No.

61 november 2010

Lisa Bruttel
Urs Fischbacher
Learning and Peer Effects
Taking the initiative. What characterizes leaders?

Research Paper Series

Thurgau Institute of Economics and Department of Economics
at the University of Konstanz
 Taking the initiative. What characterizes leaders?∗

Lisa Bruttel† Urs Fischbacher‡

This version: June 2013 | First Version: December 2010

Abstract

Taking the initiative is a crucial element of leadership and an important asset for many
jobs. We assess this element of leadership in a game in which it emerges spontaneously
since people have a non-obvious possibility to take the initiative. We can show that lead-
ership in this game correlates with real life activities associated with taking the initiative.
Combining this game with other experimental games and with questionnaires, we inves-
tigate the personality characteristics that entail leadership. We find efficiency concerns
and generosity to be important determinants of leadership. Leaders have an internal lo-
cus of control and are more patient than non-leaders, but they are not different from the
non-leaders with respect to risk attitude. Response time patterns and the results from
the cognitive reflection test show that cognitive resources are relevant in the decision to
lead.

Keywords: leading-by-example, social preferences, experiment

JEL-Classification: A13, C92, D03, D83

∗
We thank Katharine Bendrick, Paul Heidhues, Dorothea Kübler, Lydia Mechtenberg, Daniele Nosenzo and
participants of the ESA world meeting 2010 in Copenhagen, the 2011 Florence Workshop on Behavioural and
Experimental Economics, the GfeW meeting 2011 in Nuremberg, and of research seminars in Berlin, Konstanz
and Kreuzlingen for very helpful suggestions. Nathalie Hock and Natalie Zimmer provided valuable research
assistance. For providing laboratory resources we kindly thank LERN of the University of Erlangen-Nuremberg.
†
University of Konstanz, Department of Economics, Box 131, 78457 Konstanz, Germany. E-mail:
[email protected]. Phone: 0049-7531-88-3214.
‡
University of Konstanz, Department of Economics, Box 131, 78457 Konstanz, Germany. E-mail:
[email protected]. Phone: 0049-7531-88-2652. Thurgau Institute of Economics, Hauptstrasse
90, 8280 Kreuzlingen, Switzerland. Phone: 0041-71-677 05 12

1
1 Introduction

Imagine the following situation that recently occured in the department of the authors of this
paper. A group of researchers is sitting in a lecture room, the presentation of a guest of the
department is about to start. Trying to connect the speaker’s laptop with the projector, people
notice that the latter is defect. Some people hectically start pressing buttons and switches,
but the projector continues not to be working. Everybody knows that a fast solution is needed
so that the presentation can start in time, but only one postdoc actually solves the problem.
He inspects the neighboring rooms, finds one in which the lecturer does not use the projector,
enters that room and kindly asks the people there to switch rooms. Two minutes later the
presentation can start.

The behavior of the postdoc is named taking-the-initiative and it has two distinct features,
which generalize to many other situations in our everyday life. First, the situation concerns
several people who are generally all capable of solving the problem. However, only one or a few
of them actually do take the initiative. Second, in many of these situations, the best solution
initially requires some creativity or innovation, even though with hindsight it seems quite
obvious that this action was needed. Further examples are citizens’ initiatives or neighbourhood
centers creating cultural or sportive offers for disadvantaged young people to keep them busy
and out of trouble, or enterprises arranging flexible childcare solutions for their employees when
sufficient public childcare is not available. Often this initiative also has the third feature that it
encourages others to follow the good example. The open source community is an economically
relevant example. Without people like Linus Torwalds who initiate a project and contribute a
significant code base, open source projects will never start (Lerner and Tirole 2002). Taking the
initiative is a channel through which cooperation can arise in any bad state of social interaction,
be it mutual freeriding in team work or a negotiation with parties irreconcilably opposed. In
such situations, it is highly desirable that someone breaks the vicious circle by giving a good
example, but only some people actually do so. We are interested in the determinants of this
behavior. What characterizes people who take the initiative?

We build on a game introduced by Dufwenberg and Gneezy (2000) to study the Bertrand
paradox and interpreted by Bruttel (2009a) in terms of leading-by-example. In this game, pairs
of two people choose a number between 2 and 100. The person setting the lower number gets

2
the number she chose as a payoff; the other gets nothing. In the case of a tie both receive
half the price. This game is repeatedly played with changing pair composition within groups
of 12 subjects in total, with all 12 subjects in a group being informed about the decisions of all
players in their group after the end of each round.

In this game, typically a cyclical movement of average numbers is observed. In the beginning,
the 12 numbers within a group are more or less arbitrarily selected between 2 and 100. During
the first few rounds, average numbers decrease, because all participants try to choose a number
slightly lower than most of the others. After some rounds, the group reaches a relatively low
level of numbers. Eventually though, one player raises the number to a very high level. Many
of the other players follow, so a temporary increase of the numbers occurs. This behavior is in
no way induced by the experimental instructions, but appears to be very robust. It crucially
depends on the feedback condition, because such signaling to the group is only possible because
of the group feedback. If subjects only get feedback from their own pair, the chosen numbers
continuously decline. We use this game with 8 participants per group, and we consider the
initiative to coordinate at a higher level to be endogenous leading-by-example. The person who
takes the initiative, we call a “leader”. We connect this classification to the decisions in other
games and questionnaires eliciting other-regarding preferences, beliefs, risk attitude, patience,
cognitive abilities, and other personality characteristics.

According to our results, leaders are characterized by above-average cognitive skills and
are predominantly male. They have strong preferences for efficiency, generosity, and against
advantageous inequality, and do not primarily seek to maximize their personal monetary benefit
or to obtain a positive public image. They have accurate beliefs about the extent to which others
will follow their example. Leaders in the experiment have an internal locus of control and are
more patient than non-leaders, but we do not find an impact of the big five personality traits
or risk attitude on leading-by-example. Finally, we can show that leaders as identified in the
experiment are also more likely than non-leaders to engage in activities associated with taking
the initiative in their real lives.

The leaders we are interested in have the ability and willingness to improve the outcome of
the behavior of a group of people who are “stuck” in a bad situation, but they do not primarily
give the good example because they are or want to be a formal leading authority. This makes

3
our definition different from most concepts of leadership discussed in the literature (see Yukl
2009 or Kouzes and Posner 2007), where the focus is on advising designated leaders. The notion
most closely related to our research topic is charismatic or transformational leadership (Bass
and Avolio 1994). However, unlike in situations in organizations, where charismatic leaders
inspire followers by their impressive personality (see, e.g., Conger and Kanungo 1987), leaders
in our experiment have relatively weak measures to communicate their “vision” of a cooperative
solution to others as face-to-face interaction is ruled out in the computerized experiment. It is
particularly the motivation of people taking the initiative which we borrow from this part of
the leadership literature.

Hermalin (1998) defines leadership in a very similar way as we do, but the theoretical model
of leadership he develops is different from our approach. In the model he discusses, a leader
is the only member of the team to have information about their common effort return. By
choosing effort before the other group members, the leader can signal this information to the
followers, inducing them to provide high effort as well. The experiment of Potters et al. (2005)
illustrates that such an informational setup in fact yields coordination on sequential moves
to the benefit of efficiency. In our experiment, the action of the leader serves rather as a
coordination device but not as a measure to resolve information asymmetries between leaders
and followers as all players in our framework receive exactly the same information.

Because of the incentive structure of leading-by-example, this kind of leadership is often

studied by introducing a sequential move structure in public good experiments (Gächter and
Renner 2006; Güth et al. 2007; Gächter et al. 2012; Drouvelis and Nosenzo 2012; Moxnes and
van der Heijden 2007; Pogrebna et al. 2011; Potters et al. 2007; Levati et al. 2007). These
studies focus on the mechanism of leadership and typically show that groups with leaders on
average contribute more than groups without, but only due to the higher contributions of the
leaders.

Public good games capture nicely the incentive structure of leading-by-example. However,
in these games it is obvious to all players what constitutes the good example and, therefore, they
do not cover the innovative facet of the act of taking the initiative. Furthermore, it is always
clear to the subjects that the experiment they are participating in is about leading and following.
This may induce experimenter demand effects (Zizzo 2010) possibly manipulating leadership

4
in either direction. It may reduce leadership because involuntary leaders perform worse; or it
may enhance leadership, because even natural non-leaders infer from the experimental design
that leadership is socially desirable. In our design, there is no predefined leader. Different from
other experiments about leadership, the leading behavior to initiate a number increase in this
experiment is neither explicitly nor implicitly induced by the experimenter. There is no explicit
assignment of the leading role to a certain subject. Decision making occurs simultaneously, so
no player has a distinct role. So we do not address the question of whether a person accepts to
be a leader when she is assigned the role but whether a person decides to take the initiative.
Thus, our setup avoids the experimenter demand effect problem and includes the innovative
element of leadership.

Our research question is related to a recent study by Arbak and Villeval (2013). They
investigate the motivations of leaders by combining different variants of a two-stage public good
game with personality tests. Similar to our setup, leadership is voluntary in their experiment,
as subjects decide themselves whether they would like to make their contribution to the public
good in the first or in the second stage of a round. However, the basic experimenter demand
effect argument with respect to the two-stage structure still holds. In line with our results, they
find that social concerns (measured as donations to a charity) are a driving force for at least
some of their leaders and that men are more likely to lead than women. Furthermore, Arbak
and Villeval (2013) argue that a positive social status1 of leading drives the decision to give a
good example, a finding which is rather not supported by our data.

The strategic interaction of our experiment is different to the one in a public good game and
more related to price setting games. However, while cooperation of firms setting prices is usually
inefficient for society because it implies a welfare loss at a cost for consumers, cooperation in our
setup is socially desirable, because there is no consumer side which may suffer from collusion.
Nevertheless, the behavioral patterns in price setting games often resemble similar cyclical
movements as the numbers in our experiment, in particular if price setting occurs sequentially.
For example, Leufkens and Peters (2011) and Bruttel (2009b) report cyclical price patterns for
sequential price setting duopolies, which are initiated by a drastic price increase of one of the two

1
While in Arbak and Villeval’s (2013) study status arises endogenously from a high contribution to the public
good, Kumru and Vesterlund (2010) show that also an ex ante induced high status makes leaders contribute
more, because they anticipate that low-status followers will closely imitate the behavior of a high-status leader.

5
firms. Similarly, Selten and Apesteguia (2005) report “cooperative attempts”, price increases
with the intention to make others follow this step, for a simultaneous-move game with price
competition on a circle. Comparing sequential with simultaneous price setting, Datta Mago
and Dechenaux (2009) report higher average prices for the sequential game when capacities are
asymmetric, while Kübler and Müller (2002) find average prices to be lower in the sequential
case with differentiated products. In our experiment, the main feature triggering coordination
is the public feedback within matching groups.

