Essays on Leadership: Incentives, Legitimacy, and Goal Setting
Von der Fakultät für Wirtschaftswissenschaften der Rheinisch-Westfälischen Technischen Hochschule Aachen zur Erlangung des akademischen Grades eines Doktors der Wirtschafts- und Sozialwissenschaften genehmigte Dissertation
vorgelegt von
Martin Scheuermann, M.Sc.
Berichter: Univ.-Prof. Dr. rer. pol. Christine Harbring Univ.-Prof. Dr. rer. pol. Christian Grund
Tag der mündlichen Prüfung: 28.05.2020
Diese Dissertation ist auf den Internetseiten der Universitätsbibliothek online verfügbar.
Der Abschnitt „Part I: Leadership in combination with individual rewards and punishments“ ist veröffentlicht unter:
Gürerk, Özgür; Lauer, Thomas; Scheuermann, Martin (2018). Leadership with individual rewards and punishments. Journal of Behavioral and Experimental Economics 74, 57–69. DOI: 10.1016/j.socec.2018.03.007.
The true method of knowledge is experiment.
(William Blake, 1757 - 1827)
i
ACKNOWLEDGMENTS
During my time as a PhD student I had not only one, but two supervisors to whom I am very grateful. For most of the time I was supervised by Özgür Gürerk, who guided me through my first two research projects and enabled me to publish my first scientific paper. Özgür, thank you very much for your patience and support. When I needed a new supervisor, I was glad that Christine Harbring agreed to step into the role for the remainder of the time. Thank you, Christine, your supporting feedback and encouragement in the final phase really helped me a lot.
Special thanks go to Thomas Lauer for collaborating on my first two projects and helping me with his profound expertise in econometrics. My thanks also go to Christian Grund, who made himself available as a reviewer for this dissertation.
As a research associate, I had the pleasure of being surrounded by wonderful colleagues. I am thankful to Andreas Staffeldt, Britta Butz, Frederik Graff, Jan Wilhelm, Luca Carduck- Eick, Lucas Braun, Maik Theelen, and Nicolas Meier, for creating such a pleasant atmosphere and many joyful memories.
Regarding Part III of this dissertation, I express my gratitude to Jan Wilhelm, who initially approached me with the underlying idea. Although it was intended as a joint project, Jan left research to pursue other goals, so I wrote the paper on my own. Jan, I hope you are satisfied with the result.
I further thank our secretary Anja Uttich who assisted in many ways and Christine Stibbe for transforming my English into a comprehensible form.
Finally, I would like to thank my friends and family on whose support I could always rely.
ii
TABLE OF CONTENTS
Acknowledgements ...... i
Table of Contents...... ii
Index of Tables ...... v
Index of Figures ...... vii
Index of Abbreviations ...... ix
INTRODUCTION ...... 1
1 Project Overview ...... 7 1.1 Part I - Leadership in combination with individual rewards and punishments ...... 7 1.2 Part II - Legitimacy in the lab: The effects of leader selection and information on team performance ...... 9 1.3 Part III - Goal setting and team performance ...... 11 1.4 Discussion and outlook ...... 13
PART I: LEADERSHIP IN COMBINATION WITH INDIVIDUAL REWARDS AND
PUNISHMENTS ...... 16
2 Introduction ...... 16
3 Related Literature ...... 18 3.1 The mechanics of leading-by-example ...... 18 3.2 Leading-by-example experiments without rewards or punishments .. 19 3.3 Experiments with centralized rewards or punishments in simultaneous settings ...... 20 3.4 Experiments with centralized rewards or punishments in sequential settings ...... 21
4 Experimental Design ...... 22 4.1 Treatments ...... 22 4.2 The game ...... 22 4.3 The procedures ...... 24
Table of Contents iii
5 Hypotheses ...... 24
6 Results ...... 25 6.1 The effect of leadership on contributions in the absence of rewards or punishments ...... 25 6.2 The effect of rewards and punishments on contributions in the absence of leaders ...... 26 6.3 The effect of leaders on contributions in the presence of rewards or punishments ...... 27 6.4 Evolution of contributions and individual contribution behavior ...... 28 6.5 The use of rewards and punishments ...... 31 6.6 Payoffs ...... 37
7 Discussion and Conclusion ...... 39
Appendix A...... 41
PART II: LEGITIMACY IN THE LAB: THE EFFECTS OF LEADER SELECTION AND
INFORMATION ON TEAM PERFORMANCE ...... 47
8 Introduction ...... 47
9 Related Literature ...... 49 9.1 Leader appointment and cooperation in experiments ...... 49 9.2 Information and cooperation in experiments ...... 51 9.3 Theoretical considerations and hypotheses ...... 51
10 The Game and Experimental Design ...... 53 10.1 Leader appointment ...... 54 10.2 Procedures...... 55
11 Results ...... 55 11.1 The effect of endogenous leader election ...... 57 11.2 The effect of information about the leader’s past behavior ...... 58 11.3 The combined effect of both sources of legitimacy ...... 59 11.4 Leader and follower behavior ...... 60 11.5 Ex-post legitimization ...... 64
12 Discussion ...... 65
13 Conclusion ...... 67
Table of Contents iv
Appendix B ...... 68
PART III: GOAL SETTING AND TEAM PERFORMANCE ...... 72
14 Introduction ...... 72
15 Related Literature ...... 73 15.1 Goal setting theory ...... 73 15.2 Meta-analyses ...... 74 15.3 Experimental studies ...... 75
16 Experimental Design and Procedures ...... 77 16.1 The game ...... 77 16.2 The treatments ...... 79 16.3 Procedures...... 79
17 Hypotheses ...... 80
18 Results ...... 83 18.1 Average contributions ...... 83 18.2 Individual contributions ...... 85 18.3 Voting behavior and success rates ...... 86 18.4 Beliefs about other agents’ contributions ...... 87
19 Follow-up Study ...... 88 19.1 Experimental design and procedures ...... 88 19.2 Results...... 89
20 Discussion ...... 94
21 Conclusion ...... 97
Appendix C ...... 98
22 Bibliography ...... 105
v
INDEX OF TABLES
Table 1: PGG parameters of the three parts...... 5
Table 2: Treatment averages over all periods. Standard deviations in parentheses...... 26
Table 3: Treatment differences - the effects of leadership, punishment, and rewards...... 27
Table 4: Determinants of contributions (all periods)...... 30
Table 5: Determinants of punishment and reward decisions (all periods)...... 34
Table 6: Contribution change after receiving reward or punishment...... 36
Table 7: Treatment overview...... 54
Table 8: Average contributions. Standard deviations in parentheses...... 55
Table 9a-c: Average contributions and differences with and without endogenous leader selection for the pooled sample (9a) and separate sub-samples (9b, 9c). Standard deviations in parentheses...... 57
Table 10a-c: Average contributions and differences with and without information about leader behavior for the pooled sample (10a) and separate sub- samples (10b, 10c). Standard deviations in parentheses...... 58
Table 11: Average contributions and differences with one or two sources of legitimacy. Standard deviations in parentheses...... 60
Table 12: Regression analysis: (I) treatment effects on contribution difference between leaders and followers, (II) and (III) structural equation models for contributions in Phase 2...... 62
Table 13: Regression analysis: (I) reelection effect on contribution difference between leaders and followers and (II) structural equation models for contributions in Phase 3...... 71
Table 14: Project choices and success rates. Absolute frequencies in parentheses...... 86
Table 15: Average beliefs and contributions...... 87
Table 16: Average contributions to and frequencies of P1 and P2...... 91
Index of Tables vi
Table 17: Determinants of contributions...... 92
Table 18: Average payoffs for agents and principals...... 93
vii
INDEX OF FIGURES
Figure 1: Course of the experiment in Part II...... 10
Figure 2: Goal choices in the experiments in Part III...... 12
Figure 3: Average contributions...... 29
Figure 4: Average received reward or punishment...... 32
Figure 5: Average contributions and the sum of allocated reward points within single groups of the L-REW treatment...... 44
Figure 6: Average contributions and the sum of allocated reward points within single groups of the P-REW treatment...... 44
Figure 7: Average contributions and the sum of allocated punishment points within single groups of the L-PUN treatment...... 45
Figure 8: Average contributions and the sum of allocated punishment points within single groups of the P-PUN treatment...... 45
Figure 9: Average contributions and average allocated rewards and punishments...... 46
Figure 10: Average relative change in leaders’ and followers’ contributions from Phase 1 to Phase 2, attributed to treatment effects...... 61
Figure 11: An agent’s best response, depending on the other agents’ contributions...... 80
Figure 12: Average contribution decisions in AV and PD...... 83
Figure 13: Proportion of contribution amounts for the projects in AV (left) and PD (right)...... 85
Figure 14: Average contribution (line chart, left scale) and P2 ratio (bar chart, right scale) for AP...... 89
Figure 15: Average contribution (line chart, left scale) and P2 ratio (bar chart, right scale) for PA...... 89
Figure 16: Scenarios for an agent’s payoff in P1, depending on her contribution and the other agents’ contributions...... 103
Index of Figures viii
Figure 17: Scenarios for an agent’s payoff in P2.L, depending on her contribution and the other agents’ contributions...... 104
Figure 18: Scenarios for an agent’s payoff in P2.H, depending on her contribution and the other agents’ contributions...... 104
ix
INDEX OF ABBREVIATIONS
BPT Binomial probability test
DID Difference in differences
DV Dependant variable e.g. Exempli gratia (for example) et al. Et alii (and others)
GST Goal setting theory i.e. Id est (that is)
MWU Mann-Whitney-U test
OLS Ordinary least squares
PGG Public good game
ORSEE Online Recruitment System for Economic Experiments
RWTH Rheinisch-Westfälische Technische Hochschule
SD Standard deviation
SOEP Sozio-ökonomisches Panel (German socio-economic panel) vs. Versus
WMP Wilcoxon-matched-pairs test
1
INTRODUCTION
Leadership is a core element of human interaction. Similarities between the leadership behavior of humans and animals suggest that the concept of leadership is deeply rooted in the evolutionary development of humankind (King et al. 2009). Today, leaders are present in almost any area of our life, e.g., in politics, religion, and organizations. The importance of leadership research is emphasized in the literature, where leadership is described as “permeating all aspects of human social affairs” (King et al. 2009, p. 911) and “perhaps the single most important issue in the human sciences.” (Hogan and Kaiser 2005, p. 169)
In this dissertation, I investigate leadership in the context of a social dilemma using economic experiments. What exactly does leadership mean, and why is it important? Exemplary for the many definitions of leadership, John P. Kotter defined it as follows: “Leadership defines what the future should look like, aligns people with that vision, and inspires them to make it happen despite the obstacles.” (Kotter 2012, p. 28)
The importance of leadership in the organizational context, which builds the framework for this dissertation, has often been scientifically examined. Empirical studies found evidence that CEOs could explain some of the variances in firm performance (e.g., Barrick et al. 1991, Joyce et al. 2003) and that their influence became more significant over the last decades (Quigley and Hambrick 2015). This “CEO effect” could be even higher than previously thought, as it may have been underestimated in the past (Mackey 2008, Hambrick and Quigley 2014).
As a scenario to investigate the effects of leadership, I use the so-called social dilemma. A social dilemma is characterized by the fact that individual and collective interests are at odds. More specifically, an individual can maximize her (often short-term) benefit by not cooperating with her group members, whereas the group as a whole would be better off if every individual would cooperate (Dawes 1980). As an illustration, consider the use of a limited, regenerative resource, e.g., forests or fishing areas. If all those involved adhere to a certain limit on use, the resource can recover and continue to benefit all in the future. If individuals do not adhere to the limit and consume more, their short-term benefit (or profit) will be higher. As the resource will no longer recover fully, however, less of it will be available to everyone in the future. Other examples of social dilemmas include, e.g., environmental pollution, an arms race between nations, or cooperation in work teams.
Introduction 2
In Part I, we examine if a leader can improve cooperation in a social dilemma, either just by setting a good example, or in combination with rewards or punishments at her disposal. In Part II, we test how much groups cooperate when leaders are appointed with different degrees of legitimacy. Finally, in Part III, we investigate if the cooperation of group members with a self-chosen goal differs from their cooperation when the goal is chosen by their principal and if the goal level is decisive for cooperative behavior.
The remainder of this introduction is structured as follows. After a short note on leadership theory, I introduce the research method of experimental economics and its frequently used terms. Then I explain the public good game, which we used in all experiments. Following these introductions, chapter 1 provides an overview of the three projects of this thesis.
Leadership theory
Leadership has gained popularity as a topic for scientific research. Similar to the increase in empirical and experimental leadership studies (Gardner et al. 2010), numerous papers on leadership theory have been published. To get an impression of the extensive amount of leadership theories, one may refer to the review of Dinh et al. (2014). Dinh and colleagues analyze the articles from 10 top-tier journals from the years 2000 to 2012 and find 66 different types of leadership theory for this period alone.
An overview of what different leadership theories contain and to what extent they are related to each other is given in the article by Hernandez et al. (2011). In their approach, the theory types are classified along two dimensions: the locus of leadership (leader, context, followers, collectives, dyads) and the mechanism of leadership (traits, behaviors, cognition, affect). The classification of the different theories in the resulting 5x4 grid offers an orientation within the leadership theories.
Due to the high number of different approaches, a comprehensive presentation of leadership theory in this introduction is not feasible. Furthermore, the aim of this dissertation is by no means to provide a complete overview of leadership research, but rather a detailed description of incremental contributions to this field. However, in the three parts of this thesis, relevant theoretical work is covered in the respective Related Literature section.
Introduction 3
Experimental economics
In economics, experiments are used to validate or develop economic theories, find solutions to economic problems or test possible effects of laws and regulations before implementation, e.g., taxing rules (Torgler 2002), auction forms (Cummings et al. 2004), or even health-related issues such as vaccination recommendations (Böhm et al. 2017).
While the first economic experiments were conducted as early as the middle of the last century (Chamberlin 1948), this research method remained relatively neglected for a long time. It was not until the 1990s that this form of economic research became more popular and was considered an established method by 2002 when Vernon Smith received the Nobel Prize “for having established laboratory experiments as a tool in empirical economic analysis.”1
In general, economic experiments can be classified as either lab or field experiments. In a lab experiment, the participants are invited to a specially equipped room, usually at a university, where they enter their decisions on a computer or use pen and paper. In a field experiment, on the other hand, participants are unaware of their participation while their behavior is observed in a natural environment, e.g., to test different prizing schemes (Gneezy et al. 2010).
Recently, a mixture of lab and field experiments has become established as well, the so- called lab-in-the-field experiment, or extra-laboratory experiment (Charness et al. 2013). For this variant, the controlled setup of a lab experiment is used, e.g., to experiment with a specific pool of participants (Baldassarri and Grossman 2011, Jack and Recalde 2015, Grieco et al. 2016), or to investigate whether participants behave differently when the stakes are relatively high (Ariely et al. 2009, Andersen et al. 2011).
One major advantage of lab experiments is the high degree of experimental control. If only one aspect is different between two experimental conditions (treatments), differences in observed behavior can be directly attributed to this aspect. Thus, causal relations can be derived, which is not possible in field experiments. For instance, Jack and Recalde (2015) find in their experiment that a leaders’ observable characteristics (e.g., gender, education, wealth) do have an impact on the followers’ behavior. In a lab experiment, where subjects
1 https://www.nobelprize.org/prizes/economic-sciences/2002/smith/facts/, retrieved on January 16, 2020.
Introduction 4 interact anonymously, these characteristics can be excluded as a potential influence on behavior.
In general, the different types of experiments have specific advantages and disadvantages and should be used complementary. For example, field experiments can be used to check whether findings from the laboratory are also valid in a specific context. Conversely, observed behavior in field experiments, of which the exact cause may be unknown, can be investigated in the laboratory for potential reasons.
The course of an experiment
In the following, I describe the typical procedure of a lab experiment, as it was carried out for the three studies of this dissertation and explain some of the frequently used terms. Before the experiment starts, each of the participants, who are also called players or subjects, is randomly assigned a place in the laboratory. At the beginning of the experiment, the participants receive instructions for the task at hand, either by the experimenter or by an explanatory text on the screen. Then the participants start working on the decision task, often referred to as game. Participants can neither observe the screens of other participants, nor do they know with which of the other participants they interact.
If a game is played in groups for more than one round, players are either kept in a group with the same group members (partner matching) or assigned to a different group (stranger matching) for the following round. After all tasks have been completed, and questionnaires filled out, participants receive payment for their participation in the experiment, the payoff, which can usually be influenced by their own decisions, the decisions of other participants, and by chance.
Public Good Game
The Public Good Game (PGG) is a game frequently used in experimental economics to investigate behavior in a social dilemma. The exact versions of the PGG used for the studies of this dissertation are described in the respective Experimental Design sections. Among others, Zelmer (2003) and Chaudhuri (2011) provide an overview of various research results obtained with PGG experiments. In the following, I present the basics of the game as well as possible modifications.
Introduction 5
In a common form of the PGG, players are assigned to groups. Each group member receives an endowment and needs to decide if they want to contribute to the public good. The sum of contributions of all players is then increased by a factor known to the players. Finally, the resulting amount of multiplied contributions is equally distributed among all group members, independent of their individual contribution size. Players may also refrain from contributing, or only contribute a part of their endowment and keep the non-contributed endowment as personal profit.
As a tool for social dilemma research, the PGG is very versatile due to many possible modifications. For instance, one could change the group size, provide players with unequal endowments, or implement a specific payout restriction. The game can be played repeatedly for a certain number of rounds, or just a single time (one-shot game), and players may make their contribution decision either simultaneously or sequentially. Additional elements, such as punishments or rewards, which either de- or increase a players’ payoff, or any form of interaction between the group members are further options to adjust the PGG for specific research questions, resulting in countless potential variations of the game. Table 1 gives an overview of how some of these parameters differ between the studies of this dissertation.
Table 1: PGG parameters of the three parts.
2 Part I Part II Part III L-treatments P-treatments First study Follow-up Group One leader, Four equal One leader, One principal, composition three followers group members three followers three agents Contribution Sequential Simultaneous Sequential Simultaneous mode
Duration 20 rounds 3x10 rounds One-shot 2x5 rounds
Additional Different forms of Punishments and rewards Contribution goals elements leader appointment
2 In Part II, leaders are appointed after ten rounds. For the first ten rounds, groups consist of four equal members who make simultaneous contribution decisions.
Introduction 6
In Part I and Part II, leadership in the PGG is implemented by using a leading-by- example mechanism. For this, one of the group members (the leader) makes the contribution decision before the others. The remaining members of the group (the followers) then can observe the leader’s decision before deciding on their contributions. Leadership in Part III is carried out by the principal who chooses a project for the agents (see section 16.1 for a description of the study design).
Although the players are described as leaders and followers (Part I and II), or principals and agents (Part III), these terms were never used in the experiments to avoid framing effects (Cookson 2000). Instead, in the instructions and on-screen, the different player types are just described as type A and type B.
Introduction 7
1 Project Overview
This chapter provides a summary of the three projects to give the reader an overview of the scope of this dissertation. Sections 1.1 to 1.3 each describe one project and are structured as follows. At first, I describe the research question and results from the related literature. This is followed by a brief description of the experiment and a presentation of the main results. Finally, section 1.4 discusses some of the findings, the scientific contributions of the individual studies, and possibilities for further research.
1.1 Part I - Leadership in combination with individual rewards and punishments
Regarding the question of how cooperation in a social dilemma could be improved, various possible solutions have already been tested in lab experiments. In this context, the cooperation-enhancing effect of rewards and punishments represents a robust result, with the latter having a stronger impact than the former (Balliet et al. 2011, Milinski and Rockenbach 2012). The introduction of leadership via leading-by-example is also a frequently investigated approach, although in this case, the results are less conclusive: while some studies report a positive effect of leading-by-example (Güth et al. 2007, Levati et al. 2007, Pogrebna et al. 2011), others do not (Sturm and Weimann 2007, Haigner and Wakolbinger 2010, Sahin et al. 2015, Gächter and Renner 2018).
Although both elements, the use of incentives, as well as leading-by-example, have been extensively researched, almost no combined analysis of these factors has been conducted. This is even more surprising as leaders in the real world usually can use incentives towards their followers. For example, a manager may offer rewards by promoting an employee or granting a bonus payment and punish by assigning unpleasant tasks, denying a vacation request, or even laying off an employee.
We are aware of only two studies investigating the combined use of leadership and incentives in a PGG. In Güth et al. (2007), leaders can punish a team member by excluding her from the team for the next round, therefore also from the team output. If leaders have this possibility, contributions are significantly higher than when leaders do not have this option. Next to this punishment by exclusion, Sutter and Rivas (2014) also implement a
Introduction 8 treatment where a randomly determined leader may assign a fixed reward to one of the group members. Contributions in these groups are higher compared to a control treatment with neither leadership nor rewards.
To test whether leadership or the use of incentives is decisive for improvement of cooperative behavior and if a combination of the two might affect contributions even more, the first part of this dissertation offers a systematic analysis of these two factors. For this, we vary if one of the group members leads by example, i.e., contributes first, and the availability of incentives (rewards, punishment, or no incentives), yielding a 2x3 experimental design. Unlike in Güth et al. (2007), leaders in our setting cannot exclude group members but instead may reduce their payoff by assigning punishment points (see section 4.2 for a detailed description of the game). Further crucial differences, also in comparison with Sutter and Rivas (2014), are that leaders can punish or reward not only one, but any number of team members (including none) and that the extent of the reward or punishment is not fixed, but is at the discretion of the leader.
Our results confirm the positive effect of rewards or punishments on cooperation. Regarding the presence of a leader, however, we could not find such an effect. The lacking leader effect coincides with the results of the studies which find either little (e.g., Levati et al. 2007) or no positive leader effect (e.g., Gächter and Renner 2018). Hence, leading-by- example per se does not seem to be a promising means of promoting cooperation in a social dilemma. Although leadership does not improve group outcomes in this particular setting, we nevertheless derive a few interesting insights regarding the cooperative behavior in groups with a leader, as described in section 1.4 and chapter 6.
Introduction 9
1.2 Part II - Legitimacy in the lab: The effects of leader selection and information on team performance
In the second part, we investigate how a leader’s legitimacy influences cooperative behavior in a team. Remember that in Part I, we find that leadership per se does not improve cooperation in a social dilemma. Neither the example set by the leader nor the additional use of rewards or punishments results in higher contributions compared to a setting without a leader. Previous research already yielded mixed results regarding the effectiveness of leading-by-example (see chapter 3). Hence, in Part II, we want to deepen our understanding of how leadership can improve group outcomes in a social dilemma.
Looking at real-life examples, it becomes evident that the determination of a leader is usually linked to at least one criterion: a king is the son of a former king (or queen), a manager is hired based on past performance, and a political leader is appointed if she receives the most votes. Linking the selection of the leader to specific criteria grants legitimacy to the leader, improving the odds that her role is accepted and her example followed (Hollander 1992, van Vugt and Cremer 1999, Brandts et al. 2014). A possible reason for the lack of the leaders’ impact in Part I could thus be the way they were chosen: since leaders were determined by chance, they lacked any form of legitimacy, and therefore, the other group members may not have felt obliged to follow their leader’s example.
So how can we bestow legitimacy upon a leader in a lab experiment? A common finding from the literature discussed in chapter 9 is that the way a leader is selected is crucial for her legitimacy. We implement different degrees of legitimacy by using different ways of leader determination. For this, we vary if a leader is appointed endogenously or exogenously and if the information on past cooperation is available.
