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

Stocar

Full Member
Joined
Dec 30, 2011
Messages
699
Thats debatable. Playing good football isn't about sustainability, its about consistently playing good football. If a "reactive" teams outperforms their xG due to having good forwards, whereas a possession based team creates better clear cut chances, the "high quality football" still has to take place somewhere.

In a possession based team, that usually happens around elsewhere on the pitch where they open teams up with their quick movement and passing. However we've seen times where they (possession based teams) aren't quick enough in their passing, they end up not creating as much clear cut chances.

Same for the "reactive" teams. The good plays have to happen somewhere on the pitch to open teams up and score goals. If its your forwards mostly doing that, you'll outperform your xG. If its your midfielders doing that with their movement, you'll have a higher xG and might not need to outperform it. But the dangers for "non-sustainability" are the same IMO in both approaches.
You're basically saying that xG doesn't do good enough job in estimating the quality of chances, because it doesn't take certain factors into account. I agree that the system isn't perfect, but not sure what your argument here is. I guess you want to say that it possibly wrongly estimates the quality of chances because it apparently doesn't distinguish between chances "created from midfield" and those "created by forwards", also by not taking into account the players's quality. Maybe there are possible fine tweaks in that sense, I'm not that familiar with the whole thing. But it wouldn't make a huge difference as there would still be huge over/underachievers, and their change of fortune would still be easy to predict. Same rules are applied for every team, and while not perfect, it's a rather consistent metric.
 

Andrew~

Joined
Feb 21, 2008
Messages
6,190
After having a little think about it, xG seems like a good stat. All it is saying (judging by the way it is computed) is that it puts greater weight on the quality of chances. So looked at in a team sense: a team that creates a higher amount of good quality chances is more likely to score goals, ipso facto, more likely to win games.

How does this relate to United? It probably means, on average we are not creating as many high quality chances, but still winning matches because (i) we have better strikers than most teams, (ii) defensively we're pretty solid and (iii) David De Gea.

I don't think that means we'll regress to the mean because that involves the strikers going completely off form all at the same time, the defense turning off and De Gea packing it in. So far, Ole has set the team up as a counter-pressing team so most of our goals in big games have come from forcing errors on the opposition, and I don't think xG reflects that very well because it is weighted towards creating a higher volume of good quality chances.
 

cyberman

Full Member
Joined
May 26, 2010
Messages
37,331
We had a higher Xg than PSG last night (according to ESPN)
PSG were, by far, the better side last night?
 

Rish Sawhney

New Member
Newbie
Joined
Mar 15, 2015
Messages
619
Location
State College
OK, how often do teams consistently outperform the xG over 2-3 seasons? Examples in England (only top teams)?

What is this xG tool good for, according to you? Does it show something interesting about United since Ole?
This is my last post for today so here goes:

It’s hard to tell because no team has actually defended their title since we last did it. But you can also interpret the fact differently. You can say that the fact they they couldn’t outperform it the next year means they were playing above themselves and just couldn’t sustain it. Or it could be they had a few very good players in key positions that either left or couldn’t reproduce that form. I’m more tempted with the latter.

I think xG is very useful for analyzing the finishing abilities of your strikers. For example if you’re consistently underperforming your xG stays and you’re losing, you need better finishers.

Similarly if you’re outperforming it, it means your forwards are in good form and are converting harder chances. You might say that they can’t do that consistently over a long period of time, but then they’ll just move on and become less prolific.

Now if a team is regularly creating easy chances for their strikers so they can pretty much be level with their xG’s you could say those chance creating stats are unsustainable (I’m not saying that cos if your system or players and good at doing that then more power to them).

My point is that xG stats measure finishing performance, but can be filled by very good midfield or complete team performances which need not be any more sustainable than outperforming the xG.

I dunno if that made sense.

Regarding @chromepaxos post below (sorry I don’t have any more posts left) what team has ever maintained the form we’re in long term? If we maintain that form that we’ll win everything in sight.

The point is that people have been saying it would regress all the way to the mean, hence the twitter guy saying Ole shouldn’t take the job cos he’s over performing. If our regression is only going to be slight, then why is this a factor in the first place? I’d assume all teams when they have a purple patch outperform the expected stats.


