Two weeks ago, I wrote an article discussing whether midtable teams perform worse at the end of the Premier League season because they have nothing to play for. But what about those teams who are desperately trying to avoid relegation? Are they taking points from those midtable teams? Does desperation set in and lead to relegation teams crumbling? Or does this vary by team and average itself out in the long run? Everyone remembers Sunderland’s great escape in 2014 or Leicester in 2014/2015 winning seven of their last nine matches to avoid relegation, but are those examples an anomaly or part of a larger trend of teams improving their results at the end of the season?
The methodology used was very similar to the methodology used in the post about midtable teams. To start, I constructed the Premier League table 5 weeks before the end of the season for every season from 1996/97 to 2016/17. If a team was less than 6 points above the relegation zone, then they were considered to be a relegation contender. However, if a team in the relegation zone was more than 6 points from safety, then they were considered to be doomed and not considered in this analysis (think teams like West Brom this year and Sunderland last year) as they also had very little to play for. For example, let us examine the table with 5 weeks to go in the 2015/16 season to help illustrate how we are identifying relegation threatened teams:
The three teams highlighted in red were in danger of relegation, while Aston Villa were too far behind to still have anything to play for.
We found that in the 20 seasons studied, there were 100 teams that fit the criteria described above. We then conducted the following analysis:
1.) We calculated the average points per match for each team in the final five weeks of the season and subtracted it from their average points per match in all other matches during that season. We then calculated the average difference across all teams in the dataset. Our average difference was -.22, which meant that relegation threatened teams on average attained .22 points per match more at the end of the season than in their previous matches.
2.) For each of the teams in the dataset, we sampled a random subset of consecutive matches equal to the number of matches used in the analysis in step 1. We calculated the average points per match in those matches and subtracted it from the average points per match in all other matches during the season. We then calculated the average of this difference for all teams in the dataset.
3.) We repeated step 2 10,000 times to create a null distribution for average difference in points over a random sequence of matches.
After running this analysis, we were able to produce a histogram of our average difference of our null distribution across the 10,000 trials:
We found that of the 10,000 simulations, 0 of them resulted in a test statistic that was less than the -.22 found by our 100 teams. This implies that there is incredibly strong evidence to conclude that teams fighting relegation start improving their performance at the end of the season in a last ditch effort to avoid the drop.
There are a couple of possible explanations for this. It is likely that these teams who have everything to play for are taking more points from teams that are in midtable and have nothing to play for. Another possible explanation is teams that are fighting relegation are playing weaker teams at the end of the season (and had fewer points to begin with because of a harder schedule) and are able to easily pick up more points. However, by creating a large sample size for simulation, it is much more likely that the first explanation is accurate.
With only the relegation places and the final Europa League place still to be decided in this Premier League season, do not be surprised to see teams like Stoke City, Southampton, Huddersfield and Crystal Palace start improving in these final few weeks of the season.
If you have any questions for Andrew, please reach out to him on Twitter @andrew_puopolo or by email at andrewpuopolo@college.harvard.edu.