Back to the Future (of defensive football statistics)
Imagine, for a moment, that the 2017 version of Doc Brown arrives, emerging from his gullwing doors, in front of your house. “Wait a minute. Wait a minute, Doc,” you say “Are… Are you telling me that you built a time machine… out of a DeLorean?”.
He tells you that yes, yes he has and that you can use it. You can travel anywhere in time, see anything you want. Do you want to see the future? Do you want to go back and see England win the World Cup? No, you decide, you want to do something far more important. You want to go back in time and re-write the way that football stats are collected.
It is no secret that using statistics in meaningful ways for defenders is, to say the least, difficult. Players who make more tackles, for example, do not find themselves strongly correlated with victory in football games (see here) and while it’s possible to determine style from them (see here) it’s not an easy task.
Let’s start all over again. Let’s wipe the whiteboard clean, grab a fresh sheet of paper, press the reset button, and do things for ourselves.
First, we should consider what the currently available useful stats do. Shot-based data, either counting shots or using them to make something like an Expected Goals model, is basically a proxy for lots of different skills. It’s a proxy to quantify some amalgamation of positioning, physical capabilities, good game sense or attacking instincts, and maybe some other things too.
Work focusing on more specific skills is also being worked on, with plenty of passing ability and finishing ability models entering the public sphere in the past couple of years, although they’re still partly amalgamations of several different things.
‘Finishing ability’ models tend to lump together all shots that a player makes, whereas the colloquial term ‘finishing’ is more in line with the type of side-footed finish which comes to mind whenever one thinks of Thierry Henry. Passing ability is measuring a combination of technical ability and vision.
Let’s go back to defensive things, and tackles. They’re similar to ‘shots’ in that they are an event which happens, and which we apply thinking or context to in order to gain useful information. With shots, it was discovered that good strikers generally take more shots (because people pass them the ball and because they get themselves into good opportunities, probably). With tackles, one can use them to try and understand a player’s tactical role (although this is made easier when combined with other defensive ‘event’ statistics).
You could, theoretically, go down the same route for tackles as one does for shots and try and create a ‘tackling ability’ model, along with similar lines as ‘finishing models’. But:
- Like with shots, there are different types of tackles
- Unlike with shots, we don’t tend to have data which differentiates between these different types of tackles
- Would you really want a ‘tackling ability’ model?
There is huge footballing value to something which models finishing ability. Finishing is how goals are scored, and goals are extremely rare and extremely valuable in football. Tackles happen quite a lot and are significantly less valuable, and are not the principal focus of defenders.
However, how a defender wins the ball – whether through tackling or anticipating to intercept, whether won principally through physical prowess or through intelligent movement – might be useful to know, and perhaps could be combined into a ‘finishing ability’ type of thing. ‘Obtaining the ball’ ability or something.
You would still have the problem of different ways of defending though, even when the action is to win the ball (as opposed to shepherding an attacker away from the goal or marking a slippery forward – these would be much more difficult to quantify).
There are different types of interception, for example, (such as passive, anticipatory, and reactive); there are different types of tackle, such as a slide tackle after having chased alongside an attacker or poking the ball away when squaring on to them before they try and accelerate away.
Clearances create a problem because not only are there different types, but a clearance is largely dependent on what the player does AFTER the ball is with them (namely lumping the ball away) rather than what the player does BEFORE. The rise in playing out from the back could create a drop in the number of clearances, as defenders put their foot on the ball and pass it out instead of hoofing it to the halfway line, but this would not change the essential part of what the defender had done in order to be in the position to receive the ball in the first place.
Going back to the DeLorean, perhaps you decide to leave an email in [insert data collection company name]’s office (did they have email back then?) telling them to separate defensive actions into the different ‘types’ as well as the different events like tackles etc that have always been used.
That’s still an issue, and it’s a good thing you have a time machine because it’s going to take you a long time to come up with a solid theoretical basis to understand defensive actions. Does one exist at the moment?
A defender is at a striker’s back as they receive a pass. As the ball is almost at the striker’s feet, the defender’s leg hooks around and pokes it away. What is this tackle called? Is it a tackle, or an interception?
A forward receives the ball between the lines and turns. A defender steps up and is square onto them, not quite within touching distance. Yet. The forward picks a lane to dribble down and tries to fool the defender, but they pick the right direction and stick their foot in the way with their full body weight behind it. What is this tackle called?
This is the kind of process I would go through if I were re-writing the way that defensive football statistics were collected and analyzed. Like with so much about soccer stats, it starts with the football itself.
Eastbridge Soccer Betting Broker, bespoke sports betting for the discerning client.