Less than 4 weeks ago, France defeated Croatia 4-2 in Moscow to claim the 2018 FIFA World Cup. Over the last few weeks, soccer fans have been treated to a series of uninspiring preseason friendlies where many of the top players from the world’s top clubs have been absent due to their exertions in the World Cup. In the most extreme example of this, Tottenham Hotspur have been without 9 of their key players for their entire four match preseason tour of the United States and Spain, and those players will only be returning to training this week ahead of their opening match on Saturday lunchtime away to Newcastle United. This has led to preseasons being disrupted heavily across the Premier League as well as transfer plans being put on hold, leading to a frantic last week for many clubs before the season opens on Friday at Old Trafford between Manchester United and Leicester City.
Newspaper journalists speculate the existence of something called a “World Cup Hangover”, where players who have had their off-season’s cut short due to the World Cup perform worse in the following season. I wanted to see if there was any empirical evidence of this hypothesis being true, and if fans of teams who had many players playing in the final stages of the World Cup, like Manchester United and Tottenham, have reason for concern?
To help try answer this question, I decided to calculate the number of players each Premier League club has had at each World Cup since 1998 as well as the number of games that team was represented in (to account for certain nations going further in each World Cup and disrupting preseason plans more). I then calculated the change in ELO rating for each team from the start of the season to certain intervals during the season. The intervals chosen were the ELO ratings 1,2,3 months from the start of the season, as well as the ELO rating on New Year’s Day and again at the end of the season.
If there truly is a “World Cup Hangover”, then teams that had more players and games would perform worse than a team that had fewer players, and this would be reflected by the change in their ELO ratings.
I started by running a linear regression of the number of games a club participated in at the World Cup with the change in their ELO rating during the next Premier League Season.
From this, we see a very slight negative correlation between games played at a World Cup and the change in ELO. However, R^2 of this relationship is only .02 and the p value of the predictor is .17, which implies that this is an incredibly weak relationship.
Based on the plot, we can see that there are many outliers present. A majority of the Premier League clubs participated in between 0 and 20 matches at the World Cup, while some clubs participated in as many as 74. These outliers provide leverage to the plot and can seriously affect the regression. Because of this, I devised a couple of ways to account for this.
The first thing I tried was isolating the effect of a World Cup hangover to teams that played a large number of matches. I decided to take out all the clubs that participated in more than 20 games in the World Cup, and constructed a similar plot to above.
From this, we can see that the slope of the regression line is incredibly close to 0, and there is essentially no correlation between the two variables.
However, there is a confounding variable at play here. As you might expect, there is an extremely high correlation (.81) between a team’s preseason ELO and the number of games they participated in at the World Cup. The team’s that had high number of players participating were teams like Manchester United, Arsenal and Liverpool and thus we were dealing with a possible error of regression to the mean. Instead of looking at how many games a club participated in, I decided to look at the difference between the number of games participated in and the number of games they would be expected to participate in based on their preseason ELO ranking. To find the expected number of games, I fit a Generalized Additive Model (GAM) to predict how many games a team should have participated in given their ELO ranking, and then calculated the difference between that value and the true number of games they participated in. To account for the larger residuals between teams that played in more games compared to their expected values, I also computed the ratio between the true number of games played and the expected number of games played. The thought process behind this adjustment was that teams that had more players in the World Cup relative to their ability level would perform worse. For example, in 2014 Newcastle United participated in 29 games at the World Cup when their ELO rating of 1646 would have implied that they only were expected to participate in 11. Thus, we would expect Newcastle to suffer in 2014/15 due to their extra exertions at the World Cup, and indeed their ELO rating dropped to 1591. But is there any extra predictive power in this that was not originally picked up from using the raw games?
Again, this plot has a very low R^2 (.014) and the p value of the predictor is .23. When we ran the same regression on the ratio of games played to change in ELO rating, our results were similar.
Like with the raw number of games played, we also ran the same analysis with only the teams that played in 20 or more games at the World Cup, and found similar results to above.
Similar analyses were conducted on change in ELO rating at the various points in the season detailed above, and all of them led to similar conclusions to the three shown above.
Based on this, we find that there is very little evidence to suggest the presence of a World Cup hangover in English Premier League results. If you are a Manchester United or Tottenham Hotspur supporter, you should not have too much to be worried about… well at least in terms of a World Cup hangover.
If you have any other suggestions for trying to prove the presence of a hangover, please leave them in the comments below.
If you have any questions or comments for Andrew, please feel free to email him at firstname.lastname@example.org, or reach out to him on Twitter @andrew_puopolo.