Preference Communication and leadership in group decision-making

Journal of Behavioral and Experimental Economics 80 (2019) 130–140

Contents lists available at ScienceDirect

Journal of Behavioral and Experimental Economics journal homepage: www.elsevier.com/locate/jbee

Preference Communication and leadership in group decision-making a

b,⁎

Seda Ertac , Mehmet Y. Gurdal a b

T

Department of Economics, Koc University, Rumeli Feneri Yolu Sariyer, Istanbul 34450 Turkey Department of Economics, Bogazici University, Bebek, Istanbul 34342 Turkey

ARTICLE INFO

ABSTRACT

Keywords: Group decision-making Risk Leadership Delegation Advice Communication Experiments

We report results from a laboratory experiment that explores the effects of preference communication and leader selection mechanisms in group decision-making. In a setting where all members of a group get the same payoff based on the leader's decision of how much risk to take, we study (1) how group members communicate their preferences to the leader, (2) whether and how the leader incorporates the communicated preferences into his/ her decision. We vary the leader selection mechanism as a treatment variable and consider cases where the leader is exogenously appointed or voluntarily self-selects into the position. We find that communicated preferences have a significant effect on actual group decisions, and that leaders’ compromise between their own preferences and the preferences of others. The data also reveal that individual characteristics matter in both what kind of suggestions are communicated to the leader, and whether leaders are likely to stick to their own preferences or compromise when making group decisions: Women and individuals who are more trusting of others are more likely to manipulate their own preferences when communicating them and more likely to compromise in response to others’ preferences as leaders.

JEL Codes: C91 C92 D81 J16

1. Introduction In many situations, an individual is faced with the task of making a risky decision that will affect the payoffs of a group of people, including herself. Individuals in executive positions in organizations have the authority and responsibility to make risky and consequential decisions on a very frequent basis. In the finance domain, mutual fund managers make investment decisions that are payoff-relevant to a number of people. In a firm, a manager can make risky or safe hiring decisions that ultimately affect the whole department's payoff. In these contexts and many others, leadership involves the authority of decision-making, which in turn brings responsibility for other people's payoffs. In this paper, we use a decision context where all members of a group earn the same payoff, based on a single member's ("leader") choice in a risky decision task, to model decision-making on behalf of a group.1 The group's interests are ex-post aligned, in the sense that a good outcome is equally good for everyone and a bad outcome is equally bad, but exante group members might have different opinions about what decision

is best. Within this context, using two separate experiments, we explore two different dimensions of the institutional structure of group decision-making. The first one is the communication of group members’ preferences to the leader, in terms of how preferences are communicated and how they affect the actual group decision. The second one is the mechanism by which the group leader is determined, that is, whether the leader volunteered for the position or was exogenously appointed. In addition, we explore the role of individual characteristics such as gender and level of trust in others, both in how individuals communicate their own preferences to others and how much they take into account others’ preferences when making a decision that affects everyone's payoff. In many contexts where group members have differing preferences as to what the group decision should be, leaders have the option to find out the preferences of, or get suggestions from group members about what to do. This is usually the case in organizations with a hierarchical structure: for example, at universities deans have the responsibility and authority of making binding decisions that affect the whole college. The

We would like to thank participants at the ESA 2012 World Meeting (New York) for helpful comments. Seda Ertac thanks the European Commission (Marie Curie International Reintegration Grant, FP-7 239,529), the Turkish Academy of the Sciences (TUBA-GEBIP program) and the Science Academy (BAGEP program), and Mehmet Y. Gurdal thanks Bogazici University (BAP 7921) for generous financial support. ⁎ Corresponding author. E-mail addresses: [email protected] (S. Ertac), [email protected] (M.Y. Gurdal). 1 We use the term “leader” throughout the paper as a shortcut for “decision-maker for the group”, since it is this particular aspect of leadership that we focus on. https://doi.org/10.1016/j.socec.2019.03.004 Received 26 July 2015; Received in revised form 15 October 2018; Accepted 8 March 2019 Available online 09 March 2019 2214-8043/ © 2019 Published by Elsevier Inc.

Journal of Behavioral and Experimental Economics 80 (2019) 130–140

S. Ertac and M.Y. Gurdal

organizational culture may be such that the dean collects information about the preferences of faculty members before making the final decision. In contrast, the decision-making process could be more "autocratic", in that the leader decides without much input from other group members.2 Even when preferences of other group members are communicated to the leader, the leader can choose whether or not to incorporate it into the decision. In our treatments with communication, we explore the questions of (1) what kind of recommendations are given by group members and whether the communication of preferences is strategic, and (2) how much of the transmitted preferences or advice is taken into account by leaders. We also explore whether there are individual characteristics that predict how individuals make recommendations to the leader given their own preferences about what to do, and how they respond to others’ preferences if they are the decision-maker. Knowing the answers to these questions is important for crafting policy implications about implementing communication/information-gathering mechanisms in groups where decision-making is delegated to a single person. For example, if preference communication mechanisms are costly to build and communicated preferences will not influence the actual decision anyway, then it may not be efficient to implement such mechanisms. Similarly, preference communication may increase the group's welfare when individuals who are responsive to others’ preferences are in the leader position. Knowing the individual characteristics that affect how leaders aggregate others’ preferences is also important for leader appointment in organizations, if a participative environment is the goal of upper management. The nature of decisions made in the leadership position can also depend on the mechanism that determines who gets to be the leader in a group, team or organization. Cases where one is exogenously appointed to make binding decisions on behalf of others are not uncommon. An upper-level manager may assign a single employee the responsibility to make project-related decisions on behalf of a team of co-workers or subordinates. For example, a political party may appoint a local representative to coordinate the election campaign for a particular district, choosing between alternative strategies with uncertain outcomes. The government may assign lower-level officers (such as village heads) to be in charge of local decision-making. In perhaps a less consequential context, an employee could be asked to organize a business event and choose between a risky new venue and a tried-andtrue one, with the resulting outcome influencing everyone's payoffs. An alternative to appointment is self-selection. Especially when there is no hierarchical decision-making structure, the leader is usually endogenously determined from among a group, with people who are more willing to take the responsibility being more likely to emerge as decision-makers. Even when there is a hierarchy as in the examples above, instead of making assignments, the authority may ask group members (co-workers, local party members, villagers, group of employees) to choose someone to be in charge. The simplest one of possible choice mechanisms involves the random choice of a member among those who are willing to be in charge. This process creates a democratic environment where people self-select into leadership. Self-selection into leadership can affect leadership style and the decisions made on behalf of the group in two ways. First, leadership willingness may be trait-dependent and these traits can also correlate with decisions. In addition, the mechanism that placed the leader into the decision-making position may have a direct effect on the leader's choices. In fact, research from psychology and sociology show that the leader's source of authority, dependent on whether the leader is "appointed" or "emergent", may influence the actions of the leader as well as the evaluation of the leader's actions by other members of the group (Read, 1974; Hollander et al., 1977; Hollander, 1992). In the context of

our experiment, the leadership selection mechanism is especially relevant for understanding the response of the leader to other group members' preferences: would a leader who did not choose to be in the leadership position be more or less likely to listen to others? In order to answer these questions, we use treatments that vary the source of leadership. In the treatments with appointed leadership, one group member is exogenously assigned to the leader position, whereas in the treatments with voluntary leadership, individuals are first asked to state whether they are willing to act as leaders, and a fair procedure (random draw) determines the leader from among willing group members.3 We analyze the effects of preference communication and leadership mechanisms using two different experiments. In a three-person, withinsubject design, the first experiment provides suggestive evidence that when telling the leader what to do, people deviate from their own preferences and that the leaders take into account both their own preference and the communicated preferences of other group members. In a cleaner, two-person design that better controls for individual preference, Experiment 2 focuses on communication and the response to communication. The results show that about 20% of the time, individuals do not exactly express their own preferences to the leader. In terms of the question of how leaders respond to others’ communicated preferences, we find that about 42% of the time leaders compromise by moving away from their own preference towards the communicated preference. This tendency to shift is crucially dependent on individual characteristics—women and individuals who trust others more are more likely to compromise. Compromise is also more likely when the advice received is towards a riskier allocation than one's own preference. Overall, individuals neither suggest their own preferences to the leader, nor do they implement their own preferences when in the leader position. While the literature has documented cautious and risky shifts in groups with mixed results, our data reveals that what matters for the direction of the shift can be what others actually want. When others’ preferences are actually revealed to the leader, this manifests itself in the form of a compromise between the person's own preference and the communicated preference. The rest of the paper is organized as follows. In Section 2, we review the relevant literature. In Sections 3 and 4 we present the design, procedures and results from the two experiments. Section 5 includes a discussion and concluding remarks.