@chromepaxos, I'm gonna respond to you here:

"Who has been saying it would "regress all the way"? I haven't seen one.

I think the point of the Tweet in the OP was that we are so far above xG - more than even Barca or City can sustain - that luck is clearly playing a significant part, maybe even a major part. That seems like a reasonable point for a statistician to make."


When something counter to your model happens the reaction shouldn't be - its must be due to luck. The models themselves are partial and imperfect reflections of reality. So when you observe some odd results in your models, before you can say its luck, you have to model what such luck would look like - like a statistician would - instead of getting defensive about your existing models.

This point has been brought up before, but Manchester United has always been about out performing the xG. How many of the 13 PL titles SAF won were we the best team using xG metrics? 2012/2013 certainly I'm sure we were routinely beat on xG but ended up pulling a 2-3 win out. The idea that a regression must happen is odd to me. Sure it might happen, or it might not. The fact that it has tended to happen since 2014 (thats the earliest I could find on understat) could just be a corresponding to the fact that no team has successfully defended their title in that time.
 
Last edited:
  • Like
Reactions: KM

chromepaxos

Full Member
Joined
Jul 29, 2015
Messages
192
My only opposition is to the "regression to the mean", "its unsustainable to outperform the xG" idea. If you hold on to that (I dunno as I responded to someone else and you responded to me), then you're wrong. But you have a right to be as wrong as you want in life. And most people are wrong most of the time about most things. So its not something to be pissed about that much.
You seem to be labouring under the misapprehension that regression to the mean must result in going all the way to the mean. Referring back to the stats quoted in the OP where we have scored ~30% more goals than xG since Olly came in, and conceded 38% fewer. Since your argument is that xG has no predictive power and that regression to the mean is nonsense, is it your contention that we will sustain these differences through to the end of the season?

Wanna bet?
 

Pexbo

Winner of the 'I'm not reading that' medal.
Joined
Jun 2, 2009
Messages
68,413
Location
Brizzle
Supports
Big Days
xG is bollocks.

It’s like the opposite shite logic that MP used when he claimed that every school pupil should achieve an above average grade in their Maths GCSE.

xG doesn’t take into account who the player is.

Messi is currently on 1goal per game with an xG of 0.74. So he’s overachieving but over 33%. No shit, he’s Messi.

Look at the quality of our attacking players.

We’re achieving above a generic xG stat. Shock feckin horror.
 

Revan

Assumptionman
Joined
Dec 19, 2011
Messages
49,419
Location
London
xG is bollocks.

It’s like the opposite shite logic that MP used when he claimed that every school pupil should achieve an above average grade in their Maths GCSE.

xG doesn’t take into account who the player is.

Messi is currently on 1goal per game with an xG of 0.74. So he’s overachieving but over 33%. No shit, he’s Messi.

Look at the quality of our attacking players.

We’re achieving above a generic xG stat. Shock feckin horror.
I am disappointed in you Pexbo.
 

Ajaxsuarez

Full Member
Joined
Aug 8, 2009
Messages
928
Supports
Ajax
xG is bollocks.

It’s like the opposite shite logic that MP used when he claimed that every school pupil should achieve an above average grade in their Maths GCSE.

xG doesn’t take into account who the player is.

Messi is currently on 1goal per game with an xG of 0.74. So he’s overachieving but over 33%. No shit, he’s Messi.

Look at the quality of our attacking players.

We’re achieving above a generic xG stat. Shock feckin horror.
eh there's a lot of flaws to xG and how it's measured but what you're saying isn't really it.

This is a fairly old article http://11tegen11.net/2014/02/15/how-to-scout-a-striker/

but it has some interesting observations



The horizontal axis shows ExpG NP per 90 for the first season, and the vertical axis shows the same for the next season.
Excellent! It turns out that players with a high ExpG per 90 in one season, are also the players with a high ExpG per 90 in the next season. This is not too surprising, as several factors influencing ExpG per 90 will remain constant over time. Strikers will still be playing as strikers, and most players playing for top team will still be playing for top teams. More work needed here, but we’ll leave that for another post, as there is a far more interesting graph coming up.
The next graph shows the repeatability of non penalty goals above replacement (NPGAR). This represents the conversion of goal scoring chances into actual goals.


