Does xG tell us anything about the season?

[MyBrotherErnie]

7 minutes ago, IAmNick said:

I disagree that Goodhart's Law is applicable, because xG can't be gamed in that way you're implying can it? It's an objective measure, so it's not like possession or shots on target - where you could play it around the back or shoot from 30 yards out to game the stats.

I think it does matter if something like an xG rolling average improves, because even if results don't match then it's an indication of where we're headed - and you'd expect in the majority of cases results to eventually converge with it. So it can be a leading indicator of improvement and can show an increase in performance that isn't yet paying dividends, but should be continued with, right?

Completely agree that xG is harder to game than the other stats you mention (both of which are also objective), but I would still say Goodhart's law applies. xG still incentivises certain things, especially getting shots away, that aren't always the best course of action. 

On the second point I don't think we fundamentally disagree. It definitely matters if performances are improving even if results aren't yet. My point was just that the statistic doesn't matter in itself, but what it measures does. 

 

[IAmNick]

14 minutes ago, Tim Monaghan said:

This is why I think it’s nonsense. Sometimes it’s Way Off. Way off. You’d think the team with the higher xG would win, right? Not always the case. It just measures the quality of chances, not who’s gonna win. 
Unlike what @Davefevs will have you believe, football is played by humans. Football isn’t played by robots; Players get nervous, they make mistakes, they do random stuff. xG doesn’t really get that.

Also xG is all about past data. But football’s always changing, so what happened before might not mean squat for what’s coming next. Injuries, suspensions, new players, change of tactics, the list is endless. 

So, yeah, xG is basic of the basic nonsense. 

I would think the team with the higher xG would win if I knew nothing else about the match... and if two teams played 1000 times then statistically the one with the higher xG would almost certainly win more often there. If I told you two teams played each other and one had way better chances you'd presumably think there were more likely to win (but not definite). That's all xG is saying though. This team had better chances. But games are played once, so that doesn't happen. Your argument is just against random chance, not against xG.

And xG is nothing to do with injuries or suspensions or anything like that. Why are you even talking about them? It's nothing to do with what we're talking about.

It's literally just "Based on a load of similar chances, what's the chance this one is a goal?"... that's it. So it takes into account nerves and mistakes to some degree, because that's part of the model, right? Is a perfect model of every players exact mental state at all times? No, and nobody is claiming it is.

It just sounds like you don't really understand what it is to me. It's clearly not nonsense, because that would imply there's absolutely no value in it or correlation with reality - that's provable untrue.

Is it perfect? No. Can it predict the future? No. Is it wrong sometimes? Yeah, of course it is - quite often in fact. Does that make it nonsense? No.

 

Edited April 28 by IAmNick

[Davefevs]

Bless!

[ExiledAjax]

Fair play to @IAmNick for taking this one. Doing god's work.

[Silvio Dante]

Opened a can of worms here!

Heres my twopenneth, using examples from this season:

-xG as an arbiter of individual chance quality is wholly unreliable. IIRC, Sykes’s winner against Boro (H) had a (relatively) good xG because of where he was stationed. However, the trajectory of the ball towards him and the difficulty of the skill, and the opposing players to navigate weren’t considered. On a sample of 100 chances in the same position the xG would have been a reliable average but Skykes’ chance was at the highest difficulty/lowest xG position of those 100 chances. So as an individual chance, the xG wasn’t reflective

- However, it goes without saying that in a sample of 100 the majority of your chances will be nearer the average than outliers!

- Now take the Hull City at home game. We won 3-2 with an xG of 1.64. We had 16 shots. The penalty was one (0.76 xG) - meaning the residual 15 shots had a combined xG of 0.88. If you remember the game we barely tested the keeper, and our two open play goals were a massive deflection for Knights and Mehmetis shot where the keeper injured himself in the dive so didn’t get down properly. Inherently we had a lot of shots but the quality of chances we produced was extremely poor, hence the xG. Any coach - and I’d imagine Liams the same - would have looked at that and known that position couldn’t go on (as you won’t get the deflection/keeper injury every game). Interestingly enough the next game was Watford, where we were excellent, but then we went on a horror run up until Easter - so the xG actually gave a bit of a leading indicator

- And a great parallel to that is the Sheffield Weds away game where we lost 2-1. Liam said after the game “We had 17 shots, 12 inside the box but didn’t test the keeper” - our xG was 0.96, so both shots and xG in the same ballpark as Hull - but a markedly different outcome - the low xG relative to shots wasn’t impacted by the outcome bias 

In the “pre stats” world we all could say “we were lucky/unlucky to lose that one”. I see xG on a collective basis as being a decent arbiter of whether that’s true.