2 Autocratic and democratic/participative leadership are in fact two major leadership styles that have been long identified and highlighted in leadership research in psychology (Lewin et al., 1939).

3 Notice that in this context where individuals are ex-ante identical and only (unobserved) preferences are different, the “self-selection into leadership” procedure mimics an election based on willingness.

2. Literature review Group decision-making and the difference between individual and group/team decisions have been important topics of study for economists (see, for example, Cooper and Kagel, 2005; Charness and Sutter, 2012). Particularly relevant to the current paper is the literature that compares individual vs. group decisions in the context of risk-taking and studies so-called risky- or cautious- shifts in the group context (Shupp and Williams, 2008; Masclet et al., 2009). While decisionmaking in groups has long been of interest, the issues of leadership willingness, leader selection mechanisms and leadership styles are getting more recent attention. Within this literature, one set of studies focus on the determinants of the endogenous emergence of leaders in contexts where there may be benefits/costs to leadership. Arbak and Villeval (2013) and Rivas and Sutter (2011) study the determinants and effects of voluntary leadership in public goods games, whereby voluntary leadership is defined as being willing to act first and lead others by example. In contrast, in our context leadership means making a binding decision on behalf of a group of people who will receive a common payoff, as in Charness et al. (2007), Sutter (2009) and Ertac and Gurdal (2012a), Ertac et al. (2018). Sutter (2009) is of particular interest,

131

Journal of Behavioral and Experimental Economics 80 (2019) 130–140

S. Ertac and M.Y. Gurdal

because it implements a treatment with an intergenerational communication channel, whereby advice can be transmitted to past and future decision-makers in a risk-taking context similar to ours. The findings are that risky investments are the most prominent recommendation, and such recommendations significantly affect actual group decisions when they are backed up by expected payoff reasoning. While Sutter (2009) shows that communication to the group decision-maker can affect group decisions significantly, our paper identifies how individual characteristics, the properties of the decision environment and the leader selection mechanism affect the way recommendations are incorporated into the group decision, within a design that elicits the leader's individual preference and enables controlling for preference differences. In addition, we also look at the correlates of the recommendations themselves. The effect of advice from other players has also been documented in settings such as the ultimatum game (Schotter and Sopher, 2007), coordination games (Schotter and Sopher, 2003) and trust game (Schotter and Sopher, 2006). In a more recent paper, Çelen et al. (2010) show that subjects are more willing the follow the advice from other players than copying their choices. Charness et al. (2013) find that the advice becomes effective only when the advice-giver has incentives to convince. Unlike these papers, which study how advice influences others' individual choices, we study the effects of preference communication in a group context with common payoffs and a single decisionmaker. Several other papers study how the presence of others affect risk-taking decisions of individuals as well as third parties. Cettolin and Tausch (2015) conduct an experiment where individuals endogenously decide how much risk to share after they make individual choices over lotteries as well as after being exogenously assigned to a lottery. They find significant differences between the endogenous and exogeneous treatments in terms of risk-sharing among high risk-takers and low risktakers. Rohde and Rohde (2015) experimentally study the behavior of impartial spectators choosing public risks and document the varying levels of aversion to ex-ante and ex-post risks and inequality. Linde and Sonnemans (2012) employ a design where individuals choose between risky options while comparing their payoffs to the fixed payoff of another subject. Observing that they can earn at least as much as the other subject makes the subject less risk-averse in this experiment. Our focus on alternative leadership mechanisms relates the current paper to a different strand of the literature that studies the behavior of elected and appointed leaders. Drazen and Ozbay (2016), for example, study the decisions of elected vs. appointed leaders in terms of their decisions in a setup similar to a dictator game. They find that elected leaders more frequently follow a non-selfish policy, which is likely guided by a reciprocity motive to reward their election by vote. The paper closest to the current paper in this literature is Kocher et al. (2012), who find that elected leaders display a more democratic as opposed to an autocratic leadership style, in terms of following the majority decision of the group. Two immediate points of departure of our paper are that we consider "self-selected" rather than elected leaders, and the leader can observe the individual recommendation of each group member, rather than a single majority decision. Another important difference is in focus–the aforementioned studies implement situations where the leader's other-regarding preferences have an important role, because of either a payoff divergence between the leader and the followers, or a different risk profile faced by the leader. In contrast, our setting involves pure payoff commonality among group members ex-post, thereby shutting down social preference motives or any direct payoff incentives for becoming or refraining from becoming leaders. The absence of election by vote in our setup also prevents a reciprocity motive in leaders' decisions. Our focus, then, is specifically on the effects of leadership mechanism on the propensity of leaders to go with their own idea or others' recommendations in implementing what is best for the group.

3. Design and results of experiment 1 3.1. Experimental design and procedures In each round of this experiment, the decision context is based on the risk allocation task of Gneezy and Potters (1997), whereby subjects decide how to allocate 10 TL between a riskless option and a risky option.4 While the amount invested in the riskless option is safe, the amount invested in the risky option is doubled with probability p, and lost with probability 1-p. The experiment consists of four different group decision-making treatments that differ in terms of the leader selection mechanism and the availability of preference communication to the leader. In addition, there is an individual decision treatment that serves as a control. For each of these five treatments, three periods are run, with the probability of winning set to 0.3, 0.5 and 0.7.5 Each subject, then, goes through 15 periods of decisionmaking, with the order of the treatment configurations randomly determined in each session to neutralize order effects. The types of treatment are presented in the following table: Treatment 1: 2: 3: 4: 5:

Individual Group T1 Group T2 Group T3 Group T4

Payoff Commonality – Yes Yes Yes Yes

Leader Selection – Appointed Self-selected Appointed Self-selected

Preference Communication – No No Yes Yes

Number of Periods 3 3 3 3 3

At the start of every group decision-making period, 3-person groups are randomly (re) drawn, and every group member learns the type of treatment: the probability of winning, whether the leader will be selfselected or appointed, and whether there will be preference communication to the leader. In all group treatments, the decision-making context is such that a single subject makes an allocation decision on behalf of the whole group, composed of three members including herself. Based on the outcome of this decision, all subjects in the same group receive the same payoff. In periods with appointed leadership, the group decision-maker role is randomly assigned to one of the three group members, whereas in voluntary leadership periods each individual is asked, at the start, whether they would like to be the decision-maker for their group or not. In the latter, if more than one person wants to be the decision-maker, a random draw among those determines the leader. If no one wants to be the decision-maker, one of the three people in the group is selected randomly to make the decision. If the period is a "no communication" period, the leader proceeds to make the group decision. If the period is a "communication" period, before the leader makes her choice, the nonleader group members are individually asked how much they would allocate to the risky option, if they were the group decision-maker.6 This information is then made available to the leader, but the leader has the option to view or not view it before he/she makes the group choice. This allows us to have more observability and control over how much information is taken into account, and by which types of leaders, since not viewing the information at all is an extreme case of disregarding others’ preferences. The leader then makes her choice. In every group treatment, other group members can observe the leader's decision, but not the outcome.7 Finally, at the end of any group 4

