Xg

[Frankel]

What was it for Leicester when they won the league?

 

[riffraff]

Its a load of shite.

Last season for safc proved otherwise.

 

[Deleted member 634]

it's literally empirically the best predictor of past results as a single metric we have. But AgnewsHorseshoe from the SMB has decided it's rubbish so let's get rid of all statistics altogether

Eh? 'Best predictor of past results' f***ing shite man. Only one stat matters

 

[flusheding]

I've never understood if it takes into account quality of players mind. For example a clear shot from near the penalty spot seems to have a specific xG value assigned to it. The actual likelihood of scoring, however, varies greatly depending on whether the striker it falls to is Charlie Wyke who can't hit a cow's arse with a banjo or, say, Defoe who is lethal.

So if 2 teams had the same midfield supplying chances but one had Defoe where the other had Wyke, would they not have the same xG, even though in reality you'd expect the Defoe team to score a lot more? Maybe that's one for @The Cabbage

It is averaged out. The data used to create the expectancy is normally all players in that league, so the xG is the averaged out results from a position on the pitch by the average of all players shooting from that position.

This is why it isn't 100% for every team and player.

This season Vardy is massively overperforming but doesn't seem to be letting up as he's incredibly clinical so it doesn't matter if he only gets 1 chance from a dodgy angle, he can bury the chance even though the "average" xG player wouldn't be able to.

But just because it doesn't work for the outliers doesn't make it bad

 

[Rogue Badger]

I've never understood if it takes into account quality of players mind. For example a clear shot from near the penalty spot seems to have a specific xG value assigned to it. The actual likelihood of scoring, however, varies greatly depending on whether the striker it falls to is Charlie Wyke who can't hit a cow's arse with a banjo or, say, Defoe who is lethal.

So if 2 teams had the same midfield supplying chances but one had Defoe where the other had Wyke, would they not have the same xG, even though in reality you'd expect the Defoe team to score a lot more? Maybe that's one for @The Cabbage

That’s exactly the point and exactly the problem with xG. It removes a load of variables.

Defoe would have more chances than Wyke with the same midfield because of his movement. He’d end up scoring and missing more in the same team.

 

[football genius]

Thinking About Finishing Skill - StatsBomb | Data Champions

Apparently it’s “finishing skill” season. The debate happens every year, usually precipitated by an incredible run of goals by somebody.

statsbomb.com

Remember, shot conversion overwhelmingly depends on location, this stuff is exceedingly small potatoes by comparison. It doesn’t impact things enough to make a difference.

This is a good guide to how much finishing skill matters if anyone's genuinely interested. So's this, and it's worth noting that since then xG models have started to adjust for the variance that he talks about. Finishing skill is real

Basically, the quality of finishers you have on your team does matter, but not anywhere near as much as the basics - shot volume, shot location. Ultimately virtually every striker should be able to score a goal from 5 yards out - creating those sorts of opportunities in high numbers is more important than worrying about whether our strikers are worse than our rivals'

 

[AC Mack]

I've never understood if it takes into account quality of players mind. For example a clear shot from near the penalty spot seems to have a specific xG value assigned to it. The actual likelihood of scoring, however, varies greatly depending on whether the striker it falls to is Charlie Wyke who can't hit a cow's arse with a banjo or, say, Defoe who is lethal.

So if 2 teams had the same midfield supplying chances but one had Defoe where the other had Wyke, would they not have the same xG, even though in reality you'd expect the Defoe team to score a lot more? Maybe that's one for @The Cabbage

It doesn't account for player quality in isolation, nor does it account for a defender's position in relation to the chance. So Will Grigg on the penalty spot with four defenders between him and goal will be just as good a chance on xG as *insert top league one striker* one on one with a goalkeeper with no hands.

However, iirc it does loosely take into account how the chance came about, which might level the above out a little. It doesn't account for how good a striker is at finishing, nor how good a keeper is at saving shots, yet it's still the best barometer we have.

 

[What A Waster]

I've never understood if it takes into account quality of players mind. For example a clear shot from near the penalty spot seems to have a specific xG value assigned to it. The actual likelihood of scoring, however, varies greatly depending on whether the striker it falls to is Charlie Wyke who can't hit a cow's arse with a banjo or, say, Defoe who is lethal.

So if 2 teams had the same midfield supplying chances but one had Defoe where the other had Wyke, would they not have the same xG, even though in reality you'd expect the Defoe team to score a lot more? Maybe that's one for @The Cabbage

Players are irrelevant.
If you think of the chance you describe, the xG data is taken from the thousands of times that chance occurs, and how many result in a goal.

 

[football genius]

That’s exactly the point and exactly the problem with xG. It removes a load of variables.

Defoe would have more chances than Wyke with the same midfield because of his movement. He’d end up scoring and missing more in the same team.

but it doesn't remove that second variable - in this case we'd have higher xGs with Defoe in the team than Wyke because we'd be able to create more chances than we do at the moment - shot volume/quality would go up. It's just that xG doesn't think Defoe's finishing ability (as opposed to his ability to create opportunities for himself) is as important

 

[AC Mack]

Burnley have consistently been among the worst in the Premier League in terms of xG and xGA, even in the season they finished 7th iirc. They don't make a large number of good chances, and they give up a decent number of good chances to their opposition.

