[Football] Who actually produces and publishes xG data?

[Bry Nylon]

I get the concept of xG but I don't understand who actually publishes this data. Is it one source? It is multiple sources?

Is it a 14 year old kid with a laptop in Manilla who's never seen a PL game in his life, watching matches on a dodgy stream and feeding his view of how easy / difficult a chance is into a database, based on playing FIFA? Or is xG the output of a panel of former pros, forming a balanced judgment based on real life experience of playing at the top level? Or something in between?

I think xG has a role but who actually curates the data and forms the judgements? It seems to becoming accepted as a universal truth and valid measurement to judge teams and players and the numbers go unquestioned; but I've no idea who is actually responsible for producing, publishing and managing the data.

 

[dazzer6666]

I get the concept of xG but I don't understand who actually publishes this data. Is it one source? It is multiple sources?

Is it a 14 year old kid with a laptop in Manilla who's never seen a PL game in his life, watching matches on a dodgy stream and feeding his view of how easy / difficult a chance is into a database, based on playing FIFA? Or is xG the output of a panel of former pros, forming a balanced judgment based on real life experience of playing at the top level? Or something in between?

I think xG has a role but who actually curates the data and forms the judgements? It seems to becoming accepted as a valid measurement to judge teams and players, but I've no idea who is actually responsible for producing, publishing and managing the data.

Multiple, and it’s very subjective.

 

[Weststander]

The Gustav Potter Institute based in Ostersund.

 

[dazzer6666]

The Gustav Potter Institute based in Ostersund.

Emeritus Professor [MENTION=38333]Swansman[/MENTION] head of faculty

 

[Swansman]

Different data companies, OPTA is probably the most well-known and widely used though clubs got even more sophisticated (and expensive) tools. OPTA used data from hundreds of thousands of shots in from different positions and in different situations and based on this measures the expected/most frequent outcome of shots. Of course its far from perfect and more variables will need to be added before its really accurate but its already affecting football a lot, like killing off long shots.

 

[AlbionInUs]

these are the factors that form the basis of expected goals (xG):
Assist type - long ball, through ball etc.
Was it a header?
Was the ball struck with the foot?
Was it a big chance?
Angle of the shot
Distance of the shot
Was it a one-on-one?
Did it occur in open play or a from a set-piece?

xG is not subjective, it’s objective. Based on the history of similar shorts, said shot is assigned an xG value. For example, a shot from 30 yards out is compared with other shots from 30 yards out, and how often that shot ended in a goal.

 

[dsr-burnley]

these are the factors that form the basis of expected goals (xG):
Assist type - long ball, through ball etc.
Was it a header?
Was the ball struck with the foot?
Was it a big chance?
Angle of the shot
Distance of the shot
Was it a one-on-one?
Did it occur in open play or a from a set-piece?

xG is not subjective, it’s objective. Based on the history of similar shorts, said shot is assigned an xG value. For example, a shot from 30 yards out is compared with other shots from 30 yards out, and how often that shot ended in a goal.

xG also has one very major flaw. It's a snapshot that takes no account of the overall situation. Example - Burnley scored against Southampton last year, direct from a corner. Westwood took it, it bent in at the near post, the keeper flapped it clear, Ben Mee banged it in from less than a yard. xG gave it 0.01 because the ball crossed the line before the keeper knocked it back, and direct from a corner has very low xG. If the keeper had stopped it crossing the line or goalline technology hadn't worked, then Westwood's 0.01 xG wouldn't have counted but Mee would have had about 0.9 for his. It's pure nonsense to say that if the goalkeeper had made a save, Burnley's expected goals would have been higher than if he had let it in.

Same from a free kick. The forward has a shot with xG 0.1 and it goes in. The attacking side has no other attacks and their xG for the game is 0.1. But if the keeper makes a brilliant save and tips it onto the post, only for it to rebound to the middle of goal and the onrushing forward taps it in from half an inch, the xG goes up to 1.09 (0.1 + 0.99). That keeper's save has cost his side an expected goal; if he had let it in, his side might have conceded a goal less!

 

[Bry Nylon]

xG also has one very major flaw. It's a snapshot that takes no account of the overall situation. Example - Burnley scored against Southampton last year, direct from a corner. Westwood took it, it bent in at the near post, the keeper flapped it clear, Ben Mee banged it in from less than a yard. xG gave it 0.01 because the ball crossed the line before the keeper knocked it back, and direct from a corner has very low xG. If the keeper had stopped it crossing the line or goalline technology hadn't worked, then Westwood's 0.01 xG wouldn't have counted but Mee would have had about 0.9 for his. It's pure nonsense to say that if the goalkeeper had made a save, Burnley's expected goals would have been higher than if he had let it in.

