The Boy Crisis

The Boy Crisis: Experimental Evidence on the
Acceptance of Males Falling Behind
BY Alexander W. Cappelen, Ranveig Falch and

Bertil Tungodden
DISCUSSION PAPER
Institutt for samfunnsøkonomi

Department of Economics
SAM 06/2019
ISSN: 0804-6824
March 2019
This series consists of papers with limited circulation, intended to stimulate discussion.
The Boy Crisis: Experimental Evidence on the
Acceptance of Males Falling Behind
Alexander W. Cappelen, Ranveig Falch and Bertil Tungodden
March 1, 2019
Abstract The ‘boy crisis’ prompts the question of whether people interpret inequal-
ities differently depending on whether males or females are lagging behind. We study
this question in a novel large-scale distributive experiment involving more than 5,000
Americans. Our data provide strong evidence of a gender bias against low-performing
males, particularly among female participants. A large set of additional treatments es-
tablishes that the gender bias reflects statistical fairness discrimination. The study pro-
vides novel evidence on the nature of discrimination and on how males falling behind
are perceived by society.
JEL: C91, D63, J16

Key words: gender bias, boy crisis, statistical fairness discrimination, large-scale
experiment
Cappelen: Department of Economics, NHH Norwegian School of Economics, 5045 Bergen, alexan-
der.cappelen@nhh.no. Falch: Department of Economics, NHH Norwegian School of Economics, 5045
Bergen, ranveig.falch@nhh.no. Tungodden: Department of Economics, NHH Norwegian School of Eco-
nomics, 5045 Bergen, bertil.tungodden@nhh.no. The study was organized by FAIR/The Choice Lab.
We are greatful to Daniel Benjamin, Pedro Rey Biel, Caroline Bonn, Gary Charness, Stefano DellaVi-
gna, Catherine Eckel, Christine L. Exley, Uri Gneezy, Ingar Haaland, Alex Imas, Siri Isaksson, Elisabeth
Ivarsflaten, Nanjundi Karthick Krishnan, James Konow, Erin Krupka, Axel Ockenfels, Anders Poulsen,
Odile Poulsen, Lise Rakner, Charlotte Ringdal, Marta Serra-Garcia, Ingrid Sjursen, Charles Sprenger,
Eirik Strømland, Matthias Sutter, Lise Vesterlund, Jonas Tungodden and numerous seminar participants
for great comments and suggestions and to Sebastian Fest for valuable research assistance. We are also
thankful for the financial support from the NORFACE (New Opportunities for Research Funding Agency
Cooperation in Europe) Network, The Centre for Ethics and Economics at the Norwegian School of Eco-
nomics and the Research Council of Norway through its Centres of Excellence Scheme, FAIR project No
262675. The study was pre-registered in the AEA RCT Registry; AEARCTR-0000853 and AEARCTR-
0001027.
2
1 Introduction
Across the world, males occupy the majority of top-level jobs, head of govern-
ment positions and national parliament seats, and in all societies there is still
a significant gender wage gap. To illustrate, in the United States, women full-
time workers earned on average 79 percent of what men did on an annual basis
in 2014 (Blau and Kahn, 2017). Thus, there is still urgent need for political
action to achieve gender equality for women.
The present paper is motivated by a different gender gap, often termed the
‘boy crisis’ (Autor and Wasserman, 2013). In education, it is by now well es-
tablished that boys are lagging behind. In all but six OECD countries, a larger
proportion of boys than girls do not attain the baseline level of proficiency in any
of the core subjects; mathematics, reading and science (OECD, 2015). In the US
for instance, the average percentage of students who do not attain the baseline
proficiency level was 71% higher for boys than for girls. Boys are also drop-
ping out of high school at higher rates than girls in most OECD countries. In
higher education, females have surpassed the rate of males graduating in nearly
all OECD countries, on average by 14 percentage points (OECD, 2016).
Similarly, in high-income countries there is a growing concern about the
prospects for low-skilled males: “The decline in economic opportunities for
low-skilled men and the possible negative effects of this trend on their well-
being is a matter of increasingly urgent concern for policy makers and the gen-
eral public” (Coile and Duggan (2019), p. 2). Males with less than four-year
college education have seen a significant reduction in real income over the last
decade in the US (Autor and Wasserman, 2013; Binder and Bound, 2019), and
the percentage of idle young males and prime age males outside the labor force
has increased (Blau and Kahn, 2013; Krueger, 2017). The prospects for males
outside the labor force are dim, in particular for those from low-income house-
holds and for males with minority backgrounds. The likelihood of living in
poverty is increased and their expected future health and emotional well-being
is poor (Autor and Wasserman, 2013; Krueger, 2017; Council of Economic Ad-
visers, 2016).
These striking empirical developments make it important to study whether
people react more negatively to males falling behind than to females falling be-
hind, since this may both reinforce the negative trend of low-performing males
3
and shape our public responses to the ‘boy crisis’. In this paper, we examine this
question with a novel experimental approach, where we study peoples’ views on
income inequalities generated in a labor market. Our main focus is on whether
people show less concern for income inequality when a male is falling behind
than when a female is falling behind. We provide a simple theoretical frame-
work to guide the empirical analysis, where we show how statistical fairness
discrimination against male losers may occur if people perceive males to have a
productivity advantage. The idea is that such an advantage makes people infer
that male losers are likely to have exerted lower effort than female losers, and
thus are less deserving of assistance.
To collect experimental data on a general population sample, we combine
the infrastructure of a leading international data-collection agency and an on-
line labor market (Almås, Cappelen, and Tungodden, 2016b). In the online labor
market, we recruit more than 2,000 workers and generate inequalities by paying
two workers differently for the same assignment. In our main treatments, the in-
equality is generated by paying the more productive worker (the winner) more
than the less productive worker (the loser). We then ask a general population
sample of more than 3,000 Americans to act as impartial spectators and make a
decision on whether to redistribute earnings between the winner and the loser,
where each spectator makes a real decision that has consequences for two work-
ers. We randomly assign spectators into a treatment where the loser is a male or
a treatment where the loser is a female. Our main interest is in studying whether
the redistribution decision of the spectators depends on the gender of the loser.
We run a set of additional treatments to study the underlying mechanisms of the
spectator choices, where we vary the source of the inequality (merit or luck) and
the gender-composition of the two workers (mixed-sex or single-sex).
Our main result is reported in Figure 1. In a large general population sample
of Americans, we find a significant gender bias against male losers in the mixed-
sex treatments when merit is the source of inequality. The left bar of Panel
A shows the average share transferred by the spectators to the losing worker
when a female is the least productive, and the right bar shows the average share
transferred when a male is the least productive. The spectators transfer about
15% less to a male loser than to a female loser (23.1% versus 19.5%), and the
share of spectators transferring nothing increases by 7.3 percentage points when
the loser is a male. Panel B reports the difference in the average share transferred
4
by the gender of the spectator, which shows that the gender bias is clearly driven
by the female spectators, who transfer 25% less to the male losers than to the
female losers.
[Figure 1 about here]
We further show that the gender bias against males does not reflect taste-
based preferences among the spectators. In the treatments where the source of
inequality is luck, the spectators transfer the same amount to male losers and to
female losers. We also do not find any difference in spectator behavior between
treatments where both workers are female or both workers are male, which sug-
gests that the spectators do not find inequality more acceptable between males
than between females. The findings are in line with our theoretical framework,
and our preferred interpretation of the main result is that the gender bias in the
mixed-sex merit setting reflects statistical fairness discrimination: spectators in-
fer that male losers to a greater extent than female losers have exerted low effort
and therefore consider them less deserving of assistance. The fact that we find
the gender bias to be driven by female spectators suggests that females, to a
greater extent than males, consider males to have a productivity advantage, and,
as a result, are more likely to infer differently about the effort of male losers and
female losers.
Our paper contributes to a number of different literatures. Our results shed
light on the growing literature on the ‘boy crisis’, by providing evidence sug-
gesting that the general public may view male losers differently from female
losers. This may have important implications for the support for public policies
targeting low-performing males, and for our understanding of gender-biased
behavior. The existing literature in this area has largely focused on explain-
ing the low performance of males with structural changes and socioeconomic
developments (Almås, Cappelen, Salvanes, Sørensen, and Tungodden, 2016a;
Autor and Wasserman, 2013; Autor, Figlio, Karbownik, Roth, and Wasserman,
forthcoming; Bertrand and Pan, 2013; Binder and Bound, 2019; Krueger, 2017;
Rosin, 2012), while the present paper focuses on how low-performing males are
perceived by society. We show that people tend to interpret males falling be-
hind as having exerted less effort than females falling behind, which may result
5
in male losers being treated differently in school, at the workplace, and in the
family. This may result in low-performing males ending up in a vicious circle
where they lose their motivation to provide effort because they are not rewarded
to the same extent as females in the same situation.
We further contribute to the literature on gender discrimination (Bertrand
and Duflo, 2017), by providing novel evidence on a gender bias against low-
performing males. A number of important papers have shown discrimination
against females in different areas, including in hiring decisions (Goldin and
Rouse, 2000; Coffman, Exley, and Niederle, 2019), task allocation (Babcock,
Recalde, Vesterlund, and Weingart, 2017), bargaining (Castillo, Petrie, Torero,
and Vesterlund, 2013; Exley, Niederle, and Vesterlund, forthcoming), teaching
evaluations (Mengel, Sauermann, and Zölitz, 2018), and career development
(Reuben, Sapienza, and Zingales, 2014), but recent research has also provided
evidence of a gender bias against males in certain settings (Bohren, Imas, and
Rosenberg, forthcoming; Heikensten and Isaksson, 2016; Mengel et al., 2018;
Reynolds, Sjåstad, Howard, Okimoto, Baumeister, Aquino, and Kim, 2017;
Williams and Ceci, 2015). In particular, Bohren et al. (forthcoming) find in
a field experiment on an online mathematics forum for STEM-students and re-
searchers that females initially face significant discrimination, but over time are
favored over men. Our paper focuses on low-performing males in a distributive
context, where we show, in a large-scale study of the general population in the
United States, that people are less willing to assist males falling behind than
females falling behind.
Our results also speak to the non-experimental studies of gender discrim-
ination in schools (Breda and Ly, 2015; Cornwell, Mustard, and Van Parys,
2013; Falch and Naper, 2013; Lavy, 2008; Lavy and Sand, 2018; Lindahl, 2016;
Terrier, 2016). Albeit with exceptions, accumulated evidence suggests that in
developed countries, teacher grade setting is discriminatory against males in a
range of subjects in kindergarten, primary school, high school and higher edu-
cation. One of the challenges in this literature has been to disentangle a teacher
gender bias from the effect of gender differences in non-cognitive skills on grade
setting. The present study contributes to this literature by identifying that there
is a gender bias against males in a performance-based environment, even in a
setting in which non-cognitive skills cannot affect evaluations.
We more generally contribute to the discrimination literature by introduc-
6
ing, to our knowledge, the first study of statistical fairness discrimination, where
people make inferences about the deservingness of a person based on observable
characteristics. We believe that this mechanism speaks both to distributional set-
tings, as is the focus in this paper, but also more generally to labor market and
educational settings where a principal would like to reward effort. We also pro-
vide a new approach to disentangling statistical discrimination and taste-based
discrimination (Bohren et al., forthcoming; Cettolin and Suetens, forthcoming;
Fershtman and Gneezy, 2001; List, 2004), by randomly manipulating whether
the spectators make a distributive decision in an environment where productivity
or luck is the source of inequality.
Finally, our results contribute to the literature in behavioral economics inves-
tigating the role of gender in people’s social preferences (Croson and Gneezy,
2009; Eckel and Grossman, 2008). Previous studies have varied the salience of
recipient gender in stakeholder games such as the dictator game and the ulti-
matum game. A meta-study on dictator game experiments largely finds that fe-
males receive more than males in dictator games when recipient gender is made
salient (Engel, 2011). For ultimatum games that vary recipient gender, individ-
uals make higher offers to males compared to females (Eckel and Grossman,
2001; Solnick, 2001). In contrast to these previous studies, we examine gender
bias in an impartial spectator decision, which provides a direct expression of the
moral preferences of the participants (Cappelen, Konow, Sørensen, and Tun-
godden, 2013). In this setting, we show that there is a significant gender bias
against males reflecting statistical fairness discrimination. More broadly, we
also provide new evidence on how fairness preferences shape distributive behav-
ior (Bolton and Ockenfels, 2000; Bortolotti, Soraperra, Sutter, and Zoller, 2017;
Cappelen, Drange Hole, Sørensen, and Tungodden, 2007; Fehr and Schmidt,
1999; Konow, 2000), by showing how people’s fairness considerations differ
across contexts and depend on the source of inequality.
The paper is organized as follows: Section 2 describes the experimental
design, Section 3 introduces a simple theoretical framework that guides our in-
terpretation of the results, Section 4 outlines the main empirical strategy, Sec-
tion 5 reports the main results and the heterogeneity results, while Section 6
concludes. Additional analysis is provided in Appendix A, and the complete
instructions for both spectators and workers are provided in Appendix B.
7
2 Experimental design and participants

We first provide an overview of the general structure of the experiment, which
builds on Almås et al. (2016b), before we turn to a detailed discussion of the
participants and the experimental design.
Table 1 summarizes the main stages of the experiment. The experiment had
two types of participants, workers and spectators. First, the workers completed
an assignment. They were then matched in pairs and assigned different earn-
ings. The spectators were randomly matched to one pair of workers and decided
whether to redistribute earnings between the two workers. Finally, the workers
were paid according to the spectator decisions.
[Table 1 about here]
2.1 The workers

The workers in the experiment were recruited from the international online labor
market platform Amazon Mechanical Turk. This is a crowdsourcing web ser-
vice that specializes in recruiting anonymous workers to complete small tasks
online. When recruited, the workers were promised a participation fee of 2 USD
and told that they could earn additional money, depending on the actions they
and others would take in the experiment. We recruited 2,072 workers, 1,036
men and 1,036 women. Each worker completed three different tasks. After they
had completed all three assignments, the workers were told how they would
be paid for the assignments. Specifically, for each assignment, they were ran-
domly matched in pairs, giving us 3,108 unique pairs workers conditional on
assignment. In each such pair, one worker was initially assigned 6 USD and
the other 0 USD. The workers were not told whether they had been assigned
high or no initial earnings. They were told, however, that a third person, the
spectator, would be informed about the assignment and the initial distribution
of earnings. They were further informed that the spectator would be given the
opportunity to redistribute the earnings between the two workers in the pair and
thus determine how much they would actually be paid for the assignment. The
workers received the participation fee immediately after they had completed the
assignment and the income determined by the spectator within a few days after
the spectators made their choice.
8
2.2 The spectators

The spectators in the experiment were recruited using the infrastructure of the
data-collection agency TNS Gallup. We recruited 3,102 participants who con-
stitute a nationally representative sample of the United States (+ 18 years old)
on a limited set of observable characteristics (gender, age and geography). Each
spectator was matched with a unique pair of workers and decided whether and
how much of the initial earnings to redistribute.1 We further collected back-
ground characteristics of the spectators in terms of gender, political orientation,
income and age. Table 2 provides an overview of the background characteristics
of the spectators and a comparison with US census data. The sample is largely
gender-balanced, with 48.8% being males, and a median age of 41 years. The
median yearly gross income before taxes is 55 000 USD and 33.7% state that
they would vote Republican.2 The sample is largely representative of the US
population on these dimensions, even though we note that income is more com-
pressed in the sample than in the population at large (which partly may reflect
that self-reported income was restricted at the extremes).
2.3 General structure

In all treatments, the spectators made a decision in a situation where one worker
had earnings of 6 USD and the other had earnings of 0 USD. The spectators
were not informed about the nature of the tasks assigned to the workers, but
about the age, nationality, and gender of the two workers. It was emphasised to
the spectators that, in contrast to traditional survey questions, their choice would
have consequences for a real life situation. They were given all the information
provided to the workers. In particular, to minimize the role of worker expecta-
tions in the spectator choice, they were told that the workers would not at any
point be informed about their initial earnings.
1 Since we had 3,108 unique distributive situations and 3,102 spectators, we applied six spec-
tator decisions twice.
2 4.8% of the spectators reported that they did not know or preferred to not state their income.
Our results are robust to the removal of these participants.