It turns out that if you correct for the quality of goal scoring attempts, there is absolutely no connection between conversion in one season and the next. A high or low NPGAR in one season has zero relation with NPGAR in the next season.

Messi is the dot in the lower right hand corner
 

Wibble

In Gadus Speramus
Staff
Joined
Jun 15, 2000
Messages
88,602
Location
Centreback
The over use of these sort of stats is as irritating and pointless as people over thinking and over-debating formations.

Just stop it.
 

kundalini

Full Member
Joined
Aug 26, 2004
Messages
5,750
Expected goals is about the position of the shots a team takes.

It doesn't take into account our passing tempo and defensive shape:

  • A quick move is more dangerous than a slow move, even if they end with a shot from the same place on the pitch.

  • A compact defense will gladly give up a shot if they know a block can be easily made.

It's not just about the raw numbers. We're an outlier in terms of the way we play football. The pace we have up front and the deliberatley lopsided nature of our formation (amongst other things) makes xG meaningless.
Understat includes attack speed in their xG model. Our average xG per shot for a fast break is 0.25 compared to 0.04 for a slow attack. For normal it is 0.11 and for standard it is 0.18

They also have xG stats for formation, game state, situation (corner, free-kick, direct free-kick, open play, penalty), shot zones, timing

https://understat.com/team/Manchester_United/2018

Statsbomb's xG model has even more elements than that. I believe the latest version includes goal keeper positioning, height of ball when shot took place, extent of pressure on the shot taker, number of players behind the ball.

Of course, even then there will be some situations that the model fails to assess sensibly.

The main reason Solskjaer has so far out-performed xG models is because we have scored first in 9 out of 12 PL games, in many cases gone 2-0 up, even 3-0, then sat back and let our opponents take shots, in some cases get decent chances when the match was effectively over. In the Spurs match we were somewhat fortunate that they didn't take their chances but in other matches it barely mattered if Fulham scored late on (they missed) or Huddersfield scored to make it 3-1.

In the games in which we have conceded the first goal, our xG looks decent as we have been forced to chase the game, create opportunities.
 
Last edited:

chromepaxos

Full Member
Joined
Jul 29, 2015
Messages
192
xG is bollocks.
It’s like the opposite shite logic that MP used when he claimed that every school pupil should achieve an above average grade in their Maths GCSE.
xG doesn’t take into account who the player is.
Messi is currently on 1goal per game with an xG of 0.74. So he’s overachieving but over 33%. No shit, he’s Messi.
Look at the quality of our attacking players.
We’re achieving above a generic xG stat. Shock feckin horror.
Unbelievable how many people actually prefer to not understand what they're talking about. :lol:

Here's the thing: nobody in the stats business thinks xG is the be-all and end-all. Everybody recognizes that it is a simplified stat that doesn't account for player identity, game state, shot placement, etc. Everyone understands that good teams will do better than the mean.

But the fact that the experts accept it is an imperfect stat yet still useful should be a tell, even for the most determinedly unaware members of the Caf, that it has some value. Currently there are huge efforts underway to try and incorporate things like game state measurements, passing phases and forward progression into a more advanced stat that might replace xG in due course. But even now, xG offers real insights.

To take the current Olly run, for example, while it is true that most top teams outperform xG, most only do so by around a maximum of 10%. If Utd are well above that figure, then it's worthy of note. Yes, the explanation might be that our strikers are suddenly much better than those of City and of Pool and of Spurs, or it might be that we are benefiting from a lucky streak. I love Rashford but I'm not yet sure I want to say he is better than Kane or Aguero, so maybe there's some luck involved? And if there is, then we can expect some regression to the mean. Is that a blinding insight that you couldn't make yourself by simply assuming the current run is unlikely to be maintained at this level? Not really, but as the engineer's old saying says, it's much harder to move the needle if you can't measure what you want to change.
 

