Dissertation
The Impact of Leadership Skills, Social Pressure and
Sabotage Behavior on Individual Income and the
Performance of Teams
Referent: Prof. Dr. Bernd Frick
Vorgelegt von
Dipl.- Volkswirt Christian Deutscher
Peter-Hille-Weg 2a
33098 Paderborn
Abgabetermin: 24.02.2010
Vorwort
Dass es die Möglichkeit geben würde, sich im Rahmen einer Doktorarbeit unter ande-
rem mit der Entlohnung von professionellen Basketballspielern zu beschäftigen, wirkt
für mich selbst nach Beendigung der Arbeit surreal. Auch wenn inhaltlich spannende
Forschungsfragen vieles erleichterten, wäre deren Bearbeitung ohne die vielfältige
Unterstützung verschiedener Personen nicht möglich gewesen. Ein besonderer Dank gilt
an dieser Stelle denjenigen, die mich maßgeblich bei der Erstellung unterstützt haben:
Meinem Doktorvater Prof. Dr. Bernd Frick für die optimale Betreuung und sein Ver-
trauen. Er gab mir immer die Möglichkeit eigene Ideen umzusetzen und half mir durch
inspirierende und konstruktive Kritik diese umzusetzen.
Prof. Dr. Erik E. Lehmann, Prof. Dr. Martin Schneider und Prof. Dr. René Fahr für ihre
Bereitschaft als Zweitgutachter bzw. Promotionskommissionsmitglieder zu fungieren
und für die fachlich äußerst bereichernde Diskussion im Rahmen der Disputation.
Prof. Dr. Mattias Kräkel und Prof. Dr. Oliver Gürtler, die mich zur Weiterverfolgung
einer akademischen Laufbahn motiviert, und den Kontakt zu meinem späteren Arbeit-
geber hergestellt haben.
Meinen lieben Kollegen Arne Büschemann, Filz Şen, Linda Kurze und Marcel Battré
für eine Zusammenarbeit, die weit über die berufliche Ebene hinaus geht und den
Begriff Freundschaft verdient. Gleiches gilt insbesondere für Prof. Dr. Joachim Prinz,
der mir immer eine große Unterstützung ist. Den studentischen Hilfskräften danke ich
für ihre tatkräftige Hilfe bei der Fertigstellung von Datensätzen.
Meinen Freundeskreisen aus Düsseldorfer Schulzeiten, Bonner Studienzeiten und den
letzten drei Jahren in Paderborn. Ein besonderer Dank gilt Alexander Westhoff, Marco
Bader und beiden Familien.
Meinen Eltern und meiner Schwester. Sie haben mir mein ganze Leben lang alles
geboten, was man sich wünschen kann: Liebe, bedingungslose Unterstützung, Zuspruch
und Motivation. Jedes erdenkliche Dankeswort an dieser Stelle kann nur untertrieben
bleiben.
Paderborn, im Juli 2010 Christian Deutscher
I
Table of Contents
Table of Contents ............................................................................................................. I
List of Figures ............................................................................................................... III
List of Tables ................................................................................................................. IV
1Introduction ........................................................................................................... 1
2Rewarding Leadership Skills in Professional Team Sports – Empirical
Evidence from the German Bundesliga .............................................................. 8
2.1Introduction ............................................................................................................ 8
2.2Data Set and Descriptive Statistics ...................................................................... 10
2.3Empirical Results ................................................................................................. 15
2.4Summary and Implications .................................................................................. 26
3The Payoff to Leadership in Teams ................................................................... 27
3.1Introduction .......................................................................................................... 27
3.2Data ...................................................................................................................... 29
3.3Empirical Analyses .............................................................................................. 32
3.4Conclusions .......................................................................................................... 39
4Performance under Pressure: Estimating the Returns to Mental Strength
in Professional Basketball .................................................................................. 40
4.1Introduction .......................................................................................................... 40
4.2What Can we Learn from the Available Literature? ........................................... 42
4.2.1Personality Traits and Earnings: A Review of the Evidence ..................... 42
4.2.2Performance and Remuneration in Professional Basketball...................... 45
4.3“Choking Under Pressure”: The Fragility of Performance under Stress ............. 47
4.4Data, Estimation, and Findings ............................................................................ 50
4.5Summary and Implications .................................................................................. 61
5Productivity in Friendly and Hostile Environments: Empirical Evidence
from the National Basketball Association. ........................................................ 62
5.1Introduction .......................................................................................................... 62
5.2Literature .............................................................................................................. 63
5.3Data ...................................................................................................................... 67
5.4Empirical Results ................................................................................................. 72
5.5Conclusion and Future Research ......................................................................... 78
II
6Cut-off Dates and Their Effect on Player Selection, Salaries and Hazard
Rates in the German Bundesliga ....................................................................... 79
6.1Introduction .......................................................................................................... 79
6.2Data ...................................................................................................................... 82
6.3Birth-Distribution in the Bundesliga ................................................................... 83
6.4Salary Determinations in the Bundesliga ............................................................ 84
6.5Hazard Rates ........................................................................................................ 90
6.6Conclusions .......................................................................................................... 94
7Sabotage in Heterogeneous Tournaments: A Field Study .............................. 96
7.1Introduction .......................................................................................................... 96
7.2The Model ............................................................................................................ 98
7.2.1Description of the Model and Notation ..................................................... 98
7.2.2Solution to the Model .............................................................................. 100
7.2.3Parameterized Version of the Model ....................................................... 102
7.3Data Set and Descriptive Statistics .................................................................... 103
7.4Empirical Results ............................................................................................... 107
7.5Conclusion ......................................................................................................... 109
7.6Appendix A ........................................................................................................ 111
7.7Appendix B ........................................................................................................ 111
7.8Appendix C ........................................................................................................ 114
8The Economics of the World Cup ................................................................... 116
8.1Introduction ........................................................................................................ 116
8.2Selection of Host Countries for the World Cup Finals ...................................... 117
8.3Benefits to Local and National Economies ....................................................... 121
8.4Benefits of Hosting the World Cup Finals to Soccer Fans ................................ 124
8.5Benefits to Players from Participating in the World Cup Finals ....................... 128
8.6Club versus Country?: The Domestic Player Quota Debate ............................. 136
8.7Conclusions ........................................................................................................ 139
9Outlook ............................................................................................................... 141
10List of Literature ................................................................................................. VI
III
List of Figures
Figure2‐1:KernelDensityEstimationoftheSalaryintheGermanBundesliga.........................16
Figure2‐2:SalaryHistoryofPlayersintheBundesligaSubjecttothePositionsontheField....18
Figure3‐1:KernelDensityEstimationofCaptainsandNon‐CaptainsNHLExperience..............30
Figure3‐2:KernelDensityEstimationoftheSalaryintheNHL..................................................33
Figure3‐3:Experience‐SalaryProfile..........................................................................................36
Figure4‐1:KernelDensityEstimation“MentalStrength”..........................................................50
Figure5‐1:AverageHomeAttendanceintheNBAbetween1997/1998and2006/2007.........68
Figure5‐2:KernelDensityEstimationoftheSalaryintheNBA.................................................71
Figure5‐3:KernelDensityEstimationofFreeThrowSuccessduringHomeandAwayGames.73
Figure6‐1:SalaryHistoryofGermanPlayersintheGermanBundesliga...................................85
Figure6‐2:KernelDensityEstimationofPlayers’SalarySubjecttotheQuarterofBirth..........86
Figure6‐3:SurvivalRatesintheBundesligawithRegardtothePositionontheField..............91
Figure6‐4:SurvivalRatesintheBundesligawithRegardtotheQuarterofBirth.....................94
Figure7‐1:Epanechikov‐KernelDensityEstimationofHET
......................................................114
Figure7‐2:Epanechikov‐KernelDensityEstimationoflogOddsofHET..................................115
IV
List of Tables
Table2‐1:CaptainsintheGermanBundesligawithThreeormoreSeasonsofExperience......12
Table2‐2:DescriptiveStatistics..................................................................................................14
Table2‐3:DeterminantsofPlayerSalaryintheGermanBundesliga.........................................19
Table2‐4:DeterminantsofPlayerSalaryintheGermanBundesligaIncludingFlexibility.........22
Table2‐5:QuantileRegressionsofPlayerSalaryintheBundesliga(Model1)..........................24
Table2‐6:QuantileRegressionsofPlayerSalaryintheBundesliga(Model2)..........................25
Table3‐1:DescriptiveStatisticsofPlayerCharacteristicsandPerformanceIndicators.............32
Table3‐2:DeterminantsofPlayerSalaryintheNHL..................................................................35
Table3‐3:QuantileRegressionsofPlayerSalaryintheNHL(Model1).....................................38
Table3‐4:QuantileRegressionsofPlayerSalaryintheNHL(Model2).....................................39
Table4‐1:DescriptiveStatistics..................................................................................................53
Table4‐2:TheImpactofMentalStrengthonPlayerSalaries(OLS‐Estimation).........................57
Table4‐3:TheImpactofMentalStrengthonPlayerSalaries(RE‐Estimation)...........................58
Table4‐4:TheImpactofMentalStrengthonPlayerSalaries(QuantileRegression).................59
Table5‐1:DescriptiveStatistics..................................................................................................70
Table5‐2:DeterminantsofFreeThrowShootingSuccessintheNBA.......................................74
Table5‐3:QuantileRegressionofFreeThrowSuccessduringHomeGames(Model1)............77
Table5‐4:QuantileRegressionofFreeThrowSuccessduringAwayGames(Model2)............77
Table6‐1:DistributionofGermanPlayers’BirthDatesintheBundesliga.................................83
Table6‐2:DescriptiveStatisticsofPlayerCharacteristicsandPerformanceIndicators.............87
Table6‐3:DeterminantsofPlayerSalary....................................................................................89
Table7‐1:DescriptiveStatistics................................................................................................106
Table7‐2:DeterminantsofFairandUnfairBehavior...............................................................108
Table8‐1:HostsandWinnersoftheWorldCupFinalssinceOrigin........................................118
V
Table8‐2:TeamBonusesatthe2006WorldCupFinals...........................................................129
Table8‐3:OLSandFixedEffectsResultsfortheGermanBundesliga......................................133
Table8‐4:QuantileRegressionResultsforlogSalary...............................................................134
Table8‐5:ProbitEstimatesofMovementtoamoreHighly‐RankedTeam.............................135
1
1 Introduction
Ever since the development of modern human capital theory by Becker (1962, 1964),
Mincer (1958, 1974) and Schulz (1961) countless studies have tried to explain what
impacts individuals’ salary. Most of them rely on the enclosure of traditional human
capital predictors like workers tenure, education and social background and are able to
explain a sizeable amount of salary differences. Only recently, studies have started to
include personality traits into the salary determination, even though it is common
knowledge that cognition and personality are related to labor market participation, as
stated by Heineck and Anger (2009). This lack of empirical evidence has earlier been
mentioned by Bowles, Gintis and Osborne (2001a). In their opinion, the scarcity of cor-
responding research is mainly based on the fact that economic theory simply does not
predict which personality traits influence earnings. Additionally, the authors suggest a
dependency between the job in itself and the influence of certain personality traits. They
claim that the desired personality is subject to the according job, as for example the per-
sonality requirements for a salesperson differ vastly from those for a police officer. Lee
(2006) supports this view, as in his opinion jobs themselves hold demand on personality
traits, which leads to personality being rewarded differently in different workplace situ-
ations. Having mentioned these specifics of evaluation, the question remains in which
way personality influences salary - given individuals’ human capital, cognitive ability
and job performance. The main goal of the following research is to explain prior unex-
plained components of salary by personality traits - a task given twenty years ago by
Earl (1990). In order to be able to relate the following work to previous studies, a short
overview of the already existing literature appears to be appropriate.1
Recent research focuses on the impact of personality traits on individuals’ salary as they
describe the individual’s emotional, interpersonal, experimental, attitudinal and motiva-
tional styles. To classify personality traits into categories, the so-called big five factors
of personality, also known as the Five Factor Model, have been introduced by Digman
1 Exploring the impact of personality on salary may also help to explain salary discrimination against
woman. Men and woman rather work in certain jobs, which reward or punish sets of personality traits
differently. For studies on this subject see Filer (1981), Nyhus and Pons (2005) and Fietze, Holst and
Tobsch (2009).
2
(1990). Namely they are referred to “extraversion”, “agreeableness”, “conscientious-
ness”, “emotional stability” and “autonomy”.2 Increasing consensus developed during
the last years on these five personality factors, as reported by Nyhus and Pons (2005). In
order to collect evaluation data about ones personality traits, which are typically self-
reported, studies refer to test procedures like the five-factor inventory developed by
Hendriks, Hofstee, De Raad und Angleitner (1999). This query contains 20 statements
for each of the five personality traits, of which half are phrased positively and half are
phrased negatively, as respondents select answers out of five-level Likert scales. The
Five Factor Model helps to understand the relationship between personality and job
criteria, as certain personality traits appear to be valid predictors of job performance.
Studies usually control for job occupation, since different jobs set unique requirements
to employees. In addition, cultural differences lead to varying exigencies on employees,
as pointed out by Schmidt, Ones and Hunter (1992). To understand the denotation of the
five previously named categories, a short explanation follows for each of them, all re-
lating to the definitions provided by Nyhus and Pons (2004).
• Extraversion is related to ones preference for human contact, attention and the
wish to inspire people. It describes people’s gregariousness and assertiveness to
illustrate how outgoing people are.
• Agreeableness describes to which degree people are willing to help others and
cooperate with them. It furthermore shows how individuals act consistent to oth-
ers peoples interest.
• Conscientiousness displays the preference for following rules and schedules,
working hard and being organized.
• Emotional stability shows if people are self-confident, calm and cool.
• Autonomy illustrates individuals’ penchant to make one’s own decisions.
Besides it describes to which degree an individual takes initiative and control.
Summing up results of previous research, there is to mention that results vary consider-
ably across studies, rarely exhibiting constant results concerning the impact of the
2 Other studies show minor differences in the category labels with basically the same meaning. See for
example Costa and McCrea (1985, 1992), who determine the categories of personality traits as
“extraversion”, “agreeableness”, “conscientiousness”, “neuroticism” and “openness to experience”.
3
described personality traits on salary. Meta-studies by Salgado (1997) for Europe as
well as by Barrick and Mount (1991) and Tett, Jackson and Rothstein (1991) for North
America find a coincident result in a way that conscientiousness is the only personality
trait which serves as a valid predictor of job performance in all three analyses. A more
recent study by Bondreau, Boswell and Judge (1999) which analyzes career success of
executives in North America and Europe, supports this finding. All other personality
traits prove to be valid predictors for only some job criteria and some occupational
groups. Problematic in this regard might the broad range of jobs included as well as the
self-reported evaluations of the respondents, which are nearly impossible to verify if
being correct. To account for these problems the following work concentrates on clearly
identified markets and includes objective and measurable indicators for individuals’
personality.
In contrast to research on the influence of personality on salary, research that deals with
the development of personality traits throughout the life course offers coincident results.
Literature shows consensus that personality is developed at young age and remains very
stable throughout life.3 Costa and McCrae (1988, 1994, 1997) claim that personality
remains consistent and if it actually does change, it changes at a very slow pace. They
suggest that personality traits stop changing at the age of 30, a finding which is sup-
ported in a further study by Brandstätter (1999). Nevertheless, it should be considered
that interaction with others as well as the surroundings of an individual lead to, even
though small, changes of personality over time. In this context, Srivastave, John, Gosl-
ing and Potter (2003) show that an individual’s social and job environment affects per-
sonality traits in early and middle adulthood, a finding which is also pointed out by
Borghans, Duckworth, Heckman and Ter Weel (2008).4 Using the German Socio-Eco-
nomic Panel, Flossmann, Piatek, Wichert and August (2007) examine the role of perso-
nality traits for labor market success. Their findings suggest that personality matters
even when controlling for different aspects such as education and professional expe-
rience. They conclude that labor market success is influenced by early childhood since
3 Jang, Livesley and Vernon (1996) show positive correlations of personality traits between generations,
as their finding support the theory that they are heritable.
4In line with this result, Caspi (1998) mentions that interaction between individuals affects personality
over time.
4
the formation of personality occurs during the first years of life under the influence of
the parents and the educational system.
To empirically test the following research questions, which mainly address the influ-
ence of personality on earnings, I use different data sets from professional sports, espe-
cially from the National Basketball Association, the National Hockey League and the
German Bundesliga, whose characteristics are being described in detail in the respective
chapters. Next to providing detailed information on individuals’ performance and sal-
ary, the team sports industry itself is an expanding market with team values growing on
a yearly basis and reaching considerable magnitudes. Especially for the US team sports
industry information about team values is very well kept by Forbes Magazine.5 For the
leagues considered in this work, the National Basketball Association displays the high-
est average team value with 367.0 million US-Dollars for the league consisting of 30
teams. Team values in the National Hockey League are comparably lower, as the league
which also consists of 30 teams exhibits an average team value of 222.6 million US-
Dollars. For the German Bundesliga which is 18 teams strong, individual team values
are only reported for the six most valuable teams. One observes team values to be way
less equally distributed across the teams as Bayern Munich is the most valuable team
with 1.11 billion US-Dollars, while the sixth most valuable team is VfB Stuttgart with
264 million US-Dollars.6 As teams generate increasing revenues, individual player sala-
ries also rise. Data on this is provided in the following chapters, also including informa-
tion on salary history. As one can see, from an economic standpoint professional sports
surely display an industry analyzing worthwhile.
The next two chapters of the following work deal with a personality trait which some
individuals are asked to show in team sports as well as in other group productions. Of
the five personality traits presented above, leadership skills are best fitting with extra-
version, as Heineck and Anger (2009) state that extraverts show a higher probability of
5 Yearly information on team values is provided on the magazines official website at
http://www.forbes.com/business/sportsmoney. All following information is from the year 2009.
6 The Los Angeles Lakers are the most valuable team in the National Basketball Association with 607
million US-Dollars, while the Toronto Maple Leafs is the most valuable team in the National Hockey
League with 470 million US-Dollars.
5
taking leadership roles.7 Leadership skills are assumed to be compensated monetarily,
nevertheless there is a lack of empirical evidence for this common thesis. Since in pro-
fessional sports the team captains are expected to possess these leadership skills, the
next two chapters will explore the impact of this ability on salary. The second chapter is
a joint work with Bernd Frick about the monetary impact of leadership skills in profes-
sional soccer. Using data of 13 consecutive seasons from the German Bundesliga, we
show that team captains receive a wage premium between 25 and 67 percent, dependent
on the model specification.8 In chapter three a similar research question is applied, this
time using data of four years from the National Hockey League. Again controlling for
individual player characteristics and performance indicators I show that leadership abil-
ity is rewarded pecuniary by a premium between 21 and 35 percent.9
Following in chapter four I turn to an analysis on the returns to mental toughness in
professional basketball and present a joint work with Bernd Frick and Joachim Prinz.
We classify mental toughness to the personality trait of emotional stability, as it depicts
individuals’ self-confidence and tendency to be calm. Since many team sporting events
are decided during the last seconds of the game, the question arises if players, who are
able to maintain their performance during crucial game situations, are rewarded pecu-
niary. While looking at the performance of players from the National Basketball Asso-
ciation for a time period of four consecutive years our data suggests that individuals get
monetarily rewarded for exhibiting mental toughness. Furthermore, we show that expe-
rience does not improve the ability to maintain the individual performance level during
crucial game situations. This implies that mental toughness is a congenital skill, a result
which goes in line with the mentioned findings that personality traits stay consistent
throughout a lifespan.
7 The relationship between conscientiousness and leadership, which is of highly interest in the work to
follow, is unclear. Tagger, Hackett and Saha (1999) find that leadership is most strongly associated
with cognitive ability, followed by the personality trait conscientiousness. By contrast, Judge and
Bono (2000) find conscientiousness to be unrelated to leadership, as they relate the five traits to
transformational leadership. For a survey article concerning different approaches to relate personality
to effectiveness see Andersen (2006).
8 Chapter two is the only one written in German. For the sake of clarity, I retain English chapter headings
and legends.
9 Based on this chapter the article “The Payoff to Leadership in Teams“, has been published in the Journal
of Sports Economics (Deutscher (2009)).
6
Chapter five is closely related as it analyses the influence of the audience on the per-
formance from the free throw line in professional basketball. I distinguish between the
performance in friendly and hostile environments as in home and away games. Since
past research neglects the impact of a change of the audience, professional sports appear
to be a fitting natural experiment as players change teams between seasons to face a
new home crowd after a team switch. The results show that players who sign a contract
with a new team show comparably worse performance in front of the new supportive
audience compared to players who either do not change teams or get traded to a new
team. I conclude that signing with a new team during the offseason puts additional pres-
sure on the player during home games in the following season, while it does not impact
performance during away games. Performing quantile regressions shows that especially
bad free throw shooters performance suffers from signing with a new team.
For the last three chapters I broaden my focus to other fields of research. Chapter six is
concerned with the impact of the implementation of cut-off dates on success in profes-
sional German soccer. Cut-off dates lead to groupings of children with regard to their
birthdays. Especially at young age this leads to big differences in relative age between
members of the same cohort, which in turn might lead to a selection bias. Older children
of the cohort might be labeled as being more talented, simply due to their physical
advantage over their younger counterparts. This actually causes an overrepresentation of
German players who were born shortly after the cut-off date in the Bundesliga for a
sample of 13 seasons. Prior studies find a wage premium for young players who make it
to the professional level despite their age disadvantage. I also analyze this for my data
set to find no support for this result, as the birth date does not affect the income. This
makes sense in the way that only performance enhancing factors should result in a wage
premium and the birth date itself should not impact the performance. In addition, hazard
rates for players should be independent of the date of birth as, again, the birth date
should not impact players’ performance. For the aforementioned data set on the Bun-
desliga this claim is supported.
Following in chapter seven is a joint work with Bernd Frick, Oliver Gürtler and Joachim
Prinz. We address the problem of sabotage in tournaments, when contestants are hetero-
7
geneous in ability. Our theoretical model states that favorites exert higher legal effort
while underdogs are more tempted to engage in sabotage actions, since favorites are
assumed to be more productive with respect to legal effort and both types of effort are
substitutes. In a second step, we use data from German professional soccer to test this
prediction empirically. In line with the theoretical model, we find that favorite teams
win more tackles in a fair way, while underdog teams commit more fouls.10
In chapter eight I once again broaden my focus to another field of research, as the
chapter provides insights on the economics of the FIFA World Cup, which is the most-
viewed sporting event on earth and generates billions in revenues. In a joint work with
Rob Simmons we present a general review of the literature on financial impact of the
World Cup for the hosting country. We provide an extension to the existing literature by
analyzing the remuneration of participating in the World Cup as a player. Using data
from the German Bundesliga we find support for a World Cup shop window effect.
Participating in the World Cup raises players’ salary and increases their chance to move
to stronger teams within Europe. Finally, chapter nine concludes the dissertation by
summarizing the results and providing a short outlook for ulterior research.
10 Based on this chapter the article “Sabotage in Heterogeneous Tournaments: A Field Study“, has been
sent to the Journal of Institutional and Theoretical Economy (status: revise and resubmit).
8
2 Rewarding Leadership Skills in Professional Team Sports –
Empirical Evidence from the German Bundesliga
2.1 Introduction
Unabhängig davon, ob man nationale Ligaspiele oder internationale Ländervergleiche
von Sportmannschaften betrachtet: Immer wieder fordern sowohl Trainer, als auch
Medien von den beteiligten Mannschaftskapitänen Führungsqualitäten unter Beweis zu
stellen. Unklar ist bislang, inwieweit diese Führungsqualitäten, welche die Kapitäne
vorweisen sollten, unter Berücksichtigung weiterer individueller Charakteristika sowie
beobachtbarer Leistungsdaten, zu einem höheren Lohn für die jeweiligen Spieler, wel-
che mit den Kapitänsaufgaben betraut werden, führen.
Die bestehende Literatur analysiert zumeist die Entlohnung von Führungsqualitäten auf
Managementebene in Unternehmen. Diese Führungsqualitäten werden im Allgemeinen
als “soft skill“ angesehen, deren Bedeutung bei der Rekrutierung von Personal immer
weiter an Bedeutung gewinnt.11 In ihrer, die bestehende Literatur prägenden, Arbeit
gehen Kuhn und Weinberger (2005) der Frage nach, inwieweit die Übernahme von
Positionen als Kapitän einer Schulsportmannschaft oder als Präsident einer schulischen
Organisation Einfluss auf die spätere Gehaltshöhe nimmt. Die beiden Autoren argu-
mentieren hierbei, dass die Übernahme dieser Positionen für den späteren Karrierever-
lauf relevante Führungsqualitäten fördert. Die der Arbeit zugrunde liegenden Datensätze
lassen die Autoren zu dem Ergebnis kommen das Männer, welche eine der beiden oben
genannten Positionen während ihrer Schulzeit besetzen, eine Dekade später im Durch-
schnitt ein zwischen 4 und 24 Prozent höheres Gehalt beziehen. Eine Reihe weiterer
Publikationen beschäftigen sich mit dem Einkommenseffekt der Partizipation an außer-
schulischen Aktivitäten. So zeigen Barron, Ewing und Waddell (2000), dass Schüler,
die am Sportprogramm ihrer High-Schools teilnehmen, im Beruf ein zwischen 4 und 15
Prozent höheres Gehalt beziehen als Schüler, die an anderen schulisch organisierten
Freizeitgestaltungen teilnehmen. In ihrer empirischen Untersuchung zeigen Cornelißen
und Pfeifer (2007) für deutsche Schüler einen signifikant positiven Zusammenhang
11 Vergleiche Moss und Tilly (2001).
9
zwischen der Partizipation an außerschulischen Sportaktivitäten und dem höchsten
erreichten Bildungsabschluss.
Unter Berücksichtigung dieser Erkenntnisse ergibt sich die Frage, welche Einfluss-
faktoren für die Entwicklung von Führungscharakteren von Bedeutung sind. Die Lite-
ratur bietet eine Reihe von Arbeiten, welche die Fähigkeit, Führungsrollen zu überneh-
men mit dem Alter von Individuen zu Beginn ihrer schulischen Laufbahn in Verbin-
dung bringen. Sie argumentieren in ähnlicher Weise, dass in Bundesstaaten, in denen
ein Stichtag für die Einschulung von Kindern besteht, der relative Altersunterschied
zwischen den Ältesten und den Jüngsten in der Klasse bis zu zwanzig Prozent beträgt.
Auf diese Feststellung aufbauend formulieren sie die Hypothese, dass diejenigen Kin-
der, die kurz nach dem Stichtag Geburtstag haben aufgrund ihrer fortgeschrittenen phy-
sischen und geistigen Entwicklung die Führung im Klassenverband übernehmen. Diese
Führungsqualitäten bleiben aufgrund der Erfahrung auf der Führungsposition auch in
späteren Jahren erhalten. Dhuey und Lipscomb (2008) zeigen, dass Schüler aus dem
ältesten Quantil eines Jahrgangs mit vier bis elf Prozent höherer Wahrscheinlichkeit die
Position als Mannschaftskapitän im Sport übernehmen als die jüngsten Schüler eines
Jahrgangs. Auch in der Selbsteinschätzung bezüglich Führungsqualitäten weisen ältere
Schüler signifikant höhere Werte auf. Allen und Barnsley (1993) kommen bei ihren
Untersuchungen von schulischen Leistungen sowohl in Kanada, wie auch in Groß-
britannien zu vergleichbaren Ergebnissen: Die ältesten Schüler eines Jahrgangs gehören
mit signifikant höherer Wahrscheinlichkeit zu den Besten.12 Zudem zeigen die Autoren,
dass in kanadischen Eishockeyligen Spieler aus dem ältesten Quantil eines Jahrgangs
bis zu viermal mehr vertreten sind als Spieler, die zu den Jüngsten des Jahrgangs gehö-
ren. Im Bezug auf die Fußballbundesliga unterscheiden Ashworth und Heyndels (2007)
in ihrer Untersuchung der Bundesligagehälter von zwei Saisons zwischen zwei ver-
schiedenen diskriminierenden Effekten des Geburtsdatums von Spielern: Zum einen
ermitteln sie eine Diskriminierung von Spielern, welche in großem Abstand zum Stich-
tag (1. August) für die Zuordnung von Jugendlichen in eine Altersklasse geboren wur-
12 Bedard und Dhuey (2006) kommen in ihrer Untersuchung verschiedener OECD-Länder ebenfalls zu
der Erkenntnis, dass die Jahrgangältesten bessere schulische Leistungen vorweisen als die jüngeren
Mitschüler. Obwohl der relative Altersunterschied mit zunehmendem Alter geringer wird, beobachten
die Autoren einen persistenten Leistungsunterschied.
10
den. Diese werden im Jugendalter „übersehen“ und sind folglich in der Bundesliga
unterrepräsentiert. Des Weiteren kommen die Autoren zu der Erkenntnis, dass Spieler,
die trotz ihres Altersnachteils die Fußballbundesliga erreichen, höher entlohnt werden.13
Während die Literatur also belegen kann, dass Individuen in der Wirtschaft für ihre
Führungsqualitäten entlohnt werden und zudem Erklärungsansätze für die unterschied-
liche Ausprägung dieser Fähigkeit liefert, herrscht bisher noch keine Erkenntnis darü-
ber, inwieweit Führungsqualitäten im professionellen Mannschaftssport zu einem höhe-
ren Gehalt führen. Die vorliegende Arbeit geht dieser Fragestellung anhand der deut-
schen Fußballbundesliga nach. Gerade vom Kapitän wird gefordert, seine Mannschaft
zu führen und als “rechte Hand“ des Trainers zu agieren. Unsere Haupthypothese lautet
demnach, dass von Knappheit der Führungspersönlichkeiten auf dem „Arbeitsmarkt
Bundesliga“ die Kapitäne zusätzlich monetär vergütet werden. Wir gehen nun mit Hilfe
eines Datensatzes von 13 Saisons der ersten deutschen Bundesliga der Frage nach,
inwieweit diese Führungsrolle monetär entlohnt wird. Es zeigt sich das, für individuelle
Performance und Spielermerkmale kontrollierend, der Kapitän für seine
Führungsqualitäten mit einem Lohnaufschlag zwischen 25 und 66 Prozent vergütet
wird.
Die vorliegende Arbeit ist wie folgt aufgebaut: Im nächsten Abschnitt wird der Daten-
satz systematisch dargestellt. Im dritten Abschnitt werden die Modellschätzungen prä-
sentiert, während im vierten Abschnitt abschließend die Befunde unserer Untersuchung
diskutiert und mögliche Erweiterungen unserer Arbeit angesprochen werden.
2.2 Data Set and Descriptive Statistics
Die zentrale Fragestellung, der folgend nachgegangen wird, ist, ob und inwieweit die
Führungsposition des Mannschaftskapitäns zusätzlich monetär entlohnt wird. Zwecks
dieser Untersuchung verwenden wir einen Datensatz der deutschen ersten Fußballbun-
desliga. Dieser umfasst Leistungsdaten sowie individuelle Charakteristika der
Bundesligaspieler aus den Saisons 1995/96 bis 2007/08 und umfasst 1993 verschiedene
13Eine eigene Untersuchung zu diesem Thema findet sich im sechsten Kapitel der vorliegenden Arbeit.
11
Spieler und eine Gesamtzahl von 6147 Spieler-Jahresbeobachtungen. Die Leistungs-
daten der Bundesligaspieler wurden aus den jährlichen Sonderausgaben des Sport-
magazins „Kicker“ entnommen, während Informationen bezüglich der Mannschafts-
kapitäne aus Artikeln des zweimal wöchentlich erscheinenden „Kicker- Sportmagazins“
zusammengetragen wurden. Nachfolgend wird derjenige als Kapitän einer Mannschaft
betrachtet, der zu Beginn einer jeweiligen Saison dieses Amt übertragen bekommt. Bei
einem Beobachtungszeitraum von 13 Saisons und einer konstanten Anzahl von 18 Ver-
einen ergibt sich somit eine Gesamtzahl von 234 Kapitänsjahren. Wie bereits
angesprochen, wird vom Kapitän, welcher durch das Tragen einer Oberarmbinde
optisch von seinen Mitspielern zu unterscheiden ist, die Übernahme von
Führungsausgaben erwartet. Während ihm aufgrund seiner Position nicht mehr Rechte
zustehen, so ist er dennoch mit mehr Pflichten betreut. So obliegt ihm beispielsweise die
Pflicht, als Ansprechpartner des Schiedsrichters eventuelles Fehlverhalten seiner
Mannschaft abzustellen. Im Beobachtungszeitraum waren 123 verschiedene Spieler
Kapitän einer Mannschaft, was einer durchschnittlichen Amtsdauer von 1,90 Saisons
entspricht. Hierbei ist eine rechtsschiefe Verteilung der Amtsdauer zu beobachten, da 70
Kapitäne ihr Amt für lediglich eine Saison ausübten. Mit Stefan Effenberg gibt es auch
einen Spieler in der Bundesliga, der im relevanten Zeitraum bei zwei verschiedenen
Vereinen Kapitän war, und zwar bei Borussia Mönchengladbach und bei Bayern
München. Insgesamt können 14 Spieler eine Amtsdauer von mehr als drei Saisons
verzeichnen, wobei es keine „Mindestzugehörigkeit“ zum Verein zu scheinen gibt. Alle
diese Spieler können zum Zeitpunkt ihrer Berufung jedoch eine Mindesterfahrung von
vier Saisons im deutschen Profifußball aufweisen. Eine Übersicht der Kapitäne mit den
längsten Amtsdauern im Beobachtungszeitraum bietet Table 2-1.
12
NameVereinKapitän
ab
Jahre
Kapitän
Jahreim
Vereinbei
Erstberufung
Profisaisons
bei
Erstberufung
FrankBaumannWerderBremen1999812
JensNowotnyBayerLeverkusen1996805
OliverKahnBayernMünchen20017713
StefanReuterBorussiaDortmund19976513
ZvonimirSoldoVFBStuttgart2000644
TomaszWaldochSchalke041999505
DariuszWoszVFLBochum20025112
ArneFriedrichHerthaBSCBerlin2004425
MarcoKurz1860München1998409
AltinLalaHannover962003455
MichaelPreetzHerthaBSCBerlin19984212
OlafThonSchalke0419954112
ThorstenWohlertMSVDuisburg1996439
ChristianWörnsBorussiaDortmund20044515
Table 2-1: Captains in the German Bundesliga with Three or more Seasons of Experience
Neben der, für die vorliegende Untersuchung entscheidenden, Variablen des Kapitäns-
amtes hängen die Spielergehälter erwartungsgemäß auch von einer Reihe weiterer
Einflussfaktoren und Spielercharakteristika ab (siehe Table 2-2). Im Hinblick auf die
Humankapitaltheorie wird im Folgenden zwischen drei verschiedenen
Spielermerkmalen unterschieden, welche eine Differenzierung hinsichtlich der
Erfahrung der Individuen ermöglichen. Zunächst bildet das Alter der Spieler zu
Saisonbeginn einen Indikator für die Erfahrung. Zu erwarten sind - neben dem positiven
Einfluss von dem Alter auf das Gehalt – negative Grenzerträge des Alters aufgrund
nachlassender Leistungsfähigkeit. Aus diesem Grund wird auch das quadrierte Alter der
Individuen ermittelt und der Einfluss auf das Spielergehalt untersucht.
Als zwei weitere Indikatoren von Erfahrung werden zudem die vor Saisonbeginn absol-
vierten Spiele in der ersten Fußballbundesliga, sowie die A-Länderspiele für das jewei-
lige Herkunftsland der Spieler betrachtet.14 Die Unterscheidung ist bedeutsam, da sie
14 Vergleiche Lucifora und Simmons (2003), sowie Frick (2007b).
13
eine Differenzierung zwischen nationaler und internationaler Erfahrung ermöglicht.
Analog zu dem Alter-Gehalts-Profil sind auch hier jeweils umgekehrt U-förmige Ver-
läufe des Bundesligaspiel-Gehalts-Profils und des Länderspiel-Gehalts-Profils zu er-
warten.
Neben diesen, die Erfahrung eines Spielers abbildenden, Variablen wird mit den in der
Bundesliga erzielten Toren auch eine bedeutende Leistungsvariable berücksichtigt.
Hierbei sind die während der bisherigen Karriere erzielten Bundesligatore als Indikator
für die Offensivfähigkeiten der Spieler zu sehen. Da die in einem Spiel erzielten Tore
letztendlich den Ausgang des Spiels bestimmen, würde man einen positiven Einfluss
der individuell erzielten Tore in der ersten Fußballbundesliga auf das Spielergehalt er-
warten. Nicht unerwähnt bleiben sollte der Fakt, dass Tore vorwiegend von Stürmern
und Mittelfeldspielern erzielt werden.
Abgesehen von den obigen Erfahrungs- und Leistungsdaten kann auch die Spieler-
position, auf der die Spieler eingesetzt werden, Einfluss auf die Gehaltshöhe nehmen.
Daher wird zwischen den Positionen Torwart, Abwehr, Mittelfeld und Sturm unter-
schieden. Jede dieser Positionen stellt spezifische Anforderungen an die Bundesliga-
spieler und ein Wechsel zwischen verschiedenen Positionen innerhalb der Karriere ist
daher unüblich. Ein Wechsel von oder auf die Torwartposition war nicht zu beobachten;
bedingt durch das sich stark unterscheidende Anforderungsprofil an diese Position. In
dem untersuchten Zeitraum wechselten 241 Feldspieler die Positionen. Auf den Einfluss
von höherer Flexibilität auf das Spielergehalt wird zu einem späteren Zeitpunkt
einzugegangen.
14
VariableOperationalisierungMeanMin.Max.