9
2.4 Treatments
We implemented a between-subject design, where spectators were randomly
assigned to treatments.
In the two main treatments, the mixed-sex merit treatments, the spectators
considered a distributive situation involving a female worker and a male worker,
where the initial inequality in earnings was determined by the productivity of
the workers: the more productive worker earned 6 USD and the less productive
worker earned 0 USD. The two treatments only differed in terms of whether the
female or the male worker had been less productive, which allows us to identify
the causal effect of the gender of the loser on the amount transferred to the loser.
The experimental design thus allows us to study whether the spectators are more
inequality accepting when a male falls behind then when a female falls behind.
To investigate the underlying mechanisms of the spectator behavior, we in-
cluded six additional treatments. First, to study the role of the source of in-
equality, we implemented two additional treatments that mirrored the two main
treatments but where the inequality in earnings was determined by luck; second,
to study the role of the gender composition, we added four single-sex treatments
that mirrored the four mixed-sex treatments. Table 3 provides an overview of
the treatments and the number of participants in each treatment.3
3 Theoretical framework
We here provide a simple theoretical framework to guide the interpretation of
the results.
The spectator decides on a distribution (1 − y, y) of income between two
randomly matched workers i = j, k, where y is the share given to the worker
with no initial earnings (the loser).
3 Werecruited the spectators in two rounds of data collection. In the first round, we recruited
2,052 US participants to act as spectators, who were randomly allocated to one of eight treat-
ments. In the second round, to study the robustness of our findings, we recruited another 1,050
US participants. In this round, the participants were randomly allocated to one of the two main
treatments. Individuals who participated in the first round of data collection were not permit-
ted to participate in the second round. Table A.1 presents a balance test, where we show that
the eight treatments are not significantly different from each other on any of the background
characteristics in Table 2.
10
We assume that the spectators have the following model of how the produc-
tivity of a worker is determined, ri (g) = ei + a(g) + εi , where ei is the worker’s
effort, a(g) is a gender specific advantage, g = f , m, and εi is a normally dis-
tributed random shock, εi ∼ N(0, σ ).4 For simplicity, we consider only two
effort levels, eh > el , and we assume that the spectators believe that males and
females have the same effort distribution, where p(el ) is the share of work-
ers exerting low effort We normalize the gender specific advantage of females,
a( f ) = 0, such that a(m) > 0 would imply that males have a productivity advan-
tage relative to females. We may interpret the productivity advantage in terms
of ability, but it can also be interpreted more broadly as capturing any kind of
advantage that makes one gender more productive than the other. Finally, we
assume that the random shock is truncated, εi ∈ (−1/2a(m), 1/2a(m)), and that
the gender specific advantage is restricted, 2|a(m)| < (eh − el ), which implies
that a worker exerting high effort is always more productive than a worker ex-
erting low effort.
We assume that the spectators care about fairness and that they consider the
fair share to transfer to the losing worker, l, to be determined by the relative
effort of the two workers. The spectators may also have a partial gender prefer-
ence, as captured by the following utility function (Cappelen et al., 2013; Almås
et al., 2016b):
V (y; ·) = −(y − h(E(e)))2 − Iβ (g)y, (1)
where E(e = el /ew ) is the expected relative effort of the loser (el ) compared
to that of the winner (ew ). We assume that the fair share to the loser is increasing
in the expected relative effort of the loser. The second term captures partial
gender preferences, where I is an indicator variable taking the value one if this
is a mixed-sex pair, g is the gender of the loser, and β (g) is the strength of the
gender preference, β (m) = −β ( f ).
The optimal interior solution is given by:
β (g)
y(E(e), g, I) = h(E(e)) −
I (2)
2
The model captures the two main approaches to discrimination in the eco-
4 We do not have data that allow us to study whether the spectator’s belief about the relation-
ship between effort and productivity is misspecified, see Bohren et al. (forthcoming).
11
nomics literature. The first approach, developed by Becker (1957), introduces

taste-based discrimination, where a distaste for a group of people (or favoritism
of another) may lead to differential, negative treatment of its members. In our
theoretical framework, spectators have taste-based discrimination if β (g) 6= 0.
The second approach is statistical discrimination (Phelps, 1972; Arrow, 1973),
where observable characteristics of individuals are used to proxy unobservable,
but relevant, characteristics. Our model captures statistical fairness discrimina-
tion. The spectators’ belief about the expected relative effort of the loser and the
winner, and consequently what the spectators consider the fair share to give to
the loser, depend on the signal they receive about the gender composition in the
pair and the source of the inequality.
Let us now study the predictions of this model for the different treatments
in our study. First, consider the mixed-sex merit treatments. Assume that the
spectators believe that males have an advantage a(m) > 0. In this case, it fol-
lows from the model that a male worker will only have lower productivity than
a female worker when he has exerted low effort and she has exerted high ef-
fort.5 In all others cases, the female worker has lower productivity, including
the cases where they both have exerted the same effort. Thus, assuming that the
spectators have rational beliefs, the expected relative effort of the loser will be
lower when there is a male loser than when there is a female loser. Hence, with-
out taste-based gender preferences in favor of the male worker, the spectators
would transfer less to a male loser than to a female loser in the mixed-sex merit
treatments. More generally, we have the following observation:
Observation 1: Statistical fairness discrimination would in mixed-sex merit

treatments imply a smaller transfer to the loser of the gender that is considered
to have an advantage.
In the mixed-sex luck treatments, there is no signal of productivity that can

lead to statistical fairness discrimination. Hence, any gender bias in the amount
transferred to the loser in these treatments would have to reflect taste-based
5 This follows straightforwardly from comparing the four possible scenarios: (i) a male k
exerting low effort matched with a female j exerting low effort: rk (m) = el + a(m) + εk >
r j ( f ) = el + ε j , (ii) a male k exerting low effort matched with a female j exerting high effort:
rk (m) = el + a(m) + εk < r j ( f ) = eh + ε j , (iii) a male k exerting high effort matched with a
female j exerting low effort: rk (m) = eh +a(m)+εk > r j ( f ) = el +ε j , (iv) a male k exerting high
effort matched with a female j exerting high effort: rk (m) = eh + a(m) + εk > r j ( f ) = eh + ε j .
12
discrimination or that the spectators use gender as a signal for other morally
relevant characteristics not captured by the fairness considerations.
Observation 2: Statistical fairness discrimination would not affect spec-

tator behavior in mixed-sex luck treatments, while taste-based discrimination
would imply a smaller transfer to the loser of the gender that the spectators
have a distaste for.
Consider now the single-sex treatments. It follows by design that there can-
not be any gender bias in these treatments, since the loser and the winner are
of the same gender. However, our model still offers some predictions for the
comparison of the single-sex treatments. Given the assumption that the effort
distribution is the same for males and females, it follows that the likelihood of
the loser having relatively low effort compared to the winner is the same in the
two single-sex merit treatments.6 Hence, we should observe the same spectator
behavior in these two treatments. In the single-sex luck treatments, there is no
productivity signal and the two workers are of the same gender, and thus we
should also in these treatments expect the spectator behavior to be unaffected
by the gender of the loser.
Observation 3: Statistical fairness discrimination and taste-based discrim-

ination should not make spectators transfer differently to male losers and female
losers in the single-sex treatments.
Finally, the model provides predictions for the comparison of the mixed-
sex treatments and the single-sex treatments. In particular, statistical fairness
discrimination implies that we should observe the lowest transfer to the male
loser in the mixed-sex merit treatment, where a male loser provides the strongest
signal of the loser having exerted low effort compared to the winner.7
Observation 4: Statistical fairness discrimination would imply that we ob-

serve the lowest transfer to male losers in the mixed-sex merit treatments.
6 In the single-sex treatments, the likelihood of the loser having exerted low effort and the
winner having exerted high effort is given by 2p(el )(1 − p(el )).
7 This follows straightforwardly from the fact that the probability that the loser has exerted
low effort compared to the winner p(ew > el ) is equal to one in the mixed-sex merit treatment
with a male loser and less than one in all the other treatments.
13
4 Empirical strategy
We here outline the main empirical strategy. We specified the empirical strategy
in two pre-analysis plans, one for each round of data collection, registered at the
AEA RCT Registry.
4.1 Main analysis

Our main variable of interest is the amount transferred to the losing worker by
spectator i in the mixed-sex merit treatments. The main empirical specification
used in the analysis of these two treatments is:
ui = α + β Maleloseri + γXi + εi , (3)
where ui is the standardized amount transferred by spectator i to the losing

worker, Maleloseri is an indicator variable for spectator i belonging to the treat-
ment where the losing worker is male, Xi is a vector of control variables, and εi
is an error term. We regress (3) on the sample of spectators in the two mixed-
sex merit treatments, with and without a set of pre-specified control variables
(gender, political affiliation, income, age). The reference category in (3) is the
treatment with a female loser, and β thus provides an estimate of the causal ef-
fect on the transfer to the loser of the loser being a male instead of a female.
We also report (3) for the dependent variable being an indicator variable for
spectators giving nothing to the losing worker.
For the mixed-sex merit treatments, we study heterogeneous effects based on
the gender, political orientation, income and age of the spectators. For gender,
we use the following specification:
ui = α + β1 Maleloseri + β2 Mi + β3 Maleloser × Mi + εi , (4)
where ui is the standardized amount transferred by spectator i to the losing

worker or the indicator variable for spectators giving nothing to the losing worker,
Maleloseri is an indicator variable for spectator i being in a treatment where the
losing worker is male, Mi is an indicator variable for spectator i being male,
Maleloser × Mi is an interaction variable for spectator i being male and in a
treatment where the losing worker is male, and εi is an error term. In this analy-
14
sis, β1 provides an estimate of the causal effect for female spectators of the loser
being a male, β1 + β3 provides a corresponding estimate for the male spectators,
and β3 provides an estimate of whether the causal effect of the loser being male
differs between male spectators and female spectators. We report corresponding
regressions for the other dimensions of heterogeneity and a regression including
interaction variables for all the background variables.
4.2 Mechanisms
To study the underlying mechanisms of the spectator choices, we provide a set
of regressions involving the six additional treatments. First, we study the role
of the losing gender in the two mixed-sex treatments where luck is the source
of inequality, by reporting regressions using specifications (3) and (4). In addi-
tion, we include all the four mixed-sex treatments and run regressions where we
interact the male loser indicator variable with an indicator variable for being in
the treatment where luck is the source of inequality:
ui = α + β1 Maleloseri + β2 Lucki + β3 Maleloser × Lucki + γXi + εi . (5)
In this analysis, β1 provides an estimate of the causal effect of the loser being
a male in the mixed-sex treatment where the source of inequality is merit and
β1 + β3 provides the corresponding estimate for the luck treatment, β2 provides
an estimate of whether the loser is treated differently in the merit treatment and
the luck treatment, and β3 provides an estimate of whether the effect of being a
male loser differs between the merit and the luck treatment. We implement (5)
for all participants in the mixed-sex treatments and separately for each subgroup
defined by the background characteristics.
Second, we study the role of gender in the single-sex treatments. We report
regressions corresponding to (3) and (4) separately for the two single-sex merit
treatments and for the two single-sex luck treatments, where in both cases the
treatment variation is the gender of the workers. Further, to study whether the
male loser and the female loser in the mixed-sex merit treatments are treated
differently from the average loser in the single-sex merit treatments, we run the
following regression for the four merit treatments:
ui = α + β1 MS − Maleloseri + β2 MS − Femaleloseri + γXi + εi , (6)

15
where MS − Maleloseri and MS − Femaleloseri are indicator variables for the

spectator being in a mixed-sex merit treatment with a male loser and a female
loser, respectively. In (6), the reference category is the pooled single-sex merit
treatments, and thus β1 and β2 provide causal estimates of whether the male
loser and the female loser in the mixed-sex merit treatments are treated differ-
ently from the average loser in the single-sex treatments. We report the regres-
sions with and without control variables.
4.3 Multiple hypothesis testing

As a robustness check of our main results, we compute p-values adjusted for
multiple hypothesis testing as described in Romano and Wolf (2016). We calcu-
late unadjusted p-values as bootstrap p-values following Davison and Hinkley
(1997) and compute p-values adjusted for stepdown multiple testing following
the algorithm proposed by Romano and Wolf (2016). Bootstrapping is imple-
mented with 10,000 replications.
5 Results
We first provide an overview of the spectator choices in the experiment. We then
turn to the analysis of the main treatment effects, the heterogeneity analysis and
the mechanisms.
5.1 The distributive decisions

Figure 2 presents a histogram of the spectator decisions pooled for all treat-
ments.8 We observe that the mode is to give nothing to the losing worker (31%),
the average amount and the median amount transferred are 1.56 USD and 2
USD. A significant share of the spectators equalize the earnings (22%), while
almost half of the spectators (44%) do not fully equalize but give some to the
losing worker (1 USD or 2 USD). Interestingly, very few spectators (3%) give
more to the losing worker than to the winning worker, which provides strong
evidence against the spectators randomizing in their choices.
8 Figure
A.1 gives a disaggregated presentation by treatment. An overview of the distributive
decisions by round is provided in Figure A.2.
16
[Figure 2 about here]
5.2 Main findings

