Organization Style, Leadership Strategy and Free-Riding B´eatrice Boulu-Reshef∗1,2 , Charles A. Holt†1 and Melissa Thomas-Hunt‡2 1 2

University of Virginia, Department of Economics

University of Virginia, Darden Graduate School of Business November 25, 2014

Abstract Mitigating free-riding in organizations is of paramount importance for increasing economic output. In most organizations, leaders are matched with followers to accomplish cooperative tasks, which often conflict with followers’ material payoffs. This paper asks the following question: how does organization style affect leaders’ choices of leadership strategy, and in turn, free-riding? We provide a theoretical framework to study the effect of organization style on leaders’ leadership strategy and report the results of controlled experiments in which we ask leaders to choose messages from a message set that induces a leadership style, and then to send them to participants at the beginning of each round of a repeated and finite horizon public goods game. The design features two leadership styles-top down and collegial-and two forms of communication-public leadership with no communication and public leadership supplemented with private communication between the leader and targeted followers. Leaders provided with collegial leadership style public messages perform better than their top down counterparts, but only when targeted private communication is not allowed. This is consistent with the economics insight on collegiality as a costly way to organize cooperation, but challenges a large literature that promotes collegial leadership. When targeted private communication is possible, top down leaders perform better than their collegial counterparts due to their increased efforts and increased use of public leadership that caters to the contributors and raises the contributions of free-riders. Leadership style affects leaders’ ability to induce a sense of obligation. We find that the key strategic prerequisite to successful leadership strategy is the leader’s ability to extract information about the existence of types of contributors combined with a targeted leadership style that caters to contributors on the basis of performancebased ad hoc coalitions. These findings have important implications for organizational design. Key words: leadership strategy, organization style, free-riding, cooperation, public goods game. JEL: C92, D23, H41, L23. ∗

[email protected] (corresponding author) [email protected] ‡ [email protected] †

1

1

Introduction

Mitigating free-riding in organizations is of paramount importance for increasing economic output. Freeriding happens because collective creation value processes are characterized by the fact that individuals face a public good dilemma: the maximization of their own monetary payoff conflicts with the accomplishment of the collective task. In such contexts, the strategy that reflects the individuals’ best interest is to free-ride (Samuelson 1954) and unless individuals are provided with leaders who work towards enlisting their support in the accomplishment of the collective task, individuals free-ride. In the economics literature, leaders are first movers. Strictly defined, they can allocate their private resources, time, effort, messages, money, towards the enlistment of the followers’ support but they do not allocate resources of the organization, otherwise they are managers1 . In the management literature, leaders are characterized by the fact that they have a vision, they communicate, they have integrity, they empower others and they execute. Both paradigms attempt to identify the best set of practices for getting others (i.e., people, organizations, industries etc.) to engage in certain practices and to implement a particular strategy in order for the organization to perform effectively and efficiently. DeRue et al. (2011) report that, combined, leader traits and behaviors explain at least 31% of the variance in leadership effectiveness. In addition, in organizations, contrasting examples of leadership styles are prevalent, as in the example of how firms have tried to enlist the support of their suppliers in improving industry standards of green practices provided by Karaer et al (2014). Using a directional strategy with little communication, Walmart asked its suppliers to answer questions regarding their sustainability practices and warned that those who chose not to participate would “probably [be] less relevant [to them]” (Rosenbloom 2009). In contrast, Nike implemented a demanding but collaborative environmental program with many of its suppliers (Plambeck et al. 2012). Such contrasts highlight the role of incentives, channels of communication and leadership styles. Although largely studied in non-incentive compatible settings, the question of the efficacy of leaders in reducing free-riding in settings in which the effects of incentives are accounted for is still a largely unanswered question. In particular, the question of how organization styles will affect individual leaders’ ability to implement the leadership strategy that is promoted by an organization is unaddressed. This paper addresses the complex question of how organization styles, which we define as the channels of communication and leadership styles that are provided to the organization members, affect leaders’ ability to mitigate free-riding. We use economics experiments to analyze the effect of organization styles on leaders’ choices of leadership strategy and the effects on free-riding. The paper relies both on the findings and the methodology of experimental economics and uses core insights from the management literatures to analyze the results in a manner that is relevant to multiple literatures. Given the importance of the phenomenon, economists and management scholars have worked toward documenting aspects of it. In the economics literature, the growing economics of leadership literature has started to study how incentives and information affects leaders’ choices and performance. In usually simple non-cooperative game-theoretical settings, this literature shows that under sufficiently incentivized 1

Note that managers may also be leaders. However, the economics perspective on leadership offers a stricter definition

of what a leader than the management literature.

2

conditions, leaders may increase the provision of effort. Also, in organizational economics, much research in the recent years has shown that management styles affect firm-level productivity, managers’ compensation and actions. Bertrand and Schoar (2003) find that ”styles” in managerial decision-making are significantly related to manager fixed effects in performance and that higher-performers receive higher compensation. Bloom and Van Reenen (2007) show that better managed firms do better and that these better managed firms are found in more competitive industries. Using field experiments, Bloom et al. (2013) document that differences in management practices across firms explain differences in productivity. Other research in economics has shown that managers’ traits affect management styles. Looking at quota policies to increase the proportion of women on corporate boards, Matsa and Miller (2013) find that firms that were affected by gender quotas had different policies compared to comparison firms regarding workforce reductions, indicating that gender affects management style. In an experimental paper, Kocher et al. (2012) show that managers who prefer efficiency are more likely to exercise an autocratic management style. Although this literature shows that management styles exist and provides strong evidence that they have an effect on major economic outcomes, the role of leadership styles, understood as sets of messages with different contents, is not identified in the economics literature. Management scholars have studied the drastic differences between leadership strategies mainly via the analysis of leader effectiveness, focusing on how the different types of leaders affect group performance and follower satisfaction. Leaders are characterized by their behaviors and by their traits given demographics, task competence, and interpersonal attributes. More recently, the focus of the literature has moved towards the quality of the dyadic relationship between leaders and followers. The literature has identified that collegial leaders tend to perform better in group settings in which members possess critical task-relevant knowledge or experience and that top down leaders, who according to the literature, perform better when there is no room for error (Eagly and Johannesen-Schmidt 2001), when monetary incentives are tangible (Bass and Avolio 1993) and when situations are stressful and/or ambiguous (Eagly and Johannesen-Schmidt 2001). These investigations have relied on the traditional research methodologies of the leadership literature, mainly surveys and experiments in which tasks are not incentive-compatible2 . With more than thirty different leadership styles and contrasting schools of thoughts, this leadership literature is spread. The state of the Art is best understood as a spectrum between functional leadership and more traditional trait leadership. Functional leadership claims that a good leader will act “situationally”, i.e., will adopt the leadership style that best matches each situation. In contrast, the leadership trait school claims that leaders’ traits dominate any attempts to adapt the style to the situation: leaders do what they are good at with little variation around their trait. The literature has had two responses to this tension. First, a preponderance of leadership styles has emerged. Second, the relative utility of different leadership styles has been questioned. The absence of consensus may be interpreted as the outcome of research efforts that were undertaken using empirical strategies that are not easily commensurable and that are, as a result, not always integrated in an encompassing literature. Across management literatures, it is difficult to compare outcomes of survey research and experiments that do not integrate monetary incentives with those that do provide monetary incentives. 2

Literature reviews can be found in Avolio et al. (2003) and Bass and Bass (2008).

