xG - hype or valuable stat?

[MadAmster]

As an old school type, I obviously go for hype. Is it a more valuable stat than shots, shots on target? I don't think it is.

Here's some examples from last weekend

Luton 1.75 Sheff U 1.56. The result was 1-3
Villa 2.45 Utd 1.9. The result was 1-2
Red Dogs 1.25 Toon 0.68. The result was 2-3
Fulham 1.75 B'mouth 2.61. The result was 3-1

Shots/shots on were

Luton 20/5 Sheff U 7/3.
Villa 23/10 Utd 17/5.
Red Dogs 13/3 Toon 7/5.
Fulham 7/6 B'mouth 25/4.

Look at xG and, in all 4 cases the team with the better xG lost. Look at shots and the team with the most shots lost. Shots on target sees Luton and Villa losing despite more SOT and then Toon and Fulham having more SOT and winning.

IMO xG is no better a stat than shots/SOT. In the end, it all comes down to creating chances, making the decision to shoot and then whether it's off target, on target but saved/blocked or on target and in.

In the interest of balance, there were other games where the side with the best xG won.

[duncanjwitham]

10 minutes ago, MadAmster said:

IMO xG is no better a stat than shots/SOT. In the end, it all comes down to creating chances, making the decision to shoot and then whether it's off target, on target but saved/blocked or on target and in.

That only holds true if you think that all shots (or shots on target) are equally as likely to result in goals.  As soon as you accept that some shots are more likely to result in goals than other shots, then xG is potentially a better stat that just shots on its own.

There are absolutely some types of games where shots (on target) is going to be a very good predictor of the actual outcome, because in general the better team is more likely to have more shots and more likely to score some of them.  But there are also games where that is not going to be true.  For example, games where you've got one team dominating possession against a team sitting in a low block (which we've seen plenty of over the years!), the team on the ball is probably going to have a fair few shots, but many of them will be pot-shots from distance or snatched half chances under pressure, that are unlikely to result in goals.  The other team is going to have much fewer shots, but they are probably going to be much better chances, where they've won the ball back an counter-attacked or something. 

xG is just an attempt to factor all that stuff in.  You can see it in the scores/stats you've posted above.  Luton had nearly 3 times as many shots as Sheff Utd, but the xG is almost the same.  That's very clearly saying that Luton were creating a lot of poor chances, vs Sheff Utd creating a few very good ones.  The Fulham vs Bournemouth game is similar.  I haven't watched those games, but those xG stats are telling more about the game than just the shots ones on their own are.  xG is certainly not perfect, and there's definitely an element of opinion in it (someone is having to make a value judgment somewhere about what a 'good' chance is, even if that is just down to the choice of methodology for calculating it), but to me, it's telling me stuff I didn't already know, so its useful.

[Jubbs]

I could speak for ages on this and the importance of looking at underlying stats etc.

But as a basic view, xG in terms of individual games is good surface level but over the course of a season, gives a much better picture.

[Returning ram]

Like all stats, they can be useful with context. Unfortunately like with most stats, they are often used without it to try to prove a point.

[Day]

Absolute nonsense stat that needs to get in the bin.

[Ghost of Clough]

22 minutes ago, MadAmster said:

As an old school type, I obviously go for hype. Is it a more valuable stat than shots, shots on target? I don't think it is.

Here's some examples from last weekend

Luton 1.75 Sheff U 1.56. The result was 1-3
Villa 2.45 Utd 1.9. The result was 1-2
Red Dogs 1.25 Toon 0.68. The result was 2-3
Fulham 1.75 B'mouth 2.61. The result was 3-1

Shots/shots on were

Luton 20/5 Sheff U 7/3.
Villa 23/10 Utd 17/5.
Red Dogs 13/3 Toon 7/5.
Fulham 7/6 B'mouth 25/4.

Look at xG and, in all 4 cases the team with the better xG lost. Look at shots and the team with the most shots lost. Shots on target sees Luton and Villa losing despite more SOT and then Toon and Fulham having more SOT and winning.

IMO xG is no better a stat than shots/SOT. In the end, it all comes down to creating chances, making the decision to shoot and then whether it's off target, on target but saved/blocked or on target and in.