 

Edited April 29 by Silvio Dante

[ExiledAjax]

9 minutes ago, Silvio Dante said:

In the “pre stats” world we all could say “we were lucky/unlucky to lose that one”. I see xG on a collective basis as being a decent arbiter of whether that’s true.

In summary yes. Which is why it is most relevant and useful when we go on a good/bad run of results, and people get excited/depressed about promotion/relegation but xG and other underlying indicators of performance don't back it up.

In the early days circa 2015-18 it was also fairly useful for trying to "beat" the bookies, but their odds now all take account of xG (because it's actually really good) so it's not so useful for that now.

But anyway, I'm tired of trying to explain it to people who don't want to listen or cannot understand it.

...

As to your opening post, I said to @Davefevs on DM that it's a season where there aren't many teams that have busted the stats. Preston, Cardiff and Plymouth are the most obvious ones benefitting from lady luck/fortune. It also shows that Rotherham and the top 4 really are as good/bad as they seem. You can also see QPR and Sheffield Wednesday improve almost at exactly the point they change manager. Birmingham the opposite.

I still find it interesting, and I'll still tell people we aren't getting promoted/relegated because xG suggests we're utterly mediocre.

[LondonBristolian]

I think very often people criticise statistics when the real issue is the fact that most of us - including within the media - haven't been taught how to interpret statistics anywhere near well enough.

To use a simple example, if a team has a 60% chance of winning a cup final, that will mean there is a 4 in 10 chance they will lose. But if the team that are 60% favourites lose, everyone immediately goes "well, the stats were wrong" but actually the stats clearly implied that - whilst they were more likely to win than lose - them losing was also a strong possibility. But far too often people see a 6 in 10 chance as meaning that will definitely happen and then feel something has gone wrong with the stats if a reasonable possibility doesn't come to fruition.

Expected goals is one indicator of how well a team will do. It's certainly not the only indicator and it can only underpin a story rather than wholly tell it. Is a team that concedes less than "expected" lucky or do they have a really good goalkeeper? Is a team that scores more than "expected" riding their luck or working really hard on finishing on the training pitch? We don't know from expected goals alone. Both @Silvio Danteand @IAmNickhave provided tables that show there is very clearly (and unsurprisingly) correlation between expected goals and position in the table. In the past people have also shown there is a correlation between league position and possession, and (depressingly) league position and salary spend. (Some research suggests that the number of shots taken or faced and transfer spend are not linked anywhere near as much with league position).

Expected goals very obviously isn't nonsense. It's worth a manager and a club understanding their expected goals - expected goals against so that, if they're in a position of unexpectedly good or unexpectedly bad results - they can get a sense of whether it is likely to be a temporary anomaly or part of a deeper pattern. People relying on expected goals as the only indicator or the whole picture is nonsense but I don't see anyone on this thread doing that. 

Edited April 29 by LondonBristolian
Edited bit of about league position and transfer spend for clarity as I had worded it a bit confusingly.

[Davefevs]

12 minutes ago, LondonBristolian said:

People relying on expected goals as the only indicator or the whole picture is nonsense but I don't see anyone on this thread doing that

Indeed, only whoppers trying to twist it.  I just saw Timmy’s reply in a quote.

Edited April 29 by Davefevs

[Alessandro]

So what that XG chart tell us - we’ve just about returned under Manning to our XG under Pearson…

So if we’d not sacked Pearson, we’d be in the play offs now, right?

[ExiledAjax]

6 minutes ago, Davefevs said:

Indeed, only whoppers trying to twist it.

Aka. Cardiff City's legal team.

[David Brent]

It’s the most overrated stat in football.

