Twelve’s rating system

This post describes our work at Twelve to develop visualisations of football matches. Our idea is to show, live during matches, how the players are performing. We want to help fans both to see the contributions of individual players and give a tactical view of the game using the same data relied on by managers and coaches.

In this article I’m going to showcase our current visualisations and describe the idea behind the methodology for assigning points to actions.

We have developed a method for evaluating player performance based on everything they do on the ball during the match. Below is our rankings for Manchester City players in their game against Tottenham Hotspur.

These rankings are broken in to four categories attack (green), defence (red), off-the-ball (yellow) and shots & goals (blue).

According to our model Raheem Sterling was man of the match. This is because he scored two goals, which are worth 1000 points each.

Sterling also missed a number of chances before scoring. Some fans might think that these misses should give him negative points. But in Twelve’s system we give Sterling positive points for these misses. This is because Sterling got in to a good position to take the shot, and in the long term there is statistical evidence that creating chances is an important sign of a quality player.

The points for these shots are awarded on the basis of a model called expected goals. We measure the position of the shot and look at historical data about about the probability that shots from these position are typically goals and we assign this as the number of points for a shot. Sterlings best chance would normally be a goal in 42.7% of cases, so we give him 427 points.

So thats goals. Lets move to attack. Kevin De Bruyne is the best City player. Here are his attacking contributions.

His top attacking contributions were given 160 points, because they created a clear cut chance for a teammate. If you click on the passes, you’ll see that two of them got slightly lower scores. These passes were also in to dangerous areas, but didn’t directly result in a chance. In this case, the points are assigned in a way that is similar to expected goals. If a pass takes the ball from a part of the pitch that a team is less likely to score a goal from (e.g. out near the touchline) to a place they are likely to score a goal from (e.g. in the box) then it will get a lot of points.

If you click on ‘All’ in the attack dashboard, you’ll see every pass De Bruyne made that improved the probability of City scoring. Each of these passes got a lower number of points, because although it improved his team’s attacking position the contribution was relatively small. Clicking on these should give you an idea of how much different passes increase danger for the opposition.

We’ll now turn to defence. Here we’ll look at Tottenham Hotspur. Their defence had a lot to do and Kieran Trippier, in particular, didn’t have a good game.

In the dashboard above you can see the mistakes he made (crosses) and the successful defensive actions (stop signs). Trippier made several big mistakes, being dribbled past in and around the box. Add these up and he got a total of -48 points for the match. Again, you can see by clicking on the points that mistakes in front of goal are assigned larger numbers of points than those further away.

One thing we are still working on is a model of off-the-ball actions. The data we use only measures what happens on the ball, but we keep track of where players are and identify the areas of the pitch they typically play in. We then assign points to players when the opposition lose the ball in these areas. We confine off-the-ball points to areas away from goal, so it gives a rough measure of press. It is typically defensive midfielders who are given most credit for pressing. Here is Fernandinho.

Again, if the opposition lose the ball in a more offensive position, then the nearby players are awarded more points.

As the Twelve project develops we’ll share more details. Ultimately, the techniques developed here will become the basis for match analysis websites, a fantasy football game and tools for football analytics. This post should give you a feeling for where we are now and where we are going.