NBER WORKING PAPER SERIES

GENDER DIFFERENCES IN RESPONSE TO COMPETITIVE ORGANIZATION?

DIFFERENCES ACROSS FIELDS FROM A PRODUCT DEVELOPMENT PLATFORM
FIELD EXPERIMENT

Kevin Boudreau
Nilam Kaushik

Working Paper 30062

[Link]

NATIONAL BUREAU OF ECONOMIC RESEARCH

1050 Massachusetts Avenue
Cambridge, MA 02138
May 2022

We thank members of supporting organizations and particularly Maria Costa De Sousa, Hugh
Courtney, Mavez Dabas, Sylvain Demortier, Anthony Donaldson, Raj Echambadi, Koreen
Geisler-Wagner, Austen Keene, Afan Khan, Atif Khan, Abhinav Kharbanda, Dyan Khor,
Raghavi Kirouchenaradjou, Sreerag Sreenath Mandakathil, Patrick McGrath, Tucker Marion,
Marc Meyer, Robert Hughes, Michael Orr, Olga Ozhereleva, Kaushik Padmanabhan, Edwige
Poinssot, Fernando Suarez, Prathamesh Kisan Tajane, Nikin Tharan, Emery Trahan, Maureen
Underhill, Robert Whelan, and Katherine Wilhoit. For helpful comments we thank Victor
Bennett, Roberto Fernandez, Sharique Hasan, Olga Shurchkov, and seminar participants at Duke
University, DRUID, and the Academy of Management. We would also like to acknowledge
generous financial support from the Kauffman Foundation and Northeastern DMSB. The views
expressed here are those of the authors alone and do not necessarily reflect those of supporting
institutions, funders or supporters involved with this study. All errors are our own. The views
expressed herein are those of the authors and do not necessarily reflect the views of the National
Bureau of Economic Research.

NBER working papers are circulated for discussion and comment purposes. They have not been
peer-reviewed or been subject to the review by the NBER Board of Directors that accompanies
official NBER publications.

© 2022 by Kevin Boudreau and Nilam Kaushik. All rights reserved. Short sections of text, not to
exceed two paragraphs, may be quoted without explicit permission provided that full credit,
including © notice, is given to the source.
Gender Differences in Response to Competitive Organization? Differences Across Fields
from a Product Development Platform Field Experiment
Kevin Boudreau and Nilam Kaushik
NBER Working Paper No. 30062
May 2022
JEL No. D02,D03,D9,D91,J0,J01,J2,M0,M12,M5,O3,O31,O32,O36

ABSTRACT

Prior research, primarily based on lab experiments, suggests that females might be more averse to
competition than males and could be more inclined towards collaboration, instead. Were these
findings to generalize to adults across the workforce, there could be profound implications for
organizational design and personnel management. We report on a field experiment in which
97,678 adults from a wide range of fields and ages were invited to join a product development
opportunity. Individuals were randomly assigned to treatments framing the opportunity as either
involving competitive or collaborative interactions with other participants. Among those outside
of science, technology, engineering, and math fields (STEM), we find significant gender
differences in willingness to participate under competition. Among those in STEM fields, we
detect no statistical gender differences. These results and broader patterns documented in the
study are consistent with significant heterogeneity in competitiveness across both men and
women, with field and career sorting resulting in differences (in gender differences) across fields.

Kevin Boudreau
D'Amore-McKim School of Business
Northeastern University
360 Huntington Avenue
Boston, MA 02115
and NBER
[Link]@[Link]

Nilam Kaushik
Bannerghatta Road, Bengaluru, India
IIMB
India
[Link]@[Link]
1 Introduction

A stream of largely lab-based studies has reported that male subjects have a higher probability of
choosing to engage in tasks under competitive conditions than female subjects (e.g., Niederle and
Vesterlund 2007; Saccardo et al. 2017; Sutter and Glätzle-Rützler 2015).1 The gender differences observed
in these studies have been interpreted as consistent with gender-based differences in “preferences” for
working under competitive conditions (e.g., Croson and Gneezy 2009). Female subjects in the literature
have instead exhibited a greater willingness to participate under non-competitive piece-rate regimes and
collaborative team-based regimes, relative to male subjects within these studies (e.g., Abramo et al. 2013;
Dargnies 2012; Healy and Pate 2011; Kuhn and Villeval 2015). These provocative findings and
interpretations are supported by numerous replication studies and have even begun to diffuse to wider
popular media audiences (Guo 2015; PBS News Hour 2015; Villeval 2012). Nonetheless, the
generalizability of these findings remains little understood (Flory et al. 2015). The preponderance of
existing empirical results has been drawn from relatively narrow contexts and study populations (especially
children and students). Moreover, there is no theoretical consensus on underlying mechanisms that might
be used to extrapolate beyond the existing results (see Section 2). In this paper, we take steps towards
further understanding the generalizability and boundary conditions of gender differences in responses to
competition across a wide spectrum of working-age adults.

It is important to better understand the extent to which working-age men and women might
systematically differ in their responses to competition. For example, scholars in this literature now routinely
speculate that differences in competitiveness might explain some part of “why women are underrepresented
in many high-profile jobs and in whole professions” (Niederle and Vesterlund 2007). If girls and women
are more averse to competition, they might systematically sort out of fields and occupations that are
perceived to be competitive. This line of argumentation has been used to speculate about the causes of
disparities in gender representation in areas such as senior executive positions, STEM fields (i.e., sciences,
technology, engineering, and mathematics), and politics (e.g., Balafoutas et al. 2018; Buser et al. 2014;
Niederle and Vesterlund 2010). Clearer understanding might therefore shed light on whether “women’s
talents are sometimes wasted because they avoid competitive environments” (Kuhn and Villeval 2015).

Attaining a better understanding of systematic behavioral patterns in response to competition could

also point to ways of improving organizational designs to better harness workers’ talents and efforts. This
is because the “levers” of organizational design—internal incentive systems, hiring and selection processes,

1
See Inglehart et al. (1994) for a discussion of this hypothesis pre-dating the more recent experimental literature.

2
decision making and political processes, work norms, and values—regulate the intensity of competition
experienced by workers (Baumann et al., 2019; Lee and Puranam, 2017; Obloj and Zenger, 2017; Song et
al., 2018). Optimizing exposure to competition could be especially relevant for new digital forms of
organizing—such as crowdsourcing, gig platforms, open communities, and open innovation—where
interactions among participants can be “tuned” to be more or less competitive or collaborative (Hunt and
Samman 2019; Liang et al. 2018).

In this paper, we take steps toward theorizing and documenting possible limits and boundary
conditions regarding gender differences in competitiveness. At the core of the paper is a field experimental
approach that allows us to observe responses to competition across a wide range of college-educated,
working-age adults. We proceed by first developing theoretical predictions to guide our investigation. In
short, we predict that gender differences in competitiveness should differ according to individuals’ fields
of training, and that gender differences should be significantly smaller among those who sort into STEM
fields.

Given the current lack of theoretical consensus on underlying mechanisms shaping responses to
competition and possible gender differences (Section 2.2), our approach is to develop predictions based on
just the most consistent claims and assumptions across the prior literature. A first point we build on is that
there exists some (yet little understood) degree of heterogeneity in competitiveness across the population,
i.e., differences in propensity for or aversion to competition. The second point we build on is that
individuals’ competitiveness serves as a meaningful basis for sorting into different fields. This sorting is
especially important as young people develop and make path-dependent choices in education, culminating
into longer-term career pathways (Kahn and Ginther, 2017). A simplest possible implication of these
presumptions of prior literature is that men and women sorting into STEM fields should be relatively similar
in their competitiveness—in the sense that those sorting into STEM might each be relatively competitive.

However, whether and to what extent men’s and women’s responses to competition converge in
STEM fields and do so in a manner consistent with presumptions remains an empirical question. To test
for gender differences in responses to competition across working-age adults and to investigate possible
differences across fields, we report results from a field experiment covering 97,678 adults from a range of
fields of training, and ages spanning college years to retirement.

Each individual in our study population was invited to participate in a three-week, part-time product
development activity. This activity was designed with the intent of being both accessible and relevant to all
invitees, whatever their (technical or non-technical) field of training, experience, or background (see
Section 3.1). The experiment focused on randomly assigning individuals to treatments that framed
interactions with other participants as either competitive or collaborative. We, therefore, derive inferences

3
by investigating how the likelihood of participation depended on assignment to competition versus
collaboration treatments, along with gender, field of training (i.e., college majors in humanities, business,
sciences, computer science, etc.), and other personal characteristics.

We found that in non-STEM fields, gender differences in responses to competition mapped closely
to existing intuitions regarding gender-based differences in responses to competition. Non-STEM men’s
and women’s participation under the collaboration treatment was statistically equivalent, with roughly three
percent choosing to participate. Non-STEM men’s participation was also statistically the same under the
competitive treatment. However, the participation rate of non-STEM women was significantly lower in the
competition treatment, by about a half percentage point (or a 25.5% drop).

Baseline participation rates were higher among those in STEM fields, with overall participation at
roughly four percentage points. Our primary finding, consistent with predictions, is that responses to
competition among men and women in STEM fields was similar. Participation rates for both STEM men
and STEM women were about one percentage point lower under the competition treatment than under the
collaboration treatment. Apart from simply being more similar, the response to competition among STEM
men and women within our context is statistically indistinguishable from zero. ( Sections 5.2 and 6 provide
a fuller presentation and discussion of results.) The results are robust to a series of alternative controls and
specifications. We also validated this finding result in tests of 17 STEM more narrowly-defined subfields.

However, whereas we found evidence of similarities of STEM men and women, we did not find
this was because they were both more competitive. Both STEM men and women were in fact more willing
to work under collaboration than under competition. This result runs counter to simplest interpretations of
the sorting-on-competitiveness hypothesis suggested in prior research. As we discuss in Section 6,
willingness to work under competition (and collaboration, too) is inherent to much of the training and work
in technical problem-solving in STEM. Building on research describing gender dynamics in STEM fields,
we argue the results are consistent with those in STEM fields possessing a willingness and ability to work
under either competition or collaboration and teamwork, depending on the demands of the context.

Overall, this study takes steps towards clarifying the generalizability—along with certain limits and
boundary conditions—of previously documented gender differences in competitiveness. By doing so, this
paper continues to build bridges between research on gender and competition and questions of firms and
organizations, outside of the lab. Organizational scholarship considering the implications of possible gender
differences in competitiveness remains scarce. (See Brands and Fernandez-Mateo (2017) for an exception.)
This paper also contributes to a growing number of studies examining how organizational designs and
processes shape gender representation in STEM fields (e.g., Fernandez Campero 2017; Murciano-Goroff
2018; Wynn Correll 2017.)

4
 The paper proceeds as follows: Section 2 reviews the prior literature and develops our theoretical
hypothesis. Section 3 describes the field experimental research design. Section 4 describes the data and
variables. Section 5 reports the results. Section 6 summarizes and discusses the results. Section 7 concludes.

2 Background and Literature

This section reviews prior research on gender-based differences in responses to competition,

clarifying the limits of generalizability and extrapolation beyond the existing studies. We also build upon
the existing consensus in the literature to develop our main predictions.

2.1 Existing Evidence Suggesting Gender Differences in Competitiveness

The bulk of existing evidence on gender differences in response to competition comes from lab
experiments (e.g., Gneezy et al. 2003; Gneezy and Rustichini 2004; Niederle and Vesterlund 2011). The
canonical experimental approaches have involved either comparing whether male and female lab subjects
choose to work under competitive or noncompetitive conditions or, alternatively, evaluating the effect of
competition on the performance of male and female subjects.

To consider questions of generalizability, it is useful to review details of the existing research. For
example, Niederle and Vesterlund’s (2007) seminal lab experiment studied the behavior of 80 college
students recruited from the University of Pittsburgh. This study has emerged as the canonical experimental
protocol in the literature. The subjects sat in groups of two men and two women at a time and were asked
to add two-digit numbers for five minutes. The subjects were asked to choose whether they would be paid
according to a competitive scheme in which they would be paid $2 for each correct answer if they correctly
solved more problems than the other three subjects (i.e., a rank order tournament); alternatively, they could
choose a non-competitive piece rate of 50¢ per correct answer. The striking finding in this study was that
twice as many male subjects chose competition as did female subjects (73 percent versus 35 percent). To
help make the claim that the differences between male and female subjects could be interpreted as related
to gender differences rather than unobserved factors correlated with the recruited subjects’ gender, the
authors pointed out that they found no statistical differences in male and female subjects’ ability to solve
problems in pre-experimental trials.2 The authors also showed that a residual mean difference remained
between male and female subjects after controlling for measures they constructed to reflect risk tolerance,

2
As with any study using gender as an explanatory variable, including lab experiments, there are unavoidable challenges in
interpreting differences as all-else-being-equal “gender effects.” As it is not possible to randomly assign a subject’s sex (Holland,
1986) it is therefore not possible to unequivocally rule out spurious differences, even with “control” variables (Oster, 2019).

