econstor
Make Your Publications Visible.

A Service of

zbw
Leibniz-Informationszentrum
Wirtschaft
Leibniz Information Centre
for Economics

Darai, Donja; Grätz, Silvia

Working Paper

Facing a dilemma: Cooperative behavior and beauty

Working Paper, No. 82

Provided in Cooperation with:
Department of Economics, University of Zurich

Suggested Citation: Darai, Donja; Grätz, Silvia (2012) : Facing a dilemma: Cooperative behavior
and beauty, Working Paper, No. 82, University of Zurich, Department of Economics, Zurich,
http://dx.doi.org/10.5167/uzh-63130

This Version is available at:
http://hdl.handle.net/10419/77493

Standard-Nutzungsbedingungen:

Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen
Zwecken und zum Privatgebrauch gespeichert und kopiert werden.

Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle
Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich
machen, vertreiben oder anderweitig nutzen.

Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen
(insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten,
gelten abweichend von diesen Nutzungsbedingungen die in der dort
genannten Lizenz gewährten Nutzungsrechte.

Terms of use:

Documents in EconStor may be saved and copied for your
personal and scholarly purposes.

You are not to copy documents for public or commercial
purposes, to exhibit the documents publicly, to make them
publicly available on the internet, or to distribute or otherwise
use the documents in public.

If the documents have been made available under an Open
Content Licence (especially Creative Commons Licences), you
may exercise further usage rights as specified in the indicated
licence.

www.econstor.eu



 

 

 
 

 
Working Paper No. 82 

 
 

Facing a Dilemma:  
Cooperative Behavior and Beauty  

 
 

 
 

Donja Darai and Silvia Grätz 
 
 
 

Revised version, December 2012 
 

 

 

 

 
 
 

University of Zurich 
 

Department of Economics 
 

 
 

Working Paper Series 
  

ISSN 1664-7041 (print) 
 ISSN 1664-705X (online) 

 
 

 
 

 
 

 

 



Facing a Dilemma: Cooperative Behavior and Beauty∗

Donja Darai and Silvia Grätz†

December 7, 2012

Abstract

This paper investigates the influence of physical attractiveness on cooperative be-
havior. We survey data from 211 episodes of a television game show and combine
it with independent facial attractiveness ratings of the show’s contestants. The fi-
nal of the show represents a simultaneous one-shot modified prisoner’s dilemma.
Our results show that facially attractive contestants provoke cooperative behavior
from their counterparts, but attractive contestants themselves do not behave dif-
ferently from unattractive ones. This preferential treatment or “beauty premium”
rewards attractive contestants with substantially higher monetary gains of up to
£2 153. This finding applies to both sexes. But, the attractiveness effect is limited
to mixed-gender interactions. Contestants are only more cooperative when facing
an attractive fellow player of the opposite sex. Attractiveness also has no effect in
group-decisions made by the contestants prior to the prisoner’s dilemma. Therefore
people seem to show a preference to cooperate with someone towards whom they
are personally attracted; and this preference reaches full effect when people lack
other information on which to base their decision.

Keywords: beauty premium, cooperation, prisoner’s dilemma

JEL classification: C71, D83, Z13

∗The authors thank Ulrich Kaiser, Armin Schmutzler, Michelle Sovinsky, Roberto Weber, the seminar
participants at the University of Zurich and Florida State University, the participants at the ESA Inter-
national Meeting 2012 in New York (U.S.), the Workshop on Behavioral Economics Mainz (Germany),
the Zurich Workshop on Economics 2012 in Luzern (Switzerland), and the ESA European Meeting 2012
in Cologne (Germany) for helpful discussions and suggestions, as well as Pascal Kappeler, Christina
Richard, and Holger Scriba for excellent research assistance. Financial support of the Swiss National
Science Foundation is gratefully acknowledged. The data on “Golden Balls” were provided to the authors
by the television show producers, courtesy of Endemol UK plc, in May 2009.
†Both authors, Department of Economics, University of Zurich, donja.darai@econ.uzh.ch,

silvia.graetz@econ.uzh.ch

1



“All that glitters is not gold;

Often have you heard that told.”

William Shakespeare, Merchant of Venice (1596-1598)

1 Introduction

Beginning with the seminal paper of Biddle and Hamermesh (1998), which identifies a

wage gap based on physical attractiveness using labor market data of the U.S. and Canada,

economists started to be interested in the effects of physical attractiveness on economic de-

cision making. For instance, Mobius and Rosenblat (2006) show that the beauty premium

for attractive people is even present in a controlled laboratory labor market experiment.

Eckel and Petrie (2011) provide evidence that people have a willingness to pay for being

able to see a picture of their counterpart before making their decision in a trust game,

suggesting that valuable information is conveyed by the physical appearance of people.

However, the transmission channel of the effect of physical attractiveness on economic

decision making is still only partly understood. The most prominent theory is “beauty-is-

good” stereotyping. People assign a wide range of positive personality traits and abilities

to physically attractive people. That is, people believe that physically attractive people

are, for instance, more trustworthy, more likable, and more productive (see e.g., Dion,

Berscheid, and Walster (1972)). These beliefs then alter people’s behavior towards at-

tractive people. However, not all that glitters is gold. There is only very limited evidence

that the “beauty-is-good” stereotype is accurate.

This paper studies the relationship between attractiveness and cooperative behavior and

shows that the beauty premium is also present in a high-stakes field setting with two-sided

communication and previous interaction between players. Our results offer an alternative

explanation of the effects of attractiveness on economic decision making. We suggest a

preference based transmission channel in the sense that people’s behavior is biased when

facing someone attractive of the opposite sex. We confirm our hypothesis with existing

experimental data from Mobius and Rosenblat (2006).

Specifically, this paper combines two independent data sets. First, to study cooperation,

we survey data from 211 episodes of the British television game show “Golden Balls”, in

which the show’s finalists play for a jackpot at the end of the show. The rules of this game

follow a slightly modified version of a simultaneous one-shot prisoner’s dilemma: if both

contestants choose to cooperate, the jackpot is split equally; if one chooses to defect while

the other cooperates, the former receives the entire jackpot and the latter gets nothing;

if both defect, they both go home empty-handed. The accumulation of the jackpot takes

place in two rounds of pre-play previous to the prisoner’s dilemma and requires neither

the contestants’ effort nor cognitive ability. On average, the stake size is £12 912 indicat-

1



ing that decisions in the prisoner’s dilemma have substantial distributional consequences

for the contestants. During the two rounds of pre-play, the two finalists are selected out

of four initial contestants. Second, we collect data on the contestants’ facial appearance

from a sample of 365 independent raters. These raters are asked to evaluate portrait

photographs of the contestants along various dimensions such as physical attractiveness,

sympathy, trustworthiness, or likability.

We provide evidence that physical attractiveness affects cooperative behvaior. Contestants

of the show are significantly more cooperative towards a facially attractive opponent than

towards a facially unattractive opponent, independent of gender, age, other demographic

characteristics, stake size, and communication. Facially attractive contestants are there-

fore awarded with significantly higher earnings in the prisoner’s dilemma. Quantifying

this beauty premium, a one-standard deviation increase in facial attractiveness, at the

margin, causes the contestant’s expected earnings to rise by up to £2 153, as well as the

contestant’s probability to obtain positive earnings by up to 5.9 percentage points. This

effect is not driven by non-cooperative behavior of the attractive. With minor qualifica-

tions for younger and female contestants, we find no correlation between a contestant’s

own facial attractiveness and cooperation.

Although the attractiveness effect likewise applies for men and women, we show that it is

limited to mixed-gender interactions only. Contestants are only biased in their decision to

cooperate when facing an attractive opponent of the other sex. Thus, the attractiveness

effect is not present in same-gender interactions. This finding offers a new perspective

on the underlying mechanism of the effects caused by attractiveness. Our results are not

driven by people believing that attractive others are more likely to cooperate, but rather

by people having a preference to cooperate more with someone towards whom they are

personally attracted to. Furthermore, we cannot identify a beauty premium or plainness

penalty during the pre-play, which clearly supports our conjecture that personal attrac-

tion serves as the underlying transmission channel of physical attractiveness on people’s

behavior in pairwise interactions and that attractiveness matters most when objective

information is missing.

To validate our results, we repeat the analysis of Mobius and Rosenblat (2006) and un-

cover that their identified effects of attractiveness disappear when we parse their data

by the composition of sex, that is, when we consider interactions between employers and

workers of the same sex separately from those of the opposite sex. Again, “beauty-is-good”

stereotyping fails to consistently explain the findings. Therefore, we offer an alternative

mechanism underlying the attractiveness effects, namely a “preference-driven” mechanism.

Physical attractiveness mediates people’s behavior, because people want to behave differ-

ently towards someone whom they feel personally attracted to.

The remainder of this paper is structured as follows. In Section 2 we discuss related

2



literature. Section 3 describes the data sets and Section 4 presents the results. Potential

transmission channels are evaluated in Section 5. Finally, Section 6 concludes.

2 Literature

A long history of research on physical attractiveness in psychology, sociology, and evo-

lutionary biology and a more recent one in economics and political science shows that

physically attractive people are or behave not substantially differently from physically

unattractive people, but that physically attractive people receive a preferential treatment

by others in many regards. This “beauty premium” rewards attractive people with sub-

stantially higher monetary gains or greater economic success. Surprisingly, the mechanism

underlying the effect of physical attractiveness is still not completely uncovered.

The most prominent theory, established in sociology and psychology, is that people link

beauty and positive personality traits. In particular, people attribute a variety of positive

characteristics and higher abilities to physically attractive people, and negative ones to

physically unattractive people (see reviews by Eagly, Ashmore, Makhijani, and Longo

(1991), Feingold (1992), and Langlois, Klakanis, Rubenstein, Larson, Hallam, and Smoot

(2000)). This is referred to as “beauty-is-good” stereotyping in the literature.1 The theory

argues that stereotype beliefs cause people to treat attractive ones more favorably, and

this in turn results in higher economic gains for the attractive. Hence, beauty-is-good

stereotyping can explain why attractive people generally fare better in the labor market,

i.e., why they are more likely to be hired, promoted, and earn higher salaries. In their sem-

inal work Biddle and Hamermesh (1998) use a broad household survey of the U.S. and

Canada and show that physically attractive employees earn about 10-15 percent more

than less attractive employees, independent of their occupation. In an experiment on the

effects of beauty in experimental labor markets, Mobius and Rosenblat (2006) find that

attractive employees are awarded higher wages, although they are not more productive

than unattractive employees, measured by a real-effort task. This suggests that employers

hold (inaccurate) stereotypical expectations about the performance of physically attrac-

tive employees. Ruffle and Shtudiner (2010) explore the value of beauty in the hiring

process, examining response rates to CVs sent to companies in Israel. They uncover a

gender-specific effect of attractiveness, namely a beauty premium for males, and a beauty

penalty for females.

