Introducing my new metric: "unlucky son"
I floated the idea yesterday that we should be able to assign a positive value to players who do something good that doesn't *quite* come off.
I'm not looking to reward a good idea that has been poorly executed.
Nor a bad idea, turned good by massive amounts of luck, we already reward that.
I'm talking about a really good idea, well executed but missing by a tiny amount.
The game that comes to mind for me is Everton 0 Bolton 0 on 29th December, 2002.
I'm pretty sure everyone reading this will instantly be able to recall that famous day.....no?
It illustrated the 3 ages of analytics.
Goals and assists era
Rooney finished the game with 0 goals and 0 assists.
xG and xA, era
I'd love someone to go back and examine the footage for this but I'm pretty sure he would have finished with over 1xG and 1xA and about 10 dribbles completed
The "unlucky son" era
I cannot stress how good and plain unlucky Rooney was in this game. This was a good Bolton team with Campos, Okocha, Nolan etc. Rooney ran them ragged to the point where Campo told him he was as good as Ronaldo, Rivaldo and Del Piero after the game.
Obviously a lot of the good things he did were picked up in xG etc ratings but I think there needs to be some way of recording brilliant things that don't quite come off. One part I particularly remember is an attempted one-two with Kevin Campbell where Rooney beat about 3 players sprinted away and tried to square it for Campbell to tap into an empty net. The pass was cleared by tip of the boot of a sliding defender.
It deserves more than being classed as a misplaced pass or a failed dribble.
And this is where my "unlucky son" button comes in.
@Ryanbahia suggested it could be triggered where a player is applauded by a teammate! Alternatively AI could be trained to recongise an appreciative "ooooohhhhhh" from a crowd....
However I can imagine two ways of this operating.
At the match tagging/coding stage an additional category of "unlucky" could be added for positive attempts that don't come off. For example we all love that Benjamin Pavard line breaking pass from centre back. However I presume for every one that has come off there have been many that *just* miss and are recorded as a failed long pass.
Now it was rightly pointed out that those who attempt those type of passes will eventually appear in the data as they will eventually come off. However, in the data scouting arms race where we all end up agreeing who looks good in the data when the sample size is big enough, it might help accelerate the process.
If I am filtering defenders by their long passes and going off completion I'm going to miss that new Pavard who is playing with less aware strikers. Yes, real life scouts will see him eventually and note it but it is a race for talent.
An alternative to coding it in the start, for all I know adding an extra coding action may massively increase cost, is for associated clips on Wyscout of failed actions to be viewed by a secondary team and classed as unlucky (good intent, unfortunate), neutral (the default), terrible (what was he thinking?). A player could then have an unlucky son metric that shows whether the actions that ended in failure were a negative or positive.
So in my Rooney example rather than just having a "mistakes bucket" on Wyscout full of mistakes you would also have an "unlucky son bucket" filled with the very near misses. My Statsbomb radar would show a 99th percentile for good actions that were just unfortunate.
In a few years (or maybe now, who knows!) you may be able to train some sort of AI working on tracking data to recognise all the possible options a player has to make a pass or shoot and assign a value to their choice. What looks like dallying on the ball, and losing possession, may have been the best option when there was a chance of playing in a teammate for a 0.5xG chance rather than just recycling possession.
Any thoughts on the value, or otherwise (love you @footballfactman), of this would be appreciated.
I'm not looking to reward a good idea that has been poorly executed.
Nor a bad idea, turned good by massive amounts of luck, we already reward that.
I'm talking about a really good idea, well executed but missing by a tiny amount.
The game that comes to mind for me is Everton 0 Bolton 0 on 29th December, 2002.
I'm pretty sure everyone reading this will instantly be able to recall that famous day.....no?
It illustrated the 3 ages of analytics.
Goals and assists era
Rooney finished the game with 0 goals and 0 assists.
xG and xA, era
I'd love someone to go back and examine the footage for this but I'm pretty sure he would have finished with over 1xG and 1xA and about 10 dribbles completed
The "unlucky son" era
I cannot stress how good and plain unlucky Rooney was in this game. This was a good Bolton team with Campos, Okocha, Nolan etc. Rooney ran them ragged to the point where Campo told him he was as good as Ronaldo, Rivaldo and Del Piero after the game.
 |
| I ruined stats |
Obviously a lot of the good things he did were picked up in xG etc ratings but I think there needs to be some way of recording brilliant things that don't quite come off. One part I particularly remember is an attempted one-two with Kevin Campbell where Rooney beat about 3 players sprinted away and tried to square it for Campbell to tap into an empty net. The pass was cleared by tip of the boot of a sliding defender.
It deserves more than being classed as a misplaced pass or a failed dribble.
And this is where my "unlucky son" button comes in.
@Ryanbahia suggested it could be triggered where a player is applauded by a teammate! Alternatively AI could be trained to recongise an appreciative "ooooohhhhhh" from a crowd....
However I can imagine two ways of this operating.
At the match tagging/coding stage an additional category of "unlucky" could be added for positive attempts that don't come off. For example we all love that Benjamin Pavard line breaking pass from centre back. However I presume for every one that has come off there have been many that *just* miss and are recorded as a failed long pass.
Now it was rightly pointed out that those who attempt those type of passes will eventually appear in the data as they will eventually come off. However, in the data scouting arms race where we all end up agreeing who looks good in the data when the sample size is big enough, it might help accelerate the process.
If I am filtering defenders by their long passes and going off completion I'm going to miss that new Pavard who is playing with less aware strikers. Yes, real life scouts will see him eventually and note it but it is a race for talent.
An alternative to coding it in the start, for all I know adding an extra coding action may massively increase cost, is for associated clips on Wyscout of failed actions to be viewed by a secondary team and classed as unlucky (good intent, unfortunate), neutral (the default), terrible (what was he thinking?). A player could then have an unlucky son metric that shows whether the actions that ended in failure were a negative or positive.
So in my Rooney example rather than just having a "mistakes bucket" on Wyscout full of mistakes you would also have an "unlucky son bucket" filled with the very near misses. My Statsbomb radar would show a 99th percentile for good actions that were just unfortunate.
In a few years (or maybe now, who knows!) you may be able to train some sort of AI working on tracking data to recognise all the possible options a player has to make a pass or shoot and assign a value to their choice. What looks like dallying on the ball, and losing possession, may have been the best option when there was a chance of playing in a teammate for a 0.5xG chance rather than just recycling possession.
Any thoughts on the value, or otherwise (love you @footballfactman), of this would be appreciated.
I'm not sure about this. It will only affect players that try these things rarely. For others it should even out and show up in GF and xG.
ReplyDeleteBut that AI thing obviously would be great. Not sure if it really is doable but judging decision-making is hard while it's one of the most important skills a player has.