New Method

This is a write-up of my latest method to try and find some sense in central defender statistics (because I refuse to fail). There’s a kind of tl;dr short version at the bottom.

It starts with an idea of what central defenders (when defending) have to do:

1) Stop chances [A] (relative to when they are called upon[B])

2) Don’t give the ball away close to goal (linked to 1, but can be separated for ease of compartmentalising stuff)

The method also has to balance ‘accuracy’ and ‘simplicity’, essentially because I’ve got to be able to do it.

I split 1A up into:

  • ‘Actions’, things that can be easily measured. Nicknamed ‘Blacksmith’ (not for any particular reason, it just seemed a nice juxtaposition to the nickname for the next thing).
  • ‘Preventing passes/advancement’. This is not a great name, because it applies to the above as well. However, by this I mean the positioning side of things, the awareness to cut off passing lanes etc. Nicknamed ‘Magic’.

1B was split up (and thanks to @ASOWIGROVENFAR for helping to clear my thoughts on this):

  • Defender’s team must not have the ball
  • Opponents must be making a direct effort to score (ie not ‘sterile possession’)
  • Centre-backs must be the main line of defence (ie the opponent’s attack isn’t being primarily stopped by a midfield layer of insulation)

2 was not split up.

Regarding 1A, I used tackles, interceptions and ball recoveries per90 as ‘actions’. My reason for not including clearances is that I believe they are a lot more to do with the pressure that your team is under than the other actions. The reason for including ball recoveries, despite their problems [they can count tackles and interceptions, so these will be counted twice, and rely on the team keeping the ball so those on better teams will probably have more recoveries just because of their team-mates], is that I believe there is some skill to recoveries. Ideally, there would be a stat available like recoveries that would not count the same actions twice for a player, but this is what I have to make do with.

‘Magic’, for the moment, I have left alone, as I didn’t know where to start with it yet and wanted to logically move through the other aspects first.

I thus have the ‘stopping chances’ part of my rough definition of what a central defender should do.

‘Relative to when called upon’

For the ‘team must not have the ball’, possession is the obvious choice.

For ‘opponents must be making a direct effort to score’, I chose to use attempted deep completions conceded [kindly provided by @SaturdayOnCouch], ‘deep’ being within 30 yards of goal. This was just because it seemed the best (balancing ‘accuracy’ and ‘simplicity’) method to hand.

For ‘centre-backs must be the main line of defence’ I used the percentage of a team’s tackles and interceptions that the central defender in question makes. I didn’t include ball recoveries because I thought that tackles and interceptions are done all over the pitch and will be done by the protective midfield if a team is containing, whereas I’m not sure the same is true of recoveries. Perhaps I should include them as they’re included in the ‘Blacksmith’ actions I’m using.

I calculated three standard deviations above and below the mean of these and then ‘normalised’ to a decimal 0-1 [x – 3 SD below / (3 SD above – 3 SD below)]. At least, this is my probably quite bad maths attempt to normalise all of these. I added the three of these up and divided by 1.5, as 0.5 is the average, meaning that those above average in these three areas will have a figure of over 1.

‘Don’t lose the ball away close to goal’

I collected possession losses by ‘Unsuccessful Touches’ (described by WhoScored as ‘bad control’) and ‘Dispossessed’ (‘dispossessed on the ball by an opponent – no dribble involved’). I also got the number of ‘Inaccurate short passes’, as, of the passing stats I had easily available I thought that would best reflect what I wanted. I added the unsuccessful touches and dispossessed numbers together, as they were similar in numbers p90, and then ‘normalised’ both this number and the inaccurate short passes. Following this I added the two together, and once again anyone above average would have a number above 1.

Cooking it up

Having assembled all (most, minus the magic) component parts, I averaged the ‘normalised’ numbers from the previous two sections, then multiplied this by the number of actions.

I did this because I wasn’t sure what way was best to use these qualifiers to adjust the number of actions. I feel that losing the ball should be weighted less than stopping chances, but a) I am unsure how to do that and b) I think that the losing the ball numbers partially reflect team style, thereby also acting to an extent as another ‘relative to when called upon’ qualifier.

This method comes up with this list as the ‘top 20’ centre-backs in the Premier League for 2014/15 (slightly different to the one I posted on Twitter because I had used the wrong ‘deep completion’ numbers).


As it seems no-one really knows much about who are good central defenders and who aren’t, I can’t comment *that* much on it. Huth, Davies, and Wasilewski’s inclusion concerns me; Okore’s to a slightly lesser extent as he’s had a good reputation; the bunching of the Manchester United players is a concern too. There’s also some concerning bunching of Stoke centre-backs right at the very bottom of the list.

The Chelsea centre-backs are down in the middle of the 65 or so players in the list (all of whom played more than 10.0 90s). I’ve come to the view that Chelsea’s system last year makes their CB stats basically unworkable as they were sheltered in such a peculiar way, though this is partly because I *know* that Terry is a legitimately good defender and doesn’t deserve to be that low. As well as the raw stats for both Terry and Cahill being very low, they are very similar when compared per90, more so than at other clubs, which adds to my above theory.

I’ve not included any measure of chances (whether shots or deep completions) conceded in this, as at the moment I only have these on a per team basis, rather than per player, which would just add to the level of team effects at play here. I also haven’t included tackle % (tackles made vs tackles ‘missed’), which I also have a suspicion may have *some* useful link to player ability, just because I wasn’t sure where to fit it in.

Ideally, I’d be able to test this in multiple leagues over multiple seasons against a solid list of centre-backs that were definitely good, but for various reasons this isn’t feasible, which is a shame.

I hope the above all makes sense and any thoughts are welcome.

Short version:

[Tackles + interceptions + recoveries p90] * [AVERAGE (possession+deep completions+% of team actions), (dispossessions+misplaced passes)]


One thought on “New Method

  1. Pingback: More of a look at the new method | Every Team Needs A Ron

Leave a Reply

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

You are commenting using your 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