3

To summarize, what is known from the economics literature is that, as Bloom et al put it: “management matters”. Managerial styles can be identified and they affect economic outcomes. Apart from the works of Kocher et al. (2012) and Bloom and Van Reenen (2007) that document the link between leader behavior, incentives and productive environments, little is known on the effect of the features of the environment on leaders’ ability to enlist the support of the followers. What is also known from the management literature is that different leadership styles exist. What is still questioned is the driver of the effect of such leadership styles. More fundamentally, what is not known is the extent to which organization styles, as opposed to behaviors or traits of the leaders, enable leaders to perform. The objective of the paper is dictated by the core of the problem of leadership. In organizations, the public good dilemma is often managed as a motivational problem, where leaders are asked to work on enlisting the support of the individuals who are engaged in collective value creation tasks. The question of the efficacy of leaders in reducing free-riding is still a largely unanswered question; the booming field of the economics of leadership is one expression of this state of the literature. In addition, the question of how likely individuals are to use leadership strategies as they are promoted in organizations is unaddressed. This paper addresses both questions. Its objective is study the effect of organization styles, defined as the channels of communication and the leadership styles that are provided to leaders in an incentivized experiment, on the leaders’ ability to enlist the support of followers an mitigate freeriding. Ultimately, it will enable the understanding of how organizations can be designed to promote successful leadership practices. This paper will attempt to meet its objective by proposing a new perspective on this question by studying the three sided-dynamic of the organization-leader-followers interaction, given different organization styles. This approach relies on a premise that is closely related to the research program of resource-based view research. The paper focuses on the role of fit between employees and managers in facilitating value creation, as in, for example, Wang et al. (2007). In strategy, the requirement to have employees exhibit high effort and make investment at a private cost cannot be leveraged, unless safeguards and trust building mechanisms are provided. This paper asks the question that is underlying this approach using an experimental approach. It studies the effect that organization styles have on leaders’ ability to implement a successful leadership strategy to mitigate free-riding, while holding incentives constant. We ask leaders in controlled experiments to choose messages from a message set that induces a leadership style, and then to send them to participants at the beginning of each round of a repeated and finite horizon public goods game. The design features two leadership stylestop down and collegial-and two forms of communication-public leadership with no communication and public leadership supplemented with private communication between the leader and targeted followers. The experimental strategy adopted in this paper differs from the strategies used in leadership literature. We rely on a context free, game theoretical formulation of the problem that designs the collective task as a team problem. Each team has a leader who provides the team members with messages. Each team member decides on the amount of resources to invest in the public good. Then, the game-theoretical setup is tested using incentive compatible experiments. Finally, the paper departs from the conventional emphasis on the positive aspects of leadership styles as it studies, primarily, free-riding. 4

The intended contribution of this approach is to study how organization styles differ in their ability to enable leaders to implement leadership strategies that are successful. The paper shows that, when public leadership strategy cannot be supplemented by private communication, leaders provided with collegial leadership style perform better than their top down counterparts. But when private communication is possible, the reverse is true. This result challenges the vast literature that promotes collegiality as a mode of organization to mitigate free-riding and enlist the support of individuals in a collective value creation task. This result is consistent with the economics insight that collegiality is organizationally a costly way to promote cooperation. The nature of leadership strategies is documented and explained in the paper. The main mechanisms through which top down leaders operate is that they increase their efforts and increase their use of positive public leadership to cater to the contributors. These efforts raise the contributions of the free-riders, mainly though the sense of obligation that is promoted by the individual-oriented top down leadership style, as opposed to the collegial leadership style that is mainly group-oriented. Using insights from the strategy literature, the paper argues that the key prerequisite to successful leadership is therefore to provide the leader with the ability to extract some information about the existence of types of contributors, combined with a leadership style that enables leaders to implement a targeted leadership that supply to contributors on the basis of performance-based ad hoc coalitions. The next section presents the results of the literature that is relevant to the research setting in greater detail.

2

Related experimental literature

The effect of leadership in economics experiments has received a lot of interest in the recent years. Experimental and behavioral economists have studied leadership and have focused on characterizing the conditions that increase the leader’s effectiveness given the nature of the information structure, the incentives provided to the players and the structure of the groups. In public good games, overall, results from experiments show that leading by example has a positive influence on the level of contributions in the public good (G˝ uth et al., 2007; Haigner and Wakolbinger, 2010, Rivas and Sutter 2011). In most papers, leadership is implemented as a sequential public goods game in which the leader is the first one to contribute. When leadership is voluntary, the leader is more effective when provided with exclusion rights (G¨ uth et al. 2007). As noted by Rivas and Sutter (2011), these results show that the behavior of followers may be influenced even when contractual relationships or hierarchical authority is absent. In detail, this literature has identified the following findings. G¨ uth et al. (2007) examine the effects of leading by example when leadership is implemented by letting one group member be the first one to contribute. They find that leading by example (i.e., leadership by moving first) results in a marginally significant increase in contributions, compared to a situation without leadership. Followers condition their voluntary contributions on their leader’s earlier contribution. When leaders are equipped with the authority to exclude other group members, they achieve - and sustain - very high levels of cooperation that are significantly higher than the levels 5

prevailing with pure leadership by example. Backing up a leader’s (good) example with the punishment option of excluding another group member promotes cooperation, even though the higher levels of contributions have to be considered only partly voluntary since they are at least partly caused by the threat of exclusion. Interestingly, determining a leader either once and for all or in a predetermined and rotating order does not have a noticeable influence on cooperation levels within groups and groups with higher contributions in the exogenous part of the experiment are more likely to appoint a leader in the endogenous part of the experiment. Leaders with exclusion power typically want to remain leaders, but followers often turn them down, except when the institution has been successful in sustaining very high contribution levels in the exogenous part. Haigner and Wakolbinger (2010) and Rivas and Sutter (2011) find that voluntary leadership as letting the first mover become the leader i.e. by not restricting the leadership role to a specific group member yields higher contributions than when leadership is enforced exogenously, or when there is no leader at all. Yet, there are instances in which leadership in public good games has been found to be relatively ineffective. For instance, Sutter et al. (2007) examine its effect when group members are heterogeneously endowed and when this unequal endowment is common knowledge. They find that the presence of a leader increases average contribution levels but less so than in the case of homogeneous endowments. Leadership is almost ineffective, though, if participants do not know the distribution of endowments. Also, granting leaders with exclusion power does not lead to significantly higher contributions. Contrasting with Rivas and Sutter (2011), Arbak and Villeval (2013) find that voluntary leaders are not necessarily more influential than randomly-chosen leaders. In their setting, voluntary leadership is implemented by having group members apply to be the leader and then be randomly selected as the leader, if several group members apply to be a leader. In fact, in their experiments, followers contribute more and are more responsive when leaders are imposed. They conjecture that this outcome is most likely due to a selection bias as some of the randomly-chosen followers include the more generous individuals, who are more likely to be leader candidates in other treatments and also ”better” followers. Evidence from randomized controlled trials shows that local leaders are more effective at promoting public goods (Jack and Recalde 2014) and at promoting products that have positive externalities on group members (Dowd et al. 2014). The presence of a leader as a first mover has also been shown to affect the decisions of followers in experiments that rely on variations of the public good game. In a step-level public goods experiment, Normann and Rau (2014) find that the sequential order of decisions improves public-good provision and players’ payoffs, despite the fact that second movers tend to punish first movers when they give less than half of the threshold contribution. Moxnes and van der Heijden (2003) investigate the effect of leadership in a public bad game and find a significant within-subject effect of leadership: on average, followers invest 13% less in the public bad when there is a leader setting a good example compare to when there is no leader. However, they find that benefits are not enough to recover the costs of taking the lead. To the best of our knowledge, Brandts and Cooper (2007) is, in addition to this one, the only paper that implements leadership as the ability to send messages before actions are taken in economics 6

experiments. While implementing this form of leadership in a minimum-effort game, they find that, given a certain level of reward, targeted communication between leaders and participants is more effective in coordinating subjects than monetary rewards sent by leaders. They also find that such leadership is less costly monetarily than communication.