Maccataq

Full Member
Joined
May 22, 2016
Messages
286
Location
Manchester
They dont include penalties into the total, they separate I'm pretty sure. So really its like 36ish expected goals compared to 39 actual. (Big deal right).
No I don't think that's right. Penalties are always awarded 0.76xg, why wouldn't they be included?
 

Wibble

In Gadus Speramus
Staff
Joined
Jun 15, 2000
Messages
88,602
Location
Centreback
How are statistics overused? They are just segments of information for crying out loud
Expected goals can feck right off.

I once taught biostatictics (multivariat mainly) and use stats in my job so I don't have an aversion to them in a general sense, but this level examination and interpretation by the majority is just fecking pointless and almost completely meaningless.
 

cyberman

Full Member
Joined
May 26, 2010
Messages
37,331
Unbelievable how many people actually prefer to not understand what they're talking about. :lol:

Here's the thing: nobody in the stats business thinks xG is the be-all and end-all. Everybody recognizes that it is a simplified stat that doesn't account for player identity, game state, shot placement, etc. Everyone understands that good teams will do better than the mean.

But the fact that the experts accept it is an imperfect stat yet still useful should be a tell, even for the most determinedly unaware members of the Caf, that it has some value. Currently there are huge efforts underway to try and incorporate things like game state measurements, passing phases and forward progression into a more advanced stat that might replace xG in due course. But even now, xG offers real insights.

To take the current Olly run, for example, while it is true that most top teams outperform xG, most only do so by around a maximum of 10%. If Utd are well above that figure, then it's worthy of note. Yes, the explanation might be that our strikers are suddenly much better than those of City and of Pool and of Spurs, or it might be that we are benefiting from a lucky streak. I love Rashford but I'm not yet sure I want to say he is better than Kane or Aguero, so maybe there's some luck involved? And if there is, then we can expect some regression to the mean. Is that a blinding insight that you couldn't make yourself by simply assuming the current run is unlikely to be maintained at this level? Not really, but as the engineer's old saying says, it's much harder to move the needle if you can't measure what you want to change.
What experts use it?
 

Revan

Assumptionman
Joined
Dec 19, 2011
Messages
49,419
Location
London
Expected goals can feck right off.

I once taught biostatictics (multivariat mainly) and use stats in my job so I don't have an aversion to them in a general sense, but this level examination and interpretation by the majority is just fecking pointless and almost completely meaningless.
It seems to show decent predictive power, while in other sports which have been doing this for longer it works much better.

You look like the old man who says that cars will never be able to drive themselves.
 

chromepaxos

Full Member
Joined
Jul 29, 2015
Messages
192
The point is that people have been saying it would regress all the way to the mean, hence the twitter guy saying Ole shouldn’t take the job cos he’s over performing. If our regression is only going to be slight, then why is this a factor in the first place? I’d assume all teams when they have a purple patch outperform the expected stats.
Who has been saying it would "regress all the way"? I haven't seen one.

I think the point of the Tweet in the OP was that we are so far above xG - more than even Barca or City can sustain - that luck is clearly playing a significant part, maybe even a major part. That seems like a reasonable point for a statistician to make.
 

Wibble

In Gadus Speramus
Staff
Joined
Jun 15, 2000
Messages
88,602
Location
Centreback
It seems to show decent predictive power, while in other sports which have been doing this for longer it works much better.

You look like the old man who says that cars will never be able to drive themselves.
Not at all. Football isn't like testing a scientific hypothesis where stats are the be all and end all. You pick a team for how it fits together as a team and a manager doesn't need some bullshit meaningless individual statistics for this. Bad managers might. Bet Ole doesn't.
 

chromepaxos

Full Member
Joined
Jul 29, 2015
Messages
192
What experts use it?
Er...is your question, which statisticians use statistics that they themselves helped to develop? All of them, I think.

If you don't consider Opta and StatsBomb as experts, then I would refer you to the analytics department of pretty much every Premiership team. You think they are paying a ton of money for game data for no reason?
 

chromepaxos

Full Member
Joined
Jul 29, 2015
Messages
192
You pick a team for how it fits together as a team and a manager doesn't need some bullshit meaningless individual statistics for this. Bad managers might. Bet Ole doesn't.
Right. That's why you never see players at Carrington wearing heart rate monitors. Only bad managers want bullshit meaningless individual statistics.