In our experiment, subjects first play a PGG for ten rounds without a leader (Phase 1), which gives us a baseline level of cooperation in the absence of leadership. Before the next ten rounds (Phase 2) are played, a leader is determined in one of four different ways, depending on the treatment (see Table 7 for a treatment overview). After round 20, in all treatments, the leader for the final ten rounds (Phase 3) is selected via majority vote. Figure 1 shows the outline of the experiment; a detailed description can be found in chapter 10.
Introduction 10
Figure 1: Course of the experiment in Part II.
Implementing a leader for Phase 2 results in significantly higher contributions compared to the first ten rounds without a leader. Furthermore, we find the following differences between our treatments. If the information on a leader’s past cooperation is available, groups that choose their leader endogenously increase their contributions more than groups with exogenously appointed leaders. Similarly, groups that select their leader endogenously increase their contributions more if they have information on the leader’s past cooperation at hand, compared to when this information is not available. Overall, groups increase their contributions more if their leader derives her legitimacy from both criteria (endogenous choice and availability of information) than with a leader legitimized by only a single factor.
Separate consideration of leader and follower contributions shows that a higher legitimacy of the leader does not directly affect the contributions of the followers. Regardless of a leader’s legitimacy, followers contribute slightly less than leaders. However, leaders with high legitimacy set a particularly good example (i.e., contribute higher than leaders with less legitimacy), and followers adapt to this high contribution level. Therefore, a higher level of leader legitimacy induces a better result for the whole group.
Introduction 11
1.3 Part III - Goal setting and team performance
The final part of this dissertation deals with the effects of goal setting on team performance. More specifically, we investigate whether it matters who sets a goal and to what extent the level of the goal has an influence on cooperative behavior. While the studies of Part I and Part II use the same mechanism to implement leadership in a PGG (leading-by- example), a principal-agent setting is used in Part III, where the principal can exercise leadership by selecting a project for the agents.
Our framework represents a highly simplified form of the principal-agent problem. In the principal-agent theory by Jensen and Meckling (1979), the scope of the problem is much broader and covers additional elements such as cost and utility functions, setting incentives, and asymmetric information, which are not considered here. We nevertheless use the terms agents and principals since the efforts of the former benefit the latter and principals can (depending on the treatment) determine the goal for the agents.
The effect of goal setting on performance has been the subject of numerous studies for more than 50 years. The results of these studies have already been processed in various meta- analyses for the effect of goal setting on both individual performance (Mento et al. 1987) and group outcomes (O'Leary-Kelly et al. 1994, Kleingeld et al. 2011). Based on the experimental results, a theory which deals with the relationship between goal setting and performance has been formulated and continuously revised: goal setting theory (Locke and Latham 1990, 2002, 2006, Latham and Locke 2007). The core message of goal setting theory is that the performance of individuals or groups can be increased by setting a goal. In this regard, high goals work better than low goals, especially if they are specific compared to unspecific goals.
In the experiments of this part, groups consist of three agents and one principal. The agents contribute to a project (the public good), which either has a goal or not. If a project has a goal, the sum of contributions to this project must be at least as high as the goal level for the project to be successful and provide a payoff – otherwise, contributions are lost. While agents can contribute to and receive a payoff from the projects, the principal also profits from contributions but cannot contribute herself.
Figure 2 shows the goal choices for both experiments in Part III. In our first study, subjects make two separate choices, once between having no goal or a low goal, and once between no goal and a high goal. Depending on the treatment, these choices are either made
Introduction 12 by the agents via majority vote or by the principal. After the project choice, agents decide on their contributions. In the follow-up study, the decision of whether the agents have either a high goal or no goal is made ten times in total. For five consecutive rounds each, the selection is made by the agents or the principal, respectively.
Figure 2: Goal choices in the experiments in Part III.
The main results in Part III are as follows. Agents contribute significantly more when they have a high goal compared to a goal-free setting. However, this only applies if the agents choose the goal themselves - if the principal decides on the goal, the agents' contribution behavior is exactly the opposite: agents contribute significantly less if the principal chooses a high goal for them, compared to when she chooses a low goal or no goal at all. Although agents contribute the most if they set themselves a high goal, they strongly prefer to have no goal at all. The principals, on the other hand, show no preference for either setting.
Regarding contributions and project preferences, the results of our follow-up study are similar to those of the first study in this part. Additionally, we find that in a repeated setting, contributions decline over time if the principal chooses the project for the agents. In contrast, if agents can vote on their project, contributions can remain stable over time, but only if the agents did not experience declining contributions under the goal setting by the principal before.
Introduction 13
1.4 Discussion and outlook
In this section, the following aspects of our studies are elaborated. In Part I, our systematic analysis of a leader’s influence, either with or without incentives at hand, yields no positive effect of a leader on contributions. We propose that leading-by-example on its own might not be a useful tool for increasing cooperation. Nevertheless, we do find positive aspects of having a leader. In Part II, we test different forms of leader selection to implement leader legitimacy in a lab experiment. Our mechanisms provide a leader for each group who also achieve higher levels of cooperation than without a leader. Finally, Part III contributes to the goal setting literature in the context of leadership, demonstrating the importance of who sets the goal for the group.
The benefits of having a leader
The difference between the treatments with and without a leader in Part I lies not only in the contribution mode (sequential or simultaneous) but also in the incentive structure. If only the leader is given the possibility to reward or punish, instead of giving this possibility to all group members, it is a centralized scheme and not a decentralized one as in the treatments without a leader. Neither in Part I, nor in previous studies (O'Gorman et al. 2009, Nosenzo and Sefton 2014, Fischer et al. 2016, Harrell and Simpson 2016) a negative effect of the centralized system on cooperative behavior could be observed. However, based on our findings, the use of incentives only by a leader does offer advantages.
On the one hand, leaders pronounce significantly less anti-social punishments (see section 6.5), which could have a demotivating effect on the group members and thus a negative effect on the willingness to cooperate. On the other hand, the punishments assigned by leaders cause a stronger, contribution-increasing reaction of the followers compared to peer punishment. Since punishment is associated with costs for both the punisher and the punished, centralization increases efficiency.
In general, a leader has perfect information about the distribution of the incentives, as they are only allocated by herself. In a decentralized system, the group members do not know whether someone else is already punishing (or rewarding) a group member, which can lead to too many punishments, or no punishments at all if everybody relies on someone else to execute them. Centralization, therefore, improves the coordination of incentives.
To conclude, even if a leader’s presence does not increase the willingness to cooperate, having a leader can nevertheless be beneficial in different ways. These advantages do not
Introduction 14 come at the expense of cooperation - neither in Part I nor in the studies mentioned above disadvantages regarding cooperation are identified in a centralized system compared to a decentralized one.
The (in-)effectiveness of randomly selected leaders
In Part I we find that leading-by-example does not help to improve cooperation in a social dilemma, at least not within the framework we have chosen. Since the leader is not determined by a specific criterion but by chance, we investigate in Part II if the lack of a leader’s legitimacy may be a possible reason for the missing leader effect.
In order to investigate the possible influence of leader legitimacy, we determine the leaders in Part II in various ways. We observe that the determination of a leader causes a (weakly) significant increase in contributions in all treatments, even if the leader is randomly selected. However, why do we observe a positive effect of a randomly chosen leader in Part II, whereas this is not the case in Part I?
Even if a final answer cannot be given at this point, it seems to be of relevance that there is a crucial difference between the experimental designs of Part I and Part II. In Part I, right at the beginning of the game, a leader is determined for 20 rounds, and the contributions decrease over time. In Part II, however, the groups initially play ten rounds without a leader and experience diminishing cooperation. Against this background, it seems conceivable that the leader, as well as the followers, want to reach a higher level of cooperation in the following ten rounds in order to achieve better results than in the first ten rounds. Hence, even groups with randomly chosen leaders contribute higher than they did without a leader. For application in an organizational context, this result suggests that it could be useful, if one has a team with low cooperation, to determine one team member as the leader to improve cooperative behavior within the team.
How to choose a leader
Our methods of leader determination in Part II ensure the presence of a leader in each group. The types of leader selection used in the literature, however, often lack this feature. This aspect seems to be almost ignored or at least neglected in the general discussion. Although numerous studies (e.g., Güth et al. 2007, Levati et al. 2007, Haigner and Wakolbinger 2010, Dannenberg 2015a, Cappelen et al. 2015) report that the level of cooperation increases after a leader has been determined, this only applies to those groups in which a leader is present. However, the proportion of groups that actually have a leader
Introduction 15 and thus increase their cooperation is usually quite low and can be as little as 13% (Dannenberg 2015a), while the majority of groups remain without a leader and show no improvement of cooperation. Our leader selection methods in Part II not only determine a leader for each group but also result in higher contributions in the following rounds. Therefore, Part II contributes to an understanding of how to choose a leader in these kinds of experiments.
Goal setting as a leadership tool
The results from Part III reveal an interesting facet of goal setting in the context of leadership. On the one hand, we replicate the performance-enhancing effect of high specific goals, as described by goal setting theory. On the other hand, this effect is reversed if the goal is not determined by the agents but by the principal: compared to a goal-free situation, the agents contribute significantly less if the principal chooses a high specific goal for them.
As discussed in chapter 20, there are various possible reasons for the low contributions when the principal chooses a high goal. One of them is that the agents could feel restricted in their decision-making and regard the high target of the principal as a sign of mistrust. The principal thus experiences hidden cost of control (Falk and Kosfeld 2006). If, on the other hand, the principal chooses a goal-free setting for the agents, the agents may perceive this decision as a sign of trust, which they reward with high contributions. Interestingly, these two effects seem to cancel each other out when the principal chooses the low goal; agents then contribute a very similar amount compared to when they choose this goal themselves.
Unlike in Part I and Part II, leaders (i.e., principals) in Part III neither can use punishments or rewards nor contribute to the public good and, therefore, also cannot lead by example. In this respect, an implementation of these elements to the experiments of Part III could provide useful insights. Would the selection of a high goal by the principal perhaps lead to high contributions if she had incentives at her disposal? Would the agents be more willing to cooperate if the principal could also contribute and set a good example? Moreover, how would the agents behave if the principal was determined with a certain degree of legitimacy?
16
PART I: LEADERSHIP IN COMBINATION WITH
INDIVIDUAL REWARDS AND PUNISHMENTS
An adapted version has been published in the Journal of Behavioral and Experimental Economics (Gürerk et al. 2018). Co-authored with Özgür Gürerk & Thomas Lauer.
Abstract: Leading-by-example is considered an important means of influencing followers. In most organizations, however, to influence their followers, leaders use a variety of instruments. Most frequently, leaders possess the power to administer rewards or punishments to team members. But do individual rewards or punishments reinforce the impact of leading-by-example on team members’ contributions? Because of confounding factors, it is difficult to research leading-by-example when using field data. Here, we investigate the effects of leading-by-example and the effect of rewards or punishments on contributions in controlled lab experiments. We find that both rewards and punishments are more effective in fostering contributions than leading-by-example. When leading-by- example involves reward or punishment power, it does not improve the effects of rewards or punishments as such for increasing contributions.
2 Introduction
Leaders may induce cooperative behavior by influencing their followers and coordinating them on efficient outcomes (Foss 2001). According to conventional wisdom, one possibility of exerting such influence is through leading-by-example. If leaders exert high efforts first, their followers can be expected to mimic them. The commander of a platoon, the captain of a football team, the head of a political party: all are expected to engage themselves in a particular way and to serve as role models for the motivation of soldiers, fellow teammates, or supporters, respectively.
In addition to leading-by-example, leaders usually use a variety of other instruments simultaneously. Often, leaders are equipped with the power to reward or punish their subordinates individually. Rewards may be monetary, such as bonus payments or a salary increase, or they may be non-monetary, like the provision of a better work environment, or
Part I 17 a representative car. On the other hand, leaders can exert individual punishments, such as ending a job contract, assigning the employee less preferable tasks, or withholding expected bonuses. In the military, leaders may even use disciplinary “non-judicial punishment” and send their subordinates to military prison for several weeks.
Given that leaders can often administer individual rewards or punishments to their followers, it is surprising that no experimental study has yet systematically investigated the combined effect of leading-by-example and rewards or punishments. In this paper, we fill this gap by examining the following research question: Do individual rewards or punishments reinforce the assumed cooperation-enhancing effect of leading-by-example?
Since many other confounding factors may play a role, investigating leading-by- example in the field is not easy. The advantage of controlled lab experiments is that we can disentangle the effect of individual rewards and punishments from the effect of leading-by- example. To do this, we conduct 2x3 factorial between-subject treatments, with and without leaders.
We make two main contributions to the existing literature. First, we inquire into the leader’s influence on contributions in the absence of any rewards or punishments. Since some previous studies experimentally investigated this question, we contribute to the literature with a robustness check. Second, we ask whether rewards or punishments reinforce leading-by-example concerning the increasing of contributions. By comparing different treatments, we identify the effects of leading-by-example and individual rewards and punishments. The second point is our novel contribution to the literature.
Our main findings are as follows: First, in the absence of any rewards or punishments, leader-free teams achieve similar contribution levels as teams with leaders do. Second, rewards and punishments are helpful in order to increase contributions, both in teams with and without leaders. We find, however, that setting an example does not further increase the impact of rewards and punishments on contributions. We conclude that in our setting, it is not leading-by-example, but rather the use of rewards or punishments that seems crucial for achieving high contributions.
Part I 18
3 Related Literature
3.1 The mechanics of leading-by-example
Leading-by-example may work via (social) preferences. Empirical evidence shows that many people are conditional cooperators who prefer to match others’ contributions (Fischbacher et al. 2001). Thus, theoretically, in the presence of conditional cooperators, leaders’ high efforts can influence the followers to exert high efforts, too. Experimental evidence, however, shows that many conditional cooperators slightly undercut others’ efforts (Fischbacher and Gächter 2010), which often leads to a decline in average contributions over time.
Leading-by-example could also work by shaping the followers’ beliefs. Cartwright and Patel (2010) show theoretically that an individual may prefer to contribute first in a sequential public goods game if she believes that a sufficient number of other players will imitate her contribution. Related to this, Gächter and Renner (2018) experimentally show, in a repeated setting, that the initial behavior of leaders seems to be decisive. In the subsequent rounds, however, followers place more weight on the fellow teammates’ behavior than on the leader’s efforts.
Leading-by-example may work even better “in the dark” (Weber 2015) if followers do not have full information on the productivity of the team project. In an asymmetric information model, Hermalin (1998) considers a team where only one single member knows about the team’s overall productivity. Hermalin (1998) shows that the team can achieve higher welfare if the informed team member exerts effort – as a leader – before her teammates do. With her costly commitment, the leader can credibly signal her being in the preferred high-productivity state and can motivate her teammates to exert high effort, too. This model nicely demonstrates that the existence of a leader and her revelation of private information may result in high cooperation in teams. Komai et al. (2007) extend Hermalin (1998) by showing that some cases exist where preventing full revelation of the state may induce followers to exert higher efforts, compared to if the leader reveals the state fully through her action (as in Hermalin 1998).
In a fundraising context, Vesterlund (2003) presents a similar theoretical prediction to that of Hermalin (1998) when there is uncertainty about the quality of a charity (the value of the public good). To increase overall contributions, fundraisers of high-quality charities
Part I 19 should publicly announce the first contribution. This model again provides a rationale for the effectiveness of having a leader in order to increase contributions.
In their model, Huck and Rey-Biel (2006) consider a team with two inequality-averse players. If a leader exerts effort first, team output and payoffs are higher, compared to simultaneous play. Huck and Rey-Biel (2006) also show how leadership can arise endogenously. In their setting, to maximize team output, the least productive member should be selected as the leader.3
3.2 Leading-by-example experiments without rewards or punishments
Leading-by-example is found to be beneficial in coordination experiments (Cartwright et al. 2013), as well as in settings with asymmetric information on the value of the team project (Potters et al. 2007, Komai et al. 2011). In this paper, we focus on leading-by- example settings with symmetric information about the value of the team project.
Sequential public goods games have emerged as the “workhorse” for the experimental investigation of leaders’ influence on team cooperation. In these experiments, the leader contributes first. Since other teammates, also denoted as “followers”, are informed about the leader’s contribution before simultaneously contributing themselves, the leader can set an example.
In a public bad experiment, Moxnes and van der Heijden (2003) show that the presence of a leader improves the group outcome to a small extent, but significantly so (followers invest 13% less in the public bad). In a public goods game, Güth et al. (2007) observe higher overall contributions if a leader is present, compared to a no-leader setting. Levati et al. (2007) find a positive effect of a leader’s existence on contributions when endowments are heterogeneous, but the effect is less pronounced for the case of homogenous endowments. Pogrebna et al. (2011) also find that contributions are higher when a subject makes a binding announcement before fellow group members contribute – compared to the simultaneous decision setting.
3 To our best knowledge, there are no theoretical studies investigating leading-by-example in combination with reward or punishment power.
Part I 20
There are, however, at least as much studies reporting no significant leader effect (Sturm and Weimann 2007, Haigner and Wakolbinger 2010, Sahin et al. 2015, Gächter and Renner 2018). In a field experiment conducted in Bolivia, Jack and Recalde (2015) do not find a significant leader effect if the leader is elected randomly, compared to the simultaneous baseline treatment without a leader. Hence, for experiments with symmetric information on the value of the public good, the literature is not conclusive on leaders’ influence on overall contributions.
In general, leaders’ and followers’ efforts are highly correlated (see, e.g., Gächter and Renner 2018). The followers, however, persistently undercut the leader’s contributions, which in turn makes the leader lower her contribution in the following round. Gächter and Renner (2018) argue that a leader, after realizing that the followers have contributed less than herself, wants to avoid the feeling of being exploited again and reduces her contribution in the next round. Overall, this behavior leads to a decay of contributions over time, similarly to what one observes in simultaneous contribution experiments without a leader.
3.3 Experiments with centralized rewards or punishments in simultaneous settings
Ample evidence from public goods experiments shows that decentralized (peer) rewards or punishments can increase contributions (see, e.g., Balliet et al. 2011, Milinski and Rockenbach 2012). The absence of a coordination device, however, often hinders the efficient use of rewards or punishments.
Numerous studies have investigated the centralized use of rewards or punishments when team members contribute simultaneously. O'Gorman et al. (2009) find that centralized execution of punishment does not increase contributions more than peer punishment does. Under perfect information, Fischer et al. (2016) do not find a difference in aggregate behavior between peer and central punishment, either. Nosenzo and Sefton (2014) report a similar result in a setting with both rewards and punishments. Grieco et al. (2017) compare several centralized punishment mechanisms to a decentralized peer punishment setting. Similarly to O'Gorman et al. (2009), if only one randomly selected group member is allowed to punish, contribution levels are not different from the decentralized setting. Contributions are significantly higher, however, if the group member with the highest contribution of the
Part I 21 current round has the power to punish. In a within-subject design, Harrell and Simpson (2016) investigate whether the introduction of punishment leads to higher contributions when only the leader has punishment power, compared to peer punishment. They do not find a significant contribution difference between the leader and the peer punishment treatments. Teams with prosocial leaders, however, contribute more than teams with proself leaders, but no more than in the peer punishment treatment. In Gürerk et al. (2009), leaders can choose between a positive and a negative incentive scheme. The selected scheme is used for ten consecutive periods. While initially 19 out of 20 leaders prefer the reward scheme, many of them switch to the punishment scheme after observing decreasing contributions in their group. In the final and third phase, most leaders choose the punishment scheme, which generates higher contribution levels than the reward scheme.
In contrast to a central authority in simultaneous settings as above, the leader in our sequential setting is possibly more effective for increasing the contributions, since she can set a clear example. If teammates do not follow, she can use rewards or punishments to motivate them.
3.4 Experiments with centralized rewards or punishments in sequential settings
To our best knowledge, only two studies investigate a (sequential) leading-by-example setting in combination with a reward or punishment power of the leader. In Güth et al. (2007), the team leader can punish just one team member by temporarily excluding her from the team and the team output. When leaders have such exclusion power, contributions increase. Sutter and Rivas (2014) report a treatment in which the leader can reward just one single follower. In this setting, contributions are higher compared to a control treatment without a leader.
Our study is different from the works mentioned above in some crucial ways: In the present study, unlike in Güth et al. (2007), leaders cannot exclude players from the group but can reward (or punish) teammates by increasing (or reducing) their payoffs. Unlike Sutter and Rivas (2014), where leaders may only reward or punish one group member, we give the leaders the possibility to reward or punish each teammate individually.
Part I 22
4 Experimental Design
4.1 Treatments
We explain our experimental design in three steps with respect to treatments. First - to investigate the leader’s influence in the absence of rewards or punishments - we conduct two treatments. While in the leader treatment L, the leader contributes first, in the leader-free treatment P, peers contribute simultaneously. Second, to investigate the impact of rewards or punishments in the absence of a leader, we conduct the treatments P-REW, and P-PUN, and contrast them to the P treatment. Third, to inquire into the leader’s influence in the presence of individual rewards or punishments, we conduct two more treatments: the treatment L-REW with reward possibilities for the leader, and the treatment L-PUN with punishment possibilities for the leader. To find out the combined effect of leading-by- example and rewards or punishments, we compare these treatments to the peer treatments P- REW and P-PUN.
4.2 The game
In the treatments L and P, we deploy a one-stage public goods game with a voluntary contribution mechanism. In the other four treatments, there is an additional stage with the possibility of allocating rewards or punishments. The game is played for 20 periods, in groups of four, with identical group composition over the rounds.
Public Goods Stage: In each period, each player receives an endowment of 푒 = 20 and decides on her contribution 0 ≤ 푐푖 ≤ 푒 to the public good. The sum of contributions 퐶 = 4 ∑푖=1 푐푖 is multiplied by 1.6 and split equally among all group members. Thus, the marginal per capita return MPCR amounts to 0.4. After each period, each player is informed about individual contributions and payoffs. In the leader treatments, in each group, one player is randomly chosen to serve as the leader for all periods. Leaders contribute first, the followers after receiving feedback about the leader’s contribution.4
4 We deliberately decided to use a random-selection mechanism in order to avoid confounds between our variables of interest and the potential effect from selection procedures, and to keep our design comparable to
Part I 23
Reward and Punishment Stage: In the leader treatments L-REW and L-PUN, the leader is equipped with an additional 20 points that she can either keep in her private account or use for individual rewards or punishments. In the peer treatments P-REW and P-PUN, each subject is given 5 additional points which she can either keep in her private account or use to reward (or to punish) other teammates individually.5 In all treatments with rewards or punishments, each allocated reward point increases the receiver’s payoff by 3 points, and each punishment point decreases the receiver’s payoff by 3 points. In all treatments, players keep the “unused” points in their private accounts.
Payoffs: The payoff 휋 푖 for player i in each treatment is given below. The sum of reward or punishment points that a player i allocates to others is given by 푟푖 and 푝푖, with 푟−푖 and 푝−푖 being the sum of the points that player i receives from other players.