3

Setting

3.1

The game

Consider a group of four identical individuals who play a repeated public goods game and who are matched with a leader. For convenience, we refer to the group members as followers. All players interact for 16 periods and the finite horizon is common knowledge. Each follower receives an endowment of 10 tokens per round that is fixed and that can be either privately consumed or contributed to the group account. In each round t, followers earn $0.10 per token kept and an internal and external return of $0.04 per token contributed to the group account. The 4 X monetary payoff of a follower i is defined by ei,t − ci,t + (0.4) cj,t with ei,t the endowment, ci,t the j=0

contribution of the follower and with cj,t the contributions of all followers in the group. ci,t must satisfy 0 ≤ ci,t ≤ 10, in tokens. The dominant strategy for a selfish, payoff-maximizing player is to contribute nothing, and earn at least $1 per round from the tokens kept. The socially efficient outcome requires that all followers contribute everything and earn $1.6 at each period. The leader is asked to send messages to the followers at the beginning of each round. The leader, who receives no endowment and is paid based on the group’s performance, has a payoff that is defined by 4 X (0.5) cj,t . To replicate the common organizational feature that, in reality, leaders are more likely to be j=0

in managerial positions and to receive higher private benefits from their group’s productive output, the leader receives higher payoff from contributions. As in a typical principal-agent problem in organizations, P the leader does not see the individual output but the total number of tokens cj,t at the end of each round. The followers see the total number of tokens contributed and their own contribution.

3.2

Output-based leadership

We focus on leaders who operate in organizations that promote output-based leadership. In such organizations, leaders are instructed to take actions given the output of the followers, as described in (1). Thus, to take an action, the leader needs to decide on what message to send to the followers. The instructions portray the leadership strategy implementation problem as an informational problem that is not linked to followers’ utility but to their output. Thus, in order to decide on an action, the leader needs to relate to the output. Note that for round 1, when the leader does not know what the level of contribution is, the interface offers a neutral introduction message so that the leader can send a neutral message. 7

 take a “positive” action s if output is satisfactory. l,p take a “negative” action s , otherwise. l,n

(1)

The paper turns next to describing the organization styles in greater details.

3.3

Organization styles

To understand the effect of organization styles on leadership strategy and free-riding, the game is played in organizations of four different styles, which are selected to maximize the external validity of the setting. Organization styles are defined as two-by-two combinations of two channels of communication and two leadership styles. The channel of communication is either a “no chat” environment in which private communication is not possible or a “vertical chat” environment in which targeted private communication between the leader and targeted followers is possible. The leadership style is either a “collegial” leadership style or a “top down” leadership style. In the experiment, leadership styles are implemented by providing leaders with public messages from a finite set of messages that corresponds to the leadership style they are assigned with, as will be explained in greater details in the Experimental procedures section. 3.3.1

Channels of communication

The two main modus operandi of leaders in organizations are studied via the selection of channels of communication. The “no chat” organization corresponds to environments in which only the official leadership strategy that leaders promote in organizations is provided to the followers. It depicts organizations in which leaders may organize meetings to have all followers hear their perspective on how to go about a task. This public message is a type of one-way communication. The “vertical chat” organization corresponds to environments in which leaders lead by supplementing their public leadership messages with private communication with targeted participants. It depicts organizations in which leaders entertain inter-individual relationships with targeted followers as a form of two-way communication. The channels of communication affect the sequence of actions in the game and the problem of the leader in the following way. When private communication is not possible • Step 1: The leader selects and sends a message. • Step 2: After observing the message chosen by the leader, the followers decide simultaneously and without communicating among themselves or with the leader on their contribution level.

8

When targeted private communication is possible • Step 1: The leader selects and sends a message. • Step 2: After observing the message chosen by the leader, the leader and targeted followers can privately communicate. • Step 3: The followers decide simultaneously and without communicating among themselves or with the leader on their contribution level. 3.3.2

Leadership styles

The classification of leadership styles encompasses the majority of the leadership styles. The classification is based on the observation that, in management, leaders’ willingness to ’share’ their leadership is not a necessary condition to leadership. To have a vision, to communicate the vision, and to deliver outcomes consistent with that vision are three necessary conditions of leadership. Yet, leaders’ willingness to ’share’ their leadership differs across styles: some leadership styles rely heavily on the leaders’ aptitudes to empower others while others focus on the leaders’ function as the center of power. Leaders that share their leadership have a prominently collegial leadership style: they focus on empowering their followers, by making them feel valued and appreciated. Leaders that have a top down leadership style focus on execution, by making the decisions and by letting their followers know what decisions they made. The paper retains these two types of leadership styles3 . On the one hand, collegial leadership is portrayed in the management literature through the objective of engaging others. Collegial leaders are concerned with their personal responsibility. Inputs from followers is central to their decision-making and they seek to leverage the best ideas. Their authority is based on cultivated relationships, they think of themselves as accountable to superiors, peers and subordinates and they seek evaluations. Leadership development is a high priority and they promote those who contribute to group success. On the other hand, top down leadership is portrayed in the management literature through the objective to direct action. Top down leaders are concerned with their personal interests. Inputs from followers is tangential to their decision-making and they seek to be obeyed. Their authority is based on external controls, they think of themselves as accountable mainly, if not only, to superiors and they tend to discard evaluations. Leadership development is not a priority and they promote those who follow the instructions. 3

Collegial leadership styles include people-oriented, democratic, relations-oriented, participative, transformational, ser-

vant, shared, authentic, positive, pygmalion leadership styles. They all rely on the idea that the leader empowers others. In particular, “servant leadership” refers to leaders who lead with integrity and generosity, ”People-oriented” and ”relationsoriented” leadership styles refer to leaders who are focused on organizing and developing the people on the team and democratic and participative leadership styles refer to the leader who makes final decisions only after consulting the team members. Top down leaders include task-oriented, bureaucratic, hierarchical, transactional, autocratic, charismatic leadership styles. In particular, transactional or hierarchical leadership style corresponds to situations in which team members agree to obey the leader when they accept a job. Task-oriented leaders focus only on getting the job done, autocratic leaders provide followers with a lot of direction on task execution, bureaucratic leaders work according to some rules and charismatic leaders inspire followers with enthusiasm but are largely directing them to follow a prescribed path.

9

Collegial leadership style Building on the large literature that claims that collegial leaders generally manage teams more successfully than top down leaders (Bass and Avolio 1993), collegial leadership style is represented as group-oriented. In this style, positive messages are encouraging and negative messages are supportive. The focus on the group does not stress the responsibility of individual follower but the responsibility of the leader, as summarized in (6).  To be collegial, “encourage” the group s l,n To be collegial, “support” the group s , l,p

if output is satisfactory.

(2)

otherwise.

Top down leadership style In contrast, top down leadership style is individual-oriented and it stresses the responsibility of individuals in the group, as summarized in (7). In this style, positive messages are rewarding and negative messages are punishing.  To be top down, “reward” individual followers s if output is satisfactory. l,p To be top down, “punish” individual followers s , otherwise. l,n

(3)

This classification enables to study the two sides of the spectrum on the leadership style spectrum without relying on the two extreme ends of the spectrum, which goes from transformational leadership to transactional leadership4 . In addition, this classification of leadership styles does not designate some styles as more or less effective than others but instead describes their behavioral manifestation.

4

A theoretical framework and related hypotheses

We propose a theoretical framework to explain output-driven leadership strategy and free-riding in organizations of different styles. Our approach departs from non-cooperative game theory in that it provides insights on how to consider the problem of the leader5 . To understand the effect of organization styles on leadership strategy and free-riding, we develop an approach to the leader’s problem that relies on a two-phase decision process composed of an “editing phase” and a “solution phase”. The two phases correspond to the decision process leaders face in outputbased organizations. In the first phase, the leader gathers information about the output; in this phase, we study the effect of communication channels which affect the leader’s ability to gather information. In the second phase, the leader uses this information to choose a leadership strategy; in this phase, we study the effect of leadership styles on the leader’s choice. 4

The two extreme emphasize, for the former, the role of empowering followers through a collegial process and by inducing

identity change by getting followers to take ownership for their work and, for the latter, contractual and managerial nature of the leader/followers relationship in which leaders promotes compliance of their followers through rewards and punishments. 5 The non-cooperative game solution to the problem of the leader is that leadership will have no effect on contribution decisions: considering the incentives of the followers, the leader can only choose a leadership strategy randomly as followers play the unique Nash equilibrium.