At the time of the experiments, 1 TL corresponded to $0.62. Notice that in the original paper by Gneezy and Potters (1997), the amount invested in the risky option is multiplied by 2.5 with probability ⅓ and is lost with probability ⅔. 6 The non-leader group members know that the leader can see this information, therefore strategic advice is possible. 7 The goal was to prevent spillovers on risk-taking behavior from positive and negative outcomes of previous risky decisions of oneself and others. 5

132

Journal of Behavioral and Experimental Economics 80 (2019) 130–140

S. Ertac and M.Y. Gurdal

period, non-leader members communicate their opinion of the implemented decision to the group leader, in binary like-or-dislike messages that are not payoff-relevant and designed to mimic the nonmonetary consequences of making good or bad decisions for others. Subjects are aware of this from the start. In each new period, groups are shuffled, the leadership procedure is re-applied, that is, a (potentially) new leader is determined, subjects learn of the treatment type, and then make their decisions according to their roles. The experiments were conducted at two universities in Turkey, Koc University and TOBB ETU, using undergraduate students as subjects. The experiment was programmed using Z-tree (Fischbacher (2007)). 10 sessions were run, and we have data from 156 subjects in total: 57 females and 99 males. All subjects were paid a show-up fee of 5 Turkish Liras, in addition to the amount earned in the experiment. Subjects never learned who was in their group. One of the 15 periods was chosen randomly at the end for payment. If an individual task was chosen for payment, subjects were paid on the basis of their own decisions. If a group task was chosen for payment, all three subjects that form a group got the same payoff, based on the decision of the group leader.

treatment (Ertac and Gurdal (2012a); Arbak and Villeval (2013); Banerjee et al., (2015)). Research Question 1: Is there a difference between the extent of risk-taking on behalf of the group when the leaders are appointed versus when they self-select into leadership? Research Question 2: In settings where communication is available, do individuals communicate their own preference to the leader exactly, when not in the decision-making role? Research Question 3: Do communicated preferences affect leaders’ decisions? Do riskier suggestions lead to riskier actual group decisions, controlling for the leader's own preference? Research Question 4: Is there a difference between men and women in terms of volunteering for the decision-making role? 3.3. Results of experiment 1 We present our results in two main subSections, 3.3.1 and 3.3.2, where we focus on group risk taking decisions with and without preference communication, respectively. Before we set out to answer our research questions regarding communication and leadership mechanisms, we start by examining leadership willingness in the voluntary rounds. In the full data, we observe that individuals state a willingness to make the group decision 85% of the time (86% under no communication, and 84% under communication). As for the question of whether there are gender differences in leadership willingness, our data confirm Ertac and Gurdal (2012a): on average, men are significantly more willing to make the group decision than women, with men volunteering 89.9% of the time and women 76.3% of the time (p < 0.01 in a Mann-Whitney test).8

3.2. Research questions There are several research questions that motivates the design of Experiment 1. These questions mainly focus on the effects of the leadership mechanism (assignment vs. self-selection) on the choices made on behalf of the group, and on the patterns of preference communication; in particular, whether communicated preferences are different than one's own preferences. Below, we flesh out these research questions, in light of the literature. Given the literature that shows that individuals behave differently when their decisions affect others’ payoffs (e.g. Charness et al. (2007), Sutter (2009) and Ertac and Gurdal (2012a)), we expect that individuals’ decisions will also be responsive to communicated preferences by others, that is, we expect that communicated preferences will affect the decision made by the leader, controlling for the leader's own preference. A major research question we aim to answer is differences in risk-taking by assigned and self-selected leaders. The psychology literature on leadership has pointed to the source of leadership authority as an important determinant of how leaders act and are perceived (Eslgie et al., (1988); Hollander and Julian (1969); Hollander et al., (1977); Read (1974)). Our design aims to study this difference in the context of the leaders’ decisions, and contributes to the literature by testing whether different leader selection mechanisms (exogenous appointment vs. volunteering) are associated with different levels of risktaking on behalf of the group. As mentioned before, there are two reasons why one may expect a difference between leaders’ choices as a function of the leader selection mechanism: it may be that self-selection favors a certain type of personality, e.g. having a more self-centered attitude and less regard for others, and/or it may be that the source of authority itself (having been assigned the responsibility versus having volunteered for it while others probably did not) may directly affect decisions. Supporting the self-selection mechanism there is some evidence that individuals who would be willing to pay to make decisions on behalf of others are less altruistic (Ertac et al., (2018)). Next, we take the non-leader perspective and ask whether individuals communicate their preferences directly to the leader, that is, the recommendation they give to the leader is equal to their own risk preference, or not. This is equivalent to asking whether there are strategic concerns when reporting preferences, such as stating a more extreme preference with the expectation that the leader will curb the recommendation anyway. In addition to these main questions, we ask whether the role of individual characteristics and contextual parameters such as gender and the probability of winning affect communicating preferences, group decisions, and the willingness to lead. Based on prior literature, we expect that women will be less likely to volunteer in the self-selection

3.3.1. Group decisions without preference communication Table 1 provides summary statistics on the risk taken on behalf of the group, when the leader does not have access to recommendations by other group members. In this setting, self-selected and appointed leaders take similar levels of risk on behalf of the group (Mann Whitney test, p-value = 0.697). As expected, the probability of winning uniformly increases the risk taken on behalf of a group, as well as the risk taken individually. An important question in studying group decisions is whether individuals take more, less or similar amounts of risk when deciding on behalf of a group, as compared to how they decide individually. On average, there is no significant difference between the risk taken for a group and for oneself, when individuals receive no recommendation from other group members (Wilcoxon signed-rank test, pvalue = 0.62). We find that 41.7% make exactly the same decision they make individually, whereas 31.1% and 27.2% engage in risky and cautious shifts, respectively. Regressions (not reported here) of the within-person difference between group risk and the risk taken in the corresponding individual round show that the leader selection mechanism does not make a significant difference in the frequency of cautious vs. risky shifts. The regressions reported in Table 2 outline the determinants of leader risk-taking when preference communication is absent. We find that the risk taken by a leader in the corresponding individual decision period (termed as “own risk”) significantly affects the risk taken for the group, meaning that the leader's own preferences matter. On the other hand, the regression results show no significant effect of leadership selection mechanism (appointed versus self-selected) on the amount of risk taken on behalf of the group. 3.3.2. Treatments with preference communication 3.3.2.1. Communicated preferences. We first explore the question of what kind of preferences are communicated to the leader by group members. The first issue here is whether information transmission to the leader is strategic, that is, whether individuals report riskier or more 8

133

All non-parametric tests are conducted using subject-level averages.

Journal of Behavioral and Experimental Economics 80 (2019) 130–140

S. Ertac and M.Y. Gurdal

absolute value of the difference between the recommendation to the group leader and the actual individual risk taken in the corresponding individual decision round, we find that this difference is non-zero 57.9% of the time, with a mean of 1.51, and a t-test rejects the hypothesis that the advice given to the leader and the risk taken individually have the same means (p-value = 0.075) In other words, preference communication is overall strategic with respect to individual risk. While 30.9% give riskier recommendations, 26.9% give more cautious recommendations than the risk they take for themselves. With regards to our research question 2, we see that individuals do not simply communicate their own preferences to the leader.10 The next question is what determines the direction of deviations from individual preferences when communicating to the group leader. In particular, we look at the propensity to give the same, higher or lower advice than one's own preference. The second column of Table 3 show that as the probability of winning reaches the highest level, individuals are more likely to just state their own preferences when making suggestions to the leader. Own risk preference being higher increases the likelihood of a cautious deviation (but less so for males), whereas own risk preference being lower increases that of a risky deviation when giving an advice. That is, individuals seem to generally deviate to more moderate choices when communicating preferences, rather than exaggerating their preference so that it will be implemented “correctly” even if the leader takes it partially into account.11

Table 1 Group Risk without Preference Communication.

