United Hour - xG Gon' Give It To Ya

Rood

nostradamus like gloater
Scout
Joined
Jun 21, 2008
Messages
21,232
Location
@United_Hour
Nik @Rood , Ed @That'sHernandez and Colm @Duafc assess the first month of the season and the frustrating results against Palace and Southampton.
We go deep on 'Expected Goals' (xG) stats which suggest we have been unlucky so far - are they worthwhile or just a convenient excuse for poor performance?
A quick mention for the start of the Women's Super League and the international break.
Finally we preview our next game against Leicester City and the start of our of Europa League campaign against the champions of Kazakhstan.




Transistor (stream or download): https://unitedhour.transistor.fm/episodes/united-hour-podcast-xg-gon-give-it-to-ya

Spotify:

iTunes: https://podcasts.apple.com/gb/podcast/united-hour/id902792476
Facebook: https://www.facebook.com/manchesterunitedpodcast/
Twitter: https://twitter.com/United_Hour
 
Last edited:

overtheborder

New Member
Newbie
Joined
Jul 12, 2019
Messages
15
Interested listening to this XG stuff - i wonder what factors the algorithm takes into account, does it take into account the player's goalscoring record who had the chance, the person delivering the assist's record with them etc? Interesting stat to have around though.
 
  • Like
Reactions: Rood

ROFLUTION

Full Member
Joined
Jan 22, 2009
Messages
7,547
Location
Denmark
Interested listening to this XG stuff - i wonder what factors the algorithm takes into account, does it take into account the player's goalscoring record who had the chance, the person delivering the assist's record with them etc? Interesting stat to have around though.
Yeah can anyone elaborate on this? I Know it includes where the shot was taken and the angle of the shot
 

Ole90+3

Full Member
Joined
Apr 24, 2005
Messages
5,945
Location
Paddy's Pub with the gang
Interested listening to this XG stuff - i wonder what factors the algorithm takes into account, does it take into account the player's goalscoring record who had the chance, the person delivering the assist's record with them etc? Interesting stat to have around though.
I think the algorithm uses the success rate of all teams from those positions. Not sure what leagues are brought into the equation or the timespan.
 

Sparky Rhiwabon

New Member
Joined
Jul 10, 2013
Messages
16,946
Interested listening to this XG stuff - i wonder what factors the algorithm takes into account, does it take into account the player's goalscoring record who had the chance, the person delivering the assist's record with them etc? Interesting stat to have around though.
No, I don't think it takes into account any of that. Hence good (or on form) strikers will generally outperform their xG and vice versa
 

bosnian_red

Worst scout to ever exist
Joined
Aug 13, 2011
Messages
57,585
Location
Canada
Interested listening to this XG stuff - i wonder what factors the algorithm takes into account, does it take into account the player's goalscoring record who had the chance, the person delivering the assist's record with them etc? Interesting stat to have around though.
It doesnt take into account individual player ability, nor should it really. Its judging the quality of chances assuming average finishing for that chance. If you look at historical data over years, even the top strikers who are known as the most clinical only have minor variances from their xG. Messi is one of the only ones who has a consistently decently big variance, and even that is something like 10% only. Aguero is known as a super clinical striker, yet his xG over the past ~3 seasons is basically dead even.
 

bosnian_red

Worst scout to ever exist
Joined
Aug 13, 2011
Messages
57,585
Location
Canada
No, I don't think it takes into account any of that. Hence good strikers will generally outperform their xG and vice versa
People always say this assuming it to be true but it's not really. There is generally a very tiny difference between xG and their actual goals. Aguero is the perfect example of it, almost no difference over the past 3 seasons.

Edit: Since the start of 2014/15, Aguero scored 118 goals and his xG is 109.65. 8.45 goals over 5 seasons and 4 games... 7%.
 

Classical Mechanic

Full Member
Joined
Aug 25, 2014
Messages
35,216
Location
xG Zombie Nation
People always say this assuming it to be true but it's not really. There is generally a very tiny difference between xG and their actual goals. Aguero is the perfect example of it, almost no difference over the past 3 seasons.

