How'd we do? Reviewing our Premier League 2023/2024 Predictions

What can we learn from the results from this past season? How “just” was the Premier League? And how did we do at Zone 14 in predicting the ordered chaos that was the 2023/2024 Premier League season? Let’s find out.

The first question to ask at the end of the season is how much can we trust the final league table? Did teams get what they deserved? Who should feel lucky, and who should feel short changed?

As we have done in the past here, a good way of answering these questions is by comparing the point attained by each team to their expected points (xPTS) over the course of the season. These xPTS are generated from the xG that each team created in each match and the xG that they conceded in those same matches. It represents an average number of points that a team can expect to have gained from a match given the chances that happened. If a team dominates on xG, their xPTS from the match can be expected to be close to 3, whereas a team on the end of a hiding would expect an xPTS far closer to 0. If we tally these up, we get a value for the number of points each team would have expected to gain this season in the absence of luck and variation. The chart below looks at this for each team (as represented by the small red dots) and compares them to the actual number of points each team actually attained (the larger, green dots).

A couple of things jump out instantly from this plot. Firstly, the top 3 seemed to massively overperform their xPTS, while the bottom three massively underperformed. While this could be down to the bottom three being worse than the competition and the top 3 being in another stratosphere, it likely comes down to how xPTS assigns points. In close matches, the top 3 managed to win their games, giving them 3 points but likely between 1 and 2 xPTS. Over the course of a season, that deviation adds up but also contributes to their title charge. The same goes for the bottom three, who consistently received 0 points even though their performances likely warranted some small apportionment of the points available.

Where this data gets interesting is looking at the biggest deviators between the values. The two largest overperformances? Manchester United and West Ham. While West Ham have decided to make a change in their manager this summer, Manchester United have stuck with Erik ten Hag, even though United’s performances this season earned them enough xPTS to only put them 15th in the Justice table.

On the other hand, Brentford and Nottingham Forest can be expected to make big leaps forwards next season if their trends continue. They each deserved approximately 14 more points than they actually attained, which would have put them solidly midtable and well out of the relegation battle in which they found themselves. Of course, part of this deviation can be attributed to Nottingham Forest losing 4 points due to FFP breaches, which would bring their deviation to 10 points instead. While not as dramatic as Brentford’s underperformance, this is still a substantial difference.

While it’s useful to look at this last season and compare the final league table to its statistically backed counterpart, it’s ultimately useless unless we take it into account moving forward. We can determine some trends and differences that might help us understand what happened in 2023/2024, but how can we take this and figure out what might happen in 2024/2025?

At the beginning of the season, Zone 14 made a Premier League prediction. Let’s take a look at that prediction (middle column, below) as compared to the final league table (left) and the justice table (right) to see what went wrong and how next year’s model might be improved.

Overall, it wasn’t an awful prediction. On the bright side, we got the champion right (just like everyone else), 3 of the top 4, 6 of the top 7, and 2 of the 3 relegated teams were correct. Our model, which as a reminder used the last 3 seasons worth of expected points weighted towards the most recent season, did fairly well at predicting the final table.

Digging even deeper and looking at the predicted point values shows that the model is on the right tracks, especially for the teams in midtable. Six of the league’s teams (Newcastle, Chelsea, Crystal Palace, Manchester United, West Ham, and Nottingham Forest) finished on a points total within 3 points of the Zone 14 predicted. A simple last minute winner or bit of luck would have made our predictions perfect there. In fact, the Zone 14 prediction using previous seasons looks to be a closer approximation of the final table, at a glance, than expected points was this year, with a similar number of close calls and a similar number of big deviations.

When looking ahead to next season, however, some adjustments will need to be made. Firstly, the spread of points in the table is much higher than the model predicted. We talked about this for the top 3 and bottom 3, but if those deviations are explainable then they should be included in the model. Potentially our model for next season will look to increase the system variance a bit more to predict better, and worse, sides.

There might be a need to adjust some weights in the model here, as well. Arsenal were predicted to finish with 68 points based on their last season, but if we had weighted their strong 22/23 campaign more then we may have been able to get closer to their actual total. The same, perhaps, would apply for Bournemouth, who broke their Premier League points record for the second consecutive season this year. The model may have corrections in place for pedigree, but not necessarily for the trajectory of clubs.

Lastly, we’ll need to adjust the model’s approach to newly promoted sides. With it being Luton’s first season in the Premier League, there was no data with which to approximate their performance directly. We may run into similar issues with Ipswich this year. The other two promoted sides, Southampton and Leicester, also look very different to the sides that were relegated in previous years. Is it fair to compare the newly promoted versions of these clubs to the versions who were previously relegated? We’re not sure.

A potential solution to this would involve taking a weighted average based on where they finished in the championship. Leicester would be compared to how the last three winners of the championship performed in the Premier League, Ipswich to the runners up, and Southampton to the playoff winners. This way, we may be able to get around the comparison by comparing them to teams that were more similar to their profiles than their own history.

Overall, this prediction was a success. We have demonstrated this year that it’s easy to make a pretty good estimate of what the Premier League will look like and what points totals teams might accrue based on a handful of data. Knowing that we’re heading in the right direction is only the first step in the process. Now we can work on it, improve it, and try and get even closer next year. We’ll see you in August for a sneak preview of just how the Premier League might look this coming year.