[Albion] xG table

[RyFish]

I also think xG, or something very similar, kept Hyypia in a job longer than most of us believe he merited. There were a few soundings from the Albion boardroom that the team had been unlucky, and I'm sure that was based on one/some of Bloom's statistical measures of choice.

And likewise why Brentford sacked Warburton from a seemingly impressive league position.

 

[Bakero]

We've always talked about players missing sitters or, on the contrary, said something was a good effort but a difficult chance. Xg simply provides the data. If a team are underperforming compared bro XG, it suggests they're due an upswing (see Watford and Southampton), or vice versa (see Newcastle). If a player consistently outperforms xg, it suggests he's clinical (see Vardy).

People aren't calculating this shit for the fun of it.

 

[Mr Putdown]

We’re above palace, so I’d say it’s a pretty good indicator.

But you’re not, even playing such great football as you unquestionably are.

It’s a completely pointless stat, amongst a sea of utterly irrelevant stats.

 

[Bob!]

This.

Basically actual goals scored v Xg just tells you how good a team is at converting those chances.

What I read into about XG tables is it tells you about a teams goalscoring prowess. So if a team (like Newcastle) is much lower in the XG table it says to me they are, in the real world, doing really well with their scoring chances. Southampton on the other hand have been terrible in front of goal in the real world

Joelinton better than Ings? hmmm

So XG may, or may not tell you than you need a new striker or two, beyond that I can't really see what information it gives to fans or data analysts. I've heard it said that it provides a reflection of who has been dominant in a game, or over the course of a season and in some way reflects the amount of luck a team has, so if you lose a game but have a higher XG, you've been somehow wronged. Not sure about that either. The game is about sticking it in the goal after all

 

[Kalimantan Gull]

We've always talked about players missing sitters or, on the contrary, said something was a good effort but a difficult chance. Xg simply provides the data. If a team are underperforming compared bro XG, it suggests they're due an upswing (see Watford and Southampton), or vice versa (see Newcastle). If a player consistently outperforms xg, it suggests he's clinical (see Vardy).

People aren't calculating this shit for the fun of it.

But does it really suggest that these teams are due a change in fortune? All I've seen is a stat that says how easy or otherwise a teams chances were, and whether they took them or not. I haven't seen anything to say that this is predictive of the future, or how long it takes for the stat to stabilise, or how many games have to be played before the difference between xG and actual goals reduces to zero.

If a player consistently outperforms his xG, it says something of his quality. For shots hit from outside the box the xG of Lionel Messi and Davy Propper will be the same, the outcome will be expectedly different. Perfect for comparing players, far more problematic for comparing teams if the player to whom the chance fell isn't recorded. Still work to do.

 

[Hampster Gull]

Any bright sparks out their - What number are we in the Attacking/defensive prowess and why?

 

[Bakero]

My issue with it is that if a speculative shot from 30 yards is saved, but then falls to the feet of an unmarked striker for a tap in, the total xG of the move is quite high, but only happens because of the speculative shot from 30 yards.

Tell me if I'm wrong, because I haven't looked into the stat in great detail, but I imagine that it is assuming that all chances are independent events and can be summed to give a total xG for the match (or player, or whatever). But goal scoring chances are not independent events, as in the example above. You can also imagine a situation where a striker misses 3 easy tap ins in the same sequence, with the keeper making brilliant point blank saves, but scores with the fourth. The xG for each chance is greater than 0.5, so the total xG for a single goal in that scenario would be >2. Has that been accounted for?

I like the idea of the stat, but I don't believe that it is accurately capturing what is happening in a game.

Are they also claiming that this stat is predictive of future performance, because I also find that hard to accept?

Yes, that's been accounted for.

With regard to future performance, as I mentioned, if the xg differential is poor it shows a team has not been taking their chances. Over the course of the season you could predict that they'll eventually start putting more away. That's not to say the xg diff will eventually be zero. If there is no upswing whatsoever, there's work to do on the training pitch or more likely in the recruitment department.

 

[nwgull]

Looking at this and their performance on Saturday, makes me think Bournemouth are in big danger of going down! Norwich/Villa/Bournemouth I reckon! With a big slide on for Newcastle too! We'll be fine!

That relegation treble is 2nd favourite on Sky Bet at 9-1. May put a £10 on based on current form of all 3.

 

[Kalimantan Gull]

Yes, that's been accounted for.

With regard to future performance, as I mentioned, if the xg differential is poor it shows a team has not been taking their chances. Over the course of the season you could predict that they'll eventually start putting more away. That's not to say the xg diff will eventually be zero. If there is no upswing whatsoever, there's work to do on the training pitch or more likely in the recruitment department.

Its predictive if the xG is predicted to stay the same for a team? Is it? How long does that stat take to settle down?

As for the regression to the mean idea, why is that expected to even out over a season? Why not 2 seasons, or 5 or 10, or a 1/2? Again, how long does that take to settle down?

To me it just says that if you beat your xG you have better players. If a chance has an xG of 0.2, that should change if its Salah or Benteke taking the chance, but I assume that level of detail is hard to account for.