We now turn to a regression analysis of our two main treatments, where we
focus on whether the spectator choice is responsive to the gender of the loser.
Table 4 reports the results for specification (3), where the dependent variable is
the standardized amount transferred to the losing worker (columns 1-2) and the
share of spectators giving nothing to the losing worker (columns 3-4).
We observe a significant gender bias in the spectator choices: the average

amount transferred to the losing worker is reduced by 0.174 standard deviations
(p < 0.001) when the losing worker is a male rather than a female (column 1),
which corresponds to a 15% reduction in the amount transferred. In Table A.2,
we show that the estimated causal effect is almost identical in the two rounds of
data collection. We find the same qualitative result when we use the alternative
outcome variable (column 3): the share of spectators giving nothing to the losing
worker increases by 7.3 percentage points (p = 0.003) when the losing worker
is a male rather than a female. These results remain practically unchanged when
we include background variables for gender, political orientation, income and
age (columns 2 and 4), and are robust to multiple hypothesis adjustment (p =
0.003, Table A.15). Thus, we can state our first main result:
Result 1: We find strong evidence of a gender bias against male losers when
the source of inequality is merit.
With respect to the background variables, we observe in Table 4 that Repub-

licans on average transfer significantly less than non-Republicans to the losing
worker (0.173 standard deviations, p = 0.001), and are significantly more likely
not to transfer anything to the loser (5.8 percentage points, p = 0.026). We do
not observe any significant associations between the other background variables
and the spectator choices.
In Table 5, we report the heterogeneity analysis, as specified in (4), focusing
on the standardized amount transferred to the losing worker.9
9 The full set of results are reported in Table A.3.
17
The gender bias against males is strikingly robust across the different sub-
groups (columns 1-4): the estimated causal effect of the loser being a male
rather than a female is in all cases negative, the effect being statistically signif-
icant for all subgroups except for low income spectators and males. In particu-
lar, we observe a strong gender bias against males among the female spectators
(0.283 standard deviations, p < 0.001), which is robust to multiple hypothesis
adjustment (p = 0.001, Table A.16). The gender interaction effect is also large,
significant and robust to the inclusion of all the interaction variables in the re-
gression (0.23 standard deviations, p = 0.025) and robust to multiple hypothesis
adjustment (p = 0.085, Table A.17). As shown in Table A.4, the patterns are the
same when using the share transferring nothing to the losing worker.
Result 2: We find a significant gender bias against male losers in almost

all subgroups when the source of inequality is merit, and, in particular, among
female spectators.
The finding of a gender bias against males in the merit-sex treatments is in

line with statistical fairness discrimination by spectators who consider males to
have an advantage, as established in Observation 1 in our theoretical framework.
However, from the mixed-sex merit treatments alone, we cannot rule out that
this behavior reflects taste-based discrimination.
5.3 Mechanisms
In this section, we use our additional treatments to shed light on the underlying
mechanisms driving the observed gender bias against male losers.
First, we study whether the gender bias against male losers reflects taste-
based discrimination, by considering the distributive situations where the source
of inequality is luck. In these situations, there is no productivity signal that can
lead to statistical fairness discrimination, and thus the spectators believe that the
expected effort of the winner and the loser is the same. However, in line with
Observation 2, taste-based discrimination against males would imply a lower
transfer to a male loser, both in the mixed-sex luck treatments and in the mixed-
sex merit treatments. Table 6 reports the regressions using specification (3) for
18
the two mixed-sex luck treatments, where the dependent variable is the stan-
dardized amount transferred to the losing worker in column 1 and the share of
spectators giving nothing to the losing worker in column 4. Given that there is
a significant gender interaction effect in the spectator behavior in the mixed-sex
merit treatments, we also report this analysis separately for females (columns 2
and 5) and males (columns 3 and 6).
We do not find evidence of a gender bias in the spectator behavior in the

mixed-sex luck treatments. The estimated male loser effect is small and not sta-
tistically significant both for the amount transferred to the loser (-0.043 standard
deviations, p = 0.629) and for the share transferring nothing to the loser (0.002
standard deviations, p = 0.957). We further observe that there is no significant
gender bias among female spectators or male spectators, and this pattern extends
to all the other subgroups, see Table A.5. As shown in Table A.6, the subgroup
patterns are largely robust to using the share of spectators giving nothing to the
loser as the dependent variable.10
Result 3: We find no robust evidence of a gender bias against male losers

when the source of inequality is luck, which suggests that the spectator choices
do not reflect taste-based preferences.
In Table 7, we compare the spectator behavior in the mixed-sex merit treat-

ments and mixed-sex luck treatments, using specification (5). We again observe
that there is a highly significant male loser effect in the mixed-sex merit treat-
ments, but no gender effect in the mixed-sex luck treatments.11 The interaction
effect is highly significant for female spectators: the male loser effect increases
by 0.298 standard deviations in the mixed-sex merit treatments compared to
the mixed-sex luck treatments (-0.266 standard deviations versus 0.032 stan-
dard deviations, p = 0.020), and the share transferring nothing to the male loser
10 In Table A.6, we find some suggestive evidence of the high income group discriminating
against males and the low-income group discriminating against females even when the source of
inequality is luck, but these effects do not hold when adjusting for multiple hypothesis testing.
11 In line with previous research (Almås et al., 2016b), we also observe that there is signifi-
cantly more inequality acceptance when the source of inequality is merit than luck: on average,
the share transferred to the loser when the source of inequality is merit is 0.594 standard devi-
ations lower than when the source of inequality is luck (p < 0.001) and this pattern is robust
across subgroups (in all cases, p < 0.001).
19
increases by 15.3 percentage points (0.128 percentage points versus -0.025 per-
centage points, p = 0.020). This provides evidence of the behavior of female
spectators reflecting statistical fairness discrimination. In Table A.7 and Ta-
ble A.8, we compare the spectator behavior in the mixed-sex merit treatments
and mixed-sex luck treatments for all subgroups, where we also observe evi-
dence of statistical fairness discrimination among low income spectators and
older spectators. We should note, however, that the estimated interaction effects
are not statistically significant if we adjust for the multiple testing of interaction
effects between merit and luck for all subgroups, see Table A.18.
We now turn to an analysis of the single-sex treatments. In our theoretical
framework, Observation 3, the spectators should not infer differently about the
effort of the loser and the winner in the female and male single-sex merit treat-
ments. An essential assumption underlying this observation is that spectators
believe that males and females have the same underlying effort distribution. As
shown in Table 8, using specification (3), the evidence is very much in line with
this observation. The loser is not treated differently in the single-sex male treat-
ments than in the single-sex female treatments. This finding is largely robust
across subgroups, as shown in Tables A.9- A.12.12
Result 4: We find no evidence of spectators choosing differently in single-

sex male treatments and single-sex female treatments, which is consistent with
spectators believing that males and females have the same effort distribution.
This result thus suggests that the observed gender bias is not driven by spec-
tators in general believing that males exert less effort than females. It is rather
consistent with the idea formulated in the theory framework, namely that spec-
tators believe that males have an advantage, which then implies that a male loser
is a stronger signal of low effort than a female loser.
Finally, we compare the mixed-sex merit treatments and the single-sex merit
treatments in Table 9, where we report regressions using specification (6). The
theoretical framework, Observation 4, predicts that we should observe the low-
est transfer to the loser in the mixed-sex treatment with a male loser, which is
exactly what we observe The male losers losing to a female are disadvantaged
12 The only exception is that low income spectators transfer significantly more to the loser in
the single-sex male treatments than in the single-sex female treatments when luck is the source
of inequality.
20
relative to the losers in the single-sex merit treatments: on average, these male
losers receive 0.138 standard deviation less (p = 0.015) and female losers losing
to a male receive 0.037 standard deviations more than the average loser in the
single-sex treatments (p = 0.518). This pattern is driven by the female specta-
tors, who give the male losers in the mixed-sex merit treatments 0.296 standard
deviations less than the losers in the single-sex merit treatments (p < 0.001),
while they give the female losers in the mixed-sex treatments almost the same
as the losers in the single-sex merit treatments.13 For the male spectators, we
observe that both male and female losers in the mixed-sex merit treatments re-
ceive slightly more than the losers in the single-sex merit treatments, but these
effects are not statistically significant.
Result 5: We find strong evidence of the male losers in the mixed-sex merit
treatments receiving less than the losers in the single-sex treatments and this
pattern is driven by the female spectators.
Taken together, our findings are consistent with a model of statistical fair-
ness discrimination, where spectators perceive a male loser in a mixed-sex merit
treatment as a particularly strong signal of low effort, and thus consider it fair to
transfer less to male losers than to female losers.
6 Concluding remarks
The emergence of the ‘boy crisis’ prompts the question of whether people in-
terpret gender inequalities differently depending on whether males or females
are lagging behind. We study this question in a novel large-scale economic
experiment conducted with a general population from the United States. The
participants act as spectators and distribute earnings between two workers in a
controlled labor market environment. When initial earnings are based on merit,
we find that the spectators are gender-biased against males. We show that this
gender bias is driven by female spectators and we provide evidence suggest-
ing that the underlying mechanism is statistical fairness discrimination, where
spectators interpret a male loser as someone who has exerted less effort than a
female loser.
13 Table
A.13 and Table A.14 report this analysis for all subgroups and for the analysis where
the dependent variable is the share giving nothing to the loser.
21
Our study provides new evidence on the nature of gender discrimination, by

showing how the perception of females being disadvantaged may cause people
to infer that low-performing males have exerted less effort than low-performing
females. This mechanism speaks to the ‘boy crisis’, by calling attention to the
possibility of a gender bias against low-performing males. Both in the edu-
cational system and at the workplace, a perception of males being advantaged
may lead us to interpret a losing male, more than a losing female, as someone
who has not exerted sufficient effort, and thus may find it fair that less resources
are allocated to assist them. Notably, however, as evident from our theoret-
ical model, statistical fairness discrimination may equally well work against
females, in settings where females are perceived to have an advantage. This
may for example be the case in settings where affirmative action for women
is perceived to create an advantage for females (and not only remove a disad-
vantage), where people may infer that females falling behind have exerted low
effort. More broadly, our findings speak to how we relate ourselves to individ-
uals struggling in society. In an era where American politics has become “a
personal responsibility crusade” (Hacker, 2006), it is of great importance that
we better understand how we arrive at our views about whether individuals who
are struggling should be held personally responsible for their situation (Moffitt,
2015).
The paper opens up several avenues for future research. First, it is important
to understand whether the identified gender bias extends to other settings where
males fall behind. In particular, it would be extremely interesting to study this in
an educational setting, to link it even more closely to the existing field literature
on the discrimination of boys in schools. Second, it would be interesting to
study whether the observed gender bias of female spectators is specific to the
gender domain or whether it reflects a more general behavioral phenomenon
of disadvantaged groups. Finally, we believe that our experimental design is
well suited for a number of different applications, including the study of ethnic
discrimination and discrimination of immigrants. We believe that the idea of
statistical fairness discrimination is powerful, and we hope that future research
will further our understanding of this mechanism.
22
References
Almås, Ingvild, Alexander W. Cappelen, Kjell G. Salvanes, Erik Ø. Sørensen,
and Bertil Tungodden (2016a). “What explains the gender gap in college track
dropout? Experimental and administrative evidence,” American Economic
Review: Papers & Proceedings, 106(5): 296–302.
Almås, Ingvild, Alexander W. Cappelen, and Bertil Tungodden (2016b). “Cut-
throat capitalism versus cuddly socialism: Are Americans more meritocratic
and efficiency-seeking than Scandinavians?” NHH Department of Economics
Discussion Paper No. 18/2016.
Arrow, Kenneth J. (1973). “The theory of discrimination,” in Orley Ashenfelter
and Albert Rees (eds.), “Discrimination in labor markets,” Princeton Univer-
sity Press, pp. 3–33.
Autor, David, David Figlio, Krzysztof Karbownik, Jeffrey Roth, and Melanie
Wasserman (forthcoming). “Family disadvantage and the gender gap in be-
havioral and educational outcomes,” American Economic Journal: Applied
Economics.
Autor, David and Melanie Wasserman (2013). “Wayward sons: The emerging
gender gap in labor markets and education,” Technical report, Third Way.
Babcock, Linda, Maria P. Recalde, Lise Vesterlund, and Laurie Weingart (2017).
“Gender differences in accepting and receiving requests for tasks with low
promotability,” American Economic Review, 107(3): 714–747.
Becker, Gary S. (1957). The economics of discrimination, University of Chicago
Press, Chicago and London.
Bertrand, Marianne and Esther Duflo (2017). Handbook of economic field exper-
iments, volume 1, chapter Field experiments on discrimination, pp. 309–393.
Bertrand, Marianne and Jessica Pan (2013). “The trouble with boys: Social
influences and the gender gap in disruptive behavior,” American Economic
Journal: Applied Economics, 5(1): 32–64.
Binder, Ariel J. and John Bound (2019). “The declining labor market prospects
of less-educated men,” NBER Working Paper No. 25577.
Blau, Francine D. and Lawrence M. Kahn (2017). “The gender wage gap: Ex-
tent, trend, and explanations,” Journal of Economic Literature, 55(3): 789–
865.
Blau, Francine D. and M. Kahn, Lawrence (2013). “Female labor supply: Why
23
is the Unites States falling behind?” American Economic Review: Papers and
Proceedings, 103(3): 251–256.
Bohren, Aislinn J., Alex Imas, and Michael Rosenberg (forthcoming). “The dy-
namics of discrimination: Theory and evidence,” American Economic Re-
view.
Bolton, Gary E. and Axel Ockenfels (2000). “ERC: A theory of equity, reci-
procity, and competition,” American Economic Review, 90(1): 166–193.
Bortolotti, Stefania, Ivan Soraperra, Matthias Sutter, and Claudia Zoller (2017).
“Too lucky to be true. Fairness views under the shadow of cheating,” IZA DP
No. 10877.
Breda, Thomas and Son Thierry Ly (2015). “Professors in core science fields
are not always biased against women: Evidence from France,” American Eco-
nomic Journal: Applied Economics, 7(4): 53–75.
Cappelen, Alexander W., Astri Drange Hole, Erik Ø. Sørensen, and Bertil Tun-
godden (2007). “The pluralism of fairness ideals: An experimental approach,”
American Economic Review, 97(3): 818–827.
Cappelen, Alexander W., James Konow, Erik Ø. Sørensen, and Bertil Tungod-
den (2013). “Just luck: An experimental study of risk taking and fairness,”
American Economic Review, 103(3): 1398–1413.
Castillo, Marco, Ragan Petrie, Maximo Torero, and Lise Vesterlund (2013).
“Gender differences in bargaining outcomes: A field experiment on discrim-
ination,” Journal of Public Economics, 99(C): 35–48.
Cettolin, Elena and Sigrid Suetens (forthcoming). “Return on trust is lower for
immigrants,” The Economic Journal.
Coffman, Katherine B., Christine L. Exley, and Muriel Niederle (2019). “When
gender discrimination is not about gender,” Mimeo, Harvard Business School.
Coile, Courtney and Mark Duggan (2019). “When labor’s lost: Health, family
life, incarceration, and education in a time of declining economic opportunity
for low-skilled men,” NBER Working Paper No. 25569.
Cornwell, Christopher, David B. Mustard, and Jessica Van Parys (2013).
“Noncognitive skills and the gender disparities in test scores and teacher as-
sessments: Evidence from primary school,” Journal of Human Resources,
48(1): 236–264.
Council of Economic Advisers (2016). “The long-term decline in US prime-age
male labour force participation,” Technical report, Executive Office of the
24
President.
Croson, Rachel and Uri Gneezy (2009). “Gender differences in preferences,”
Journal of Economic Literature, 47(2): 1–27.
Davison, A.C. and D. V. Hinkley (1997). Bootstrap methods and their applica-
tion, Cambridge University Press.
Eckel, Catherine C. and Philip J. Grossman (2001). “Chivalry and solidarity in
ultimatum games,” Economic Inquiry, 39(2): 171–188.
Eckel, Catherine C. and Philip J. Grossman (2008). Differences in the economic
decisions of men and women: Experimental evidence, volume 1, chapter 57,
Handbook of experimental economic results, Elsevier, 1st edition, pp. 509–
519.
Engel, Christoph (2011). “Dictator games: A meta study,” Experimental Eco-
nomics, 14(4): 583–610.
Exley, Christine L., Muriel Niederle, and Lise Vesterlund (forthcoming).
“Knowing when to ask: The cost of leaning-in,” Journal of Political Econ-
omy.
Falch, Torberg and Linn Renée Naper (2013). “Educational evaluation schemes
and gender gaps in student achievement,” Economics of Education Review,
36(C): 12–25.
Fehr, Ernst and Klaus M. Schmidt (1999). “A theory of fairness, competition
and cooperation,” Quarterly Journal of Economics, 114(3): 817–868.
Fershtman, Chaim and Uri Gneezy (2001). “Discrimination in a segmented so-
ciety: An experimental approach,” Quarterly Journal of Economics, 116(1):
351–377.
Goldin, Claudia and Cecilia Rouse (2000). “Orchestrating impartiality: The im-
pact of “blind” auditions on female musicians,” American Economic Review,
90(4): 715–741.
Hacker, Jacob S. (2006). The great risk shift, Oxford University Press.
Heikensten, Emma and Siri Isaksson (2016). “In favor of girls: A field study of
adults’ beliefs in children’s ability,” Mimeo, Stockholm School of Economics.
Konow, James (2000). “Fair shares: Accountability and cognitive dissonance in
allocation decisions,” American Economic Review, 90(4): 1072–1091.
Krueger, Alan B. (2017). “Where have all the workers gone?” Brookings Papers
on Economic Activity, 48: 1–87.
Lavy, Victor (2008). “Do gender stereotypes reduce girls’ or boys’ human cap-
25
ital outcomes? Evidence from a natural experiment,” Journal of Public Eco-

