United and xG (now that Ole is gone will things change?)

sullydnl

Ross Kemp's caf ID
Joined
Sep 13, 2012
Messages
34,063
Why is that misleading? The whole point of xG is quantify the quality of attempts at goal so 10 attempts with xG at 0.03 each are as likely to produce a goal as 1 with 0.3 which is why they are given those values to begin with.
Not sure about misleading but just looking at the basic xG figures in that game glossed over our inefficiency quite a bit, with that inefficiency being central to our underperformance.

10 attempts with an xG at 0.03 each is indeed as likely to result in a goal as 1 with 0.3, but with us only getting so many possessions per game it raises the question as to why so many of those possessions are ending in low xG chances. In other words, why are we shooting so often from bad positions instead of being able to work those possessions into higher xG chances? Either we're making bad decisions or we're unable to break through and resorting to pot shots.

In this case I think the more telling statistic is a more basic one: over the last two games we've had 55 shots. 24 off target, 21 blocked, 10 on target, 0 goals. That (like the xG to shot ratio) tells us we're not creating good enough chances.

Though tbf just having watched the game would tell us the same. It hasn't been hard to miss.
 
Last edited:

mav_9me

Full Member
Joined
Feb 9, 2009
Messages
12,400
Are you Yoda? Because this is getting surreal. You’re not making any sense.
Haha... I suspect he means Greenwood did not square the ball, so that hypothetical chance never happened.

I suspect you are saying the shot by Greenwood should be counted as a good chance right?
 

ti vu

Full Member
Joined
Jun 16, 2015
Messages
12,799
Are you Yoda? Because this is getting surreal. You’re not making any sense.
These expected stats, quantities action. Unless Greenwood make the pass, so Bruno can either score or miss, it expect nothing for the scenario (Greenwood's chance to pass).

Edit: NVM. Just saw other posts.
 

NZT-One

Full Member
Joined
Mar 8, 2021
Messages
2,342
Location
Berlin
Are you Yoda? Because this is getting surreal. You’re not making any sense.
He means that Greenwood didn't square the ball. Hence that tap in opportunity, likely with a great xG value didn't happen. Instead what happend was Greenwood shooting from a pretty good spot, resulting in a shot worth 0.10 xG.

To put it into perspective, the chance that resulted from Maguire's and De Geas moment of brilliance resulted in a 0.31xG chance. (To be honest I think it is mentionworthy how these two shots are so different in xG.)
 

Pogue Mahone

The caf's Camus.
Joined
Feb 22, 2006
Messages
133,350
Location
"like a man in silk pyjamas shooting pigeons
He means that Greenwood didn't square the ball. Hence that tap in opportunity, likely with a great xG value didn't happen. Instead what happend was Greenwood shooting from a pretty good spot, resulting in a shot worth 0.10 xG.

To put it into perspective, the chance that resulted from Maguire's and De Geas moment of brilliance resulted in a 0.31xG chance. (To be honest I think it is mentionworthy how these two shots are so different in xG.)
That seems completely wrong. Just to clarify, if player A receives a pass with a defender between him and goal then sits the defender down and has a free shot at goal from 6 yards out is that a lower xG than player B, who takes a shot from the exact same position without needing to sit a defender down first?
 

Classical Mechanic

Full Member
Joined
Aug 25, 2014
Messages
35,216
Location
xG Zombie Nation
Even so @Pogue Mahone, let’s say that Greenwood did square the ball resulting in a big chance, would you deem one big chance adequate against a side of Villa’s stature when we’re playing at home and with the talent we have available?
 

Pogue Mahone

The caf's Camus.
Joined
Feb 22, 2006
Messages
133,350
Location
"like a man in silk pyjamas shooting pigeons
Even so @Pogue Mahone, let’s say that Greenwood did square the ball resulting in a big chance, would you deem one big chance adequate against a side of Villa’s stature when we’re playing at home and with the talent we have available?
Of course not. I just couldn’t understand why were deemed to have no big chances from open play when Greenwood had such a great chance to score.
 

NZT-One

Full Member
Joined
Mar 8, 2021
Messages
2,342
Location
Berlin
That seems completely wrong. Just to clarify, if player A receives a pass with a defender between him and goal then sits the defender down and has a free shot at goal from 6 yards out is that a lower xG than player B, who takes a shot from the exact same position without needing to sit a defender down first?
Well the distance of both shots seems pretty much to be the same.
https://understat.com/match/16431

Why there is such a difference in valuation is a good question. Maybe because of the distance between ball and defender? Let me check.
 

