xG so far this season

[And Its Smith]

I am sure a lot of people are becoming more aware of xG and I am hoping someone who is well versed on it, and believes in it, can convince me that it isn't a pile of tosh! 

I listen to a lot of podcasts and read a lot about football and increasingly people are talking about xG as a more factual way to tell how well your team has played.  Some people seem to treat it as the 'justice league' i.e. the league without luck (I guess) and think it is more truthful than the league table.

We are 7th at the moment yet the xG table has us at 20th.  Other wild differences in the xG league are Swansea 14th (2nd in real life), Birmingham 24th (8th in real) and Stoke 4th (24th in real table).

Here are our results so far this season.  First of all the real result and then the xG result:

Leeds 1-3 loss......0.76 - 2.01 loss

Birmingham 1-1 draw...….0.62 - 1.36 win

QPR 2-0 win..….1.21 - 1.58 draw

Derby 2-1 win…...2.23 - 1.13 loss

Its the latest game that I would like people who believe in xG's opinion.  I watched the game at the ground and have seen it again since.  How does xG calculate that we deserved to lose that game?  xG seems to suggest that they deserved to win it because of the chances that fell to Tom Huddlestone and Tom Lawrence in the first half.  Giving those chances as high probability ones but then it gives Brownhill's goal as a 7% chance 

Is xG reliable and how on earth does it come up with its percentages for each chance? 

[IdliketoRogerMoore]

some People got way to much time on there hands to devise a new formulas and tables! 

Mid prefer to stick to the traditional table! 

Additional analysis is great but giving points based on someone’s opinion of a chance is going a bit far for me!

[Red DNA]

Complete over-analysis, some analytics are undoubtedly useful but I struggle to see what this offers other than an opinion based on ‘ifs’ and ‘buts’. 

Sometimes you get luck, other times you don’t, it’s life.

Football used to be such a simple game

 

[Major Isewater]

13 minutes ago, RedDave said:

I am sure a lot of people are becoming more aware of xG and I am hoping someone who is well versed on it, and believes in it, can convince me that it isn't a pile of tosh! 

I listen to a lot of podcasts and read a lot about football and increasingly people are talking about xG as a more factual way to tell how well your team has played.  Some people seem to treat it as the 'justice league' i.e. the league without luck (I guess) and think it is more truthful than the league table.

We are 7th at the moment yet the xG table has us at 20th.  Other wild differences in the xG league are Swansea 14th (2nd in real life), Birmingham 24th (8th in real) and Stoke 4th (24th in real table).

Here are our results so far this season.  First of all the real result and then the xG result:

Leeds 1-3 loss......0.76 - 2.01 loss

Birmingham 1-1 draw...….0.62 - 1.36 win

QPR 2-0 win..….1.21 - 1.58 draw

Derby 2-1 win…...2.23 - 1.13 loss

Its the latest game that I would like people who believe in xG's opinion.  I watched the game at the ground and have seen it again since.  How does xG calculate that we deserved to lose that game?  xG seems to suggest that they deserved to win it because of the chances that fell to Tom Huddlestone and Tom Lawrence in the first half.  Giving those chances as high probability ones but then it gives Brownhill's goal as a 7% chance 

Is xG reliable and how on earth does it come up with its percentages for each chance? 

There’s only one table that counts .

 

[handsofclay]

We obviously sat back in the second period v Derby and but for a wonder goal comfortably absorbed everything they threw at us.  I take it the fact they were 'allowed'  ascendancy in the second half means this xG has them down as being unlucky not to have won so gives them the win by 1.1 goals, a bigger margin than the 1 goal we ACTUALLY won by.

[BobBobSuperBob]

As a dinosaur I’m amazed at the complete modern obsession with stats

Useful in some small particular areas  but personally I’d rather trust my eyes , occasionally referring back to certain stats as a comparison to what I believe in seeing

And as for XG , As a Head Coach I’d much rather see a few mins of compilation of chances than take credence from a XG Score or map

IMHO

[cityexile]

I think what it ends up measuring is simply 'how good where you at taking your chances'. Birmingham for instance it shows us as winning, and most of us would have felt 'if only Afobe had buried one of those really good openings'. Equally against Derby I certainly came away from the ground thinking we were very clinical with our finishing. Interesting at that level, but cannot see that it gives you a 'better' real picture. There is no reason you should revert to the mean if your strikers are better finishers on average.