AlterAlterzuSaisonbeginn26,391741
Alter²QuadriertesAlterzuSaisonbeginn714,752891681
BLSBundesligaspielevorSaisonbeginn55,810540
BLS²QuadrierteBundesligaspielevorSaisonbeginn9611,220291600
GPINTLänderspielevorSaisonbeginn7,540130
GPINT²QuadrierteLänderspielevorSaisonbeginn331,00016900
ToreBundesligatorevorSaisonbeginn6,340171
TorwartTorwart(Dummy;1=ja)0,1101
AbwehrAbwehrspieler(Dummy;1=ja)0,2801
MittelfeldMittelfeldspieler(Dummy;1=ja)0,3901
PositionSpielerwechselteimKarriereverlaufdiePosition
SturmStürmer(Dummy;1=ja)0,2201
Kapitän(t)SpielerinderaktuellenSaisonKapitän(Dummy;1=ja)0,0401
Kapitän(t‐1)SpielerinderletzenSaisonKapitän(Dummy;1=ja)0,0301
DeutschlandSpieleristdeutscherNationalität(Dummy;1=ja)0,5801
SüdamerikaSpielerstammtausSüdamerika(Dummy;1=ja)0,0501
NordamerikaSpielerstammtausNordamerika(Dummy;1=ja)0,0101
OsteuropaSpielerstammtausOsteuropa(Dummy;1=ja)0,1601
WesteuropaSpielerstammtausWesteuropa(Dummy;1=ja)0,1301
AfrikaSpielerstammtausAfrika(Dummy;1=ja)0,0501
Asien/Austr.SpielerstammtausAsien/Australien(Dummy;1=ja)0,0201
Table 2-2: Descriptive Statistics
Die abhängige Variable der vorliegenden Untersuchung, das individuelle Spielergehalt,
wird über die Marktwerte der Bundesligaspieler aus dem „Kicker Managerspiel“ appro-
ximiert. Bei diesem Managerspiel weist das Magazin jedem Spieler vor Beginn der
Bundesligasaison einen Marktwert zu. Die Nutzer des Spiels stehen vor der Aufgabe
unter gegebener Budgetrestriktion aus dem Pool aller Bundesligaspieler eine möglichst
gute Mannschaft zusammenzustellen, wobei die erreichte Punktzahl der eigenen Mann-
schaft von der Leistung der gewählten Spieler in der Fußballbundesliga abhängt. Das
approximierte Spielergehalt in der Saison t ermittelt sich wie folgt:
1,5
15
Es zeigt sich, im Abgleich mit den Informationen aus den Lizensierungsverfahren des
Deutschen Fußball Bundes beziehungsweise der Deutschen Fußball Liga, dass diese
Approximation hoch mit den öffentlich bekannten Gehältern korreliert und die durch-
schnittlichen Gehaltsaufwendungen der Bundesligisten gut abbildet. Nach der Darstel-
lung des der Regressionsanalyse zugrunde liegenden Datensatzes gilt es, im Rahmen
der folgenden Untersuchung der monetären Entlohnung des Kapitänsamtes nachzu-
gehen.
2.3 Empirical Results
Zunächst einmal wird auf die Verteilung der Gehälter in der ersten Fußballbundesliga
eingegangen, bevor die Einflussnahme der Kapitänsrolle auf das Spielergehalt analysiert
wird. Bei einer Gehaltsspanne zwischen 17.043 und 10.000.000 Euro pro Spielzeit ist
eine rechtsschiefe Verteilung der Gehälter zu beobachten (siehe Figure 2-1). Während
das Durchschnittsgehalt pro Saison bei 909.014 Euro liegt, lässt sich ein Mediangehalt
von 666.667 Euro ermitteln. Diese Einkommensverteilung lässt sich mittels Rosens
(1981, 1983) „Theorie der Superstars“ erklären, nach der schon geringe Talentunter-
schiede zu erheblichen Einkommensunterschieden führen.15
15 In ihrer Untersuchung zeigen Lehmann und Schulze (2008), dass die Medienpräsenz von deutschen
Fußballprofis einem statistisch signifikant positiven Einfluss auf Spielerhälter nimmt, allerdings mit
abnehmendem Grenznutzen. Es scheint schwer, dieses Ergebnis mit der “Theorie der Superstars” von
Rosen in Einklang zu bringen.
16
Figure 2-1: Kernel Density Estimation of the Salary in the German Bundesliga
Nach dieser sehr allgemeinen Analyse der Verteilung der Spielergehälter werden nun
die vermuteten Einkommensdeterminanten der individuellen Spielergehälter untersucht.
Mit Hilfe verschiedener Regressionsverfahren wird hierfür im Folgenden versucht, die
Varianz der Gehälter zu erklären. Ziel des Hauptteils der vorliegenden Untersuchung ist
es der Frage nachzugehen, welchen pekuniären Einfluss die Führungsqualitäten der
Bundesligafußballer, abgebildet durch das Besetzen des Kapitänsamtes, haben. Dies
erfolgt durch eine Schätzung der Einkommensfunktion nach Mincer (1974). Aufgrund
des Panelcharakters der vorliegenden Daten führen wir neben der OLS-Schätzung auch
eine Random Effects-Schätzung durch, welche die Inkludierung von zeitunabhängigen
Variablen erlaubt. Zudem berücksichtigt letztere Schätzung, dass die gegebenen indivi-
duellen Effekte auch auf einer Vielzahl zusätzlicher, nicht beobachtbarer oder zufälliger
Variablen beruhen können.16 Die Verwendung von einer Fixed Effects-Schätzung
16 Vergleiche Mátyás und Sevestre (1996, S. 94).
0.2 .4 .6
Dichte
0 5 10 15
Gehalt (in Millionen €)
Kerndichteschätzung
Normalverteilung
17
scheint im vorliegenden Kontext wenig angebracht, da beispielsweise die erklärenden
Variablen der regionalen Herkunft zeitinvariant sind.17
Unter der Annahme, dass sich das Gehalt aus verschiedenen Faktoren multiplikativ
zusammensetzt, führt ein Logarithmieren - unter Berücksichtigung der vermuteten
Einkommensfaktoren des Gehaltes – für die folgende Modellschätzung zu der allgemei-
nen Form:
ln
ä
Als Gehalt wird hierbei der aus dem Managerspiel approximierte Wert verwendet. Das
Alter der Spieler wird zu Saisonbeginn in Jahren gemessen, während bei den absol-
vierten Bundesligaspielen (BLS) und Länderspielen (LS), sowie den erzielten Bundes-
ligatoren (Tore) jeweils auf die ausgewiesenen Karriereleistungen vor Beginn des ersten
Spieltages der jeweiligen Saison verwiesen wird. Um weitere mögliche Einflussfaktoren
bei der Gehaltsdeterminierung abzubilden, werden Dummy-Variablen für die verschie-
denen Vereine, die Saisons und die Herkunft der Spieler inkludiert. Die Vereins-
dummies (TD) bildet hierbei die, persistent vorherrschenden, unterschiedlichen finan-
ziellen Möglichkeiten ab, welche auch Auswirkungen auf die gezahlten Gehälter bei
den verschiedenen Vereinen haben dürften. Mit Hilfe der Saisondummies (SD) wird
versucht, die allgemeine Gehaltsentwicklung in der Fußballbundesliga über die
verschiedenen Spielzeiten abzubilden. Vergleicht man das Durchschnittsgehalt der Sai-
son 1995/1996 (565.681 Euro) mit dem der Saison 2007/2008 (1.288.702 Euro), so lässt
sich ein starker Anstieg der Entlohnung von Bundesligaprofis erkennen, welcher die
Verwendung eines Saison-Dummies in die Modellschätzung rechtfertigt. Eine grafische
Darstellung der Gehaltsentwicklung unter Berücksichtigung der Position auf dem Spiel-
feld bietet Figure 2-2. Hier lässt sich eine positive Gehaltsentwicklung, unabhängig von
17 Zudem sind die Koeffizienten des Fixed-Effects Modells, insbesondere die der Kapitänsvariablen, sehr
ähnlich, sodass uns die Verwendung des Random Effects-Modells vertretbar erscheint.
18
der Position erkennen. Über den gesamten Beobachtungszeitraum hinweg werden
Spieler umso besser bezahlt, je offensiver ihre Position auf dem Spielfeld ist.
Figure 2-2: Salary History of Players in the Bundesliga Subject to the Positions on the Field
Durch das Einschließen von Informationen über die Herkunft der Spieler (HD) wird
eine mögliche Diskriminierung im Sinne von geringeren Lohnzahlungen berücksichtigt.
Nach der Liberalisierung des Spielermarktes im Profifußball im Rahmen des „Bosman-
Urteils“ aus dem Jahr 1995 gilt es zu prüfen, welchen Einfluss die Nationalität auf das
Gehalt nimmt.18 Bei der Modellschätzung wird daher nach Herkunft der Spieler zwi-
schen Deutschland, Westeuropa, Osteuropa, Afrika, Südamerika, Nordamerika und
Australien bzw. Asien unterschieden.19
Ein möglicherweise vorliegendes Kausalitätsproblem bezüglich der Kapitänsvariablen
und den Spielergehältern ist nicht zu vernachlässigen. Es wäre zu befürchten, dass die
Gehaltshöhe die Wahrscheinlichkeit der Übernahme des Kapitänsamtes beeinflusst und
bevorzugt besser verdienende Spieler zum Kapitän berufen werden. Um dieser Proble-
18 Vergleiche Antonioni und Cubbin (2000), Frick und Wagner (1996) und Frick (2008, 2009).
19 Einen Überblick über die Literatur bezüglich der ethnischen Diskriminierung von Profisportlern bietet
Kahn (1991). Darüber hinaus liefert Kalter (1999) Befunde für die deutsche Fußballbundesliga.
0€
200.000€
400.000€
600.000€
800.000€
1.000.000€
1.200.000€
1.400.000€
1.600.000€
1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007
Torwart Abwehr Mittelfeld Sturm
19
matik zu begegnen, wird die Schätzung des Lohns in Modell 2 nicht unter Berück-
sichtigung der aktuellen Besetzung (Saison t) des Kapitänspostens durchgeführt, son-
dern unter der Fragestellung, ob der Spieler in der vorherigen Spielzeit (Saison t-1)
Kapitän war. Hierdurch verringert sich die Anzahl der Spieler mit dem Merkmal Kapi-
tän von 234 auf 169. Diese Reduktion lässt sich dadurch erklären, dass im vorliegenden
Datensatz 65 Bundesligakapitäne der Vorsaison entweder in einer unterklassigen deut-
schen Liga oder im Ausland gespielt haben.
Modell1Modell2
KoeffizientenKoeffizienten
VariableOLSREOLSRE
Alter0,659(24,77)***0,690(25,09)***0,657(24,55)***0,729(28,81)***
Alter²‐0,012(‐27,49)*** ‐0,013(‐27,20)*** ‐0,012(‐27,19)***‐0,013(‐27,02)***
BLS0,004(11,67)***0,002(4,60)***0,004(11,63)***0,002(4,85)***
BLS²‐0,000(‐8,63)***‐0,000(‐4,01)*** ‐0,000(‐8,46)***‐0,000(‐4,06)***
GPINT0,017(11,44)***0,012(6,59)***0,018(11,86)***0,012(6,92)***
GPINT²‐0,000(‐6,93)***‐0,000(‐4,11)*** ‐0,000(‐7,21)***‐0,000(‐4,27)***
Tore0,002(2,15)**‐0,000(‐0,21)+0,002(2,02)**‐0,000(0,28)+
Abwehr0,173(5,28)***0,307(6,39)***0,175(5,29)***0,297(6,14)***
Mittelfeld0,278(8,65)***0,388(8,23)***0,277(8,58)***0,385(8,12)***
Sturm0,425(11,80)***0,489(9,57)***0,420(11,59)***0,479(9,34)***
Südamerika0,565(13,01)***0,616(9,66)***0,560(12,82)***0,619(9,65)***
Nordamerika‐0,125(‐1,27)+‐0,045(‐0,34)+‐0,111(‐1,13)+0,002(0,02)+
Osteuropa0,150(5,30)***0,181(4,52)***0,139(4,92)***0,170(4,22)***
Westeuropa0,287(9,29)***0,347(8,20)***0,285(9,19)***0,347(8,17)***
Afrika0,059(1,39)+0,169(2,75)***0,047(1,09)+0,155(2,51)**
Asien/Austr.0,079(1,30)+0,159(1,86)*0,068(1,11)+0,139(1,62)+
Kapitän(t)0,511(10,56)***0,396(8,64)***//
Kapitän(t‐1)//0,369(6,15)***0,224(4,05)***
Saison‐DummiesBerücksichtigt
Team‐DummiesBerücksichtigt
AdjR²0,4770,4560,4700,449
Fallzahl61476147
*p<0,10; **p<0,05; ***p<0,01; + n. s. (t-Werte bzw. z-Werte in Klammern).
Table 2-3: Determinants of Player Salary in the German Bundesliga
20
Die Ergebnisse der Regressionsschätzungen werden in Table 2-3 präsentiert. Hier lässt
sich das bereits vermutete umgekehrt U-förmige Alters-Einkommens-Profil erkennen.
Analog hierzu ist jeweils ein positiver Einfluss der Partizipation an Bundesliga- und
Länderspielen, bei gleichzeitig negativem Grenznutzen zu beobachten. Der Einfluss von
Erfahrung auf das Salär von Bundesligaprofis ist also sowohl auf nationaler, als auch
auf internationaler Ebene erwartungsgemäß, wobei die Einsätze bei Länderspielen einen
größeren Einfluss auf das Gehalt haben als Bundesligaspiele.20 Betrachtet man den Ein-
fluss der Position der Spieler auf das Gehalt, so fällt auf, dass Feldspieler, im Vergleich
zur Referenzposition Torwart, signifikant mehr verdienen. Dies lässt sich möglicher-
weise durch die, bereits angesprochene, geringe Flexibilität der Torwarte hinsichtlich
ihres Einsatzes auf anderen Positionen erklären sowie über die Berücksichtigung der
Variablen “Tore“. Hierüber hinaus wird die Bedeutung der Herkunft der Spieler für das
Gehalt deutlich. Im Vergleich zum Referenzland Deutschland verdienen Spieler aus
Südamerika, Osteuropa und Westeuropa signifikant mehr.
Um der zentralen Fragestellung, der monetären Entlohnung von Führungsqualitäten,
nachzugehen, betrachten wir die Ausprägung der beiden Kapitänsvariablen. Als Haupt-
ergebnis unserer Arbeit stellen wir fest, dass die Ausübung des Kapitänsamtes einen
signifikant positiven Einfluss auf das Spielergehalt hat. Halvorsen und Palmquist (1980)
zeigen, dass bei semilogarithmischen Gleichungen der Koeffizient der Dummy-Vari-
ablen nicht als prozentualer Einfluss interpretiert werden darf. Stattdessen sei der
prozentuale Einfluss als 100*(exp(K)-1) zu berechnen, wobei K der Koeffizient der
Dummy-Variablen ist. Hiernach erhält der Mannschaftskapitän, je nach Schätzmethode,
ein Surplus zwischen 25,61 und 66,70 Prozent. Dies bestätigt unsere eingangs formu-
lierte Hypothese, dass auch im professionellen Teamsport Führungsqualitäten pekuniär
entlohnt werden. Einem eventuellen Endogenitätsproblem der Kapitänsvariablen kann
durch eine analoge Regressionsanalyse, jetzt unter Berücksichtigung des Kapitäns des
Vorjahres, Rechnung getragen werden (Modell 2). Auch im Rahmen dieser zweiten
Modellschätzung kann der erwartete Einfluss des Kapitänsamtes auf das Spielergehalt
nachgewiesen werden. Dieser ist abermals hochsignifikant positiv und spricht dem
20Im neunten Kapitel wird hierauf detailliert eingegangen.
21
Kapitän einen Einkommensbonus zwischen 25,11 und 43,62 Prozent zu. Somit scheint
die Exogenität der hier zentralen Variablen Kapitän (t) gegeben.
Wie bereits angesprochen, wechselten im Beobachtungszeitraum viele Spieler die
Position. Es stellt sich hierzu die Frage, inwieweit diese Flexibilität Einfluss auf das
Spielergehalt nimmt. Daher wird für die folgende Analyse eine weitere Variable
generiert, die einen Positionswechsel im Karriereverlauf abbildet (Wechsel Position).
Für Table 2-4 wird die vorherige Regressionsgleichung um diese Variable erweitert. Es
zeigt sich, dass kein signifikanter Einfluss auf das Gehalt messbar ist. Dies kann den
Grund haben, dass ältere Spieler auf eine Position wechseln, die geringere
Anforderungen an die körperliche Verfassung stellt. Dies würde den erwarteten
Gehaltsbonus für die Flexibilität wettmachen.
22
Modell1Modell2
Koeffizienten Koeffizienten
VariabelOLSRE OLS RE
Alter0.660(24.77)***0.687(24.92)***0.658(24.55)***0.685(24.74)***
Alter²‐0.013(‐27.49)*** ‐0.013(‐27.03)*** ‐0.012(‐27.19)***‐0.013(‐26.80)***
BLS0.004(11.64)***0.002(4.35)***0.004(11.62)***0.002(4.39)***
BLS²‐0.000(‐8.66)***‐0.000(‐3.88)***‐0.000(‐8.50)***‐0.000(‐3.87)***
GPINT0.017(11.42)***0.012(6.63)***0.018(11.84)***0.012(6.96)***
GPINT²‐0.000(‐6.90)***‐0.000(‐4.16)***‐0.000(‐7.18)***‐0.000(‐4.41)***
Tore0.002(2.12)**‐0.000(‐0.14)+0.001(1.97)**‐0.000(‐0.25)+
Abwehr0.176(5.32)***0.299(6.20)***0.179(5.36)***0.299(6.15)***
Mittelfeld0.280(8.67)***0.385(8.17)***0.281(8.62)***0.385(8.13)***
Sturm0.427(11.82)***0.487(9.53)***0.422(11.62)***0.482(9.38)***
Südamerika0.564(12.98)***0.618(9.69)***0.559(12.79)***0.613(9.56)***
Nordamerika‐0.126(‐1.28)+‐0.043(‐0.32)+‐0.113(‐1.14)+‐0.030(‐0.22)+
Osteuropa0.148(5.27)***0.182(4.55)***0.138(4.88)***0.172(4.28)***
Westeuropa0.287(9.29)***0.348(8.22)***0.285(9.15)***0.345(8.12)***
Afrika0.060(1.39)+0.168(2.73)***0.047(1.09)+0.159(2.56)**
Asien/Austr.0.080(1.31)+0.158(1.85)*0.069(1.13)+0.150(1.75)*
WechselPos.‐0.020(‐0.69)+0.051(1.51)+‐0.026(‐0.86)+0.046(1.36)+
Kapitän(t)0.510(10.53)***0.398(8.67)***//
Kapitän(t‐1)//0.368(6.12)***0.229(4.14)***
Saison‐DummiesBerücksichtigt
Team‐DummiesBerücksichtigt
AdjR²0.4770.4560.470.449
Fallzahl61476147
*p<0,10; **p<0,05; ***p<0,01; + n. s. (t-Werte bzw. z-Werte in Klammern).
Table 2-4: Determinants of Player Salary in the German Bundesliga Including Flexibility
Aufgrund der linkssteilen und rechtsschiefen Verteilung der Spielergehälter wird in
vielen aktuellen Arbeiten bei der Gehaltsschätzung auch auf Quantilsregressionen
zurückgegriffen.21 Auch für den vorliegenden Datensatz lässt sich eine
Normalverteilung der Gehaltsvariablen widerlegen. Schätzt man nun für die beiden
Modellspezifikationen Quantilsregressionen (0,10; 0,25; 0,50; 0,75; 0,90 Quantil) um
den Einfluss der exogenen Variablen in verschiedenen Bereichen der Lohnverteilung zu
messen, so fällt auf, dass die Ergebnisse der Einkommensschätzungen denen der
21 Vergleiche Berri und Simmons (2009), Simmons und Berri (2009) und Vincent und Eastman (2009) für
aktuelle Untersuchungen für den nordamerikanischen Teamsport.
23
Random Effects-Schätzungen sehr ähnlich sind. Da die Koeffizienten, bis auf wenige
Ausnahmen, über die Perzentile sehr konstant sind, lässt sich schlussfolgern, dass der
Einfluss der erklärenden Variablen auf die Gehälter an verschiedenen Stellen der
Einkommensverteilung nahezu identisch ist. Gerade der Einfluss der Kapitänsvariablen
ist über alle Perzentile hinweg signifikant positiv und bestätigt die Vermutung, dass
Führungsqualitäten einen entscheidenden Einfluss auf die Spielergehälter nehmen. Auf
den ersten Blick verwundert der abnehmende Koeffizient des „Kapitän-Dummies“ über
die Perzentile. Hierfür bietet sich jedoch eine simple Erklärung an: Die Anzahl der
Spieler in den höchsten Perzentilen ist nicht über alle Teams gleichmäßig verteilt. Von
den 580 höchstbezahlten Spielern spielten 145 beim FC Bayern München. Da aber jeder
Verein jeweils einen Kapitän je Saison stellt, gibt es Kapitäne, die aufgrund der
Budgetrestriktionen der Vereine nicht zu den „Topverdienern“ der Liga gehören. Eine
Darstellung der Schätzergebnisse der Quantilsregressionen, jeweils für
Modellspezifikation 1 und Modellspezifikation 2, findet sich nachfolgend in Table 2-5
und Table 2-6.22
22Die Ergebnisse der Bootstrap und Jackknife Methode sind analog.
24
Variable0,1Quantil0,25Quantil 0,5Quantil0,75Quantil0,9Quantil
Alter,7721***,8306*** ,7058*** ,5726***,4362***
Alter2‐,0137***‐,0148*** ‐,0127*** ‐,0105***‐,0081***
BLS,0053***,0045*** ,0038*** ,0028***,0017***
BLS²‐,0000***‐,0000*** ‐,0000*** ‐,0000***‐,0000**
GPINT,0205***,0155*** ,0155*** ,0163***,0173***
GPINT²‐,0002***‐,0001*** ‐,0001*** ‐,0001***‐,0001***
Tore,0007+,0035*** ,0039*** ,0027***,0028**
Abwehr,3674***,3167*** ,1814*** ‐,0105+‐,1683***
Mittelfeld,4939***,3718*** ,2643*** ,1282***,0005+
Sturm,5957***,4754*** ,3918*** ,2872***,1594***
Südamerika,5142***,5330*** ,5262*** ,5200***,5013***
Nordamerika‐,0826+‐,0538+‐,0019+‐,0967+‐,1804+
Osteuropa,1854***,1685*** ,1241*** ,1130***,0701+
Westeuropa,3451***,3367*** ,3096*** ,2100***,1627***
Afrika,0892+,1264** ,0438+‐,0066+‐,0054+
Asien/Austr.,2035**,1686** ,0699+‐,0773+,1510+
Kapitän(t),6252***,5700*** ,5168*** ,4410***,3161***
PseudoR²,3150,3114,2932,2879,2966
Fallzahl61476147614761476147
RawSumofDev.2.196,53.891,54.656,63.577,41.934,0
MinSumofDev.1.504,52.679,63.291,42.547,41.360,3
*p<0,10; **p<0,05; ***p<0,01; + n. s.
Table 2-5: Quantile Regressions of Player Salary in the Bundesliga (Model 1)
25
Variable0,1Quantil0,25Quantil 0,5Quantil0,75Quantil0,9Quantil
Alter,7774***,8455*** ,7033*** ,5416***,4618***
Alter2‐,0138***‐,0150*** ‐,0126*** ‐,0099***‐,0085***
BLS,0058***,0042*** ,0039*** ,0032***,0017***
BLS²‐,0000***‐,0000*** ‐,0000*** ‐,0000***‐,0000**
GPINT,0181***,0167*** ,0165*** ,0175***,0165***
GPINT²‐,0001***‐,0001*** ‐,0001*** ‐,0001***‐,0001***
Tore,0014+,0038*** ,0040*** ,0023**,0027**
Abwehr,3830***,3087*** ,1834*** ‐,0178+‐,1530***
Mittelfeld,4695***,3630*** ,2704*** ,1178***,0178+
Sturm,5746***,4566*** ,3860*** ,2713***,1753***
Südamerika,4913***,5054*** ,5108*** ,5317***,5003***
Nordamerika‐,0727+‐,0403+‐,0078+‐,0732+‐,0632+
Osteuropa,1794***,1555*** ,1268*** ,1090***,0556+
Westeuropa,3343***,3324*** ,3051*** ,2018***,1614***
Afrika,1093+,0996+,0320+‐,0169+‐,0017+
Asien/Austr.,2510**,1539*,0579+‐,0860+,1844+
Kapitän(t‐1),3786***,4751*** ,4402*** ,3293***,3063***
PseudoR²,3108,3071,2877,2817,2921
Fallzahl61476147614761476147
RawSumofDev.2.196,53.891,54.656,63.577,41.934,0
MinSumofDev.1.513,92.696,63.316,92.569,61.369,0
*p<0,10; **p<0,05; ***p<0,01; + n. s.
Table 2-6: Quantile Regressions of Player Salary in the Bundesliga (Model 2)
26
2.4 Summary and Implications
Das aktuelle Kapitel liefert empirische Evidenz für die monetäre Entlohnung von
Führungsqualitäten im professionellen Teamsport. Da sich der Aufsatz auf die deutsche
Fußballbundesliga beschränkt, ergeben sich eine Vielzahl von denkbaren weiter-
führenden Arbeiten. Hier wäre es zum Beispiel von Interesse, inwieweit sich der mone-
täre Einfluss des Kapitänsamtes mit der Mannschaftsgröße verändert. Aus organisa-
tionsökonomischer Sicht wäre zu erwarten, dass mit steigender Anzahl an Team-
mitgliedern das Gehaltssurplus für die Übernahme der Kapitänsposition aufgrund des
größeren Koordinationsaufwands steigt. Darüber hinaus könnte der Einfluss von ver-
schieden großen Führungsteams analysiert werden. Einige Sportarten, wie beispiels-
weise der US-amerikanische Basketball, weisen unterschiedliche Anzahlen von Kapi-
tänen je Team auf. Beim Basketball schwankt diese zwischen einem und drei Kapi-
tänen. Inwieweit diese Streuung der Führungsverantwortung einen monetären Einfluss
nimmt, wäre ein weiterer interessanter Ansatzpunkt für weitere Untersuchungen. Hier
ist zu erwarten, dass die Bündelung der Führungsaufgaben auf einen einzigen Kapitän
für diesen zu dem größten Gehaltsbonus führt.
27
3 The Payoff to Leadership in Teams
3.1 Introduction
The presence of a leader is often said to be decisive for the success of a team. While the
sum of individual skills of the team members might be a good indicator for the potential
of a group, a team leader is expected to improve his teams’ performance by directing his
team mates. This ability is assumed to be compensated monetarily. In professional
sports a team captain is said to have responsibility for the strategy as well as teamwork.
Since this requires leadership skills, the analysis of the monetary reward for captaincy is
the main subject of this chapter.
Looking at the existing literature one finds many articles explaining which factors influ-
ence the players’ salary in the National Hockey League (NHL), using data sets from
different seasons and including various kinds of performance indicators. In an early
study Jones and Walsh (1988) analyze the influence of players’ skills on their respective
salaries. They also focus on the influence of penalty minutes on players’ salary, as they
predict that players who play with a higher intensity, which results in a higher number
of penalty minutes, earn significantly more. By clustering players into the two groups
“grunts” and “non-grunts”, Jones, Nadeau and Walsh (1997) provide a rational for both
types of players to find employment in the National Hockey League, as a market for
both player types exists. The authors also show that the structure of salaries differs for
both groups of players, even through the main salary for both groups does not differ
statistically.
There also exists a large body of literature concerning salary discrimination against
French-Canadian players in the National Hockey League. Longley (1995), Lavoie
(2000) and others investigate on this subject, ending up with mixed findings about this
subject.23 Using a stochastic frontier approach, Kahane (2005) picks up this topic, and
estimates an optimal relative presence of French-Canadian players for a given payroll.
23 See Kahn (1991) for a general survey on discrimination in professional sports.
28
He shows greater inefficiencies for National Hockey League teams which have an over-
representation of Francophone players.
Regarding leadership skills, most research has been done concerning big company
managers, whose ability to lead others is often concerned to be a “soft skill”. Hence the
question arises, why some individuals develop leadership skills while others do not.
Kuhn and Weinberger (2005) investigate the influence of occupying leadership posi-
tions in high school on the later salary. They identify team captains and club presidents
as individuals who take leadership positions during high school and hereby accumulate
important leadership skills for their further career. While accounting for cognitive skills
as well as psychological and physical indicators on the individual basis, the authors
show that men who possessed one of the named positions during high school receive a
wage premium between 4 and 24 percent a decade later. In a related research Barron,
Ewing and Waddell (2000) show that students who participate in extracurricular sport
activities during high school years later earn a wage premium between 4 and 15 percent
over students, who participated in non-sport extracurricular activities.
To summarize, the literature offers different approaches to measure factors that influ-
ence players’ salary in team sports, especially for the National Hockey League. Addi-
tionally, a big body of literature is devoted to leadership skills, describing the develop-
ment of leadership skills. This chapter combines these two fields as it measures the
impact of leadership skills on players’ salaries. Since team captains are expected to pos-
sess this ability, the main goal of this chapter is to measure the wage premium for team
captains in the National Hockey League. Using a data set containing four successive
seasons I show that, other things equal, players are paid an extra 21 to 35 percent for
serving as team captains.
The chapter is organized as follows: The next section describes the data and presents the
goals of the chapter. Section three presents the results obtained from the analyses of the
data set. Finally, the chapter concludes with section four.
29
3.2 Data
The data set that I analyze in this chapter contains individual statistics of all ice hockey
players who played in the National Hockey League between the 2003/2004 and the
2007/2008 season. Due to the NHL lockout in 2004/2005 which resulted in the cancel-
lation of the entire season, our data set contains four seasons. Since salary information
is not available for all players, the data set totals in 1067 players and 2773 player-year-
observations. The NHL had 30 teams during the observed period with 82 regular season
games per team. This leads to 1230 games in each of the four seasons and therefore the
data set includes statistics from a total of 4920 regular season games. The vast majority
of teams have just one team captain, while in rare cases teams report up to five captains.
For the course of four seasons 148 team captains were reported. Our data set excludes
goalkeepers, since the official rules prohibit them to act as a team captains.24 Individual
player statistics were drawn from the leagues official website at http://www.nhl.com,
while player salaries were obtained from the website of USA Today at
http://content.usatoday.com/sports/hockey/nhl/salaries/default.aspx. Finally, informa-
tion concerning the team captains was taken from the “NHL Official Guide & Record
Book 2009”.
In the National Hockey League, the team captain is selected before the start of each sea-
son and has a capital “C” sewn on the left side of his jersey to distinguish him from his
teammates.25 He is the only player who is allowed to talk to the referees about the inter-
pretation of rules during the game. To minimize delay in case of disagreement about a
call, the league disallows goalkeepers to be team captains. Beside the named compe-
tency, the main task for the team captain is to be a leader during games as well as before
and after games in the locker room. Furthermore he is the one who is responsible to
represent his teammates’ concerns to the team management.
The primary task of this chapter is to analyze how this leadership, displayed by the
function of serving as team captain, is rewarded monetarily. Observing the years of
24 See official NHL rules, section 2, rule 6 at http://www.nhl.com/ext/0708rules.pdf.
25 NHL teams also name alternate captains each season, who carry a capital “A” on their jersey. Unfor-
tunately, information concerning alternate captains is unavailable.
30
experience team captains have, one can notice a nearly normal distribution around the
average of eleven years of experience (see Figure 3-1). Hence it seems that leadership
skills requires some experience in the league, as one observes an average experience of
only 5.55 years for non team captains.
Figure 3-1: Kernel Density Estimation of Captains and Non-Captains NHL Experience
Besides the captaincy, one has got to consider other factors which influence players’
salary. The present approach hence accounts for a variety of individual characteristics
presented in Table 3-1. Players’ experience is expected to have a positive impact
according to the human capital theory. Therefore I consider years played in the National
Hockey League prior to the respective season to display players’ experience in the
league. Human capital theory lets us expect a positive impact of experience on salary
accompanied with decreasing marginal returns, which would result in an upward-slop-
ing experience-earnings profile with diminishing returns to experience. Accordingly,
one also has to consider the squared experience of the players.26 Especially in a fast-
26 See also Idson and Kahane (2000).
0.05 .1 .15
Density
0 5 10 15 20 25
Years of NHL experience
Captains Noncaptains
31
paced game like hockey decreasing speed and agility can be hardly compensated by
additional experience.27
All-star game appearances seem to be an appropriate indicator to display players’ tal-
ents. Participating in this yearly event, which had been cancelled three times during the
last 20 years due to a lockout or to simultaneously proceeding Winter Olympic Games,
clearly allows distinguishing between talent levels. This differentiation proves to be
quite important, since superstars have a big impact on media attention and hence might
influence their franchises value.28
The ability to occupy a certain position might also affect the individuals’ salary. Hence I
differentiate between four positions as I exclude goalkeepers because of aforementioned
reasons. I control for the positions defenseman (DE), center (CE), right wing (RW) and
left wing (LW). Throughout the observed period players typically do not switch their
position. Altogether, it occurred just once that a player switched positions in between
seasons.
Next to variables accounting for players’ experience, talent and position, one also has to
consider performance on the ice as an important factor which influences players’ salary.
Consequently, several individual statistics are included in the following salary determi-
nation. Considering games played during the regular season accounts for two things:
First, it illustrates how important a player is to his teams’ success. Secondly, it displays
if a player is injury prone, since this would lead to a cutback in games played. Further-
more, points scored per game serves as an indicator for the offensive ability of players.
Following Lavoie (2000) and others, the squared term of the achieved points per game
is also included in the salary regression.
27 See Fair (1994).
28 For empirical evidence from the National Basketball Association (NBA) see Hausman and Leonard
(1997).
32
VariableOperationalizationMeanMin.Max.
ExpExperience(seasons)intheleague5.83024
Exp²Squaredexperience(seasons)intheleague55.310576
ASGAll‐stargameappearances0.45015
ASG²SquaredAll‐stargameappearances2.580225
DEDefender(dummy;yes=1)0.3501
CECenter(dummy;yes=1)0.2601
RWRightwing(dummy;yes=1)0.2001
LWLeftwing(dummy;yes=1)0.1901
GPGamesplayed61.36184
PPGPointspergame0.3901.6
PPG²Squaredpointspergame0.2402.4
Captain(t)Captainincurrentseason(dummy;yes=1)0.0501
Captain(t‐1)Captaininpriorseason(dummy;yes=1)0.0401
Table 3-1: Descriptive Statistics of Player Characteristics and Performance Indicators
3.3 Empirical Analyses
To start off this chapter, I am going to analyze the salary structure of the National
Hockey League. While the average salary for the observed period is 1,609,947 dollars
one observes a right-skewed distribution of players’ salary as the mean salary is compa-
ratively low at 945,630 dollars (see Figure 3-2). Individual players’ salaries range from
150,000 up to 13,500,000 dollars for one season. In the following regression analyses
the natural logarithm of the salary is used as the dependent variable, as this is a better
approximation to normal distribution.
Since the selected captains are expected to be players who spend a lot of minutes on the
ice, their salaries range from 570,000 to 9,880,939 dollars, with an average salary of
3,903,226 dollars and a mean salary of 4,000,000 dollars. Hence the distribution of
captain salaries is not as right-skewed as for the whole sample.
33
Figure 3-2: Kernel Density Estimation of the Salary in the NHL
As mentioned before, the lockout resulted in the cancellation of the 2004-05 NHL sea-
son, because the team owners and the NHL Players Association (NHLPA) did not con-
sent on a new collective bargaining agreement. Subsequent to both sides agreeing to a
new collective bargaining agreement during the summer of 2005, the total salary
spending decreased considerably for the following season. Due to an agreed increase of
the salary cap for the following seasons the average players’ salary rose afterwards.
Following this rather general analysis of the players’ salary I turn to the investigation of
the impact of leadership on players’ salary. For this purpose different regression models
are being run, which also regard the individual characteristics and performance indica-
tors described in the previous section. Hereby I distinguish the impact of leadership on
our dependent variable, denoted by the players’ log wage rate, from other influencing
factors. On the basis of the standard Mincer (1974) wage equation the following equa-
tion to estimate the impact on salary is being suggested:
0.2 .4 .6 .8
Density
0 5 10 15
Salary (in million $)
Kernel density estimate
Normal density
34
²
Due to the panel character of our data I apply a random effects model as well as a con-
ventional OLS model. The random effects model accounts for unobservable factors
which might influence the given individual effects.29 One might argue that there could
be an endogeneity problem, since the direction of causality between the dependent sal-
ary variable and the independent captain variable is not clear, as players might be
appointed to be captains because they receive a large salary. In this regard a second
regression analysis is being run, replacing the current team captains by the captains of
the prior season. Hence the information on team captains in the 2002/03 season is
included in the second model that uses salary information for 2003/04. Thus, the regres-
sion analysis for the 2003/04 season includes the captaincy information from the
2002/03 season (and so forth).
To account for further influencing factors, season dummies (SD) as well as team dum-
mies (TD) are included in both models. The season dummies account for the discon-
tinuous salary history within the league. As stated earlier, salaries dropped considerably
after the lockout season and this is reflected by the season dummies. Furthermore, the
team dummies depict prevailing differences in financial power. Teams’ spending on
salary ranged from 18,932,830 dollar by the Washington Capitals for the 2005/06 sea-
son to 77,856,109 dollar by the Detroit Red Wings for the 2003/04 campaign, display-
ing this heterogeneity between teams.
29 See Mátyás / Sevestre (1996, p. 94).
35
Model1Model2
CoefficientsCoefficients
VariableOLSREOLSRE
Exp0.143(21.81)***0.164(21.59)***0.143(21.73)***0.164(21.63)***
Exp²‐0.006(‐13.85)***‐0.008(‐15.26)*** ‐0.006(‐13.82)*** ‐0.008(‐15.36)***
ASG0.236(13.34)***0.280(12.04)***0.238(13.49)***0.283(12.21)***
ASG²‐0.017(‐8.89)***‐0.017(‐6.97)***‐0.017(‐9.02)***‐0.017(‐7.13)***
DE0.266(10.60)***0.164(4.89)***0.271(10.78)***0.168(4.99)***
RW‐0.036(‐1.34)+‐0.062(‐1.67)*‐0.031(‐1.14)+‐0.058(‐1.57)+
LW‐0.021(‐0.77)+‐0.052(‐1.40)+‐0.020(‐0.74)+‐0.051(‐1.37)+
GP0.002(3.47)***0.001(1.09)+0.002(3.68)***0.001(1.29)+
PPG1.492(13.38)***1.224(10.48)***1.517(13.60)***1.253(10.75)***
PPG²‐0.279(‐2.91)***‐0.330(‐3.34)***‐0.297(‐3.10)***‐0.361(‐3.66)***
Captain(t)0.314(7.04)***0.211(4.74)***//
Captain(t‐1)//0.355(7.37)***0.278(5.83)***
SeasonDum.Included
TeamDum.Included
AdjR²0.6240.6130.6250.614
Numberofobs2773277327732773
***, ** and * denote statistical significance at the 0.01, 0.05 and 0.1 level, + denotes insignificance (z-values in brackets).
Table 3-2: Determinants of Player Salary in the NHL
Estimations are reported in Table 3-2. As one would expect the data goes in line with
the human capital theory, as the experience-salary profile is upward sloping and con-
cave (see Figure 3-3).