And how does it account for multiple shots /saves in a single move?

Sorry lots of questions I know, do you have links to where this is discussed? Because without knowing that kind of stuff a statistic is pretty useless if we want to predict performance off it.

 

[sussex_guy2k2]

I was having a bit of a row about xG on saturday with a Liverpool supporting friend. I argued it was an absolute load of complete rubbish - just a stat for the sake of another stat, but doesn't really give you anything of any great value. He argued that it's the most representative stat of how good a team is at creating chances. The underlying point he made was that Leicester have been incredibly lucky this season having scored so many goals from difficult/low percentage chances. I argued it means nothing - if Vardy wants to spank one in the top corner from his only chance every game, he will. Good lad.

He’s right, that’s exactly what it is. What it doesn’t take account of is the quality of the individual players to take chances.

A prime example was Burnley for a couple of seasons. The xG against them wasn’t particularly high because they funnelled teams into low risk areas where the opposition would take a lot of shots but they were in a prime position to be blocked. So the shots against stat was high, but the xG against was low relative to their position. So it’s a useful stat.

 

[Bakero]

Its predictive if the xG is predicted to stay the same for a team? Is it? How long does that stat take to settle down?

As for the regression to the mean idea, why is that expected to even out over a season? Why not 2 seasons, or 5 or 10, or a 1/2? Again, how long does that take to settle down?

To me it just says that if you beat your xG you have better players. If a chance has an xG of 0.2, that should change if its Salah or Benteke taking the chance, but I assume that level of detail is hard to account for.

And how does it account for multiple shots /saves in a single move?

Sorry lots of questions I know, do you have links to where this is discussed? Because without knowing that kind of stuff a statistic is pretty useless if we want to predict performance off it.

There's a bit about multiple shots in one phase here https://fbref.com/en/expected-goals-model-explained/

Teams with better players don't necessarily beat their xg. Man City are about bang on. Chelsea and Man U have a negative xg diff.

Norwich are slightly out-performing xg now but earlier in the season they were way above it. I had a bet with a mate that Norwich would go down and Pukki wouldn't get 20 goals- at the time almost every shot he was taking was going in, which seemed unsustainable to me.

As for reversion to the mean, I'm not sure anyone has said it'll balance out completely over the season. In Klopp's final season at Dortmund they were under performing compared to xg and even flirted with relagation. The upswing came and they finished 7th, which was still below where they'd have expected to come pre-season. They bounced back to finish 2nd the next year.

 

[Kalimantan Gull]

There's a bit about multiple shots in one phase here https://fbref.com/en/expected-goals-model-explained/

Teams with better players don't necessarily beat their xg. Man City are about bang on. Chelsea and Man U have a negative xg diff.

Norwich are slightly out-performing xg now but earlier in the season they were way above it. I had a bet with a mate that Norwich would go down and Pukki wouldn't get 20 goals- at the time almost every shot he was taking was going in, which seemed unsustainable to me.

As for reversion to the mean, I'm not sure anyone has said it'll balance out completely over the season. In Klopp's final season at Dortmund they were under performing compared to xg and even flirted with relagation. The upswing came and they finished 7th, which was still below where they'd have expected to come pre-season. They bounced back to finish 2nd the next year.

Thanks for the link. Would still be interested to know how long it takes to settle down, its a bit like a cricketers average that a couple of matches don't tell you his true skill, it takes time to find a stable mean.

 

[Mtoto]

Its predictive if the xG is predicted to stay the same for a team? Is it? How long does that stat take to settle down?

As for the regression to the mean idea, why is that expected to even out over a season? Why not 2 seasons, or 5 or 10, or a 1/2? Again, how long does that take to settle down?

To me it just says that if you beat your xG you have better players. If a chance has an xG of 0.2, that should change if its Salah or Benteke taking the chance, but I assume that level of detail is hard to account for.

And how does it account for multiple shots /saves in a single move?

Sorry lots of questions I know, do you have links to where this is discussed? Because without knowing that kind of stuff a statistic is pretty useless if we want to predict performance off it.

There's no such thing as a definitive xG for a given match, there are plenty of versions around using a variety of methodologies (eg. some count penalties as 0.8, some don't). Some also take account of which player gets a chance, so Messi getting the ball with a sight of goal inside the box might be 0.3 when it would be 0.15 for a less gifted player.

Individual conversion rates, and fluctuations, will also feed in to the overall rating for both teams - ie. it's not just about how good a player is at scoring chances, it's also about how good the defenders are at stopping them, and how good both the attackers and defenders might be next weekend when they are up against better/lesser opposition.

This is where the predictive side comes in, because it's how the odds-setters go about pricing matches up in the first place and overall, the odds are a very good guide to the likely outcome.

Personally I'm a fan of the 538 numbers, because they are a. free, b. downloadable, c. cover every major league and perhaps most importantly, d. the Racing Post's stats guru reckons they're the best around and he's a man who'd know.

They discuss their way of doing things here:

https://fivethirtyeight.com/methodology/how-our-club-soccer-predictions-work/