5
aversion to receiving feedback, and self-confidence.3 Additionally, men with low ability were interpreted
as being over-confident, as they chose competition too often relative to their odds of winning. Women with
high ability choose competition too infrequently relative to their odds of winning, and thus were interpreted
as being averse to competition.4

These findings of male subjects exhibiting willingness to participate under competition, or

otherwise performing better under competitive conditions than female subjects, have been replicated over
other experimental tasks, subject pools, and variants on the original protocols (Buser 2016; Comeig et al.
2016; Dato and Nieken 2014; Datta Gupta et al. 2013; Dittrich et al. 2014; Healy and Pate 2011; Iriberri
and Rey-Biel 2019; Morin 2015, 2015; Paserman 2010; Reuben et al. 2015; Saccardo et al. 2017; Shurchkov
2012; Sutter and Glätzle-Rützler 2015).

In relation to the questions of this study, there are yet few studies concerning working-age adults
and in relation to roles in the workforce. Reuben et al. (2015) found that a lab-based measure of the
competitiveness in a set of MBA students differed between men and women and that this measure was
correlated with entering high-paying industries after graduation. Additionally, Buser et al. (2020) used
online survey data and found that a measure of competitiveness among those responding to the survey was
associated with participants’ completed level of education, field of study in college, occupation, and
income. Research by Flory et al. (2015) is especially notable, as they reported on a field experiment that
tested how rates of applications by administrative assistants to online job postings in multiple US cities
vary depending on the posted compensation scheme. The study finds that in cities where there are higher
wages, women are less likely to apply to competitive pay schemes. The authors pointed out that this result
can be interpreted as consistent with female administrative assistants avoiding competitive jobs if they have
outside options.

2.2 Existing Evidence on Competition versus Collaboration

While many studies contrast competition with a lack of competition, particularly piece-rate
regimes, a closely-related stream of papers has contrasted female subjects’ aversion to competition with an
inversely-related preference for collaborative or team-based work (Abramo et al. 2013; Dargnies 2012;

3
NB. From a practical perspective, one’s risk tolerance, willingness to receive feedback, and self-confidence might alternatively
simply be viewed as part and parcel of one’s willingness to compete.
4
Gneezy et al. (2003) illustrates an alternative canonical protocol, recruiting 324 students to solve maze problems for 15 minutes.
Students were randomly assigned to competitive, or piece-rate pay schemes in groups of six. Men outperformed women in all
treatments, but the differences were not statistically significant in the noncompetitive regime. The authors interpreted the results as
indicating that men are more competitive, providing a possible explanation for “gender differences… in competitive high-ranking
positions.”

6
Healy and Pate 2011; Kuhn and Villeval 2015). Much of this work follows the same basic outlines of the
Niederle-Versterlund (2007) experimental protocol. In addition to comparing forms of competition versus
non-competitive (piece-rate) payment, these protocols include tests contrasting collaboration with
competition or collaboration with piece-rate.

For example, in a study representative of this work, Dargnies (2012) evaluated 78 people recruited
through an online system. She reported that women were more likely than men to choose to “team,” where
“team” was operationalized as combining (averaging) scores when solving arithmetic problems with an
anonymous unobserved “teammate,” as opposed to payoffs based on individual performance. Specifically,
she found that men with high skills were less likely to accept this version of “teaming” when they were
uncertain of the skills of their “teammate.” Healy and Pate (2011) carried out a similar protocol in an
experiment involving 192 students from Loyola Marymount University, finding similar patterns. Kuhn and
Villeval (2015) also implemented a similar protocol in a study of 174 students interacting online at a French
University. The protocol was similar, in this case, except the choice between individual pay and pooled
average outcomes was made without their being competition among teams. Again, female students
disproportionately choose the nominally “cooperative” (shared payoffs) regime. The authors emphasize an
interpretation of women experiencing inequity aversion and that “the same confidence deficit that pushes
women out of competitions pulls women into teams, where it is beneficial to have an abler teammate.”
Flory et al.’s (2015) field experiment involving online applicants for administrative assistant jobs included
a treatment that mentioning “teams” in the compensation description for an administrative assistant job had
a positive influence on the likelihood of women applying.

2.3 Limits to Extrapolating beyond Existing Findings

Notwithstanding a considerable number of replication studies, there are important limitations to

generalization that are well-known to authors in this literature. A first challenge to extrapolating beyond
the existing results is that the evidence is drawn from narrow contexts and narrow study populations,
concentrated on school children and campus lab experiments with synthetic tasks. Important advances in
field studies themselves remain somewhat pinpointed in coverage, as the pathbreaking study by Flory et al.
(2015), which focused on administrative assistants.

Narrow sampling of evidence places an especially large burden on theory to interpret

generalizability. However, there is yet little consensus on underlying theoretical explanations and
mechanisms. The prior research has suggested numerous plausible explanations for observed gender
differences, emphasizing “attitudes” toward or “preferences” and “tastes” for competition (Croson and
Gneezy 2009; Dohmen and Falk 2011; Kleinjans 2009), with an implication of possible biological and/or

7
evolutionary factors playing some role (Apicella et al. 2011; Buser 2009; Kivlighan et al. 2005; Knight
2002; Lippa 1998; Vugt et al. 2007; Wozniak et al. 2014). Other explanations include possible learned,
socialized, or innate differences in overconfidence and self-efficacy (e.g., Comeig et al. 2016; Niederle and
Vesterlund 2007), socialization of the stereotypes of gender roles (e.g., Gneezy et al. 2006), gender
differences in subjective beliefs about the implications and meaning of competition (e.g., Kesebir et al.
2019), and differing psycho-physiological tendencies of men and women to feel stress (Buckert et al. 2017;
Holt-Lunstad et al. 2001). We might also interpret related streams of lab studies reporting analogous
residual differences in risk attitudes, tastes for high-powered incentives, and aversion to feedback on
performance as alternatives or complementary explanations for differences in preferences for competition
(e.g., Byrnes et al. 1999; Croson and Gneezy 2009; Eckel and Grossman 2008). Despite the best efforts of
the authors in this literature, the inability to randomize subjects’ gender means it is not possible to entirely
rule out the possibility of spurious factors, especially in small and highly selective study groups.

Apart from the limitations, above, a third challenge is the number of published results that run
counter to findings of gender differences (many by the same authors). The authors of these contrarian
studies have interpreted results to hypothesize that gender differences could depend on any number of added
factors, including socialization, context, individual characteristics, and even how competition is
operationalized (e.g., Andersen et al., 2013; Boschini et al., 2018; Dreber et al., 2011; Gneezy et al., 2003,
2009; Günther et al., 2010). More generally, these contrarian results point to a need for further research into
possible boundary conditions and contingencies. At the same time, these contrary studies are not themselves
immune to the same limitations as the earlier-described studies, described above. A number of these
findings that are contrarian to popular intuitions are also contrarian to one another (see Appendix A.)

2.4 Differences Across STEM and non-STEM Fields?

Given the above limits to generalization in the literature, Flory et al. (2015) note: “Whether, and to
what extent, gender differences in attitudes toward competition lead to differences in naturally occurring
labor markets remains an open question.” Our empirical analysis to follow will make progress by reporting
more widely sampled evidence on adults across the workforce than has been possible in the prior literature.

However, we first build on earlier research to develop theoretical predictions to better guide and
support a generalizable interpretation of the analysis to follow. Below, we enumerate multiple arguments
and behavioral mechanisms that suggest that competitiveness among men and women should be
significantly more similar among those in STEM fields than those in non-STEM fields and the wider
population.

8
 Despite a lack of consensus on the theoretical mechanisms (Section 2.2), there are two widely
accepted presumptions across much of the prior research that allow us to develop simple predictions. The
first presumption is that individuals across the population are heterogeneous and somehow distributed in
their propensity toward or aversion to competition. Where prior research has focused on mean gender
differences, the presumption also implies there could potentially be considerable variance more generally,
both among males and females. Some degree of variance of behavior among men and among women is
observed in all prior studies, even small studies.

A second key presumption of the prior literature is that competitive individuals tend to sort
themselves into competitive fields (Antonovics et al., 2009; Balafoutas et al., 2018; Buser et al., 2014;
Niederle and Vesterlund, 2010a), and particularly within education and within careers thereafter. The
greatest available evidence supporting this presumption relates to the mathematics and STEM fields.
Specifically, the literature has gone as far to report correlations between measures of competitiveness and
sorting into math and STEM fields (Antonovics et al., 2009; Buser et al., 2014; Pregaldini et al., 2020;
Reuben et al., 2015). Strictly speaking, this evidence cannot rule out the possibility that sorting occurs based
on unobserved characteristics that are merely correlated with competitiveness. This point remains a
fundamental limitation in interpreting empirical results, even those of experiments, in this area.

To the extent that these premises are correct and produce economically significant effects, the
competitiveness of working-age adults in the economy should reflect some degree of sorting into fields
based on competitiveness. Thus, inasmuch as STEM fields are perceived to be more competitive, they
should tend to attract more competitive men and women, leading to greater similarities among the men and
women selecting these fields. This is an empirical question—and the one studied here.

3 Experimental Research Design

The goal of the latter half of this paper is to empirically investigate gender differences in
competitiveness across a widely sampled study population of working-age adults; and to test our prediction
that gender differences in response to competition should vary by field (Section 2.3).5

3.1 Research Context

This research was carried out in partnership with a large R1 research-oriented university in the
United States with roughly 20,000 undergraduate students, 8,000 graduate students, and several hundred

5
Other experiments and A/B tests were embedded in the launch campaign. However, these assignments were orthogonal to the
assignments described here and balanced across all subjects, and therefore have no effect on the results reported here.

9
thousand alumni in its network. The university is in the top-40 ranked colleges and universities according
to the US News Reports list of the best US universities with ranked undergraduate programs in engineering,
computer science, and business.

The university runs an online platform-based program to allow alumni and students to gather in
teams and compete with each other to design new products related to Internet-of-Things (IoT).6 The
objective of the program is to provide participants with opportunities to design IoT products
(“applications”), to gain an understanding of IoT technologies and opportunities, and to learn how these
technologies can be relevant to participants’ work organizations and sectors. The program also provides
experience in increasingly standard product development work practices from “use case” design to technical
design and business case design (see Appendix B). These goals are achieved through a series of part-time
learning-by-doing product development activities. These exercises are organized as team-based competitive
“challenges” or rank-order tournaments. While people sign up as individuals, they are assigned to teams by
the platform. To make the program as readily and widely accessible as possible, all activities are hosted on
an online platform that allows individuals to participate from any location during off-hours outside of work
or school.

The program involves a minimum of three weeks of part-time commitment of one's off-hours,
participating in at least one product development exercise. For this opportunity cost, individuals can benefit
from participating in interesting projects, learning, interacting with members of the university community,
receiving feedback on designs, and receiving cash prizes. Several teams also benefited by taking the ideas
and designs they developed on the platform into follow-up entrepreneurial development (off the platform).
The opportunity is promoted as highly relevant to all students and alumni, in terms of developing multi-
disciplinary product development skills, and in terms of exposing both nontechnical and technical
individuals to technical and business opportunities made possible by the Internet-of-Things. The platform
work and learning process was designed with the philosophy of “low floors and high ceilings” (Boaler,
2016) to engage a wide range of participants. Thus, relative neophytes could productively engage in
learning about the technology and contributing to designs, by following defined steps and platform tools.
Meanwhile, the platform also provides pathways for relative experts in the fields to channel their expertise
to develop designs. The platform has grown to more than 5,500 participants and includes participants from
all fields of training and career stages in all US states and five continents.

6
The Internet-of-Things or “IoT” refers to a wide set of largely yet-to-be-imagined products, services, and systems that connect
machines, infrastructure, consumer products, and other things while making use of the intelligence created by data collection and
networking. IoT is widely expected to generate an impact as wide as the Internet, impacting occupations and firms across the
economy in a “Fourth Industrial Revolution” or “Second Machine Age” (Brynjolfsson and Mcafee, 2014). These technologies are
predicted to affect most sectors across the economy (Patel et al., 2017).

10
3.2 Timeline and Procedure

Embedding our experiment in the initial launch of the platform allowed us to be part of a one-time
process of mass communication with 97,678 potential adopters under highly controlled conditions. By
tracking responses of this known set of individuals for whom descriptive data are available, the effect of
random assignment to the competition collaboration treatments can be assessed.

