Per Possession - A new way of normalising data for Scottish football

Written by: Jamie Kilday @FitbaInScotland

When looking at football statistics, analysts are always looking for ways of normalising data to make comparisons as fair as possible. You may be familiar with ‘per 90’ as a metric for comparing players, by which you measure a player not by their number of appearances but by how many minutes they have played. A good reason for doing this is that a player who routinely comes off the bench could have the same amount of appearances as a starter but significantly fewer minutes so that must be taken into consideration. 

What if we could apply this sort of data manipulation to team stats. This is where Per Possession comes in, by breaking down a teams actions into minutes of possession you can get a better idea of how the team plays. 

So why use minutes and not the percentage of possession? This is largely due to how unreliable percentage of possession can be. The fact that in most matches the ball is in play anywhere between 40-60 mins means there can be great discrepancies in how many minutes a percentage can represent, for example, Rangers had 62% possession against Ross County in December which came to 35:49 minutes, yet their 63% possession against St Johnstone in October was only 29:33 minutes.  

To see how much possession can effect the statistics we’ll break teams playing styles to their absolute simplest form, passing vs dribbling. This article will be using InStat data that describes a dribble as;

“ active action performed by a player in order to leave an opponent behind. Dribbles can be when a player passes the opponent by kicking the ball past them and can also be when a player lets the ball go through without touching it, but thus leaves the opponent behind. Dribbles can be performed not only forward, but also across the field, and even backwards.”

Now, this is not a definitive way of representing a teams playing style, more a way of showing how much possession can skew the data. For this, I will only look at attempted actions rather than successful actions as this gives a better idea of how the team attempts to play.

Without taking possession into consideration you get this;


Broadly speaking, teams that pass a lot also dribble a lot. Celtic seem to dominate in both and Hamilton, Motherwell and St Johnstone do very little. All the teams are clumped fairly close together with the exceptions being Hibs high dribbling, Rangers high passing and Celtic’s high everything. How possession has skewed this data is exemplified in the next graph;


With the number of passes a team attempts per game versus their number of minutes on the ball having an r2 of 0.9788, you can see why we might want to take this strong correlation into consideration.

With this in mind here is how teams passing and dribbling stats compare when you apply Per Possession;


Here we have a better idea of team styles as the teams have fallen into more defined categories. 

The Passers

Rangers, Ross County, Thistle & Celtic

It may come as a surprise that the bottom two sides in the league are in a similar position to two of this season’s best teams. All four teams are above the league average for short passes, preferring a less direct game than their counterparts.

Ross County and Thistle both average around 24 mins of possession per game which is five minutes less than Rangers and a whopping thirteen minutes less than Celtic but move the ball around at a similar rate, all be it, with less success. It will shock nobody to know that there is also a correlation between a teams ability to pass and their minutes of possession.

A notable difference between this and the non Per Possession stats are Celtics dribbling statistics, you would perhaps have expected their dribbling to have maintained a high position considering their wingers, Forrest, Sinclair and Roberts tend to retain the ball and Kristopher Ajer’s tendency to carry the ball out from defence but they are now below the league average. 

The Dribblers

Kilmarnock & Hibs

John McGinn, Martin Boyle and Jordan Jones are highest in the league for dribbles attempted with the latter leading the way with 11.98 dribbles per 90 so its unsurprising that the two sides are close to each other. Kilmarnock have the 3rd lowest possession in the league at just 22.57 minutes yet average 30 attempted dribbles per game. Compare this to Hibernian's 35 per game but with 4 minutes more possession and Celtic's 36 from 14 minutes more and you can see that dribbling is a big part of Kilmarnock’s attacking game.

The Motherwell


With only one player in the top 30 for passes attempted in the league and one player in the top 40 for dribbles attempted it is no surprise that Motherwell are out on their own. Motherwell’s direct approach doesn’t require them to be prolific in either field, averaging only 2.7 passes per goal scoring attack (3rd in the league).

One thing that I found interesting here is that before adjusting for possession, Hamilton and St Johnstone were very much in the Motherwell camp of being way below average at both metrics when this isn't the full story. 


This article has by no means set a definitive on how teams play but I hope it has highlighted how much possession can skew the statistics.  

Per Possession is by no means a new thing, basketball statisticians have been using Points Per Possession for the best part of a decade but there is no reason why it can't be applied to football. Over the summer, and into the coming season, Modern Fitba will be looking more into per possession stats and how they can effect not only team statistics but player statistics as well. 

Things that cropped up while putting this together...

  • Celtic actually have a tendency to attempt more Dribbles Per Possession in games where they have less possession.

  • Most teams passing and dribbling statistics are not directly affected by playing home/away.

  • Rangers attempt the second fewest Dribbles Per Possession yet have the best dribbling success rate in the league with an impressive 60% of their dribbles being successful.

All data correct as of 15/5/2018 and is supplied courtesy of InStat.