The eects of status competition on social ties structures

University of Amsterdam

Master Thesis Behavioural Economics

Faculty of Economics and Business

The eects of status competition on

social ties structures

Author:

Casper Eimers

Student number:

5922097

Supervisor:

Prof. Dr. T.J.S. Offerman

August 8, 2014

Contents

1 Introduction 3

2 Competition for social status 6

2.1 Reciprocity, altruism and social ties . . . 6

2.2 Social status . . . 8

2.2.1 Competitive altruism . . . 9

2.2.2 Constructive and destructive envy . . . 10

2.2.3 Predictions . . . 12

3 Experimental design and procedures 13 3.1 Methodology . . . 13

3.2 The circle-test . . . 15

3.3 A public good game including status competition . . . 17

3.3.1 Payo structure . . . 18

3.3.2 Treatments . . . 21

3.4 Questionnaire . . . 22

4 Results 23 4.1 Attitudes and social ties . . . 23

4.2 Contributions in the public good game . . . 25

4.2.1 General contribution results . . . 26

4.2.2 In-group contribution dynamics . . . 26

4.3 Status competition and social ties . . . 29

4.3.1 Social ties formation . . . 29

4.3.2 Prediction testing . . . 32

4.3.3 Social ties structures . . . 33 5 Conclusion and discussion 35

CONTENTS 2 A Instructions and decision forms 38 B Individual contributions 41

Chapter 1

Introduction

Meet Jack, an ambitious quantitative analyst, employed by one of the major strategy consul-tancies in the world and currently working on a group project with a great deal at stake. Jack and his colleagues are nearing a deadline but so far his team failed to come up with a decent solution to solve the problem. Everybody is starting to fear the worst, until one of Jack's group members, Pete, provides the critical input needed to nd the solution. In this scenario, Jack is likely to develop a tie with him, as the critical input provided by Pete benets the team as a whole, including Jack. However, ambitious as Jack is, he might envy Pete at the same time, especially when Pete is individually rewarded by a salary raise, a promotion or even a simple compliment from the boss. Moreover, Pete receives general recognition and appreciation from his colleagues for saving the team, which acquires him social status. Ultimately, this aects the positive social tie that Jack originally developed with Pete, potentially suppressing the positive attitude toward his colleague.

There is no doubt that humans are social animals (Aronson, 1999; Buss, 2004). Where most mammals help and care for each other only within small groups of their own kin, humans have the unique ability to form and cooperate within large social groups, which include many generalized unknown individuals (McAndrew, 2002). However, economic-related behavior often takes place in relatively small groups like the working group described above. For this type of group interactions, social relationships are much more evident and aective inuences like personal chemistry, emotions and attitudes become more relevant. Van Dijk et al. (2002) and Sonnemans et al. (2006) examine social dynamics in an economic context, focusing on the formation of social ties structures in groups through interaction in a public good dilemma. Various factors, including the individual contribution levels in the public good game, are found to have a signicant eect on the development of social ties between group members. But both studies seem to have overlooked one important factor which is illustrated by the example above: the competitive element of group interactions.

CHAPTER 1. INTRODUCTION 4 Motivated by the studies of Van Dijk et al. (2002) and Sonnemans et al. (2006), this thesis examines the role of competition in the development of social ties in group interactions. However, competition can be dened in many dierent ways depending on the context of the group interaction. One type of competition that frequently occurs, but whose eects on social ties formation have been generally unexplored, is (social) status competition (Hardy and van Vugt, 2006). Returning to Jack, the importance of status competition and the potential conse-quences for the development of social ties are illustrated. Even though Jack and his colleagues work together on a project in order to collectively achieve the best result, each colleague might simultaneously strive for individual rewards, which can be either formal (incentives) or informal (status acquisition). Hence, status competition could play an important role in ties formation and group performance not only in this example, but also in economic-related group interactions in general. Therefore, this study focuses on the eects of status competition on social ties development and group performance in a public good game.

One type of emotion that often emerges in the presence of competition and which is likely to aect social ties formation and group performance is envy. In the context of status competition, a small stream of literature argues that envious feelings could lead to invest more in order to achieve a certain status that is similar to the envied person, indicating constructive envy is at play when competition for status is present (Marglin, 2002). However, a much larger body of experimental evidence exists on the eects of destructive envy on the collective performance within groups (Kirchsteiger, 1994; Zizzo and Oswald, 2001; Beckman et al., 2002; White et al., 2006). For example, White et al. (2006) demonstrate the relationship between frequent social comparisons and the destructive emotions and behavior that is elicited as a result of these comparisons. They showed that police ocers who claimed to make frequent social comparisons were more likely to show in-group bias and be less satised with their jobs, as a result of envious feelings.

While most of these studies focus on the eects related to group performance, this study looks specically at the development of social ties. Some research on this particular topic has been done by Tsaur and Wang (2011), who demonstrate the roles of reciprocity and competition in the relationship between personal ties and performance of strategic alliances occurring in Taiwan's travel industry. Senior executives of tour companies located in Tapei who participated in horizontal strategic alliances were invited to participate in the study through a structural questionnaire. Using equation models, Tsaur and Wang (2011) show that personal ties aect the performance of strategic alliances positively, as expected. However, when the intensity of competition is increased, this eect becomes insignicant, which suggests that competition has a moderating eect on the relationship between personal ties and group performance.

In order to understand how social interaction aects economic behavior and outcomes, the notion of a social tie needs to be formalized. Baumeister and Leary (1995) argue that the

CHAPTER 1. INTRODUCTION 5 aective components, i.e. sentiments, form the key element of a tie, which closely relates to the extent to which individuals care about the wealth of others. Nevertheless, it should be emphasized that a social tie is not just given by the value or weight an individual attaches to the wealth of another agent. This would only hold in case the individual is completely neutral towards any random stranger. However, when an agent holds a specic attitude towards a random stranger, a social tie with an interaction partner can only be said to exist when the value attached to the partner's being diers from the value attached to a stranger's well-being. Hence, in order to establish the existence of a social tie, two dierent attitudes need to be measured: the attitude of an individual towards the interests of a random stranger, and the attitude of the same individual towards the interests of a specic other (e.g. group member). During the experiment, social ties measurement is done using a so-called "circle-test" (Sonnemans et al., 2006), which is discussed in Section 3.2. Of course, there needs to be some form of social interaction for the social ties to (potentially) develop. Conrmed by the ndings of various studies (Coleman, 1990; Feld, 1981; Levine and Moreland, 1998), social ties are acknowledged to develop through prolonged interaction, and disappear when the interaction ends. For this reason, and similar to Sonnemans et al. (2006), a repeated public good game is used to allow for the formation of social ties. To summarize, the experiment starts with a circle-test, which measures the attitude towards a random stranger, then continues with a repeated public good game, and ends with a second circle-test, which measures the attitudes towards the group members each subject interacted with during the public good game. The experiment consist of two treatments, one in which competition for status is present, and one in which it is absent.