Yet they somehow do better than xG and xGA suggest they should.

xG is a good metric, but it doesn't tell the whole story, Burnley might make a load of half chances and this barrage results in more goals than if those half chances came in isolation. And they've been blessed with very good 'keepers.

In Burnley's case, xG doesn't factor in how defensive they are - the chances against them aren't as clear cut as they seem on xG because there are lots of defenders in the way.

 

[What A Waster]

It doesn't account for player quality in isolation, nor does it account for a defender's position in relation to the chance. So Will Grigg on the penalty spot with four defenders between him and goal will be just as good a chance on xG as *insert top league one striker* one on one with a goalkeeper with no hands.

However, iirc it does loosely take into account how the chance came about, which might level the above out a little. It doesn't account for how good a striker is at finishing, nor how good a keeper is at saving shots, yet it's still the best barometer we have.

A one on one from the penalty spot is different to a ball crosses into a crowded penalty area where the shot comes from the spot. They will have a different xG rating.

but it doesn't remove that second variable - in this case we'd have higher xGs with Defoe in the team than Wyke because we'd be able to create more chances than we do at the moment - shot volume/quality would go up. It's just that xG doesn't think Defoe's finishing ability (as opposed to his ability to create opportunities for himself) is as important

You wouldn't, because xG measures the quality of chances, rather than players ability.
The data is based on 1000s of times that occurs so accounts for good and bad players to give an average number of times that chance is scored, irrespectiveof players ability

Thinking About Finishing Skill - StatsBomb | Data Champions

Apparently it’s “finishing skill” season. The debate happens every year, usually precipitated by an incredible run of goals by somebody.

statsbomb.com

This is a good guide to how much finishing skill matters if anyone's genuinely interested. So's this, and it's worth noting that since then xG models have started to adjust for the variance that he talks about. Finishing skill is real

Basically, the quality of finishers you have on your team does matter, but not anywhere near as much as the basics - shot volume, shot location. Ultimately virtually every striker should be able to score a goal from 5 yards out - creating those sorts of opportunities in high numbers is more important than worrying about whether our strikers are worse than our rivals'

This is partly why Man City have high xG, because they frequently work the ball into the area between the posts inside the box which is where about 80% of their goals come from. They make it easy as possible for the strikers. Who are also very lethal finishers.

 

[AC Mack]

A one on one from the penalty spot is different to a ball crosses into a crowded penalty area where the shot comes from the spot. They will have a different xG rating.


You wouldn't, because xG measures the quality of chances, rather than players ability.
The data is based on 1000s of times that occurs so accounts for good and bad players to give an average number of times that chance is scored, irrespectiveof players ability

That's why I said they take into account how the chance came about.

Also, football genius is saying that as Defoe has a much better all round game than Wyke, a team would create more and better chances with him in the team, and therefore their xG would go up. It's nothing to do with the finishing ability of either.

 

[What A Waster]

That's why I said they take into account how the chance came about.

Yes sorry i was mainly replying to your first paragraph, and sort of disregarded you second one. We agree, I think.

 

[AC Mack]

Yes sorry i was mainly replying to your first paragraph, and sort of disregarded you second one. We agree, I think.

Yes, I counter argued my first paragraph with my second. It's good but not flawless.

 

[Rogue Badger]

but it doesn't remove that second variable - in this case we'd have higher xGs with Defoe in the team than Wyke because we'd be able to create more chances than we do at the moment - shot volume/quality would go up. It's just that xG doesn't think Defoe's finishing ability (as opposed to his ability to create opportunities for himself) is as important

I agree. We still have the same midfield so when hypothetical Defoe gets injured and we replace him with hypothetical Wyke we’re screwed.

I’m not knocking stats. It’s just more complex than looking at one number which I’m sure you understand.

I hope with the investment coming in we might get a lot better at analysis of large data.

 

[gipetto]

Which is something xG doesn't appreciate, unfortunately.

Surprisingly I think it does. I'm pretty sure the stat also assesses who is taking the shot by measuring the player's data over x number of games i.e. Aguero is more likely to score a one on one than Wyke

 

[123SunTan]

One of the modifications 538 did over the original ESPN was to include team value at the start. So in League One we are miles above the other teams in valuation.

With each match the plays cause SPI to evolve.

 

[flusheding]

Surprisingly I think it does. I'm pretty sure the stat also assesses who is taking the shot by measuring the player's data over x number of games i.e. Aguero is more likely to score a one on one than Wyke

One of the things this thread doesn't mention clearly is there is no one "true" xG model.

Different websites and different companies may end up with different xG numbers because you can start with a very basic data set of "shots from within the box are x% likely to go in" or a very advanced data set of "shots from within the box on the left side, with 2 defenders in the way are x% likely to go in". Using models with both types of data will result in an xG number but the second scenario (should) give a much more accurate number.

You have to take the xG numbers as rough if they are from a casual blogger who has pulled their stats from public websites compared to someone like Opta.

Without knowing what data source someone is using and what different variables are being included to calculate the probabilities it is anyones guess how accurate the xG number is.

If anyone wants to create their own xG model then there's a simple guide here using WhoScored? and Excel - An xG Model for Everyone in 20 minutes (ish)

 

[zebediah]

XG accentuates the reason why good goal scorers cost so much.

Nonce behaviour if you’re into this like

I would have little hope of any management team that thought this in modern football.