L and P: 휋푖 = 20 − 푐푖 + 0.4퐶
퐿푒푎푑푒푟 퐹표푙푙표푤푒푟 L-REW: 휋푖 = 20 − 푐푖 + 0.4퐶 + 20 − 푟푖, 휋푖 = 20 − 푐푖 + 0.4퐶 + 3 푟−푖
퐿푒푎푑푒푟 퐹표푙푙표푤푒푟 L-PUN: 휋푖 = 20 − 푐푖 + 0.4퐶 + 20 − 푝푖, 휋푖 = 20 − 푐푖 + 0.4퐶 − 3 푝−푖
P-REW: 휋푖 = 20 − 푐푖 + 0.4퐶 + 5 − 푟푖 + 3푟−푖
P-PUN: 휋푖 = 20 − 푐푖 + 0.4퐶 + 5 − 푝푖 − 3푝−푖
the previous studies mentioned in section 2. We are also aware of potential effects of the selection modus (random) of the leader on contributions. Previous studies show that a rotating leader (Güth et al. 2007) who is chosen exogenously does not significantly affect leader contributions or the followers’ propensity to follow the leader’s example. On the other hand, if a randomly selected group member voluntarily decides to lead, team contributions are higher compared to teams with involuntary leaders (Haigner and Wakolbinger 2010). The focus of this study, however, is on the effectiveness of leading-by-example as such, and on the additional effects of individual rewards or punishments. 5 To clearly separate contribution behavior from the use of rewards or punishments, we provide additional tokens for the reward and punishment stage. We do this to keep the reward and punishment mechanisms comparable, and to ensure that leaders and peers face the same costs for a given level of reward or punishment. A number of studies apply a similar design (Gürerk et al. 2009, Choi and Ahn 2013, Nosenzo and Sefton 2014).
Part I 24
4.3 The procedures
We programmed the experiment with z-Tree (Fischbacher 2007) and conducted it at RWTH Aachen University using ORSEE (Greiner 2015) for subject recruitment. 288 subjects participated in 72 independent observations (12 per treatment). After the experimenter had read the instructions (see Appendix A.1) aloud, subjects could privately ask any clarifying questions. Sessions lasted about 60 minutes. Each subject was paid privately. The average payoff was €12.70.
5 Hypotheses
Assuming money-maximizing actors with self-centered preferences, and applying backward-induction, it is straightforward to see that no leader or follower should engage in costly punishment or reward. Following the same rationale, it is also apparent that no player contributes to the public good.
In contrast, models assuming social preferences (e.g., Bolton and Ockenfels 2000; Fehr and Schmidt 1999) show that players may contribute to public goods and forego parts of their earnings in order to punish uncooperative players.
Based on the evidence from the experimental literature discussed in Section 2, we state the following three hypotheses corresponding to our three main research questions.
Hypothesis 1: In the absence of rewards or punishments, teams with leaders achieve different levels of contributions than leader-free teams do. (Comparison of L and P treatments)
Hypothesis 2: In the absence of leaders, teams with reward or punishment options achieve higher contributions than teams without reward or punishment possibilities do. (Comparison of P-REW and P, and P-PUN and P treatments, respectively)
Hypothesis 3: In the presence of rewards or punishments, teams with leaders achieve different levels of contributions than leader-free teams do. (Comparison of L-REW and P-REW, and L-PUN and P-PUN treatments, respectively)
Part I 25
6 Results
We first focus on the test of our three hypotheses (Results 1-3). After that, we report results for the evolution of contributions and individual contribution behavior. We then investigate the use of rewards and punishments and teammates’ reactions in terms of contribution changes. Finally, we analyze the payoffs.
6.1 The effect of leadership on contributions in the absence of rewards or punishments
Table 2 shows the average contributions, payoffs, and the average rewards and punishments for all treatments. What is the effect of leadership on contributions in the absence of any punishment and reward possibilities? Note that in our setting, being a leader always means contributing first. As can be seen over all periods, the leader treatment L induces lower contributions than the peer treatment P, the difference, however, is not statistically significant (9.8 and 11.3 points, Mann-Whitney-U test (MWU), two-sided, p = 0.564).6
Result 1. In the absence of rewards or punishments, teams with leaders do not achieve different levels of contributions than leader-free teams.
As can also be seen from Table 2, the followers in the L treatment contribute significantly less than the leaders do (9.3 and 11.2 points, Wilcoxon-matched-pairs test (WMP), two-sided, p = 0.021). This result is in line with previous studies. We do not find a difference between the followers’ contributions in the leader treatment L and the teammates’ contributions in the peers treatment P (9.3 and 11.3, MWU, two-sided, p = 0.453).
6 This result may be considered to be somewhat surprising, since some of the previous studies (mentioned in Section 2.2.1.) report that groups with leaders tend to have higher than or at least similar contribution levels to treatments without leaders. We are not the first to find insignificantly lower average contributions in a leader setting compared to a no-leader environment (Gächter and Renner 2018: no-leader treatment 10.24, leader treatment 9.64, p = 0.603).
Part I 26
Table 2: Treatment averages over all periods. Standard deviations in parentheses.
Contributions Treatment L P L-REW P-REW L-PUN P-PUN All 9.8 (5.7) 11.3 (4.6) 14.5 (5.0) 15.9 (4.2) 13.8 (6.8) 16.9 (4.4) Leaders 11.2 (5.6) - 13.1 (5.9) - 14.6 (6.5) - Followers 9.3 (5.9) - 15.0 (5.4) - 13.5 (7.2) -
Payoffs from the public goods stage excluding rewards and punishment Treatment L P L-REW P-REW L-PUN P-PUN All 25.9 (3.4) 26.8 (2.8) 28.7 (3.0) 29.5 (2.5) 28.3 (4.1) 30.1 (2.6) Leaders 24.4 (4.2) - 30.1 (5.2) - 27.5 (5.7) - Followers 26.3 (3.4) - 28.2 (3.1) - 28.5 (3.9) -
Net Payoffs including rewards and punishments Treatment L P L-REW P-REW L-PUN P-PUN All 25.9 (3.4) 26.8 (2.8) 38.8 (5.1) 40.4 (4.9) 31.6 (4.4) 32.6 (4.1) Leaders 24.4 (4.2) - 40.0 (1.9) - 45.8 (5.9) - Followers 26.3 (3.4) - 38.4 (6.9) - 26.9 (4.3) -
Average allocated rewards and punishments (average of the team’s total) Treatment L P L-REW P-REW L-PUN P-PUN All - - - 11.7 (5.1) - 2.5 (2.3) Leaders - - 10.2 (5.1) - 1.6 (1.4) -
6.2 The effect of rewards and punishments on contributions in the absence of leaders
To see, in the absence of leaders, whether teams with reward or punishment options achieve higher contributions than teams without, we compare P to P-REW, and P to P-PUN, respectively. As can be seen in Table 2, teams with peer rewarding possibilities (P-REW) achieve significantly higher contributions than the teams in the peer treatment P (15.9 and 11.3, MWU, two-sided, p = 0.043). Similarly, teams with peer punishment options achieve higher contributions than the teams in P (16.9 and 11.3, MWU, two-sided, p = 0.006).
Result 2. In the absence of leaders, teams with reward or punishment options achieve higher contributions than teams without such options.
Part I 27
6.3 The effect of leaders on contributions in the presence of rewards or punishments
To investigate whether leaders influence contributions in the presence of reward or punishment possibilities, we compare the average contributions of the L-REW treatment to P-REW, and the contributions in L-PUN treatment to P-PUN, respectively. Based on our finding that leading-by-example does not increase contributions in the absence of reward or punishment opportunities (Result 1), we argue that any potential difference in contribution levels between L-REW and P-REW or L-PUN and P-PUN should come from the combined effect of having a leader with the power to reward or punish. Over all 20 periods, we observe no significant differences between the treatments L-REW and P-REW (14.5. and 15.9 points, MWU, two-sided, p = 0.237), and between L-PUN and P-PUN (13.8 and 16.9 points, MWU, two-sided, p = 0.106). The average contributions even tend to be higher in the peer treatments (see Table 2).
To separate the (pure) effect of having a leader from the effect of having a leader with reward or punishment power, we run two panel regressions. The results are presented in Table 3.
Table 3: Treatment differences - the effects of leadership, punishment, and rewards.
(I) (II) Dependent variable: L-PUN, L-REW, Teammate’s contribution P-PUN, P-REW, L & P L & P −0.553 −0.634 (1) Leader dummy (0.560) (0.685) 2.111*** 1.677*** (2) Punishment (I) / reward (II) treatment dummy (0.558) (0.595) −0.543 0.526 (3) Punishment (I) / reward (II) treatment dummy x Leader dummy (0.878) (0.931) 0.732*** 0.692*** (4) Other followers’ (peers’) average contribution in t-1 (0.031) (0.037) 2.596*** 3.065*** Constant (0.505) (0.598) Observations 3192 3192 R² overall 0.615 0.491 *** Significant at 1%. Panel regression with cluster-robust standard errors in parentheses (clustered on group level). Note: To check for potential collinearity problems we run the same specifications without (4). The results remain qualitatively the same.
Part I 28
In column (I), we compare individual contributions in treatments L, P, L-PUN, and P- PUN to see whether a leader can increase contributions in a punishment environment. Neither the dummy that indicates whether the groups have a leader (1) nor the independent variable for the combined effect of having a leader with punishment power (3) is significantly different from zero. In contrast to that, the dummy for being in a punishment treatment (2) is highly significant. Column (II) compares the treatments L, P, L-REW, and P-REW, and shows a very similar result for the reward environment. There is no effect from having a leader (1) or from having a leader with reward power (3). Being in a reward environment (2), however, results in significantly higher contributions. In both regressions, the other followers’ average contribution in the previous period (4) has a positive and significant effect on contributions in the current period.
Result 3. In the presence of rewards or punishments, leaders have no (additional) effect for increasing the contributions.
6.4 Evolution of contributions and individual contribution behavior
Evolution of contributions
Panel a) of Figure 3 shows the evolution of contributions in treatments L and P. Note, for the leader treatment L, that we plot the overall team average, and additionally, the average contribution for the followers. Over time, contributions both in L and P treatments show a decreasing trend. Compared to the first half of the experiment (periods 1-10), average contributions both in L and P are significantly lower in the second half (average of the periods 1-10 in L: 12.2 points, in periods 11-20: 7.3 points, WMP, two-sided, p = 0.007), (P treatment: 13.7, and 8.8, WMP, two-sided, p = 0.006). Both in L-REW and P-REW treatments (see Figure 3, panel c), contributions in the second half are also significantly lower than in the first half (L-REW: 15.3 points and 13.6, WMP, two-sided, p = 0.023, P- REW: 16.9 and 14.8, WMP, two-sided, p = 0.034). Interestingly, this decreasing trend is absent in both punishment treatments (see Figure 3, panel b). While we observe a non- decreasing trend in the L-PUN treatment, we even see a clearly increasing trend in the P- PUN treatment: We find no difference in average contributions in L-PUN (13.8 in periods 1-10 and 13.8 in periods 11-20, WMP, two-sided, p = 0.969), whereas contributions are significantly higher in the second half in P-PUN (16.3 and 17.5, WMP, two-sided, p = 0.050).
Part I 29
Figure 3: Average contributions.
Individual contribution behavior: does leading-by-example work?
To analyze teammates’ contribution behaviors, we ran a series of panel regressions that we present in Table 4. In column (I), we compare the behavior in treatments L and P to see whether leaders make a difference in the absence of any rewards or punishments. In columns (II) and (III), we compare L-REW and P-REW, and L-PUN and P-PUN, respectively. Column (IV) looks at the three leader treatments aiming at disentangling the effects of leaders’ contributions and the effect of their use of reward or punishment. Column (V) displays the effects of a reward or punishment setting in the absence of the leader.
The independent variable “leader dummy” (1) indicates whether the group has a leader or not. Interestingly, the mere existence of a leader has no significant effect on teammates’ contributions in any of the comparisons. In all leader treatments, however, the leader’s contribution (2) does have a significant influence on followers’ contributions. The magnitude of this effect, however, is not different from the effect of the other followers’
Part I 30 contributions (3) which is also significant.7 The interaction between these two influencing factors (2) and (3) shows that the existence of a leader significantly reduces the impact of teammates’ average contribution (4).
Table 4: Determinants of contributions (all periods).
(IV) (V) (I) (II) (III) Dependent variable: L-REW, P-REW, L & P L-REW & L-PUN & Teammate’s contribution L-PUN & P-PUN & P-REW P-PUN L P −0.952 1.423 −0.904 (1) Leader dummy (0.608) (2.639) (2.316) 0.441*** 0.142** 0.212*** 0.404*** (2) Leader contribution (0.034) (0.058) (0.082) (0.036) (3) Other followers’ (peers’) average 0.818*** 0.645*** 0.680*** 0.430*** 0.738*** contribution in t-1 (0.032) (0.090) (0.078) (0.045) (0.036) (4) Leader x Other followers’ (peers’) −0.447*** −0.225* −0.255** average contribution in t-1 (0.067) (0.121) (0.105) 6.208*** 1.463*** (5) Reward treatment dummy (1.777) (0.511) (6) Reward treatment dummy x −0.255***
Leader contribution (0.082) 3.724** 2.079*** (7) Punishment treatment dummy (1.752) (0.555) (8) Punishment treatment dummy x −0.164*
Leader contribution (0.085) 1.585*** 5.501*** 5.578*** 0.472 2.524*** Constant (0.376) (1.584) (1.613) (0.481) (0.543) Observations 1596 1596 1596 2052 2736 R² overall 0.549 0.341 0.650 0.580 0.514 * Significant at 10%, ** at 5%, *** at 1%. Panel regression with cluster-robust standard errors in parentheses (clustered on group level). In column (IV) we only look at the followers’ contributions (3 per group), in column (V) we look at the contributions of all team members (4 per group), hence the different number of observations (2052*4/3=2736).
A closer look at the relation of leaders’ contributions and the availability of rewards (6) and punishments (8) reveals that the mere possibility of punishment or reward reduces the impact of leaders’ contributions. It seems that leading-by-example becomes less important
7 We tested whether the coefficients (2) and (3) are different. To do so, we calculated the chi-square for (2)- (3)-(4)=0. The p-value is 0.405 (Wald test).
Part I 31 as soon as the leader is equipped with additional means to influence followers’ contributions. The following correlations also support this claim. While leaders’ and followers’ contributions are highly correlated (average Spearman’s rho = 0.746) in the L treatment, the leaders’ impact is much smaller in L-PUN (average Spearman’s rho 0.504), and even less so in L-REW (average Spearman’s rho 0.205). The average correlation coefficients (rho) obtained in the L treatment are significantly higher than the coefficients in the L-PUN and L-REW (L and L-PUN: MWU, one-sided, p = 0.002, L and L-REW: MWU, one-sided, p = 0.001). Comparing the impact of the mere possibility of reward (5) and punishment (7) shows that a centralized use of reward or punishment power by the leader (IV) affects contribution behavior much more strongly than the same possibilities in a decentralized environment (V) (Wald-test, rewards: p = 0.003, punishment: p = 0.002).
Result 4. The effect of leading-by-example (measured as the correlation between leader and follower contributions) is less strong if the leader has rewards or punishments at hand.
6.5 The use of rewards and punishments
As can be seen in Table 2, leaders, as well as peers, allocate more reward points than punishment points (MWU, two-sided, p < 0.001 in both cases). On average, leaders do not allocate more points than peers do, either rewards (MWU, two-sided, p = 0.544) or punishments (MWU, two-sided, p = 0.299).8
Figure 4 shows the average points that a teammate received, dependent on the contribution difference between her and the rewarding/punishing leader; in the peers treatments, the difference between her and the rewarding/punishing team member, respectively (averages include zeros if no reward or punishment point was assigned).9 Let us first look at the
8 If we only consider cases with positive reward and punishments, there are still no differences between L- REW and P-REW (11.9 and 12.2, MWU, two-sided, p = 0.71) or between L-PUN and P-PUN (5.9 and 5.6, MWU, two-sided, p = 0.85). Figure 9 shows the evolution of rewards and punishments over time (Appendix A.4). 9 Recall that in the leader treatments, the leader may assign a maximum of 20 points, in a discretionary way, to one single teammate or to several teammates. In the peer treatments, each teammate may allocate up to 5 points.
Part I 32 rewarding behavior. In L-REW, leaders allocate the highest average rewards (4.76 points) to those teammates who contributed more than the leaders themselves. Interestingly, all leaders rewarded (at least one time) followers who contributed the same amount (3.32 points) or even less (0.72 points). In the P-REW treatment, teammates on average allocate the most reward points to peers who contributed the same amount as they did. On average, they allocated 1.20 points to those who contributed the same amount, 0.68 to those who contributed more, and 0.31 to those who contributed less than they did.
Figure 4: Average received reward or punishment.
Based on single rewards and punishments, rather than the totals a subject might have received in a period. As peers only have 5 points to allocate, in the panels on the right-hand side, we only see bubbles for the received reward/punishment categories zero and the interval [1, 5]. The diameter of a bubble reflects the relative frequency of the respective reward or punishment level.
Part I 33
Now let us look at punishments. In both PUN treatments, on average, those subjects who contributed less than the person allocating the points were punished more heavily (L- PUN 1.58 points, P-PUN 0.94 points), compared to teammates who contributed equal (L- PUN 0.11 points, P-PUN 0.04 points) or more (L-PUN 0.45 points, P-PUN 0.19 points). Teammates punish others more harshly, the greater the contribution differential between the punisher and the teammate is. Nevertheless, some subjects also punish teammates who contributed more than they did. Interestingly, this “anti-social” punishment is significantly less frequent in L-PUN than in P-PUN (3.3% and 13.3% of all punishments, MWU, two- sided, p = 0.046).
Result 5. Leaders exert anti-social punishment less frequently than peers.
In public goods experiments with peer punishment, a non-negligible fraction of peers punishes those who have contributed more than themselves (see, e.g., Nikiforakis 2008). One possible motivation for anti-social punishment is retaliation punishment. The fact that followers cannot retaliate after the leader has punished them may at least partly explain why we observe less anti-social punishment in L-PUN.
Do leaders use rewards and punishments differently than peers do?
To investigate how reward and punishment are used on an individual level, and whether there are differences in behavior between leaders and peers, we run a series of hurdle regressions. In the probit columns (I)-(IV) depicted in Table 5, we estimate how the probability of receiving a reward or punishment depends on a teammate’s contribution. The Tobit columns (V)-(VIII) provide information on the magnitude of the points received from the teammate, given that she received a positive amount of reward or punishment.
Part I 34
Table 5: Determinants of punishment and reward decisions (all periods).
Probit Tobit Dependent variable: probability of subject i of being punished (rewarded) amount of punishment (reward) that subject i receives (I) (II) (III) (IV) (V) (VI) (VII) (VIII) L-PUN L-REW P-PUN P-REW L-PUN L-REW P-PUN P-REW −0.030* 0.151*** −0.020 0.136*** −0.153 0.287*** −0.155* 0.159*** (1) i’s contribution (0.018) (0.017) (0.037) (0.019) (0.113) (0.036) (0.086) (0.027) (2) i contributes less (REW: more) than the punishing 0.803** 0.598*** 0.846*** -1.093***
(rewarding) subject (0/1, binary) (0.338) (0.225) (0.231) (0.205) (3) i’s contribution is lower (REW: higher) than the average 1.713*** −0.250 0.594*** −0.002 contribution of the other two subjects in the group (0/1) (0.292) (0.340) (0.203) (0.153) (4) Difference between i’s and the punishing (rewarding) −0.130 0.104*** 0.014 −0.103*** subject’s contribution (0.167) (0.025) (0.085) (0.012) (5) Difference between i’s and the average contribution of −0.690*** 0.026* −0.257*** 0.076*** the other two subjects in the group (0.213) (0.049) (0.077) (0.016) −1.980*** −1.457*** −1.372* −1.591*** −5.033* −1.723*** −1.196 −1.992*** Constant (0.371) (0.166) (0.704) (0.264) (2.853) (0.520) (1.859) (0.495) Observations 720 720 2880 2880 720 720 2880 2880 Pseudo R² 0.513 0.518 0.242 0.368 0.222 0.242 0.152 0.192 * Significant at 10%, ** at 5%, *** at 1%. Since in P-treatments, each subject can reward/punish the peers, we have four times more observations than in the L-treatment. Cluster-robust standard errors in parentheses (clustered on group level).
Part I 35
Let us look first at the probit columns (I) and (III) to see which variables increase the chances of being punished. In L-PUN, a higher contribution decreases the chances (1), but in P-PUN, the effect is not significant. In both punishment treatments, both the relative contribution compared to others (3) and that compared to the punishing teammate (2) increase the chances of receiving punishment significantly. Interestingly, for a leader’s decision to punish a follower, it is more important how much a follower contributes compared to the other followers than compared to the leader herself, as the respective coefficient (3) is considerably higher than (2). The Tobit columns (V) and (VII) show, in both punishment treatments, that the amount of punishment decreases with the contribution level (1) by a similar magnitude, but only significantly so in the P-PUN treatment. In both PUN treatments, teammates receive less punishment, the more they contributed compared to the other two teammates (5).
Now let us turn to rewards. As columns (II) and (IV) show, in both reward treatments the own contribution level increases the probability of being rewarded (1). In L-REW, contributing more than the leader (2) increases the chances of a reward significantly, but not the relative contribution compared to other teammates. Interestingly, in P-REW, contributing more than others is decreasing the chances of getting a reward. As the Tobit columns (VI) and (VIII) show, in both treatments, if rewarded, the magnitude of reward increases with the own contribution (1). In L-REW, the reward increases with the contribution difference between the reward-receiving teammate and the rewarding teammate (leader) while in P-REW, the reward is decreasing with the contribution differential between the rewarding and the rewarded peer (4). It seems that peers use rewards to encourage those who have contributed less than themselves, while leaders reward in order to compensate those who have already contributed a lot. In P-REW, the reward increases with the contribution difference between the rewarded and the other two teammates (5). Overall, we find a larger effect (regarding magnitude) from falling behind the other two team members’ average contribution in PUN treatments (coefficients (5) L-PUN: -0.690, P-PUN: -0.257) compared to the REW treatments (coefficients (5) L-REW: 0.026, P-REW: 0.076).
Leaders seem to have different reference contributions for evaluation when allocating rewards and punishments. While in the case of punishment, the relative contribution of the teammate to other followers is decisive, when giving rewards, leaders evaluate a teammate’s contribution to her own (the leader’s) contribution. In particular, a teammate is more likely to be punished if she has contributed less than other followers. In the case of rewards, a teammate who contributes more than the leader is more likely to get a reward.
Part I 36
Teammates’ reaction to rewards and punishments (in the next round)
To what extent do teammates change their contribution in period t after being rewarded or punished in period t-1 (given that they received positive rewards or punishments)? To answer this question, we run regressions as depicted in Table 6, with the contribution differential (contribution in t – contribution in t–1) being the dependent variable.
Table 6: Contribution change after receiving reward or punishment.
Dependent variable: (I) (II) (III) (IV) Contribution change (Contribution in t minus L-REW P-REW L-PUN P-PUN contribution in t-1) 0.048 0.211 Leader contribution (0.047) (0.181) Received reward -0.037 -0.145 points in t-1 (0.194) (0.145) Received punishment 1.172*** 0.350*** points in t-1 (0.199) (0.088) -0.146*** -0.296*** 0.087 -0.079* Period (0.050) (0.045) (0.112) (0.041) -0.505 2.280*** -4.687 1.610** Constant (1.554) (0.736) (3.564) (0.747) Observations 515 764 90 204 R² overall 0.028 0.072 0.221 0.107 * Significant at 10%, ** at 5%, *** at 1%. OLS with cluster-robust standard errors in parentheses (clustered on group level). We only looked at the team members’ contribution change after they received a reward or punishment. Since the frequency of rewards or punishments executions varies across treatments, the number of observations is different in each treatment.
We ran separate regressions for each treatment. As columns (I) and (II) show, perhaps unsurprisingly, receiving a reward does not significantly change a teammate’s contribution, either in L-REW or in P-REW. On the other hand, as columns (III) and (IV) show, punishments have a positive and highly significant effect on the contribution change of the punished teammate. In the leader treatment L-PUN, the effect is much larger (Wald test p = 0.015) than in P-PUN.