Same from a free kick. The forward has a shot with xG 0.1 and it goes in. The attacking side has no other attacks and their xG for the game is 0.1. But if the keeper makes a brilliant save and tips it onto the post, only for it to rebound to the middle of goal and the onrushing forward taps it in from half an inch, the xG goes up to 1.09 (0.1 + 0.99). That keeper's save has cost his side an expected goal; if he had let it in, his side might have conceded a goal less!

Interesting examples. Of course, statistics will only ever reveal patterns or predictions based on historic events and come with no guarantee of future outcomes; but it does seem some are choosing to see them as the beginning and the end, in a sport where there that just isn't possible.

I was reading up about xG earlier, in an attempt to answer my own question, before I thought it would be easier and possibly more insightful to simply post it on here. One sentence that caught my eye was "the thing that makes football great is its fundamentally chaotic nature" which I guess is underlined by your examples above.

 

[Bry Nylon]

these are the factors that form the basis of expected goals (xG):
Assist type - long ball, through ball etc.
Was it a header?
Was the ball struck with the foot?
Was it a big chance?
Angle of the shot
Distance of the shot
Was it a one-on-one?
Did it occur in open play or a from a set-piece?

xG is not subjective, it’s objective. Based on the history of similar shorts, said shot is assigned an xG value. For example, a shot from 30 yards out is compared with other shots from 30 yards out, and how often that shot ended in a goal.

It is completely subjective.

Take one of your examples above: "was it a header?"

Okay, so:

Was it a free header?
Was it a challenged header?
Is the first team goalie in goal or the reserve / No:2 keeper?
Was a header from a player with a history of scoring headed goals or one with a head like a 50p piece?
Did a defender shout "leave it!" or "mine!" just before he headed it?
Was it a wet, greasy ball?
Was it a night game with the ball coming out of the floodlights?
Had the player just played a Europa game 3 days before and it's in the last minute of added time and he's dead on his feet?
Has the player just come back from injury?
Has he only been on the pitch for 3 minutes and is still getting up to speed with the game?
Did he have a massive row with his Doris before the game?

There are loads of factors that mean it is subjective and not objective because it is a dynamic game played by humans.

I still go with the concept of xG and understand that it says "for a header from this position, the likelihood of a goal is X based on accumlated data of headers from the same position" but I still stuggle with not seeing each event as unique, in its own beautiful chaotic way.

I'm conflicted

 

[Swansman]

It is completely subjective.

Take one of your examples above: "was it a header?"

Okay, so:

Was it a free header?
Was it a challenged header?
Is the first team goalie in goal or the reserve / No:2 keeper?
Was a header from a player with a history of scoring headed goals or one with a head like a 50p piece?
Did a defender shout "leave it!" or "mine!" just before he headed it?
Was it a wet, greasy ball?
Was it a night game with the ball coming out of the floodlights?
Had the player just played a Europa game 3 days before and it's in the last minute of added time and he's dead on his feet?
Has the player just come back from injury?
Has he only been on the pitch for 3 minutes and is still getting up to speed with the game?
Did he have a massive row with his Doris before the game?

There are loads of factors that mean it is subjective and not objective because it is a dynamic game played by humans.

I still go with the concept of xG and understand that it says "for a header from this position, the likelihood of a goal is X based on accumlated data of headers from the same position" but I still stuggle with not seeing each event as unique, in its own beautiful chaotic way.

I'm conflicted

True.

But I give it tops ten years before they know worryingly much about what the player thought, felt and interacted about/with his Doris before the game.

 

[Perfidious Albion]

Are we not in danger of becoming more concerned with computer based predictions than with the actual happenings on green grass by human blokes ?

 

[McTavish]

It is completely subjective.

Take one of your examples above: "was it a header?"

Okay, so:

Was it a free header?
Was it a challenged header?
Is the first team goalie in goal or the reserve / No:2 keeper?
Was a header from a player with a history of scoring headed goals or one with a head like a 50p piece?
Did a defender shout "leave it!" or "mine!" just before he headed it?
Was it a wet, greasy ball?
Was it a night game with the ball coming out of the floodlights?
Had the player just played a Europa game 3 days before and it's in the last minute of added time and he's dead on his feet?
Has the player just come back from injury?
Has he only been on the pitch for 3 minutes and is still getting up to speed with the game?
Did he have a massive row with his Doris before the game?

There are loads of factors that mean it is subjective and not objective because it is a dynamic game played by humans.

I still go with the concept of xG and understand that it says "for a header from this position, the likelihood of a goal is X based on accumlated data of headers from the same position" but I still stuggle with not seeing each event as unique, in its own beautiful chaotic way.

