1 Preference Communication and Leadership in Group Decision-Making Seda Ertac Koc University Mehmet Y. Gurdal 1 Bogazici University August 2016 Abstract 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 individuals frequently deviate from their own preferences when they suggest to the leader what to do. We also 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 whether leaders are likely to keep 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. Keywords: group decision-making, risk, leadership, delegation, advice, communication, experiments. JEL Codes: C91, C92, D81, J16. 1 Corresponding author. Department of Economics, Bogazici University, Bebek, Istanbul 34342 Turkey (email: [email protected], phone: +90 (212) 3597554, fax: +90 (212) 287 24 53). Ertac: Department of Economics, Koc University, Rumeli Feneri Yolu Sariyer, Istanbul 34450 Turkey (email: [email protected]). 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 239529) and the Turkish Academy of the Sciences (TUBA-GEBIP program), and Mehmet Y. Gurdal thanks Bogazici University (BAP 7921) for generous financial support.
Transcript
1
Preference Communication and Leadership in Group Decision-Making
Seda Ertac
Koc University
Mehmet Y. Gurdal1
Bogazici University
August 2016
Abstract
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 individuals
frequently deviate from their own preferences when they suggest to the leader what to do. We
also 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 whether leaders are likely to keep 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.
Keywords: group decision-making, risk, leadership, delegation, advice, communication,
experiments.
JEL Codes: C91, C92, D81, J16.
1Corresponding author. Department of Economics, Bogazici University, Bebek, Istanbul 34342 Turkey (email:
[email protected], phone: +90 (212) 3597554, fax: +90 (212) 287 24 53). Ertac: Department of Economics,
Koc University, Rumeli Feneri Yolu Sariyer, Istanbul 34450 Turkey (email: [email protected]). 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 239529) and the Turkish Academy of
the Sciences (TUBA-GEBIP program), and Mehmet Y. Gurdal thanks Bogazici University (BAP 7921) for generous
financial support.
2
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.2 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
ex-ante 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 organizational culture may be such that the dean
2 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.
3
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.3 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
3 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).
4
choose between a risky new venue and a tried-and-true 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.4
4 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.
5
We analyze the effects of preference communication and leadership mechanisms using two
different experiments. In a three-person, within-subject design, the first experiment provides
suggestive evidence that appointed leaders are more likely to take others’ preferences into
account when they make group decisions. We also find that when telling the leader what to do,
people deviate from their own preferences. In a cleaner, two-person design that better controls
for individual preference, Experiment 2 focuses on communication and the response to
communication. The results confirm that individuals do not exactly express their own
preferences when communicating them. 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 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. 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
decision-making in groups has long been of interest, the issues of leadership willingness, leader
6
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, Rigotti and Rustichini
(2007), Sutter (2009) and Ertac and Gurdal (2012), Ertac, Gumren and Gurdal (2016). Sutter
(2009) is of particular interest, 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 decision-maker. 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
7
lottery. They find significant differences between the endogenous and exogeneous treatments in
terms of risk-sharing among high risk-takers and low risk-takers. 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
(2012), 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.
8
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.5 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. 6 Each
subject, then, goes through 15 periods of decision-making, 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 Payoff
Commonality
Leader
Selection
Preference
Communication
Number of
Periods
1: Individual - - - 3
2: Group T1 Yes Appointed No 3
3: Group T2 Yes Self-selected No 3
4: Group T3 Yes Appointed Yes 3
5: Group T4 Yes Self-selected Yes 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
5 At the time of the experiments, 1 TL corresponded to $0.62. 6 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.
9
leader will be self-selected 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 non-leader group members are individually asked how much they would
allocate to the risky option, if they were the group decision-maker.7 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.8 Finally, at the end of any group 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 non-monetary 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.
7 The non-leader group members know that the leader can see this information, therefore strategic advice is possible. 8 The goal was to prevent spillovers on risk-taking behavior from positive and negative outcomes of previous risky
decisions of oneself and others.
10
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.
Based on this design, our hypotheses are as follows:
H1: Without preference communication, assigned and self-selected leaders make similar choices.
H2: Without preference communication, assigned leaders change their decisions more than self-
selected leaders in the group context with respect to the individual context.
H3: In settings where communication is available, individuals communicate their own preference
to the leader exactly, when not in the decision-making role.
H4: Communicated preferences affect leaders’ decisions: the riskier the suggestion is, the riskier
the actual group decision, controlling for the leader’s own preference.
H5: Assigned leaders’ decisions are more affected by the communicated preference, when they
see the preferences communicated by other group members.