1In their pioneering study, Dion, Berscheid, and Walster (1972) claim “what is beautiful is good” by
demonstrating that attractive people are believed to have better career prospects, to possess socially
desirable traits, to lead happier lives and to be happier overall. This paper spawned a large literature on
the physical attractiveness stereotype, demonstrating for both sexes a robust association between physical
attractiveness and cognitive ability, competence, sociability, popularity, dominance, sexual experience,
mental health, and social skills.

3



Also, research in political science shows that physical attractiveness has a significant im-

pact on the evaluation of candidates and thereby on electoral outcomes (see the survey by

Ottati and Deiger (2002)). For instance, attractiveness increases the candidate’s number

of votes and thereby the likelihood of being elected (Berggren, Jordahl, and Poutvaara

(2010) and Rosar, Klein, and Beckers (2008)). Antonakis and Dalgas (2009) suggest that

underlying the effect of physical attractiveness is the voters’ belief that attractive can-

didates are more competent. Recruiting adults and children raters in Switzerland, they

find that even children can predict results of the 2002 French parliamentary election ret-

rospectively by rating the competence of the candidates from their photographs.

Furthermore, the theory of beauty-is-good stereotyping in combination with interdepen-

dent social preferences can explain findings in various experimental settings with strategic

interactions. These findings suggest that people behave more generously or cooperatively

towards physically attractive people. Solnick and Schweitzer (1999) show that attractive

responders receive significantly higher offers than unattractive responders in the ultima-

tum game. In dictator games, Rosenblat (2008) shows that allocators treat physically

and vocally attractive recipients more generously. Andreoni and Petrie (2008) observe a

beauty premium in a repeated public goods game, which is not caused by lower contri-

butions of attractive players, but can be attributed to an increase of the other players’

contributions triggered by an attractive group member. In a repeated trust game, Wilson

and Eckel (2006) demonstrate that attractive trustees are trusted more and that players

expect attractive players to be more trustworthy than unattractive ones. The failure to

meet these expectations leads to a beauty penalty.2 Mulford, Orbell, Shatto, and Stockard

(1998) examine a repeated prisoner’s dilemma, pointing out that players are more coop-

erative if matched with an attractive partner. In contrast to the other studies, they use

attractiveness ratings based on the perception of the player who also makes the decision

in the experiment, rather than ratings elicited from independent third-party judges.

However, all above mentioned studies provide only very limited evidence for the accuracy

of the beauty-is-good stereotype. Physically attractive people themselves do not behave

differently than physically unattractive people.3

Besides the theory of stereotyping based upon peoples’ visual attractiveness, Zuckerman

and Driver (1989) and Zuckerman, Hodgins, and Miyake (1990) proclaim the existence

2In another experimental study on the trust game, Eckel and Petrie (2011) investigate the informa-
tional value of a photograph and the differential desire to acquire this information. Subjects are willing to
pay to see the photograph of their partner whom they transact with, indicating that a face has a positive
informational value which is used to discriminate between players in their choices.

3Jackson, Hunter, and Hodge (1995) show in their meta-analysis that there is a modest correlation
between attractiveness and intelligence for children. Mueller and Mazur (1997) use data from a cohort
of military officers and find that recruits with a high ranked facial appearance are also high ranked in
their physical fitness. Concerning social skillfulness and likability, Goldman and Lewis (1977) and Erwin
and Calev (1984) find that physically attractive people indeed possess better social skills and are more
likable.

4



of a vocal attractiveness stereotype by showing that physical attractiveness is positively

correlated with vocal attractiveness. Supporting a vocal-attractiveness mechanism, Mo-

bius and Rosenblat (2006) find that the beauty premium even exists in treatments in

which employers are able to talk to the employee without seeing the employee’s picture.

Rosenblat (2008) provides evidence that physically attractive people only achieve signifi-

cantly better outcomes in dictator games when the dictator sees the recipient’s picture in

addition to hear her pre-recorded voice. This suggests voice to be one of the transmission

channels of the beauty premium, at least in two-sided communication settings.

Finally, one of the oldest and most well-known theories underlying the effects of attrac-

tiveness is the one of taste-based discrimination, proposed by Becker (1957). Physically

attractive people are favored because people enjoy being or working with them more than

with plain looking people. Discrimination is based upon prejudices correlated with peo-

ple’s personal characteristics and is rational in the sense that interactions with such a

person generate a (dis)utility for the discriminator in case of positive (negative) discrimi-

nation. Belot, Bhaskar, and van de Ven (2012) find that attractive contestants of a Dutch

television game show are positively discriminated against unattractive ones in proceeding

to the final stage of the show.4 However this theory fails to explain why attractive people

are also treated more favorably in one-shot interactions.

Summarizing, in line with the literature we expect physically attractive people to obtain a

beauty premium in the one-shot prisoner’s dilemma. Physically attractive people will be

favored by others in the sense that people show greater cooperativeness towards them. But

physically attractive people themselves will not behave differently compared to physically

unattractive people.

3 Data

3.1 The television game show “Golden Balls”

We analyze people’s behavior in a simultaneous one-shot prisoner’s dilemma of 211 episodes

of the British television game show “Golden Balls”.5 The dilemma is played in the final

4In particular attractive contestants are much more likely to reach the final round of the show, even
though they are not performing better or are not more confident than unattractive contestants. Perfor-
mance in this show means being the first to correctly answer trivia questions. Attractive contestants are
believed to be more confident and to be more cooperative. Besides, at the end of the show a prisoner’s
dilemma – like the one we study – is played. Belot, Bhaskar, and van de Ven (2012) also test for effects
of attractiveness on cooperative behavior, however they find no significant effects. We suspect that the
reason why they are not able to identify any attractiveness effects is that their sample of finalists becomes
too homogeneous with regard to attractiveness, due to the selection bias towards attractive contestants
during the pre-play.

5The game show was aired in June 2007 and ended in December 2009. We use records of 211 episodes
of the first four series of the show.

5



round of the show to allocate the jackpot between the two finalists: each player is assigned

two balls, indistinguishable from the outside, but one contains the word “steal” and one

contains the word “split”. Both contestants choose one of their two balls and then open it

simultaneously. If both choose the split-ball (cooperation), the jackpot is split equally; if

one chooses the steal-ball (defection) while the other chooses the split-ball (cooperation),

the former receives the entire jackpot and the latter gets nothing; if both choose the steal-

ball (defection), they both go home empty-handed (see Figure 1). Thus, defection is a

weakly-dominant strategy.

Figure 1: Prisoner’s dilemma game

split (cooperate) steal (defect)

split (cooperate) 1/2 jackpot ,
1/2 jackpot 0 , jackpot

steal (defect) jackpot , 0 0 , 0

Note: Defection is a weakly dominant strategy

Before the prisoner’s dilemma is played, the contestants have to pass two rounds of pre-

play. In these rounds, the finalists are selected and the jackpot is accrued. Accumulating

the jackpot does not require contestant’s cognitive ability or effort.6 On average, the jack-

pot amounts to £12 912 and ranges from a minimum of £3 to a maximum of £93 250. The

two finalists are selected through two voting decisions. After each round, the contestants

need to cast a vote against one of them to leave the show. The contestant who receives

the majority of votes is eliminated. Throughout the show the contestants face each other

and are allowed to freely communicate with each other (see supplementary material for a

detailed description of the game show).

Table A.1 in the appendix presents summary statistics of the outcomes of the prisoner’s

dilemma and of the contestants’ personal characteristics. The unilateral cooperation rate

is 54%, and contestants mutually cooperate (defect) in 32% (25%) of cases. 46% of finalists

are male and the mean age is 37 years.

3.2 Survey on facial appearance

We evaluate the contestants’ facial appearance using a panel of independent raters. The

raters are recruited at the Euro-Airport Basel, at the University of Zurich, and at the

6Each pre-play round, the contestants are randomly assigned a certain cash value. For the detailed
description of the show, see online appendix.

6



University of Zurich for the elderly.7 All 844 contestants are judged by 728 raters and,

from those, 365 raters judge the 422 finalists. Each rater was asked to individually rate

the facial appearance of five randomly assigned contestants, of which two or three were

male (female). On average, a finalist is judged by 4.3 raters.

Table I reports summary statistics for the finalists’ raters and Table A.2 in the appendix

for all contestants’ raters. The mean age of the 365 raters is 41 years and 50% are male.

Table I: Summary statistics of the finalists’ raters

Rater’s Variable Mean Std. Dev. Min. Max. N

Male 0.5 0.5 0 1 365
Age (in years) 40.88 15.54 17 93 361
Age of Male (in years) 41.58 15.60 17 93 183
Age of Female (in years) 40.15 15.50 18 86 178

Age of Female (≥ 40.15 years) 53.51 10.36 41 86 86
Age of Female (< 40.15 years) 27.66 6.38 18 40 92
Age of Male (≥ 41.58 years) 55.21 11.42 42 93 85
Age of Male (< 41.58 years) 29.76 6.28 17 41 98

The survey is questionnaire based. For an illustration of a sample questionnaire see Fig-

ure 2. Each questionnaire contains two portrait photographs of the same contestant and

is divided into three parts. To receive non-biased evaluations of the contestants and to

reduce measurement error, all photographs are selected from the same two sequences of

the game show such that one photograph shows a neutral facial expression with a view

to the camera and the other a neutral facial expression with a view to the side of the

camera.8

In the first part of the questionnaire, the raters are asked to judge the contestant with

respect to her age by fitting the contestant into one of seven age categories, “<20”, “20-30”,

“30-40”, “40-50”, “50-60”, “60-70”, or “>70” years. The second part includes assessments

of the contestant’s appearance using four opposite word pairs, i.e., “attractive - unattrac-

tive”, “likable - unlikable”, “trustworthy - untrustworthy”, and “honest - dishonest”. These

items are rated on a 1-to-7 point Likert scale, where 1 equals very unattractive, 4 com-

prises a neutral position and 7 equals very attractive.9 In the last part of the questionnaire

we asked the raters to give a binary response (yes/no) to two statements: “this person’s

appearance helps him/her in life” and “this person strikes me as calculating”. These state-

ments are included to capture a rater’s overall impression of the respective contestant’s

7The University of Zurich provides lectures for a senior audience, that are mainly attended by retired
people. We only recruited raters who either had a very high proficiency or were native speakers of
German, French, or English and provided the questionnaires in the respective language.