3.2.1 Randomized Assignment to Competition vs. Collaboration Treatments

An introductory email introduced individuals to the opportunity, inviting them to the platform to
learn more details and join. Invitation emails were sent over 60 business days, with individuals randomly
assigned to each day. Spreading invitations over time allowed the engineering team to monitor platform
scaling while minimizing the risk that any one day could produce eccentric results.

Individuals were randomly assigned to competition or collaboration treatments with the aim of
influencing subjects’ perceptions of interactions they would encounter with other participants. Subjects
were assigned to receive different emails depending on their treatment. Within the email invitation they
could click on a link in the email to send them to the platform (“Click HERE to learn more and to sign up
to the platform”). The platform landing page also provided additional details including high-level
descriptions of the technical nature of IoT and the opportunity to create new products and applications. The
version of the platform landing page to which they were sent also depended on whether they were in the
competition or collaboration treatments, where these platforms only differed in the sense of repeating the
distinct language used in the emails.

The details of both the emails and landing pages varied in the same way, depending on the
treatment. What follows are the relevant subset of distinguishing words from both the email and the
platform for the competition treatment:

“This is a two-sided competitive platform to ideate and innovate ‘smart’ Internet-

of-Things (IoT) products and services—using hardware, software, networking, data and
algorithms.

On one side of the platform, companies seek solutions to their IoT innovation
challenges. On the other side, you will compete with other participants to solve IoT
innovation problems of companies for cash and other benefits.”

For the collaboration treatment, the bolded words were replaced with “collaborative platform” and
“you will work within a team to solve IoT innovation.” Therefore, the content and length were virtually
identical for both treatments, except for these key bolded words.

11
 Individuals who did not act on the email were sent a single reminder seven days later (also at 10
am EST). We explicitly added a line in the email stating that the invitation was not transferable or to be
forwarded. Nonetheless, there were 26 instances of forwarding of emails; we were able to detect such
forwarding, and we removed these instances from our analyses.

Once on the platform, individuals could choose to participate by clicking on a red “join and
participate” button prominently displayed on the site. Users were prompted to log in with their LinkedIn
passwords. This sign-in and authentication allowed their LinkedIn picture, name, link, and brief blurb to be
used to create their platform profiles.

The analytics generated by the email management system (Mailchimp) and the platform, allowed
us to determine the number of people among the 97,678 receiving invitations who positively double-clicked
to open the invitation they received and then proceeded to the platform. Positive confirmation of opening
rates for females and males were statistically identical, with 46.5 percent of females and 45.9 percent of
males opening. The true rate at which individuals read emails will be higher, as the email management
system tracked active clicking on emails. However, most modern browsers to read an email without double-
clicking on it to open the email. Further, even the title of the email mentioned the institution’s “competitive
platform” or “collaborative platform” for IoT product development. Therefore, even reading the title
implied exposure and treatment. We cannot be sure of the true share of emails that were not read. To the
extent there is some fraction of our study population who did not, in fact, read the invitation, we should
expect weaker statistical results in our analysis.

3.2.2 Controlled and Independent Bilateral Communications

A crucial aspect of running this experiment was to carefully control the information environment.
Running the experiment during the initial launch campaign, at a time when prospective adopters had no
prior experience or preconceived notions of the platform, allowed us to selectively emphasize the
competitive or collaborative aspects of the platform in our communications. By implementing alternative
emails and entirely separate platforms with distinct communications, we could present separate and
controlled information to each subject, depending on the treatment. Importantly, the information presented
in both the competitive and collaboration treatments was also truthful, as the platform and program within
which subjects would eventually participate would involve competition among teams.

In controlling the information environment, it was also important that the only form of
communication and marketing of this initial 60-day campaign be this sequence of bilateral interactions with
the platform that would occur prior to opening the platform to development events. For example, there was
no simultaneous broad-based marketing campaign. We were also aware of the possibility that individuals

12
within the experiment might potentially, themselves, share messages with their peers. If messages were to
be shared with other subjects in our study population, resulting in the mixing of treatments, this would
likely weaken any treatment effect by mixing messages. We attempted to minimize such interactions by
adding a line to all communications indicating that the subjects should not forward any messages. Using
tracking and analytics, we found that among the tens of thousands of communications, there were only 26
instances of forwarding. We removed these instances from our data set; doing so had no effect on the results.
We also considered the possibility that any interactions and information diffusion occurring between
individuals could be more likely toward the end of the 60-day campaign. However, including time controls,
even individual-day fixed effects, showed no effect on results.

3.3 Links with the Prior Experimental Literature

The experimental protocol here is based directly on previous research in the literature on
competition and gender (Section 2.1). To observe whether individuals participate conditional on the
organizational regime, our experimental protocol follows prior work from Samek (2019), for example. The
advantage of this approach is that it allows us to assess preferences to participate under each regime
independently (relative to an outside option of not participating). The alternative in the literature is to recruit
only those subjects who are willing to participate (as when recruiting to a campus lab experiment) and then
ask individuals to choose the regime they would prefer, effectively providing a rank-order preference for
one regime versus the other, rather than an independent evaluation of each regime. Another advantage of
studying whether people participate or not under either regime independently is that we are able to explicitly
observe the selection process into participation in the exercise, rather than having the selection process be
part of recruitment to the experiment.

Our approach also follows the approach of prior studies that compare willingness to work under
competition with willingness under collaborative regimes (e.g., Abramo et al. 2013; Dargnies 2012; Healy
and Pate 2011; Kuhn and Villeval 2015), as opposed to comparing preferences with a piece-rate regime
(e.g., Niederle and Vesterlund, 2007). This approach has natural advantages in our context, as it is simply
not feasible to implement a piece-rate for innovation, development, and uncertain learning and problem-
solving in our context of developing novel IoT projects. In practice, we also exploited the organization of
competitions on the platform between teams as a means of focusing on competitive or collaborative and
team-based treatments while scrupulously avoiding deception of any of our subjects in any of our
communications.

Prior research tends to operationalize competition and collaboration as a matter of either a small
(often two-person) winner-take-all rank-order tournament for competition or averaging performance

13
payoffs for collaboration (see Section 2.1). Here, we simply directly use the words “competitive” and
“collaborative” to describe the organizational context, allowing subjects to discern for themselves what
should be meant by these terms in an organizational context where they carry out work and learning akin
to activities in industry.

The largest and most important departure from the bulk of previous research in the literature
focused on gender-based differences in competitiveness is that ours is a field experiment rather than a lab
experiment involving synthetic tasks. Among the field experiments in this study, this paper comes closest
to that of Samek (2019), in the sense of examining who applies for an office support job, conditional on
organizational regime. She focuses on a study population of students. This paper is also arguably more
closely related to Flory et al. (2015), as it also involved a large field experiment, involving nearly nine
thousand subjects; although that study focused on only a market for administrative assistants. Here, we
examine a considerably larger number of college-educated adults from all fields.

4 Data, Variables, and Descriptive Statistics

Our study population included 97,678 people in the risk set of potential platform adopters. (A
broader population was targeted by the platform, but this number includes the individuals identified and
assigned to treatments during the experiment.) Data for these individuals was kindly provided by the
institution, matched, and anonymized before the launch. The data set included data on majors and
graduation year—and academic performance (cumulative undergraduate GPA) for 13,926 individuals from
cohorts after 2009. Further, we used an online API-based service ([Link] to
probabilistically assign the vast majority of names in our dataset to genders. Table 1 describes the main
variables used in the analysis. Figure 1 summarizes the distribution of individuals and the gender
breakdowns by GPA and cohort (reported in terms of age).

<Table 1>

The participation rates of men and women in STEM fields were 3.97 percent and 4.47 percent,
respectively; in non-STEM fields, they were 2.62 percent and 1.98 percent, respectively. Participation was
higher for men than for women, and this was true across ages and academic performance, shown in Figure
1.7

<Figure 1>

7
Gender differences in participation are also observed when stratifying between undergraduate, graduate, and alumni subgroups.

14
 Therefore, the initial platform launch attracted thousands of adopters from the risk set. According
to university officials, these participation rates were (very) high relative to most other university programs
and clubs directed at both students and alumni. These levels of participation are also higher than most
comparable prior studies. For example, Samek (2019) advertised a research assistant position to 35,462
university students and received applications from 1.38 percent, from which she derived inferences by
comparing differences in participation across treatments. These adoption rates might also be regarded to be
high in relation to, say, a typical platform launch by a private company, who might typically not expect to
achieve multiple percentage points of adoption among the general population in the first couple months of
operation. By contrast, those studies that non-randomly select their participants conditional on an
expression of interest in participating, of course, tend to find higher eventual participation rates (e.g., Flory
et al., 2015; Niederle and Vesterlund, 2007).

Although randomization should ex-ante lead to equal groups in treatment and control, Table 2 tests
and confirms that observable characteristics of those assigned to competition and collaboration treatments
were virtually identical.

<Table 2>

5 Results

5.1 Baseline Study Population Average Gender Differences

This first subsection begins by replicating findings from previous research.8 We show that response to
competition (in terms of percentage difference in participation rates between the competition treatment and
collaboration treatment) is greater for women than for men in the study population average. The magnitude
of the gender difference is only marginally statistically significant and smaller than prior lab experimental
research (Section 2).

Model (1) of Table 3 reports gender differences in participation rates, regressing the participation
indicator on the gender dummy Female and constant in a linear probability model. Heteroskedasticity robust
standard errors are reported.9 Coefficients are expressed as percentage points (i.e., 100 = 100 percent). The
constant term, or probability of males participating, is 3.19. The coefficient in the female dummy is highly

8
Instructions for accessing data and code to replicate the analysis reside on the Open Science Framework at the following link
[Link]
9
Linear models ease the interpretation of interactions that are central to this investigation.

15
significant -0.67 percentage points (p = 0.0001).10 This result means that women are 21 percent (i.e., -.67 /
3.19) less likely to participate on average than are men, overall.11

<Table 3>

Model (2) adds the indicator for the dummy variable for Competition to the model, revealing that
those assigned to the competition treatment are 0.71 percentage points (p = 0.0001) less likely to participate
than those assigned to the collaboration treatment. Model (3) introduces an interaction between competition
and gender. The coefficient of this interaction is statistically zero, which means that there is no absolute
change difference between men and women. These basic differences between men and women’s responses
are summarized an equivalent comparison of unconditional participation rates in Panel I of Figure 2.

<Figure 2>

Therefore, the absolute difference in participation differences across competition and collaboration
treatments is statistically the same for men and women. However, as baseline levels of participation differ,
we also test whether the percentage differences across competition and collaboration for men and women
statistically differ. A comparison of percentage changes arguably also provides a more direct comparison
with prior experimental results (Section 2), in which subjects necessarily proceed to participate, and the
percentage of subjects choosing one work regime or another is reported. Figure 2 summarizes the results of
model (3) (equivalently, Figure 2) in terms of percentage changes i.e.,:
!"#$%&%'"$%()(+(,'-$%$%())/!"#$%&%'"$%()(+(00"1(#"$%())
!"#$%&%'"$%()(+(00"1(#"$%())
. The difference in women’s participation from
2!"#$%&'&'"( 32)%#*+% × !"#$%&'&'"(
collaboration to competition treatments is &()4$")$32)%#*+%
= −24.86 percent. For men, this
2!"#$%&'&'"(
difference in participation rates is &()4$")$
= −20.08 percent. A Wald test of the nonlinear restriction
2!"#$%&'&'"( 2!"#$%&'&'"( 32)%#*+% × !"#$%&'&'"(
that these percentages are the same, 05 : = , is significant at p =
&()4$")$ &()4$")$32)%#*+%

0.097. These differences are summarized graphically in Panel II of Figure 2

These population-average results would appear to be robust to numerous alternative controls and
specifications (aside from accounting for field differences, as in Section 5.2). The pattern of negative
coefficient estimates on Female and Competition., implying a greater negative percentage response to
competition among women, does not change when controlling for a series of dummies corresponding to the

10
We ran an analysis in which we replaced participation here with a measure of participation in other university-sponsored activities
(counts of university clubs and activities listed on Linkedin). The patterns are starkly different, confirming that the patterns observed
here do not simply reflect some general inclination to participate in university activities, in general.
11
A low adjusted-R^2 statistic is consistent with high variance with a low mean incidence of 1’s.

16
day an individual was invited to join (model 4), whether subjects are students or graduates (model 5), or
age (equivalently, cohort) (model 6). We also find similar patterns when controlling for college cumulative
grade point average (GPA) (model 8) or all controls at once (model 9). Note that, since tests with college
GPA data include only the 13,926 individuals from cohorts after 2009, for which we have GPA data, we
also report main model results with just this subset of observations (model 7) to provide a more direct
comparison of results in models (8) and (9). The coefficients in model (7), estimated in the subsample and
those of model (3), are the same in sign and similar in the relative size ratios between the model coefficients.
However, the magnitudes are larger in model (7), as this subsample comes from more recent graduates,
where overall participation is higher in younger cohorts (see Figure 1). This is consistent with possible
higher returns or lower opportunity costs to participate in this exercise among younger cohorts.