Result 6. (a) While rewards do not have a significant effect on contribution adjustments, punishments do. (b) Punishment from leaders has a much higher impact on contribution adjustments compared to punishment from peers.
Part I 37
A leader’s reward or punishment decisions may have higher acceptance from the other members since by contributing first, the leader exposes herself to the risk of being exploited. The higher potential acceptance, in turn, may lead followers to demonstrate stronger contribution reactions to punishments.
We also observe that except L-PUN, there is a significant negative trend of the contribution change over the periods, as the respective coefficients indicate. In other words, over time, the magnitude of adjustment from the previous round to the current round decreases.
Appendices A.2 and A.3 show group-level data on contributions and the use of reward and punishment points. In most of the groups in L-REW, the level of reward points assigned by the leader is rather stable, while in P-REW in most groups, the amount of reward points issued by the peers decreases over time. In treatments with the possibility of punishing, leaders, as well as peers in most cases, decide not to punish at all. In P-PUN, we observe that peers often use punishment in the first rounds until every group member contributes the maximum amount of 20 points, and then refrain from punishment in the following rounds. If someone deviates from the maximum contribution (usually in the last period), punishment is used again.
6.6 Payoffs
We focus on net payoffs, i.e., payoffs including rewards and punishments. 10 Not surprisingly, due to the payoff-increasing reward mechanism (recall that each allocated reward point generates two more points), net payoffs in L-REW and P-REW are higher than in L-PUN and P-PUN, respectively (MWU, two-sided, p = 0.002, p = 0.005). We do not observe significant differences, either between L-REW and P-REW (MWU, two-sided, p = 0.453) or between L-PUN and P-PUN (MWU, two-sided, p = 0.564). Payoffs in L are significantly lower compared to L-REW (L: 25.9 and L-REW: 38.8 points, MWU, two- sided, p < 0.001), and compared to L-PUN (L: 25.9 and L-PUN: 31.6 points, MWU, two- sided, p = 0.004), perhaps not surprisingly, since in L-REW and L-PUN, leaders have an additional endowment.
10 In Table 2, we also depict payoffs from the public goods stage, which are directly proportional to the contributions.
Part I 38
Leaders in L-REW use about half of their 20 additional points to increase the payoffs of the followers. This leads to similar net payoffs for the leaders (40.0 points) and the followers (38.4, WMP, two-sided, p = 0.908). In L-PUN, by allocating points, leaders cannot increase followers’ payoffs; but they can decrease the payoffs of their subordinates. Indeed, leaders in L-PUN receive much higher payoffs than followers do (45.8 and 26.9, WMP, two-sided, p < 0.001).
Which mechanism pays off more for leaders and which for followers? As the numbers above indicate, leaders with punishment possibilities obtain significantly higher payoffs than leaders with rewards (MWU, two-sided, p = 0.028), since leaders in the reward treatments invest a higher portion of their budget in rewards. Due to the payoff increasing reward mechanism, it is not surprising that followers in the reward treatment L-REW have much higher incomes than the followers in the L-PUN treatment (MWU, two-sided, p < 0.001). In other words, as a leader, it pays off to be in the punishment setting. For a follower, the rewards setting is more advantageous.
Part I 39
7 Discussion and Conclusion
In this study, we disentangle the effects of individual rewards and punishments on contributions from the pure influence of leadership. In the absence of rewards and punishments, teams with a leader do not achieve higher contributions than leader-free teams (Result 1). Thus, in our setting, leading-by-example as such is not sufficient for increasing contributions. This result is not in line with the early influential experimental literature on leading-by-example that reports a positive effect (Moxnes and van der Heijden 2003, Güth et al. 2007, and Levati et al. 2007). Other studies, however, report a non-significant result, (Sturm and Weimann 2007, Haigner and Wakolbinger 2010, Sahin et al. 2015, Gächter and Renner 2018). Why are there these discrepancies between the early papers and the followers? Let us look at the results of the early studies in more detail. Moxnes and van der Heijden (2003) investigate a “within subjects” design; they report a small but significant effect. In Güth et al. (2007), teams with leadership contribute only weakly significantly more than the control teams without leadership (p = 0.08, two-sided). Lastly, Levati et al. (2007) report a less substantial effect of leadership in the case of homogenous endowments. Thus, the effects found in these studies are rather small (Moxnes and van der Heijden 2003, Levati et al. 2007), or only weakly significant (Güth et al. 2007). Taken together, in cases of randomly chosen leaders in symmetric settings, the positive effects of leading-by-example on contributions might not be as robust as commonly believed. This conclusion renders our first important contribution to the literature.
In the absence of leaders, teams with rewards or punishments possibilities achieve higher contributions than teams without these options (Result 2). Thus, in our setting rewards and punishments are sufficient for raising contributions. This result supports previous findings from the literature on the effectiveness of rewards or punishment options for increasing contributions (see, e.g., Milinski and Rockenbach 2012).
Finally, if reward or punishment possibilities are available, teams with leaders do not achieve a higher level of contributions than leader-free teams (Result 3). In other words, in our setting, powerful leaders who not only lead by example but can administer individual rewards or punishments cannot make an (additional) improvement concerning the increase of contributions. This finding is our novel contribution to the literature.
What are the implications of the above results summarized above? They indicate that to increase contributions effectively, leaders need some additional power. In our setting, individual rewards or punishments are the driving force behind the increase of contributions,
Part I 40 and not the leader’s example, as the contrasts of our L-PUN and L-REW treatments to the corresponding P-treatments show. We observe, however, that contributions in teams with leaders who can reward or punish are not higher than the contributions in leader-free teams with peer rewarding or peer punishment options. Thus, our results indicate that in small teams with mutual monitoring possibilities, a leader may not be necessary in order to increase contributions. The teammates’ willingness to invest in costly bilateral rewards and punishments may suffice for maintaining high contributions. Indeed, some real organizations rely on flat hierarchies, some even with temporary leaders or no leaders at all. One of the most prominent examples is Valve, a major game developer and digital distribution company, which is organized into self-managing “boss-free” teams.11 Other examples with similar, non-rigid hierarchies and flexible organizational structures are W.L. Gore or Partake.12
How does our study relate to the large and growing experimental leadership literature? The present study sheds light on leading-by-example as such, and the allocation of individual rewards or punishments. A growing number of experimental studies investigate the role of other potentially important aspects of leadership with regard to contributions, such as communication (see, e.g., Pogrebna et al. 2011, Koukoumelis et al. 2012), or the modus of leader appointment (see e.g., Haigner and Wakolbinger 2010, Baldassarri and Grossman 2011, Bruttel and Fischbacher 2013), or the role of leader characteristics (see, e.g., Gächter et al. 2012). Future work should inquire into the interplay of these and other aspects with leading-by-example, and with rewards and punishments in order to obtain a complete picture.
11 http://www.valvesoftware.com/company/people.html, retrieved on June 2, 2017.
12 https://www.gore.com/about/our-beliefs-and-principles, retrieved on June 2, 2017. http://www.partake.de/en/was-ist-partake, retrieved on June 2, 2017.
Part I 41
Appendix A
A.1 Instructions (L-PUN)
General information
We welcome you to this economics experiment. It is very important for you to read the following instructions carefully. If you have any questions, please let us know.
In this experiment, you can earn money. The exact amount of your payout will depend on your decisions and on the other participants’ decisions.
While the experiment is running, you are not allowed to communicate with other participants. Non-compliance will lead to your exclusion from the experiment and from all payments. All decisions are anonymous, i.e., no other participant gets to know the identity of the participant who makes a specific decision. Anonymity is also ensured during the payout process, i.e., no participant gets to know the amount of other participants’ payouts.
During the experiment, your income will be calculated in points. The earned amount of points will be converted to Euro at the following exchange rate:
80 points = 1 Euro.
At the end of this experiment, you will receive your payout according to the total number of accumulated points as well as 2.50 Euro for showing up.
In the following, we will provide you with a detailed description of the experiment.
Rounds and groups
• The experiment consists of 20 rounds with each round having the same structure.
• You are a member of a group with 4 members in total. During the experiment, the group composition will always stay the same.
• One group member will randomly be assigned the role of a type A participant; the remaining three members will be type B participants.
• During the experiment, you will maintain your role and only interact with members of your group.
• Each participant receives a starting capital of 100 points.
Part I 42
The course of the experiment
Each round consists of two stages:
Stage 1: Contributions of the group members
• In every round, each group member receives 20 points.
• Each group member must decide how many of the 20 points he or she wants to contribute for the group. Points which are not contributed remain with the group member. Possible amounts to contribute are integral numbers from 0 to 20. First, the type A member decides how much to contribute for the group.
• After being informed about the contribution of the type A member, type B members decide on their contribution.
• The sum of the contributions of all group members (type A and type B) is multiplied by 1.6 and forms the group result.
(sum of contributions x 1.6 = group result)
• Each group member (type A and type B) receives a quarter of the group result independently of their own contribution (group result / 4 = individual share of the group result).
Stage 2
• The type A member gets to see how much each group member has contributed. • The type A member now receives 20 additional points and must decide whether - and if so - how many of these 20 points he or she wants to assign to each type B member. • With each point which the type A member assigns to a type B member, the income of the type B member is reduced by 3 points. • The type A member keeps points which have not been assigned.
Please notice: The order in which type B members are displayed will be determined randomly for each round. Therefore, it is not possible to identify a type B member over the rounds by a member’s position on the displayed lists.
Part I 43
Calculation of your round income
Round income for type A members =
20 (endowment for the round) - your contribution + 1.6 x sum of the contributions of all group members / 4 + 20 (points which can be assigned to type B members) - sum of points which are assigned to type B members
Round income for type B members =
20 (endowment for the round) - your contribution + 1.6 x sum of the contributions of all group members / 4 - 3 x number of received points
Information at the end of each round
At the end of each round, you will be provided with an overview of the group results. For each group member you will find out: contribution for the group, income after stage 1, assigned or received points, round income.
Please note: The order in which type B members are displayed will be determined randomly for each round. Therefore, it is not possible to identify a type B member over the rounds by the member’s position on the displayed lists.
Total income
The total income will result from the starting capital of 100 points plus the sum of the earnings from each of the 20 rounds. At the end of the experiment, your total income will be paid out at an exchange rate of 1 Euro per 80 points. As already mentioned, you will additionally receive 2.50 Euro for showing up.
Part I 44
A.2 Contributions and reward points within single observations (groups).
Figure 5: Average contributions and the sum of allocated reward points within single groups of the L-REW treatment.
Figure 6: Average contributions and the sum of allocated reward points within single groups of the P-REW treatment.
Part I 45
A.3 Contributions and punishment points in single observations (groups).
Figure 7: Average contributions and the sum of allocated punishment points within single groups of the L-PUN treatment.
Figure 8: Average contributions and the sum of allocated punishment points within single groups of the P-PUN treatment.
Part I 46
A.4 Contributions and reward or punishment points.
Figure 9: Average contributions and average allocated rewards and punishments.
Reward treatments show a negative trend in contributions. While in treatments with punishment possibilities, contributions remain stable in the second half (round 11-20) of the experiment compared to the first half (round 1-10), contribution declines significantly in treatments with rewards. The decline in contributions in reward treatments compared to punishment treatments is in line with findings of other studies (see, e.g., Gürerk et al. 2014, Sutter et al. 2010).
47
PART II: LEGITIMACY IN THE LAB: THE EFFECTS OF
LEADER SELECTION AND INFORMATION ON TEAM
PERFORMANCE
Co-authored with Özgür Gürerk & Thomas Lauer.
Abstract: Legitimacy is crucial for leaders. Research has identified several important sources of leader legitimacy. With observational and field data, however, it is difficult to isolate the unconfounded influence of a specific source of legitimacy on leaders’ and followers’ behaviors. Controlled lab experiments allow clear relationships between the cause and the effect to be derived. In a team setting, we study the specific and combined influence of two sources of leader legitimacy on subsequent team performance, namely endogenous leader selection and information about leader behavior. We find that if a leader derives her legitimacy from both sources, team cooperation increases more than when the leader is legitimized by a single source only. This positive effect on team performance is driven by the higher contributions of the leader herself.
8 Introduction
The legitimacy of an organization and of its authorities has been found to significantly influence employee adherence to organizational policy, leading to more compliance and deference, and less rule breaking (Tyler and Blader 2005). Legitimate leaders can evoke positive feelings among employees and enhance employee well-being, which is positively correlated with various measures such as productivity or loyalty (Keyes et al. 2000).
Legitimacy is also important outside typical organizations. Field studies provide evidence for the importance of a leader’s legitimacy in naturally existing groups. In an experiment with Ugandan farmers, Baldassarri and Grossman (2011) find team cooperation in a public good game to be higher if the group members elect a player who is equipped with punishment power, compared to a setting with a randomly chosen player with punishment power. Another field experiment, conducted in rural Bolivian communities, suggests that a legitimate leader can significantly increase public good contributions compared to a
Part II 48 randomly selected leader (Jack and Recalde 2015). In South African townships, Grieco et al. (2016) observe that if the legitimate leader of a group decides how to use contributions to a public good, contributions are much higher than when the purpose is discussed within the group or privately voted on. These findings suggest a connection between a leader’s legitimacy and cooperative behavior in groups.
In general, legitimacy can be described as a state perceived as appropriate and in accordance with norms and values (Dowling and Pfeffer 1975). Possible sources of the legitimacy of an individual and her actions can be propriety, endorsement, or authorization (Walker et al. 1986), the level of her compensation, or the availability of information about her actions (Dickson et al. 2015), as well as the way the individual is selected for her position (van Vugt and Cremer 1999, Brandts et al. 2014). French and Raven (1959) name elections as a common source of legitimacy.
In this study, we focus on two important sources of legitimacy, namely how the leader is appointed, and information about the leader’s past cooperation. In a series of experiments, we investigate how legitimacy influences team cooperation when one member of a team is appointed as the leader of that team for a subsequent experimental phase. In a 2x2 experimental design, we vary whether the leader is elected endogenously by the team members or appointed exogenously. We also vary whether team members are informed about a leader’s past cooperation or not.
We find that leader appointment as such increases team cooperation in the subsequent phase of the experiment, independent of the appointment modus (endogenous election or exogenous appointment). Cooperation increases the most, however, if the leader is elected by team members and information on the leader’s past cooperation is available to them.
A note on methodology
While field experiments generate valuable insights on leaders’ attitudes and characteristics, different leadership styles and more, they also come with a lack of control over potential covariates. Since many aspects can potentially influence the followers’ perception of the leader, with field data the identification of a causal relationship between various forms of legitimate leadership and cooperative behavior is virtually impossible.
In a comprehensive recent article, Podsakoff and Podsakoff (2019) underline the renewed interest in experiments in leadership research providing “strong evidence of causal relationships between independent and dependent variables” (Podsakoff and Podsakoff
Part II 49
2019, p. 21). To identify causal relations between a leader’s legitimacy and team cooperation, we conduct a lab experiment where individual leader characteristics are unknown to followers, excluding these characteristics as a potential influence on followers’ behavior.
9 Related Literature
To study leaders’ influence on team cooperation in the lab, scholars frequently use a variant of the public good game.13 In this game, teamwork is modeled as a public good. Each player (team member) can have an impact on the success of teamwork by contributing to the public good. The productivity of the team (the added value of teamwork) is implemented so that each player benefits from other team members’ contributions to the public good.
Leadership is usually modeled such that one player moves first, setting an example with her contribution decision before the remaining team members simultaneously contribute. Economists frequently use this setting, which is sometimes called the leading-by-example framework (see, e.g., Gächter and Renner 2018).
9.1 Leader appointment and cooperation in experiments
Exogenous appointment of leaders
One way of appointing a leader to the team is to select one member by chance. Previous literature shows that random leaders have virtually no positive effect on team cooperation, compared to settings without a leader. Typically, leaders contribute more to the public good than other team members do, who are called followers. Over rounds, however, leaders decrease their contributions, as observed in settings without a leader (Gächter and Renner 2018, Haigner and Wakolbinger 2010, Levy et al. 2011, Gürerk et al. 2018). A randomly appointed leader may even have detrimental effects on team cooperation (Sutter and Rivas 2014).
Leaders may also be exogenously appointed based on their cooperation preferences. Gächter et al. (2012) provide evidence that groups may benefit from (highly) cooperative leaders who contribute high amounts, partly because the latter falsely hold the belief that
13 We focus on literature using public good games. Other lab studies investigating the leader’s role on coordination are based on coordination games with different strategic properties (see, e.g., Brandts et al. 2014).
Part II 50 other members would contribute high amounts as well (false consensus effect). However, one cannot take it for granted that cooperative types are present in small groups.
Endogenous appointment of leaders
In some experimental studies, leaders were selected with unanimous acceptance. Güth et al. (2007) let subjects decide with regard to each group member, including themselves, whether they wanted that person as a leader in the next phase of the experiment. Only unanimously accepted members could become a leader. A similar procedure is used by Levati et al. (2007).
Other studies utilize different ask mechanisms for voluntary leadership. Haigner and Wakolbinger (2010) ask a randomly chosen group member whether she wants to lead (similarly to Dannenberg 2015a); Cappelen et al. (2015) ask all group members. In Sutter and Rivas (2014), whoever contributed first within a given time frame is automatically declared the leader. Except for Cappelen et al. (2015), who apply a stranger matching, all “ask” studies find that appointing a leader results in higher team cooperation compared to a situation without a leader.
Both the unanimous acceptance rule as well as the voluntary leadership procedure share one important issue: groups may end up having no leader at all. In fact, in the above studies, the share of groups which successfully appoint a leader is typically below 30%, with numbers reaching as low as 13.3% (Dannenberg 2015a). Therefore, only a minority of groups end up having a leader and can benefit from the selection methods above, while most groups remain without leadership. How, then, should a leader be appointed to ensure both the presence of a leader as well as a rise in contributions for all groups?
One way to guarantee the endogenous appointment of a leader is the majority vote. Levy et al. (2011) implement this method and allow leaders to suggest a contribution level to the group. In groups with endogenous leaders, team members contribute more and follow leaders’ suggestions more closely compared to groups with randomly determined leaders. However, analyzing the communication that preceded the leader selection, Levy et al. (2011) find that the leaders of successful teams (both endogenously elected and random leaders) were likely to describe reasonable strategies to reach efficient outcomes. Hence, it is not clear whether a group’s success depends on the leader selection method or the communicated strategies.
Part II 51
9.2 Information and cooperation in experiments
In experimental games, a common manipulation is to change the availability of relevant information, e.g., on the state of nature or on the behavior of group members. Faillo et al. (2013) implement different levels of information provision in a public good game where players can punish team members who contribute less than themselves. They find that contributions to the public good are significantly higher in a treatment with full information regarding group members’ contributions, compared to when the contribution information is only partially provided.
Using a similar experimental design, Fischer et al. (2016) test the effect of information on contributions in a public good game by modifying the preciseness of the given information. Information on the team members’ contribution amounts is either always accurate or players receive a noisy signal which is equally likely to show the correct contribution amount or any other possible level of contribution. Fischer et al. (2016) find that under the former condition, cooperation and total earnings are significantly higher compared to the latter setting.
Dickson et al. (2015) vary the availability of information about a punishment decision taken by a group leader. Depending on the treatment, all other team members know the individual contributions to a public good and which player is targeted for a possible punishment, or they do not have this information. Although Dickson et al. (2015) find no difference in contributions, team members show higher support for the leader’s punishment decisions if they have the information regarding teammates’ contributions, compared to when they do not have it. The authors consider information to be an essential aspect of legitimacy and conclude that the “lack of transparency undercuts legitimacy” (Dickson et al. 2015, p. 125).
9.3 Theoretical considerations and hypotheses
We expect that the legitimacy which group members associate with their leaders depends on the framework in which these leaders emerge. A legitimated leader may contribute more due to a heightened sense of responsibility for the group outcome. Furthermore, if a leader has legitimacy, other team members may be more inclined to follow her with their contributions than in the case of a leader without legitimacy. Hence, appointing a leader equipped with legitimacy may result in an overall positive shift of team cooperation.
Part II 52
Legitimacy based on endogenous appointment
Hollander (1992) points out that the way in which a leader attains her role has an important influence on how the followers perceive her, and therefore on her legitimacy. He describes appointment and election as different forms of inducing legitimacy since they affect followers in different ways.
Based on our argumentation that groups cooperate more when they have a legitimated leader compared to a leader without legitimacy and evidence from previous literature on the endogenous appointment of leaders, as reported in section 9.1, we state the following hypothesis.
Hypothesis 1. Teams with endogenously elected leaders increase their cooperation more than teams with exogenously appointed leaders.
Legitimacy based on the information about past cooperation
Extending traditional leadership theories, in newer theories attention is increasingly paid to the motivation and perceptions of followers as essential factors for successful leadership. In an extensive analysis of these more recent approaches, Hannah et al. (2014) elaborate that a leader’s influence on her followers not only depends on her own behavior and competence but also on the followers’ ideas of effective leadership and the extent to which they perceive her as credible. The absence of information on a leader’s previous behavior might be detrimental to her credibility, while full information can lead to higher cooperation and more support for the leader’s decisions, as shown in Section 9.2. If followers do not perceive their leader as credible, they should also assign little legitimacy to her, since they might not believe her authority to be appropriate (Tyler 2006). We therefore believe that the availability of relevant information on a leader’s contributions in a previous phase of the game should positively influence the legitimacy of a leader and result in higher cooperation within the team.
Hypothesis 2. Teams with leaders who were appointed/elected with their previous behavior disclosed increase their cooperation more than teams with leaders without this information.
Part II 53
Legitimacy based on two sources
For several forms of power, including that of legitimate power, French and Raven (1959) conclude that “the stronger the basis of power, the greater the power” (French and Raven 1959, p. 156). We assume that the more sources that provide a leader with legitimacy, the more impact this will have on team cooperation. The effect on cooperation is not likely to be linear in the number of sources, but we expect that the combined impact of two sources of legitimacy on team cooperation will be greater than the impact of a single source.
Hypothesis 3. Teams with leaders deriving their legitimacy from two sources increase their cooperation more than teams with leaders deriving their legitimacy from a single source.
10 The Game and Experimental Design
The game consists of three phases with ten rounds in each phase. In each phase, four players play a public good game (PGG). In Phase 1, all players decide simultaneously on contributions, and there is no leader. In Phase 2, depending on the treatment, a leader is appointed either exogenously or endogenously. In this phase, the leader contributes first, the other group members (followers) observe the leader’s contribution, and then they simultaneously decide on their own contributions. In Phase 3, in all treatments, leaders are appointed endogenously. In Phase 3, again, the leader contributes before the group members do.
In each round, each player receives an endowment of 20 points and decides on her/his contribution 푐푖 to the public good. The sum of contributions C is then multiplied by 1.6 and equally distributed among all group members. Hence, the payoff 휋푖 for each member equals:
휋푖 = 20 − 푐푖 + 0.4퐶 Before Phase 1 starts, players are informed that the experiment consists of three phases, but they are unaware that there will be leader appointments in Phase 2 and Phase 3. Only before Phase 2 begins are players informed about the leader appointment procedures. At the beginning of each phase, players also get to know that five of the following ten rounds would be randomly determined to be payoff relevant.
Part II 54
10.1 Leader appointment
Depending on the treatment, the leader for Phase 2 is appointed in one of four different ways. In a 2x2 design, we vary whether the leader is appointed exogenously or endogenously, and whether information on contributions in Phase 1 is available or not. In all treatments, subjects are informed at the beginning of Phase 2 of how the leader will be selected. In EXO, chance determines the leader. In EXO-Info, the group member with the highest average contribution in Phase 1 automatically becomes the leader for Phase 2. In the other two treatments, the leader gets elected by majority vote from within the group, where subjects have one vote and can vote for one of the group members, including themselves. If two or more subjects receive the highest number of votes, one of these subjects is randomly assigned the leader role.14 In ENDO, each subject can vote for one of four symbols, which are randomly assigned to the four group members. In ENDO-Info, subjects see the contributions of all group members for every round in Phase 1, as well as their average contributions.Table 7 shows our treatments and the different ways of leader appointment.