8If possible we chose a neutral facial expression of the contestant, otherwise a positive one was chosen,
but never a negative or disadvantageous one.

9The use of a 7-point Likert scale, which includes a neutral element, allows for sufficient diversification
and is standard in the relevant literature, see e.g., Alreck and Settle (1995), pp.113-114.

7



appearance, i.e., whether a contestants is either attributed a beauty premium or a strate-

gic intention per se.

Figure 2: Sample Questionnaire

3.3 Evaluation of facial appearance ratings

The raters’ evaluations are used to construct a facial appearance measure for each of the

four items: attractiveness, honesty, likability, and trustworthiness. We account for rater

specific variation in the perception of the respective appearance item in the construction

of each measure.10 For each rater j we calculate the average across all contestants rated

by j, x̄j. Then we mean-center the ratings by subtracting the respective rater’s mean

rating, x̄j, from her individual rating of contestant i, xij. These demeaned ratings are

10Raters may differ in the interpretation of the 7-point scale, and may vary in anchoring their average
rating above, below or close to 4. Student t-tests of differences between the means of the facial appearance
measure by the raters’ gender are: t = 2.799, p = 0.005 for attractiveness, t = 4.746, p = 0.000 for
likability, t = 4.913, p = 0.000 for trustworthiness, and t = 4.591, p = 0.000 for honesty. This shows
that female raters tend to give higher ratings than male raters. This phenomenon is also persistent with
respect to the raters’ age. Further, it is known that images of females tend to be rated higher than images
of males, see e.g., Jackson, Hunter, and Hodge (1995) and Wilson and Eckel (2006). Since we control
directly for gender-related effects in the regressions, the attractiveness measure can be used uncorrected
for the gender of the contestants. As a robustness check we construct a measure by making use of the
rater-specific fixed effects, rather than correcting for them. In particular, we attach higher weight to
ratings made by raters that are similar to the contestant with respect to age and gender, and we attach
lower weight to ratings made by raters that are unsimilar to the contestant. All results remain unchanged.

8



now anchored at 0 for each rater and are therefore corrected for rater-specific fixed effects

(see Figure A.1 in the appendix). Finally we take the mean of all mean-centered ratings

of contestant i, resulting in the particular facial-item measure for each contestant.

[Facial-item]i = Ej[xij − x̄j] with

{
i = i-th contestant

j = j-th rater

Additionally, we construct two variables from the binary statements “appearance helps

in life” and “appearance strikes as calculating” by taking the mean rating for each con-

testant.11 In the following we refer to these variables as “statement variables”. Table II

provides summary statistics of all appearance measures.

Table II: Summary statistics facial appearance

Variable Mean Std. Dev. Min. Max. N

Attractiveness (mean-centered, cont.) 0 0.79 -1.8 2.65 422
Honesty (mean-centered, cont.) 0 0.6 -1.8 1.45 422
Likability (mean-centered, cont.) 0 0.66 -1.95 1.65 422
Trustworthiness (mean-centered, cont.) 0 0.6 -1.75 1.65 422

Appearance Helps In Life (cont.) 0.58 0.30 0 1 422
Appearance Strikes As Calculating (cont.) 0.35 0.25 0 1 422

All four facial appearance variables are highly positively correlated (see Table III). This

is also reflected in a sufficiently high Cronbach coefficient alpha (α = 0.82).12

Table III: Correlation matrix for facial appearance variables, N=422

Attractiveness Honesty Likability Trustworthiness Helps in Life

Attractivenessa 1
Honesty 0.286∗∗∗ 1

Likability 0.461∗∗∗ 0.651∗∗∗ 1
Trustworthiness 0.340∗∗∗ 0.760∗∗∗ 0.680∗∗∗ 1

Helps in Lifeb 0.643∗∗∗ 0.229∗∗∗ 0.389∗∗∗ 0.299∗∗∗ 1
Strikes as Calculating 0.007 −0.313∗∗∗ −0.288∗∗∗ −0.315∗∗∗ 0.034
a Mean-centered, continuous variables (for attractiveness, honesty, likability, trustworthiness).
b Averaged, continuous variables (for “appearance helps in life”, “appearance strikes as calculating”).

11Again, there may be differences between the means of the statement variables’ ratings by the raters’
gender. Student t-tests w.r.t. gender are t = 0.851, p = 0.395 for “appearance helps in life” and
t = −5.437, p = 0.000 for “appearance strikes as calculating”, i.e., male raters are more likely to rate a
contestant as calculating than female raters.

12We use Cronbach’s alpha for standardized variables to measure the inter-item reliability for facial
appearance. The measure adjusts for item specific mean and variance. Also a nonparametric test for
testing whether samples originate from the same distribution cannot be rejected (Kruskal-Wallis K =
0.9868).

9



The statement variable “appearance helps in life” is also positively correlated with all

four facial appearance variables. However, the statement variable “appearance strikes as

calculating” is not correlated with attractiveness, and even negatively correlated with the

remaining facial appearance variables.

In the following, we focus on facial attractiveness, since it is a crucial part of the first

impression of a person and a stable characteristic which is almost impossible to mimic

(see e.g., Grammer, Fink, Møller, and Thornhill (2003)). The two statement variables are

used as additional controls.

There are many ways to define an attractive person. We use the following classifications:

First, we classify a contestant as facially attractive if her facial attractiveness rating lies

above or is equal to the mean over all facial attractiveness ratings, and as facially unattrac-

tive if her facial attractiveness rating lies below this mean. The average mean-centered

rating of facially attractive contestants is 0.599, and the one of facially unattractive con-

testants is −0.650. We also used the median over all facial attractiveness ratings as a
classification device. The average mean-centered rating of a facially attractive contestant,

who is rated above or equal to (below) the median is 0.613 (−0.635). Second, we define
extreme measures of facial attractiveness. A contestant is classified as most attractive if

her facial attractiveness rating lies within the top 10% (25%) percentile of the distribution

of facial attractiveness and as least attractive if her facial attractiveness rating lies within

the bottom 10% (25%) percentile. Contestants who are rated as most attractive receive,

on average, a mean-centered rating of 1.389 (0.993), and those who are rated as least

attractive receive a mean-centered rating of −1.386 (−1.003). For an illustration of the
distribution of facial attractiveness see Figure A.1 in the appendix.

Table IV: Summary statistics facial attractiveness

Variable Mean Std. Dev. Min. Max. N

Attractiveness (mean, d) 0.52 0.50 0 1 422
Attractiveness (median, d) 0.51 0.50 0 1 422
Most attractive (90% percentile, d) 0.11 0.31 0 1 422
Most attractive (75% percentile, d) 0.25 0.44 0 1 422
Least attractive (10% percentile, d) 0.10 0.30 0 1 422
Least attractive (25% percentile, d) 0.25 0.43 0 1 422

Helps in Life (d) 0.50 0.50 0 1 422
Strikes as Calculating (d) 0.47 0.50 0 1 422
(d) for dummy variable.

Third, we define a contestant’s appearance to be helpful in life (to strike as calculating ) if

her “helps in life”-rating (“strikes as calculating”-rating) lies above or is equal to the mean

over all “helps in life”-ratings (“strikes as calculating”-ratings), and as not to be helpful

in life (not to strike as calculating ) if her “helps in life”-rating (“strikes as calculating”-

10



rating) lies below this mean. Table IV summarizes the binary attractiveness and statement

variables. A detailed description of the distribution of attractive and unattractive finalists

with respect to gender and age is provided by Table A.3 in the appendix.

4 Results

4.1 Facial attractiveness

In order to investigate the relationship between cooperative behavior and facial attractive-

ness we use several binary probit models with the decision to cooperate as the dependent

variable (with yi = 1 equal cooperate; yi = 0 equal defect). Throughout the analysis,

we control for effects and interactions related to the contestant’s gender and age, other

demographic characteristics, as well as variables of stake size, communication, and vari-

ables describing the course of events of the game show previous to the prisoner’s dilemma

(pre-play), see Table A.1 in the appendix.

Own attractiveness The results depicted in Table V, model (1) to (5), show that

facially attractive contestants do not behave differently with respect to cooperativeness

than facially unattractive contestants, independent of the specification of the attractive-

ness measure.13 This finding is in line with our conjecture from the literature (see e.g.,

Eagly, Ashmore, Makhijani, and Longo (1991)): physically attractive people are not more

pro-social and therefore not more cooperative than physically unattractive people.

Whereas, overall, we find no difference between the cooperative behavior of attractive

and unattractive people, there are some qualifications with respect to gender and age.14

Both gender and age seem to mediate the effect of a contestant’s own attractiveness on

13Table V reports the regression results including a dummy variable for the attractive contestant. For
robustness of all our results, we also estimate the regressions including (i) the continuous measure for
attractiveness, (ii) the mean over the four appearance variables, (iii) the predicted factors obtained in a
confirmatory factor analysis of the four appearance variables, (iv) a normalized attractiveness measure
in line with Mobius and Rosenblat (2006), in which our constructed attractiveness measure is normalized
across all contestants, (v) a normalized attractiveness measure in line with Biddle and Hamermesh (1998),
in which the normalization is across all raters, and (vi) the mean and median attractiveness ratings of
the raw data, where the mean (median) is 0.443 (0.728) with an average rating of facially attractive
contestants above the mean (median) of 5.552 (4.914), and below the mean (median) of 3.196 (2.357).
All measures produce qualitatively the same results.