nomics, 92(10-11): 2083–2105.
Lavy, Victor and Edith Sand (2018). “On the origins of gender gaps in human
capital: Short- and long-term consequences of teachers’ biases,” Journal of
Public Economics, 167: 263–279.
Lindahl, Erica (2016). “Are teacher assessments biased? Evidence from Swe-
den,” Education Economics, 24(2): 224–238.
List, John A. (2004). “The nature and extent of discrimination in the market-
place: Evidence from the field,” Quarterly Journal of Economics, 119(1):
49–89.
Mengel, Friederike, Jan Sauermann, and Ulf Zölitz (2018). “Gender bias in
teaching evaluations,” Journal of the European Economic Association.
Moffitt, Robert A. (2015). “The deserving poor, the family, and the U.S. welfare
system,” Demography, 52(3): 729–749.
OECD (2015). “OECD stat export, htt p :
//stats.oecd.org/index.aspx?datasetcode = idd,” downloaded Febru-
ary 23, 2015.
OECD (2016). “Education at a glance 2016: OECD indicators,” Technical re-
port, OECD.
Phelps, Edmund S. (1972). “The statistical theory of racism and sexism,” Amer-
ican Economic Review, 62(4): 659–661.
Reuben, Ernesto, Paola Sapienza, and Luigi Zingales (2014). “How stereotypes
impair women’s careers in science,” Proceedings of the National Academy of
Sciences, 111(12): 4403–4408.
Reynolds, Tania, Hallgeir Sjåstad, Chuck Howard, Tyler Okimoto, Roy F.
Baumeister, Karl Aquino, and JongHan Kim (2017). “Man up and take it:
Greater concern for women’s than men’s suffering,” Mimeo, Indiana Univer-
sity Bloomington.
Romano, Joseph P. and Michael Wolf (2016). “Efficient computation of adjusted
p-values for resampling-based stepdown multiple testing,” Statistics & Prob-
ability Letters, 113(1): 38–40.
Rosin, Hanna (2012). The end of men and the rise of women, New York: River-
head Books.
Solnick, Sara J. (2001). “Gender differences in the ultimatum game,” Economic
Inquiry, 39(2): 189–200.
26
Terrier, Camille (2016). “Boys lag behind: How teachers’ gender biases affect
student achievement,” IZA Discussion Paper No. 10343.
Williams, Wendy M. and Stephen J. Ceci (2015). “National hiring experiments
reveal 2:1 faculty preference for women on STEM tenure track,” Proceedings
of the National Academy of Sciences, 112(17): 5360–5365.
27
Figure 1: Main findings

.25 .08
Difference in share transferred ± s.e.
.2 .06
Share transferred ± s.e.
.15 .04
.1 .02
.05 0
0 −.02
Female loser Male loser Male spectator Female spectator
Panel A: Share transferred to loser Panel B: Difference by spectator gender
Note: Panel A shows the mean share transferred to the female loser and the male loser in
the mixed-sex merit treatments, while Panel B shows the difference in the mean share
transferred in the same two main treatments by the spectator’s gender. The standard
errors are indicated by the bars.
28
Figure 2: Transfer to loser

.3
.2
Fraction
.1
0
0 1 2 3 4 5 6
USD
Note: The figure shows the distribution of transfers (in USD) to the losing worker,
pooled for all treatments.
Table 1: The stages of the experiment
1. Work stage: Workers complete an assignment.

2. Earnings stage: Workers matched in pairs. Assigned initial earnings according to treatment.
3. Redistribution stage: Each spectator decides for one pair of workers whether and how much to redistribute.
4. Payment stage: Workers in the pair paid according to the decision of the spectator.
29
30
Table 2: Descriptive statistics
Spectator sample US
Male (share) 0.488 0.492
Age (year)
Median 41 46
p10 23 23
p90 59 72
Income (USD)
Median 55000 57500
p10 19999 12500
p90 125000 167500
Republican (share) 0.337 0.270

Note: The table provides the descriptive statistics for
the spectator sample and the US population. Sample
(self-reported): The income variable is combined yearly
income in USD (gross income before taxes are de-
ducted) and given in standard categories where we use
the mid-point in the category (see Appendix B.3 for a
listing of income categories). A participant is classified
as Republican if he or she would have voted for the Re-
publican party if there was an election tomorrow. US:
The share of males and the median age (+18) in the US
are from the US Census Bureau, Population Division
(2016 and 2017) (https://www.census.gov/quickfacts/
and https://www.census.gov/data/). The income data
are based on the US Census Bureau, Current Population
Survey together with the 2016 and 2017 Annual Social
and Economic Supplement. Political affiliation is from
Gallup (http://news.gallup.com/poll/).
31
Table 3: Treatments
Mixed-sex Single-sex
Female loser Male loser Female loser Male loser
MERIT N=782 N=782 N=257 N=256
LUCK N=256 N=256 N=256 N=257
Note: The table provides an overview of the eight treatments in the experiment
and the number of spectators in each treatment. The two main treatments are
highlighted.
32
Table 4: Regression analysis: Transfer to loser
Mixed-sex merit
Amount to loser (std) Nothing to loser
Male loser -0.174 -0.173 0.073 0.073
(0.050) (0.050) (0.024) (0.024)
Male spectator 0.078 -0.011
(0.050) (0.024)
Republican -0.173 0.058
(0.054) (0.026)
Low income -0.011 0.009
(0.052) (0.025)
Low age -0.005 -0.000
(0.051) (0.024)
Constant 0.066 0.092 0.335 0.318
(0.051) (0.068) (0.024) (0.033)
Observations 1564 1564 1564 1564
R2 0.008 0.016 0.006 0.010
Note: The table reports OLS regressions, where the dependent
variable is the standardized amount transferred to the losing worker
(columns 1 and 2) or an indicator variable for transferring nothing
to the losing worker (columns 3 and 4). The sample and the basis
for the standardization are the mixed-sex merit treatments. Male
loser is an indicator for the spectator being in a treatment where
the male has lost. Male spectator is an indicator variable for being
male. Republican is an indicator for voting Republican. Low in-
come is an indicator for having an income below $55,000 (which
is the median income per year in the sample). Low age is an in-
dicator for being below 41 years old (which is the median age in
the sample). Included in all regressions, but not reported, is an in-
dicator for the spectator participating in the second round of data
collection. Standard errors in parentheses.
33
Table 5: Heterogeneity analysis: Amount to loser (std)
Mixed-sex merit
Gender Politics Income Age All
Male loser -0.283 -0.171 -0.228 -0.206 -0.378
(0.070) (0.061) (0.065) (0.071) (0.105)
Male spectator*Male loser 0.228 0.230
(0.101) (0.101)
Republican*Male loser -0.013 0.005
(0.107) (0.108)
Low income*Male loser 0.131 0.137
(0.103) (0.103)
Low age*Male loser 0.065 0.071
(0.101) (0.101)
Constant 0.108 0.141 0.110 0.098 0.219
(0.050) (0.044) (0.047) (0.051) (0.076)
Controls Yes Yes Yes Yes Yes
(Male loser+interaction) -0.055 -0.184 -0.096 -0.141
(0.072) (0.088) (0.079) (0.072)
Observations 1564 1564 1564 1564 1564
R2 0.012 0.014 0.009 0.008 0.020
Note: The table reports OLS regressions, where the dependent variable
is the standardized amount transferred to the losing worker. The sample
and the basis for the standardization are the mixed-sex merit treatments.
Male loser, Male spectator, Republican, Low income and Low age are de-
fined in Table 4. Male spectator*Male loser, Republican*Male loser, Low
income*Male loser and Low age*Male loser are interactions between the
respective characteristic and Male loser. Male loser+interaction is a linear
combination of Male loser and the respective interaction. Each regression
also includes the respective indicator variable interacted with Male loser
(see Table A.3 for the full table). Standard errors in parentheses.
34
Mixed-sex luck
All Female Male All Female Male
Male loser -0.043 0.012 -0.113 0.002 -0.023 0.034
(0.089) (0.120) (0.132) (0.038) (0.052) (0.057)
Male spectator -0.108 0.036
(0.089) (0.038)
Republican -0.140 -0.309 0.058 0.070 0.149 -0.020
(0.094) (0.126) (0.142) (0.041) (0.054) (0.061)
Low income -0.027 -0.078 0.032 0.007 -0.004 0.020
(0.092) (0.120) (0.140) (0.039) (0.052) (0.060)
Low age -0.108 -0.148 -0.026 -0.000 0.009 -0.030
(0.089) (0.120) (0.133) (0.038) (0.052) (0.057)
Constant 0.184 0.256 -0.014 0.198 0.184 0.257
(0.104) (0.127) (0.137) (0.045) (0.055) (0.059)
Observations 512 266 246 512 266 246
R2 0.010 0.031 0.004 0.007 0.030 0.003
Note: The table reports OLS regressions, where the dependent variable is the stan-
dardized amount transferred to the losing worker (columns 1-3) or an indicator
variable for transferring nothing to the losing worker (columns 4-6). The sample
and the basis for the standardization are the mixed-sex luck treatments: all specta-
tors (columns 1 and 4), female spectators (columns 2 and 5), and male spectators
(columns 3 and 6). Male loser, Male spectator, Republican, Low income and Low
age are defined in Table 4. Standard errors in parentheses.
35
Mixed-sex merit vs. mixed-sex luck

Female Male All Female Male All
Male loser -0.266 -0.052 -0.163 0.128 0.016 0.073
(0.064) (0.074) (0.048) (0.033) (0.034) (0.024)
Luck 0.580 0.606 0.594 -0.062 -0.129 -0.095
(0.100) (0.119) (0.077) (0.051) (0.055) (0.038)
Luck*Male loser 0.298 -0.074 0.118 -0.153 0.017 -0.070
(0.128) (0.149) (0.098) (0.065) (0.069) (0.047)
Male spectator 0.027 0.000
(0.042) (0.021)
Republican -0.225 -0.085 -0.158 0.099 0.020 0.060
(0.060) (0.069) (0.045) (0.031) (0.032) (0.022)
Low income -0.048 0.021 -0.013 0.016 -0.000 0.008
(0.056) (0.067) (0.043) (0.029) (0.031) (0.021)
Low age -0.064 0.008 -0.031 0.008 -0.011 -0.000
(0.055) (0.065) (0.042) (0.028) (0.030) (0.021)
Constant 0.063 -0.106 -0.035 0.255 0.371 0.312
(0.077) (0.088) (0.062) (0.040) (0.041) (0.030)
Male loser (luck) 0.032 -0.126 -0.045 -0.025 0.033 0.003
(0.111) (0.129) (0.085) (0.057) (0.060) (0.041)
Observations 1070 1006 2076 1070 1006 2076
R2 0.131 0.053 0.085 0.039 0.009 0.019
Note: The table reports OLS regressions, where the dependent variable is the standard-
ized amount transferred to the losing worker (columns 1-3) or an indicator variable for
transferring nothing to the losing worker (columns 4-6). The sample and the basis for
the standardization are the mixed-sex treatments: female spectators (columns 1 and 4),
male spectators (columns 2 and 5), and all spectators (columns 3 and 6). Male loser,
Male spectator, Republican, Low income and Low age are defined in Table 4. Luck is
an indicator for spectators being in one of the luck treatments. Luck*Male loser is an
interaction between Luck and Male loser. Male loser (luck) is a linear combination of
Male loser and Luck*Male loser. Included in all regressions, but not reported, is an in-
dicator for the spectator participating in the second round of data collection. Standard
errors in parentheses.
36
Single-sex
Merit Luck
Amount to Nothing Amount to Nothing
loser (std) to loser loser (std) to loser
Male loser -0.040 -0.031 -0.007 -0.010 0.087 0.091 -0.001 -0.003
(0.088) (0.089) (0.041) (0.041) (0.088) (0.088) (0.036) (0.036)
Male spectator -0.126 0.071 -0.020 -0.004
(0.088) (0.041) (0.088) (0.036)
Republican -0.181 0.088 -0.259 0.091
(0.095) (0.044) (0.092) (0.038)
Low income 0.035 -0.018 -0.033 0.012
(0.092) (0.043) (0.091) (0.037)
Low age 0.113 -0.037 -0.125 0.029
(0.088) (0.041) (0.089) (0.037)
Constant 0.020 0.072 0.323 0.285 -0.044 0.136 0.215 0.165
(0.062) (0.100) (0.029) (0.047) (0.063) (0.102) (0.026) (0.042)
Observations 513 513 513 513 513 513 513 513
R2 0.000 0.017 0.000 0.017 0.002 0.020 0.000 0.012
Note: The table reports OLS regressions, where the dependent variable is the standardized
amount transferred to the losing worker (columns 1-2 and 5-6) or an indicator variable for
transferring nothing to the losing worker (columns 3-4 and 7-8). The sample and the basis for
the standardization are the single-sex merit treatments (columns 1-4) and the single-sex luck
treatments (columns 5-8). Male loser, Male spectator, Republican, Low income and Low age
are defined in Table 4. Standard errors in parentheses.
37
Table 9: Single-sex vs. mixed-sex merit: Amount to loser (std)
Merit
All All Females Females Males Males
MS-Male loser -0.138 -0.137 -0.296 -0.292 0.028 0.024
(0.057) (0.057) (0.075) (0.075) (0.085) (0.085)
MS-Female loser 0.037 0.038 -0.012 -0.009 0.084 0.079
(0.057) (0.057) (0.076) (0.076) (0.084) (0.084)
Male spectator 0.028
(0.044)
Republican -0.175 -0.172 -0.175
(0.047) (0.063) (0.070)
Low income -0.001 -0.051 0.055
(0.045) (0.059) (0.068)
Low age 0.023 0.044 -0.003
(0.044) (0.058) (0.066)
Constant 0.038 0.070 0.108 0.157 -0.030 0.016
(0.044) (0.059) (0.059) (0.073) (0.065) (0.082)
MS-Male loser - -0.175 -0.174 -0.284 -0.283 -0.055 -0.055
MS-Female loser (0.050) (0.050) (0.067) (0.067) (0.076) (0.076)
Observations 2077 2077 1057 1057 1020 1020
R2 0.006 0.013 0.022 0.030 0.001 0.009
Note: The table reports OLS regressions, where the dependent variable is
the standardized amount transferred to the losing worker. The sample and
the basis for the standardization are the merit treatments: all spectators
(columns 1 and 2), female spectators (columns 3 and 4), and male specta-
tors (columns 5 and 6). The reference sample is the sample of spectators
in the single-sex merit treatments. MS-Male loser is an indicator for the
spectator being in a treatment where the male has lost to a female. MS-
Female loser is an indicator for the spectator being in a treatment where
the female has lost to a male. Male spectator, Republican, Low income
and Low age are defined in Table 4. MS-Male loser - MS-Female loser
is a linear combination of MS-Male loser and MS-Female loser. Standard
38
A Online Appendix: Additional analysis
A.1 Figures and tables
Figure A.1: Distribution of transfers to loser by treatment
Mixed-sex merit, female loser Mixed-sex merit, male loser