36
Figure 3-3: Experience-Salary Profile
The number of All-star game appearances enables one to separate between very good
players and so-called star players. According to Rosen’s (1981, 1983) theory of
superstars even a marginal difference in talent leads to a considerable difference in
salary. The squared number of All-star game appearances is also considered, as one ex-
pects decreasing marginal returns of appearances. Data supports the expected influence
of All-star game experience, independent of the chosen model.
As I take the center position as the reference position there is no significant difference
in salary between it and the other offensive positions right wing and left wing. Other
things equal, defensemen earn significantly more than offensive players. This goes in
line with Idson and Kahane (2000), who state that a defensemen with the same offen-
sive skills as an offensive player earns more due to his additional defensive abilities.
Looking at the individual performance data, one finds the expected influence on the
players’ salary. Independent of the selected model, the number of games played during
the regular season have a significant positive impact on the salary, as it displays players’
0 1 2 3 4
Salary (in million US-Dollar)
0 5 10 15 20
Years of NHL experience
37
skills as well as physical characteristics like fatigue and frequency of injuries. Servicing
as offensive skill indicators, points per game and squared points per game both have a
highly significant impact on players’ salary. While points scored per game have a posi-
tive effect on players’ salary, one can observe decreasing marginal returns.
To test the main hypothesis which claims that leadership skills in teams are rewarded
monetarily, I look at the impact of the captain variable on players’ salary. All regression
analyses support the thesis that leadership is indeed rewarded, since the impact of the
captain variables is significantly positive for all models. As the main result of the thesis
I state that the teams’ current captain, other things equal, is rewarded with a monetary
bonus between 20 and 30 percent (Model 1) for obtaining his position.30 To accommo-
date for the potential heterogeneity problem of the captain variable I take a look at the
team captains of the previous season (Model 2). As expected this does not change the
result, since the impact of this captain variable is also highly significant.
Since salaries in the NHL are not normally distributed which is supported by a
skewness-kurtosis test for normality, I continue by performing quantile regressions for
both models. This way, one can measure the impact of the independent variables at
different parts of the salary distribution. As tables 3-3 and 3-4 report, coefficients are
relatively constant throughout the quantiles. As in the previous chapter, the impact of
the variable depicting the team captain decreases by the quantile. The argument goes in
line with the one presented in the prior chapter. The number of players belonging to the
highest quantiles is not equally distributed over all teams, but since every team has to
name a team captain the impact of captain variable decreases over the quantiles.
30Following the line of argumentation by Halvorsen and Palmquist (1980) as in the second chapter, the
surplus ranges between 23.5 and 36.9 percent.
38
Variable0.1Quantile0.25Quantile 0.5Quantile 0.75Quantile0.9Quantile
Exp.0661***.1124*** .1437*** .1548***.1522***
Exp²‐.0032***‐.0048*** ‐.0057*** ‐.0056***‐.0056***
ASG.4058***.3284*** .2742*** .2116***.1604***
ASG²‐.0378***‐.0264*** ‐.0214*** ‐.0146***‐.0100***
DE.0704***.1874*** .3032*** .3309***.3518***
RW‐.0069+‐.0091+‐.0102+‐.0810**‐.0330+
LW.0245+.0049+.0280+.0851**.0401+
GP.0020***.0028*** .0016*** .0009+‐.0004+
PPG.5698***1.1247*** 1.4418*** 1.6098***1.9771***
PPG²‐.0387+‐.0875+‐.2249*‐.3113**‐.4847***
Captain(t).4950***.4180*** .2584*** .1783***.1965***
PseudoR²0.20330.31670.41840.47630.4773
Fallzahl27732773277327732773
RawSumofDev.533.91200.41792.31566.5856.0
MinSumofDev.425.3820.31042.4820.5447.4
*p<0,10; **p<0,05; ***p<0,01; + n. s.
Table 3-3: Quantile Regressions of Player Salary in the NHL (Model 1)
39
Variable0.1Quantile0.25Quantile 0.5Quantile 0.75Quantile0.9Quantile
Exp.0649***.1125*** .1416*** .1580***.1524***
Exp²‐.0031***‐.0048*** ‐.0055*** ‐.0058***‐.0057***
ASG.4063***.3282*** .2596*** .2056***.1744***
ASG²‐.0389***‐.0265*** ‐.0192*** ‐.0142***‐.0112***
DE.0674**.1968*** .3104*** .3153***.3686***
RW.0075+‐.0053+.0036+‐.0833**‐.0173+
LW.0097+.0034+‐.0218+‐.0835**‐.0392+
GP.0020***.0027*** .0015** .0001+‐.0002+
PPG.6193***1.1680*** 1.5120*** 1.6081***1.930***
PPG²‐.0088+‐.1012+‐.2786** ‐.3145**‐.4349***
Captain(t‐1).5729***.4229*** .3306*** .2167***.2863***
PseudoR²0.19790.31610.41820.47670.4788
Fallzahl27732773277327732773
RawSumofDev.533.91200.41792.31566.5856
MinSumofDev.428.2821.01042.8819.7446.2
*p<0,10; **p<0,05; ***p<0,01; + n. s.
Table 3-4: Quantile Regressions of Player Salary in the NHL (Model 2)
3.4 Conclusions
This article has investigated the impact of leadership in teams on individuals’ salary.
The empirical work has focused on the National Hockey League. Using data from four
seasons it was shown that leadership skills, displayed by the captain variable, have a
significant positive impact on players’ salary. Depending on the chosen model, the
pecuniary surplus for the team captains range between 21 and 35 percent. Further
research might concentrate on other sports, possibly with a different team size and
varying performance indicators. Theory of organization would lead us to the expectation
of rising benefit of leadership as team size increases. A differentiation between roster
size and actual number of players on the field might be necessary to distinguish between
leadership on the field and leadership off the field. Next to this, further research on per-
sonal background of players might indicate under which circumstances individuals
develop into leaders. In this content further information on players, concerning origin,
education and earlier performance would be of great interest.
40
4 Performance under Pressure: Estimating the Returns to Mental
Strength in Professional Basketball
4.1 Introduction
Human capital theory as developed by Gary S. Becker (1962, 1964) is clearly one of the
cornerstones of modern labor and personnel economics. Since the term “human capital”
refers to the stock of knowledge and expertise an individual brings to a job, any per-
sonal attributes and characteristics that are rewarded in a competitive labor market must
be “skills”. Building on this concept, Jacob Mincer (1974) a few years later developed
and estimated what is now called the “earnings function”. Here, an individual’s earnings
are modeled as a function of her formal qualification, i.e. the years of schooling and
experience. Since the publication dates of these two seminal contributions, hundreds of
studies using data from different countries and/or time periods have appeared, each
seeking to identify the impact of age and highest school degree obtained on hourly,
monthly or annual wages (for an overview see e.g. Card 1999).31 Over the years, other
(potential) determinants of an individual’s wage have been added to the regression
models: Apart from schooling and experience most studies now include variables like
tenure with the current employer and the duration of non-employment spells, both of
which are likely to affect an individual’s stock of knowledge and expertise, too.
In two recent contributions Bowles, Gintis and Osborne (2001a, 2001b) have challenged
the traditional view by arguing that the “standard” human capital variables explain little
of the observable variance in earnings. Moreover, they find that variables that have been
omitted so far in the empirical literature very often prove to be important determinants
of an individual’s labor market success. Most important among the omitted variables are
“non-cognitive skills”, which are usually defined and measured in terms of work habits
31 Most of the more recent studies include additional variables that are considered valid measures of an
individual’s productivity (see e.g. Krueger 1993, DiNardo and Pischke 1997). Moreover, the adequacy
of the initial specification has been questioned in a number of studies. Murphy and Welch (1990) for
example argue that the standard formulation understates early career wage growth by about 30-50
percent and overstates midcareer earnings growth by 20-50 percent. They present a number of
alternative specifications that seem to fit the data better. Depending on the treatment of economy-wide
trends, the dating conventions for tenure and wages, the handling of wage observations that span
multiple jobs and the estimation approaches used Altonji and Shakotko (1987), Topel (1991) and
Altonji and Williams (2005) find huge variations in the returns to tenure.
41
(such as effort, determination and discipline) or in terms of personality traits (such as
self-confidence, sociability and emotional stability) (ter Weel 2008: 729).
Although the literature on the impact of work habits and personality traits on labor mar-
ket careers in general and on salaries in particular has grown rapidly over the last years
(for an overview see e.g. Borghans et al. 2008), very few robust results have been pre-
sented so far. Most of the available studies suffer from the fact that the data used for
estimation purposes cover a broad range of jobs where the returns to different work
habits and personality traits are likely to vary considerably. This, in turn, is likely to
bias the estimated coefficients downwards. Moreover, most of the available evidence is
based on questionnaire responses to an often large number of statements intended to
describe the respective individuals’ personality structures.
This chapter tries to overcome these deficits by, first, restricting the empirical analysis
to a particular dimension of an individual’s personality that we henceforth either call
“mental strength” or “mental toughness”. Second, we avoid using subjective evaluations
of individuals about their own personalities by restricting our analysis to observable
behavior. Thus, we derive an “objective” measure of mental strength/toughness that we
enter as an additional explanatory variable in our estimations. Contrary to most of the
available literature, we use a data set with detailed information on workers who are
employed in a quantitatively rather small and at the same time very specific labor mar-
ket, the “National Basketball Association” (NBA). While the data precludes generali-
zation of the findings beyond the professional team sports industry, it has the advantage
that we have “clean” measures of individual productivity and performance as well as
reliable information on individual salaries and that we can construct an unbiased
measure of mental strength without having to ask the players to respond to a question-
naire to reveal their work habits and/or their personality traits.32 Summarizing, the goal
of this chapter is to analyze the impact of a particular (and presumably highly valued)
capability of professional basketball players – to avoid “choking under pressure” or to
score “when it really counts” – on the individuals’ remuneration.
32 For a detailed discussion of the advantages of sports data to test different labor market theories see
Kahn (2000).
42
The chapter proceeds as follows: Section two reviews the available literature, section
three describes the setting from which we derive the data used in the estimations (the
NBA). In section four we present our findings and section five concludes.
4.2 What Can we Learn from the Available Literature?
4.2.1 Personality Traits and Earnings: A Review of the Evidence
While a large number of studies have documented the impact of cognitive ability (as
measured by e.g. IQ test scores) on earnings, the impact of non-cognitive skills (such as
motivation, persistence, dependability, etc.) has remained virtually unexplored (early
examples include Andrisani 1978, Filer 1981 and Jencks 1979). 33 In the last five years,
however, a large and still growing number of papers have studied the influence of
personality traits on individuals’ incomes. Irrespective of considerable differences in the
independent variables used, in sample sizes and in cultural contexts all the studies sug-
gest that wages are not only determined by “traditional” human capital measures and by
cognitive abilities but are also affected by non-cognitive skills. Summarizing, these
psychological traits have been found to account for as much as one third of the impact
the traditional human capital variables have on individual earnings (Manning and Swaf-
field 2008).
The studies by e.g. Braakman (2009), Fortin (2008), Heckman et al. (2006), Mueller
and Plug (2006), Nyhus and Pons (2005) as well as Osborne Groves (2005) are based
on slightly different versions of the “five factor model” of personality structure, a
widely accepted instrument among psychologists. The five personality traits are extro-
version (a preference for human contact and attention and the wish to inspire other
people), agreeableness (the willingness to help other people and to act in accordance
with other people’s interests), conscientiousness (a person’s preference for following
33 A further branch of literature examines the impact of physical attributes on earnings. First, a number of
papers have examined the possible effects of physical height on earnings (see Persico, Postlewaite and
Silverman 2004, Heineck 2005, Gautschi and Hangartner 2006, Hübler 2009). Second, a number of
papers look at the impact of handedness on earnings (see Denny and O’Sullivan 2007, Ruebeck,
Harrington and Moffitt 2007). Finally – and perhaps most interestingly – there are a number of papers
looking at the impact of physical attractiveness (“beauty”) on labor market outcomes (see e.g. Biddle
and Hamermesh 1998, French 2002, Hamermesh and Biddle 1994 and Hamermesh, Meng and Zhang
2002).
43
rules and schedules), emotional stability (self-confidence, coolness) and auton-
omy/openness to experience (a person’s propensity to make his or her own decisions
and the degree of initiative and control).34
While all the papers quoted above interpret an individual’s personality as a “bundle of
productive attributes valued in the labor market” (Mueller and Plug 2006: 4), they differ
in the data they use for estimation purposes: Nyhus and Pons (2005) use a represent-
ative sample of the Dutch population collected in 1996 while the study by Mueller and
Plog (2006) is based on the “Wisconsin Longitudinal Study”. Braakmann (2009) uses
the 2005 wave of the German “Socio-Economic Panel”, Fortin refers to the “National
Longitudinal Study of the High School Class of 1972“ and the “National Education
Longitudinal Study” of 1988. Osborne Groves (2005) relies on the US “National
Longitudinal Survey of Young Women” and the British “National Child Development
Study” while Heckman et al. (2006) again use the “National Longitudinal Survey of
Youth”.
Not surprisingly, the findings presented in these studies are quite heterogeneous: Nyhus
and Pons (2005) find that extraversion as well as agreeableness is both associated with
significantly lower wages while emotional stability is associated with significantly
higher wages.35 These results are partly confirmed and partly rejected by Mueller and
Plog (2006) who find that extraversion has no statistically significant impact on the
wages of either men or women and that agreeableness is associated with a wage pre-
mium only for women. For men (but not for women), neuroticism (the opposite of
emotional stability) is associated with a wage penalty. Finally, openness is found to
have a significantly positive influence on wages for both, men and women. However,
even the statistically significant coefficients are of marginal economic importance only.
Braakmann (2009), in turn, finds that for both men and women, only conscientiousness
has a statistically significant negative impact on individual wages while the remaining
34 Apart from the papers quoted above, Drago (2008) studied the impact of “self-esteem” on wages, Liu
and Wong (2005) looked at the influence of “loyalty” on individual wages, Ippolito (1996) at the
impact of “reliability” on individual remuneration and Laband and Lentz (1999) at the role of having a
mentor on subsequent earnings.
35 The remaining two dimensions of the five factors inventory (autonomy and conscientiousness) proved
to be statistically insignificant.
44
four coefficients are all insignificant in the wage equation for men. In the case of
women, agreeableness as well as neuroticism is also associated with a significant wage
penalty. Using somewhat different measures of personality traits Fortin (2008), Heck-
man et al. (2006) and Osborne Groves (2005) find that individuals reporting a high
degree of self-esteem and/or a high degree of control over their life earn significantly
higher wages than individuals believing that luck and fate determine success and
achievement.
The most important common element of the remaining three studies is that they all use a
more “objective” measure of another highly valued personal characteristic, i.e. leader-
ship skills. Using three different data sets from 1960, 1972 and 1982 (with 8,000, 3,000
and 2,000 observations respectively) Kuhn and Weinberger (2005) find that among
white men individuals who occupied leadership positions in high school (as president or
as captain of a varsity team, for example) earn significantly more as adults. The leader-
ship-wage effect varies, depending on model specification and time period, from 4-33
percent. Moreover, high school leaders are more likely to occupy managerial positions
as adults, and leadership skills are associated with a higher wage premium in managerial
positions than elsewhere. The two other studies looking at the influence of leadership
skills on wages use data from two different professional team sports leagues: Deutscher
(2009) relies on data from the National Hockey League over a period of five seasons
(2003/04-2007/08) with some 2,800 player-year-observations while Battré, Deutscher
and Frick (2009) analyze comparable data from the German “Bundesliga” over a period
of thirteen consecutive seasons (1995/96-2007/08) with more than 6,500 player-year-
observations. Both studies reveal that players who have been appointed team captain by
the respective head coach earn significantly higher wages: Other things equal, hockey
and football players who are captains earn a wage premium of some 20-35 percent.
Evaluated at the respective mean this implies that captains earn about 300,000-500,000
$ (hockey) and 350,000-400,000 € (soccer) more per year than their teammates with a
comparable performance on the ice/pitch.
Summarizing, it appears that, first, the studies using “representative” data are likely to
underestimate the “true” impact of personality traits on earnings, because the returns to
45
personality traits are likely to vary across occupations (see Cobb-Clark and Tan 2009).
If, for example, extraversion is beneficial for sales people and teachers, but not for
accountants, it is problematic to use a sample that includes sales people and teachers as
well as accountants. Second, the significantly positive correlation between formal edu-
cation and various measures of self-esteem, extraversion, agreeableness, emotional sta-
bility, etc. found in most studies is likely to result in a downward bias of the coefficients
of either the former or the latter variables. Given these methodological problems we
restrict our analysis to a particular labor market, where one of the five dimensions of an
individual’s personality, his emotional stability (something we call “mental strength” or
“mental toughness” instead) is likely to be of prime importance for that person’s per-
formance (and, therefore, his income). Before explaining the institutional set-up of that
particular labor market we will now briefly summarize the available literature analyzing
the wage determinants of professional basketball players.
4.2.2 Performance and Remuneration in Professional Basketball
Since we use data from the professional team sports industry in the empirical section of
this chapter it is helpful to take a brief look at the available studies that seek to identify
the determinants of player salaries in that particular industry, i.e. professional
basketball. Due to the availability of detailed and reliable information on player salaries,
contract duration and performance/ productivity, a large number of studies have been
published recently (see e.g. Kahn and Scherer 1988, Koch and Vander Hill 1988,
Wallace 1988, Brown, Spiro and Keenan 1991, Jenkins 1996, Dey 1997, Hamilton
1997, Gius and Johnson 1998, Escker, Perez and Siegler 2004, Hill 2004, Prinz 2005).
Perhaps surprisingly, none of the available studies has so far included a measure of
“non-cognitive skills”. Thus, the evidence only documents the returns to “standard”
performance measures.36
36 Significant impacts of experience, performance and peer reputation on salary can also be found in
studies of other North American sports, see Kahn (1993) for baseball, Berri and Simmons (2009) and
Simmons and Berri (2009) for American football and Idson and Kahane (2000) for hockey. Moreover,
similar effects have been found in studies on European soccer (see Lehmann and Weigand 1999,
Huebl and Swieter 2002, Lucifora and Simmons 2003, Lehmann and Schulze 2008, Garcia-del-Barrio
and Pujol 2007, Frick 2007, Battré, Deutscher and Frick 2009).
46
Irrespective of the time period37 and the size of the sample38 these studies unanimously
agree that player wages are a function of, first, a player’s human capital (i.e. his ability/
potential as measured by his draft position and his years of experience), second, a
player’s “fan appeal” (as measured by his number of all star appearances) and, finally,
his productivity (i.e. an individual’s contribution to the “glamour statistics” such as
scoring, rebounding and shot-blocking). While draft position, experience, tenure with
the current team and number of all-star appearances all have a positive, yet decreasing
impact on wages, minutes per game, points, rebounds, assists and blocks per minute
have a positive and strictly linear influence. Moreover, height (Eschker et al. 2004, Hill
2004 and Prinz 2005) and contract length (Jenkins 1996) seem to have a positive influ-
ence on player wages too, while the impact of the number of previous teams is negative
(Gius and Johnson 1998, Eschker et al. 2004, Prinz 2005).
Summarizing, it appears that the salaries of professional sports players are not just ran-
dom, but that systematic factors determine these salaries to a large extent and that these
systematic factors e.g. age, experience and performance are very similar to those found
in other occupations.39 However, the variation in salaries that can be explained with a
small set of precisely measured right-hand side variables is quite large compared to the
respective values in the studies that use data from representative labor market surveys
(25-35 percent vs. 60-70 percent).
37 Some studies use data from the mid 1980s (Koch and Vander Hill 1988, Kahn and Sherer 1988,
Wallace 1988 and Brown, Spiro and Keenan 1991), some from the mid 1990s (Hamilton 1997, Gius
and Johnson 1998). Most interesting, however, are the papers that use longitudinal data covering a
period of five to ten consecutive seasons (Jenkins 1996, Dey 1997, Eschker, Perez and Siegler 2004,
Hill 2004, Prinz 2005).
38 Some of the samples are rather small (with slightly more than 200 observations) while others are quite
large (with up to 4,500 player-year-observations).
39 Where sports teams differ is that they apply more stringent selection procedures into occupations. For
example, poor performance by a player results in being dropped from team squad and very quickly
being discarded; there are high levels of mobility within the industry (between teams) and into and out
of the industry, with shorter careers than in most occupations.
47
4.3 “Choking Under Pressure”: The Fragility of Performance under Stress
The term “choking under pressure” is widely used in the academic literature as well as
in the popular press to describe poor performance in response to what an individual
perceives as an important and stress-filled situation:
“Choking under pressure is not just poor performance. Rather, choking is sub-
optimal performance – worse performance than expected given what the perfor-
mer is capable of doing and what this performer has achieved in the past. This
less-than-optimal performance does not reflect a random fluctuation in skill level
(…), but rather occurs in response to a high-pressure situation” (Beilock and Gray
2007: 426).
The pressure experienced by many individuals – be it athletes in a competition, students
in an exam or executives in a business meeting – can be the result of the level of
rewards that the individuals expect to receive in the case of success (e.g. Baumeister
1984) or the level of punishment they fear in the case of failure (see e.g. Paulus 1983).
Given the well-documented influence of pressure on performance40, we seek to answer
the question whether and to what extent an individual’s ability to deal with pressure
situations has an impact on that person’s remuneration. More specifically, we analyze
40 A significant amount of research has examined choking under pressure in laboratory settings rather
than in actual game situations. While a laboratory environment provides a controlled setting in which
performance failures can be studied while the amount and the players’ perceptions of pressure can be
manipulated. For example Leith (1988) found that individuals shooting free throws who were made
aware of the fact that “some people have the tendency to choke at the free throw line” performed
significantly worse than those who had not received that kind of information. Dohmen (2008a) finds
that professional football players are more successful in penalty shot-outs when playing away games,
i.e. when not playing in front of their fans (a finding that supports the related ‘social pressure
hypothesis’).
48
the performance under pressure of professional basketball players41 and the monetary
returns to the ability to handle pressure situations.
To measure a player’s “stress resistance” or “mental toughness” we compare his
performance from the free throw line in “crucial game situations”, i.e. during crunch-
time with his performance during the rest of the game. As crunch-time we define the
last five minutes of the 4th quarter of a match or overtime, when no team is ahead by
more than five points. In that situation, the preconditions for “choking under pressure”
are likely to be fulfilled: First, missing an opportunity to score is particularly proble-
matic in a close match and, second, making a mistake that cannot be corrected as the
match is coming to an end is particular problematic, too. Thus, pressure and stress are
likely to increase towards the end of a match when the score is close.
Why are free throws particularly suited to study the impact of “stress resistance” and
“mental strength” on individual performance (and, subsequently, on remuneration)? The
free throw is a quite unique situation in sports: It is a routine task for a player without an
opponent being able to deter him from hitting the basket. The distance from the free-
throw line to the basket is 15 feet in every arena and there is no influence apart from
crowd size and behavior. However, average attendance in the NBA is very similar
across the teams42 and the distance between the crowd and the player is the same in
every NBA arena. Thus, the free throw is the only instance during a match where a
player’s output is independent of the efforts made by players of the opposing team. A
player is awarded one or more free throws if he gets fouled while taking a shot, if he
41 Anecdotal evidence abounds: With 90 seconds left to play in the national championship game in the
2008 NCAA men’s division I basketball tournament, the Memphis Tigers had a six point lead over the
Kansas Jayhawks. Kansas fouled strategically to send Memphis to the free throw line, hoping to see
them fail. Memphis, with a previous 59 percent completion rate, missed four out of five free throws,
helping Kansas to reach overtime and finally win the game. Another example: In the first game of the
1995 finals of the National Basketball Association (NBA) between the Orlando Magic and the
Houston Rockets, Nick Anderson, a 70 percent career free throw shooter of the Orlando Magic, had
four straight free throw attempts with a few seconds left to expand the team’s lead of three points. He
missed all four shots enabling the Houston Rockets to tie the game with a last second shot and finally
win the match in overtime. Later on, the Rockets won the series in four games. Nick Anderson’s free
throw percentage declined significantly after this incidence to a level of 40 percent in the two seasons
after that particular game.
42 During our observation period the Detroit Pistons had the highest average attendance (with 20,335 in
the 2005/06 season) while the Atlanta Hawks had the smallest crowd on average (with an average of
15,026 spectators in the 2003/04 campaign). This difference is far smaller than the difference between
the strong and the weak drawing teams in European football.
49
gets fouled while the opposing team is over the team foul limit for the quarter, or if the
opponent receives a technical or flagrant foul.
Given the importance of free throws in close matches particularly during crunch-time
the question arises whether players who are better able to maintain their performance
level in “critical situations” benefit in terms of higher salaries. We assume that being
able to avoid “choking under pressure” is a scarce talent that is particularly rewarded in
the labor market. To measure an individual player’s response to pressure we compare
his performance from the free throw-line in crunch-time with the same player’s perfor-
mance during the rest of the match. Thus, our measure of “mental strength” is computed
as follows43:
Thus, for players whose performance suffers under pressure we end up with a value of
less than one while for those doing better (i.e. improving) in pressure situations, we
have a value above one.44
On average, players successfully complete 77.8 percent of their free throws in pressure
situations, while the respective percentage share for the rest of the time on the court is
78.6 percent. Thus, players seem to perform worse in pressure situations, but this differ-
ence is not statistically significant.45 Since the individual players’ response to pressure
varies considerably (see Figure 4-1) we now take a closer look at the impact of mental
strength on player remuneration.
43 Data on player performance under pressure were obtained from the website http://www.82games.com.
44 We also take into account that it is harder for a good free throw shooter to improve further during
crunch-time situations compared to a mediocre shooter by multiplying the respective player’s
improvement/decline during crunch-time by his non-crunchtime free throw percentage. This does not
change any of the results presented in this chapter.
45 If, as one might argue, a player’s performance suffers due to fatigue, we should find in our data a
negative correlation between our measure of “mental toughness” and the number of minutes played
per game. The resulting correlation coefficient is close to zero and not statistically significant. Thus,
the relative performance of players from the free throw line does not suffer from having played more
minutes. Moreover, one might also expect to observe a significantly positive correlation between a
player’s mental strength and his minutes on the court during crunch-time. Again, our data does not
support this hypothesis, because the correlation is again insignificant.
50
Figure 4-1: Kernel Density Estimation “Mental Strength”
4.4 Data, Estimation, and Findings
The data set analyzed in this chapter includes the standard performance statistics as well
as the annual salaries of all professional basketball players who appeared in at least one
regular season match in the National Basketball Association (NBA) in the period
between 2003/04 and 2006/07. We start with a data set that includes 697 different play-
ers and 2,008 player-year-observations that have been compiled from all 4,879 regular
season games that have been played during that period.46 Statistics on player perfor-
mance were drawn from the league’s official website (http://www.nba.com), while
player salaries were obtained from http://www.eskimo.com/~pbender. Our final data set
includes only players who were awarded at least 15 free throw attempts during crunch-
time in the 82 regular season games in the 2003/04-2006/07 seasons. This leaves us
with 208 different players and 458 player-year-observations. These players were
awarded a total of 147,518 free throws, of which 14,325 occurred during crunchtime.
46 The NBA had 1,189 regular season games during the 2003/04 season and, due to the arrival of one
expansion team (the Charlotte Bobcats) 1,230 games during following three seasons.
01234
Density
.5 11.5 2
Mental Strength
Kernel density estimate
Normal density
51
Apart from mental strength, a number of additional factors are likely to affect an indi-
vidual’s annual salary. As suggested by the previous literature, we control for player
experience and “potential” (measured by the number of years in the NBA at the start of
the respective season and the draft number). Since we expect decreasing marginal
returns to experience, we also entered experience squared in our estimations. A low
draft number, in turn, indicates a player of high ability, as he has been selected at an
early stage of the annual recruiting event of the NBA. We therefore expect a negative
influence of draft position on annual earnings (the higher the draft number, the lower
the ability level and, consequently, the lower the salary). Since there are 30 teams in the
NBA and each team has two picks, the highest possible draft number is 60. Players who
were not selected during a draft, but still made it into the league and into our sample,
were coded as pick number 61.
In addition to talent and experience a player’s performance on the court is likely to have
a strong impact on his remuneration. We therefore control in our estimations for the
number of minutes played per match as well as the individual’s performance in the
offense and the defense by including, first, the number of points scored per game and,
second, the non-scoring performance that we measure as follows:
However, since individual performance statistics might not tell the whole story, we
additionally control for the impact that individual players have on their team’s perfor-
mance. We therefore measure the different teams’ performance in defense and offense
per 100 ball possessions when a particular player is either on or off the court. The vari-
able “ONOF” measures the difference between these values and was obtained from the
website http://www.82games.com.
The number of all-star appearances serves as our measure of “superstardom” and/or
“fan appeal” and allows distinguishing between exceptional players and very good
52
ones.47 Since we expect decreasing marginal returns to the number of all-star appear-
ances we include the squared term of that variable in our estimations too. Moreover,
left-handed players might be harder to defend, since players are accustomed to defend
right-handed players during practice and during the majority of the games played. One
might therefore assume that left-handed players, due to their short supply, earn more
than their right-handed counterparts.48 Our data set contains twelve left-handed players
who account for 35 player-year observations. Finally, an individual’s ability to play on a
certain position is expected to have an influence on his salary, too. We therefore control
for the five positions point guard (PG), shooting guard (SG), small forward (SF), power
forward (PF) and center (CE). Due to the short supply of individuals who are able to
play the center position, we expect centers to be among the highest paid individuals in
professional basketball. We also control for the different clubs’ financial situation by
including the natural log of the team wage bill in the estimations.
Table 4-1 displays the descriptive statistics of our sample. Since our sample includes
only players with a minimum of 15 free throw attempts during crunch-time, we have a
highly selected population of athletes who are among the highest paid in the league. The
average salary is about 6.9 million dollars with a median of 5.5 million dollars.49 Indi-
vidual salaries vary from 366,931 dollars that were paid to Udonis Haslem in the
2003/04 season to 28 million dollars that went to Kevin Garnett during the same cam-
paign.50
47 Superstars have been shown to increase the public interest in a franchise and raise its market value
considerably (see Hausman and Leonard 1997).
48 Bryson, Frick and Simmons (2009) show that left-footed soccer players are paid significantly higher
salaries.
49 The average salary in our complete (“starting”) sample is slightly below 3 million dollars. Moreover,
the players in our final sample spend more than 32 minutes per match on the court (compared to 12
minutes in the whole sample).
50 To account for possible discontinuous salary history and heterogeneous financial powers of the teams,
season dummies (SD) as well as team dummies (TD) are included in the regression analysis.
53
VariableDefinitionMeanMin.Max.
LNTPNaturallogarithmofteampayroll17.9016.9718.65
EXPExperience5.08017
EXP²Experiencesquared35.940289
DNDraftnumber19.27161
DN²Draftnumbersquared716.3913721
ASGAll‐stargameappearances1.28014
ASG²All‐stargameappearancessquared7.420196
MINMinutesplayedin%ofmaximumminutespossible60.9518.3685.19
PPGPointsscoredpergame15.744.235.4
NSPNon‐scoringperformancepergame8.53320
ONOFDifferenceinteamperformanceifplayerisoncourt 3.0‐9.820.2
LHLeft‐handedplayer(dummy;left=1)0.0801
PGPointguard(dummy;yes=1;referencecategory)0.2701
SGShootingguard(dummy;yes=1)0.2401
SFSmallforward(dummy;yes=1)0.2101
PFPowerforward(dummy;yes=1)0.1901
CECenter(dummy;yes=1)0.1001
MESTRelativeperformanceinpressuresituation0.990.531.64
Table 4-1: Descriptive Statistics
Before presenting our estimation results we want to emphasize that experience and
mental strength are not correlated. Neither the time a player has managed to survive in
the NBA – a certainly highly competitive labor market – nor the time per match he is on
the court have any discernible influence on “performance under pressure”. Thus, mental
strength seems to be an innate skill and ability rather than a personal characteristic that
can be improved upon by systematic training. We have in our sample young players
who perform well under pressure as well as older players who tend to choke under
stress. These findings confirm sportswriters as well as fans, both usually stating that
mental toughness is an innate skill, which enables some players to respond to pressure
situations better than others.
We take this result as the starting point to test our initial hypothesis stating that players,
who maintain or exceed their performance level during crucial game situations, are paid
54
better than observationally similar players who tend to choke under pressure. To esti-
mate the impact of the aforementioned individual characteristics and performance
measures on player salaries, we estimate different Mincer-type earnings functions that
all have the following general form:
ln
TD
We admit that the models based on the equation above may suffer from an endogeneity
problem, because the direction of causality between the dependent variable (log of an-
nual salary) and our key independent variable (mental strength; MEST) is not com-
pletely clear. It may well be the case that players with higher incomes – who are finan-
cially better endowed with private wealth – are mentally stronger due to an increase in
self-confidence. In order to test for potential endogeneity we performed a Durbin-Wu-
Hausman test (see Hausman, 1978), which analyzes whether there is sufficient differ-
ence between the coefficients of the instrumental variables regression (2SLS) and those
of the conventional OLS specification. The Prob >chi2 statistic of our regression model
(χ2= 14.36, p > .1) clearly demonstrates that we cannot reject the null hypothesis that
the OLS specification is a consistent unbiased estimator, supporting the assumption that
an instrumental approach is not necessary.
Estimation results are reported in Tables 4-2 to 4-4 below. We start with our OLS
model, followed by the random effects estimation and the quantile regressions. Several
studies of salary determination in professional team sports use quantile regression esti-
mation since log salary measures tend to have even greater kurtosis values than standard
occupations (Hamilton 1997, Berri and Simmons 2009, Vincent and Eastman, 2009). Of
course, ordinary least squares is the best linear unbiased estimator provided that the
error distribution is homoscedastic. Moreover, ordinary least squares parameters tend to
a normal distribution around true values even if the individual residuals are not nor-
mally distributed. The particular advantage of quantile regression is that it facilitates
examination of salary returns to characteristics at different points in the salary distri-
55
bution (Koenker 2005). Ordinary least square estimates constrain marginal effects of
covariates to be the same at the mean and elsewhere. But in salary models, and more so
in sports than in standard labor markets, the average salary is greater than the median
due to excess kurtosis of the distribution. Marginal effects at the median are not neces-
sarily the same as at the mean or anywhere else in the distribution. The presence of
player outliers – the “superstars” – may well cause marginal effects of covariates, such
as mental strength, to differ through the distribution. However, we have no prior on the
pattern of this variation. It does appear, though, from evidence on some other North
American sports, that marginal effects of covariates on player salaries do differ in mag-
nitude, sometimes substantially, over the salary distribution (Simmons and Berri 2009,
Vincent and Eastman 2009).51
The OLS and the RE estimations include three different models with model (1) being as
close as possible to the ones that have been used in the literature so far. Models (2) and
(3) differ from model (1) insofar as they both include our measure of mental strength as
an additional right hand-side variable. The difference between models (2) and (3) is that
the former uses points per game and non-scoring performance per game as explanatory
variables while model (3) uses the “on-off” statistic as the main performance measure.
The reason for replacing the more traditional performance measures by the “on-off”
metric is twofold: First, individual statistics are not the only possible way to measure a
players’ contribution to his team’s performance. Comparing the team’s performance
when a certain player is on the court to the team’s performance when he is off the court
provides an alternative measure of that player’s efficiency. Second, the different esti-
mations allow us to evaluate the robustness of our findings with respect to our key inde-
pendent variable, mental strength.
Since the data we use is an unbalanced panel it is possible (and necessary) to exploit
this additional information and account for unobserved player heterogeneity by using
51 Presence of non-normality in the dependent variable is indicated by a large kurtosis value and in our
case the D’Agostino (1990) test is performed by the sktest command in Stata 10.1. We can investigate
the impacts of mental strength at any quantile of the salary distribution, not just the conditional mean.
Moreover, the quantile regression approach is semi-parametric in that it avoids assumptions about the
parametric distribution of the regression error term, an especially suitable feature where the data are
heteroskedastic as in our case.
56
the random effects estimation technique. In order to select the most appropriate model
we performed a Lagrange-Multiplier test (Breusch and Pagan 1980) that analyzes
whether the null hypothesis that the pooled OLS model would work just as well as the
random effects model is rejected or not. The Prob > chi2 statistic of our two regression
models (χ2 = 12.74, p <. 01 (Model 1) and χ2 = 24.24, p < .01 (Model 2)) clearly indi-
cates that unobserved player heterogeneity is present which suggests that applying the
random effects model is more appropriate than using the pooled OLS.52 Nevertheless,
the coefficients obtained via OLS and RE are virtually identical suggesting that the bias
in the OLS-estimation resulting from unobserved heterogeneity may be negligible.
Tables 4-2 and 4-3 reveal that around 70 percent of the variation in player salaries can
be explained by our set of independent variables. This is surprisingly high, as our sam-
ple includes only players whose performance is above average, i.e. who differ far less in
their “glamour statistics” than those excluded from the sample. Moreover, with regard
to the “standard” performance measures our results corroborate the findings that have
been presented in the literature already: Earnings increase with experience, but at a
decreasing rate with player incomes reaching their maximum after nine years and
declining thereafter (Gius and Johnson 1998: 704, Kahn and Sherer 1988: 50, Hamilton
1997: 292, Hill 2004: 88). The draft number also has the expected influence on player
salaries53: The earlier a player is picked out of the pool of applicants ready to join the
NBA, the higher is his income (Gius and Johnson 1998: 704, Wallace 1988: 305, Hill
2004: 88). This effect, too, declines as the resulting profile exhibits a downward-sloping
and convex form (Koch and Vander Hill 1988: 88, Prinz 2005: 119).
52 Estimation of a fixed effects model is not possible, as some of the right hand side variables (such as
draft number and left-handedness) are constant over time.
53 Moreover, Staw and Hoang (1995) find that the draft number has a statistically significant impact on
the individuals’ playing time in the NBA.
57
Variable1.12.13.1
CoefficientsCoefficientsCoefficients
LNTP0.295(1,54)+0.299(1.54)+0.319(1.68)*
EXP0.348(14.24)***0.372(15.08)***0.350(14.33)***
EXP²‐0.018(‐9.07)***‐0.021(‐10.47)***‐0.018(‐9.22)***
DN‐0.024(‐3.83)***‐0.027(‐4.45)***‐0.025(‐3.98)***
DN²0.000(1.89)*0.000(2.24)**0.000(2.06)**
ASG0.110(3.42)***0.187(6.36)***0.112(3.60)***
ASG²‐0.008(‐2.34)**‐0.012(‐3.60)***‐0.008(‐2.73)***
MIN0.000(0.15)+0.007(3.14)***0.000(0.18)+
PPG0.028(4.06)***‐‐‐0.028(4.08)***
NSP0.034(2.81)***‐‐‐0.033(2.81)***
ONOF‐‐‐0.007(1.27)+‐‐‐
LH0.220(2.06)**0.278(2.48)**0.221(2.08)**
SG0.123(1.65)+0.127(1.79)*0.114(1.53)+
SF0.075(0.97)+0.092(1.17)+0.067(0.87)+
PF0.035(0.45)+0.111(1.47)+0.026(0.33)+
CE0.400(4.12)***0.436(4.57)***0.402(4.20)***
MEST‐‐‐0.456(2.17)**0.432(2.16)**
SeasondummiesIncluded
TeamdummiesIncluded
Const8.51(2.41)**8.21(2.26)**7.636(2.15)**
R²0.7300.7180.733
Numberofobs458458458
***, ** and * denote statistical significance at the 0.01, 0.05 and 0.1 level, + denotes insignificance (z-values in brackets).