Table 7: Treatment overview.
Leader appointment Treatment Leader appointment for Phase 2 for Phase 3 EXO Random EXO-Info Player with the highest average contribution in Phase 1 Majority vote with ENDO Majority vote with randomly assigned symbols contribution information Majority vote with contribution information for every for every group member ENDO-Info group member
Leader appointment for Phase 3 does not differ across treatments and is done by majority vote with full information on contributions in Phase 2. For EXO, EXO-Info, and ENDO, a change of the leader selection method might induce a different contribution level in Phase 3 compared to Phase 2, since the leader in Phase 3 has a different form of legitimacy compared to the leader in Phase 2.
14 Another possibility would be to do another vote and only randomly assign a leader if the group keeps failing to produce a majority decision. In this case, however, Kocher et al. (2013) find that groups are worse off compared to having no vote and randomly assigning a leader right away.
Part II 55
10.2 Procedures
We used z-Tree (Fischbacher 2007) for programming and ORSEE (Greiner 2015) for subject recruitment, and we experimented in the AIXperiment economics lab at RWTH Aachen University. In total, 168 subjects participated (49% female; mean age 24.4) in groups of four with partner matching, resulting in 42 independent observations (11 each for EXO and EXO-Info, 10 each for ENDO and ENDO-Info). Before each phase, the experimenter read the instructions for that phase to the subjects, who then could privately ask any questions. Sessions lasted about 75 minutes with an average payoff of 15.65 €.
11 Results
We start our analysis by comparing the increase in cooperation levels across the four treatments, which gives us a general assessment of how the two sources of legitimacy affect team performance. We then separate the analysis to investigate to what extent these differences can be ascribed to leader behavior and follower behavior, as well as investigating the relation between the two. Table 8 reports the average contributions in each phase for all treatments. To keep track of individual behavior across the different phases, we refer to team members who are leaders or followers in Phase 2 as leaders or followers also in Phase 1.
Table 8: Average contributions. Standard deviations in parentheses.
EXO EXO-Info ENDO ENDO-Info (N=44) (N=44) (N=40) (N=40) Phase 1⁺ 9.0 (7.0) 7.9 (6.4) 7.0 (5.2) 6.9 (5.7) Leader 9.3 (6.9) 10.8 (6.4) 7.4 (5.1) 8.4 (5.2) Follower 9.0 (7.2) 7.0 (6.6) 6.9 (5.4) 6.4 (6.3)
Phase 2 13.2 (6.2) 11.5 (8.0) 11.4 (6.0) 13.7 (5.7) Leader 14.4 (5.5) 13.5 (6.8) 13.0 (5.9) 15.7 (5.1) Follower 12.8 (6.4) 10.8 (8.5) 10.9 (6.1) 13.0 (6.0)
Phase 3⁺⁺ 13.0 (7.1) 10.7 (8.7) 12.5 (6.9) 11.8 (6.3) Leader 14.7 (7.2) 12.2 (8.5) 14.3 (6.5) 13.8 (5.8) Follower 12.4 (7.4) 10.3 (9.0) 12.0 (7.1) 11.1 (6.5) ⁺ Note that in Phase 1, contributions were made simultaneously. ⁺⁺ In all treatments, leader appointment in Phase 3 was determined by majority vote with information on previous contributions (the same method as used in Endo-Info for Phase 2).
Part II 56
In our first treatment, EXO, average contributions amount to 9.0 (45%), which is consistent with the cooperation rates of between 35% and 50%, which are typically observed in similar settings (Zelmer 2003). The same holds for the other treatments. To provide some more detail, Table 8 also reports the average contributions separately for players in the leader role and those in the follower role.
In Phase 1, average team cooperation is not significantly different across treatments (Mann-Whitney-U test (MWU), two-sided, p > 0.42 for all comparisons, leaders, followers, and combined). While not statistically different from each other, average contributions still vary across treatments. To have a consistent measure of our treatment effects, we compare the differences in relative contribution changes from Phase 1 to Phase 2.
In all treatments, average contributions increase remarkably with the introduction of a leader in Phase 2. In EXO and EXO-Info, contributions increase by a similar rate (45.9% and 44.2%, respectively, Wilcoxon-matched-pairs test (WMP), two-sided, p = 0.075 each), in ENDO the participants contribute 62.4% (WMP, two-sided, p = 0.059) more and in ENDO-Info contributions almost double (97.9%, WMP, two-sided, p = 0.005). In all treatments combined, contributions increased by 59% (from 7.8 to 12.4 points, WMP, two- sided, p < 0.001).
Part II 57
11.1 The effect of endogenous leader election
As the first test of Hypothesis 1, we conduct a linear difference-in-differences (DID) analysis, comparing the contribution change from Phase 1 to Phase 2 using the pooled data of EXO and EXO-Info against the pooled data of ENDO and ENDO-Info. Table 9a shows the results of this analysis.
Table 9a-c: Average contributions and differences with and without endogenous leader selection for the pooled sample (9a) and separate sub-samples (9b, 9c). Standard deviations in parentheses.
Phase 1 Phase 2 Difference
9a EXO & EXO-Info 8.5 (8.3) 12.3 (8.6) 3.8** ENDO & ENDO-Info 7.0 (7.6) 12.6 (7.9) 5.6*** Difference / DID 1.5*** (0.4) -0.3 (0.4) 1.8*** 9b EXO 9.0 (8.2) 13.1 (8.1) 4.1* ENDO 7.1 (7.6) 11.4 (8.1) 4.3* Difference / DID 1.9*** (0.5) 1.7*** (0.6) 0.2 9c EXO-Info 7.9 (8.4) 11.5 (9.0) 3.6* ENDO-Info 6.9 (7.5) 13.7 (7.6) 6.8*** Difference / DID 1.0* (0.5) -2.2*** (0.6) 3.2***
Note: row differences are tested with Wilcoxon signed rank test (two-sided); column differences and difference-in-differences (DID) are based on linear regression with robust bootstrapped standard errors, DID p-value is correct for multiple hypotheses testing in Table 9a-c using the adjustment proposed by Holland and Copenhaver (1987). * Significant at 10%, ** at 5%, *** at 1%.
The differences in the first (3.8, 45% increase) and second (5.6, 80% increase) row confirm the positive effect of appointing a leader in both sub-samples. In row three, the difference-in-differences (DID = 1.8) shows that this effect is stronger in the ENDO sub- sample. This result supports our first hypothesis stating that teams with an endogenously elected leader increase their contributions significantly more (p < 0.01) than those with an exogenously selected one.
Next, to see whether this result holds for both information conditions, we run two separate diff-in-diff analyses. In the first, we compare EXO and ENDO, and in the second we compare EXO-Info and ENDO-Info. For the first comparison, the results in Table 9b show that the increase in contributions in ENDO (62%) is about 16 percentage points higher than it is in EXO (46%). This difference, however, is not statistically significant (DID = 0.2,
Part II 58 p = 0.744). It seems that the endogenous election of a leader itself does not increase cooperation more than a random leader selection does. In contrast, the difference in contribution increases is significantly higher (DID = 3.2, p < 0.01) in ENDO-Info (98%) than in EXO-Info (44%). Table 9c presents the results of this second comparison. Considering all three diff-in-diff analyses, we summarize our first result as follows.
Result 1: Groups that appoint leaders endogenously increase their cooperation more than groups following a randomly selected leader, but only if information about the leaders’ past behavior is available.
11.2 The effect of information about the leader’s past behavior
We now turn to our second hypothesis to see whether the disclosure of the leader’s previous contributions has a positive effect on the team’s contributions. Following the same procedure as in the preceding section, we start with a diff-in-diff analysis that compares the pooled data from both information treatments with the pooled data from both treatments without information. Table 10a presents the results of this comparison.
Table 10a-c: Average contributions and differences with and without information about leader behavior for the pooled sample (10a) and separate sub-samples (10b, 10c). Standard deviations in parentheses.
Phase 1 Phase 2 Difference
10a EXO & ENDO 8.1 (8.0) 12.4 (8.1) 4.3* EXO-Info & ENDO-Info 7.4 (8.0) 12.5 (8.0) 5.1*** Difference / DID 0.7* (0.4) -0.1 (0.4) 0.8 10b EXO 9.0 (8.2) 13.2 (8.1) 4.2* EXO-Info 7.9 (8.4) 11.5 (9.0) 3.6* Difference / DID 1.1** (0.6) 1.7*** (0.6) 0.6 10c ENDO 7.1 (7.6) 11.4 (8.1) 4.3* ENDO-Info 6.9 (7.5) 13.7 (7.6) 6.8*** Difference / DID 0.2 (0.5) -2.3*** (0.6) 2.5*** Note: row differences are tested with Wilcoxon signed rank test (two-sided); column differences and difference-in-differences (DID) are based on linear regression with robust bootstrapped standard errors, DID p-value is correct for multiple hypotheses testing in Table 10a-c using the adjustment proposed by Holland and Copenhaver (1987). * Significant at 10%, ** at 5%, *** at 1%.
Part II 59
While the increase in contributions is stronger in information treatments (68%) than in the non-information treatments (53%), this difference is small (DID = 0.8) and not statistically significant (p = 0.147). The separate analyses for the unpooled data reveal that there is also no difference in contribution increase between EXO and EXO-Info (DID = 0.6, p = 0.428, see Table 10b). As with endogenous elections (the first source of legitimacy that we tested), information by itself does not affect contribution increases. In the two treatments with endogenous leader election, however, we find a strong effect from information on the increase of contributions. We already know from the previous section that average contributions in ENDO increase by 62% compared to 98% in ENDO-Info. Table 10c confirms that this difference in the increase is highly significant (DID=2.5, p < 0.001). Hence, we conclude that information only fosters cooperation increase if the team members have elected the leader. Taking the results from the three diff-in-diff analyses together, we state our second result as follows.
Result 2: Teams with leaders whose past behavior is known increase their cooperation more than teams without information on their leader’s past behavior, but only if the leader is selected endogenously.
11.3 The combined effect of both sources of legitimacy
In this section, we focus on our third hypothesis to see whether a combination of two sources of legitimacy fosters the increase in cooperation more than a single source does. We pool the data from the two single-source treatments (EXO-Info and ENDO) to compare them against the two-source ENDO-Info treatment. The row differences in Table 11 show that a single source of legitimacy (election or information) leads to an average increase in contributions by about 52%. In ENDO-Info, where both sources are combined, the average relative increase is almost twice as high (98%). The difference-in-difference estimator confirms that the cooperation-enhancing effect from two sources is significantly stronger (DID = 2.9, p < 0.01) than from a single source. Based on this support for H3, we state our third result as follows.
Result 3: Teams with a leader who derives her legitimacy from two sources increase their cooperation more than teams with a leader whose legitimacy is based on a single source.
Part II 60
Table 11: Average contributions and differences with one or two sources of legitimacy. Standard deviations in parentheses.
Phase 1 Phase 2 Difference
EXO-Info & ENDO 7.5 (8.0) 11.4 (8.6) 3.9**
ENDO-Info 6.9 (7.6) 13.7 (7.6) 6.8***
Difference (abs) 0.6 (0.4) -2.3*** (0.4) 2.9***
Note: Deviation of DID-value is due to rounding. Row differences are tested with Wilcoxon signed rank test (two-sided); column differences and difference-in-differences (DID) are based on linear regression with robust bootstrapped standard errors. ** Significant at 5%, *** at 1%.
11.4 Leader and follower behavior
To further explore the effects of endogenous election and information about leader behavior, we analyze the contributions of leaders and followers separately. In addition, we estimate a structural equation model that provides a better understanding of the process by which our ENDO-Info treatment affects team cooperation.
Figure 10 shows a descriptive analysis of the potential effects on leaders’ and followers’ contributions. The first bar in Figure 10 shows the leader effect, i.e., the relative increase in average contributions from Phase 1 to Phase 2 in EXO. While the first bar in the leader panel is somewhat higher than the corresponding bar in the follower panel (54.8% and 42.2%), this difference is not statistically significant (MWU, two-sided, p = 0.369). In other words, the increase in contributions that we can attribute to the mere introduction of a leader is not different for leaders and followers.
The second bar in each panel in Figure 10 shows the ENDO effect, which we define as the relative contribution increase in ENDO minus the leader effect. Again, the difference between leaders (20.8%) and followers (15.8%) is not significant (Wald-test, p = 0.216).
The INFO effect is depicted by the third bar in each panel and defined as the relative increase in EXO-Info minus the leader effect. As for the first two effects, there is no difference between leaders and followers (Wald-test, p = 0.254). The last bar shows the combined effect, defined as the relative contribution increase in ENDO-Info minus the Leader effect, the ENDO effect, and the INFO effect. Also, for this comparison, we find no difference between leaders’ and followers’ reactions (Wald-test, p = 0.253).
Part II 61
Figure 10: Average relative change in leaders’ and followers’ contributions from Phase 1 to Phase 2, attributed to treatment effects.
The obvious conclusion from these findings is that leaders and followers are similarly affected by our treatments. The sequential nature of the game, however, provides at least one alternative explanation. Given that our treatments are intended to enhance the legitimacy that followers grant their leaders, they might follow a leader’s example more closely when her position is based on an endogenous election and the information about her past behavior. To test this alternative explanation, we run a series of regressions. Table 12 presents the results of our regression analysis.
Part II 62
Table 12: Regression analysis: (I) treatment effects on contribution difference between leaders and followers, (II) and (III) structural equation models for contributions in Phase 2.
(I) (II) (III) OLS regression Structural equation model
DV: contleader-contfollower DV: contfollower (1) Leader’s contribution in Period t 0.635*** 0.636***
(0.033) (0.033) (2) Endogenous election (dummy): 0.282 -0.299
ENDO, ENDO-Info (0.435) (0.405) (3) Information on leader behavior (dummy): 0.059 0.037
EXO-Info, ENDO-Info (0.405) (0.382) (4) Endogenous and information (dummy): 0.368 0.125
ENDO-Info (0.598) (0.569) (5) Period 0.145** -0.358*** -0.358*** (0.060) (0.059) (0.059) (6) Other followers’ contribution in Period t-1 -0.065*** 0.266*** 0.265*** (0.021) (0.030) (0.030) (7) Own average contribution in Phase 1 -1.021*** 0.965*** 0.964*** (0.39) (0.099) (0.097) (8) Other group members’ average 0.999*** -0.869*** -0.855*** contribution in Phase 1 (0.116) (0.106) (0.102) (9) One source of legitimacy (dummy): -0.127
EXO-Info, ENDO (0.335) (10) Two sources of legitimacy (dummy): -0.140
ENDO-Info (0.389) Constant 0.267 5.201*** 5.179*** (1.046) (1.077) (1.072)
DV: leader’s contribution
(11) Endogenous election (dummy): -0.049
ENDO, ENDO-Info (0.422) (12) Information on leader behavior (dummy): 0.264
EXO-Info, ENDO-Info (0.424) (13) Endogenous and information (dummy): 1.350***
ENDO-Info (0.595) (14) Period -0.584*** -0.584***
(0.055) (0.055) (15) Followers’ contributions in Period t-1 0.551*** 0.551***
(0.021) (0.021) (16) Own average contribution in Phase 1 -0.152* -0.163*
(0.090) (0.087) (17) Group’s average contribution in Phase 1 0.357*** 0.370***
(0.106) (0.092) (18) One source of legitimacy (dummy): 0.111
EXO-Info, ENDO (0.364) (19) Two sources of legitimacy (dummy): 1.565***
ENDO-Info (0.387) Constant 14.991*** 14.978***
(0.924) (0.923) N 1,260 1,260 1,260 * Significant at 10%, ** at 5%, *** at 1%.
Part II 63
In the first model (I) in Table 12, we estimate the treatment effects on the difference between leaders’ contributions (contleader) and followers’ contributions (contfollower). We restrict the analysis to followers’ decisions from periods 11 to 20 (Phase 2). To account for the effects of past cooperation, we include the other followers’ contributions in the previous period (see row (6) in Table 12) as well as the average individual (7) and group (8) contributions in Phase 1 (period 1 to 10). None of three treatment dummies (2-4) is significantly different from zero (p > 0.5 for all dummies), i.e., the distance that followers keep to the leader’s example does not seem to be affected by one of the sources of legitimacy or by the combination of both. As for the control variables, we find a small but significant increase over time (5) and an even smaller negative effect from others’ past contributions (6). The negative effect from a player’s average contribution in Phase 1 (7) and the positive effect from average group contribution in Phase 1 (8) cancel each other out (Wald-test (7) + (8) = 0, p = 0.418).
If the extent to which the team members follow their leader’s example is not affected by our treatments, it must be the leader’s example itself that increases with additional legitimacy. In column (II) in Table 12, we estimate a structural equation model to test this conclusion. The first equation describes the followers’ contributions as a function of their leader’s contribution (1), the treatment effects (2-4), and the control variables we introduced above (5-8). The leader’s contribution, in turn, is described as a function of the treatment effects (11-13) and the same control variables (14-17). The coefficients of the treatment dummies show that neither one of the two sources alone (2 and 3), nor the combination of both (4) has a significant effect on followers’ contributions (p > 0.460 for all dummies). Joint estimation of the control variables (5-8) reveals that their overall effect is also not significantly different from zero (Wald-test, p = 0.939). The predominant factor that explains the followers’ contribution choices is the example set by the leader (1).
Turning to the second part, in which we estimate the effects on the leader’s contributions, shows that the combination of both sources of legitimacy in ENDO-Info (13) has a tremendously positive effect on the example set by the leader. While the other two treatments do not affect the leader’s contributions (11 and 12), there is a positive effect from
Part II 64 team cooperation in Phase 1 (p < 0.001, joint estimate of 16 and 17)15. We summarize the results of the structural equation model as follows.
Result 4: Increasing the leader’s legitimacy does not directly affect the followers’ cooperation.
As a robustness check, we estimate an alternative structural equation model in column (III). Here we combine the two dummies that indicate either endogenous election (2) or information disclosure (3) to a new dummy that indicates the presence of one of the two sources (9). For reasons of clarity and comprehensibility, we also replace the dummy for ENDO-Info (4) by a new dummy indicating the presence of both sources (10). We do the same for the second equation and replace (11) and (12) with (18), and (13) with (19). The model in (III) replicates the main results from the first structural equation model.
11.5 Ex-post legitimization
In this last part of the results section, we incorporate the data from Phase 3 into our analysis. Remember that, in this final phase, all leaders are endogenously elected and that information on average individual contributions in Phase 2 is disclosed to each group member. In other words, the ENDO-Info procedure is applied to all treatments. These additional data allow us to analyze whether it is possible to legitimize a leader ex-post after she has already been appointed without (or with less) legitimization. 16 Since ex-post legitimization is only possible if a group reelects their leader from the second phase also for the final phase, we introduce a new dummy variable to indicate that a group has the same leader in Phase 2 and Phase 3. A descriptive analysis shows that the groups who reelected their leader contribute on average 13.1 (SD=8.6) in Phase 3, while those groups that elect a new leader contribute only 10.5 (SD=8.7) on average. Across all treatments, 59.5% of all groups reelected their leader. To estimate the reelection effect, we include this dummy in a new structural equation model like the one in the previous section, using contribution data from Phase 3. The detailed results are presented in Table 13 (II) in appendix B.2. We find that being reelected does not affect the leader’s contributions (see (11) in Table 13, coef. =
15 The negative effect from period (14) and the positive effect from followers’ contributions in the previous period (15) cancel each other out (Wald-test (14) + (15) = 0, p = 0.548). 16 A change in the leader appointment method for Phase 3 only occurs in EXO, EXO-Info, and ENDO. Hence, we only analyze the Phase 3 contributions of these treatments.
Part II 65
0.552, p = 0.122), but significantly increases the followers’ contributions (see (6) in Table 13, coef. = 1.561, p < 0.01). Consequently, team members who reelected their leader follow the leader’s example more closely than those who elected a new leader. The corresponding OLS regression (see (I) in Table 13) confirms this. In groups with a reelected leader, the distance between followers’ and leader’s contributions is on average about 1.6 points smaller than in groups with a new leader.
Result 5: Groups that reelect, i.e., legitimize their leader ex-post based on an endogenous election with information about the leader’s past behavior, contribute more than group members who elect a new leader.
12 Discussion
Leadership has frequently been proposed as a possible remedy for the lack of cooperation in social dilemmas. Numerous studies (see section 9.1) report a positive effect of leadership, although only a minority of groups have a leader, whereas most groups cannot benefit from leadership. On the other hand, studies where a leader is present in every group often find no positive effect of leadership. Evidence from field data suggests that a crucial factor for a leader’s effectiveness is the leader’s legitimacy. Using a lab experiment, we systematically test different leader selection mechanisms which guarantee a leader for each group and vary possible sources of legitimacy. We find that a majority vote with information on past behavior yields the highest increase in cooperation among leaders and followers.
A majority vote has also been used by Levy et al. (2011) and Brandts et al. (2014). Both first implement a part without a leader and then use a vote for leader appointment. The positive shift to a more efficient group outcome, however, may not only be attributed to the election, but rather to the content of the leader communication. In our setting, apart from the signal a leader makes with her binding contribution, no communication is possible. Hence, a difference in cooperative behavior can be assigned to the way a leader is chosen and, therefore, to her legitimacy.
Leader behavior Interestingly, different sources of a leader’s legitimacy not only have effects on overall group behavior but in some cases also work differently for the leaders and her followers, respectively. In contrast to the other selection methods tested here, leaders do not increase
Part II 66 their contributions if they are exogenously selected based on their high performance (EXO- Info) – but why do they not? One possible explanation is that since the selected leaders are already the highest contributors in their group, little room for improvement is left. Looking at their average contribution of 10.8 in Phase 1, this is not the case. Speculating about possible reasons, the selected group members perhaps feel that they have already done their best and that further effort would not result in better results. Alternatively, since they have contributed most to their group and therefore improved the group outcome more than any other group member, they might now want another group member to do the job.
Now, let us look at the leaders who were elected by their group with information about the leaders’ contributions at hand (ENDO-Info). Here, group members with the highest average contributions were elected in seven out of ten groups. In contrast to the leaders in EXO-Info, who were also selected because of their high performance, leaders in ENDO-Info significantly increased their contributions in Phase 2.17 Although the appointment is based on the same criterion in both cases, elected leaders further increase their contributions despite already being the highest contributors, whereas exogenously selected leaders remain at their previous contribution level. Since an elected leader is determined by her group, she might regard her new role as a sign of trust or a distinction and therefore be motivated to contribute even more.
Follower behavior Followers significantly increase their contributions if the highest contributor of the group becomes their leader (EXO-Info). They also contribute more if they can determine their leader themselves (ENDO, ENDO-Info). This increase in followers’ contributions, however, is not due to a change in their leader perception, i.e., that they follow their leaders more closely with their own contributions. Legitimized leaders set a better example than non-legitimized leaders, and since team members follow the example, the overall team cooperation increases with additional legitimacy. Followers do not significantly increase their contributions if the leader is randomly selected (EXO). On the other hand, followers do contribute higher when they elect their leader based on randomly assigned symbols which carry no informational value (ENDO), which seems peculiar since this can also be considered a random leader appointment. The
17 This is also the case if we only look at the seven groups in ENDO-Info, where the highest contributors were elected as leaders (13.9 vs. 6.9 points, p = 0.028, WMP, two-sided).