Good grief.
 

Paul_Scholes18

Full Member
Joined
Sep 13, 2014
Messages
13,891
Right. That's why you never see players at Carrington wearing heart rate monitors. Only bad managers want bullshit meaningless individual statistics.

Good grief.
Stats are good, but context need to be used as well. The key thing for football teams is to win games and not get good stats. They can do this in different ways. Styles can change depending on results and this will effect the stats a lot.
 

Revan

Assumptionman
Joined
Dec 19, 2011
Messages
49,419
Location
London
Not at all. Football isn't like testing a scientific hypothesis where stats are the be all and end all. You pick a team for how it fits together as a team and a manager doesn't need some bullshit meaningless individual statistics for this. Bad managers might. Bet Ole doesn't.
Neither is driving a car or finding cancer from blurry images, or deciding to give a mortgage to a person or not, playing Go or Chess or video games etc, but it is working. There is no doubt that xG can be extremely useful, just that now we are at the beginning of their usage and in all likelihood the models are a bit primitive.

Pretty sure people would have said the same for chess a few decades ago. Thousands of years of human knowledge, emotion, art cannot be replicated. Guess what, even a primitive chess program beats infinite to 0 best grandmasters. Football is obviously much more complicated and it will require more work, but using advanced stats to predict results in long term, scout players and so on will be extremely useful and those who won't do it will fall behind.
 

Pexbo

Winner of the 'I'm not reading that' medal.
Joined
Jun 2, 2009
Messages
68,413
Location
Brizzle
Supports
Big Days
Neither is driving a car or finding cancer from blurry images, or deciding to give a mortgage to a person or not, playing Go or Chess or video games etc, but it is working. There is no doubt that xG can be extremely useful, just that now we are at the beginning of their usage and in all likelihood the models are a bit primitive.

Pretty sure people would have said the same for chess a few decades ago. Thousands of years of human knowledge, emotion, art cannot be replicated. Guess what, even a primitive chess program beats infinite to 0 best grandmasters. Football is obviously much more complicated and it will require more work, but using advanced stats to predict results in long term, scout players and so on will be extremely useful and those who won't do it will fall behind.
If you understood the technology behind the examples you have listed here and then compared them to xG, you would realise how absolutely ridiculous and incomparable they are.
 

cyberman

Full Member
Joined
May 26, 2010
Messages
37,331
Er...is your question, which statisticians use statistics that they themselves helped to develop? All of them, I think.

If you don't consider Opta and StatsBomb as experts, then I would refer you to the analytics department of pretty much every Premiership team. You think they are paying a ton of money for game data for no reason?
But you're saying they use xG as a fact. I'm wondering how you know this? It seems to be used by the average fan trying to make a point but I can't see where managers would use it?
Would Ole look at it and say we need to score more goals? I don't get it, especially when it's validated by saying its not exact, its flawed but will improve etc.
It seems to be a possession like stat. Interesting to know but ultimately meaningless.
 

Revan

Assumptionman
Joined
Dec 19, 2011
Messages
49,419
Location
London
If you understood the technology behind the examples you have listed here and then compared them to xG, you would realise how absolutely ridiculous and incomparable they are.
Considering that I am working in the field, I probably know a bit or two about the technology behind it.

I also admitted that current xG models are relatively primitive, but dismissing them is still stupid.
 

cyberman

Full Member
Joined
May 26, 2010
Messages
37,331
Considering that I am working in the field, I probably know a bit or two about the technology behind it.

I also admitted that current xG models are relatively primitive, but dismissing them is still stupid.
Surely reading too much into it while in its primative state is just as bad? How useful are flawed statistics in reality?
 

Green_Red

New Member
Joined
May 29, 2013
Messages
10,296
How could xG be lower than goals scored? Surely that makes zero sense statistically.
 

