DO WOMEN SHY AWAY FROM COMPETITION? DO MEN
COMPETE TOO MUCH?*
MURIEL NIEDERLE AND LISE VESTERLUND
We examine whether men and women of the same ability differ in their
selection into a competitive environment. Participants in a laboratory experiment
solve a real task, first under a noncompetitive piece rate and then a competitive
tournament incentive scheme. Although there are no gender differences in performance,
men select the tournament twice as much as women when choosing
their compensation scheme for the next performance. While 73 percent of the men
select the tournament, only 35 percent of the women make this choice. This
gender gap in tournament entry is not explained by performance, and factors such
as risk and feedback aversion only play a negligible role. Instead, the tournamententry
gap is driven by men being more overconfident and by gender differences in
preferences for performing in a competition. The result is that women shy away
from competition and men embrace it.
I. INTRODUCTION
A series of psychology studies suggest that men are more
competitive than women. While boys spend most of their time at
competitive games, girls select activities where there is no winner
and no clear end point. This difference increases through puberty,
and by adulthood more men than women describe themselves as
competitive (see Campbell [2002] for a review of the literature).
The objective of this paper is to investigate whether men and
women differ in their preferences for competition and how such
differences impact economic outcomes. We study whether men
and women differ in the type of compensation scheme they prefer
to receive for their work when holding other job characteristics
constant. Specifically we examine whether for a given performance
level more women than men prefer to work under a noncompetitive
piece rate than under a competitive tournament compensation
scheme.
If women are less likely to compete, this not only reduces the
number of women who enter tournaments, but also those who
win tournaments. Hence, it decreases the chances of women
* We thank Scott Kinross, who conducted all the experiments reported in this
paper, for his excellent research assistance. We thank the editors and the referees
who helped us improve the paper. We also thank Liran Einav, Jean Franc¸ois
Richard, Alvin Roth, and Carmit Segal for comments, and we are grateful to the
NSF for generous support. We thank the Institute for Advanced Study at Princeton
(Niederle) and CEBR at the Copenhagen Business School (Vesterlund) for
their hospitality.
© 2007 by the President and Fellows of Harvard College and the Massachusetts Institute of
Technology.
The Quarterly Journal of Economics, August 2007
1067
succeeding in competitions for promotions and more lucrative
jobs. Bertrand and Hallock [2001] show that in a large data set of
U. S. firms women only account for 2.5 percent of the five highest
paid executives. Ability differences can only explain part of this
occupational difference and it is commonly argued that preferences
and discrimination can account for the remaining difference.
Women may not select into top level jobs because they do
not enjoy the responsibilities associated with a managerial position.
Or they may avoid these jobs because they tend to have long
work hours, which may conflict with the desire or necessity for
child rearing. Second, discrimination or anticipated discrimination
may cause women and men with equal abilities to hold
different occupations.1
Gender differences in preferences for competition
may be an additional explanation for differences in labor
market outcome; in particular, it may help explain the absence of
women in top-level and very competitive positions.2
To study how gender differences in preferences for competition
impact choices of compensation scheme, we want to eliminate
other factors that may cause women to be underrepresented in
competitive jobs. To do so, we use a controlled laboratory experiment
to examine individual choices between competitive and
noncompetitive compensation schemes in a nondiscriminatory
environment. This environment enables us to objectively measure
performance and secures that the time commitment is the same
under both compensation schemes. Attaining measures of performance
under both compensation schemes is crucial for determining
the extent to which choices of compensation scheme are
driven by differences in performance. Prior research suggests
that performance measures are particularly important in this
environment as men and women who perform similarly in noncompetitive
environments can differ in their performance when
they have to compete against one another (see Gneezy, Niederle,
and Rustichini [2003], Gneezy and Rustichini [2004], and Larson
[2005]).
We have groups of two women and two men perform a real
task, namely adding up sets of five two-digit numbers for five
minutes, a task where we expect no gender differences in performance.
Participants first perform the task under a piece-rate
1. See Black and Strahan [2001], Goldin and Rouse [2000], Altonji and Blank
[1999], and references therein.
2. See Blau and Kahn [2004] for a general overview of gender differences in
labor market outcomes.
1068 QUARTERLY JOURNAL OF ECONOMICS
compensation and then under a tournament. While they are
informed of their absolute performance after each task, they do
not receive any feedback on their relative performance. Having
experienced both compensation schemes, participants then
choose which of the two schemes they want to apply to their
performance of the next task, either a piece rate or a tournament.
Despite there being no gender difference in performance under
either compensation scheme, we find that twice as many men
as women select the tournament. While 73 percent of men prefer
the tournament, this choice is only made by 35 percent of the
women. This gender gap persists when we compare the choices of
men and women of equal performance. Compared to payoff-maximizing
choices, low-ability men enter the tournament too much,
and high-ability women do not enter it enough.
We consider a number of possible explanations to understand
what may give rise to such gender differences in tournament
entry. One explanation is simply that preferences for performing
in a competitive environment differ across gender. Other more
general explanations are that women have lower expectations
about their relative ability, are more averse to risk, or are more
reluctant to be in an environment where they receive feedback on
their relative performance. We determine the extent to which
these potential differences can explain the gender gap in tournament
entry.
We find that men are substantially more overconfident about
their relative performance than women and that the beliefs on
relative performance help predict entry decisions. Although gender
differences in overconfidence are found to play an important
role in explaining the gender gap in tournament entry, these
differences only account for a share of the gap.
To assess whether general factors, such as overconfidence,
risk, and feedback aversion by themselves cause a gap in choices
of compensation scheme, we also determine if, absent the thrill or
fear of performing in a competition, a gender gap in choice of
compensation scheme still occurs. We find that combined such
factors cause men and women of equal performance to select
different compensation schemes. This difference appears to be
largely explained by gender differences in overconfidence, while
risk and feedback aversion seem to play a negligible role.
Finally, controlling for gender differences in general factors
such as overconfidence, risk, and feedback aversion, we estimate
the size of the residual gender difference in the tournament-entry
1069GENDER DIFFERENCES IN COMPETITIVENESS
decision. Including these controls, gender differences are still
significant and large. Hence, we conclude that, in addition to
gender differences in overconfidence, a sizeable part of the
gender difference in tournament entry is explained by men and
women having different preferences for performing in a competitive
environment.
We first present a brief discussion of the factors that may
cause women and men to make different choices over compensation
schemes. We then present our experimental design. The
empirical results are presented in Sections IV and V. In Section
IV we determine if, conditional on performance, the choices of
compensation scheme of women and men differ; then in Section V
we consider alternative explanations for such differences. We
report only the most important of our results and refer the interested
reader to Niederle and Vesterlund [2005] for a more extensive
analysis of the data. Finally, Section VI concludes and discusses
the results in connection to the existing literature.
II. THEORY
To determine what may cause women and men of equal
ability to differ in their propensity to enter a competitive environment
we consider four different explanations.
Explanation 1: Men enter the tournament more than women
because they like to compete. Women may be more reluctant to
enter a competitive environment simply because they dislike
performing when they are competing against others. While the
prospect of engaging in a future competition may cause women to
anticipate a psychic cost and deter them from tournaments, men
may anticipate a psychic benefit and instead be drawn to them.3
Nurture as well as nature may cause women to be relatively
more reluctant to perform in a competition. First, we tend to raise
girls and boys differently. Parents, teachers, and peers encourage
gender-typed activities in children while cross-gender activities
are discouraged. While boys are encouraged to be assertive, girls
are encouraged to show empathy and be egalitarian [Ruble, Martin,
and Berenbaum 2006]. Second, nature may also cause a
gender difference in preferences for competition. Evolutionary
psychology proposes two theories that suggest that men have
3. While “psychic” costs and benefits of a tournament may affect entry, they
need not affect tournament performance.
1070 QUARTERLY JOURNAL OF ECONOMICS
evolved to enjoy competition. Both of these are tied to the reproductive
strategies of the two sexes. One argues that since men can
have many more children than women the potential gain in
reproductive success from winning a competition is much greater
for men, and men have therefore evolved to be more competitive
than women [Daly and Wilson 1983]. The second theory focuses
on one gender being responsible for parental care. While a man’s
death does not influence his current reproductive success, a woman’s
death may cause the loss of her current offspring [Campbell
2002]. Thus differences in potential losses as well as potential
gains from competition may make males more eager to compete.
In addition to suggesting that men hold a stronger preference
for competition these evolutionary explanations are also used to
explain why men often are more confident in their relative performance
and less averse to risk. Such gender differences may
also influence tournament-entry decisions.
Explanation 2: Men enter the tournament more than women
because they are more overconfident. Psychologists often find that
while both men and women are overconfident about their relative
performance, men tend to be more overconfident than women
(e.g., Lichtenstein, Fischhoff, and Phillips [1982], Beyer [1990],
and Beyer and Bowden [1997]). Consistent with a greater male
overconfidence, Barber and Odean [2001] show that in financial
markets men trade more excessively than women. If in our experiment
men are more overconfident about their relative performance,
then the probability of selecting the competition is expected
to be larger for a man than a woman with the same
performance.