[And Its Smith]

2 minutes ago, handsofclay said:

We obviously sat back in the second period v Derby and but for a wonder goal comfortably absorbed everything they threw at us.  I take it the fact they were 'allowed'  ascendancy in the second half means this xG has them down as being unlucky not to have won so gives them the win by 1.1 goals, a bigger margin than the 1 goal we ACTUALLY won by.

On the contrary actually.  xG gives Derby no positives from their second half display as they created no clear cut chances.  

Their four big chances were all in the first half.  What I find weird is it doesn't give any importance (or doesn't seem too) for a good save or block tackle.  It only seems to concentrate on the attacking team.  I'm genuinely confused by it and want to understand it, even if I don't end up agreeing with it.

So far it seems everyone thinks its nonsense which is hard to argue with!!

[And Its Smith]

2 minutes ago, cityexile said:

I think what it ends up measuring is simply 'how good where you at taking your chances'. Birmingham for instance it shows us as winning, and most of us would have felt 'if only Afobe had buried one of those really good openings'. Equally against Derby I certainly came away from the ground thinking we were very clinical with our finishing. Interesting at that level, but cannot see that it gives you a 'better' real picture. There is no reason you should revert to the mean if your strikers are better finishers on average.

Weirdly it gives Afobe's two chances as exactly the same probability of a goal at 35% but then gives Nagy chance as 37%.  

[Mr Popodopolous]

It can be useful IMO, in showing underlying trends- for example I am quite sure we were outperforming our xG average in 2016/17 in the early stages- and this was reflected on the pitch. On the flipside, some underlying numbers during the awful run that LJ was lucky to survive showed an improvement.

Leeds however xG based should have finished if not top last season- had most shots on goal, some impressive numbers shots conceded, possession, dominance- yet finished 3rd and lost it towards the end.

In summary, think it is a useful guide to underlying form, performance level and perhaps a guide of future things to come- but that it has its limits.

[semblar]

Another point with this is whether you can actually apply this kind of statistical analysis to a single game with any expectation of accuracy. For a large number of games (not necessarily a whole season) I can see the approach, but I would not call four games a large enough pool to make a comparison with, especially given the change in personnel after the Leeds match 

[cidered abroad]

xG?

What is this mumbo jumbo?

No, please don't try to explain it to me. Leave me to my, old age, well past best before date, ignorance.

[And Its Smith]

3 minutes ago, Mr Popodopolous said:

It can be useful IMO, in showing underlying trends- for example I am quite sure we were outperforming our xG average in 2016/17 in the early stages- and this was reflected on the pitch. On the flipside, some underlying numbers during the awful run that LJ was lucky to survive showed an improvement.

Leeds however xG based should have finished if not top last season- had most shots on goal, some impressive numbers shots conceded, possession, dominance- yet finished 3rd and lost it towards the end.

In summary, think it is a useful guide to underlying form but that it has its limits.

My argument against it though would be that the xG percentages are complete nonsense which makes a mockery of the whole thing.

If the basis is flawed then it is useless.  How can they say we should have lost on Tuesday?  That's what I'm mainly trying to understand

[mozo]

19 minutes ago, RedDave said:

I am sure a lot of people are becoming more aware of xG and I am hoping someone who is well versed on it, and believes in it, can convince me that it isn't a pile of tosh! 

I listen to a lot of podcasts and read a lot about football and increasingly people are talking about xG as a more factual way to tell how well your team has played.  Some people seem to treat it as the 'justice league' i.e. the league without luck (I guess) and think it is more truthful than the league table.

We are 7th at the moment yet the xG table has us at 20th.  Other wild differences in the xG league are Swansea 14th (2nd in real life), Birmingham 24th (8th in real) and Stoke 4th (24th in real table).