10

Our approach contributes to the literature on intrinsic motivation initiated by B´enabou and Tirole (2003) in that we study how leaders’ leadership messages affect the followers’ perception of the leader’s expectations of their contribution levels when choosing their leadership strategy6 . Before proceeding with the description of the decisions in the two phases, we characterize the utilities of the players and use this characterization to improve the definition of their actions. Consider two types of players, a leader (the principal), and followers (the agents).

4.1

Preliminaries

Leader utility The leader’s gains utility from the monetary output that is generated by the group out4 X put. In this setting, the leader’s utility is denoted for a leader l at time t by Ul,t = (0.5) cj,t . j=0

Followers utility The followers’ utility accounts for the followers’ valuation of leadership: it allows to factor in the utility gain from following the leader. In this setting, the followers’ utilities are denoted, at 4 X time t and for a follower i, by Ui = ei + (vi | m)ci − ci + (0.4) cj . vi ∈ [0; 1] is the value the individual j=0

places on the message. The followers’ utility characterization accounts for the fact that followers’ valuation of leadership may lead them to decreasing their monetary payoffs. Followers types Suppose there is a proportion ex ante of λ contributors who have a high propensity to value the presence of the leader and to contribute and a proportion of 1 − λ contributors who have a low propensity to value the presence of the leader and to contribute. Leader decision The leader chooses a message m at time t in the hope of increasing the group output. The leader can choose between two types of messages: positive messages and negatives messages. The leader will choose a positive message when thinking that λ contributors have a propensity α of contributing that is greater than (1 − λ) contributors’ propensity of contributing β such that λ × α ≥ (1 − λ) × β. The leader will choose a negative message when thinking that the gain from having 1 − λ contributors’ propensity of contributing β increase will be compensated by the potential decrease in λ contributors’ propensity of contributing β such that λ × α ≤ (1 − λ) × β. This leads to the following trade-off:  The leader will choose a positive message, if the leader thinks that λ × α ≥ (1 − λ) × β. The leader will choose a negative message, if the leader thinks that α × λ ≤ β × (1 − λ).

(4)

We study now how organization styles affect the leader’s leadership strategy. 6

Our approach differs in that we focus only on how leadership strategy is chosen and how, in turn, it affects followers’

expectations of their contribution levels. B´enabou and Tirole’s framework applies to a broader range of policies that may be chosen by a principal. In particular, they include monetary incentives which are not considered here

11

4.2

The effect of communication channels

Prior to making a decision, the leader needs to gather information about the output of the group. The problem of the leader is to determine whether the level of contributions reflects a group of followers that is composed of high contributors or a group of followers that is mainly composed of low contributors. The former reflects a “good state” of nature in which λ ≥ (1 − λ), labeled for simplicity (Λ). The latter reflects a “bad state” of nature in which λ ≤ (1 − λ), labeled for simplicity (1 − Λ). Getting information about the types of followers is crucial as experimental results have established that contributors’ types are instrumental in shaping cooperative outcomes. For instance, when sorting is not possible, cooperation collapses when contributors decide to reduce their contributions (Gunnthorsdottir et al. 2007). Communication channels affect the leader’s ability to gather information in the following way. When private communication is not possible The leader cannot gain information about the types of contributors who are part of the group of followers. In fact, quotes in the introspection box suggest that leaders understand that there are followers of different types and that they cannot gain information about their types7 . When targeted private communication is possible The leader can gain information about the types of followers who compose the group when communicating with them8 . When the leader communicates with a targeted follower, the leader receives a signal S. As the leader cannot see which state of nature is true, a Bayes’ rule may be used to depict how priors P r(Λ) and P r(1 − Λ) of the leader may evolve as in P r(Λ | S) =

P r(S | Λ)(P r(Λ)) P r(S | Λ)(P r(Λ))+P r(S | Λ)(P r(1−Λ))

)9 .

The improvement of the leader’s ability to evaluate the states of nature allows for the implementation of targeted leadership strategies. The leader can think along the following lines10 . First, the followers cannot create value alone, as all their endowment is privately consumed. Second, if all followers came together to interact without a leader, they would also consume their endowment and not create value. Third, a leader cannot create value, unless matched with followers. Finally, in an ideal setting in which 7

Some leaders write: “I probably pushed them too far to think they were the only ones investing” or “My last statement

increased profits by $0.15 so I want to continue that trend and have the people who were too nervous to do it before, do it this time”. Although the leader may realize that there may be contributors and free-riders, the leader is constrained in that information about potential heterogeneity in the group is not available. 8 As information is noisy, we assume that the case of a tie is not relevant to this setting. Rather, we focus on the “good” or “bad” dichotomous formulation that is explained above. 9 Consider the following example. If the prior of the leader is that the probability of both states is equal, P r(Λ) = 1/2 and there are 2 λ-type followers and 1 1 − λ-type follower in the group i.e. P r(λ | Λ) = 2/3 and P r(λ | (1 − Λ)) = 1/3. Similarly, the probability of (1 − Λ) is given by: P r((1 − Λ) | λ) = (1/6)/(2/6 + 1/6) = 1/3. Consider that the prior deteriorates to, for example, P r(Λ) = 1/3. In this case, P r(Λ | λ) = 1/2, which is a decrease in probability that (Λ) is the relevant state of the world. 10 The following formulation may be used as an analogy to understand how the leadership strategy may be conceived in organizations in which information about different types of contributors may be gathered. To be clear, binding coalitions are not possible in the interface and the leader knows that such coalitions may not be enforced, but to shed light on the leader’s strategy to provide a targeted leadership, consider the formulation above as a thought experiment.

12

coalitions could be enforced, all players and the entire endowment would be contributed. Knowing this, a leader could attempt to apply the marginal-contribution principle and exclude from the coalition the players who contribute less than the marginal contribution as, in an ideal setting in which coalitions could be formed, leaders would be better off by forming coalitions with the contributors and excluding free-riders. In this setting, leaders cannot cast aside free-riders11 , but yet, they can try to form ad hoc coalitions and choose targeted leadership strategies as if they were managing a sorted team. We next consider how leadership styles affect the leader’s choice of a leadership strategy.

4.3

The effect of leadership styles

After gathering information about the output, the leader needs to choose messages. We analyze the effect of collegial leadership style and top down leadership style on the leader’s choice of leadership strategy and on free-riding. The effect of collegial leadership style Collegial leadership style does not allow the leader to target a specific type of followers because the style always refers to followers as a group. Thus, even when leaders can distinguish high contributors from low contributors and even with varying the type of messages across positive and negative messages, the leader’s ability to target high contributors is limited. The effect of top down leadership style By design, negative messages have stronger effects in the top down leadership style: λ-type contributors will have a higher propensity to decrease their contribution and (1 − λ)-type contributors will have a higher propensity to increase their contribution in the top down treatment than in the collegial treatment. In contrast to collegial leadership style, the effect of top down leadership style is that, if leaders can distinguish contributor types, they may adapt their leadership strategy in order to target contributor types12 . The analysis leads to the following hypotheses.

4.4

Hypotheses

We formulate research hypotheses about the leader strategy, as in (H1) and (H3), and the effect of the leaders’ strategy on followers’ contribution decisions, as in (H2) and (H4). The two sets of hypotheses are organized by types of channels of communication: (H1) and (H3) refer to the situation in which 11

Bridoux et al. (2011) specify the conditions under which self-selection and sorting lead to implementing such reasoning.