.5
.5
.4
.4
Fraction
Fraction
.3
.3
.2
.2
.1
.1
0
0 2 4 6 0 2 4 6
USD USD
Mixed-sex luck, female loser Mixed-sex luck, male loser

.5
.5
.4
.4
Fraction
Fraction
.3
.3
.2
.2
.1
.1
0
0 2 4 6 0 2 4 6
USD USD
Single-sex merit, female loser Single-merit, male loser

.5
.5
.4
.4
Fraction
Fraction
.3
.3
.2
.2
.1
.1
0
0 2 4 6 0 2 4 6
USD USD
Single-sex luck, female loser Single-sex luck, male loser

.5
.5
.4
.4
Fraction
Fraction
.3
.3
.2
.2
.1
.1
0
0 2 4 6 0 2 4 6
USD USD
Note: The figure shows the distribution of transfers (in USD) to the losing worker by
treatment.
39
Figure A.2: Distribution of transfers to loser in the mixed-sex merit treatments,

by round
Male loser, Round 1 Male loser, Round 2

.5
.5
.4
.4
Fraction
Fraction
.3
.3
.2
.2
.1
.1
0
0
0 2 4 6 0 2 4 6
USD USD
Female loser, Round 1 Female loser, Round 2

.5
.5
.4
.4
Fraction
Fraction
.3
.3
.2
.2
.1
.1
0
0 2 4 6 0 2 4 6
USD USD
Note: The upper panels (lower panels) show the distribution of transfers (in USD) to
the male loser (female loser) in the mixed-sex merit treatments by round.
40
Table A.1: Balance test
Male Republican Income Age

Mixed-sex merit -0.028 -0.005 2455.043 0.921
male loser (0.025) (0.024) (2114.984) (0.667)
Mixed-sex luck -0.020 0.026 538.868 -0.075
female loser (0.040) (0.038) (3356.926) (1.063)
Mixed-sex luck 0.011 -0.013 8095.877 1.332
male loser (0.040) (0.038) (3365.135) (1.063)
Single-sex merit 0.017 -0.010 3076.967 0.869
female loser (0.040) (0.038) (3394.785) (1.062)
Single-sex merit 0.027 0.007 2361.561 1.652
male loser (0.040) (0.038) (3373.458) (1.063)
Single-sex luck -0.001 0.030 3184.619 -0.571
female loser (0.040) (0.038) (3373.458) (1.063)
Single-sex luck -0.002 0.075 657.073 2.153
male loser (0.040) (0.038) (3365.135) (1.062)
Included control:
Round dummy Yes Yes Yes Yes
Observations 3102 3102 2954 3102
R2 0.001 0.002 0.005 0.004
Prob > F 0.8637 0.4264 0.3692 0.1718
Note: The table reports OLS regressions with the different spectator
background characteristics as dependent variables. For each background
characteristic, we have reported the p-value of the joint F-test testing
whether the eight treatments are significantly different from each other
with respect to that background characteristic. The reference category
across all regressions is the mixed-sex merit treatment where a female has
lost. The dependent variable in column 1 is male, an indicator for being
a male. The dependent variable in column 2 is republican, an indicator
for wanting to vote republican. The dependent variable in column 3 is
income. Income is the spectator’s reported combined, yearly household
income (gross income before taxes are deducted). The nine different
income categories we used are specified in Appendix B.3. The spectators
who did not know or want to state their income are not included in the
sample in column 3. The dependent variable in column 4 is age, which
reports the spectator’s age in number of years. Included in all regressions,
but not reported, is an indicator for the spectator participating in the
second round of data collection. Standard errors in parentheses.
41
Table A.2: Regression analysis: Transfer to loser by round
A. Amount to loser (std) Mixed-sex merit

Round 1 Round 2
Male loser -0.170 -0.169 -0.176 -0.175
(0.088) (0.088) (0.062) (0.061)
(0.088) (0.061)
Republican -0.207 -0.161
(0.095) (0.066)
Low income -0.018 -0.017
(0.092) (0.063)
Low age 0.127 -0.069
(0.089) (0.062)
Constant 0.064 0.049 0.098 0.147
(0.062) (0.103) (0.044) (0.072)
Observations 514 514 1050 1050
R2 0.007 0.022 0.008 0.016
B. Nothing to loser Mixed-sex merit
Round 1 Round 2
Male loser 0.058 0.062 0.080 0.080
(0.043) (0.043) (0.029) (0.029)
Male spectator 0.020 -0.026
(0.043) (0.029)
Republican 0.077 0.050
(0.046) (0.032)
Low income 0.036 -0.001
(0.045) (0.030)
Low age -0.031 0.014
(0.043) (0.030)
Constant 0.342 0.308 0.309 0.299
(0.030) (0.050) (0.021) (0.034)
Observations 514 514 1050 1050
R2 0.004 0.011 0.007 0.010
Note: The table reports OLS regressions, where the dependent variable is
the standardized amount transferred to the losing worker (Panel A) or an
indicator for the spectator transferring nothing to the losing worker (Panel
B). The sample and the basis for the standardization are the mixed-sex merit
treatments. Male loser, Male spectator, Republican, Low income and Low
age are defined in Table 4. Standard errors in parentheses.
42
Table A.3: Heterogeneity analysis: Amount to loser (std)
Mixed-sex merit
Male loser -0.283 -0.171 -0.228 -0.206 -0.378
(0.070) (0.061) (0.065) (0.071) (0.105)
(0.101) (0.101)
Republican*Male loser -0.013 0.005
(0.107) (0.108)
(0.103) (0.103)
(0.101) (0.101)
Male spectator -0.042 -0.036
(0.071) (0.071)
(0.075) (0.077)
Low income -0.056 -0.078
(0.072) (0.073)
Low age -0.020 -0.045
(0.071) (0.072)
Constant 0.108 0.141 0.110 0.098 0.219
(0.050) (0.044) (0.047) (0.051) (0.076)
(Male loser+interaction) -0.055 -0.184 -0.096 -0.141
(0.072) (0.088) (0.079) (0.072)
Observations 1564 1564 1564 1564 1564
R2 0.012 0.014 0.009 0.008 0.020
is the standardized amount transferred to the losing worker. The sample
and the basis for the standardization are the mixed-sex merit treatments.
Male loser, Male spectator, Republican, Low income and Low age are
defined in Table 4. Male spectator*Male loser, Republican*Male loser,
Low income*Male loser, Low age*Male loser and Male loser+interaction
are defined in Table 5. Standard errors in parentheses.
43
Table A.4: Heterogeneity analysis: Nothing to loser
Mixed-sex merit
Male loser 0.127 0.070 0.099 0.065 0.146
(0.034) (0.029) (0.031) (0.034) (0.050)
Male spectator*Male loser -0.112 -0.114
(0.048) (0.048)
Republican*Male loser 0.010 0.007
(0.051) (0.052)
Low income*Male loser -0.065 -0.068
(0.049) (0.050)
(0.048) (0.049)
(0.034) (0.034)
(0.036) (0.037)
Low income 0.034 0.041
(0.035) (0.035)
Low age -0.013 -0.006
(0.034) (0.035)
Constant 0.297 0.303 0.306 0.326 0.265
(0.024) (0.021) (0.022) (0.024) (0.037)
(Male loser+interaction) 0.015 0.080 0.034 0.081
(0.035) (0.042) (0.038) (0.034)
Observations 1564 1564 1564 1564 1564
R2 0.009 0.009 0.007 0.006 0.014
is an indicator for transferring nothing to the losing worker. The sample is
the mixed-sex merit treatments. Male loser, Male spectator, Republican,
Low income and Low age are defined in Table 4. Male spectator*Male
loser, Republican*Male loser, Low income*Male loser, Low age*Male
loser and Male loser+interaction are defined in Table 5. Standard errors
in parentheses.
44
Mixed-sex luck
No Gender Politics Income Age All
interaction
Male loser -0.033 0.051 -0.045 -0.187 0.069 -0.035
(0.088) (0.123) (0.108) (0.114) (0.124) (0.183)
(0.177) (0.178)
(0.188) (0.189)
(0.181) (0.183)
Low age*Male loser -0.219 -0.252
(0.177) (0.178)
(0.125) (0.126)
(0.131) (0.131)
Low income -0.200 -0.215
(0.126) (0.127)
Low age 0.003 0.017
(0.125) (0.126)
Constant 0.017 0.026 0.067 0.103 0.015 0.174
(0.063) (0.085) (0.078) (0.083) (0.090) (0.133)
(Male loser+interaction) -0.118 -0.026 0.197 -0.150
(0.128) (0.154) (0.140) (0.126)
Observations 512 512 512 512 512 512
R2 0.000 0.005 0.004 0.009 0.006 0.024
dardized amount transferred to the losing worker. The sample and the basis for the
standardization are the mixed-sex luck treatments. Male loser, Male spectator, Re-
publican, Low income and Low age are defined in Table 4. Male spectator*Male
loser, Republican*Male loser, Low income*Male loser, Low age*Male loser and Male
loser+interaction are defined in Table 5. Standard errors in parentheses.
45
Mixed-sex luck
interaction
Male loser 0.000 -0.032 -0.019 0.082 -0.032 0.003
(0.038) (0.053) (0.046) (0.049) (0.053) (0.078)
(0.076) (0.076)
(0.081) (0.081)
(0.077) (0.079)
(0.076) (0.076)
(0.054) (0.054)
(0.056) (0.056)
(0.054) (0.054)
Low age -0.035 -0.042
(0.054) (0.054)
Constant 0.242 0.241 0.229 0.200 0.260 0.199
(0.027) (0.037) (0.033) (0.035) (0.039) (0.057)
Male loser + interaction 0.032 0.046 -0.125 0.033
(0.055) (0.066) (0.060) (0.054)
Observations 512 512 512 512 512 512
R2 0.000 0.003 0.007 0.014 0.001 0.025
Note: The table reports OLS regressions, where the dependent variable is an indicator
variable for transferring nothing to the losing worker. The sample is the mixed-sex luck
treatments. Male loser, Male spectator, Republican, Low income and Low age are de-
fined in Table 4. Male spectator*Male loser, Republican*Male loser, Low income*Male
loser, Low age*Male loser and Male loser+interaction are defined in Table 5. Standard
Table A.7: Regression analysis: Amount to loser (std)

Female Male Republican Non- High income Low income High age Low age All
Republican
Male loser -0.266 -0.052 -0.172 -0.164 -0.212 -0.086 -0.203 -0.127 -0.163
(0.064) (0.074) (0.086) (0.059) (0.063) (0.076) (0.068) (0.069) (0.048)
Luck 0.580 0.606 0.611 0.575 0.663 0.498 0.625 0.550 0.594
(0.100) (0.119) (0.135) (0.094) (0.102) (0.119) (0.110) (0.108) (0.077)
Luck*Male loser 0.298 -0.074 0.159 0.114 0.009 0.285 0.280 -0.046 0.118
(0.128) (0.149) (0.173) (0.118) (0.126) (0.155) (0.137) (0.140) (0.098)
Male spectator 0.123 -0.012 0.006 0.068 0.004 0.066 0.027
(0.075) (0.051) (0.055) (0.067) (0.059) (0.060) (0.042)
Republican -0.225 -0.085 -0.184 -0.103 -0.246 -0.059 -0.158
(0.060) (0.069) (0.057) (0.076) (0.062) (0.067) (0.045)
Low income -0.048 0.021 0.043 -0.035 0.009 -0.041 -0.013
(0.056) (0.067) (0.080) (0.051) (0.062) (0.061) (0.043)
Low age -0.064 0.008 0.105 -0.092 -0.009 -0.075 -0.031
(0.055) (0.065) (0.076) (0.051) (0.055) (0.067) (0.042)
Constant 0.063 -0.106 -0.343 0.041 -0.006 -0.078 -0.055 -0.042 -0.035
(0.077) (0.088) (0.103) (0.073) (0.077) (0.092) (0.082) (0.082) (0.062)
Male loser (luck) 0.032 -0.126 -0.014 -0.050 -0.203 0.199 0.077 -0.173 -0.045
(0.111) (0.129) (0.150) (0.103) (0.110) (0.135) (0.119) (0.122) (0.085)
Observations 1070 1006 686 1390 1235 841 1046 1030 2076
R2 0.131 0.053 0.087 0.082 0.095 0.080 0.112 0.069 0.085
Note: The table reports OLS regressions, where the dependent variable is the standardized amount transferred to the losing worker.
46
The sample and the basis for the standardization are the mixed-sex treatments, with the different subgroups (columns 1-8) and the full
sample (column 9). Male loser, Male spectator, Republican, Low income and Low age are defined in Table 4. Luck is an indicator for
spectators being in the mixed-sex luck treatments. Luck*Male loser is an interaction between Luck and Male loser. Male loser (luck) is
a linear combination of Male loser and Luck*Male loser. Included in all regressions, but not reported, is an indicator for the spectator
participating in the second round of data collection. Standard errors in parentheses.
Table A.8: Regression analysis: Nothing to loser