Note however that overconfidence as well as gender differences
in overconfidence are task dependent. Studies have found
that overconfidence is sensitive to how easy a task is [Moore and
Small 2004], and the gender difference in overconfidence has
primarily been found in masculine tasks. For example, Lundeberg,
Fox, and Puncochar [1994] argue that the reason why some
studies do not find gender differences in confidence on general
knowledge is because it is not in the masculine domain. Thus,
depending on the perception of our addition task, gender differences
in overconfidence may or may not help explain potential
gender differences in tournament entry.
Explanation 3: Men enter the tournament more than women
because they are less risk averse. As tournaments involve uncertain
payoffs, potential gender differences in risk attitudes are
1071GENDER DIFFERENCES IN COMPETITIVENESS
likely to also affect the choice of compensation scheme. Studies
examining gender differences in risk attitudes over monetary
gambles find either that women are more risk averse than men or
that there is no gender difference. Eckel and Grossman [2002a]
summarize the experimental literature in economics and conclude
that women exhibit greater risk aversion in choices. A
summary of the psychology literature is presented by Byrnes,
Miller, and Shafer [1999]. They provide a meta-analysis of 150
risk experiments and demonstrate that while women in some
situations are significantly more averse to risk, many studies find
no gender difference.
Explanation 4: Men enter the tournament more than women
because they are less averse to feedback. One consequence of
entering the tournament is that the individual will receive feedback
on relative performance. The psychology literature suggests
that men and women may respond differently to such feedback.
First, there is evidence that women tend to incorporate negative
feedback more than men (see, e.g., Roberts and Nolen-Hoeksema
[1989]). Second, women, more than men, may view a negative
signal as indicative of their self-worth rather than simply their
one-time performance on a task. Women may therefore fall into
“confidence traps” from which they do not recover easily (see, e.g.,
Dweck [2000] and references therein). If participants benefit from
holding positive beliefs about themselves, then both of these
factors may cause women to avoid environments where they
receive feedback on relative performance.
Our experiment is designed to shed light on the role played
by these alternative explanations. Of particular interest is
whether a gender difference in tournament entry is explained by
general factors such as overconfidence, risk, and feedback aversion
(Explanations 2–4) or if part of such a difference is accounted
for by preference differences for performing in a competition
(Explanation 1). What distinguishes Explanation 1 from the other
three is that it relies critically on the tournament-entry decision
resulting in a subsequent competitive performance. The other
explanations are more general and should be present in other
decisions as well. To jointly determine the role played by these
three general factors we consider an environment that is as close
as possible to the tournament-entry decision, without involving
an actual tournament performance. Specifically, we ask participants
to choose between a competitive and noncompetitive compensation
scheme for a past noncompetitive performance; that is,
1072 QUARTERLY JOURNAL OF ECONOMICS
the choice of tournament does not require participants to subsequently
perform in a competition. While the potential thrill, anxiety,
or fear of performing in a competition is absent from this
choice, this decision will show whether general factors such as
overconfidence, risk, and feedback aversion in and of themselves
can cause a difference in choices of compensation scheme.
To determine whether a gender difference in preferences for
competition (Explanation 1) plays a role when controlling for
overconfidence, risk, and feedback aversion, we use the choice of
compensation scheme for past performance along with the participants’
beliefs on their relative performance ranking as controls
in the tournament-entry decision.
III. EXPERIMENTAL DESIGN
We conduct an experiment in which participants solve a real
task, first under a noncompetitive piece-rate scheme and then
under a competitive tournament scheme. Participants are then
asked to select which of these two compensation schemes they
want to apply to their next performance. This provides participants
with experience of both compensation forms and enables us
to determine if men and women of equal performance choose the
same compensation scheme.
The task of our experiment is to add up sets of five two-digit
numbers. Participants are not allowed to use a calculator, but
may use scratch paper. The numbers are randomly drawn and
each problem is presented in the following way, where participants
fill in the sum in the blank box:
Once the participant submits an answer on the computer, a
new problem appears jointly with information on whether the
former answer was correct.4
A record of the number of correct and
wrong answers is kept on the screen. Participants have five
minutes in which they may solve as many problems as they can.
4. The program was written using the software zTree [Fischbacher 2007].
21 35 48 29 83
1073GENDER DIFFERENCES IN COMPETITIVENESS
The final score is determined by the number of correctly solved
problems. We selected this five-minute addition task because it
requires both skill and effort and because research suggests that
there are no gender differences in ability on easy math tests.5
This will enable us to rule out performance differences as an
explanation for gender differences in tournament entry.
The experiment was conducted at the University of Pittsburgh,
using standard recruiting procedures and the subject pool
at the Pittsburgh Experimental Economics Laboratory (PEEL).
Two or three groups of four participants participated in each
session. Participants were seated in rows and informed that they
were grouped with the other people in their row. A group consisted
of two women and two men. Although gender was not
discussed at any time, participants could see the other people in
their group and determine their gender. A total of twenty groups
participated in the experiment (forty men and forty women).
Each participant received a $5 show-up fee, and an additional
$7 for completing the experiment. Participants were told
that they would be asked to complete four tasks and that one of
these tasks would be randomly chosen for payment at the end of
the experiment. By paying only for one task, we diminish the
chance that decisions in a given task may be used to hedge
against outcomes in other tasks. Participants were informed of
the nature of the tasks only immediately before performing the
task. While they knew their absolute performance on a task, i.e.,
how many problems they solved correctly, they were not informed
of their relative performance until the end of the experiment and
did not know if they performed better or worse than the other
participants in their group. The specific compensation schemes
and order of tasks were as follows.
Task 1—Piece Rate: Participants are given the five-minute
addition task. If Task 1 is randomly selected for payment, they
receive 50 cents per correct answer.
Task 2—Tournament: Participants are given the five-minute
addition task. If Task 2 is randomly selected for payment, the
participant who solves the largest number of correct problems in
the group receives $2 per correct answer while the other partic-
5. While males often score better on abstract math problems, there is no
gender difference in arithmetic or algebra performance. Women tend to score
better than men on computational problems (see Hyde, Fennema, and Lamon
[1990] for a meta-analysis of 100 studies on gender differences in math
performance).
1074 QUARTERLY JOURNAL OF ECONOMICS
ipants receive no payment (in case of ties the winner is chosen
randomly among the high scorers).
The tournament is designed so that for a given performance
a participant with a 25 percent chance of winning the tournament
receives the same expected payoff from the tournament as from
the piece rate.6
In the third task participants once again perform
the five-minute addition task but this time select which of the two
compensation schemes they want to apply to their future performance,
a piece rate or a tournament.
Task 3—Choice of Compensation Scheme for Future Performance:
Before performing the five-minute addition task, participants
select whether they want to be paid according to a piece
rate, i.e., 50 cents for each correct answer, or a tournament. When
the participant chooses the tournament she receives $2 per correct
answer if her score in Task 3 exceeds that of the other group
members in the Task-2 tournament they just completed; otherwise
she receives no payment (in case of ties the winner is chosen
randomly).
Winners of the Task-3 tournament are determined based on
the comparison relative to the other group members’ Task-2 and
not their Task-3 performance. One can think of this as competing
against other participants who already performed.7
This has several
advantages; first, the performance of a player who enters the
tournament is evaluated against the performance of participants
who also performed under a tournament compensation. Second,
while beliefs regarding relative performance in a tournament
may affect the decision to enter the tournament, beliefs regarding
the choices of others will not. Thus, we avoid a potential source of
error through biased beliefs about other participants’ choices.8
Furthermore, since a participant’s choice does not affect the payment
of any other participant we can rule out the possibility that
women may shy away from competition because by winning the
6. By paying the tournament winner per correct problem we avoid the problem
of choosing a high enough fixed prize to ensure that even high-performing
participants benefit from entering the tournament.
7. Many sports competitions are not performed simultaneously, e.g., downhill
skiing.
8. For example, the odds of winning a simultaneous competition would be
greatly changed if men believed that women would not enter the tournament,
causing them to face only one rather than three competitors. Note that our design
allows for the possibility that there is no winner among participants who choose
the tournament (if none of those entering the tournament beat the high score of
their opponents). Conversely, all participants can win the tournament if everyone
increases their performance beyond the highest Task-2 performance in that
group.
1075GENDER DIFFERENCES IN COMPETITIVENESS
tournament they impose a negative externality on others.9
Effectively
in Task 3 participants face an individual decision problem,
which depends only on their ability to beat the Task-2 performance
of others and their preference for performing in a
tournament.
To determine whether the gender gap in tournament entry is
caused by gender differences in preferences for performing in a
competitive environment, or if it is accounted for by general
factors such as differences in overconfidence, risk, or feedback
aversion, we present participants with one last task. Here participants
face a choice similar to that of Task 3 but without using
a tournament performance and having to subsequently perform
in a tournament.
