So the recent discussion on Twitter about using analytics to help choose who to pick and how to set-up for an upcoming match got me thinking. I remember some stuff on pressing, and this by @SaturdayonCouch on using some of the stats work he’s done to make changes in match which I like. The discussion also mentioned tactical theory, something which isn’t my strong point.
I did however think of profiling players with their stats (not new, I know), looking at who creates chances against your next opponents and then looking at your own squad list to see if you have anyone similar (kind of similar to this by Neil Charles).
I was working with what WhoScored have on their site so put into excel dribbles, crosses, detailed passing stats (which on WS are accurate/inaccurate long/short passes), offsides, crosses blocked, and through balls. Through balls and offsides are per game, whereas everything else is per 90.
From these, and doing other things like looking at how many pass attempts they make per dribble attempt (where my Carrick vs Lennon tweet came from), I could kind of identify rough types of players by stats (ish, not really to a standard any rigourous data types would like, and its by eye which also has its dangers).
For example, players who have a very low percentage of their passes being crosses, and low amount of crosses blocked, are likely to be central. Coupled with a fairly low amount of pass attempts per 90, probably relatively high offsides per game, and probably a low number of pass attempts per dribble attempt, and it looks like you have yourself a striker.
Blocked crosses help for identify full-backs, but they can still sometimes ‘look’ like defensive midfielders and the line between centre-back and defensive midfielder is also a bit blurry. It should also be noted that I’ve only got 60 players in my spreadsheet.
So, in a simple example, I’ve brought up the two players who created more than one chance (Surman 2, Smith 3) against Arsenal in their last game along with all of the Leicester players that I have in my spreadsheet.
|Player||Club||Dribble %||% Long||% Acc||Pass Att||Pas/Drb||Offsides*||Pass-Cross||Crosses blocked|
|King, A||Leicester City||71.43%||12.70%||76.76%||37||52.8571428571429||0||2.16%||0.2|
Surman has barely any crosses and high pass attempts per dribble attempt and pass attempts p90 numbers, suggesting to me he plays deep-ish in central midfield (his dribble success% is also fairly high, which I think can be a bit of a suggester of the same thing through dribbling in less congested areas).
Smith blocks quite a few crosses, comparative to most other players, so he probably plays on the wing. He doesn’t cross often in his passes, so he’s not an out-and-out winger, though does dribble quite a bit. Maybe an attacking full-back.
Surman, as a deeper lying CM, looks a bit like Andy King or Drinkwater as central players who don’t dribble as much as the rest of the team.
Smith doesn’t have an easy comparison in this small sample. Fuchs, blocking 2 crosses a game, is an obvious full-back, but his high amount of pass attempts per dribble attempt is a far cry from Smith. Albrighton is similar on those two numbers, but crosses a hell of a lot more, but even without picking a single like-for-like you can extrapolate from Smith’s success that Arsenal conceded chances down the flank.
Ideally a lot would be different from this example, which I deliberately made quite basic. You’d of course want to look at a lot more chances that your next opponents conceded, ideally using some way to combine or highlight similar patterns in the players that created those chances, so you’re not trawling through thirty-odd lines of players (for example, there were 7 Bournemouth players who created one chance vs Arsenal).
In an ideal world you’d also have more data to work with (but then everyone always wants more data), and you’d come up with some data-led, objective, algorithmic way of clumping players, instead of the more by-eye and intuitive way that I’m doing.
You’d also probably need to look at the league average for the amount of chances that, say, full-backs created against teams. There’s no point going “oh wait, these guys concede LOADS of chances to full-backs, let’s focus everything through them” if they’re actually just league average for this.
It’s still potentially interesting, I think, though because like I mentioned above you can make educated guesses as to an opponent’s weakness based on what type of players are creating chances against them.
Of course, you wouldn’t use this to just pick like-for-like players who create chances against your opponent, playing with 4 wingers and 5 deep-lying central midfielders, and team dynamics will make some decisions for you, but it might help you decide a couple of positions.
I mentioned when I first posited this kind of thing on Twitter though that I thought it was both too simplistic and too complex. Too simplistic because there are lots of factors that go into football and winning; too complex because creating adequate player profiles that you can group people in for this kind of thing could get ridiculously hard. Also, I suspect that some people have already done this kind of thing, it’s just that I haven’t seen much of it in the public sphere – maybe there’s a reason for that. Also, I don’t really think this is what the conversation was aiming at, more at combining analytics and tactics or tactical theory instead.
Thoughts and advice is very much welcome 🙂