In addition, we set to explore the role of individual characteristics and contextual parameters
such as gender and the probability of winning in communicating preferences, group decisions,
and willingness to lead.
3.2. Results of Experiment 1
We present our results in two main subsections, 3.2.1 and 3.2.2, where we focus on group risk
taking decisions with and without preference communication, respectively. Before we set out to
test the hypotheses regarding communication and leadership mechanisms, we start by examining
11
leadership willingness in the voluntary rounds. In the full data, we observe that individual state a
willingness to make the group decision 85% of the time (86% under no communication, and 84%
under communication). Motivated by the prior literature (Ertac and Gurdal (2012), Arbak and
Villeval (2013)), we also examine whether there are gender differences in leadership willingness.
Our data confirm Ertac and Gurdal (2012): 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).9
3.2.1 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 = 0.68), corroborating Hypothesis 1. As expected, the probability of winning uniformly
increases the risk taken on behalf of a group, as well as the risk taken individually.
Table 1: Group Risk without Preference Communication
p = 0.3 p = 0.5 p = 0.7
Self-selected 2.692 4.942 7.577
(2.356) (1.984) (2.562)
Appointed 2.538 5.423 7.192
(2.173) (2.585) (2.686)
All 2.615 5.183 7.385
(2.257) (2.306) (2.619)
N 104 104 104
Notes: Amount invested (out of 10 TL) as main number, standard errors in parentheses
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
9 The availability of preference communication mechanisms does not seem to affect the leadership propensity for
females.
12
signed-rank test, p-value = 0.62). We find that 41.7% make exactly the same decision they make
individually, whereas 31.1 % and 27.2% engage in cautious- and risky-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. However, the
leadership mechanism has a crucial effect on the likelihood of leaders to "shift" at all.
The regressions reported in Table 2 show that appointed leaders are 10 percentage points more
likely to make a different decision in a group context (Column 1, p-value = 0.064), and the size
of the absolute difference from individual decisions is also significantly larger for appointed
leaders (Column 2, p-value = 0.009). 10 Note that, these regressions, and the other ones
conducted for Experiment 1, control whether a corresponding individual period with the same
probability of winning was played or not (the variable Earlier Alone Period).
Table 2: Choice-shifts by Leaders in Periods without Preference Communication
Shift
Size of the Shift
Appointed 0.0983*
0.505***
(0.0522)
(0.192)
Male -0.0367
0.0926
(0.0648)
(0.25)
p = 0.5 0.105
0.406
(0.0685)
(0.275)
p = 0.7 -0.0538
-0.0589
(0.0666)
(0.246)
Period -0.00657
-0.0454
(0.00899)
(0.0331)
Earlier Alone Period -0.00273
0.308
(0.0623)
(0.264)
Constant
1.211***
-0.343
N 312 312 Notes: Column 1 reports marginal effects from a logistic regression of shift (1=shift, 0=no-shift). Column 2
reports results from an OLS regression of the size of the shift. Robust standard errors in parentheses. * p < 0.1,
** p < 0.05, *** p < 0.01
10 Throughout the paper, in all regressions with multiple observations from the same subject, standard errors are
clustered at the subject level in order to correct for dependence. The reported results are robust to using a random
effects specification and clustering at the session level.
13
3.2.2. Treatments with Preference Communication:
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 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). 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 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 is
significantly different from zero in a t-test (p<0.01). 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. These results
provide evidence against the hypothesis (H3) that individuals simply communicate their own
preferences to the leader.
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 first 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,
14
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 3: Preference Communication to the Leader in Experiment 1
Same Advice Higher Advice Lower Advice
Appointed 0.00415 -0.003 -0.00279
-0.0347 -0.0314 -0.0282
Male 0.0575 0.0353 -0.123***
-0.0476 -0.0434 -0.0382
Own risk -0.0059 -0.0667*** 0.0828***
-0.00867 -0.00766 -0.00677
P = 0.5 0.0686 0.128*** -0.238***
-0.0557 -0.0481 -0.0453
P = 0.7 0.129* 0.198*** -0.391***
-0.0666 -0.0637 -0.0562
Period 0.00998* -0.00803* -0.00228
-0.00551 -0.00485 -0.00431
Earlier alone period 0.00251 0.0053 -0.0262
-0.0508 -0.0433 -0.0415
N 624 624 624 Notes: Marginal effects from probit regressions reported. Dependent variable: 1 if advice=own risk, 0
otherwise in column 1, 1 if advice>own risk, 0 otherwise in column 2, 1 if advice<own risk, 0
otherwise in column 3. Robust standard errors in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01
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. 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
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.