They argue that in firms that rely on “benevolent cooperation”, selection mechanisms are required to exclude free-riders and prevent strong cooperators to select out. They argue that managers can influence the firm’s mix of motives by putting in place selection mechanisms. 12 The success of top-down strategies depends highly on the leaders’ ability to activate positive emotions. Huy (2011) explain that the top down strategy implementation literature has overlooked social-emotional factors, and especially groupfocus emotions. Huy further argues that group-focus emotions are important as they are activated in the tasks that are undertaken in the organizations.

13

private communication is not possible and (H3) and (H4) refer to the situations in which private targeted communication is possible. Leadership styles affect the leader’s decision in the following way. Leadership strategy (H1) In organizations without private communication, public leadership strategy implementation leads to the adoption of the promoted leadership style: leaders assigned to the top down leadership style use more negative messages than their collegial counterparts. Free-riding (H2) In organizations without private communication, public collegial leadership leads to higher contribution levels than public top down leadership. Leadership strategy (H3) In organizations in which public leadership may be supplemented by private communication, leaders will implement a targeted leadership strategy in performance-based ad hoc coalitions. Free-riding (H4) In organizations in which public leadership may be supplemented by private communication, top down leaders perform better at enlisting the support of their followers than their collegial counterparts. The paper now turns to explaining the experimental procedures that was used to test the hypotheses.

5

Experimental procedures

The setting and hypotheses were tested in laboratory experiments. Participants were seated in separate cubicles so they could not communicate during the experiment. Subjects were randomly assigned to either the role of a leader or a participant based on their time of login in the interface. The experimenters distributed the login information according to a random order that was scheduled prior to the session. The first subjects to receive the login information and were asked to login first. They became the leaders. The other subjects were assigned to the role of participants. The randomization procedure was implemented very discretely and using decision support documents that were on the experimenters’ desk that the participants could not see to ensure that participants cannot outguess who the leaders are in the room. In the interface, followers were named “participants” and leaders were named “leaders”. They all played 16 rounds of a linear public good game in fixed groups of five subjects composed of one leader and four participants. All specifications were known to all participants, with the exception of the initial cash endowment which differed between the leader and participant roles. This difference was introduced in order to guarantee minimal earnings to the leaders in case contribution decisions of the participants were very low. The amount of the initial endowment was not disclosed but its existence was: the initial endowment of the participants was stated as non-zero so that the instructions could be read out loud by the experimenters. The specifications of the game are explained in Table 1. 14

[Table 1 about here.] The sequence of events in the experiment was as follows. After reading the instructions and answering the questions related to the payoff structure, leaders were shown a description of their role given the list of messages that could be sent to participants. Leaders in the collegial treatments received messages corresponding to the collegial leadership style, and leaders in the top down treatments received messages corresponding to the top down leadership style. The interface instructed: “You begin the round by sending a message to all participants in your group.” Leaders assigned to the Collegial treatment were provided with the following additional instructions: “In general, you should project a collegial, helpful manner and avoid direct commands”. In contrast, Top down leaders were provided with the following instructions: “In general, you should project a strong, decisive leadership posture, with a clear expectation that participants will follow your mandates”. After these statements, all leaders saw the following message: “The options below are templates that may help you with this. Each option is associated with a description of when the message may be appropriate, e.g. after an increase in investments”. In the interface, the leader sees the description of nine situations and nine messages, with each message behind associated with a situation. This feature of the interface aims at depicting the rules of thumb that are promoted by organizations. The message sets are designed so that the list of messages reflect the leadership styles as described in the organizational and management literatures. In the Collegial leadership style, messages are maneuvering or encouraging as collegial leaders guide their followers carefully and give them support and confidence. In the Top down leadership style, messages are directional or rewarding as top down leaders rely on carrot and stick leadership strategies. The list of the messages that were provided to the leaders is displayed in Table 2. The leader would select a message and submit it to the participants. Participants received the message and had to acknowledge that they read it. [Table 2 about here.] In the No chat treatment, leaders and participants were given a one minute-period to record their thoughts about their strategy and the decisions of others or their decisions, the decision of others and the leader, respectively. In the Vertical chat treatment, participants were given a two minute-period to send messages to the leader. Leaders could say anything: they were not limited to the same set of messages provided by the experimenter. Communication between participants was not permitted. During that period, leaders were able to send messages to targeted participants. An introspection box was added to the interface in the No chat treatments. This introspection feature added time between each round, which reduced the difference between the Vertical chat and the No chat treatments on that dimension. Thanks to this feature, the difference between the two treatments is only the difference in ability to communicate and not the time allocated to each decision. A one-minute introspection mode was chosen over a two-minute one in order to avoid loss in attention to the game. Experiments were conducted in the Veconlab at the University of Virginia and using the Veconlab software. The instructions were embedded in the computerized interface and the comprehension of the 15

payoff structure was checked via a computerized questionnaire. Subjects were recruited from a large subject pool of undergraduate and graduate students. Treatments were assigned randomly to subjects and treatments were randomized over morning and afternoon time slots and over days of the week.

6

Experimental tests of the research hypotheses

6.1

Test of (H1): The effect of organization style on public leadership strategy when private communication is not possible

To characterize the effect of organization style on public leadership strategy, the public messages chosen by the leaders are classified using two classifications to ensure a comprehensive description of the choices of the leaders. The classifications sort messages as “neutral”, “positive” or “negative”. The neutral message is the introduction message that states the role of the leader. Two classifications are proposed: a “situation-based” classification and a “style-based” classification. Note that the decision to associate messages to situations was motivated by the fact that leaders need some guidance on when they may want to use specific messages. The words “neutral”, “positive” and “negative” were not stated in the interface. The leaders could only see the description of the situations and the messages. The situation-based classification is a strict classification that relies on the factual content of the messages as described in the “situations” that are provided to the leaders in the interface. By design, this classification leads to an equal number of positive and negative messages. It retains messages 2, 3, 5 and 8 as positive messages and messages 4, 6, 7 and 9 as negative messages. This classification is not leadership style-based. The style-based classification relies on the description of the two leadership styles in the organizational behavior and management literatures. In these literatures, collegial leadership style is predominantly positive and top down leadership style is predominantly negative. By design, this classification leads to an unequal number of positive and negative messages. In the collegial leadership style message sets, it retains messages 2, 3, 4, 5 and 8 as positive messages and messages 6, 7 and 9 as negative messages. In the top down leadership style message sets, it retains messages 2, 5 and 8 as positive messages and messages 3, 4, 6, 7 and 9 as negative messages. This classification is leadership style-based. The choice of leadership strategies is depicted using both classifications in Figure 1. The persistence of the patterns is also salient at the individual level, as depicted in Figure 2, which displays the choice of public messages chosen by individual leaders using both classifications, for consistency. [Figure 1 about here.] [Figure 2 about here.] In line with (H1), the experimental results show that the leadership strategy of leaders provided with the collegial leadership message set is consistent with the findings of the management literature. In addition, the strategies of the collegial leaders are not affected by the change in communication 16

environments. Also aligned with (H1), leaders provided with top down leadership style messages use predominantly negative messages. This leads to the following result, which is aligned with the Bayes’ explanation of public leadership strategy and which validates (H1). Result 1 When public leadership strategy cannot be supplemented by private communication, collegial public leadership strategy relies predominantly on positive communication and top down public leadership strategy relies predominantly on negative communication.