Part II 67 involvement in the leader selection process, however, might be a possible reason for the ensuing positive shift in contributions.
13 Conclusion
Whenever a leader is required, one should be regardful of the way in which the leader is appointed. Depending on the selection method, followers may assign different degrees of legitimacy to their leader. Since the legitimacy of a leader not only plays a crucial role for the followers’ behavior but also for how the leaders act themselves, adequate leader selection must be taken into consideration.
This study contributes to the understanding of leader legitimacy on cooperation, induced by different leader selection methods in a lab experiment. Comparing various procedures, we find the biggest increase in cooperation for groups where the members can determine their leader endogenously by majority vote and have information about the group members’ past behavior. Using a majority vote for leader appointment not only grants the leader high legitimacy but also ensures the presence of a leader for every single group. Future research could deepen our understanding of possible long-term effects of the different leader selection methods as well as the different effects of a leader’s legitimacy on her followers and on the leader herself.
Part II 68
Appendix B
B.1 Instructions (EXO)
General information
We welcome you to this economics experiment. It is very important for you to read the following instructions carefully. If you have any questions, please direct them to us.
In this experiment, you can earn money. The exact amount of your payout depends on your decisions and the other participants' decisions.
While the experiment is running, it is not allowed to communicate with other participants. Non-compliance leads to the exclusion from the experiment and all payments. All decisions are anonymous, i.e., no other participant gets to know the identity of the participant who makes a specific decision. Anonymity is also ensured during the payout process, i.e., no participant gets to know the amounts of other participants’ payouts.
During the experiment, your income will be calculated in points. The earned amount of points will be converted to Euro with the following exchange rate:
30 points = 1 Euro.
At the end of this experiment, you will receive your payout according to the total number of accumulated points as well as 3 Euro for showing up.
In the following, we will provide you with a detailed description of the experiment.
Parts of the experiment and group
• The experiment consists of 3 parts. At this point, you only receive the instructions for part 1. Instructions for part 2 and part 3 will be handed out before those parts will start.
• You are a member of a group with 4 members in total. During the experiment, the group composition will always stay the same.
Part II 69
The course of a round
• Part 1 consists of 10 rounds.
• In every round, each group member receives an endowment of 20 points.
• Each group member has to decide how many of the 20 points he or she wants to contribute to the group. Points which are not contributed remain with the group member. Possible amounts to contribute are integral numbers from 0 to 20.
• The sum of the contributions of all group members (type A and type B) gets multiplied with 1.6 and forms the group result.
(sum of contributions x 1.6 = group result)
• Each group member receives a quarter of the group result independently from their own contribution (group result / 4 = individual share of the group result).
Calculation of round income
퐺푟표푢푝 푟푒푠푢푙푡 푅표푢푛푑 푖푛푐표푚푒 = 퐸푛푑표푤푚푒푛푡 − 푌표푢푟 푐표푛푡푟푖푏푢푡푖표푛 + 4
Information at the end of each round
At the end of each round, you will be provided with an overview of the group results. For each group member, you will get to know: contribution for the group, round income.
Please notice: The order in which group members are displayed will be determined randomly for each round. Therefore, it is not possible to identify a group member over the rounds by the position on the displayed lists.
Total income
The total income will result from the sum of the earnings from each of the three parts. For part 1, 5 rounds will be chosen randomly. Your income from part 1 is the sum of points from these five rounds. At the end of the experiment, your total income will be paid out with the exchange rate of 1 Euro per 30 points. As already mentioned, you will additionally receive 3 Euro for showing up.
Part II 70
Part 2
In part 2, you play 10 rounds of the same game as in part 1 with the following changes:
• One group member will randomly be assigned the role of a type A participant, the remaining three members will be type B participants. • In every round, first the type A member decides how much to contribute for the group. • After being informed about the contribution of the type A member, type B members decide on their own contribution.
During the 10 rounds of part 2, all group members will maintain their role (type A or type B).
For part 2, 5 rounds will be chosen randomly. Your income from part 2 is the sum of points from these five rounds.
Part 3
In part 3, you play 10 rounds of the same game as in part 2 with the following changes:
The decision which group member becomes the type A participant is carried out by a vote within your group. Each group member receives information on the contributions of all group members in part 2. Please indicate for every group member including yourself if he or she should become a type A participant.
The participant who receives the most votes gets assigned the role of type A participant. If more than one group member receive the highest amount of votes, one of these group members gets randomly selected as type A participant.
During the 10 rounds of part 3, all group members will maintain their role (type A or type B).
For part 3, 5 rounds will be chosen randomly. Your income from part 3 is the sum of points from these five rounds.
Part II 71
B.2 Regression Analysis (reelection effect)
Table 13: Regression analysis: (I) reelection effect on contribution difference between leaders and followers and (II) structural equation models for contributions in Phase 3.
(I) (II) OLS regression Structural equation model
DV: contleader-contfollower DV: contfollower (1) Leader’s contribution in Period t 0.820*** (0.018) (2) Endogenous election (dummy): 0.143 0.059 ENDO, ENDO-Info (0.323) (0.249) (3) Information on leader behavior (dummy): 0.262 -0.541** EXO-Info, ENDO-Info (0.309) (0.229) (4) Period 0.221*** -0.223*** (0.061) (0.047) (5) Other followers’ contribution in Period t-1 -0.071*** 0.075*** (0.017) (0.016) (6) Leader reelected (dummy) -1.644*** 1.561*** (0.336) (0.257) Constant -1.731 5.083*** (1.569) (1.258)
DV: leader’s contribution
(7) Endogenous election (dummy): 0.575*
ENDO, ENDO-Info (0.328) (8) Information on leader behavior (dummy): -1.103***
EXO-Info, ENDO-Info (0.349) (9) Period -0.332***
(0.062) (10) Followers’ contributions in Period t-1 0.492***
(0.017) (11) Leader reelected (dummy) 0.552
(0.357) Constant 17.870***
(1.618) N 1,260 1,680 * Significant at 10%, ** at 5%, *** at 1%.
72
PART III: GOAL SETTING AND TEAM PERFORMANCE
Abstract: Goal setting has been shown to affect performance in various settings (Latham and Locke 2007). In this study, we focus on how the mechanism of goal setting might influence its effect on team performance and how such an effect might vary with goal level. Using a threshold public good game, we test how the agents’ contributions are influenced by the goal setting party, i.e., the agents or the principal, and the difficulty of a goal. In a follow-up study, we further explore how the different goal setting mechanisms and a change thereof influence performance in a repeated setting. Although agents prefer a goal- free setting, they contribute significantly more if they choose a high goal for themselves. We find the opposite contribution pattern, however, if the principal chooses a goal for the agents.
14 Introduction
The effects of goal setting on performance have been studied thoroughly in the past decades. A robust result from these studies and core element of goal setting theory (GST) is that setting a specific and challenging goal enhances the performance of an individual or a group, compared to a goal-free environment (Mento et al. 1987, O'Leary-Kelly et al. 1994, Latham and Locke 2007).
In today’s working environment, projects are often worked on by groups rather than individuals, and there is uncertainty about the project’s success. While some goals may be easy to reach, others may be hard but result in a substantial payoff if completed successfully. If an ambitious goal is not reached, however, time and effort spent on achieving the goal may be lost without generating a positive outcome for the contributing agents. In our study, we investigate how contributions to such a risky project might be influenced by the way the goal is chosen. Does goal setting, within this framework, maybe prove to be a viable leadership tool to increase the team members’ contributions?
In a principal-agent setting, we test how goal setting affects contribution decisions in a threshold public good game, where contributions are lost if a specific goal is not met. For this, we vary the goal level and use two goal setting mechanisms, where the decision rights
Part III 73 are either assigned to the agents or the principal. In the former case, agents determine by majority vote if they want to contribute to a project with a specific goal, or rather to a project without a goal. When the principal can decide, she picks one of these two options for the agents.
In our first study, we investigate the effects of goal setting when the team members only cooperate once and how different goal levels might induce different contribution decisions. We also conduct a follow-up study to further improve our understanding of two more aspects of goal setting. First, we investigate how goal setting influences performance if team members interact repeatedly. Second, we analyze how a change of the goal setting mechanism affects team performance.
Our results confirm the performance-enhancing effect of a specific, challenging goal as predicted by GST, but only if the agents themselves decide for this goal by majority vote. In contrast, if the principal chooses such a goal for the agents, they contribute significantly less compared to when she chooses either a low goal or even a goal-free setting for the agents. Interestingly, agents strongly prefer a goal-free setting, while principals show no such preference.
15 Related Literature
The subject of goal setting has been the focus of research for many years, and correspondingly many studies concerning this topic have been published. The numerous publications have been the subject of several reviews and meta-analyses. Before we describe the essential findings of those superordinate articles, in the following, we first present the goal setting theory by Edwin Locke and Gary Latham. We conclude this part looking at experimental studies that are particularly close to our specific research questions.
15.1 Goal setting theory
Goal setting theory (GST), as formulated by Edwin Locke and Gary Latham, deals with the relationship between goal setting and task performance and how this relationship is affected by different factors (Locke and Latham 1990, 2002, 2006, Latham and Locke 2007). An underlying assumption of GST is the goal difficulty function, which describes a positive
Part III 74 relationship between goal difficulty and task performance: the higher the goal, the higher the performance, until the limit of someone’s ability is reached.
Another essential element of GST is that the specificity of a goal can also enhance performance. Goal specificity per se does not yet lead to an improvement in performance; compared to 'do your best' goals, however, specific and challenging goals lead to a significant improvement in performance.
The results of the studies that contributed to the formation of GST are mainly based on experiments in which the participants had to expend a real effort, e.g., in tasks involving calculations or reading, or by assembling objects. In our setting, there is only an abstract form of effort – the decision about a project contribution makes no difference in the physical effort but results in a cost for the subjects depending on the amount they contribute. Hence, subjects in our study are not limited by their physical ability or mental capacity but can freely decide how much they want to contribute to a project.
15.2 Meta-analyses
In an early meta-analytic study, Mento et al. (1987) investigate the influence of goal setting on individual task performance, analyzing research published between 1966 and 1984. They focus on two strings of literature: on the one hand, studies that deal with the influence of goal difficulty on performance. On the other hand, studies that examine the impact of specific goals on performance in comparison to non-specific or missing goals. Overall, Mento et al. (1987) analyzed 119 studies, including both laboratory experiments and field studies. In contrast to more recent studies, where participants usually received performance-dependent compensation for taking part, in about 84% of the analyzed studies, subjects had not been incentivized.
Regarding goal difficulty and goal specificity, the analysis of Mento et al. (1987) is in line with goal setting theory: hard and specific goals lead to higher performance than low specific goals, unspecific goals, or no goals at all. Mento et al. (1987) further look at a small number of studies comparing participative versus assigned goal setting but only find inconclusive results.
O'Leary-Kelly et al. (1994) also investigate the influence of goal setting on performance, but for groups instead of individuals. A separate analysis of goal setting for groups seems necessary because there are additional factors that can influence performance in the group
Part III 75 context. Among these factors are, e.g., the group composition, the degree of anonymity, or possible conflict of individual and group goals. For their meta-analysis, O'Leary-Kelly et al. (1994) analyze ten studies published between 1978 and 1991. Comparing the performance of groups with and without goals, they find a significant difference, suggesting that goal setting improves performance for groups even more than for individuals.
Apart from the quantitative meta-analysis, O'Leary-Kelly et al. (1994) also conduct a qualitative review, including the results of 19 additional studies from 1950 to 1992, for a total of 29 studies. In this review, the authors investigate the influence on the group goal effect of eight variables, among them goal difficulty and goal source.18 For goal difficulty, O'Leary-Kelly et al. (1994) find no difference between settings with a challenging goal and those with an unclear goal difficulty. Both goal types enhance group performance in 81% and 85% of the cases, respectively. Comparing assigned to endogenously set goals, the latter seems to work better, leading to improved group performance in every single case. This improvement rate of 100%, however, is based on a low number of only nine studies with participatively set goals. Assigned goals, on the other hand, also improve group performance in 78% of the cases (14 out of 18 studies).
In a more recent meta-analysis, Kleingeld et al. (2011) build on the work of O'Leary- Kelly et al. (1994) and analyze 38 studies from the period from 1978 to 2010. They find a positive effect of specific group goals on group performance and an even stronger effect of specific difficult group goals. In their investigation of potential moderator variables of the group goal effect, Kleingeld et al. (2011) expected that more participation in goal setting resulted in a larger effect on group performance. They find, however, that different degrees of participative goal setting do not moderate its effect on performance and name participation in group goal setting as a topic for further investigation.
15.3 Experimental studies
In a game with one principal and one agent, Falk and Kosfeld (2006) discover that setting a binding minimum can be detrimental to the agent’s contribution. In their experiment, the principal has no endowment, and her payoff is entirely dependent on the agent’s contribution decision. Before the agent contributes, the principal decides if she wants
18 The other variables are goal specifity, task type, subject type, setting, group type, and time.
Part III 76 to control the agent by imposing a binding minimum contribution or let the agent freely decide how much to contribute. On average, agents contribute significantly less if their principal demands a binding minimum contribution, compared to agents whose principal did not impose such a restriction on them. Falk and Kosfeld (2006) further find that agents most frequently perceive the principal’s decision to control them as a sign of distrust and a lack of autonomy in their contribution decision. One significant difference to our experimental design is that the principal can force a minimum contribution on the agent. In contrast, in our study, the principal may choose a project with a goal, but agents are not obligated to contribute. Furthermore, it is only a two-player game, i.e., when deciding on a contribution, the agent has complete information about her resulting payoff since it will not be affected by other agents’ contribution decisions or the loss of contributions due to a missed goal.
Similar to our study, Engel and Rockenbach (2014) let subjects play a public good game in which there are active players who can contribute to the public good and passive players who cannot contribute. All players receive an endowment of 20 points, and before they know their player type, vote for a non-binding contribution recommendation. Contribution amounts are restricted to either the whole endowment, half of it, or nothing at all. Depending on the treatment, contributions by the active players can affect passive players’ payoffs.
Engel and Rockenbach (2014) find that voting leads to higher contributions, but only if the passive players‘ payoffs are unaffected by the contributions, whereas voting does not result in higher contributions if passive players would benefit from them. The authors identify the payoff structure as a reason for this behavior since, for the case of the positive externality, an active players’ payoff can often be lower than that from a passive player. To avoid these unfortunate payoff comparisons as best as possible, we endowed principals with only half of the endowment of the agents. Additionally, we do not restrict agents regarding their contribution amounts and their votes do not determine a specific amount, but instead, if the project they contribute to has a goal or not.
Dannenberg (2015b) also tests how non-binding agreements may result in higher contributions in a public good game. She finds that agreeing on a medium contribution goal leads to significantly higher contributions than agreements to contribute the whole endowment. Nevertheless, in both cases, contributions are not significantly different from the baseline treatment without any agreement. Dannenberg (2015b) observes the opposite contribution pattern, however, if players can punish each other after the contribution phase; then, agreements on the high goal lead to significantly higher contributions than agreements
Part III 77 on the medium goal. In contrast to our study, agreements are always formed endogenously by the group members, whereas we compare contributions to projects which are either voted for by the agents or exogenously selected by the principal.
Rauchdobler et al. (2010) investigate if implementing thresholds can increase contributions in a public good game. They find that neither exogenously set nor endogenously chosen thresholds improve cooperation, and the former is often detrimental for the subjects’ payoffs. On the other hand, when group contributions below a threshold are not refunded, subjects reach the thresholds more often if they are exogenously set compared to thresholds determined by vote. Subjects also showed a preference for low or medium thresholds, compared to a high threshold or no threshold.
We took the experimental design of Rauchdobler et al. (2010) as a starting point for our design. Hence, there are some similarities: we also use a majority vote as the mechanism for endogenous goal setting and work with similar goal levels. The most relevant differences lie in the way an exogenous goal comes into place and, related to this, the group composition. While exogenous goals in Rauchdobler et al. (2010) are predetermined by the experimenter, in our study, there is an active choice by the principal, who also can profit from the agents’ contributions.
16 Experimental Design and Procedures
16.1 The game
We use a modified threshold public goods game with two independent rounds. At the start of the experiment, players are randomly matched to groups of four and are assigned one of two roles, agent or principal19, which they keep for both rounds. Between the rounds, however, players are reassigned to different groups, ensuring that no player stays in a group with a group member from the first round (perfect stranger matching).
Within a group, the three agents each receive an endowment of 푒퐴 = 20 points and can contribute an amount 0 ≤ 푐푖 ≤ 푒 to one of two projects, either Project 1 (P1) or Project 2
19 We used a neutral wording. See Appendix C.1 for the instructions.
Part III 78
20 (P2). The principal receives an endowment of 푒푃 = 10 points but cannot contribute to a project. In P1, contributions to the project are doubled and then split among the group members, providing each agent with 30% and the principal with 10% of the payoff. In order to receive a payoff from P2, the sum of contributions must be at least as high as a specific goal. The agents know the number of points needed to reach the goal before deciding on their contributions. If a goal is met in P2, the payoff scheme is identical to that of P1. If the goal is missed, however, players receive no payoff from the public good, and the agents’ contributions are not refunded. Hence, the payoffs for an agent 휋퐴푖 and the principal 휋푃 for P1 and P2 (if the goal is met) are:
휋퐴푖 = 푒퐴 − 푐푖 + 0.3 ∗ 2 ∑ 푐푖 휋푃 = 푒푃 + 0.1 ∗ 2 ∑ 푐푖
If the sum of contributions is below the goal in P2, payoffs are:
휋퐴푖 = 푒 − 푐푖 휋푃 = 푒푃
The difference between the two rounds is the minimum amount of points that need to be contributed to realizing a payoff from the public good. In one round, this amount is 39 points (low goal, project P2.L); in the other round, it is 54 points (high goal, project P2.H). To avoid possible order effects, half of the subjects played the round with the low goal first and then the round with the high goal and the other half vice versa. Apart from the different goals in P2, both rounds are identical.
Similar to Rauchdobler et al. (2010), we use specific amounts of 39 and 54 points for two reasons. First, goal levels should be divisible by three to ease coordination, letting agents calculate their “fair share” as a focal point for contributions. Second, the medium goal should be attainable even if one agent refrains from contributing, whereas the high goal should only be attainable if all three agents contribute.
20 When equipping the principal with the same endowment as the agents, the latter might refrain from contributing to not be worse off than the non-contributing principal, as observed for contributing players by Engel and Rockenbach (2014). Giving the principal no endowment at all, on the other hand, might let agents contribute out of pity to spare the principal to leave the experiment empty-handed. We decided for an endowment of 10 points to mitigate both of these effects.
Part III 79
After the first round, players receive no feedback about the other players’ decisions but immediately play the second round. Players then get to fill out a questionnaire21 with which we elicit sociodemographic data as well as attitudes regarding trust, reciprocity, and risk, using questions from the German socio-economic panel study (Richter et al. 2013). In the end, players are informed about their payoffs from both rounds.
16.2 The treatments
The determination of the project is made in one of two ways, depending on the treatment. In treatment AV, the agents determine the project by an anonymous majority vote. In treatment PD, on the other hand, the principal decides whether the agents contribute to P1 or P2. In both treatments, agents are not informed right away about the result of the vote or the principal’s decision. Instead, we used a strategy method to elicit an agent’s contribution decisions for all possible voting outcomes of the other two agents in AV or both possible decisions of the principal in PD, respectively.22
Furthermore, we ask for the agents’ beliefs for all possible scenarios on how much the other two agents would contribute on average. We incentivize beliefs so that agents can earn up to 5 additional points for a good estimation.
16.3 Procedures
We programmed the experiment with z-Tree (Fischbacher 2007) and recruited participants with ORSEE (Greiner 2015). The experiment took place in the Aachen laboratory for experimental economics (AIXperiment), with a total of 212 participants (mean age 24.9, 41% female). The average payoff was €11.10, and sessions lasted around 60 minutes.
At the beginning of the experiment, instructions for the first round were read out loud by the experimenter, after which participants could privately ask questions. After the first round, instructions for the second round were given in the same manner. At the end of the experiment, participants were paid out individually.
21 See Appendix C.2 for the original questionnaire used in this study.
22 Appendix C.3 shows the different scenarios and how we used them for our analysis.
Part III 80
17 Hypotheses
The goal level in P2 provides an additional point of reference for contributions between contributing nothing or everything. In the following, we call this amount the “fair share,” i.e., the project-specific goal divided by the number of agents in the group (13 and 18 points in P2.L and P2.H, respectively). Figure 11 shows an agent’s best response function for the different projects, i.e., the payoff maximizing contribution amount contingent on the other agents’ contributions.
Figure 11: An agent’s best response, depending on the other agents’ contributions.
If an agent wants to maximize her payoff, her best strategy in P1 is to contribute nothing. This is also true for P2.L and P2.H if the other agents contribute nothing as well. However, if they each contribute their fair share, the agent will maximize her payoff in P2.L and P2.H by also contributing her fair share. Appendix C.4 further illustrates the different payoff scenarios for an agent, depending on the other agents’ contributions.
Although it is payoff-maximizing to contribute nothing in public good games without threshold, like P1, often contributions of about half the endowment are made in these games, due to other-regarding preferences such as pro-social behavior (see, e.g., Andreoni 1995,
Part III 81
Fischbacher et al. 2001). Still, an agent could perceive the other agents’ preference for P1 as a signal for their unwillingness to contribute, so that she also refrains from contributing to P1.
While payoff-maximizing and pro-social motives are at odds for an agent in P1, they are aligned in P2 if she believes the other two agents contribute their fair share. While contributing the fair share then not only maximizes her payoff, the agent also ensures that the other agent’s contributions are not lost and that the principal receives a payoff from the project as well. So even if an agent prefers P1 and initially did not intend to contribute, she might contribute her fair share if P2 is implemented. We, therefore, hypothesize that if agents can vote for their project, they contribute more to P2 compared to P1.
Hypothesis 1: If agents set a goal for themselves, they contribute more compared to a no- goal setting. (Comparison of contributions to P1 and P2.L/P2.H in AV)
If the principal decides which project agents contribute to, her choice might not reflect the preference of the agents. Furthermore, even if P2 provides a focal point for contributions, each agent knows nothing about the project preference of the other agents. If the other agents do not contribute to P2, an agent’s contribution is lost. Even if one other agent contributes to P2, this might not suffice to generate a payoff from the project, making contributions to P2 especially risky.
Additionally, the principal’s choice of P2 can be perceived by the agents as a sign of distrust since the principal seems to demand them to contribute at least their fair share to reach the goal, compared to a free contribution decision in P1. Although the principal cannot enforce a minimum contribution by choosing P2, the agents nevertheless might feel controlled by the principal and contribute less in P2 than they would otherwise have contributed in P1 (Falk and Kosfeld 2006).
Hypothesis 2: If the principal chooses a goal for the agents, they contribute less than if she chooses a no-goal setting. (Comparison of contributions to P1 and P2.L/P2.H in PD)
Part III 82
Based on our argumentation above that agents contribute more if they set a goal for themselves compared to a goal-free setting, and that they contribute less if the principal chooses a goal for them compared to a goal-free setting, we derive the following hypotheses.
