# Introducing Per Possession Numbers

Per 90 minute numbers don’t tell us everything about players.

While per 90s help compare players who’ve played different amounts of minutes, not all players get the same amount of opportunities on the ball. Players who play in bad teams, or even players who play in teams that keep less possession and attack directly, won’t get the ball as much as players in teams in the same league as Barcelona or Bayern Munich.

So, I tried to go in the same path that basketball analytics has been going for quite a lot of time and decided to divide player actions by the number of possessions the player has had, which is basically the number of times a player had the ball at his or her feet.

## The Process

Just to be clear, a player possession and a team possession are different. Additionally, the way I’ve defined a possession doesn’t have anything to do with the Opta definition, which you can find towards the bottom of this page. Here’s how I defined a team possession (also called a possession chain or a sequence):

A sequence of actions in which a team has the ball. It is ended when a team turns over the ball or takes a shot.

And here’s how I defined a player possession:

An action(s) when a player has the ball at his feet. It is ended when a player passes the ball, turns it over, or takes a shot.

With this definition, I divided player attacking contributions by the number of possessions the player ended. The first time I tried this was in November, where I got these results when I looked at the how many shots a player contributed to (shots taken + chances created).

Here’s the updated table for the top 20 forwards who’ve played at least 500 minutes in Europe’s big-5 leagues for shot contributions (shots + key passes) per possession.

The top looks okay, but a few names that should definitely, definitely be somewhere at the top are missing. Moreover, some players at the top don’t really look elite. This brings us to a problem:

What if adjusting player attacking stats by possessions returns false positives and false negatives? Does this method not work?

Let’s start with false negatives. Lionel Messi puts up 5.9 shots per 90 and creates 2.7 key passes per 90, which adds up to 8.6 shots created per 90. This number is astronomically high. However, his shot contribution numbers don’t stand out when adjusted for possessions – 0.127 shots created per possession, which is the 90th-highest in the big-5 leagues. I refuse to believe that Messi is only the 90th-best shot contributor in the big-5 leagues, and Messi isn’t the only elite player whose shots per possession numbers look average.

Let’s move on to false positives. The person who contributes to the highest number of shots per possession, with 0.267 shots and key passes per possession, is Marseille’s Kostas Mitroglou. I must admit that I haven’t really watched enough of the Greek striker, but I don’t think he’s the best at contributing to shots in Europe’s big-5 leagues.

This gives us a clear conclusion: per possession numbers (or at least the way I define it) can’t replace per 90 numbers.

What may be the reasons for this? Well, firstly, good players tend to get the ball more in football. Furthermore, good players don’t necessarily always take a shot when they get the ball, and choose to pass it around. There are many excellent players, like Harry Kane and Cristiano Ronaldo, who put up high shots per possession numbers, but these numbers indicate roles and tendencies rather than ability.

This sent me in a new direction: why not use per possession numbers to find out more about player tendencies?

## Application:  How do players play?

Knowing what players do when they get the ball definitely has value in recruitment, tactics, and coaching. And instead of looking at per possession numbers as decimal fractions, why not look at them as percentages? So if a player passes 0.8 times per possession, we can say that s/he passes the ball in 80% of his/her possessions.

For example, if a scout is in pursuit of a full-back who can take part in the build-up and play a lot of passes, like Joshua Kimmich or Sergi Roberto, the scout can take a look at players who pass a lot per possession. Number one on the list for full-backs with the highest percentage of passes per possession is Bayern Munich’s 32-year-old right-back Rafinha, with 98%, who, unfortunately, may not be a practical target because of his age and the team he plays for. If we keep going down, we find players like Paul Dummett (?!) at 2nd, Joe Gomez at 3rd, Mario Sampirisi at 10th, and Mario Rui at 15th. These numbers don’t tell you to sign them immediately, but it definitely makes you take a closer look and thus provide a starting point.

Or, if a club wants someone who can just take the ball in midfield and dribble, they can look for center-midfielders who attempt a dribble most of the time when they have the ball. There are some interesting players here with Amine Harit (1st; attempts a dribble in 11.4% of his possessions), Samu Castillejo (2nd, 9.8%), and Tanguy N’Dombele (9th; 7%). The club’s scouts can then move onto more advanced metrics, meticulous video analysis, and live scouting.

If a coach wants to see why his star signing has flopped, he could maybe start with looking at what a player did with his possessions in the season(s) before he moved, and then he could compare it with his current season’s numbers. If there’s a change, the reason for the player’s performances could be a result of being used in a different role, and being given different duties.

Using possessions as a denominator for player attacking stats doesn’t tell us a lot about player ability, but it does give us insight on player styles. And that’s definitely important.

One thing I’d like to do in the future is looking at opposition team possessions and defensive actions, but I can’t do it now because of the lack of data in hand. This could help tell us a lot about defender styles, and it could give us alternative ways of adjusting defensive actions.

This concept has also been written about by Tom Harrison. You can read his article here.

Feel free to message me on Twitter about any questions or feedback regarding this, or completely different ways of doing the things I did.

All data from WhoScored.com. Data correct upto 15th February 2018.