Here are our results so far this season.  First of all the real result and then the xG result:

Leeds 1-3 loss......0.76 - 2.01 loss

Birmingham 1-1 draw...….0.62 - 1.36 win

QPR 2-0 win..….1.21 - 1.58 draw

Derby 2-1 win…...2.23 - 1.13 loss

Its the latest game that I would like people who believe in xG's opinion.  I watched the game at the ground and have seen it again since.  How does xG calculate that we deserved to lose that game?  xG seems to suggest that they deserved to win it because of the chances that fell to Tom Huddlestone and Tom Lawrence in the first half.  Giving those chances as high probability ones but then it gives Brownhill's goal as a 7% chance 

Is xG reliable and how on earth does it come up with its percentages for each chance? 

I think this is a fascinating post @RedDave. Firstly because I didn't know what our xg scores have been, but secondly because you raise a great point about the efficacy of this analysis. 

My intuition is that xg doesn't analyse who the chances fall to and a myriad of variables that are possibly not factored in such as did the ball fall kindly to the player on the their stronger foot, did it roll cleanly on the turf, was their sight of goal obstructed, how many seconds did they have to set themself etc etc. I need to learn more about what details are and aren't considered.

[Davefevs]

19 minutes ago, RedDave said:

I am sure a lot of people are becoming more aware of xG and I am hoping someone who is well versed on it, and believes in it, can convince me that it isn't a pile of tosh! 

I listen to a lot of podcasts and read a lot about football and increasingly people are talking about xG as a more factual way to tell how well your team has played.  Some people seem to treat it as the 'justice league' i.e. the league without luck (I guess) and think it is more truthful than the league table.

We are 7th at the moment yet the xG table has us at 20th.  Other wild differences in the xG league are Swansea 14th (2nd in real life), Birmingham 24th (8th in real) and Stoke 4th (24th in real table).

Here are our results so far this season.  First of all the real result and then the xG result:

Leeds 1-3 loss......0.76 - 2.01 loss

Birmingham 1-1 draw...….0.62 - 1.36 win

QPR 2-0 win..….1.21 - 1.58 draw

Derby 2-1 win…...2.23 - 1.13 loss

Its the latest game that I would like people who believe in xG's opinion.  I watched the game at the ground and have seen it again since.  How does xG calculate that we deserved to lose that game?  xG seems to suggest that they deserved to win it because of the chances that fell to Tom Huddlestone and Tom Lawrence in the first half.  Giving those chances as high probability ones but then it gives Brownhill's goal as a 7% chance 

Is xG reliable and how on earth does it come up with its percentages for each chance? 

As a stats man, I love my data....but you have to take it with a pinch of salt....especially if you’re going to use a 4 game sample.  Even 46 games is a statistically small sample.

xG is evolving its scoring accuracy, but until it can take lots of other things into account like, positioning if defenders, strong or weak foot, distance from ground, keeper’s save ratios from that spot etc etc, it’s always gonna be a “useful” metric but not the be-all and end-all.

What should we take from xG?

That Benik Afobe’s two one-on-ones v Brum are not gilt-edged chances, xG suggests both were a 1:5 ratio, so anyone saying a striker should take every one-on-one is not understanding that it is difficult.

A penalty is about 0.78 xG....so circa 1 in 5 is missed.

I do like the Stoke data....there is a guy, big Stoke analyst, who is trying to convince everyone that Stoke and Nathan Jones is about to turn everything around.  He might be right, but we know as football fans, when your team is on a downer, you concede all manner of shit goals, e.g. OGs, keeper errors, deflections etc that don’t reflect the balance of the game, and at the other end, keepers have worldies, the ball hits the wrong side of the post and bounces out rather than in etc.

Football will never be data driven, it’s too fluid a game....but data can be used smartly to make good / better decision either on recruitment or tactics.

[chinapig]

A reasonably good explanation, made more so by the fact that it is historic so you can see the extent to which it has any predictive capability.

https://www.telegraph.co.uk/football/2017/07/25/expected-goals-xg-does-show-man-city-should-win-premier-league/

Not that any sensible person relies on a single metric, or indeed solely on metrics, in any business. Though I often find that managers want the analytics to make decisions for them, there is no magic number.