Revan

Assumptionman
Joined
Dec 19, 2011
Messages
49,419
Location
London
Surely reading too much into it while in its primative state is just as bad? How useful are flawed statistics in reality?
Of course. I don't think that anyone is making decisions purely on xG, but I think that it helps to keep an open mind about it, especially knowing that it is virtually certain that they will improve a lot in the next few years.

For scouting stats have always been used, but it has been much more primitive stats like height of parents, speed etc, while now we have a bit more advanced ones.
 

Paxi

Dagestani MMA Boiled Egg Expert
Joined
Mar 4, 2017
Messages
27,678
This XG twattery annoys the feck out of me.

For years I have rolled my eyes at NFL fans who spend more time looking at PFF spreadsheets than actually watching games. Now this same butllshit has crawled into football.


This tweet (probably already posted) says it perfectly:


Exactly. I'll let my eyes decide rather than reading some cnut spreadsheet.
 

ArmchairCritic

You got pets me too mines are dead
Joined
Jan 4, 2011
Messages
16,152
I think any rational person can reasonably deduct that at some point United might not win every single away game they play in. Don't need xG to tell me that.
 

shabadu84

Mint? Berry?
Joined
Jan 12, 2010
Messages
4,741
Location
Muppet Treasure Island
It doesn't need to be binary. Nobody uses sports analytics alone. And you can bet even the most steadfast, old-school proponent is also relying on sports science and advanced statistics in some fashion whether they know it or not.
 

Anustart89

Full Member
Joined
Jun 6, 2013
Messages
15,912
I've just been singing United songs all day at work, when I clearly should have been more concerned about xG.
Ole's at the wheel
Tell me what is his xG
We score more than we're supposed to score
And we should be conceding some more
Du du du du du du du du du du du
Regression to the mean to come
This song is dumb
 

bosnian_red

Worst scout to ever exist
Joined
Aug 13, 2011
Messages
57,585
Location
Canada
No I don't think that's right. Penalties are always awarded 0.76xg, why wouldn't they be included?
Caley for example shows open play xG and then in brackets has (1 pen). The main xG total is pretty much always not including pens. I went game by game to double check, they aren't included. Obviously so they can try and drive their agenda.
 

nistleloy

Full Member
Joined
May 7, 2009
Messages
1,013
Location
Republic of Mancunia
This looks like at a Peter Storey table type stat. It is trying to tell you something that has limited basis in reality of a football game and fails to look at the variables of the actual contest on the pitch. It is as meaningless as the possession stat taken in isolation of the rest of the game.

Does it adjust for red cards, injuries, goalies suddenly have a worldie, different set ups, teams parking the bus? It all seems rather pointless. Have fun discussing it but these stats always seem to me that individuals look at other sports data and get envious of it so sudo science the shit out of football.
 
Last edited:

chromepaxos

Full Member
Joined
Jul 29, 2015
Messages
192
But you're saying they use xG as a fact. I'm wondering how you know this? It seems to be used by the average fan trying to make a point but I can't see where managers would use it?
Would Ole look at it and say we need to score more goals? I don't get it, especially when it's validated by saying its not exact, its flawed but will improve etc.
It seems to be a possession like stat. Interesting to know but ultimately meaningless.
xG is just a starting point that serves as part of the foundation of the football analytics industry. I don't think anyone has suggested in this thread even - let alone in Man Utd's analytics dept - that xG on its own would indicate, "oh, we need to score more goals." xG is just a quick and dirty way to get a very rough picture.

But xG is now being overlaid and extended to provide much better insights by adding other data such as measurements of game states, passing in different areas of the pitch, number of opposition players bypassed by a pass, pace of the ball, skill of the individual player involved, etc. Which of these additional data will add most value is still a matter of intense discussion - and a lot of secrecy. Most teams won't talk about their own analytics efforts and where statisticians are under a team contract they typically do so under NDA.

I'm trying to think of an analogy. How about blood pressure? Nobody would only use blood pressure to diagnose someone - it's unreliable, it varies hour by hour, it's hard to get repeatable measurements over the short term, there are multiple causes, etc. But is it fundamental as an indicator of...something? Absolutely.

So, does anyone only use xG? Of course not. But it's still important.