TMDaines

Fun sponge.
Joined
Sep 1, 2014
Messages
13,906
It’s not great in isolation, but when you look at what those expected goals and assists mean in terms of expected points from actual match scenarios. and how that ranks us against the opposition, we’re better off now. Ole’s two whole seasons had us deserved top 4 each time. Even his half season did. Mourinho’s final whole season and half season did not and had us quite short.

Ole’s not smoking it, but he at least has us now in a place where I feel confident about getting top 4 and looking up.

Whether he can take us to place where we can contend for the title, I don’t know. I’m not sure what move I would make to further that. Changing the manager isn’t always for the better.
 

Chipper

Adulterer.
Joined
Oct 25, 2017
Messages
5,467
Maybe "misleading" isn't the most accurate word to describe it but showing that we have a higher xG than Villa makes it look like we a) were productive (quantity and quality) in attack and b) were unlucky not to win. All that isn't a good representation of the game I saw, apart from maybe being productive in quantity but you get that from the number of shots which is just as worthless because only 4 of 28 shots were on goal.
I get what you mean, and it is an interesting one. To our eyes we want to see some big chances created, restricting the opponents to none/few, dominating possession, territory and some entertainment thrown in in the form of dribbles, flicks, turns and nice skills etc. That's a classical good performance.

When it comes down to it xG doesn't really give a crap about any of that I suppose. It's just really trying to rate the likelihood of goals being scored by the teams and that's where we are today to a large degree. It's interesting how it throws up stuff like that sometimes. It could even make us think about what actually is a good performance, providing the numbers are correct of course.

I remember the Everton 3-3 game from last season where we did a lot of what you really want to see right but ended up about even on the real scoreboard and the xG as we made some really big mistakes in defence, gifting Everton some big chances when they didn't really do much all match and we played classically well for a lot of the game. That was a bit of a wild one, and I remember some saying it showed some flaws xG when to me it actually showed why it was a good result predictor based on what happened even though you couldn't but feel a bit robbed. We just made those massive errors, and they didn't really earn them through good play or endeavour of their own.

Today, I don't think we played well as a team, and I also think it's interesting when a game has penalties in them too. Michael Caley, one of the big proponents of xG on Twitter, and one of the first guys to run a model publicly lists penalties separately which I find interesting and often with merit. Think Fbref show non-penalty xG as a separate thing too if you delve into their numbers? Not all penalties are equal after all. Some are awarded when a defender handballs it on the line to prevent a certain goal following a good attacking move, some in the corner of the box when there wasn't much threat, some are controversial decisions by refs etc.

All penalties end up with the same chance of scoring from12 yards but we definitely didn't get there the same way. Who knows, one day in the future when they think they can assess things accurately we might even see teams penalised differently for the fouls they commit in the box based on the actual amount of danger at the time?

Anyway, so today we had that penalty to throw in there with the xG too, which came out of not much and if deducting it, which I'm not even sure we should do entirely we actually had lower xG than they did over the course of the match. So yes, there's that too.

Football of course is a low scoring game anyway with variance so the best way to play is to play in a way that you pummel the opponent so that even if things don't quite go your way that day then you still get a narrow win instead of maybe the comfortable or handsome win that you feel you deserved. We were absolutely nowhere near doing that today and cannot exactly feel robbed. If you narrowly "win" the xG or narrowly "lose" it you could very much win, lose or draw. At home to Villa with our resources we'd like to think we'd win both the xG and the actual scoreboard battle by some way.
 

NZT-One

Full Member
Joined
Mar 8, 2021
Messages
2,342
Location
Berlin
Well the distance of both shots seems pretty much to be the same.
https://understat.com/match/16431

Why there is such a difference in valuation is a good question. Maybe because of the distance between ball and defender? Let me check.
Actually I don't really see why there is such a difference between the shot from Greenwood (17th minute, xG 0.10) and the shot from Watkins (21th minute, xG 0.31. It might be because De Gea is caught on the wrong foot while Martinez was able to setup himself. But I am not sure about it, maybe more experienced users of that stat can answer?

