Predicting the final Eredivisie standings

Despite the early season fuzz about the progressive equalization of the league, the Eredivisie has a clear top-4. It has had that in terms of underlying stats for some time, but presently it also does in the league table. This post will use these underlying stats to identify front runners for the title, the Europa League play-offs, and of course relegation football.

 

Stats, what stats?

The model used to create these predictions has been explained before, but I’ll explain the most important elements once more.

It’s a shots based model, where each shot is rated by its chance of resulting in a goal. Hence the term ‘Expected Goals’, or ExpG model. To convert a goal scoring attempt to an ExpG we use shot location as the most important factor, but also shot type and assist type have their influence. Game state also plays a role, as it’s easier to score when leading due to the other team taking more risks to come back into the game.

So, with those element we construct an idea of each team’s relative strength, and from there on we simulate each match to get the appropriate spread of chances for a home win, a draw and an away win.

The final step is to simulate the remainder of the season to achieve an estimate for the number of points, and the league placements come the end of the season.

Stats are all collected by OPTA and presented at Squawka. I can’t thank these guys enough, as without their input… no model.

 

Predictions

In the following graph, each team’s predicted points are plotted. The boxes are color coded for the relevant league places, with the code explained at the bottom. The box represents half of the simulated results, with the think black line the mean of the simulated results.

Vitesse is expected to win on average some 67 points, but anywhere from 65 to 70 should cover 50% of simulations. On both sides of the box a wide spread can be seen, representing 95%  of the simulations, with some extreme outliers on both sides.

Please take careful note of the wide spread, and if you’re willing to learn one thing from this post, please make it this! Even with the exact same underlying performance, potential outcomes have a very wide spread. Anywhere between 58 and 77 is reasonably possible with the exact same underlying performance that lead Vitesse to their current points tally of 40 points after 19 matches.

Boxplot projected league table Eredivisie 2013-14 20 januari 2014In terms of predicted points, Vitesse has the highest mean, though slightly against intuition this does not automatically mean they have the best chance of winning the league. Their spread is quite small compared to Feyenoord and Twente, who happen to have more open games, which leads to a higher ceiling, but a lower bottom.

 

Ajax

Ajax’ fourth place may come as a surprise, but, and I cannot stress this enough, the wide margins predict them to finish anywhere from first to fourth. Ajax currently shares the lead of the table with Vitesse, on 40 points, being trailed by Twente at three and Feyenoord at four points. Therefore, this prediction indicates the Ajax is expected to have the weakest performance over the remaining matches of all four title candidates. How come?

Well, there are two things that are weak in Ajax’ game in terms of ExpG. First of all, they concede 1.11 ExpG per 90 minutes while playing at level score (0-0, 1-1, etc.). For now, they’ve come away with that, but Ajax simply have a bigger chance of going a goal down than their rivals have. Cillessen’s save percentage of 95% sounds fantastic, but there is no defense in the world that is going to keep that going.

The second ‘problem’ at Ajax is defending small leads. While most top teams improve their ExpG when leading by a single goal, this is certainly not the case at Ajax. This problem also existed during the past season, when I wrote about it for ‘De Zestien’. As a consequence, Ajax will allow more teams a comeback goal than their rivals.

 

Predicted final standings

Besides predicting the final points total, the exact same simulation also predicts league ranks. The next graph shows the spread in predicted positions. Despite the fact that Vitesse produced a higher average predicted points number, they are on equal terms with Feyenoord and Twente due to Vitesse’s narrow spread compared to the other two.

One caveat, though. Ajax is predicted with just 14% chance of winning the title, but during both the past and the present season, they have managed to overachieve in comparison with the model. And they are the only team to do so. This means that either Ajax do something very good that is not part of my model, but in this case they would be the only team to do so, or Ajax has been lucky for quite some time. Only time will tell…

Boxplot projected League positions Eredivisie 2013-14 20 januari 2014

Europa League

The Dutch Cup final will not be contested between two top-4 teams, so places three and four will provide EL qualification to two teams dropping out of the title race. Place 5 to 8 will lead to EL play-off football. PSV, Groningen and Heerenveen are currently up there and are widely expected to stay, while AZ may well drop out as they are predicted about level with PEC Zwolle, with an outside chance for Heracles.

 

Not much to play for

Each league always has a bunch of teams with not much to play for in the second half of the season, but the Eredivisie is still quite open at this stage. Utrecht can still make the play-offs – just 11% despite currently holding the ninth place – but stand an equal chance of battling relegation. Yes, luck and random effects are very very important in football.