In section 2 of this paper, we present the design we use in the experiment. In section 3 we
provide the behavioral predictions. Section 4 presents our findings and section 5 concludes.

2 Design and Procedures

2.1 Design

In order to determine the characteristics of leaders, we combine a game in which spontaneous

leadership occurs regularly with a series of experiments and questionnaires that allow the mea-
surement of beliefs, risk and social preferences as well as other characteristics. We start with
the explanation of the game that we use as our leadership game.

The basic design of the game in our experiment is a variant of the stylized Bertrand pricing
game in Dufwenberg and Gneezy (2000). In this game, two participants simultaneously choose
a number from the interval [2,100]. The participant choosing the lower number wins the game.
The prize is equal to the winning number. In case of a tie, each participant gets half of the
prize. The game is repeated for 30 rounds. Players are divided into groups of eight participants.
In each round, the eight participants in one group are randomly matched in pairs of two. Thus,
four pairs play the game simultaneously in a group of eight participants. After each round the
subjects were informed about their own number and the number of their partner in this period.
Furthermore, all eight numbers were made publicly known in the group, ordered by size of the
number. Thus, subjects received both, individual feedback and group feedback about the other
participants whom they did not meet in the current round but might meet in the next rounds.

6
 Each number choice decision in this main part of the experiment was surrounded by a belief
formation stage and a publicity choice stage. In the belief formation stage before the number
choice, players had to submit beliefs about the minimum, maximum, and average number of
the other seven players in the next round. For each of the three values, they had to submit
a probability distribution over the intervals 2-20, 21-40, 41-60, 61-80, 81-100. To facilitate
submission of their beliefs, they were provided a graphical tool on the computer screen. Figure
9 in the Appendix shows a screen shot. The bars of the single intervals could be moved with
mouse clicks. A click on “update” next to one of the distributions automatically increased or
decreased all five bars proportionally to balance the sum of weights to 100 percentage points. If
participants were done with their belief formation, they had to click “next”. In the beginning
of the next round, their past estimates were shown as default values and could be adapted with
the same procedure. The quality of their prediction for each of the three values was determined
with the quadratic scoring rule (Brier, 1950). They received a payment proportional to this
measure.

By asking subjects before their decision for their belief about the probability distribution of
the maximum number of the other group members in the next period, we learn how likely they
think it is that someone else will lead. In particular, we need the probability weight leaders
assign to the categories equal to or larger than their own leading number. The estimated
average number of the other group members one round after a leading number provides an
approximation of the leader’s belief on the extent to which the others will follow. For our
analysis, we re-calculate the estimated average from the submitted probability distribution. By
comparing the beliefs of leaders to the beliefs of other not leading participants, we learn whether
leaders are different from others with respect to their estimate of the benefits of leadership. The
stronger the increase in the average number after a leading number, the higher are the potential
gains to a participant undercutting opponents by a small amount. If leaders systematically
overestimate others’ average numbers after a leading number, this would indicate that leaders
lead because they overestimate their monetary benefits from leading.

In addition to the maximum and the average number, we asked players to submit their
belief about the minimum number of the other group members which we do not need for the
analysis at all. We elicit beliefs in such a detailed way to receive an accurate belief of leaders on
whether there will be another leader. Asking for the probability of the maximum interval only,

7
however, could introduce the experimenter demand effects again which we were able to avoid
by the design of the main part. In order not to lead subjects into thinking about leadership, we
therefore included the minimum belief and applied the distributional belief elicitation procedure
to all three values, minimum, maximum, and average.

In the publicity stage after the number choice, we allowed players in the given round to
give up anonymity and publish their seat number on the other participants’ computer screens
beside their own chosen number. Publication of the seat number in one round cost 10 points
and could be decided upon by ticking a box on a separate screen after the number choice. Use
of this feature allows us to control for whether appreciation by others motivated extraordinary
number choices. Paying for publishing the seat number might not only be due to the leader’s
desire to become publicly known as a leader but also be used to strengthen the signaling effect
of the leading number. It certainly emphasizes the leading number on the other participants’
computer screens if the additional seat number entry is displayed as well. To test whether
leaders do not want attention for themselves but for their number, we added a highlight option
to the publicity feature in the second half of the sessions we conducted. In addition to the option
to display their seat number at a cost of 10 points, participants could also choose to highlight
their number anonymously on the others’ screens by displaying three exclamation marks aside
their number in that round at a cost of 5 points. To make sure that the highlighting effect of
the publicity option is now not weaker than the one of the highlight option, in these sessions
also the display of the seat number included three exclamation marks.

Treatment Description
1 Distribution games (efficiency, inequality, generosity)
2 Risk elicitation
3 Belief trial phase (only in series 1)
4 Number choice game
5 Feedback about outcomes and payoffs
6 Questionnaires

Table 1: Order of Treatments

Before the main part of the experiment, we conducted some short games to elicit preferences
for efficiency, generosity, inequality aversion and risk attitude. Table 1 includes an overview
of the order of the different games in the experiment. We applied the same order of these

8
experiment to all the subjects. This procedure has the disadvantage that there might be
spillovers between the games for which we do not control. However, we are interested in
the difference between leaders and non-leaders and as long as there is no interaction between
the type and the spillover, we can draw valid conclusions. The first part of the experiment
was a series of six simple two-player distribution games using the strategy method, similar to
Engelmann and Strobel (2004). As Bruttel (2009a) argues, there seem to be spillover wealth
effects from the main experiment to the decisions in such distribution games. For this reason,
we conducted these games before the main part of the experiment and not afterwards. In
each game, participants had to choose between two distributions of money between themselves
and another player. Table 2 shows the payoffs of the options between which player 1 could
choose. The six games were designed in order to create tradeoffs between selfishness, equality
and efficiency.2 In the first column, there is a tradeoff between selfishness and equality on
the one hand and efficiency on the other. The second column contains games with a tradeoff
between selfishness on the one hand and equality and efficiency on the other hand. In the third
column there is one game. In this game there is a tradeoff between equality in the form of
envy and efficiency. The roles of players 1 and 2 were randomly assigned to the players after
they had decided for both roles. Afterwards, one out of the six games was randomly selected
for payment by the computer program. After completion of the six choices, we elicited risk
attitude using the Holt and Laury (2002) procedure. The random draws from these initial parts
of the experiment and the corresponding payoffs were revealed only after the main part of the
experiment.

After the main part of the experiment, players first had to answer a questionnaire about
their decisions in the number choice game. After that, they were asked to fill in several ques-
tionnaires, including the BFI-S big five questionnaire as used in the German Socioeconomic
Panel (see Dehne and Schupp 2007), a locus of control3 questionnaire according to the Rotter
(1966) scale, a shortened version of the scale of patience4 developed by Dudley (2003), the risk

2
Originally, we presented one more choice measuring inequality aversion to the subjects. It was excluded af-
terwards because of a typo on the computer screen in the first series of the experiment leading to an inconsistency
in the presentation of this choice.
3
Broadly speaking, the locus of control measures the extent to which an individual feels to have control
about the things happening in her life.
4
The questions are included in Appendix B.

9
 Selfishness and Equality Selfishness vs. Equality
Envy vs. Generosity
vs. Efficiency and Efficiency

Option A Option B Option A Option B Option A Option B

Player 1 2 1 7 6 1 1
Player 2 2 6 2 6 1 3
Player 1 2 1 6 5
Player 2 2 5 2 5
Player 1 2 1
Player 2 2 4

Table 2: Parameters in the distribution games

questionnaire used in Dohmen et al. (2011), a cognitive reflection test (Frederick 2005), and
a short sociodemographic questionnaire. Among these questions, we asked for the following
seven variables, referring to actual behavior in participants’ real lives, which we interpret as
taking the initiative: whether this person has ever been to a foreign country for an extended
period of time, is working in an voluntary capacity, receives a student scholarship, took part
in a youth research competition, has a function in an unincorporated association, has ever
organized an event, or was class representative at school. Out of these seven variables, taking
either the value zero or one, we create a “real life index” for participants’ inclination to take
the initiative. Furthermore, we used a variant of the personal initiative questionnaire in Meyer
(2006) which builds on the questions of Bledow and Frese (2009). In this questionnaire, par-
ticipants are given five hypothetical situations with four possible ways each to behave in these
situations, where only some of the possible reactions represent personal initiative, but also the
others are presented in a way that they can all be considered socially appropriate. Participants
have to state which of the four reactions would most and least likely describe the way they
would behave in such a situation. The questions and their coding are presented in detail in
Appendix C.

2.2 Procedures

The experiment was computerized using z-Tree (Fischbacher 2007). A total of 224 students,
95 males and 129 females, from various disciplines took part in the experiment, divided into 28

10
groups of 8 participants each. They were recruited via ORSEE (Greiner 2004). The experiment
took place in the Lakelab, the laboratory for experimental economics at the University of
Konstanz between December 2009 and June 2010 (series 1, 13 groups), and in the LERN
at the University of Erlangen-Nuremberg in June 2011 (series 2, 15 groups).5 Sessions in
series 1 lasted between 2 and 2.5 hours, sessions in series 2 about 1.5 hours. The sessions in
the second series were shorter because they did not contain the belief elicitation procedure.
We did not elicit beliefs in the second series to save time for filling out the large number of
questionnaires. We did so because we learned after conducting the experiments in the first series
that many participants were tired by the long duration of the experiment and rushed through
the questionnaires without giving thoughtful answers.6 In the second series we shortened the
experiment by leaving out the belief elicitation so that they started filling out the questionnaire
approximately 45 minutes after the beginning of the experiment. We also announced orally
to the subjects that they may expect filling out the questionnaires to take about 45 minutes
and emphasized that their answers are of great importance for the scientific evaluation of the
experiment.7

The experimental currency was points. In the number choice game and in the belief elicita-
tion procedure 30 points were converted into 1 euro after the experiment. In the distribution
games and in the risk elicitation procedure there was a one-to-one exchange rate. On aver-
age, participants earned about 29 euros in the experiment with belief elicitation and 23 euros
without. The questionnaires were not incentivized. The protocol during the experiment was
as follows: After welcoming participants and explaining the main rules for participation in the
experiment, they were randomly assigned seats in the laboratory. At their place, they read

5
We had to move to another place for the second series, because while running some sessions (which are not
included in this paper) in the Lakelab in April and May 2011 information about this experiment had become
known in the subject pool in Konstanz.
6
We used two measures to identify careless answers. The first one was an item “you can trust my answers” in
the big five questionnaire, according to which 18 percent of the participants cannot be trusted in their answering
behavior, i.e. they select an answer of less than the maximum minus 1 unit to this item. The second one was
an analysis of obvious patterns (e.g. choosing the very left answer for all items) in the answers to the locus
of control questionnaire identifying 27 percent as very unlikely to be honest. As the two measures were also
virtually uncorrelated (correlation coefficient 0.04), we decided that we needed a second series of experiments
to obtain usable answers in the questionnaires.
7
The two indices of careless answers in the second series were 4 percent and 8 percent, respectively.

