Matching playing styles with primary school level maths
I'm aware (because I've remember reading it but can't find it) that some people have already developed models for finding similar playing styles across teams. This is my attempt to sketch out how you might do something similar with free resources and basic maths.
In yesterdays blog I suggested that teams should research very carefully where they send their players on loan.
My initial theory is that attacking players should only be sent on loan to attacking teams if you want to increase their value. This is the Ademola Lookman / Serge Gnarby situation where a year in WBA reserves decreased the value of Gnarby whereas 4 months in Leipzig doubled Lookman's market value.
But also that if you are sending a player on loan with a view to bringing them into your first team then the best thing would be to send them to a team who play a similar style of football to you.
There are two ways of doing this; my preferred method would be forming a lose organisation of collaborating clubs to commit to a way of playing football and share best practice and players to achieve this. To pick some random clubs this could see Fulham, Bologna and Dijon working together on a pressing 4-3-3 model with Fulham loaning a striker to Bologna and Bolgona a defender to Dijon (I may have chosen these team names due to hunger).
An alternative would be to look for patterns of play that show similar tendencies.
I'm a big believer in the Pareto principle (or at least saying 80% of this comes from 20% of that) - so I reckon, without evidence, you can get 80% of the benefit of analytical thinking with 20% of the brain power of other bloggers using free resources on places like whoscored.com and understat.com
So taking that I'm going to look through the stats and see what I'd look at to find similar teams. Keep in mind I am trying to find the minimum viable method for doing this. I'm sure there are loads of extras that could be done to improve this. But I want to get something useful for barely any effort.
I'll start with passes.
Here whoscored.com provides us with quite a few useful things:
Total passes per game sees Chelsea and Man City way ahead of the pack and the rest of the teams pretty much where their reputation would see them. Bournemouth maybe slightly lower and Leicester slightly higher?
If we sort the table by Accurate long balls per game we see Sarri's hoof merchants Chelsea in 5th whilst Warnock's quick interchanging Cardiff sit in 19th. Err hang on....sorting by Inaccurate long balls shows us Cardiff top and Chelsea bottom. A quick lesson that the interpretation is more important than the number. Chelsea play long when a long pass is on and perform it accurately, Cardiff play long for tactical reasons.
So I think a good starting point for a team comparison model would be:
Total passes per game.
Long balls (whether accurate or not) as a percentage of total passes
Crosses (whether accurate or not) as a percentage of total passes
A good start but that isn't telling me which players in the teams are making the passes, big difference between passes across the back 4 and passes around the opposition penalty box. I'll come to this later.
Shots next:
Total shots probably just tells me how good a team is so I'll go over to Understat.com for their expected goal data. I love the detail on this site but comparing across teams is painful (unless I am doing it wrong).
Of their vast array of stats I think:
Open play shots as a percentage of total shots
Open play expected goals per shot
Will probably give me the best overall impression of the type of chance a club is creating. A team with a high level of xG per shot will tell me about the quality of chances they are creating. A shot map would be nice but I suspect would yield similar results (and remember I am going for the minimum data)
Pressing
Passes per defensive action in the opposition half is something my brain struggles with. I understand it to mean "how many times do you let the opposition pass the ball before you try and nab it off them when they are in their own half"
A low number means you swarm all over them, a high number means you sit back and set up whilst they knock it around.
The other stat "opposition passes per defensive action in the opposition half" seems to be being called press resistance now? I think that means that you get to pass it around a lot before anyone trys to call you down. I'm not sure if that is because of "press resistance" or because the poorer teams tend to get back into position quickly against the better teams and are happy for them to knock the ball around miles from their goal. @Efaroh has done some great work on this.
However, despite making my brain hurt, I think both stats are potentially useful for making comparisons, although this stats seems to be very league specific
I'm including them for now though:
Passes per defensive action
Opponents passes per defensive action
Key question: Do I need to adjust these figures for leagues? Perhaps a above or below league average type thing? Or a rank?
Formation
It also seems wise to pick a side with a similar formation.