Self-selected Appointed All N

p = 0.3

p = 0.5

p = 0.7

2.692 (2.356) 2.538 (2.173) 2.615 (2.257) 104

4.942 (1.984) 5.423 (2.585) 5.183 (2.306) 104

7.577 (2.562) 7.192 (2.686) 7.385 (2.619) 104

Notes: Amount invested (out of 10 TL) as main number, standard errors in parentheses. Table 2 Leader Risk-taking without Communication in Experiment 1. Appointed Own Risk Male p = 0.5 p = 0.7 Period Constant N No. of Subjects

Preference −0.116 (0.22) 0.511⁎⁎⁎ (0.0661) 0.165 (0.265) 1.254⁎⁎⁎ (0.361) 2.160⁎⁎⁎ (0.516) 0.0207 (0.037) 1.112⁎⁎ (0.434) 312 143

3.3.2.2. Group risk decisions with preference communication. Table 4 provides summary statistics on the overall risk taken by leaders across leadership treatments under the availability of preference information. In this case, similar to the periods where individuals receive no recommendation (see section 3.3.1), there is no significant difference between the average risk taken for the group vs. for oneself (Wilcoxon signed-rank test, p-value = 0.83). Noting that communicated preferences are viewed by leaders 85% of the time, we turn to the questions of (1) whether and how strongly this information affects decisions, and (2) whether its effects interact with the leader selection mechanism.12 The first question is whether the two pieces of information/recommendations received from the other group members have a significant impact on the risk taken on behalf of the group. In a regression where we pool all data (Table 5), we find that the average communicated risk preference indeed has a significant effect on the group risk taken: when the members recommend taking higher risk, actual group decisions are riskier. These results are robust to using the maximum and/or minimum of the recommendations, instead of the average.13 On the other hand, as in the case of periods without preference communication (Table 2) the effect of leadership mechanism and gender on the risk taken for the group is insignificant.

Notes: Amount invested as dependent variable, standard errors (clustered at subject level) in parentheses. *p < 0.1. ⁎⁎ p < 0.05. ⁎⁎⁎ p < 0.01.

cautious preferences to the leader compared to their own true risk preferences. Such deviations from own preferences may happen for two reasons: First, individuals with other-regarding preferences may not want to impose their own preference on others, and may therefore shape their suggestion in line with what they think others may want (for example, an individual who is very risk-loving may not want to recommend what he/she himself would do to others).9 Alternatively, individuals who want their own preference implemented and who believe that the leader will only partially compromise may exaggerate their preference (someone who wants a high level of risk taken and who believes others are more risk averse than him/her may recommend more risk than they would take for him/herself). When we look at the

4. Design and results of experiment 2 The results of the first study show that (i) leaders are responsive to

9

Although final payoffs are the same for everyone in our setup, at the point where the uncertainty has not yet been resolved, the expected utilities of people with different preferences are different, and if they honestly communicate their preferred allocation, that is the choice that maximizes their expected utility. That is, a person who is very risk-averse will attain the highest expected utility when all is invested into the safe option, while a more risk-tolerant individual will attain the highest expected utility with a more risky allocation, although both will get the same ex-post payoff no matter what. Someone who changes her own choice and moves closer to the other person's choice (or makes a recommendation that is more in line with what others may want) may be aiming to minimize expected utility differences between herself and the other person. Therefore, in expectation, it may still be that the compromising individual is behaving in an “inequality averse” way. Alternatively, there may be a regret cost to having made the “wrong” choice or wrong recommendation ex-post, and this cost may be higher for individuals who are more other-regarding (Mengarelli et al. (2014)).

10 Note, however, that the frequency of cases where individual and communicated preferences differ (57.9%) goes down to about 20% in Experiment 2 where individual preferences are more precisely measured, suggesting that some of the differences here are probably due to noisy measurement of individual preferences. 11 In unreported regressions, we show that this move towards more moderate choices still holds if we take out the extreme cases where there may not be much room to move to one direction. Since advice responds to the probability of winning in the expected way as well, it seems unlikely that the result is driven by random behavior. 12 In unreported regressions, we find that the likelihood of viewing the advice does not depend on our treatment variables. 13 When we include both the maximum and the minimum of the preferences together, both of them turn out to have significant effects on leader's decision.

134

Journal of Behavioral and Experimental Economics 80 (2019) 130–140

S. Ertac and M.Y. Gurdal

where the leader is appointed and where the leader is selected, which might induce spillover effects across different treatments (for example, once an individual is put into the leader role exogenously, he/she may be more willing to decide in later periods). In addition to this, individual preferences of the subjects could be changing across periods as they get recommendations from other group members, and the information collected in the individual decision rounds, which may come earlier or later than the corresponding group round, might not be sufficient to control for individual preferences that dynamically change. Finally, in the three-person setup it is difficult to examine how much the leader shifts away from her own preference towards the preferences of the others as there are two stated preferences that may conflict with each other, and it is not clear how one should conceptualize the strength of the leader's response to these preferences. Motivated by these concerns, we design and conduct a second experiment where (i) the mechanism used to determine the group leader is fixed throughout a given session and (ii) information about individual preferences is collected in all periods. In the second study, there are two between-subjects treatments: appointed leaders, and self-selected (voluntary) leaders. Given that we would like to further explore, based on the first study results, how leaders respond to recommendations by others, all treatments involve communication by the non-leader. In addition, we use a two person-setup which allows us to cleanly measure and analyze whether and how the leader incorporates the other person's preference into her decision. In particular, this setup lets us conceptualize “compromising” behavior as whether the leader moves toward the (single) suggestion and away from his/her own preference, and study what kind of individual and institutional characteristics are associated with the propensity to compromise.

Table 3 Preference Communication to the Leader in Experiment 1.

Appointed Male Own Risk p = 0.5 p = 0.7 Period N Number of Subjects

Lower Advice

Same Advice

Higher Advice

−0.00441 (0.0282) −0.122⁎⁎ (0.0384) 0.0820⁎⁎⁎ (0.00700) −0.239⁎⁎⁎ (0.0457) −0.387⁎⁎⁎ (0.0576) −0.00420 (0.00319) 624 155

0.00841 (0.0347) 0.0843 (0.0480) −0.0126 (0.00845) 0.0977 (0.0545) 0.166* (0.0656) 0.0121⁎⁎ (0.00424) 624 155

−0.00400 (0.0312) 0.0372 (0.0431) −0.0694⁎⁎⁎ (0.00767) 0.142⁎⁎ (0.0472) 0.221⁎⁎⁎ (0.0639) −0.00795* (0.00396) 624 155

Notes: Marginal effects from multinomial-logistic regression reported. Outcomes: Lower Advice = 1 if Advice < Own Risk, Same Advice = 1 if Advice = Own Risk, Higher Advice = 1 if Advice > Own Risk. Standard errors (clustered at subject level) in parentheses. ⁎ p < 0.1. ⁎⁎ p < 0.05. ⁎⁎⁎ p < 0.01. Table 4 Group Risk with Preference Communication in Experiment 1.