11
short general instructions about the sequence of experiments they would participate in.8 For
the distribution games and the risk elicitation, subjects received instructions on their computer
screen and made decisions immediately after reading the instructions. For the main part of the
experiment, they received written instructions explaining the decisions and their consequences
as well as the belief formation stage including the payment method with the quadratic scoring
rule and the publicity choice stage. Next they were given the possibility to familiarize them-
selves with the computer screen for the belief formation. Then the experiment started. At the
end of the session, the participants were asked to complete several questionnaires.

3 Behavioral predictions

In this section, we focus on our research question - what characterizes leaders. At the beginning
of the next section, we will give the exact description of how we classify leaders. For now, we
just note that if there is common knowledge about rationality and selfishness, subjects should
choose 2 as their number. So, even when subjects try to coordinate on a higher number at the
beginning, directing the behavior towards the best reply of the previous period will cause a
decline in the numbers and, hence, in the payoff (Selten and Stöcker 1986). A subject displays
leadership when she breaks out of this vicious circle and increases her number. In this section,
we discuss the potential motivations for this behavior and the likely characteristics of such
leaders.

First, let us consider the selfish motivation to lead. Some leaders in our experiment might
initiate a number increase not for the purpose of the benefit of the group, but rather because
they intend to undercut others at a higher level in the next round. Such selfishly motivated
leadership crucially hinges on the belief on extent to which the other players will follow. Actu-
ally, Gächter et al. (2012) find that cooperative leaders have over-optimistic beliefs about the
cooperativeness of followers, and that this can (aside from social motivations of leaders) explain
their high contribution as first mover in a sequential public good game. These over-optimistic
beliefs might be a consequence of the false consensus effect (Ross et al. 1977). It seems likely
that such over-optimism is not only present for the randomly assigned leaders in the sequential

8
The Appendix of this paper contains a translation.

12
public good game in Gächter et al. (2012) but also in the context of our number choice game.
Thus, our first prediction is:

Hypothesis 1 Leaders over-estimate the average numbers of their group members after lead-
ing.

Our second hypothesis refers to other-regarding preferences. Successful leadership will pro-
vide a higher payoff for the group, but it is potentially costly for the leader and it could in
particular create inequality that is disadvantageous for the leader. In the framework of a se-
quential public good game Arbak and Villeval (2013) find that voluntary leadership is related
to preferences for efficiency and generosity. In Table 2, option B is always the efficient outcome.
If leaders care more about efficiency than non-leaders, we expect that leaders more frequently
choose option B. In the first column, efficiency does not only go against selfishness, it also
creates disadvantageous inequality, which envious people will dislike. Since leaders risk disad-
vantageous inequality, we expect them to be more tolerant towards disadvantageous inequality
and to choose option B more frequently that non-leaders in particular in the games in the first
column of Table 2.

Hypothesis 2 Leaders have stronger pro-social attitudes than non-leaders. They attach a
higher value to efficiency, they are more generous and they are more willing to accept disad-
vantageous inequality.

Leadership could be a signal of prosociality. As a prosocial attitude is generally seen as a

positive trait it might be that leaders lead because they want to signal their “good character”
to others. In our experiment, we offer participants an opportunity to make their seat number
publicly know. We expect that leaders use this option more frequently than non-leaders.

Hypothesis 3 Leaders are more likely to give up anonymity than non-leaders. Leaders are
more likely to give up anonymity in their leading rounds than in other rounds.

The considerations a leader makes before deciding to lead are relatively complex. In the
beginning of the game most players, including the later leaders, follow a best reply strategy

13
against the distribution of numbers in their group in the previous round. This best reply
dynamic leads to decreasing numbers from round to round, because all players try to choose
a lower number than their representative opponent. The leader first has to understand the
dynamic that all players react in a similar way to the group feedback. Second, the leader must
be innovative in exploiting this behavior. By drastically increasing the own number, the leader
manipulates the distribution of numbers the others are reacting to and reaches a temporary
coordination of the group at a higher level of numbers. This understanding and manipulation of
the dynamic decisions in this game requires a lot of innovation, creativity, and cognitive ability,
as well as the willingness to use them. It also requires the willingness to break out of the simple
responding to the other players’ behavior. The cognitive reflection test described in Frederick
(2005) captures the essence of these abilities. This is summarized in the next hypothesis.

Hypothesis 4 Leaders have a higher score in a cognitive reflection test than non-leaders.

According to Frederick (2005), the score in a cognitive reflection test is on average higher for
males than for females. Furthermore, Arbak and Villeval (2013) hypothesize that particularly
male participants may be concerned with maintaining a positive public image as men in their
sample act more often as voluntary leaders than women. The latter result is also found in
Gächter et al (2012), though it is not significant there. Matched with the information about
participants’ gender, we can also test whether male leaders are more publicity seeking than
female leaders. We formulate this as our next hypothesis.

Hypothesis 5 Men are more likely to lead than women. In particular, men are more likely
than women to give up anonymity as leaders.

Leadership is risky. When deciding to provide a good example, the leader hopes that the
others will follow the example. In this case, future social welfare and potentially the leader’s
individual payoffs will increase. However, the leader cannot be sure that others will follow. A
risk averse player might therefore be reluctant to lead even if this person was willing to set
the example if it were guaranteed that others would follow. Similarly, the leader has to be
patient, because for the leader any potential monetary gains from leading are realized only in

14
the future while initial costs (foregone possible gains) materialize immediately. We derive our
next hypothesis:

Hypothesis 6 Leaders are less risk averse and more patient than non-leaders.

Leading-by-example is an optimistic act to improve the inefficient situation the group is

in after a phase of mutual underbidding. Taking the initiative, the leader has to trust in his
ability to change the circumstances of the interaction. We therefore hypothesize that leaders
have an internal locus of control.

Hypothesis 7 Leaders have an internal locus of control.

The literature on charismatic or transformational leadership reports mixed evidence on

the impact of the Big Five personality factors on these leadership theories. While Crant and
Bateman (2000) and Judge and Bono (2000) both report a positive impact of extraversion on
charismatic and transformational leadership, De Hoogh et al. (2005) find no such effect, and
that neither in a stable nor in a dynamic work environment. Similarly, both De Hoogh et al.
(2005) and Judge and Bono (2000) report a positive effect of agreeableness on charismatic and
transformational leadership, but Crant and Batemann do not. So we control for the Big Five
personality factors, but have no specific hypotheses concerning their impact on leadership.

Finally, the power of our results depends on the external validity of the leader classification
with respect to taking the initiative outside the lab. The experiment contains two measures
for this aspect, the “real life index” and the “index stories”. We predict that these two indices
correspond to leading behavior in the experiment. In contrast, we expect no significant corre-
lation between taking the initiative in the experiment and the self-stated attitude of being a
leader. Leadership in the experiment is likely to be driven by an implicit power motive (Fodor
2010), i.e. a concern for having an impact. The explicit power motive, in contrast, reflects the
self-proclaimed goal to be in a leading position. According to a review by McClelland et al.
(1989) the implicit and the explicit power motive are not correlated. In our context this means
that people who take the initiative do not differ from others with respect to their inclination
to seek a formal leading position.

15
Hypothesis 8 Leaders in the experiment are also more likely to take the initiative in their
real life outside the lab. A self-stated goal to be a leading personality does not correspond to
leadership in the experiment.

4 Results

We start the review of our results with an overview of the average numbers in all 28 groups.
Figure 1 illustrates them, ordered by the timing of conducting the session. In all groups,
average numbers fluctuate quite considerably, indicating dynamics within the groups. Average
winnings numbers follow a very similar pattern. Looking at the initial phase of the game, we
see that average numbers in most groups decrease from round to round, while in some groups
(groups 2, 5, 9, 17, 24 and 25) they start by increasing. In these groups, at least one player
chooses the number 100 in the first round which triggers the first upward movement of average
numbers right in the beginning of the game. The number 100 in the first round of the game
already seems to be an instrument of leadership.

Let us next consider groups with decreasing average numbers in the first rounds. In almost
all of these groups the downward trend of average numbers stops after at most 10 rounds and
turns into an increase instead. This later increase is always initiated by one player (sometimes
also two at the same time) increasing the number substantially. Different from leadership in
the first round, these leaders in later rounds do not necessarily increase their number to 100. In
order to systematically disentangle intended leadership from casual number increases without
a leading purpose, we use a refinement of the definition introduced by Bruttel (2009a). There,
a leading number has to be more than 30 points larger than the leader’s number in the round
before and it has to be larger than all numbers of all other players in this group in the previous
round.9 We base our classification also on these criteria. In our understanding leadership con-
tains the expectation that others will follow. We therefore define leadership to only be possible
up to two rounds before the end of the game. Later high numbers may be observed for other
reasons but cannot be motivated by the intention to lead. The value of 30 points is necessarily
arbitrary. Bruttel (2009a) explains that 30 is a relatively low threshold including almost all

9
In Bruttel (2009a) no leading numbers in the first round of the game were considered.

16
 100 100 100 100
Group 1 Group 2 Group 3 Group 4
80 80 80 80

60 60 60 60

40 40 40 40

20 20 20 20

0 0 0 0
0 5 10 15 20 25 30 0 5 10 15 20 25 30 0 5 10 15 20 25 30 0 5 10 15 20 25 30

100 100 100 100

Group 5 Group 6 Group 7 Group 8
80 80 80 80

60 60 60 60

40 40 40 40

20 20 20 20

0 0 0 0
0 5 10 15 20 25 30 0 5 10 15 20 25 30 0 5 10 15 20 25 30 0 5 10 15 20 25 30

100 100 100 100

Group 9 Group 10 Group 11 Group 12
80 80 80 80

60 60 60 60

40 40 40 40

20 20 20 20

0 0 0 0
0 5 10 15 20 25 30 0 5 10 15 20 25 30 0 5 10 15 20 25 30 0 5 10 15 20 25 30
100 100 100 100
Group 13 Group 14 Group 15 Group 16
80 80 80 80

60 60 60 60

40 40 40 40

20 20 20 20

0 0 0 0
0 5 10 15 20 25 30 0 5 10 15 20 25 30 0 5 10 15 20 25 30 0 5 10 15 20 25 30

100 100 100 100

Group 17 Group 18 Group 19 Group 20
80 80 80 80

60 60 60 60

40 40 40 40

20 20 20 20

0 0 0 0
0 5 10 15 20 25 30 0 5 10 15 20 25 30 0 5 10 15 20 25 30 0 5 10 15 20 25 30

100 100 100 100

Group 21 Group 22 Group 23 Group 24
80 80 80 80

60 60 60 60

40 40 40 40

20 20 20 20

0 0 0 0
0 5 10 15 20 25 30 0 5 10 15 20 25 30 0 5 10 15 20 25 30 0 5 10 15 20 25 30

100 100 100 100

Group 25 Group 26 Group 27 Group 28
80 80 80 80

60 60 60 60

40 40 40 40

20 20 20 20

0 0 0 0
0 5 10 15 20 25 30 0 5 10 15 20 25 30 0 5 10 15 20 25 30 0 5 10 15 20 25 30

Figure 1: Average numbers (black line) and average winning numbers (grey line) in the 28
groups over all 30 rounds.