The results indicate a positive eect of status competition on the development of social ties in groups, which is in conict with the expectations. No proof was found of envious feelings towards high-status individuals. Instead, these individuals were generally more appreciated by other group members. Also, in the presence of status competition, signicantly less negative social ties were formed compared with the condition in which competition was absent. These ndings are broadly consistent with the existing literature (Eckel et al., 2010; Hardy and van Vught, 2006). Furthermore, they support the ndings by Van Dijk et al. (2002) and Sonnemans et al. (2006) who found that the value of social ties depended on the success of interaction during the public good game.

The structure of this thesis is as follows. Chapter 2 gives an overview of the existing litera-ture on social ties and status competition. Additionally, based on experimental evidence some predictions are made with regard to the experiment conducted in this thesis. Next, Chapter 3 presents the methodology and the experimental design, including an extensive explanation of the circle-test, the public good game and the dierent treatments. The results are pre-sented and analyzed in Chapter 4, while Chapter 5 concludes and provides some suggestions for future research on this topic.

Chapter 2

Competition for social status

Given the existing evidence (Samuelson, 1993; Sen, 1995; Bolton and Ockenfels, 2000; Fehr and Schmidt, 2006), the real question is not whether human beings have other-regarding preferences, but under which conditions these preferences have important eects from an economic and social point of view. This chapter reviews the existing literature on other-regarding preferences in a competitive environment and it discusses the predictions made with regard to the experiment conducted in the thesis. First, Section 2.1 gives a small introduction on other-regarding preferences together with a link to the concept of social ties (van Dijk en van Winden, 1997). Then, Section 2.2 discusses competitive altruism in relation to social status, which results in a number of predictions with respect to the experiment.

2.1 Reciprocity, altruism and social ties

Until the 1980s economists were convinced that material self-interest is the sole motivation of all humans, often referred to as the self-interest hypothesis. While some economists still rely on this hypothesis, a large and growing amount of evidence is gathered by experimental economists and psychologists which indicates that a substantial percentage of the human pop-ulation is strongly motivated by other-regarding preferences such as reciprocity, fairness and altruism (Samuelson, 1993; Sen, 1995; Bolton and Ockenfels, 2000; Fehr and Schmidt, 2006). A strand of literature closely related to the set mentioned above focuses on economic behavior in networks that include social interaction and interpersonal relationships. Sociologists like Coleman (1984) and Uzzi (1996) emphasize the importance of this behavior in social networks, as it bears consequences for economic interaction in markets and organizations in general.

In accordance with the social ties model by van Dijk and van Winden (1997): "A social tie between two individuals is dened in terms of the extent to which they care about each other's well-being." Hence, social ties are closely related to the theory of other-regarding and

CHAPTER 2. COMPETITION FOR SOCIAL STATUS 7 interdependent preferences. Moreover, ties develop through continued interaction and they disappear when the interaction ends (Coleman, 1990; Levine and Moreland, 1998). Finally, dependent on whether the interaction between two individuals was valued positively or neg-atively, positive or negative ties are formed, which may inuence future interactions between these two individuals (Feld, 1981).

Relating this to the concept of other-regarding preferences, it can be argued that positive ties enhance other-regarding preferences which lead to reciprocal and altruistic behavior, while negative ties lead to more selsh behavior. This is supported by the paper of Sonnemans et al. (2006), who analyze the development of social ties in a four-player repeated public good game. As one of their ndings shows, better performing groups (in terms of contribution levels and payos) developed more positive social ties compared with worse performing groups. As several other studies have shown, both from a theoretical (van Dijk and van Winden, 1997) and an empirical perspective (Bault et al., 2014; van Dijk et al., 2002), an important factor in the development of these ties is the payo of the interaction. However, a factor that has been mostly neglected in these studies, is competition.

Competition

Many scenarios similar to the example of Jack and his team of colleagues in the introduction can be thought of, e.g. a football team where the entire team benets in case one team member scores a goal, while other players in similar positions might have had to desire to score that goal in order to avoid the possibility of being substituted. Essentially, both the examples boil down to the same conclusion: while social ties are developed through group interactions, they are simultaneously aected by a competitive element often involved in group interactions, either by in a positive or negative way. Hence, it is important to take this competitive element into account when analyzing the development of social ties during group interactions, even more because it potentially inuences the (reciprocal and interdependent) preferences of individuals. While several studies have looked at the eect of competitive pressure on reciprocal and interdependent preferences (Bolton and Ockenfels, 2000; Fehr and Schmidt, 2006; Falk and Fishbacher, 2006), the eect of competition with respect to social ties development has received little attention so far. There is one study by Tsaur and Wang (2011), which is relatively close related to the social ties model, wherein the authors conduct a eld experiment and demonstrate the roles of reciprocity and competition in the relationship between personal ties and performance of strategic alliances occurring in Taiwan's travel industry. Their ndings show that personal ties aect the performance of strategic alliances positively, as expected. Nevertheless, when the intensity of competition is increased, this eect becomes insignicant, which suggests that competition has a negative eect on the relationship between personal ties and group performance.

CHAPTER 2. COMPETITION FOR SOCIAL STATUS 8 However, competition can be dened in many dierent ways depending on the context of the group interaction. The study by Tsaur and Wang (2011) was conducted in a market environment, hence it was about market competition. For group interactions that take place in non-market environments, dierent types of competition exist. One type that often occurs and has been studied on a broad scale, is (social) status competition (Hardy and van Vugt, 2006). The next section digs further into social status and the competitive element related to it by discussing several experimental studies of the eects of social status on decision behavior in groups.

2.2 Social status

Since Becker's (1974) theory of social interaction, economists have recognized the importance of social status and status competition in economic decision making. According to Frank (1985), status is a motive in itself. Individuals engage in numerous activities in order to acquire and display status, sometimes at the cost of monetary compensation. Moreover, the behavior of agents with high statuses may function as guidance for surrounding individuals, which aects the choices made by lower-status individuals. So individuals are generally tended to mimic the decisions made by high-status agents, which is conrmed by experimental research. For example, Eckel and Wilson (2001) provide evidence that a commonly observed agent can inuence the equilibrium selection in a coordination game. Also, Kumru and Vesterlund (2005) show that high-status rst-movers are more likely to be mimicked in a two-person sequential voluntary contribution game compared with low-status rst-movers.

More interesting in light of the research conducted in this thesis, however, is the question whether social status can serve as a useful mechanism to solve public goods problems, i.e. increase the level of contributions in a public good game. In fact, some attempts to answer this question have already been made. For example, Munoz-Garcia (2011) examines the role of status acquisition as a motive for contributing in a public good dilemma, where every agent's status is given by the dierence between his contribution and that of the other agent. He nds that agents contribute more than in standard models where status is absent and that the contribution level is increasing in the value agents assign to status. In another study, Eckel et al. (2010) explore the impact of social status on behavior in public good games with a specic network structure. The network has a central player that is observed by a group of peripheral players who, in turn, only observe the central player. They nd that high-status central players are more likely to be followed in the key situation where the peripheral player is contributing less than the central player, which is in line with the ndings of existing literature on this topic.