Female Male Republican Non- High income Low income High age Low age All
Republican
Male loser 0.128 0.016 0.081 0.072 0.100 0.036 0.072 0.081 0.073
(0.033) (0.034) (0.042) (0.028) (0.031) (0.037) (0.033) (0.034) (0.024)
Luck -0.062 -0.129 -0.107 -0.082 -0.114 -0.069 -0.092 -0.090 -0.095
(0.051) (0.055) (0.066) (0.046) (0.049) (0.058) (0.054) (0.053) (0.038)
Luck*Male loser -0.153 0.017 -0.047 -0.090 -0.015 -0.160 -0.103 -0.047 -0.070
(0.065) (0.069) (0.085) (0.057) (0.061) (0.076) (0.067) (0.068) (0.047)
Male spectator -0.058 0.024 0.002 -0.006 0.003 -0.009 0.000
(0.037) (0.025) (0.026) (0.033) (0.029) (0.029) (0.021)
Republican 0.099 0.020 0.086 0.013 0.113 0.001 0.060
(0.031) (0.032) (0.027) (0.037) (0.030) (0.032) (0.022)
Low income 0.016 -0.000 -0.039 0.027 0.023 -0.006 0.008
(0.029) (0.031) (0.039) (0.025) (0.030) (0.030) (0.021)
Low age 0.008 -0.011 -0.075 0.033 0.011 -0.014 -0.000
(0.028) (0.030) (0.037) (0.025) (0.027) (0.033) (0.021)
Constant 0.255 0.371 0.455 0.271 0.275 0.373 0.302 0.321 0.312
(0.040) (0.041) (0.051) (0.035) (0.037) (0.045) (0.040) (0.040) (0.030)
Male loser (luck) -0.025 0.033 0.034 -0.017 0.085 -0.124 -0.031 0.034 0.003
(0.057) (0.060) (0.074) (0.050) (0.053) (0.066) (0.058) (0.059) (0.041)
Observations 1070 1006 686 1390 1235 841 1046 1030 2076
R2 0.039 0.009 0.026 0.020 0.029 0.018 0.030 0.017 0.019
Note: The table reports OLS regressions, where the dependent variable is an indicator variable for transferring nothing to the losing
47
worker. The sample is the mixed-sex treatments, with the different subgroups (columns 1-8) and the full sample (column 9). Male loser,
Male spectator, Republican, Low income and Low age are defined in Table 4. Luck is an indicator for spectators who belong to either
of the luck treatments. Luck*Male loser is an interaction between Luck and Male loser. Male loser (luck) is a linear combination of
Male loser and Luck*Male loser. Included in all regressions, but not reported, is an indicator for the spectator participating in the second
round of data collection. Standard errors in parentheses.
48
Single-sex merit
interaction
Male loser -0.040 -0.108 -0.075 -0.132 0.088 -0.109
(0.088) (0.126) (0.107) (0.111) (0.122) (0.181)
(0.177) (0.177)
(0.188) (0.190)
(0.184) (0.185)
(0.177) (0.177)
(0.125) (0.125)
(0.134) (0.134)
Low income -0.064 -0.076
(0.133) (0.133)
Low age 0.244 0.240
(0.125) (0.124)
Constant 0.020 0.124 0.102 0.041 -0.104 0.099
(0.062) (0.088) (0.075) (0.076) (0.089) (0.125)
(Male loser+interaction) 0.028 0.043 0.106 -0.165
(0.124) (0.155) (0.147) (0.128)
Observations 513 513 513 513 513 513
R2 0.000 0.006 0.010 0.005 0.008 0.026
standardization are the single-sex merit treatments. Male loser is an indicator for the
spectator being in a treatment of two males. Male spectator, Republican, Low income
and Low age are defined in Table 4. Male spectator*Male loser, Republican*Male
loser, Low income*Male loser and Low age*Male loser are interactions between the
respective characteristic and Male loser. Male loser+interaction is a linear combination
of Male loser and the respective interaction. Standard errors in parentheses.
49
Single-sex merit
interaction
Male loser -0.007 -0.001 -0.003 0.020 -0.065 -0.033
(0.041) (0.059) (0.050) (0.052) (0.057) (0.084)
(0.082) (0.083)
Republican*Male loser -0.015 -0.019
(0.088) (0.089)
(0.086) (0.087)
(0.083) (0.083)
(0.058) (0.058)
(0.063) (0.063)
(0.062) (0.062)
Low age -0.098 -0.097
(0.058) (0.058)
Constant 0.323 0.281 0.290 0.322 0.373 0.300
(0.029) (0.041) (0.035) (0.035) (0.042) (0.059)
Male loser + interaction -0.013 -0.019 -0.046 0.053
(0.058) (0.072) (0.069) (0.060)
Observations 513 513 513 513 513 513
R2 0.000 0.007 0.010 0.002 0.006 0.022
variable for transferring nothing to the losing worker. The sample is the single-sex
merit treatments. Male loser is an indicator for the spectator being in a treatment of
two males. Male spectator, Republican, Low income and Low age are defined in Table
4. Male spectator*Male loser, Republican*Male loser, Low income*Male loser, Low
age*Male loser and Male loser+interaction are defined in Table A.9. Standard errors in
parentheses.
50
Single-sex luck
interaction
Male loser 0.087 0.087 0.059 -0.079 -0.028 -0.194
(0.088) (0.123) (0.111) (0.112) (0.124) (0.183)
Male spectator*Male loser 0.000 -0.032
(0.177) (0.177)
(0.181) (0.184)
(0.181) (0.181)
(0.177) (0.178)
(0.125) (0.125)
(0.130) (0.131)
Low income -0.250 -0.245
(0.129) (0.129)
Low age -0.202 -0.205
(0.125) (0.125)
Constant -0.044 -0.030 0.060 0.048 0.062 0.269
(0.063) (0.087) (0.077) (0.078) (0.090) (0.126)
Male loser + interaction 0.087 0.161 0.360 0.194
(0.127) (0.143) (0.143) (0.126)
Observations 513 513 513 513 513 513
R2 0.002 0.002 0.016 0.013 0.007 0.033
standardization are the single-sex luck treatments. Male loser is an indicator for the
spectator being in a treatment of two males. Male spectator, Republican, Low income
and Low age are defined in Table 4. Male spectator*Male loser, Republican*Male
loser, Low income*Male loser, Low age*Male loser and Male loser+interaction are
defined in Table A.9. Standard errors in parentheses.
51
Single-sex luck
interaction
Male loser -0.001 0.006 0.000 0.065 0.048 0.112
(0.036) (0.051) (0.046) (0.046) (0.051) (0.076)
(0.073) (0.073)
Republican*Male loser -0.013 -0.026
(0.075) (0.076)
(0.075) (0.075)
(0.073) (0.074)
(0.052) (0.052)
(0.054) (0.054)
(0.053) (0.053)
Low age 0.066 0.066
(0.051) (0.051)
Constant 0.215 0.212 0.182 0.179 0.180 0.110
(0.026) (0.036) (0.032) (0.032) (0.037) (0.052)
Male loser + interaction -0.008 -0.013 -0.108 -0.050
(0.052) (0.059) (0.059) (0.052)
Observations 513 513 513 513 513 513
R2 0.000 0.000 0.010 0.011 0.004 0.024
variable for transferring nothing to the losing worker. The sample is the single-sex
luck treatments. Male loser is an indicator for the spectator being in a treatment of
two males. Male spectator, Republican, Low income and Low age are defined in Table
4. Male spectator*Male loser, Republican*Male loser, Low income*Male loser, Low
age*Male loser and Male loser+interaction are defined in Table A.9. Standard errors in
parentheses.
Table A.13: Single-sex vs. mixed-sex merit: Amount to loser (std)
Merit
All Females Males Republican Non- Low income High income Low age High age
Republican
MS-Male loser -0.137 -0.292 0.024 -0.120 -0.143 -0.122 -0.142 -0.170 -0.098
(0.057) (0.075) (0.085) (0.099) (0.069) (0.092) (0.072) (0.083) (0.077)
MS-Female loser 0.038 -0.009 0.079 0.066 0.028 -0.030 0.084 -0.035 0.114
(0.057) (0.076) (0.084) (0.099) (0.069) (0.091) (0.072) (0.083) (0.078)
Male spectator 0.028 0.029 0.033 0.100 -0.011 0.005 0.059
(0.044) (0.076) (0.054) (0.069) (0.056) (0.063) (0.060)
Republican -0.175 -0.172 -0.175 -0.104 -0.215 -0.128 -0.226
(0.047) (0.063) (0.070) (0.080) (0.058) (0.071) (0.063)
Low income -0.001 -0.051 0.055 0.096 -0.038 -0.081 0.084
(0.045) (0.059) (0.068) (0.082) (0.054) (0.065) (0.062)
Low age 0.023 0.044 -0.003 0.106 -0.012 -0.079 0.090
(0.044) (0.058) (0.066) (0.077) (0.054) (0.070) (0.056)
Constant 0.070 0.157 0.016 -0.188 0.109 0.091 0.058 0.165 -0.001
(0.059) (0.073) (0.082) (0.099) (0.068) (0.088) (0.072) (0.080) (0.074)
MS-Male loser - -0.174 -0.283 -0.055 -0.186 -0.172 -0.092 -0.225 -0.135 -0.212
MS-Female loser (0.050) (0.067) (0.076) (0.087) (0.062) (0.079) (0.065) (0.073) (0.070)
Observations 2077 1057 1020 682 1395 824 1253 1021 1056
R2 0.013 0.030 0.009 0.011 0.007 0.008 0.023 0.010 0.023
Note: The table reports OLS regressions, where the dependent variable is the standardized amount transferred to the losing worker. The sample
and the basis for the standardization are the merit treatments, with the full sample (column 1) and all the subgroups (columns 2-9). The reference
52
sample is the spectators in the single-sex merit treatments. MS-Male loser is an indicator for the spectator being in a treatment where a male
has lost to a female. MS-Female loser is an indicator for the spectator being in a treatment where a female has lost to a male. Male spectator,
Republican, Low income and Low age are defined in Table 4. MS-Male loser - MS-Female loser is a linear combination of MS-Male loser and
MS-Female loser. Standard errors in parentheses.
Table A.14: Single-sex vs. mixed-sex merit: Nothing to loser
Merit
All Females Males Republican Non- Low income High income Low age High age
Republican
MS-Male loser 0.073 0.141 0.002 0.050 0.083 0.074 0.073 0.095 0.052
(0.027) (0.038) (0.039) (0.049) (0.032) (0.044) (0.034) (0.039) (0.037)
MS-Female loser -0.000 0.015 -0.013 -0.033 0.012 0.040 -0.027 0.014 -0.018
(0.027) (0.038) (0.038) (0.048) (0.032) (0.044) (0.034) (0.039) (0.038)
Male spectator 0.009 0.011 0.004 -0.022 0.027 0.021 -0.008
(0.021) (0.037) (0.025) (0.033) (0.027) (0.030) (0.029)
Republican 0.065 0.060 0.069 0.026 0.086 0.009 0.114
(0.022) (0.032) (0.032) (0.038) (0.028) (0.033) (0.031)
Low income 0.002 0.024 -0.023 -0.044 0.019 0.005 -0.006
(0.021) (0.030) (0.031) (0.040) (0.025) (0.031) (0.030)
Low age -0.008 -0.020 0.006 -0.081 0.025 0.003 -0.016
(0.021) (0.029) (0.030) (0.038) (0.025) (0.034) (0.027)
Constant 0.297 0.263 0.338 0.428 0.267 0.302 0.294 0.284 0.305
(0.028) (0.036) (0.037) (0.049) (0.032) (0.042) (0.034) (0.037) (0.036)
MS-Male loser - 0.073 0.127 0.015 0.083 0.072 0.034 0.100 0.080 0.070
MS-Female loser (0.024) (0.033) (0.035) (0.043) (0.029) (0.038) (0.031) (0.034) (0.034)
Observations 2077 1057 1020 682 1395 824 1253 1021 1056
R2 0.010 0.023 0.006 0.014 0.007 0.005 0.018 0.008 0.017
Note: The table reports OLS regressions, where the dependent variable is an indicator variable for transferring nothing to the losing worker. The
sample is the merit treatments, with the full sample (column 1) and all the subgroups (columns 2-9). The reference sample is the spectators in the
53
single-sex merit treatments. MS-Male loser is an indicator for the spectator being in a treatment where a male has lost to a female. MS-Female
loser is an indicator for the spectator being in a treatment where a female has lost to a male. Male spectator, Republican, Low income and Low
age are defined in Table 4. MS-Male loser - MS-Female loser is a linear combination of MS-Male loser and MS-Female loser. Standard errors
in parentheses.
54
A.2 Multiple hypothesis testing

We here report the p-values adjusted for multiple hypothesis testing. We calcu-
late unadjusted p-values as bootstrap p-values following Davison and Hinkley
(1997) and compute p-values adjusted for stepdown multiple testing following
the algorithm proposed by Romano and Wolf (2016). Bootstrapping is done
with 10000 replications.
Table A.15: Multiple hypothesis adjustments: Average treatment effects (male

loser vs. female loser)
Differences Unadjusted Romano-Wolf

p-values adjusted
p-values
Mixed-sex
Merit (male vs. female loser) 0.164 0.001 0.003
Luck (male vs. female loser) -0.044 0.624 0.855
Single-sex
Merit (male vs. female loser) -0.035 0.665 0.855
Luck (male vs. female loser) 0.092 0.279 0.641
Note: The multiple hypothesis adjustment is based on the following OLS regression
specification
ui = α + β1 T 2i + β2 T 3i + β3 T 4i + β4 T 5i + β5 T 6i + β6 T 7i + β7 T 8i + γXi + εi
where ui is the standardized amount transferred by spectator i to the losing worker

(based on the full sample). Xi is a vector of control variables and includes indicator
variables for gender, political party, income and age as defined in Table 4, and εi is
an error term. Column 1 reports the estimated treatment effects (the difference in the
standardized amount transferred to a male loser and a female loser). Column 2 reports
the unadjusted p-values and column 3 reports the Romano-Wolf adjusted p-values.
55
Table A.16: Multiple hypothesis adjustments: Treatment effect (male loser

vs. female loser) in mixed-sex merit for all subgroups

p-values adjusted
p-values
Female -0.283 0.000 0.001
Male -0.055 0.477 0.477
Republican -0.184 0.035 0.136
Non-Republican -0.171 0.006 0.028
High income -0.228 0.001 0.003
Low income -0.096 0.223 0.400
High age -0.206 0.003 0.022
Low age -0.141 0.051 0.140
Note: The multiple hypothesis adjustment is based on specification (4), one for
each dimension: gender, political party, income and age. The dependent variable is
the standardized amount transferred to the losing worker. The sample and the basis
for the standardization are the mixed-sex merit treatments. Column 1 reports the
estimated treatment effects (the difference in the standardized amount transferred
to a male loser and a female loser). Column 2 reports the unadjusted p-values and
column 3 reports the Romano-Wolf adjusted p-values.
56
Table A.17: Multiple hypothesis adjustments: Subgroup differences in treatment ef-

fect (male loser vs. female loser) in mixed-sex merit

p-values adjusted
p-values
Males vs. females 0.230 0.025 0.085
Republicans vs. non-republicans 0.005 0.959 0.959
Low inc. vs. high inc. 0.137 0.188 0.459
Low age vs high age 0.071 0.490 0.731
Note: The multiple hypothesis test is based on specification (4), including interaction vari-
ables and background variables for gender, political party, income and age as specified in
Table 4 and Table 5. The dependent variable is the standardized amount transferred to the
losing worker. The sample and the basis for the standardization are the mixed-sex merit treat-
ments. Column 1 reports the estimated subgroup differences in treatment effect (the difference
in the standardized amount transferred to a male loser and a female loser). Column 2 reports
the unadjusted p-values and column 3 reports the Romano-Wolf adjusted p-values.
57
Table A.18: Multiple hypothesis adjustments: Difference in treatment effect (male

loser vs. female loser) in mixed-sex luck vs. mixed-sex merit, for all subgroups:

p-values adjusted
p-values
Female 0.305 0.028 0.187
Male -0.076 0.627 0.923
Republican 0.135 0.479 0.878
Non-Republican 0.110 0.374 0.853
High income 0.008 0.953 0.953
Low income 0.286 0.084 0.335
High age 0.274 0.061 0.288
Low age -0.052 0.723 0.923
Note: The multiple hypothesis test is based on OLS regressions with the standardized
amount transferred to the losing worker as the dependent variable, one regression for each
dimension of background characteristics; gender, political party, income and age. The
sample and the basis for the standardization are the mixed-sex treatments. Each regression
includes indicator variables Male loser, Male spectator, Republican, Low income and
Low age as defined in Table 4, Luck and Luck*Male loser as defined in Table 7, and
interaction variables between Male loser, between Luck and between Luck*Male loser
and the relevant background characteristic. Included in all regressions, but not reported, is
an indicator for the spectator participating in the second round of data collection. Column
1 reports the estimated differences in treatment effect (the difference in the standardized
amount transferred to a male loser and a female loser) between the mixed-sex luck and the
mixed-sex merit treatments for each subgroup, column 2 reports the unadjusted p-values,
and column 3 reports the Romano-Wolf adjusted p-values.
58
B Online Appendix: Instructions
B.1 Instructions, main treatments