Edit: Since the start of 2014/15, Aguero scored 118 goals and his xG is 109.65. 8.45 goals over 5 seasons and 4 games... 7%.
Messi's is surely over 10% for 5 years?
 

bosnian_red

Worst scout to ever exist
Joined
Aug 13, 2011
Messages
57,585
Location
Canada
Messi's is surely over 10% for 5 years?
Just checked, yeah Messi is actually decently bigger. Has 176 goals, 144.82 xG, so his xG is 17.6% off of his actual total.
Ronaldo is at 156 goals, 151.95 xG for example.
2 of the best goalscorers of all time, 2 of the best players of all time, one has an xG of only 2.5% off of his total, while the other is 17%. So obviously there will be a range, but the best goalscorers aren't necessarily players who outperform xG. It's usually players who just get themselves in big chances often, as xG for each individual is pretty close to being spot on over a long period of time.
 

Classical Mechanic

Full Member
Joined
Aug 25, 2014
Messages
35,216
Location
xG Zombie Nation
Just checked, yeah Messi is actually decently bigger. Has 176 goals, 144.82 xG, so his xG is 17.6% off of his actual total.
Ronaldo is at 156 goals, 151.95 xG for example.
2 of the best goalscorers of all time, 2 of the best players of all time, one has an xG of only 2.5% off of his total, while the other is 17%. So obviously there will be a range, but the best goalscorers aren't necessarily players who outperform xG. It's usually players who just get themselves in big chances often, as xG for each individual is pretty close to being spot on over a long period of time.
What's Kane, probably somewhere closer to Aguero I'd guess?
 

bosnian_red

Worst scout to ever exist
Joined
Aug 13, 2011
Messages
57,585
Location
Canada
Another good example is Lewandowski. 134 goals, 141.2 xG. Won't see many who claim he isn't one of the best goalscorers in the world.
 

bosnian_red

Worst scout to ever exist
Joined
Aug 13, 2011
Messages
57,585
Location
Canada
What's Kane, probably somewhere closer to Aguero I'd guess?
Kane was +20 over 5 years, Lewandowski is -7, Aguero +8, Ronaldo +4, Salah +20, Suarez +6.7, Salah +15, Martial +8 (though with much smaller numbers, so much higher percentage), Rashford -0.75, Aubameyang -6.

Can look on understat.com for a nice and easy format.
 

Rood

nostradamus like gloater
Scout
Joined
Jun 21, 2008
Messages
21,232
Location
@United_Hour
Interested listening to this XG stuff - i wonder what factors the algorithm takes into account, does it take into account the player's goalscoring record who had the chance, the person delivering the assist's record with them etc? Interesting stat to have around though.
Yeah can anyone elaborate on this? I Know it includes where the shot was taken and the angle of the shot
It doesnt take into account individual player ability, nor should it really. Its judging the quality of chances assuming average finishing for that chance. If you look at historical data over years, even the top strikers who are known as the most clinical only have minor variances from their xG. Messi is one of the only ones who has a consistently decently big variance, and even that is something like 10% only. Aguero is known as a super clinical striker, yet his xG over the past ~3 seasons is basically dead even.
The dark arts behind xG are not entirely clear, worth noting that there is no single accepted measure of this stat and there are a few different options out there all with their own secret algorithm. As far as I can tell they all give similar results in the end.

The numbers I used on this episode come from https://understat.com/ who state that "we trained neural network prediction algorithms with the large dataset (>100,000 shots, over 10 parameters for each)." They do not tell us what those 10 parameters are though.

Meanwhile the big boys at OPTA tell us that "Expected goals (xG) measures the quality of a shot based on several variables such as assist type, shot angle and distance from goal, whether it was a headed shot and whether it was defined as a big chance."
 

Crashoutcassius

Full Member
Joined
Oct 10, 2013
Messages
10,295
Location
playa del carmen
For a striker, important to remember that xg per game is as Important as whether they run over and under and arguably more important in the short run. Getting in positions to score is more important since xg tends to equal goals scored over time.
 

bosnian_red

Worst scout to ever exist
Joined
Aug 13, 2011
Messages
57,585
Location
Canada
For a striker, important to remember that xg per game is as Important as whether they run over and under and arguably more important in the short run. Getting in positions to score is more important since xg tends to equal goals scored over time.
Yep, your general level of play and getting in chances regularly is what should be repeatable for players, while finishing will always blow hot and cold really. Martial by the stats is one of the most clinical players around, it's just he doesn't find himself in anywhere near enough chances. Doesn't matter how clinical he is if he only gets the chances to get 10-15 goals a season. Needs to bump that number up to be the main striker.
 