17
potential leading bids. Thus, the criterion provides a conservative classification ensuring that
the differences in characteristics between leaders and non-leaders are not artificially amplified.
Variations of the leadership criterion, such als shifting the value of 30 points to 20 or 40 do not
have a qualitative impact on the results. Appendix D provides an overview of changes in the
main results if different criteria of leader definition are applied.

Definition: A number nit of player i in round t is called a “leading number” if one of the
following conditions is satisfied:

1. nit = 100 if t = 1 or

2. nit (t) > nj,t−1 ∀j ∈ [1; 8] and nit (t) > ni,t−1 + 30 if t ∈ [2; 28]

The player i who places the leading number is called a “leader”.

Once leaders are identified according to the above definition we group them into the two
subcategories “early” and “late” leaders. Early leaders are the first leaders in their group, late
leaders are all subsequent leaders. Late leadership is a weaker variant of leading for several rea-
sons. The innovative aspect of leadership disappears if the leader has already observed someone
else leading. Thus, late leaders do not necessarily have to have above-average cognitive skills.
Late leaders have also already observed the reaction of their group to leadership. Therefore,
they have an easier task in forming a belief about the potential gains and losses of leadership
for the leader and the consequences for group efficiency.

early late no leader

Series 1 1.33 1.92 4.75
Series 2 1.15 2.23 4.62

Table 3: Number of leaders per matching group.

With this definition, we identify 31 out of 200 subjects10 as early leaders11 and 52 as late

10
We excluded the data from matching groups number 12, 14 and 15 from the analysis. In these groups, one
participant continuously set the number 100 over almost the whole duration of the game. This disabled us from
classifying the remaining seven participants in these groups into late leaders and non-leaders, because they had
no chance to lead during the whole experiment, even if they wanted to.
11
There are more early leaders than matching groups, because it happened several times that two subjects
led early in the same round.

18
leaders. Accordingly, 117 subjects are classified as non-leaders. Table 3 summarizes the average
number of early, late and non-leaders per matching group across the two series’ of sessions. A
chi-squared test clearly rejects differences in the two distributions (p-value > 0.7). In each
group but one, we are able to identify at least one early and one late leader. The exception
is group 7, where we have two early leaders in the first round, but no late leader. More than
half of the leading numbers had the value 100. Out of the 31 early leaders, 42 percent were
leading at least once more in later rounds, among the 52 late leaders this holds for 33 percent.
Roughly half of the leaders keep or even increase their number in the next round after leading
(43 percent if we double-count leaders leading more than once, 51 percent if we count them
only the first time they are leading), about one third (32 or 40 percent) do not lower their
number for at least two subsequent rounds. The longer the first leader in a group stays at or
above the leading number, the stronger is the effect on followers’ behavior. On average, an
additional round of early leading increases the average number of the other subjects in that
group in all subsequent rounds up to round 28 by 2.3 points (see the GLS regression in Table
4). Furthermore, the scatterplot in Figure 2 illustrates that the higher a leading number is, the
larger is the reaction of the other group members.

Number
Duration of early leadership 2.32***
(0.544)
Period -0.106
(0.188)
Constant 41.18***
(5.532)

Table 4: Regression coefficients: subsequent numbers of the other group members depending on
the duration of initial leadership. Standard errors in brackets. *** denotes significance at the
1% level, ** at the 5% level and * at the 10% level. Standard errors are clustered by matching
group. Subject random effects are included.

4.1 Does leadership pay for the leader - and what do they expect?

We start our analysis of the characteristics of leaders with the question of whether leadership
is profitable for the leaders. On average, early leaders earn 498 points in the experiment, late
leaders 533 points, and non-leaders 595 points. Thus, leaders earn significantly less than non-

19
 70

Change of group average (excl. leader) from t to t+1

60

50

40

30

20

10

-10

-20
0 10 20 30 40 50 60 70 80 90 100
Leading number in t

Figure 2: Average reaction of other group members to different leading numbers.

leaders (p-value = 0.01).12 13

This finding is not surprising given that leaders deliberately
forgo the possibility to win their match while leading and also in other rounds choose higher
numbers (56.58) than non-leaders (45.15) on average (p-value = 0.00). More relevant for the
motivation of leaders (and much more difficult to answer) is the question whether leaders
benefit from leading compared to the counterfactual situation in which they do not lead - and
whether they correctly anticipate their net monetary loss or benefit from leading. We cannot
answer the first question because we do not have a reference point which we could compare
leaders’ profits with. However, we can say that they are very good in anticipating the extent
to which others will follow their good example, which is the basis for their own expected loss
or gain from leading. We use the beliefs submitted for the average number of the seven fellow

12
For a statistical comparison of leaders and non-leaders we treat each matching group of eight participants
as one independent observation. Thus, we consider 25 independent observations, 12 for the first series and 13
for the second series. Within each group, we average the scores for each measure, e.g. the profit, over all early
leaders, late leaders and non-leaders separately. All reported significance levels in this paper are then obtained
(if nothing else is stated) in one-sided Wilcoxon signed rank tests testing the measures of (early or late) leaders
against the non-leaders in each matching group. We do not correct for multiple hypothesis testing since we have
ex ante hypotheses for almost all tests that we conduct.
13
Average profits per matching group are not significantly different in the two series’ (p-value = 0.28, two-sided
test).

20
participants.14 If leaders overestimated the average number of the others, it was very likely
that they overestimated their gain from leading as well. Leaders are generally very good in
estimating the reaction of others after their leading bid. If at all, they slightly under- rather
than overestimate the average number of the seven other participants in the round after their
leading bid (by about 4 units on the scale from 2 to 100). The quality of their estimate does not
depend on whether they were leading in the previous round or not, and it is also not different
from the quality of the estimates of the non-leaders. We conclude that over-optimistic beliefs
as in Gächter et al. (2012) are not driving leadership in the framework of our number choice
game. This makes it unlikely that selfish motives are the major driving force for taking the
initiative. Nevertheless, we cannot exclude that some leaders are leading because they expect
to earn more by leading than in the counterfactual situation without any leader.

Result 1 (i) Leaders earn less than non-leaders. (ii) Leaders have realistic beliefs about how
much the followers respond to their leading decision.

Beliefs that others will lead

0.6
Belief Actual value
0.5

0.4

0.3

0.2

0.1

0.0
Leading leaders Not leading leaders No leaders
(N=58) (N=553) (N=821)

Figure 3: Belief that others will lead.

The decision to lead might not only depend on the belief whether others will follow but also
on the belief whether someone else in the group will take the initiative instead. To capture

14
Given the simplifying assumption that beliefs b are uniformly distributed within each of the five intervals
2-20, 21-40, 41-60, 61-80, and 81-100, we computed a point estimate for the predicted average of the other seven
numbers as (b2−20 ∗ 11 + b21−40 ∗ 30.5 + b41−60 ∗ 50.5 + b61−80 ∗ 70.5 + b81−100 ∗ 90.5).

21
this belief, players had to submit an estimate for the probability distribution over the intervals
2-20, 21-40, 41-60, 61-80, 81-100 of the maximum of the other seven numbers in their group in
each round. We approximate the belief on whether there is another leader with the percentage
weight given to the upper interval from 81 to 100.15 Figure 3 contrasts this belief with the actual
frequency of at least one other group member choosing a number larger than 80, separating
between leaders who are currently leading, leaders who are currently not leading and non-
leaders. The data behind this illustration contains only values from rounds where leading was
generally possible, i.e. rounds in which the maximum number in the round before was smaller
than 100 and the minimum number was smaller than 70. In line with the argumentation of a
false consensus effect (Ross et al., 1977) leaders would generally overestimate others’ willingness
to lead (p-value = 0.12, two-sided test) while non-leaders underestimate the probability that
there would be a leader (p-value = 0.11, two-sided test). However, in their leading round, leaders
underestimate others’ willingness to lead (p-value = 0.03, two-sided test). In our interpretation,
this pessimistic belief about the probability that others will lead additionally motivates leaders
to take the initiative.

4.2 Leaders attach a high value to efficiency

Figure 4 summarizes the decisions of all participants in the role of player 1 in the distribution
games. Leaders have stronger preferences for efficiency and they are more generous. We find the
most notable difference for early leaders when there is a conflict between efficiency and equality.
In this game, early leaders have an average efficiency score16 of 1.52, while late leaders and the
followers have an efficiency score less than 1. Table 5 summarizes the significance levels. For
efficiency concerns it holds that early leaders choose the efficient option more often than all

15
This approximation does not perfectly match actual leading bids, which could also be lower than 81.
Eliciting beliefs perfectly fitting our leadership criterion would have meant using an even more detailed and
complex elicitation procedure which we abstained from for practical reasons.
16
This score takes the highest value 3 if a player chooses the efficient option in the game where the efficiency
gain of this option compared to the selfish option is the smallest. If a player does not choose the efficient option
in this game but does so in the next game where the efficiency gain increases by one unit, this player gets a
score of 2. Similarly, a score of 1 is assigned to players choosing the efficient option only in the third game.
Otherwise, the score is zero. If we simply take the sum of efficient choices as a subject’s score, total scores are
slightly smaller, but the relative differences between them and the statistical significance levels remain almost
the same.

22
other players, and late leaders score higher than non-leaders. For generosity we observe a
similar pattern, only the difference between late and non-leaders is just not significant. With
respect to the games testing a preference for equality and efficiency versus selfishness, the major
difference is between leaders and non-leaders, with no significant distinction between early and
late leaders.