I'm conflicted

The factors in bold are completely irrelevant (and a couple of others probably are as well). The skill or otherwise of the players is irrelevant to xG and to a large extent is what xG is trying to measure. The Xg for a chance that falls to a bad or tired player is exactly the same as that which falls to a good or sharp player; the bad player will not score more often and so actual goals will be below the xG, cf. BHA.

 

[happypig]

When they start giving out points for xG, then I will take notice, until then it's bollocks.
Take a scenario where players break into the opposition half and it's 2 on 1 ; would xG give the same likelihood of scoring whether the players are Salah/Firmino, Son/Kane or Burn/Maupay ?

 

[McTavish]

When they start giving out points for xG, then I will take notice, until then it's bollocks.
Take a scenario where players break into the opposition half and it's 2 on 1 ; would xG give the same likelihood of scoring whether the players are Salah/Firmino, Son/Kane or Burn/Maupay ?

Yes, that's the whole point.

A good example is penalties.

A penalty in the Premier League has a standard xG because, on average, a certain number of penalties are scored. Some players are better at scoring from penalties, some goalkeepers are better at saving them but the xG is exactly the same. It doesn't matter if it is a vital game deciding penalty in the last minute or the fourth goal in a 6-0 rout; the xG is the same.

 

[Audax]

xG also has one very major flaw. It's a snapshot that takes no account of the overall situation. Example - Burnley scored against Southampton last year, direct from a corner. Westwood took it, it bent in at the near post, the keeper flapped it clear, Ben Mee banged it in from less than a yard. xG gave it 0.01 because the ball crossed the line before the keeper knocked it back, and direct from a corner has very low xG. If the keeper had stopped it crossing the line or goalline technology hadn't worked, then Westwood's 0.01 xG wouldn't have counted but Mee would have had about 0.9 for his. It's pure nonsense to say that if the goalkeeper had made a save, Burnley's expected goals would have been higher than if he had let it in.

Same from a free kick. The forward has a shot with xG 0.1 and it goes in. The attacking side has no other attacks and their xG for the game is 0.1. But if the keeper makes a brilliant save and tips it onto the post, only for it to rebound to the middle of goal and the onrushing forward taps it in from half an inch, the xG goes up to 1.09 (0.1 + 0.99). That keeper's save has cost his side an expected goal; if he had let it in, his side might have conceded a goal less!

I don't see any problem with the examples you provide here.

I think what many may be missing with their assumptions about xG is exactly how much goes into the analysis. I'm guessing the big providers keep their calculations secret (the likes of OPTA), but it's reasonable to believe that they are looking at a lot more than just the obvious (range, angle, header or not, etc). That much is clear from the xG map of our match vs Crystal Palace. They're clearly taking into account the position of all players (is the shot going through a wall of defenders or not), and I would suspect they are also taking into account the lead up to the shot itself given this is something they also analyse.

I would expect that these companies (if they haven't already) to also be working up a method for rating alternative options that players had. So for example, if you have a break away counter where you have a 2 vs 1 on the keeper and Attacker 1 takes the shot with a low xG, they may also attempt to analyse what the xG would have been had Attacker 1 instead passed to Attacker 2 for them to take the shot.

 

[zefarelly]

its already affecting football a lot, like killing off long shots.

along with VAR, slowly sapping the excitement and life out of football like as spider sucking its vitim dry.

 

[dsr-burnley]

I don't see any problem with the examples you provide here.

I think the problem is that xG purports to represent each side's chance of victory, and it doesn't. In a hypothetical match played in the centre circle where each side has only one attack, a free kick with an xG of 0.1 Both sides score from the free kick. And yet the xG doesn't say a 0.1-0.1 so a draw is a fair result; the xG says Burnley (the team with the better goalkeeper ) had an xG of 0.1 while Brighton (whose goalkeeper didn't make the fabulous save) had an xG of 1.09. Is it really fair to say that Pope's hypothetical save made a big improvement to Brighton's chances of winning?

 

[A1X]

Is this the prelude to the fact we're so rubbish under the terms of xG meaning we're going to form an NSC lynch-mob to find out who publishes the xG stuff and physically stop them? Because I am here for this.

 

[Swansman]

along with VAR, slowly sapping the excitement and life out of football like as spider sucking its vitim dry.

In some ways it is / is going to make the game more streamlined so I agree to some extent. But still pretty far from reaching that stage, teams still play with different styles and unless you get into it, its pretty avoidable to get bothered by it, unlike VAR.

 

[RossyG]

Football is a mixture of skills quickly applied and the butterfly effect.

Trying to apply mathematical formulae to it is pretty pointless.