Task 4—Choice of Compensation Scheme for Past Piece-Rate
Performance: Participants do not have to perform in this task.
Rather, if this task is randomly selected for payment, their compensation
depends on the number of correct answers they provided
in the Task-1 piece rate. Participants choose which compensation
scheme they want to apply to their past piece-rate
performance: a 50 cent piece rate or a tournament. They win the
tournament and receive $2 per correct answer if their Task-1
piece-rate performance is the highest of the participants in their
group; otherwise they receive no payment (in case of ties the
winner is chosen randomly). Before making their choice, participants
are reminded of their Task-1 piece-rate performance.
As in the Task-3 choice, a participant’s decision does not
affect the earnings of any other participant, nor does it depend on
the entry decisions of others. Thus Task 4 is also an individualdecision
task. This final task allows us to see whether gender
differences in choice of compensation scheme appear even when
no future and past tournament performance is involved. That is,
we can determine whether general factors such as overconfidence,
risk, and feedback aversion (Explanations 2–4) by themselves
cause a gap in tournament entry. While these effects can influence
the Task-3 choice, they are not unique to the decision of
performing in a competition; in particular, they are likely to affect
the Task-4 choice as well.
9. For a discussion on possible gender differences in altruism, see, e.g.,
Andreoni and Vesterlund [2001]. See Ledyard [1995] for gender differences in
social dilemma and public good games, as well as Eckel and Grossman [2002b]
and Croson and Gneezy [2004] for a review of gender differences in experimental
settings.
1076 QUARTERLY JOURNAL OF ECONOMICS
Finally, we elicit the participants’ beliefs on their relative
performance to determine their relation to choices of compensation
scheme. We elicit these beliefs both for performances in Task
1 and Task 2, where all participants had the same incentive
scheme. These will help us determine whether gender differences
in overconfidence about tournament performance affect the decision
to enter a tournament, i.e., we can assess the role Explanation
2 plays in closing the gender gap.
Belief-Assessment Questions: At the end of the experiment
participants are asked to guess their rank in the Task-1 piece rate
and the Task-2 tournament. Each participant picks a rank between
1 and 4 and is paid $1 for each correct guess.10
To determine whether gender differences in preferences for
competition (Explanation 1) cause a gender gap in the Task-3
tournament entry we include the elicited beliefs on tournament
ranking and the Task-4 choice of compensation scheme to control
for general factors such as confidence, risk, and feedback
aversion.
At the end of the experiment, a number from one to four is
drawn to determine which of the four tasks is selected for earnings.
The experiment lasted about forty-five minutes, and participants
earned on average $19.80.
IV. BASIC EXPERIMENTAL RESULTS
In this section we examine whether, conditional on ability,
women and men differ in their preference for performing under a
piece rate versus a tournament scheme. To eliminate ability
differences as an explanation for potential gender differences in
tournament entry, we selected a task for which we anticipated
that women and men would have similar performances under the
two compensation schemes. We start by confirming that we succeeded
in selecting such a task. We then examine the participants’
compensation scheme choices and determine whether they
differ conditional on performance.
10. In case of ties in the actual ranks, we counted every answer that could be
correct as correct. For example, if the performance in the group was 10, 10, 11, 11,
then an answer of last and third was correct for a score of 10, and an answer of
best and second was correct for a score of 11.
1077GENDER DIFFERENCES IN COMPETITIVENESS
IV.A. Performance in the Piece Rate and the Tournament
As expected we find no gender difference in performance
under the piece rate or under the tournament. In the piece
rate, the average number of problems solved is 10.15 for
women and 10.68 for men. Using a two-sided t-test, this difference
is not significant (p ϭ .459). The gender difference in
performance is also not significant in the tournament where,
on average, women correctly solve 11.8 problems, and men 12.1
(p ϭ .643). Throughout the paper the reported test statistics
refer to a two-sided t-test, unless otherwise noted. The conclusions
of the reported t-test do not differ from those of a Mann–
Whitney test.
The cumulative distributions for the number of correct answers
in the piece rate (Task 1) and the tournament (Task 2) are
shown in Panels A and B of Figure I, respectively. For every
performance level the graphs show the proportion of women or
men who solved that many or fewer correct problems. In both
tasks the performance distributions are very similar for women
and men.
Although the piece rate and tournament performances are
highly correlated (Spearman rank correlations of .69 for
women and .61 for men), both genders perform significantly
better under the tournament than the piece rate (one-sided p Ͻ
.01 for each gender separately). This improvement may be
caused by learning or by the different performance incentives
FIGURE I
CDF of Number of Correctly Solved Problems
Panel A: Piece rate (Task 1); Panel B: Tournament (Task 2)
1078 QUARTERLY JOURNAL OF ECONOMICS
under the tournament.11
The increase in performance varies
substantially across participants. While noise may be one explanation
for this variance, another may be that some participants
are more competitive than others. The increase in performance
from the piece rate to the tournament, however, does
not differ by gender (p ϭ .673).
The similar performances of men and women result in
there being no gender difference in the probability of winning
the Task-2 tournament. Of the twenty Task-2 tournaments,
eleven were won by women and nine by men. To assess the
probability of winning the tournament we randomly create
four-person groups from the observed performance distributions.
Conditioning only on gender, the probability of winning
the tournament is 26 percent for a man and 24 percent for a
woman; in a sample of forty men and forty women this difference
is not significant (p ϭ .836). Similarly there are no gender
differences when we instead examine the probability of winning
the tournament conditional on performance. For both men
and women who solve thirteen problems the chance of winning
the tournament is 26.6 percent. If instead they solve fourteen
problems, the probability of winning increases to 47.8 percent
for women and 47.7 percent for men. The change in the probability
of winning is quite dramatic with the chance being less
than 2 percent for those solving ten problems and more than 70
percent for those solving fifteen problems.12
Whether we use a piece-rate or tournament compensation
scheme, our task is one for which there appears to be no gender
differences in performance. After completing the first two
tasks, women and men have therefore had similar experiences,
and based on performance alone, we would not expect a gender
difference in the subsequent Task-3 choice of compensation
scheme.
11. DellaVigna, Malmendier, and Vesterlund [in preparation] have participants
perform six rounds of three-minute tournaments and find a significant
increase in performance from round one to round two but no significant increase
in performance in subsequent rounds. This suggests that initial learning may
have some effect.
12. For any given performance level, say fifteen for a woman, we draw 10,000
groups consisting of two men and one other woman, where we use the sample of
forty men and women with replacement. We then calculate the frequency of wins.
The exercise is repeated 100 times, and we report the average of these win
frequencies. For more details on the probability of winning the tournament for a
given performance, see Niederle and Vesterlund [2005].
1079GENDER DIFFERENCES IN COMPETITIVENESS
IV.B. Gender Differences in Tournament Entry (Task-3 Choice)
Having experienced both the 50-cent piece rate and the $2
tournament, participants are asked which of the two they want to
apply to their Task-3 performance. A participant who chooses the
tournament wins the tournament if his or her number of correct
answers in Task 3 exceeds the number of correct answers by the
other three members in the group in the Task-2 tournament.
Thus, choosing the tournament depends on beliefs regarding own
ability and the other players’ past tournament performance, but
it does not depend on beliefs about the choice of compensation
scheme of other participants.
For a given performance level a risk-neutral participant who
only aims to maximize monetary earnings is indifferent between
the two incentive schemes when the chance of winning the tournament
is 25 percent. According to our analysis above, all players
with a given performance of fourteen or more have higher expected
monetary earnings from the tournament. If the participant’s
Task-3 performance is exactly like the Task-2 performance,
this corresponds to 30 percent of the women and 30
percent of the men. When we include participants who solve
thirteen problems—and are virtually indifferent between the two
schemes—the percentages are 40 percent for women and 45 percent
for men.
Despite the similar performances of women and men, their
choices of compensation scheme are very different. While the
majority of women prefer the piece rate, the majority of men
prefer the tournament. Specifically, we find that 35 percent of
women and 73 percent of men select the tournament. This
observed gender gap in tournament entry is both substantial
and significant (a Fisher’s exact test yields p ϭ .002).
IV.C. Tournament-Entry Decisions Conditional on Performance
To examine how performance affects the propensity of
women and men to enter a tournament, we first compare the
mean past performance characteristics of participants by choice
of compensation scheme. Table I reports, by gender and the
chosen compensation scheme, three performance measures; the
average number of problems solved correctly under piece rate
(Task 1) and tournament (Task 2), as well as the average increase
in performance between the two.
For women there is no significant difference in perfor-
1080 QUARTERLY JOURNAL OF ECONOMICS
mance between those who do and do not enter the tournament
(p Ն .35 for each of the three performance measures). For men,
only the tournament performance is marginally higher for
those who enter the tournament (p ϭ .14 for the Task-2 tournament).
