher compared to the average destruction level in the Lottery Treatment. This hypothesis is added to provide further evidence whether type of competition influences the destruction level in a JOD game. For example, imagine if destruction frequency were lower in the Ability Treatment and higher in the Lottery Treatment, this would indicate that in a lottery individuals are more likely to engage in spiteful behaviour. However, it could still be the case that average destruction levels are actually higher in the Ability Treatment than in the Lottery Treatment. This would result in conflicting results. So, both hypotheses are tested. The second

hypothesis (H1.2) is tested with a Mann–Whitney U test (one-sided). The relevant variables for this test are average destruction levels in both the Lottery Treatment and the Ability Treatment. The Mann-Whitney U test is conducted for both the

conditional level and the aggregate level. The conditional average level only includes the average of the subjects that did destroy. The aggregate average level includes the

(30)

average level of all subjects (subjects that did not destroy and subjects that did destroy).

The second set of hypotheses focus on the level of spiteful behaviour between in-group or out-group members. In the experiment, subjects were randomly matched with someone who was an in-group member or an out-group member. Muller et al. (2012) and Jauernig et al. (2016) demonstrate that winners are more likely to engage in spiteful behaviour towards losers. This forms the basis for the second hypothesis. The first hypothesis (H2.1) is tested with the following variables: results of part 1 (Winner), in-group or out-group treatment (whether they were matched with a Loser or a Winner) and the dummy variable Destroy. Hypothesis 2.1 is tested with an Exact Fischer test (one-sided) that tests whether the frequency of destruction is higher towards Losers than the frequency of destruction towards Winners. Also, a Mann-Whitney U test (one-sided) is conducted to test whether Winners choose higher destruction levels towards Losers than towards Winners. This will be done for the conditional level (only those who did destroy) and for the aggregate level (all subjects). These two tests are included to avoid conflicting results (see previous paragraph). The relevant variables to conduct the Mann Whitney U test are results of part 1 (Winner), in- and out-group sub-treatment (whether they were matched with a Loser or a Winner) and the destruction rate.

The similar tests are conducted to the test hypothesis 2.2. Jauernig et al. (2016) also provided evidence that losers choose lower levels of destruction towards losers than towards winners. First, an Exact Fischer test (one-sided) is used to test whether the frequency of destruction towards Winners is higher than the frequency of destruction towards Losers. The relevant variables for this test are: results of part 1 (Loser), in- and out-group sub-treatment (whether a subject was matched with a Loser or a Winner) and the dummy variable Destroy. Similar to hypothesis 2.1 a Mann-Whitney U test is conducted. The relevant variables of the Mann Mann-Whitney U test are results of part 1 (Loser), in- and out-group sub-treatment (whether they were matched with a Loser or a Winner) and the destruction rate. The Mann Whitney U test is conducted at the conditional and the aggregate level.

The third method that is used is an OLS Regression. This regression model allows us to control for variables such as age, gender and educational background. The following formula is tested:

(31)

𝑦 = 𝛼 + 𝛽_!𝐴𝑏𝑖𝑙𝑖𝑡𝑦 + 𝛽_!𝑌𝑜𝑢𝑛𝑔 + 𝛽_!𝑆𝑜𝑐𝑖𝑎𝑙𝑆𝑐𝑖𝑒𝑛𝑐𝑒𝑠 + 𝛽_!𝐸𝑥𝑎𝑐𝑡𝑆𝑐𝑖𝑒𝑛𝑐𝑒𝑠 + 𝜀 The variable y stands for the Destruction Rate and is a continuous variable. Ability is a dummy variable, which is 0 when participating in the Lottery Treatment and 1 when subjects participate in the Ability Treatment. Second the variable Young is also a dummy variable that is 0 when subjects are older than 40 and 1 when subjects are younger than 40. The education variables (Exact Sciences, Social Sciences, Arts, Economics/Business/Finance, Humanities, Natural Sciences, Medicine, Other and No Study) are grouped together, because this reduces the number of observations we need to conduct a regression. As a rule of thumb, a regression model should have

approximately 10~20 observations for each explanatory variable (Austin and Steyerberg, 2015). Grouping the variables together allows us to include all the relevant variables with a sample size of 48. This resulted in three categories: Exact Sciences/Natural Sciences/Medicine, Humanities/Social Sciences/Economics and Other/No Study. Only two variables (Social Sciences and Exact Sciences) are included to avoid multicollinearity.

