Something to piss a few off , XG table (19 Viewers)

[shmmeee]

Looks like a load of Bollocks to me. Overall as an average, our xG for is 1.46 per game and our XG against is 1.25 per game. The difference is negligible really. We've score slightly less than what we are expected 1.19 goals per game, and conceding slightly more than we would expect 1.31 goals per game

https://footystats.org/clubs/coventry-city-fc-239

Sent from my Pixel 7 using Tapatalk

That’s like saying our goal difference is roughly 0 so we must have drawn every game.

 

---

[Flying Fokker]

The table is showing how many wins draws and losses we would be on based on expected goals for and against .

Long story short , we're creating alot

Of angst?

 

[SBAndy]

Oh yeah definitely, like I said there’s clearly been some opportunities our strikers have missed which should have been put away but I think quite a few of them have been tricky angles or the ball’s bobbled around or the defence has made it awkward to get a clear shot away. I think it’s probably fine margins but I don’t completely blame our strikers, I definitely blame them somewhat but imo a lot of the time they’re not being given the easiest route to goal from our midfielders.

That was my hunch as well, thinking our xG might be inflated due to the number of long shots we take. Our xG per shot is 0.1, same as Ipswich and marginally behind Leicester (0.11).

 

[BlueSkiesForever]

That was my hunch as well, thinking our xG might be inflated due to the number of long shots we take. Our xG per shot is 0.1, same as Ipswich and marginally behind Leicester (0.11).

Exactly, and yet when you watch our games it doesn’t look like we’d ever score anywhere near the same amount of goals as Ipswich or Leicester. I think that’s down to the quality of the shot as you said, they’re not proper chances a lot of the time, sort of almost chances and yet for some reason someone is counting them as actual expected goals

 

[SBAndy]

Exactly, and yet when you watch our games it doesn’t look like we’d ever score anywhere near the same amount of goals as Ipswich or Leicester. I think that’s down to the quality of the shot as you said, they’re not proper chances a lot of the time, sort of almost chances and yet for some reason someone is counting them as actual expected goals

Don’t think you’re quite understanding the data. If you shot from the halfway line then it would still generate an xG, likely 0.01 i.e you’d score 1 in 100 (I think it just rounds to 2 decimal places as I think you could try that 100,000 times and maybe one comes off). If you’re on the line with no keeper then it might generate 0.99 xG i.e. almost impossible to miss.

The fact our xG vs shots is comparable to Ipswich and Leicester continues to point the finger at poor finishing. Our top 3 players for ‘Big Chances Missed’ (Godden, Wright, Simms) equates to 21 big chances missed. Leicester’s is 15. Ipswich’s is also 15. We’re missing a lot more comparatively easier chances.

 

[BlueSkiesForever]

Don’t think you’re quite understanding the data. If you shot from the halfway line then it would still generate an xG, likely 0.01 i.e you’d score 1 in 100 (I think it just rounds to 2 decimal places as I think you could try that 100,000 times and maybe one comes off). If you’re on the line with no keeper then it might generate 0.99 xG i.e. almost impossible to miss.

The fact our xG vs shots is comparable to Ipswich and Leicester continues to point the finger at poor finishing. Our top 3 players for ‘Big Chances Missed’ (Godden, Wright, Simms) equates to 21 big chances missed. Leicester’s is 15. Ipswich’s is also 15. We’re missing a lot more comparatively easier chances.

I know what you mean but just from watching the games it doesn’t feel that way, it doesn’t feel like we’ve created that many big chances to me. I realise I’m arguing against data but is it proper solid data? I’m guessing it’s still done by someone watching the game and counting chances? Or is some AI thing doing it somehow?

I’m just not convinced that our xG v shots is comparable with Leicester or Ipswich, it’s clear that our strikers have been under performing but have they been underperforming so much that we’d be scoring as many goals as Ipswich and Leicester if they’d been on form? I’m not sure, I personally think it’s a whole team thing, we have had “big chances” but have they actually been “big chances”, I get that it sounds thick, I know, but I’m just not entirely convinced that the blame should be put solely on our strikers.