To find out whether social preferences are purely coincidental attributes of leaders or in fact
a motive for leading, we analyzed their answers to the open questions in the post-experimental
strategy questionnaire. In this questionnaire 69 percent of the early leaders explain that they
started choosing a very high number because they wanted their group to coordinate on the
socially optimal outcome, compared to null early leaders stating that they were leading in
order to increase their own profit in future rounds (p-value = 0.00, one-sided test).17 For the
late leaders, the efficiency motive occurs in 29 percent and the selfish motive in 20 percent of
the statements (p-value = 0.05). Taken together, the results from the distribution games and
the answers from the questionnaires indicate that concerns for others’ outcomes are a driving
force for leading-by-example. Leaders are more pro-socially minded than non-leading players.
2.0 2.0 1.00

1.5 1.5 0.75

1.0 1.0 0.50

0.5 0.5 0.25

0.0 0.0 0.00

Early Late No leader Early Late No leader Early Late No leader
(N=31) (N=52) (N=117) (N=31) (N=52) (N=117) (N=31) (N=52) (N=117)
Efficiency vs. equality Efficiency and equality Generosity

Figure 4: Valuation of efficiency, equality and generosity.

early vs. no leader early vs. late late vs. no leader

Equality vs. Efficiency 0.00*** 0.01** 0.07*
Equality and Efficiency 0.02** 0.17 0.09*
Generosity 0.02** 0.03** 0.12

Table 5: Wilcoxon signed rank tests (one-sided) for decisions in the distribution games. ***
denotes significance at the 1% level, ** at the 5% level and * at the 10% level.

17
We had an independent sample of 27 subjects classifying the answers into categories predetermined by us.
Each of them received one third of the statements for classification. The statements were presented in random
order to the different subjects. They were paid 15 euros for this task. A certain statement is defined to belong
to a category if the majority of these subjects classified it accordingly.

23
Result 2 Leaders attach a high value to the maximization of others’ welfare. Early leaders are
more willing to accept disadvantageous inequity than others when it is in conflict with efficiency.

4.3 Do leaders like attention?

Leaders might lead not because they want to do something good for their group, but because
they want their group to see that they are doing something good. Comparing the use of the
publicity and the highlight option after the number choice allows us to disentangle these two
motives. Figure 5 shows that leaders indeed use the publicity feature. In leading rounds,
around 17 percent of the leaders opt for publicity; one round later (when the leaders often still
have the highest number in their group), this holds for 12 percent of them. In comparison,
non-leaders pay for publicity in less than 1 percent of their decisions and currently not leading
leaders in only 3 percent. The differences between currently leading leaders (in t or t − 1) and
non-leaders as well as between currently leading leaders and currently not leading leaders are
statistically significant (the p-values are 0.02 and 0.03, respectively). As expected, men use the
publicity feature on average more often than women, but the differences between them are far
from being significant.

0.45
Publicity
0.40
Highlight
0.35

0.30

0.25

0.20

0.15

0.10

0.05

0.00
Leader in t Leader in t-1 Leader, but not Non-leader Leader in t Leader in t-1 Leader, but not Non-leader
in t and t-1 in t and t-1
(N=58) (N=58) (N=1015) (N=1653) (N=70) (N=70) (N=1136) (N=1740)
Publicity only Highlight and Publicity

Figure 5: Frequency of payment for giving up anonymity and for highlighting a number.

Adding the highlight option remarkably reduces the use of the publicity feature by leaders
in the second series while the highlight option, if available, is used by 40 percent of the leaders.
This indicates that leaders in the first series mainly used the publicity option to strenghthen the

24
effect of giving a good example while the benefit from the positive image of being a leader plays
if at all a minor role. When the highlight option is available, the difference between currently
leading leaders and currently not leading leaders in the use of the publicity feature is no longer
significant (p-value = 0.11) and the difference between currently leading leaders and non-leaders
is only weakly significant (p-value = 0.08). Again, men use both the publicity feature and the
highlight option more often than women without the difference being statistically significant.

Both highlighting and giving up anonymity have the desired effect to attract attention and
strenghthen the signal for coordination within the group. More participants increase their
number after a leading bid with (61 percent) than without (44 percent) three exclamation
marks aside (p-value = 0.01). Publication of the leader’s seat number has a similar, though
not significant effect.18

Result 3 Leaders want to direct other’s attention to the good example they give, but they are
only slightly more likely to reveal their identity than non-leaders.

The finding that leaders in our experiment are rather not status-seeking seems to contradict
the related finding of Arbak and Villeval (2013). However, the results are in fact very much
in line with each other. Arbak and Villeval (2013) observe that subjects who volunteered to
contribute in the first stage often contribute a much lower amount when assigned to make their
contribution in the second stage. More specifically, they say that subjects, who contribute
much to the public good when others can see their contribution before contribution themselves,
are driven by status concerns. However, this behavior is perfectly comparable to anonymous
leading in our experiment, that is, choosing a high number to make others imitate. Thus, what
Arbak and Villeval name a “positive social image” for the leader could actually be very similar
to “attention for my number” in our experiment.

4.4 Leaders show a high degree of cognitive reflection

After the main experiment, all participants had to answer three questions from a cognitive
reflection test (CRT). There was no incentive for giving a correct answer and no feedback.

18
In only 5 of 25 groups we observe both a leading bid with and without the leader using the publicity option.

25
Each correct answer gives one point in our evaluation so that participants could get between
zero and three points in this task. Figure 6 shows that leaders have a significantly higher
score on the cognitive reflection test than non-leaders (p-value = 0.00). The difference is also
significant between early and late leaders (p-value = 0.01) and between late and non-leaders
(p-value = 0.02).

2.5
Early
2.0 Late
No leader
1.5

1.0

0.5

0.0
All subjects Male Female
(N=200) (N=87) (N=113)

Figure 6: Scores in the cognitive reflection test.

As in Frederick (2005), men score higher in the cognitive reflection test than women. Men
are significantly more often early leaders than women (24 percent of the males, 9 percent of
the females, p-value = 0.01) and women are significantly more often classified as non-leaders
than men (49 percent of the males, 65 percent of the females, p-value = 0.03). However, the
relative difference of the CRT scores for leaders and non-leaders does not depend on gender
as can be seen in the second and third group of bars in Figure 6 (p-values leaders vs. non-
leaders: males = 0.03, females = 0.10). Thus, the special kind of intelligence measured by the
cognitive reflection test and more frequently observed with men, seems to be associated with
taking-the-initiative in the experiment. The probit regression in Table 6 reveals that this result
is statistically significant. While the variable male significantly affects the leading probability
without further control, it does not when we control for the performance in the CRT task.

Result 4 Leaders perform better than non-leaders in a cognitive reflection task.

Result 5 Men are more likely leaders. The difference between men and women disappears
when controlling for performance in the cognitive reflection task.

26
 Leader Leader
Male 0.413** 0.214
(0.188) (0.234)
CRT-score 0.254**
(0.0991)
Constant -0.398*** -0.694***
(0.0907) (0.126)

Table 6: Regression coefficients: leadership depending on crt score and gender. Standard errors
in brackets. *** denotes significance at the 1% level, ** at the 5% level and * at the 10% level.
Standard errors are clustered by matching group.

The result that women take the initiative in this experiment less often than men is in contrast
to the empirical finding that female leaders apply a transformational leadership style more often
than male leaders (see the meta-study by Eagly at al. 2003). The importance of the depth
of reflection - as measured by the CRT score - for leadership in our experiment may explain
this discrepancy. Furthermore, the difference could stem from a selection effect in empirical
studies considering only established real-life leaders, where the strength of selection into leading
positions may be stronger for women than for men. As our student sample of subjects is more
general than samples consisting of real-life leaders only, we have no such selection.

Average time in belief stage Average time in number choice stage Average time in number choice stage
sec sec (with belief stage) sec (no belief stage)
30 30 30

20 20 20

Current leader Current leader Current leader

10 10 10
Average over all leaders and rounds Average over all leaders and rounds Average over all leaders and rounds

0 0 0
t-3 t-2 t-1 Leader t+1 t+2 t-3 t-2 t-1 Leader t+1 t+2 t-3 t-2 t-1 Leader t+1 t+2

Figure 7: Reaction times.

A second piece of evidence from the recording of reaction times supports the interpretation
that cognitive skills are a determinant of leadership in the experiment. Figure 7 illustrates the
average time which leaders spend in the belief formation stage and in the decision stage of the
experiment. Given the cognitive effort needed before a player decides to lead, we would expect
that leading decisions take longer than other decisions.19 In fact, leaders’ belief formation times
slow down significantly before their leading decision while the actual decision making gets even

19
For an interesting application of response time to economic decision making see Rubinstein (2007).

27
faster in the leading round. Similarly, in the session without the belief stage, decision times
slow down before the leading round and get faster in the leading round itself. The regressions
in Table 7 show that reaction times of leaders slow down before the decision to lead. “Ever
leader?” distinguishes leaders from non-leaders, because it might be that leaders are generally
slower or faster than non-leaders in their decisions. The variables “Leader in t”, “Leader in
t + 1?”, and “Leader in t + 2?” are dummy variables equal to one if the subject is a leader in
the respective round. Using them, the regression captures changes in the response times in the
leading round and two rounds before compared to rounds in which the person in consideration
acts as a leader neither in the current nor in the two subsequent rounds. As can be seen in
the first regression, the formation of beliefs lasts significantly longer in the leading round and
already one round before. The time for the actual number choice in the second and third
regression slows down in the two rounds before leading and quickens in the actual leading
round, significantly only in series 1 with the preceding belief stage.

Log belief time Log decision time

Series 1 Series 2
Round -0.0308*** -0.0114*** -0.0166***
(0.00308) (0.00169) (0.00201)
Ever leader? -0.0211 0.110* 0.0214
(0.0620) (0.0566) (0.0449)
Leader in t? 0.269** -0.187** -0.112
(0.0989) (0.0822) (0.0779)
Leader in t + 1? 0.310*** 0.0653 0.0786
(0.0922) (0.0593) (0.0743)
Leader in t + 2? 0.115 0.0688 0.0621
(0.119) (0.0558) (0.0647)
Constant 3.443*** 3.066*** 3.008***
(0.0710) (0.0444) (0.0418)

Table 7: Regression coefficients: log of time spend in belief stage and decision stage. Standard
errors in brackets. *** denotes significance at the 1% level, ** at the 5% level and * at the
10% level. Standard errors are clustered by matching group.

28
 Our understanding of this change in reaction times is that the decision to lead develops
while participants are forced to think about others’ behavior in the belief formation stage. The
fast leading decision itself could be interpreted to be a self-commitment not to rethink the
courageous decision to take the initiative. As the decision to lead seems to be formed already
during the belief stage, it could also be that finally entering the leading number simply gets
faster than selecting a number potentially maximizing profit against the numbers of the other
players shown on the decision screen.