14Irrespective of a contestant’s own attractiveness, we find a very strong and significant correlation
between age and cooperative behavior (see Table V): older contestants (≥ 37 years) are much more likely
to cooperate than younger contestants (< 37 years), regardless of the age of the opponent. This result is
in line with List (2006) who finds that contestants ≥ 31 years are significantly more likely to cooperate
than younger contestants. Concerning gender, we find no direct effect, which is contrary to the studies
of e.g., Kahn, Hottes, and Davis (1971) and Ortmann and Tichy (1999). But we can show that younger
males are significantly more likely to defect; and, as age increases, males are more likely to cooperate
than females (see Table A.4 in the appendix).

11



cooperation, see Table A.5 in the appendix. Attractive females (model (1)) and attrac-

tive younger contestants (model (3)) show more cooperative behavior, whereas attractive

males (model (1)) and attractive older contestants (model (3)) cooperate less.

Opponent’s attractiveness We now turn to the impact of the opponent’s facial at-

tractiveness on a contestant’s willingness to cooperate. The regression results in Table V

show that contestants are 10−16 percentage points more likely to cooperate when facing
an attractive opponent than when facing an unattractive opponent. Hence, attractive

contestants are rewarded with greater cooperativeness, and this provides attractive con-

testants a beauty premium. The premium is independent of the opponent’s gender and

age, see Table A.5, model (2) and (4) in the appendix. Furthermore, our results show that

least attractive contestants suffer a beauty penalty due to lower cooperativeness towards

them. As model (5) reports, a contestant is less likely to cooperate if the opponent is

rated to be least attractive than if the opponent is neither rated to be most nor least

attractive. There is no significant effect on cooperative behavior if the opponent is rated

to be most attractive. The result holds independently of the definition of the extreme

measure of attractiveness. These findings suggest that contestants rather focus on the

opponent’s “negative” than “positive” appearance in the decision to cooperate.

The results remain unchanged when adding the two binary statement variables (see model

(4) of Table V). There is additional evidence that a contestants is less likely to cooperate

if the opponent’s appearance is rated as to “help her in life” than if it is not. But we

find no interaction effect of the binary statement variables and our attractiveness mea-

sures, which indicates that the statement variables have not much additional explanatory

power.15

Similarity Since contestants behave more cooperatively towards the attractive counter-

part, the question arises whether pairs of attractive contestants behave differently in the

prisoner’s dilemma than pairs of unattractive contestants or pairs who are mixed in terms

of facial attractiveness. We denote pairs of contestants as similar team if either both

contestants are facially attractive or both are facially unattractive, and we identify pairs

of contestants as an attractive team if both are facially attractive, and as an unattractive

team if both are facially unattractive. We find no evidence that contestants who are sim-

ilar with respect to attractiveness are more or less likely to cooperate or defect, see model

(1) of Table A.6 in the appendix. However, we find that an unattractive contestant is less

likely to cooperate with her unattractive counterpart compared to teams of contestants

15Note, the effect of the variable “the contestant’s appearance helps her in life” without the inclusion
of the facially attractiveness variables is positive, but not significant. We also tested for interaction
effects between the two statement variables and gender or age (table unreported). We find that a female
opponent, whose appearance is rated as helping her in life, lowers the contestant’s propensity to cooperate;
but this effect is not robust with respect to different specifications of the model.

12



T
a
b
le

V
:

R
es

u
lt

s
fr

o
m

b
in

a
ry

p
ro

b
it

re
g
re

ss
io

n
s

o
f

th
e

d
ec

is
io

n
to

“
co

o
p

er
a
te

”
(y
i

=
1
)

o
r

“
d
ef

ec
t”

(y
i

=
0
)

in
th

e
p
ri

so
n
er

’s
d
il
em

m
a

M
a
rg

in
a
l

E
ff

e
c
ts

M
o
d
e
l

(1
)

M
o
d

e
l

(2
)

M
o
d

e
l

(3
)

M
o
d

e
l

(4
)

M
o
d

e
l

(5
)

A
tt

ra
c
ti

v
e
n

e
ss

(m
e
a
n

-c
e
n
te

re
d

,
d
)

(d
)

0
.0

6
7

(0
.0

4
9
)

0
.0

6
7

(0
.0

5
2
)

0
.0

4
8

(0
.0

5
3
)

0
.0

2
6

(0
.0

6
0
)

O
p
p

o
n
e
n
t

A
tt

ra
c
ti

v
e
n
e
ss

(m
e
a
n
-c

e
n
te

re
d
,

d
)

(d
)

0
.1

1
2
∗
∗

(0
.0

4
6
)

0
.1

2
5
∗
∗

(0
.0

4
9
)

0
.1

0
1
∗
∗

(0
.0

5
0
)

0
.1

6
2
∗
∗
∗

(0
.0

5
7
)

A
p

p
e
a
ra

n
c
e

H
e
lp

s
In

L
if

e
(d

)
0
.0

6
2

(0
.0

6
6
)

0
.0

7
5

(0
.0

6
1
)

O
p
p

o
n
e
n
t

A
p

p
e
a
ra

n
c
e

H
e
lp

s
In

L
if

e
(d

)
-0

.1
1
0
∗

(0
.0

6
4
)

-0
.0

5
0

(0
.0

6
0
)

A
p

p
e
a
ra

n
c
e

S
tr

ik
e
s

A
s

C
a
lc

u
la

ti
n

g
(d

)
0
.0

1
7

(0
.0

5
6
)

0
.0

1
2

(0
.0

5
5
)

O
p
p

o
n
e
n
t

A
p

p
e
a
ra

n
c
e

S
tr

ik
e
s

A
s

C
a
lc

u
la

ti
n
g

(d
)

-0
.0

0
1

(0
.0

5
6
)

0
.0

0
0

(0
.0

5
6
)

M
o
st

a
tt

ra
c
ti

v
e

(9
0
%

p
e
rc

e
n
ti

le
)

(d
)

-0
.1

3
4

(0
.0

8
6
)

O
p
p

.
m

o
st

a
tt

ra
c
ti

v
e

(9
0
%

p
e
rc

e
n
ti

le
)

(d
)

-0
.0

9
6

(0
.0

8
5
)

L
e
a
st

a
tt

ra
c
ti

v
e

(1
0
%

p
e
rc

e
n
ti

le
)

(d
)

-0
.0

7
7

(0
.0

9
0
)

O
p
p

.
le

a
st

a
tt

ra
c
ti

v
e

(1
0
%

p
e
rc

e
n
ti

le
)

(d
)

-0
.2

3
0
∗
∗
∗

(0
.0

8
1
)

M
a
le

(d
)

-0
.0

3
6

(0
.0

4
8
)

-0
.0

4
6

(0
.0

5
0
)

-0
.0

5
8

(0
.0

5
1
)

-0
.0

5
2

(0
.0

5
6
)

-0
.0

6
0

(0
.0

5
1
)

A
g
e

(c
o
n
t.

)
0
.0

8
1
∗
∗
∗

(0
.0

2
3
)

0
.0

8
2
∗
∗
∗

(0
.0

2
5
)

0
.0

8
8
∗
∗
∗

(0
.0

2
7
)

0
.0

9
3
∗
∗
∗

(0
.0

3
0
)

0
.0

8
5
∗
∗
∗

(0
.0

2
8
)

O
p
p

.
M

a
le

(d
)

-0
.0

0
8

(0
.0

5
6
)

O
p
p

.
A

g
e

(c
o
n
t.

)
0
.0

1
6

(0
.0

2
8
)

P
ro

m
is

e
o
r

V
o
w

(d
)

0
.2

4
0
∗
∗
∗

(0
.0

5
1
)

0
.2

5
4
∗
∗
∗

(0
.0

5
2
)

0
.2

5
8
∗
∗
∗

(0
.0

5
2
)

O
p
p

.
P

ro
m

is
e

o
r

V
o
w

(d
)

0
.0

0
6

(0
.0

5
4
)

-0
.0

0
2

(0
.0

5
4
)

-0
.0

1
3

(0
.0

5
4
)

H
a
n

d
sh

a
k
e

(d
)

-0
.1

0
2
∗

(0
.0

5
9
)

-0
.0

9
8

(0
.0

6
0
)

-0
.1

1
7
∗
∗

(0
.0

5
9
)

D
e
m

o
g
r
a
p
h
ic

s
–

X
X

X
X

S
t
a
k
e

S
iz

e
–

X
X

X
X

P
r
e
-P

la
y

–
–

X
X

X

W
a
ld
χ

2
1
8
.5

9
∗
∗
∗

4
3
.0

1
∗
∗
∗

6
3
.9

6
∗
∗
∗

6
9
.2

1
∗
∗
∗

6
8
.2

1
∗
∗
∗

L
o
g
-L

ik
e
li
h
o
o
d

-2
8
2
.7

5
-2

6
4
.3

9
-2

5
0
.7

4
-2

4
8
.5

3
-2

4
6
.7

9
A

d
ju

st
e
d

R
2

0
.0

1
3

0
.0

3
9

0
.0

5
5

0
.0

4
2

0
.0

4
8

N
4
2
2

4
1
9

4
1
9

4
1
9

4
1
9

N
u

m
b

e
r

o
f

c
lu

st
e
rs

2
1
1

2
1
1

2
1
1

2
1
1

2
1
1

S
ta

n
d

a
rd

e
rr

o
rs

in
p

a
re

n
th

e
se

s
a
re

c
o
rr

e
c
te

d
fo

r
e
p

is
o
d

e
c
lu

st
e
rs

;
*
p
<

0
.1

0
,

*
*
p
<

0
.0

5
,

*
*
*
p
<

0
.0

1
(d

)
fo

r
d

is
c
re

te
ch

a
n
g
e

o
f

d
u
m

m
y

v
a
ri

a
b

le
fr

o
m

0
to

1

13



who are both attractive or mixed, see model (2) and (3) of Table A.6 in the appendix.

This improves our attractiveness-results: contestants not only behave more cooperatively

towards an attractive partner, but also more deceitfully towards an unattractive partner,

and this unattractive-penalty is likely to dominate.16

Thus, our results provide evidence for a causal relationship between the opponent’s at-

tractiveness and cooperative behavior. Facially attractive contestants are able to provoke

cooperation from their counterpart, independent of their gender or age. But we do not

find a significant difference in behavior between facially attractive and facially unattractive

contestants.