Table 4-2: The Impact of Mental Strength on Player Salaries (OLS-Estimation)
Apart from experience and draft number, the number of all-star appearances is also a
valid measure of a player’s talent. Entering the linear as well as the squared number of
all-star appearances in our estimation results in the well-known picture: First, the mar-
ginal returns to all-star appearances are decreasing and, second, the number of all-star
appearances at which the income maximum is reached (the turning point) is about nine
– a number that very few players ever reach (Prinz 2005: 119).
58
Variable1.22.23.2
CoefficientsCoefficientsCoefficients
LNTP0.352(2.02)**0.371(2.13)**0.371(2.14)**
EXP0.372(11.60)***0.389(12.53)***0.374(11.63)***
EXP²‐0.020(‐7.65)***‐0.025(‐8.72)***‐0.020(‐7.78)***
DN‐0.030(‐3.77)***‐0.033(‐4.22)***‐0.031(‐3.91)***
DN²0.000(1.74)*0.000(1.99)**0.000(1.87)*
ASG0.116(2.94)***0.122(3.25)***0.122(3.25)***
ASG²‐0.007(‐1.75)*‐0.008(‐2.28)**‐0.008(‐2.28)**
MIN0.000(0.10)+0.002(1.15)+0.000(0.09)+
PPG0.013(1.67)*‐‐‐0.012(1.58)+
NSP0.016(1.05)+‐‐‐0.017(1.13)+
ONOF‐‐‐0.005(1.02)+‐‐‐
LH0.268(1.64)+0.298(1.77)*0.274(1.67)*
SG0.126(1.35)+0.116(1.26)+0.117(1.23)+
SF0.014(0.14)+0.005(0.05)+0.002(0.02)+
PF0.126(1.23)+0.142(1.40)+0.101(1.08)+
CE0.377(3.09)***0.389(3.28)***0.383(3.20)***
MEST‐‐‐0.427(2.44)**0.407(2.33)**
SeasondummiesIncluded
TeamdummiesIncluded
Const7.80(2.44)**7.261(2.24)**7.052(2.19)**
R²0.7070.6980.710
Numberofobs458458458
***, ** and * denote statistical significance at the 0.01, 0.05 and 0.1 level, + denotes insignificance (z-values in brackets).
Table 4-3: The Impact of Mental Strength on Player Salaries (RE-Estimation)
Perhaps surprisingly, minutes played, points scored per game, non-scoring performance
and the on-off statistic do not have pronounced effects on player salaries once unob-
served heterogeneity is taken into account. This is certainly due to the fact that we have
a highly selected population, where the variance in these performance measures is far
lower than in the total player population.54
54 This is again in line with the available literature: Using a large unbalanced panel of NBA players in the
seasons 1990/91-1999/2000 with 4,072 player-year-observations, Prinz (2005: 125) finds that when
controlling for other (potential) determinants of individual salaries, few of the measures of scoring and
non-scoring performance are statistically significant.
59
Variable0.1Quantile0.25Quantile 0.5Quantile0.75Quantile0.9Quantile
LNTP.5993***.3280** .2621** .1815+.1680+
EXP.4206***.3966*** .3746*** .2972***.2536***
EXP²‐.0291***‐.0208*** ‐.0185*** ‐.0141***‐.0125***
DN‐.0525***‐.0034*** ‐.0200*** ‐.0164***‐.0045+
DN².0005***.0003** .0001+.0001+‐.0000+
ASG.1307***.1373*** .0442+.0512+.0800*
ASG²‐.0034+‐.0115*** ‐.0036+‐.0043+‐.0058+
MIN.0026+.0020+.0002+.0018+‐.0031+
PPG.0053+.0158*.0252*** .0334***.0271***
NSP.0101+.0420** .0397*** .0258**.0311**
LH‐.0060+‐.2048+.2039** .3367***‐.2210*
SG.0499+.1836*.1382*.0455+.0670+
SF.0198+.1060+.1340*.0967+.0941+
PF.0388+‐.0160+.0748+.0451+.0204+
CE.1307+.3496*** .4379*** .4205***.3705***
MEST.2816+.7282*** .4389** .1547+.3352+
Seasondummiesincluded
Teamdummiesincluded
Const2.892+6.783** 8.483*** 10.82***11.29***
PseudoR²0.5270.5230.4900.4670.426
Numberofobs.458458458458458
***, ** and * denote statistical significance at the 0.01, 0.05 and 0.1 level, + denotes insignificance (z-values in brackets).
Table 4-4: The Impact of Mental Strength on Player Salaries (Quantile Regression)
Interestingly – and in line with the evidence from the general labor market (see e.g.
Rueback, Harrington and Moffitt 2007; Denny and O’Sullivan 2007) – we find that left
handed players earn significantly more than players who shoot the ball with their right
hand. As only a small fraction of the players is left handed and as these players appear
to be harder to defend, the short supply of that scarce talent leads to a higher remune-
ration of these particularly “gifted” players.
Of the four position dummies (reference category: point guard) only the coefficient for
centers turned out to be statistically significant, indicating that centers earn significantly
more than players at any other position on the court (see e.g. Hamilton 1997: 292, Dey
60
1997: 86).55 This result is in line with Berri et al. (2005), who find that particularly tall
basketball players earn significantly more than their observationally similar smaller
counterparts. Finally, the log of the teams’ total wage bill has a significantly positive
impact on the individual players’ salary, indicating that large market teams share their
revenues with players and that the players on their rosters benefit to the same extent
from the rich teams’ “ability to pay” (Wallace 1988: 305).
We now turn to the interpretation of the coefficient that we are most interested in, our
measure of “mental strength”. It appears from Tables 4-2 and 4-3 above that players
who are able to maintain their performance from the free throw line in “critical” game
situations receive a statistically significant and economically sizeable pay premium. An
increase in mental strength of one standard deviation (i.e. by about 10 percent) causes a
player’s annual salary to increase by about 40 percent. Thus, mental strength – a perso-
nality trait that we assume to be an innate skill that neither increases with experience
nor declines as fatigue and exhaustion set in – has a pronounced effect on individual
salaries. In our data, the p-value for the test statistic of the null hypothesis that kurtosis
does not depart from the value associated with a normal distribution is 0.000 and hence
our log salary data depart from normality, a result that is similar to those found in other
studies of North American sports too (e.g. Berri and Simmons 2009 on NFL and Vin-
cent and Eastman 2009 on NHL). We therefore report quantile regression estimates in
Table 4-4. We find that the salary premium for mental strength is significant at 5 per
cent or better only at the 0.25 and the 0.50 quantile. At the median the premium for
mental strength is again estimated at about 40 percent. Thus, the quantile regression
results corroborate those from the OLS and the RE estimations. Summarizing, it appears
that the “median player” enjoys a significant pay rise in case of pronounced mental
strength while the “marginal players” as well as the “real superstars”56 do not benefit at
all from that particular ability.
55 Throughout, we derive percentage impacts of changes in dummy variable from coefficients as
exp ( β ) – 1, where β is an estimated coefficient (Halvorsen and Palmquist 1980).
56 This term has first been used by Alan B. Krueger (2005) who analyzes the revenues generated by
particularly successful rock bands and musicians.
61
Our finding has a number of practical implications: First, we can think of no justi-
fication to invest additional resources in mental training in professional sports with the
goal of improving players’ self-confidence. Since the ability to respond to pressure
(something that is ubiquitous in professional sports) seems to be an innate ability that
cannot be improved by training and practice and does not increase with experience, the
clubs’ scouts should try to identify “stress resistant” players when they are still in col-
lege. When successful, this could lead to another “moneyball” story (Lewis 2003).
4.5 Summary and Implications
In this chapter we have tried to answer the question whether the ability to maintain
one’s performance level “under pressure”, i.e. in situations that people perceive as
“stressful” is particularly rewarded by employers. Due to the deficits of the available
data sets we use an unbalanced panel of players from a particular team sports industry,
the NBA. Although this precludes generalization of our findings to the “real world”, we
are able to measure the distribution of mental toughness in the population under study.
Moreover, we identify the impact of mental strength – the ability to perform well from
the free throw line during “crunch-time” – on player salaries. Athletes whose
performance “when it really counts” is one standard deviation above the mean of the
sample are paid about 40 percent more than their observationally similar teammates.
Moreover, we find that the ability to maintain one’s performance level is an innate skill
that can hardly be improved by training and practice. Experienced players do not per-
form significantly better under pressure and tired players do not perform worse. Thus,
the ability to avoid “choking under pressure” is a valuable and scarce resource that is in
particularly short supply.
Since this chapter is the first to identify the impact of mental toughness on individual
salaries, additional research is urgently required to either document or question the
robustness of our results. First natural candidates are certainly other team sports indus-
tries like soccer or hockey as the players here very often experience similar feelings of
pressure when taking penalty kicks to finally decide a match.
62
5 Productivity in Friendly and Hostile Environments: Empirical Evi-
dence from the National Basketball Association.
5.1 Introduction
Recent studies have shown that the presence of spectators influences the aggressive
stance as well as the performance of individuals, as shown by Charness, Rigotti and
Rustichini (2007) and Dohmen (2008b), respectively. In order to categorize the impact
of the audience on performance, social psychologists distinguish between the “social
support hypothesis” and the “social pressure hypothesis”. The social support hypothesis
claims that performance of an individual increases once he is surrounded by spectators.
The social pressure hypothesis however alleges the opposite, stating that individuals
performance declines due to spectators expectations. Past research neglected the impact
of changing audiences on the individual performance due to a change of group member-
ship. Professional sports appear to be a fitting natural experiment as players change
teams within the league between seasons and face a new home crowd after a team
switch.
In the present chapter I analyze the impact of the crowd on the free throw shooting per-
formance by professional players from the National Basketball Association, by distin-
guishing between performance in front of the home crowd and performance during
away games. The longitudinal character of the data set enables me to analyze the influ-
ence of a change of the team between seasons on the individual players. After changing
teams a player has to play in front of a new home audience during the new season which
might have an impact on performance. Hence I distinguish between players who
remained with their team, players who signed as free agents for new teams and players
who were traded during the offseason. The main difference between being traded or
signing as a free agent is the choice of a player on his new team: Players who get traded
usually do not have a say on the team they get traded to, while free agents can sign with
the team of their choice.57 Results suggest that for away games, the performance is inde-
57 Devean George, former player of the Dallas Mavericks, is one of very few players with a right to veto a
trade embedded in his contract. He used this right to inhibit being traded to the New Jersey Nets in
February of 2008.
63
pendent of a possible team change in between seasons by a player, which is not sur-
prising since away games are virtually the same. Playing schedules are nearly the same
for every team, so a player still has to play away games against practically the same
teams as before. Data for home games reveals that players who get traded to a new
team, compared to players who stay with their team, maintain their performance level
from the free throw line. In contrast, players who sign with a new team exhibit a signifi-
cantly lower performance level during home games. I conclude that only players who
are able to select a new team worsen their performance due to social pressure expe-
rienced at home games, as their performance during away games is unaffected.
5.2 Literature
With this research I add a natural experiment to the literature concerning the impact of
spectators on sports performance for both, individual and team sports. For Olympics
Games, a home advantage has been found for Winter Olympics as well as Summer
Olympics. Overall, Balmer, Nevill and Williams (2001) report a statistical significant
higher winning percentage for contestants from the host country during Winter Olym-
pics between 1908 and 1998. By examining the different events in greater detail, the
authors illustrate that while some sports such as figure skating and alpine skiing provide
evidence of home advantage, others like bobsled and biathlon do not. In their related
work, Balmer, Nevill and Williams (2003) also analyze the home advantage for the
Summer Olympics from 1896-1986. They group sports according to the impact that
judges have on the outcome of the event. As one of their main results, the authors state
that a highly significant home advantage can be found for events where performance
was predominantly judged subjectively.
For professional team sports in the United States, studies coincidently report a home
advantage, leading to more home wins than away wins under “balanced home and away
schedule”, equivalent to an identical number of home and away matchups between two
teams. In their works concerning the home advantage in all major leagues in the United
States as well as in four levels of professional soccer in England, Pollard and Pollard
(2005) show that the winning percentage for home teams in the National Basketball
64
Association for the relevant time period of the present data set was around 60 percent.58
Comparing this with the numbers the authors provide for the other major league sports,
one can see that it is the highest percentage. Major League Baseball reports a winning
percentage of home teams of around 54 percent, while the National Hockey League and
the National Football League display average home winning percentages of 55 percent
and 59 percentage, respectively.59
These studies display the influence of the audience on regular season games, but espe-
cially in decisive playoff games the atmosphere might impact the outcome of a game
even stronger: Baumeister and Steinhilber (1984) present a work concerning the per-
formance of home teams in decisive games in professional sports. Looking at data from
the National Football League as well as the National Basketball Association, they find
that home teams win early playoff games rather than decisive ones. Related to the
present chapter they find that players from the home team rather choke shooting free
throws, as the performance from the free throw stripe declines. Using updated infor-
mation a decade later, Schlenker, Philipps, Boniecki and Schlenker (1995b) find chok-
ing by home teams to be ceased for their more comprehensive data.60 In their opinion
chocking is rather associated with the anticipation of failure than the distraction of
potential success, as they go in line with Baumeister and Steinhilber (1984).
Given these results for team performance, the question arises what personality traits
lead to individual excelling or choking in pressure situations. Heaton and Sigall (1991)
analyze which personality type is most prone to be influenced by pressure situations.
Performing a laboratory experiment which includes performing a psychomotor task in
front of a supportive and non supportive audiences, they find that individuals low in
self-consciousness perform relatively well when they are expected to please the
audience. On the other hand they choke, when there is a high chance that the audience
was disappointed by the outcome. The influence of self-consciousness on performance
under pressure was earlier explored by Baumeister (1984), who finds that self-conscious
58 For more detailed results on the emergence of home court advantage in the National Basketball Asso-
ciation see Jones (2007) and Jones (2008).
59 For a review and further evidence of home advantage in team sports see Nevill and Holder (1999).
60 For a review on why supportive audience might have a positive or negative impact on performance, see
Butler and Baumeister (1998).
65
persons are more susceptible to choking, while Brockner (1979) shows the opposite as
he observes a positive significant correlation between self-consciousness and perfor-
mance under pressure.
The influence of pressure on free throw shooting in basketball has been illustrated in
several studies. Dandy, Brewer and Tottman (2001) show that Australian basketball
players performed significantly worse in games than during practice which might indi-
cate influence of the audience on the performance. Leith (1998) had some basketball
players talking within the team about choking in pressure situations before the compe-
tition while others did not. It turned out that groups that did not discuss choking per-
formed significantly better than groups that did. In a related study, Wang, Marchant,
Morris and Gibbs (2004) let individuals perform repeated free throw shooting tasks,
first without pressure and later with methods adding pressure on them. This resulted in
reduced performance in pressure situation while it displays that high self-conscious
persons were more likely to choke under pressure. For the National Basketball Asso-
ciation Deutscher, Frick and Prinz (2009) show that players receive an additional mon-
etary reward if they are able to perform relatively well from the free throw line during
pressure situations. These pressure situations are defined as being during the last five
minutes of a game when neither team is ahead by more than five points.
Players are not the only individuals in professional sports whose performance is influ-
enced by social pressure, as many studies illustrate the influence of the spectators on
individuals’ decision making. Garicano, Palacious-Huerta and Prendergast (2005) find
proof for social pressure on behavior of individuals. Using a data set from professional
soccer in Spain they show that referees lengthen the amount of stoppage time in close
games when the home team is behind, while they shorten the time in close games when
the home team is leading, supporting the thesis that social pressure indeed impacts indi-
viduals’ behavior.61 Similar results are presented by Sutter and Kocher (2004) as they
analyze a data set from the German Bundesliga from the 2000/2001 season. They find
that referees award more extra time at the end of the match when the home team is be-
61The problem with refereeing decisions is based on the circumstance that they result from subjective
judgment (See Dawson, Dobson, Goddard and Wilson (2007) and Boyko, Boyko and Boyko (2007)).
66
hind. In addition there is a bias in favor of home teams in regard to the number of pe-
nalties being awarded. Dohmen (2005) comes to a very similar finding, also for the
German Bundesliga for the seasons 1992/1993 to 2003/2004. He emphasizes a stronger
bias toward the home team when the crowd largely consists of home team supporters.
Scoppa (2008) also reports similar results for two seasons in Italy’s Serie A.62 As a
main finding he reports injury time being rewarded if the home team is losing and a
greater bias towards the home team if there is no running track between the field and the
stance. Also working with data from Italian professional soccer, Pettersson-Lidbom and
Priks (2009) compare the decision making by referees when games are played in an
empty stadium to the situation when games are played in front of spectators, as some
teams were punished to play in empty stadiums due to hooligan violence. The authors
find that referees punish away players less harshly in these games compared to when
spectators are attending. The authors show that teams play with the same level of inten-
sity in front of spectators compared to games in empty stadiums and conclude that
social pressure rather affects referees than players. In an earlier study concerning the
influence of the crowd on referees Nevill, Balmer and Williams (2002) confront refe-
rees with video tapes of game situations from a Premier League game. Results show
more favorable calls for the home team with crowd noise audible than when watching
the videos in silent. Once more, this underlines the thesis that referees are indeed influ-
enced by the atmosphere in the arena.
Furthermore, next to studies concerning sports, experimental research on the impact of
group membership on decision making of individuals has also been done in the recent
past. Eckel and Grossman (2005) show in their work how team identification impacts
the willingness to cooperate. This in turn leads to a higher team output in their setting.
In a related work by Charness, Rigotti and Rustichini (2007) individuals increase their
aggressive stance if they have members of their group in the audience. In the Battle of
the Sexes game this leads to more coordination, while in the Prisoner’s Dilemma it
leads to less cooperation. Supporting the assumption that social pressure might impact
productivity, Mas and Moretti (2009) reveal in their empirical work that social pressure
can partially internalize problems of free-riding when considering team outputs.
62 For a recent empirical work on the referee bias toward home teams in European cup football see
Dawson and Dobson (2009).
67
Observing performance of workers at a large supermarket chain, they find that replacing
a worker with a below average permanent productivity with a worker who shows above
average permanent productivity eventually increases the effort of the other workers in
that same working shift.
Summarizing, the abovementioned literature analyzes the impact of audience on indi-
viduals’ performance, providing support for both, the social support hypothesis as well
as the social pressure hypothesis. As performance improvement as well as performance
decline in front of a supportive audience has been shown by these studies, this present
chapter adds to the aforementioned literature by presenting a large database to measure
performance in front of changing audiences. In particular, the free throw shooting per-
formance in the National Basketball Association is the main subject of this work as I
analyze the impact of changing teams between two seasons. The chapter is organized as
follows: The next section presents the data set used for the research. While section three
provides the empirical results which display the impact of a change of teams on players’
performance, section four concludes.
5.3 Data
To measure the impact of the crowd on players’ performance from the free throw line, I
analyze a data set from the National Basketball Association. It reports the outcomes of
all 584,396 free throw attempts during regular season games between the 1997/1998
season and the 2006/2007 season on an individual player basis and differentiates
between the performance during home games and away games. Studies concerning team
sports usually pose the problem of interaction, where success of one side is also
depending on the effort of the other side. Contrary, free throw shooting in basketball
provides a rare situation in sports, where success of the performing player is indepen-
dent of opponents’ or teammates’ actions. The fact that the average attendance in arenas
is relatively constant throughout the league displays another advantage over other lea-
gues since players might be influenced by the size of the attending crowd. For the given
period, average home attendance in the National Basketball Association was 17,123,
68
while 80 percent of the average home attendance was between 14,369 and 19,954. For
an illustration of the average home attendance during the relevant period see Figure 5-1.
Figure 5-1: Average Home Attendance in the NBA between 1997/1998 and 2006/2007
Furthermore, the distance between the crowd and the players is virtually the same in
every NBA arena, which is an additional advantage compared to games like soccer.
Distinguishing between skill based performance and effort based performance, I classify
basketball free throw shots to belong to the first category. The motion sequence of
shooting free throws becomes increasingly automatic and less conscious due to repe-
tition during practice. I consider the shooting procedure as skill based performance
which is largely controlled by procedures outside of the working memory.63 Baumeister
(1984) notes, that skill based performance is rather influenced by pressure than effort-
based performance. In addition to that, Muraven, Tice and Baumeister (1998) state that
skill based performance is rather not as much subject to cognitive or physical fatigue as
effort based performance. These two remarks clearly support the free throw shot as a
63 See Anderson (1993).
0.05 .1 .15
Density
10 15 20 25
Average Attendance (in 1,000)
69
feasible test subject, as it is a skill based performance which allows us to neglect the
playing time of the individual player.
Analyzing the data set, I set a minimum of 10 free throw attempts for both home and
away games. By doing so, I minimize the number of players who exhibit a perfect suc-
cess rate due to the low number of attempts.64 Hereby, the number of players is 630,
while the data set totals in 2,675 player-year-observations and 502,317 free throw
attempts, including 257,144 attempts for home teams and 245,173 for away teams. Data
concerning the free throw performance has been taken from the league’s official website
at http://www.nba.com and http://www.basketball-reference.com. Data concerning the
players’ salary was obtained from Patricia Bender’s webpage at
http://www.eskimo.com/~pbender/. Information on the average home and away atten-
dance in the NBA has been taken from the website of the Entertainment and Sports Pro-
gramming Network at http://www.sports.espn.go.com/nba/attendance.
To measure the impact of the audience on free throw performance, I distinguish
between free throw shooting percentages during home games and free throw shooting
percentages during away games which allows estimating the influence of the crowd on
the players’ performance. The two dependent variables that I analyze in the following
are the free throw percentages in a given season T, both during home games (denoted
as: FT (T) Home) and during away games (denoted as: FT (T) Away). Since signing
with a new team apparently is not the only explanatory variable for performance from
the free throw line, I also include a number of other variables in the following regres-
sion analyses. Table 5-1 shows the descriptive statistics for the dependent as well as the
independent variables used in the following estimations.
64 Out of 2,675 player-year observations, only five players show a perfect success rate during home
games, while four did not miss a single free throw during away games.
70
VariableOperationalizationMeanMin.Max.
FT(T)HomeFreethrowperc.athomegamescurrentseason0.740.221
FT(T)AwayFreethrowperc.atawaygamescurrentseason0.740.141
FT(T‐1)HomeFreethrowperc.athomegamespreviousseason0.740.211
FT(T‐1)AwayFreethrowperc.atawaygamespreviousseason0.740.211
ExpExperience5.68119
Exp²Squaredexperience45.81361
AttemptsHomeChangeoffreethrowattemptsathome1.180.0414.2
AttemptsAwayChangeoffreethrowattemptsataway1.150.0510.3
lnSalaryNaturallogarithmofplayerssalary15.09.4617.2
StayerPlayerremainedwithhisoldteam0.7401
TradePlayergottradedbeforetheseason0.1201
FreeAgentPlayersignedwithanewteambeforetheseason0.1401
Table 5-1: Descriptive Statistics
As the free throw routine is a task a player has to perform throughout his amateur and
professional career, the free throw percentage of a particular player from the previous
season should be a decent indicator to estimate his ability. Hereby, I distinguish
between the previous season free throw percentages during home games (denoted as: FT
(T-1) Home) and during away games (denoted as: FT (T-1) Away). As players’ expe-
rience might also impact his performance and his response to social pressure, the expe-
rience is also included in the following regression analyses (denoted as: Exp). It is
measured in years as a professional player in the National Basketball Association. As I
also take the performance from the previous season T-1 into account, the minimum
experience of the players included in the data set is one year. Since one might expect
decreasing marginal returns to experience, the squared experience is included as well
(denoted as: Exp²). As the number of free throw attempts sometimes differs dramati-
cally between two seasons, the players’ performance might be subject to change due to
a higher or lower number of attempted free throws and resulting in higher or lower
recent experience in this particular game situation. I therefore account for an increase or
decrease in attempts between the most recent season T and the previous season T-1 by
71
dividing the number of attempts in season T by the number of attempts in season T-1.65
Once more, I differentiate between the change in attempts during home games (denoted
as: Attempts Home) and away games (denoted as: Attempts Away).66 As players’ salary
might also impact the pressure put upon a player it is also included as an independent
variable. Since Figure 5-2 clearly shows a right-skewed distribution of players’ salary,
the natural logarithm of the salary is included in the following analyses (denoted as: ln
Salary).
Figure 5-2: Kernel Density Estimation of the Salary in the NBA
Since the aim of this chapter is to measure the effect of playing for a new team on the
players’ performance, I distinguish between three different situations a player might
65 If the number of free throw attempts is the same in T as in T-1, the variable has a value of one. Due to
the lockout in 1998/1999, the number of regular season games was reduced from 82 to 50 in this
particular season. I weigh the number of free throws attempted and free throws made for this
particular season by the factor 1.64.
66 The apparently low numbers of 0.04 and 0.05 result from Orlando’s Grant Hills’ injury during the
2000/2001 season and the resulting decline of attempts in that particular year. The maximum values
display the vast increment of Gerald Wallaces’ playing time after being picked by Charlotte during the
expansion draft in 2004 and the return of Miami’s Alonzo Mourning in 2001/2002 after an injury
during the previous season.
0.05 .1 .15 .2
Density
010 20 30
Salary (in million $)
Kernel density estimate
Normal density
72
face in season T. The first and most common situation is that a player still plays for the
team he has already played for it in the previous season T-1. 74 percent of the players in
the data set remained with their teams (denoted as: Stayer). The second situation,
switching teams between season T-1 and season T, can be subdivided into two possible
events: A player either gets traded by his former team to another team (denoted as:
Trade) or he signs with the team of his choice as a free agent (denoted as: Free Agent).
Since also the contract of a player gets traded between season T-1 and T, any player
who is traded at least has to have a valid contract for season T and does not have a
choice on the team he gets traded to. A player who signs as a free agent does not have a
running contract for season T and is free to sign with a team of his choice after season
T-1. In the present data set, 12 percent of the players are traded between seasons, while
14 percent of the players sign as free agents with a new team. A player who has signed
as a free agent with his old team is also denoted as being a Stayer, since the home crowd
he is playing in front of is the same in season T as it has been in season T-1.
5.4 Empirical Results
The main purpose of this section is to determine what impacts the players’ performance
from the free throw line during home and away games. I particularly want to determine
the impact of the audience on the performance as some players change teams between
two seasons while others do not. The data set shows that 184,856 out of 245,173
attempted free throw during away games were successful, which equals a success rate of
75.40 percent. During home games, 194,736 of 257,144 free throws attempts were
made, equivalent to a success rate of 75.73 percent. Applying a simple t-test while
controlling for the number of attempts by the players, this difference proves to be insig-
nificant (t=0.1657+). While cumulated performance shows that players on average hit
three out of four free throws, individual performances range between a 26.2 and 98.8
percent success rate. Kernel density estimations for players’ free throw percentages are
shown in Figure 5-3, again distinguishing between free throw success during home and
free throw success during away games and supporting the finding that individual
performance indeed differs between players.
73
Figure 5-3: Kernel Density Estimation of Free Throw Success during Home and Away Games
Even though it might be seen as a bit disappointing that performance appears to be
independent of playing at home or away, I go on by running four regression analyses on
individual player basis to estimate the impact of the variables introduced in the previous
section on free throw success during home and away games in season T, which are the
dependent variables. I suggest the two equations to look as follows:
FTTHome
1
² ln
FTTAway
1
² ln
Due to the panel characteristics of the data it is possible to account for some unobserved
player specific characteristics via the random effects estimation technique. Hence, next
0.01 .02 .03 .04
Density
20 40 60 80 100
FT (T) Home *100 FT (T) Away*100
74
to the conventional OLS specification two random effects models are run.67 Estimations
of both models are reported in Table 5-2, where the results of the regression analyses
are shown for home games (Model 1) and away games (Model 2).68
Home(Model1)Away(Model2)
CoefficientsCoefficients
VariableOLSREOLSRE
FT(T‐1)0.679(45.10)***0.442(24.23)*** 0.653(41.82)***0.424(22.95)***
Attempts0.010(5.74)***0.009(5.61)***0.014(7.16)***0.013(6.71)***
Exp0.000(0.48)+0.000(1.25)+0.000(0.52)+0.000(1.27)+
Exp²‐0.000(‐0.14)+‐0.000(‐0.76)+‐0.000(‐0.25)+‐0.000(‐0.92)+
lnSalary‐0.001(‐0.47)+‐0.001(‐0.29)+0.002(0.89)+0.002(0.76)+
Trade‐0.005(‐0.88)+‐0.004(‐0.81)+‐0.006(‐1.21)+‐0.008(‐1.49)+
FreeAgent‐0.017(‐3.11)***‐0.017(‐3.32)*** ‐0.004(‐0.75)+‐0.005(‐0.93)+
AdjR²0.4390.4380.4060.405
Observations2675267526752675
***, ** and * denote statistical significance at the 0.01, 0.05 and 0.1 level, + denotes insignificance.
Table 5-2: Determinants of Free Throw Shooting Success in the NBA
Already introduced in the previous section, the main concern of this work is the impact
of the audience on the performance of the players. Observing the results from the
regression analyses, the performance from the free throw line during the previous sea-
son T-1, separately reported for home and away games, has a highly significant positive
impact on the performance during the most recent season T. This is not very surprising,
since the free throw shot is a standardized situation, practiced at a high volume. Hence,
the assumption that prior performance is a good indicator for the ability to shoot free
67 Even though the Hausman-test suggests performing a fixed effects model, I prefer to present the
random effects model, as robustness checks support the results presented in Table 5-2. Furthermore,
fixed effects results are not very convincing, as they suggest no significant impact of the free throw
performance from the previous season on the most recent performance. This appears to be odd, since
the correlation between FT (T-1) Home and FT (T) is 0.6574 for home games and 0.6276 for away
games, respectively. Furthermore, applying fixed effects models for Model 1 and Model 2 does not
impact the significant levels of the Free Agent variables.
68 As Wallace, Baumeister and Vohs (2005) note, not all audiences show the same type of support for
players and hence some home audiences might put more pressure on players than others. Including
data on average home and away attendance does not impact the significance level of any other
variable. Including team dummies to control for the intenseness of the atmosphere in the different
arenas does not affect them either.
75
throws is supported by the regression analyses. Likewise, the change of free throw
attempts during home and away games in season T compared to season T-1 has a sig-
nificantly positive impact, meaning that shooting more free throws increases the success
rate of a player. While these variables, depicting recent experience during games have a
positive impact on the shooting percentage, experience measured as seasons in the
NBA, does not.69 A possible explanation for this is the fact that players have already
faced pressure situations during their basketball career prior to playing in the NBA, as
college games as well as games in Europe attract a similar number of spectators.70 Since
the players’ income might put additional pressure on him, the natural logarithm of the
salary is included as a control variable in the analyses, showing that it has no impact on
the performance.
To answer the main research question of this chapter one has to look at the key inde-
pendent variables, denoting the change of teams between season T-1 and season T. For
the regression analyses, the players who remain with their teams (denoted as: stayers)
serve as the reference group. Compared to those, neither players who were traded nor
players who sign as free agents with a new team perform significantly different during
away games. This does not surprise, since away games are basically the same as in the
previous season, except for those for the old and the new team. On the other hand, the
free throw shooting performance during home games is clearly influenced by the change
of teams during the offseason. Compared to the players who stay with their team, play-
ers who are traded to a new team do not show a significantly different performance.
This result can be justified by the fact that players commonly do not have a say on the
team that they are traded to, which consequently does not create additional pressure on
them. Taking a look at players who signed as free agents before the start of the new sea-
son, the results show that those perform significantly worse during home games com-
pared to players who stayed with their old team. As a main result of this chapter, I con-
clude that social pressure, generated by selecting a new supporting audience due to
signing with the team as a free agent, leads to a reduction in performance.
69 Running the regression analyses without the variable Attempts does not change the significance level of
any other variable.
70 This result goes in line with a finding in the previous chapter. In chapter four we show that experience
does not increase players’ ability to remain their performance level during crucial game situations.
76
Furthermore, one might be interested whose performance suffers most under social
pressure, targeting the question if rather good or bad free throw shooters are influenced
by a changing home audience. Several laboratory experiments analyze the impact of
distraction on performance by novices and experienced players in sports. These appear
to be comparable to the quantile regressions in this chapter, as they analyze how indi-
viduals of different skill levels respond to pressure situations. They coincidently find
that distraction from a main task hurts performance of novices more than it hurts per-
formance by experts. Leavitt (1979) was the first to support this hypothesis by demon-
strating how speed skating by novices, while stick handling through pylons, was signifi-
cantly slowed down under dual-task condition, while this result does not hold for
experts. Studies by Beilock, Carr, MacMahon and Starkes (2002) and Gray (2004) sup-
port this finding as they study the performance on soccer dribbling and baseball batting,
respectively. Coincidently, they find experts to be significantly less prone to distraction
than novices.
As the present data set only includes professionals one can distinguish between different
levels of free throw shooting within the cohort. Performing quantile regressions for
home and away free throw performances serves as a decent tool to answer the question
if rather good or rather bad shooters suffer from playing for a new team. Results are
presented in Table 5-3 for home games and in Table 5-4 for away games. One can see
that the results for the quantile regressions (.10, .25, .50, .75, .90 quantile) are pretty
similar to the OLS and random effects results presented in Table 5-2 for most variables.
Bearing in mind that players who signed as free agents prior to the season perform
significantly worse during home games one might be interested if rather good or rather
bad free throw shooters at home games are affected by playing in front of a new home
audience. Table 5-3 illustrated that only comparably bad shooters, belonging to the .1,
.25 or .5 percentile, are affected negatively by the new audience, as the performance of
players belonging to the .75 and .9 percentile remains unchanged. Performance during
away games is again independent of changing teams between seasons.
77
Variable.1Quantile.25Quantile.5Quantile.75Quantile.9Quantile
FT(T‐1).8414***.7971*** .7118*** .6020***.4298***
Attempts.0121***.0128*** .0102*** .0049***.0002+
Exp‐.0042+‐.0029+‐.0003+.0026+.0048*
Exp².0002+.0002+.0000+‐.0001+‐.0002+
lnSalary.0158***.0065*** ‐.0010+‐.0043**‐.0123***
Trade‐.0198+‐.0068+.0008+‐.0004+‐.0048+
FreeAgent‐.0287**‐.0240*** ‐.0116** ‐.0064+‐.0045+
PseudoR².3146.3052.2644.2150.1689
NofCases726752675267526752675
RawSumofDev.127.7210.7234.9168.085.6
MinSumofDev.87.5146.4172.8131.871.1
***, ** and * denote statistical significance at the 0.01, 0.05 and 0.1 level, + denotes insignificance.
Table 5-3: Quantile Regression of Free Throw Success during Home Games (Model 1)
Variable.1Quantile.25Quantile.5Quantile.75Quantile.9Quantile
FT(T‐1).7974***.7727*** .6665*** .5766***.4603***
Attempts.0193***.0143*** .0118*** .0082***.0026+
Exp‐.0021+‐.0013+.0008+.0029+.0040+
Exp².0000+.0000+‐.0000+‐.0001+‐.0001+
lnSalary.0208***.0082*** .0037+‐.0076***‐.0129***
Trade‐.0143+‐.0140** ‐.0016+‐.0143+‐.0095+
FreeAgent‐.0120+‐.0078+.0074+‐.0040+‐.0067+
PseudoR².2841.2736.2357.2011.1663
NofCases26752675267526752675
RawSumofDev.129.4211.4232.9166.484.8
MinSumofDev.92.6153.4178.0132.970.7
***, ** and * denote statistical significance at the 0.01, 0.05 and 0.1 level, + denotes insignificance.
Table 5-4: Quantile Regression of Free Throw Success during Away Games (Model 2)
78
5.5 Conclusion and Future Research
This chapter presents an empirical work about the effect of performing in front of
supportive audiences, focusing on the effect of changing teams and supportive
audiences between two seasons. Observing data of free throw shooting in the National
Basketball Association, the outcome is contrary to studies which are based on outcomes
of interaction, as I do not find reduced performance in away games on an aggregated
level. Due to the panel characteristic, the data provides the possibility to test how
individuals respond to changing audiences. The key result presented in this chapter
shows that players who changed teams after free agency suffer decline of performance
at home games compared to those who stay with their old team or are traded. Therefore,
I conclude that the possibility to select the team to play for by oneself has a negative
impact on the performance during home games as players feel additional pressure
during the following season. By performing quantile regression it has been shown that
especially bad free throw shooters suffer from facing this additional pressure, while
good free throw shooters performance does not decline after signing as a free agent with
a new team.
Future research might take a look at the impact of trading a player within the season.
Players are allowed to be traded from one team to another up until the end of February,
so one might want to analyze the impact of changing teams within the season. Since
decent players rarely sign within the season, the control group would be players who
were not traded within the season. Further future research might subdivide the stayers
into two groups: Players with a running contract and players who resigned as free
agents with their old team. There might be problems with the implementation, since
players sometimes sign extensions of the contract long before the old contract runs out.
Still, this data might provide further insights on the impact of the audience on players.
Another future research question could go in line with Baumeister and Steinhilber
(1984) and Schlenker, Philipps, Boniecki and Schlenker (1995) to analyze if free throw
performance of players from the home team suffers in decisive playoff games if they
joined the team as free agents prior to the season. Especially, since the results from the
aforementioned papers contradict, this research would shed more light on the individual
impact of social pressure on individuals.
79
6 Cut-off Dates and Their Effect on Player Selection, Salaries and
Hazard Rates in the German Bundesliga
6.1 Introduction
The implementation of cut-off dates is said to have a severe impact on the personal
development of the affected individuals. Grouping children according to the date of
their birth leads to a high difference in relative age between the oldest and the youngest
of the relevant cohort. The relative age difference between a child who just turned six
years old and another who is close to his seventh birthday is around 15 percent. In Ger-
man soccer the cut-off date is 1st of August for all youth leagues. One possible effect of
introducing a cut-off date is a selection bias, favoring children born shortly after this
cut-off date as they have a physical advantage over the children born afterwards, who
are nevertheless put into the same age cohort. Once these children reach the age which
enables them to become professionals, this in turn would lead to an overrepresentation
of German players in the Bundesliga who were born shortly after the cut-off date.