 

[shmmeee]

I know what you mean but just from watching the games it doesn’t feel that way, it doesn’t feel like we’ve created that many big chances to me. I realise I’m arguing against data but is it proper solid data? I’m guessing it’s still done by someone watching the game and counting chances? Or is some AI thing doing it somehow?

I’m just not convinced that our xG v shots is comparable with Leicester or Ipswich, it’s clear that our strikers have been under performing but have they been underperforming so much that we’d be scoring as many goals as Ipswich and Leicester if they’d been on form? I’m not sure, I personally think it’s a whole team thing, we have had “big chances” but have they actually been “big chances”, I get that it sounds thick, I know, but I’m just not entirely convinced that the blame should be put solely on our strikers.

But that’s why we have data, because “feels” aren’t reliable.

 

[BlueSkiesForever]

But that’s why we have data, because “feels” aren’t reliable.

I realise that but I watch us play almost every week, I do not think our strikers are solely to blame for the fact we haven’t scored many goals. It’s a team sport and our team has not given them many gilt-edge chances to score, there’s definitely been some, I’m not arguing that but there’s not been enough. A lot of the time the strikers are getting the ball pumped up to them and they’re having to somehow dribble past 2 or 3 defenders to score, that’s not always the case but that seems to be how a lot of our chances are created.

I’m happy with the stats being facts but it’s how you interpret those facts, can we all sit here and actually say we’ve created as many solid opportunities to score as Leicester and Ipswich etc? I certainly don’t think that’s the case, some numbers might say so but in the real world, on the pitch, we are miles away from creating the number of clinical opportunities the likes of Leicester, Leeds, Ipswich regularly do.

 

[Grendel]

But that’s why we have data, because “feels” aren’t reliable.

The data though is different depending on which data source you have on xg

 

[shmmeee]

I realise that but I watch us play almost every week, I do not think our strikers are solely to blame for the fact we haven’t scored many goals. It’s a team sport and our team has not given them many gilt-edge chances to score, there’s definitely been some, I’m not arguing that but there’s not been enough. A lot of the time the strikers are getting the ball pumped up to them and they’re having to somehow dribble past 2 or 3 defenders to score, that’s not always the case but that seems to be how a lot of our chances are created.

I’m happy with the stats being facts but it’s how you interpret those facts, can we all sit here and actually say we’ve created as many solid opportunities to score as Leicester and Ipswich etc? I certainly don’t think that’s the case, some numbers might say so but in the real world, on the pitch, we are miles away from creating the number of clinical opportunities the likes of Leicester, Leeds, Ipswich regularly do.

I dunno, I don’t watch Leicester and Ipswich. But I do think we’ve missed more than usual and let in more silly goals than usual. Which is basically what xG is saying. Our general play hasn’t been as bad as our results, despite the wailing and gnashing on here.

 

[hill83]

I’m guessing it’s still done by someone watching the game and counting chances? Or is some AI thing doing it somehow?

The data though is different depending on which data source you have on xg

Can someone answer these? I’m sick to death of hearing about XG, so if you can’t beat them join them. I want to learn. This would be a start.

 

[shmmeee]

Can someone answer these? I’m sick to death of hearing about XG, so if you can’t beat them join them. I want to learn. This would be a start.

So like any model of the real world it’s boiled down into a set of features the modeller thinks is important.

At a basic level that feature is “what’s the probability that this attempt at goal goes in?” And you might for a super simple model take just the x, y coordinates on the pitch. But we can all see issues with this like a shot from the penalty spot in a crowded area isn’t the same as a penalty isn’t the same as a free header, etc.

So different models add different features in to try and tease out a more accurate probability. They might take into account whether it’s a set piece, or even a counter attack. They might take into account if it’s the players “wrong” foot, or the number of defenders in a given cone of influence. They might also train their probabilities on different data sets. Maybe only the PL, maybe the full pyramid depending on the resources available for data collection.

The aim is to get a model that best summarises how “easy” a chance is, like all models it is obviously wrong and cannot model every feature that impacts, but we hope that by modelling the most important and impactful features we can glean something useful, if not exact.