However, in the study by Eckel et al. (2010) status is awarded based on scores on a general knowledge trivia quiz, where the central position is given to either the high scorer or the low

CHAPTER 2. COMPETITION FOR SOCIAL STATUS 9 scorer depending on the treatment. Hence, there is no actual competition for social status in this case. The experimental design used in this thesis diers in the fact that status competition is introduced by letting the rank (and status) depend on the contribution level, which may shift as the game progresses. For this particular design, the same question arises as before: can social status competition serve as a useful tool to increase the level of contributions in a public good dilemma? More importantly, how does status competition aect the development of social ties in a public good game?

The studies discussed so far all indicated positive eects of status on group performance. However, in a study by Berger et al. (1972) group members actively attempted to raise their status, while status was unrelated to task competence, and thus hindering the group's performance. Indeed, substantial evidence has been provided in evolutionary anthropology (Barkow, 1975; Chapais, 1991; de Waal, 1996) and sociology (Kemper and Collins, 1990) that status can be considered as an intrinsically valued emotional good. The pleasure of status can serve as an end in itself. A good example is the study by Huberman et al. (2004), who demonstrated that subject were willing to trade real money for short-lived status recognition that had no further benets.

Hence, status-seeking behavior that is purely motivated by the intrinsic preference for status encourages competition in group interactions. Every member of the group seeks the highest rank and will make an eort to obtain it. But as there is only one highest rank, the result of status-seeking behavior is competition for status in a group. Earlier ndings regarding the eect of status competition on group performance and relationships/social ties are rather divergent. For example, Wu et al. (2011) show that status and relationship maintenance hinder one another: status reduces the collaboration benet from relationships, and increase only individual, but not collaborative eort. On the other hand, Hardy and van Vught (2006), who examined the relationship between altruistic behavior and the emergence of status hierarchies within groups, present evidence that people were more altruistic in a reputation environment where contributions were made public. The authors referred to this as the competitive altruism hypothesis, which is further discussed in the section below.

2.2.1 Competitive altruism

Throughout the last few decades, two main theories of altruism have emerged: the kin selection theory by Hamilton (1964) and the reciprocal altruism theory by Trivers (1971). Both theories have received a substantial amount of empirical support (Axelrod, 1984; Neyer and Lang, 2003; van Lange and Semin-Goosens, 1998). Still, as with almost any theory, question remain about to what extent they can fully account for the diversity of altruistic patterns in human behavior (Fehr and Fischbacher, 2003). Hardy and van Vught (2006) present a novel theory of altruism, the so-called competitive altruism hypothesis, which they believe can account for a range of

CHAPTER 2. COMPETITION FOR SOCIAL STATUS 10 altruistic behaviors among humans that theories of kinship and reciprocity have a hard time explaining.

As stated by Hardy and van Vught (2006): "Competitive altruism is the process through which individuals attempt to outcompete each other in terms of generosity." It emerges because altruism, if publicly displayed, enhances the status and reputation of the altruists, which makes them more attractive interaction partners, providing them benets that are not attainable by non-altruistic individuals. Based on this theory, one would expect a positive eect of status competition on contribution levels (and social ties) in a public good game. But why would groups reward high-status altruists in the rst place instead of exploiting them?

One explanation is given by costly signaling theory (Zahavi and Zahavi, 1997), with the peacock's tail as a classic example. The tail of male peacock handicaps the animal because it is very dicult to grow and severely limits his movement, making it an easy catch for predators. On the other hand, the tail is critical for the peacock's reproduction, as it advertises the animal's quality as a mate. Hence, altruism might qualify as a handicap: the altruist benets by increasing his or her social status and thus the likelihood of being chosen as a partner for reproduction. Another explanation comes from indirect reciprocity theory (Alexander, 1987), which states that groups may compensate altruists by giving them status because then they can continue to benet from the presence of these individuals. Hence, by generously contributing in a public good game, an individual might build up a reputation for being altruistic, which in return might make them more attractive as future exchange partners, developing positive social ties with the other players. Disregarding the precise mechanism, Hardy and van Vught (2006) argue that it is likely that competition is induced when reputation and status are at stake, because individuals will be competing with their group members in terms of generosity to advertise themselves as future exchange partners.

Based on the competitive altruism hypothesis, Hardy and van Vught (2006) make a number of predictions about the emergence of altruism in a public good game where status competition is present. First, they expect high contributors to be seen as higher in status. Second, "nice guys should nish rst", meaning high contributors are expected to benet in the long from their altruistic behavior in a way that they are more likely to be chosen as future interaction partners by other group members. Finally, altruists should be awarded with more status the greater the costs of their contributions to the group. The results are consistent with their predictions and the competitive altruism hypothesis: individuals may behave altruistically for reputation reasons because of the selective benets that accrue to the generous.

2.2.2 Constructive and destructive envy

Although the ndings by Hardy and van Vught (2006) are in favor of the competitive altruism hypothesis, the authors seem to have overlooked a commonly observed human characteristic

CHAPTER 2. COMPETITION FOR SOCIAL STATUS 11 which often plays a role in competitive environments: envy. Envy is a complex emotion that occurs as soon as multiple individuals engage in mutual comparison, inuencing the behavior of both envious and envied individuals. A substantial amount of studies have looked into the idea that interpersonal comparisons matter for an individual's subjective well-being (Frank, 1985; Mui, 1995; Solnick and Hemenway, 2005; Zizzo, 2007). In the presence of a gap between two individuals (e.g. in income), an envious person is likely to undertake eorts such that this gap is reduced. As stated by Grolleau et al. (2009), reducing the gap may be done in two distinct ways. In the rst way, the envious agent invests in order to reach the state enjoyed by the envied individual, referred to as constructive envy. In the second way, however, the envious agent invests in order to reduce the state enjoyed by the envied individual, sometimes even at a cost, which Grolleau et al. (2009) dene as destructive envy. While both forms of envy lead to the same result, i.e. a gap reduction between the two individuals, the economic implications greatly dier.

In the context of status competition, a small stream of literature argues that envious feelings could lead to invest more in order to achieve a certain status that is similar to the envied person, indicating constructive envy is at play when competition for status is present. According to Marglin (2002, p. 23), envy can be described as a virtue that "serves an economic purpose in motivating individuals to maximum exertion and eort. The desire to improve one's relative status is the engine that drives the economic train." But the stream of literature that supports the presence of constructive envy in economic decision making processes within groups is rather contrasting.