[CiderCraig]

18 minutes ago, Major Isewater said:

There’s only one table that counts .

 

Is it this one? 

[mozo]

3 minutes ago, Davefevs said:

As a stats man, I love my data....but you have to take it with a pinch of salt....especially if you’re going to use a 4 game sample.  Even 46 games is a statistically small sample.

xG is evolving its scoring accuracy, but until it can take lots of other things into account like, positioning if defenders, strong or weak foot, distance from ground, keeper’s save ratios from that spot etc etc, it’s always gonna be a “useful” metric but not the be-all and end-all.

What should we take from xG?

That Benik Afobe’s two one-on-ones v Brum are not gilt-edged chances, xG suggests both were a 1:5 ratio, so anyone saying a striker should take every one-on-one is not understanding that it is difficult.

A penalty is about 0.78 xG....so circa 1 in 5 is missed.

I do like the Stoke data....there is a guy, big Stoke analyst, who is trying to convince everyone that Stoke and Nathan Jones is about to turn everything around.  He might be right, but we know as football fans, when your team is on a downer, you concede all manner of shit goals, e.g. OGs, keeper errors, deflections etc that don’t reflect the balance of the game, and at the other end, keepers have worldies, the ball hits the wrong side of the post and bounces out rather than in etc.

Football will never be data driven, it’s too fluid a game....but data can be used smartly to make good / better decision either on recruitment or tactics.

I love the data too, but you've done a great job there of explaining why xg is deeply flawed!

[And Its Smith]

5 minutes ago, Davefevs said:

As a stats man, I love my data....but you have to take it with a pinch of salt....especially if you’re going to use a 4 game sample.  Even 46 games is a statistically small sample.

xG is evolving its scoring accuracy, but until it can take lots of other things into account like, positioning if defenders, strong or weak foot, distance from ground, keeper’s save ratios from that spot etc etc, it’s always gonna be a “useful” metric but not the be-all and end-all.

What should we take from xG?

That Benik Afobe’s two one-on-ones v Brum are not gilt-edged chances, xG suggests both were a 1:5 ratio, so anyone saying a striker should take every one-on-one is not understanding that it is difficult.

A penalty is about 0.78 xG....so circa 1 in 5 is missed.

I do like the Stoke data....there is a guy, big Stoke analyst, who is trying to convince everyone that Stoke and Nathan Jones is about to turn everything around.  He might be right, but we know as football fans, when your team is on a downer, you concede all manner of shit goals, e.g. OGs, keeper errors, deflections etc that don’t reflect the balance of the game, and at the other end, keepers have worldies, the ball hits the wrong side of the post and bounces out rather than in etc.

Football will never be data driven, it’s too fluid a game....but data can be used smartly to make good / better decision either on recruitment or tactics.

Thanks for this.  Confirming to me that it is not really worth looking at.  I've not seen Stoke this season but I get the impression that if Butland was even playing to 75% of his capabilities they would be a lot higher. 

The Derby game and what you have written have confirmed my suspicion that xG isn't good enough a model to put faith in.  I guess the only thing it can be used for is to show you how many chances you are creating as a team, and how good those chances are.  To use to it say that Derby should have won is nonsense as Bentley's save and Kalas tackle in the first half are two of the big reasons that we won and cant just be ignored. 

[Davefevs]

3 minutes ago, RedDave said:

My argument against it though would be that the xG percentages are complete nonsense which makes a mockery of the whole thing.

If the basis is flawed then it is useless.  How can they say we should have lost on Tuesday?  That's what I'm mainly trying to understand

That is partly the problem, I don’t think the stats purists are saying we should have lost on Tuesday.  However stats amateurs (like myself) are using one metric to try to do just that.

To simplify, If we take a shot from the edge of the box centrally, it will have a low xG, because of all of the goals from that spot are added up and divided by the total of the shots.  Let’s say the xG was 0.1.  However if on Tuesday that shot from that spot was an open goal because Afobe had nipped the ball around the onrushing keeper, you’d expect him to roll it into the net 99/100.  The xG isn’t sophisticated enough to know that, so it’s still counted as 0.1.