Treatment 1: Mixed-sex merit, losing female
In contrast to traditional survey questions that are about hypothetical situations,
we now ask you to make a choice that has consequences for a real life situation.
A few days ago two workers were recruited via an online labor market to con-
duct an assignment. They were both from the US; a man and a woman of the
same age.
They were each paid a participation compensation of 2 USD regardless of

what they would end up being paid for the assignment. After completing the
assignment, they were told that their earnings from the assignment would be
determined by their productivity. The most productive worker would earn 6 USD
for the assignment and the other worker would earn nothing for the assignment.
They were not informed about who was the most productive worker. However,
they were told that a third person would be informed about the assignment and
who was the most productive worker. They were also told that this third person
would be given the opportunity to redistribute the earnings and thus determine
how much they were paid for the assignment.
You are the third person and we now want you to choose whether to redis-
tribute the earnings for the assignment between the two workers. Your decision
is completely anonymous. The workers will receive the payment that you choose
for the assignment within a few days, but will not receive any further informa-
tion.
The man was most productive and earned 6 USD for the assignment. The
woman was least productive and earned nothing for the assignment.
Please state which of the following alternatives you choose:
I do not redistribute:
• The most productive worker is paid 6 USD and the least productive worker
is paid 0 USD.
I do redistribute:
59
is paid 1 USD.
is paid 2 USD.
is paid 3 USD.
is paid 4 USD.
is paid 5 USD.
is paid 6 USD.
B.2 Instructions, additional treatments

Treatment 3: Luck, mixed-sex, unlucky female
duct an assignment. They were both from the US; a man and a woman of the
same age.

assignment, they were told that their earnings from the assignment would be de-
termined by a lottery. The worker winning the lottery would earn 6 USD for the
assignment and the other worker would earn nothing for the assignment. They
were not informed about the outcome of the lottery. However, they were told that
a third person would be informed about the assignment and the outcome of the
lottery. They were also told that this third person would be given the opportunity
to redistribute the earnings and thus determine how much they were paid for the
assignment.
60
tion.
The man was lucky, won the lottery and earned 6 USD for the assignment.
The woman was unlucky and earned nothing for the assignment.
• The lucky worker is paid 6 USD and the unlucky worker is paid 0 USD.
I do redistribute:
Treatment 5: Merit, single-sex, two females

duct an assignment. They were both from the US; two women of the same age.

assignment, they were told that their earnings from the assignment would be
determined by their productivity. The most productive worker would earn 6 USD
61
for the assignment and the other worker would earn nothing for the assignment.
They were not informed about who was the most productive worker. However,
they were told that a third person would be informed about the assignment and
who was the most productive worker. They were also told that this third person
would be given the opportunity to redistribute the earnings and thus determine
how much they were paid for the assignment.
tion.
One of the women was most productive and earned 6 USD for the assign-
ment. The other woman was least productive and earned nothing for the assign-
ment.
is paid 0 USD.
I do redistribute:
is paid 1 USD.
is paid 2 USD.
is paid 3 USD.
is paid 4 USD.
is paid 5 USD.
62
is paid 6 USD.
B.3 Background questions

In addition, the spectators will answer the following set of background ques-
tions:
• (...) we would be grateful if you could type in your actual age below?
• Are you?
– Male
– Female
• In which region do you live? (State in the United States)
• What is your household’s combined yearly income (gross income before

taxes are deducted)?
– Less than $20.000

– $20.000 - $29.999
– $30.000 - $39.999
– $40.000 - $49.999
– $50.000 - $59.999
– $60.000 - $74.999
– $75.000 - $99.999
– $100.000 - $149.999
– $150.000 or more
– Do not know/prefer not to state
• Which political party would you vote for if there was an election tomor-
row?
– Republican
– Democratic
– Other
63
B.4 Beliefs and attitudes questions

Question 1) US 8th graders were tested in mathematics. How do you think the
male students performed relative to the female students in
• a) mathematics?
– Males much better

– Males somewhat better
– Equal performance
– Females somewhat better
– Females much better
We then asked the same question about reading (1b), by inserting reading
instead of mathematics in the question structure above.
Question 2) Do you generally favor or oppose affirmative action programs

for women?14
• Generally favor
• Generally oppose
B.5 HIT announcement and worker instructions

We recruited the workers in two rounds of data collection. The complete in-
structions for the workers in each of the two rounds are provided below. In
the first round of data collection, we recruited 1370 workers, 685 men and 685
women living in the US. In the second round, we recruited 702 workers, 351
men and 351 women. In total, we ended up with 2055 unique pairs of assign-
ments/workers in the first round of data collection and 1053 in the second.
14 Question 1a and 1b are based on a nationally representative assessment conducted in 2015
by The National Assessment of Educational Progress (NAEP). Here, male and female 8th
graders performed equally well on the mathematics test (each group with an average of 282
points), while females on average performed slightly better than males on the reading test (271
vs. 260 points). For more details, see http://www.nationsreportcard.gov/reading math 2015/.
Question 2 is taken from Gallup’s Minority Rights and Relations survey conducted in 2015 with
more than 2000 US adults. They found that 67% of Americans are in favor of affirmative action
for women, with females being more prone to support it than males (72% vs. 62%).
64
B.5.1 Round 1
Qualtrics Survey Software Page 1 of 18
Introduction
Please read the instructions below carefully
General instructions:
The results from this experiment will be used in a research project. It is therefore important that you carefully
read and follow all instructions. Note that you will remain anonymous throughout the experiment. We will only
use your Worker ID to assign payments and check that you have not participated in this experiment before.
You will be paid a fixed participation fee of 2 USD and you may, depending on the actions you and others
take, earn additional money.
You will be given detailed instructions on your screen before each part of the experiment. Please read the
instructions to each part carefully.
If you have any questions regarding this experiment, you may contact thechoicelab@nhh.no
I have read and understood the above and want to participate in this study:
Yes
No
Part 1 - Production phase
Part 1 — Production phase
The first part of the experiment is a production phase where you are given three assignments to work on.
Go on to the next page to receive instructions for the first assignment.
Assignment 1 - Sentence unscrambling
Assignment 1:
In the first assignment you are asked to work on a sentence unscrambling task for 5 minutes. Your
performance will not be measured as there is no right or wrong answer, but we do ask you to work
continuously on this assignment.
Description of the assignment:

You will be shown five English words and are asked to form a sentence or an expression by using four of
these words. This means that each sentence or expression must only contain four words.
For example, if the words given to you are "wings, sun, the, has, jet" then you can construct the sentence:
the jet has wings
Write the sentence or expression that you form into the blank space using your keyboard. Your answer will
https://nhh.eu.qualtrics.com/ControlPanel/Ajax.php?action=GetSurveyPrintPreview&... 24.09.2015
be submitted automatically after 20 seconds and you will auto-advance to five new words.
This assignment will last for 5 minutes and we ask you to work continuously. When you have read and
understood the instructions press >> to start the assignment.
These page timer metrics will not be displayed to the recipient.

First Click: 0 seconds
Last Click: 0 seconds
Page Submit: 0 seconds
Click Count: 0 clicks
STRONG WAS LONG THE HIKE

GLASSES ARE PICTURE DIRTY HIS

BASICS SHE SEDUCED THE INTRODUCED

FINALLY IS ALWAYS HERE FRIDAY
You have now completed the first out of three assignments.
On the next page you will receive instructions for the second assignment.
Assignment 2 - Black cells
Assignment 2:
In the second assignment you are asked to work on a counting task for 5 minutes.
You are asked to count the number of black cells in a matrix similar to the one below (but bigger). You will
auto-advance from each matrix after 60 seconds. On the following page you will then get to submit the
number of black cells you counted, before moving on to another matrix. Your performance will be measured.
The aim is to get the closest to the true number of black cells in each of the matrices. In the example matrix
below, the correct number of black cells is five.
After this 5-minute assignment you will advance to the last assignment.
When you have read and understood the instructions press >> to start the assignment.

#QuestionText, TimingPageSubmit#: 0 seconds
Press >> to start go on to the next page.
You have now completed the second assignment.
On the next page you will receive instructions for the third and final assignment.
Assignment 3 - Code recognition
Assignment 3
In the third assignment you are asked to work on a code recognition task for 5 minutes. For this assignment
we will measure your performance by the number of points you receive. You will be informed about your
score on assignment 3 at the end of the assignment.

On top of the page you will be shown a 3-digit code that you must find and check off from a matrix of 3-digit
codes in random order. The assigned code will occur multiple times in the same matrix and you will be given
1 point for each correct marking. You will be subtracted 1 point if you check off a wrong code, but you will not
lose any points for failing to check off all occurrences of the correct code.
Your matrix will be submitted automatically after 60 seconds and you will auto-advance to the next page.
This assignment will last for 5 minutes and after 5 minutes you will be taken to the last part of the survey.
Below you are shown a simplified example to make sure you understand the assignment. When you have
read and understood the instructions press >> to start the assignment.
This is an example:
The code you must check off is: 123

123 283
231 123
952 641
864 820
123 462
791 123

#QuestionText, TimingPageSubmit#: 0 seconds
#QuestionText, TimingClickCount#: 0 clicks
180 847 843 738 302 301 191 379 881 632 821 534 525 191 584 302 276
808 918 678 555 656 559 281 720 390 834 557 116 234 229 157 302 616
787 478 856 412 832 848 305 406 746 288 843 997 926 302 427 302 302
616 500 549 345 587 844 185 409 591 302 964 353 350 709 916 398 998
908 227 302 219 424 372 738 800 356 326 408 782 273 898 730 628 116
825 675 260 661 585 109 690 677 760 419 536 282 364 114 516 731 213
574 810 394 113 578 909 302 509 588 907 197 302 256 160 607 302 971
843 578 711 718 595 869 562 652 980 387 332 745 664 236 308 259 341
170 154 302 839 438 597 102 150 336 118 434 378 941 750 309 244 626
705 729 161 158 749 302 871 706 220 964 280 460 848 225 302 692 437
959 293 206 434 302 319 321 655 428 390 269 302 735 671 738 326 331
401 302 880 352 450 919 547 673 302 254 158 614 302 596 519 472 984
675 852 857 180 593 340 869 146 772 182 885 302 786 899 302 376 302
You have now completed the third and final assignment. Your total score on Assignment 3 is
${gr://SC_af1QZozPqCuXWND/Score}.
Press >> to continue to the next part of the experiment.
Part 2 - Determination payments - productivity
Part 2 – Determination of payments
You have now completed your work on all three assignments. We will now explain how you will be paid for
this work. After you have completed this HIT, we will for each assignment match you with another participant
who has completed the same assignment. The payments to you and the other participant is determined by a
two-stage process. Below we explain this process in more detail.
First stage:
Assignment 1: For this assignment, your earnings are determined by a lottery where each of you with equal
probability earns 6 USD or 0 USD.
Assignment 2: For this assignment, your earnings are determined by how productive you are. Your
productivity in assignment 2 is measured by how close you got to the true number of black cells in each of
the matrices. The participant who got closest to the true number of black cells in each of the matrices earns 6
USD and the other participant earns 0 USD. If you both have the same productivity, you will be matched with
another participant.
Assignment 3: For this assignment, your earnings are again determined by how productive you are. Your
productivity in assignment 3 is measured by your score on assignment 3. The participant with the
highest score earns 6 USD and the other participant earns 0 USD. If you both have the same score, you will
be matched with another participant.
Second stage:
For each assignment, a randomly selected third person will be given the opportunity to redistribute the
earnings between you and the other participant. This person will not know the identity of you or the other
participant, but will be informed about the nature of the assignment and your earnings for this assignment.
For each assignment, either you or the other participant earns 6 USD and the other participant earns 0 USD.
If the third person chooses not to redistribute, each of you will be paid your earnings from the assignment. If
the third person chooses to redistribute earnings for any of the assignments, increasing the payment of the
participant with the low earnings by 1 USD decreases the other participant’s payment by 1 USD.
You will receive your payments for the three assignments within three weeks and it will be paid separately
from your fixed participation fee of 2 USD.
Please click >> to continue.
Part 2 - Determination payment - lottery
You have now completed your work on all three assignments. We will now explain how you will be paid for
this work. After you have completed this HIT, we will for each assignment match you with another participant
who has completed the same assignment. The payments to you and the other participant is determined by a
two-stage process. Below we explain this process in more detail.
First stage:
Assignment 1: For this assignment, your earnings are determined by a lottery where each of you with equal
probability earns 6 USD or 0 USD.
Assignment 2: For this assignment, your earnings are determined by how productive you are. Your
productivity in assignment 2 is measured by how close you got to the true number of black cells in each of
the matrices. The participant who got closest to the true number of black cells in each of the matrices earns 6
USD and the other participant earns 0 USD. If you both have the same productivity, you will be matched with
another participant.
Assignment 3: For this assignment, your earnings are determined in the same way as for assignment 1.
Second stage:
For each assignment, a randomly selected third person will be given the opportunity to redistribute the
earnings between you and the other participant. This person will not know the identity of you or the other
participant, but will be informed about the nature of the assignment and your earnings for this assignment.
For each assignment, either you or the other participant earns 6 USD and the other participant earns 0 USD.
If the third person chooses not to redistribute, each of you will be paid your earnings from the assignment. If
the third person chooses to redistribute earnings for any of the assignments, increasing the payment of the
You will receive your payments for the three assignments within three weeks and it will be paid separately
from your fixed participation fee of 2 USD.
Comment
Finally, if you have any comments or suggestions related to this experiment please write them down in the
blank space below. Your feedback is very important to improve our research.
72
B.5.2 Round 2
4.2.2016 Qualtrics Survey Software
Background questions
What is your current age?

Less than 16
1619
2024
2534
3544
4554
5564
65 or over
What is your gender?

Male
Female
Introduction
Please read the instructions below carefully
General instructions:
The results from this experiment will be used in a research project. It is therefore important that you carefully
read and follow all instructions. Note that you will remain anonymous throughout the experiment. We will only
use your Worker ID to assign payments and check that you have not participated in this experiment before.
You will be paid a fixed participation fee of 2 USD and you may, depending on the actions you and others take,
earn additional money.
You will be given detailed instructions on your screen before each part of the experiment. Please read the
instructions to each part carefully.
If you have any questions regarding this experiment, you may contact thechoicelab@nhh.no
I have read and understood the above and want to participate in this study:
Yes
No
Part 1 Production phase
Part 1 — Production phase
The first part of the experiment is a production phase where you are given three assignments to work on.
Go on to the next page to receive instructions for the first assignment.
https://nhh.eu.qualtrics.com/ControlPanel/Ajax.php?action=GetSurveyPrintPreview 1/18
Assignment 1 Black cells runde 2
Assignment 1:
In the first assignment you are asked to work on a counting task for 5 minutes.
You are asked to count the number of black cells in a matrix similar to the one below (but bigger). You will auto
advance from each matrix after 60 seconds. Please remember to submit the number of black cells you counted
within the 60 seconds. Your performance will be measured. The aim is to get the closest to the true number of
black cells in each of the matrices. In the example matrix below, the correct number of black cells is five.
After this 5minute assignment you will advance to the second assignment.

01 00
How many black cells are there in matrix 1?
You have now completed the first assignment.
On the next page you will receive instructions for the second assignment.
Assignment 2 Code recognition
Assignment 2
In the second assignment you are asked to work on a code recognition task for 5 minutes. For this assignment
we will measure your performance by the number of points you receive. You will be informed about your score
on assignment 3 at the end of the assignment.