A much larger body of experimental evidence exists on the eects of destructive envy within groups (Kirchsteiger, 1994; Zizzo and Oswald, 2001; Beckman et al., 2002; White et al., 2006). Zizzo and Oswald (2001) nd experimental evidence that people dislike other individuals' wealth to such an extent that they are willing to sacrice some of their own money in order to reduce that wealth. Moreover, based on an international setting Beckman et al. (2002) show that factors like envy and malice can lead to an opposition to Pareto optimal gains that may reach up to 60% in some transitional countries. More relevant in the context of status competition and social comparison is the study by White et al. (2006). Although social comparisons may seem to serve various benecial features like self-enhancement, White et al. (2006) argue that frequent social comparisons have a "dark side". In two dierent studies they demonstrate the relationship between frequent social comparisons and the destructive emotions and behavior that is elicited as a result of these comparisons. In the rst study, people who claimed to make frequent social comparisons were more likely to experience destructive envy, guilt and regret. In addition, the second study showed that police ocers who claimed to make frequent social comparisons were more likely to show in-group bias and be less satised with their jobs, again as a result of envy.

CHAPTER 2. COMPETITION FOR SOCIAL STATUS 12

2.2.3 Predictions

To a great extent, the ndings reported by White et al. (2006) are consistent with the example in the introduction, where Jack beneted from his colleague's critical input, but nonetheless envied this colleague as a result of social comparison. Consequently, this (destructive) envy might aect the positive social tie that Jack originally developed with his colleague throughout the group project. Intuitively, this makes sense: an individual may dislike it when her group member's actions outperform her (in terms of social status), even though she simultaneously benets from these specic actions. Relating this to the article by Hardy and van Vught (2006), an individual might even dislike it when another agent outperforms her in terms of being more altruistic. While Hardy and van Vught (2006) claim that this competitive altruism enhances group performance, one could argue that this competition for status (and altruism) works counterproductive. One might consider the behavior of the high-status altruist to be ashy or poshy, potentially leading to destructive envy.

Based on this line of reasoning, a number of hypotheses related to the eects of status competition on social ties development and contribution levels in a public good game is ex-plored. First of all, it is expected that individuals who are altruistic by nature (social value orientation) will also be the highest contributors in the public good game and therefore obtain the highest status:1

H1. The greater an individual's social value orientation, the greater his or her status.

Second, the competition for status potentially leads to destructive envy, especially when certain individuals exhibit (overly) altruistic behavior:

H2. The higher an individual's status, the more group members will (destructively) envy the individual, leading to attempts to degrade the individual.

The hypotheses will be explored using experimental data. The experimental design and pro-cedures are extensively discussed in the following chapter.

Chapter 3

Experimental design and procedures

3.1 Methodology

In order to test the hypotheses laid down in Section 2.2.3, an experimental design was used which consisted of three parts. First, a circle-test was applied to measure the social ties.1

Second, subjects played a four-person repeated public good game for six rounds, in which each group member received a monetary endowment and had to decide how much to contribute to the group and how much to keep for themselves. As with most public good games, the payo structure was such that it was nancially better for the individuals to keep the tokens themselves, but if every member did this, they would each be worse o than if they all contribute to the group (Dawes, 1980). When all group members had made their decision for a specic round, their individual contributions to the group for that round were made public by ways of a feedback table, which is discussed in more detail in Section 3.3 Finally, a second circle-test was applied, where subjects had to perform the test three times: one time for each group member they were linked with in the public good game. Details regarding the purpose of the two circle-tests follow in Section 3.2.

The hypotheses in 2.2.3 were tested by making a distinction between two dierent condi-tions: a control condition and a status condition. The essential dierence between the two conditions was the way of representing the feedback table during the public good game. In the control condition, information on the contributions of each player was alphabetically ordered. However, in the status condition, this information was rank-ordered, meaning the player with the highest rank appeared as rst in the feedback table. More details on the two conditions are provided in Subsection 3.3.2. At the end of the experiment, participants in the status condition had to ll out a questionnaire about their feelings and attitude towards the Star Player. Table 1 provides an overview of the three dierent parts of the experiment, including

1_{Details regarding the circle-test follow in Section 3.2}

CHAPTER 3. EXPERIMENTAL DESIGN AND PROCEDURES 14 Control condition Status condition

Part 1 Circle-test in wich each subject is coupled with an (randomly determined) unknown other

Part 2 Public good game in which each participant, randomly assigned to a group of 4 subjects, has to divide 10 tokens between a private and public allocation. After each round feedback is provided, which is alphabetically ordered.

Between 5 and 12 rounds are announced; 6 rounds are played.

Public good game in which each participant, randomly assigned to a group of 4 subjects, has to divide 10 tokens between a private and public allocation. After each round feedback is provided, which is rank-ordered. The highest rank is assigned status of "Star Player". Between 5 and 12 rounds are announced; 6 rounds are played.

Part 3 Second circle-test in which each subject is coupled with the three subjects he or she was matched with in Part 2.

Table 3.1: The design of the experiment, including the two treatments.

the two conditions just explained.

Procedures

Forty-eight subjects participated in two sessions, of which about two-thirds were students from the department of Economics. Each participant was randomly allocated to one of two conditions: the control group or the status condition. In total, there were six groups of 4 participants in each condition, resulting in 24 participants per condition. The experiment was completely computerized and was conducted in the CREED laboratory of the University of Amsterdam. Furthermore, the experiment was part of a collective event, hosted by Joep Sonnemans, Professor of Behavioral Economics at the University of Amsterdam. Two sessions took place, where each session consisted of a sequential series of seven experiments conducted within a time frame of two hours. Hence, each experiment lasted between 15 and 20 minutes, which required some restrictions regarding the experimental design. Upon arrival in the lab, participants randomly picked a card containing an identication number and took place at the seat corresponding to that number. This way, participants were randomly and anonymously assigned to a group of four people. Once everyone was seated, the tasks were introduced.

CHAPTER 3. EXPERIMENTAL DESIGN AND PROCEDURES 15 table that showed the payo structure of the game. This is not a standard procedure, but it was done in order to save time as the payo structure was non-linear and thus not very straightforward. Subjects otherwise might have spent too much time trying to gure out the exact payo structure. Furthermore, the number of rounds announced was kept exible, namely between 5 and 12, in order to avoid the so-calledend-eect which often occurs in these type of games (Ledyard, 1995).

After the public good game, the circle-test was applied for the second time. For this second circle-test subjects had to perform the test three times, where the interaction partner (the "Other") in these tests concerned each one of their group members in the public good game. While performing the test, subjects had access to the same feedback table provided during Part 2, which contained information on the contributions of all players during the 6 rounds played. At the end of the experiment, subjects in the status condition had to ll out a questionnaire regarding their feelings towards the Star Player. Examples of the instruction and decision sheets provided during the experiment are included in Appendix A.