Furthermore, as previous studies show, players who were among the youngest in their
cohort and still managed to make it to the professional level earn a wage premium. This
might be seen as a surprise, as the market for soccer players should only take perfor-
mance into account which is relevant for the output of players and this would certainly
not include the players’ birth date. In addition, market efficiency would not lead to dif-
ferent hazard rates for different birth dates if players’ performance is indeed indepen-
dent of birth dates. This article tries to shed light on aforementioned possible impacts of
cut-off dates on player selection and player salary in the German Bundesliga. It goes in
line with a big body of literature, which analyzes the effect of cut-off dates and the rela-
tive age effect in education and sports.
In regard to education, research results are ambiguous. Jinks (1964) is among the first to
find evidence for a better school performance of children who are born within the first
six months of the school year, compared to peers who are born in the second half of the
school year. Closely related, Pidgeon (1965) points out that while children born in the
later month of the school year score higher points in intelligence tests, they later per-
80
form comparatively badly due to their disadvantageous date of birth. In a study covering
a wide range of countries, Bedard and Dhuey (2006) analyze long-run effects of cut-off
dates. They find a persistence of better school performance by children who belong to
the oldest of their cohort for the course of their school career. In a recent study, Billardi
and Pellizzari (2008) find contrary results for students at an Italian university, as they
show that the younger students of a cohort perform better than their older counter-
parts.71
Angrist and Krueger (1991) were the first to study the impact of cut-off dates on the end
of school careers at compulsory schooling. They argue that children born shortly after
the cut-off date start school at an older age and hence are allowed to drop out after less
school years than their younger counterparts. Plug (2001) finds that this leads to an
increase in the probability of receiving a university degree by 12 to 16 percent for child-
ren who are born shortly after the cut-off date and hence are the oldest in their cohort.
With regard to professional sports, Barnsley, Thompson and Barnsley (1985) were the
first to discover the relative age effect, as they observed that a disproportionally high
number of hockey players in Canada were born in the first three months after the cut-off
date.72 In a more recent study by Baker and Logan (2007), the authors show that the
relative age plays a role at the annual NHL draft. Furthermore, Barnsley, Thompson and
Legault (1992) find additional evidence of the relative age effect analyzing the rosters
of national teams in the 1990 soccer world cup as well as two soccer youth world tour-
naments. In a related work, Helsen, Van Winckel and Williams (2005) examine the rel-
ative age effect for a number of European national youth teams playing at international
tournaments. Analyzing a total of more than 2000 players, they find a highly significant
effect for nearly all teams, resulting in an overrepresentation of players who were born
shortly after the cut-off date. Musch and Hay (1999) find cross-cultural evidence for a
selection bias of players’ born shortly after the cut-off date as they analyze data from
professional soccer leagues located on different continents. In addition, they observe a
71 See Fertig and Kluve (2005), Puhani and Weber (2007) and Puhani and Mühlenweg (2010) for analyses
of educational outcomes from Germany.
72 See Boucher and Mutimer (1994) for a review of the early literature regarding the impact of cut-off
date in professional sports.
81
shifted peak in the distribution of the birthdays of Australian players once the cut-off
date changed. Verhulst (1992) observes cross-country data on professional soccer play-
ers in Europe to also find a relative age effect for the professional leagues in the Neth-
erlands, Belgium and France. Dudink (1994) observes a skewed distribution of the birth
dates in Dutch youth tennis leagues as well as in the Premier League and the first three
divisions in British professional soccer.
Concerning the player numeration, a number of papers using data from different Euro-
pean soccer leagues analyze the determinants which help explain the players’ salary.73
For the German Bundesliga, Lehmann and Weigand (1999) as well as Lehmann (2000)
were among the first to determine influencing factors of players’ salary in the Bundes-
liga. Both studies use data from one season, as they also take into account the players
origin as well as the performance of their team. Being the first to use longitudinal data,
Huebl and Swieter (2002) and Frick (2007a) support the human capital theory, as they
find positive, yet decreasing impact of several variables depicting players’ experience.74
As hazard rates in the Bundesliga are going to be discussed in the following work, a
short overview of the existing literature seems to be appropriate. Frick, Pietzner and
Prinz (2007) are to my best knowledge the only authors who analyze hazard rates in the
Bundesliga. They investigate the influence of individual performance indicators and
players’ origin on the probability to remain in the league. Analyzing the careers of all
players who played in the Bundesliga between the season 1963/64 and 2002/03, they
find evidence of the position of players as well as their age and experience impacting
the probability of staying in the league. Further studies on hazard rates in professional
sports include careers of Japanese baseball players by Ohkusa (2001) as well as Kura
and Matsuzawa (2006).
For studies about American professional sport leagues Atkinson and Tschirhart (1986)
analyze which determinants influence players’ careers in the National Football League,
while in their work concerning matching models, Chapman and Southwick (1991)
73 See Frick (2007b) for an overview.
74 For similar studies concerning other European leagues see Garcia-del-Barrio and Pujol (2005, 2006)
and Lucifora and Simmons (2003).
82
address the hazard rates for American baseball players. In a more recent study Dilger
and Prinz (2004) show that players from the National Basketball Association have to
leave the league rather due to poor performance than on a voluntarily basis.75 To con-
clude, a growing body of literature has developed during the last years, displaying the
interest of economics in player remuneration and career tracks in professional sports.
The present chapter combines the field of analyzing the impact of cut-off dates on play-
ers’ careers as well as on their salary determination. It is organized as follows: The next
section presents the data while the birth-distribution of German players in the Bundes-
liga is presented in section 3. Section four contains the salary determination of the
soccer players. Hazard rates in the Bundesliga are presented in section five and section
six concludes the chapter.
6.2 Data
Analyzing the impact of the implementation of cut-off dates on players’ career perspec-
tive and players’ salary is the main subject of this chapter. The data set that I analyze in
this chapter contains individual statistics as well as approximated salaries of 2011
players who played in the German Bundesliga between the 1995/96 and the 2007/08
season and totals in 6146 player-year-observations.76 To my knowledge this is the first
longitudinal analysis run with such a large number of observed seasons, as previous
studies include a time frame of at most two seasons. Individual players’ performance
and statistics were obtained from the yearly published special issue of the “Kicker
Sportmagazin”. During the course of the relevant period the cut-off date in German
youth leagues was 1st of August. Since one has to assume that only players born in
Germany were subject to this rule and following Musch and Hay (1999), the number of
observed players decreases to 1090. Consequently the number of player-year-
observations decreases to 3550.
75 Further studies concerning hazard rates in the NBA include Staw and Hoang (1995) as well as Hoang
and Rascher (1999).
76To explain the number of players, which differentiates from chapter 2 and 8, one should note that this
chapter was written with big timely distance and using a then up-to-date data set.
83
6.3 Birth-Distribution in the Bundesliga
Following Ashworth and Heyndels (2007), I will initially take a look at a possible
selection bias of German players entering the Bundesliga. The reasoning behind such a
selection bias is as follows: Due to their physical predominance over the later born
members of the cohort, adolescent soccer players who are born shortly after the cut-off
date are rather labeled as talents and hence receive stronger support. In other words, the
youngest players of the cohort are easily overlooked in favor of their older counter-
parts.77 Looking at the distribution of players’ birth dates in Table 6-1 one can see that
over 29 percent of the German players in the Bundesliga are born in the first three
months after the cut-off date and hence between 1st of August and 31st of October.
Furthermore, 55 percent of the players are born in the first half of the year following
this date which includes 1st of August to 30th of January. Since the data set includes the
precise birth date of every player one can compare it to the birth dates in the population.
Following Barnsley, Thompson and Legault (1992) in assuming an equal distribution of
birth throughout the course of the year, one would expect the players to be born, on
average, 182 days after the cut-off date 1st of August. The 1090 players in the sample
are born, on average, 170 days after 1st of August. Applying a simple t-test, one can see
that this difference proves to be highly significant (t=-4.76).
PlayersFractionCumulated
1stQuarter3170.2910.291
2ndQuarter2820.2590.550
3rdQuarter2470.2270.776
4thQuarter2440.2241.000
Table 6-1: Distribution of German Players’ Birth Dates in the Bundesliga
77 Helsen, Starkes and Van Winckel (1999) find that youth players born shortly after the cut-off date are
identified as being talented by their coaches with a higher probability than their peers. Helsen, Starkes
and Van Winckel (2000) furthermore show for the Belgian youth soccer that a change of the cut-off
date immediately changed the distribution of players’ birth in teams for the following season. In a
study of the Olympic Development Program for youth soccer players in United States, Glamser and
Vincent (2004) find a disproportional high number of players born in the first half of the year after the
cut-off date.
84
The above shown results go in line with Allen and Barnsley (1993), even though the
impact of the cut-off date in Canadian hockey seems to be more drastic. After showing
the expected selection bias, I proceed by estimating the influence of the players’ birth
date on their salary.
6.4 Salary Determinations in the Bundesliga
Concerning the influence of the players’ birth date on salary one is confronted with the
thesis by Ashworth and Heyndels (2007), who claim that late born players receive a
wage premium when they achieve to play in the Bundesliga. They explain this thesis by
stating that everything else equal, these players are more productive and hence earn
more money. In the following, I will show that including goals as a performance para-
meter in my data set also supports the salary bias in favor of late born players while
differentiating between recent career performance and experience revokes the impact of
the birth date on players’ salary.78 Before looking at the detailed results I will start by
presenting the data set.
The players’ salary is approximated by the data from the “Kicker Managerspiel”, which
is run every year. Before the season, every player is assigned a value for this virtual
management game. Comparing these player values with information available from the
licensing procedure by the Deutsche Fußball Liga (DFL), one finds that these values are
very good estimators for players’ salary in season t which are approximated as follows:
1.5
Before giving information on which individual player characteristics and performance
indicators might impact the salary, I start by presenting the salary structure of German
players in the Bundesliga. As one can see in Figure 6-1 the average salary nearly
doubled in the observed period, increasing from 513,327 Euro in 1995/1996, to
78 This goes in line with the finding of Du, Gao and Levi (2008). They show that, after controlling for
relevant information like firm performance and size, the season of birth of CEOs of S&P 500
companies has no impact on their compensation.
85
1,062,903 Euro in 2006/2007. This rapidly increased salary suggests that season dum-
mies should be included in the following salary regression.
Figure 6-1: Salary History of German Players in the German Bundesliga
While individual player salaries range between 17,043 and 10,000,000 Euro per year,
one observes a right-skewed distribution of players’ salary as the medium salary in the
Bundesliga is 769,208 Euro. Looking at the salaries subject to the quarter the players
were born in, one observes the highest average salary for players born in the fourth
quarter (839,694 Euro) and the lowest for players born in the third quarter (707,889
Euro). The distribution of players’ salary subject to their birth date is shown in Figure 6-
2. As there is obviously is no normal distribution of the players’ salary, the following
salary regressions are performed using the natural logarithm of the salary.
0€
200.000€
400.000€
600.000€
800.000€
1.000.000€
1.200.000€
1.400.000€
Year 199519961997199819992000200120022003200420052006
Salary
STDIV
86
Figure 6-2: Kernel Density Estimation of Players’ Salary Subject to the Quarter of Birth
In order to perform the salary estimation and to measure the effect of birth dates on the
salary a variety of other factors have to be considered, which also explain the variation
in players’ payoff. To begin with the differentiation between the players’ position on the
field, I distinguish between goalkeepers, defenders, midfielders and forwards. As
human capital suggests that experience increases productivity, even though with
decreasing marginal returns, I introduce several parameters which depict experience. I
control for players’ age at the start of the particular Bundesliga season as well as for the
squared age, expecting an upward-sloping and concave age-earnings profile. For the
other experience indicators, I distinguish between recent and career experience: Games
played in the Bundesliga also serves as a good indicator for experience. In line with the
age variable, I also include the squared number of Bundesliga games, expecting
decreasing marginal returns. Experience in international matches serves as the third and
final experience indicator and the squared number is included once again. Finally, goals
scored in the Bundesliga serve as an indicator for offensive performance by players.
Following Lucifora and Simmons (2003), I distinguish between recent and career per-
formance. Therefore, I run a second salary regression, in which I differentiate between
0.2 .4 .6 .8 1
Density
0 2 4 6 8 10
Salary (in million €)
1st Quarter 2nd Quarter
3rd Quarter 4th Quarter
87
the aforementioned players’ experience and performance indicators observed in the pre-
vious season (labeled as: t-1) and the cumulated performance before the previous season
(labeled as: prior t-1).
To measure the impact of the birth date on the salary of the player, the variable Birthday
is introduced, denoting the number of days the respective player is born after the cut-off
date 1st of August. Descriptive statistics of the experience and performance indicators
are presented in Table 6-2.
VariableOperationalizationMeanMin.Max.
GoalkeeperGoalkeeper(Dummy;1=yes)0.1401
DefenderDefender(Dummy;1=yes)0.3001
MidfielderMidfielder(Dummy;=yes)0.3801
ForwardForward(Dummy;1=yes)0.1801
AGEAgeatthebeginningoftheseason24.21741
AGE²Squa.ageatthebeginningoftheseason6042891681
GPBLCareerBundesligagames69.20540
GPBL²Squa.Bundesligacareergames121550291600
GPINTCareerinternationalgames8.970137
GPINT²Squa.careerinternationalgames418018769
GSCareergoals7.930182
GPBL(t‐1)Bundesligagames(previousseason)14.9034
GPBL(t‐1)²Squa.Bundesligagames(previousseason)376.8201156
GPBL(priort‐1)Bundesligagamespriortopreviousseason54.30512
GPBL(priort‐1)²Squa.Bundesligagamespriortopreviousseason93890262144
GPINT(t‐1)Internationalgamesinpreviousseason1.43025
GPINT(t‐1)²Squa.internationalgamesinpreviousseason11.50625
GPINT(priort‐1)Internationalgamespriortopreviousseason7.540130
GPINT(priort‐1)²Squa.internationalgamespriortopreviousseason 331016900
GS(t‐1)Goalsinpreviousseason1.63028
GS(priort‐1)Goalspriortopreviousseason6.340171
BirthdayDaysbornafterthecut‐offdate1750365
Table 6-2: Descriptive Statistics of Player Characteristics and Performance Indicators
88
Season dummies (SD) and team dummies (TD) are included in all regression analyses
to account for further influencing factors. As shown above, the inclusion of season
dummies appears to be reasonable. As teams in the Bundesliga exhibit differences in
financial power, team dummies are also included in the salary regressions.
In order to estimate the impact of the date of birth on the players’ salary, the time gap
between the respective birthday and the cut-off date is included in the salary regression.
Different regression analyses are run, which include experience as well as performance
indicators described above. Next to performing a salary regression which refers to total
experience and performance, a second model is run in which I distinguish between
recent and career experience and performance. On basis of the standard Mincer wage
equation (1974) the following equation of players’ salary is suggested:
ln
²
² ²
I run two models, the first includes overall career player information (Model 1), while
the second differentiates between recent and previous performance indicators (Model
2). Due to the panel characteristics of the data it is possible to exploit this additional
information and account for some unobserved player specific characteristics via the
random effects estimation technique. Hence, a Breusch/Pagan (1980) Lagrange-multip-
lier test is performed, which analyzes whether the pooled OLS model would work as
well as the random effects model. The Prob > chi2 statistic of the two regression models
(χ2 = 344.06, p < .01 (Model 1)) and (χ2 = 163.29, p < .01 (Model 2)) clearly indicate
that unobserved player heterogeneity is present. This suggests that applying the random
effects model is more appropriate than the pooled OLS. Estimations of both models are
illustrated in Table 6-3.
89
Model1Model2
CoefficientsCoefficients
VariableOLSREOLSRE
Defender.020(0.55)+.020(0.51)+‐.061(‐1.88)*.100(2.08)***
Midfielder.127(3.49)***.127(3.22)***.021(0.65)+.151(3.25)***
Forward.276(6.16)***.276(5.68)***.092(2.34)**.235(4.34)***
AGE.612(2.06)***.612(17.72)***.451(17.04)***.491(17.26)***
AGE²‐.011(‐2.32)***‐.011(‐17.87)*** ‐.008(‐17.18)***‐.009(‐17.06)***
GPBL.008(2.12)***.008(16.42)***//
GPBL²‐.000(‐15.79)*** ‐.000(‐11.32)*** //
GPINT.026(1.84)***.026(11.01)***//
GPINT²‐.000(‐5.89)***‐.000(‐6.17)***//
GS.001(1.07)+.001(1.06)+//
GPBL(t‐1)//.058(16.83)***.049(14.96)***
GPBL(t‐1)²//‐.001(‐7.55)***‐.001(‐6.83)***
GPBL(priort‐1)//.002(3.93)***.002(3.39)***
GPBL(priort‐1)²//‐.000(‐2.50)**‐.000(‐1.65)*
GPINT(t‐1)//.111(7.86)***.086(6.49)***
GPINT(t‐1)²//‐.005(‐4.09)***‐.003(‐3.00)***
GPINT(priort‐1)//.012(5.05)***.005(1.85)*
GPINT(priort‐1)²//‐.000(‐2.37)**‐.000(‐0.36)+
GS(t‐1)//.040(8.40)***.043(9.36)***
GS(priort‐1)//‐.002(‐1.62)+‐.002(‐1.97)**
Birthday.000(2.30)**.000(2.35)**.000(0.78)+.000(0.92)+
SeasondummiesIncluded
TeamdummiesIncluded
AdjR²0.5060.5140.6440.640
Observations35503550
***, ** and * denote statistical significance at the 0.01, 0.05 and 0.1 level, + denotes insignificance (t/z-values in brackets).
Table 6-3: Determinants of Player Salary
Looking at the results of both models, one can see that the impact of most variables on
players’ salary is the same for all four regression analyses. Supporting the findings of
Frick (2007a) and Garcia-del-Barrio and Pujol (2005, 2006), goalkeepers earn rather
low salaries. This might be explained by their high degree of specialization. Going in
line with human capital theory, all variables indicating the experience of a player have a
90
significantly positive impact on the players’ salary with decreasing marginal returns.
Additionally, the non-impact of career goals scored in the Bundesliga in Model 1
changes to a highly significant impact of goals scored in the previous season. To test if
German players who make it to the Bundesliga despite being born shortly before the
cut-off date earn a wage premium, the results differ between the models. Supporting the
findings by Ashworth and Heyndels (2007), Model 1 suggests that players who make it
into the league despite their unfavorable birth date are indeed additionally rewarded
monetarily. But once I distinguish between recent and career performance parameters,
this wage premium vanishes. Comparing which fraction of salary variance can be
explained by the independent variables, I clearly prefer Model 2 over Model 1 as the
adjusted R-square increases considerably. In addition, it seems to be very reasonable to
distinguish between recent and previous performance expecting decreasing impact as
the time lag increases. Furthermore, as the birth date of a player should not impact his
performance on the field, there should be no impact on his salary, once I control for his
performance. Therefore, efficient markets should not give a monetary reward or
punishment for the players’ birth date.
6.5 Hazard Rates
After observing a selection bias for the market for professional soccer players in section
three, the question arises if there is a selection bias once a player has achieved to obtain
a spot on the roster of a Bundesliga team. This section will shed light, applying a hazard
rate model. For this purpose, I observe if players remain in the Bundesliga after a season
to estimate the survival rates depending on the players’ quarter of birth. Following
Frick, Pietzner and Prinz (2007), I refer to “spell duration” as the numbers of consecu-
tive seasons spent in the Bundesliga by a certain player.79 On average, German players
exhibit spell durations of 4.04 seasons, while the mean duration is considerably lower at
just 2 seasons. Figure 6-3 provides the hazard rates of German players in the Bundesliga
with regard to their position on the field. Since this data set only includes players who
left the Bundesliga before the start of the 2008/2009 season, it reduces the number of
79 Frick, Pietzner and Prinz (2007) show that estimating a similar hazard model leads to nearly identical
results, regardless of including or excluding left-censored spells.
91
observed exits to 931. For players who were already in the league before the 1995/1996
season, information about their previous career length had been added and thus I control
for left truncated cases. This leads to players exhibiting career length of more than 13
seasons.
It can be observed that forwards have on average shorter careers in the Bundesliga than
players on the other positions. Frick, Pietzner and Prinz (2007), who come to a similar
finding, explain this result by stating that the performance by forwards is easy to meas-
ure as it is often equated with the number of goals scored during the season. Hence,
their average career length would be shorter than for players playing on the other posi-
tions. For the other positions on the field there is not such a clear indicator for perfor-
mance.
Figure 6-3: Survival Rates in the Bundesliga with Regard to the Position on the Field
For the following analysis I keep up the distinction from the third section as I assort
players in four different groups according to their date of birth. As stated earlier, I
expect players’ performance on the field to be independent of their birth date. This
indeed would lead to an expected stay in the league which is not dependent on the date
of birth. Calculating the survival rates of individuals born in the first quarter after the
cut-off date, one can observe that around 28 percent of the players leave the Bundesliga
0
0,1
0,2
0,3
0,4
0,5
0,6
0,7
0,8
0,9
1
12345678910111213141516171819
Survivalrate
YearsintheBundesliga
Goalkeeper
Defender
Midfielder
Forward
92
after just one season. This value is the lowest compared to the other quarters of birth,
even though just by a small margin. The same holds true for the percentage of players
who leave the league after just two years. But as one analyzes the survival rates for
more than two seasons the percentage of players who manage to stay in the league for a
certain period of time appears to be independent of the date of birth.
To check for potential statistical significant difference in the survival rates, a Cox
(1972) semi-parametric proportional hazard model for censored data is applied. The
Cox-Model is the most general regression model developed to investigate survival data,
because it does not impose any assumptions concerning the nature of the shape of the
underlying distribution. It assumes that the underlying hazard rate is a function of the
independent variables and it solves the problem of censored observations. One
additional feature of the model is that exogenous variables can be time-constant
variables, but also, time-varying variables such as performance or age. As I expect the
survival rates to be independent of the date of birth, I test whether the survival rates for
players are dependent on the quarter of birth. This turns out not to be the case, as the
quarter of birth does not have a significant effect on the spell duration in the Bundesliga
while controlling for the performance indicators included in the salary regression.80
Results of the Cox semi-parametric proportional hazard model are presented in Table 6-
4.81
80 These results do not change as one omits the team- and season dummies.
81ResultsofaloglogisticmodellsuggeststhatresultsareinlinewiththeresultspresentedinTable6‐5.
93
Model1Model2
VariableHazardRatioz‐ValueHazardRatioz‐Value
Defender2.83(8.24)***3.27(9.01)***
Midfielder2.53(7.23)***2.98(8.22)***
Forward2.82(7.07)***3.50(8.21)***
AGE0.55(‐7.28)***0.62(‐5.69)***
AGE²1.01(7.84)***1.01(6.39)***
GPBL0.98(‐16.37)***//
GPBL²1.00(8.90)***//
GPINT1.00(‐0.35)+//
GPINT²1.00(0.63)+//
GS1.00(‐0.89)+//
GPBL(t‐1)//0.94(‐5.30)***
GPBL(t‐1)²//1.00(2.40)***
GPBL(priort‐1)//0.98(‐14.45)***
GPBL(priort‐1)²//1.00(7.18)***
GPINT(t‐1)//0.91(‐1.20)+
GPINT(t‐1)²//1.01(‐1.67)*
GPINT(priort‐1)//1.00(‐0.04)+
GPINT(priort‐1)²//1.00(1.00)+
GS(t‐1)//0.97(‐1.89)*
GS(priort‐1)//1.00(1.97)**
Birthday1.01(0.26)+1.01(0.44)+
Observations28502850
No.OfFailure931931
WaldChi21178.531227.47
***, ** and * denote statistical significance at the 0.01, 0.05 and 0.1 level
Table 6-4: Determinants of Spell Duration in the Bundesliga
For visual support of these results, Figure 6-4 shows the survival rates for players in the
Bundesliga dependent on their quarter of birth. At this stage I do not control for the
aforementioned performance indicators as I analyze the survival rates contingent on the
quarter of birth. As it can be observed that the lines in Figure 6-4, which represent the
hazard rates contingent on the quarters of birth, cross multiple times, the survival rates
can be seen as independent of the date of birth.
94
Figure 6-4: Survival Rates in the Bundesliga with Regard to the Quarter of Birth
6.6 Conclusions
The present chapter is subdivided into three parts concerning the relative age effect in
the German Bundesliga. The first part offers an analysis concerning the unequal
distribution of players in the league subject to their birth date. The data supports the
popular thesis that players who are born shortly after the cut-off date are more often
considered to be talented and hence get more support during their early playing years.
This again improves their probability to become professional soccer players in later
years and goes in line with the results of the majority of previous studies. The second
part of the chapter deals with the salary determination of the players, taking the birth
date of the players into consideration. Here I state that since the birth date of the players
should not affect the performance it should not impact their salary. Separating between
recent and career performance my data supports this hypothesis, contradicting the thesis
stated by Ashworth and Heyndels (2007), who claim that players who belong to the
youngest in their cohort and still make it to the professional level receive an additional
monetary reward. The third and final part of this chapter deals with hazard rates of
players in the Bundesliga. Going along with the proposition of the second part of the
0
0,1
0,2
0,3
0,4
0,5
0,6
0,7
0,8
0,9
1
123456789101112131415161718
Survivalrate
YearsintheBundesliga
1stQuarter
2ndQuarter
3rdQuarter
4thQuarter
95
present chapter, the birth date should not impact the players’ performance once he
achieves to obtain a roster spot in the Bundesliga and hence should not have an impact
on his chance of survival in the league. Data also supports this thesis, as the length of
stay in the Bundesliga proves to be independent of the date of birth.
In summary, the selection into the Bundesliga is affected by the date of birth, while the
implementation of cut-off dates does not impact the players’ salary or the length of stay
in the league. Still the question arises how the discriminating effect of cut-off dates can
be resolved. The literature offers a number of possible solutions like the yearly change
of the cut-off date82 or allowing the age difference in a cohort to be maximally six
month.83 These ideas would probably reduce the problem, still the organizational excess
work does not justify the change of the existing rules.
Further work using the present data set might include observing the career track of the
included athletes. Since data concerning the switching between youth teams is available,
it might be of interest at what age young soccer players change from the youth team of
their small home town to the youth team of a “big club” to be provided with better
practice. In this context the date of birth might be a factor, as I would expect players
who are born closely after to the cut-off date to be identified as talents at an earlier stage
than late-born players. This proposition is based on the fact that the relative age differ-
ence decreases as the age of the players increases. Furthermore, the date of the first-time
nomination to a German national youth team might also be influenced by the players’
birth date, especially since these teams are selected from a considerably larger cohort.
82 See Ashworth and Heyndels (2007).
83 See Billari and Pellizzari (2008).
96
7 Sabotage in Heterogeneous Tournaments: A Field Study
7.1 Introduction
In practice, tournaments or contests are ubiquitous. A typical example is an internal
labor market tournament, in which employees compete for a bonus or a promotion
(Lazear and Rosen 1981). Other examples include R&D races, litigation contests, rent-
seeking contests, political campaigns or sports contests.84
In tournaments, only the relative performance of the contestants matters. The partici-
pants therefore have an incentive to decrease the performance of their opponents. This is
typically called the “sabotage problem”. In the meantime, a number of theoretical con-
tributions on sabotage in tournaments have appeared (e.g. Lazear 1989, Konrad 2000,
Chen 2003, 2005, Kräkel 2005, Münster 2007, Gürtler 2008 or Gürtler & Münster
2008). Unfortunately, however, there are only a few empirical studies. Garicano and
Palacios-Huerta (2005) for instance show that an increase in the price difference in pro-
fessional soccer (from two to three points for a win) has led to more sabotage (as
measured by the number of defenders and the number of disciplinary actions). Drago
and Garvey (1998) find that in Australian companies employees help each other less if
they are paid according to their relative performance.85 Harbring et al. (2007) find in a
laboratory experiment that subjects tend to retaliate against sabotage activities if the
identity of the saboteur is known. Moreover, they are able to show that sabotage is more
pronounced if the identity of the saboteur is unknown since retaliation is then less
strong.86
The aim of this chapter is to extend both, the theoretical and the empirical literature on
sabotage in tournaments. We consider a tournament between two heterogeneous play-
ers, both of whom may take legal and illegal actions to increase their probability of
winning. We impose two assumptions: First, the favorite is more productive than the
underdog with respect to legal actions. Second, the players’ cost function is such that
84 See Frick (2003) or Konrad (2007) for surveys.
85 Note that help can be interpreted as the opposite of sabotage.
86 See Harbring and Irlenbusch (2004, 2005) for further experiments on sabotage in tournaments.
97
both activities are substitutes. We then show that the favorite chooses a higher level of
legal and a lower level of illegal actions than the underdog. In other words, the favorite
is sabotaged more strongly than the underdog. This complements the theoretical lite-
rature on sabotage in asymmetric tournaments, where the favorite is sabotaged the most
because he is the truly dangerous rival (Chen 2003, Münster 2007). This latter finding,
however, requires at least three contestants. In our model, we show that the favorite is
subject to most illegal activities even if there are only two players, a result that has not
been derived before.
In a second step, we perform an empirical analysis to test our derived results. Similar to
Garicano and Palacios-Huerta (2005) we use data from professional soccer. As a proxy
for legal activities we use the percentage of fair, i.e. successful, tackles of a team, while
sabotage is measured by the percentage of fouls a team has committed. We further use
betting odds to determine the favorite and the underdog in a particular match. Control-
ling for a number of additional factors that may have an impact on legal and illegal
activities we confirm both our theoretical results. In particular, we find that a team fouls
less often and wins more tackles in a fair way the more it is favored by the betting odds.
Altogether, we find that sabotage is a serious problem in heterogeneous tournaments.
We show that players indeed take actions to reduce the performance of their opponents.
Apart from this direct negative effect, this shifts their focus away from own legal
actions. Finally, if tournaments are also used for selection purposes, sabotage increases
the probability of selecting (and promoting) the wrong player. As the favorite is subject
to most illegal actions, sabotage increases the likelihood for the underdog to win the
tournament. For an employer designing an intra-company promotion tournament it is
thus important to prevent the employees from engaging in sabotage activities. One way
to achieve this is to punish detected sabotage harshly. Another way is to make it more
difficult for the employees to sabotage each other. The employer may for instance not
reveal the identities of the contestants. Moreover, he may organize a tournament
between employees who are separated from each other (i.e. they are working in differ-
ent business units or departments in different cities).
98
The chapter is organized as follows: The next section presents our model and the theo-
retical findings. In section 3, we describe our data. Section four contains the empirical
results, and section five concludes.
7.2 The Model
7.2.1 Description of the Model and Notation
Consider a situation, in which two risk-neutral players, i=1, 2, compete in a tournament.
The player with the higher performance is declared the winner and receives the winner’s
prize 0>w, while the loser’s prize is normalized to zero. We focus on actions that are
intended to decrease the opposing player’s performance. This can either be done
„fairly”, i.e. by actions that are in accordance with the rules of the game, or „unfairly”,
i.e. by violating the rules of the game.87 We denote by 0≥
i
e a player’s fair effort and
by 0≥
i
s his unfair or sabotage effort. The players’ performances are then given by
(1) 12211
ε
+
−
−= bseyy and 21122
ε
+
−
−
=
bsaeyy ,
with i
y as the players’ gross performance from their (unmodeled) productive decisions,
a, b as productivity parameters and i
ε
as a random term being independently
distributed according to pdf )(⋅
i
f. Let the pdf of 12
:
ε
ε
ε
−
=
be defined as )(⋅g and
denote the corresponding cdf as )(
⋅
G.
We assume the players to be heterogeneous. In particular, we have 1>a, ba > and
0: 21 ≥−= yyy
Δ
. This means that player 1 is better in reducing his opponent’s output
in a fair way and also has a weakly higher productive performance.88 In the following,
we therefore speak of player 1 as the favorite and player 2 as the underdog.
87 We introduce two efforts decreasing the opponent’s performance to keep the model in line with the
empirical part of the chapter, where the available proxy variables also decrease the opposing team’s
performance. Note, however, that the model results would continue to hold, if e would describe a
productive effort increasing the own performance. This is demonstrated in Appendix A.
88 As we will show, the latter assumption does not affect the qualitative results of our model.
99
Both types of effort are costly to a player and these costs are given by the function
),( iii seCC =. The function ),( ii seC is assumed to be symmetric89 and to satisfy90
0
),( ≥
∂
∂
i
ii
e
seC , 0
),(
2
2
>
∂
∂
i
ii
e
seC and 0
),(
2
>
∂∂
∂
ii
ii
se
seC , for 0, >
ii se . The last assumption
implies that the players’ actions are substitutes. Thus, if a player decides to increase his
fair effort, he decreases his unfair effort and vice versa. For the application we address,
this assumption seems very suitable. When trying to stop the opponent, a player can
either exert effort to do this fairly or he can just foul the opponent, in which case he has
to accept the consequences (e.g. a free kick, a yellow or a red card). The more fouls a
player commits in a game, the less fairly he behaves. Thus, it is reasonable to assume
costs to be such that both kinds of effort are substitutes.
Finally91, we impose the Inada-conditions 0
),0( =
∂
∂
i
i
e
sC and ∞=
∂
∂
i
ii
e
seC ),( , for
∞→
i
e.
These conditions imply that we have an interior solution, where optimal efforts are
strictly positive, but finite.
89 One may argue that the marginal costs of fair and unfair actions are different. Different marginal costs,
however, have a similar effect as different marginal productivities of the actions in the performance
functions. It is, therefore, sufficient to assume that b is different from 1.
90 Note that the symmetry of the cost function implies that we obtain the same conditions if we consider
the partial derivatives with respect to the sabotage activity.
91Again, the symmetry of the cost function implies that we obtain the same conditions for the sabotage
activity.
100
7.2.2 Solution to the Model
In the tournament, each player chooses i
e and i
s so as to maximize his expected payoff.
This payoff is given by
(2)
{
}
{}
),())((
),()(
),(
112121
1121212112
11211212211
seCwssbeaeyG
seCwssbeaeyyrobP
seCwbsaeybseyrobPEU
−−+−+=
−−+−+−<−=
−
+
−
−
>
+
−−=
Δ
εε
ε
ε
for player 1 and
(3) ),())]((1[ 2221212 seCwssbeaeyGEU
−
−
+
−+−=
Δ
for player 2.
The optimality conditions are given by92
(4) 0
),(
))((
1
*
1
*
1
*
2
*
1
*
2
*
1
1
1=
∂
∂
−−+−+=
∂
∂
e
seC
wssbeaeyag
e
EU
Δ
(5) 0
),(
))((
1
*
1
*
1
*
2
*
1
*
2
*
1
1
1=
∂
∂
−−+−+=
∂
∂
s
seC
wssbeaeybg
s
EU
Δ
92 A sufficient condition for this first-order approach to be valid is that the players’ objective functions are
strictly concave. This requires 0
2
2
<
∂
∂
i
i
e
EU and 0)( 2
2
2
2
2
2
>
∂∂
∂
−
∂
∂
∂
∂
ii
i
i
i
i
i
se
EU
s
EU
e
EU (see Sydsaeter,
Hammond, Seierstad and Strom 2005: 55). For player 1, for instance, these conditions can be written
as
0
),(
))((' 2
1
11
2
2121
2<
∂
∂
−−+−+ e
seC
wssbeaeyga
Δ
and
.0]
),(
))(('[
]
),(
))(('][
),(
))(('[
2
11
11
2
2121
2
1
11
2
2121
2
2
1
11
2
2121
2
>
∂∂
∂
−−+−+−
∂
∂
−−+−+
∂
∂
−−+−+
se
seC
wssbeaeyabg
s
seC
wssbeaeygb
e
seC
wssbeaeyga
Δ
ΔΔ
This is e.g. fulfilled, if 2
1
11
2),(
e
seC
∂
∂ is everywhere high enough. In Section 2.3, we consider a specific
example where these conditions are always fulfilled.
101
(6) 0
),(
))((
2
*
2
*
2
*
2
*
1
*
2
*
1
2
2=
∂
∂
−−+−+=
∂
∂
e
seC
wssbeaeyg
e
EU
Δ
(7) 0
),(
))((
2
*
2
*
2
*
2
*
1
*
2
*
1
2
2=
∂
∂
−−+−+=
∂
∂
s
seC
wssbeaeybg
s
EU
Δ
Using the optimality conditions (4) to (7), we can derive the following proposition:
Proposition 1: In equilibrium, the favorite exerts higher fair effort than the underdog
but sabotages less (i.e. *
2
*
1ee > and *
2
*
1ss <). Moreover, if 1≥b ( 1<b), we have
*
2
*
2se ≤ ( *
2
*
2se >).
Proof: See Appendix B.
According to Proposition 1 the favorite always chooses a higher fair effort than the
underdog, but sabotages less. Two effects drive this result. First, the favorite has a
higher return on fair effort which, in turn, induces him to engage more strongly in the
legal activity. Second, both types of effort are substitutes. This means that a player
engaging more strongly in one activity engages less strongly in the other one. Accor-
dingly, player 1, choosing a higher fair effort, sabotages less.
In Sections three and four, we are going to test these results using data from German
professional soccer. Before we do so, however, we consider a parameterized version of
the tournament model to develop a better understanding of the sabotage problem.
102
7.2.3 Parameterized Version of the Model
In this section, we assume specific forms for our functions to be able to derive some
closed-form solutions for the optimal efforts. To keep the model tractable, we assume
the composed random variable 12
ε
ε
−
to be uniformly distributed on ],[ uu−, with
u>093, and effort costs to be given by )5.05.0(),( 22
iiiiii skesecseC ++= , with 0>c
and )
1
,min,0( ⎭
⎬
⎫
⎩
⎨
⎧
∈ba
b
k94 . Then, the optimality conditions simplify to95
(4’) 0
2
*
1
*
1=−− cksce
u
aw
(5’) 0
2
*
1
*
1=−− ckecs
u
bw
(6’) 0
2
*
2
*
2=−− cksce
u
w
(7’) 0
2
*
2
*
2=−− ckecs
u
bw
Solving these conditions simultaneously, optimal efforts can be written as96
(8) )1(2
)(
,
)1(2
)1(
,
)1(2
)(
,
)1(2
)(
2
*
2
2
*
2
2
*
1
2
*
1kcu
kbw
s
kcu
bkw
e
kcu
akbw
s
kcu
bkaw
e−
−
=
−
−
=
−
−
=
−
−
=
From these conditions, it is straightforward to see that all efforts are increasing in the
prize w and decreasing in u (which measures how strongly the tournament outcome is
influenced by factors beyond the players’ control) and the cost parameter c. Moreover,
*
1
e ( *
1
s) is increasing in a (b) and decreasing in b (a). This is intuitive. If player 1
93 If
ε
1 is a constant v and
ε
2 is uniformly distributed on [-u+v,u+v], the composed random variable
ε
2-
ε
1
would be uniformly distributed on [-u,u].
