Trying to rank the players

NB – please try to take this in the spirit it’s intended: I’m exploring.  Trying to see where I can get.

You may recall that recently I tried to find a way to objectively assign credit to a football team’s players based on the team’s overall performance. I ran this through the Leeds United team of 1972-73 and was happy enough with the conclusions.

The method is driven by a simple enough ‘engine’: in each game, did Fulham perform better against an opponent than would be expected given that opponent’s record across the season? Or worse.

It’s calculated as follows: say Fulham played a team that averaged 10 goals a game and kept a clean sheet. Whoever played in that match would get a +10 for that game defensively, as this attacking giant was kept quiet. If Fulham scored 10 and our team averaged a clean sheet then they’d get +10 there, too, for a total of +20.

If Fulham had lost 10-0 then this would have been a zero, as Fulham kept the opponent exactly to their seasonal average.

We add up the totals for the season and see which players played when opposing teams were kept above or below seasonal tallies.

Phil Magnus took my Leeds spreadsheet and filled in the latest Fulham season. If we apply the above literally we get the following ranking of players:

Christensen
Ruiz
Burn
Bettinelli
Rodallega
Stafylidis
Parker
Hoogland
McCormack
Hutchinson
Bodurov
Tunnifcliffe

This is all players who played 20 games or more. Which is to say that the team performed MUCH better with Christensen in the side than without him, and a fair bit worse with Tunnicliffe in the side than without him.

Okay, the elephant in the room is McCormack, who played so much he had to take one for the team, which is to say that we couldn’t get a with or without you for McCormack because there’s no without you part. He’s largely alone in this, although Betinelli (39 games) does well because Kiraly and Joronen had bad happenings when he was absent.

Among non-qualifiers Elsad Zverotic (5 games) was actually top, having played his part during a relatively unspectacular but almost unbeaten run halfway through the season. Turner was up there between Ruiz and Burn, and we did okay in games when Richards played. The team did badly in games when Fofana started and also Woodrow, Amorebieta and Hydman (Magath effect!).

I did a second version of this in which defenders got more credit for defensive performance and attackers more credit for attacking performance. So, if we lost a game 6-4 the defenders who played would get fried and the forwards would get credit. That’s probably reasonable enough.

For context, I put these numbers up against Whoscored.com’s ratings and against a subjective view from Mike Gregg, who is good at these things and, I find, takes a balanced approach that neither swings too far towards the ‘populist view’ nor tries to be too contrarian (which I, of course, am prone to do).  These are all rankings. The numbers are meaningless in this context, really:

umbers

RAW is just the pure data: how did the team do when he was playing?  POS is when I adjust this for position played; WS is WhoScored; MJG is Mike.

Hmmm. I think Whoscored is too high on our defenders (top ratings generally going to these people) which can’t be right under the circumstances. Mike’s views, I think, cover a consensus quite well.

So I don’t know. My initial ratings are essentially facts (facts that are open to all kinds of distortion and factors beyond folks’ control, but facts regardless.. kind of): when LVC played, teams did worse against Fulham than they did against all other opponents. Ruiz and Burn had similar impacts (as did Turner later on). You can slice this around a bit to try to account for responsibility and maybe that works, too. Not sure. Clearly I have a way to go here. But the key messages are:

Christensen is really good
Dan Burn did about as well as anyone could in that defence (which is a slightly circular argument, I realise)
But both look like the pillars on which this team needs to be rebuilt
If you take out the Whoscored defensive bias (?) then Hoogland, Hutchinson and Bodurov get panned. Ryan Tunnicliffe doesn’t come out of any of this well.
McCormack was unfairly held back by the raw approach to all this because he played so much, so he must be seen as a success (duh!)
Rodallega’s probably a fair bit better than the fans give him credit for
Parker did okayish

12 thoughts on “Trying to rank the players

  1. If I understand this correctly, this isn’t telling you who is absolutely a good player – it tells you the difference between them and the back-up player who replaced them when they weren’t in the team. So where we have a poor player on the MJG rating, who does well on the RAW, we can see that this the position where we’ll have the highest return on bringing in a new player.

    Or, to put it another way, ‘Bye-bye, Stafylidis’

    1. That ties in with The Numbers Game, doesn’t it – seem to recall that one of its key arguments is that a team is only as good as its weakest player i.e. the one whose replacement will bring the greatest return.

  2. Well done you have confirmed for me just what a lot of nonsense statistics are when applied to football. Any list that has Zverotic at the top and McCormack so far down is fundamentally flawed.

  3. Further to explore, let me add a question that adds to the notion about which Douglas M has reminded us. Let me put it like this…Is a player as strong as his weakest moments?

    WhoScored.com notes positive contributions, but not when a defender is horribly caught out for pace or concentration, even when providing the Assist, so to speak, to a conceded goal. For forwards it records goals scored, but not excellent chances spurned. But of course spectators clock these things and tend to attach significance to them.

    They are right to do so if a typical match is seen as 85 minutes of non-consequence plus 5 with successes or failures around either goal on which the result hinges.

    As you have shown, there is no single way of looking at things that works in all cases and this one certainly won’t. But if we are talking in the context of how to improve, then finding players whose worst moments are less bad/ less frequent does seem a candidate. It’s a judgement only possible subjectively on the basis of viewing and it doesn’t have a great fit with midfielders. So no more a universal tool than the metrics you have presented. And not so much for ranking players (which is interesting but mainly apples versus oranges) but for looking to improve a particular position (apples versus apples). That is one of the two real-life tasks, after all, the other — getting organised — being covered by the article of yours preceding this.

  4. If anyone’s still interested, if I take these through a thoroughly fudged bayesian approach I get:

    Christensen
    Rodallega
    Ruiz
    McCormack
    Burn
    Parker
    Stafylidis
    Hutchinson
    Tunnifcliffe
    Hoogland
    Bettinelli
    Bodurov

    This is achieved with a starting assumption that all players are equal, adjusting for Mike’s subjective view, adjusting for the whoscored control, adjusting this for the raw WOWY and adjusting again for the adjusted (!) WOWY.

    So if I had to end this sorry episode with a list this might be it.

        1. Dunno. The *reason* that Rodders is lifted above Ross (in an apples v apples comparison) is that he flourished when LVC was having his golden couple of months. He lost form after that and after a while (probably too long) lost his place. Can’t see how any set of pure metrics can cater for that, but at the same time it’s not a mystery to the human observer. A bit like robots are unprogrammable for physical actions that a toddler can perform just fine.

          1. Well your human eyes can see this. Others would assert that Rodallega was a waste of space, full stop. But yes, my vision for this has been somewhat frustrated. at least by working things through I can see that I can’t do things this way, which is valuable in some ways. Probably the shortlist approach from a while back would suffice, although that would have missed LVC.

  5. Statistica can prove anything you like to be honest – what you really don’t take into account is the opposition for god’s sake – this is a total waste of time !

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 )

Twitter picture

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

Facebook photo

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

Google+ photo

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

Connecting to %s