4.2 Beauty premium

The results of the previous section should also translate into a monetary beauty premium,

i.e., into higher earnings for the attractive than for the unattractive. In order to quantify

the marginal beauty premium we use a standard censored tobit model. The outcome

“taking no money home” from the prisoner’s dilemma is interpreted as a corner solution

outcome, where the response variable yi describes the observable outcome of a contestant,

which takes on the value zero with positive probability (if the opponent defects), and

which is a continuous variable over strictly positive values (if the opponent cooperates).

We estimate marginal effects for contestant i’s expectation of receiving positive earnings

yi > 0, conditional on i’s own and her opponent’s facial attractiveness (standardized mea-

sures), demographic characteristics, the log value of the stake size, as well as opponent

characteristics, which are summarized in Xi, E(yi|Xi,yi > 0).
We find that a one standard deviation increase in facial attractiveness significantly in-

creases a contestant’s expected positive earnings by £617 to £741 depending on the

controls included, see Table A.7 in the appendix. Figure 3 provides an illustration of

the beauty premium based on model (2) of Table A.7 in the appendix, and depicts the

predicted positive earnings for attractive and unattractive contestants as a function of

stake size. The figure shows that the predicted earnings of facially attractive contestants

are always clearly above the earnings of unattractive contestants, given the size of the

jackpot.17

Decomposing the beauty premium with respect to gender and age, we can show that

there are no differences in receiving positive earnings between attractive male and female,

16For robustness of our results, we repeat the analysis using ordered probits on the contestants’ likeli-
hood to mutually cooperate. Again, similar teams, as well as teams of attractive contestants are no more
likely to reach a certain outcome. Only pairs of unattractive contestants are significantly less likely to
reach mutual cooperation (tables unreported). Additionally, we estimate regressions including the relative
difference between both final contestants’ attractiveness, i.e., the distance in attractiveness between both
contestants, and including an index of the contestants’ similarity with respect to facial attractiveness,
age, and gender, weighting each component by one-third. All measures do not matter for the contestant’s

14



Figure 3: Predicted positive earnings by attractive and unattractive contestants, given
stake size

0

2000

4000

6000

8000

10000

12000

14000

16000

18000

20000

pr
ed

ic
te

d 
ea

rn
in

gs

0 10000 20000 30000 40000 50000 60000 70000 80000 90000100000
stake size

95% CI attractive unattractive

as well as attractive older and younger contestants. However, a contestant’s expected

earnings increase the older the opponent, and a contestants expected earnings decrease

when the opponent is male (see model (2) of Table A.7 in the appendix).

5 Transmission channels

Our results show that attractive people are able to provoke cooperative behavior from their

opponent and, since they are not more or less cooperative than unattractive people, they

obtain a beauty premium in the prisoner’s dilemma. In this section we will address the

potential transmission channels suggested by the literature (see Section 2) and evaluate

their explanatory content for our observed effects of attractiveness.

5.1 Beauty-is-good stereotyping and taste-based discrimination

The most prominent theory is that people hold stereotype beliefs about attractive people.

People believe that attractive ones behave more pro-socially than less attractive ones. In

the presence of interdependent social preferences, people may belief that attractive ones

are more cooperative and behave more cooperatively towards them with the intention to

reciprocate cooperation. Hence, these stereotype beliefs can mediate people’s behavior.

Recall, however, that we do not find any difference between attractive and unattractive

contestants regarding their likelihood to cooperate. If “beauty-is-good” stereotyping is

decision to cooperate.
17We obtain a similar result for the overall conditional expectation that a contestant receives the

outcome yi, given Xi, E(yi|Xi). A one standard deviation increase in facial attractiveness significantly
increases a contestant’s expected earnings by £1 784 to £2 153. Further, we estimate that the probability
of obtaining positive earnings, given Xi, Pr(yi > 0)|Xi), increases by up to 5.9 percentage points (tables
unreported).

15



driving our result, then we should observe a higher likelihood of cooperation towards at-

tractive contestants in the prisoner’s dilemma and the effect should be independent of the

selected sample, e.g., the effect should be present in interactions between contestants of

the same sex as well as between contestants of the opposite sex.

However, theories from evolutionary psychology argue that effects of physical attractive-

ness on behavior originate in primeval partner selection and therefore predict that effects

of attractiveness are more prevalent in mixed-gender interactions (see e.g., Cosmides and

Tooby (1987)). To scrutinize this argument, we parse the data in mixed-gender and

same-gender interactions (63% and 37% of all interactions) and run several probits to

evaluate the influence of facial attractiveness on a contestant’s propensity to cooperate,

controlling for demographic characteristics and stake size. The two subsamples are not

different regarding the observed cooperation rate (54% in mixed- vs. 53% in same-gender

interactions, see Table VI).

Table VI: Results from binary probit regressions on outcomes in the prisoner’s dilemma
(subsamples)

Marginal Effects

(1) Mixed-gender (2) Same-gender

Attractiveness (mean-centered, d) 0.047 (0.067) 0.101 (0.094)
Opp. Attractiveness (mean-centered, d) 0.151** (0.062) 0.013 (0.092)
Male (d) -0.049 (0.059) -0.104 (0.107)
Age (cont.) 0.076** (0.033) 0.111*** (0.042)

Demographics X X
Stake Size X X

Wald χ2 47.27*** 20.01*
Log-Likelihood -161.07 -96.46
Adjusted R2

N 265.00 154.00
Number of clusters 133.00 78.00

Note: Binary probit regressions of the decision to “cooperate” (yi = 1) or “defect” (yi = 0) in the
prisoner’s dilemma, restricting the sample to (1) mixed-gender and (2) same-gender interactions.
The marginal effect of the respective explanatory variable determines the effective change of this
variable on player i’s predicted probability to “split”. Standard errors are reported in parentheses
and are corrected for episode clusters. * p < 0.10, ** p < 0.05, *** p < 0.01

As the regression results of Table VI show, attractiveness only matters in mixed-gender

interactions, but not in same-gender ones. Males and females are about 15% percentage

points more likely to cooperate if facing a facially attractive opponent of the other sex

than if they face an unattractive opponent.18 The finding applies across sexes, both fe-

males and males are biased towards attractiveness if facing someone of the other sex. This

18This result is also displayed in the fact that facially attractive contestants enjoy a beauty premium
in mixed-gender interactions, but not in same-gender ones (tobit regression tables unreported). A one
standard deviation increase in attractiveness results in a significantly higher gains for the attractive in
mixed-gender (p=0.003), but not in same-gender interactions (p=0.840).

16



extends the finding of Ashmore and Longo (1995), who note that only attractive females

tend to have the ability to make males more likely to do them a favor. We show that

males have the same ability.

According to the theory of taste-based discrimination (Becker (1957)), contestants might

favor to be in the final round with a facially attractive fellow player. Hence, taste-based

discrimination could be the underlying mechanism if we identify a beauty premium in the

pre-play, but not in all interactions in the prisoner’s dilemma. Using binary and ordered

probit models, we estimate the effect of attractiveness on the likelihood to be voted off

the game and on the likelihood to receive a certain number of votes after the first and

second round. We find no significant effect of attractiveness on the voting outcome (tables

unreported).19 This is also reflected in the almost equal share of attractive contestants in

the final and the one in the initial round of the game show (52.4% vs. 51.8%).

The absence of a beauty premium or plainness penalty in the pre-play and in same-gender

interactions in the final suggests that neither beauty-is-good stereotyping nor taste-based

discrimination can explain our results consistently. It rather seems that the ability of

attractiveness to mediate behavior is entangled to the opponent’s sex and that people

have a preference to cooperate with someone to whom they are personally attracted.

5.2 Vocal attractiveness and social skills

In addition to the theories discussed above, the literature reports that physical and vocal

attractiveness are highly correlated and thus suggests the existence of a vocal rather than

a visual attractiveness stereotype (see, e.g., Zuckerman and Driver (1989)). Furthermore,

physically attractive people are also assigned stronger verbal and social skills (e.g., Gold-

man and Lewis (1977) and Erwin and Calev (1984)). In the presence of other regarding

preferences, vocal attractiveness and potentially strong verbal and social skills might en-

able attractive people to trigger cooperative behavior off their opponents. Although our

data does not allow us to directly test for the impact of (perceived) vocal attractiveness on

cooperative behavior, we can indirectly test for a correlation between facial attractiveness

and communication which comprises verbal and social skills.

Shortly before the prisoner’s dilemma is played both contestants are given some extra time

to talk to each other. In these short conversations which are, on average, 38 seconds long,

each contestant tries to convince her opponent to cooperate. Since it has been shown that

19In all regressions we control for demographic characteristics such as gender, age, race and place of
residence as well as for objective voting criteria such as the stake size a contestant accumulated, and
whether a contestant lied in a previous round of the pre-play. We also follow Belot, Bhaskar, and van de
Ven’s approach and rank the contestants by their facial attractiveness to explain the likelihood to be
voted off during the game, which yields no significant effects either (table unreported). Our finding is
in contrast to Belot, Bhaskar, and van de Ven (2012), who find that attractive people are positively
discriminated against unattractive people and hence pass the pre-play more easily.

17



promises effect people’s behavior in experiments (e.g., Charness and Dufwenberg (2006)

and Vanberg (2008)) and in the field (e.g., Belot, Bhaskar, and van de Ven (2010)), we

code whether a contestant explicitly promises her opponent to cooperate.20 Furthermore,

we observe that contestants use handshakes to corroborate their mutual intention to co-

operate and therefore we also code whether two contestants shake hands. As Table V

shows, promises have a significantly positive impact on cooperative behavior, whereas

handshakes have a negative one.21 Comparing models (1) and (3) of Table V shows that

the effect of attractiveness on cooperation remains almost unchanged when we add the

variables of communication as controls.