Hypothesis 3a: If agents vote for a no-goal setting, they contribute less than when the principal chooses the no-goal setting. (Comparison of contributions to P1 in AV and PD)
Hypothesis 3b: If agents vote for a goal, they contribute more than when the principal chooses the goal for them. (Comparison of contributions to P2 in AV vs. PD)
Note that we do not form hypotheses regarding the different goal levels and their (potentially distinct) effects on group performance. We refrain from doing so since there is no clear indication in the literature to form a basis for such a statement. In general, as described in goal setting theory (Latham and Locke 2007) and observed for individual goal setting (Mento et al. 1987), higher goals lead to higher performances. For groups, however, this relation is not so clear. The meta-analysis of Kleingeld et al. (2011) already revealed that there is no significant difference in the performance-enhancing effect of moderately difficult and difficult group goals. Furthermore, experimental studies closely related to our study either also report no significant difference between these goal levels (Rauchdobler et al. 2010) or even find that groups contribute less with high goals, compared to medium goals (Dannenberg 2015b). Although we do not anticipate how the different goal levels might affect group performance in our specific setting, based on the previous literature, we assume that the difficulty of a goal can be a decisive factor in this regard and therefore implemented a medium goal as well as a high goal in our first study.
Part III 83
18 Results
In this section, we test our hypotheses by comparing the agents’ contribution decisions for the different projects, both within and across treatments. We furthermore analyze the agents’ voting behavior, the success rates for P2, and the agents’ beliefs about the contributions of their team members.
18.1 Average contributions
First, we look at the agents’ contributions and compare them for settings with and without a goal. Figure 12 shows the average contribution decisions in AV and PD, for P1, P2.L, and P2.H, respectively.
Figure 12: Average contribution decisions in AV and PD.
Part III 84
When agents can vote for a project, they contribute more if the elected project has a goal, compared to a no-goal setting. On average, they contribute 8.9 points in P1, which is less than in P2.L, where they contribute 10.7 points. This difference in contributions, however, is not significant (Wilcoxon-matched-pairs test (WMP), two-sided, p = 0.175), but contributions are even higher in P2.H with 11.1 points, which is significantly more than in P1 (WMP, two-sided, p = 0.005).
Result 1: Agents contribute significantly more if they set themselves a high goal, compared to a no-goal setting.
When a principal decides which project agents contribute to, they contribute less if the chosen project has a goal, compared to a no-goal setting. In P1, agents contribute 11.0 points, whereas they contribute 10.3 points in P2.L and only 7.2 points in P2.H. The difference in contributions between P1 and P2.L is not significant (WMP, two-sided, p = 0.484), but highly significant between P1 and P2.H (WMP, two-sided, p = 0.001). Furthermore, the difference in contributions is also weakly significant between P2.L and P2.H (10.3 vs. 7.2 points, WMP, two-sided, p = 0.058).
Result 2: Agents contribute significantly less if the principal chooses a high goal for them, compared to when she chooses a low goal or a no goal at all.
Now, we look at the possible effects of the goal setting mechanism. Do agents behave differently if they choose a project themselves compared to if the principal decides for them? In the case of a no-goal setting, agents contribute less in AV compared to PD, with the difference being weakly significant (8.9 vs. 11.0 points, Mann-Whitney-U test (MWU), two- sided, p = 0.063).
Result 3a: If agents select a no-goal setting by vote, they contribute less than when the principal chooses a no-goal setting for them.
If the chosen project has a goal to reach, the agents’ cooperative behavior depends on the goal level. In P2.L, there is almost no difference in contributions between AV and PD (10.7 vs. 10.3 points, MWU, two-sided, p = 0.873). In P2.H, on the other hand, agents contribute significantly more in AV compared to PD (11.1 vs. 7.2 points, MWU, two-sided, p = 0.003).
Result 3b: If agents select a high goal by vote, they contribute significantly more than when the principal chooses a high goal for them.
Part III 85
18.2 Individual contributions
Figure 13 shows the proportion of agents who made specific contributions in P1, P2.L, and P2.H, respectively, separately for both treatments. The general pattern of higher (lower) contributions when there is a high goal in AV (PD) compared to a goal-free setting is also reflected in the individual contributions, e.g., the share of contributions at the lower or upper end of possible contribution amounts. Looking at free-riders who do not contribute anything to the project and agents who contribute their whole endowment, we find the following discrepancies. While the proportion of free-riders in AV is similar for P1 and P2.H (27% and 33%), the share of agents who contribute all their 20 points is about twice as high in P2.H with 37% compared to only 18% in P1. In PD, on the other hand, 30% of the agents contribute all their points in P1, while only 20% do so in P2.H. The most prominent contrast can be observed in the share of free-riders, which is only 19% in P1 compared to 52% in P2.H. In other words, if the principal chooses a high goal for the agents, more than half of them refrain from contributing even a single point.
Figure 13: Proportion of contribution amounts for the projects in AV (left) and PD (right).
Part III 86
Overall, the contribution patterns look quite similar for both treatments. Apart from the most frequent contribution amounts of 0 and 20 points, agents also show a preference for a few other amounts, depending on the project. In P1, these amounts are 5, 10, and 15 points, which seem to act as natural focal points. In P2.L and P2.H, more agents are contributing either nothing or everything, and there are fewer contributions in between, with a slight increase of contributions at the fair share amount of the projects (13 points in P2.L and 18 points in P2.H).
18.3 Voting behavior and success rates
Table 14 shows the distribution of the agents’ votes and the principals’ decisions for either P1 or P2 and how often they reached the goal if P2 was carried out. Looking at the voting behavior, we see that agents prefer P1, while principals seem to have no preference for either P1 or P2. When agents can vote, 75% of the groups (21 out of 28) vote for P1 if the alternative is P2.L, with a quite similar proportion (71%, 20 of 28 groups) voting for P1 if the alternative is P2.H. Principals, on the other hand, show no such preference, deciding for P1 in 56% and 48% of the cases when the alternative is P2.L or P2.H, respectively. Overall, agents show a significant preference for P1 over P2 (Binomial probability test (BPT), two-sided, p < 0.001), while principals show no inclination for either project (BPT, two-sided, p = 0.888).
Table 14: Project choices and success rates. Absolute frequencies in parentheses.
Treatment P1 P2.L Success rate P1 P2.H Success rate
75% 25% 57% 71% 29% 13% Agents vote (AV) (21) (7) (4) (20) (8) (1)
Principal decides (PD) 56% 44% 45% 48% 52% 0% (14) (11) (5) (12) (13) (0)
66% 34% 50% 60% 40% 5% Overall (35) (18) (9) (32) (21) (1)
Part III 87
Across treatments, we observe similar success rates for P2.L and P2.H, respectively. Comparing the overall success rates for P2.L and P2.H, however, we observe a stark contrast. While the goal could be reached in P2.L half of the time (9 out of 18 cases for both treatments), the goal in P2.H was only reached once in 21 cases.
18.4 Beliefs about other agents’ contributions
In each of the two rounds, agents were asked how much they would contribute to P1 or P2 and should furthermore submit their belief on how much they think the other agents would contribute to these projects. Combining the data allows checking for possible relations between an agent’s belief and her contribution, and how these relations might be different for P1 and P2. Table 15 shows the corresponding data for AV and PD.
Table 15: Average beliefs and contributions.
Agents vote (AV) Principal decides (PD) P1 P2_L P2_H P1 P2_L P2_H Belief 9.5 12.2 13.2 9.9 10.8 8.7 Contribution 8.9 10.7 11.1 11 10.3 7.2 Difference -1.6 -1.5 -2.1 1.1*** -0.5 -1.5* Correlation 0.829 0.772 0.651 0.666 0.671 0.679
* Significant at 10%, *** at 1%.
In both treatments and for all projects, we find a moderate to strong positive correlation (0.65 < 푟 < 0.83) between an agent’s belief about the other agents’ contributions and her contribution. Interestingly, agents in PD contribute significantly more in P1 than they believe their fellow agents would contribute (WMP, two-sided, p < 0.001), whereas in P2.H, they contribute less (WMP, two-sided, p = 0.059). In P2.L as well as in all projects in the AV treatment, agents also contribute on average less than they believe their fellow agents would contribute, but not significantly.
Part III 88
19 Follow-up Study
So far, we have investigated how the cooperative behavior of agents might differ if a principal chooses a project for them, compared to when they can decide for themselves, and how such differences might depend on the difficulty of the project’s goal. The precedent analysis, however, is based on two one-shot decisions by the agents. More realistically, members of a team cooperate repeatedly. Also, a decision rule for goal setting might already be in place. How, then, do different goal setting mechanisms work if used repeatedly, and how might a change of the mechanism influence cooperative behavior? To tackle these questions, we conducted an additional study described in this chapter.
19.1 Experimental design and procedures
We derived the experimental design from our main study described in section 16, with the following adjustments. Instead of two independent rounds without feedback and a rematching of the subjects to the groups between rounds, subjects now play ten consecutive rounds, remain in the same group and receive the following information in each round: the chosen project, total contributions to the project, payoff from the project, and total earnings from the current round.
In each round, subjects choose between the projects P1 (no goal) and P223 (high goal), using the mechanisms according to the treatments AV and PD of the main study, with each mechanism being used for five consecutive rounds. Our two new treatments differ in the order in which the mechanisms are implemented. In treatment AP, the agents vote for the project in rounds 1 to 5, and the principal chooses a project in rounds 6 to 10, whereas the order in the treatment PA is vice versa.
23 As P2.H of the main study, P2 has a goal of 54 points. Since we found little or no effects for P2.L in within or across treatment comparisons, we decided to only use a P2.H equivalent in this follow-up study.
Part III 89
19.2 Results
Contributions
Figure 14 and Figure 15 show the average contribution for every round, as well as the proportion of groups in which P2 was carried out.
Figure 14: Average contribution (line chart, left scale) and P2 ratio (bar chart, right scale) for AP.
Figure 15: Average contribution (line chart, left scale) and P2 ratio (bar chart, right scale) for PA.
Part III 90
First, we investigate if and how the goal setting mechanisms influence cooperative behavior when used repeatedly. To do so, we compare the average contributions from the first and last round in which each mechanism is used. In AP, average contributions in rounds 1 and 5 are on the same level (11.6 points), but there is a significant difference between those in rounds 6 and 10 (13.2 vs. 10.0 points, WMP, two-sided, p = 0.024). In both parts of PA, contributions are significantly lower in the last round a mechanism is used compared to the first round: agents contribute 12.8 points in round 1 and only 7.2 points in round 5 (WMP, two-sided, p = 0.014), with a similar decline in contributions from round 6 to 10 (11.8 vs. 7.1 points, WMP, two-sided, p = 0.006).
Comparing rounds 1 and 10, we observe that contributions in AP remain on a similar level (11.6 vs. 10.0 points, WMP, two-sided, p = 0.470), while they significantly decline in PA (12.8 vs. 7.1 points, WMP, two-sided, p = 0.047).
Result 4a: When agents can determine their goal by vote, contributions remain stable over time if no other goal setting mechanism was used before.
Result 4b: When the principal chooses a goal for the agents, contributions decline over time.
Looking at average contributions in part 1 and part 2 of the game (rounds 1-5 and 6-10, respectively), we find that in AP, contribution amounts are almost identical in both parts (11.7 vs. 11.8 points, WMP, two-sided, p = 0.875). In PA, we observe slightly lower contributions in the second part, although the difference is not significant (10.1 vs. 9.1 points, WMP, two-sided, p = 0.333). Overall, average contributions in AP are higher than in PA, but not significantly (11.8 vs. 9.6 points, MWU, two-sided, p = 0.305).
Result 5: Changing the goal setting mechanism does not result in different contribution levels.
Part III 91
So far, we reported average contributions independent from the project the contributions were made to since groups could choose either P1 or P2 and therefore, each group may have played a unique sequence and/or frequency of the two projects over rounds. However, to cover the agents’ behavior thoroughly and compare it to the results of our first study, we also look at the contributions separately for P1 and P2. Table 16 shows the average contributions and frequencies for both projects in Part 1 (rounds 1-5), Part 2 (rounds 6-10), as well as for all ten rounds combined.
Table 16: Average contributions to and frequencies of P1 and P2.
Part 1 Part 2 Overall Contribution Frequency Contribution Frequency Contribution Frequency P1 10.0 70% 8.6 30% 9.6 50% AP P2 15.8 30% 13.2 70% 14.0 50% P1 8.8 54% 6.8 74% 7.7 64% PA P2 11.5 46% 15.4 26% 12.9 36%
The preferences for either project in this repeated setting of our follow-up study are similar to those observed at the one-shot decisions in our first study. When agents can vote, i.e., in Part 1 of AP and Part 2 of PA, they vote for P1 in 72% of the cases, compared to 71% in our first study (see Table 14). Principals, on the other hand, show a preference for P2 and choose it with a frequency of 58%, compared to 52% in our first study. While the agents’ preference for P1 is almost identical in both treatments, project preference differs for principals, who choose P1 with a frequency of 54% in PA and only 30% in AP.
On average, agents contribute more to P2 than to P1, overall as well as in the different parts of each treatment. To further explore possible influences on agents’ contributions, we conduct a regression analysis for both parts of each treatment. Table 17 shows the corresponding results for the first (I) and second (II) part of AP, as well as for the first (III) and second (IV) part of PA.
Part III 92
Table 17: Determinants of contributions.
(I) (II) (III) (IV) Dependant variable: AP AP PA PA agent’s contribution Part 1 Part 2 Part 1 Part 2
(1) Other agents‘ average 0.308*** 0.782*** 0.114 0.260*** contribution in t-1 (0.0729) (0.0993) (0.133) (0.0965) (2) Project 2 dummy 6.925*** -2.257 -0.466 6.466** (2.026) (1.440) (1.604) (2.538) (3) Risk 1.560*** -0.263 0.0804 0.129 (0.553) (0.525) (1.213) (0.620) (4) Trust -0.413 -0.102 1.012 0.420 (1.487) (0.974) (2.472) (2.307) (5) Positive reciprocity -1.763*** -0.352 2.495 -0.298 (0.597) (0.507) (2.193) (1.562) (6) Negative reciprocity -2.142*** -1.122** -1.091 -0.847 (0.514) (0.537) (1.092) (0.734) (7) Round -0.0456 -0.368 -1.134*** -0.763 (0.394) (0.315) (0.384) (0.545) (8) Constant 17.41*** 13.15 -2.825 13.55 (6.282) (8.837) (19.81) (11.16)
Observations 168 168 120 120 R² overall 0.474 0.688 0.106 0.349 ** Significant at 5%, *** at 1%. Robust standard errors in parentheses.
Columns (I) and (IV) show how an agent’s contribution is influenced when agents can vote on their goal. Agents’ contributions correlate positively with their group members’ contributions in the previous round (1) and they contribute significantly more to P2 than to P1 (2). Risk (3) and reciprocity (5, 6) also influence an agent’s contribution, but only when no other goal setting mechanism was used before (I).
When the principal decides on a goal for the agents (columns II and III), the negative coefficient of the round (7) reflects the significant decline in contributions over rounds if the agents experienced no other goal setting mechanism beforehand (III). If they already had the opportunity to vote on their goal before (II), the other agents’ contributions (1), as well as an agent’s negative reciprocity (6), still have a significant impact on an agent’s contribution decision.
Part III 93
Voting behavior
As already observed in the main study, agents significantly prefer P1 over P2 (BPT, two-sided, p < 0.001). In 86 out of the total 120 votes in AP and PA, the majority voted for P1, compared to only 34 votes for P2 (72% and 28%, respectively). Principals, on the other hand, show a significant preference for P2 (BPT, two-sided, p = 0.035), choosing it in 72 out of the 120 decisions (60%), and opting for P1 only in 48 cases (40%). This disparity in project preference is especially noticeable in the AP treatment, where the principals receive the decision right in the second half of the game. While agents voted for P1 in 70% of the cases in the first half, principals afterward only chose P1 30% of the time.
Success rates in P2
In the AP treatment, agents reached the goal in P2 in 52.4% of the cases when they voted for P2. The success rate remained on a similar level when the principal could decide on the project and only slightly increased to 57.1%. In PA, on the other hand, the goal in P2 was only reached in 21.7% of the cases when the principal decided for P2, whereas the success rate was much higher at 53.9% when agents could determine the project by majority vote.
Payoffs
Comparing the average payoffs per round in AP and PA, we find no significant difference, neither overall (24.0 vs. 21.1 points, MWU, two-sided, p = 0.349), nor separately for agents (26.6 vs. 23.3 points, MWU, two-sided, p = 0.349) or principals (16.1 vs. 14.3 points, MWU, two-sided, p = 0.292). The lowest payoffs for both player types result from the PA treatment when the principals decide for P2. Table 18 gives a detailed overview of the payoffs for agents and principals, depending on the goal setting mechanism, the project, and the treatment.
Table 18: Average payoffs for agents and principals.
Agents vote Principal decides
Agents Principals Agents Principals
P1 P2 P1 P2 P1 P2 P1 P2
AP 28.0 22.5 16.0 16.1 26.9 26.8 15.2 16.7 PA 25.5 22.4 14.1 15.9 27.1 15.9 15.3 12.5
Part III 94
20 Discussion
Contributions
Compared to many other studies on goal setting for groups (for a meta-analysis, see Kleingeld et al. 2011), our experimental setup contains several crucial differences potentially hindering cooperation: goals come with the risk of losing contributions, contributions are made anonymously, and the project choice takes place indirectly by vote and without communication. The fact that we still observe significantly higher contributions when agents choose a high goal for themselves underlines the strong effect goal setting can have on team performance. Surprisingly, we observe the opposite when the principal sets the goal for the agents: if she chooses the high goal, agents contribute significantly less than if she chooses a goal-free setting. What are the possible reasons for this detrimental effect of the principal’s goal setting?
In our study, goal assignment is not previously determined by the experimenter or a random selection but executed by the principal. Agents might refrain from trying to reach a goal set by the principal for various reasons. First, the principal does not contribute to the project but still profits from its payoff. Agents might consider this unfair and may not want to support the principal with their contributions. Second, the principal is not legitimized in any form but randomly chosen for the role; therefore, agents might not feel obliged to follow her goal setting. Furthermore, unlike with the majority vote, an agent does not know anything about the project preferences of her fellow agents if the principal chooses the project. With such a high degree of uncertainty, the agent might rather keep her endowment than risk losing her contribution if the goal is not met.
Additionally, if a principal chooses a high goal for the agents, they could perceive this choice as a sign of distrust, since it entails a demand of the principal to contribute at least a certain amount. Feeling mistrusted and potentially limited in their freedom of choice, agents, as a result, might contribute less than without that goal. In this way, principals suffer from a form of the hidden cost of control (Falk and Kosfeld 2006).
Contrary to the principal’s goal setting, we observe a positive effect on contributions if agents can vote for their goal: contributions are higher if the agents set a high goal for themselves, compared to when they vote for a goal-free project. What may be possible causes for this effect? First, through the election result, each agent learns something about the other agents’ goal preferences; it has a signaling aspect. Independent of her preference, an agent
Part III 95 might contribute high if a goal is voted for by the other agents or contribute low to waste no points if the others voted against the goal. Second, even if the agents are free in their contribution decision, the result of a joint vote can have a binding character, which may lead them to adjust their contribution according to the voting result.
We also observe a noteworthy difference between treatments if agents contribute to the project without a goal (P1). When agents vote for P1 themselves, they contribute on average less than they believe their team members will do. If the principal chooses P1 for them, however, contributions are even higher than their beliefs about the other agents’ contributions. An explanation could be that agents want to reward the leaders’ trust and possibly compensate for the presumably low contributions of the others.
In our follow-up study, we investigate the effects of different goal setting mechanisms if they are used repeatedly. Interestingly, we find in both treatments that the contribution pattern of the second half almost exactly mirrors the first half, although the pattern itself is different between treatments. In AP, contributions are quite stable and on a similar level in the first and second half, although the relation of chosen projects is quite the opposite: while agents strongly prefer P1, principals choose P2 most of the time. In PA, on the other hand, contributions substantially decline in both halves. Principals initially prefer P2, but in rounds 3-5, the majority chooses P1 for their agents. Again, as in AP, most groups implement P1 when agents can vote in rounds 6-10.
Considering that contribution patterns within both halves are very much alike, one could speculate if the goal setting mechanism is the decisive factor for the agents’ cooperative behavior. An alternative explanation is that the learning from the first half shapes the agents’ expectations and behavior for the second half. If they already experienced either stable or declining contributions in the first half, agents could feel inclined to repeat these patterns in the second half as well.
We further observe a difference in success rates for P2. When the principal chooses the goal for the agents in PA, where the groups have no experience with the game, they reach the goal only in about 22% of the cases. In AP, on the other hand, after the agents voted for the goal themselves and reached it more than half of the time, the success rate even slightly increases when the principal chooses the goal for the agents in the second part. Interestingly, when the agents can vote for the goal, success rates are almost identical in AP and PA,
Part III 96 although in the former they have no prior experience at all, and in the latter agents beforehand failed to reach the goal most of the time.
Voting behavior
Choosing P2 is risky since there is no payoff from the project if the goal is not met. Agents seem to anticipate this risk, with more than 70% of the groups voting for P1, even if the goal in P2 is low and relatively easy to reach. This preference for P1 could be based on a lack of trust in the group members to contribute their fair share, which appears reasonable since they are anonymous and only cooperate once. The fear of being the “sucker” who contributes to reaching a goal while the other group members do not might explain the preference for P1.
Choosing P2 is even riskier for the principal since she would already profit from low contributions in P1 but would receive no payoff in P2 if the goal is not met. Particularly the choice of the high goal alternative of P2 seems not very comprehensible at first, offering little chance of being successful and resulting in a payoff. Maybe principals wanted the choice of the high goal to be motivating to the agents to contribute very high and therefore generating the biggest possible payoff for the principals. Whatever the motivation, choosing P2.H backfired, and not one of the twelve groups where the principal chose the high goal succeeded in reaching it, resulting in zero payoffs for the principals.
Limitations
This study follows an explorative approach to shed some light on the effects of different ways of goal setting in a principal-agent setting with risky projects. Although we observe significant differences in group performance, a clear connection to the underlying elements of goal setting is not feasible. We cannot attribute the difference in agents’ contributions to either the decision mechanism (majority vote vs. single decider) or the source (contributing vs. non-contributing player) of goal setting since both aspects are different between our treatments. In their meta-analysis on goal setting and group performance, Kleingeld et al. (2011) already found that participation did not influence the effect of goal setting, suggesting that the source of goal setting might be the driving force behind the contribution differences we observe in our study. To disentangle the effects of the mechanism and the source of goal setting, further research is necessary.
Part III 97
21 Conclusion
In our first study, we investigated how different goal setting mechanisms influence cooperative behavior in a principal-agent setting and how this influence might depend on the difficulty of the goal. In line with goal setting theory, we find that high goals lead to high contributions. Interestingly, this is only the case if the agents vote for the goal themselves. If the principal chooses a high goal for the agents, they contribute less than in a goal-free setting. Furthermore, we observe these significant differences in contributions solely when agents contribute to a goal that is hard to reach. When the goal is only moderately difficult, we find no significant differences in contributions compared to a goal-free setting.
A change from one goal setting mechanism to the other yields no change in contribution levels, as shown by our follow-up study. When the decision rights are transferred from one party to the other, agents contribute in a similar way they did before the change. The results further indicate that goal setting by the principal does not lead to stable contribution levels over time. If the agents of a newly formed team can vote for their goals, however, their contributions remain stable.
In general, transferring results from a lab experiment to a real-life scenario should be done very cautiously. Important factors potentially influencing the variable of interest are cut out in a lab experiment to derive causal relationships. Based on our results, one could argue that leaders or managers should be careful with setting overly ambitious goals for their teams, for it may be detrimental to their performance. Putting trust in the team members and not setting a specific goal, on the other hand, might result in a good performance and keep the number of free riders quite low.
Part III 98
Appendix C
C.1 Instructions
[Basic instructions. Identical for both AV and PD.]