Conditional on the choice of compensation scheme,
there is, however, no gender difference in Task-1 and Task-2
performance or in the increase between the two (p Ն .28 for
each of the six tests).
A probit regression reveals that while the participant’s performance
under the two compensation schemes does not significantly
affect the decision to enter the tournament, the participant’s
gender does. The reported marginal gender effect of Ϫ.380
in Table II shows that a man with a performance of thirteen in
TABLE I
PERFORMANCE CHARACTERISTICS BY CHOICE OF COMPENSATION SCHEME (TASK 3)
Compensation scheme
Average performance
Piece rate Tournament
Tournament–
piece rate
Women Piece rate 10.35 11.77 1.42
(0.61) (0.67) (0.47)
Tournament 9.79 11.93 2.14
(0.58) (0.63) (0.54)
Men Piece rate 9.91 11.09 1.18
(0.84) (0.85) (0.60)
Tournament 10.97 12.52 1.55
(0.69) (0.48) (0.49)
Averages with standard errors in parentheses. Sample is forty women and forty men.
TABLE II
PROBIT OF TOURNAMENT CHOICE IN TASK 3
Coefficient p-value
Female Ϫ.380 .00
Tournament .015 .41
Tournament–piece rate .015 .50
Dependent variable: Task-3 choice of compensation scheme (1-tournament and 0-piece rate). Tournament
refers to Task-2 performance, tournament–piece rate to the change in performance between Task 2 and
Task 1. The table presents marginal effects of the coefficient evaluated at a man with thirteen correct answers
in the tournament and twelve in the piece rate. Sample is forty women and forty men.
1081GENDER DIFFERENCES IN COMPETITIVENESS
the tournament (and twelve in the piece rate) would have a 38
percentage point lower probability of entering the tournament if he
were a woman. Thus, controlling for past performance, women are
much less likely to select a competitive-compensation scheme.13
A possible explanation for the observed gender difference in
choice of compensation scheme may be that there is a gender
difference in performance following the choice—and that our participants
correctly anticipate such a difference. However, this
does not appear to be the case, as the results from the Task-3
performance parallel those of the tournament performance. Conditional
on gender, the performance in Task 3 does not differ
between those who do and do not enter the tournament (p Ն .288).
Similarly, the participants who enter the tournament do not have
a significantly different increase in performance in the choice
task (Task 3) relative to the former (Task 2) tournament (p Ն .88).
Thus, not only is it not true that only participants with a high
past performance enter the tournament, it is also not true that
those who entered the tournament performed better than those
who did not. We find that performance in Task 3 cannot explain
the gender gap in tournament entry just like the performance
before the tournament-entry decision.14
Figure II shows the proportion of women and men who enter
the tournament conditional on their performance quartile in Task
2 (Panel A) and Task 3 (Panel B). In both cases, performance has
at most a small effect on tournament entry, and for every performance
level, men are more likely to enter the tournament. Furthermore,
in each case, we see that even women in the highest
performance quartile have a lower propensity to enter the tournament
than men in the lowest performance quartile.
Among participants who for a given performance have higher
expected earnings in the tournament than the piece rate (i.e.,
those solving thirteen or more problems) significantly more men
than women enter the tournament (a two-sided Fisher’ exact test
yields p ϭ .004 and .015 for Tasks 2 and 3, respectively). Similarly
13. Probit regressions show the p-value associated with the coefficient. However,
to ease interpretation, we do not show this overall coefficient, but rather the
marginal effect at a specific point. This evaluation point is selected because a
risk-neutral individual solving thirteen problems in the tournament is indifferent
toward entering the tournament. The average piece-rate performance for this
group was twelve.
14. A probit analysis of the tournament-entry decision yields marginal effects
Ϫ.357 on female (p ϭ .00) and .015 on Task-3 performance (p ϭ .31) evaluated at
a man with thirteen correct answers in Task 3.
1082 QUARTERLY JOURNAL OF ECONOMICS
men are more likely to enter the tournament among participants
whose expected earnings are lower in the tournament (a twosided
Fishers exact test yields p ϭ .15 and p ϭ .05 for Tasks 2 and
3, respectively). Thus, whether we use the Task-2 or Task-3
performances, from a payment-maximizing perspective, low-performing
men enter the tournament too often, and high-performing
women enter it too rarely.
Before evaluating possible explanations for gender differences
in tournament entry, it is worth considering the magnitude of the
gap we are trying to explain. For example, it is easily seen that a
gender difference in risk aversion alone is an unlikely explanation
for the observed tournament-entry gap. Consider participants with
fourteen or more correct answers in the Task-2 tournament who
have a 47 percent or higher chance of winning the tournament.
Ignoring performance costs and presuming that one knows the performance
distribution and maintains the exact same performance in
Task 3, the decision to enter the tournament is a gamble of receiving,
per correct answer, either $2 with a probability of 47 percent (or
more), or receiving 50 cents for sure. For participants who have
fourteen correct answers, that means a gamble of a 47 percent
chance of $28 (i.e., an expected value of $13) versus a sure gain of $7.
Of the participants who solve fourteen problems or more, eight out
of twelve of the women and three out of twelve of the men do not take
this or a better gamble. This difference is marginally significant with
a two-sided Fisher’s exact test (p ϭ .100). Similarly, for participants
who have eleven or fewer correct answers, the chance of winning the
tournament is 5.6 percent or less. Thus, entering the tournament
FIGURE II
Proportion of Participants Selecting Tournament for Task 3 Conditional on
Task-2 Tournament Performance Quartile (Panel A) and Task-3 Choice
Performance Quartile (Panel B)
1083GENDER DIFFERENCES IN COMPETITIVENESS
means receiving $2 per correct answer with a probability of 5.6
percent (or less) versus receiving 50 cents for sure. For participants
who solve eleven correct answers, this is a choice between
a 5.6 percent chance of winning $22 (i.e., an expected
value of $1.23) compared to receiving $5.5 for sure. Of the
participants, who solve eleven problems or less, eleven out of
eighteen of the men and only five out of seventeen of the women
take this or a worse gamble. This difference is marginally
significant with a two-sided Fisher’s exact test (p ϭ .092). To
explain these choices, women would have to be exceptionally
risk averse and men exceptionally risk seeking. We are not
aware of any studies that find such extreme gender differences
in risk attitudes.
To further assess the magnitude of the gap in tournament
entry, we determine the costs associated with payoff-inferior
choices of compensation scheme. To do so, we ignore performance
costs (which we cannot measure) and assume that performance
is independent of the chosen compensation scheme.
The expected costs of over- and under-entry into the tournament
are the difference between the potential earnings under
the two incentive schemes. We calculate the expected monetary
losses using either performance in Task 2 (as earlier) or in
Task 3. Table III, columns (1) and (2) report the costs for
women and men using the Task-2 performance as a predictor of
future performance. We can think of this as reporting the
ex-ante costs. Columns (3) and (4) report instead the costs
based on the actual Task-3 performance; this corresponds to
the ex-post costs. In each case we also report the number of
people who, for a given performance, in expectation would have
been better off making a different choice.
While the magnitude of the costs is sensitive to the precise
assumptions we make, the qualitative results are not. More
women than men fail to enter when they should, and more men
than women enter when they should not. The total cost of
under-entry is higher for women, while the cost of over-entry is
higher for men. Since over-entry occurs for participants of low
performance and under-entry for those with high performance,
by design the cost of under-entry is higher than that of overentry.
So, although the number of men and women who make
payoff-inferior decision is the same, the total costs of doing so
are higher for women than for men.
1084 QUARTERLY JOURNAL OF ECONOMICS
V. EXPLANATIONS FOR THE GENDER GAP IN TOURNAMENT ENTRY
Our results thus far show that men and women with similar
performances differ substantially in their tournament-entry decisions.
While women shy away from competition, men are drawn to
it. From a payoff-maximizing perspective, high-performing women
enter the tournament too rarely, and low-performing men enter the
tournament too often. In this section we try to determine the causes
for these differences in tournament entry. We start in Section V.A by
examining the possibility that greater overconfidence of men can
cause a gender gap in tournament entry.
We then examine the broader set of explanations from Section
II. Specifically, we use Task 4 to distinguish between the role
played by gender differences in preferences for performing in a
competition, and the more general explanations such as gender
differences in overconfidence and risk and feedback aversion. In
Task 4 participants choose between a competitive and a noncompetitive
compensation scheme for their past Task-1 piece-rate
TABLE III
EXPECTED COSTS OF OVER- AND UNDER-ENTRY IN TASK-3 TOURNAMENT
Calculation
based on Task-2
performance
Calculation
based on Task-3
performance
Women Men Women Men
Under-entry
Number who should enter 12 12 9 20
Of those how many do not enter 8 3 6 4
Expected total cost of under-entry 99.4 34.5 84.6 49.6
Average expected cost of underentry
12.4 11.5 14.1 16.5
Over-entry
Number who should not enter 24 22 24 19
Of those how many do enter 9 14 8 12
Expected total cost of over-entry 32.9 56.5 28.9 43.8
Average expected cost of overentry
3.7 4.0 3.6 3.6
Total expected costs 132.3 91.0 113.5 93.3
Participants solving fourteen or more problems should enter the tournament, and those with twelve and
fewer problems should select the piece rate. Participants with thirteen problems (who are virtually indifferent
between the two compensation schemes) are not included in the analysis.
