Do Midtable Premier League Teams Perform Worse At The End Of The Season?

By Andrew Puopolo

With the international break coming to a close this weekend, we are heading into the homestretch of the Premier League season. With seven or eight matches left for every Premier League team, the table is roughly split into three major subgroups (four, if you consider how far ahead Manchester City are). At the top end of the table, you have six teams fighting for places in next year’s UEFA Champions League and UEFA Europa League. Behind them, you have about five teams (Burnley, Leicester, Everton, Bournemouth and Watford) who are fighting for a possible spot in the third qualifying round of next year’s Europa League but are not in danger of relegation. Behind that, you have 8 teams fighting to avoid two relegation places (with West Bromwich Albion cut adrift at the bottom of the table and destined for the Championship).

Last year, FourFourTwo wrote an article about how teams started to achieve fewer points after reaching the 40 point barrier that is the commonly accepted magic number to ensure safety in the Premier League using data from the 2016/17 Premier League season. Their two most jarring examples were West Bromwich Albion, who reached the 40 point barrier after only 26 matches but only attained 5 from their final 12 matches and Watford who hit the 40 point barrier after 32 matches and failed to pick up a point the rest of the season. Arsene Wenger complained last season that “some teams turn up, some teams are on holiday” at the end of the season after West Ham lost 4-0 at home to Liverpool, which severely damaged any hopes Arsenal had of qualifying for the Champions League for a 20th consecutive season.

This got me wondering; was last season an aberration where teams were noticeably worse when they had nothing to play for or has this trend been present for a longer period of time?

I set out to answer this question in the following way. For each season dating back to 1996/7, I constructed the Premier League table five weeks before the end of the season.

Using that table, I identified teams that had nothing to play for. If a team was more than 6 points below from the last European qualifying position and more than the number of matches left ahead of the relegation zone, then they would be classified as having nothing to play for.

For example, let us use the 2015/16 Premier League table on April 11th as an example:

The top seven teams in this season qualified for Europe, so Chelsea still had a Europa league spot to play for. Down at the bottom end, all teams that had more than 33 points were assumed safe as they were more than 6 points above the relegation zone. From this, we get the teams that were 11th to 16th had nothing to play for.

In these seasons, we identified 132 teams that had nothing to play for at the end of the season. To figure out if these teams performed better or worse on average in their final matches of the season, we conducted the following analysis:

1.) For each team in the group, we calculated the average number of points attained in the final matches of the season and subtracted it from the average points in the first part of the season. We then averaged all these values to obtain a test statistic of .113.
2.) For each team in the group, we randomly selected a random consecutive sequence of matches from the first part of the season and calculated the average points per match in that sequence.
3.) We calculated the average points per match in all other matches during that season outside of the sequence, and subtracted it from our result in step 2.
4.) We took the average difference across all teams.
5.) We repeated Steps 2 and 3 10,000 times to generate a null distribution of average difference.

For each of the 10,000 trials, we calculated the difference for each of the 132 teams in the dataset and then averaged those differences. Our null hypothesis is that the average difference in points per match between each subset of matches is 0.

After producing a histogram of our 10,000 means, we get the following results:

We find that of the 10,000 trials, only 16 of them result in an average that is greater than our test statistic of .118, which gives us a p-value of .0016 for our hypothesis. This means that we can conclude that teams that have little to play for at the end of the season perform significantly worse at the end of the season.

One potential confounding factor at play here is that teams reach safety by playing easier matches earlier and thus there is selection bias in their final matches by playing against the top teams. However, by sampling random consecutive sets of matches, this issue should be mostly alleviated.

But how do these results affect this year’s Premier League relegation battle? Which teams have the most hidden benefit from playing midtable teams in these last 7 or 8 games? To do that, I took the remaining schedules of the 8 teams fighting for relegation. Matches highlighted in red are matches against the top six teams in the Premier League, matches in yellow are against the other teams on this list and matches in green are against teams with little to play for.

The team that most sticks out here is Crystal Palace. Of their seven remaining matches, this weekend’s match against Liverpool at Selhurst Park is their only match against one of the Premier League’s elite. After that, they have four of their final six matches against teams that will have little to play for and their last two matches are against the two bottom clubs. Another team that sticks out is Crystal Palace’s M23 rivals Brighton and Hove Albion. They play 4 of the top 6 in their last 8 matches, while only two against mid table sides. Brighton are already on 34 points, and likely only need two wins from their last eight to stay up. Things look less promising for Stoke City, who find themselves three points from safety with seven to play. Their only match against a midtable side is Burnley, so in all likelihood their three relegation six pointers (especially the West Ham match) will seal their fate.

It will be interesting to see how this plays out, but don’t be surprised if some teams start playing quite differently to the formbook.

If you have any questions for Andrew, please reach out to him at andrewpuopolo@college.harvard.edu or on Twitter @andrew_puopolo.

Special thanks to Laurie Shaw at the EightyFivePoints for supplying the data and helping to edit this article.

About the author

harvardsports

View all posts

3 Comments

Leave a Reply

Your email address will not be published. Required fields are marked *