6.2

Test of (H2): The effect of public leadership strategy on free-riding when private communication is not possible

When private communication is not possible, contributions start at 40% of the endowment in the top down treatment and at above 50% in the collegial treatment and end below 20% of the endowment. Figure 3 depicts the pattern of mean contributions per round and per treatment. There is a declining pattern but contributions do not converge. Collegial public leadership leads to higher contributions than top down public leadership. The mean contribution in the collegial treatment is 3.87 and the mean contribution in the top down treatment is 2.74. Earnings are higher in the collegial treatment. The descriptive statistics of the contribution results are summarized in Table 3. Participants’ earnings are displayed in Figure 4. The statistical significance of the results is reported in Table 4. The statistical significance of the results are very strong: at the overall level, the p-value comparing the two treatments is 0.0000, per round 0.0167, by group 0.0821 and by subject 0.0143. [Figure 3 about here.] [Table 3 about here.] [Figure 4 about here.] [Table 4 about here.] The relationship between public leadership and free-riding is documented in Table 5: negative messages are associated with a decrease in average group contributions and positive messages are associated with an increase in average group contributions, both in a statistically significant fashion and using the stricter situation-based classification method. The analysis accounts for group random effects and is performed on group contribution per round, not on individual contributions to avoid the noise generated by mean individual contributions. Positive messages do not have systematic effects on group contributions, which is due to the overall declining pattern of contribution. Table 6 further reports that it is the core negative messages that have a significantly negative effect on group contributions. These findings validate (H2) and further identify the role of negative messages when there is a decrease in contributions. [Table 5 about here.] 17

[Table 6 about here.] Result 2 In organizations without private communication, public collegial leadership leads to higher contribution levels than public top down leadership. Result 2a In organizations without private communication, decrease in contributions are associated with negative public leadership.

6.3

Test of (H3): The effect of organization style on leadership strategy when private communication is possible

Public leadership The effect of organization style on public leadership when private communication is possible is that, in contrast with collegial leadership treatments, the change in communication environments affects the choice of public messages made by leaders in the top down treatment. Figure 1 shows that leaders assigned to the top down treatments use more positive messages than when in an environment in which they can privately communicate with the followers. They use more rewarding public messages and fewer credit taking public messages. Using a Kruskal-Wallis, which is the nonparametric test that is the extension of the Mann-Whitney U test that allows for comparison of more than two independent groups, the comparison between the choice of public messages in the collegial treatments across two communication environments shows no statistically significant difference with a p-value of 0.2736. Using the same test, in contrast, the comparison of the top down treatments across communication environments exhibit a statistically significant difference with a p-value of 0.0126. The difference is also visually noticeable as in Figure 5, which displays the choice of public messages by treatment. [Figure 5 about here.] This leads to the following result. Result 3 Top down public leadership strategy is predominantly negative in the absence of private communication and predominantly positive in the presence of private communication. Cooperative types and public leadership Using the methodology developed in Gunnthorsdottir et al. (2007), the paper studies cooperative types. Free-riders are defined as players who contribute 3 tokens or less on the first round and contributors are those who contributed more than 3 tokens during the first round. The distinction between free-riders and cooperators allows to identify the effect of the presence of different types on public leadership. Figure 6 shows that when there are two or more free-riders in the group, which is the case for 4 groups in the top down treatment and for 5 groups in the collegial treatment, top down leaders are more likely to use negative style messages than their collegial counterparts, with a p-value of 0.0453. This means that top down leaders implement a 18

targeted leadership strategy to establish performance-based ad hoc coalitions when they are matched with contributors, and they do not when they are matched with two or more free-riders. [Figure 6 about here.] Result 3a Top down leaders implement a targeted leadership strategy to establish performance-based ad hoc coalitions when they are matched primarily with contributors, and they do not if they are matched with two or more free-riders. Private communication Furthermore, the results show that more communication is initiated by leaders assigned to the Top down treatment than by those assigned to the Collegial treatment, as reported in Table 7. Specifically, the sum of occurrences of private messages and the length of the messages in the top down when private communication is possible is greater than in the collegial vertical chat treatment. This table reports the occurrences of a message i.e. the occurrence of a message or several messages sent by the leader to a targeted participant, or from a specific participant to the leader. [Table 7 about here.] The difference of amount of private communication across treatments is driven by leaders’ effort. Leaders’ effort, as the sum of messages sent by the leader, is significantly greater in the top down treatment than in the collegial treatment. In contrast, participants’ effort is not significantly different. The comparison of the sum of messages sent by leaders to participants indicates that the number of messages sent by leaders in the top down treatment with a p-value of 0.0008, while the sum of the number of messages sent by participants to leaders by round and by treatment is not significantly different across treatments, as indicated by a p-value of 0.1295. Result 3b When private communication is possible, leaders assigned to the top down leadership style communicate more privately than their collegial counterparts. Content of private communication The analysis of the content of the messages shows that leaders in the top down treatment are more likely to use private communication to talk about motivation and to send personal messages than in the collegial treatment. The content of the messages sent by the followers enable verification that followers matched with top down leaders are more likely to make neutral statements and they are more likely to use a tone that is top down. Followers in the collegial treatment are more likely to make negative statements and they are more likely to use a collegial tone than their top down counterpart. These findings are aligned with (H3) which states that leaders will implement targeted leadership strategies that establish ad hoc coalitions that are performance-based. They are also aligned with (H4) which states that leaders in the top down treatment will be better than their collegial counterpart at enlisting the support of their followers. Some quotes illustrate the overall results. A top down leader, in a private conversation with a follower, states: “I have told others to invest more”. Another one talking to a targeted participant, says: “I’ll 19

talk to them. Keep investing.” and then saying to the other followers, including the former one: “You really should invest more. When one of the group members’ keeps investing and gets no return, that hurts everyone.” Followers in the top down treatment, for example, state: “ID3: (to Leader): my guess is this round someone will invest 0”, or “ID2: (to Leader): What are other people saying?”, or “ID4: (to Leader): I contributed 10”, or “ID2: (to Leader): I’m going in at none this round”. Some time passes, and then, the follower says “I’ll meet you halfway. I break even, or I quit”. [Figure 7 about here.] Result 3c When private communication is possible, leaders assigned to the top down leadership style communicate privately using messages that are related to the motivation of their followers and they are more likely to use personal messages. These differences in leadership strategies lead to differences in free-riding patterns, as explained next.

6.4

Test of (H4): The effect of public leadership strategy on free-riding when private communication is possible

Contributions level and patterns Contributions start around 60% of the endowment and stay above 40% of the endowment, demonstrating there is no sharp decline in the contribution patterns in the presence of vertical chat. The most noticeable result is that contributions are the highest in the top down treatment. As for the the results when private communication is not possible, the results are summarized in Table 3 and Table 4, as depicted in Figure 3 and Figure 4. The mean contribution in the top down treatment is 6.35 and is 5.57 in the collegial treatment. The difference between the two treatments is highly statistically significant, with a p-value of 0.0014 for the overall sample, of 0.0059 by round, 0.0059 by group and 0.0003 by subject. Further analysis of the contribution patterns shows that mean standard deviations of contribution are higher in the collegial treatment than in the top down treatment, as depicted in Figure 8. The difference across treatment is statistically significant with a p-value of 0.0899 in the no chat treatment and of 0.0075 in the vertical chat treatment. The results are aligned with (H4). Result 4 When private communication is possible, contributions are higher when leaders are provided with the top down leadership style than when they are provided with collegial leadership style.