1085GENDER DIFFERENCES IN COMPETITIVENESS
performance. Although this choice is very similar to that of Task
3, it eliminates the prospect of having to subsequently perform in
a competition. Thus, while general factors can influence the compensation
choices in Tasks 3 and 4, only in Task 3 can preference
differences for performing in a competition play a role.
In Section V.B we use Task 4 to simultaneously assess whether
gender differences in general factors such as confidence and risk and
feedback aversion by themselves cause differences in compensation
scheme choices. In Section V.C we then determine if the act of
performing in a competition creates a gap in tournament entry that
cannot be explained by these general factors. To do so we use the
elicited beliefs as well as the Task-4 decision as controls in the
Task-3 tournament-entry decision. This helps us determine whether
an explanation for the tournament-entry gap may be that women,
relative to men, are more averse to choices that require a future
performance in a competitive environment.
V.A. Does Greater Male Confidence about Relative Performance
Explain the Tournament-Entry Gap?
To elicit participants’ beliefs on their relative tournament
performance we asked them at the end of the experiment to guess
how their performance in Task 2 ranked relative to the other
members of their group. Participants received $1 if their guess
was correct, and in the event of a tie they were compensated for
any guess that could be deemed correct.15
We first examine whether men and women of equal performance
have different beliefs about their relative performance. We
then investigate whether conditional on these beliefs there is a
gender gap in tournament entry. That is, we determine the extent
to which gender differences in confidence can account for the
gender gap in tournament entry.
Given the absence of a gender difference in performance, the
distributions of relative performance ranks within actual groups
as well as expected rank within randomly formed groups are the
same for men and women. Accounting for rewards in the event of
ties, this implies that participants who know the performance
distributions of men and women will maximize their payoffs by
15. While the payment for the guessed rank is not very high, it still offers
participants the opportunity of using their guess as a method of hedging against
their tournament-entry decision. The strong positive correlation between elicited
ranks and tournament entry (Figure III) suggests that hedging was not a dominant
motive.
1086 QUARTERLY JOURNAL OF ECONOMICS
guessing that they ranked second or third.16
However, participants
believe that they are ranked substantially better than that.
Table IV shows the participants’ believed rank distributions and
the number of incorrect guesses.
Relative to their actual rank, both men and women are overconfident.
A Fisher’s exact test of independence between the
distribution of guessed rank and actual rank yields p ϭ .00 for
both men and women. However, men are more overconfident
about their relative performance than women. While 75 percent
of the men think they are best in their group of four, only 43
percent of the women hold this belief. The guesses of women and
men differ significantly from one another, a Fisher’s exact test of
independence of the distributions for men and women delivers
p ϭ .016.
An ordered probit of the guessed rank as a function of a
female dummy and performance shows that, conditional on performance,
women are significantly less confident about their relative
ranking than men and that participants with a higher
tournament performance think they have higher relative performance
(see Table V).17
Can the greater overconfidence by men explain why, conditional
on performance, they enter the tournament more fre-
16. Based on 10,000 artificially generated groups, the likelihood of a woman
being ranked first is .223, second .261, third .262, and last .255; the corresponding
probabilities for a man for first is .243, second .288, third .278, and last .199. For
more details, see Niederle and Vesterlund [2005].
17. We eliminate guessed ranks of four, as we have only one man and two
women with such guesses. The results are similar when we code guesses of three
and four as guesses of rank three. Furthermore, the results are also similar when
we include the guesses of four.
TABLE IV
DISTRIBUTION OF GUESSED TOURNAMENT RANK
Men Women
Guessed rank Incorrect guess Guessed rank Incorrect guess
1: Best 30 22 17 9
2 5 3 15 10
3 4 2 6 5
4: Worst 1 1 2 1
Total 40 28 40 25
1087GENDER DIFFERENCES IN COMPETITIVENESS
quently than women? Figure III shows for each guessed rank the
proportion of women and men who enter the tournament.18
While
tournament entry-decisions are positively correlated with the
participants’ beliefs on relative performance, there are still substantial
gender differences. Looking at the more than 80 percent
of participants who think they are first or second best in their
group, there is a gender gap in tournament entry of about 30
percentage points.
The probit regression in column (2) of Table VI shows that
conditional on actual performance, participants who are more
confident about their relative tournament performance are more
likely to enter the tournament.19
Furthermore, women remain
significantly less likely to enter the tournament when controlling
for both absolute and believed relative performance.
How important are gender differences in overconfidence in
explaining the gender gap in tournament entry? Evaluated at a
man who solves thirteen problems in the tournament and twelve
in the piece rate, we previously found that controlling only for
performance, the gender effect was 38 percentage points (column
(1)). Column (2) shows that including a control for guessed tournament
rank, the gender effect reduces to 28 percentage points.
That is, about 27 percent of the gender gap in tournament entry
can be attributed to men being more overconfident than women,
with a remaining 73 percent of the overall gender effect being
unaccounted for.
18. Note that a participant with a point prediction of a guessed rank of 2 may
still optimally choose to enter the tournament if, for example, the participant
believes that she has a 40 percent chance of being best, and a 60 percent chance
of being second.
19. The two performance measures are included in the regression because we
are interested in examining gender differences in tournament entry conditional on
performance. The results in column (1) correspond to those of Table III with the
exception that guesses of four are not included.
TABLE V
ORDERED PROBIT OF GUESSED TOURNAMENT RANK
Coefficient Standard error p-value
Female .75 0.30 .01
Tournament Ϫ.19 0.06 .00
Tournament–piece rate Ϫ.08 0.07 .27
Ordered probit of guessed rank for guesses of ranks 1, 2, and 3. Sample is thirty-eight women and
thirty-nine men.
1088 QUARTERLY JOURNAL OF ECONOMICS
While greater male overconfidence helps explain why equally
able women and men select different compensation schemes, the
majority of the gender gap remains.
V.B. Do General Factors Cause Gender Differences in Choice of
Compensation Scheme?
To better understand the gender gap in tournament entry we
need to consider the other explanations we proposed in Section II.
We start by determining the effect general factors such as overconfidence
and risk and feedback aversion have on the tournaFIGURE
III
Proportion Selecting the Tournament for Task 3 Conditional on Guessed Rank
TABLE VI
PROBIT OF TOURNAMENT-ENTRY DECISION (TASK 3)
Coefficient (p-value)
(1) (2)
Female Ϫ.379 Ϫ.278
(.01) (.01)
Tournament .015 Ϫ.002
(.39) (.90)
Tournament–piece rate .008 Ϫ.001
(.72) (.94)
Guessed tournament rank Ϫ.181
(.01)
Dependent variable: Task-3 choice of compensation scheme (1-tournament and 0-piece rate). The table
presents marginal effects evaluated at a man with a guess of one and who has thirteen correct answers in the
tournament and twelve in the piece rate. Guesses of four are eliminated, resulting in a sample of thirty-eight
women and thirty-nine men.
1089GENDER DIFFERENCES IN COMPETITIVENESS
ment-entry decision. We use Task 4 to examine whether a gender
gap in compensation scheme choices still is observed when the
tournament choice does not require a subsequent competitive
performance. Participants in Task 4 select one of two compensation
schemes for their past piece-rate performance (Task 1), either
the 50-cent piece rate or the $2 tournament. If the tournament
is chosen, the piece-rate performance is submitted to a
competition against the piece-rate performances of the other participants
in the group (independent of their choice of compensation
scheme). A tournament is won if an individual’s performance
exceeds that of the other three players.
Before examining the participants’ choices, we use the
Task-1 performance to determine the performance level at which
participants have higher monetary earnings from submitting the
piece rate to a tournament scheme. In the piece rate men and
women have similar but not exactly the same probability of being
the highest performer in a randomly drawn group of two men and
two women. While the chance of having the highest piece-rate
performance is 29 percent for a man, it is 21 percent for a woman.20
In our twenty groups, eleven women and eleven men were the
highest performers in their group (including two cases of ties).
The gender differences are also small when we examine for each
gender the probability of winning the tournament conditional on
performance. While women who solve eleven problems have a
21.6 percent chance of wining, the chance for men is 24.4 percent.
When solving twelve problems, the chances are 33 and 39.3
percent for women and men respectively.21
Thus, the per-problem
compensation under the two schemes implies that for a given
performance, individuals who solved twelve or more problems
have higher expected earnings from submitting to a tournament.
This corresponds to 30 percent of the women and 40 percent of the
men. Including participants who solve eleven problems—and are
virtually indifferent between the two schemes—the percentages
are 40 percent for the women and 45 percent for the men.