We coped without it. Not interested if it’s used by betting companies or coaches either. It appears its biggest use is by football fans on innumerable football podcasts and forums. 

It’s no more an important stat than shots on target. 

[mozo]

1 hour ago, Silvio Dante said:

Opened a can of worms here!

Heres my twopenneth, using examples from this season:

-xG as an arbiter of individual chance quality is wholly unreliable. IIRC, Sykes’s winner against Boro (H) had a (relatively) good xG because of where he was stationed. However, the trajectory of the ball towards him and the difficulty of the skill, and the opposing players to navigate weren’t considered. On a sample of 100 chances in the same position the xG would have been a reliable average but Skykes’ chance was at the highest difficulty/lowest xG position of those 100 chances. So as an individual chance, the xG wasn’t reflective

- However, it goes without saying that in a sample of 100 the majority of your chances will be nearer the average than outliers!

- Now take the Hull City at home game. We won 3-2 with an xG of 1.64. We had 16 shots. The penalty was one (0.76 xG) - meaning the residual 15 shots had a combined xG of 0.88. If you remember the game we barely tested the keeper, and our two open play goals were a massive deflection for Knights and Mehmetis shot where the keeper injured himself in the dive so didn’t get down properly. Inherently we had a lot of shots but the quality of chances we produced was extremely poor, hence the xG. Any coach - and I’d imagine Liams the same - would have looked at that and known that position couldn’t go on (as you won’t get the deflection/keeper injury every game). Interestingly enough the next game was Watford, where we were excellent, but then we went on a horror run up until Easter - so the xG actually gave a bit of a leading indicator

- And a great parallel to that is the Sheffield Weds away game where we lost 2-1. Liam said after the game “We had 17 shots, 12 inside the box but didn’t test the keeper” - our xG was 0.96, so both shots and xG in the same ballpark as Hull - but a markedly different outcome - the low xG relative to shots wasn’t impacted by the outcome bias 

In the “pre stats” world we all could say “we were lucky/unlucky to lose that one”. I see xG on a collective basis as being a decent arbiter of whether that’s true.

 

I think that's a great summary, and really, the more additional metrics you add into that collective, the more sense you'll get out of it. On its own, it's pretty useless.

Imagine a cross is drilled into the box, takes two deflections and lands to the left foot of Zak Vyner, who has an open goal in front of him. Surely that would be a high xG? But realistically the ball would end up in the stands. Context is everything.

[ExiledAjax]

20 minutes ago, David Brent said:

It’s the most overrated stat in football.

Not interested if it’s used by betting companies or coaches either. It appears its biggest use is by football fans on innumerable football podcasts and forums.

If you ever have the fortune to speak to Matthew Benham, Thomas Frank, Rasmus Ankersen, or anyone else involved with Brentford in the last ten years then please record yourself telling them this and send me the recording of their answer.

I could really do with a laugh.

Edited April 29 by ExiledAjax

[Davefevs]

And lo and behold, todays update of Opta / The analyst shows:

• the swing from negative xG GD into positive xG GD for the first time pre-Leeds (h)

reflecting upon the improved performances and results since Easter.

• an upturn in rolling xG for
• a stabilising of xG against, maybe a slight trend of improvement (tbc)

Since Easter the data shows:

• More build-up attacks than at any other period in LM’s reign, but importantly circa 25% of long pass sequences ending with a shot or touch in the box (previously 14%)
• A consistent number of turnovers in opposition final third
• pressing at its best level since LM was appointed, despite game-state meaning we can sit-off when ahead (we’ve led for 177 mins / drawing 444 / losing 9 (total 630))

You can use data to present both positive and negative trends, benchmark / baseline it for future reference too.

Rasmus’s TedTalk might be too much!

Edited April 29 by Davefevs

[mozo]

8 minutes ago, Davefevs said:

And lo and behold, todays update of Opta / The analyst shows:

• the swing from negative xG GD into positive xG GD for the first time pre-Leeds (h)

reflecting upon the improved performances and results since Easter.