The reaction times of followers after a leading number further support the idea that reaction
times provide a measure for the intensity of thought before a decision. The decision times for
the number choice of followers significantly (p-value = 0.00) slow down in the round after a
leading number (21.99 seconds) compared to rounds where no leading number was set in the two
previous rounds (19.00 seconds). In the second round after a leading number, average decision
times are with 19.83 seconds still slower (p-value = 0.00). The differences remain significant
when considering the data from the first and second series separately. Belief formation times
in series 1 get slower as well. In the round immediately after a leading number, the average
belief formation time is 27.60 seconds, compared to 24.33 seconds in rounds without a leading
number in the two previous rounds (p-value = 0.03). Two periods after a leading number, the
belief formation lasts 28.59 seconds, which is again significantly (p-value = 0.01) more than in
normal rounds.

4.5 Personality measures

Risk aversion, as measured by the Holt and Laury (2002) lottery procedure, has no significant
effect on leadership. Early leaders are a little less (average number of safe choices 5.35), late
leaders a little more (5.81) risk averse than non-leaders (5.70), but the p-values in Table 8 are
far from any reasonable level of significance. Also the self-stated risk attitude does not differ
significantly between leaders and non-leaders. This contradicts our hypothesis that leaders
have a more positive attitude towards risk than non-leaders. The reason might be that the risk
of leadership is different from (and hardly correlated with) the risk measured with the Holt
and Laury (2002) lotteries. Their procedure generates risk as random draws between lotteries
while the risk of leadership is a behavioral risk depending on the reaction of followers. The

29
former requires calculation of expected values while the latter depends on the ability to deal
with strategic uncertainty.

early vs. no leader early vs. late late vs. no leader

Holt & Laury risk aversion 0.22 0.17 0.47
Self-stated risk aversion 0.14 0.50 0.17
Patience 0.03** 0.35 0.02**
Locus of control 0.08* 0.03** 0.50

Table 8: Statistical tests for decisions in the distribution games, risk attitude, and gender. ***
denotes significance at the 1% level, ** at the 5% level and * at the 10% level.

Patience is required from both early and late leaders as they have to wait for future rounds
to regain the profits waived while leading. As can be seem from Figure 8 and Table 8, both
types of leaders score significantly higher on the scale of patience than non-leaders,20 but the
absolute difference is rather small.

Result 6 Leaders are more patient than non-leaders. Risk aversion does not play a role for
the decision to take the initiative.

Following Piatek and Pinger (2010), we consider the locus of control as a unidimensional
concept. The distinction into an internal and an external dimension of the locus of control
as implemented, for example, by Caliendo, Cobb-Clark and Uhlendorff (2010) does not seem
convincing to us, because we cannot imagine what else a non-internal locus of control should
be if not an external locus of control and vice versa. The results of a factor analysis for the ten
items of the locus of control questionnaire point in the same direction as the intutive argument
above: It indicates that item 1 has a negative loading on the main factor, while items 4, 6
and 9 neither load on the same factor as the other seven items nor on a joint second factor.
Consequently, our index of the locus of control recodes item 1 and excludes the items 4, 6
and 9. This procedure also delivers the highest value for internal consistency as measured by

20
The questionnaires for patience, locus of control and big five personality traits included the statement “You
can trust my answers.” as the last item. Subjects selecting an answer of less than the maximum minus 1 unit
to this item were not considered in the analysis. This concerns four subjects in the big five questionnaire, six in
the locus of control questionnaire, and three in the patience questionnaire, with considerable overlap between
questionnaires.

30
Cronbach’s Alpha, 0.637. For the patience and the big five questionnaire no such adjustments
are necessary as already the full scales deliver very reasonable values for Cronbach’s Alpha.
4.0 6.0
5.0
3.0
4.0
3.0
2.0
2.0
1.0 1.0
Early Late No leader Early Late No leader
(N=15) (N=29) (N=58) (N=15) (N=29) (N=58)
Patience Locus of control

Figure 8: Patience and locus of control (a high value indicates an internal locus of control).

Figure 8 illustrates that an internal locus of control corresponds to leadership in the ex-
periment. The main (but nevertheless rather small) difference for the locus of control in our
experiment is between early leaders and late leaders with rather no difference between late lead-
ers and non-leaders. Intuitively, only early leaders need intrinsic confidence that their behavior
can change their environment while late leaders have already observed that leading behavior
by other players does have the desired effect.

None of the big five personality traits has a significant effect on leadership. The average
scores of early leaders, late leaders and non-leaders with respect to all five traits show almost
no differences. If at all, a low score on the scale for neuroticism seems to be associated with
leadership in the experiment (p-value early vs. non-leaders = 0.09, two-sided test), a result
which is also reported by Arbak and Villeval (2013).

Result 7 Leaders have a more internal locus of control. Other personality traits do not play a
role for leadership in the experiment.

4.6 External validity

Table 9 reports two measures for the external validity of our classification of subjects into leaders
and non-leaders, reflecting the implicit power motive. The data stems from the second series
of the experiment. The “real life index” consists of the sum of the values of the seven variables

31
revealing participants’ actual propensity to take the initiative in real life. “Index stories” refers
to the sum of points received for the stated behavior in the hypothetical situations.21 This
index can vary between +5 and −5.

Early Late Non-leader

Implicit power motive Real life index 3.20 2.17 2.35
Index stories 2.47 2.31 1.45
Explicit power motive Self-stated personal initiative 0.68 0.66 0.66
Leader in a student work group 0.53 0.38 0.42
Wants to be in a leading position 0.60 0.45 0.55

Table 9: Measures for the implicit and explicit power motive.

Both indices show that being classified as a leader in the number choice game corresponds
to taking the initiative in life outside the lab. For the real life index, early leaders on average
receive a score of 3.20 while non-leaders have a score of only 2.35 (p-value = 0.02). Late leaders
are not significantly different from non-leaders with respect to this index. For the story index,
we find that both early and late leaders score higher than non-leaders (p-value leaders vs.
non-leaders = 0.05).

In addition to the implicit power motive revealed in the real life index we also assessed
participants’ self-evaluation of their proneness to take the initiative, reflecting their explicit
power motive. They had to classify themselves as more or less “initiative-taking”, they were
asked whether they would like to work in a leading position, and what role they typically have
in a student work group. None of these self-stated measures corresponds significantly to taking
the initiative in the experiment. We interpret this finding as evidence that taking the initiative
is a facet of personality which is different from the self-stated claim to be a leader.

Result 8 (i) Leadership according to the behavior in the number choice game corresponds to
the degree of taking the initiative in participants’ real lives. (ii) The self-stated claim to be a
leader does not correspond to leadership according to the behavior in the number choice game.

21
We excluded the data of two subjects (number 202 and 223) from this part of the analysis, because they
stated orally during the experiment that they misunderstood the question and answered the first questions in
this part randomly before asking for an explanation of the question.

32
5 Conclusion

What are the characteristics behind leadership? We address this question for a particular type
of leadership, taking the initiative. So far, the implicit power motive behind such behavior
seemed difficult to assess. In a simple questionnaire, people who just want to be a leader
without having a particularly social motivation would also state that they take the initiative
whenever possible. The most common alternative measurement method is the picture story
exercise (Pang 2010). In this test, ambiguous pictures of everyday situations are presented to
participants, who have to imagine and write stories explaining the picture. The written stories
are interpreted by the experimenter as representing a stronger or weaker implicit power motive.
In this paper we use an experiment which can identify people who take the initiative in an
incentive compatible way.

We combine the experimental design in which leadership develops endogenously with several
other small games and questionnaires to explore possible characteristics of leaders. Our design
permits classification of subjects as leaders and non-leaders and to study the determinants of
leadership. We find that traditional personality measures are not predictive for leading-by-
example, but characteristics such as an internal locus of control and patience are. Further
main determinants are that leaders attach a high value to efficiency, are not envious and they
have better cognitive abilities than non-leaders. We can show that taking the initiative in the
experiment corresponds to similar behavior in participants’ real lives outside the lab, but it
does not correlate with the self-stated intention to be a leader.

Our results have implications for the creation of an environment, be it in firms or public
administration, where initiative is desired. As we find that leaders have above-average cognitive
abilities and are willing to use them, we recommend creating a stimulating environment, as for
example, the company Google is known for.22 The rather internal locus of control leaders have
calls for measures fostering trust in self-efficacy such as the assignment of responsibility. From
the finding that leaders attach a high value to efficiency we conclude that it is worthwhile to
give potential leaders reason to believe in the successful implementation of their suggestions.

22
See, e.g. “A Place to Play for Google Staff”, New York Times, March 16, 2013, p. B1.

33
 Leadership, and in particular leading-by-example, is an important and desired trait for
many jobs. Thus, measurement devices and the pattern of determinants of this behavior are
highly desirable. Our results suggest that traditional personality traits are not very predictive
for this behavior. This implies that leadership has to be assessed in a different way. We do
not claim that our experiment provides the only way to do so. For example, it is not deception
proof, and the measure of leadership depends on the comparison group. Once participants
know about the purpose of the game, the classification into leaders and non-leaders according
to behavior in that game is hardly possible. Nevertheless, it provides interesting insights into
the mechanism of taking the initiative and suggests a new way to measure a disposition for
taking the initiative.

Appendix A: instructions

General instructions at the beginning

Welcome and thank you for participating in this economic experiment.

This experiment consists of multiple parts. The instructions for the first two parts of the
experiment will be displayed on your computer screen. The instructions for the third part will
be handed out later in hard copy. All instructions are identical for all participants.

Please read the instructions carefully. If you have any questions regarding the experiment
please raise your hand. We will then come directly to your place. Please be quiet during the
experiment and do not talk to other participants. Failure to comply with these rules will result
in an exclusion from the experiment. If this occurs you will not receive any payment.

After you have completed all three parts of the experiment please fill out the following
questionnaires on your computer screen. Afterwards you will receive your payment for the entire
experiment. The order in which participants receive their payments is already determined,
beginning with the participant sitting at the computer “lakelab 1”. So take your time to fill in
the questionnaires. Your speed will have no influence on the timing of your payment.

34
Instructions for the number choice game

Now we will start with the third part of the experiment. After this part the experiment will
be over and we will ask you to fill out some questionnaires.

Your gains and losses during the experiment are counted in points. The exchange rate is 30
points for 1 euro. Your payment in this part of the experiment depends on your decisions and
on the decisions of other participants.

This experiment will last for 30 rounds. In each round you will be asked to choose a number
between 2 and 100. Subsequently, the computer will randomly determine one participant out
of a group of eight and compare the numbers you and the other participant have chosen. The
participant who selected the smaller number receives as many points as her number. The other
participant receives zero points in this round. If both of you selected the same number, each of
you gets half of the points. At the end of each round you are informed about your payment in
points and about the numbers all participants of your group have chosen. The composition of
your group of eight does not vary during the 30 rounds. Out of this group in each round one
participant will be randomly chosen and your numbers will be compared.