@Borys @sullydnl
 

TheMagicFoolBus

Full Member
Joined
Nov 5, 2016
Messages
6,502
Location
Lisboa, Portugal
Supports
Chelsea
Actually I don't really see why there is such a difference between the shot from Greenwood (17th minute, xG 0.10) and the shot from Watkins (21th minute, xG 0.31. It might be because De Gea is caught on the wrong foot while Martinez was able to setup himself. But I am not sure about it, maybe more experienced users of that stat can answer?

@Borys @sullydnl
Understat's xG model is pretty unsophisticated - it certainly doesn't get into the nuances of goalkeeper positioning much less where his weight is shifted!

I'm guessing this is down to two factors - first, Watkins taking a right footed shot from the right channel means that he can viably shoot at the near post - generally it's harder to shoot near post with your right foot when you're left of the goal, as Greenwood was. The other key difference is that Greenwood is outside the width of the 6 yard box, whilst Watkins is inside it - from that wider position it's quite difficult to curl a far post shot on target, much less have a genuine scoring chance.
 

Eddy_JukeZ

Full Member
Joined
Aug 21, 2012
Messages
16,977
There's no adaptive xG model right?

If a player throughout the season is outperforming their xG, the xG reflecting that player isn't going to differ from another player who's under-performing their xG?
 

NZT-One

Full Member
Joined
Mar 8, 2021
Messages
2,342
Location
Berlin
I get what you mean, and it is an interesting one. To our eyes we want to see some big chances created, restricting the opponents to none/few, dominating possession, territory and some entertainment thrown in in the form of dribbles, flicks, turns and nice skills etc. That's a classical good performance.

When it comes down to it xG doesn't really give a crap about any of that I suppose. It's just really trying to rate the likelihood of goals being scored by the teams and that's where we are today to a large degree. It's interesting how it throws up stuff like that sometimes. It could even make us think about what actually is a good performance, providing the numbers are correct of course.

I remember the Everton 3-3 game from last season where we did a lot of what you really want to see right but ended up about even on the real scoreboard and the xG as we made some really big mistakes in defence, gifting Everton some big chances when they didn't really do much all match and we played classically well for a lot of the game. That was a bit of a wild one, and I remember some saying it showed some flaws xG when to me it actually showed why it was a good result predictor based on what happened even though you couldn't but feel a bit robbed. We just made those massive errors, and they didn't really earn them through good play or endeavour of their own.

Today, I don't think we played well as a team, and I also think it's interesting when a game has penalties in them too. Michael Caley, one of the big proponents of xG on Twitter, and one of the first guys to run a model publicly lists penalties separately which I find interesting and often with merit. Think Fbref show non-penalty xG as a separate thing too if you delve into their numbers? Not all penalties are equal after all. Some are awarded when a defender handballs it on the line to prevent a certain goal following a good attacking move, some in the corner of the box when there wasn't much threat, some are controversial decisions by refs etc.

All penalties end up with the same chance of scoring from12 yards but we definitely didn't get there the same way. Who knows, one day in the future when they think they can assess things accurately we might even see teams penalised differently for the fouls they commit in the box based on the actual amount of danger at the time?

Anyway, so today we had that penalty to throw in there with the xG too, which came out of not much and if deducting it, which I'm not even sure we should do entirely we actually had lower xG than they did over the course of the match. So yes, there's that too.

Football of course is a low scoring game anyway with variance so the best way to play is to play in a way that you pummel the opponent so that even if things don't quite go your way that day then you still get a narrow win instead of maybe the comfortable or handsome win that you feel you deserved. We were absolutely nowhere near doing that today and cannot exactly feel robbed. If you narrowly "win" the xG or narrowly "lose" it you could very much win, lose or draw. At home to Villa with our resources we'd like to think we'd win both the xG and the actual scoreboard battle by some way.
Yeah you are right. I agree with everything you said. I think, xG is a wonderful thing in the right hands. It is a reflection of reality but only of one aspect of it. And you have to be aware of that limitation - but that flaw is shared with mostly everything, the number of shots, the possession stat - in isolation they don't give you anything.