Whoscored seems to have a formation summary for each team which is the best source I've found for this.
But not every 4-2-3-1 is the same, so perhaps we would need to combine the data with player level data? Like passes per player per 90?
That is getting complicated - could we sample it?
Is there a quick way of sampling a squad to see who the play makers are?
I guess that is what xGoalchain and XGoalBuildup are for on Understat.com but they don't pass the eye test for me with Gana at almost double every other Everton player.
I'll leave this for now.
So who is the nearest to Everton in style in the Bundesliga?
Everton stats:
Short passes per 90: 364
Long ball %: 12
Cross %: 5
Open play shots as a %: 67%
Average xG per open play shot: 0.11
Passes per defensive action: 10.14
Opponent passes per defensive action: 10.14
Formation: 4-2-3-1
I'll do this by a process of elimination.
Passes per game - Bayern Munich and Dortmund pass it way more so they are out. Freiburg, Frankfurt and Dusseldorf too for passing far less.
Long ball % rules out - Wolfsburg (hoof!) and Augsburg
Cross % rules out - Gladbach and Hertha
Average xG per open play shot: Hannover 96
Passes per defensive action: Mainz and Nuernberg
Opponent passes per defensive action: Werder Bremen
Teams left: Hoffenheim, RB Leipzig, Bayer Leverkusen, Schalke and VFB Stuttgart
Now we have a manageable list of outcomes similar to my Everton list I'll apply the formation filter:
Looking for similar to 4-2-3-1
Gone then are Hoffenhein, RB Leipzig, Schalke and VFB Stuttgart who don't line up like that.
Which leads me to the remaining team: Bayer Leverkusen.
Stats:
Short passes per 90: 417
Long ball %: 13%
Crosses: 4%
Open play shots as a %: 71%
Average xG per open play shot: 0.10
Passes per defensive action: 12.02
Opponents passes per defensive action: 10.72
Formation: 4-2-3-1
So there we go using my back of the envelope calculation style I see Everton are most similar to Bayer Leverkusen in playing style.
Does that pass the eye test? Let me know.
In yesterdays blog I suggested that teams should research very carefully where they send their players on loan.
My initial theory is that attacking players should only be sent on loan to attacking teams if you want to increase their value. This is the Ademola Lookman / Serge Gnarby situation where a year in WBA reserves decreased the value of Gnarby whereas 4 months in Leipzig doubled Lookman's market value.
But also that if you are sending a player on loan with a view to bringing them into your first team then the best thing would be to send them to a team who play a similar style of football to you.
There are two ways of doing this; my preferred method would be forming a lose organisation of collaborating clubs to commit to a way of playing football and share best practice and players to achieve this. To pick some random clubs this could see Fulham, Bologna and Dijon working together on a pressing 4-3-3 model with Fulham loaning a striker to Bologna and Bolgona a defender to Dijon (I may have chosen these team names due to hunger).
An alternative would be to look for patterns of play that show similar tendencies.
I'm a big believer in the Pareto principle (or at least saying 80% of this comes from 20% of that) - so I reckon, without evidence, you can get 80% of the benefit of analytical thinking with 20% of the brain power of other bloggers using free resources on places like whoscored.com and understat.com
So taking that I'm going to look through the stats and see what I'd look at to find similar teams. Keep in mind I am trying to find the minimum viable method for doing this. I'm sure there are loads of extras that could be done to improve this. But I want to get something useful for barely any effort.
I'll start with passes.
Here whoscored.com provides us with quite a few useful things:
Total passes per game sees Chelsea and Man City way ahead of the pack and the rest of the teams pretty much where their reputation would see them. Bournemouth maybe slightly lower and Leicester slightly higher?
If we sort the table by Accurate long balls per game we see Sarri's hoof merchants Chelsea in 5th whilst Warnock's quick interchanging Cardiff sit in 19th. Err hang on....sorting by Inaccurate long balls shows us Cardiff top and Chelsea bottom. A quick lesson that the interpretation is more important than the number. Chelsea play long when a long pass is on and perform it accurately, Cardiff play long for tactical reasons.