Self-selected Appointed All N

p = 0.3

p = 0.5

p = 0.7

2.731 (2.466) 3.019 (2.429) 2.875 (2.44) 104

5.442 (2.296) 4.846 (2.296) 5.144 (2.304) 104

8.077 (1.867) 7.577 (2.208) 7.827 (2.05) 104

4.1. Experimental design and procedures As mentioned above, the second experiment was run after analyzing the data from the first experiment, with the goal of presenting a cleaner set of results regarding leader selection and the response to others’ preferences. While the basic decision context was the same as in Experiment 1 (the Gneezy-Potters task with varying probabilities of winning in each round), there were a few crucial differences. The experiment lasted for 12 periods and in each period, subjects were randomly matched to form 2-person groups, with changing pairings every period. The main task, which involved allocating a constant sum between a safe and a risky option, was the same as in the first study, except that the amount to be allocated was increased to 15 TL from 10 TL. The probability of the good outcome could be 0.3, 0.5 or 0.7 as in the first study, and each of these values was used for a total of 4 periods, with the order being random across periods. The order of the tasks was as follows:

Notes: Amount invested (out of 10 TL) as main number, standard errors in parentheses. Table 5 Leader Risk-taking with Preference Communication in Experiment 1. Appointed Average Advice Own Risk Male p = 0.5 p = 0.7 Period Constant N Number of Subjects

−0.161 (0.195) 0.259⁎⁎⁎ (0.0752) 0.343⁎⁎⁎ (0.0711) 0.275 (0.233) 0.487 (0.386) 2.054⁎⁎⁎ (0.531) −0.0504* (0.0258) 1.560⁎⁎⁎ (0.447) 312 129

– Subjects made an investment decision for themselves in every period. – With a probability of 0.25, this decision was used to determine the payments in that period. With probability 0.75, this decision did not count and one of the subjects in the group was determined as the group leader to make an investment decision for both members. In treatment “appointed”, this person was randomly determined. In treatment “selected”, this person was determined using the same mechanism as in the voluntary leadership periods of Experiment 1. That is, each participant expressed whether he/she would like to be the group leader. If both group members said yes and both said no to leadership, the leader was randomly determined. If only one group member wanted to be the leader, that person would be appointed as the leader. – The person who was not designated as the group leader made a nonbinding recommendation to the group leader. – The leader saw this recommendation and made an investment decision for the group.

Notes: Amount invested as dependent variable, standard errors (clustered at subjects level) in parentheses. ⁎ p < 0.1. **p < 0.05. ⁎⁎⁎ p < 0.01.

the information about group members’ preferences and (ii) these preferences are communicated strategically to the leader. However, several aspects of the design might be of concern in interpreting the results of this study. In particular, the same subjects go through treatments both 135

Journal of Behavioral and Experimental Economics 80 (2019) 130–140

S. Ertac and M.Y. Gurdal

– The other subject was informed about the leader's decision, but was not given any feedback regarding the outcome of the draw for the risky option.

4.2. Results of experiment 2 We start by noting some observations regarding leadership willingness. In the voluntary leadership treatment, we observe that subjects are willing to make the group decision 85.8% of the time. The corresponding ratios are 94% for male and 75.5% for female subjects. In an unreported logistic regression we find that male subjects are around 17% more likely to be willing to lead the group, controlling for probabilities of winning and period. This result is highly significant, and lends support to both what we observed in Experiment 1 and the result reported in Ertac & Gurdal (2012a), indicating a robust gender difference in leadership willingness. Individuals who do not agree with the statement of “most people can be trusted” are willing to be the decisionmaker 87.3% of the time, whereas this fraction is 76.8% for those who agree with this statement. However, a logistic regression that controls for characteristics such as gender and the probability of winning fails to find a significant difference among two groups. In Table 6 we provide summary statistics regarding the investment decisions of subjects during the individual rounds and leadership rounds. As in Experiment 1, we do not find a systematic difference between appointed and self-selected leaders in terms of the risk taken on behalf of the group (p = 0.68, Mann-Whitney test). This finding holds both when decisions are compared in the aggregate and when they are compared separately across different probabilities of winning. As expected, probability of winning has a positive effect on the amount invested and the investment decisions are similar to those of Experiment 1 in terms of the percentages of the endowment. Compared to their individual decisions for a given round, appointed leaders make a cautious (risky) shift around 16% (25.7%) of the time, whereas they make the same decision during the remaining rounds. The frequencies are very similar for self-selected leaders who make a cautious (risky) shift around 19.9% (24.8%) of the time.

At the end of the experiment, one period was randomly selected to determine the payments of the subjects. Subjects were also asked to complete a post-experiment survey that included demographics as well as a question on the propensity to trust others, taken from the World Values Survey.14 The average earnings were about 24.5 TL.15 The experiment was conducted at the Bogazici University Economics Laboratory and we had 9 sessions and 142 subjects in total with each session lasting around 40–50 min. Subjects were invited by an e-mail that was sent to students who previously indicated an interest in participating economics experiments and they could register online for a given session. No subject participated more than once. This design allows us to answer these additional research questions: Research Question 5: Are assigned leaders’ decisions are more affected by the communicated preference than self-selected leaders. In particular, are assigned leaders more likely to compromise (move towards the suggested choice and away from their own), controlling for the preference difference? Research Question 6: Controlling for the difference between the communicated preference and the leader's own, does riskier advice has the same likelihood of being taken into account as more cautious advice? Research Question 7: Do personal characteristics such as gender and interpersonal trust affect the level of leader's compromise? We believe that compared to Experiment 1, the new design features in Experiment 2 such as the better measurement of individual risk preferences and the two-person setup allow us to understand how leadership selection mechanism interacts with decision environment, and also observe whether leaders are more affected by riskier or more cautious recommendations. As in Experiment 1, we also explore the role of individual characteristics and contextual parameters such as gender and the probability of winning in communicating preferences, group decisions, and the willingness to lead. Similarly to Experiment 1, we hypothesize that females will be less likely to volunteer. We also test whether women are more likely to move away from their own preferences to others’ preferences (compromise more) when they are leaders. It is probable that women would be responsive to others’ preferences when they are leaders, as research in psychology shows that female leaders adopt a more democratic leadership style (see Eagly and Johnson (1990) for a review). In addition, we expect that as in the case of females, more trusting individuals will also be (1) less likely to volunteer for leadership, because they think it is not likely that others will make an irrational or extreme choice, (2) more likely to compromise (move away from her own preferences towards others’ stated preferences) once in the leader role, because they are more likely to believe that others’ preferences have merit. In fact, the literature shows that trust is positively correlated with the Big-Five personality trait of agreeableness (Evans et al. (2008); Dohmen et al., (2008)), which has in turn, been shown to reduce the willingness to take on the leader role in our setup (Ertac and Gurdal (2012b)). Trust is very likely to affect whether one compromises with others’ requests as well (in fact, Barry and Friedman (1998) show that more agreeable bargainers tend to compromise more in response to others’ demands in negotiation). Finally, trust may also be relevant on both the volunteering and compromise fronts because of its relation to prosocial traits in general (Ertac et al., (2018) show that altruistic individuals have a lower preference for making decisions on behalf of others).