3.2 The circle-test

As dened in the social ties model by van Dijk and van Winden (1997): "A social tie between two individuals is dened in terms of the extent to which they care about each other's well-being." However, van Dijk et al. (2002) argue it is important to realize that a social tie is not simply given by the value or weight an individual attaches to the well-being of another person. This would only hold in case the individual is completely neutral towards any given person he or she doesn't know. However, when an agent holds a specic attitude towards a random stranger, a social tie with an interaction partner can only be said to exist when the value attached to the partner's well-being diers from the value attached to a stranger's well-being. The existence of very dierent types of players within public good dilemmas is conrmed by various studies including Fischbacher et al. (2001), who categorize them as condition cooperators and free-riders. These dierent types of behavior may well be related to dierences in attitudes towards strangers. Hence, in order to establish the existence of a social tie, two dierent attitudes need to be measured:

1. The attitude of an individual towards the well-being of an unknown, generalized other. 2. The attitude of an individual towards the well-being of a specic person he or she is

interacting with.

Similar to the studies by van Dijk et al. (2002) and Sonnemans et al. (2006), a so-called "ring-test of social value orientation" is used to measure the two types of attitudes mentioned

CHAPTER 3. EXPERIMENTAL DESIGN AND PROCEDURES 16

Figure 3.1: An example of how the circle-test is displayed to the subjects. Allocations can be changed using either the mouse or the keyboard. When selecting a dierent allocation, the numbers in the upper-right corner are automatically updated. The subject conrms the chosen allocation by clicking the "send" button.

above. This test, introduced by Liebrand (1984), derives the preferences of individuals for distributions of outcomes for themselves and others using a specic game technique. The predictive validity of the ring-test has been shown by various studies (Rusbult and van Lange, 1996), making it a useful tool to measure the attitudes towards a random stranger (see Of-ferman et al. (1996) for an application). However, despite its predictive power, the ring-test is not very practical when the attitudes towards multiple specic others have to be measured. As Sonnemans et al. (2006) argue, repeated measurements would result in too many questions which would require too much time, especially considering the limited amount of time available to conduct the experiment (15-20 minutes). For this reason, Sonnemans et al. (2006) devel-oped the circle-test, in which the attitude of a subject towards a specic other is measured by only one decision. As the experiment in this thesis requires measurements of attitudes towards multiple others, the circle-test is favored over the ring-test.

In the circle-test subjects are asked to select a point on a circle, where each point represents a certain distribution or allocation of money between the person him/herself (S) and a specic other (O). Depending on the selected point, allocations can be positive or negative, where S2_{+ O}2 _{= 1000}2_{, denoted as tokens. Fig. 3.1 provides an illustration of the circle-test. Note}

that when starting the test, the circle is displayed without any vertical or horizontal lines (excluding the axes) and pre-set allocation, as that could have led to anchoring (Kahneman and Tversky, 1974), biasing the results. Once the participant clicks somewhere on the line of the circle, two lines parallel to both axes appear with the corresponding allocation for Self and Other as depicted in Fig. 3.1.

The rst circle-test was used to measure the attitude of the subject towards an unknown generalized other. For this test, each subject was randomly paired with another participant, who was not among the three group members that the respective subject was linked with in part 2. Participants were notied of this on the instruction sheet for part 2 (see Appendix A).

CHAPTER 3. EXPERIMENTAL DESIGN AND PROCEDURES 17 In part 3, after the public good game had been played, three more circle-test were performed. In each of these tests, the "specic other" was one of the group members the subject interacted with in the public good dilemma. The four players in the public good game were denoted players A, B, C and D. Considering player D, he or she needed to perform three circle-tests, with players A, B and C respectively. While performing the tests, all subjects had access to the feedback table provided during the public good game, which showed all contributions made by the subjects throughout the entire game. This way, the circle-tests were able to measure whether the attitude towards a specic group member was inuenced by the decisions made by that group member during the public good game.

The payos for both circle-tests were determined by the choice of the subject him/herself as well as the choice of the other(s). The earnings were given by:

Pi = Si+ O−i,

where Pi denotes the payo for individual i; Si reects the amount of points kept by the

agent i him/herself, while O−i reects the amount of points distributed to the other by

the interaction partner(s) −i. In the rst circle-test, −i reects the unknown generalized other. However, for the second circle-test, −i represents all three group members in the public good game, so O−i becomes the sum of three allocations. Until the end of the experiment,

participants did not receive any information about the choice made by the other(s) in both circle-tests.

3.3 A public good game including status competition

Before the eect of status competition on social ties development can be examined, there is a number of conditions that must be met for status competition to emerge in the rst place (McAndrew, 2002; Zahavi and Zahavi, 1997). First, the behavior needed to obtain the higher status must be costly for the agent to display. Second, the behavior displayed by the agent must be easily observable to the other group members. Third, the decisions or signals made by the individual must be a reliable indicator of some of the underlying characteristics of that individual, e.g. wealth, health or intelligence. Fourth, the behavior must be benecial in the long term to the displaying agent.

Considering these conditions, an ideal framework to test for eects of status competition on the development of social ties would be the public good game. Moreover, by providing a feedback table which publicly presents the individual contributions of each group member per round, participants can easily compare among various contributors, which facilitates the judgment of the underlying quality of the contributors and also provides a competitive en-vironment for those involved. Additionally, the feedback table helps subjects in forming an attitude towards specic group members, which is measured by the second circle-test.

CHAPTER 3. EXPERIMENTAL DESIGN AND PROCEDURES 18 For the public good dilemma task, subjects were assigned to four-person groups, of which the members were identied as players A, B, C and D. Groups consisted of 4 persons rather than 2 or 3, as it was expected that envious feelings were more likely to emerge in larger groups. Throughout the 6 rounds of the game, the composition of the groups as well as the identication numbers of the players remained the same. In each round, participants were given the task to distribute an endowment of 10 tokens over two allocations, which were denoted as "Private" and "Public". In accordance with its name, allocation Private provided a payo exclusively for the subject him- or herself, whereas allocation Public provided a payo received by all group members. The endowments per round were homogeneous among the subjects.

The rst round started after all group members had nished the computerized instructions. For each period participants had to decide how they wanted to distribute their 10 tokens over the allocations "Public" and "Private". During the decision making process, subjects received a hand-out called "Table of earnings per period" which contained the individuals payos per round. After all members had made their decision for that round, they received feedback on the number of tokens distributed by each member to "Public" and their own earnings for the specic round.2

3.3.1 Payo structure

The payo structure in the public good game was non-linear, such that the standard Nash equilibrium was interior in the action space. Also, using a non-linear structure would ensure consistency with the designs of van Dijk et al. (2002) and Sonnemans et al. (2006), allowing for a proper comparison of results. Similar to these two studies, the payo in each round for individual i was of the general form:

Ui(gi, g−i) = Vpublic· 4

X

k=1

gk+ Vprivate· (wi− gi) − (wi− gi)2− C, (3.1)

where gi denotes the contribution to allocation "Public" by individual i and g−i the

contribu-tions to "Public" by the others. Moreover, wi denotes the endowment of individual i in each

round, so (wi − gi) reects the amount of tokens that individual i distributes to "Private";

Vpublic and Vprivate denote the values of each token assigned to allocations "Public" and

"Pri-vate" respectively, and C represents xed costs that are subtracted each round. Note that Vpublic is multiplied by the sum of tokens allocated to "Public" by all four players, as those

tokens generate value for each member of the group.