If attractive contestants are better in terms of verbal and social skills, a promise or hand-

shake of a facially attractive contestant might be more convincing than a promise or

handshake of a facially unattractive contestant. Testing for interaction effects between

facial attractiveness and the communication variables, reveals no additional effects (table

unreported). Further, we could expect that facially attractive contestants are more likely

to elicit a promise from their opponent and less likely to engage in a handshake. Using

binary probit regressions on the contestant’s propensity to promise or to shake hands, we

find that facially attractive contestants are not more likely to state a promise or to shake

hands than facially unattractive contestants (table unreported). However, we find limited

evidence that contestants are significantly more likely to state a promise if the opponent

is attractive (table unreported). Since promising to cooperate significantly increases a

contestant’s likelihood to cooperate, we cannot exclude that promises are the underlying

transmission channel of facial attractiveness on cooperative behavior. The absence of the

beauty premium in the pre-play can also not be explained by the theory of better social

and verbal skills or vocal attractiveness.

6 Conclusion

This paper analyzes the relationship between attractiveness and cooperative behavior in

a high-stakes field setting with two-sided communication. Two independent data sets

are combined. One on cooperation, collected from decisions made in a slightly modified

prisoner’s dilemma played in the final round of a television game show. The other one

on the physical attractiveness of the game show’s contestants using a sample of indepen-

20We count all statements as a promise when they contain either the word ”promise” or ”swear” or they
are a statement of intent. Examples are ”I promise to split”, ”I promise I will not steal”, ”I swear I will
split”, ”I swear I will not steal”, ”I will split”, or ”I will not steal”.

21The effect of the communication variables is robust to various specifications of the regression model.
Including only demographics, the likelihood to cooperate of a contestant who made a promise to co-
operate is 25 percentage points (p=0.000) higher than the one of a contestant who did not promise to
cooperate. When contestants use a handshake they are actually 13 percentage points (p=0.031) less likely
to cooperate than when they do not use a handshake (table unreported).

18



dent third-party raters. In the prisoner’s dilemma two finalists play for an accumulated

jackpot by deciding either to “split” the jackpot or to “steal” it and to keep the entire

amount for themselves. The results show a strong and robust effect of attractiveness:

contestants are significantly more likely to cooperate with a facially attractive opponent.

But, facially attractive contestants are not more or less likely to be cooperative com-

pared to facially unattractive contestants. Hence, attractive contestants are rewarded by

a beauty premium, which, at the margin, amounts to up to £2 153 for an increase in

attractiveness of one standard deviation. The attractiveness effect is robust to demo-

graphic characteristics, including gender, stake size, and communication. However, the

effects of attractiveness might be amplified by the attractive contestant’s ability to talk

their opponent into promising to cooperate, which has a significantly positive effect on a

contestant’s likelihood to cooperate.

However, the decomposition of the data into same- and mixed-gender interactions reveals

a new picture. The ability of attractive contestants to elicit cooperative behavior from

their opponent vanishes in interactions between two contestants of the same sex. That

is, contestants are only biased by the facial attractiveness of their opponent when the op-

ponent is of the other sex. This suggests that stereotype beliefs about attractive people,

such as them being more pro-social, cannot explain our results. Since for this explanation

to hold, the effect should prevail in all interactions. In line with our finding are rather

theories from evolutionary psychology arguing that effects of physical attractiveness origi-

nate in primeval partner selection and should therefore be only or at least more present in

mixed-gender interactions (see e.g., Cosmides and Tooby (1987)). Also, the absence of a

beauty premium or plainness penalty in the pre-play suggests that physical appearance is

particularly important as soon as people are lacking objective information and moreover

it allows us to exclude taste-based discrimination as the underlying transmission channel

of attractiveness on cooperative behavior.

To confirm our hypothesis of a preference based mechanism, we test it with the data

from Mobius and Rosenblat’s experiments on the beauty premium in the labor market.

The authors argue that stereotype beliefs about attractive people’s productivity and the

higher confidence of attractive people in their own productivity are driving the results.

Attractive people earn significantly higher wages even though they do not display a higher

productivity than unattractive people. We re-estimate Mobius and Rosenblat’s full model

(Table 6, p.234 in Mobius and Rosenblat (2006)) parsing the data into mixed- and same-

gender interactions. The results are presented in Table A.8 in the Appendix. As predicted

by our hypothesis the effects of attractiveness remain only prevalent in interactions be-

tween either a female worker and a male employer or a male worker and a female employer

(Model (2)). In model (3) which uses the sample of same-gender interactions, all effects

of attractiveness disappear.

Our results are relevant and applicable to many, if not all, one-shot face-to-face inter-

19



actions. The better performance of attractive people in the labor market, for instance,

may be enforced by the fact that they also benefit from greater cooperativeness towards

them. And facing someone attractive of the other sex seems to bias people when making

economic decisions, which may help beautiful people in negotiations when cooperation is

required.

References

Alreck, P. L., and R. B. Settle (1995): The Survey Research Handbook. Chicago:

Irwin.

Andreoni, J., and R. Petrie (2008): “Beauty, Gender and Stereotypes: Evidence

from Laboratory Experiments,” Journal of Economic Psychology, 29, 73–93.

Antonakis, J., and O. Dalgas (2009): “Predicting Elections: Child’s Play!,” Science,

323, 1183.

Ashmore, R. D., and L. C. Longo (1995): “Accuracy of Stereotypes: What Research

on Physical Attractiveness can Teach us,” in Stereotype Accuracy: Toward Appreciating

Group Differences, ed. by Y. Lee, L. Jussim, and C. McCauley, chap. 3, pp. 63–86.

American Psycological Association.

Becker, G. (1957): The Economics of Discrimination. University of Chicago Press.

Belot, M., V. Bhaskar, and J. van de Ven (2010): “Promises and Cooperation:

Evidence from a TV Game Show,” Journal of Economic Behavior and Organizations,

73, 396–405.

(2012): “Beauty and the Sources of Discrimination,” Journal of Human Re-

sources, 47(3), 851–872.

Berggren, N., H. Jordahl, and P. Poutvaara (2010): “The Looks Of A Winner:

Beauty And Electoral Success,” Journal of Public Economics, 94, 8–15.

Biddle, J. E., and D. S. Hamermesh (1998): “Beauty, Productivity, and Discrimina-

tion: Lawyers’ Looks and Lucre,” Journal of Labor Economics, 16(1), 172–201.

Charness, G., and M. Dufwenberg (2006): “Promises and Partnership,” Economet-

rica, 74(6), 1579–1601.

20



Cosmides, L., and J. Tooby (1987): “From Evolution to Behavior: Evolutionary

Psychology as the Missing Link,” in The Lates on the Best: Essays on Evolution and

Optimality, ed. by J. Dupré, chap. 13, pp. 277–306. Cambridge: MIT Press.

Dion, K., E. Berscheid, and E. Walster (1972): “What Is Beautiful Is Good,”

Journal of Personality and Social Psychology, 24, 285–290.

Eagly, A. H., R. D. Ashmore, M. G. Makhijani, and L. C. Longo (1991): “What

is Beautiful is Good, But...: A Meta-Analytic Review of Research on the Physical

Attractiveness Stereotype,” Psychological Bulletin, 110(1), 109–128.

Eckel, C. C., and R. Petrie (2011): “Face Value,” American Economic Review, 101,

1497–1513.

Erwin, P. G., and A. Calev (1984): “Beauty: More than Skin Deep?,” Journal of

Social and Personal Relationships, 1, 359–361.

Feingold, A. (1992): “Good-Looking People Are Not What We Think,” Psychological

Bulletin, 111(2), 304–341.

Goldman, W., and P. Lewis (1977): “Beautiful is Good: Evidence that the Physically

Attractive are More Socially Skillful,” Journal of Experimental Social Psychology, 13,

125–130.

Grammer, K., B. Fink, A. P. Møller, and R. Thornhill (2003): “Darwinian

Aesthetics: Sexual Selection and the Biology of Beauty,” Biological Review, 78, 385–

407.

Jackson, L. A., J. H. Hunter, and C. N. Hodge (1995): “Physical Attractiveness

and Intellectual Competence: A Meta-Analysis Review,” Social Psychology Quarterly,

58(2), 108–122.

Kahn, A., J. Hottes, and W. L. Davis (1971): “Cooperation and Optimal Respond-

ing in the Prisoner’s Dilemma Game: Effects of Sex and Physical Attractiveness,”

Journal of Personality and Social Psychology, 17(3), 267–279.

Langlois, J. H., L. Klakanis, A. J. Rubenstein, A. Larson, M. Hallam, and

M. Smoot (2000): “Maxims or Myths of Beauty? A Meta-Analysis and Theoretical

Review,” Pssychological Bulletin, 126(3), 390–423.

List, J. A. (2006): “Friend or Foe? A Natural Experiment of the Prisoner’s Dilemma,”

The Review of Economics and Statistics, 88(3), 463–471.

21



Mobius, M. M., and T. S. Rosenblat (2006): “Why Beauty Matters,” The American

Economic Review, 96(1), 222–235.

Mueller, U., and A. Mazur (1997): “Facial Diominance in Homo Sapiens as Hinest

Signaling of Male Quality,” Behavioral Ecology, 8, 569–579.

Mulford, M., J. Orbell, C. Shatto, and J. Stockard (1998): “Physical Attrac-

tiveness, Opportunity and Success in Everyday Exchange,” The American Journal of

Sociology, 103(6), 1565–1593.

Ortmann, A., and L. K. Tichy (1999): “Gender differences in the laboratory: evidence

from prisoner’s dilemma games,” Journal of Economic Behavior and Organization, 39,

327–339.

Ottati, V. C., and M. Deiger (2002): “Visual Cues and the Candidate Evaluation

Process,” in Social Psychology of Politics, ed. by V. C. Ottati, and R. S. Tindale, pp.

75–87. New York: Kluwer Academic/Plenum.

Rosar, U., M. Klein, and T. Beckers (2008): “The Frog Pond Beauty Contest: Phys-

ical Attractiveness and Electoral Success of the Constituency Candidates at the North

Rhine-Westphalia State Election of 2005,” European Journal of Political Research, 47,

64–79.

Rosenblat, T. S. (2008): “The Beauty Premium: Physical Attractiveness and Gender

in Dictator Games,” Negotiation Journal, 24(4), 465–481.

Ruffle, B. J., and Z. Shtudiner (2010): “Are Good-Looking People More Employ-

able?,” Working Paper.

Solnick, S. J., and M. E. Schweitzer (1999): “The Influence of Physical Attractive-

ness and Gender on Ultimatum Game Decisions,” Organizational Behavior and Human

Decision Process, 79(3), 199–215.