94 See for this kind of cost function e.g. Itoh (1994). Note that the function does not fulfill the Inada-
conditions described above. We will nevertheless see that the solution is an interior one.
95 The sufficient conditions for optimality are –c<0 and c2-(ck)2>0. Under the assumptions imposed on the
parameters c and k, these conditions are always fulfilled.
96 Note that the assumption ⎭
⎬
⎫
⎩
⎨
⎧
<ba
b
k1
,min ensures that all equilibrium efforts are strictly positive.
103
becomes more productive with respect to his fair (unfair) activity, he increases his fair
(unfair) effort and, since both efforts are substitutes, decreases his unfair (fair) effort. A
similar argument applies to the second player. Finally, it is not clear how optimal efforts
react to changes in k. On the one hand, an increase in k leads to higher effort costs, thus
reducing efforts. On the other hand, it may yield a substitution of one kind of effort by
the other. Therefore, efforts may either increase or decrease.
From the optimal efforts, we are able to compute the players’ winning probabilities. We
can show that player 1 wins with probability
(9) ))21(
)1(2
)1(
())(( 2
*
2
*
1
*
2
*
1bka
kcu
aw
yGssbeaeyG −+
−
−
+Δ=−+−+Δ
It is straightforward to see that player 1 is more likely to win, if y
Δ
or a gets higher and
c or u gets lower. All this is intuitive. Moreover, we can see that player 1’s winning
probability is also increasing in w: while both players increase their efforts as a reaction
to a higher price, the first player’s increase is more effective in influencing the winning
probability. As k does not have a clear-cut effect on the efforts, its effect on the win-
ning-probability is ambiguous, too. Finally, player 1 becomes less likely to win the
tournament if b gets higher. This has the following implication: If the tournament orga-
nizer were able to completely rule out unfair behavior (which is the same as setting b
equal to zero), the more able player were most likely to win. Hence, if tournaments are
also used for selection purposes, sabotage increases the probability of selecting the
wrong player. As the favorite is subject to more illegal actions, sabotage increases the
likelihood for the underdog to win the tournament.
7.3 Data Set and Descriptive Statistics
In the following section we empirically test our proposition that in a contest with two
heterogeneous players the favorite prefers legal (“fair”) actions while the underdog pre-
fers illegal (“unfair”) activities to increase his probability of winning. Our empirical
investigation is based on a data set from the German “Bundesliga” covering a 2.5 year
104
period from the start of the 2005/06 season until the middle of the 2007/08 season. For
that period we have compiled complete and detailed information on all the 765 matches
(2005/06: 306 matches; 2006/07: 306 matches; 2007/08: 153 matches). The data come
from the league’s homepage (www.bundesliga.de), the highly respected soccer maga-
zine “Kicker” (www.kicker.de) and from the largest betting company in the country
(www.oddset.de).
Most important in our context is the question of how to operationalize the dependent
and independent variables. Since the act of taking the ball away from an opposing
player by either kicking it away or by stopping that player with one’s own feet is either
considered a successful tackle or a foul, we distinguish between these two outcomes as
being the results of either “fair” or “unfair” efforts. Hence we define as “constructive”
or “fair” effort the home team’s number of successful tackles divided by the total num-
ber of tackles during a match tacklessuccessfulofnumbertotal
teamomehoftacklessuccessfulofn . Our preferred meas-
ure of “destructive” or “unfair” effort is the number of fouls committed by the home
team divided by the total number of fouls during a particular match
foulsofnumbertotal
teamomehoffoulsofn .97
Following Fama (1970), we assume the betting market to be efficient in the sense that
prices (the odds) fully reflect all available information. He distinguishes between three
different forms of test depending on the information used: “Weak form” tests use past
prices only, “semi-strong” tests use all publicly available information and “strong” tests
also include information that is only accessible to certain people (see Kuypers 2000). In
betting markets semi-strong efficiency implies that the incorporation of publicly avail-
able information on the two teams’ playing strength, their current form, player injuries,
etc. should not improve the accuracy of outcome predictions based on odds. In the
empirical part of this chapter we therefore use betting odds to distinguish between the
favorite and the underdog to analyze the effect of heterogeneity (i.e. match uncertainty)
on the choice of fair and unfair effort. Our betting odds are from “Oddset”, which pub-
97Note that the number of tackles and the number of fouls committed by the away team is accounted for
in the respective denominator.
105
lishes these figures a few days before the weekend (usually on Tuesday). Thus, all bets
are fixed in the sense that information regarding the fitness of key players, injuries,
playing strategy, etc. that becomes available between Tuesday and Saturday (when most
of the matches are played) will have no impact on the odds.
For example, in week 16 of the 2007/08 season, Bayern Munich played MSV Duisburg
at home and was heavily favored by the bookmakers. In case of a home win bettors
received an amount of 1.10 € for every € they had placed on the home team. In the case
of a draw the payoff to bettors was 5.50 € and in the unlikely case of an away win the
payoff would have been 10.00 €. Summing up the inverse of the quotes not only yields
the mark-up of the betting company, but also allows computation of the implicit proba-
bilities of a home win, a draw and an away win.98
The payout ratio is determined as follows:
winawayPayoffdrawPayoffwinmehoPayoff
ratioPayout 111
1
++
=,
which in the case of the above mentioned game resulted in a payout ratio of 0.8397.99
To calculate the implicit probabilities of the possible outcomes, we divide the payout
ratio by the payoffs associated with the respective outcomes. This leads to an implicit
probability of a home win of 76.3 percent, while the implicit draw and away win proba-
bilities are 15.3 percent and 8.4 percent respectively. To measure the heterogeneity
between the opponents, we now introduce “HET” which will later serve as our main
exogenous variable. HET is calculated as the difference between the implicit winning
probabilities of the opposing teams:
AH PPHET
−
=
,
98 Sauer (1998) provides a detailed analysis of the economics of wagering markets. Forrest, Goddard and
Simmons (2005) demonstrate the (increasing) effectiveness of the betting market using data from
English football.
99 This implies that the bookmaker’s mark-up is about 16 percent - a value that is quite high compared to
most online bookmakers. Using data from www.gamebookers.com Stadtmann (2006) calculates an
average mark-up of only 12 percent.
106
where PH and PA represent the implicit winning probabilities of home and away team.
Hence HET can theoretically vary between -1 and +1. Positive values indicate that the
home team is the favorite while in the case of negative values the away team is the favo-
rite. The further away HET is from zero, the more heterogeneous the two teams facing
each other are. The overall (expected) home advantage is obvious from our Figure 7-1,
as the mean of HET is 0.152.
As stated in our proposition in the previous section, we expect the favorite to exert a
higher level of fair effort and a lower level of sabotage compared to the underdog, as we
measure fair effort as the percentage of successful tackles and sabotage as the percen-
tage of fouls. Obviously, we do not only have to control for the degree of heterogeneity
of the opposing teams but also for some other factors that may have an influence on the
effort levels chosen by the two clubs. These controls are necessary in order to eliminate
alternative explanations for either team’s choice of fair and unfair effort. Thus, we
include the natural logarithm of the number of spectators (ATT) to control for the
atmosphere in the arena as well as a dummy-variable DERBY, which takes a value of 1
if the respective match is one between two long-time rivals whose home grounds are
located closely to each other.
VariableDescriptionMeanMin.Max.
FOULSPercentageofFoulsbyHomeTeam(UNFAIR)0.4780.2080.792
TACKLESPercentageofTacklesWonbyHomeTeam(FAIR)0.5140.3110.683
ATTAttendance401051091481264
DERBYMatchisaDerby0.02401
HWPWinningProbabilityofHomeTeam0.4370.1490.763
AWPWinningProbabilityofAwayTeam0.2850.0840.609
HETHWP–AWP0.15‐0.460.68
ODDSLogoddsofHWP–AWP0.06‐0.790.77
Table 7-1: Descriptive Statistics
107
Table 7-1 presents the descriptive statistics of the variables used in the estimations
(Kernel density estimates of the two dependent variables (HET and log odds of HET)
are displayed in Appendix C)).
7.4 Empirical Results
The main purpose of the empirical section of this chapter is to investigate whether
weaker teams have an incentive to choose more unfair and unconstructive effort, while
stronger clubs have a preference for fair and constructive effort. We employ two
different estimation techniques which produce more or less identical results,
documenting the robustness of our findings.
We start with two conventional OLS-estimations where the percentage of fouls by the
home team (FOULS) and the percentage of successful and fair tackles won by the home
team (TACKLES) respectively are the dependent variables. These specifications have
the advantage that we can control for systematic effects on the residual errors; i.e. the
estimation errors are robust (White 1980) and clustered at the match level. While this
procedure seems to produce conclusive results, we have not yet paid attention to the fact
that the two endogenous variables (FOULS/TACKLES) are proportions or percentage
variables with values must ranging by definition from zero to one. In order to obtain
unbiased estimates a traditional solution to this problem is to perform a logit transfor-
mation of the variables ((y → log(y/(1-y)))). Following such a transformation it is then
possible to use the OLS method.
108
Variable
Model1Model2
(OLS)LOGIT‐TRANS(OLS)
TACKLESFOULSTACKLESFOULS
log(ATT)0.004(0.96)+‐0.002(‐0.27)+0.017(0.95)+‐0.010(‐0.32)+
DERBY‐0.001(‐0.09)+0.003(0.16)+‐0.004(‐0.10)+0.012(0.18)+
HET0.027(3.00)***‐0.070(‐4.65)***0.111(3.02)***‐0.290(‐4.65)***
REFEREE‐
DUMMIESIncludedIncluded
CONST0.491(10.25)***0.459(5.97)***‐0.321(‐0.17)+‐0.150(‐0.47)+
Adj.R20.040.050.040.05
F‐Test1.30+1.97*1.37+1.98**
NofCases765765765765
***, ** and * denote statistical significance at the 0.01, 0.05 and 0.1 level
Table 7-2: Determinants of Fair and Unfair Behavior
Note: Both models are estimated with White robust standard errors. 21 referee dummies
are included in the estimations (Manuel Graefe is the reference referee. The full results
are available upon request).
Table 7-2 displays the contribution of the independent variables in explaining the
variance of the two left-hand side variables. It appears that only a small fraction of that
variation can be explained by our set of exogenous parameters. The number of specta-
tors and the character of a match as a “neighborhood duel” do not have any statistically
significant impact on the “intensity” of the match. Since a number of recent studies have
documented a tendency of referees to favor the home team (Sutter and Kocher 2004,
Garicano, Palacios-Huerta and Prendergast 2005, Petterson-Lidbom and Priks 2009,
Buraimo et al. 2007, Scoppa 2008, Dohmen 2005)100, we initially expected to find that
the higher the attendance in a stadium, the more biased the referee’s decision-making
towards the home team will be.101 Moreover, we expected that type of bias to be partic-
ularly strong in local derbies with their peculiar “atmosphere”. It turns out, however,
that the coefficient of these two match-variables have the “correct” sign in all of our
100 Moreover, a number of recent studies have demonstrated that referees react to financial incentives in
the same way as “ordinary” people by adjusting their effort levels (see Frick et al. 2009, Rickman and
Witt 2008).
101 Replacing the number of spectators by the capacity utilization leaves the findings virtually unaffected.
The results are, of course, available from the authors upon request.
109
estimations, but their magnitude is not statistically significant or economically rele-
vant.102
Most important in our context is the coefficient of our key independent variable, the
heterogeneity measure HET. Since HET measures the difference in the playing strength
of the two opposing teams, the highly significant coefficients in the TACKLES- as well
as in the FOULS-model support our theoretically derived proposition that favorites
usually exert higher levels of “legal” effort while the respective underdogs’ best answer
is to engage in “illegal” activities.103
7.5 Conclusion
The present chapter consists of two different, yet closely related parts. In the first (theo-
retical) section we develop a formal model demonstrating that the favorite in a tourna-
ment chooses legal activities, while the underdog is tempted to engage in illegal activi-
ties. The reasons for this are twofold: First, the favorite is more productive with respect
to legal activities and, second, the two types of activities are substitutes. In a second
(empirical) section we demonstrate the plausibility of the theoretical model by using
match-level data from German professional soccer. In particular, we find that teams that
are more likely to win a match (as measured by the respective betting odds) win signifi-
102 Since both left-hand side variables (FOULS and TACKLES) are determined by the same set of
explanatory variables, it may be advisable to estimate a seemingly unrelated regression model (Zellner
1962). The SURE approach takes into account the potential correlation of the residuals in the two
models. If such a correlation exists, simple OLS-regressions are likely to produce inefficient estimates
(Frick 2004). Although the findings of the SURE specification are identical to the ones from the OLS-
models, the Lagrange-Multiplier Test (Breusch and Pagan 1980) of (χ2 = 210.1 p<.001) indicates that
the correlation of the residuals in the FOULS and the TACKLES-equations is highly significant.
Nevertheless, the relevant coefficients are not at all affected by that correlation. Moreover, we have
also estimated a random-effects version of the SURE approach as developed by Biörn (2004). While
most coefficients remained at their initial level, the statistical significance of our heterogeneity
measure decreased in the TACKLES-model (which is due to some unobserved referee specific
heterogeneity, such as differences in cognitive skills). The findings are available from the authors
upon request.
103 It may also be argued that the logit-transformation is not always feasible and the coefficients are
difficult to interpret if the dependent variable takes on values of 0 and 1 (Wooldrigde 2001: 661). We
therefore also employed the “fractional logit estimator” developed by Papke and Wooldridge (1996)
to get rid of that particular problem. In that latter estimation the coefficient of our heterogeneity
measure increases dramatically compared to the OLS-estimation (as well as the SURE-model). These
results are also available from the authors upon request.
110
cantly more tackles in a fair way and, at the same time, commit significantly fewer
fouls.
Intra-company tournaments are typically used to induce incentives and to select the
most able contestants. Our findings indicate that either of these two objectives may be
difficult to reach. On the one hand, we find that weaker contestants (“underdogs”) tend
to engage in sabotage which shifts their focus away from legal activities and thus
reduces the incentive effects of tournaments. On the other hand, stronger contestant
(“favorites”) are sabotaged more heavily which, in turn, reduces the selection efficiency
of tournaments. For an employer designing an intra-company promotion tournament it
is therefore important to prevent the employees from engaging in sabotage activities.
One way to achieve this is to punish detected sabotage harshly. Another way is to make
it more difficult for the employees to sabotage each other. The employer may for
instance not reveal the identities of the contestants. Moreover, he may organize a tour-
nament between employees that are locally separated, for example in units in different
cities.
From time to time the organizers of professional team sports leagues in general and of
soccer leagues in particular (i.e. the respective national associations) instruct their refe-
rees to punish certain types of fouls more harshly.104 In this sense, sabotage should
become more costly and the teams should sabotage less. An interesting array for future
research would therefore be to analyze whether these measures indeed affect behavior.
This may offer some indication on the effectiveness of punishments in tackling the
sabotage problem.
104 For instance, referees are nowadays advised to punish tackling from behind with a red card, the
strongest punishment a referee can inflict on a player.
111
7.6 Appendix A
In this appendix, we show that all the results of our model continue to hold, if e
describes a productive effort increasing the own performance. To see this, consider the
same model as before with the exception that the players’ performances are now given
by
(1’) 12111
ε
+−+= bsaeyy and 21222
ε
+
−
+
=
bseyy
Then, player 1’s winning-probability can be written as
(10)
{}
{}
))((
)(
2121
21212112
21221211
ssbeaeyG
ssbeaeyyrobP
bseybsaeyrobP
−+−+=
−+−+−<−=
+
−
+
>+−+
Δ
εε
ε
ε
This is the same winning probability as in the original model. As nothing else was
changed, all our results continue to hold. Q.E.D.
7.7 Appendix B
In this appendix, we present the proof of Proposition 1. We restrict attention to the case
where 1<b. The proof in the case where 1≥b is completely analogous.
From (4) and (5), we have
1
*
1
*
1
1
*
1
*
1),(
1
),(
1
s
seC
be
seC
a∂
∂
=
∂
∂. As ba >, we cannot have
*
1
*
1se =. Similarly, a comparison of (6) and (7) shows that we cannot have *
2
*
2se =.
Note further that we cannot have *
1
*
1se < since then player 1 could deviate to *
11
~se =
and *
11
~es =. Such a deviation would leave the player’s effort cost unchanged, while the
player’s winning probability would increase from ))(( *
2
*
1
*
2
*
1ssbeaeyG −+−+
Δ
to
112
))(( *
2
*
1
*
2
*
1sebeasyG −+−+
Δ
. Thus, the deviation would be profitable. We therefore
have *
1
*
1se > and with a similar argument *
2
*
2se >.
Suppose now that *
2
*
1ee = and *
2
*
1ss =. This, however, contradicts the condition
2
*
2
*
2
1
*
1
*
1),(),(
1
e
seC
e
seC
a∂
∂
=
∂
∂, which we obtain from (4) and (6). The ordering *
2
*
1ee ≤ and
*
2
*
1ss ≤ (with one inequality being strict) also contradicts this condition. Similarly,
*
2
*
1ee ≥ and *
2
*
1ss ≥ (with one inequality being strict) contradicts
2
*
2
*
2
1
*
1
*
1),(),(
s
seC
s
seC
∂
∂
=
∂
∂, which follows from a comparison of (5) and (7).
Consider now the case *
2
*
1
*
1
*
2ssee ≥>≥ . As each player’s choice is (expected) utility-
maximizing, a player prefers his own choice to the other player’s choice. Hence, it must
be that
(11) ),()(),())(( *
2
*
2
*
2
*
2
*
1
*
1
*
2
*
1
*
2
*
1seCweaeyGseCwssbeaeyG −−+≥−−+−+
ΔΔ
and
(12) ).,()](1[),())]((1[ *
1
*
1
*
1
*
1
*
2
*
2
*
2
*
1
*
2
*
1seCweaeyGseCwssbeaeyG −−+−≥−−+−+−
ΔΔ
Transforming (11) and (12) yields
(13) )]())(([),(),( *
2
*
2
*
2
*
1
*
2
*
1
*
2
*
2
*
1
*
1eaeyGssbeaeyGwseCseC −+−−+−+≤−
ΔΔ
and
(14) )]())(([),(),( *
1
*
1
*
2
*
1
*
2
*
1
*
2
*
2
*
1
*
1eaeyGssbeaeyGwseCseC −+−−+−+≥−
ΔΔ
.
Combining these conditions, we obtain
(15)
)()(
)]())(([
)]())(([
*
2
*
2
*
1
*
1
*
1
*
1
*
2
*
1
*
2
*
1
*
2
*
2
*
2
*
1
*
2
*
1
eaeyGeaeyG
eaeyGssbeaeyGw
eaeyGssbeaeyGw
−+≥−+⇔
−+−−+−+
≥−+−−+−+
ΔΔ
ΔΔ
ΔΔ
113
Assuming *
1
*
2ee ≥, this condition can only hold if *
2
*
1ee =. Together with
2
*
2
*
2
1
*
1
*
1),(),(
s
seC
s
seC
∂
∂
=
∂
∂, this implies *
2
*
1ss =. As before this leads again to a contra-
diction.
Hence, the only case that remains has *
1
*
2
*
2
*
1ssee ≥>≥ . Here, we have to show that nei-
ther *
2
*
1ee ≥ nor *
1
*
2ss ≥ can be binding (recall that both conditions cannot be binding at
the same time). First, let *
2
*
1ee =, but *
1
*
2ss >. Again, this contradicts the condition
2
*
2
*
2
1
*
1
*
1),(),(
1
e
seC
e
seC
a∂
∂
=
∂
∂. Second, let *
2
*
1ee > and *
1
*
2ss =. This contradicts the con-
dition
2
*
2
*
2
1
*
1
*
1),(),(
s
seC
s
seC
∂
∂
=
∂
∂. Thus, in equilibrium it must be the case that
*
1
*
2
*
2
*
1ssee >>> . Q.E.D.
114
7.8 Appendix C
Figure 7-1: Epanechikov-Kernel Density Estimation of HET
0.2 .4 .6
Density
-2 -1 0 1 2
HET
Kernel density estimate
Normal density
115
The D’Agostino et al. (1990) test (implemented in Stata 10.1 as “sktest”) reveals that in
the case of HET the distribution deviates from normality. However, presence of nor-
mality is indicated by a rather small kurtosis and an insignificant value of the
D’Agostiono et al. (1990) test in the case of log odds of HET.
0.5 11.5 2
Density
-1 -.5 0.5 1
odds
Kernel density estimate
Normal density
Figure 7-2: Epanechikov-Kernel Density Estimation of log Odds of HET.
116
8 The Economics of the World Cup
8.1 Introduction
The FIFA World Cup is an international soccer tournament played every four years in a
different country to the previous location. FIFA stands for Fédération Internationale de
Football Association, and is the governing body of world soccer. In addition to holding
the rights to host the World Cup, and owning the World Cup brand, FIFA is responsible
for the rules of soccer, as played at both professional and amateur levels worldwide. The
World Cup is competed by teams representing countries, more specifically national
soccer associations that are officially recognized by the organizers. The first World Cup
was in 1930 and the tournament has been held every four years since, with breaks for
wartime disruption in 1942 and 1946. Although the specific format of the Finals has
varied, the basic concept remains that a large number of national teams compete in a
qualifying tournament organized around regional football associations, for the right to
participate in the World Cup Finals. In 2010, 32 teams were awarded places in the
Finals, played in South Africa between June 11 and July 11 2010. One place is tradi-
tionally reserved for the host nation (two places if there are co-hosts as in the case of
Japan and South Korea in 2002). The 2010 Finals comprised a round-robin league com-
petition of eight divisions of four teams each, with places allocated by complex seeding
principles; this was followed by a knock-out tournament involving single games, lead-
ing eventually to the Final. The World Cup Final is the planet’s most-viewed sports
event. The 2006 Final in Germany played between Italy and France, drew an estimated
75 million television viewers. FIFA estimated that an overall total of $3.4 billion would
be generated from proceeds of the most recent World Cup Finals (www.sportcal.com).
This would represent an increase from $2.6 billion in the 2006 World Cup held in Ger-
many (Maennig and du Plessis, 2007).
This Chapter will present a general review of the economics of the FIFA World Cup.
We shall proceed as follows. Section 1 explores the method by which host countries are
selected by FIFA for the rights to organize the World Cup finals. Section two explores
the benefits to host countries from organizing the World Cup finals. Since the literature
117
consensus that the direct benefits are at best modest we move on to consider intangible
benefits to host country residents from the World Cup finals. In Section three we
consider the later benefits to soccer fans in a host country from new stadium
infrastructure and other legacies of hosting the World Cup finals. Section four turns our
attention to the players participating in the World Cup finals, examining their direct
remuneration from participation plus later benefits in terms of career advantages. This
Section will present new evidence from the German Bundesliga of salary premia to
players associated with World Cup participation. In Section 5, we offer some views on
attempts by some national soccer associations to restrict imports of foreign players,
sometimes with the explicit goal of using import controls as a means of promoting
national success in tournaments such as the World Cup.
8.2 Selection of Host Countries for the World Cup Finals
FIFA is essentially an umbrella organization, comprising six confederations which in
turn represent national associations. The six confederations are Africa (CAF), Asia
(AFC), Europe (UEFA), North America, Central America and the Caribbean
(CONCACAF), Oceania (OFC) and South America (CONMEBOL). Places for the
World Cup Finals are allocated by confederations so, for example, UEFA sent nine
teams to the 2010 Finals. On the executive committee, UEFA has the largest represen-
tation with nine seats out of 24, followed by CONMEBOL, AFC and CAF with four
each.
The choice of host country for the World Cup Finals is made by the FIFA Executive
Committee from a set of national association bids. In several respects, the choice
mechanism has some similarities with that of host city for the Olympic Games, also
held every four years, but not in the same year as the World Cup Finals. Hosting the
World Cup Finals is a right that the local soccer federation buys off FIFA. Similar to the
Olympic Games, FIFA sets up an auction for the rights to hold the World Cup Finals
every four years. Also similar to the Olympic Games, FIFA extracts economic rent due
to its position as monopoly provider of the tournament. FIFA does not take a fee from
the successful bidder but instead insists on a number of favorable contract provisions
118
which can be judged to be components of economic rent. For example, Maennig and du
Plessis (2007) report that FIFA insisted that, for the 2010 Finals in South Africa, adver-
tising billboards within 1 kilometer of stadia where Finals games were played, and
along access roads to such stadia, should be restricted to FIFA-endorsed enterprises.
Profits from such advertising went to FIFA. A similar rule applied to the 2006 Finals in
Germany. From 1958 to 2006, FIFA followed a rotation policy, although this was only
ever explicit from 2000. Under this policy, the Finals would be held every eight years in
Europe, alternating with a venue from another continent (see Table 8-1). Ostensibly,
this policy was designed to promote soccer in continents and countries where soccer
leagues and soccer participation were less well-established. The choice of South Korea
and Japan as co-hosts in 2002 could be rationalized in this way.
YearHostWinner
1930UruguayUruguay
1934ItalyItaly
1938FranceItaly
1950BrazilUruguay
1954SwitzerlandWestGermany
1958SwedenBrazil
1962ChileBrazil
1966EnglandEngland
1970MexicoBrazil
1974WestGermanyWestGermany
1978ArgentinaArgentina
1982SpainItaly
1986MexicoArgentina
1990ItalyGermany
1994USABrazil
1998FranceFrance
2002SouthKorea/JapanBrazil
2006GermanyItaly
2010SouthAfrica
2014Brazil
Table 8-1: Hosts and Winners of the World Cup Finals since Origin
119
A counter-argument to the rotation policy is that FIFA has taken the World Cup Finals
to countries where soccer is always likely to be a minority sport (USA) or where a club
league is always going to be small in size and audience interest (South Africa). Con-
versely, the continent where soccer is most intensely followed as the dominant team
sport is of course Europe. So why does FIFA not locate the World Cup Finals in Europe
at every opportunity so as to maximize attendances, broadcast audiences and possibly
revenues? One answer to this objection is that the stadium audience is much smaller
than the global television audience, now enhanced by other media such as cell phones
and internet. With a huge global broadcast audience available it may not matter where
the games are actually played. It is notable that part of the revenue from sales of the
2010 Finals broadcast rights came from the USA. FIFA negotiated a deal with the Eng-
lish-language networks, ABC and ESPN for a total of $100 million for the two tourna-
ments in 2010 (South Africa) and 2014 (Brazil). This sum was exceeded, however, by
the amount of $325 million paid by the Spanish-language network, Univisión, over the
same period. The reason for this large fee is the sizeable and growing Hispanic/Latino
minority group in the United States, and their passionate interest in soccer.
There may be a deeper reason for FIFA’s rotation policy than an ambassadorial function
for the Finals. UEFA, itself one of the FIFA confederations, hosts two tournaments, the
prestigious club-level UEFA Champions’ League contested by the most successful
teams across national leagues within Europe, and the European Championship, a
national team tournament contested by European national sides every four years and
two years apart from the World Cup Finals. Locating the World Cup Finals on every
occasion in Europe, would dilute the FIFA brand and would present UEFA with
enhanced negotiating power within FIFA. Understandably, therefore, FIFA avoids the
option of locating the Finals in Europe on every occasion.
After the 2018 finals, FIFA will operate a new system for allocation of host nation(s).
Rotation will be formally abandoned in favor of a new rule whereby the confederations
whose associations have hosted the two preceding World Cups are not eligible to bid.
However, all the associations from Asia, North and Central America and the Caribbean,
Oceania and Europe could bid for the 2018 FIFA World Cup. The reason for this
120
change in policy came about because Brazil emerged as sole (and successful) bidder for
the rights to host the 2014 Finals. As shown in the case of the Olympic Games, rights
holders gain more rents from a larger number of bids.
Competition between bidders drives up the sums of money committed to stadium and
other infrastructure and these sums are an important part of any successful bid. For the
2006 Finals in Germany, many club stadia needed renovation and redevelopment rather
than being built as new. The German Organizing Committee spent $1.9 billion over 12
locations with over 60 per cent of this figure coming from clubs and private investors,
meaning that the share of public spending on stadia development programs was rela-
tively low. In contrast, the spending on stadia in South Africa was estimated by Maen-
nig and du Plessis (2007) at $1.4 billion over 10 venues, with five new and five reno-
vated stadia. The key difference from the German case was that in South Africa vir-
tually all the financial commitment was made by the Government. Another difference is
that South Africa does not have a well-developed and vibrant team-level soccer league,
which Germany clearly has.
In addition to stadia and related infrastructure, such as roads and public transport net-
works, host associations must commit to various other expenses which include security,
advertising and cultural programs. Maennig and du Plessis (2007) estimate the profit for
the German organizing committee in 2006 to be $206m, aided by near capacity sales of
match tickets. This surplus was distributed among the German Soccer Association
(DFB), the German Premier League (DFL) and the German Olympic Federation. In
contrast, the estimated profit to FIFA from the 2006 Finals was $1.9 billion, although
part of this was redistributed to FIFA development programs worldwide.
Despite the substantial costs involved, which extend to non-trivial costs of the bidding
process itself, the likelihood of cost overruns and projections of fairly modest net profits
to the host organizing committee, there has been no shortage of bids to host the 2018
and 2022 World Cup Finals. FIFA decided to allocate both sets of Finals simultaneously
and by deadline of March 2009, had received seven bids for 2018 or 2022, featuring
Australia, Belgium and Netherlands as co-hosts, England, Japan, Russia, Spain and
121
Portugal as co-hosts and USA. A further four bids were received for 2022 only, includ-
ing Indonesia, South Korea and Qatar. As at November 2009, the betting odds offered
by Sky Bet for host of the 2018 tournament where 11/8 for England, 3/1 for Spain and
Portugal and 7/2 for Australia. However, the short odds on England may be driven by
national sentiment and odds do vary in response to news stories, such as internal argu-
ments within the England bid team which moved the odds from 11/10 to 11/8 in
November 2009.
Given the high costs of stadium infrastructure, security and advertising and the ability
of FIFA to extract economic rents as tournament rights holder, it is worth considering in
more detail the likely pecuniary and non-pecuniary benefits from hosting the World
Cup Finals. Benefits to host country populations are considered next.
8.3 Benefits to Local and National Economies
The general consensus of the economic benefits from hosting large-scale sports events
is that these are both exaggerated in ex ante studies and small ex post (Baade, 2003;
Matheson, 2008). This consensus covers the Olympic Games (Baade and Matheson,
2002; Hotchkiss et al. 2003; Humphreys and Zimbalist, 2008) and the National Football
League’s Superbowl (Baade and Matheson, 2006). Ex ante estimates of expected eco-
nomic benefits of large-scale sporting events tend to be optimistic, partly through
booster studies undertaken by consultants who have an incentive to report large bene-
fits. The events themselves are subject to a number of leakages and diversion effects.
For example, hotel room rates rise around locations of large sporting events. Local resi-
dents may leave the area to avoid congestion and nuisance associated with the event.
Projected employment gains may be misleading as the jobs involved may well be low-
skilled and temporary and many services are actually performed by volunteer workers
whose activity will not form any part of Gross Domestic Product.
In contrast, ex post studies of the economic benefits to local economies from hosting a
large scale sports studies are more downbeat. There are two methods typically used to
evaluate economic benefits ex post. The first is applied by Baade and Matheson (2004)
122
to the World Cup Finals hosted by the USA in 1994 and estimates a regression model of
income growth with national GDP growth, demographic variables and time trend as
control variables. This study compared income growth in the 75 largest population cen-
ters (Metropolitan Standard Areas) before and after the World Cup, spanning a period of
1970 to 2000. The main finding was that actual growth in incomes was less than would
have been expected prior to the World Cup event. Moreover, 9 out of thirteen host cities
suffered lower growth after the World Cup compared to before. Using a similar
approach, Hagn and Maennig (2008) could not find any evidence of statistically signifi-
cant positive benefits to German regions (in the old Federal Republic) from hosting the
1974 World Cup, whether assessed by GDP growth, income growth or unemployment
reductions.
The second approach involves use of local employment data and a difference-in-differ-
ence methodology in which employment levels in localities which hosted World Cup
games are compared with those that did not host, using differences in employment as
the dependent variable in a regression model. This method was applied by Hotchkiss et
al (2003) in a study of employment effects of the Atlanta Olympic Games on counties in
the Atlanta area. Applying this method to German cities before and after the 2006
World Cup, Feddersen et al (2009) find a lack of significant short-run or long-run
employment or unemployment effects attributable to hosting the World Cup in Ger-
many. The procedure was to consider 12 World Cup venues as a treatment group in
amongst the 118 largest population urban districts in Germany. Noting that construction
projects on host city stadia began several years before the World Cup actually took
place, and using a time span of 1995 to 2005, the authors compare differences in out-
come measured as per capita income, employment and unemployment levels (sepa-
rately) before and after the intervention, defined as beginning of a large stadium con-
struction project. The hypotheses of zero income and employment effects of the stadia
construction projects in urban districts with completed work could not be rejected at the
conventional five per cent significance level.
At the microeconomic level, some particular industries may benefit from a country
hosting the World Cup Finals. For example, the German beer industry enjoyed high
123
levels of sales during the 2006 World Cup Finals. But it is difficult to separate the
effects of hot weather in the June and July period from the effects of the World Cup
Finals. Also, Hagn and Maennig (2009) report that hotel occupancy rates actually fell
during the World Cup Finals with substantial reductions in Berlin and Munich as host
cities. Revenues held up as room rates rose. These results are very much line with the
skeptical analyses offered by Baade (2003) and Matheson (2008).
The absence of substantial direct economic benefits from hosting the World Cup has led
several researchers to examine possible intangible or ‘feelgood’ effects. Heyne, Maen-
nig and Sussmuth (2007) conducted a before-and-after contingent valuation study to
extract willingness to pay values for 500 people, before and after the 2006 World Cup
Finals in Germany. The authors found that the average ex ante willingness to pay figure
for respondents offering a positive value was $30.4. But fewer than 20 per cent of res-
pondents had a positive willingness to pay so the overall average was $5.66. After the
event, 43 per cent of respondents reported a positive willingness to pay while the over
the whole sample before and after the event, willingness to pay had a mean of $13.4 per
person. Many respondents switched from zero willingness to pay to positive willingness
to pay, particularly respondents from the eastern Germany and also the less educated.
The authors suggest that this is indicative of the World Cup as an experience good and
also that ex ante studies of willingness to pay might be biased downwards.
As shown in the 2010 South Africa case, hosting the World Cup Finals can entail consi-
derable spending by the public sector on soccer stadia and associated infrastructure. The
case for such spending is stronger if it can be shown that the World Cup generates a net
increase in social welfare. But welfare is not observed, and is imperfectly correlated
with objective measures such as Gross Domestic Product. This problem has led to use
of self-reported happiness indicators in questionnaire studies. The question typically put
is “Taking all things together, how would you say things are these days- would you say
that you are happy, quite happy or nor very happy”. Responses are then coded on a
Likert scale and applied ordinally in econometric analysis.
124
Kavetsos and Szymanski (forthcoming) use happiness data taken from the Euro-
barometer Survey covering 1,000 people per country over 12 countries for the period
1974-2004. In this case there were four responses to the life satisfaction question: very,
fairly, not very and not at all satisfied. The authors propose two hypotheses: i) better
than expected national athletic performance raises happiness and ii) hosting major
sporting events increases happiness. These hypotheses are tested over the World Cup
years 1978, 1982, 1986, 1990 and 1994. Thus, they included a dummy variable for the
only host country in the data, Italy, 1990. By comparing national team ratings just
before and just after the World Cup Finals the authors tested the impact of team perfor-
mance on happiness using an ordinal logit model. This model had a long list of control
variables. Macroeconomic control variables comprised GDP per capita, unemployment
rate and inflation rate. Personal control variables included employment status, sex, age,
age squared, marital status, household income quartiles and educational level attained.
The results showed that hosting the World Cup Finals by Italy in 1990- who did not win
the tournament- was positively and significantly correlated with reported happiness for
the population as a whole and for a series of subgroups, with the notable exception of
females. The authors also considered anticipation and legacy effects. A set of post-event
dummies for two and four years after was jointly significant for all subgroups consi-
dered. Also, dummy variables for one year before and one year after gave positive
effects on happiness, for the population as whole, for individuals under the age of 50,
males, the unemployed and those who had not benefitted from higher education.
8.4 Benefits of Hosting the World Cup Finals to Soccer Fans
A criticism of public expenditure on purpose-built stadia constructed for the World Cup
Finals is that they are under-used after the event. In the case of the 2006 Finals in Ger-
many, all 12 stadia used for the Finals reverted to use by Bundesliga clubs after the
event. This includes the venue for the Final itself, the Olympiastadion in Berlin,
currently occupied by Hertha Berlin. This club is a good example of how a team can
successfully raise its attendance following a move into a new stadium. In the season
directly after the World Cup Finals, 2006/07, Hertha Berlin’s average league attendance
125
was 25,000. But in 2008/09 average attendance had grown to 52,300 including three
sell-out games at a capacity limit of 74, 000.
The German Bundesliga is notable for the enthusiasm of its fans and it is no surprise
that several teams which had renovated or new stadia after hosting World Cup Finals
have experienced rising league attendances. As put in the baseball literature, the ques-
tion is ‘if you build it, will they come?’. A thorough empirical investigation of this
question is offered, again for Germany, by Feddersen et al. (2006). Their analysis of
‘novelty’ effects of new or renovated stadia follows the literature on baseball atten-
dances (Coates and Humphreys, 2005) and distinguishes three effects:
1. Fans quickly get used to newly built or renovated stadia. This is a short-term
immediate novelty effect confined to the first season directly after opening, with
a dummy variable set equal to one for this period only.
2. There may be a long-term novelty effect, as long as five years but then ending
abruptly. Hence, a dummy variable is set equal to one for time periods t+1 to t+5
after stadium opening.
3. The novelty effect begins on date of stadium opening, has a maximum value
after opening but then decays over the following years; the dummy variable is
then D = aT where T is a five year time trend and a<0.
Looking first at the 1974 Finals, Feddersen et al. (2006) find that there was an average
increase in attendance for all clubs playing home games in new or renovated stadia of
47 per cent one season after the Finals. This novelty effect shows some persistence:
after five years, eight out of nine new or renovated stadia had higher fan numbers than
in the year prior to the completion of construction. For the 2006 Finals, new stadium
projects began as early as 1998. For these stadia, the authors find some evidence of
novelty effects in the descriptive data. The authors combine impacts of new and reno-
vated for the 1974 and 2006 Finals in an econometric analysis of Bundesliga club atten-
dances over 1963-2003 using club and season fixed effects and controls for regional
income and team performance. In Germany, stadium capacity is rarely binding so Tobit
estimation is not required. Of the three novelty effects noted above, only the second is
126
significant at the five percent level. This long-term novelty effect is estimated at 2,700
extra fans per game, for five years, or 10.7 per cent on a mean value of 25,000 fans.
