The great/terrible thing about being in the stats community on Twitter but not having the necessary skills to ‘obtain’, organise, manipulate, and magic data is that you can throw ideas around without (being able to) test them in actual data. I do, however, big-headed though this may sound and probably is, think that my ideas may add something to the echo chamber of analytics.
One of the problems with assessing defenders through data is that a lot of defending is team reliant, there’s a decent extent to which you need to eliminate the work that is being done by the rest of the team. For example, Vincent Kompany will concede fewer shots than Russell Martin partly because he’s a better player but partly because his team have more of the ball and they’re more capable of defending without it than Norwich.
Building off the idea behind Thom Lawrence’s defensive hulls, I figure that you need to look at what reaches them – their ‘territory’ – and what goes through them, not solely what goes through them. 2014-15 Chelsea for example seemed to defend very compactly, so there might only be a small amount that passes through their territory compared to Manchester City defenders, but this could be down to the position and size of their territories.
I looked at some touch maps from the Arsenal (boooo) vs Leicester game. I drew a by-eye shape around Mertesacker’s touches (this includes passes and defensive actions, so it isn’t perfect by any means for measuring defense, but then defenders occupy different spaces on different types of attacks).
I put this over the Leicester touch map and added vague shading for areas that would probably be in front of Mertesacker and the area that’s probably behind where he would be and for which he would be responsible for. I count, for what little it might be worth, ~17 touches in the green area in front of him and ~11 in the red behind him. This seems pretty good.
I wanted to do the same for Koscielny as well but he came off just after half time so it didn’t really work. There were also difficulties when I tried to do the same for Wes Morgan, who played the full match, for Leicester.
There are two definite lines of defending here (in itself this is quite interesting, though not necessarily useful for this). I took the lower part as his main defensive territory, which may be wrong.
With this territory so close to goal perhaps more of his area should be coloured in red seeing as it’s so close to goal, but it depends how the team is playing. It could well be that in these situations his midfield line is between the penalty spot and 18 yard line, and it’s just that the team as a whole is very, very deep.
Anyway, I tried to keep the area of each colour pretty similar. Totting up the totals in each now I count ~25 in the green and ~24 in the red. It should also be noted that touches in the penalty area may be the result of corners, which makes Morgan’s in this match even worse to consider, but is nonetheless fairly useful in looking at a worst-case scenario for this idea.
The point though is that you’re seeing the defender as a sieve. On one side is the stuff that wanted to be poured through the sieve, and the other is the stuff that got through the sieve. By looking at the difference, you might be able to work out how big the holes are.
Kind of related is just a wish that more people would have a crack at defensive statistics. It’s a bit harder than whizzing up an xG model, but it seems like in a lot of other areas in analytics discussion and differing techniques – and talking about why the techniques differ and whether that offers any benefits – can really help things along.
Maybe it’s just that attacking stats are easy and creating a rough expected goals model is the kind of level of skill that most people are at, and perhaps people are working on more defensive measures behind the scenes. Basically what I’m saying is that if more people start thinking about this then I can think about it less, and I’d really like to be able to think about this less.