I guess you are right, as soon as somebody finds a way to amass even more data in football matches, I think, there is quite a potential to lose some of the magic of the game. It is a complex game but it doesn't have and infinite number of variables even though the number is certainly huge. As long as there are ways to measure data, to pool it together and you have enough processor power, I think, anything is possible in terms of usage. (actually I think, on a higher level there is even more stuff done in that regard, I don't think, it is coincidence that for example on understat the possibilities to export the data is somewhat limited, also that there seems to be no way to get a hand on distance covered stats indicates, that this is stuff some companies make serious money with.)

Even though I really like the stats, I know it only tells a part of the story. But over a number of games and across multiple teams, it gives you some sort of overview you would only get otherwise if you would watch all games. I see it as a way to display things, make them more "attainable".
 

NZT-One

Full Member
Joined
Mar 8, 2021
Messages
2,342
Location
Berlin
Understat's xG model is pretty unsophisticated - it certainly doesn't get into the nuances of goalkeeper positioning much less where his weight is shifted!

I'm guessing this is down to two factors - first, Watkins taking a right footed shot from the right channel means that he can viably shoot at the near post - generally it's harder to shoot near post with your right foot when you're left of the goal, as Greenwood was. The other key difference is that Greenwood is outside the width of the 6 yard box, whilst Watkins is inside it - from that wider position it's quite difficult to curl a far post shot on target, much less have a genuine scoring chance.
Ah, yes that sounds pretty plausible. You are right, even though the position of the shot are quite close to each other, Watkins is indeed a little more central. So I guess it is just "coincidence" that more than 30 shots from this position shot by a right footer went in while only 10 shots from Greenwoods position went in. But I also find it already pretty sophisticated, I mean, are we sure that understat takes into consideration where the shot is aimed at? And actually I didn't mean the factor of weight shift more the factor of being suprised but your idea is even better, imagine having boots to measure that as well :D

I certainly understand the question of Pogue because I also think it is mentionworthy that one shot seems to be 3 times as likely to go in than the other.

Would love to know all the factors that go into those models.

There's no adaptive xG model right?

If a player throughout the season is outperforming their xG, the xG reflecting that player isn't going to differ from another player who's under-performing their xG?
As far as I know, you are right, there is no adaptive model yet. It is also not taken into consideration, that there are different levels of players. A player like Kane is probably expected to score more goals from a certain position than for example Fred.

But I hope, somebody is working on that already. It would be invaluable to scouting departments I guess.
 
Last edited:

TheMagicFoolBus

Full Member
Joined
Nov 5, 2016
Messages
6,502
Location
Lisboa, Portugal
Supports
Chelsea
There's no adaptive xG model right?

If a player throughout the season is outperforming their xG, the xG reflecting that player isn't going to differ from another player who's under-performing their xG?
I know some teams are tracking this information on a proprietary basis to use in part to evaluate potential transfers but nothing like that is publicly available. There certainly are sample size concerns though and these data are probably only useful to gauge trends overall (and even then with a large pinch of salt).
 

Eddy_JukeZ

Full Member
Joined
Aug 21, 2012
Messages
16,977
I know some teams are tracking this information on a proprietary basis to use in part to evaluate potential transfers but nothing like that is publicly available. There certainly are sample size concerns though and these data are probably only useful to gauge trends overall (and even then with a large pinch of salt).
It is a shame this sport is behind others in publicly available stats.
 

TheMagicFoolBus

Full Member
Joined
Nov 5, 2016
Messages
6,502
Location
Lisboa, Portugal
Supports
Chelsea
Ah, yes that sounds pretty plausible. You are right, even though the position of the shot are quite close to each other, Watkins is indeed a little more central. So I guess it is just "coincidence" that more than 30 shots from this position shot by a right footer went in while only 10 shots from Greenwoods position went in. But I also find it already pretty sophisticated, I mean, are we sure that understat takes into consideration where the shot is aimed at? I certainly understand the question of Pogue because I also think it is mentionworthy that one shot seems to be 3 times as likely to go in than the other.

Would love to know all the factors that go into those models.
Sorry if I was unclear - understat definitely does not take this into consideration. What they do is they look at past outcomes of comparable situations to calculate the % chance of scoring - I was extrapolating from there to make some inferences as to what might be driving those percentages. Hope this makes sense!