In the interest of balance, there were other games where the side with the best xG won.

xG is a useful tool if used correctly. Give Chris Martin and Connor Sammon exactly the same chance and they'll have the same xG, but one of them will be more liekly to score as he's a better finisher. Unsurprisingly, the teams who create a high proportion of chances for their best finishers also outperform their xG as a team, whilst the team creating chances for poorer finishers underperform.

xG is a better metric than shots as it gives meaning to the shot - the chance of scoring from 1 yard out will be significantly greater than 40 yards out.

[AndyinLiverpool]

28 minutes ago, Jubbs said:

I could speak for ages on this and the importance of looking at underlying stats etc.

But as a basic view, xG in terms of individual games is good surface level but over the course of a season, gives a much better picture.

A much better picture of what? Whether I'm likely to enjoy a game? It seems to me it's only useful for coaching and betting. If you do neither, it gives a picture of nothing.

Yet Live broadcasters yum this stuff up, despite the fact that anyone who wants this info can go to their bookie's website.

[Ian Buxton's Bat]

I quite like XG and as @Ghost of Clough says, if you understand it, it can tell you quite a bit about how a team is performing in terms of creating chances and how they (and individuals within the team) are performing in terms of taking them.

And........it keeps me reasonably balanced re Warne and the in/out and awful/brilliant debates.

My position re Warne is that I think the football is pretty grim to watch but he definitely knows what he is doing.

Our XG for is 1.46 per game and we're scoring 1.74 per game.........outperforming it with only one striker available

Our XG against is 1.07 per game and we're conceding 0.94 per game........again goalies and defenders must be doing a lot right

If you then take Peterborough....

Their XG for is 1.83.......clearly demonstrating they are playing more attacking football over the season so far.....and they are scoring 1.9....just outperforming XG

Their XG against is 1.16 but they are conceding 1.26

What does that add up to?

I'd currently rather be Derby than Peterborough in terms of league position (+ stadium, toilets, kit, chairman etc etc) but, if were a neutral, wanting to be entertained, I'd go to Peterborough without a shadow of doubt.

And you a kick-the-ball-into-a-skip halftime competition thrown in.

I think XG is more of an entertainment factor on any given match day but becomes more meaningful over time.

[Gaspode]

Load of cobblers to satisfy the 'Football Manager' fanboys - this chap can explain it far better than me....

[Day]

24 minutes ago, Ghost of Clough said:

xG is a useful tool if used correctly. Give Chris Martin and Connor Sammon exactly the same chance and they'll have the same xG, but one of them will be more liekly to score as he's a better finisher. Unsurprisingly, the teams who create a high proportion of chances for their best finishers also outperform their xG as a team, whilst the team creating chances for poorer finishers underperform.

xG is a better metric than shots as it gives meaning to the shot - the chance of scoring from 1 yard out will be significantly greater than 40 yards out.

With football though, it’s not like snooker where you can line the balls up and have 2 players take the same shot.

xG fails to take into account the ability of the opposition defenders and keeper.

It fails to take into account weather conditions.

It fails to take into account the current situation in so many way, home, away, comfortable lead, behind, fitness level, carrying knocks.

If it was a game played between robots in an indoor arena I could see the value.

Comparing it to shots as a metric is nonsensical, shots and shots on target are facts. Indisputable facts.

Nothing more than a stat that with possession gives you a rough idea as to how the game as gone.

Never really see anyone give it much value as you see others try and give xG.

Just watch the game. You will know how it’s gone and who played better, who created the better chances to win the game and who’s finishing or mistakes let them down.

[Returning ram]

28 minutes ago, Ghost of Clough said:

xG is a useful tool if used correctly. Give Chris Martin and Connor Sammon exactly the same chance and they'll have the same xG, but one of them will be more liekly to score as he's a better finisher. Unsurprisingly, the teams who create a high proportion of chances for their best finishers also outperform their xG as a team, whilst the team creating chances for poorer finishers underperform.

xG is a better metric than shots as it gives meaning to the shot - the chance of scoring from 1 yard out will be significantly greater than 40 yards out.