15
Table 4: Group Risk with Preference Communication
p = 0.3 p = 0.5 p = 0.7
Self-selected 2.731 5.442 8.077
(2.466) (2.296) (1.867)
Appointed 3.019 4.846 7.577
(2.429) (2.296) (2.208)
All 2.875 5.144 7.827
(2.44) (2.304) (2.05)
N 104 104 104
Notes: Amount invested (out of 10 TL) as main number, standard errors in parentheses.
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, 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 (see Table 5, Column 1).
Further results emerge when we consider the interaction of preference communication with the
leadership selection mechanism. Using separate regressions for the impact of preference
information on the decisions of appointed and selected leaders, we find that the average risk
recommendation from group members has a significant effect on decisions only when the leader
is appointed--that is, self-selected leaders' decisions do not seem to be affected by the
communicated preferences of the group members (Table 5, Columns 2 & 3). These results are
robust to using the maximum and/or minimum of the recommendations, instead of the average.13
These findings suggest that communication mechanisms will have a significant impact on
implemented decisions in groups only if leadership responsibility is exogenously assigned.
13 When we include both the maximum and the minimum of the preferences together, both are significant for
appointed leaders, and neither is significant for voluntary leaders.
16
Table 5: Leader Risk-taking with Preference Communication
All Appointed Self-selected
Appointed -0.15
(0.195)
Average Preference 0.258*** 0.370*** 0.0784
(0.0759) (0.106) (0.0984)
Own risk 0.350*** 0.407*** 0.253***
(0.0697) (0.0958) (0.0954)
Male 0.262 0.202 0.524
(0.235) (0.312) (0.342)
p = 0.5 0.498 -0.296 1.609**
(0.388) (0.454) (0.637)
p = 0.7 2.012*** 0.812 3.751***
(0.531) (0.616) (0.801)
Period -0.0311 -0.0721 0.0274
(0.0317) (0.0471) (0.0461)
Earlier alone period -0.294 -0.0667 -0.778**
(0.246) (0.328) (0.348)
Constant 1.525*** 1.358* 1.606***
(0.456) (0.69) (0.523)
N 265 129 136
Notes: OLS regression. Robust standard errors in parentheses, * p < 0.1, ** p < 0.05, *** p < 0.01
4. Design and Results of Experiment 2
The results of the first study provide suggestive evidence that the selection mechanism used in
determining leaders can have an effect on the leader’s responsiveness to the information about
group members’ preferences. 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
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.
17
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 order
to more cleanly study how the leader incorporates the other group member’s recommendation
along with his/her own preference in making a decision, we use a 2-person context rather than 3-
person groups.
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:
- 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
18
“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 non-binding
recommendation to the group leader.
- The leader saw this recommendation and made an investment decision for the group.
- 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.
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 minutes. 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 test the following hypotheses:
H1: Assigned and self-selected leaders make similar decisions on behalf of the group.
H2: Individuals communicate their own preference to the leader exactly, when not in the
decision-making role.
H3: Communicated preferences affect leaders’ decisions: the riskier the suggestion is, the riskier
the actual group decision, controlling for the leader’s own preference.
14 We hypothesized that trust would be an important individual characteristic in leadership and the response to
others’ recommendations. 15 At the time of the experiments, 1TL corresponded to $0.33.
19
H4: Assigned leaders’ decisions are more affected by the communicated preference than self-
selected leaders. In particular, assigned leaders are more likely to compromise (move towards the
suggested choice and away from their own), controlling for the preference difference.
H5: Controlling for the difference between the communicated preference and the leader’s own,
riskier advice has the same likelihood of being taken into account as more cautious advice.
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.
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 (2012), indicating a robust gender difference in leadership willingness. In
addition, the data reveal that individuals who are less trusting of others are more likely to wish to
be the decision-maker for the group.
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=0.6, Mann-Whitney test), in line with H1. 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
20
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.
Table 6: Group Risk without Preference Communication
p = 0.3 p = 0.5 p = 0.7
Self-selected 3.167 7.47 11.49
(3.013) (2.98) (2.362)
Appointed 2.713 7.758 11.77
(2.854) (3.735) (2.553)
All 2.886 7.632 11.65
(2.918) (3.42) (2.472)
N 220 228 238
Notes: Amount invested (out of 15 TL) as main number, standard errors in parentheses
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 still deviates from the
individual preferences in a substantial number of instances. The average absolute difference
between advice and individual preference is around 2.82, which is significantly higher than zero
(p<0.01, t-test), providing evidence against H2.