Higher novelty effects, estimated at an extra 10,300 fans per game, are found for two
clubs, Hamburg and Schalke 04, who introduced the innovation (for Germany) of
building new stadia without the traditional running track to separate crowd from the
pitch.
Feddersen et al point out that a large part of the revenue gains from new or renovated
stadia comes from VIP and corporate seats, with around €8 millions accruing to Bun-
desliga clubs after the 2006 World Cup. It appears that extra revenue from new stadium
developments comes from absorbing the purchasing power of a small group of affluent
fans, a point that echoes the luxury box strategy of new stadium construction by Major
League Baseball and National Football League franchises in North America. Since these
affluent fans are willing to pay higher ticket prices then consumer surplus rises, but
extra corporate boxes and seating may reduce the capacity available for less affluent
fans, leading to ‘social exclusion’ from games. Supporters’ organizations often voice
equity concerns as a result.
FIFA operates two key restrictions on stadium development for World Cup Finals. First,
there must be a minimum capacity of 40,000. The capacity minimum is designed to
ensure sufficient revenue from ticket sales but has the consequence that smaller teams
using a World Cup stadium for regular League games may find they often have empty
seats. This is a particular problem for countries such as South Korea and South Africa
whose Leagues are less mature than European Leagues. In the case of South Africa, two
new stadia were built in Durban and Cape Town, holding 70,000 and 68,000 fans
respectively. The two stadia in Johannesburg were renovated to reach capacities of
94,700 (Soccer City) and 65,000 (Ellis Park). These are most unlikely to sell out save
for a few international rugby and possibly soccer matches and certainly not for club
fixtures. Second, FIFA prohibits the sponsorship of stadia in the form of naming rights
unless the stadium sponsors are also official sponsors. Thus, the AOL Arena in Ham-
burg became the FIFA Football World Cup Arena for the duration of the 2006 Finals.
127
A total of seven out of 12 German stadia were affected in this way. This is another
example of rent extraction by FIFA.
Regardless of newness of stadia, hosting the World Cup Finals may lead to a boost to
soccer attendances at League level. In his analysis of English Football League, Bird
(1982) used a World Cup dummy to capture the combined effects of England hosting
and winning the tournament on annual League attendances after the 1966 Finals. Con-
trolling for admission prices, travel costs and real incomes, Bird found that the World
Cup contributed to a significant upward shift of 9.7 per cent in the 1966-67 season
immediately following the Finals. This was in the context of long-term decline in soccer
attendances in England over the period 1946 to 1985. However, Bird could not diffe-
rentiate between a World Cup effect and a possible uplift to attendances due to the extra
publicity brought about by the introduction, for the first time, of a popular TV show
Match of the Day featuring edited highlights of Saturday matches.
In 1998, France repeated England’s achievement of hosting and winning the World
Cup. Falter et al. (2008) argue that the period after France’s success was one of ‘over-
whelming joy’ which led to a positive network externality on club attendances in the
French League. In terms of raw data, average League attendances rose from 16,600 in
the 1997/98 season just before the Finals to 19,800 in 1998/99, just after, and again to
22,300 in 1999/2000. Falter et al present an econometric demand model containing a
host city dummy, home and away standings, home team payroll as a proxy for team
quality, last score of the home team, transport costs, and dummies for seasons, matches
involving local rivalry and sunshine all as control variables alongside year dummies
intended to capture World Cup effects. Unlike Bird’s (1982) study, Falter et al use
match level data. Their World Cup dummy variables show statistically significant
increases in club League attendances of 14 per cent and 23 per cent for the two years
following France’s World Cup success, ceteris paribus. There is also a positive and sig-
nificant host city effect. That is, the World Cup effect on club attendances is stronger
for cities that hosted tournament games. This is apparently a combination of new sta-
dium (novelty) and advertising effects.
128
The authors make two robustness checks that do not affect their results. They deduct the
more habitual season ticket holds from their match attendance figures and they also
adopt tobit estimation to deal with sell-out fixtures. The authors also note that the rise in
attendances in the French League was not matched by other Leagues such as England,
Germany, Italy and Netherlands. But the rise in League attendances following a World
Cup victory is also apparent in England (1966), Germany (1974, 1990) and Italy (1982).
Hence, the authors conclude that average club League attendance tends to increase after
a World Cup win and moreover, this effect persists for several years after the victory.
8.5 Benefits to Players from Participating in the World Cup Finals
Players, who appear in national team squads in the World Cup Finals, and in the earlier
qualifying tournament, are selected by national associations, primarily by the national
team head coach. These players will have employment contracts with clubs, with whom
they play their regular soccer games in Leagues, domestic Cup competitions and in
other competitions such as the UEFA Champions’ League. But national associations
have the right to demand release of players from club duties for national selections. This
right does not amount to conscription as players can ‘retire’ from international soccer.
Also, players may be injured with their club teams and hence temporarily unavailable
for selection for national teams. Clubs receive some compensation for release of players
for national team games but this is nevertheless a source of complaint, especially when
players return from national games with injuries. Conversely, national associations
complain that players withdraw from less important national team games, especially ad
hoc ‘friendly’ matches, with injuries that are not as serious as first appears. The inter-
national fixture calendar, including the World Cup Qualifying competition, is carefully
set so that club fixtures are removed from weekends or midweek periods when inter-
national games are played. Also, the Finals themselves are always played in the close
season for the vast majority of Leagues.
In terms of career prestige playing in, and better still, winning the World Cup Finals
offers a huge boost to player career incomes and also offers a high non-pecuniary
reward. A player who appears in the winning team in the World Cup Finals can expect
129
both national adulation for a long time to come and immense off-field earnings pros-
pects in the form of endorsements and lucrative after-dinner speaking engagements.
Hence, players typically regard the possibility of playing in the World Cup Finals as an
important career goal. The size of direct payments to players from participating in the
World Cup Finals has been assessed by Coupé (2007). He finds for the 2006 World Cup
that most national associations had bonus schemes that rewarded performance but
incentives did not rise monotonically as the tournament progressed. Table 8-2 shows the
total and marginal (by tournament stage) team bonus in millions of Euros paid to com-
petitors in the 2006 Finals (Source: Coupé (2007)). Prior to 2006, FIFA and the national
associations had used a fixed bonus per match regardless of the stage of the tournament,
presumably under the assumption that prestige effects dominated pecuniary considera-
tions in terms of player effort and performance.
RoundBonusesMarginalbonus
Eliminationround3.79
Reach8thfinals5.381.59
Reachquarterfinal7.281.90
Semifinal13.616.33
Final14.240.63
Winner15.511.27
Table 8-2: Team Bonuses at the 2006 World Cup Finals
Across the competitors in the 2006 Finals, Coupé finds that different bonus schemes
were used. Croatia distributed a fixed percentage of FIFA prize money. Germany oper-
ated a fixed increase in prize for reaching the next stage together with a double bonus
for winning the Final. Spain applied a rising bonus level at each stage and this was the
method closest in spirit to the predictions of tournament theory. Generally though,
Coupé finds that bonuses did not have the convex structure that tournament theory
would predict. There was no discernable relationship between size or structure of
bonuses and either match results or quality ratings of games played. Typically, all play-
ers in a Finals squad (fixed at 22 by FIFA) got the same bonus regardless of playing
time.
130
There are several episodes of disputes between players and national associations over
the size of bonuses e.g. Germany in 2002 and Ghana in 2006. In Ghana’s case, player
pressure forced their national association to raise the level of bonuses. Bonuses can vary
according to performance in previous World Cup Finals. The bonus paid to the French
national team squad was €244,000 in 1998 when they won, rose to €300,000 in 2002, as
a reflection of player bargaining power, and fell back after a poor 2002 Finals to
€240,000 in 2006 (all values are nominal).
At individual level, a number of studies have shown that soccer player basic salary
offered by clubs (before bonuses) can be successfully modeled as a Mincer-type earn-
ings function in which experience (or age) and its square and performance assessed as
goals scored and possibly assists to goals are regressors that deliver statistically signifi-
cant coefficients (Lucifora and Simmons, 2003; Frick, 2006). But these models lack a
full set of performance indicators for defenders and midfield players. Lucifora and
Simmons (2003) had a sample of 533 players in the 1993-94 season for Italy’s Series A
and B with salary data gleaned from the players’ association. Using dummy variables to
denote players who had recently appeared for their national teams, they find that Italian
international players received a 52 per cent salary premium against non-internationals
while other internationals obtained a higher premium of 75 per cent. The study predates
the Bosman ruling of 1995 which led to enhanced player mobility within the European
Union. Using a suitable proxy measure of player salaries and with data spanning 1995
to 2005, therefore after the Bosman ruling, Frick (2006) shows, again using a standard
Mincer earnings function, that sizeable nationality premia for country of birth applied in
the German Bundesliga over the period 1995-2003. On top of these ethnicity premia are
further salary returns for each career international appearance made. Frick enters the
career international appearance variable in quadratic form and finds that salary is max-
imized at 50 appearances.
Both Lucifora and Simmons (2003) and Frick (2006) find large salary premia accruing
to players who represent their national teams. We suspect that much of these sizeable
premia is a consequence of omitted variable bias since the only performance measure in
Frick’s study is goals scored, itself mostly a product of forwards, while Lucifora and
131
Simmons consider goals and assists. Future data sets may well resolve this omitted vari-
able problem as more detailed performance statistics become publicly available,
including those for defenders. However, we do have detailed data on player perfor-
mances in the Bundesliga top division for one season. These data include the number of
tackles won or lost and both completed and incomplete passes made, plus several other
performance measures. We find that the impact of these additional variables on player
salary is statistically not significant, separately or jointly considered.
Neither Lucifora nor Simmons or Frick attempt to separate international status or
appearances into participation in World Cup Finals and other World Cup games.
Clearly, ad hoc ‘friendly’ matches and World Cup Finals are quite different in prestige
and are likely to have different impacts on salary. At least, this is a proposition to be
tested. Within the soccer industry, there has been much discussion of the World Cup
Finals as a ‘shop window’ effect for players, especially less well-known players from
third world countries. Rather than view misleading video clips or make expensive
scouting trips to remote countries, clubs and their agents can view and assess players in
the highest-level competition on a world stage.
At issue here is whether participation in the World Cup Finals adds to a players’ prod-
uctivity at club level, that is, augments his human capital, or whether participation in the
Finals is a simply a signal. In the labor and education economics literatures, there is a
long-standing debate over whether college education adds to human capital or is a sig-
nal to employers. In the sports economics literature, we know of no attempt to distin-
guish signaling from human capital explanations of player salary. In the major North
American sports, appearances for a national team are rare. Playing for the USA national
basketball team at the Olympic Games is a notable exception. Therefore, international
soccer offers an excellent opportunity to discriminate empirically between human capi-
tal and signaling explanations of player salary. Arguments can be made for the conjec-
ture that participation in World Cup Finals augment a player’s productivity at club
level. At the Finals, players must pit their wits against the best players from opposing
national teams and their experience of competition at this highest level could well spill
over into more successful League performance. Learning effects may also result from
132
playing with excellent peers in one’s own national team. Moreover, the World Cup
Finals put great pressure on players due to the burden of expectations placed upon
national teams by citizens and media alike. The experience of playing under such
intense pressure could also aid the mental development of a player.
On the other hand, the World Cup Finals could be argued to be a unique exercise in
competition between nations that is unlikely to be replicated at League level, where
teams play each other repeatedly, save for entry and exit occasioned by promotion and
relegation. It is quite possible that participation in the World Cup Finals adds little or
nothing to a player’s productivity at club level, yet has the effect of raising a player’s
salary. This would be consistent with the signaling hypothesis. Indeed, Szymanski and
Kuper (2009), in their entertaining account of world soccer, argue that clubs tend to
overpay for players who have recently appeared in World Cup Finals.
To properly discriminate between human capital and signaling explanations of soccer
player salary one would require more detailed performance measures than current pub-
licly available data sources permit. As an interim step, we can simply assess whether
salary premia for World Cup appearances are larger than for other appearances.
We have data on salaries for all players with positive appearances in the German Bun-
desliga top division from 1995/06 to 2007/08. The salary measure is a market value
measure collected by Kicker magazine that is known to be a good proxy (in the sense of
well correlated) with a subsample of actual salaries released by the German Football
Association (Frick, 2006; Torgler and Schmidt, 2007). This sample comprises 1,993
players for a total number of 6,147 observations. We regress log salary against the fol-
lowing control variables: age and its square, number of appearances in the Bundesliga
in the previous season, goals scored last season in the Bundesliga, career appearances in
Bundesliga and its square, career goals scored in the Bundesliga and its square and
dummy variables for position played in the club team, seasons and region of birth. Over
and above these controls we add our focus variables which are number of World Cup
Finals games played in the previous season, number of non-World Cup Finals games
played in the previous season, career World Cup Finals appearances up to the previous
133
season and career non-World Cup appearances up to the previous season. Non-World
Cup games include World Cup qualifying matches, European Championship qualifying
and Finals games and friendly matches. Table 8-3 shows OLS and fixed effects (for
players) results of our focus variable log salary.
Focusvariable
OLScoefficient
(tstatistic)
Fixedeffectscoeffi‐
cient(tstatistic)
WorldCupappearances0.087(6.16)0.072(5.58)
Otherinternationalappearances0.033(9.52)0.015(4.02)
CareerWorldCupappearances0.029(2.70)0.004(0.20)
Careerotherinternationalappearances0.007(4.33)‐0.005(1.32)
Careerotherinternationalappearancessquared‐0.0001(4.02)0.000(1.05)
Table 8-3: OLS and Fixed Effects Results for the German Bundesliga
The results clearly show that one extra appearance in World Cup Finals games delivers
a greater salary benefit compared to one extra appearance in other international games.
This result applies to both ordinary least squares and fixed effects estimates, although
the latter may not be reliable due to the small of number of observations per player
(three on average). Also, an extra career appearance in World Cup Finals games up to
the previous season delivers a greater salary increment than an extra game played in
non-World Cup Finals matches. Hence, we have support for a World Cup shop window
effect on player salaries.
The ordinary least squares results show impacts of international appearances at the
means of variables. But the salary distribution for soccer players is highly skewed, even
more so than in standard occupations. Evaluating impacts at the mean may then be
misleading and it is useful to consider impacts over the whole salary distribution.
Recently, a number of studies have looked at determinants of player salaries in various
sports using quantile regression (for example, see Berri and Simmons, 2009 on National
Football League quarterbacks and Vincent and Eastman, 2009 on the National Hockey
League). This method allows us to estimate differing salary impacts of covariates
through the salary distribution. Our own quantile regression results for soccer player in
the Bundesliga are shown in Table 8-4.
134
Focusvari‐
able0.1quantile0.25quantile0.5quantile0.75quantile0.9quantile
World
Cupappear‐
ances
0.107***0.102***0.079***0.069***0.062**
Otherinter‐
national
appearances
0.029***0.022***0.031***0.037***0.044***
CareerWorld
Cupappear‐
ances
0.029*0.039**0.030**0.039**0.035**
Careerother
international
appearances
0.004+0.005**0.007***0.008***0.009***
Careerother
international
appearances
squared
0.000+‐0.0001**‐0.0001***‐0.0001***‐0.0001***
***, ** and * denote statistical significance at the 0.01, 0.05 and 0.1 level, + denotes insignificance (z-values in brackets).
Table 8-4: Quantile Regression Results for log Salary
The results of the quantile regression broadly support those of OLS. At the median sal-
ary, it is clear that salary returns to an extra World Cup Finals match exceed returns to
other international matches. However, above the median the gap in returns is not sig-
nificantly different at five per cent level, despite the apparent gap in point estimates.
The results point to diminishing returns of salary to recent World Cup Finals appear-
ances through the salary distribution. For career appearances, the results are more clear-
cut with salary returns to career World Cup Finals games being always greater than
returns to other international games, at all quantiles.
At club level it is apparent that some national Leagues are more prestigious and gen-
erate more revenues than others. Hence, one would expect the best players to gravitate
towards those Leagues where these players’ marginal revenue product would be at their
highest. The Leagues where revenues and average player salaries are highest at present
are the English Premier League and Spain’s La Liga. UEFA gives each European team
a score, based on standings in national Leagues and progress in European competitions,
essentially the UEFA Champions’ League and the European Cup (previously UEFA).
135
The score variable ranges from one to 31. We create a variable, mover, which states if
players switch from lower ranked teams to higher ranked teams or the other way
around. Positive numbers suggest a change to a higher ranked team. We then estimate a
probit regression for this variable using the same control variables as for the salary
model above. This model is estimated for all players who switch teams within Europe,
giving us 1,638 observations. Using our World Cup dummy variables already created,
we can determine whether participation in World Cup games raises the probability of
movement to a more highly ranked team. The estimated coefficients from our probit
model are shown in Table 8-5 below.
Focusvariablecoefficient(tstatistic)
WorldCupappearances0.335(3.00)
Otherinternationalappearances0.026(1.57)
CareerWorldCupappearances0.045(0.81)
Careerotherinternationalappearances0.026(2.92)
Careerotherinternationalappearancessquared‐0.0004(3.38)
Table 8-5: Probit Estimates of Movement to a more Highly-Ranked Team
It is clear that an extra recent appearance in a World Cup Finals match raises the proba-
bility of a move to a more highly ranked team, conditional on moving at all. In contrast,
an extra recent appearance in other internationals has no statistically significant effect
on a move to a better team. Looking at the career variables, however, the results switch
so that an extra career World Cup Finals appearance has no effect on probability of a
move to a more highly ranked team while an extra career appearance in other inter-
national matches does have a significant and positive effect on transition to a better
team. Nevertheless, the marginal effect is smaller than for recent appearances in World
Cup Finals matches.
To summarize, recent appearances in World Cup Finals matches do appear to have shop
window effects, both by raising player salaries paid by clubs and by helping players
secure transitions to more highly ranked teams. This still leaves open the intriguing
research question as to whether participation in World Cup Finals matches genuinely
136
raises player productivity or whether it is just a signal that need not necessarily
represent player ability.
8.6 Club versus Country?: The Domestic Player Quota Debate
Up to the Bosman ruling of 1995, it was possible to restrict the number of foreign-born
players in a national League by means of various rules. In the English Football League
in the 1950s and 1960s, ‘foreigner’ meant a player from Scotland or Wales and not
someone from Africa or South America. Many European Leagues, such as England and
Italy, operated a rule such that a club team could consist of a maximum of three foreign-
born players plus two ‘assimilated’ players who were foreign-born but had played at
club level in their adopted country for at least two years.
After the Bosman ruling, such restrictions were deemed to be counter to principle of
free movement of labor, at least as far as European Union countries were concerned.
Restrictions remained in place for players attempting to move into European Union
Leagues from outside the European Union. But the increasing demands by clubs for
quality players from outside the European Union grew, and restrictions became more
relaxed over time. We now have a situation where each major soccer League in Europe
(England, France, Germany, Italy and Spain) has a cosmopolitan mix of players from all
continents. Indeed, it is quite possible for a team to field a starting eleven consisting
entirely of non-domestic players, as practiced in recent years by Arsenal and Chelsea in
the English Premier League.
Increased immigration of player talent has led to concerns that the displacement of
domestic talent by foreign-born talent, a form of import substitution, might undermine
the prospects for the national team. In England, these concerns were documented in a
set of proceedings from the ‘feet drain’ conference held at Birkbeck College, London in
April 2008. These proceedings feature an address by Gordon Taylor, Chief Executive of
the Professional Footballers’ Association which summarizes the arguments in favor of
quotas of domestic nationals in team squads, using the term ‘meltdown’ to describe the
internationalization of playing talent in the English Premier League in particular
137
(Taylor, 2009). Recently, UEFA has implemented a ruling whereby the squads of teams
playing in its Champions’ League football competition should comprise 25 players, 8 of
which should be ‘home grown’, meaning having trained for three seasons at a domestic
club between the ages of 15 and 21, and of these at least half should have been trained
at the club itself for the same period. Similar rules will be applied by the English Prem-
ier League from the 2010/11 season. One of the arguments in favor of such quotas is
that they encourage the development of promising young ‘home grown’ players who
may eventually be selected for their national teams. Conversely, the present arrange-
ments in several European Leagues, where starting teams in England, in particular, fea-
ture relatively few English players and youth academies also feature a high proportion
of ‘imported’ foreign players, are alleged to stifle the development of the national team.
In England, this translates into the notion of reduced prospects of winning the World
Cup.
Looking at the general arguments surrounding fixed quotas for playing rosters at club
level, we make several predictions drawing on simple economic theory:
• The supply curve for talent gets steeper as there is more competition for stars,
giving richer clubs an advantage;
• There will be rising salaries for domestic stars;
• There will be less competition for international players means owners get more
profits;
• Playing quality falls as some stars leave for Leagues that do not operate restric-
tions;
• Broadcasting deals become less attractive as the total pool of talent is less;
• Big clubs will be less able to win the UEFA Champions’ League;
• Small countries will be less successful in national competitions such as the
World Cup.
In a literature that is analogous to that on Olympic medal success, a number of papers
have empirically modeled World Cup success using World Cup Finals rankings or
something similar as a dependent variable in a regression analysis. Monks and Husch
(2009) present a panel data analysis of World Cup success, over 1982 to 2006, using
138
tournament Finals standings as their dependent variable. They find that playing on one’s
own continent and being seeded are each significant determinants of World Cup suc-
cess. In purely practical terms, of course, the strong records of Brazil and Germany
(previously West Germany) at the World Cup Finals, as noted in Table 8-1, means that
a national team aspiring to win the tournament has to be capable of beating one or both
of these teams, or hope that some other team does so.
Leeds and Leeds (2009) adopt a cross-country (not panel) analysis with FIFA points
assessed at 2006 as their dependent variable. They find a statistically significant positive
effect of the international success of a country’s club teams on national team success. A
specific example supporting this result is France in 1998, who managed to win the
World Cup with a majority of players in the national team being attached to clubs out-
side France, many of which had made successful progress in the Champions’ League.
As we showed earlier, in section three on impacts of World Cup Finals on fans, Falter et
al (2008) report positive externalities from French national team success on club atten-
dances and revenues. Szymanski and Kuper (2009) take the club versus country argu-
ments a stage further by suggesting that England’s prospects for World Cup success
would be enhanced if more of the higher-quality English players actually played abroad,
a suggestion that is directly counter to the arguments noted above in favor of domestic
quotas. However, Szymanski and Kuper also suggest that the natural Finals standing for
an England team at the World Cup is the quarter-finals stage and it is somewhat opti-
mistic, based on historical data, to expect anything better.
Overall, the attempt to re-introduce domestic quotas in European League soccer appears
to be a protectionist device with little merit. The case for such quotas to actually contri-
bute to enhanced World Cup performance has yet to be made, and does not come with
any supporting empirical observation. It is notable that the ‘internationalization’ of
teams in European soccer Leagues has been accompanied by rising attendances in four
out of five major European Leagues since the mid 1990s (Italy being the exception,
which can be explained by a host of specific circumstances relating to hooliganism and
corruption scandals). Thus fans, and also television viewers, appear to like the increa-
singly cosmopolitan make-up of their club sides.
139
8.7 Conclusions
The FIFA World Cup can make a legitimate claim to be the world’s largest sporting
tournament, even bigger than the Olympic Games in terms of broadcast audience and
worldwide interest. The rights to hold this lucrative and prestigious tournament reside
with FIFA and it is no surprise to find that FIFA succeeds in rent extraction, even
though FIFA does not take a direct payment from local organizing committees. FIFA
does earn substantial revenues from sponsors and advertisers and employs restrictions to
ensure that official FIFA partners receive favorable treatment on access to advertising
space e.g. on billboards.
The consensus of the sports economics literature is that the World Cup, in common with
other large-scale sports events, does not generate substantial benefits in terms of local
real income and employment growth and unemployment reductions. In contrast, non-
pecuniary benefits, in terms of enhanced satisfaction and happiness, have been identi-
fied by some recent research. Further research is needed to corroborate these findings.
There is little doubt that hosting the World Cup Finals is costly in terms of stadium
development, infrastructure and security. Where public spending is involved in hosting
the Finals, more cost-benefit studies on the net benefits to the host country would be
welcome.
There are two impacts on soccer fans worth noting. First, there are novelty effects of
new stadia that persist for some time after new stadia are constructed, at least as far as
Germany 2006 was concerned. Second, hosting and winning the World Cup can lead to
a boost to club attendances through a ‘warm glow’ effect.
Players who appear in the World Cup Finals appear to benefit in terms of enhanced sal-
ary, although whether this is due to a signaling effect or to genuine improvement in
productivity is a matter for future research to resolve. Finals participants also benefit
from increased probability of a transfer to a better club, as measured by UEFA rankings
for European clubs.
140
The FIFA World Cup Finals is no doubt here to stay as a vital component of the sport-
ing calendar. Avoiding excessive and wasteful expenditures is difficult to achieve, given
the emotion and sentiment that surrounds the event. But the World Cup does offer
excellent opportunities for economists to make substantial contributions to the various
public choice and policy issues surrounding the World Cup, including a more sober
assessment of employment and income effects that is present in consultancy and booster
studies.
141
9 Outlook
To address the main question stated in the introduction, the present work analyzes the
influence of personality traits on individuals’ income. As I relate leadership ability to
the personality trait “extraversion”, I find clear support for the claim that leadership
ability is indeed rewarded monetarily. Research for the German Bundesliga as well as
the National Hockey League reveals that team captains in professional sports receive a
considerable wage premium. Controlling for individual player characteristics and per-
formance indicators it shows that team captains in soccer earn a wage premium between
25 and 67 percent, while team captains in hockey receive a wage premium between 21
and 35 percent. To evaluate these results, comparable research concerning other sports,
possibly with different team sizes and varied numbers of team captains, would be of
great interest. Theory of organization would lead us to the expectations of rising bene-
fits of leadership ability as team size increases and as the number of team captain
decreases.
The following two chapters address the influence of pressure on individuals’
performance. Relating to the personality trait of emotional stability, we show that
professional basketball players who are able to maintain their performance level during
pressure situations are rewarded monetarily. Furthermore, I present an empirical work
about the effect of performing in front of friendly and hostile audiences, focusing on the
effect of changing teams and therefore supportive audiences between two seasons.
Research shows that players’ performance suffers during home games after they sign
with a new team as they feel additional social pressure. This is most notably true for
players who display comparably bad performances. Consentaneously, both studies show
that emotional stability does not increase with age and therefore appears to be innate.
This finding clearly goes in line with the literature presented in the introduction which
states that personality traits rather not change over time. Future research might combine
both studies to analyze the influence of supportive or hostile audiences on performance
during pressure situations.
142
For my research on the relative age effect in the German Bundesliga, I summarize
results as follows: Grouping children according to their age leads to labeling older
children of the cohort as being more talented which leads to more support during their
early playing years. My data set on professional soccer supports the claim that this
selection improves the probability of the oldest members of the cohort to become pro-
fessional soccer players in the Bundesliga. Along with my thesis that the birth date itself
should not affect playing performance, I find no proof for a monetary reward of reach-
ing the professional level despite an age disadvantage. Providing hazard rates this
statement is supported as the length of stay in the Bundesliga proves to be independent
of the date of birth. A similar research for the German second division would provide
the possibility to compare career tracks dependent on the date of birth.
Concerning the sabotage behavior of heterogeneous contestants, our theoretical model
and empirical work offer concurrent results. Our formal model suggests that the favorite
in a tournament chooses legal activities, while the underdog is tempted to engage in
illegal activities. The reasoning is that the favorite is more productive with respect to
legal activities and that both types of activities are substitutes. In our empirical section,
we demonstrate the plausibility of the theoretical model by using match-level data from
the German Bundesliga. We find that teams that are more likely to win a match, win
significantly more tackles in a fair way while they commit significantly fewer fouls. As
betting odds, which enable us to compute the implicit winning percentages, are
available for many other professional team sports a similar research for another league
would be of great interest.
Summarizing previous research on the economics of the FIFA World Cup we conclude
that the FIFA itself earns considerable revenues from sponsors as well as advertisers.
For the host country research does not find substantial benefits for the local economy as
real income and employment rates are not affected positively. In contrast, players bene-
fit monetarily from participating in the World Cup as our work supports the thesis of a
shop window effect. In addition, we show that players from the German Bundesliga
who participate in the World Cup exhibit an increased probability of a transfer to a bet-
ter club, as measured by UEFA rankings for European clubs. Future research might try
143
to find a shop window effect for the European Championship, which we expect to be
significantly lower than the shop window effect for the FIFA World Cup.
After providing implications and ideas for future analysis separately, I conclude by pre-
senting an outlook for ulterior research. Data used for research in the present work pro-
vides the basis for some yet unanswered questions in the field of sport economics. One
work soon to follow concerns nomination contests in professional sports, especially for
German soccer players in the German Bundesliga and relates to Spence (1973). Players
signal their abilities during games in the Bundesliga to be nominated for international
caps. As previously shown, being nominated to the national team has a significantly
positive impact on the players’ salary. The question arises which performance indicators
influence the chances of being nominated to the national team. Preliminary results show
that players’ age, experience in national and international competition as well as most
recent performance in the Bundesliga have a significant impact on the chances of being
nominated.
Also to follow soon is a joint work with Arne Büschemann concerning the efficiency of
teams in the four major league sports in the United States. We have information con-
cerning the economic performance of all teams such as the value of teams, revenues and
operating income. In addition, we have specific information for all teams concerning
average attendance, market size, sporting achievements of the team, the fan cost index
and the tradition of a club. By performing frontier analysis we are going to test whether
teams in bigger markets, which face competition by other teams nearby, exhibit higher
efficiencies due to competition. In addition, we would like to reveal if the lockout in the
National Hockey League in 2004/05 and the following new collective bargaining
agreement led to higher efficiencies of the teams in the National Hockey League.
A third subsequent work will concern violence in professional hockey. It follows a work
by Dennis Coates and Andrew Grillo (2009) who relate the influence of fighting,
violence and penalties to team and league success in the National Hockey League. In a
joint work with Marcel Battré and Dennis Coates we transfer this approach to European
professional hockey leagues to test whether the influence of the aforementioned factors
144
are the same. We possess comparable data from the German and the Finish professional
hockey league and are able to relate rule changes, which occurred during the observa-
tion period, to the impact of violence on aforementioned measurements of team and
league success.
To summarize: Even though the present work sheds light on different parts of personal
economics there is still a lot of work to be done. Sports data offers the basis to answer
aplenty of research questions, waiting to be worked on.
VI
10 List of Literature
Allen, J.; Barnsley, R. (1993): Streams and Tiers: The Interaction of Ability, Maturity, and
Training in Systems with Age-Dependent Recursive Selection. The Journal of Human
Resources, vol. 28, pp. 649–659.
Altonji, J.; Shakotko, R. (1987): Do Wages Rise with Job Seniority? Review of Economic
Studies. vol. 54, pp. 437-459.
Altonji, J.; Williams, N. (2005): Do Wages Rise with Job Seniority? A Reassessment.
Industrial and Labor Relations Review, vol. 58, pp. 370-397.
Andersen, J. (2006): Leadership, Personality and Effectiveness. The Journal of Socio-
Economics, vol. 35, pp. 1078-1091.
Anderson, J. (1993): Rules of Mind. Hillsdale: Erlbaum.
Andrisani, P. (1978): Work Attitudes and Labor Market Experience: Evidence from the
National Longitudinal Surveys. Springfield, VA: Praeger Publishers.
Angrist, J.; Krueger, A. (1991): Does Compulsory School Attendance Affect Schooling and
Earnings? Quarterly Journal of Economics, vol. 106, pp. 979-1014.
Antonioni, P.; Cubbin, J. (2000): The Bosman Ruling and the Emergence of a Single
Market in Soccer Talent. European Journal of Law and Economics, vol. 9, pp. 157-173.
Ashworth, J.; Heyndels, B. (2007): Selection Bias and Peer Effects in Team Sports: The
Effect of Age Grouping on Earnings of German Soccer Players. Journal of Sports
Economics, vol. 8, pp. 355-377.
Atkinson, S.; Tschirhart, J. (1986): Flexible Modeling of Time to Failure in Risky Careers.
Review of Economics and Statistics, vol. 68, pp. 558-566.
Baade, R. (2003): Evaluating Subsidies for Professional Sports in the United States and
Europe: A Public Sector Primer. Oxford Review of Economic Policy, vol. 19, pp. 585-
597.
VII
Baade, R.; Matheson, V. (2002): Bidding for the Olympics: Fool’s Gold? In: Barros; C.
Ibrahim, M.; Szymanski, S. (eds.): Transatlantic Sport. Cheltenham: Edward Elgar.
Baade, R.; Matheson, V. (2004): The Quest for the Cup: Assessing the Economic Impact of
the World Cup. Regional Studies, vol. 38, pp. 341-352.
Baade, R.; Matheson, V. (2006): Padding Required: Assessing the Economic Impact of the
Super Bowl. European Sport Management Quarterly, vol. 6, pp. 353-374.
Baker, J.; Logan, A. (2007): Developmental Contexts and Sporting Success: Birthdate and
Birthplace Effects in National Hockey League Draftees 2000-2005. British Journal of
Sports Medicine, vol. 41, pp. 515-517.
Balmer, N.; Nevill, A.; Williams, A. (2001): Home Advantage in the Winter Olympics
(1908-1998). Journal of Sports Sciences, vol. 19, pp. 129-139.
Balmer, N.; Nevill, A.; Williams, A. (2003): Modelling Home Advantage in the Summer
Olympic Games. Journal of Sports Sciences, vol. 21, pp. 469-478.
Barnsley, R.; Thompson, A.; Barnsley, P. (1985): Birthdate and Success in Minor League
Hockey: The Key to the NHL. Canadian Association for Health, Physical Education, and
Recreation Journal, vol. 51, pp. 23-28.
Barnsley, R.; Thompson, A.; Legault, P. (1992): Family Planning: Football Style. The
Relative Age Effect in Football. International Review for the Sociology of Sport,
vol. 27, pp. 77-87.
Barron, J.; Ewing, B.; Waddell, G. (2000): The Effects of High School Athletic
Participation on Education and Labor Market Outcomes, The Review of Economics
and Statistics, vol. 82, pp. 409-421.
Barrick, M.; Mount, M. (1991): The Big Five Personality Dimensions and Job
Performance: A Meta-Analysis. Personal Psychology, vol. 44, pp. 1-26.
Battré, M.; Deutscher, C.; Frick, B. (2009): Salary Determination in the German
Bundesliga: A Panel Study. University of Paderborn, Department of Management.
VIII
Baumeister, R. (1984): Choking under Pressure: Self-Consciousness and Paradoxical
Effects of Incentives on Skillful Performance. Journal of Personality and Social
Psychology, vol. 46, pp. 610-620.
Baumeister, R.; Steinhilber, A. (1984): Paradoxical Effects of Supportive Audiences on
Performance under Pressure: The Home Field Disadvantage in Sports Championships.
Journal of Personality and Social Psychology, vol. 47, pp. 85-93.
Becker, G. (1962): Investing in Human Capital: A Theoretical Analysis. Journal of
Political Economy, vol. 70, pp. 9-49.
Becker, G. (1964): Human Capital: A Theoretical and Empirical Analysis with Special
Reference to Education. New York: Columbia University Press.
Bedard, K.; Dhuey, E. (2006): The Persistence of Early Childhood Maturity: International
Evidence of Long-Run Age Effects. The Quarterly Journal of Economics, vol. 121,
pp. 1437-1472.
Beilock, S.; Carr, T. (2001): On the Fragility of Skilled Performance: What Governs
Choking under Pressure? Journal of Experimental Psychology: General, vol. 130,
pp. 701-725.
Beilock, S.; Carr, T.; MacMahon, C.; Starkes, J. (2002): When Paying Attention Becomes
Counterproductive: Impact of Divided Versus Skill-Focused Attention on Novice and
Experienced Performance on Sensorimotor Skills. Journal of Experimental Psychology:
Applied, vol. 8, pp. 6-16.
Beilock, S.; Gray, B. (2007): Why Do Athletes Choke Under Pressure? In: Tenenbaum, G.;
Eklund, R. (eds.): Handbook of Sports Psychology, 3rd ed., Hoboken, NJ: John Wiley
and Sons, pp. 425-444.
Berri, D.; Brook, S.; Frick, B.; Fenn, A.; Vicente-Mayoral, A. (2005): The Short Supply of
Tall People: Competitive Imbalance and the National Basketball Association. Journal of
Economic Issues, vol. 39, pp. 1029-1041.
Berri, D.; Simmons, R. (2009): Race and the Evaluation of Signal Callers in the National
Football League, Journal of Sports Economics, vol. 10, pp. 23-43.
IX
Biddle, J.; Hamermesh, D. (1998): Beauty, Productivity, and Discrimination: Lawyers’
Looks and Lucre. Journal of Labor Economics, vol. 16, pp. 172-201.
Billari, F.; Pellizzari, M. (2008): The Younger, the Better? Relative Age Effects at
University. IZA Discussion Paper No. 3795.
Biørn, E. (2004): Regression Systems for Unbalanced Panel Data: A Stepwise Maximum
Likelihood Procedure. Journal of Econometrics, vol. 122, pp. 281-291.
Bird, P. (1982): The Demand for League Football. Applied Economics, vol. 14, pp. 637-
649.
Borghans, L.; Duckworth, A.; Heckman, J.; ter Weel, B. (2008): The Economics and
Psychology of Personality Traits. Journal of Human Resources, vol. 43, pp. 972-1059.
Bowles, S.; Gintis, H.; Osborne, M. (2001a): Incentive-Enhancing Preferences:
Personality, Behavior, and Preferences. American Economic Review, vol. 91, pp. 155-
158.
Bowles, S., Gintis, H.; Osborne, M. (2001b): The Determinants of Earnings: A Behavioral
Approach. Journal of Economic Literature, vol. 39, pp. 1137-1176.
Boucher, J.; Mutimer, B. (1995): The Relative Age Phenomenon in Sport: A Replication
and Extension with Ice Hockey. Research Quarterly for Exercise & Sport, vol. 65,
pp. 377-381.
Boudreau, J.; Boswell, W.; Judge, T. (2001): Effects of Personality on Executive Career
Success in the United States and Europe. Journal of Vocational Behavior, vol. 58,
pp. 53-81.
Boyko, R.; Boyko, A.; Boyko, M. (2007): Referee Bias contributes to home advantage in
English Premiership Football. Journal of Sports Sciences, vol. 25, pp. 1185-1194.