General information
Welcome to today's experiment. Keep calm during the whole experiment. Switch off your mobile phones in order not to disturb the procedure. Please never ask questions out loud. If you have any questions, please raise your hand. We will then come to your place. Please read the following carefully.
The experiment
In this experiment there are two roles, Player A and Player B. At the beginning of the experiment, the role you will have for the duration of the whole experiment is determined randomly.
All participants of the experiment are randomly assigned to groups, each consisting of three Player A and one Player B. In this part of the experiment, two rounds will be played. After the first round, all groups are reassembled, ensuring that none of the participants will be in a group with players who have already played with him in the first round.
At the beginning of each round, each Player A receives 20 points, each Player B 10 points.
The following conversion rate applies: 5 points = 1 Euro
Your income in this part results from the sum of the two round incomes:
Income = Income from round 1 + Income from round 2
Part III 99
[Instructions for AV treatment with low goal.]
At the beginning of this round, the three Player A vote for Project 1 or Project 2. The project that receives the majority of the votes is carried out. Neither player will know how the other two players decided.
In Project 1, all Player A can contribute an integer amount between 0 and 20 of their endowment to a project. Points that you do not invest in the project are kept to yourself. Each point contributed to the project by the members of a group is multiplied by 2 - from this group result, each Player A receives 30%, Player B receives 10%.
In Project 2 all Player A can contribute an integer amount between 0 and 20 of their endowment to a project. Points that you do not invest in the project are kept to yourself. The project will only be successful if at least 39 points have been contributed.
In the case of a successful Project 2, each point contributed by the members of the group to the project is multiplied by 2 - of this group result each Player A receives 30%, Player B receives 10%. If less than 39 points are invested in Project 2, the project will not be realized. In this case, the contributions will be lost.
Player A only: Please indicate how many points you would contribute to each project if you knew how the other two Player A in your group would have voted.
Player A and Player B: At the end of the round, we will ask you to estimate the average contribution of the (other) Player A in your group. Here, you can earn up to 5 additional points.
Part III 100
[Instructions for PD treatment with low goal.]
At the beginning of the round, Player B decides whether Project 1 or Project 2 will be carried out. All Player A indicate which of the projects they prefer.
In Project 1, all Player A can contribute an integer amount between 0 and 20 of their endowment to a project. Points that you do not invest in the project are kept to yourself. Each point contributed to the project by the members of a group is multiplied by 2 - from this group result, each Player A receives 30%, Player B receives 10%.
In Project 2 all Player A have the possibility to contribute an integer amount between 0 and 20 of their endowment to a project. Points that you do not invest in the project are kept to yourself. The project will only be successful if at least 39 points have been contributed.
In the case of a successful Project 2, each point contributed by the members of the group to the project is multiplied by 2 - of this group result each Player A receives 30%, Player B receives 10%. If less than 39 points are invested in the Project 2, the project will not be realized. In this case the contributions will be lost.
Player A only: Please indicate how many points you would contribute to each project if you knew how the other two Player A in your group would have voted.
Player A and Player B: At the end of the round, we will ask you to estimate the average contribution of the (other) Player A in your group. Here, you can earn up to 5 additional points.
Part III 101
C.2 Questionnaire
Please fill out the following questionnaire.
• Age: • Sex: • Nationality: • Course of study: • Desired degree: • Semester: • Disposable income per month: • Number of acquaintances in this experiment: • Number of friends in this experiment:
What is your opinion on the following statements? [Trust. Scale from 1 (Completely agree) to 4 (Completely disagree)]
• In general, you can trust people. • You can't rely on anyone these days. • When dealing with strangers, it's better to be careful before you trust them.
To what extent do the following statements apply to you personally? [Reciprocity. Scale from 1 (Not at all) to 7 (Totally true)]
• If anyone does me a favor, I'm willing to return it. • If I am severely wronged, I will take revenge at all costs at the next opportunity. • When someone gets me into a difficult situation, I'm gonna do the same to him. • I am willing to incur costs to help someone who has helped me before. • If someone insults me, I will also behave insultingly towards them.
How do you personally assess yourself: are you generally a risk-seeking person or do you try to avoid risks? [Risk. Scale from 1 (Not willing to take any risks) to 7 (Very risk-seeking)]
How would you assess your attitude towards risk in relation to the following areas?
• Driving? • Investing? • Leisure and sports? • Your professional career? • Your health? • Trusting in strangers?
Part III 102
C.3 Elicitation of Contribution Decisions
AV treatment:
(1) Project 1 is carried out. Both other Player A voted for Project 1.
How much would you contribute in this case?
(2) Project 1 (or 2, depending on the player’s choice) is carried out. One Player A
voted for Project 1, the other voted for Project 2.
How much would you contribute in this case?
(3) Project 2 is carried out. Both other Player A voted for Project 2.
How much would you contribute in this case?
PD treatment:
(1) Project 1 is carried out. How much would you contribute in this case?
(2) Project 2 is carried out. How much would you contribute in this case?
In our analysis, we used the answers to scenarios (1) and (3) of the AV treatment and the scenarios (1) and (2) of the PD treatment.
Part III 103
C.4 Payoff Scenarios
Figure 16Figure 18 show possible payoffs for an agent in the different projects. Within the figures, each line refers to one of three contribution behaviors of the other two agents:
• No contribution: The other agents contribute nothing (0 points each)
• Fair share: The other agents contribute the project specific goal divided by the number of agents in the group (13 and 18 points each in P2.L and P2.H, respectively)
• Full contribution: the other agents contribute their whole endowment (20 points each)
Figure 16: Scenarios for an agent’s payoff in P1, depending on her contribution and the other agents’ contributions.
Part III 104
Figure 17: Scenarios for an agent’s payoff in P2.L, depending on her contribution and the other agents’ contributions.
Figure 18: Scenarios for an agent’s payoff in P2.H, depending on her contribution and the other agents’ contributions.
105
22 Bibliography
Andersen, Steffen; Ertaç, Seda; Gneezy, Uri; Hoffman, Moshe; List, John A. (2011). Stakes matter in ultimatum games. American Economic Review 101 (7), 3427– 3439.
Andreoni, James (1995). Cooperation in public-goods experiments. Kindness or confusion? American Economic Review 85 (4), 891–904.
Ariely, Dan; Gneezy, Uri; Loewenstein, George; Mazar, Nina (2009). Large Stakes and Big Mistakes. Review of Economic Studies 76 (2), 451–469.
Baldassarri, Delia; Grossman, Guy (2011). Centralized sanctioning and legitimate authority promote cooperation in humans. Proceedings of the National Academy of Sciences 108 (27), 11023–11027.
Balliet, Daniel; Mulder, Laetitia B.; Van Lange, Paul A M (2011). Reward, punishment, and cooperation: a meta-analysis. Psychological bulletin 137 (4), 594–615.
Barrick, Murray R.; Day, David V.; Lord, Robert G.; Alexander, Ralph A. (1991). Assessing the utility of executive leadership. The Leadership Quarterly 2 (1), 9–22.
Böhm, Robert; Meier, Nicolas W.; Korn, Lars; Betsch, Cornelia (2017). Behavioural consequences of vaccination recommendations. An experimental analysis. Health Economics 26, 66–75.
Bolton, Gary E.; Ockenfels, Axel (2000). ERC. A theory of equity, reciprocity, and competition. American Economic Review 90 (1), 166–193.
Brandts, Jordi; Cooper, David J.; Weber, Roberto A. (2014). Legitimacy, Communication, and Leadership in the Turnaround Game. Management Science 61 (11), 2627– 2645.
Bruttel, Lisa; Fischbacher, Urs (2013). Taking the initiative. What characterizes leaders? European Economic Review 64, 147–168.
Cappelen, Alexander W.; Reme, Bjørn-Atle; Sørensen, Erik Ø.; Tungodden, Bertil (2015). Leadership and Incentives. Management Science 62 (7), 1944–1953.
Cartwright, Edward; Gillet, Joris; van Vugt, Mark (2013). Leadership By-Example in the Weak-Link Game. Economic Inquiry 51 (4), 2028–2043.
Bibliography 106
Cartwright, Edward; Patel, Amrish (2010). Imitation and the Incentive to Contribute Early in a Sequential Public Good Game. Journal of Public Economic Theory 12 (4), 691–708.
Chamberlin, Edward H. (1948). An Experimental Imperfect Market. Journal of Political Economy 56 (2), 95–108.
Charness, Gary; Gneezy, Uri; Kuhn, Michael A. (2013). Experimental methods. Extra- laboratory experiments-extending the reach of experimental economics. Journal of Economic Behavior & Organization 91, 93–100.
Chaudhuri, Ananish (2011). Sustaining cooperation in laboratory public goods experiments. A selective survey of the literature. Experimental Economics 14 (1), 47–83.
Choi, Jung-Kyoo; Ahn, T. K. (2013). Strategic reward and altruistic punishment support cooperation in a public goods game experiment. Journal of Economic Psychology 35, 17–30.
Cookson, Richard (2000). Framing effects in public goods experiments. Experimental Economics 3 (1), 55–79.
Cummings, Ronald G.; Holt, Charles A.; Laury, Susan K. (2004). Using laboratory experiments for policymaking. An example from the Georgia irrigation reduction auction. Journal of Policy Analysis and Management 23 (2), 341–363.
Dannenberg, Astrid (2015a). Leading by example versus leading by words in voluntary contribution experiments. Social Choice and Welfare 44 (1), 71–85.
Dannenberg, Astrid (2015b). Non-binding agreements in public goods experiments. Oxford Economic Papers 68 (1), 279–300.
Dawes, Robyn M. (1980). Social dilemmas. Annual review of psychology 31 (1), 169–193.
Dickson, Eric S.; Gordon, Sanford C.; Huber, Gregory A. (2015). Institutional Sources of Legitimate Authority. An Experimental Investigation. American Journal of Political Science 59 (1), 109–127.
Dinh, Jessica E.; Lord, Robert G.; Gardner, William L.; Meuser, Jeremy D.; Liden, Robert C.; Hu, Jinyu (2014). Leadership theory and research in the new millennium.
Bibliography 107
Current theoretical trends and changing perspectives. The Leadership Quarterly 25 (1), 36–62.
Dowling, John; Pfeffer, Jeffrey (1975). Organizational Legitimacy:Social Values and Organizational Behavior. Pacific Sociological Review 18 (1), 122–136.
Engel, Christoph; Rockenbach, Bettina (2014). Give everybody a voice! the power of voting in a public goods experiment with externalities. Working Paper (2014/16).
Faillo, Marco; Grieco, Daniela; Zarri, Luca (2013). Legitimate punishment, feedback, and the enforcement of cooperation. Games and Economic Behavior 77 (1), 271–283.
Falk, Armin; Kosfeld, Michael (2006). The Hidden Costs of Control. American Economic Review 96 (5), 1611–1630.
Fehr, Ernst; Schmidt, Klaus M. (1999). A theory of fairness, competition, and cooperation. The quarterly journal of economics 114 (3), 817–868.
Fischbacher, Urs (2007). z-Tree. Zurich toolbox for ready-made economic experiments. Experimental Economics 10 (2), 171–178.
Fischbacher, Urs; Gächter, Simon (2010). Social Preferences, Beliefs, and the Dynamics of Free Riding in Public Goods Experiments. American Economic Review 100 (1), 541–556.
Fischbacher, Urs; Gächter, Simon; Fehr, Ernst (2001). Are people conditionally cooperative? Evidence from a public goods experiment. Economics Letters (71), 397–404.
Fischer, Sven; Grechenig, Kristoffel; Meier, Nicolas (2016). Monopolizing sanctioning power under noise eliminates perverse punishment but does not increase cooperation. Frontiers in behavioral neuroscience 10, p. 180.
Foss, Nicolai J. (2001). Leadership, Beliefs and Coordination: An Explorative Discussion. Industrial and Corporate Change 10 (2), 357–388.
French, John R. P.; Raven, Bertram (1959). The bases of social power. Classics of organization theory 7, 311–320.
Gächter, Simon; Nosenzo, Daniele; Renner, Elke; Sefton, Martin (2012). Who makes a good leader? Cooperativeness, optimism, and leading‐by‐example. Economic Inquiry 50 (4), 953–967.
Bibliography 108
Gächter, Simon; Renner, Elke (2018). Leaders as role models and ‘belief managers’ in social dilemmas. Journal of Economic Behavior & Organization 154, 321–334.
Gardner, William L.; Lowe, Kevin B.; Moss, Todd W.; Mahoney, Kevin T.; Cogliser, Claudia C. (2010). Scholarly leadership of the study of leadership. A review of The Leadership Quarterly's second decade, 2000–2009. The Leadership Quarterly 21 (6), 922–958.
Gneezy, Ayelet; Gneezy, Uri; Nelson, Leif D.; Brown, Amber (2010). Shared social responsibility. A field experiment in pay-what-you-want pricing and charitable giving. Science 329 (5989), 325–327.
Greiner, Ben (2015). Subject pool recruitment procedures. Organizing experiments with ORSEE. Journal of the Economic Science Association 1 (1), 114–125.
Grieco, Daniela; Braga, Michela; Bripi, Francesco (2016). Cooperation and leadership in a segregated community. Evidence from a lab-in-the-field experiment in a South African township. WIDER Working Paper 2016/76.
Grieco, Daniela; Faillo, Marco; Zarri, Luca (2017). Enforcing cooperation in public goods games. Is one punisher enough? Journal of Economic Psychology 61, 55–73.
Gürerk, Özgür; Irlenbusch, Bernd; Rockenbach, Bettina (2009). Motivating teammates. The leader’s choice between positive and negative incentives. Journal of Economic Psychology 30 (4), 591–607.
Gürerk, Özgür; Irlenbusch, Bernd; Rockenbach, Bettina (2014). On cooperation in open communities. Journal of Public Economics 120, 220–230.
Gürerk, Özgür; Lauer, Thomas; Scheuermann, Martin (2018). Leadership with individual rewards and punishments. Journal of Behavioral and Experimental Economics 74, 57–69.
Güth, Werner; Levati, M. Vittoria; Sutter, Matthias; van der Heijden, Eline (2007). Leading by example with and without exclusion power in voluntary contribution experiments. Journal of Public Economics 91 (5-6), 1023–1042.
Haigner, Stefan D.; Wakolbinger, Florian (2010). To lead or not to lead. Endogenous sequencing in public goods games. Economics Letters 108 (1), 93–95.
Bibliography 109
Hambrick, Donald C.; Quigley, Timothy J. (2014). Toward more accurate contextualization of the CEO effect on firm performance. Strategic Management Journal 35 (4), 473–491.
Hannah, Sean T.; Sumanth, John J.; Lester, Paul; Cavarretta, Fabrice (2014). Debunking the false dichotomy of leadership idealism and pragmatism. Critical evaluation and support of newer genre leadership theories. Journal of Organizational Behavior 35 (5), 598–621.
Harrell, Ashley; Simpson, Brent (2016). The dynamics of prosocial leadership. Power and influence in collective action groups. Social Forces 94 (3), 1283–1308.
Hermalin, Benjamin E. (1998). Toward an Economic Theory of Leadership. Leading by Example. American Economic Review 88 (5), 1188–1206.
Hernandez, Morela; Eberly, Marion B.; Avolio, Bruce J.; Johnson, Michael D. (2011). The loci and mechanisms of leadership. Exploring a more comprehensive view of leadership theory. The Leadership Quarterly 22 (6), 1165–1185.
Hogan, Robert; Kaiser, Robert B. (2005). What we know about leadership. Review of General Psychology 9 (2), 169–180.
Holland, Burt S.; Copenhaver, Margaret DiPonzio (1987). An Improved Sequentially Rejective Bonferroni Test Procedure. Biometrics 43 (2), p. 417.
Hollander, Edwin P. (1992). Leadership, followership, self, and others. The Leadership Quarterly 3 (1), 43–54.
Huck, Steffen; Rey-Biel, Pedro (2006). Endogenous Leadership in Teams. Journal of Institutional and Theoretical Economics (JITE) / Zeitschrift für die gesamte Staatswissenschaft 162 (2), 253–261.
Jack, B. Kelsey; Recalde, María P. (2015). Leadership and the voluntary provision of public goods. Field evidence from Bolivia. Journal of Public Economics 122, 80– 93.
Jensen, Michael C.; Meckling, William H. (1979). Theory of the firm. Managerial behavior, agency costs, and ownership structure. In: Economics social institutions: Springer, 163–231.
Bibliography 110
Joyce, William F.; Nohria, Nitin; Roberson, Bruce (2003). What really works: The 4+2 formula for sustained business success. New York: Harper Business.
Keyes, Hysom, Lupo (2000). The Positive Organization. Leadership Legitimacy, Employee Well-Being, and the Bottom Line. The Psychologist-Manager Journal 4 (2), 143–153.
King, Andrew J.; Johnson, Dominic D. P.; van Vugt, Mark (2009). The origins and evolution of leadership. Current Biology 19 (19), R911-6.
Kleingeld, Ad; van Mierlo, Heleen; Arends, Lidia (2011). The effect of goal setting on group performance. A meta-analysis. The Journal of applied psychology 96 (6), 1289–1304.
Kocher, Martin G.; Pogrebna, Ganna; Sutter, Matthias (2013). Other-regarding preferences and management styles. Journal of Economic Behavior & Organization 88, 109– 132.
Komai, Mana; Grossman, Philip J.; Deters, Travis (2011). Leadership and information in a single-shot collective action game. An experimental study. Managerial and Decision Economics 37, 119-134.
Komai, Mana; Stegeman, Mark; Hermalin, Benjamin E. (2007). Leadership and Information. American Economic Review 97 (3), 944–947.
Kotter, John P. (2012). Leading change. Boston, Mass.: Harvard Business Review Press.
Koukoumelis, Anastasios; Levati, M. Vittoria; Weisser, Johannes (2012). Leading by words. A voluntary contribution experiment with one-way communication. Journal of Economic Behavior & Organization 81 (2), 379–390.
Latham, Gary P.; Locke, Edwin A. (2007). New Developments in and Directions for Goal- Setting Research. European Psychologist 12 (4), 290–300.
Levati, M. Vittoria; Sutter, M.; van der Heijden, Eline (2007). Leading by Example in a Public Goods Experiment with Heterogeneity and Incomplete Information. Journal of Conflict Resolution 51 (5), 793–818.
Levy, David M.; Padgitt, Kail; Peart, Sandra J.; Houser, Daniel; Xiao, Erte (2011). Leadership, cheap talk and really cheap talk. Journal of Economic Behavior & Organization 77 (1), 40–52.
Bibliography 111
Locke, Edwin A.; Latham, Gary P. (1990). A theory of goal setting & task performance: Prentice-Hall, Inc.
Locke, Edwin A.; Latham, Gary P. (2002). Building a practically useful theory of goal setting and task motivation. A 35-year odyssey. American Psychologist 57 (9), 705–717.
Locke, Edwin A.; Latham, Gary P. (2006). New Directions in Goal-Setting Theory. Current directions in psychological science 15 (5), 265–268.
Mackey, Alison (2008). The effect of CEOs on firm performance. Strategic Management Journal 29 (12), 1357–1367.
Mento, Anthony J.; Steel, Robert P.; Karren, Ronald J. (1987). A meta-analytic study of the effects of goal setting on task performance. 1966–1984. Organizational Behavior and Human Decision Processes 39 (1), 52–83.
Milinski, Manfred; Rockenbach, Bettina (2012). On the interaction of the stick and the carrot in social dilemmas. Journal of theoretical biology 299, 139–143.
Moxnes, Erling; van der Heijden, Eline (2003). The effect of leadership in a public bad experiment. Journal of Conflict Resolution 47 (6), 773–795.
Nikiforakis, Nikos (2008). Punishment and counter-punishment in public good games. Can we really govern ourselves? Journal of Public Economics 92 (1-2), 91–112.
Nosenzo, Daniele; Sefton, Martin (2014). Promoting Cooperation. The distribution of reward and punishment power. Reward and punishment in social dilemmas. Series in human cooperation, 87–114.
O'Gorman, Rick; Henrich, Joseph; van Vugt, Mark (2009). Constraining free riding in public goods games: designated solitary punishers can sustain human cooperation. Proceedings. Biological sciences / The Royal Society 276 (1655), 323–329.
O'Leary-Kelly, Anne M.; Martocchio, Joseph J.; Frink, Dwight D. (1994). A review of the influence of group goals on group performance. Academy of Management Journal 37 (5), 1285–1301.
Podsakoff, Philip M.; Podsakoff, Nathan P. (2019). Experimental designs in management and leadership research. Strengths, limitations, and recommendations for improving publishability. The Leadership Quarterly 30 (1), 11–33.
Bibliography 112
Pogrebna, Ganna; Krantz, David H.; Schade, Christian; Keser, Claudia (2011). Words versus actions as a means to influence cooperation in social dilemma situations. Theory and Decision 71 (4), 473–502.
Potters, Jan; Sefton, Martin; Vesterlund, Lise (2007). Leading-by-example and signaling in voluntary contribution games. An experimental study. Economic Theory 33 (1), 169–182.
Quigley, Timothy J.; Hambrick, Donald C. (2015). Has the “CEO effect” increased in recent decades? A new explanation for the great rise in America's attention to corporate leaders. Strategic Management Journal 36 (6), 821–830.
Rauchdobler, Julian; Sausgruber, Rupert; Tyran, Jean-Robert (2010). Voting on thresholds for public goods. Experimental evidence. FinanzArchiv: Public Finance Analysis 66 (1), 34–64.
Richter, David; Metzing, Maria; Weinhardt, Michael; Schupp, Jürgen (2013). SOEP scales manual. SOEP Survey Papers.
Sahin, Selhan Garip; Eckel, Catherine; Komai, Mana (2015). An experimental study of leadership institutions in collective action games. Journal of the Economic Science Association 1 (1), 100–113.
Sturm, Bodo; Weimann, Joachim (2007). Unilateral emissions abatement. CESifo Working Paper 10 (1152).
Sutter, Matthias; Haigner, Stefan D.; Kocher, Martin G. (2010). Choosing the Carrot or the Stick? Endogenous Institutional Choice in Social Dilemma Situations. Review of Economic Studies 77 (4), 1540–1566.
Sutter, Matthias; Rivas, M. Fernanda (2014). Leadership, reward, and punishment in sequential public goods experiments. In: Reward and Punishment in Social Dilemmas, 133–160.
Torgler, Benno (2002). Speaking to Theorists and Searching for Facts. Tax Morale and Tax Compliance in Experiments. Journal of Economic Surveys 16 (5), 657–683.
Tyler, Tom R. (2006). Psychological perspectives on legitimacy and legitimation. Annual review of psychology 57, 375–400.
Bibliography 113
Tyler, Tom R.; Blader, Steven L. (2005). Can Businesses Effectively Regulate Employee Conduct? The Antecedents of Rule Following in Work Settings. Academy of Management Journal 48 (6), 1143–1158. van Vugt, Mark; Cremer, David de (1999). Leadership in social dilemmas. The effects of group identification on collective actions to provide public goods. Journal of Personality and Social Psychology 76 (4), 587–599.
Vesterlund, Lise (2003). The informational value of sequential fundraising. Journal of Public Economics 87 (3-4), 627–657.
Walker, Henry A.; Thomas, George M.; Zelditch, Morris (1986). Legitimation, Endorsement, and Stability. Social Forces 64 (3), 620–643.
Weber, Roberto A. (2015). The Economics of Effective Leadership: UBS International Center of Economics in Society at the University of Zurich.
Zelmer, Jennifer (2003). Linear Public Goods Experiments. A Meta-Analysis. Experimental Economics 6 (3), 299–310.