In each round before choosing a number you will be asked to make an estimate about
the numbers which the other seven participants of your group are going to choose in this
round. More specifically, you have to submit your belief about what is going to be the highest,
the lowest and the average number of the other seven participants. We ask you to forecast
the probability of these three numbers (maximum, minimum and average) being within the
following intervals: 2-20, 21-40, 41-60, 61-80 and 81-100. For each of the five intervals you
have to indicate the percentage value of the three numbers (maximum, minimum, average)
being within these intervals. The five percentage values add up to a total of 100%, because the
numbers have to be within one of the intervals no matter what. We place a graphical computer
program at your disposal so you can enter your beliefs. You will have the opportunity to
familiarize yourself with the program before the experiment begins. Here you can see what the
program looks like:

35
 Figure 9: Belief formation tool.

You can change the height of the bars implemented in the program by clicking on a bar,
holding the left mouse button and moving the mouse. Do not worry about whether the per-
centage values add up to 100 or not. Just change the heights of the bars until their proportions
match the relative probability you propose. Then click on the button “update” next to the
diagram. The bars are automatically adjusted so the values of your estimates sum up to 100.
After entering your belief for minimum, maximum and average please click on “next”. Next
you can choose your number for the coming round.

There will also be a payment for the accuracy of your guess. The exact computation of this
accuracy-dependent payment is described in detail in the appendix. If you have no interest in
the details, feel free to ignore the explanations concerning this matter. The only important
thing you have to know is that you maximize your payment by indicating your true beliefs.

From the second round on, your previous estimates will be the default setting, so you only
have to indicate new numbers in case you want to adjust your previous estimates.

Your decisions in this experiment are always anonymous. The other participants of your
group can only see the number you (and all the other participants) have chosen, but not the
number of the computer you are sitting at. The numbers are ordered by size. So it is not

36
possible to draw conclusions about participants’ seats from the numbers. If in a particular
round you want the other participants not only to know the number you have chosen, but also
the number of the computer you are sitting at, you can determine so with a mouse click on
your computer screen. To disclose the number of your seat you have to pay 10 points.

Before the experiment begins you have the opportunity to familiarize yourself with the
computer program. After the experiment please fill in the questionnaires. You will be paid in
cash directly after the end of the experiment and after you have finished the questionnaires.

If you have any further questions regarding the conducting of the experiment, please give a
short notice to the supervisors of the experiment. We will then come directly to your place.

Payment for probability estimates

As previously described, for the three numbers maximum, minimum and average you allo-
cate five probability values pi to the five intervals 2-20, 21-40, 41-60, 61-80 and 81-100. The
actual number (for example the minimum) lies later in one of these intervals. For one prob-
ability estimate you can earn 2 points at most. If your estimate is not accurate there will
be subtractions from the 2 points. The probabilities you have assigned to intervals in which
the actual number does not lie, will be squared and subtracted from your maximal payment.
For example, if you set 70% on the lowest interval but the actual number does not lie in this
interval, 0.49 = 0.70 ∗ 0.70 points will be subtracted from your payment. Furthermore, it is dis-
advantageous if the probability value you distributed to the interval in which the actual number
lies deviates significantly from 100%. This deviation will also be squared and subtracted from
your payment. If you set 60% on the right interval, (1 − 0.60) ∗ (1 − 0.60) = 0.16 points would
be subtracted.

The smaller the sum of the squared wrong estimates is, the better was your guess. For those
who are interested, here is the mathematical formula to calculate the quality Q of your guess:

p2wrong,j − (1 − pright )2
P
Q=2−

In each round the computer will calculate the quality Q of your estimate for minimum,
maximum and average number. The higher the quality Q is, the better was your guess in
that particular round. At the end of the 30 rounds of the experiment your 30 values of Q

37
for minimum, maximum and average will be summed up. This value will be added to your
payment in points.

Examples

In the following we will describe some examples of the calculation of the quality of your
estimate and demonstrate some useful tips on how to improve your estimate.

If you think that the smallest of the seven numbers of the other participants of your group
definitely is equal to or smaller than 20, you say the probability of the minimum being within
the interval 2-20 is 100% and the probability for the minimum being within one of the other
intervals is 0%. In this case you gain 2 points if your guess is correct and no points are
subtracted for false estimations, because you were 100% right. If you had distributed 20% to
each of the five intervals, you would have scored only 1.2 points. In general: if you are sure
about the actual number not being within a certain interval, it is better for you to assign a
probability of 0% to this interval. Intentional probability “dispersion” does not pay off.

If you think that the highest of the seven numbers of the other participants of your group is
either in the interval 61-80 or is higher than 80, but you are sure that the maximum definitely
lies above 60, you should assign the value 50% to both intervals 61-80 and 81-100. In this case,
your expected payoff is higher than in case you assigned 100% probability to only one of the
intervals: If you assign 50% to both of the intervals, you surely gain 1.5 points. If you assigned
the entire 100% to one of the intervals, you gained 2 points in case you were right and 0 points
in case you were wrong. So your expected payoff would be only 1 point. In general: If you
think that a given number is possibly within several intervals and the probability of the number
being in each of these intervals is equal, it is best for you to enter equal probabilities to these
intervals.

38
Appendix B: patience questionnaire

In the following you find a list with statements. You will probably agree completely to some
and not at all to others. To some others you might be undecided.

Please answer according to the following scale. If you do not agree to the statement at all, then
select the button to the left. If you agree to a statement completely, then mark the button to
the right. In between you can grade your opinion.

Please state what you really think. Nobody is here you have to impress. The results can only
be used scientifically if you answer honestly.

I frequently feel like hurrying others.

If I want something I get it.
I always have something to do in case I have to wait.
I am often in a hurry.
I often lose track of what people are saying if they go on for too long.
I consider myself as easy going.
I have trouble finding time to get my hair cut.
I wait too long to act.
I get things accomplished without undue stress.
I have enough time to do the things that are important to me.
I work fast.

Remark: Items 1, 2, 4, 5, 7, and 11 were recoded in the analysis.

Appendix C: questionnaire “index stories”

In the following different scenarios are going to be presented to you.

For each scenario we offer you different action alternatives.

39
Please select for each scenario which action alternative you would pick most likely and which
you would pick the least.

Question 1:

The bus you have to take to the university every day is overcrowded. Since your stop is near
the beginning, everyone has a spot on the bus. However, this is not true for later stops, some
people had to wait for the next bus. What would you do?

a) I am going to write a letter to the bus company and ask them to reduce the problem by
putting in another bus on this line. (-1; 1)23

b) As long as I get in I do not care. (1; -1)

c) Because these many people bother me in the morning, I decide from now on to take a bus
earlier or after the busy times whenever possible. (0; 0)

d) If it goes on like this the bus driver will soon realize that a change is necessary - and after
all it is his task to make sure to transfer all passengers. (1; -1)

Question 2:

A good friend of yours is celebrating his birthday in two days. Among your friends it is common
to buy a present from all of you. It is in the middle of February and exams are right ahead.
Since everybody is studying nobody volunteers to get the present. What would you do?

a) As everybody knows I am going to write one more exam than the others. The others will
consider this for sure and are going to leave me out of the organisation of the present. (1;
-1)

b) I propose that I will think about a present and somebody else will organize it. (-1; 1)

c) I will go to the city after my class and check if I can find something suitable. (-1; 1)

23
In brackets after each statement are the points for answering: (fits the least; fits the most).

40
 d) Since I have to study and I am hesitating to go to the party anyway I will keep out of it.
(1; -1)

Question 3:

Since the introduction of tuition fees the university library has more financial resources. But
there are still not enough copies of a standard reference which is needed by the second term
students for their exam. What would you do in this situation?

a) I will buy the book at Amazon. (0; 0)

b) I will organize a study group with fellow students. So we can study together with one book.
(-1; 1)

c) I will go to the information desk of the library and ask them to get another copy of the
book. (-1; 1)

d) I assume that the professor knows about the shortage of the books and that he will not ask
too many details in the exam. (0; 0)

Question 4:

Recently you moved in with two friends. So far there are not any agreements about the cleaning
of the shared rooms (kitchen, bathroom). What would you do?

a) I will get an organizer in which I will list who will have cleaning duty in which week. I will
start. (-1; 1)

b) I will clean the kitchen and bathroom when the rooms become too dirty for me. (-1; 1)

c) Since I am at the university all day and going home at the weekends, I make little dirt and
do not feel responsible for cleaning. (1; -1)

d) I plan to talk to my roommates on the next occasion. (0; 0)

41
Question 5:

The cleaning staff did not refill the soap in the washing room of your working place for some
days. How would you react?

a) I will post a note for the cleaning staff at the door to the washing room, they should
remember refilling. (-1; 1)

b) I will bring my own soap and put it at the sink. (-1; 1)

c) That does not bother me. I rarely wash my hands with soap.(0; 0)

d) The cleaning staff gets controlled regularly at a random basis, somebody who is responsible
for it will realize it soon. (1; -1)

Appendix D: variations of the leadership criterion

The following table provides an overview how variations in the leadership criterion affect our
main results. The first block repeats data from the criterion used in the paper. The second
block adds the condition that the leader’s number has to exceed the previous maximum by at
least 10 points. Blocks 3 to 5 vary the required increase of the leader’s number from round
t − 1 to the leading round t from its original value of 30 to 10, 20, and 40. Finally, the sixth
block applies the same criterion as in the paper but includes only leading numbers which have
the value 100.

The values for efficiency, CRT score, and risk aversion are computed based on the whole
dataset of 200 participants. For the stories index and the real life index, the data comes from
the second series only.

Qualitatively, the results of the different criteria look very similar to the definition used in
the main text, and also the results of statistical tests are robust to the variation, with only
one major exception. When only increases to 100 are allowed as leading numbers, differences
between leaders and non-leaders with respect to the real life index turn out to be no longer
significant.

