Monday, 13 October 2014

The Value of Corners - 94 clubs and 25,000 corners analysed

I was planning on looking at effectiveness from corners anyway but Swansea's stats in this regard make it even more pertinent.

From the 7 Premier League games so far this season, Swansea have had 12 corners with the next lowest team being Burnley with 25 (West ham are highest with 53). Why Swansea have had so few is for another blog but this one is concerned with the question of do corners count and who does them well.

Ben Mayhew from the Press Association tweeted me this image looking at Average Number of Corners per game and final position:
As with any stat there will be exceptions but averaging over 6 corners generally comes from a team in the top 8 not one fighting relegation.  Swansea's average so far of 1.7 per game is off the scale (but not in a good way) 
There appears to be some link between corners and final position although I'd expect it to be more 'Good teams win lots of corners' rather than 'Winning lots of corners makes a team good'.

There's a good piece by James Horncastle from a couple of years ago looking at a guy in Italy called Gianni Vio who is considered a set-piece guru.  He's recently joined AC Milan so is still making a career out of being a set-piece specialist.

There have been a number of studies looking at the value of a corner such as by Anderson/Sally in 'The Numbers Game' and on Statsbomb among many others.  One of the issues is how to credit a goal to a corner? It's easy enough if the goal comes straight from a corner crossed in, but what about the secondary phase after a corner is cleared or if the corner is initially played short?

Most previous analysis credits a goal to a corner if it is scored within 3 touches, what I've tried to look at is to use the match reports on the BBC that are generated using Opta data to try and analyse activity in the period after a corner is awarded. The data isn't perfect as the timestamp relates to when it was logged that a corner was awarded and not when taken, but is good enough for some rough calculations.

From this data I've gathered information on over 25,000 corners that have been taken in games involving clubs from the English leagues (includes League, Domestic and European cups to try and get as much data as possible even if it risks adding unfair comparisons).  These games took place between 1st Oct 2013 and mid-Sep 2014.

The first thing I looked at was corners that produce other corners, the aim of this was to remove double counting so at a later stage you could say 'If I get a corner, I'd expect this phase of play to finish in a goal x% of the time'.

Overall, from the 25,892 corners analysed 3,079 (12%) are awarded within 60 seconds of the previous corner being awarded so could crudely be classed as part of the same phase of play.  The linked corners could either be as the result of a shot or due to a defensive clearance (it's there in the data but that's for another day).

No Surprises - Time to next corner is most likely to be shortly after you've been awarded a corner.  There will be a variable time between the awarding and actual taking of a corner but things level out around the 60 second mark
As you'd expect, the 12% isn't a uniform figure and there are some major differences by team, ranging from 18% down to 7.5%:
10 Highest and 10 lowest clubs in terms of % of corners that are awarded within 60 seconds of previous corner. Everton's and Arsenal's corners per game figures are much closer when factoring out linked phases with multiple corners (I'm guessing Everton favour in-swinging corners towards near post more than Arsenal)
When charting average number of corners per game by % of corners that are awarded within 60 sec of a previous corner a few teams stand out:
In terms of raw average Corners per game Burnley and Scunthorpe are pretty similar but Scunthorpe have far more 'Linked' corners. Overall Chesterfield have the highest number of 'Corner Phases' per match with Man City in second with Leeds the bottom of the 94 teams analysed (the 92 from 2013/14 along with Luton and Cambridge)
Corners are all well and good, but ultimately it's goals that count, you know you're clutching at straws if you say '..but we won 8-2 on corners'.

If we look at the last corner in a sequence (where no corner awarded in previous 60 seconds), we are left with 22,813 corners from which to analyse the subsequent goal return.
The presence of 3 goals occurring the same time the corner is awarded shows the data isn't perfect but similarly to corners, goal response tends to tail off just after the minute mark.
Again as with subsequent corner activity, there's huge variation in number of corner phases that lead to a goal:
Top/Bottom 10 clubs by % of their corner phases that end with a goal within 70 seconds
When plotting Number of Corner Phases by Conversion rates you get some major differences:
Performance ranges from Man City who were had the 2nd highest number of Corner phases and 5th best conversion to Dagenham & Redbridge who had 3rd lowest number of corner phases and lowest conversion
All of the above looks at the attacking impact of a team but this can be reversed to see how well each team fares when facing corners:
These figures just look at Goals Scored/Conceded per match without taking into account volume of corners.  Reading show the ideal combination of High scoring and Low conceding from corners while Dagenham & Redbridge are the worst of 94 clubs in both aspects.  Those below the blue line are 'Corner Positive' in that they score more than they concede from Corners
One of the main questions from anything like this is 'Is this repeatable?' are clubs that were performing particularly well/badly in this analysis doing the same previously (and more importantly are they likely to do so in the future)?  I only have approx 1 season's worth of data in this format but it's interesting to see that the Stasbomb piece that is from Aug 2013 also mentions Newcastle at being bad from corners over a totally separate time period.

In an ideal world if you had the Prozone data for example, you could code up all corner scenarios where at least 1 centre back enters the opposition penalty area and link all activity as part of the same phase until that player has returned to his own half.

Also, all the timings of events relate to when events entered in to the system and the 60/70 second cut-offs are arbitrary as is the decision to include all games and not just league but the analysis shows the kind of things you could look out for when creating your own bespoke metrics at a club.  There's also the risk of conceding on the counter-attack from your own corner that needs to be factored in, Stoke's winner and Chelsea's goal v Man City at the Etihad this season were both breakaway's from Man City Corners.

The next stage will be to try and look at what teams that have a higher scoring % do.  If the answer to the question 'How do I score as many goals from corner situations as Man City' is 'Buy Silva, Aguero and Dzeko' then that's no use to anybody.

Similarly if Reading's strength in scoring from and not conceding from corners is due to having centre backs who are terrible at everything else that's not much use either, but given there are plenty of less glamorous clubs with high returns from these situations suggests that it could also be process and not just personnel that matters.

Corners alone won't win you a title or get you relegated but in a game of fine margins, if everything else remains unchanged, well executed corners (and set pieces in general) can make a difference.

Other Posts: Premier league Shot LocationWorld Cup Shot Location World Cup Distance + Sprint Stats
Twitter: @we_r_pl
Stats: Created using BBC Sport