4.2.1. Communicated preferences The design of the second experiment allows us to see whether the subjects deviate from their individually stated preferences when giving advice to the group leader in a dynamically accurate way, by eliciting individual preferences in every round. We find that the advice given to assigned and self-selected leaders do not exhibit a significant difference (p = 0.98, Mann-Whitney test) and this is also true when we break up the decisions based on different probabilities. We observe that subjects’ advice to the leader and their individual preference are the same around 78.5% of the time when leaders are appointed and 81.1% of the time when leaders are self-selected, indicating that advice to the leader is a highly reliable proxy for the individual preference of the non-leader subject in the group. At the same time, the suggestion to the leader deviates from the individual preferences in 140 out of 686 cases where the group decision was delegated to the leader. The average absolute difference between advice and individual preference is around 2.82, but using a paired t-test, we cannot reject the hypothesis that the advice given to the leader and the risk taken individually have the same means (p-value = 0.504). A relevant question here is how likely subjects are to give a suggestion that is the same as/lower than/greater than their own preference, which is stated at the beginning of each period. In a multinomial logistic regression, reported in Table 7, we find that subjects’ own risk preference or the mechanism with which the leader is chosen has no effect on the likelihood of giving a suggestion that is equal to the individual preference (column 2). When it comes to the likelihood of giving an advice lower than the individual preference, we find that individual risk preference has a highly significant positive effect (column 1). In addition, we observe just the opposite and a highly significant effect when we examine the likelihood of giving an advice greater than the individual preference (column 3). This corroborates the finding in Experiment 1 that individuals do not simply communicate their own preference to the leader, and tend to move towards a

14 We hypothesized that trust would be an important individual characteristic in leadership and the response to others’ recommendations, as explained below. 15 At the time of the experiments, 1TL corresponded to $0.33.

136

Journal of Behavioral and Experimental Economics 80 (2019) 130–140

S. Ertac and M.Y. Gurdal

Table 6 Group Risk without Preference Communication in Experiment 2.

Self-selected Appointed All N

Table 8 Leader Risk-taking with Preference Communication in Experiment 2.

p = 0.3

p = 0.5

p = 0.7

3.167 (3.013) 2.713 (2.854) 2.886 (2.918) 220

7.47 (2.98) 7.758 (3.735) 7.632 (3.42) 228

11.49 (2.362) 11.77 (2.553) 11.65 (2.472) 238

Appointed Advice Own risk Male

Notes: Amount invested (out of 15 TL) as main number, standard errors in parentheses.

p = 0.5 p = 0.7

Table 7 Preference Communication to the Leader in Experiment 2.

Appointed Male Own risk P = 0.5 P = 0.7 Period Interpersonal Trust N Number of Subjects

Period

Lower Advice

Same Advice

Higher Advice

−0.0117 (0.0264) −0.0552⁎⁎ (0.0264) 0.0165⁎⁎⁎ (0.00402) −0.0812⁎⁎ (0.0371) −0.103⁎⁎⁎ (0.0399) −0.0103⁎⁎⁎ (0.00363) 0.0660⁎⁎ (0.0287) 686 141

−0.031 (0.0417) 0.109⁎⁎⁎ (0.0409) 0.00453 (0.00523) 0.0059 (0.044) −0.0893 (0.0614) 0.0107⁎⁎ (0.00483) −0.0953* (0.0506)

0.0427 (0.0276) −0.0537* (0.0275) −0.0211⁎⁎⁎ (0.00436) 0.0753⁎⁎ (0.0328) 0.193⁎⁎⁎ (0.0469) −0.00039 (0.00369) 0.0293 (0.0336)

Constant N Number of Subjects

All

Appointed

Self-selected

0.117 (0.155) 0.171⁎⁎⁎ (0.028) 0.722⁎⁎⁎ (0.0475) −0.0813 (0.145) 0.912⁎⁎⁎ (0.32) 1.315⁎⁎ (0.54) −0.0181 (0.0189) 0.237 (0.26) 686 141

0.208⁎⁎⁎ (0.04) 0.750⁎⁎⁎ (0.0613) −0.241 (0.194) 0.858* (0.465) 0.85 (0.719) −0.00657 (0.0255) 0.0735 (0.292) 400 80

0.111⁎⁎⁎ (0.0374) 0.679⁎⁎⁎ (0.0755) 0.149 (0.223) 1.071⁎⁎ (0.448) 2.030⁎⁎ (0.841) −0.0224 (0.0276) 0.598 (0.471) 286 61

Notes: OLS regression. Standard errors (clustered at subject level) in parentheses. ⁎ p < 0.1. ⁎⁎ p < 0.05. ⁎⁎⁎ p < 0.01.

preference. In table 9, we summarize the leader's response to advice (i.e. leader's choice shift with respect to her own preference), conditional on the relative comparison of the advice and the leader's own preference for that round. The results reveal an expected, though to our knowledge, a previously undocumented pattern. When a leader receives a recommendation that points to higher (lower) investment compared to her own individual preference, she reduces (increases) the amount of risk she takes or keeps it the same while deciding for the group, behaving as if she uses a convex combination of the advice and his individual preference. In order to analyze this result more rigorously by controlling for other variables related to the decision context and studying the role of individual characteristics, we construct a “compromise” variable that takes the value of 1 if the leader shifts his/her choice towards the suggested decision, in cases where the suggested decision is different than what the leader would like to do herself. Table 10 shows the results from probit regressions of the propensity to compromise. Similar to the results in Table 8, we find no significant difference between the propensities of assigned and selected leaders to compromise (Column 1). Instead, individual characteristics seem to strongly predict the propensity to compromise. Specifically, female leaders are 20 percentage points more likely than males to change their decisions in response to the other person's recommendation, controlling for the distance between their own preference and the

Notes: Marginal effects from multinomial-logistic regression reported. Outcomes: Lower Advice = 1 if Advice < Own Risk, Same Advice = 1 if Advice = Own Risk, Higher Advice = 1 if Advice > Own Risk. Standard errors (clustered at subject level) in parentheses. ⁎ p < 0.1. ⁎⁎ p < 0.05. ⁎⁎⁎ p < 0.01.

less (more) risky option if they are individually more (less) risk averse. We also document an interesting gender effect: women are about 10 percentage points more likely to change their own preference when making a recommendation to the leader. Individuals who are more trusting of others are also less likely to keep to their own preference when making a recommendation, and are likely to make more cautious recommendations (column 1). 4.2.2. Group decisions and the response to communicated preferences We first look at the effects of the suggestion of the other group member on the decision the leader makes. In Table 8, we report results from a set of regression models where the dependent variable is the decision made on behalf of the group. Using the full sample, Column 1 shows that both the received recommendation and the leader's individual preference have a highly significant effect on the group risk decision. There is no difference between assigned and self-selected leaders in terms of the risk they take for the group. In Columns 2 and 3, we separately do the analysis for assigned and self-selected leaders, respectively, and we find that advice has a significant effect on the decisions of both types of leaders. A particular strength of the design in Experiment 2 is that we are able to consider a single communicated preference, which makes it easier to clearly identify its effect on the decision along with the leader's own preference. If we conceptualize the leader's choice as a potential compromise between her own preferences and the preferences of the other individual, then an important object of interest is whether the leader moves toward the suggestion and away from his/her own

Table 9 Preference Communication and Leader's Choice Shift in Experiment 2.