Or the other team has better defenders or better keeper or if the ball takes a bobble, or etc etc etc.

It was designed for betting purposes, it's not accurate, watching a game is a better reflection than any metrics.

As per their website....Expected goals predicts the correct home team result 66% of the time and away results 58% of the time. This is slightly better than shots on target on the away results and slightly worse on the home results.

[nottingram]

Think I’ve said it before but if it was called something like “chance quality” instead of expected goals then you’d have a lot less people poo-pooing it for whatever reason they do.

It’s not the be all and end all but can add context if you want it to. Does make me laugh when people will say it’s meaningless and then happily sit there quoting shots or shots on target.

Edited February 13 by nottingram

[Ghost of Clough]

23 minutes ago, David said:

With football though, it’s not like snooker where you can line the balls up and have 2 players take the same shot.

xG fails to take into account the ability of the opposition defenders and keeper.

It fails to take into account weather conditions.

It fails to take into account the current situation in so many way, home, away, comfortable lead, behind, fitness level, carrying knocks.

If it was a game played between robots in an indoor arena I could see the value.

Comparing it to shots as a metric is nonsensical, shots and shots on target are facts. Indisputable facts.

Nothing more than a stat that with possession gives you a rough idea as to how the game as gone.

Never really see anyone give it much value as you see others try and give xG.

Just watch the game. You will know how it’s gone and who played better, who created the better chances to win the game and who’s finishing or mistakes let them down.

You're right - there are many variables and it is 'just a stat'. But, it does take account of these somewhat by giving the average liklihood of a player scoring agaisnt an average defence with an average goalkeeper, etc.. It still comes down to how you use the stat, just like any other stat - be in possession, passing, tackles, whatever tickles your fancy...

Nothing beats watching a game and it's the only realy way you can enjoy the game. But, what if you can't watch all 552 League 1 games to make a fair judgement? How do you compare how you're doing with other teams? You could compare us against them based on how they played against us, but that's when you get some people saying Peterborough are the best L1 side of all time, whilst some of their own fans are worried they won't even finish top 6. In a lot of cases, stats assist with making a more balanced opinion than just what you see with your eyes. We see "XXXX doesn't run enough" all the time. Well if you saw their running stats, you may think differently.

 

Interpretation is also important. Which is better: 10 shots or 2 shots? Now what if both had a total xG of 1 (10 shots of 0.1 xG vs 2 shots of 0.5 xG)? 
Whilst the xG suggests both teams should have both scored 1, there's actually a 75% probability that the team with just 2 shots has scored, whilst there's only a 65% probability that the team with 10 shots has scored. It's not until that team has had 13 shots worth 0.1 xG (1.3 xG total) where they both have a 75% probability of a goal.
Team 1 only need 5 shots (2.5 xG) to have more than a 95% probability of having scored. Team 2 needs 30 shots (3.0 xG) for the same probability.

[IlsonDerby]

1 hour ago, David said:

Absolute nonsense stat that needs to get in the bin.

Rubbish. 
 

Shots on target/shots assume each shot is just as likely to go in which leads to people assuming ‘more shots = better performance’. xG rightfully builds on that over simplified view and looks at the quality of chances rather than just quantity of chances. I can’t see how anyone can say it’s not a useful metric. 
 

Surely moved past smarmy pundits giving it the ‘the only stat that matters is the goals scored stat’ large. 
 

It was clear from xG that despite derby not having the results we’d like under Rosenior we were creating quality chances that we simply weren’t taking. (The eye sight test would tell you that Collins had a howler of a start to last season but xG took the emotion out of those claims that we were a boring team that didn’t attack. We did. We attacked very well. We just couldn’t hit a barn door). 

Whereas Rotherham were massively over performing against their xG at the start of last season and eventually started falling like a stone with only the over performance at the start of the season keeping them up ultimately. They regressed back to where their underlying stats said they should have been. 

[May Contain Nuts]

1 hour ago, David said:

With football though, it’s not like snooker where you can line the balls up and have 2 players take the same shot.

xG fails to take into account the ability of the opposition defenders and keeper. It fails to take into account weather conditions.It fails to take into account the current situation in so many way, home, away, comfortable lead, behind, fitness level, carrying knocks.