The actual difference in compensation scheme choices is substantially
larger. With 25 percent of the women and 55 percent of
20. This difference is not significant in a sample of forty men and forty
women (p ϭ .408).
21. For any given performance level, say fifteen for a woman, we draw 10,000
groups consisting of two men and one other woman, where we use the sample of
forty men and women with replacement. We then calculate the frequency of wins.
The exercise is repeated 100 times, and we report the average of these win
frequencies. For details, see Niederle and Vesterlund [2005].
1090 QUARTERLY JOURNAL OF ECONOMICS
the men submitting their piece-rate performance to the tournament,
we find a significant gender difference (p-value ϭ .012 by
Fisher’s exact test).22
That is, men and women differ in their
compensation scheme choices even when the decision does not
involve the prospect of having to subsequently perform in a competition.
Next we examine if this difference remains when we
condition first on performance and then on beliefs about relative
piece-rate performance. By controlling for beliefs we can determine
the extent to which the gap in compensation scheme choices
is accounted for by overconfidence versus risk and feedback
aversion.
There is no significant difference in the piece-rate performance
of the women who do versus do not submit the piece rate
(10.7 vs. 10.0 problems, p ϭ .48). In contrast, men who submit to
the tournament solved significantly more problems (12.05) than
those who did not submit to the tournament (9), p ϭ .004. Figure
IV panel A shows the propensity of women and men to submit to
the tournament for each piece-rate performance quartile. While
for men a higher piece-rate performance is correlated with a
higher propensity to submit to a tournament, this is not the case
for women. The gender gap in choice of compensation scheme is
largest among participants in the top performance quartile. Of
the participants who have about equal or higher expected earn-
22. While participants seem more reluctant to submit the piece rate result to
a tournament than they were to enter a tournament and then competing, these
differences are not significant either for women (a Fisher’s exact test yields p ϭ
.465) or for men (a Fisher’s exact test yields p ϭ .162).
FIGURE IV
Proportion of Participants Who Select Tournament for Task 4 Conditional on
Task-1 Performance Quartile (Panel A) and Guessed Piece-rate Rank (panel B).
1091GENDER DIFFERENCES IN COMPETITIVENESS
ings from submitting to the tournament (eleven or more correct
answers), significantly more men (fourteen out of sixteen) than
women (three out of twelve) select the tournament (Fisher’s exact
test p ϭ .001). Of those who have lower expected earnings from
the tournament (less than eleven correct answers) there is no
significant difference in the proportion of men and women who
submit to the tournament (eight out of twenty-two and five out of
twenty-two, respectively) (Fisher’s exact test p ϭ .33).
A probit regression confirms that participants with a higher
performance are more likely to submit to the tournament and
that conditional on piece-rate performance men are more likely to
do so than women.23
The significant gender difference in choices
of compensation scheme seems driven by high-performing participants
with twelve or more correct answers. While the female
dummy is significant in a probit regression on this subsample of
participants, it is not significant in the subsample of participants
with a performance of ten or less.24
Absent future competition we
see that gender differences in general factors such as confidence,
risk, and feedback aversion cause a gap in choice of compensation
scheme among high performing participants.
To distinguish the impact of confidence from that of risk and
feedback aversion we elicited the participants’ beliefs on relative
performance in the piece rate. The characteristics of the elicited
beliefs on relative performance in the Task-1 piece rate are very
similar to what we found for the Task-2 tournament. Both women
and men are overconfident and their believed rank distributions
differ significantly from the actual rank distributions (a Fisher’s
exact test of independence yields p ϭ .00 for men and for women).
As for the tournament performance, men are significantly more
confident about their relative performance than women (a Fisher’s
exact test yields p ϭ .02). An ordered probit of guessed
piece-rate rank as a function of the piece-rate performance and a
23. A probit regression of decision to submit the piece rate to a tournament
yields marginal effects of Ϫ.31 on female (p ϭ .01) and .06 on piece-rate performance
(p ϭ .01), evaluated at a man with eleven correct answers in Task 1 (this
is the performance at which the expected payoff is the same from the piece rate as
from the tournament).
24. A probit regression of the decision to submit to a tournament on the
piece-rate performance and a female dummy yields, for participants who solve ten
or less in the piece rate, a coefficient on the piece rate of .03 (p ϭ .60), and on the
female dummy of Ϫ.17 (p ϭ .23) evaluating the marginal effects at a man who
solves ten problems. For participants who solve twelve or more, the coefficient on
the piece-rate performance is .03 (p ϭ .42) and on the female dummy Ϫ.63 (p ϭ
.002) evaluated at a man who solves twelve.
1092 QUARTERLY JOURNAL OF ECONOMICS
female dummy confirms that women are significantly less confident
than men and that participants with a higher absolute
performance think they have a higher relative performance.25
Even though we find a gender gap in beliefs about the relative
piece rate performance, this gap is less substantial than the
gap we found when women and men rank their tournament
performance. Beliefs on relative performance in the piece rate
and tournament are correlated, but beliefs in the piece rate cannot
fully explain those in the tournament. While men are more
overconfident than women about their piece-rate performance,
the difference in overconfidence is even greater when it comes to
the tournament performance.26
This could be because of a stereotype
that women are not so competitive or that women may be
more stressed during the tournament [Steele 1997].27
While the characteristics of the elicited beliefs for Task 1 are
similar to those of Task 2, the effect of beliefs on choice of compensation
scheme is very different. Panel B of Figure IV shows,
for each guessed piece-rate rank, the proportion of women and
men that submits their piece-rate performance to a tournament.
In contrast to our Task-3 choice of compensation scheme, the
gender gap in the Task-4 decision is very small when we condition
on the participant’s believed ranking. More confident participants
are much more likely to submit to the tournament, and women
and men are both about 60 percentage points more likely to
submit to a tournament when they think they are the highest
performer in their group, rather than the second highest.
The probit analysis in Table VII confirms that, conditional on
performance and guessed rank, the gender gap in choice of compensation
scheme is small. While gender plays a substantial and
significant role when controlling only for piece-rate performance
25. An ordered probit of guessed piece-rate rank yields coefficients of .77 on
female (SE 0.27, p ϭ .01) and Ϫ.19 on piece-rate performance (SE 0.05, p ϭ .00).
Guesses of four are eliminated, leaving thirty-nine women and thirty-eight men.
26. The result of an ordered probit of relative tournament rank generates the
following coefficients: .74 on female (SE 0.33, p ϭ .03), Ϫ.07 on tournament (SE
0.09, p ϭ .35), Ϫ.25 on tournament-piece rate (SE 0.09, p ϭ .00), and .82 on
guessed piece rate rank (SE 0.28, p ϭ .00). The six participants who guessed a
rank of four in either the tournament or the piece rate are omitted leaving
thirty-seven men and thirty-seven women.
27. Stereotype threat theory suggests that stereotyped individuals (e.g.,
women who are supposed to be poor competitors) who find themselves in a
situation where they run the risk of confirming the stereotype (i.e., in a tournament
where they may lose) may feel additional performance anxiety for fear of
confirming the stereotype. This additional threat may harm female performance,
as they may “choke” under the pressure.
1093GENDER DIFFERENCES IN COMPETITIVENESS
(column (1)), this effect is to a large extent accounted for by
gender differences in overconfidence (column (2)). When the compensation
scheme choice does not require that participants subsequently
perform in a competition, the relative overconfidence of
men appears to explain most of the gender difference. Thus,
conditional on beliefs, general factors such as risk and feedback
aversion have a negligible effect on the Task-4 choice of compensation
scheme.
V.C. Do Preferences for Performing in a Competition Cause
Gender Differences in Choice of Compensation Scheme?
The decision to submit a past piece-rate performance to a
tournament (Task-4 choice) and the decision to enter a tournament
and perform in a competition (Task-3 choice) have similar
characteristics. In both cases the choice is between a piece rate
versus a tournament payment, and in both cases the decision
depends on the participants’ beliefs about their relative performance.
Furthermore, in both cases a choice of tournament will
provide participants with feedback on their relative performance.
The difference between the two decisions is that only when participants
enter the tournament do they have to perform subsequently
in a tournament. In this section we determine whether
gender differences in tournament entry are driven largely by
general factors or if there are additional gender differences when
it comes to entering a tournament. Specifically, is the tournament-entry
gap in part explained by different preferences for
TABLE VII
PROBIT OF DECISION TO SUBMIT THE PIECE RATE TO A TOURNAMENT (TASK 4)
Coefficient (p-value)
(1) (2)
Female Ϫ.327 Ϫ.13
(.01) (.21)
Piece rate .05 .00
(.02) (.80)
Guessed piece rate rank Ϫ.32
(.00)
Dependent variable: Task-4 choice of compensation scheme (1-tournament, 0-piece rate). The table
presents marginal effects evaluated at a man with a guess of one and eleven correct answers in Task 1.