Vanberg, C. (2008): “Why Do People Keep Their Promises? An Experimental Test of

Two Explanations,” Econometrica, 76(6), 1467–1480.

Wilson, R. K., and C. C. Eckel (2006): “Judging a Book by its Cover: Beauty and

Expectations in the Trust Game,” Political Research Quarterly, 59(189), 188–202.

Zuckerman, M., and R. E. Driver (1989): “What sounds beautiful is good: The

vocal attractiveness stereotype,” Journal of Nonverbal Behavior, 13, 67–82.

Zuckerman, M., H. Hodgins, and K. Miyake (1990): “The vocal attractiveness

stereotype: Replication and elaboration,” Journal of Nonverbal Behavior, 14, 97–112.

22



Appendix

A Figures and tables

Table A.1: Summary statistics prisoner’s dilemma

Variable Mean Std. Dev. Min. Max. N

Decision Variables
Cooperate 0.54 0.5 0 1 422
Mutual Decision 1.07 0.76 0 2 422
(0=“steal-steal”, 1=“steal-split”, 2=“split-split”)
Amount Money Taken Home 4 614.8 10 799.03 0 93 250 422

Demographics
Male 0.46 0.5 0 1 422
Agea(cont.) 3.21 1.1 1 6.5 422
Age of Male (cont.) 3.26 1.07 1 6.4 422
Age of Female (cont.) 3.19 1.09 1.25 6.6 422
White 0.94 0.25 0 1 422
London 0.11 0.32 0 1 422
England (1 = ENG, 0 = SCO, WAL, NIR, IRL) 0.85 0.36 0 1 420
Social Job (Reputation)b 0.16 0.36 0 1 421
Unexperienced (contestants of series 1) 0.19 0.39 0 1 422
Experienced (contestants of series 4) 0.19 0.39 0 1 422

Stake size
Jackpot 12912.33 18213.95 3 93250 422
Potential Jackpotc 50329.69 29946.46 5000 168100 422

Communication
Promise or Vow 0.42 0.494 0 1 422
Handshake 0.33 0.47 0 1 422

Pre-play
Accumulated Most Money 0.5 0.5 0 1 422
Selected Higher Values in Bin/Win 0.5 0.5 0 1 422
Selected Most Killers in Bin/Win 0.33 0.47 0 1 422
“Should Have Left The Game”d 0.26 0.44 0 1 422
Lied during Pre-Play 0.62 0.49 0 1 422
a Age is judged on a 7-item scale (see questionnaire and section 3.2), where 3=“30-40”, and 4=“40-50” implying that

the scale average of 3.21 equals a mean age of 37 years.
b A social job is defined as a job in which people care for other people, e.g., doctors, nurses, child minders, social

workers, teachers, police officers, firemen, soldiers.
c The potential jackpot is announced by the show host before the actual jackpot is determined. It displays the

maximal amount the actual jackpot could potentially comprise of.
d The variable “should have left the game” points at the player who is the “weakest” in material terms in round 2.

23



Figure A.1: Raw [left] and demeaned [right] variables of facial appearance, N=422

1

2

3

4

5

6

7

attractive honest likable trustworthy

−3

−2

−1

0

1

2

3

attractive honest likable trustworthy

Table A.2: Summary statistics all raters

Variable Mean Std. Dev. Min. Max. N

Age (in years) 41.76 18.46 17 93 720
Age of Male (in years) 39.57 17.70 17 93 371
Age of Female (in years) 44.10 19.00 18 86 349
Male 0.51 0.5 0 1 728

Female (≥ 44.1 years) 62.07 10.10 45 86 162
Female (< 44.1 years) 28.52 7.84 18 44 187
Male (≥ 39.6 years) 58.02 11.88 40 93 153
Male (< 39.6 years) 26.62 5.17 17 39 218

Table A.3: Distribution of attractive and unattractive finalists

facially attractive facially unattractive

Male (< 37 years) 21% 27%
Male (≥ 37 years) 15% 30%
Female (< 37 years) 41% 18%
Female (≥ 37 years) 23% 25%

100% (N=221) 100% (N=201)

24



T
a
b
le

A
.4

:
R

es
u
lt

s
fr

o
m

b
in

a
ry

p
ro

b
it

re
g
re

ss
io

n
s

o
f

th
e

d
ec

is
io

n
to

“
co

o
p

er
a
te

”
(y
i

=
1
)

o
r

“
d
ef

ec
t”

(y
i

=
0
)

in
th

e
p
ri

so
n
er

’s
d
il
em

m
a
,

in
cl

u
d
in

g
th

e
in

te
ra

ct
io

n
te

rm
“
g
en

d
er

*
a
g
e”

M
a
rg

in
a
l

E
ff

e
c
ts

M
o
d
e
l

(1
)

M
o
d

e
l

(2
)

M
o
d

e
l

(3
)

M
o
d

e
l

(4
)

M
o
d

e
l

(5
)

A
tt

ra
c
ti

v
e
n

e
ss

(m
e
a
n

-c
e
n
te

re
d

,
d
)

0
.0

6
2

(0
.0

5
2
)

0
.0

4
8

(0
.0

5
8
)

0
.0

2
5

(0
.0

5
9
)

O
p
p

.
A

tt
ra

c
ti

v
e
n
e
ss

(m
e
a
n
-c

e
n
te

re
d

,
d

)
0
.1

2
4
*
*

(0
.0

4
9
)

0
.1

6
8
*
*
*

(0
.0

5
6
)

0
.1

6
1
*
*
*

(0
.0

5
7
)

A
p

p
e
a
ra

n
c
e

H
e
lp

s
In

L
if

e
(d

)
0
.0

3
6

(0
.0

6
2
)

0
.0

5
6

(0
.0

6
5
)

0
.0

6
7

(0
.0

6
1
)

O
p
p

.
A

p
p

e
a
ra

n
c
e

H
e
lp

s
In

L
if

e
(d

)
-0

.0
9
0

(0
.0

6
1
)

-0
.1

0
8
*

(0
.0

6
4
)

-0
.0

4
9

(0
.0

6
0
)

A
p

p
e
a
ra

n
c
e

S
tr

ik
e
s

A
s

C
a
lc

u
la

ti
n

g
(d

)
0
.0

1
1

(0
.0

5
4
)

0
.0

1
8

(0
.0

5
7
)

0
.0

5
6

(0
.0

5
6
)

O
p
p

.
A

p
p

e
a
ra

n
c
e

S
tr

ik
e
s

A
s

C
a
lc

u
la

ti
n

g
(d

)
0
.0

1
8

(0
.0

5
5
)

0
.0

0
1

(0
.0

5
6
)

0
.0

0
2

(0
.0

5
6
)

M
o
st

a
tt

ra
c
ti

v
e

(9
0
%

p
e
rc

e
n
ti

le
,

d
)

-0
.1

4
1

(0
.0

8
7
)

O
p
p

.
m

o
st

a
tt

ra
c
ti

v
e

(9
0
%

p
e
rc

e
n
ti

le
,

d
)

-0
.0

9
0

(0
.0

8
7
)

L
e
a
st

a
tt

ra
c
ti

v
e

(1
0
%

p
e
rc

e
n
ti

le
,

d
)

-0
.0

8
8

(0
.0

8
9
)

O
p
p

.
le

a
st

a
tt

ra
c
ti

v
e

(1
0
%

p
e
rc

e
n
ti

le
,

d
)

-0
.2

3
5
*

(0
.0

8
1
)

M
a
le

*
A

g
e

(c
o
n
t.

)
0
.0

7
1
*
*

(0
.0

4
2
)

0
.0

6
7
*

(0
.0

4
2
)

0
.0

6
5
*

(0
.0

4
2
)

0
.0

4
7

(0
.0

4
3
)

0
.0

5
9
*

(0
.0

4
3
)

M
a
le

(d
)

-0
.3

0
4
*
*

(0
.1

4
7
)

-0
.2

8
6
*

(0
.1

4
9
)

-0
.2

7
3
*

(0
.1

4
9
)

-0
.2

3
2

(0
.1

6
5
)

-0
.2

8
8
*

(0
.1

6
0
)

A
g
e

(c
o
n
t.

)
0
.0

3
7

(0
.0

3
3
)

0
.0

4
8

(0
.0

3
4
)

0
.0

5
0

(0
.0

3
4
)

0
.0

6
6
*

(0
.0

3
9
)

0
.0

7
3

(0
.0

5
0
)

O
p
p

.
M

a
le

(d
)

-0
.0

0
4

(0
.0

5
6
)

O
p
p

.
A

g
e

(c
o
n
t.

)
0
.0

1
5

(0
.0

2
8
)

P
ro

m
is

e
o
r

V
o
w

(d
)

0
.2

5
2
*
*
*

(0
.0

5
1
)

0
.2

5
5
*
*
*

(0
.0

5
2
)

O
p
p

.
P

ro
m

is
e

o
r

V
o
w

(d
)

-0
.0

0
8

(0
.0

5
5
)

-0
.0

2
0

(0
.0

5
5
)

H
a
n

d
sh

a
k
e

(d
)

-0
.0

9
2

(0
.0

6
0
)

-0
.1

0
9
*

(0
.0

5
9
)

D
e
m

o
g
r
a
p
h
ic

s
X

X
X

X
X

S
t
a
k
e

S
iz

e
X

X
X

X
X

P
r
e
-P

la
y

–
–

–
X

X

W
a
ld
χ

2
4
0
.3

5
*
*
*

4
7
.2

1
*
*
*

4
8
.5

6
*
*
*

6
9
.1

7
*
*
*

7
0
.2

6
*
*
*

L
o
g
-L

ik
e
li
h
o
o
d

-2
6
6
.4

3
-2

6
3
.1

0
-2

6
1
.7

4
-2

4
7
.8

7
-2

4
5
.7

5
A

d
ju

st
e
d

R
2

0
.0

3
5

0
.0

4
0

0
.0

3
0

0
.0

4
0

0
.0

4
8

N
4
1
9

4
1
9

4
1
9

4
1
9

4
1
9

N
u

m
b

e
r

o
f

c
lu

st
e
rs

2
1
1

2
1
1

2
1
1

2
1
1

2
1
1

S
ta

n
d

a
rd

e
rr

o
rs

in
p

a
re

n
th

e
se

s
a
re

c
o
rr

e
c
te

d
fo

r
e
p

is
o
d

e
c
lu

st
e
rs

;
*
p
<

0
.1

0
,

*
*
p
<

0
.0

5
,

*
*
*
p
<

0
.0

1
(d

)
fo

r
d

is
c
re

te
ch

a
n
g
e

o
f

d
u
m

m
y

v
a
ri

a
b

le
fr

o
m

0
to

1

25



T
a
b
le

A
.5

:
R

es
u
lt

s
fr

o
m

b
in

a
ry

p
ro

b
it

re
g
re

ss
io

n
s

o
f

th
e

d
ec

is
io

n
to

“
co

o
p

er
a
te

”
(y
i

=
1
)

o
r

“
d
ef

ec
t”