 

[shmmeee]

To give a toy example. Imagine all shots are equal. We live in a fantasy land with no defenders or goal keepers and each ball is hit from a dead ball. All players are equally good with whatever foot they’re using as each other. So the only variable that matters is where the shot is taken from.

Now we get 100 players to take a shot from various positions (edge of the D, centre circle, by the corner flag) and record how many go in. This gives you a probability for each location you tried. You could then use various statistical techniques like a voronoi diagram or some form of clustering to workout the probability for any future shot in this god awful one sided dead ball game we’ve invented.

 

[hill83]

Cheers. How is the data collated and how does our XG in one game have any bearing to the next game. I understand XG now in isolation but I can’t get to grips with for example how many easier shots we had against one team has any bearing on what shots we might have against another team.

 

[shmmeee]

Cheers. How is the data collated and how does our XG in one game have any bearing to the next game. I understand XG now in isolation but I can’t get to grips with for example how many easier shots we had against one team has any bearing on what shots we might have against another team.

I don’t know the ins and outs of each model, I’m not sure if that info is available. Someone like @Frostie might know.

I doubt the models update game by game, though maybe.

But generally there’s no link. The idea is to capture whether you’re missing/scoring more than you “should” over a period. For me totals are where things get tricky as 7 0.1 attempts might not really be equal to one 0.7 attempt (eg a pen). But overall on average if you’re consistently having better chances than you’re conceding you should be doing OK.

 

[stupot07]

Our xG has been superior in 13 of the 16 games

Sent from my Pixel 7 using Tapatalk

How superior though? Are we talking significant or negligible? If it is. 1.3 Vs 1.1 that's negligible and we cou could have 20 poor chances, yet the other team has a pan and 2 other big chances. XG is saying we should have won that game.

Sent from my Pixel 7 using Tapatalk

 

[stupot07]

I don’t know the ins and outs of each model, I’m not sure if that info is available. Someone like @Frostie might know.

I doubt the models update game by game, though maybe.

But generally there’s no link. The idea is to capture whether you’re missing/scoring more than you “should” over a period. For me totals are where things get tricky as 7 0.1 attempts might not really be equal to one 0.7 attempt (eg a pen). But overall on average if you’re consistently having better chances than you’re conceding you should be doing OK.

This is where it falls down, unless you've got Hawkeye and loads of other technology to calculate trajectory of cross, speed of cross, angle the player is approaching from, the speed the player is approaching, the height of the cross, the part of then body used, the positioning of the keeper, defenders, pressure applied on the ball, where there any outfield players between the ball and the goa, etc, etc.

Sent from my Pixel 7 using Tapatalk

 

[Frostie]

I don’t know the ins and outs of each model, I’m not sure if that info is available. Someone like @Frostie might know.

I doubt the models update game by game, though maybe.

But generally there’s no link. The idea is to capture whether you’re missing/scoring more than you “should” over a period. For me totals are where things get tricky as 7 0.1 attempts might not really be equal to one 0.7 attempt (eg a pen). But overall on average if you’re consistently having better chances than you’re conceding you should be doing OK.

Yeah, that's basically it. The whole point of xG is to be utilised over a period of time, the longer the better really to analyse. Single game xG & single chance xG will be massively flawed for the reasons everyone is pointing out in this thread. Over a long enough dataset though they are proven to be very accurate hence being adopted by pretty much every pro club nowadays.

These xG Tables take your xG For & Against in each game & decide who the most likely winner was to award Expected Points. If the xG Totals are close, they award it as a draw.
Again, on an individual game basis there is obvious flaws to this i.e. one team having a penalty or very high probability chance but little else vs an opponent having lots of lower probability chances. Also, biggest factor would be game state, i.e. a team taking an early lead & then sitting back will usually result in the opponent amassing more chances, even if not particularly good ones. Again though, over a long enough period of time there is obvious merit to them & we can start to see trends emerging.

 

[stupot07]

Yeah, that's basically it. The whole point of xG is to be utilised over a period of time, the longer the better really to analyse. Single game xG & single chance xG will be massively flawed for the reasons everyone is pointing out in this thread. Over a long enough dataset though they are proven to be very accurate hence being adopted by pretty much every pro club nowadays.