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 series
of probit regressions, 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
21
that is equal to the individual preference (column 1). 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 2). 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 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.
Table 7: Preference Communication to the Leader in Experiment 2
Same Advice Higher Advice Lower Advice
Appointed -0.0298 0.0444* -0.0112
(0.0406) (0.0269) (0.0263)
Male 0.0982** -0.0548** -0.0507**
(0.0393) (0.0274) (0.0252)
Own Risk 0.00364 -0.0207*** 0.0152***
(0.00539) (0.00437) (0.00385)
p = 0.5 0.00476 0.0821** -0.0746**
(0.0463) (0.0332) (0.0369)
p = 0.7 -0.0807 0.189*** -0.0907**
(0.0654) (0.0475) (0.0398)
Period 0.00971** 0.000121 -0.0102***
(0.00474) (0.00371) (0.00341)
Trust -0.0957* 0.0307 0.0656**
(0.0506) (0.0329) (0.0288)
N 686 686 686 Notes: Marginal effects from probit regressions reported. Dependent variable: 1 if advice=own risk, 0
otherwise in column 1, 1 if advice>own risk, 0 otherwise in column 2, 1 if advice<own risk, 0 otherwise
in column 3. Robust standard errors in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01
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
22
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
similar effects on the decisions of both types of leaders, in contrast to Experiment 1. That is,
results of Experiment 2 do not seem to corroborate H3.
Table 8: Leader Risk-taking with Preference Communication in Experiment 2
All Appointed Self-selected
Appointed 0.117
(0.155)
Advice 0.171*** 0.208*** 0.111***
(0.028) (0.04) (0.0374)
Own risk 0.722*** 0.750*** 0.679***
(0.0475) (0.0613) (0.0755)
Male -0.0813 -0.241 0.149
(0.145) (0.194) (0.223)
p = 0.5 0.912*** 0.858* 1.071**
(0.32) (0.465) (0.448)
p = 0.7 1.315** 0.85 2.030**
(0.54) (0.719) (0.841)
Period -0.0181 -0.00657 -0.0224
(0.0189) (0.0255) (0.0276)
Constant 0.237 0.0735 0.598
(0.26) (0.292) (0.471)
N 686 400 286
Notes: OLS regression. Robust standard errors in parentheses, * p < 0.1, ** p < 0.05, *** p < 0.01
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 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
23
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.
Table 9: Preference Communication and Leader’s Choice Shift
Leader’s Choice
Shift
Treatment
Appointed Self-Selected
Advice > Leader’s
Preference
Risky Shift 89 56
No Shift 102 64
Cautious Shift 8 4
Advice = Leader’s
Preference
Risky Shift 6 4
No Shift 49 27
Cautious Shift 0 1
Advice < Leader’s
Preference
Risky Shift 8 11
No Shift 82 67
Cautious Shift 56 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.
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 23 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 recommendation. Individuals
who have a high level of trust in others are (an additional) 16 percentage points more likely to
24
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.
We find 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.
Table 10: Leader's Compromise
I II III
Appointed -0.0102 -0.0173 -0.0162
(0.0508) (0.0502) (0.0504)
Male -0.206*** -0.204*** -0.202***
(0.0475) (0.0471) (0.0473)
Own Risk 0.00478 0.00925* 0.00384
(0.00471) (0.00506) (0.00919)
Period -0.00655 -0.00565 -0.00549
(0.00582) (0.00569) (0.00589)
Interpersonal Trust 0.163*** 0.167*** 0.164***
(0.0585) (0.0576) (0.0573)
Preference Difference 0.0114 0.0111 0.012
(0.00829) (0.00827) (0.00819)
Risky Advice 0.0954** 0.0742
-0.0437 -0.0516
p = 0.5
0.0256
-0.0677
p = 0.7
0.0675
-0.0975
N 599 599 599 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
25
5. Discussion and Concluding Remarks
While Experiment 1 provides evidence that appointed leaders may be more likely to change their
own decisions without communication and also respond more to communicated preferences by
others, the latter result (that appointed leaders are more responsive to others’ preferences) does
not hold in Experiment 2. There could be two reasons why appointed and self-selected leaders
might behave differently in response to preference communication. First, voluntary leadership
might select certain attitudinal characteristics that induce a person to not put much weight on
others' preferences. 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.16 Alternatively, or in addition, the institutional environment may make
individuals behave in certain ways. Appointed leaders may feel that since leadership is bestowed
upon them in a situation where no group member was ex-ante different in terms of (1) possessing
the right of leadership, or (2) having expressed a preference for/against leadership, they have a
responsibility to take the preferences of others into account. In contrast, since our setup partially
implies delegation of responsibility to the leader by other group members in voluntary rounds (if
somebody is a leader in a voluntary round, it is more likely than assigned rounds that others in
the group did not wish to be leaders), self-selected leaders may feel more entitled to go with their
own preferences when making the group decision. Similar mechanisms might be behind our
finding that when the leader acts without knowing the preferences of others, self-selected leaders
are less likely to change their individual risk-taking decisions, i.e. are less responsive to the
group context.17
The high rate of self-selection in both of our studies (85% in Experiment 1, 86% in Experiment
2) potentially weakens the self-selection channel discussed above, in that the samples that end up
making the decision will be more similar, the higher the rate of volunteering for leadership. On
the flipside, a high rate of self-selection makes it easier to identify the pure effect of the leader
selection mechanism on individuals’ behavior if there is such an effect, i.e., to see whether
16 Arbak and Villeval (2011) find a small effect of agreeableness in the same direction in their study on voluntary
leadership in public good contributions. 17 Our results are partially in line with Kocher et al. (2009) in that autocratic leadership in response to the team’s
opinion is positively correlated with a lower likelihood to change one's individual decision in a team context.
