I remember when we started the 2019-20 season really well, beating Blackburn, Stoke and Brentford, but were considerably outperforming our expected goals. That was a sign that regression was on the way, and of course that's what happened (though I don't think any of us predicted the extent of the mess that followed!).
As others have said, it's one piece of data that gives a reasonably good indication of how teams are playing, especially in this early stage of the season when the league table often bears little resemblance to the final standings.
It's a good indication of which team had the better of the play in terms of chances.
Statistically minded people will like it because looking purely at the scoreline will often not paint the picture of the match. For example, if Manchester City were shellacking Carlisle Utd's goal in the FA Cup for 90 mins, but it just won't go in, while Carlisle get a cheeky goal from their only shot, xG will tell us how the match played out aside from the 1-0 win for Carlisle.
When matches are decided by the odd goal, statistically there is a lot of "luck", or at least chance, involved. It's not perfect by any means, but xG attempts to show what the scoreline would be based on standard probability of the chances resulting in goals.
After 3 matches, we're still 21st for xG, and slipped to 4th for xGA.
The stats look encouraging for a Rotherham resurgence as unbelievably they've created the most goal worthy chances but scored 0.
Stockport look good for their high position in the table suggesting they are dominating their games.
Please be patient with me. What does 1.07 mean for us? We create or have goal scoring opportunities 1.07 per game/half/minute per something else or is it not a 'per' calculation?
EDIT: Forget it the video above helps a lot thanks
No hang on. The video says you can't have an xG of 1 because that's a guaranteed goal. But @balham red s table shows everyone but poor Wigan got somehow more than a guaranteed goal. Once again patience is appreciated.
It's a good indication of which team had the better of the play in terms of chances.
Statistically minded people will like it because looking purely at the scoreline will often not paint the picture of the match. For example, if Manchester City were shellacking Carlisle Utd's goal in the FA Cup for 90 mins, but it just won't go in, while Carlisle get a cheeky goal from their only shot, xG will tell us how the match played out aside from the 1-0 win for Carlisle.
When matches are decided by the odd goal, statistically there is a lot of "luck", or at least chance, involved. It's not perfect by any means, but xG attempts to show what the scoreline would be based on standard probability of the chances resulting in goals.
After 3 matches, we're still 21st for xG, and slipped to 4th for xGA.
The stats look encouraging for a Rotherham resurgence as unbelievably they've created the most goal worthy chances but scored 0.
Stockport look good for their high position in the table suggesting they are dominating their games.
Please be patient with me. What does 1.07 mean for us? We create or have goal scoring opportunities 1.07 per game/half/minute per something else or is it not a 'per' calculation?
It means we have created opportunities to score an expected 1.07 goals per game.
So if for example we had an xg of 1.00 we would expect to score 1 goal exactly every game.
Rotherham on this is a brilliant example of how it doesn’t always tally but can give a good indicator of form outside the actual result. They’ve not scored yet this season, but they’re creating enough chances to have expected to have scored an average of 1.8 goals a game. So roughly 5 goals so far this season.
So a punter would look at that and think there’s value in backing Rotherham to score a couple in their next game because the betting markets will focus on the fact they’ve not scored yet so the odds will be longer.
No hang on. The video says you can't have an xG of 1 because that's a guaranteed goal. But @balham red s table shows everyone but poor Wigan got somehow more than a guaranteed goal. Once again patience is appreciated.
What that means is that no individual chance will have an xg of 1 because they’re is always a chance they’ll miss. A sitter on the line with an open goal will have say an xg of 0.95 but there is always a chance they’ll miss so it can’t be 1.
I find xG a useful stat to give more background to how a team is playing, but in this case I'm confused how Bolton ended up with a higher xG than we did, as it implies they had a couple of decent chances, better ones than the ones we had.
I find xG a useful stat to give more background to how a team is playing, but in this case I'm confused how Bolton ended up with a higher xG than we did, as it implies they had a couple of decent chances, better ones than the ones we had.
The xG a team gets during a game could be made up of a few very good chances or many poor ones.
For example 1.5xG could be two 0.75 chances or ten 0.15 chances. How that looks in a game could be two great opportunities in the box with only the GK to beat vs ten long range shots.
Create a few of the former and you’ll likely score at least one do them. Do a lot of the latter and eventually one of them will go in.
From memory our main chances were 1) Docherty goal 2) Docherty double shot 3) Chuks missed header 4) Godden goal
For that to only come to 1.10 goals suggests that both of our goals had a low xG value, which slightly surprises me as while both were smart finishes, neither were of the worldie variety.
Then you have to really rack your brain to list their chances, McAtee had an early one that went wide, the Charles swivel and shot, the one that Mannion got down low to... All in the first half though. Did they really test imus much after that?
It’s only a model and it won’t be perfect - it’ll fit to the data over thousands of samples (shots) but there’s going to be individual matches or shots that people can pick out and question. It’s never going to be an exact science but the cleverest people who have the best models are making a lot of money from them.