Advice > Leader's Preference Advice = Leader's Preference Advice < Leader's Preference

Leader's Choice Shift

Treatment Appointed

Self-Selected

Risky Shift No Shift Cautious Shift Risky Shift No Shift Cautious Shift Risky Shift No Shift Cautious Shift

89 102 8 6 49 0 8 82 56

56 64 4 4 27 1 11 67 52

Notes: Number of observations reported. Leader's choice shift is classified according to her own preference for the corresponding period and the amount of risk she takes for the group. 137

Journal of Behavioral and Experimental Economics 80 (2019) 130–140

S. Ertac and M.Y. Gurdal

When we precisely control for individual preferences (as in Experiment 2), we find that about 20% of the time, individuals deviate from their own preferences in making recommendations to the leader. This suggests that some individuals may not want to impose their own preferences on others even when making recommendations. Results from Experiment 2 provide further evidence in favor of this, by documenting individual characteristics that predict how individuals make recommendations. Specifically, we find that women and individuals who are more trusting of others are less likely to stick to their own preference when making a recommendation to the leader. Individuals who are more trusting of others are also more likely to deviate from their own preference, and shift in a cautious way when making recommendations. Experiment 2 also highlights that these two individual characteristics (being female and more trusting of others) independently predict whether leaders compromise when they receive a recommendation that is different from their own preference.17 These results suggest that women will have a more democratic style in terms of incorporating others’ preferences when in the leadership role, in addition to being more sensitive to the group context when they are making recommendations to others. This may have implications for assignment of decision-making responsibility in organizations that aim to construct a more democratic environment in teams or groups. Under self-selection, leaders in our experiment are determined by a random draw among those who indicate a willingness to decide on behalf of the group. This process might not be fully representative of the way that leaders get to be elected in real-life, but we believe it involves an essential aspect of this process, namely, the assertion for being the leader. While the election of a leader often involves a vote between those who choose to run as candidates, our focus is on leadership in a context with preference differences but no inherent heterogeneity in informational advantages or leadership capacity. A vote in our case would therefore be somewhat unnatural, because the candidates would have no reasonable record for evaluation by others. Moreover, voting might also trigger reciprocal responses which we would like to abstract from in focusing on the integration of others’ preferences into the leader's decision. Still, we believe that it would be fruitful to study the effects of recommendations in settings where leaders are chosen on the basis of individual characteristics in further research. Finally, we should note that our results may be specific to the risky decision context. In other decision contexts, where the leader makes allocation decisions or choices over time, for example, individuals may behave differently, both in terms of volunteering and in terms of taking into account others’ preferences. We believe these differences are important to explore in future research.

Table 10 Leader's Compromise in Experiment 2.

Appointed Male Own Risk Period Interpersonal Trust Preference Difference Risky Advice

I

II

III

−0.0102 (0.0508) −0.206⁎⁎⁎ (0.0475) 0.00478 (0.00471) −0.00655 (0.00582) 0.163⁎⁎⁎ (0.0585) 0.0114 (0.00829)

−0.0173 (0.0502) −0.204⁎⁎⁎ (0.0471) 0.00925* (0.00506) −0.00565 (0.00569) 0.167⁎⁎⁎ (0.0576) 0.0111 (0.00827) 0.0954⁎⁎ (0.0437)

599 140

599 140

−0.0162 (0.0504) −0.202⁎⁎⁎ (0.0473) 0.00384 (0.00919) −0.00549 (0.00589) 0.164⁎⁎⁎ (0.0573) 0.012 (0.00819) 0.0742 (0.0516) 0.0256 (0.0677) 0.0675 (0.0975) 599 140

p = 0.5 p = 0.7 N Number of Subjects

Notes: Marginal effects from probit regressions reported. Dependent variable: 1 if leader's choice shifted towards the suggestion (compromise), 0 otherwise. ⁎ p < 0.1. ⁎⁎ p < 0.05. ⁎⁎⁎ p < 0.01.

recommendation. Individuals who have a high level of trust in others are (an additional) 16 percentage points more likely to compromise. In Column 2, we consider whether the advice being riskier or more cautious with respect to the leader's own preference has an effect on the leader's propensity to compromise. Wefind that controlling for their individual preference (Own Risk) and the absolute distance between their preference and the suggested decision (Preference Difference), leaders are more likely to compromise in response to a suggestion to take higher risk (Risky Advice) than their own. This effect becomes insignificant when we control for the probability of winning (Column 3), although the sign of the effect doesn't change. In all models, the effect of the absolute difference between the advice and the leader's own risk has the expected (positive) sign but this effect is insignificant. 5. Discussion and concluding remarks In two separate experiments, we demonstrate that leaders’ risktaking on behalf of the group depends both on the leader's own preference and information about group members’ preferences. This information, communicated by the members to the leader is oftentimes strategic, as it can show substantial deviations from the true individual preferences of group members. We do not document systematic difference between the risk decisions made on behalf of the group by appointed and self-selected leaders. While the pure (treatment) effect of leadership may be important, our data also reveal some insights about leadership willingness based on individual characteristics. We find that although volunteering rates are in general quite high, women are still less frequently willing to become leaders than men in both experiments, consistently with the prior literature on gender differences. As we will discuss below, individual characteristics also seem to matter for how individuals behave once in the leadership role. In fact, Ertac and Gurdal (2012b) study the correlation of personality traits with the willingness to be a leader, and find agreeableness to be negatively correlated with leadership willingness, and Ertac et al., (2018) find that more altruistic individuals are more likely to refrain from making decisions on behalf of others.16 16

Supplementary materials Supplementary material associated with this article can be found, in the online version, at doi:10.1016/j.socec.2019.03.004. Appendix Instructions (Experiment 1) Welcome to this study on economic decision-making. The study will last for 15 rounds, and in each round, you will make certain decisions. Your monetary earnings from the study will be based on these decisions, the decisions of others, and on chance. These earnings will be added to the participation fee and will be paid to you in cash, in private at the end of the study. The decisions that you make will be recorded (footnote continued) direction in their study on voluntary leadership in public good contributions. 17 This result holds both in the assigned and in the selected treatments.

Arbak and Villeval (2011) find a small effect of agreeableness in the same 138

Journal of Behavioral and Experimental Economics 80 (2019) 130–140

S. Ertac and M.Y. Gurdal

with a participant ID number randomly assigned to you by the computer. Therefore, your decisions will not be linked to your name. If you have any questions throughout the study, please raise your hand.

Communication by Other Group Members In some of the group rounds, the group members who are not making the group decision will be asked to state what they would have done, had they been the one deciding on behalf of the group. The group decision-maker will have access to this information, and will be able to observe it at no cost, if he/she likes. In some other group rounds, the decision-maker will decide without access to the preferences of other group members. Throughout the study, in each round, the rules for that particular round will be displayed on your computer screen. In other words, in each round, you will know:

The Decision Task At every round of the study, the decision to be made is to split 10 Turkish Liras into a risky and a riskless option. What happens to the amount allocated to the risky option depends on whether the "good outcome" or the "bad outcome" occurs. In the case of a good outcome, the amount put in the risky option is multiplied by 2. In the case of a bad outcome, it is lost. The probabilities of the good and bad outcome will change from round to round. In particular, in some rounds the likelihood of the good outcome will be 70% (and that of the bad outcome 30%), in some rounds the likelihood of the good outcome will be 50% (and that of the bad outcome 50%), and in some rounds the likelihood of the good outcome will be 30% (and that of the bad outcome 70%). In every round, the relevant probabilities will be shown to you on your screen, before you make your decision. Any amounts put into the riskless option stays as it is. Example: Suppose that in a round, the likelihood of the good outcome is 50% (and that of the bad outcome 50%). Suppose that one puts 5 TL into the riskless and 5 TL into the risky option. In the case of the good outcome, this person would get 5 × 2 + 5 = 15 TL. In the case of the bad outcome, this person would get: 0 + 5 = 5 TL.

– The likelihoods of the good and bad outcome for the risky option – Whether that round is an individual decision round or a group decision round. – (In case it is a group round) Whether the decision-maker will be randomly selected by the computer or will be based on the group members' willingness to be decision-makers – (In case it is a group round) Whether the decision-maker can see information about what other group members would have done, had they been the decision-maker. As mentioned before, there will be 15 rounds in the study. Your earnings will be determined based on a single round, which will be selected randomly by the computer. All rounds have the same likelihood of being selected. Therefore, it makes sense to pay attention to every round of decision-making. After the decision-making part of the study ends, you will be asked to fill out a questionnaire on your screen. After you are done, your earnings will be displayed on your screen. You will add the participation fee to this amount and write the total on your earnings sheet, and then you will be paid.

