Abstract
In this article, we investigate the pay–performance relationship of soccer players using individual data from eight seasons of the German soccer league Bundesliga. We find a nonlinear pay–performance relationship, indicating that salary does indeed affect individual performance. The results further show that player performance is affected not only by absolute income level but also by relative income position. An additional analysis of the performance impact of team effects provides evidence of a direct impact of team-mate attributes on individual player performance.
Acknowledgements
For helpful comments and suggestions, thanks are due to Doris Aebi, Julia Angelica, Bruno S. Frey, the editor Mark Taylor and an anonymous referee.
Notes
1 For instance the FIFA World Cup has become a major spectacle and one of the world's largest sporting events, having been broadcast in 2002 in more than 200 countries and regions around the world, covering over 41 100 h of programming and reaching an estimated 28.8 billion television viewers (see FIFA Media Information, 21 November 2002, http://fifaworldcup.yahoo.com).
2 In England, for example, clubs listed on the stock exchange must publish their annual reports (Kern and Süssmuth, Citation2003).
3 It was impossible to include 1997 because player salary information was unavailable.
4 These data are then used in a so-called manager game (Kicker-Managerspiel), in which individuals (aged 18+) can participate. Compared to the performance data, the salary variable has some missing values. However, imputing missing values on the basis of the other independent variables produces similar results.
5 See also discussion in Section IV.
6 Historical data are not provided by Transfermarkt.de, as the site has only just begun to collect this information.
7 In general, both arts and sports have many similarities such as community impacts, demand interdependencies and the presence of superstars (Seaman, Citation2003).
8 The advantage of this class of estimators is that they do not require a precise modelling of the source of heteroscedasticity. Clustering tend to increase the reported SDs, which reduces the statistical significance levels of the estimated coefficients.
9 In both cases yes = value of 1.