Think of it like a trend line fitting the dots of a scatter graph. We don’t have all the inner workings of the model to know why the shots that resulted in goals were given 0.01 and 0.15 specifically.
The other thing that’s good to remember is that shots that turn into goals always appear easier to score from than they actually are, because of the fact they ended up in the net - plenty of times a player would’ve got into Docherty’s position and blasted the ball over the bar or directly into the defender or down the keeper’s throat.
Godden’s one will have been automatically marked down just for the fact it came from a header, which are statistically harder to score with than your feet. Like Chuks’ free header from a corner 10 minutes earlier which was given only 0.03.
Are the values right all the time? Almost definitely not, it’s just a computer making an educated guess. If you think you can do better on average than Opta’s boffins then I think the owners of Brighton and Brentford might have a well paying job lined up for you..!
I remember reading that to the neutral, headers are marked especially low on xg. That Aneke header is a prime example. 0.03 for what fans would consider a sitter.
I find xG a useful stat to give more background to how a team is playing, but in this case I'm confused how Bolton ended up with a higher xG than we did, as it implies they had a couple of decent chances, better ones than the ones we had.
I find xG a useful stat to give more background to how a team is playing, but in this case I'm confused how Bolton ended up with a higher xG than we did, as it implies they had a couple of decent chances, better ones than the ones we had.
A team that goes 1-0 up early will usually lose the xG. We didn’t do much in the second half because we were 1-0 up so didn’t have to. That’s where it’s important to consider game state and that xG doesn’t tell the full story.
Looking at the xG you’d think Bolton were hard done by but in reality we kept them at arms length and limited them to low quality chances
It’s only a model and it won’t be perfect - it’ll fit to the data over thousands of samples (shots) but there’s going to be individual matches or shots that people can pick out and question. It’s never going to be an exact science but the cleverest people who have the best models are making a lot of money from them.
Think of it like a trend line fitting the dots of a scatter graph. We don’t have all the inner workings of the model to know why the shots that resulted in goals were given 0.01 and 0.15 specifically.
The other thing that’s good to remember is that shots that turn into goals always appear easier to score from than they actually are, because of the fact they ended up in the net - plenty of times a player would’ve got into Docherty’s position and blasted the ball over the bar or directly into the defender or down the keeper’s throat.
Godden’s one will have been automatically marked down just for the fact it came from a header, which are statistically harder to score with than your feet. Like Chuks’ free header from a corner 10 minutes earlier which was given only 0.03.
Are the values right all the time? Almost definitely not, it’s just a computer making an educated guess. If you think you can do better on average than Opta’s boffins then I think the owners of Brighton and Brentford might have a well paying job lined up for you..!
0.01 though makes the Docherty goal a 1 in 100 chance, which is an odd interpretation, as that would imply it was either a goal of the month bit of finishing or a bad bit of keeping, when it was a decent, but not exceptional finish. It was in the D (so central) and Doc was under no pressure either, yet was ranked a much worse chance than others where the attacking player was under pressure or had lots of defenders in front of him.
I wonder if the stats are skewed too much by Premier League and top level football. For example the Collins free kick being ranked as 0.16 states that 1 in 6 free kicks in that position will go in, which seems way too high lower down the leagues, where without the likes of a Ward-Prowse or KDB, the number of direct free kicks scored during a season is tiny when compared with headers from set pieces. Even if headers are statistically harder than shots, 0.03 for the Chuks chance, a free header, seems far too low. It's the sort of goal which you see up and down the country, a free header from a corner.
There’s elements that the model that’s being used probably isn’t taking into account, if I had to guess.
The shot map shows that Chuks’ header came from a corner, from outside the six yard box, slightly wide of the frame of the goal, with his head.
Putting all of those pieces together the computer is searching a database of previous headers from corners from 8-9 yards out and determined that ~3% of them were goals and will spit out a number based on that. It likely isn’t taking into account the fact that in this particular instance it was a free header which would have instantly boosted its likelihood.
But again, it’s just a model and it’s using averages. It means there’ll be other, contested headers, from that area of the pitch, that are assigned a number that’s slightly too high - and it will all average out to 3% over thousands of shots.
So these 'stats' are actually based on subjective opinions then?
Hmmm.
Next we'll be having people tell us that a goal has 1.4 chance of going to VAR where it will have a 0.82 chance of being disallowed, and trying to work out the outcome of the game on that.
For both of our goals at the weekend, I doubt that the model being used is taking into account the high turnover of the ball that led to the chance in the first place.
Both chances are described as being from “regular play” which would indicate that it’s being lumped in with 20-25 yard pot shots with lots of men behind the ball (in Docherty’s case) or a somewhat difficult headed chance between defenders from outside the six yard box (in Godden’s case).
In both cases, you’re far more likely to score because the defence is caught on their heels and out of shape.