Now play that out with Weimann’s goal.  It doesn’t know that the keeper is on his arse, no one is challenging him, so in real terms it was a gilt-edged chance.  But in xG it’s a left foot shot from 6 yards.

[And Its Smith]

3 minutes ago, Davefevs said:

That is partly the problem, I don’t think the stats purists are saying we should have lost on Tuesday.  However stats amateurs (like myself) are using one metric to try to do just that.

To simplify, If we take a shot from the edge of the box centrally, it will have a low xG, because of all of the goals from that spot are added up and divided by the total of the shots.  Let’s say the xG was 0.1.  However if on Tuesday that shot from that spot was an open goal because Afobe had nipped the ball around the onrushing keeper, you’d expect him to roll it into the net 99/100.  The xG isn’t sophisticated enough to know that, so it’s still counted as 0.1.

Now play that out with Weimann’s goal.  It doesn’t know that the keeper is on his arse, no one is challenging him, so in real terms it was a gilt-edged chance.  But in xG it’s a left foot shot from 6 yards.

So does it just look at the shot?  If so that is pretty poor.

Weimann still had a bit to do and his turn put the keeper on his arse.  It gives it as a 70% chance which seems high.  He still had a bit of work to do. 

Do you listen to Not The Top 20 podcast?  Its a superb listen but they put a lot of faith in xG 

[phantom]

17 minutes ago, cidered abroad said:

xG?

What is this mumbo jumbo?

No, please don't try to explain it to me. Leave me to my, old age, well past best before date, ignorance.

I'm glad that it is not just me !

I have no idea what xG is either. . . 

[WarksRobin]

How’s the packing?

[BobBobSuperBob]

8 minutes ago, Davefevs said:

That is partly the problem, I don’t think the stats purists are saying we should have lost on Tuesday.  However stats amateurs (like myself) are using one metric to try to do just that.

To simplify, If we take a shot from the edge of the box centrally, it will have a low xG, because of all of the goals from that spot are added up and divided by the total of the shots.  Let’s say the xG was 0.1.  However if on Tuesday that shot from that spot was an open goal because Afobe had nipped the ball around the onrushing keeper, you’d expect him to roll it into the net 99/100.  The xG isn’t sophisticated enough to know that, so it’s still counted as 0.1.

Now play that out with Weimann’s goal.  It doesn’t know that the keeper is on his arse, no one is challenging him, so in real terms it was a gilt-edged chance.  But in xG it’s a left foot shot from 6 yards.

I remember LJ at the first fans forum he did talking about data and stating after each game he received a pack which he indicated was a couple of inches thick  to digest

id be fascinated to know how much weight he puts to this data and whether as his experience grows , if he’s changed in the amount he looks at or digests

I fear, due to an obvious interest in learning ,  he may or may have been be vulnerable to data overload  

He grew up playing in a professional academy , has been in the game his whole life, the added bonus of his Dad being in management , played in hundreds and seen thousands of games ..... I’d be amazed if theirs many stats that would tell him , JM , DH something , or certainly anything significant , they hadn’t already noticed or were aware of 

[Davefevs]

2 minutes ago, RedDave said:

So does it just look at the shot?  If so that is pretty poor.

Weimann still had a bit to do and his turn put the keeper on his arse.  It gives it as a 70% chance which seems high.  He still had a bit of work to do. 

Do you listen to Not The Top 20 podcast?  Its a superb listen but they put a lot of faith in xG 

It has a bit more info than the simplicity in which I explain it.

I have started listening to NTT20, and although I get more intel about Lg1/2 than I would from watching Quest highlights and following our loanee player club’s twitter, I find them a bit “superior” in their views.  Maybe because I don’t necessarily agree with some of their thoughts / comments.

There are a few data people on twitter who are excellent, but there are some who are no better than those ill-informed ex-pro pundits....using stats as the be-all and end-all....you wonder if they are watching the game.

I had a friendly argument on Twitter with a Brum guy (Gabriel Sutton), who’s quite respected.  In our 1-1 draw his vlog said it was an even game and a draw was fair, they had the better of the play, we had the better chances.  I then looked at the player ratings and Brum were all 7/8/9 - totalled something like 85, ours were all 6s/odd-7, totalled something like 67.