3.5 Data

3.5.1 Sample Characteristics

Before making any statistical inferences about the result, the analysis will focus on a general summary of the experimental results. The relevant variables that are discussed in the following section are: results of the Ability/Lottery Treatment, in/out-group matching, destruction rate, gender, age and field of study. Ultimately, the following variables are important to test the hypotheses: results of the Ability/Lottery Treatment (Winners or Losers), in/out-group matching (with whom subjects are matched) and the destruction rate.

The experiment consisted of two treatments: The Ability Treatment and the Lottery Treatment. In total 48 individuals participated in the experiment and 24 of those individuals were randomly assigned to the Ability Treatment and the other 24 subjects were assigned to the Lottery Treatment. A summary of the characteristics of the subjects can be found in figure 3.3. The graph shows that the experiment consisted of an equal number of male and female (Female: N=24 and Male: N=24).

Furthermore, the sample consisted of one subject that was younger than 20, 34 subjects that were between 20-40 years old and 13 subjects that were over 40 years

(32)

old. The education background differed across the sample: 12 subjects that

studied/study Social Sciences, nine subjects that studied/study Economics, Business or Finance, six subjects that study/studied Art, six subjects that study/studied

Humanities, four subjects that had a background in Natural Sciences, three subjects that had a background in Exact Sciences. Eight subjects did not study or studied something else that was not included in the questionnaire (for example, Hospitality).

3.5.2 Destruction in the Ability and the Lottery Treatment

The experiment consisted of two parts. In the first part subjects participated in a lottery (Lottery Treatment) or they participated in an ability-based competitive task (Ability Treatment). In the second part, the subjects participated in a JOD game. In this game, the subjects had the possibility to destroy the earnings of their partner with a minimum of 0 ECU and a maximum of 100 ECU. In total 15 subjects chose a destruction rate larger than zero (31.25%) and 33 chose a destruction level of zero (68.75%). The aggregate average level of destruction was 23.33 (SD=38.99 ECU). The average mean of those who did destroy was (which will be called conditional mean/average from this point forward) 74.67 ECU (SD=31.59 ECU).

In the Lottery Treatment, 8 individuals (33.33%) chose a destruction rate larger than zero. In the Ability Treatment, 7 individuals chose a destruction rate larger than zero (29.17%). The mean of the aggregate level of destruction in the Lottery

Treatment was 25.83 ECU (SD=40.21 ECU) and the conditional mean was 77.5 ECU (SD=27.12 ECU). In the Ability Treatment, the aggregate average destruction level was 20.83 ECU (SD=38.44 ECU) and the conditional mean was 71.43 ECU

(33)

(SD=38.05). Figure 3.4 and 3.5 provide an overview of both the conditional as the aggregate average of the destruction level and the frequency of destruction between the two treatments.

3.5.3 Destruction in the Sub-treatments

The experiment also consisted of two sub-treatments: The In-group or Out-group sub-treatment. This ultimately lead to four groups in each treatment: Winner vs. Winner, Winner vs. Loser, Loser vs. Loser and Loser vs. Winner. Each group consisted of six subjects.

In the Winner vs. Winner sub-group, only one person chose a destruction rate higher than zero in both the Lottery and the Ability Treatment. The aggregate average

Figure 3.5 The graph shows the aggregate average destruction level for each sub-treatment and for each sub-treatment.

Hard work versus luck : a study on competitiveness and spiteful behaviour