These xG Tables take your xG For & Against in each game & decide who the most likely winner was to award Expected Points. If the xG Totals are close, they award it as a draw.
Again, on an individual game basis there is obvious flaws to this i.e. one team having a penalty or very high probability chance but little else vs an opponent having lots of lower probability chances. Also, biggest factor would be game state, i.e. a team taking an early lead & then sitting back will usually result in the opponent amassing more chances, even if not particularly good ones. Again though, over a long enough period of time there is obvious merit to them & we can start to trends emerging.

Thanks for this. So basically the purpose of the OP, it's bollocks. But if you look at it over half a season or a season as a whole/total/average it is pretty useful!


Sent from my Pixel 7 using Tapatalk

 

[fernandopartridge]

Thanks for this. So basically the purpose of the OP, it's bollocks. But if you look at it over half a season or a season as a whole/total/average it is pretty useful!


Sent from my Pixel 7 using Tapatalk

16 games is a decent sample size.

Like my post said yesterday, look at the league table, the teams with a superior xg in more of their games are generally in the top half, teams with a superior xg in less games are generally in the bottom half.

Sent from my Pixel 7 using Tapatalk

 

[Calista]

Looks like a load of Bollocks to me. Overall as an average, our xG for is 1.46 per game and our XG against is 1.25 per game. The difference is negligible really.

It's far from negligible over 16 games though. It's easily enough to suggest we 'should' have got the extra couple of wins that would have us around 11th place.

 

[Calista]

To give a toy example. Imagine all shots are equal. We live in a fantasy land with no defenders or goal keepers and each ball is hit from a dead ball. All players are equally good with whatever foot they’re using as each other. So the only variable that matters is where the shot is taken from.

Now we get 100 players to take a shot from various positions (edge of the D, centre circle, by the corner flag) and record how many go in. This gives you a probability for each location you tried. You could then use various statistical techniques like a voronoi diagram or some form of clustering to workout the probability for any future shot in this god awful one sided dead ball game we’ve invented.

I'm starting to worry about you shmmeee. Are you getting enough "feels"?

 

[stupot07]

It's far from negligible over 16 games though. It's easily enough to suggest we 'should' have got the extra couple of wins that would have us around 11th place.

Depends on where you score those goals, it may have been turning a few more losses into drawers rather than draws into wins. Quick maths xG is 23 goals instead of the 19 we have scored, we could have conceivably scored them and still have the same points.

Win Blackburn 2-0 instead of 1-0
Win Middlesbrough 4-0 instead of 3-0
Lost Rotherham's 2-1 instead of 2-0
Lost WBA 2-1 instead of 2-0

We meet our xG and still have the same points.

Some other interesting facts is we have lost all 6-7 games where we have dominated possession (60%+), yes in all 3 wins the opposition have had the most possession.

Sent from my Pixel 7 using Tapatalk

 

[Frostie]

Depends on where you score those goals, it may have been turning a few more losses into drawers rather than draws into wins. Quick maths xG is 23 goals instead of the 19 we have scored, we could have conceivably scored them and still have the same points.

Win Blackburn 2-0 instead of 1-0
Win Middlesbrough 4-0 instead of 3-0
Lost Rotherham's 2-1 instead of 2-0
Lost WBA 2-1 instead of 2-0

We meet our xG and still have the same points.

Sent from my Pixel 7 using Tapatalk

Which is why it's calculated by xPts from each game.

 

[Frostie]

I always find this interesting & a good explanation from Rasmus Ankersen (then DoF at Brentford) for those new to data analytics:

 

[Legia Sky Blue]

Cheers. How is the data collated and how does our XG in one game have any bearing to the next game. I understand XG now in isolation but I can’t get to grips with for example how many easier shots we had against one team has any bearing on what shots we might have against another team.