Individual and Group Decisions In some rounds, you will make the allocation decision only for yourself, and your decision will determine only your earnings. In some rounds, on the other hand, you will be assigned to a group of 3, with two other participants. In such rounds, one group member will make the decision on behalf of the group, and this decision will be binding for all group members. That is, the earnings of all three participants in the same group will be based on the decision of a single group member. In every group round, the groups will be randomly formed by the computer, and will not stay the same. That is, in each group round, a new set of groups will be formed. You will never know the identities of the participants that you are in a group with.

References Arbak, E., Villeval, M.C., 2013. Voluntary leadership: motivation and influence. Soc. Choice Welf. 40 (3), 635–662. Banerjee, D., Ibañez, M., Riener, G., Wollni, M., 2015. Volunteering to take on power: experimental evidence from matrilineal and patriarchal societies in India. DICE Discussion Paper, No. 204 Retrieved March 11, 2019, from. http://www.dice. hhu.de. Barry, B., Friedman, R.A., 1998. Bargainer characteristics in distributive and integrative negotiation. J. Pers. Soc. Psychol. 74 (2), 345. Cettolin, E., Tausch, F., 2015. Risk taking and risk sharing: does responsibility matter? J. Risk Uncert. 50 (3), 229–248. Charness, G., Karni, E., Levin, D., 2013. Ambiguity attitude and social interactions: an experimental investigation. J. Risk Uncert. 46 (1), 1–25. Charness, G., Sutter, M., 2012. Groups make better self-interested decisions. J. Econ. Perspect. 26 (3), 157–176. Charness, G., Rigotti, L., Rustichini, A., 2007. Individual behavior and group membership. Am. Econ. Rev. 97 (4), 1340–1352. Cooper, D.J., Kagel, J.H., 2005. Are two heads better than one? Team versus individual play in signaling games. Am. Econ. Rev. 95 (3), 477–509. Çelen, B., Kariv, S., Schotter, A., 2010. An experimental test of advice and social learning. Manag. Sci. 56 (10), 1687–1701. Dohmen, T., Falk, A., Huffman, D., Sunde, U., 2008. Representative trust and reciprocity: prevalence and determinants. Econ. Inq. 46 (1), 84–90. Drazen, A., Ozbay, E., 2016. Does ‘Being Chosen’ to Lead Induce Non-Selfish Behavior? Experimental Evidence on Reciprocity. CEPR Discussion Paper No. DP11338 Available at SSRN: https://ssrn.com/abstract=2798127. Eagly, A.H., Johnson, B.T., 1990. “Gender and leadership style: a meta-analysis. Psychol. Bull. 108 (2), 233–256. Eslgie, D.M., Hollander, E.P., Rice, R.W., 1988. Appointed and elected leader responses to favorableness of feedback and level of task activity from followers. J. Appl. Soc. Psychol. 18 (16), 1361–1370. Ertac, S., Gurdal, M.Y., 2012a. Deciding to decide: gender, leadership and risk-taking in groups. J. Econ. Behav. Organ. 83 (1), 24–30. Ertac, S., Gurdal, M.Y., 2012b. “Personality, Group Decision-Making and Leadership”. Koç University-TUSIAD Economic Research Forum Working Papers 1227. Ertac, S., Gumren, M., Gurdal, M.Y., 2018. “The Demand for Autonomy and Willingness to Take Responsibility in Risky Decisions: Experimental Evidence”. Unpublished Manuscript Available at. https://www.econ.boun.edu.tr/gurdal. Evans, A.M., Revelle, W., 2008. Survey and behavioral measurements of interpersonal trust. J. Res. Personal. 42 (6), 1585–1593. Fischbacher, U., 2007. z-Tree: zurich toolbox for ready-made economic experiments.

Selection of the Participant who makes the group decision In some of the group rounds, the participant who makes the decision on behalf of the group will be randomly selected by the computer. In this case, one participant in each group will see a statement on his/her screen that reads: "in this round, you will be the one making the decision on behalf of your group". In some other group rounds, you will be asked whether you would like to be the group member making the decision on behalf of the group. In such rounds, if only one group member wanted to be the decision-maker, that participant will make the group decision. If more than one group member wanted to be the decision-maker, one of the willing members will be randomly selected to be the decision-maker. If none of the group members wanted to be the decision-maker, one group member will be randomly selected to be the decision-maker. At the end of any group round, the decision that was made by the group decision-maker will be shown to the two other group members. These group members will be asked to state whether they like the decision made by the decision-maker or not, through clicking a like/dislike button. Information about how many of the group members liked or disliked the decision will be shown to the group decision-maker. However, these statements will not have an impact on the implementation of the decision made by the group decision-maker, nor will they have any monetary consequence for any group member. 139

Journal of Behavioral and Experimental Economics 80 (2019) 130–140

S. Ertac and M.Y. Gurdal Exper. Econ. 10 (2), 171–178. Gneezy, U., Potters, J., 1997. An experiment on risk taking and evaluation periods. Q. J. Econ. 112 (2), 631–645. Hollander, E.P., Julian, J.W., 1969. Contemporary trends in the analysis of leadership processes. Psychol. Bull. 71 (5), 387. Hollander, E.P., Fallon, B.J., Edwards, M.T., 1977. Some aspects of influence and acceptability for appointed and elected group leaders. J. Psychol. 95 (2), 289–296. Hollander, E.P., 1992. Leadership, followership, self, and others. Lead. Q. 3 (1), 43–54. Kocher, M., Pogrebna, G., Sutter, M., 2012. “Other-regarding preferences and leadership styles. J. Econ. Behav. Organ. 88, 109–132. Lewin, K., Lippitt, R., White, R.K., 1939. Patterns of aggressive behavior in experimentally created social climates. J. Soc. Psychol. 10 (2), 269–299. Linde, J., Sonnemans, J., 2012. Social comparison and risky choices. J. Risk Uncert. 44 (1), 45–72. Masclet, D., Colombier, N., Denant-Boemont, L., Lohéac, Y., 2009. Group and individual risk preferences: a lottery-choice experiment with self-employed and salaried workers. J. Econ. Behav. Organ. 70 (3), 470–484. Mengarelli, F., Moretti, L., Faralla, V., Vindras, P., Sirigu, A., 2014. Economic decisions

for others: an exception to loss aversion law. PLoS One 9 (1), e85042. Read, P.B., 1974. Source of authority and legitimation of leadership in small groups. Sociometry 37 (2), 189–204. Rivas, M.F., Sutter, M., 2011. The benefits of voluntary leadership in experimental public goods games. Econ. Lett. 112 (2), 176–178. Rohde, I.M., Rohde, K.I., 2015. Managing social risks–tradeoffs between risks and inequalities. J. Risk Uncer. 51 (2), 103–124. Schotter, A., Sopher, B., 2003. Social learning and coordination conventions in intergenerational games: an experimental study. J. Polit. Econ. 111 (3), 498–529. Schotter, A., Sopher, B., 2007. Advice and behavior in intergenerational ultimatum games: an experimental approach. Games Econ. Behav. 58 (2), 365–393. Schotter, A., Sopher, B., 2006. Trust and trustworthiness in games: an experimental study of intergenerational advice. Exper. Econ. 9 (2), 123–145. Shupp, R.S., Williams, A.W., 2008. Risk preference differentials of small groups and individuals. Econ. J. 118 (525), 258–283. Sutter, M., 2009. Individual behavior and group membership: comment. Am. Econ. Rev. 99 (5), 2247–2257.

140