As football gradually makes its way into a brave new world of data and metrics, it is telling that one of the statistics that was most readily accepted by the community (football, rather than analytics) remains one of the least well-defined and understood.
Possession statistics are commonplace in football broadcasting and reporting. They are regularly used to demonstrate one team's superiority over another, to confirm dominance or otherwise when the scoreboard remains deadlocked.
However, there is often very little thought given to how this statistic is calculated. Is it time spent in possession? If so, who is in possession when the ball is in the air? What if there is a stoppage in play? Does this get added to the possession log? If it depends on which team takes control of the ball when it lands, then when do you stop the clock? This sounds trivial, but it could make a significant difference to possession statistics for teams that play in a more direct style. It also requires extra manpower and a reliance on inconsistent manual methodology, both of which mean that the 'chess clock' approach would be impossible to apply effectively across the thousands of matches Opta analyse every season.
However, if it is calculated solely using passes, this doesn't account for time spent on the ball. It also potentially overstates the influence of teams who hold on to the ball and play short passes between defenders or defensive midfielders.
As detailed here Opta uses this second method to calculate possession, but we accept that neither system is perfect.
The recent Champions League game between Celtic and Barcelona at Celtic Park highlighted the vagaries in possession recording. Depending on the source, Celtic's 'possession' was recorded as 11%, 16% or 28%. It is clear that what is a useful way for broadcasters, other media sources and fans to easily demonstrate the apparent state of a game, isn't necessarily accurately representing the actual fact.
Therefore we have looked at a different way of defining and calculating possession by creating a new framework, with the aim of improving our understanding and analysis of this notoriously complicated and misinterpreted subject.
"Half of the people can be part right all of the time,
Some of the people can be all right part of the time.
But all the people can't be all right all of the time"
As Messrs Zimmerman & Lincoln conclude, this may not be to everybody's taste but we will attempt to set out the rationale behind the possession(s) framework we have employed:
- (a) possession starts upon a player controlling the ball
This should be pretty straightforward. Specifically, winning a tackle or header does not, of itself, indicate the start of a possession. Only the first controlled action thereafter does.
- (a) possession ends when a player no longer has - or tries to have - control of the ball.
This is where it becomes a bit more open to debate. While it is straightforward to define a possession ending when a player's pass is intercepted or they dribble out of play there are some grey areas.
For the purposes of this exercise we are considering launches* (including long goal kicks) and crosses as 'zonal' passes, into areas rather than to a specific player. By doing so, the ballplayer is ceding full control of the ball for the chance of improving their team's territorial advantage &/or of scoring. Similarly, any shot is deliberately ending possession with a view to trying to score - whether the team recovers any potential rebound that ensues is another matter. Finally, any play-ending event (free kick, throw-in, corner) is also defined as ending a 'possession' as the defence can reorganise in the break in play.
At this point, some may be questioning the merits of this definition. A team could take a goal kick long (1), win the header, be fouled (2), have a shot hit the post from the ensuing free kick (3) and score from the rebound (4). At no point has the opponent touched the ball - never mind possessed it - but this would be designated as four 'possessions' by the above rules, which may not sit with a classical definition of possession.
Now, however we can introduce the idea of a 'sequence' - a series of possessions uninterrupted by an opposition possession. This allows us to define the above example as one sequence (at this point it's nomenclature, you can call a sequence a possession and what we've defined a possession something else if it'd make you happier), starting with a goal kick and resulting in a goal.
It also allows us more information, now though. A long-ball and cross-heavy team may take a number of possessions to move the ball up the pitch, while a short-passing team will likely achieve the same end in fewer possessions (but more time).
* For a full explanation of launches, please have a look at our definitions page.
With a defined methodology in place, we can now look at the time of possession share for both teams. By this method, Celtic possessed the ball for around 18% of the time. That is, 18% of the time when either team had possession of the ball. Opta's in-game measurement using the simplified pass completion method was 16.4%, whereas measurements from other data suppliers that used the 'chess clock' approach were around the 30% mark.
Example: Premier League 2012/13
This is obviously only one game and, of itself, doesn't tell us too much, so we proceed to run the numbers on a group of games. As teams have had only four matches in the Champions League to date, we'll try the Premier League on for (sample) size.
For those that don't like staring at a list of numbers, we can turn this into a pretty(ish) graphic, with the size of the point representing the teams' time of possession percentage:
As expected, the pass-heavy teams spend the longest on the ball, both in terms of time per possession and overall time of possession (which are not quite interchangeable). It is not especially surprising either to see the likes of Stoke and West Ham giving up the ball the most readily.
From the other perspective, Liverpool lead the way in limiting opponents' time per possession, suggesting Brendan Rogers' high-pass volume/pressing game is getting results, even if it's not actually getting results just yet. By contrast, West Brom, Fulham & Norwich are allowing their opponents a Chelsea-like amount of time on the ball, possibly indicating a willingness to absorb pressure and counter-attack, or just that they aren't very good at imposing themselves on the opposition.
It is worth noting as well, that the spread of the teams' own time per possession numbers (as opposed to their oppositions) is a lot larger, suggesting - again, not shockingly - that teams have considerably more control over their own playing style than their opponents'.
We can also look at how teams' possession numbers changed based upon the score at the time. To further boost the sample, we've gone back to the 2011/12 season and looked at each team's figures when leading, level and behind. Below is a selection of some of the more interesting teams:
City, and to a lesser extent, United fit the expected - by this author at least - pattern. That is, taking progressively longer on the ball when the scoreline is in their favour but at the same time allowing the opponent more time, with the urgency to make something happen obviously greater when trailing than when ahead.
Swansea and Wolves display the reverse trend however, while Fulham's time per possession remained virtually unchanged in all three states. This suggests that it is simply a function of the opposition, with the sides spending more time trailing against the better, more possession-heavy sides. The counter argument to that is Norwich and Stoke, with both teams' figures remaining consistent regardless of the score.
A full overview of the 2011/12 season can be seen below. Obviously the current season is still not even half way through so any comparison can't be exhaustive. However, the numbers appear to indicate that Arsenal have been pressing the opposition less (possibly due to the departures of several players including Alex Song) and Manchester Utd appear to have been slightly more pedestrian while on the ball. Again though, be wary of drawing definitive conclusions from such as small sample.
Clearly, this represents only the first stage in redefining the measurement of possession in a football match. The graphed results above seem to tally pretty closely with what we'd expect from watching the teams play, but there may be better data-led solutions that we haven't even considered yet.
We'll be releasing more blogs on this subject over the coming weeks and months. However, please feel free to respond to the initial work with any comments or criticism. At OptaPro, we believe the best way to advance any work in this field is by taking a community approach, and opening our work up for analysis and critique. Feel free to use the comments section underneath, our LinkedIn group or indeed Twitter (@OptaPro) to send any feedback.