Go Ahead Eagles, NAC and Roda JC still have to worry a bit about relegation, but will in all likelihood stay up.

 

Relegation

The R-monster is quickly sneaking up on RKC. The model identifies them as the main relegation candidate, despite their current 15th spot, two points off the bottom. The bottom five teams are just two points apart in the current league table, and it’s an equal battle between ADO, N.E.C. and Cambuur to try and escape the relegation play-offs, while Roda is part of that five-team-group at the bottom in terms of league points, but not in terms of underlying performance.

 

Final words

“Prediction is very difficult, especially when it’s about the future”

These famous Niels Bohr words should probably be in a piece like this. And the 1922 Physics Nobel Prize winner makes a point that is often quickly interpreted as an open door. Yet, it’s fundamental to understand that predicting the future is estimating chances of events occurring. I don’t know if Vitesse will make exactly 67 points, just that 67 points would be my best estimate. And that 70, 65, or even 60 is quite possible too. Don’t believe exact predictions, they will disappoint you.

5 thoughts on “Predicting the final Eredivisie standings

  1. Kevin Grove

    What might be the case with Ajax compared to the other teams is the density of their position occupency. As probably the only team in the league they have decent players as backups for their starting 11.

    When a decent player that does well in the current setup of his team gets knocked out for a few weeks this could create a large dent in the teams performances. It is to be expected that more injuries will take place in the second part of the season, thus having a good backing for injured players becomes more important. Ajax was this covered in a better way than others.

    Moreover, Ajax is the team in the Dutch league that shows a positive line in stats that show high correlation with points per game. In the last few seasons their second season halves have surpassed every other team. This has been attributed to the team being built during the summer and their peak starting around the winter break.

    Maybe it is interesting to separate pre and post winter break statistics from the past years, to see whether Ajax’s game has substantially improved during the second season halves, or that they indeed have been overachieving the past three seasons.

    Reply
    1. 11tegen11 Post author

      Thanks for this reply, Kevin!

      Your point on the backup players sounds like a very valid one. My feeling is that Twente (Tadic), Vitesse (Piazon) en Feyenoord (Pellè) are seriously hurt as a team when their start individual should fall away through injury.
      This is less the case at Ajax where qualities are more spread throughout the squad. I have some ideas in mind on how to graphically illustrate this…

      And the ‘Ajax / second half of the season’ topic is a known thing too. I can’t get my head around the extreme discrepancy that they’ve shown year one year. Definitely on the agenda to look at that, not just results wise, but also in terms of the underlying process…

      Reply
  2. DFFF

    Excellent post, although the conclusions will hardly appeal to all fan bases…

    On your usage of Exp.G. though: could it be that the points won by clubs like Vitesse, Twente and Feyenoord may be hampered by their inefficient strikers (Havenaar, Castaignos and Pelle respectively), as you argued in your previous post? Somehow Ajax’ players (bar Fischer) seem to perform less terribly in that aspect.

    As for Ajax not creating better chances in +1 game states: my guess is that their complete aversion against counter attacking negates any advantage created by the extra space teams give up when trailing a goal. Do they compensate by keeping it extra tight at the back at GS +1, compared to others, or is this just foolish tactically?

    Final point: I think that Ajax second-half-of-the-season improvement has a lot to do with little to no European fixtures, and De Boer finally finding his preferred starting line-up after the inevitable Summer transfers.

    Reply
    1. 11tegen11 Post author

      Thanks for this excellent reply.
      I think you make some very valid points here…

      First off, linking the two posts together is something I’ve been thinking about too. I might indeed be that inefficient striking of Feyenoord, Twente and Vitesse leads to overestimation in these predictions. But then again, I need to assess whether how much this inefficient striking regresses to the mean, before I could implement it in the predictive model.

      Your second point is a nice thought, but unfortunately for Ajax, it is not the case that they are tight as the back at GS +1 at all. It seems their style is more important, or less flexible, or whatever you would call it, so they don’t take advantage of the openings that playing at GS +1 offers. It may be a case of preferring possession over quick counters as you suggest.

      Finally, both the lack of European fixtures and time to find a regular line-up do indeed correlate with Ajax’ improved performance. I’m afraid we can’t assess the causation in a statistical model, as these events will play out every year. Probably, in a few more seasons, we should know…

      Reply
  3. Xarius Desai

    Been massively interested in what you and James Grayson have been doing and am trying to emulate it with the team I’m currently working with to see what comes of it.

    A thought popped into my head though, given that the shots on target ratio isn’t better than TSR, how does a ExpGR model pan out? Do you have the data to test it?

    Reply

Leave a Reply