If it was a game played between robots in an indoor arena I could see the value.

Comparing it to shots as a metric is nonsensical, shots and shots on target are facts. Indisputable facts.

Shots and shots on target also fail to take into account any of the above.

It's Key Passes I really hate. RB gives ball to RW 70 yards from goal. RW runs forward 30 yards and balloons a ball 20 foot higher than the bar, closer to the corner flag than the goal. RB adds a key pass to their figures. Yes, this is based on two players with the initials CC and TI!.

Edited February 13 by May Contain Nuts

[MadAmster]

Some interesting and varying views on stats and their worth. xG is calculated by several different companies/organisations, each with their own model and each with their own algorithms. They come up with different xG figures. As do those who "calculate" the number of shots and SOT. How do they handle blocked shots? How far does a shot have to travel (or how close to the goal line does it have to get) in order to be registered as a shot/SOT?

As with the models and  algorithms for xG, I have no idea what, in the minds of those various organisations who do such calculations, constitutes a shot so it's difficult to reconcile any of those stats with what I saw.

At the end of the day, there is but 1 stat (or does it count as 2 stats?) that counts and that's goals scored (for/against which could be seen a 2 stats).

[richinspain]

"Statistics are like a mini skirt, very nice to look at but hide what's really important" - Jimmy Sirrel.

[Bris Vegas]

If I haven’t watched a game, I will generally look at the following stats to get an idea of who was the better side.

- Final third entries

- xG

- Shots on goal

- Possession

With those stats you can pretty much see the overall picture of the game.

The Newcastle xG against Forest was low because you only need to see the source of the goals.

- Newcastle’s first two goals came from set-pieces. The success rate of goals from corners is something like 2%. 

- Newcastle’s third goal was a shot from outside the box.

The three goals didn’t come from chances created yet xG measures the quality of chance created - hence Newcastle’s overall tally is low.

The problem with just using the shots metric is you can’t tell the quality of those chances. You could have 30 shots from outside the box which has a small chance of success.

Another example was City’s away defeat to Wolves. They lost 2-1 and the shot count was about 25-4 in their favor. But they didn’t create any quality chances, with City’s only goal coming from an Alvarez free-kick.

On paper City were by far the better side with more possession, final third entries and shots. But their xG was low because of their lack of real high quality goal scoring chances created.

[ap04]

Far more representative than goals/points, which are just cr*p for reasons that should be obvious to everyone.

For individual games it can give the wrong picture as it has several flaws eg.

Ignores shot/goalkeeping quality - how wide the shot or how comfortable the save
Goals should always count as "1" (perfect attacking/worst defending)
Ignores disallowed goals and close offsides/no-penalties
Ignores key blocked crosses/passes
Counts several shots in same sequence more than once

But combined with shots (with which there is usually good correlation) it's the safest bet if one hasn't watched a match.

After a few games and over a season the above tend to even out and the table becomes quite reliable.

The perfect metric is post-shot chances everything included (how close to a goal every phase was), this works well for a single team but for an entire league one needs to watch all games/highlights which is impractical. This is where xG tables become useful, basically other than following every team there is nothing better in the public domain.

[AndyinLiverpool]

16 minutes ago, ap04 said:

Far more representative than goals/points, which are just cr*p for reasons that should be obvious to everyone.

For individual games it can give the wrong picture as it has several flaws eg.

Ignores shot/goalkeeping quality - how wide the shot or how comfortable the save
Goals should always count as "1" (perfect attacking/worst defending)
Ignores disallowed goals and close offsides/no-penalties
Ignores key blocked crosses/passes
Counts several shots in same sequence more than once

But combined with shots (with which there is usually good correlation) it's the safest bet if one hasn't watched a match.

After a few games and over a season the above tend to even out and the table becomes quite reliable.

The perfect metric is post-shot chances everything included (how close to a goal every phase was), this works well for a single team but for an entire league one needs to watch all games/highlights which is impractical. This is where xG tables become useful, basically other than following every team there is nothing better in the public domain.

I don't understand this. Expected goals is more representative than actual goals? What are they more represetative of?