If you are curious about better understanding where shots are aimed (especially as it relates to goalkeeper performance), this is best addressed by post-shot xG. Essentially what it does is it strips away all shots that are off-target and looks only at those that are on target, using the same factors. Effectively, you can then look to see which goalkeepers have conceded fewer than they'd be expected to given the shots they faced on target.
 

TheMagicFoolBus

Full Member
Joined
Nov 5, 2016
Messages
6,502
Location
Lisboa, Portugal
Supports
Chelsea
It is a shame this sport is behind others in publicly available stats.
Definitely. It's also a question of complexity - I love baseball and the analytics associated with it, but it's far easier to quantify given that it's a game of isolated and discreet events. Football fundamentally is the antithesis of that in many ways, so it stands to reason that there's a lot more secrecy.

https://www.theguardian.com/books/2013/may/24/numbers-game-everything-football-wrong

This was a book I quite enjoyed, despite the irritating subheading. It did a lot to help me better understand how and where data might be able to be applied to address some of the dogma within football.
 

Classical Mechanic

Full Member
Joined
Aug 25, 2014
Messages
35,216
Location
xG Zombie Nation
@Pogue Mahone

I've watched the Greenwood chance back. His shot isn't classed as a big chance because of the angle and the defensive pressure on him. If he squared the ball back to Bruno then that may well have been a big chance although there was a defender between Bruno and the goal so it would depend on when Greenwood had played the ball to Bruno and how the defenders reacted to the play developing like that. That's the problem with talking about things that didn't happen. You can also get into arguments about the players being too selfish and taking shots on in suboptimal positions whilst ignoring teammates as being a symptom of Ole's poor coaching and the style that we play.
 

Pogue Mahone

The caf's Camus.
Joined
Feb 22, 2006
Messages
133,350
Location
"like a man in silk pyjamas shooting pigeons
@Pogue Mahone

I've watched the Greenwood chance back. His shot isn't classed as a big chance because of the angle and the defensive pressure on him. If he squared the ball back to Bruno then that may well have been a big chance although there was a defender between Bruno and the goal so it would depend on when Greenwood had played the ball to Bruno and how the defenders reacted to the play developing like that. That's the problem with talking about things that didn't happen. You can also get into arguments about the players being too selfish and taking shots on in suboptimal positions whilst ignoring teammates as being a symptom of Ole's poor coaching and the style that we play.
It’s all academic and I don’t really care but the fact remains that - after beating the defender - Greenwood had a very good chance to score. If a player dribbles past several defenders and ends up on one one with the keeper it seems weird to say his team had no big chances.
 

Classical Mechanic

Full Member
Joined
Aug 25, 2014
Messages
35,216
Location
xG Zombie Nation
It’s all academic and I don’t really care but the fact remains that - after beating the defender - Greenwood had a very good chance to score. If a player dribbles past several defenders and ends up on one one with the keeper it seems weird to say his team had no big chances.
So what you're saying is that we rely on individual brilliance to create chances. That becomes unsustainable over a longer period of time and is a problem in my opinion. The argument here is that we're not a well coached team that creates big chances through team play.
 

TheRedDevil'sAdvocate

Full Member
Joined
Dec 7, 2013
Messages
3,632
Location
The rainbow's end
From his position and the way he places his body and takes a touch to prepare for the shot, Greenwood can only aim at one target. The AV keeper is in a good position on his line and he's anticipating exactly that shot. The Villa chance was better: Better angle, one touch/first time finishing available, and the defence nowhere to be found. De Gea reacts well but again, all Watkins had to do was to not hesitate and place the shot just beyond De Gea's reach (which shouldn't be that difficult since he had a lot of room to aim at).

To put it otherwise, Watkins would still need a hell of a shot to finish Greenwood's chance, but Greenwood, as much better quality forward, would have probably scored from Watkins' opportunity.

Mason's reaction was the kind of stuff that would result in one of Fergie's famous hairdryers during the half-time break.
 

Pogue Mahone

The caf's Camus.
Joined
Feb 22, 2006
Messages
133,350
Location
"like a man in silk pyjamas shooting pigeons
So what you're saying is that we rely on individual brilliance to create chances. That becomes unsustainable over a longer period of time and is a problem in my opinion. The argument here is that we're not a well coached team that creates big chances through team play.
I know all of this and agree with it. I’m just curious about the semantics of all these stats.

