Predicting the final Eredivisie league table

Let me first protect myself and fire a short warning up front. This post will wade into the deep waters of prediction territory, an area I normally try to stay away from as it is generally easy to slip up, and in the current digital era all predictions made here will be available for comparison with the final outcome come the end of the season. And no matter how accurate your predictions may have been on the basis of the little information available beforehand, the end result will be different.

Not that differing from the final outcome would mean the prediction was bad though. It may just have been a season full of outliers, while the predictive model might still have been the wisest things to say given the information available at that moment in time.

That being sad, in this post I will try to build up a prediction model that will be revisited a certain times during the present season. The model will use the best known variable, or rather the best known variable of all currently available, in order to give a prediction of the final Eredivisie league table come the end of the season: the total shot rate.

 

Shot rates

Shot rates are obviously strongly correlated to goal difference, which in turn is strongly correlated to points, which, needless to say, is strongly correlated to league table position. There is total shot rate (TSR) and shots on target rate (SOTR). I prefer TSR because it performs equally well to SOTR, but gives higher numbers, thus a better base to make predictions on.

The method is rather straightforward and I will explain it in full here, show the predicted final Eredivisie league table and then provide some issues and bias that may cause the prediction to be slightly off here and there.

I’ve used the past two seasons to correlate TSR over a full season with the number of points obtained in that season and they correlate very well. All teams above the black trend line had a more or less lucky season, with PDO’s over 1000 and all teams below that line had PDO’s below 1000, reflective of bad luck that season.

 

Now, let’s rank the team’s TSR’s over the first four games of the present season and see how many points that TSR would have brought the teams over a full season in the past two years of data. Obviously, there’s not a full season left to play, so I’ve taken the number of points teams have already obtained from the first 4 matches and used a 30/34 fraction of the predicted number of points according to the TSR. Here we go…

So, based on this simple model, PSV are expected to obtain 82 points and run away laughing with the league championship far before the season is over. At some distance, Twente is expected to come in second with 70 points and Ajax take third place with 65. Obviously, this prediction of PSV’s dominance is somewhat inflated due to the fact that they played 10-men AZ for over 80 minutes on match day 4.

At the lower end of the table, difficult times are predicted for NAC, who are expected to get relegated with only 18 points, should their trend in TSR continue this way.

AZ holds a surprisingly low 16th spot, but they obviously suffer from the fact that they played PSV away with 10 men for nearly the full match. Though even when disregarding that match, their TSR over the other three matches (0.464) would predict just 37 points and a low mid-table league position.

A positive surprise is Marco van Basten’s Heerenveen side. Despite currently holding a disappointing 13th position, the model has them obtaining 50 points and finishing 6th come the end of the season. Newly promoted PEC Zwolle, predicted 13th with 39 points, and Willem II, 14th with 37 points are presently expected to stay up, but should AZ improve their biased result so far, they will hold a close fight with VVV and Groningen.

 

In the end

“All model are wrong, but some are useful.”

This predictive model should be seen as just that. Of course it’s wrong, but give it a few more match rounds and teams will probably settle more on TSR’s that better reflect their true qualities and the predictive quality of the model will improve. Of course, TSR is not the only factor involved in winning matches, but it remains the only factor that teams have shown capable of repeating.

Team have their best season’s when suddenly their strikers start finishing a lot of chances. Think Heerenveen 2011/12 for this one, when a strike force led by Bas Dost suddenly converted 16.7% of all scoring chances against a league wide average of 11.9%. Or when their defense suddenly allows only 9.0% of shots to be converted into goals, like 2010/11 Ajax did. It’s just that these teams are never capable of repeating high shooting or saving percentages, as represented in the PDO.

This year too, certain teams will have outlier seasons in terms of shooting or saving percentages. League leaders Twente and second placed Vitesse are nice examples so far, but, believe it or not, Mihaylov and Velthuizen won’t maintain their 94.9% and 96.6% saves rate. And Ajax and RKC won’t keep converting over 20% of their chances.

All these results will shake out a bit, but a single season is just too short to fully correct for these outliers and surprises will occur, this season too. Let’s see how our model holds up in a few weeks time…

 

Total Shot Rates were calculated with data provided by Infostrada Sports.

Leave a Reply