42
 Early Late Non-leader
nit = 100 if t = 1 or nit (t) > nj,t−1 ∀j ∈ [1; 8] and nit (t) > ni,t−1 + 30 if t ∈ [2; 28]
Absolute Number 31 52 117
Efficiency 1.52 0.77 0.56
CRT score 2.06 1.62 1.26
HL risk aversion 5.35 5.81 5.7
Stories index 2.47 2.31 1.45
Real life index 3.2 2.17 2.35
nit = 100 if t = 1 or nit (t) > nj,t−1 + 10∀j ∈ [1; 8] and nit (t) > ni,t−1 + 30 if t ∈ [2; 28]
Absolute Number 31 51 118
Efficiency 1.52 0.78 0.56
CRT score 2.06 1.65 1.25
HL risk aversion 5.35 5.84 5.69
Stories index 2.47 2.39 1.42
Real life index 3.2 2.14 2.36
nit = 100 if t = 1 or nit (t) > nj,t−1 ∀j ∈ [1; 8] and nit (t) > ni,t−1 + 10 if t ∈ [2; 28]
Absolute Number 31 61 108
Efficiency 1.39 0.80 0.56
CRT score 1.94 1.59 1.29
HL risk aversion 5.29 5.75 5.74
Stories index 2.75 2.09 1.42
Real life index 3.25 2.18 2.33
nit = 100 if t = 1 or nit (t) > nj,t−1 ∀j ∈ [1; 8] and nit (t) > ni,t−1 + 20 if t ∈ [2; 28]
Absolute Number 31 58 111
Efficiency 1.48 0.74 0.58
CRT score 2.00 1.60 1.27
HL risk aversion 5.23 5.79 5.74
Stories index 2.67 2.13 1.45
Real life index 3.20 2.16 2.37
nit = 100 if t = 1 or nit (t) > nj,t−1 ∀j ∈ [1; 8] and nit (t) > ni,t−1 + 40 if t ∈ [2; 28]
Absolute Number 31 49 120
Efficiency 1.52 0.71 0.59
CRT score 2.06 1.63 1.27
HL risk aversion 5.35 5.78 5.72
Stories index 2.47 2.33 1.47
Real life index 3.02 2.15 2.35
nit = 100 and nit (t) > nj,t−1 ∀j ∈ [1; 8] and nit (t) > ni,t−1 + 30 if t ∈ [2; 28]
Absolute Number 27 30 143
Efficiency 1.44 0.67 0.66
CRT score 2.00 1.73 1.33
HL risk aversion 5.52 5.57 5.73
Stories index 2.46 1.21 1.85
Real life index 2.85 2.36 2.36

Table 10: Impact of variations of the leadership criterion on the main results.

43
References
Arbak, E., and M.-C. Villeval (2013). Voluntary Leadership: Motivation and Influence, Social
Choice and Welfare 40(3), 635-662.
Bass, B.M. and B.J. Avolio (1994). Improving Organizational Effectiveness Through Transfor-
mational Leadership, Sage Publications, Thousand Oaks, CA.
Bledow, R. and M. Frese (2009). A Situational Judgment Test of Personal Initiative and its
Relationship to Performance, Personnel Psychology 62(2), 229-258.
Brier, G.W. (1950). Verification of Forecasts Expressed in Terms of Probability, Monthly
Weather Review 78(1), 1-3.
Bruttel, L. (2009a). Group Dynamics in Experimental Studies - The Bertrand Paradox Revis-
ited, Journal of Economic Behavior and Organization 69(1), 51-63.
Bruttel, L. (2009b). The Critical Discount Factor as a Measure for Cartel Stability?, Journal
of Economics 96(2), 113-136.
Caliendo, M., D.A. Cobb-Clark and A. Uhlendorff (2010). Locus of Control and Job Search
Strategies, IZA Discussion Paper No. 4750.
Conger, J.A. and R.N. Kanungo (1987). Toward a Behavioral Theory of Charismatic Leadership
in Organizational Settings, Academy of Management Review 12(4), 637-647.
Crant, J.M. and T.S. Bateman (2000). Charismatic Leadership Viewed from Above: The
Impact of Proactive Personality, Journal of Organizational Behavior 21, 63-75.
Datta Mago, S. and E. Dechenaux (2009). Price Leadership and Firm Size Asymmetry: An
Experimental Analysis, Experimental Economics 12, 289-317.
De Hoogh, A.H.B., D.N. Den Hartog, and P.L. Koopman (2005). Linking the Big Five-Factors
of Personality to Charismatic and Transactional Leadership; Perceived Dynamic Work En-
vironment as a Moderator, Journal of Organizational Behavior 26, 839-865.
Dehne, M. and J. Schupp (2007). Persönlichkeitsmerkmale im Sozio-oekonomischen Panel
(SOEP) - Konzept, Umsetzung und empirische Eigenschaften, DIW Research Notes 2007-
26.
Dohmen, T., A. Falk, D. Huffman, U. Sunde, J. Schupp, and G.G. Wagner (2011). Individual
Risk Attitudes: Measurement, Determinants, and Behavioral Consequences, Journal of the
European Economic Association 9(3), 522-550.

44
Drouvelis, M. and D. Nosenzo (2012). Group Identity and Leading-by-Example, CEDEX Dis-
cussion Paper No. 2012-05.
Dudley, K.C. (2003), Empirical Development of a Scale of Patience, Dissertation, West Virginia
University.
Dufwenberg, M. and U. Gneezy (2000). Price Competition and Market Concentration: An
Experimental Study, International Journal of Industrial Organization 18(1), 7-22.
Eagly, A.H., M.C. Johannesen-Schmidt, and M.L. van Engen (2003). Transformational, Trans-
actional, and Laissez-Faire Leadership Styles: A Meta-Analysis Comparing Women and
Men, Psychological Bulletin 129(4), 569-591.
Engelmann, D. and M. Strobel (2004). Inequality Aversion, Efficiency, and Maximin Prefer-
ences in Simple Distribution Experiments. American Economic Review 94(4), 857-869.
Fischbacher, U. (2007). z-Tree: Zurich Toolbox for Ready-made Economic Experiments. Ex-
perimental Economics 10(2), 171-178.
Fodor, E.M. (2010). Power Motivation, in: Schultheiss, O.C., and Brunstein, J.C. (Eds.)
(2010). Implicit Motives, Oxford University Press, New York.
Frederick, S. (2005). Cognitive Reflection and Decision Making, Journal of Economic Perspec-
tives 19(4), 25-42.
Gächter, S., D. Nosenzo, E. Renner, and M. Sefton (2012). Who Makes a Good Leader?
Cooperativeness, Optimism and Leading-by-Example, Economic Inquiry 50(4), 953-967.
Gächter, S. and E. Renner (2006). Leaders as Belief Managers for the Voluntary Provision of
Public Goods, Working Paper.
Greiner, B. (2004). An Online Recruitment System for Economic Experiments. In: Kremer,
K. and V. Macho (Eds.): Forschung und wissenschaftliches Rechnen 2003. GWDG Bericht
63, Göttingen: Gesellschaft für Wiss. Datenverarbeitung, 79-93.
Güth, W., M.V. Levati, M. Sutter, and E. Van Der Heijden (2007). Leading by Example with
and without Exclusion Power in Voluntary Contribution Experiments, Journal of Public
Economics 91(5-6), 1023-1042.
Hermalin, B. (1998). Toward an Economic Theory of Leadership: Leading by Example, Amer-
ican Economic Review 88(5), 1188-1206.
Holt, C.A. and S.K. Laury (2002). Risk Aversion and Incentive Effects, American Economic
Review 92(5), 1644-1655.

45
Judge, T.A. and J.E. Bono (2000). Five-Factor Model of Personality and Tranformational
Leadership, Journal of Applied Psychology 85(5), 751-765.
Kouzes, J.M., and Posner, B.Z. (2007). The Leadership Challenge, Wiley, San Francisco, CA.
Kübler and Müller (2002). Simultaneous and Sequential Price Competition in Heterogeneous
Duopoly Markets: Experimental Evidence, International Journal of Industrial Organization
20, 1437-1460.
Kumru, C.S. and L. Vesterlund (2010). The Effect of Status on Charitable Giving, Journal of
Public Economic Theory 12(4), 709-735.
Lerner, J. and J. Tirole (2002). Some Simple Economics of Open Source, Journal of Industrial
Economics 50(2), 197-234.
Levati, M.V., M. Sutter, and E. van der Heijden (2007). Leading by Example in a Public Goods
Experiment with Heterogeneity and Incomplete Information, Journal of Conflict Resolution
51(5), 793-818.
Leufkens, K. and R. Peeters (2011). Price Dynamics and Collusion Under Short-run Price
Commitments, International Journal of Industrial Organization 29, 134-153.
McClelland, D.C., R. Koestner, and J. Weinberger (1989). How Do Self-Attributed and Implicit
Motives Differ? Psychological Review 96(4), 690702.
Meyer, J. (2006). Eigeninitiative – Eine mikroökonomische und experimentelle Analyse, Diploma
Thesis, University of Zürich.
Moxnes, E. and E. van der Heijden (2007). The Effect of Leadership in a Public Bad Experi-
ment, Journal of Conflict Resolution 47(6), 773-795.
Pang, J.S. (2010). Content Coding Methods in Implicit Motive Assessment: Standards of
Measurement and Best Practices for the Picture Story Exercise, in: Schultheiss, O.C., and
Brunstein, J.C. (Eds.) (2010). Implicit Motives, Oxford University Press, New York.
Piatek, R. and P. Pinger (2010). Maintaining (Locus of) Control? Assessing the Impact of
Locus of Control on Education Decisions and Wages (2010). ZEW - Centre for European
Economic Research Discussion Paper No. 10-093.
Pogrebna, G., D.H. Krantz, C. Schade, and C. Keser (2011). Words Versus Actions as a Means
to Influence Cooperation in Social Dilemma Situations, Theory and Decision 71(4), 473-502.
Potters, J., M. Sefton, and L. Vesterlund (2007). Leading-by-Example and Signaling in Volun-
tary Contribution Games: An Experimental Study, Economic Theory 33(1), 169-182.

46
Potters, J., M. Sefton, and L. Vesterlund (2005). After you - Endogenous Sequencing in Vol-
untary Contribution Games, Journal of Public Economics 89, 1399-1419.
Ross, L., D. Greene, and P. House (1977). The “False Consensus Effect”: An Egocentric Bias
in Social Perception and Attribution Processes, Journal of Experimental Social Psychology
13(3), 279-301.
Rotter, J. (1966). Generalized Expectancies for Internal Versus External Control of Reinforce-
ment, Psychological Monographs, 80.
Rubinstein A. (2007). Instinctive and Cognitive Reasoning: A Study of Response Times. The
Economic Journal 117(523), 1243-1259.
Selten, R. and J. Apesteguia (2005). Experimentally Observed Imitation and Cooperation in
Price Competition on the Circle, Games and Economic Behavior 51, 171-192.
Selten, R. and Stoecker, R. (1986). End Behavior in Sequences of Finite Prisoner’s Dilemma
Supergames, Journal of Economic Behavior and Organization, 7(1), 47-70.
Yukl, G. (2009). Leadership in Organizations, Prentice Hall, Upper Saddle River, New Jersey.
Zizzo, D.J. (2010). Experimenter Demand Effects in Economic Experiments, Experimental
Economics 13(1), 75-98.

47
No. 21 september 2007

Gerald Eisenkopf

Learning and Peer Effects

Research Paper Series

Thurgau Institute of Economics