Frank Grimes

Full Member
Joined
Aug 1, 2014
Messages
8,608
Location
Newbies 15/16 FPL Champion.
I don't mind xG as it relates to players but I find it a bit flawed the way it relates to games, as the timing of goals and the scoreline at the time can give a false interpretation of a game.

If a team is 3-0 up and cruising and their opponents miss a host of chances to pull one back they could end up looking better on the xG but in reality they are well beaten.
 

Withnail

Full Member
Joined
Jan 5, 2019
Messages
29,333
Location
The Arena of the Unwell
Yep, your general level of play and getting in chances regularly is what should be repeatable for players, while finishing will always blow hot and cold really. Martial by the stats is one of the most clinical players around, it's just he doesn't find himself in anywhere near enough chances. Doesn't matter how clinical he is if he only gets the chances to get 10-15 goals a season. Needs to bump that number up to be the main striker.
I think it's important to remember that Martial hasn't played regularly in the team for a full season since his first one at the club. Since then he's only ever played half a season if you look at his total minutes. Last year he got 10 goals in ~1600 mins which is the equivalent of 18 games or so.

If he can stay fit and gets a decent run of games as our main number 9 this season his numbers should be higher, given how clinical he is.
 

GhastlyHun

Full Member
Joined
Dec 19, 2015
Messages
12,653
Location
Bavaria
Supports
Bayern München
I don't mind xG as it relates to players but I find it a bit flawed the way it relates to games, as the timing of goals and the scoreline at the time can give a false interpretation of a game.

If a team is 3-0 up and cruising and their opponents miss a host of chances to pull one back they could end up looking better on the xG but in reality they are well beaten.
And on another day they might have mounted an epic comeback. That's basically exactly the point of xG in games. It tells you something about how well the teams performed to create chances, even when the scoreline doesn't reflect that because for some reason or other they didn't make them count. (City - Tottenham 2-2 is a prime example of this)
 

slyadams

Full Member
Joined
Sep 20, 2012
Messages
2,189
My understanding is that each shot is evaluated against all previous efforts from a very similar situation, so a probability is generated for that chance. However, you cannot just sum up probalities.

For example, let's say a team has 3 chances and each has an XG of 1/3. These get summed up as an XG of 1. However, the probably of scoring at least 1 goal from these 3 chances is actually only 70%:

p(at least 1 goal) =1 - p(scoring no goals)
p(at least 1 goal) =1 - p(0.6666^3) = 0.70
 

Rood

nostradamus like gloater
Scout
Joined
Jun 21, 2008
Messages
21,232
Location
@United_Hour
My understanding is that each shot is evaluated against all previous efforts from a very similar situation, so a probability is generated for that chance. However, you cannot just sum up probalities.

For example, let's say a team has 3 chances and each has an XG of 1/3. These get summed up as an XG of 1. However, the probably of scoring at least 1 goal from these 3 chances is actually only 70%:

p(at least 1 goal) =1 - p(scoring no goals)
p(at least 1 goal) =1 - p(0.6666^3) = 0.70
Well are you sure about that? As I mentioned earlier, the exact workings behind xG are not clear and as I understand it the formula is constantly evolving anyway
 

slyadams

Full Member
Joined
Sep 20, 2012
Messages
2,189
Well are you sure about that? As I mentioned earlier, the exact workings behind xG are not clear and as I understand it the formula is constantly evolving anyway
I'm sure about my maths yes. The only thing I don't know is if they still just add up the individual chance XGs, but I believe that is the case.

I had a google to see if there was info on how the XG is currently aggregated and found this: https://medium.com/@dannypage/expected-goals-just-don-t-add-up-they-also-multiply-1dfd9b52c7d0
 

slyadams

Full Member
Joined
Sep 20, 2012
Messages
2,189
Agree with @Duafc , Feck XG... using my eyes, we didn't look like scoring in that Southampton game at all.
Another good example. If you have 20 shots from outside the box where each shot has an XG of 0.05, then the quoted team XG will be 1. The actual probability of scoring would only be 64%.

In general a lot of crappy chances will overstate your XG.
 