26
someone who would be willing to decide, acts differently depending on whether she was
exogenously appointed or volunteered first. The reason why we do not find any effect of the
leadership selection mechanism in Experiment 2 but in Experiment 1 may be because it is less
likely for someone who said yes to the leadership decision to actually become the leader in the
three-person than the two-person context. This may amplify the feeling of having been delegated
the decision and feeling entitled to go with one’s own preferences in the voluntary treatment in
three-person groups 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. Individuals who trust others are also less willing to take on the leadership
role. As we will discuss below, these characteristics also seem to matter for how individuals
behave once in the leadership role.
In both experiments, we find that recommendations to the leader do not exactly reflect own
preferences. Instead, as own risk tolerance increases, the recommendation becomes more
cautious. This suggests that some individuals may not want to impose their own preferences on
others even when making recommendations. Results from Experiment 2 also provide 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.18 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
18 This result holds both in the assigned and in the selected treatments.
27
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.
Self-selected 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.
References
Arbak, E., & Villeval, M. C. (2013). “Voluntary leadership: motivation and influence”, Social
Choice and Welfare, 40(3), 635-662.
Cettolin, E., & Tausch, F. (2015). Risk taking and risk sharing: Does responsibility
matter?. Journal of Risk and Uncertainty, 50(3), 229-248.
Charness, G., Karni, E., & Levin, D. (2013). “Ambiguity Attitude and social interactions: An
Experimental Investigation”, Journal of Risk and Uncertainty, 46(1), 1–25.
Charness, G., Sutter, M. (2012) "Groups Make Better Self-Interested Decisions", Journal of
Economic Perspectives, 26(3), 157-176.
Charness, G. , Rigotti, L., Rustichini, A. (2007). "Individual Behavior and Group Membership."
American Economic Review, 97(4): 1340-52.
Cooper, D. J., & Kagel, J. H. (2005). “Are two heads better than one? Team versus individual
play in signaling games”, American Economic Review, 95(3), 477-509.
28
Çelen, B., Kariv, S., & Schotter, A. (2010). “An experimental test of advice and social
learning”, Management Science, 56(10), 1687-1701.
Drazen, A., & Ozbay, E. Y. “How Do Elected and Appointed Policymakers Act When in Office?
Experimental Evidence on Citizens, Candidates, and Leaders”, Mimeo.
Eslgie, D.M., Hollander, E.P. and Rice, R.W. (1988) "Appointed and elected leader responses to
favorableness of feedback and level of task activity from followers", Journal of Applied Social
Psychology, 18(16), 1361-1370.
Ertac, S., Gurdal, M.Y. (2012a) "Deciding to Decide: Gender, Leadership and Risk-Taking in
Groups", Journal of Economic Behavior and Organization, 83(1), 24-30.
Ertac, S., Gurdal, M.Y. (2012b) "Personality, Group Decision-Making and Leadership", ERF
Working Paper, #1227.
Ertac, S., Gumren, M., Gurdal, M.Y. (2016) “The Demand for Autonomy and Willingness to
Take Responsibility in Risky Decisions: Experimental Evidence”, working paper.
Fischbacher, U. (2007). “z-Tree: Zurich toolbox for ready-made economic