(y
i

=
0
)

in
th

e
p
ri

so
n
er

’s
d
il
em

m
a
,

in
cl

u
d
in

g
in

te
ra

ct
io

n
te

rm
s

w
.r

.t
.

g
en

d
er

a
n
d

a
g
e

M
a
rg

in
a
l

E
ff

e
c
ts

M
o
d
e
l

(1
)

M
o
d

e
l

(2
)

M
o
d

e
l

(3
)

M
o
d

e
l

(4
)

A
tt

ra
c
ti

v
e
n

e
ss

*
M

a
le

(d
)

-0
.1

5
3
*

(0
.1

0
4
)

O
p
p

.
A

tt
ra

c
ti

v
e
n
e
ss

*
M

a
le

(d
)

0
.0

9
9

(0
.1

0
3
)

A
tt

ra
c
ti

v
e
n

e
ss

*
A

g
e

(c
o
n
t.

)
-0

.0
7
9
*
*

(0
.0

4
4
)

O
p
p

.
A

tt
ra

c
ti

v
e
n
e
ss

*
A

g
e

(c
o
n
t.

)
-0

.0
1
1

(0
.0

4
4
)

A
tt

ra
c
ti

v
e
n

e
ss

(m
e
a
n

-c
e
n
te

re
d

,
d
)

0
.1

3
3
*

(0
.0

7
8
)

0
.3

3
4
*
*

(0
.1

5
0
)

O
p
p

o
n
e
n
t

A
tt

ra
c
ti

v
e
n
e
ss

(m
e
a
n
-c

e
n
te

re
d
,

d
)

0
.0

6
8

(0
.0

7
3
)

0
.1

5
4

(0
.1

6
1
)

M
a
le

(d
)

0
.0

4
4

(0
.0

7
2
)

-0
.0

3
6

(0
.0

5
0
)

-0
.0

5
4

(0
.0

5
0
)

O
p
p

.
M

a
le

(d
)

-0
.0

5
5

(0
.0

7
5
)

A
g
e

(c
o
n
t.

)
0
.0

8
2
*
*
*

(0
.0

2
7
)

0
.0

7
5
*
*
*

(0
.0

2
4
)

0
.1

2
3
*
*
*

(0
.0

3
3
)

O
p
p

.
A

g
e

(c
o
n
t.

)
0
.0

0
6

(0
.0

3
3
)

D
e
m

o
g
r
a
p
h
ic

s
X

X
X

X
S
t
a
k
e

S
iz

e
X

X
X

X

W
a
ld
χ

2
3
6
.0

1
*
*
*

4
4
.6

3
*
*
*

4
3
.9

3
*
*
*

3
5
.3

2
*
*
*

L
o
g
-L

ik
e
li
h
o
o
d

-2
6
6
.0

8
-2

6
5
.2

9
-2

6
5
.6

5
-2

7
0
.0

2
A

d
ju

st
e
d

R
2

0
.0

3
3

0
.0

3
5

0
.0

3
4

0
.0

1
9

N
4
1
9

4
1
9

4
1
9

4
1
9

N
u

m
b

e
r

o
f

c
lu

st
e
rs

2
1
1

2
1
1

2
1
1

2
1
1

S
ta

n
d

a
rd

e
rr

o
rs

in
p

a
re

n
th

e
se

s
a
re

c
o
rr

e
c
te

d
fo

r
e
p

is
o
d

e
c
lu

st
e
rs

;
*
p
<

0
.1

0
,

*
*
p
<

0
.0

5
,

*
*
*
p
<

0
.0

1
;

(d
)

fo
r

d
is

c
re

te
ch

a
n
g
e

o
f

d
u

m
m

y
v
a
ri

a
b

le
fr

o
m

0
to

1

26



Table A.6: Results from binary probit regressions of the decision to “cooperate” (yi = 1)
or “defect” (yi = 0) in the prisoner’s dilemma when teams of contestants are considered

Marginal Effects

Model (1) Model (2) Model (3)

Similar Team (d) -0.043 (0.054)
Attractive Team (d) 0.045 (0.067) 0.027 (0.070)
Unattractive Team (d) -0.154** (0.064) -0.113* (0.067)

Male (d) -0.056 (0.050) -0.041 (0.050) -0.064 (0.056)
Opp. Male (d) -0.014 (0.056)
Age (d) 0.166*** (0.050) 0.187*** (0.051) 0.184*** (0.057)
Opp. Age (d) -0.034 (0.056)

Demographics X X X
Stake Size X X X
Pre-Play & Communication – – X
Wald χ2 40.40*** 46.68*** 68.32***
Log-Likelihood -266.77 -263.49 -250.48
Adjusted R2 0.034 0.042 0.049
N 419 419 419
Number of clusters 211 211 211

Note: The “team variables” are indicators and equal 1 if the team is so composed and 0 otherwise, e.g., “Similar
Team” equals 1 if both contestants are either attractive or unattractive, and 0 otherwise. Standard errors are
reported in parentheses and are corrected for episode clusters. * (p < 0.10), ** (p < 0.05), *** (p < 0.01)

27



Table A.7: Censored tobit regressions results for E[yi|Xi,yi > 0]

Marginal Effects

Model (1) Model (2)

Attractiveness (standardized) 616.927∗ (334.835) 741.419∗∗ (325.057)
Opp. Attractiveness (standardized) 457.207 (285.004)

Male (d) -209.774 (525.072) -328.794 (540.835)
Age (cont.) -218.725 (283.775) -8.920 (268.037)
Opp. Male (d) -1126.281∗ (573.093)
Opp. Age (cont.) 1087.797∗∗∗ (363.794)

log(Jackpot) 689.636∗∗∗ (157.020) 615.929∗∗∗ (146.146)

Demographics X X
Opp. Demographics – X

F-Statistic 3.186*** 2.843***
Log-Likelihood -2593.35 -2578.46
Adjusted R2 0.003 0.006
σ̂ 15059.81 14483.33
N 419 419
Number of clusters 211 211

Note: Censored tobit regression for the conditional expectation, that player i receives a positive earn-

ings yi > 0 from the prisoner’s dilemma game: E[yi|Xi,yi > 0] = Xiβ + σ
[
φ(Xiβ/σ)
Φ(Xiβ/σ)

]
, where the in-

verse Mills ratio is evaluated at
Xiβ
σ

. The marginal effect for the jth independent (continuous) variable

on E[yi|Xi,yi > 0] is computed as MEj =
∂E[yi|Xi,yi>0]

∂xj
= βj

[
1 − φ(Xiβ/σ)

Φ(Xiβ/σ)

(
Xiβ
σ

+
φ(Xiβ/σ)
Φ(Xiβ/σ)

)]
,

and quantifies the expected increase in earnings, conditional on being positive. (d) for discrete change
of dummy variable from 0 to 1. Standard errors are reported in parentheses and are corrected for
episode clusters. * p < 0.10, ** p < 0.05, *** p < 0.01

28



Table A.8: Fixed effects regression of Mobius and Rosenblat (2006)

Model (1) Model (2) Model (3)
All interactions Mixed gender Same gender

(Table 6, p. 234) interactions interactions

LNPROJECTED 0.409∗∗∗ (0.043) 0.423∗∗∗ (0.066) 0.337∗∗∗ (0.066)
LNPROJECTED*VISUAL 0.007 (0.059) 0.022 (0.088) 0.047 (0.089)
LNPROJECTED*AUDIO -0.129∗∗ (0.059) -0.101 (0.122) -0.091 (0.081)
LNPROJECTED*VISUAL*AUDIO 0.056 (0.084) -0.092 (0.155) 0.039 (0.117)
LNPROJECTED*FTF -0.069 (0.060) 0.003 (0.104) 0.038 (0.079)
LNACTUAL -0.004 (0.027) -0.094∗ (0.048) 0.032 (0.045)
BEAUTY -0.010 (0.031) -0.076∗ (0.046) 0.073 (0.055)
BEAUTY*VISUAL 0.094∗∗ (0.043) 0.170∗∗ (0.066) -0.007 (0.080)
BEAUTY*AUDIO 0.103∗∗∗ (0.035) 0.162∗∗ (0.067) 0.030 (0.059)
BEAUTY*VISUAL*AUDIO -0.097∗ (0.050) -0.161∗ (0.093) 0.076 (0.085)
BEAUTY*FTF 0.052 (0.035) 0.105 (0.070) 0.019 (0.053)
LNESTIMATED 0.018 (0.065) 0.157 (0.096) 0.048 (0.116)
LNESTIMATED*VISUAL 0.034 (0.083) -0.126 (0.118) 0.131 (0.149)
LNESTIMATED*AUDIO 0.265∗∗∗ (0.083) 0.084 (0.142) 0.205 (0.138)
LNESTIMATED*VISUAL*AUDIO -0.056 (0.117) 0.110 (0.200) -0.102 (0.194)
LNESTIMATED*FTF -0.116 (0.083) 0.065 (0.141) -0.253∗∗ (0.126)

F-Statistic 21.939∗∗∗ 8.018∗∗∗ 11.773∗∗∗

R2 0.627 0.664 0.709
N 812 384 423

Note: Significance levels: ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01. The dependent variable is LNWAGE, standard errors
are in parentheses. The base university is UNIVERSITY1.The regression includes the following resume controls:
demographic variables (sex, age and age squared, Internet at home, participation in team sports, choice of college
major, hobby variables, and previous job market experience). The regression also includes SETWAGE, and SETWAGE
interacted with BEAUTYand LNESTIMATED.

29