I challenged his could all the Brum boys be so high and ours so low in an even game.

Its all a bit of bollox in some regards.  Use data, fine, but trust your eyes too.

[And Its Smith]

3 minutes ago, Davefevs said:

It has a bit more info than the simplicity in which I explain it.

I have started listening to NTT20, and although I get more intel about Lg1/2 than I would from watching Quest highlights and following our loanee player club’s twitter, I find them a bit “superior” in their views.  Maybe because I don’t necessarily agree with some of their thoughts / comments.

There are a few data people on twitter who are excellent, but there are some who are no better than those ill-informed ex-pro pundits....using stats as the be-all and end-all....you wonder if they are watching the game.

I had a friendly argument on Twitter with a Brum guy (Gabriel Sutton), who’s quite respected.  In our 1-1 draw his vlog said it was an even game and a draw was fair, they had the better of the play, we had the better chances.  I then looked at the player ratings and Brum were all 7/8/9 - totalled something like 85, ours were all 6s/odd-7, totalled something like 67.

I challenged his could all the Brum boys be so high and ours so low in an even game.

Its all a bit of bollox in some regards.  Use data, fine, but trust your eyes too.

Exactly.  

Agree that they are a bit superior. Better than The Second Tier podcast though who just seem clueless. 

I think a lot of people will use stats so they can have opinions on games they haven’t seen.

i am sure a lot of people over the age of 30 has watched City over 300 times and probably watched 1000+ games either live or on TV.  Eyes beat stats any day 

[the red rules]

40 minutes ago, cidered abroad said:

xG?

What is this mumbo jumbo?

No, please don't try to explain it to me. Leave me to my, old age, well past best before date, ignorance.

Mumbo Jumbo to me too. But I must admit to looking it up reading this thread and trying to feel not past it.              Wish I hadn't bothered, you're not missing much remaining ignorant.

 

 

[Davefevs]

9 minutes ago, RedDave said:

Exactly.  

Agree that they are a bit superior. Better than The Second Tier podcast though who just seem clueless. 

I think a lot of people will use stats so they can have opinions on games they haven’t seen.

i’m amazed at the “detailed” opinions of some people on so many players.  The guy I refer to above was suggesting Tommy Rowe would be better suited to LCB than Nathan Baker.  FFS!!  I have a pretty photographic memory, but you’d have to consume a helluva lot of games to really have any real say in how good a player is or isn’t, and then suggest he could play a position better than someone else, especially as he’s never played LCB!  I suspect a lot have played a lot of FM / FIFA.

i am sure a lot of people over the age of 30 has watched City over 300 times and probably watched 1000+ games either live or on TV.  Eyes beat stats any day 

The problem is that our memory starts to fade!!

 

[chinapig]

6 minutes ago, RedDave said:

Exactly.  

Agree that they are a bit superior. Better than The Second Tier podcast though who just seem clueless. 

I think a lot of people will use stats so they can have opinions on games they haven’t seen.

i am sure a lot of people over the age of 30 has watched City over 300 times and probably watched 1000+ games either live or on TV.  Eyes beat stats any day 

The problem with your eyes is that they are connected to your brain!

We are all subject to bias so if we have a preconceived opinion of a player or team we are likely to ignore evidence that doesn't support our opinion. So if we want to be objective we need more than just the evidence of our own eyes.

Not a big deal if you are just an interested observer but more so if you are responsible for making decisions.

[cidered abroad]

21 minutes ago, RedDave said:

Exactly.  

Agree that they are a bit superior. Better than The Second Tier podcast though who just seem clueless. 

I think a lot of people will use stats so they can have opinions on games they haven’t seen.

i am sure a lot of people over the age of 30 has watched City over 300 times and probably watched 1000+ games either live or on TV.  Eyes beat stats any day 

I've never counted them  but it could be over 2,000.

My problem now is that I clearly remember fifty years ago but yesterday is a bit foggy.

How did we get on Tuesday evening at Derby? Stuffed again and probably had Norman Hunter red carded for chopping Franny Lee?