5.2 Main Results: Differences Between Fields

This subsection investigates our main prediction that gender responses to competition differ by
field (Section 2.3).

5.2.1 Controlling for Fields

As a first step, we investigate whether simply controlling for a series of dummies for field of
training affects the earlier result of Section 5.1. In Table 4, models (1) and (3) re-report population average
results, both without and with the competition treatment effects. Models (2) and (4) then re-estimate these
models with field dummies. Whereas results of Section 5.1 were highly robust to a wide range of alternative
controls and tests, here we see that including controls for field fundamentally alters key coefficient
estimates. Most striking, comparing models (1) to (2) shows that controlling for field differences leads the
coefficient on Female to reverse its sign. Comparing models (3) and (4) also finds this reversal of the sign
on the Female coefficient, once controlling for field-fixed effects.

The coefficients on the field dummies are reported in Table 4 ordered by their magnitude. The
estimates indicate that those in the STEM fields of computer science, engineering, and science are more
willing to participate than those in other fields. This is consistent with product development in the Internet-
of-Things perhaps being more immediately accessible or interesting to those in STEM fields (even despite
the considerable efforts to make this activity relevant, accessible, and interesting to all fields, see Section
3.1).

<Table 4>

17
5.2.2 Gender Differences in STEM Versus Non-STEM Fields

We now test our specific hypothesis of Section 2.3, that gender differences in STEM fields should
be significantly smaller than those in non-STEM fields. Our approach is simply to stratify results from the
above model by those in STEM versus non-STEM fields. (Recall, STEM is defined in this context to include
engineering, computer science, and sciences.) Model (1) of Table 5 begins by simply restating overall
results, controlling for fields, as a basis for comparison. Models (2) and (3) re-estimate stratified results
stratified by subsets of non-STEM and STEM observations. Model (2) reports that in non-STEM fields, the
difference between participation under competition versus collaboration for women is 0.47 percentage
points more negative than for male counterparts in the same fields, significant at p = 0.05 (i.e., the
coefficient on Female ⨉ Competition.) Model (3) reports that in STEM fields this difference of
participation versus competition is not statistically different between women and men. Note too, in model
(3), that those in STEM fields are generally more willing to participate in the collaboration regime than
they are in the competition regime (i.e., negative coefficient estimated on the Competition variable). These
responses, stratified by men and women and by non-STEM and STEM fields are graphically summarized
in Panel I of Figure 3. Model (4) summarizes the same information of model (2) and (3) or Panel I of Figure
3 within a single regression specification. (This alternative compact way of summarizing results will be
referred in robustness results, below.)

<Table 5>

<Figure 3>

We also test whether the percentage differences across competition and collaboration for men and
women in STEM and non-STEM fields are significantly different from one another. Panel II of Figure 3
presents our results of models (2) and (3) as percentage differences in participation rates, stratified by
!"#$%&%'"$%()(+(,'-$%$%())/!"#$%&%'"$%()(+(00"1(#"$%())
gender and STEM versus non-STEM fields, i.e., !"#$%&%'"$%()(+(00"1(#"$%())
.

As summarized in Figure 3, men and women in non-STEM fields show large differences in their
response to the competition treatment, with a difference in participation of -3.93 percent and -25.49 percent
in relation to the collaboration treatment, respectively. A Wald test of the nonlinear restriction that these
2!"#$%&'&'"( 2!"#$%&'&'"( 32)%#*+% × !"#$%&'&'"(
percentages are the same, 05 : 6 = 6."(/0123 )%#*+%,!"++*5"6*&'"(
in model (4),
."(/0123 )%#*+%,!"++*5"6*&'"(

is significant at p = 0.01.

Consistent with our main prediction, and in stark contrast to the above patterns in non-STEM fields,
men and women in STEM fields respond similarly to the competition treatment. The difference in
participation for men and women in response to the competitive treatment is -19.62 percent and -23.63

18
percent, in relation to participation in the collaboration treatment. A Wald test of the nonlinear restriction
2!"#$%&'&'"( 2!"#$%&'&'"( 32)%#*+% × !"#$%&'&'"(
that these percentages are the same, 05 : 6 = 6."(/0123 )%#*+%,!"++*5"6*&'"(
, is
."(/0123 )%#*+%,!"++*5"6*&'"(

not statistically significant in the least (i.e., p = 0.73).

5.2.3 Robustness of Main Results

A series of additional tests reported in Table 6 assesses whether the earlier established patterns of
results of the earlier section, and particularly those summarized in model (4) of Table 6, are robust to
controls for various factors influencing benefits and costs of participating. We indeed find the results to be
robust.

For example, model (1) begins by re-reporting the earlier patterns for comparison. Model (2) tests
the possibility that the age and cohort distribution of men and women could vary in some way between
fields in ways that lead to spurious findings, adding age as a control. There is no effect on the coefficients
of the main model. Controlling for a quadratic time trend or distinct time trends for each gender does not
affect results (not reported).

We also test whether variation in opportunity costs could be a spurious factor. To test for this
possibility, we include a measure of an important source of opportunity costs, being in one’s child-rearing
years.12 Model (3) includes an indicator switched on for ages 29 through 35 along with an interaction with
gender. The years of child rearing are associated with lower participation, but the main results remain
unchanged.

<Table 6>

Model (4) tests whether a measure of aptitude (academic achievement) might somehow be a
spurious factor in the model. We first re-estimate the main model on the subsample for which we have GPA
data to confirm similar patterns (Model 4). Model (5) then re-estimates the model while including GPA and
GPA interacting with gender. A first point to observe is that accounting for GPA does not alter main
coefficients, comparing coefficient estimates in models (4) and (5). In this sense the results are robust to
including GPA. We remark, too, that the considerably larger coefficient on the three-way interaction, STEM
Field ⨉ Female ⨉ Competition, suggests even greater (more positive) differences in response to

12
The average college-educated American women has her first child at the age of 30.3 (Bui and Miller, 2018). We did not find a
definitive source of the distribution over time, but results are insensitive to shifting this age range by 3 years on upper or lower
bounds. Note, in this cross-section, the age is also the graduation cohort—and therefore open to multiple interpretations.

19
competition among women in STEM for whom we have GPA data (i.e., younger cohorts).

5.3 Gender Differences Across STEM Subfields

This section re-estimates our model over 17 more narrowly-defined STEM subfields in order to
determine whether we continue to see similarities in men’s and women’s responses to competition at the
level of individual subfields.

In short, despite considerable variation in overall participation rates across individual STEM
subfields, we see that responses to competition among men and women are similar when examined at the
level of subfields. To study patterns at the level of subfields , we worked with university staff familiar with
the dozens of degrees offered in these fields to map them to subfields, listed in Table 7. Degrees with
ambiguous mapping were removed from this subfield analysis. (Computer science during the study period
does not break down into separate degrees or subfields.)13 We kept within the analysis here all STEM
subfields with at least 200 individuals.

Table 7 reports model estimates of participation rates for each subfield. Figure 4 reports the
corresponding percent changes in response to competition for both men and women. We must be somewhat
cautious in interpreting the results, as this fine-grained stratification of the analysis reduces statistical
power, and we might expect large differences in gender dynamics between STEM subfields.

<Table 7>

<Figure 4>

Consistent with previous findings of an average negative response to competition of approximately

20 percent, the coefficient estimates in Table 7 are largely negative, and the percent responses to
competition shown in Figure 4 are also largely negative. Examining the percentage changes in response to
competition in Figure 4, we found no evidence of statistical significance of the subfields in 16 of 17
subfields. The only gender difference that is statistically significant at conventional levels in computer
engineering (p = 0.01), and women are much more averse to competition (-83 percent change in response
to competition for women, versus -15 percent change for men).14

Although we must be cautious in interpreting statistical significance, particularly now as we stratify

our analysis into fine-grained subsets, what is most strikingly consistent with the earlier results and

13
Estimating the model separately for computer science, engineering, and sciences shows similar results as STEM, overall, with
men and women responding with statistically similar percentage changes in participation under the competition treatment relative
to the collaboration treatment.
14
The next most statistically significant gender difference was in biology, where differences are significant at p = 0.18.

20
theoretical hypotheses is that it is overwhelmingly the case that men’s and women’s responses to
competition appear to be much more similar within subfields than between subfields. As can be seen in
Figure 4, in most cases, the responses of men and women follow the same sign, and even the magnitudes
of the responses are very similar within the individual subfields. These patterns are broadly consistent with
the general claim and earlier documented patterns suggesting that men and women sorting into STEM fields
are more likely to respond similarly to competition.

6 Summary and Discussion of Results

6.1 Main Finding: Smaller (Negligible) Gender Differences in STEM Fields

We found that in non-STEM fields, gender differences in responses to competition mapped closely
to findings of gender differences reported in the prior research (Section 2). Non-STEM men and women
participated at similar rates in the collaboration treatment, but differed significantly in responses to the
competition treatment. Whereas non-STEM men participated at statistically similar rates under competition
as under collaboration (with participation rates just 3.93 percent lower under competition), non-STEM
women significantly reduced their participation rate by 25.49 percent. These patterns are presented in detail
in Figure 3 and in Table 5. The observation of relative aversion to competition among women in non-STEM
fields is consistent with findings of the prior lab research.

In Section 2.3, we hypothesized that gender differences in STEM fields should be smaller than
those in non-STEM fields. Indeed, we found that differences in participation for STEM men and women in
the competition treatment, relative to their participation rates in the collaboration treatment were
statistically identical (19.62 percent lower participation for STEM men under competition; and 23.63
percent lower participation for STEM women under competition). Again, these patterns are presented in
detail in Figure 3 and in Table 5.

Should we interpret the statistically identical responses of men and women in STEM fields as
reflecting truly small differences, or just an inability to detect differences with sufficient statistical power?
A variety of patterns suggests our tests are sufficiently powered and the differences are indeed negligible.
Most starkly, in instances where we do in fact find effects, these effects are large and highly statistically
significant—including differences between men’s and women’s responses in non-STEM fields. In the case
where we do not find a significant difference—between men’s and women’s responses in STEM—this is
not because of losing precision of estimates and standard errors “blowing up.” Rather, the point estimates
are simply small (e.g., compare models (3) and (4) in Table 3). Moreover, given the overall participation

21
rates are higher in STEM, the treatment effect might have been expected to mechanically generate greater
variance across women and men in STEM fields, making it easier to detect gender differences had these
been large.15

Of course, these findings of no statistically discernible gender differences in the responses to

competition in STEM fields were established in data covering 31,923 STEM subjects. Just this subset of
our data is orders of magnitude larger than the bulk of prior research (Section 2). We also found estimates
to be robust to a series of controls (Table 4). Re-estimating the model with 17 narrower STEM subfields
also revealed mostly similar results (Section 5.3). Moreover, it is only precisely where we theoretically
predict to find small(er) differences that we indeed do find small differences. Therefore, all evidence and
arguments point to gender differences being small among STEM subjects.

Nonetheless, we should be cautious in interpreting these results as a finding of true zero average
gender differences in STEM, or that this difference is necessarily zero generally. The point estimates of
percent changes in participation in response to competition were still lower for women than for men (Figure
4), despite the lack of statistical significance. Further, while the nature of the work was intended to be
representative of the sort of work carried out in the industry (Section 3.1), these results must be understood
in relation to the activity under study here. The experiment was not designed to examine the effects of
variation in the task. Further, while our examination of 17 subfields showed remarkably similar
participation rates of men and women within particular fields, there were isolated exceptions.

6.2 Secondary Finding: STEM Subjects Prefer Collaboration over Competition

Although our primary research question focuses on relative gender differences in responses to
competition, our research design also allowed us to gain insight on general preferences for working under
competition versus collaboration in a widely sampled study population, more generally. Here, we find that
both men and women in STEM fields have an absolute preference for working under a collaborative regime,
as summarized in Figure 3 and Table 5. STEM men participated 19.62 less frequently under competition—
a 3.60 percent participation rate under competition versus 4.48 percent participation rate under
collaboration. STEM women participated 23.63 percent less frequently under competition—a 4.37 percent
participation rate under competition versus 5.56 percent participation rate under collaboration.

15
We thank the editors and a referee for this point.

22
6.3 Interpretation of Results

Here we review several explanations for the reported patterns.