Again, it’s just a model and it has flaws but on average it’s going to fit the data 99%+
Comments
As others have said, it's one piece of data that gives a reasonably good indication of how teams are playing, especially in this early stage of the season when the league table often bears little resemblance to the final standings.
EDIT: Forget it the video above helps a lot thanks
Rotherham on this is a brilliant example of how it doesn’t always tally but can give a good indicator of form outside the actual result. They’ve not scored yet this season, but they’re creating enough chances to have expected to have scored an average of 1.8 goals a game. So roughly 5 goals so far this season.
So a punter would look at that and think there’s value in backing Rotherham to score a couple in their next game because the betting markets will focus on the fact they’ve not scored yet so the odds will be longer.
A team can have an xg of 1 in a game.
I find xG a useful stat to give more background to how a team is playing, but in this case I'm confused how Bolton ended up with a higher xG than we did, as it implies they had a couple of decent chances, better ones than the ones we had.
For example 1.5xG could be two 0.75 chances or ten 0.15 chances. How that looks in a game could be two great opportunities in the box with only the GK to beat vs ten long range shots.
Create a few of the former and you’ll likely score at least one do them. Do a lot of the latter and eventually one of them will go in.
1) Docherty goal
2) Docherty double shot
3) Chuks missed header
4) Godden goal
For that to only come to 1.10 goals suggests that both of our goals had a low xG value, which slightly surprises me as while both were smart finishes, neither were of the worldie variety.
2’ McAtee (right foot, regular play, miss): 0.02 xG
10’ Docherty (right foot, regular play, goal): 0.01 xG
14’ Collins (right foot, free kick, blocked): 0.16 xG
16’ Charles (right foot, regular play, miss): 0.02 xG
19’ Charles (right foot, regular play, miss): 0.02 xG
21’ Schon (right foot, regular play, blocked): 0.05 xG
21’ Sheehan (right foot, regular play, blocked): 0.04 xG
21’ Charles (right foot, regular play, saved): 0.10 xG
24’ Collins (right foot, regular play, saved): 0.08 xG
28’ Docherty (right foot, regular play, saved): 0.11 xG
28’ Docherty (right foot, regular play, saved): 0.08 xG
48’ Adeboyejo (left foot, regular play, saved): 0.02 xG
52’ Schon (left foot, corner, miss): 0.02 xG
68’ Toal (header, corner, miss): 0.02 xG
76’ Aneke (header, corner, miss): 0.03 xG
77’ McAtee (right foot, regular play, miss): 0.02 xG
82’ Aneke (right foot, regular play, miss): 0.03 xG
88’ Godden (header, regular play, goal): 0.15 xG
90+2’ Osei-Tutu (right foot, regular play, miss): 0.03 xG
I think I may have spotted a bit of an issue there.
That Collins free kick was in perfect position to be fair, I was half expecting (0.5xG 😉 ) it to go in.
Godden's goal should surely be higher than 0.15 as well? Free header, centre of the goal, unmarked, about 8 yards out, would've been harder to miss.
The other thing that’s good to remember is that shots that turn into goals always appear easier to score from than they actually are, because of the fact they ended up in the net - plenty of times a player would’ve got into Docherty’s position and blasted the ball over the bar or directly into the defender or down the keeper’s throat.
Are the values right all the time? Almost definitely not, it’s just a computer making an educated guess. If you think you can do better on average than Opta’s boffins then I think the owners of Brighton and Brentford might have a well paying job lined up for you..!
I wonder if the stats are skewed too much by Premier League and top level football. For example the Collins free kick being ranked as 0.16 states that 1 in 6 free kicks in that position will go in, which seems way too high lower down the leagues, where without the likes of a Ward-Prowse or KDB, the number of direct free kicks scored during a season is tiny when compared with headers from set pieces. Even if headers are statistically harder than shots, 0.03 for the Chuks chance, a free header, seems far too low. It's the sort of goal which you see up and down the country, a free header from a corner.
anyone who thinks central feee kicks go in one in six is deluded and that Chuks should only score that header every 33 times is nonsense
The shot map shows that Chuks’ header came from a corner, from outside the six yard box, slightly wide of the frame of the goal, with his head.
But again, it’s just a model and it’s using averages. It means there’ll be other, contested headers, from that area of the pitch, that are assigned a number that’s slightly too high - and it will all average out to 3% over thousands of shots.
Hmmm.
Next we'll be having people tell us that a goal has 1.4 chance of going to VAR where it will have a 0.82 chance of being disallowed, and trying to work out the outcome of the game on that.
Both chances are described as being from “regular play” which would indicate that it’s being lumped in with 20-25 yard pot shots with lots of men behind the ball (in Docherty’s case) or a somewhat difficult headed chance between defenders from outside the six yard box (in Godden’s case).
In both cases, you’re far more likely to score because the defence is caught on their heels and out of shape.
Again, it’s just a model and it has flaws but on average it’s going to fit the data 99%+