Tuesday, 27 August 2013

New Fantasy Football Stats (Primus Inter Pares)


What Alan Shearer Says
I was watching Match of the Day on the opening day of the season and they flashed up a graphic for Alan Shearer to talk about it.  The graphic contained stats for Daniel Sturridge and Coutinho's excellent performances at the end of last season. As well as Appearances, Goals and Assists they talked about shots on target.

This time last year they wouldn't have done this but over the subsequent twelve months underlying data like shots on target has slowly crept into the mainstream game as a meaningful stat to indicate what's going on under the bonnet of a football match.  I'm obviously pleased to see this, because it is important, but I am now ready to take it on a level or two.

Not All Shots Are Equal
Spending so much time looking at shots on target, comparing these against other shot data (like total shots, shots in box) and evaluating them against actual goals and, more importantly for us, fantasy football points I've become fully aware of the limitations. My FSCORE metric and Point Projections last season were predominately based on team's and player's shots on target.  I discussed the limitations  in this post, identifying a crop of important fantasy football players where the model just didn't fit. 

I've spent the summer months researching new data sources and am pleased to be able to vastly increase the quality of the data. Instead of looking at just shot totals, whether on target or not, or in the box or not I will now be incorporating the following:
  • Shot Zone: 6 yard box, central penalty box, side of penalty box, outside of box
  • Assist Type - attempt from a cross or pass
  • Attempt Type - Header or Shot (with foot)
  • Situation - Set Piece of Open Play
I have to again bow to 11tegen11 for his initial analysis of this earlier this summer. Shot Zone is the most important of these. It's common sense. The closer you are to the goal, the better the chance you have of hitting the target and scoring a goal. Shot Zone's is the most important single improvement to the model this season but the other factors should not be overlooked. 

For headed attempts the chance of scoring in the 6 yard box is relatively very good but the conversion rate for headers drops more quickly further from goal than it does for regular shots (with a foot that is). Similarly for set pieces, if the resultant shot or header is not close to the goal it's harder to score than it is for a shot from open play from a similar position.

Here's a table of just some of the most important "shot criteria":

Shot Criteria Typical
Conversion
Rate
Any shot or header in the 6 yard box (Zone A) 35-50%
Shot from a pass in open play from central penalty box (Zone B) 25%
Shot from a set piece in 6 yard box or central penalty box 25%
Header from a cross into 6 yard box (open play) 18%
Header from a set piece in 6 yard box or central penalty box 15%
Shot from a cross in open play from central penalty box 10%
Shot from either side of penalty box (open play) (Zone C) 10%
Header from centre of penalty box (from a cross) 9%
Shots from outside the box (Zone D) 5%

Using the Model
Once the data is collated I add up each team's and player's shots to give an expected Goals value (xGoals)and an expected assists value (xAssist) and then combine both of these with the point scoring rules in FPL to determine expected points (xPts). You can see this in action already over at InsideFPL. I'll be doing weekly team and player ratings using xG for Teams and xPts for Players. 

xG and xA will also contribute significantly for improving the Player Point Projections. Assists will be much better modelled. Knowing whether a player is creating good chances or not is something that simply counting key passes cannot tell you. Leighton Baines is a great example of this. He racks up lots of key passes every match but looking closer at this new data you can see most of his crosses result in headers in shot zone B (9% conversion). Compare this to Theo Walcott who doesn't create half as many key passes but when he does they typically result in shots in the 6 yard box (35% conversion). 

There's also real super potential in drilling deeper into the type of shots a team creates or concedes, and then using this to determine what type of shots are most likely in any given fixture and which players will likely get these chances. 

There's a tonne more I could waffle about but this is it for now (like what is Zone BD?). I'll be doing Match Centre for GW1 and GW2 very shortly using this data and as mentioned you can take a look at Player Ratings (xG, xA, xPts) for GW2 over at InsideFPL.

1 comment:

  1. Good stuff Ste!

    Categorizing the shots based on their quality should definitely improve the accuracy of F.SCORE.

    I wonder if we can do something similar to improve the the defensive strength analysis of all teams. Eg: A team like West Ham although concedes quite a few shots, very few of them are in Zone A, which should improve their defensive metric score.

    ReplyDelete