Excluding guesses of 4, the sample is thirty-nine women and thirty-eight men.
1094 QUARTERLY JOURNAL OF ECONOMICS
performing in a competition or is it fully accounted for by gender
differences in general factors such as confidence and risk and
feedback aversion?
Our results thus far suggest that the decision to submit a
past performance to a tournament differs from the decision to
enter the tournament and then perform. While for high-performing
participants there is a significant gender difference in the rate
by which participants submit to the tournament, these differences
are not significant among low-performing participants. In
contrast, the gender difference in tournament entry is independent
of performance. Furthermore, while gender differences in
beliefs about relative piece-rate performance are sufficient to
eliminate the gender gap in the decision to submit to a tournament,
beliefs on tournament performance only account for part of
the gender gap in tournament entry.
To account for gender differences in general factors such as
overconfidence and risk and feedback aversion, we use the participants’
guessed tournament ranks along with their Task-4
choice as controls in the tournament-entry decision (see column
(3) of Table VIII).28
28. We omit the three participants who guessed a rank of four in the tournament,
leaving thirty-nine men and thirty-eight women.
TABLE VIII
PROBIT OF TOURNAMENT-ENTRY DECISION (TASK 3)
Coefficient (p-value)
(1) (2) (3)
Female Ϫ.379 Ϫ.278 Ϫ.162
(.01) (.01) (.05)
Tournament .015 Ϫ.002 Ϫ.009
(.39) (.90) (.42)
Tournament–piece rate .008 Ϫ.001 .011
(.72) (.94) (.44)
Guessed tournament rank Ϫ.181 Ϫ.120
(.01) (.01)
Submitting the piece rate .258
(.012)
Dependent variable: Task-3 compensation scheme choice (1-tournament and 0-piece rate). The table
presents marginal effects evaluated at a man with thirteen correct answers in the tournament and twelve in
the piece rate, who submits to the tournament, and with a guess of one in the Task-2 tournament. Guesses
of four are eliminated, resulting in a sample of thirty-eight women and thirty-nine men.
1095GENDER DIFFERENCES IN COMPETITIVENESS
As anticipated, we see that participants who are confident
and who submit to a tournament (Task 4) are significantly more
likely to enter a tournament (Task 3). However, despite these
controls, a significant and large gender gap in tournament entry
still remains. While controlling for beliefs on relative performance
reduced the gender gap in tournament entry from 37.9 to
27.8 percentage points (columns (1) and (2)), this gender effect is
reduced to 16.2 percentage points when controlling for the decision
to submit the piece rate (column (3)). This decrease may be
explained both by the control for risk and feedback aversion and
by the fact that the decision to submit the piece rate serves as an
additional measure of the individual’s general degree of confidence.
It is therefore not surprising to see that the coefficient on
guessed tournament rank decreases as we move from column (2)
to column (3).
Overall, we find that about 57 percent of the original gender
effect can be accounted for by general differences in overconfidence
and risk and feedback aversion while the residual “competitive”
component is 43 percent. This makes clear that the gender
gap in choice of compensation scheme is exacerbated when individuals
subsequently have to perform under the selected compensation
scheme. Controlling for the Task-4 decision as well as
believed tournament rank, the marginal effect of gender on the
decision to enter the tournament is still 16 percent. This suggests
that the gender gap in tournament entry is influenced by men and
women differing in their preference for performing in a competitive
environment.
VI. CONCLUSION AND DISCUSSION
This paper contributes to a literature that tries to understand
why women are underrepresented in many high-profile jobs
and across whole professions. While gender differences in preferences
and ability or discrimination are likely to play an important
role in answering this question, we argue that another reason
may be that men and women respond differently to competitive
environments.
Past research on gender differences toward competition has
shown that in some mixed gender competitions women do not
perform as well as men. For example, Gneezy, Niederle, and
Rustichini [2003] examine performances when participants are
asked to solve mazes for fifteen minutes. Although men and
1096 QUARTERLY JOURNAL OF ECONOMICS
women perform equally well in a piece-rate scheme, there are
large gender differences in performance in a tournament. While a
few women perform extremely well, many women do poorly, and
the bottom performance quintile is almost entirely comprised of
women. Gneezy and Rustichini [2004] and Larson [2005] find
similar gender differences in competitive performance.
Rather than examining gender differences in performance
under an exogenously given incentive scheme, the focus of this
paper is instead one of self-selection. For a given performance we
examine whether men and women are equally willing to select
into a competition. Since an inferior performance of women may
make them more reluctant to compete, we chose a task for which,
even in tournaments, men and women perform equally well.
Specifically we selected a short task, where men and women are
thought to have the same abilities, a task that is not exciting, but
rather requires participants to be very careful not to make simple
mistakes. Our study demonstrates that despite there being no
gender differences in performance, men are more than twice as
likely to enter the tournament.
Combined, these studies on gender differences in competitive
environments suggest that there may be two additional reasons why
women may not be well represented in competitive jobs. First, in
mixed-gender competitions there are circumstances where the performance
of men is superior to that of women. Second, even when
women and men are equally successful in the competitive environment,
when given a choice, women may not enter the competition at
the same rate as their male counterparts.
We find that the gender gap in tournament entry is primarily
caused by two factors. One is that men are substantially more
overconfident than women, and the other is that men and women
differ in their preferences for performing in a competition. To identify
the preference for performing in a competition, we examine the
decision to perform in a tournament when controlling for gender
differences in general factors such as overconfidence, risk and feedback
aversion. Specifically we regress the tournament entry decision
on performance, belief on tournament ranking, and the decision to
submit a past performance to the tournament. As the decision to
submit a past performance to a tournament is very similar to the
decision to enter a tournament and then performing, this decision
serves as a control for gender differences in general factors such as
overconfidence and risk and feedback aversion. Although the believed
ranking and the decision to submit a past performance both
1097GENDER DIFFERENCES IN COMPETITIVENESS
have explanatory power, a substantial portion of the gender gap in
tournament entry remains. We interpret this unexplained gender
gap as evidence that men and women differ in their preferences for
entering and performing in a competition.
A few words of caution are warranted when assessing the
effect gender differences in preferences for competition may have
on tournament entry. First, while we use Task 4 to control for the
role played by gender differences in risk and feedback aversion,
these factors may play an even larger role when it comes to
performing in a competition. For example, if the gender difference
in feedback aversion is larger when actively competing, then the
Task-4 choice of compensation scheme does not fully account for
that gender difference. As a result, the additional gender difference
in feedback aversion in competitive environments, which is
present in Task 3 will be attributed to gender differences in
preferences for performing in a competition.
Second, the effect of gender differences in preferences for
competition may be overestimated if men are relatively more
optimistic about their future performance. For example, women
may act differently than men for a given believed Task-2 ranking
if they differ in how good a predictor they feel their past performance
is for a future performance. Indeed, women are more prone
to attribute past successes to luck than to inner attributes (and
past failures less to bad luck) while men do the opposite.29
However,
as we saw in Table II, there is no evidence that participants
view an increase in prior performance as indicative of a future
performance. Rather, the increase in performance between Task
1 and Task 2 has no effect on tournament entry. While we cannot
rule out that women feel that their past performance is a bad
predictor of their future performance, actual performance increases
do not predict tournament entry. Furthermore, with 75
percent of men thinking they are best in Task 2, this proportion
will only increase marginally if men expect future performances
to be even better. However to the extent that there are gender
differences in the participants’ beliefs about their future performance
and that these influence tournament entry, our study
incorrectly attributes such an effect to men and women having
different preferences for performing in a competition.
Our finding that men and women differ in their choice of
compensation scheme appears to be a rather robust one, and it
29. Beyer [1990] and Felder et al. [1994].
1098 QUARTERLY JOURNAL OF ECONOMICS
has been demonstrated by other researchers as well. For example,
Gneezy and Rustichini [in progress] and Gupta, Poulsen, and
Villeval [2005] have similar findings. Both papers focus on performance
of participants after their choice of incentive scheme
and replicate our finding that, conditional on a chosen incentive
scheme, there are no large gender differences in performance. In
contrast to our study, they do not assess performances prior to the
participants’ compensation scheme choice; thus, they cannot predict
the choices of participants based on performance and are
unable to determine payoff-maximizing choices.
While these laboratory studies replicate our general finding,
there is also evidence to suggest that gender differences in behavior
under competition may extend to other domains. For example,
Babcock and Laschever [2003] explore the possibility that
gender differences in labor market outcomes may arise because
women are poor negotiators and generally dislike the process of
negotiating. To the extent that a negotiation can be seen as a
two-person competition, their results appear consistent with
those on competition. Once again there are two effects: of women
first avoiding the competitive scheme altogether and when forced
to do so, sometimes failing to compete.
