A splurge of footballing and footballing statistic theory

“Stop. Calling. Over. Or. Under. Performance. Of. Expected. Goals. ‘Finishing. Skill’.”

Is what I would have liked to have tweeted had I not decided that that would have been annoying to almost everyone who follows me.

The point, though, is a good and meaningful one, even if I do say so myself.

The narrative around stats and finishing skill appears to have gone something like this:

  1. Finishing skill does not exist
  2. The people who said ‘finishing skill does not exist’ are dummies. Finishing skill does exist, obvs, we just can’t detect it
  3. Not only does finishing skill exist, but we can detect it, but it doesn’t do very much

And 4., more recently, is a good article from Bobby Gardiner which says

  • Not only does finishing skill exist, but we can detect it, and it does more than we think, but it’s still less important than actually getting in good positions in the first place.

This is all good, fine, and fascinating to people interested in stats.

It’s also consistently incorrect in the most important way, which is that what is commonly known as ‘finishing’ to, like, normal people, isn’t what stats people mean when they talk about finishing.

Which is a problem.

What stats people mean when they talk about finishing skill is that if a player takes 2000 shots from anywhere on the pitch over their career and the Expected Goals value of those shots is 228 and the player scores 248 then they might be a good finisher.

What normal people mean when they talk about finishing skill is usually a player being one-on-one with a keeper and either bottling the chance or slotting it past them.

This is something that’s bugged me for a long while, but it has a wider point to it.


There’s a big problem in football with regards to statistics.

Nothing happens.

Football is a game where hardly any of the pitch matters and, what’s more, the parts that matter shift throughout the game. At one point, an area of the pitch thirty yards from goal is crucial to a team scoring a goal – for the rest of the game, it could mean nothing at all.

Not even the eighteen-yard box matters all of the time – when a defence retreats to a few yards out from goal, it’s that area that they’re protecting, not the eighteen-yard line. There might, to be generous, be an area, a bubble, six to twelve yards away from goal that teams will never cede to their opponents.

To make matters worse for the statisticians and the theorists, the things that matter don’t happen very often. The ball rarely goes in this bubble. Shots rarely take place, relatively.

So, football isn’t an event game. It’s not basketball, it’s not cricket.

Football isn’t a territory game, unless we’re counting two bubbles on the pitch which are rarely entered and make up a small percentage of the pitch. It’s not American football or rugby.

If anything, it’s a tug of war, where two sides compete against each other and, for the VAST majority of the time cancel each other out completely. How do you tell which side is better at a tug of war?

Well, one side might look like they’re edging the contest for most of the time. It should indicate that they’re stronger, and the other side are digging in for their life. But if you don’t get things over the line, does it matter? Maybe?

Anyway, I digress, but it was a worthy digression.


In theory, stats people could easily try and quantify ‘finishing skill’ by asking normal people for a consensus on what they mean by ‘finishing’ and look at all of the criteria in the data to come up with a bag of shots that they can designate as ‘finishing’.

However, by divvying up shots into smaller categories and designating one of them ‘finishing’, the difficulty that exists with football being a low-event game just gets EVEN BIGGER.

A basic knowledge of statistics and variance (basically, flip a coin ten times and it could come up heads 7 times, but that doesn’t mean you’ve got a rigged ten pence) would suggest that reducing your sample of shots would make an already difficult task even moreso.

The most convincing public articles that have quantified finishing skill – one of which here from Marek Kwiatowski – are really complicated, and you imagine that it would be even more complicated with less data to work from.

Data is like pixels. If you have a million pixels, you can probably tell pretty easily what a picture is. If you have 5, you could have a Lamborghini or a lemon.

A counterargument – in this specific example – is that by whittling down the sample of shots SPECIFICALLY to ones which are relevant to ‘finishing skill’, a more obvious pattern might emerge.


Fade to black

Fade back in

A more obvious pattern might emerge. It probably, and most likely, won’t. But that’s no reason to ignore the fact that finishing skill has different meanings in normal parlance and to stats people, and that this is a problem.


Pernickety Postscript:

A team, or a player, ‘getting shots’ is not a good thing. A team, or player, ‘getting shots which are worth taking’ is a good thing, and stops you thinking of Andros Townsend, which happens when you say the former.


Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s