Team performance and race: evidence from the English and French national soccer teams

Abstract

This study examines the performance of the men’s senior national soccer teams of England and France between the years 1996 and 2011. The article models team performance using an array of match-specific variables including the racial composition of the starting XI. We find no definitive evidence that the racial composition of the national teams exerts an influence on match outcome for either country over the period reviewed. In addition, while there is no statistical support for the view that the identity of the England national team manager is an important determinant of performance, there is mild evidence that the converse is the case for France.

Keywords:

• soccer
• match outcome
• ordered probit
• race

JEL Classification:

• L83
• C25
• J71

Acknowledgements

The authors are grateful to a referee of this journal for a set of constructive comments and to Hector Gutierrez Rufrancos for research assistance. The standard disclaimer applies.

Notes

¹ Diagne was originally from French Guyana and of Senegalese descent. He represented France in the 1938 World Cup and after his playing career ended became the first coach of the Senegal national team.

² It is sometimes argued that the England national side’s first black player should have been Plymouth Argyle’s prolific London-born inside-left forward John Francis (Jack) Leslie, who scored 132 goals in 384 club appearances between 1921 and 1934. He was apparently selected to play for England but, it is alleged, his call-up was rescinded when English FA officials discovered he was black.

³ This account was reported in the Guardian newspaper (issue dated 24 June 2004) by journalist Vivek Chaudhary.

⁴ For example, the then National Front leader Jean Marie Le Pen protested that the side that won the 1998 World Cup was not sufficiently French and returned to the same theme in 2006 arguing that the national coach had selected too many black players.

⁵ These channels obviously owe their intellectual origin to the work of Becker (1957). Of course, it is possible that the desire for success on the field of play will trump the racist card across all of these four dimensions. This may be particularly the case for England given its protracted period of failure since 1966.

⁶ The chi-squared test of independence of match outcome with respect to venue location is computed as 3.142 (df = 4 and p-value = 0.534) for England and 6.244 (df = 4 and p-value = 0.182) for France. In addition, the win rate for France is statistically higher than that recorded for England on the basis of a z-score (|z| = 2.69 and p-value = 0.007).

⁷ The average French nonwhite rate was 0.61, while the comparable English rate was 0.29. A test for the statistical differences between these rates yields a |z| = 6.2 and a p-value = 0.000.

⁸ Reilly and Witt (2011) reported, using data from five recent English Premiership League seasons, that about one-third of players in this league were nonwhite with the rate for players that qualify for England reported as 0.325. The sample average proportion of nonwhites in the national team’s starting XI from Table A1 of the appendix is 0.289 and the differential in these point estimates is not found to be statistically significant at a conventional level (|z| = 0.869 and p-value = 0.385).

⁹ If we exclude the two caretaker managers, the parametric ANOVA test yields a statistically significant F(4,166) = 34.2, whereas the nonparametric Kruskall–Wallis test also returns a statistically significant chi-squared value of 81.0 with four degrees of freedom.

¹⁰ The t-tests for the proposition of no mean difference in team racial composition across these two broad categories of manager is 8.01, the z-score for the nonparametric Mann–Whitney U-test is 6.96.

¹¹ Using the French data, the ANOVA test yields a statistically significant F(4,195) = 31.0, whereas the Kruskall–Wallis test provides a statistically significant chi-squared value of 74.4 with four degrees of freedom.

¹² The FIFA country rankings are obtained from: http://www.fifa.com/worldranking/rankingtools/index.html. If a ranking is not available in the month the game is played, then the previous month’s ranking is used instead. The only missing rank was for Serbia and Montenegro in the case of one of England’s fixtures, and so the sample average was used to avoid excluding this particular observation from our analysis.

¹³ The ordered probit model has been used extensively in the literature to model soccer match outcomes. For example, see Kuypers (2000), Goddard and Asimakopolous (2004) and Goddard (2005).

¹⁴ In this case, the negative of the estimated constant provides the first threshold or cut-off value (θ₀), while the estimated threshold itself now represents the difference between the second and the first threshold parameters (i.e., θ₁–θ₀).

¹⁵ The signs on the pseudo-residuals for the ordered probit model are known a priori for the bottom (loss) and top (win) categories being negative for the former and positive for the latter. Nevertheless, the test provides evidence on whether the temporal sequence of these signs is random, which is taken to imply that the pseudo-residuals are uncorrelated across time periods. A Spearman rank order correlation is also used to determine the extent of any correlation between the pseudo-residuals in the current and selected previous periods. The sampling variance for this statistic is computed as [n − k]⁻¹, where k is the number of estimated ordered probit parameters.

¹⁶ In addition, there is no evidence of second-, third- and fourth-order serial correlation effects in the pseudo-residuals using Spearman rank order correlation coefficients.

¹⁷ The findings of Kuper and Szymanski (2009) suggest that England underperforms at tournament finals but in noncompetitive fixtures is often able to beat higher ranked teams. A set of interactions between fixture type and country ranking is introduced to test if this proposition finds support in our data. However, the set of interactions proved jointly statistically insignificant.

¹⁸ This is obtained by multiplying the marginal effect by 1/11 = 0.091, the proportional change in a soccer team’s composition corresponding to a single player change. The results obtained are invariant to using the number of nonwhite players in lieu of the racial composition variable.

¹⁹ The positive point estimate we detect, though, has some resonance with the work of Szymanski (2000) who, using data for 39 English football league clubs between 1978/79 and 1992/93 seasons, found that clubs with an above average proportion of nonwhite players systematically out-performed teams with a below average proportion of nonwhite players after controlling for players wages.

²⁰ The inclusion of the national coach dummies in the French case is marginal as the p-value for the test of their joint significance is 0.107.

²¹ For instance, Simon Kuper, journalist and a respected author on football matters, made a strong case in the Financial Times (issue dated 9 February 2012) for casting Capello as ‘…England’s best boss’. In contrast, Richard Williams’ assessment in the following day’s Guardian was more circumspect and emphasized the need to control for opposition quality prior to providing a conclusive judgement on the matter.

²² The findings in regard to managerial performance for England cast some doubt on the wisdom of awarding such a generous remuneration package to a recent incumbent, Fabio Capello, who performed no better, ceteris paribus, than his four immediate full-time predecessors. Capello was estimated to have earned over £6 million per year during his employment with the FA prior to his premature departure in early 2012.