Further evidence that our findings in the laboratory may extend
to the real world is that the factors that we identify as causing
women to shy away from competition correspond to those emphasized
by women in these environments.30
For example, a report
entitled “Women’s Experiences in College Engineering” writes that
the exit of many young women is not driven by ability but rather
that this decision is influenced by women negatively interpreting
their grades and having low self-confidence. Furthermore these
women mention that negative aspects of their schools’ climate, such
as competition, lack of support and discouraging faculty and peers,
cause them to reevaluate their field of study (Goodman, Cunningham,
and Lachapelle [2002] and Felder et al. [1994] find similar
effects).
It is generally agreed that ability alone cannot explain the
absence of women in male-dominated fields. In natural settings,
30. While it is not always easy to directly measure external validity, our
results seem to have struck a chord and are used as arguments for workplace
adjustments. For example, our paper has been cited in a Submission to the
Senate’s Employment, Workplace Relations and Education Legislation Committee’s
Inquiry into the Workplace Relations Amendment (WorkChoices) Bill 2005
in Australia (see http://www.hreoc.gov.au/legal/submissions/workplace_
relations_amendment_2005.html).
1099GENDER DIFFERENCES IN COMPETITIVENESS
issues such as discrimination, the amount of time devoted to the
profession, and the desire for women to raise children may provide
some explanation for the choices of women. However, in this
paper we have examined an environment where women and men
perform equally well and where issues of discrimination or time
spent on the job do not have any explanatory power. Nonetheless
we find large gender differences in the propensity to choose competitive
environments. It appears that these differences are
driven by gender differences in confidence and preferences for
entering and performing in a competition. These differences seem
sufficiently strong to call for greater attention of standard economics
to explanations of gender differences that so far have
mostly been left in the hands of psychologists and sociologists.
Much may be gained if we can create environments in which
high-ability women are willing to compete.
STANFORD UNIVERSITY
NBER
UNIVERSITY OF PITTSBURGH
REFERENCES
Altonji, Joseph G., and Rebecca Blank, “Race and Gender in the Labor Market,” in
Handbook of Labor Economics, Vol. 3c, O. Ashenfelter, and D. Card, eds.
(Amsterdam, The Netherlands: Elsevier Science B.V., 1999), 3144–3259.
Andreoni, James, and Lise Vesterlund, “Which is the Fair Sex: Gender Differences
in Altruism,” Quarterly Journal of Economics, CXVI (2001), 293–312.
Babcock, Linda, and Sara Laschever, Women Don’t Ask: Negotiation and the
Gender Divide (Princeton, NJ: Princeton University Press, 2003).
Barber, Brad M., and Terrance Odean, “Boys will be Boys: Gender, Overconfidence,
and Common Stock Investment,” Quarterly Journal of Economics,
CXVI (2001), 261–292.
Bertrand, Marianne, and Kevin F. Hallock, “The Gender Gap in Top Corporate
Jobs,” Industrial and Labor Relations Review, LV (2001), 3–21.
Beyer, Sylvia, “Gender Differences in the Accuracy of Self-Evaluations of Performance,”
Journal of Personality and Social Psychology, LIX (1990), 960–970.
——, and Edward M. Bowden, “Gender Differences in Self-Perceptions: Convergent
Evidence from Three Measures of Accuracy and Bias,” Personality and
Social Psychology Bulletin, XXIII (1997), 157–172.
Black, Sandra, and Philip E. Strahan, “The Division of Spoils: Rent-Sharing and
Discrimination in a Regulated Industry,” American Economic Review, XCI
(2001), 814–831.
Blau, Francine D., and Lawrence M. Kahn, “The US Gender Pay Gap in the 1990s:
Slowing Convergence,” NBER Working Paper No. 10853, 2004.
Byrnes, James P., David C. Miller, and William D. Schafer, “Gender Differences
in Risk Taking: A Meta-Analysis,” Psychological Bulletin, LXXV (1999),
367–383.
Campbell, Anne, A Mind of Her Own: The Evolutionary Psychology of Women
(Oxford, UK: Oxford University Press, 2002).
Croson, Rachel, and Uri Gneezy, “Gender Differences in Preferences,” Wharton
Working Paper, 2004, http://opim.wharton.upenn.edu/ϳcrosom/research/
%5B59%5D.pdf.
Daly, Martin, and Margo Wilson, Sex, Evolution, and Behavior, 2nd
ed. (Belmont,
CA: Wadsworth Publishing Company, 1983).
1100 QUARTERLY JOURNAL OF ECONOMICS
DellaVigna, Stefano, Ulrike Malmendier, and Lise Vesterlund, “How Robust is
Overconfidence: The Effect of Noisy and Non-Noisy Signals,” in preparation.
Dweck, Carol S., Self-Theories: Their Role in Motivation, Personality, and Development
(Philadelphia, PA: Psychology Press, 2000).
Eckel, Catherine C., and Philip J. Grossman, “Sex and Risk: Experimental Evidence,”
in Handbook of Experimental Economics Results, C. Plott, and V.
Smith, eds. (Amsterdam, The Netherlands: Elsevier Science B.V./North-Holland,
forthcoming, 2002a).
——, “Differences in the Economic Decisions of Men and Women: Experimental
Evidence,” in Handbook of Experimental Economics Results, C. Plott, and V.
Smith, eds. (Amsterdam, The Netherlands: Elsevier Science B.V./North-Holland,
forthcoming, 2002b).
Felder, Richard M., Gary N. Felder, Meredith Mauney, Charles E. Hamrin, and
Jacquelin Dietz, “A Longitudinal Study of Engineering Student Performance
and Retention,” Report NCSU-94A, 1994.
Fischbacher, Urs, “z-Tree Zurich Toolbox for Ready-made Economic Experiments,”
Experimental Economics, X (2007), 171–178.
Gneezy, Uri, and Aldo Rustichini, “Gender and Competition at a Young Age,”
American Economic Review Papers and Proceedings, XCIV (2004), 377–381.
Gneezy, Uri, “Executive versus Teachers: Gender, Competition and Self-Selection,”
Working Paper, in progress.
Gneezy, Uri, Muriel Niederle, and Aldo Rustichini, “Performance in Competitive
Environments: Gender Differences,” Quarterly Journal of Economics, CXVIII
(2003), 1049–1074.
Goldin, Claudia, and Cecilia Rouse, “Orchestrating Impartiality: The Impact of
‘Blind’ Auditions on Female Musicians,” American Economic Review, XC
(2000), 715–742.
Goodman, Irene F., Christine M. Cunningham, and Cathy Lachapelle, “The Women’s
Experience in College Engineering,” Goodman Research Group, Inc.,
Cambridge, MA, 2002.
Gupta, Nabanita D., Anders Poulsen, and Marie-Claire Villeval, “Male and Female
Competitive Behavior—Experimental Evidence,” GATE Working Paper
05-12, Ecully, France, 2005.
Hyde, Janet S., Elizabeth Fennema, and Susan J. Lamon, “Gender Differences in
Mathematics Performance: A Meta-Analysis,” Psychological Bulletin, CVII
(1990), 139–155.
Larson, Lia C., “When Girls Stop Competing Against Boys: An Experimental
Analysis of the Competitive Behavior of Young Children,” Harvard Honors
Thesis, 2005.
Ledyard, John O., “Public Goods: A Survey of Experimental Research,” in Handbook
of Experimental Economics, John H. Kagel, and Alvin E. Roth, eds.
(Princeton, NJ: Princeton University Press, 1995).
Lichtenstein, Sarah, Baruch Fischhoff, and Lawrence Phillips, “Calibration and
Probabilities: The State of the Art to 1980,” in Judgment under Uncertainty:
Heuristics and Biases, Daniel Kahneman, Paul Slovic, and Amos Tversky,
eds. (New York: Cambridge University Press, 1982).
Lundeberg, Mary A., Paul W. Fox, and Judith Puncochar, “Highly Confident but
Wrong: Gender Differences and Similarities in Confidence Judgments,” Journal
of Educational Psychology, LXXXVI (1994), 114–121.
Moore, Don A., and Deborah Small, “Error and Bias in Comparative Social
Judgment: On Being Both Better and Worse than We Think We are,” Tepper
Working Paper 2004-E1, Pittsburgh, PA, 2004.
Niederle, Muriel, and Lise Vesterlund, “Do Women Shy Away From Competition?
Do Men Compete Too Much?,” NBER Working Paper No. 11474, 2005.
Roberts, Tomi-Ann, and Susan Nolen-Hoeksama, “Sex Differences in Reactions to
Evaluative Feedback,” Sex Roles, XXI (1989), 725–747.
Ruble, Diane N., Carol L. Martin, and Sheri A. Berenbaum, “Gender Development,”
in Handbook of Child Psychology, 6th
ed., William Damon and Richard M.
Lerner, eds. (New York: Wiley, 2006), Vol. 3, Chapter 14, pp. 858–932.
Steele, Claude M., “A Threat in the Air: How Stereotypes Shape Intellectual
Identity and Performance,” American Psychologist, LII (1997), 613–629.
1101GENDER DIFFERENCES IN COMPETITIVENESS