Quickly assessing the ability of individual players isn’t a particularly easy task; firstly because a player’s overall effectiveness isn’t necessarily the sum of a series of data points (goals, assists, passes, tackles etc) and secondly – good quality, manipulable, player-data is difficult to get hold of.
There are a few good sources of Team data – including the excellent football-data.co.uk, which allows access to continually updated CSV files, containing a treasure-trove of match information, across a variety of leagues. I use this for my shot-based Adjusted Goals team rating system. But it’s a different story for player data – it’s accessible but (deliberately) only in a format unfriendly to robust analysis.
However, I’m keen to explore the impact of individual players on team performance in more detail. Whilst my shot based Team assessment (Adjusted Goals) is proving an excellent starting point for individual match modelling, it doesn’t take account of team changes – including the absence of key players. A change of player personnel obviously does affect a team’s overall ability and tactical make-up, but by how much? For betting purposes I suspect that there’s a rich vein of potential value to be found in properly adjusting for the absence of key players – essentially, because it’s so difficult to make an objective assessment, meaning that the market may over (or under) react.
As an extreme example of this, when Liverpool played at Bournemouth on 17 April 2016, their opening Pinnacle odds were 2.10. These odds already allowed for the fact that Liverpool were likely to field a weakened team, due to European commitments. However, by kick-off their odds increased by 32% to 2.77 – reflecting that Klopp made more changes than expected.The true effect on likelihood of these changes was clearly difficult to assess – but I suspect that there was true value in these odds somewhere along the line.
My initial attempt at Premier League player assessment uses the same method I deployed to rate International teams for Euro 2016. This takes a variety of (reasonably accessible) data points to assess each players effectiveness score for their particular team – for both attacking and defensive strength.
For reference, the components are:
|Attacking Component per 95 minutes|
|Defensive Component per 95 minutes|
Now, sadly, I can’t point to any scientific basis for these weightings. These were based on trial and error to get – what looked right – using 2015/16 data. But it’s a starting point.
Using these, I assess each player’s relative effectiveness for their particular team, by dividing by the average score for that team. This is then multiplied by that team’s relative attacking or defensive strength (from my adjusted goals rating method) – to give an overall rating – where the average Premier League player has a rating of 1.
So far, for the 2016/17 season the ratings are as follows:
|Rank||Attacker Rating||Team||Attack Rating|
|3||De Bruyne||Man City||3.30|
|Rank||Defence Rating||Team||Defence Rating|
|Rank||Overall rating||Team||Overall Rating|
|5||De Bruyne||Man City||1.96|
These aren’t perfect, for example James Milner’s top position is significantly influenced by his penalty haul. Also, seven matches isn’t really long enough to properly assess effectiveness.
However, it does given me a starting point for objective assessment for player changes, that I can use in conjunction with my team ratings.
There are player rankings available on some of the football stats sites – but I haven’t a clue how these are complied, so I can’t judge whether they’re robust. It’s preferable to construct your own rating method, where it’s much easier to understand the shortcomings, and can be improved over time (which it what I’ll be attempting to do with these ratings).
So how does the much maligned Wayne Rooney rate using this assessment method? He’s ranked 99th in the Premier League and 9th at Man U this season – with a rating of 1.14. Just above the average Premier League player – which seems about right.