The effect of wage dispersion on team outcome and the way team outcome is produced

Abstract

The impact of intra-team pay dispersion on team productivity is a highly discussed issue. On the one hand, wage differentials provide incentives for higher employee effort. On the other hand, pay inequality may reduce team cohesiveness and increase feelings of relative deprivation leading to lower performance. Analysing nonlinear effects of wage dispersion in professional soccer, we find empirical evidence that team performance is strongest when there is either very high or very low wage inequality. Medium levels produce the weakest team performance. In addition, we show that the pay structure affects the team's playing style even after controlling for team and coach heterogeneity. We discuss the theoretical and managerial implications as well as the limits of generalization.

Acknowledgements

We are grateful to David Forrest, Jaume Garcia, Michael Kosfeld, Andrew Oswald, the editor Mark Taylor, three anonymous referees, and to the seminar participants of the Western Economic Association conference 2007 in Seattle, US, of the Scientific Commission Organization 2008 in Munich, and of the 11th Colloquium of Personnel Economics 2008 in Bonn for helpful comments. Nicolai Grüter provided excellent research assistance. Remaining errors, are of course, our own.

Notes

¹ Exceptions are Winter-Ebmer and Zweimüller (1999), Bingley and Eriksson (2001), Brown et al. (2003), Grund and Westergaard-Nielsen (2008), which include a squared term of wage dispersion in their econometrical models.

² The following parameterization builds on the parameterization proposed by Grund and Westergaard-Nielsen (2005).

³ Allison (1978) discusses several measures of inequality and finds that both the Gini index and the coefficient of variation are advantageous in many respects. Harrison and Klein (2007) also recommend the same two measures to capture the effects of pay disparity. We did not use the Herfindahl index as measure of a team's wage dispersion because the potential range of the Herfindahl index is affected by the team's roster size.

⁴ Regarding the reliability of the Kicker salary proxies, see also Frick (2007) and Torgler and Schmidt (2007).

⁵ However, this does not contradict our hierarchical pay hypothesis. Even players, whose market values are at the low end of the distribution in a given season, may be motivated by greater pay dispersion to display better field performances in order to receive a higher base salary next season.

⁶ Bingley and Eriksson (2001), Lallemand et al. (2004) and Heyman (2005) address the issue of simultaneity by using income tax information excluding bonuses or lagged predetermined values of wage dispersion as instruments of the current pay inequality.

⁷ Wage expenditures are expressed in 2003 Euro and adjusted for inflation.

⁸ An anonymous referee correctly noted that the expert evaluations may be endogenous. Since we measure both individual playing talent and team performance on the same time interval, we cannot exclude the possibility that the expert evaluations are shaped by team productivity rather than vice versa. However, two points are worth making. First, match-specific analyses show that whereas the average of expert evaluations strongly depends on the game's result, the SD of individual expert evaluations is very similar for the winning and the losing teams. Thus, there is no significant relation between talent heterogeneity and team performance on a match-level basis. Second, due to the longitudinal nature of our data set, we can test whether current team performance affects the future talent heterogeneity of a team. The assumption of strict exogeneity of talent heterogeneity is not rejected (see next section).

⁹ See e.g. Kyriazidou (1997) for a procedure to also account for nonconstant selection effects.

¹⁰ Detailed regression results are available form the authors upon request.

¹¹ In the season 2004/05 the qualified clubs received in total €414.1 million of broadcasting income and generated substantial extra match day turnover.

¹² We use gate attendance instead of gate revenue as instrument because revenue data is partly not available for precedent seasons.

¹³ Since we do not have multiple endogenous variables, we do not need to report the Shea partial R ² measure that takes the intercorrelations among the instruments into account (Shea, 1997).

¹⁴ Team level effects are jointly significant in most of the models, implying that a pooled regression is not suitable. The Hausman specification test reveals that the team level effects correlate with the regressors, implying that the fixed-effects approach is appropriate. We do not find evidence for potential nonlinearity.

¹⁵ We are grateful to an anonymous referee for raising this issue.