6.3.1 STEM Sorting, Socialization, Training and Experience

It is widely presumed sin research on gender and competition that individuals may sort into distinct
fields and occupations based on competitiveness. We hypothesized that we should observe smaller gender
differences in competitiveness in STEM. A simplest interpretation that follows from discussion in the prior
research (Section 2) is that men and women in STEM should be more similar, as they will each tend to have
a relatively high preference for competition. But, rather than find that STEM men and women generally
preferred competition, we found that opposite: that both STEM men and STEM women had a greater
willingness to participate under collaboration in our context. Therefore, a more nuanced explanation is
required than simply a “preference” for competition.

One possible explanation for similar responses to competition among STEM men and women, but
a general greater willingness to participate under collaboration is suggested by prior research on dynamics
in STEM fields. For example, we might expect that student and workers in STEM fields could acquire more
moderated or contingent responses to both competition and to collaboration. On the one hand, competition
is well documented to be a regular feature of work in industrial firms, such as in laboratories, both within
and between organizations (e.g., Ali et al., 2018; Lam, 2011). On the other hand, the demanding nature of
problem-solving and the advance of knowledge demands in STEM fields has made it increasingly common
for technical problem-solving to be carried out in teams—whether in the lab, in the product development
department, within the software development community, and so forth (e.g., Jones 2009; Mortensen 2014;
Shah 2006). In this sense, the ability to work under both competition and collaboration is a regular aspect
of modern STEM work (e.g., Menekse et al., 2017). It is also plausible in a context where both competition
and collaboration are regular work modes that STEM workers might perhaps, to some degree, be able to
recognize the organizational approach most suited to a certain context and set of circumstances. For
example, in our context the task was one of (inherently multi-disciplinary) product conception and
development, which could have favored team-based collaboration. Therefore, it is possible that those in
STEM could be more competitive in a sense of being open to and able to perform under competition (and
collaboration, too), rather than simply inherently “preferring” one work mode over another.

Other research documenting gender dynamics in STEM fields suggests still other reasons for
smaller gender differences in STEM. Descriptive case study evidence suggests that STEM women may be
more ambivalent to both competition and collaboration. For example, female engineers have been reported
to exhibit ambivalence or aversion to teams and collaboration where they are subject to gender-stereotyped

23
expectations and work role assignments (Cech et al., 2011; Cheryan et al., 2017; Choi, 2013; Coffman,
2014a, 2014b; Foor et al., 2013; McIlwee and Robinson, 1992b; Moss-Racusin et al., 2018; Silbey, 2015,
2016), and competitions have been reported to be an important means of engaging and socializing women
and girls into STEM fields (Friesel and Timcenko 2011; Kuyath and Yoder 2004; Notter 2010). On their
own, these arguments imply that STEM females may be relatively moderated or contingent in their
willingness to work under collaboration and competition.

Other strands of research directly focused on STEM students and workers argue that STEM fields
exhibit cultures and norms promoting practices and norms associated with masculinity, including
competitiveness and assertiveness (Cheryan et al. 2009; Etzkowitz et al. 2000; Kvande 1999; McIlwee and
Robinson, 1992). Indeed, women scientists and engineers have been documented to sometimes adopt
practices to conform with their field’s culture (Kvande 1999; Ong 2005; Powell et al. 2009), including
emulating competitive behaviors (Cheryan et al. 2017). These arguments, on their own, would suggest that
women who remain in STEM fields may then potentially converge towards norms of competitiveness or
assertiveness. While there could be some convergence of men and women’s predispositions—either
through sorting or socialization—again, it is possible that the factors shaping willingness to work under
competition and collaboration could be more manifold, nuanced, and contingent, as noted above.

6.3.2 “Intellectually Distant” Non-STEM Men and Over-Confidence

Apart from our main finding related to STEM subjects, our results also allow us to consider the
more regular results observed in non-STEM fields and discern plausible explanations. For example, men in
non-STEM fields responded to competition and collaboration treatments in statistically identical ways,
suggesting an indifference between the two treatments; whereas non-STEM women participated less under
competition (model (2) of Table 5). On the one hand, these results might simply be considered consistent
with women responding more negatively to competition than men do, as in the prior research (Sections 1
and 2).

However, considering other patterns we find here, we might also observe that this indifference of
non-STEM men between competition and collaboration is consistent with over-confidence. This is
especially so if there were advantages to collaboration in this context of (inherently multi-disciplinary)
product development. Further, working in the Internet-of-Things might have created particular
disadvantages for those without technical training. Over-confidence among men has been previously
suggested as an explanation for gender differences in competitiveness, particularly so among lower-skilled
men (e.g., Niederle and Vesterlund 2007). Of course, men in non-STEM fields cannot as a group be judged
to be lower-skilled; but in this instance, they may be less familiar with or more “intellectually distant” from

24
the task at hand. Therefore, it is possible that more intellectually distant males are more susceptible to
misconstruing the work to be performed and to exhibit overconfidence in their choices.

This explanation of “intellectually distant” men being over-confident is consistent with prior
research finding that domain-specific knowledge and familiarity with a task can be highly important in
shaping responses to competition (Wieland and Sarin 2012). Comeig et al. (2016) also show that experience
and expertise in an area alters attitudes toward competition by moderating confidence. Kamas and Preston
(2012) found in their lab study of 310 students solving mathematical problems that while women were less
willing to choose competition than men, the differences were not statistically significant among science and
engineering students, who presumably have greater training in mathematics.

6.3.3 Non-STEM Women, Gender-Role Congruity, and Self-Efficacy

Apart from possible over-confidence of non-STEM men, it is possible that non-STEM gender
differences also arise from a range of previously hypothesized reasons for an aversion to competition, at
least among non-STEM women. For example, women in non-STEM fields might perceive the
technologically intensive product development to be a male-oriented task (Nosek et al. 2002, 2009; Su et
al. 2009), and consequent stereotype threat could result in a diminution in self-confidence and resulting
willingness to participate under competition (Dreber et al. 2014; Woodcock et al. 2012). Gender
stereotyping of the task might also have led to an expectation of disproportionate participation by men,
which could itself reduce willingness to compete (Booth and Nolen 2012; Gneezy et al. 2003; Lundeberg
et al. 1994).

We emphasize, too, that apart from these prior hypotheses from the literature, it is also possible
that non-STEM women are simply responding somewhat sensibly to the available choices in our context.
Consider again that if there are advantages to collaborating in this product development task, there might
especially be advantages to non-technically trained individuals to team-up and collaborate. Under this
interpretation, it should not then be a surprising that non-STEM women have a lower participation rate in
the competition treatment.

6.3.4 Demand-Side Explanations

Another more general point to raise in interpreting our results is that worker participation in the
workforce reflects two-sided matching between worker supply and employer demand. In contrast, this
experiment here (as in others in the literature) was designed to focus on the supply-side willingness of
individuals to participate or not. (See Murciano-Goroff [2018] and Price [2011] for analyses of demand-
side employers.) Nonetheless, it remains plausible that subjects’ decisions to participate could have
anticipated demand-side features of the workplace (Fernandez-Mateo and Kaplan 2018).

25
 For example, Goldin (2014) hypothesized that women’s participation and success in the workplace
is often curtailed by rigid and long work hours and a need for “facetime.” The collaborative product
development platform used in our context was designed with the intent to minimize the costs of inflexible
time requirements, including an ability to interact asynchronously. Closer to the crux of this experiment, if
collaboration was viewed as requiring more inflexible “facetime” costs, this might have had the effect of
dissuading female participation. In Table 4 of Section 5.3, we found no evidence that the competition
treatment was affected by controlling for a measure of opportunity costs, providing some assurance that
there is no such demand-side effect.

More generally, any perceptions of costs or benefits that apply asymmetrically to men and women
who interact with the treatment effect could alter our results. For example, women in STEM could
potentially have expected higher returns for participating in this developmental activity if women have
disproportionately higher returns on education, receiving more call-backs and job offers (Williams and Ceci
2015). However, it is not clear that such an effect should necessarily vary with the competition and
collaboration treatments. Nonetheless, we cannot unequivocally rule out this possibility.

6.4 Additional Considerations in Interpreting Results

Of course, there are inherent challenges to interpreting studies on gender and competition, including
ours. As per the discussion above, there may be multiple nuanced mechanisms at work, even beyond those
typically considered within the prior literature (Section 2) or those reviewed here. We, therefore, emphasize
the demonstration of fact of (i) smaller gender differences in STEM; and (ii) STEM men and women
preferring collaboration over competition, above all.

It is also important to emphasize distinguishing research design choices here, relative to prior
research, and their potential implications on our interpretation. For example, the canonical approach in the
prior lab experimental research (Section 2) is to recruit lab participants, who then necessarily participate,
and then record under which regimes subjects prefer to work. By contrast, here, we randomly assigned each
subject to either competition or collaboration. Whereas the prior lab experiments effectively measure the
percentages of lab subjects who prefer one regime over the other; here, we measure whether individuals
choose to participate—at all—conditional on the regime. Therefore, we observe decisions on a different
“margin” than lab subjects who rank order which regime they prefer. Therefore, while our reporting of
percentage differences (i.e., Panel II in Figures 2 and 3) may allow slightly closer comparison with prior
lab studies, these are not the same outcome being observed.

In attempting to translate ideas of competition and collaboration to a field context from the lab, we
also differ in how we operationalized these concepts. For example, the prior experimental research (Section

26
2) tends to focus on small cash payments depending on rank order performance relative to one or few other
subjects to operationalize competition. This prior literature has also tended to operationalize collaboration
as a matter of determining payoffs in relation to averaged performance across one or few other subjects.
Given that competition and collaboration in the field might imply a wide range of things from the internal
culture to payoffs, to modes of work organization—we operationalized these concepts by simply making
the word “competition” and “collaboration” salient in the treatments.

Related, when attempting to isolate “preferences” for competition, the prior lab experimental
literature at times controls for closely related attributes by controlling, for example, particular measures of
risk aversion and other attributes. Controlling for these measures in analyzing lab experimental results is
typically done with the goal of showing that doing so does not upend results related to competition. While
our analysis goes further in controlling for and stratifying for several characteristics of individuals related
to knowledge and training, experience, age, and other factors, it should be noted that we did not control for
these sorts of behavioral variables.

As reviewed in Section 2.3, there are also inherent limits to studies on gender and competition,
even those featuring experiments, in unambiguously establishing the causal effect of gender. This is because
it is not possible to randomly assign gender, all-else-being equal. Although we benefitted here from a wide
range of controls and tests, and the ability to stratify on especially narrow subsamples, our own study cannot
avoid this criticism and should be interpreted with caution, as all studies in this literature. Therefore, strictly
speaking, we cannot rule-out the possibility that other factors causing sorting to STEM fields (e.g.,
confidence, aptitude, interest in puzzle solving, interest in “things,” etc.) are imperfectly controlled. This
limitation is necessarily true of all research on gender sorting and competition, and might be especially
important in highly selected study groups.

Perhaps most fundamentally, there are inherent limitations to making claims about gender
differences based on studies on small subsets of humanity. Although our study population is unusually large
and widely sampled, and covers adults rather than children or students, it is nonetheless only a population
of tens of thousands of individuals, with a relatively narrow range of human experience and circumstances,
educated and working in a particular society, in a particular period in history.

Finally, while we chose a product development activity and process that closely maps to such
activities in industry, we should underline that our field experimental protocol was not geared to examining
effects of variation in work task.

27
7 Conclusion
Are females more averse to working within competitive organizations than males? We investigated
this question by studying the willingness of individuals across all fields to participate in a platform-based
product development activity depending on whether individuals were randomly assigned to competition or
collaboration treatments. These treatments were based on framing the event in ways that either emphasized
competitive or collaborative interactions with other participants.

Our main novel contribution was to present evidence (Section 5) that gender differences in
competitiveness differ by field (Sections 5.2). Specifically, we theorized and found evidence that gender
differences are smaller in STEM fields than in other fields. In our context, not only were gender differences
appreciably smaller in STEM than in non-STEM fields, but these differences were statistically
indistinguishable from zero (see Section 6.1 summary). Our results were robust to cohort and age and
various controls for factors shaping benefits and costs of participation (Section 5.2 Table 4), and similar
patterns were found when examining finer-grained STEM subfields (Section 5.3).

It is also the case that STEM men and women were both more willing to participate under
collaboration than under competition. This result is inconsistent with a simplest characterization of sorting
on “preferences” for competition, where individuals who simply prefer competition sort into STEM fields.
As we discussed in Section 6, these results might reflect a more nuanced combination of, say, tolerance and
skills for competing under competition—rather than a simple preference. Further, given that work and
problem-solving and training in STEM fields routinely involves both competition and collaboration, it is
possible that we might expect a more nuanced, moderated and contingent willingness to work under either
competition or collaboration. In our case of an inherently multi-disciplinary product development task,
those in STEM perhaps expected superior performance when organizing collaboratively.

