Premier League forecasters wrong-footed

At the beginning of the 2015/2016 Premier League season, we at Scoreboard Journalism collected 92 pre-season forecasts from journalists, statistical modellers and fans. Steve Lawrence carried out an initial analysis and discovered that this season’s forecasts are way off the mark due to the slightly unusual nature of the current season.

In previous blogs we first analysed the predicted shape of the points distribution before turning to the order and I have repeated the process for this season’s predictions.

Most forecasters are careful with the predicted mean points accumulated and the majority of  estimates are between 51 and 53 with just a few outliers including one at 58.3.

Predicted ranges differ quite substantially from 32 to 66 and given that none of the last ten seasons of the Premier League has finished with a range lower than 47 it means that about a third of the forecasts lie outside two standard deviations of the ranges of the last 10 years.

The developing range on a round by round basis is not out of the ordinary and it looks destined for a final position within 1 SD of the last 10 years results. Having said that the top position is very low in comparison to previous seasons.


Looking at the standard deviation versus the range shows that the former has already developed to the point at which it is passing some of the outlier forecasts and on its current trajectory it will pass at least half of the 92 predictions.


Successive Premier League seasons have shown that the coefficient of variation stabilises somewhere between 0.3 and 0.4 after around nine match rounds and remains fairly constant until the conclusion of the season. This season has been no different so far in following that pattern although the current coefficient of variation is 0.304 which is a little lower than we have seen in previous years.


In the last 10 years there has only been one (2010/2011) in which the coefficient of variation ended below 0.3. There are, however, similarities between this season and that of five years ago


If we compare the points per game v goal ratio charts at this stage for 2010/11 and this season with 2009/10, which has a more common profile, we can see that 2015/16 could be shaping up to be similar to 2010/11 with perhaps only one club achieving more than 2.0 PPG and a very close relegation battle. Five years ago, Manchester United won the league title with 80 points and there were only 10 points between 8th and 19th.


The coefficient of variation captures the forecasters mean and standard deviation in a single number and it’s a useful measure of the shape of each forecast but it does not capture the predicted order so to do that we plot the coefficient of variation against Pearson’s r and this is where things get quite interesting for 2015/16.

If we look at Round 20 for last season it’s clear that the forecasts were already, at that stage, proving quite accurate with the best forecast having a 0.87 correlation coefficient with actual points and the worst having a correlation coefficient of 0.62 with the league table. The bookmakers were also comfortably at the front of the pack and would go on to make even more ground by the end of the season.


This season, the best forecast is @soccermetric with 0.69, not dissimilar to last season’s last place. This season’s last place has only a 0.28 correlation with the league table while the bookmakers are uncomfortably towards the back half of the field.


The numbers bear out the commentary in the media and the reasons for this curious season are well known to be the extraordinary performance of Leicester City and the diabolical performance of Chelsea along with a few other unusual performances elsewhere in the league.

If at the end of the season we are still looking at a CV v r graph similar to Round 22 there will probably also be a major media debate asking if we are experiencing the demise of the Premier League’s big clubs along with the emergence of canny scouting, the use of data analytics and changes to player training, development and performance analysis at smaller more efficient clubs?

Arsenal and Tottenham Hotspur have already recruited members of last season’s scouting staff from Leicester City and clubs like West Bromwich Albion, Southampton, Norwich City and Reading have experimented with bio-banded competition at youth level to help them fish in a different part of the talent pond to the big clubs. Are these approaches working or will it be back to normal in 2016/2017? Whatever the answer to that question is, it is a very interesting Premier League season.

This entry was posted in Uncategorized and tagged , , , , , , . Bookmark the permalink.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s