Braakmann, N. (2009): The Role of Psychological Traits for the Gender Gap in Full-Time
Employment and Wages: Evidence from Germany. Unpublished manuscript,
Department of Economics, Leuphana University Lueneburg.
X
Brandstätter, H. (1999): Veränderbarkeit von Persönlichkeitsmerkmalen – Beiträge der
Differentiellen Psychologie. In: Sonntag, K. (eds.), Persönlichkeitsentwicklung in
Organisationen. Psychologische Grundlage, Methoden und Strategien. Göttingen, Bern,
Toronto, Seattle: Hogrefe Verlag, pp. 51-76.
Breusch, T.; Pagan, A. (1980): The Lagrange-Multiplier Test and its Applications to Model
Specification in Econometrics. Review of Economic Studies, vol. 47, pp. 239-253.
Brockner, J. (1979): Self-esteem, Self-consciousness, and Task Performance: Replications,
Extensions, and Possible Explanations. Journal of Personality and Social Psychology,
vol. 37, pp. 447-461.
Brown, E.; Spiro, R.; Keenan, D. (1991): Wage and Nonwage Discrimination in
Professional Basketball. Do Fans Affect It? American Journal of Economics and
Sociology, vol. 50, pp. 333-345.
Bryson, A.; Frick, B.; Simmons, R. (2009): The Returns to Scarce Talent: Footedness and
Player Remuneration in European Soccer. Discussion Paper No. 948, Centre for
Economic Performance, London.
Buraimo, B.; Forrest, D.; Simmons, R. (2007): The Twelfth Man? Refereeing Bias in
English and German Soccer. Mimeo, Lancaster University Management School.
Butler, J.; Baumeister, R. (1998): The Trouble with Friendly Faces: Skilled Performance
With a Supportive Audience. Journal of Personality and Social Psychology, vol. 75,
pp. 1213-1230.
Card, D. (1999): The Causal Effect of Education on Earnings. In: Ashenfelter, O.; Card, D.
(eds.): Handbook of Labor Economics, vol. 3, Amsterdam: Elsevier, pp. 1801-1863.
Caspi, A. (1998): Personality Development Across the Life Course. In: Damon, W.;
Eisenberg, N. (eds.), Handbook of Child Psychology, New York: John Wiley and Sons,
vol. 3, pp. 311-388.
Chapman, K.; Southwick, L. (1991): Testing the Matching Hypothesis: The Case of Major-
League Baseball. American Economic Review, vol. 81, pp. 1352-1360.
XI
Charness, G.; Rigotti, L.; Rustichini, A. (2007): Individual Behavior and Group
Membership. American Economic Review, vol. 97, pp. 1340-1352.
Chen, K.-P. (2003): Sabotage in Promotion Tournaments. Journal of Law, Economics, and
Organization, vol. 19, pp. 119-140.
Chen, K.-P. (2005): External Recruitment as an Incentive Device. Journal of Labor
Economics, vol. 23, pp. 259-277.
Coates, D.; Grillo, A. (2009): Does Crime Pay? Evidence from the NHL. Working paper,
University of Maryland, Baltimore County.
Coates, D.; Humphreys, B. (2005): Novelty Effects of New Facilities on Attendance at
Professional Sporting Events. Contemporary Economic Policy, vol. 23, pp. 436-455.
Cobb-Clark, D.; Tan, M. (2009): Noncognitive Skills, Occupational Attainment, and
Relative Wages. IZA Discussion Paper No. 4289, Bonn.
Cornelißen, T.; Pfeifer, C. (2007): The Impact of Participation in Sports on Educational
Attainment: New Evidence from Germany, IZA Discussion Paper No. 3160.
Costa, P.; McCrae, R. (1985): The NEO-PI Personality Inventory. Odessa, FL:
Psychological Assessment Resources.
Costa, P.; McCrae, R. (1988): Personality in Adulthood: A Six-Year Longitudinal Study of
Self-Reports and Spouse Ratings on the NEO Personality Inventory. Journal of
Personality and Social Psychology, vol. 54, pp. 269-290.
Costa, P.; McCrae, R. (1992): The NEO-PI Personality Inventory. Odessa, FL:
Psychological Assessment Resources.
Costa, P.; McCrae, R. (1994): Set Like Plaster: Evidence for the Stability of Adult
Personality. In: Heatherton, T.; Weinberger, J. (eds.), Can Personality Change?
Washington, DC: American Psychological Association, pp. 21-40.
Costa, P.; McCrae, R. (1997): Longitudinal Stability of Adult Personality. In: Hogan, R.;
Johnson, J.; Briggs, S. (eds.), Handbook of Personality Psychology. San Diego:
Academic Press, pp. 269-290.
XII
Coupé, T. (2007): Incentives and Bonuses: The Case of the 2006 World Cup. Kyklos,
vol. 60, pp. 349-359.
Cox, D. (1972): Regression Modells and Life-Tables. Review of Economics and Statistics,
Series B, pp. 187-220.
D’Agostino, R.; Balanger, A.; D’Agostino, R. (1990): A Suggestion for Using Powerful
and Informative Tests of Normality. The American Statistician, vol. 44, pp. 316-327.
Dandy, J.; Brewer, N.; Tottman, R. (2001): Self-Consciousness and Performance
Decrements Within a Sporting Context. The Journal of Social Psychology, vol. 141,
pp. 150-152.
Dawson, P.; Dobson, S. (2009): The Influence of Social Pressure and Nationality on
Individual Decisions: Evidence from the Behavior of Referees. Journal of Economic
Psychology, forthcoming.
Dawson, P.; Dobson, S.; Goddard, J.; Wilson, J. (2007): Are Football Referees Really
Biased and Inconsistent? Evidence from the English Premier League. Journal of the
Royal Statistical Society, Serias A 170, pp. 231-250.
Denny, K.; O’Sullivan, V. (2007): The Economic Consequences of Being Left-Handed:
Some Sinister Results. Journal of Population Economics, vol. 42, pp. 353-374.
Deutsche Fussball-Liga (1995-2007), Bundesliga-Report, Frankfurt.
Deutscher, C. (2009): The Payoff to Leadership in Teams. Journal of Sports Economics,
vol. 10, pp. 429-438.
Deutscher, C.; Frick, B. (2009): Entlohnung von Führungsqualitäten im Professionellen
Teamsport – Eine Empirische Untersuchung Anhand der 1. Fußballbundesliga.
Working Paper, University of Paderborn.
Deutscher, C.; Frick, B.; Prinz, J. (2009): Performance under Pressure: Estimating the
Returns to Mental Strength in Professional Basketball. Working Paper, University of
Paderborn.
XIII
Dey, M. (1997): Racial Differences in National Basketball Association Players’ Salaries:
A New Look. The American Economists, vol. 41, pp. 84-90.
Dhuey, E.; Lipscomb, S. (2008): What Makes a Leader? Relative Age and High School
Leadership, Economics of Education Review, vol. 27, pp. 173-183.
Diamond, D. (2008): The National Hockey League Official Guide & Record Book.
Chicago: Triumph Books.
Digman, J. (1990): Personality Structure: Emergence of the 5-Factor Model. Annual
Review of Psychology, vol. 41, pp. 417-440.
Dilger, A.; Prinz, J. (2004): Hazard-Raten in der NBA. Sportwissenschaft, vol. 34, pp. 327-
340.
DiNardo, J.; Pischke, J.-S. (1997): The Returns to Computer Use Revisited: Have Pencils
Changed the Wage Structure Too? Quarterly Journal of Economics, vol. 112, pp. 291-
303.
Dohmen, T. (2005): Social Pressure Influences Decisions of Individuals. Evidence from
the Behavior of Football Referees. IZA Discussion Paper No. 1595.
Dohmen, T. (2008a): Do Professionals Choke under Pressure? Journal of Economic
Behavior and Organization, vol. 65, pp. 636-653.
Dohmen, T. (2008b): The Influence of Social Forces: Evidence from the Behavior of
Football Referees. Economic Inquiry, vol. 46, pp. 411-424.
Drago, F. (2008): Self-Esteem and Earnings. IZA Discussion Paper No. 3577, Bonn.
Drago, R.; Garvey, G. (1998): Incentives for Helping on the Job: Theory and Evidence.
Journal of Labor Economics, vol. 16, pp. 1-25.
Du, Q.; Gao, H.; Levi, M. (2008): Born Leaders: The Relative-Age Effect and Managerial
Success. University of Columbia – Working Paper Series.
Dudink, A. (1994): Birth Date and Sporting Success. Nature, vol. 368, p. 592.
XIV
Earl, P. (1990): Economics and Psychology: A Survey. Economic Journal, vol. 100,
pp. 256-277.
Eastman, B.; Vincent, C. (2009): Determinants of Pay in the NHL: A Quantile Regression
Approach. Journal of Sports Economics, vol. 10, pp. 256-277.
Eckel, C.; Grossman, P. (2005): Managing Diversity by Creating Team Identity. Journal of
Economic Behavior and Organization, vol. 58, pp. 371-392.
Eschker, E.; Perez, S.; Siegler, M. (2004): The NBA and the Influx of International
Basketball Players. Applied Economics, vol. 36, pp. 1009-1020.
Falter, J.-M.; Pérignon, C.; Vercruysse, O. (2008): Impact of Overwhelming Joy on
Consumer Demand: The Case of a Soccer World Cup Victory. Journal of Sports
Economics, vol. 9, pp. 20-42.
Fair, R. (1994): How Fast do Old Men Slow Down? Review of Economics and Statistics,
vol. 76, pp. 103-118.
Fama, E. (1970): Efficient Capital Markets: A Review of Theory and Empirical Work,
Journal of Finance, vol. 25, pp. 383-417.
Feddersen, A.; Maennig, W.; Borcherding, M. (2006): The Novelty Effect of the New
Football Stadia: The Case of Germany. International Journal of Sport Finance, vol. 1,
pp. 174-188.
Feddersen, A.; Grötzinger, A.; Maennig, W. (2009): Investment in Stadia and Regional
Economic Development: Evidence from FIFA World Cup 2006. International Journal of
Sport Finance, vol. 4, pp. 221-239.
Fertig, M.; Kluve, J. (2005): The Effect of Age at School Entry on Educational Attainment
in Germany. IZA Discussion Paper No. 1507.
Fietze, S.; Holst, E.; Tobsch, V. (2009): Persönlichkeit und Karriere – She’s got what it
takes. SOEPpapers on Multidisciplinary Panel Data Research Nr. 220, DIW Berlin.
Filer, R. (1981): The Influence of Affective Human Capital on the Wage Equation.
Research in Labour Economics, vol. 4, pp. 367-416.
XV
Flossmann, A.; Piatek, R.; Wichert, L. (2007): Going Beyond Returns to Education: The
Role of Noncognitive Skills on Wages in Germany. Working Paper, University of
Konstanz.
Forrest, D.; Goddard, J.; Simmons, R. (2005): Odds-Setters as Forecasters: The Case of
English Football. International Journal of Forecasting, vol. 21, pp. 551-564.
Fortin, N. (2008): The Gender Wage Gap among Young Adults in the United States: The
Importance of Money versus People. Journal of Human Resources, vol. 43, pp. 884-918.
French, M. (2002): Physical Appearance and Earnings. Further Evidence. Applied
Economics, vol. 34, pp. 569-572.
Frick, B. (2003): Contest Theory and Sport. Oxford Review of Economic Policy, vol. 19,
pp. 512-529.
Frick, B. (2004): Does Ownership Matter? Empirical Evidence from the German Wine
Industry. Kyklos, vol. 57, pp. 357-386.
Frick, B. (2007a): Salary Determination and the Pay-performance Relationship in
Professional Soccer: Evidence from Germany. In: Rodriguez, P.; Kesenne, S. Garcia, J.
(eds.), Sports Economics After Fifty Years: Essays in Honour of Simon Rottenberg.
Oviedo: Ediciones de la Universidad de Oviedo, pp. 125-146.
Frick, B. (2007b): The Football Players’ Labor Market: Empirical Evidence From the
Major European Leagues. Scottish Journal of Political Economy, vol. 54, pp. 422-446.
Frick, B. (2008): Globalisierung und Faktormobilität: Der Einfluss des „Bosman-Urteils“
auf die Internationalisierung des Arbeitsmarktes für Fußball-Profis.
Wirtschaftspolitische Blätter, vol. 55, pp. 113-127.
Frick, B. (2009): Globalization and Factor Mobility: The Impact of the “Bosman-Ruling“
on Player Migration in Professional Soccer, Journal of Sports Economics, vol. 10,
pp. 88-106.
XVI
Frick, B.; Gürtler, O.; Prinz, J. (2008): Anreize in Turnieren mit heterogenen Teilnehmern:
Eine empirische Untersuchung mit Daten aus der Fußball-Bundesliga. Zeitschrift für
betriebswirtschaftliche Forschung, vol. 60, pp. 385-405.
Frick, B.; Pietzner, G.; Prinz, J. (2007): Carrer Duration in a Competitive Environment:
The Labor Market For Soccer Players in Germany. Eastern Economic Journal, vol. 33,
pp. 429-442.
Frick, B.; Wagner, G. (1996): Bosman und die Folgen: Das Fußball-Urteil des
Europäischen Gerichtshofes aus ökonomischer Sicht, Wirtschaftswissenschaftliches
Studium, vol. 25, pp. 611-615.
Frick, B.; Gürtler, O.; Prinz, J.; Wiendl, A. (2009): Einkommens- oder
Reputationsmaximierung? Eine empirische Untersuchung der Vergütung und Leistung
von Bundesliga-Schiedsrichtern. Die Betriebswirtschaft, vol. 69, pp. 69-83.
Garcia-Del-Barrio, P.; Pujol, F. (2005): Pay and Performance in the Spanish Soccer
League: Who Gets the Expected Monopsony Rents? Working Paper No. 05/04, Faculty
of Economics, University of Navarra.
Garcia-Del-Barrio, P.; Pujol, F. (2006): Hidden Monopsony Rents in Winner-take-all
Markets: Sport and Economics Contribution of Spanish Soccer Players. Mimeo, Faculty
of Economics, University of Navarra.
Garcia-Del-Barrio, P.; Pujol, F. (2007): Hidden Monopsony Rents in Winner-Take-All
Markets: Sport and Economic Contribution of Spanish Soccer Players. Managerial and
Decision Economics, vol. 28, pp. 57-70.
Garicano, L.; Palacios-Huerta, I. (2005): Sabotage in Tournaments: Making the Beautiful
Game a Bit Less Beautiful. CEPR Discussion Paper: 5231, Centre for Economic Policy
Research, London.
Garicano, L.; Palacios-Huerta, I.; Prendergast, C. (2005): Favoritism under Social
Pressure. The Review of Economics and Statistics, vol. 87, pp. 208-216.
Gautschi, T.; Hangartner, D. (2006): Size Does Matter? Körpergröße, Humankapital und
Einkommen. Soziale Welt, vol. 57, pp. 273-294.
XVII
Gius, M.; Johnson, D. (1998): An Empirical Investigation of Wage Discrimination in
Professional Basketball. Applied Economics Letters, vol. 5, pp. 703-705.
Glamser, F.; Vincent, J. (2004): The Relative Age Effect Among Elite American Youth
Soccer Players. Journal of Sport Behavior, vol. 27, pp. 31-38.
Gray, R. (2004): Attending to the Execution of a Complex Sensorimotor Skill: Expertise
Differences, Choking and Slumps. Journal of Experimental Psychology, vol. 10, pp. 42-
54.
Gürtler, O. (2008): On Sabotage in Collective Tournaments. Journal of Mathematical
Economics, vol. 44, pp. 383-393.
Gürtler, O.; Münster, J. (2008): Sabotage in Dynamic Tournaments. Mimeo, University of
Bonn and Free University of Berlin.
Hagn, F.; Maennig, W. (2008): Employment Effects of the Football World Cup 1974 in
Germany. Labour Economics, vol. 15, pp. 1062-1075.
Hagn, F.; Maennig, W. (2009): Large Scale Events and Unemployment: The Case of the
2006 Soccer World Cup in Germany. Applied Economics, vol. 41, pp. 3295-3302.
Halvorsen, R.; Palmquist, R. (1980): The Interpretation of Dummy Variables in
Semilogarithmic Equations. American Economic Review, vol. 70, pp. 474-475.
Hamermesh, D.; Biddle, J. (1994): Beauty and the Labor Market. American Economic
Review, vol. 84, pp. 1174-1194.
Hamermesh, D.; Meng, X.; Zhang, J. (2002): Dress for Success – Does Primping Pay?
Labour Economics, vol. 9, pp. 361-373.
Hamilton, B. (1997): Racial Discrimination and Basketball Salaries in the 1990s. Applied
Economics, vol. 29, pp. 287-296.
Harbring, C.; Irlenbusch, B. (2004): Anreize zu produktiven und destruktiven
Anstrengungen durch relative Entlohnung. Zeitschrift für betriebswirtschaftliche
Forschung, vol. 56, pp. 546-576.
XVIII
Harbring, C.; Irlenbusch, B. (2005): Incentives in Tournaments with Endogenous Prize
Selection. Journal of Institutional and Theoretical Economics, vol. 161, pp. 636-663.
Harbring, C.; Irlenbusch, B.; Kräkel, M.; Selten, R. (2007): Sabotage in Corporate
Contests - An Experimental Analysis. International Journal of the Economics of
Business, vol. 14, pp. 367-392.
Hausman, J. (1978): Specification Tests in Econometrics. Econometrica, vol. 46, pp. 1251-
1271.
Hausman, J.; Leonard, G. (1997): Superstars in the National Basketball Association:
Economic Value and Policy. Journal of Labor Economics, vol. 15, pp. 586-624.
Heaton, A.; Sigall, H. (1991): Self-Consciousness, Self-Presentation, and Performance
under Pressure: Who Chokes, and When? Journal of Applied Social Psychology,
vol. 21, pp. 175-188.
Heckmann, J.; Stixrud, J.; Urzua, S. (2006): The Effects of Cognitive and Noncognitive
Abilities on Labor Market Outcomes and Social Behavior. Journal of Labor Economics,
vol. 24, pp. 411-482.
Heineck, G. (2005): Up in the Skies? The Relationship between Body Height and Earnings
in Germany. Labour, vol. 19, pp. 469-489.
Heineck, G.; Anger, S. (2009): The Returns to Cognitive Abilities and Personality Traits in
Germany. Labour Economics, forthcoming.
Helsen, W.; Starkes, J.; Van Winckel, J. (1999) The Influence of Relative Age on Success
and Dropout in Male Soccer Players. American Journal of Human Biology, vol. 10,
pp. 791-798.
Helsen, W.; Starkes, J.; Van Winckel, J. (2000): Effect of a Change in Selection Year on
Success in Male Soccer Players. American Journal of Human Biology, vol. 12, pp. 729-
735.
Helsen, W.; Van Winckel, J.; Williams, M. (2005): The Relative Age Effect in Youth
Soccer Across Europe. Journal of Sports Sciences, vol. 23, pp. 629-636.
XIX
Hendriks, A.; Hofstee, W.; De Raad, B; Angleiter, A. (1999): The Five-Factor Personality
Inventory (FFPI). Personality and Individual Differences, vol. 27, pp. 307-325.
Heyne, M.; Maenning, W.; Süssmuth, B. (2007): Mega-Sporting Events as Experience
Goods. International Association of Sports Economists Working Paper 0706.
Hill, J. (2004): Pay Discrimination in the NBA Revisited. Quarterly Journal of Business
and Economics, vol. 43, pp. 81-92.
Hoang, H.; Rascher, D. (1999): The NBA, Exit Discrimination, and Career Earnings.
Industrial Relations, vol. 38, pp. 69-91.
Hotchkiss, J.; Moore, R.; Zobay, S. (2003): Impact of the 1996 Summer Olympic Games on
Employment and Wages in Georgia. Southern Economic Journal, vol. 69, pp. 691-704.
Huebl, L.; Swieter, D. (2002): Der Spielermarkt in der Fußball-Bundesliga. Zeitschrift für
Betriebswirtschaft, Ergänzungsheft 4, Sportökonomie, vol. 72, pp. 105-125.
Humphreys, B.; Zimbalist, A. (2008). The Financing and Economic Impact of the Olympic
Games. In Humphreys, B.; Howard, D. (eds.): The Business of Sports Volume 1:
Perspectives on the Sports Industry. Westport CT: Praeger, pp. 101-124.
Hübler, O. (2009): The Non-linear Link between Height and Wages in Germany, 1985-
2004. Economics and Human Biology, 9, in print.
Idson, T.; Kahane, L. (2000): Team Effects on Compensation: An Application to Salary
Determination in the National Hockey League. Economic Inquiry, vol. 38, pp. 345-357.
Ippolito, R. (1996): A Study of Wages and Reliability. Journal of Law and Economics,
vol. 39, pp. 149-189.
Itoh, H. (1994): Job Design, Delegation and Cooperation: A Principal-Agent Analysis.
European Economic Review, vol. 38, pp. 691-700.
Jang, K.; Livesley, W.; Vernon, P. (1996): Heritability of the Big Five Personality
Dimensions and Their Facets. Journal of Personality, vol. 64, pp. 577-591.
Jencks, C. (1979): Who Gets Ahead? New York: Basic Books.
XX
Jenkins, J. (1996): A Reexamination of Salary Discrimination in Professional Basketball.
Social Science Quarterly, vol. 77, pp. 594-608.
Jinks, C. (1964): An Investigation Into the Effect of Date of Birth on Subsequent School
Performance. Educational Research, vol. 6, pp. 220-225.
Jones, J.; Walsh, D. (1988): Salary Determination in the National Hockey League: The
Effects of Skills, Franchise Characteristics, and Discrimination. Industrial and Labor
Relations Review, vol. 41, pp. 592-604.
Jones, J.; Nadeau, S.; Walsh, D. (2001): The Wages of Sin: Employment and Salary Effects
of Violence in the National Hockey League. Atlantic Economic Journal, vol. 25,
pp. 191-206.
Jones, M. (2007): Home Advantage in the NBA as a Game-Long Process. Journal of
Quantative Analysis in Sports, vol. 3, Art. 2.
Jones, M. (2008): A Note on Team-Specific Home Advantage in the NBA. Journal of
Quantative Analysis in Sports, vol. 4, Art. 5.
Judge, T.; Bono, J. (2000): Five-Factor Model of Personality and Transformational
Leadership. Journal of Applied Psychology, vol. 85, pp. 751-765.
Kahn, L. (1991): Discrimination in Professional Sports: A Survey of the Literature.
Industrial and Labor Relations Review, vol. 44, pp. 395-418.
Kahn, L. (1993): Free Agency, Long-Term Contracts and Compensation in Major League
Baseball: Estimates from Panel Data. Review of Economics and Statistics, vol. 75,
pp. 157-164.
Kahn, L. (2000): The Sports Business as a Labor Market Laboratory. Journal of Economic
Perspectives, vol. 14, pp. 75-94.
Kahn, L.; Sherer, P. (1988): Racial Differences in Professional Basketball Players’
Compensation. Journal of Labor Economics, vol. 6, pp. 40-61.
Kalter, F. (1999): Ethnische Kundenpräferenzen im professionellen Sport? Der Fall der
Fußballbundesliga, Zeitschrift für Soziologie, vol. 28, pp. 219-234.
XXI
Kavetsos, G.; Szymanski, S. (2009): National Well-Being and International Sports Events.
Journal of Economic Psychology, forthcoming.
Kicker (1995-2007), Sonderhefte „Bundesliga”, Nürnberg: Olympia-Verlag.
Kahane, L. (2005): Production Efficiency and Discriminatory Hiring Practices in the
National Hockey League: A Stochastic Frontier Approach. Review of Industrial
Organization, vol. 27, pp. 47-71.
Koch, J.; Vander Hill, C. (1988): Is There “Discrimination” in the “Black Man’s Game”?
Social Science Quarterly, vol. 69, pp. 83-94.
Koenker, R. (2005): Quantile Regression, Cambridge: Cambridge University Press.
Konrad, K. (2000): Sabotage in Rent-Seeking Contests. Journal of Law, Economics and
Organization, vol. 16, pp. 155-165.
Konrad, K. (2007): Strategy in Contests - an Introduction. WZB Discussion Paper SP II
2007 – 01, Berlin.
Kräkel, M. (2005): Helping and Sabotaging in Tournaments. International Game Theory
Review, vol. 7, pp. 211-228.
Krueger, A. (1993): How Computers Have Changed the Wage Structure: Evidence from
Microdata, 1984-1989. Quarterly Journal of Economics, vol. 108, pp. 33-60.
Krueger, A. (2005): The Economics of Real Superstars: The Market for Rock Concerts in
the Material World. Journal of Labor Economics, vol. 23, pp. 1-30.
Kuhn, P.; Weinberger, C. (2005): Leadership Skills and Wages, Journal of Labor
Economics, vol. 23, pp. 395-436.
Kura, T.; Matsuzawa, Y. (2006): Birth Month and Sporting Success in Japan. Memoriam
Faculty of Science, Kyoto University, vol. 18, pp. 83-85.
Kuypers, T. (2000): Information and Efficiency: An Empirical Study of a Fixed Odds
Betting Market. Applied Economics, vol. 32, pp. 1353-1363.
XXII
Laband, D.; Lentz, B. (1999): The Impact of Having a Mentor on Earnings and Promotion:
Evidence from a Panel Study of Lawyers. Applied Economics Letters, vol. 6, pp. 785-
787.
Lavoie, M. (2000): The Location of Pay Discrimination in the National Hockey League.
Journal of Sports Economics, vol. 1, pp. 401-411.
Lazear, E. (1989): Pay Equality and Industrial Politics. Journal of Political Economy,
vol. 97, pp. 561-580.
Lazear, E.; Rosen, S. (1981): Rank-order Tournaments as Optimum Labor Contracts.
Journal of Political Economy, vol. 89, pp. 841-864.
Leavitt, J. (1979): Cognitive Demands of Skating and Stickhandling in Ice Hockey.
Canadian Journal of Sports Sciences, vol. 4, pp. 46-55.
Lee, K. (2006): Personality and Earnings. The Levy Economics Institute Working Paper
No. 443, Bard College.
Leeds, M.; Leeds, E.M (2009): International Soccer Success and National Institutions.
Journal of Sports Economics, vol. 10, pp. 351-390.
Lehmann, E. (2000): Verdienen Fußballspieler was sie verdienen? In H.-M. Schellhaaß
(ed.) Sportveranstaltungen zwischen Liga- und Medieninteresse. Karl Hofmann:
Schorndorf, pp. 97-121.
Lehmann, E.; Schulze, G. (2008): What Does it Take to be a Star? The Role of
Performance and the Media for German Soccer Players. Applied Economics Quarterly,
vol. 54, pp. 59-70.
Lehmann, E.; Weigand, J. (1999): Determinanten der Entlohnung von Profi-
Fußballspielern – Eine empirische Analyse für die deutsche Bundesliga.
Betriebwirtschaftliche Forschung und Praxis, vol. 51, pp. 124-155.
Leith, L. (1988): Choking in Sports: Are We Our Own Worst Enemies? International
Journal of Sports Psychology, vol. 19, pp. 59-64.
XXIII
Lewis, M. (2003): Moneyball: The Art of Winning an Unfair Game. New York: W.W.
Norton & Company.
Liu, H.-K.; Wong, Y. (2005): Do Loyal Subordinates Get Higher Wages? Evidence from
China. Applied Economics Letters, vol. 12, pp. 649-652.
Longley, N. (1995): Salary Discrimination in the National Hockey League: The Effects of
Team Location. Canadian Public Policy, vol. 21, pp. 413-422.
Lucifora, C.; Simmons, R. (2003): Superstar Effects in Sport: Evidence from Italy. Journal
of Sports Economics, vol. 4, pp. 35-55.
Maennig, W.; du Plessis, S. (2007): World Cup 2010: Economic Perspectives and Policy
Challenges Informed by the Experience of Germany 2006. Contemporary Economic
Policy, vol. 25, pp. 578-590.
Manning, A.; Swaffield, J. (2008): The Gender Gap in Early-Career Wage Growth.
Economic Journal, vol. 118, pp. 983-1024.
Mas, A.; Moretti, E. (2009): Peers at Work. American Economic Review, vol. 99, pp. 112-
145.
Matheson, V. (2008): Mega-Events: The Effect of the World’s Biggest Sporting Events on
Local, Regional and National Economies. In: Humphreys, B.; Howard, D. (eds.): The
Business of Sports Volume 1 Perspectives on the Sports Industry. Westport CT:
Praeger, pp. 81-100.
Mátyás, L.; Sevestre, P. (1996): The Econometrics of Panel Data. Dordrecht / Boston /
London: Springer.
Mincer, J. (1958): Investing in Human Capital and Personal Income Distribution. Journal
of Political Economy, vol. 66, pp. 281-302.
Mincer, J. (1974): Schooling, Experience, and Earnings. New York: Columbia University
Press.
Monks, J.; Husch, J. (2009): The Impact of Seeding, Home Continent and Hosting on FIFA
World Cup Results. Journal of Sports Economics, vol. 10, pp. 391-408.
XXIV
Moss, P.; Tilly, C. (2001): Stories Employers Tell: Race, Skill, and Hiring in America.
New York: Russell Sage.
Mueller, G.; Plug, E. (2006): Estimating the Effect of Personality on Male and Female
Earnings. Industrial and Labor Relations Review, vol. 60, pp. 3-22.
Münster, J. (2007): Selection Tournaments, Sabotage, and Participation. Journal of
Economics and Management Strategy, vol. 16, pp. 943-970.
Muraven, M.; Tice, D.; Baumeister, R. (1998): Self-control as a Limited Resource:
Regulatory Depletion Patterns. Journal of Personality and Social Psychology, vol. 74,
pp. 774-789.
Murphy, K.; Welch, F. (1990): Empirical Age-Earnings Profiles. Journal of Labor
Economics, vol. 8, pp. 202-299.
Musch, J.; Hay, R. (1999): The Relative Age Effect in Soccer: Cross-Cultural Evidence for
a Systematic Discrimination Against Children Born Late in the Competition Year.
Sociology of Sport Journal, vol. 16, pp. 54-64.
Nevill, A.; Balmer, N.; Williams, A. (2002): The Influence of Crowd Noise and Experience
Upon Refereeing Decisions in Football. Psychology of Sport and Exercise, vol. 3,
pp. 261-272.
Nevill, A.; Holder, R. (1999): Home Advantage in Sport. An Overview of Studies on the
Advantage of Playing at Home. Sports Medicine, vol. 28, pp. 221-236.
Nyhus, E.; Pons, E. (2005): The Effect of Personality on Earnings. Journal of Economic
Psychology, vol. 26, pp. 363-384.
Ohkusa, Y. (2001): An Empirical Examination of the Quit Behavior of Professional
Baseball Players in Japan. Journal of Sports Economics, vol. 2, pp. 80-88.
Osborne Groves, M. (2005): How Important is Your Personality? Labor Market Returns to
Personality for Women in the US and UK. Journal of Economic Psychology, vol. 26,
pp. 827-841.
XXV
Papke, L.; Wooldridge, J. (1996): Econometric Methods for Fractional Response Variables
with an Application to 401(k) Plan Participation Rates. Journal of Applied
Econometrics, vol. 11, pp. 619-632.
Paulus, P. (1983): Group Influence on Individual Task Performance. In: Paulus, P. (ed.):
Basis Group Processes, New York, Springer, pp. 97-120.
Persico, N.; Postlewaite, A.; Silverman, D. (2004): The Effect of Adolescent Experience on
Labor Market Outcomes: The Case of Height. Journal of Political Economy, vol. 112,
pp. 1019-1053.
Pettersson-Lidbom, P.; Priks; M. (2009): Behavior under Social Pressure: Empty Italien
Stadiums and Referee Bias. CESifo Working Paper No. 1960, Munich.
Pidgeon, D. (1965): Date of Birth and Scholastic Performance. Educational Research,
vol. 8, pp. 3-7.
Plug, E. (2001): Season of Birth, Schooling and Earnings. Journal of Economic
Psychology, vol. 22, pp. 641-660.
Pollard, R.; Pollard, G. (2005): Long-term Trends in Home Advantage in Professional
Team Sports in North America and England (1876-2003). Journal of Sports Sciences,
vol. 23, pp. 337-350.
Prinz, J. (2005): Why Are Wages Upward Sloping with Tenure? Munich and Mering:
Hampp.
Puhani, P.; Mühlenweg A. (2010): The Evolution of the School Entry Age Effect in a
School Tracking System. Journal of Human Resources, forthcoming.
Puhani, P.; Weber, A. (2007): Does the Early Bird Catch the Worm? Instrumental
Variable Estimates of Early Educational Effects of Age of School Entry in Germany.
Empirical Economics, vol. 32, pp. 359-386.
Rickmann, N.; Witt, R. (2008): Favouritism and Financial Incentives: A Natural
Experiment. Economica, vol. 75, pp. 296-309.
XXVI
Rosen, S. (1981): The Economics of Superstars. American Economic Review, vol. 71,
pp. 845-858.
Rosen, S. (1983): The Economics of Superstars: Reply. American Economic Review,
vol. 73, pp. 460-461.
Ruebeck, C.; Harrington, J.; Moffitt, R. (2007): Handedness and Earnings. Laterality,
vol. 12, pp. 101-120.
Salgado, J. (1997): The Five Factor Model of Personality and Job Performance in the
European Community. Journal of Applied Psychology, vol. 82, pp. 30-43.
Sauer, R. (1998): The Economics of Wagering Markets. Journal of Economic Literature,
vol. 36, pp. 2021-2064.
Schlenker, B.; Philipps, S.; Boniecki, K.; Schlenker, D. (1995): Championship Pressures:
Chocking or Triumphing in One’s Own Territory. Journal of Personality and Social
Psychology, vol. 68, pp. 632-643.
Schmidt, F.; Ones, D.; Hunter, J. (1992): Personnel Selection. In: Rosenzweig, R.: Porter,
L. (eds.), Annual Review of Psychology. Palo Alto, CA: Consulting Psychologists
Press, pp. 627-670.
Schultz, T. (1961): Investment in Human Capital. American Economic Review, vol. 51,
pp. 1-17.
Scoppa, V. (2008): Are Subjective Evaluations Biased by Social Factors or Connections?
An Econometric Analysis of Soccer Referee Decisions. Empirical Economics, vol. 35,
pp. 123-140.
Simmons, R.; Berri, D. (2009): Gains from Spezialization and Free Agency: The Story
from the Gridiron. Review of Industrial Organization, vol. 34, pp. 81-98.
Spence, A. (1973): Job Market Signaling. Quarterly Journal of Economics, vol. 87,
pp. 355-374.
XXVII
Srivastava, S.; John, O.; Gosling, S.; Potter, S. (2003): Development of Personality in
Early and Middle Adulthood: Set like Plaster or Persistent Change? Journal of
Personality and Social Psychology, vol. 85, pp. 1041-1053.
Stadtmann, G. (2006): Frequent News and Pure Signals: The Case of a Publicly Traded
Football Club. Scottish Journal of Political Economy, vol. 53, pp. 485-504.
Staw, B.; Hoang, H. (1995): Sunk Costs in the NBA: Why Draft Order Affects Playing
Time and Survival in Professional Basketball. Administrative Science Quarterly,
vol. 40, pp. 474-494.
Sutter, M.; Kocher, M. (2004): Favouritism and Agents: The Case of the Referees’ Home
Bias. Journal of Economic Psychology, vol. 25, pp. 461-469.
Szymanski, S.; Kuper, S. (2009): Why England Lose; and other Curious Phenonema
Explained. London: HarperCollins.
Tagger, S.; Hackett, R.; Saha, S. (1999): Leadership Emergence in Autonomous Work
Teams: Antecedents and Outcomes. Personnel Psychology, vol. 52, pp. 591-620.
Taylor, B.; Trogdon, J. (2002): Losing to Win: Tournament Incentives in the National
Basketball Association. Journal of Labor Economics, vol. 20, pp. 23-41.
Taylor, G. (2009): Meltdown: The Nationality of Premier League Players and the Future
of English Football. In: Walters, G.; Rossi, G. (eds.): Labour Market Migration in
European Football: Key Issues and Challenges. Birkbeck Sport Business Centre
Research Paper Series. Volume 2.
Ter Weel, B. (2008): The Noncognitive Determinants of Labor Market and Behavioral
Outcomes. Journal of Human Resources, vol. 43, pp. 729-737.
Tett, R.; Jackson, D.; Rothstein, M. (1991): Personality Measures as Predictors of Job
Performance: A Meta-Analytic Review. Personnel Psychology, vol. 44, pp. 703-742.
Topel, R. (1991): Specific Capital, Mobility, and Wages: Wages Rise with Job Seniority.
Journal of Political Economy, vol. 99, pp. 145-176.
XXVIII
Torgler, B.; Schmidt, S. (2007): What Shapes Players’ Performance in Soccer? Empirical
Findings from a Panel Analysis. Applied Economics, vol. 39, pp. 2355-2369.
Verhulst, J. (1992): Seasonal Birth Distribution of West European Soccer Players: A
Possible Explanation, Medical Hypotheses, vol. 38, pp. 346-348.
Vincent, C.; Eastman, B. (2009): Determinants of Pay in the NHL. A Quantile Regression
Approach. Journal of Sports Economics, vol. 10, pp. 256-277.
Wallace, H.; Baumeister, R.; Vohs, K. (2005): Audience Support and Choking under
Pressure: A Home Disadvantage? Journal of Sports Sciences, vol. 23, pp. 429-438.
Wallace, M. (1988): Labor Market Structure and Salary Determination among
Professional Basketball Players. Work and Occupations, vol. 15, pp. 294-312.
Wang, J.; Marchant, D.; Morris, T.; Gibbs, P. (2004): Self-consciousness and Trait Anxiety
as Predictors of Choking in Sport. Journal of Science and Medicine in Sport, vol. 7,
pp. 174-185.
White, H. (1980): A Heteroscedasticity-Consistent Covariance Matrix Estimator and a
Direct Test for Heteroscedasticity. Econometrica, vol. 48, pp. 817-838.
Wooldridge, J. (2001): Econometric Analysis of Cross Section and Panel Data. Cambridge
and London, MIT Press.
Zellner, A. (1962): An Efficient Method of Estimating Seemingly Unrelated Regressions
and Tests for Aggregation Bias. Journal of the American Statistical Association, vol. 57,
pp. 348-368.
Internet sources:
http://www.82games.com
http://www.basketball-reference.com
http://www.bundesliga.de
http://www.eskimo.com/~pbender/
XXIX
http://www.forbes.com
http://www.kicker.de
http://www.nba.com
http://www.nhl.com
http://www.oddset.de