These findings begin to outline systematic boundary conditions in relation to the increasingly well-
known literature on gender and competition (Section 2.1 and 2.3). While our results contrast with many
prior replication studies showing gender differences in competitiveness in smaller studies, our theoretical
predictions were in fact derived from generally accepted presumptions of the prior research (Section 2.4).
Further, our results are consistent with prior results finding systematic gender differences in lab-based
studies if a majority of those subjects came from non-STEM fields of training. Indeed, even in broad study
population averages in our study, we found that competition effects were more negative for women. It was
only when accounting for fields that we observed striking differences. We detailed caveats and cautions
regarding our analysis and emphasized distinctions in our research approach in Section 6.4.

28
7.1 Organizational Implications

We began the introduction by observing that any behavioral regularities in competitiveness across
working-age adults could—if better understood—create opportunities for unlocking the talents of workers
through organizational design. Prior research, for example, has shown that in the context of gender
differences in a lab experiment, affirmative action can increase female productivity along with
representation (Niederle et al. 2013), and priming females subjects or alternative framings of participation
choices can also increase female participation under competitive treatments (Balafoutas et al., 2018; He let
al., 2019). We ourselves suggested that, in field settings, “tuning” the competitiveness of internal
organizational environments might also better harness workers’ talents and willingness to participate
(Section 1).

Our findings suggest that any practical organizational interventions will need to consider
contingencies (across fields, discussed in Section 6.1) and possible nuances (regarding mechanisms,
discussed in Section 6.3), as we find evidence contrary to claims of general gender differences. For
example, our results provided no indication that altering the competitiveness of a work environment in the
case of STEM workers will affect gender composition—at least among those already sorted into STEM
fields of training and careers. It remains plausible that managing the intensity of the competition faced by
boys and girls in STEM education earlier in life could affect sorting into STEM fields and thus affect the
composition of the STEM pipeline and workforce (Buser et al. 2014, 2016; Niederle and Vesterlund 2010).

29
References

Abramo G, D’Angelo CA, Murgia G (2013) Gender differences in research collaboration. Journal of
Informetrics. 7(4):811–822.
Almås I, Cappelen AW, Salvanes KG, Sørensen EØ, Tungodden B (2016) Willingness to compete: Family
matters. Management Science. Aug;62(8):2149-62.
Altonji JG, Arcidiacono P, Maurel A 2016 The analysis of field choice in college and graduate school:
Determinants and wage effects. In Handbook of the Economics of Education Jan 1 (Vol. 5, pp. 305-
396). Elsevier.
Andersen S, Ertac S, Gneezy U, List JA, Maximiano S (2013) Gender, competitiveness, and socialization
at a young age: Evidence from a matrilineal and a patriarchal society. Review of Economics and
Statistics. 95(4):1438–1443.
Antonovics K, Arcidiacono P, Walsh R (2009) The effects of gender interactions in the lab and in the field.
Review of Economics and Statistics. 91(1):152–162.
Apicella CL, Dreber A, Gray PB, Hoffman M, Little AC, Campbell BC. Androgens and competitiveness
in men. Journal of Neuroscience, Psychology, and Economics. 2011 Feb;4(1):54.
Balafoutas, L, Fornwagner, H and Sutter, M, 2018. Closing the gender gap in competitiveness through
priming. Nature Communications, 9(1), pp.1-6.
Baumann, O, JP Eggers, N. Stieglitz. "Colleagues and competitors: How internal social comparisons shape
organizational search and adaptation." Administrative Science Quarterly 64, no. 2 (2019): 275-309.
Boaler J (2016) Mathematical Mindsets: Unleashing Students’ Potential Through Creative Math, Inspiring
Messages and Innovative Teaching (John Wiley and Sons, New York).
Booth A, Nolen P (2012) Choosing to compete: How different are girls and boys?. Journal of Economic
Behavior Organization. Feb 1;81(2):542-55.
Boschini A, Dreber A, von Essen E, Muren A, Ranehill E (2019) Gender, risk preferences and willingness
to compete in a random sample of the Swedish population✰. Journal of Behavioral and
Experimental Economics. Dec 1;83:101467.
Breda T, Jouini E, Napp C, Thebault G (2020) Gender stereotypes can explain the gender-equality paradox.
Proceedings of the National Academy of Sciences. 2020 Dec 8;117(49):31063-9.
Brenoe AA, Zölitz U. (2020) Exposure to more female peers widens the gender gap in stem participation.
Journal of Political Economy 38(4).
Brynjolfsson E, McAfee A (2014) The Second Machine Age: Work, Progress, and Prosperity in a Time of
Brilliant Technologies (W. W. Norton and Company).
Buckert M, Schwieren C, Kudielka BM, Fiebach CJ (2017) How stressful are economic competitions in
the lab? An investigation with physiological measures. Journal of Economic Psychology. Oct
1;62:231-45.
Bui Q, Miller C (2018) The age that women have babies: how a gap divides America. The New York Times.

30
Buser T (2009) The Impact of Female Sex Hormones on Competitiveness. Discussion Paper, Tinbergen
Institute.
Buser T (2016) The impact of losing in a competition on the willingness to seek further challenges.
Management Science. 62(12):3439–3449.
Buser T, Niederle M, Oosterbeek H (2014) Gender, competitiveness, and career choices. The Quarterly
Journal of Economics. 129(3):1409–1447.
Buser T, Niederle M, Oosterbeek H (2020) Can competitiveness predict education and labor market
outcomes? Evidence from incentivized choice and survey measures. Evidence From Incentivized
Choice and Survey Measures (January 20, 2020).
Buser T, Yuan H (2019) Do women give up competing more easily? Evidence from the lab and the Dutch
math Olympiad. American Economic Journal: Applied Economics. 11(3):225–252.
Byrnes JP, Miller DC, Schafer WD (1999) Gender differences in risk taking: A meta-analysis.
Psychological Bulletin. 125(3):360.
Casari M, Ham JC, Kagel JH (2007) Selection bias, demographic effects, and ability effects in common
value auction experiments. American Economic Review. Sep;97(4):1278-304.
Cech E, Rubineau B, Silbey S, Seron C (2011) Professional role confidence and gendered persistence in
engineering. American Sociological Review. 76(5):641–666.
Cheryan S, Ziegler SA, Montoya AK, Jiang L (2017) Why are some STEM fields more gender balanced
than others? Psychological Bulletin. 143(1):1.
Choi KS (2013) Evaluating gender significance within a pair programming context. 2013 46th Hawaii
International Conference on System Sciences, 4817–4825.
Comeig I, Grau-Grau A, Jaramillo-Gutiérrez A, Ramírez F (2016) Gender, self-confidence, sports, and
preferences for competition. Journal of Business Research. 69(4):1418–1422.
Croson R, Gneezy U (2009) Gender differences in preferences. Journal of Economic Literature. 47(2):448–
474.
Dargnies M-P (2012) Men too sometimes shy away from competition: The case of team competition.
Management Science. 58(11):1982–2000.
Dittrich M, Knabe A, Leipold K (2014) Gender differences in experimental wage negotiations. Economic
Inquiry. 52(2):862–873.
Dohmen T, Falk A (2011) Performance pay and multidimensional sorting: Productivity, preferences, and
gender. American Economic Review. Apr;101(2):556-90.
Dreber A, Von Essen E, Ranehill E (2011) Outrunning the gender gap—Boys and girls compete equally.
Experimental Economics. 14(4):560–582.
Dreber A, von Essen E, Ranehill E (2014) Gender and competition in adolescence: task matters.
Experimental Economics. Mar 1;17(1):154-72.
Eckel CC, Grossman PJ (2008) Men, women and risk aversion: Experimental evidence. Handbook of
Experimental Economics. 1:1061–1073.
Etzkowitz H, Kemelgor C, Uzzi B. (2000) Athena unbound: The advancement of women in science and
technology. Cambridge University Press.
Fernandez, R. M., Campero, S. (2017). Gender sorting and the glass ceiling in high-tech firms. ILR Review,
70(1), 73–104.

31
Fernandez-Mateo I, Kaplan S (2018) Gender and organization science: Introduction to a virtual special
issue. Organization Science. 29(6):1229–1236.
Flory JA, Gneezy U, Leonard KL, List JA (2018) Gender, age, and competition: A disappearing gap?
Journal of Economic Behavior and Organization. 150:256–276.
[Link]
Flory JA, Leibbrandt A, List JA (2015) Do competitive workplaces deter female workers? A large-scale
natural field experiment on job entry decisions. Review of Economic Studies. 82(1):122–155.
Foor CE, Walden SE, Shehab RL, Trytten DA (2013) “We weren’t intentionally excluding them… just old
habits”: Women (lack of) interest and an engineering student competition team. 2013 IEEE
Frontiers in Education Conference (FIE), 349–355.
Friesel A, Timcenko O (2011) Competition-motivated teamwork and narratives to motivate girls for
engineering. 22nd EAEEIE Annual Conference.
Gee LK (2019) The more you know: Information effects on job application rates in a large field experiment.
Management Science. 65(5):2077–2094.
Gneezy U, Leonard KL, List JA (2009) Gender differences in competition: Evidence from a matrilineal and
a patriarchal society. Econometrica. 77(5):1637–1664.
Gneezy U, Niederle M, Rustichini A (2003) Performance in competitive environments: Gender differences.
Quarterly Journal of Economics. 118:1049–1074.
Gneezy U, Rustichini A (2004) Gender and competition at a young age. American Economic Review.
94(2):377–381.
Goldin C (2014) A grand gender convergence: Its last chapter. American Economic Review. 104(4):1091–
1119.
Günther C, Ekinci NA, Schwieren C, Strobel M (2010) Women can’t jump?—An experiment on
competitive attitudes and stereotype threat. Journal of Economic Behavior and Organization.
75(3):395–401.
Guo J (2015, January 2) Why do some studies show that women are less competitive than men? The
Washington Post, [Link]
studies-show-that-women-are-less-competitive-then-men/
Healy A, Pate J (2011) Can teams help to close the gender competition gap? Economic Journal.
121(555):1192–1204.
Holland PW (1986) Statistics and causal inference. Journal of the American Statistical Association.
81(396):945–960.
Holt-Lunstad J, Clayton CJ, Uchino BN (2001) Gender differences in cardiovascular reactivity to
competitive stress: The impact of gender of competitor and competition outcome. International
Journal of Behavioral Medicine. Jun 1;8(2):91-102.
Hunt A, Samman E. Gender and the gig economy. ODI Working Paper 546, 2019.
Inglehart M, Brown DR, Vida M (1994). Competition, achievement, and gender: A stress theoretical
analysis. Student Motivation, Cognition, and Learning: Essays in Honor of Wilbert. J. McKeachie,
311–330.
Iriberri N, Rey-Biel P (2019) Competitive pressure widens the gender gap in performance: Evidence from
a two-stage competition in mathematics. Economic Journal. 129(620):1863–1893.

32
Jetter M, Walker J. (2016) Gender in Jeopardy!: The Role of Opponent Gender in High Stakes Competition.
IZA Discussion Paper No. 9669, Available at SSRN: [Link]
Lundeberg MA, Fox PW, Punćcohaŕ J (1994) Highly confident but wrong: Gender differences and
similarities in confidence judgments. Journal of Educational Psychology. Mar;86(1):114.
Jurajda Š, Münich D (2011) Gender gap in performance under competitive pressure: Admissions to Czech
universities. American Economic Review. 101(3):514–518.
Kahn S, Ginther D 2017 Women and STEM. National Bureau of Economic Research; Jun 15.
Kamas L, Preston A (2012) The importance of being confident; gender, career choice, and willingness to
compete. Journal of Economic Behavior and Organization. 83(1):82–97.
Kesebir S, Lee SY, Elliot AJ, Pillutla MM (2019) Lay beliefs about competition: Scale development and
gender differences. Motivation and Emotion. 43(5):719–739.
Kivlighan KT, Granger DA, Booth A (2005) Gender differences in testosterone and cortisol response to
competition. Psychoneuroendocrinology. 30(1):58–71.
[Link]
Kleinjans KJ (2009) Do gender differences in preferences for competition matter for occupational
expectations?. Journal of Economic Psychology. 2009 Oct 1;30(5):701-10.
Knight J (2002) Sexual Stereotypes (Nature Publishing Group).
Kuhn P, Villeval MC (2015) Are women more attracted to co-operation than men? Economic Journal.
125(582):115–140.
Kuyath SJ, Yoder L (2004) Diversity in engineering technology: Competitions. Age. 9:1.
Kvande E (1999) In the Belly of the Beast' Constructing Femininities in Engineering Organizations.
European Journal of Women's Studies. Aug;6(3):305-28.
Lazear EP, Malmendier U, Weber RA (2012) Sorting in experiments with application to social preferences.
American Economic Journal: Applied Economics. Jan;4(1):136-63.
Lee, S, P Puranam. "Incentive redesign and collaboration in organizations: Evidence from a natural
experiment." Strategic Management Journal 38, no. 12 (2017): 2333-2352.
Liang C, Hong Y, Gu B, Peng, J., 2018. Gender wage gap in online gig economy and gender differences in
job preferences. Available at SSRN 3266249.
Lippa R (1998) Gender-related individual differences and the structure of vocational interests: The
importance of the people–things dimension. Journal of Personality and Social Psychology.
74(4):996.
McIlwee JS, Robinson JG (1992) Women in Engineering: Gender, Power, and Workplace Culture (SUNY
Press).
Moakler Jr MW, Kim MM (2014) College major choice in STEM: Revisiting confidence and demographic
factors. The Career Development Quarterly. Jun;62(2):128-42.
Morin L-P (2015) Do men and women respond differently to competition? Evidence from a major education
reform. Journal of Labor Economics. 33(2):443–491.
Mortensen M. (2014) Constructing the team: The antecedents and effects of membership model divergence.
Organization Science. Jun;25(3):909-31.
Moss-Racusin CA, Sanzari C, Caluori N, Rabasco H (2018) Gender bias produces gender gaps in STEM
engagement. Sex Roles. 79(11–12):651–600.