So I think a good starting point for a team comparison model would be:
Total passes per game.
Long balls (whether accurate or not) as a percentage of total passes
Crosses (whether accurate or not) as a percentage of total passes
A good start but that isn't telling me which players in the teams are making the passes, big difference between passes across the back 4 and passes around the opposition penalty box. I'll come to this later.
Shots next:
Total shots probably just tells me how good a team is so I'll go over to Understat.com for their expected goal data. I love the detail on this site but comparing across teams is painful (unless I am doing it wrong).
Of their vast array of stats I think:
Open play shots as a percentage of total shots
Open play expected goals per shot
Will probably give me the best overall impression of the type of chance a club is creating. A team with a high level of xG per shot will tell me about the quality of chances they are creating. A shot map would be nice but I suspect would yield similar results (and remember I am going for the minimum data)
Pressing
Passes per defensive action in the opposition half is something my brain struggles with. I understand it to mean "how many times do you let the opposition pass the ball before you try and nab it off them when they are in their own half"
A low number means you swarm all over them, a high number means you sit back and set up whilst they knock it around.
The other stat "opposition passes per defensive action in the opposition half" seems to be being called press resistance now? I think that means that you get to pass it around a lot before anyone trys to call you down. I'm not sure if that is because of "press resistance" or because the poorer teams tend to get back into position quickly against the better teams and are happy for them to knock the ball around miles from their goal. @Efaroh has done some great work on this.
However, despite making my brain hurt, I think both stats are potentially useful for making comparisons, although this stats seems to be very league specific
I'm including them for now though:
Passes per defensive action
Opponents passes per defensive action
Key question: Do I need to adjust these figures for leagues? Perhaps a above or below league average type thing? Or a rank?
Formation
It also seems wise to pick a side with a similar formation.
Whoscored seems to have a formation summary for each team which is the best source I've found for this.
But not every 4-2-3-1 is the same, so perhaps we would need to combine the data with player level data? Like passes per player per 90?
That is getting complicated - could we sample it?
Is there a quick way of sampling a squad to see who the play makers are?
I guess that is what xGoalchain and XGoalBuildup are for on Understat.com but they don't pass the eye test for me with Gana at almost double every other Everton player.
I'll leave this for now.
So who is the nearest to Everton in style in the Bundesliga?
Everton stats:
Short passes per 90: 364
Long ball %: 12
Cross %: 5
Open play shots as a %: 67%
Average xG per open play shot: 0.11
Passes per defensive action: 10.14
Opponent passes per defensive action: 10.14
Formation: 4-2-3-1
I'll do this by a process of elimination.
Passes per game - Bayern Munich and Dortmund pass it way more so they are out. Freiburg, Frankfurt and Dusseldorf too for passing far less.
Long ball % rules out - Wolfsburg (hoof!) and Augsburg
Cross % rules out - Gladbach and Hertha
Average xG per open play shot: Hannover 96
Passes per defensive action: Mainz and Nuernberg
Opponent passes per defensive action: Werder Bremen
Teams left: Hoffenheim, RB Leipzig, Bayer Leverkusen, Schalke and VFB Stuttgart
Now we have a manageable list of outcomes similar to my Everton list I'll apply the formation filter:
Looking for similar to 4-2-3-1
Gone then are Hoffenhein, RB Leipzig, Schalke and VFB Stuttgart who don't line up like that.
Which leads me to the remaining team: Bayer Leverkusen.
Stats:
Short passes per 90: 417
Long ball %: 13%
Crosses: 4%
Open play shots as a %: 71%
Average xG per open play shot: 0.10
Passes per defensive action: 12.02
Opponents passes per defensive action: 10.72
Formation: 4-2-3-1
So there we go using my back of the envelope calculation style I see Everton are most similar to Bayer Leverkusen in playing style.
Does that pass the eye test? Let me know.
Comments
Post a Comment