Red cards, referee home bias and social pressure: evidence from English Premiership Soccer

Abstract

This article uses player/match-level data from five seasons of the English Premiership League (EPL) to test for home bias and referee susceptibility to social pressure in the application of the most stringent disciplinary sanction available to a referee. We find persuasive evidence for the former but none for the latter.

Keywords:

• English Premiership Football
• disciplinary cards
• referee bias
• probability models
• player fixed effects

JEL Classification:

• D80
• J20I

Acknowledgements

We are grateful to an anonymous referee of this journal for a set of useful and constructive comments, which have enhanced this article. We are also grateful to the Nuffield Foundation for financial support (under award SGS/36059). The excellent research assistance provided by Edgar Cooke and Kalle Hirvonen is acknowledged. The usual disclaimer applies.

Notes

¹ Nevill et al. (2002) present experimental evidence suggesting that some decisions made by English Premiership League (EPL) referees are subject to the influence of crowd noise. The possibility that player behaviour on the field of play is also amenable to the influence of social pressure is acknowledged. This emphasizes the importance, when assessing the impact of social pressure on referees, to control for player-specific effects in the empirical analysis.

² Baltagi (2008, pp. 237–40) provides an accessible account of Chamberlain's estimation procedure. We also experimented with the use of a random effects logit model, but these effects were found to be insignificant in all cases suggesting such a model was never preferred statistically to a standard nonpanel logit. In addition, team and season effects were not included in any of the reported specifications given their statistical insignificance.

³ The data do not distinguish between dismissals for violent conduct and those for other (less serious) misdemeanours. The former behaviour invariably incurs a straight red card. However, the inclusion of the yellow card variable in the empirical specifications allows us to partly control for this distinction as it captures whether or not a player has already received a caution prior to dismissal.

⁴ The Chamberlain logit coefficients can be scaled by 0.00586 in this application to obtain a rough estimate of the impact effect for the ‘home fixture’ variable. This yields an estimate of −0.0018. Thus, relative to the sample average of 0.0059 (see Table 1), the dismissal rate for ‘home’ team players is lower by about one-third compared to visiting team players, on average and ceteris paribus.

⁵ The Chow version of the likelihood ratio test confirms that data separation by fixture status is permissible for the Chamberlain model but not for the nonpanel logit model. The prob-values for the relevant test statistics for these two models are 0.000 and 0.224, respectively.