2_{Throughout this thesis, the terms "tokens distributed to allocation Public" and "contributions" are used}

CHAPTER 3. EXPERIMENTAL DESIGN AND PROCEDURES 19 The next step was to determine the values of the parameters Vpublic, Vprivate and C.

Son-nemans et al. (2006) use the following conguration:

(Vpublic, Vprivate, C) = (7, 21, 60) .

In this study the the conguration of the parameters was carefully considered, as it could substantially inuence the behavior of the participants. In order to elicit status competition and envy within the public good game, the payo structure should satisfy the following two conditions:

1. The level of improvement from the Nash equilibrium payo to the Pareto Optimal payo should be sucient in order to elicit altruistic behavior.

2. The level of "sacrice" needs to be sucient in order to elicit heroism and/or envy. When considering condition 1, it is important to look at the dierence between the payo that is obtained when each player would contribute the Nash amount and the payo received when each player contributes the full amount of tokens (10), which is Pareto-optimal (PO) here. Using a mathematical notation: U(NE)−U(PO). Dening the level of improvement in relative rather than in absolute terms:

IMP = U (NE)−U (PO)

U (PO) . (3.2) For condition 2, a formal denition of "sacrice" is required before the condition can be rewritten using a mathematical notation. Here, sacrice is dened as the loss in earnings incurred when an individual decides to contribute all tokens (10) to "Public", instead of the Nash amount, given that the other three group members stick to the Nash amount. Using a formula, this translates to: U (NE) − U (ALL | NE). Again, dening the level of sacrice in relative terms:

SAC = U (NE) − U (ALL | NE)

U (ALL | NE) . (3.3) Various calibrations of the parameters have been considered. Table 3.2 gives an overview of the options, including the calibration used by Sonnemans et al. (2006).

Based on the calibration by Sonnemans et al. (2006), a value of 0.80 for the level of improvement (IMP) is obtained, while the level of sacrice (SAC) equals 0.67. In other words, the relative improvement in payo from the Nash equilibrium (each player 3 tokens) to the Pareto-optimal allocation (each player all tokens) is 80 percent. Furthermore, the individual that contributes all his/her tokens could have earned 67% more in case (s)he contributed the Nash amount, given that the other group members contribute the Nash amount. While a

CHAPTER 3. EXPERIMENTAL DESIGN AND PROCEDURES 20

Table 3.2: An overview of the dierent options considered for the calibration of the parameters Vprivate, Vpublic

and C. The most suitable candidates are highlighted in blue (options 2 and 3). The calibration used by Sonnemans et al. (2006) is highlighted in orange (option 6).

value of 0.80 for IMP looks sucient to elicit altruistic behavior and cooperation, the 0.67 for SAC doesn't seem to satisfy condition 2, i.e. elicitation of heroism and envy. Therefore, dierent options were considered. Options 2 and 3 were potential candidates, as the level of sacrice (SAC) is substantially higher, 1.09 and 0.91 respectively. However, as the level of improvement for option 2 is somewhat on the poor side (0.60), option 3 was selected. So a token assigned to allocation "Public" yielded 6 points for all group members, while the value of x tokens assigned to "Private" was: 20x − x2_{. For the public good game, 250 points were}

equal to 1 euro. To summarize, the payo in each round for individual i is given by: Ui(gi, g−i) = 6 · 4 X k=1 gk+ 20 · (wi− gi) − (wi− gi) 2 − 60.

So under the usual benchmark assumption of own-payo maximization, the single-stage Nash equilibrium is the allocation of 3 tokens to "Public" and 7 tokens to "Private" for each group member. To see why, consider the payo matrix denoted as "Table of earnings per round", which is included in Appendix A. Suppose none of the group members including you would contribute (i.e. allocate tokens to "Public"), then every member would receive 40 points, which is the minimum payo. Given that the three other members stick to a zero contribution, you are better o contributing 3 tokens yourself, resulting in a payo of 49 points. In fact, regardless of the contribution levels of the other group members, the best reply/action is to always contribute 3 tokens. The same holds for the others, meaning the Nash equilibrium is reached when everyone contributes 3 tokens. When the Nash equilibrium is played, each participants received 103 points.

CHAPTER 3. EXPERIMENTAL DESIGN AND PROCEDURES 21 The Pareto-ecient (PE) allocation entails that subjects allocate all their tokens to "Pub-lic", which gives them a payo of 180 points each. A derivation similar to the one used for the Nash equilibrium can be used to show that this allocation is indeed PE. Suppose all members play the Nash equilibrium resulting in a payo of 103 tokens and that every member decides to increase his/her contribution to 4 tokens, such that the collective amount of tokens equals 16. Now each group members receives 144 points, making everyone better o. This can be re-peated up to 10 tokens per member, when everyone receives 180 points each and the collective payo equals 720 points. Now suppose one member decides to deviate and decrease his/her contribution to 9 tokens. This individual will now receive a payo of 193 points and is better o, while the three remaining members are worse o as they receive 174 points each. As a result, the collective payo is 715 points, which is lower compared with the scenario in which each member contributed the full amount of tokens (720). Hence, all members contributing 10 tokens is the only Pareto-ecient allocation.

3.3.2 Treatments

The hypotheses in 2.2.3 were tested by making a distinction between two dierent conditions: a control condition and a status condition. As discussed in Section 3.1, the essential dierence between the two conditions was the way of representing the feedback table during the public good game. In the control condition, the information on the contributions of each player was alphabetically ordered. However, in the status condition, this information was rank-ordered, meaning the player with the highest rank appeared as rst in the feedback table, followed by the player with the second-highest rank and so on.

The rank of each individual depended on the amount of tokens distributed to allocation "Public" by that individual. More specically, the rank of each group member was determined by the cumulative amount of tokens distributed to "Public" over the three most recent rounds. Moreover, the group member with the highest rank was awarded with the status of Star Player. Only the three most recent rounds were selected in order to stimulate competition among the group members. This can be illustrated with a counterexample: suppose a specic player (e.g. player A) has contributed relatively large amounts of tokens compared with the others during the rst ve rounds of the game. In case contributions over all the rounds would matter for the rank, other players are less likely to exhibit altruistic behavior in order to compete for the rank, as they have too much catching up to do in order to reach the cumulative amount of contributions of the Star Player. Hence, it makes sense to take into account only the three most recent rounds in determining which player is the Star Player.

Examples of the feedback tables in both conditions are depicted in Fig. 3.2. Note that the examples are based on the same input. Finally, dierent titles/statuses have been considered for the player with the highest rank, including "Helper", "Benefactor" and "Most generous

CHAPTER 3. EXPERIMENTAL DESIGN AND PROCEDURES 22

Figure 3.2: Examples of two feedback tables as displayed on the computer screen during the decision phase of Part 2 and 3 of the experiment, one for each condition. Both tables are based on the same input.

player". However, it was decided that "Star Player" is most appropriate, as it has the potential to elicit envious feelings, while being relatively neutral (in comparison with the other titles).

