Category: Correlations

Predicting the winter champion in the Eredivisie

Here is a challenge: predict the number of points teams will have in the first half of the season the moment the transfermarkt closes. Here is the catch: you are only allowed to use statistics of individual players. No team statistics like wins, goals scored, goals conceded or historical team records are allowed. The reason why no historical team data is allowed is that if you are able to predict sufficiently accurately how many points each team scores, you have established a clear predictive relationship between the statistics of individual players and the number of points the team score in the league.

That is also the reason why we only look at the prediction half way through the season. Otherwise your statistic is more likely to correlate with the richness of the club, rather than the quality of the players. For rich clubs who disappoint in the first half of the season, can buy themselves better players and improve their situation. 

Football Behavior Management (FBM) predicted on September 1st 2019 for the Dutch Eredivisie using only statistics of individual players. Even though the Eredivisie had quite a different season than usual, here are the correlations between our prediction and the actual points scored:

  • Correlation = 80%
  • R² = 64%

This establishes a strong and clear relationship between how well players do in the FBM system and how many points the clubs get that employ them. If you want more points, hire players who do well in the FBM system. That doesn’t mean that if a player does bad in the FBM system, that he is automatically a bad player. The FBM system is set up with a strong bias to underestimate players, rather than overestimate them. That means that a player who does badly according to us, could very well play better next season. But more importantly, it does mean that hiring that player increases the risk of hiring the wrong players. Whereas hiring a player who does well in the FBM system lowers this risk while at the same time increase the chance of winning more points!

Prediction & evaluation

Here is our original prediction and what actually happened:

1Ajax43441We predicted the performance of Ajax quite well.
2AZ294112We predicted AZ strength, but underestimated how strong AZ was.
3PSV3534-1PSV weakness is remarkable this season and we are very happy that we predicted PSV weakness so well. 
4Willem II25338We predicted Willem II strength, but underestimated how strong Willem II was
5Feyenoord28313Feyenoord weakness is remarkable this season and we are very happy that we predicted Feyenoord weakness so well
6Vitesse26304We predicted the performance of Vitesse quite well.
7Utrecht29290We predicted the performance of Vitesse quite well.
8Heerenveen23285We predicted the performance of Heerenveen quite well.
9Heracles162610Just like last year we underestimated Heracles.
10Groningen18257We underestimated Groningen.
11Sparta20233We predicted the performance of Sparta quite well.
12Twente2419-5We overestimated the performance of FC Twente.
13Fortuna17192We predicted the performance of Fortuna Sittard quite well.
14Emmen2018-2We predicted the performance of FC Emmen quite well.
15Zwolle1816-2We predicted the performance of PEC Zwolle quite well.
16VVV2215-7We overestimated the performance of VVV.
17ADO2013-7We overestimated the performance of ADO.
18RKC1511-4We predicted the performance of RKC quite well.

Strong correlation between FBM score and red cards

At FBM we divide players into four categories, namely:

  • Excellent players, who have a FBM score of 8 or higher.
  • Good players who have a FBM score between 5.5 and 8.
  • Average players who have a FBM score between 2 and 5.5.
  • Bad players who have a FBM score of below 2.

Our assumption is that bad players are more likely to get a red card than excellent players. For example, here is the FBM graph for Faes from KV Oostende of the match before he got a red card:

Continue reading “Strong correlation between FBM score and red cards”