On top of the page you will be shown a 3digit code that you must find and check off from a matrix of 3digit
codes in random order. The assigned code will occur multiple times in the same matrix and you will be given 1
point for each correct marking. You will be subtracted 1 point if you check off a wrong code, but you will not lose
any points for failing to check off all occurrences of the correct code.
Your answer will be submitted automatically after 60 seconds and you will autoadvance to the next page. This
assignment will last for 5 minutes and after 5 minutes you will be taken to the last assignment of the survey.
Below you are shown a simplified example to make sure you understand the assignment. When you have read
and understood the instructions press >> to start the assignment.
This is an example:

123 283
231 123
952 641
864 820
123 462
791 123

01 00
407 559 917 522 459 293 743 241 778 241 303 234 951 807 637 454 583
743 538 330 265 816 661 998 678 269 241 578 241 308 233 464 749 495
602 241 602 121 241 314 241 850 144 518 241 494 354 247 258 957 777
537 914 241 340 241 410 274 674 721 711 971 290 606 265 783 775 674
144 942 723 922 241 873 337 474 630 241 574 615 695 388 241 174 926
435 146 618 219 980 674 391 749 795 380 340 859 882 210 912 703 707
265 241 943 723 843 241 924 218 241 607 876 757 160 427 925 234 255
689 795 416 622 233 508 648 602 223 589 701 393 372 942 124 241 377
617 705 572 891 524 634 456 975 874 241 966 729 730 216 900 241 241
809 763 874 180 241 187 241 891 603 881 405 241 389 510 130 268 739
350 241 806 833 585 205 623 567 241 341 843 560 546 810 796 180 842
948 303 274 173 361 273 241 533 446 590 280 759 334 205 307 654 447
408 221 818 938 997 241 216 554 566 300 495 472 360 641 543 431 549
764 365 241 926 542 395 355 674 241 197 191 653 527 172 140 884 225
220 882 979 108 932 919 883 354 358 744 545 809 241 661 968 317 355
881 347 609 537 241 809 879 334 540 213 121 555 596 527 241 702 906
149 375 858 801 550 241 965 628 388 163 477 989 553 840 494 809 605
384 384 862 384 775 655 644 846 367 719 411 148 773 998 258 685 778
269 458 449 384 491 298 323 384 446 270 133 384 199 108 800 114 230

01 00
210 454 384 833 302 226 508 328 302 842 302 427 930 790 464 932 302
898 592 900 871 409 302 428 379 708 586 201 428 160 301 710 145 550
583 669 199 465 443 252 474 547 473 945 904 337 501 269 300 847 498
825 638 334 863 302 299 302 469 426 903 566 189 244 333 208 302 202
180 847 843 738 302 301 191 379 881 632 821 534 525 191 584 302 276
808 918 678 555 656 559 281 720 390 834 557 116 234 229 157 302 616
787 478 856 412 832 848 305 406 746 288 843 997 926 302 427 302 302
616 500 549 345 587 844 185 409 591 302 964 353 350 709 916 398 998
908 227 302 219 424 372 738 800 356 326 408 782 273 898 730 628 116
825 675 260 661 585 109 690 677 760 419 536 282 364 114 516 731 213
574 810 394 113 578 909 302 509 588 907 197 302 256 160 607 302 971
843 578 711 718 595 869 562 652 980 387 332 745 664 236 308 259 341
170 154 302 839 438 597 102 150 336 118 434 378 941 750 309 244 626
705 729 161 158 749 302 871 706 220 964 280 460 848 225 302 692 437
959 293 206 434 302 319 321 655 428 390 269 302 735 671 738 326 331
401 302 880 352 450 919 547 673 302 254 158 614 302 596 519 472 984
675 852 857 180 593 340 869 146 772 182 885 302 786 899 302 376 302
You have now completed the second assignment. Your total score on Assignment 2 is
${gr://SC_af1QZozPqCuXWND/Score}.
Press >> to continue to the third and final assignment.
Assignment 3 Black cells
Assignment 3:
In the third and final assignment you are asked to once again work on a counting task for 5 minutes.
In the same way as in assignment 1, you are asked to count the number of black cells in a matrix similar to the
one below (but bigger). You will autoadvance from each matrix after 60 seconds. Please remember to submit
the number of black cells you counted within the 60 seconds. Your performance will be measured. The aim is to
get the closest to the true number of black cells in each of the matrices. In the example matrix below, the correct
number of black cells is five.
After this 5minute assignment you have completed the final assignment and will advance to the last part of the
experiment.

01 00
How many black cells are there in matrix 1?
You have now completed the third and final assignment.
Please press next >> to advance to the last part of the experiment.
Part 2 Determination payments productivity
You have now completed your work on all three assignments. We will now explain how you will be paid for this
work. After you have completed this HIT, we will for each assignment match you with another participant who
has completed the same assignment. The payments to you and the other participant is determined by a two
stage process. Below we explain this process in more detail.
First stage:
Assignment 1: For this assignment, your earnings are determined by how productive you are. Your productivity
in assignment 1 is measured by how close you got to the true number of black cells in each of the matrices. The
participant who got closest to the true number of black cells in each of the matrices earns 6 USD and the other
participant earns 0 USD. If you both have the same productivity, you will be matched with another participant.
Assignment 2: For this assignment, your earnings are again determined by how productive you are. Your
productivity in assignment 2 is measured by your score on assignment 2. The participant with the
highest score earns 6 USD and the other participant earns 0 USD. If you both have the same score, you will be
matched with another participant.
Assignment 3: For this assignment, your earnings are determined by how productive you are in assignment 3 in
the same way as for assignment 1.
Second stage:
For each assignment, a randomly selected third person will be given the opportunity to redistribute the earnings
between you and the other participant. This person will not know the identity of you or the other participant, but
will be informed about the nature of the assignment and your earnings for this assignment.
For each assignment, either you or the other participant earns 6 USD and the other participant earns 0 USD. If
the third person chooses not to redistribute, each of you will be paid your earnings from the assignment. If the
third person chooses to redistribute earnings for any of the assignments, increasing the payment of the
You will receive your payments for the three assignments within three weeks and it will be paid separately from
your fixed participation fee of 2 USD.
Comment
Finally, if you have any comments or suggestions related to this experiment please write them down in the blank
space below. Your feedback is very important to improve our research.
Issued in the series Discussion Papers 2018
2018
01/18 January, Øystein Foros, Mai Nguyen-Ones, and Frode Steen, “Evidence on
consumer behavior and firm performance in gasoline retailing”
02/18 January, Agnar Sandmo, “A fundamental externality in the Labour Market?

Ragnar Frisch on the socially optimal amount of work”
03/18 February, Pierre Dubois and Morten Sæthre, “On the Effect of Parallel
Trade on Manufacturers’ and Retailers’ Profits in the Pharmaceutical
Sector”
04/18 March, Aline Bütikofer, Julie Riise, and Meghan Skira, “The Impact of Paid
Maternity Leave on Maternal Health”
05/18 March, Kjetil Bjorvatn and Bertil Tungodden, “Empowering the disabled
through savings groups: Experimental evidence from Uganda”
06/18 April, Mai Nguyen-Ones and Frode Steen, “Measuring Market Power in
Gasoline Retailing: A Market- or Station Phenomenon?
07/18 April, Chang Koo Chi and Trond Olsen, “Relational incentive contracts and
performance”
08/18 April, Björn Bartling, Alexander W. Cappelen, Mathias Ekström, Erik Ø.

Sørensen, and Bertil Tungodden, «Fairness in Winner-Take-All Markets»
09/18 April, Aline Bütikofer, Sissel Jensen, and Kjell G. Salvanes, «The Role of
Parenthood on the Gender Gap among Top Earners»
10/18 May, Mathias Ekström, “The (un)compromise effect”
11/18 May, Yilong Xu, Xiaogeng Xu, and Steven Tucker, «Ambiguity Attitudes in
the Loss Domain: Decisions for Self versus Others»
12/18 June, Øivind A. Nilsen, Per Marius Pettersen, and Joakim Bratlie, “Time-
Dependency in producers’ price adjustments: Evidence from micro panel
data”
13/18 June, Øivind A. Nilsen, Arvid Raknerud, and Diana-Cristina Iancu, “Public
R&D support and firms’ performance. A Panel Data Study”
14/18 June, Manudeep Bhuller, Gordon B. Dahl, Katrine V. Løken, and Magne
Mogstad: «Incarceration, Recidivism, and Employment”
15/18 August, Manudeep Bhuller, Gordon B. Dahl, Katrine V. Løken, and Magne
Mogstad: «Incarceration Spillovers in Criminal and Family Networks”
16/18 August, Pedro Carneiro, Kai Liu, and Kjell G. Salvanes: “The Supply of
Skill and Endogenous Technical Change: Evidence From a College
Expansion Reform”
17/18 August, Chang Koo Chi, “An analysis of the two-bidder all-pay auction
with common values”
18/18 August, Alexander W. Cappelen, Cornelius Cappelen, and Bertil

Tungodden, “Second-best fairness under limited information: The trade-off
between false positives and false negatives”
19/18 September, Aline Bütikofer, Antonio Dalla Zuanna, and Kjell G.

Salvanes: “Breaking the Links: Natural Resource Booms and
Intergenerational Mobility”
20/18 September, Juan Pablo Atal, José Ignacio Cuesta, and Morten Sæthre,
“Quality regulation and competition: Evidence from Pharmaceutical
Markets”
21/18 October, Orazio Attanasio, Agnes Kovacs, and Krisztina Molnar, “Euler
Equations, Subjective Expectations and Income Shocks”
22/18 October, Antonio Mele, Krisztina Molnár, and Sergio Santoro, “On the
perils of stabilizing prices when agents are learning”
23/18 November, Bjørn-Atle Reme, Helene Lie Røhr, and Morten Sæthre, “The
Poking Effect: Price Changes, Information, and Inertia in the Market for
Mobile Subscriptions”
24/18 November, Ingrid Hoem Sjursen, “Accountability and taxation:

Experimental evidence”
25/18 November, Liam Brunt and Antonio Fidalgo, “Why 1990 international
Geary-Khamis dollars cannot be a foundation for reliable long run
comparisons of GDP”
26/18 November, Ola Honningdal Grytten, “A continuous consumer price index

for Norway 1492-2017”
27/18 December, Liam Brunt and Antonio Fidalgo, “Feeding the people: grain
yields and agricultural expansion in Qing China”
28/18 December, Kurt R. Brekke, Chiara Canta, Luigi Siciliani and Odd Rune
Straume, “Hospital Competition in the National Health Service: Evidence
from a Patient Choice Reform”
29/18 December, Richard Friberg, Frode Steen and Simen A. Ulsaker, “Hump-
shaped cross-price effects and the extensive margin in cross-border
shopping”
30/18 December, David Jaume and Alexander Willén, “Oh Mother: The
Neglected Impact of School Disruptions”
31/18 December, Jesús Crespo Cuaresma, Gernot Doppelhofer, Martin

Feldkircher and Florian Huber, “Spillovers from US monetary policy:
Evidence from a time-varying parameter GVAR model”
32/18 January, Patrick Bennet and Amine Ouazad, “ Job Displacement,

Unemployment, and Crime: Evidence from Danish Microdata and
Reforms”
2019
01/19 January, Aline Bütikofer, Christopher Cronin, Meghan Skira,

“Employment Effects of Healthcare Policy: Evidence from the 2007 FDA
Black Box Warning on Antidepressants”
02/19 February, Ingar Haaland and Cristopher Roth “Beliefs About Racial
Discrimination and Support for Pro-Black Policies “
03/19 February, Astrid Kunze and Xingfei Liu, “Universal Childcare for the
Youngest and the Maternal Labour Supply”
04/19 February, Ingvild Almas, Alexander W. Cappelen, Bertil Tungodden.

“Cutthroat capitalism versus cuddly socialism: Are Americans more
meritocratic and efficiency-seeking than Scandinavians?”
05/19 Chang Koo Chi, Kyoung Jin Choi. “Performance Measurement in Agency
Models”
06/19 Alexander W. Cappelen, Ranveig Falch and Bertil Tungodden. “The Boy
Crisis: Experimental Evidence on the Acceptance of Males Falling Behind”
NORGES HANDELSHØYSKOLE
Norwegian School of Economics
Helleveien 30
NO-5045 Bergen
Norway
T +47 55 95 90 00
E nhh.postmottak@nhh.no
W www.nhh.no

The Boy Crisis

Uploaded by

Copyright:

Available Formats

The Boy Crisis

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

The Boy Crisis

Uploaded by

Copyright:

Available Formats

The Boy Crisis: Experimental Evidence on the

Acceptance of Males Falling Behind

BY Alexander W. Cappelen, Ranveig Falch and

Institutt for samfunnsøkonomi

JEL: C91, D63, J16

[Figure 1 about here]

2 Experimental design and participants

[Table 1 about here]

2.1 The workers

2.2 The spectators

[Table 2 about here]

2.3 General structure

Our results are robust to the removal of these participants.

[Table 3 about here]

V (y; ·) = −(y − h(E(e)))2 − Iβ (g)y, (1)

nomics literature. The first approach, developed by Becker (1957), introduces

Observation 1: Statistical fairness discrimination would in mixed-sex merit

In the mixed-sex luck treatments, there is no signal of productivity that can

Observation 2: Statistical fairness discrimination would not affect spec-

Observation 3: Statistical fairness discrimination and taste-based discrim-

Observation 4: Statistical fairness discrimination would imply that we ob-

4.1 Main analysis

ui = α + β Maleloseri + γXi + εi , (3)

where ui is the standardized amount transferred by spectator i to the losing

ui = α + β1 Maleloseri + β2 Mi + β3 Maleloser × Mi + εi , (4)

where ui is the standardized amount transferred by spectator i to the losing

ui = α + β1 Maleloseri + β2 Lucki + β3 Maleloser × Lucki + γXi + εi . (5)

ui = α + β1 MS − Maleloseri + β2 MS − Femaleloseri + γXi + εi , (6)

where MS − Maleloseri and MS − Femaleloseri are indicator variables for the

4.3 Multiple hypothesis testing

5.1 The distributive decisions

[Figure 2 about here]

5.2 Main findings

[Table 4 about here]

We observe a significant gender bias in the spectator choices: the average

With respect to the background variables, we observe in Table 4 that Repub-

[Table 5 about here]

Result 2: We find a significant gender bias against male losers in almost

The finding of a gender bias against males in the merit-sex treatments is in

[Table 6 about here]

We do not find evidence of a gender bias in the spectator behavior in the

Result 3: We find no robust evidence of a gender bias against male losers

In Table 7, we compare the spectator behavior in the mixed-sex merit treat-

Result 4: We find no evidence of spectators choosing differently in single-

Our study provides new evidence on the nature of gender discrimination, by

ital outcomes? Evidence from a natural experiment,” Journal of Public Eco-

Figure 1: Main findings

Female loser Male loser Male spectator Female spectator

Panel A: Share transferred to loser Panel B: Difference by spectator gender

Figure 2: Transfer to loser

1. Work stage: Workers complete an assignment.

Table 2: Descriptive statistics

Republican (share) 0.337 0.270

Table 4: Regression analysis: Transfer to loser

Table 5: Heterogeneity analysis: Amount to loser (std)

Table 6: Regression analysis: Transfer to loser

Table 7: Regression analysis: Transfer to loser

Mixed-sex merit vs. mixed-sex luck

Table 8: Regression analysis: Transfer to loser

Table 9: Single-sex vs. mixed-sex merit: Amount to loser (std)