• an upturn in rolling xG for
• a stabilising of xG against, maybe a slight trend of improvement (tbc)

Since Easter the data shows:

• More build-up attacks than at any other period in LM’s reign, but importantly circa 25% of long pass sequences ending with a shot or touch in the box (previously 14%)
• A consistent number of turnovers in opposition final third
• pressing at its best level since LM was appointed, despite game-state meaning we can sit-off when ahead (we’ve led for 177 mins / drawing 444 / losing 9 (total 630))

You can use data to present both positive and negative trends, benchmark / baseline it for future reference too.

Rasmus’s TedTalk might be too much!

How does this relate to possession Dave? It's interesting that since Easter we've had a variety of different possession stats. Is the output similar regardless? Is there a relative trend?

[MarcusX]

13 hours ago, Tim Monaghan said:

This is why I think it’s nonsense. Sometimes it’s Way Off. Way off. You’d think the team with the higher xG would win, right? Not always the case. It just measures the quality of chances, not who’s gonna win. 
Unlike what @Davefevs will have you believe, football is played by humans. Football isn’t played by robots; Players get nervous, they make mistakes, they do random stuff. xG doesn’t really get that.

Also xG is all about past data. But football’s always changing, so what happened before might not mean squat for what’s coming next. Injuries, suspensions, new players, change of tactics, the list is endless. 

So, yeah, xG is basic of the basic nonsense. 

On a slight tangent, but along the lines of your point, one of my biggest pet hates in football is when comparisons are made to previous meetings between teams and the relevance of that on the next result.

[Silvio Dante]

12 minutes ago, MarcusX said:

On a slight tangent, but along the lines of your point, one of my biggest pet hates in football is when comparisons are made to previous meetings between teams and the relevance of that on the next result.

@Mr Popodopolous is going to burn you.

[ExiledAjax]

Just now, MarcusX said:

On a slight tangent, but along the lines of your point, one of my biggest pet hates in football is when comparisons are made to previous meetings between teams and the relevance of that on the next result.

Ha! Now here's a stat that really can be ignored in my opinion. 

[Davefevs]

38 minutes ago, mozo said:

How does this relate to possession Dave? It's interesting that since Easter we've had a variety of different possession stats. Is the output similar regardless? Is there a relative trend?

It’s something I don’t really track, only cursory glances in relation to whether it’s changed in relation to total passes.  It’s remained around the 46-48% mark all season.

[ExiledAjax]

24 minutes ago, Davefevs said:

It’s something I don’t really track, only cursory glances in relation to whether it’s changed in relation to total passes.  It’s remained around the 46-48% mark all season.

Wait until people find out that different methods of calculating possession have been used over the past ten years, and the methods continue to be developed all the time. Even more awful is that the same fundamental principle that underpins xG and xA - that we should try and qualify the importance of an event rather than simply counting the number of times something happens (basically why xG is superior to SoT) - is now being used to calculate possession.

Shocking and vile to some, and I apologise to those of a philistine nature who may be unnerved by reading this.

[DaveInSA]

And when the big AI companies get in on xG, that’s when you’ll see (IMHO) a marked improvement on the one-offness of actual chances. The capability to pull in every televised goal ever scored into the models will make them way more accurate (not perfect but better)

And that’s not really important…actually…

What @Davefevs says in the new opta stats, is that we’re playing a different style to that of Mannings dip. Which is encouraging. Because it means that Manning has adapted to the squads strengths. This pleases me immensely. Because based on the results playing possession only football, we were absolute trousers.

Will the improvement continue into next season?

Well, we’ll need the XGf and XGa stats to tell us at games 6-10. 

[red panda]

I don't think anyone's mentioned the elephant in the room on the chart 

[Sturny]

19 hours ago, IAmNick said:

Thread: does anyone find this chart interesting?

Answer: stats in football make me physically sick and trying to find any meaning in anything beyond a result is pointless why bother life is hopeless and meaningless anyway so let's all just stop even trying

Overreaction much!

Honestly think some people hate on it because it's something new and looks like moving away from the "proper football" they know.

It's a mildly intriguing stat. It's not black and white. More of a subtle guage open to interpration and discussion. 

I will however agree it was pushed on us a couple of years ago more than it should've.  

Edited April 29 by Sturny