I'm cetainly no expert, but as far as I'm aware the XG from one game has no bearing on when you play someone else. The league table at the start of this thread therefore gives us wins for games where we had a higher XG in a match directly against our opponents, and losses for those we had as lower XG. In terms of particular games, I presume for instance with the Stoke game they would have us expected to win 1-0 based on the Godden chance when he had a free header. I know that it is more complicated than that, so that other chances in the games for both sides would also have had some mathematical input, but the Godden chance probably had the highest XG because it was the one you would normally be more likely to score from.

The most interesting aspect to me is the individual XG for our forwards which shows they have all been profligate in front of goal so far this season, and that ordinarily on the chances they've had should already have scored about 20 between them.

 

[stupot07]

I'm cetainly no expert, but as far as I'm aware the XG from one game has no bearing on when you play someone else. The league table at the start of this thread therefore gives us wins for games where we had a higher XG in a match directly against our opponents, and losses for those we had as lower XG. In terms of particular games, I presume for instance with the Stoke game they would have us expected to win 1-0 based on the Godden chance when he had a free header. I know that it is more complicated than that, so that other chances in the games for both sides would also have had some mathematical input, but the Godden chance probably had the highest XG because it was the one you would normally be more likely to score from.

The most interesting aspect to me is the individual XG for our forwards which shows they have all been profligate in front of goal so far this season, and that ordinarily on the chances they've had should already have scored about 20 between them.

Even that chances was only a 0.23 xG, so would only be scored once in every 4. Andre Vidigals chances in the 59th minute was the best of the game at 0.29 xGA. I couldn't even remember it

Sent from my Pixel 7 using Tapatalk

 

[shmmeee]

I'm starting to worry about you shmmeee. Are you getting enough "feels"?

I am and I can show you the excel tracking sheet to prove it!

 

[stupot07]

Interesting looking at Wright's 4 goals. (xG 6.38)

Middlesbrough 0.15 xG
Blackburn 0.37
Preston 0.95
Preston 0.86

That's 2 sitters

Godden (xG 6.37, 5 goals)

Middlesbrough 0.31
Swansea 0.08
Watford 0.14
Cardiff 0.09
Cardiff 0.52

Simms (xG 3.19, 2 goals)

QPR 0.28
QPR 0.41




Sent from my Pixel 7 using Tapatalk

 

[Legia Sky Blue]

Even that chances was only a 0.23 xG, so would only be scored once in every 4. Andre Vidigals chances in the 59th minute was the best of the game at 0.29 xGA. I couldn't even remember it

Sent from my Pixel 7 using Tap

If that is the case then I seriously question the value of the data being entered. I presume Vidigal's chance was the overhead kick where had to twist and turn after the cross went over his head and behind him? That was a difficult chance at best, not quite like Godden's free header in the middle of the goal where he shouldered it wide, which was a glaring miss.

 

[Frostie]

If that is the case then I seriously question the value of the data being entered. I presume Vidigal's chance was the overhead kick where had to twist and turn after the cross went over his head and behind him? That was a difficult chance at best, not quite like Godden's free header in the middle of the goal where he shouldered it wide, which was a glaring miss.

Vidigal is 3 yards out & unmarked with keeper out of position. He could have just headed it into a pretty empty net tbf but, for some reason, decided to turn his back & go for the spectacular. Couldn't believe what I was seeing. Surprised his manager didn't hook him there & then.

 

[Legia Sky Blue]

Ok, I only saw it live and haven't seen any replays of either chance. As it happened I thought their player was at the back post but that the cross was played behind him hence the overhead kick, while Godden's was a free header in the middle of the 6 yard box in the centre of the goal.

 

[Macca]

So basically are strikers are shit. We’re creating enough chances to be in the playoffs but we’re 5th bottom because we can’t score.

Yep. not sure it required any fancy science :-D

 

[mmttww]

So basically are strikers are shit. We’re creating enough chances to be in the playoffs but we’re 5th bottom because we can’t score.

...and give away goals that teams don't have to work hard to create by making dumbsh*t mistakes at the back.

 

[Frostie]

Ok, I only saw it live and haven't seen any replays of either chance. As it happened I thought their player was at the back post but that the cross was played behind him hence the overhead kick, while Godden's was a free header in the middle of the 6 yard box in the centre of the goal.

Either way, brings us back to the point of why single shot xG is not reliable.