[Figure 8 about here.] Cooperative types Using the methodology developed in Gunnthorsdottir et al. (2007), the paper reports separately the mean contribution per round of the free-riders and the contributors in Figure 9. As explained above, free-riders are defined as players who contribute 3 tokens or less on the first round and contributors are those who contributed more than 3 tokens during the first round. The figure shows that top down leaders manage to increase the contributions of the free-riders when they are provided with 20

the ability to communicate privately. This result is important because, as explained Gunnthorsdottir et al. (2007), cooperation collapses usually when contributors decide to stop contributing, in experiments in which free-riders usually do not increase their contributions. This result identifies the characteristics of an environment in which top down leaders can successfully implement a leadership strategy. [Figure 9 about here.] Result 4a When private communication is possible, free-riders in the top down leadership style organization increase their contributions. Further analysis of the contribution patterns shows that another mechanism behind the overall level of contribution is that the distance between the highest contributor and the lowest contributor in each treatment is highest in the top down treatment, i.e. the treatment in which the mean contribution is highest. The differences across treatment are significant with a p-value of 0.0303 in the no chat treatment and of 0.0002 in the vertical chat treatment. Figure 10 depicts these results and makes visible that this distance declines over time when private communication is not possible and increases when private communication is possible. The results show that more free-riding is observed in treatments in which high mean levels of contributions are achieved. Leaders in the top down treatment when communication is possible effectively manage to extort high contributions from the highest contributors when others free-ride on their contributions, meaning that high contributors have a high tolerance for the free-riding of others. Another way to phrase this result is to say that top down leaders are more effective at exploiting the contributions of high contributors. This contributes to explaining the mechanisms behind the validation of (H4). Result 4b When private communication is possible, contributors in the top down organization style have a higher tolerance for free-riding than those in the collegial one.

[Figure 10 about here.]

7

Concluding remarks

This paper investigates how leaders who are matched with a group of followers decide on a leadership strategy to induce the followers to allocate the resources that they are endowed with towards the public good, thereby increasing the pay-off of the leader and of their fellow participants, despite the fact that this conflicts with their individual interest. The analysis highlights that organization styles affect the leadership strategy chosen by leaders and, in turn, the free-riding of followers. It focuses on two main leadership styles, Collegial and Top down styles, and on two prominent forms of communication in organizations between leaders and followers, i.e. one-way communication in the form of public messages and two-way communication in the form of public messages combined with private leader-follower communication. 21

The paper investigates this question using controlled experiments that rely on a linear public goods game. This methodology allows us to create an environment in which the effects of organization styles on leadership strategy can be identified. This setup allows to control for the effect of monetary incentives that are hard to account for in non-experimental data. While the question of the external validity of such findings may be raised, much evidence from organizational economics and strategy support the patterns that the findings discern. The findings suggest that in organizations in which leaders and individual followers cannot communicate privately, negative messages, especially if they are targeting individuals and not the group, as with top down leadership styles, will discourage contributors and drive down the level of contributions. In organizations that give voice to the followers by enabling leaders and individual followers to engage in private conversations, individual-oriented leadership strategies induce a higher sense of obligation and motivate both contributors and free-riders. The paper identifies that the key strategic prerequisite to successful leadership is to provide the leader with the ability to extract some information about the type of contributors, combined with the provision of a leadership style that enables leaders to implement a form of targeted leadership that is contributor type-based, and that focuses on contributors as opposed to free-riders. Successful leaders are those who work towards implementing performance-based ad hoc coalitions, which target the contributors and disregard the free-riders. Implementing a successful leadership strategy depends on the leader’s ability to extract information about the existence of types, on the main focus of the leadership style as an individual-oriented versus group oriented and, on the followers’ perception that their contributions are acknowledged. Such sharp differences can have several important consequences for organizational design. Organizations that cannot guarantee a minimum level of targeted communication may prefer to stay away from directional leadership strategies, as top down strategies drive down the contributions of all followers, but noticeably that of contributors, who are the ones who carry the cooperative efforts when such efforts are undertaken. Organizations that can guarantee some level of private targeted communication between leaders and followers may consider the value of individual-oriented leadership strategies, as these raise the contributions of all followers, contributors as well as free-riders. In many cases, however, organizations may face difficulties in guaranteeing conditions under which high contributors will tolerate the sustained existence of free-riders. In particular, it is possible that the social planners who decide on the leadership strategy to be implemented by their organization may have to compensate high contributors for the resource loss that is generated by being matched with free-riders. Such intricate policies are at the core of the social planner’s exercise. In that respect, the identification of the key strategic prerequisite that is the ability to get cues about the existence of different contributor types provides a ground to shape leadership strategies. This paper identifies how organization style, leadership strategy and free-riding affect each other. Although this all-embracing aspect of the paper allows the study of the leaders’ strategy, it stops short of addressing some interesting issues. An important limitation of this paper is that it does not study leadership strategy as a function of organization style, in a context in which cooperation levels are 22

relatively fixed. Another important limitation is that it does not model the behavior of the followers given a leadership style that is fixed. Clearly, such two-way relationships are very important, and they should be addressed in separate settings, in which the dynamic aspect of the three-way interaction is controlled for. Such interesting research questions will be pursued in future work.

23

Acknowledgment Earlier versions of the manuscript were presented at the Young Scholars Workshop of the 2014 Behavioral Operations Conference held at the University of Cologne, at the Economics and Psychology Seminar of the Paris School of Economics and at the Brownbag Seminar Series of the University of Virginia Frank Batten School of Leadership and Public Policy. We thank the participants for their comments at these events. We gratefully acknowledge the research support provided by the Darden Graduate School of Business at the University of Virginia.

24

References Arbak E., M.C. Villeval (2013) ‘Endogenous leadership: Selection and influence’ Social Choice and Welfare, 40(3): 635-662. Avolio, B. J., J. J. Sosik, D.I. Jung and Y. Berson (2003) ‘Leadership models, methods, and applications: Small steps and giant leaps’ In W. C. Borman, R. Klimoski, D. R. Ilgen and B. Weiner (Eds.), Handbook of psychology, 12: 277-307. New York: John Wiley & Sons. Bass, B. M. and B.J. Avolio (1993) ‘Transformational leadership: A response to critiques’ In M. M. Chemers & R. Ayman (Eds.), Leadership theory and research: Perspectives and directions (pp. 49-80). San Diego, CA: Academic PressBass, B., and Bass, R. (2008) The Bass handbook of leadership: Theory, research, and managerial applications, New York, NJ: Free Press. B´enabou R. and J. Tirole (2003) ‘Intrinsic and Extrinsic Motivation,’ Review of Economic Studies, 70(3): 489-520. Bergstrom, T., L. Blume H. and Varian (1986) ‘On the private provision of public goods,’ Journal of Public Economics, 29: 25-49. Bernheim, B.D. (1986) ‘On the voluntary and involuntary provision of public goods’ American Economic Review. 76(4): 789-793. Bloom N. and J. Van Reenen (2007) ‘Measuring and explaining management practices across firms and countries,’ Quarterly Journal of Economics, 122 (4): 1351-1408. Bloom N., B. Eifert, A. Mahajan, D. McKenzie and J. Roberts 128 (2013) ‘Does management matter? Evidence from India,’ Quarterly Journal of Economics, 128(1): 1-51. Bochet, O., T. Page and L. Putterman (2006) ‘Communication and punishment in voluntary contribution experiments,’ Journal of Economic Behavior & Organization, 60(1): 11-26. Brandts J. and D. Cooper (2007) ‘It’s what you say, not what you pay: An experimental study of manager-employee relationship in overcoming coordination failure’, Journal of the European Economic Association, 5(6): 1223-1268. Bridoux, F., R. Coeurderoy and R. Durand, (2011) ‘Heterogeneous motives and the collective creation of value’, Academy of Management Review, 36: 711-730.