3.4 Questionnaire

After subjects in the status condition had nished the three parts laid down in Table 3.1, they were asked to ll out a debrieng questionnaire about their feelings and attitudes towards the Star Player, disregarding whether they had been Star Player themselves or not. Answers were based on a 7-point scale, varying from 1 ("not at all") to 7 ("very much so"). For each question regarding the Star Player status, the average score was calculated based on the scores given by the three non-SP group members. In order to check the validity of the social ties measurement, subjects were asked to what extent they would want to continue interacting with the Star Player in the future (varying from "not at all" to "very much so"). Simultaneously, this also allowed to test the hypothesis by Hardy and van Vught (2006) that "nice guys nish rst", i.e. high contributors are expected to benet in the long from their altruistic behavior in a way that they are more likely to be chosen as future interaction partners by other group members. Participants in the control condition did not receive a questionnaire.

Chapter 4

Results

This chapter provides a behavioral analysis based on the data obtained from the experiment. First, Section 4.1 discusses the measurement of social ties and presents an overview of the general results obtained with respect to the social ties development in both treatments. Sec-ond, the contribution levels for both treatment are compared in Section 4.2. This section also provides an analysis of the inuence of the Star Player status on the dynamics of the contribution levels. Third, Section 4.3 digs further into the development and impact of social ties, including a discussion on the eect of status competition within this framework.

4.1 Attitudes and social ties

The rst and second circle-test, from now on referred to as the pre- and post-test, were used to measure the social value orientation of an individual and his or her attitude towards specic group members respectively. Combining the results obtained from these two tests allowed the social tie to be measured, which is dened as the dierence between the angles measured by post- and pre-test, i.e. Tie = Anglepost− Anglepre. Moreover, to facilitate the analysis of

social ties development, the social value orientations (pre-test), attitudes (post-test) and social ties were categorized into three segments: "Positive, "Neutral" and "Negative". In accordance with van Dijk et al. (2002) and Sonnemans et al. (2006), the categorization was based on angle intervals of the tests. Angles between −5◦ _{and 5}◦ _{fell into segment "Neutral", while all}

other positive (negative) angles outside that interval were dened as "Positive" ("Negative"). For the remainder of this chapter, the control condition and status condition will be referred to as CC and SC respectively.

Table 4.1 presents the descriptive statistics on the results of the pre- and post-test in the two conditions. The means of the angles measured in the post- and pre-tests do not signicantly dier in both treatments (Wilcoxon, two-tailed, P=0.562 (CC); P=0.851 (SC)). However, the

CHAPTER 4. RESULTS 24

CC SC Pre Post Pre Post Mean 12.6 9.4 4.4 5.6 Median 4.5 5.5 0 0.7 Max 45 39 45 45 Min 0 -13.3 0 -30 Std. dev. 14.6 13.2 10.2 14.9

Table 4.1: Descriptive statistics on the angles measured by the pre- and post-tests in the control group (CC) and status group (SC) .

conditions show opposite drifts: while the angle for CC decreases by approximately 3◦_{, SC}

shows an increase of 1◦ _{between pre- and post-test. Looking at the standard deviations, the}

two conditions show again opposite patterns: CC shows a small decrease from 14.6 to 13.2, while the standard deviation increases from 10.2 to 14.9 in SC. This signicant increase in SC might be explained by subjects substantially dierentiating between group members in the post-test. When testing for dierences in the means between the two treatments, both the pre- (Welch's t-test, P=0.03) and post-test angles (t-test, P=0.07) are signicantly dierent (using a signicance level of 10%). Hence, subjects in the control condition seem to have a generally more positive attitude towards an unknown generalized other. On the other hand, the attitude (post-test) deteriorates with respect to specic group members.

Among the 48 subjects that participated, 21 chose an angle of approximately 0◦ _(i.e.

between −5◦ _{and 5}◦_{) in both the pre- and post test: 8 out of 24 for CC, and 13 out of 24}

for SC. Hence, the fraction of individuals consistently exhibiting purely selsh behavior is, especially in the status condition, relatively large. This is unfortunate, as it impedes the development of social ties. Moreover, all the observations fell within the interval [−45◦_{, 45}◦_],

which indicates that subjects never attached a weight larger than one to the generalized other's utility.

Furthermore, Tables 4.2 and 4.3 present an overview of the composition of social value ori-entations, attitudes and social ties for the control (CC) and status (SC) condition respectively. Although there is already a substantial dierence in social value orientations between the two conditions (SC is more neutrally social oriented), the dierences between the development of social ties is quite striking: 32% of the social ties formed in CC is negative against only 8% in SC. There exist several reasons that could explain this nding. First, because subjects in the control condition had an initially more positive attitude towards generalized strangers (pre-test), their "benchmark" was already quite high, leading to negative social ties while they actually had positive experiences and attitudes towards their group members. This is generally in line with Sonnemans et al. (2006), who nd a similar mean angle for the pre-test

CHAPTER 4. RESULTS 25 Control condition

Positive Neutral Negative Observations Social value orientation 50% 50% 0% 24 Attitudes towards group members 37% 53% 10% 72 Social ties 21% 47% 32% 72

Table 4.2: An overview of the results on social value orientations, attitudes and social ties formed in the control condition. The variables are categorized as positive, neutral and negative, based on the angle measured by the respective test.

Status condition

Positive Neutral Negative Observations Social value orientation 17% 83% 0% 24 Attitudes towards group members 24% 72% 4% 72 Social ties 18% 74% 8% 72

Table 4.3: An overview of the results on social value orientations, attitudes and social ties formed in the status condition.

(13.4◦), together with 30% negative ties.

On the other hand, the presence of status competition could be responsible for the signi-cant dierence in the development of negative social ties. Although the formation of positive ties does signicantly dier between the two treatments (21% vs 18%), it could be that the competition for status leads to more group cohesion and an enhanced group performance, preventing the development of negative ties. A further discussion of these results is provided in Section 4.3. However, it needs be emphasized that these ndings may also be attributable to the fact that individuals in the status condition exhibited more selsh behavior in general, impeding the development of social ties.

4.2 Contributions in the public good game

This section provides further insights into the relation between the development of social ties and the contribution levels in a four-person public good game. Moreover, the eect of the competition for status on contributions is examined by analyzing the in-group dynamics of the contribution levels between the two conditions.

CHAPTER 4. RESULTS 26

4.2.1 General contribution results

First, Fig. 4.1 shows the average contributions to allocation "Public" per round in both treatments. The total average contribution over all rounds is 4.7 and 4.4 for CC and SC respectively. Hence, the overall average contribution does not signicantly dier between two conditions (Wilcoxon, two-tailed, P=0.543), which is in accordance with Fig. 4.1: the (purple) solid and (yellow) dashed line move very close to each other during all six rounds. This is surprising considering the signicant dierences in pre- and post-test means between the two conditions (see Table 4.1). One would expect greater pre-test angles to go hand in hand with higher contribution levels, this is however not the case.

