Wednesday 11 December 2013

Manchester United 2012/13 - Less Shots, More Quality

I hope many readers are now be familiar with the expected goals model that I’m using on InsideFPL. Rather than counting up a team’s shots on target as I focused on last year, or look at total shots, or shots in the box, I am now able to give every shot an independent ‘value’ based on the location of the shot, the type of shot (foot/header), the pass type (cross/headed/regular) and the situation (set piece/open play). The goal value for each shot is determined from the typical shot:goal conversion rate from all similar shots and is called “Expected Goals” or xG.

Shots En Vogue

The xG model is in vogue right now and is a good step up from just assuming many shots are equal.  You may be familiar with similar models from folks like Paul Riley (@footballfactman) and Colin Trainor (@colinttrainor) as well as my personal inspiration over the summer 11Tegen11 (@11tegen11). 

Anyway, to the point of this post, my model stands up very well for all team’s goals scored last season with a correlation value of 0.933 (R^2=0.870) but as you can see from the image below breaks down when it tries to deal with the top scoring clubs. This analysis excludes penalties. I also need to ‘fess up I've noticed five of 2012/13’s games missing from my data currently.

Basic xG model vs. Actual Goals Scored for 2012/13 seasons (exc. penalties)

Model Break No. 1

5 teams - Chelsea, Liverpool, City, Spurs and Arsenal  - scored within a few goals of each other last season (excluding penalties) but the model has up to a 10 goal difference them all.

Model Break No. 2

Could not explain how Man United managed to score 12 more goals than their nearest rival despite a lower xG.

(You’ll notice also Aston Villa veering from expectations but I haven’t looked at them in much more detail yet).

Do We Need A Fergie Factor?

Not being able to explain the goal scoring of these top clubs, especially the team that wins the league, is a big kick in the teeth any stats model. It’s all well and good having a nice straight line through all the middle of the table clubs but if you need to start using words like ‘luck’ or ‘world class’ or ‘Ferguson’ to explain why a team can do something others clubs can’t the model starts to lose value.

What did United Excel At?

I looked at the numbers to see if I could find a particular element of the game that Man Utd could claim they bossed and quickly saw that they had the highest ratio of Expected Goals to Shots of any team (xG/S), and by a good margin too. This could be conceived as “Shot Quality”, i.e. given your team takes 1 shot what is it’s expected goal value, or how good is the average chance you create?

I biased each team’s xGT value from the first chart with the amount it’s “shot quality” xG/S value was above or below the league average to get the following results:

Biased xG model vs. Actual Goals (exc. penalties) for 2012/13

Fixed or Fudged?

The correlation is only slightly improved but the good news is it’s “fixed” United - i.e put htem on top, as well as the group of CHE/LIV/TOT/ARS. However, rather than fixed I’m tempted to say “fudged”. but have I biased the data to get what I wanted to see?  Thinking about what xG per shot actually means on the pitch however - it’s a very important thing. Imagine your team just created 1 or 2 “golden” chances per game. I’d certainly take that over 10 frustrated or speculative shot from outside the box.  

You can also imagine what it means if your team cannot achieve a high xG/S.You’re forced into lots of shots that perhaps you don’t want to or shouldn’t take (bad decisions/no options), you don’t have the skill to pick the lock on a stubborn defence, or you cannot counter attack effectively.

Did Mancini Break City?

Looking at the chart above Man City are the team that are really below the line and should perhaps have ended the season with a lot more goals (and points) than they did.  I can’t explain this any further at this time than to point out Man City “looked” like they had problems last season, Mancini was sacked and this year, with just Fernandinho different really (Negredo = just another striker), they are storming teams. 

It’ll take a deeper look into the numbers for City but having just one team to fix rather than 6 is much more reassuring. Could an unfit Aguero struggling with injury make a difference? I’m also tempted to speculate that attack pace/tempo might be what lifts Villa above the line and drops City below it.

Quick Wrap-Up

Dist. xG/S across Lge 12/13
Back to United, their xG/S was way above the league average last season and a good way above even 2nd place City’s as this little graphic shows (click to zoom). Could this be what a deep playmaker on his game like Michael Carrick brings to a team?

To quickly conclude, I haven’t done anything further with xG/S yet other than to look  briefly at this season’s data and can tell you Arsenal and City and then Liverpool are highest so far the season, Sunderland and Hull are bottom. United are doing okay, above average, but not great.


  1. I think I've seen a few articles about the analytics community that point to City massively overperforming their xG in 2011/12 and underperforming in 2012/13, which makes it seem like they were incredible one year and awful the next. Perhaps there's a reason why that all changed, but I'd lean to the argument it's just been random variation, and a bizarre one. Would be interesting to see how they've done over the 2 seasons combined.

    I think the key to City's progression this year is some key additions to add more balance to the team, Navas adds width, Fernandinho offers allround class and reliability in midfielder and Negredo is a good link up man for Aguero. Add in Aguero's fitness and Nasri's form and Man City have a much complete team, and don't have the disruptive influences of Balotelli or Tevez either.

    1. I think I will try and grab 2011/12's data and see how it shakes up. I tend to feel there's an explanation for most things so I must look.