33
Murciano-Goroff R (2018) Missing women in tech: The labor market for highly skilled software engineers.
Work in progress. [Link] Edu/∼ Ravivmg/Papers/JMP. Pdf.
Niederle M, Vesterlund L (2007) Do women shy away from competition? Do men compete too much?
Quarterly Journal of Economics. 122(3):1060–1101.
Niederle M, Vesterlund L (2010) Explaining the gender gap in math test scores: The role of competition.
Journal of Economic Perspectives. 24(2):129–144.
Niederle M, Vesterlund L (2011) Gender and competition. Annu. Rev. Econ. 3(1):601–630.
Notter KB (2010) Is competition making a comeback? Discovering methods to keep female adolescents
engaged in STEM: A phenomenological approach.
Nosek BA, Smyth FL, Sriram N, Lindner NM, Devos T, Ayala A, Bar-Anan Y, Bergh R, Cai H,
Gonsalkorale K, Kesebir S. (2009) National differences in gender–science stereotypes predict
national sex differences in science and math achievement. Proceedings of the National Academy
of Sciences. Jun 30;106(26):10593-7.
Obloj, T, Zenger, T., 2017. Organization design, proximity, and productivity responses to upward social
comparison. Organization Science, 28(1), pp.1-18.
Ong M (2005) Body projects of young women of color in physics: Intersections of gender, race, and science.
Social problems. Nov 1;52(4):593-617.
Ors E, Palomino F, Peyrache E (2013) Performance gender gap: Does competition matter? Journal of Labor
Economics. 31(3):443–499.
Oster E (2019) Unobservable selection and coefficient stability: Theory and evidence. Journal of Business
and Economic Statistics. 37(2):187–204.
Paserman MD (2010) Gender Differences in Performance in Competitive Environments? Evidence from
Professional Tennis Players. NBER Working Paper 27154.
PBS News Hour (2015) Why we think the way we do about men, women and work | PBS NewsHour.
[Link]
gender-and-work
Powell A, Bagilhole B, Dainty A (2009) How women engineers do and undo gender: Consequences for
gender equality. Gender, work organization. Jul;16(4):411-28.
Pregaldini D, Backes-Gellner U, Eisenkopf G (2020) Girls’ preferences for STEM and the effects of
classroom gender composition: New evidence from a natural experiment. Journal of Economic
Behavior and Organization. 178:102–123.
Price CR (2011) Gender, competition, and managerial decisions. Management Science. 58(1):114–122.
[Link]
Puranam, P. (2018) The microstructure of organizations. Oxford University Press.
Reuben E, Sapienza P, Zingales L (2015) Taste for competition and the gender gap among young business
professionals. NBER Working Paper No. 21695.
Saccardo S, Pietrasz A, Gneezy U (2017) On the size of the gender difference in competitiveness.
Management Science. 64(4):1541–1554. [Link]
Samek A (2019) Gender differences in job entry decisions: A university-wide field experiment.
Management Science. 65(7):3272–3281.
Shurchkov O (2012) Under pressure: Gender differences in output quality and quantity under competition
and time constraints. Journal of the European Economic Association. 10(5):1189–1213.

34
Silbey SS (2015) The elephant in the room: Constraints and consequences of a four-year undergraduate
engineering degree. Engineering Studies. 7(2–3):164–160.
Silbey SS (2016) Why do so many women who study engineering leave the field. Harvard Business Review.
1–2.
Slack Website. (2019). In good company: How to create a culture of collaboration at work.
[Link]
Su R, Rounds J, Armstrong PI (2009) Men and things, women and people: a meta-analysis of sex
differences in interests. Psychological bulletin. Nov;135(6):859.
Sutter M, Glätzle-Rützler D (2015) Gender differences in the willingness to compete emerge early in life
and persist. Management Science. 61(10):2339–2354.
van Tuijl C, van der Molen JH 2016 Study choice and career development in STEM fields: an overview
and integration of the research. International journal of technology and design education. May
1;26(2):159-83.
Vugt M van, Cremer DD, Janssen DP (2007) Gender differences in cooperation and competition: The male-
warrior hypothesis. Psychological Science. 18(1):19–23.
Wieland A, Sarin R (2012) Domain specificity of sex differences in competition. Journal of Economic
Behavior and Organization. 83(1):151–157.
Williams WM, Ceci SJ (2015) National hiring experiments reveal 2: 1 faculty preference for women on
STEM tenure track. Proceedings of the National Academy of Sciences. Apr 28;112(17):5360-5.
Woodcock, A., Hernandez, P. R., Estrada, M., Schultz, P. (2012). The consequences of chronic stereotype
threat: domain disidentification and abandonment. Journal of personality and social psychology,
103(4), 635.
Wozniak D, Harbaugh WT, Mayr U (2014) The menstrual cycle and performance feedback alter gender
differences in competitive choices. Journal of Labor Economics. 32(1):161–198.
Wuchty S, Jones BF, Uzzi B (2007) The increasing dominance of teams in production of knowledge.
Science. May 18;316(5827):1036-9.
Wynn, A. T., Correll, S. J. (2017). Gendered perceptions of cultural and skill alignment in technology
companies. Social Sciences, 6(2), 45.
Zafar B 2013 College major choice and the gender gap. Journal of Human Resources. Jul 1;48(3):545-95.

35
 Tables

Table 1 Descriptive Statistics of Main Variables

Variable Mean Std. Dev. Description

Participation 0.03 0.17 An indicator variable that is switched to one for those
subjects who chose to participate
Competition 0.50 0.50 An indicator variable that is switched to one for those
randomly assigned to the competitive treatment (as opposed
to the collaborative comparison group)
Female 0.44 0.50 An indicator variable that is switched to one for female
subjects
STEM 0.33 0.47 An indicator variable that is switched to one for those in
Computer Science, Engineering, or Sciences
Student 0.08 0.27 An indicator variable that is switched to one for subjects
currently enrolled as students
CohortYear 1998.7 16.0 Year of matriculation
GPA 3.01 1.17 Cumulative undergraduate college grade point average
Day 30.5 17.3 Count variable from 1 to 60 for each of the days in which
invitations were sent (not including weekends and
Engineering 20% 40% holidays)
Indicator variable for Business major
Law 4% 19% Indicator variable for Computer Science major
Computer Science 4% 19% Indicator variable for Design & Media major
Health & Nursing 17% 37% Indicator variable for Engineering major
Design & Media 7% 25% Indicator variable for Health & Nursing major
Business 25% 43% Indicator variable for Humanities major
Humanities 14% 35% Indicator variable for Law major
Sciences 9% 28% Indicator variable for Sciences major

Note. Num Obs. = 97,678 for all variables with the exception of CohortYear for which we have

90,087 observations; data for GPA were available for only a subset of 13,926 observations in post-2009

data.

36
Table 2 Balance of Assignment Groups

Competition Collaboration
Treatment Comparison Group
Variable Mean Std. Dev. Mean Std. Dev.
Female 44.63% 49.71% 44.33% 49.68%
STEM 32.68% 46.91% 32.68% 46.91%
Student 7.88% 26.95% 7.65% 26.57%
CohortYear 1998.8 16.0 1998.7 16.0
GPA 3.01 1.17 3.00 1.18
Day 30.5 17.3 30.4 17.3
Engineering 20.44% 40.32% 20.17% 40.13%
Law 3.70% 18.87% 3.76% 19.02%
Computer Science 3.65% 18.76% 3.62% 18.67%
Health & Nursing 16.82% 37.40% 16.95% 37.52%
Media & Design 6.67% 24.95% 6.57% 24.77%
Business 24.74% 43.15% 25.02% 43.31%
Humanities 14.22% 34.93% 13.98% 34.68%
Sciences 8.59% 0.104401
0.118418 28.03% 8.89%
0 28.46%
0.098255

37
Table 3 Baseline OLS Model Estimates of Participation Rates for Study Population-Averages

Dep. Var.: Participation [Percentage Points]

Model: (1) (2) (3) (4) (5) (6) (7) (8) (9)
Day Student Cohort Subsample GPA
Gender Treatment Interaction Controls Control Controls with GPA Controls All Controls
Female -0.67*** -0.67*** -0.67*** -0.67*** -0.81*** -1.04*** -1.15*** -1.13*** -1.31***
(0.11) (0.11) (0.16) (0.16) (0.16) (0.15) (0.28) (0.28) (0.28)
Competition -0.71*** -0.71*** -0.71*** -0.73*** -0.61*** -1.22*** -1.22*** -1.09***
(0.11) (0.15) (0.15) (0.15) (0.14) (0.27) (0.27) (0.27)
Female × Competition 0.00 -0.01 0.02 -0.06 0.30 0.30 0.19
(0.21) (0.21) (0.21) (0.20) (0.37) (0.37) (0.37)
Student 6.75*** 2.95***
(0.33) (0.50)
CohortYear 0.078*** -0.09***
(0.00) (0.01)
GPA -0.16* -.030
(0.08) (0.08)
Day FEs Y Y
Constant 3.19*** 3.54*** 3.54*** 3.09*** -151.89*** 4.35*** 4.81***
(0.08) (0.10) (0.11) (0.11) -(5.83) (0.21) (0.33)
Adjusted R^2 0.000 0.001 0.001 0.001 0.012 0.011 0.002 0.002 0.008

Note. OLS linear probability model coefficients are reported as percentage points. Coefficients reported as percentage points (i.e., 100 percent = 100).
Heteroskedasticity-robust standard errors reported in brackets. *** p<0.01, ** p<0.05, * p<0.1.

38
Table 4 OLS Model Estimates of Participation Rates, Controlling for Fields

Dep. Var.: Participation [Percentage Points]

Model: (1) (2) (3) (4)
Population Field Population Field
Avg. Controls Avg. Controls
Female -0.67*** 0.25** -0.67*** 0.40**
(0.11) .12 (0.16) (0.17)
Competition -0.71*** -0.43***
(0.15) (0.15)
Female × Competition 0.00 -0.29
(0.21) (0.21)
Computer Science 7.03*** 7.25***
(0.43) (0.44)
Engineering 3.97*** 4.18***
(0.14) (0.16)
Sciences 2.98*** 3.18***
(0.20) (0.22)
Business 2.75*** 2.96***
(0.11) (0.14)
Design & Media 2.59*** 2.80***
(0.22) (0.24)
Humanities 2.07*** 2.28***
(0.14) (0.16)
Law 1.43*** 1.63***
(0.22) (0.23)
Health & Nursing 1.06*** 1.27***
(0.13) (0.16)
Constant 3.19*** 3.54***
(0.08) (0.11)
Adjusted R^2 0.000 0.030 0.000 0.030

Note. OLS linear probability model coefficients are reported as percentage points. Coefficients reported as
percentage points (i.e., 100 percent = 100). Heteroskedasticity-robust standard errors reported in brackets.
*** p<0.01, ** p<0.05, * p<0.1.

39
Table 5 OLS Model Estimates of Participation Rates Stratified by STEM and non-STEM

Dep. Var.: Participation [Percentage Points]

Model: (1) (2) (3) (4)
Non-STEM STEM Pooled
Subset Subset Interactions
Female 0.40** 0.20 1.13*** 0.20
(0.17) (0.19) (0.40) (0.19)
Competition -0.43*** -0.11 -0.87*** -0.11
(0.15) (0.18) (0.26) (0.18)
Female × Competition -0.29 -0.47** -0.32 -0.47**
(0.21) (0.23) (0.50) (0.23)
STEM Field × Female 0.93**
(0.44)
STEM Field × Competition -0.77**
(0.32)
STEM Field × Female × Competition 0.16
(0.55)
Field FEs Y Y Y Y
Adjusted R^2 0.03 0.02 0.05 0.03

Note. OLS linear probability model coefficients are reported as percentage points. Coefficients reported as
percentage points (i.e., 100 percent = 100). Heteroskedasticity-robust standard errors reported in brackets.
*** p<0.01, ** p<0.05, * p<0.1. Note: Among the 97,678 observations, 31,923 are associated with those
trained in STEM fields and 65,755 are associated with non-STEM fields.