I also think this game is a good example of the failings of xG. Watching the game it seemed obvious that Villa had the better, clearer, bigger chances to score. Yet United “won” the xG.
 

Classical Mechanic

Full Member
Joined
Aug 25, 2014
Messages
35,216
Location
xG Zombie Nation
I know all of this and agree with it. I’m just curious about the semantics of all these stats.

I also think this game is a good example of the failings of xG. Watching the game it seemed obvious that Villa had the better, clearer, bigger chances to score. Yet United “won” the xG.
xG still has plenty of blind spots of course but will get more and more accurate as time goes on. We did win the xG yes but if you look at it closer that’s because we had a penalty a lots of lower quality chances. You can also look at it as a classic game of low block and counter attack by Villa where they will have bigger chances on the counter because we commit more men forward trying break them down. Villa we’re a bit more live in the game than that though and the point remains that we’re very poor at breaking compact teams down in contrast to City or Liverpool. I think we’ll do better than most people think in this tough run coming up because we’ll get to be the counter attacking side more often which suits us.
 

Anustart89

Full Member
Joined
Jun 6, 2013
Messages
15,912
I know all of this and agree with it. I’m just curious about the semantics of all these stats.

I also think this game is a good example of the failings of xG. Watching the game it seemed obvious that Villa had the better, clearer, bigger chances to score. Yet United “won” the xG.
I think it just shows that all data need context and add-on data to describe them. For example, if you were to exclude the penalty (or at least look at it separately) and then add non-penalty xG divided by the amount of xG contributions those two data points would be enough to describe the game as you saw it.
 

Zen86

Full Member
Joined
Jul 1, 2007
Messages
13,871
Location
Sunny Manc
What I want to know is can we hire a coach to improve our xG?? Like we did for set pieces?? It seems like that’s our big weakness right now.
 

JustAGuest

Full Member
Joined
Jun 19, 2013
Messages
742
What I want to know is can we hire a coach to improve our xG?? Like we did for set pieces?? It seems like that’s our big weakness right now.
Set pieces seems the only obvious one. Other than that I'm not so sure - it would probably be easier to bring in a new manager.
 

Rajma

Full Member
Joined
Sep 8, 2012
Messages
8,571
Location
Lithuania
xG still has plenty of blind spots of course but will get more and more accurate as time goes on. We did win the xG yes but if you look at it closer that’s because we had a penalty a lots of lower quality chances. You can also look at it as a classic game of low block and counter attack by Villa where they will have bigger chances on the counter because we commit more men forward trying break them down. Villa we’re a bit more live in the game than that though and the point remains that we’re very poor at breaking compact teams down in contrast to City or Liverpool. I think we’ll do better than most people think in this tough run coming up because we’ll get to be the counter attacking side more often which suits us.
Don’t think it does suit us that much now that we have Ronaldo and Rashford is not playing. I think Young Boys game have shown that Ronaldo although still relatively fast without the ball can’t outpace defenders with the football at his feet which is a major issue if you’re looking to be on the backfoot all game. I think with Ronaldo in the team to get the best out of him we need to asserting our dominance on the teams.
 

JustAGuest

Full Member
Joined
Jun 19, 2013
Messages
742
He is being facetious. He thinks xG is nonsense.
Okay, haven't read much else of this thread. :lol:

Creating more xG from set pieces is really just creating more or better quality chances from them. I don't think it's nonsense in that context.
 

Siorac

Full Member
Joined
Sep 1, 2010
Messages
23,789
Just out of curiosity, I looked at some of the 'expected' stats of the last five seasons. Here are the expected goals, expected goals against, and expected points metrics - sorted based on expected points, NOT by chronological order. Can you tell which season is which without looking it up?

xG: 66.19, xGA: 38.06, xPts: 70.99

xG: 57.57, xGA: 31.62, xPts: 70.89

xG: 63.17, xGA: 41.92, xPts: 65.64

xG: 59.04, xGA: 43.54, xPts: 62.33

xG: 68.62, xGA: 52.30, xPts: 61.86

(Just for comparison: the expected points tallies of Liverpool and Chelsea, respectively, in the last five seasons:

- 69.83 and 75.74
- 79.38 and 68.46
- 83.45 and 71.45
- 74.28 and 73.49
- 68.80 and 77.41

I'm not going to post City's xPts stats. They are way too depressing)