That'sHernandez

Ominously close to getting banned
Joined
Oct 30, 2010
Messages
24,545
Yeah that came across quite clearly. Do some research next time lads!
I do really know what it is and I'm aware of it, I just think it's bollocks! If you play incisive attacking football, you're going to score goals like we did when Ole first came in (despite our xG being lower than the amount of goals we were scoring) and vice versa, if you play crap football your xG will drop. It's a reactive stat as much as it is a predictive stat and a waste of time if you ask me; I'd rather watch the football.
 

Rood

nostradamus like gloater
Scout
Joined
Jun 21, 2008
Messages
21,232
Location
@United_Hour

bosnian_red

Worst scout to ever exist
Joined
Aug 13, 2011
Messages
57,585
Location
Canada
My understanding is that each shot is evaluated against all previous efforts from a very similar situation, so a probability is generated for that chance. However, you cannot just sum up probalities.

For example, let's say a team has 3 chances and each has an XG of 1/3. These get summed up as an XG of 1. However, the probably of scoring at least 1 goal from these 3 chances is actually only 70%:

p(at least 1 goal) =1 - p(scoring no goals)
p(at least 1 goal) =1 - p(0.6666^3) = 0.70
Another good example. If you have 20 shots from outside the box where each shot has an XG of 0.05, then the quoted team XG will be 1. The actual probability of scoring would only be 64%.

In general a lot of crappy chances will overstate your XG.
Could be wrong but isnt your logic more "the odds of scoring at that point if a goal hasnt been scored yet". You're looking at each chance being dependent on the other, when summing is more used if each is independent of another (like how its judged for xG). You aren't any more or less likely to score a chance depending on previous chances in a game, so they're all independent of another, so each chance gets taken as a sum.
 

Frank Grimes

Full Member
Joined
Aug 1, 2014
Messages
8,608
Location
Newbies 15/16 FPL Champion.
Yes
And on another day they might have mounted an epic comeback. That's basically exactly the point of xG in games. It tells you something about how well the teams performed to create chances, even when the scoreline doesn't reflect that because for some reason or other they didn't make them count. (City - Tottenham 2-2 is a prime example of this)
Yes but also when a team is cruising and they get a jolt they often regather.
 

bosnian_red

Worst scout to ever exist
Joined
Aug 13, 2011
Messages
57,585
Location
Canada
Some xG models also add up the xG from 1 move. So chances from rebounds get counted. Kinda dumb.
He vast majority don't and just take the highest xG chance I think. Not sure which xG model just totals each chance in 1 series as that's easy to prove how dumb it is.
 

Bastian

Full Member
Joined
Jul 16, 2015
Messages
18,444
Supports
Mejbri
Cheers for this! Great to have that regularity now :)

On the xG front, I thought Castle's take was interesting yesterday, when he says that xG fails to address a few things. Such as winning a game in the first 30 minutes and not needing to create chances or falling behind in a match and desperately trying to create any sort of an attempt (which is what we've often had to do). And of course, xG for the small sample size of 4 games is almost irrelevant.

For me there are two fundamental problems we've seen this season that are quite coaching specific: We are nowhere near clinical enough and we don't come up with a way to play that mitigates that. So we're relying on individual brilliance there rather than a team philosophy. This is also painfully apparent in set pieces that seem to be not coached at all. Like, really, not at all. We counter. We press, at times. But we have no idea how to build meaningful attacks when we have the lion's share of possession. Secondly, we are still conceding sloppy goals. It's maybe not an easy problem to fix. Do we play a high line with two non-pacey centre backs? Do we solidify midfield to provide more than the appearance of defensive cover? Do we play Lindelof against a team that poses aerial threats or against teams with rapid counter attacks (like Leicester on Saturday)?

Isn't Kieran McKenna supposed to be some brilliant attacking coach? I worry that Ole is more a positive motivator - with an eye for talent, who has a coaching staff that's got virtually no experience bar Phelan when he really needs the expertise there to bring this team on. We can't develop these youngsters if all they get (I assume) is go out there and express yourself.
 

Frank Grimes

Full Member
Joined
Aug 1, 2014
Messages
8,608
Location
Newbies 15/16 FPL Champion.
I remember we beat Arsenal 3-1 in the Emirates a couple of seasons back and Arsenal battered us but for De Geas heroics. But despite this Utd still led by 2 clear goals for over 90% of the game. Context is lost sometimes with xG.