40
Table 6 Robustness Tests

Dep. Var.: Participation [Percentage Points]

Model: (1) (2) (3) (5) (6)
Opportunity GPA
Main Result Cohort GPA
Costs Subsample
Female 0.20 -0.10 0.01 -0.17 0.79
(0.19) (0.19) (0.20) (0.35) (0.60)
Competition -0.11 -0.01 -0.01 -0.40 -0.42
(0.18) (0.18) (0.18) (0.34) (0.34)
Female × Competition -0.47** -0.57** -0.58** -0.73* -0.71
(0.23) (0.23) (0.23) (0.44) (0.44)
STEM Field × Female 0.93** 0.44 0.38 -0.89 -0.87
(0.44) (0.44) (0.44) (0.66) (0.66)
STEM Field × Competition -0.76** -0.84*** -0.83*** -1.33** -1.31**
(0.32) (0.31) (0.31) (0.56) (0.56)
STEM Field × Female × 0.15 0.20 0.22 2.10** 2.09**
Competition (0.55) (0.55) (0.55) (0.84) (0.84)
Childrearing Years -0.98***
(0.22)
Childrearing Year × Female -0.43
(0.27)
GPA 0.12
(0.12)
GPA × Female -0.33**
(0.16)
Age -0.09*** -0.09*** -0.08*** -0.08***
(0.00) (0.00) (0.01) (0.01)
Field FEs Y Y Y Y Y
Adjusted R^2 0.03 0.04 0.04 0.04 0.04
No. Obs. 97,696 97,696 97,696 36,345 36,345

Note. OLS linear probability model coefficients are reported as percentage points. Coefficients reported as
percentage points (i.e., 100 percent = 100). Heteroskedasticity-robust standard errors reported in brackets.
*** p<0.01, ** p<0.05, * p<0.1.

41
Table 7 OLS Model Estimates of Participation Rates, Separately Estimated on STEM Subfields

Dep. Var.: Participation [Percentage Points]

Female Competition Female × Competition Constant
No. Obs. coeff. s.e. coeff. s.e. coeff. s.e. coeff. s.e.
Behavioral Neuroscience 603 1.49 (2.01) -0.95 (1.98) -0.77 (2.56) 2.20 (1.54)
Biology 2420 -0.79 (0.96) -2.26 (0.84) 1.55 (1.15) 3.13 (0.75)
Chemical Eng 1337 1.11 (1.48) -0.03 (0.88) 1.20 (2.24) 1.94 (0.61)
Chemistry 782 1.57 (1.78) 0.45 (1.37) -1.58 (2.39) 1.95 (0.87)
Civil Eng 2567 3.52 (1.65) -0.61 (0.45) -2.16 (1.99) 1.42 (0.36)
Computer Eng 1621 1.10 (2.59) -0.87 (1.26) -4.98 (2.86) 5.98 (0.93)
Computer Science 3662 0.87 (1.55) -2.24 (0.99) -0.84 (2.00) 8.11 (0.75)
Electrical Eng 4756 1.64 (1.31) -0.86 (0.47) -0.60 (1.70) 2.80 (0.36)
Eng Mgmt 680 2.21 (3.18) -0.72 (1.80) -0.27 (4.21) 4.45 (1.32)
Environmental Sciences 298 2.12 (3.23) -1.57 (2.48) -2.19 (3.74) 2.94 (2.06)
Industrial Eng 1172 2.71 (2.14) 0.79 (1.16) 0.62 (3.29) 2.72 (0.78)
Information Systems 1156 -0.91 (2.00) 0.99 (1.94) -2.48 (2.74) 5.97 (1.33)
Mathematics 770 -2.49 (1.47) -0.94 (1.64) 6.40 (2.89) 3.59 (1.25)
Mechanical Eng 4418 0.56 (1.17) -0.20 (0.51) 0.02 (1.61) 2.64 (0.37)
Physics 469 -4.26 (1.48) -2.65 (1.74) 4.87 (2.81) 4.26 (1.48)
Psychology 2121 -0.82 (1.08) -1.90 (1.11) 1.38 (1.31) 2.89 (0.95)
Telecommunications Eng 505 -3.18 (3.69) 3.32 (3.64) -1.73 (5.77) 10.40 (2.33)

Note. The table reports percent participation changes in the competition treatment for both men and women, stratified by STEM subfield. Statistical significance
is reported as *** p<0.01, ** p<0.05, * p<0.1.

42
 Figures

10
I. II.

10
Participation [Percent]

Participation [Percent]
5

5
0

0
20 40 60 80 2 2.5 3 3.5 4
Age Grade Point Average

Male Male
Female Female

Figure 1 Gender Differences in Participation by Age and Academic Performance

Note. In Panel I, age is used rather than graduation year in the interest of more direct interpretation. Age is based on
graduation year and approximated according to the approximation that graduation occurs at the age of 22 or
thereabouts. Panel III summarizes differences in participation across males and females in relation to GPA with a
quadratic fitted curve.
 I. Participation Rates II. Differences in Participation
4% under Competition
10%

3.58%

3%
0%
2.86% 2.87%

% Participation Difference
in Competition Treatment
Participation Rate

2.15%
2%
-10%

1% -20%
-20%

-25%

0% -30%
Men Men Women Women Men Women
Collab Comp Collab Comp (N = 53487) (N = 43366)

Figure 2 Differences in Participation and Response to Treatment by Sex, Overall Averages

Note. Panel I is the graphical equivalent of regression results presented in model (3) of Table 3. Panel II presents the
equivalent information in terms of percentage differences in participation under competition, i.e.,:
!"#$%&%'"$%()(+(,'-$%$%())/!"#$%&%'"$%()(+(00"1(#"$%())
. The differences between percentage differences in responses of
!"#$%&%'"$%()(+(00"1(#"$%())
men and women in Panel II is weakly significant at p = 0.097, based on a non-linear Wald-test of their equivalence.
 I. Participation Rates
6%
5.56%
Participation Rate

4% 4.48% 4.37%
3.6%
2.95% 3.13%
2.83%
2% 2.55%

0%
Men Men Women Women Men Men Women Women
Collab Comp Collab Comp Collab Comp Collab Comp
Difference in Participation Rate

II. Differences in Participation under Competition

10%
under Competition

0%
-3.93%
-10%

-20%
-19.62%
-23.63%
-25.49%
-30%
non-STEM non-STEM STEM STEM
Men Women Men Women
(N = 31322) (N = 34067) (N = 22165) (N = 9299)

Figure 3 Treatment Effects, Conditional on Field, by STEM and Non-STEM

Note. Panel I is the graphical equivalent of regression results presented in models (2) and (3) or, equivalently, model
(4) of Table 6. Panel II presents the equivalent information in terms of percentage differences in participation under
!"#$%&%'"$%()(+(,'-$%$%())/!"#$%&%'"$%()(+(00"1(#"$%())
competition, i.e.,: !"#$%&%'"$%()(+(00"1(#"$%())
. The difference in non-STEM men’s and
women’s percentage differences between competition and collaboration (i.e., Panel II) is significant at p = 0.01, based
on a non-linear Wald-test of their equivalence. This gender difference in percentage differences STEM fields is
significant at p = 0.73—or, not statistically significant in the least.
 50% Math (1) Physics (2)
(N = 770) (N = 469)
40%
Chemical Eng

30% (N = 1337)

Industrial Eng
Differences in Participation
20% Telcm Eng(N = 1172)
(N = 505)

10%
Under Competition

0%
-10% Mechanical Eng
(N = 4418)

-20% Eng Mgmt

(N = 680)

Psychology
-30% Biology Info Sys
(N = 2121)
EEng Chemistry(N = 2420)(N = 1156)
Comp Sci
-40% (N = 4756) (N = 782)
(N = 3662)

-50% Beh'l NeuroSci

(N = 603)

-60% Civil Eng

(N = 2567)

-70%
Env Sciences
-80% (N = 298)
Computer Eng
-90% (N = 1621)

0 10 20 30 40 50

STEM Men STEM Women

Figure 4 Percent Difference in Participation Under Competition Treatment in STEM Subfields

Note. The graph presents percent differences in participation between those in the competition treatment and those in
the collaboration comparison group for both men and women, i.e.,:
!"#$%&%'"$%()(+(,'-$%$%())/!"#$%&%'"$%()(+(00"1(#"$%())
, stratified by both sex and STEM subfields. Only gender
!"#$%&%'"$%()(+(00"1(#"$%())
differences for computer science are found to be statistically significant at conventional levels (p = 0.01), based on
3!"#$%&'&'"( 3!"#$%&'&'"( 53)%#*+% × !"#$%&'&'"(
nonlinear Wald test of !2 : = , where $%&'()(*%'(+,% = )+,-'%,' +
&()4$")$ &()4$")$53)%#*+%
/6-,"0- × +(,'-$- ∙ Female% × Compete% + /6-,"0- ∙ Female% + / +(,'-$- ∙ Compete% + <% . The estimated percent
change for females in Physics (1) and Mathematics (2) are truncated in the graph. That for Math is 497 percent (p =
0.27) and that for Physics (2) is undefined (zero women participated in the collaboration treatment).
Appendix A: Examples of Studies on Gender and Competition Reporting
Contradictory Findings
As noted in Section 2.2, one challenge to extrapolating beyond existing results suggesting that
females might be relatively averse to competition is a growing number of published results that run counter
to these original findings. This appendix provides a brief review to illustrate the important contributions of
this research, while also highlighting how it is difficult to derive a simple set of lessons from these studies.

For example, whereas Gneezy et al.'s (2003) seminal study found that 30 male students solving
mazes at an Israeli university were more positively affected by competition than 30 female Israeli university
students who were recruited to the experiment, Günther et al. (2010) find similar results for solving similar
maze problems but not for problems involving writing out words beginning with a certain letter or for
memory games in a group of Spanish students. Günther et al. (2010) interpret their results as possibly
explained by a theory of “stereotype threat,” whereby gendered associations of tasks negatively impact task
performance of the negatively stereotyped group.

However, whereas Gneezy and Rustichini's (2004) highly influential study of 140 fourth grade
Israeli schoolchildren finds that the boys ran faster when facing competition, Dreber et al. (2011) find no
such gender differences in a group of 149 Swedish school children in either running, dancing, or skipping
rope—clearly gender stereotyped activities. We might then perhaps consider possible differences between
Swedish versus Israeli culture. However, Boschini et al. (2018) had a market research firm randomly
telephone Swedish adults, performing laboratory-style verbal and math tasks to assess the effects of
competition; collecting the responses of 997 Swedes between the ages of 18 and 73, they find some evidence
of gender differences in responses to competition in the case of the math task while controlling for income,
age, and level of education.

Gneezy et al. (2009) further examine the possible role of cultural context, recruiting 155 people
from the Maasai in Tanzania and the Khasi in India, asking them to throw tennis balls into a bucket,
choosing either competitive or noncompetitive monetary regimes, i.e., the Niederle-Vesterlund protocol.
The Maasai men more frequently chose competition, whereas the Khasi women more frequently chose
competition. The authors interpret the results as evidence that gender differences are tied to societal culture.
They speculate further that gender differences in prior research (i.e., Sections 2.1 and 2.3) could be a fixture
of Western societies.

Other results suggest that narrower socialized differences could shape gender differences. Andersen
et al. (2013) recruit 318 Indian school children, 9–15 years old, to similarly throw tennis balls into a bucket
in a Niederle-Vesterlund protocol. Most subgroups (by gender, age, and village type) do not statistically
differ, when stratifying the data by sex, village, and for ages 9–12 or 13–15. They find differences between
boys and girls in the case of patriarchal villages for the 13–15 group. The authors interpret the results to
mean that gender differences emerge only in the older girls within patriarchal societies are socialized for
some years. In other research, Gneezy and Rustichini (2004) report gender differences occurring before the
age of 13.

(A series of findings on gender differences in auctions, where female subjects tend to bid higher
than males (e.g., Casari et al. 2007), also appears to run counter to the findings in research on gender and
competition.)
 Appendix B: Product Development Platform

Fig. A1 Description of IoT Project Work: Design of Use Case, Technology Architecture, and Business
Case for Prototype
Fig. A2 Sample Section of Technology Architecture Design Steps
Fig. A3 Sample Technology Design Drag-and-Drop Design Tool