Furthermore, the signicant dierence in pre-test angles between the two conditions is rather unfortunate, as this makes it dicult to isolate the eect of status competition on the contribution levels. Although the contribution levels are now relatively close to each other, this dierence might have been much bigger in case the pre-test angles of the two conditions were similar. Hence, the dierence in pre-test angles might obscure the potential eect of social status competition on the contribution levels. In order to control for this dierence, the average contribution in each treatment was scaled by the logarithm of the condition's average pre-test angle: C =b C

log(pre). Based on this scale, the average contribution level over all rounds

is 4.3 and 6.9 for CC and SC respectively. Thus, when the signicant dierence in pre-test angles is controlled for, the average contribution level in the status condition is much higher compared with the control group.

Also, it is remarkable that both contribution levels in Fig. 4.1 move in a very similar pattern across the various rounds, with the maximum and minimum reached in rounds 2 and 5 respectively. Also, subjects contribute consistently more than the single-stage Nash allocation, even those with a pre-test angle of 0◦_{. According to Keser and van Winden (2000)}

one explanation might be that some selsh subjects anticipate the presence of conditional cooperators who only contribute if a sucient number of other members will. Note that there is no end-eect in both treatments (Ledyard, 1995), due to the announcement that between 5 and 12 rounds would be played. Nevertheless, it seems that the inclusion of status competition in the public good game does not have any eect on the actual contribution levels when the dierence in pre-test angles is not controlled for. The next section will elaborate on this.

4.2.2 In-group contribution dynamics

At rst glance, the presence of status competition does not seem to aect the contribution levels. However, the results presented in Fig. 4.1 are aggregated and might conceal substantial between-group dierences in each condition. Therefore, this subsection provides an analysis of the inuence of the Star Player status on the dynamics of the contribution levels within specic groups. Herein, the following questions are relevant:

CHAPTER 4. RESULTS 27

Figure 4.1: Average individual contributions to allocation "Public" per round in the control condition (CC) and status condition (SC).

• Does the Star Player (SP) status aect the contribution levels in a public good dilemma? If so, how and to what extent?

• How do individuals with no status (i.e. non-SP) behave? Does the presence of an SP lead to constructive envy (increase in contributions) or destructive envy (decrease in contributions)?

However, it might be dicult to answer these questions, because the SP status might shift from player to player depending on their cumulative contributions over the three most recent rounds. For this reason, only the groups in which a single group member is assigned the SP status throughout the entire game will be considered in the analysis. The relevant question then becomes: How does the average contribution of the other, non-SP group members move relative to the contribution of the Star Player?

Luckily, three out of six groups in the status treatment oer the conditions (i.e. one unique Star Player) that allow the dynamics of the contribution levels of the non-SP players to be analyzed. Fig. 4.2 shows the patterns of the contribution levels for these three groups, which were denoted as Groups 1, 3 and 6 during the experiment. The levels for the SP and the non-SP players have been separated: the contribution level for the Star Player is depicted as a solid line, while the dashed line depicts the average contribution level of the three non-SP group members.

Fig. 4.2 shows the dynamics of the contribution levels are much alike among the three dierent groups. At the start of the game, there is a gap between the levels of the SP and the non-SP group members. However, as the rounds progress, contribution levels converge between rounds 3 and 5 and then diverge again at the end in round 6. Most importantly, the graphs show that the SP status does not lead to a destructive type of envy: non-SP players

CHAPTER 4. RESULTS 28

Figure 4.2: The contribution levels for groups 1,3 and 6 in the status condition with one unique Star Player throughout the full game. The solid line depict the contribution level of the Star Player, while the dashed line depict the average contribution level of the three remaining non-SP group members.

are more likely to increase their contribution towards the level of the SP rather than to reduce it. On the other hand, forms of constructive envy are also hard to detect, because the average contribution level of the non-SP players does not substantially increase (except for group 1). Hence, the claim by Eckel et al. (2010) that high-status player are more likely to be mimicked does not nd strong support in this case.

However, before being able to conclude that these dynamics are characterized by the pres-ence of status competition, it is important to check whether the same dynamics emerge when the competition for status is not present, i.e. absence of a Star Player. For the control treat-ment the same conditions apply, meaning there needs to be a single group member which contributes the largest amount consistently throughout the game. The essential dierence is that this group member is not highlighted as the Star Player here.

Similar to the status condition, the control condition oers three groups in which the best contributing player of the group is the same subject throughout the full game. Fig. 4.3 shows the patterns of the contribution levels for these three groups, which were denoted as Groups 1, 2 and 5 during the experiment. Again, the levels for the best contributor (BC) and the other players (Others) have been separated: the contribution level for BC is depicted as a solid line, while the dashed line depicts the average contribution level of the three other group members. In contrast to Fig. 4.2, the patterns in Fig. 4.3 dier quite substantially from each other. But more importantly, they dier signicantly from the patterns shown in Fig. 4.2. In the status condition (SC), contributions converged to the same level between round 3 and 5. However, Fig. 4.3 shows that hardly any convergence of contribution levels takes place within the groups. The levels in group 1 converges to some degree, and the levels in group 5 follow approximately the same dynamics. Still, the dierences in contribution levels between the best contributor (BC) and the Others are very large, especially in groups 2 and 5.

So it appears that the (relatively) high contribution level of BC is not as salient as in the status condition, where this player receives the Star Player status and is ranked as rst

CHAPTER 4. RESULTS 29

Figure 4.3: The contribution levels for groups 1,2 and 5 in the control condition with a single best contributor throughout the full game. BC indicates the best contributor (solid line).

in the feedback table. Hence, the presence of status and a rank-ordered system seem to substantially aect the dynamics of the contribution levels within groups. More specically, contribution levels of high and low contributors are more likely to converge in the presence of status competition, potentially leading to a better group performance. An overview of all individual contribution levels per group in each condition are included in Appendix B.

4.3 Status competition and social ties

Tables 4.2 and 4.3 have already given a rst peek of the ndings regarding the development of social ties structures as a result of interaction in a public good game. This section provides a more granular analysis of the results, including a number of regressions and tests for corre-lations in order to check whether to the hypotheses presented in Subsection 2.2.3 really hold. However, it should be emphasized that some of the results should be interpreted with caution, due to the limited amount of observations on which they are based.

4.3.1 Social ties formation

Before testing the hypothesis on the eect of status competition on social ties structures, it is necessary to check whether social ties are formed in the rst place as a result of interaction in the public good dilemma. A rst step is to check whether there exists a correlation be-tween an individual's average contribution and the level of appreciation this individual receives by his/her group members. One would expect that higher contribution levels would lead to more gratitude by the other group members. In order to test this, the level of appreciation of individual i is calculated as the average attitude (i.e. post-test angle) of the three other group members towards individual i. Indeed, this average attitude, from now on referred to as the evaluation score, correlates positively and signicantly with the average contribution