25

Burger, N. and C.D. Kolstad (2009) ‘Voluntary Public Goods Provision, Coalition Formation, and Uncertainty,’ National Bureau of Economic Research Working Paper W15543. DeRue, D. S., J.D. Nahrgang, N. Wellman and S.E. Humphrey (2011) ‘Trait and behavioral theories of leadership: A meta-analytic test of their relative validity’ Personnel Psychology, 64, 7-52. Dowd R., Lipscomb M. and Pankov D. (2014) ’The Cross, the Crescent or the Crown: A randomized controlled trial testing the effectiveness of local religious and secular leaders,’ unpublished draft. Eagly, A. H. and M.C. Johannesen-Schmidt (2001) ‘The leadership styles of women and men’ Journal of Social Issues, 57: 781-797. Gunnthorsdottir, A., D. Houser and K. McCabe (2007) ‘Dispositions, history and contributions in public goods experiments’, Journal of Economic Behavior and Organization, 62(2): 304-315. G¨ uth, W., V. Levati, M. Sutter and E van der Heijden (2007) ‘Leading by example with and without exclusion power in voluntary contribution experiments’, Journal of Public Economics, 91(5-6): 1023-1042. Hackman, J. R., and R. Wageman (2007) ‘Asking the right questions about leadership’, American Psychologist, 62: 43-47. Haigner, S. and F. Wakolbinger (2010) ‘To lead or not to lead - Endogenous sequencing in public goods games’, Economics Letters, 108(1), 93-95. Huy, Q. (2011) ‘How middle managers’ collective emotions and social identities influence strategy implementation’, Strategic Management Journal, 32: 1387-1410. Jack B.K. and M. P. Recalde (2014) ‘Local leadership and the voluntary provision of public goods: Field evidence from Bolivia’, (accepted) Journal of Public Economics. Hirschleifer, J. (1983) ‘From weakest link to best shot: The voluntary provision of public goods’, Public Choice, 41(3): 371-386. Kahneman, D. and A. Tversky (1979) ‘Prospect theory: An analysis of decision under risk’ Econometrica, 47(2): 263-291. Karaer O., J. Khawam, T. Kraft (2014) ‘Improving Supplier Environmental Performance’, unpublished draft. 26

Kocher, M., G. Pogrebna and M. Sutter (2012) ‘Other-regarding preferences and management styles’, Journal of Economic Behavior & Organization, 88: 109-132. Ledyard, J.O. (1995) ‘Public Goods’ (in eds) John H Kagel and Alvin E. Roth, The Handbook of Experimental Economics, Princeton: Princeton University Press. Matsa, D. A. and A. R. Miller (2013) ‘A female style in corporate leadership? Evidence from quotas,’ American Economic Journal: Applied Economics, 5(3): 136-69. Moxnes, E. and E. van der Heijden (2003) ‘The effect of leadership in a public bad experiment,’ Journal of Conflict Resolution 47(6): 773-795. Normann H.-T. and H. A. Rau (2014) ‘Simultaneous and sequential contributions to step-level public goods: One versus two provision levels’, Journal of Conflict Resolution, Online First Version of Record. Plambeck, E.L., H.L. Lee and P. Yatsko (2012) ‘Improving environmental performance in your Chinese supply chain’, MIT Sloan Management Review, 53(2) 43-52. Rivas, M.F. and M Sutter (2011) ‘The benefits of voluntary leadership in experimental public goods games’ Economics Letters, 112: 176-178. Rosenbloom, S. (2009) ‘At Wal-Mart, labeling to reflect green intent’ NYTimes.com (retrieved: Aug. 3, 2013), July 15. Samuelson, P. A. (1954) ‘The Pure Theory of Public Expenditure”, Review of Economics and Statistics, 36 (4): 387-389. Sutter, M., V. Levati and E. van der Heijden (2007) ‘Leading by example in a public goods experiment with heterogeneity and incomplete information’, Journal of Conflict Resolution, 5: 793-818. Wang, H., J. He and J. Mahoney (2009) ‘Firm-specific knowledge resources and competitive advantage: The roles of economic- and relationship-based employee governance mechanisms’, Strategic Management Journal, 30: 1265-1285.

27

Table 1: Features of the experiment and treatments Treatment name

Collegial no chat

Top down no chat

Collegial vertical chat

Top down vertical chat

Leadership style

Collegial

Top Down

Collegial

Top Down

Communication

No chat

Vertical chat

Show-up reward

$6

Undisclosed leader initial cash endowment

$6

Undisclosed participant initial cash endowment

$2

Leader endowment per round

0

Participant endowment per round

10

Value of a token kept

$0.10

Value of a token invested for the participants

$0.04

Value of a token invested for the leader

$0.05

Size of groups

Ten groups of five subjects per treatment, composed of one leader and four participants

Number of rounds, announced

16

28

Table 2: Message sets provided to the leaders in the experiments by leadership style Number

Situation, as described in the interface

Collegial public messages

Top down public messages

An initial message to set the tone and expectations:

It is my role to suggest ways that, working together, we can achieve high levels of investments that will generate high earnings for all.

It is my role to tell you how you can achieve high levels of investments and earnings.

After investments have increased:

Please keep it up, I'll do what I can to encourage everyone to continue increasing investments. You really have done a good job, I am learning from you.

You must keep it up. We have been successful last round, and you should continue to follow the high investment path that I laid out.

3

After investments have decreased:

We need to reverse this downward trend. My role is to support you so that we can all trust each other and switch to high investments that benefit us all.

You must reverse this downward trend. I have a clear vision of a strategy for achieving a better outcome if we all invest more.

4

After some increased and some decreased their investments:

We are getting there, but we need for everyone to pitch in and invest so that we all benefit.

You haven’t achieved as must as you should. You must follow the highinvestment guidelines.

5

After high target investment levels have been reached:

We can all take credit for reaching our goal of high investments and high earnings, which we should strive to sustain.

You have achieved my goal of high investments and high earnings, which you must sustain.

6

After investments have become stuck at a minimal, disappointing level:

This situation allows us to learn from our mistakes and we must be sure not to repeat these mistakes in the future. Think about ways to increase your investments.

I have done my best to map out a way to achieve high earnings, but you are not following my advice to invest at high levels. Now is the time for everyone to change course.

7

Desire to Punish:

Are you unclear on my instructions for you? We're trying to invest more tokens.

My instructions have been clear: you should invest more tokens.

8

Desire to Reward:

You implemented a good strategy while investing. I congratulate you.

I am happy that you followed my instructions: this led us to success.

9

Desire to Exclude:

We have to learn from our mistakes. Think about where your investment strategy went wrong.

We are a results-driven group. Unless you deliver, you are not part of the group.

1

2

29

Table 3: Summary statistics Treatment name

Statistic

Obs

Mean

Std. Dev.

Collegial no chat

Participant contribution

640

3.87

3.39

0

10

Participant earnings

640

1.23

0.29

0.52

2.08 1.85

Top down no chat

Collegial vertical chat

Top down vertical chat

Min

Max

Leader earnings

160

0.77

0.44

0

Participant contribution

640

2.74

3.17

0

10

Participant earnings

640

1.16

0.27

0

2.04 1.65

Leader earnings

160

0.55

0.42

0

Participant contribution

640

5.57

4.23

0

10

Participant earnings

640

1.33

0.33

0

2.2

Leader earnings

160

1.11

0.67

0

2

Participant contribution

640

6.35

3.95

0

10

Participant earnings

640

1.38

0.32

0

2.2

Leader earnings

160

1.27

0.57

0

2

30

Table 4: Mann-Whitney (Two-sample Wilcoxon rank-sum) test of contributions by leadership style Test

Communication

Obs.

P-value

Overall contributions

No chat

1280

0.0000

Vertical chat

1280

0.0014

No chat

32

0.0167

Vertical chat

32

0.0059

No chat

20

0.0821

Vertical chat

20

0.0059

No chat

80

0.0143

Vertical chat

80

0.0003

Contributions by round

Contributions by group

Contributions by subject

31

Table 5: Situation-based classification and group contributions (with group random effect) No chat Variables

Vertical chat

Collegial

Top down

Collegial

Top down

Positive messages

-0.209

-0.563

0.831**

1.258**

-0.549

-0.498

-0.395

-0.518

Negative messages

-1.256**

-1.054**

-0.559

-0.128

-0.597

-0.476

-0.458

-0.513

4.319***

3.539***

5.100***

5.807***

-0.629

-0.665

-0.685

-0.767

Observations

160

160

160

160

Number of groups

10

10

10

10

Constant (neutral message)

Standard errors in parentheses *** p