xG limitations

[hmchan]

Now you're just being facetious.

As I said I don't know the factors involved in Leicester's PL winning campaign but over the course of that season they outperformed xga by something like 9 goals which is not unheard of. Just look at Newcastle, Wolves and Sheffield United this year. Although many people were wondering if Leicester could keep it up that year. As it turns out that they managed to stay ahead of the other teams.

In the early part of the season both Leicester and Vardy were outperforming xg at an unprecedented rate and Rodgers was the second coming of Christ. After 11 games or so they were +12 on xg and -5 on xg against. This appeared to be unsustainable especially for a club like Leicester, given their over-reliance on Vardy, squad depth etc. They weren't creating a huge amount of chances and teams were creating chances against them.

Of course Vardy could have continued this performance and finished the campaign with 35 - 40 league goals and they also could have continued in the same vein defensively and ended up with a lower goals against figure than Liverpool with a much higer xga but that was unlikely.

Xg is just a metric like any other and gives us more information. I don't know why you're so bothered by it.

As others have said it's not perfect and should be taken in conjunction with all the other data to give us a better overall picture of what is going on.

Like I said in my previous posts, I'm also a fan of xG and I quote it occassionally myself. It is useful in proving a certain point (e.g. a more clinical finisher, a better shot stopper etc.) but it's inadequate to summarize a team's overall performance. It's annoying to see people misuse/misinterpret it, view it as a sacred prediction tool, and jump to conclusion solely based on it. You, in particular, suggest you could easily predict Leicester's fall early in the season, without any reference to their recent switch in formation, absence of players and dipped form of players which are the actual reasons to their late drop.

Now you're just being facetious. For example (Leicester this season) that favors your point, you have plenty of explanations; for example (Leicester in 15/16) that suggests the opposite, you don't know the factors. The reason your argument falls flat is that you completely neglect the possibility a team/player could outperform the xG/xGA for a long enough period of time, and it is not highly unlikely. It is also not uncommon for midtable sides to finish the season overperforming by >10 xG, such as West Ham in 17/18 and Everton in 16/17. Tottenham had also tried twice in recent years with limited spending. I don't see why Leicester couldn't do the same if they hadn't lost their good players and momentum at the end of the season.

At last I'm glad that you finally acknowledge the limitation of xG and you start to incorporate some context of the game to your argument.

 

[Withnail]

Like I said in my previous posts, I'm also a fan of xG and I quote it occassionally myself. It is useful in proving a certain point (e.g. a more clinical finisher, a better shot stopper etc.) but it's inadequate to summarize a team's overall performance. It's annoying to see people misuse/misinterpret it, view it as a sacred prediction tool, and jump to conclusion solely based on it. You, in particular, suggest you could easily predict Leicester's fall early in the season, without any reference to their recent switch in formation, absence of players and dipped form of players which are the actual reasons to their late drop.

Now you're just being facetious. For example (Leicester this season) that favors your point, you have plenty of explanations; for example (Leicester in 15/16) that suggests the opposite, you don't know the factors. The reason your argument falls flat is that you completely neglect the possibility a team/player could outperform the xG/xGA for a long enough period of time, and it is not highly unlikely. It is also not uncommon for midtable sides to finish the season overperforming by >10 xG, such as West Ham in 17/18 and Everton in 16/17. Tottenham had also tried twice in recent years with limited spending. I don't see why Leicester couldn't do the same if they hadn't lost their good players and momentum at the end of the season.

At last I'm glad that you finally acknowledge the limitation of xG and you start to incorporate some context of the game to your argument.

You're completely misconstruing what I said. At no point did I say anything about easily predicting anything and I think you may be confusing me with someone else.

We're also not talking about a +10 for the season. It was around +12 with an xg of after 11 games. If that had continued it would be something ridiculous like +40/50 by the end of the season.

I felt at the time that they could drop out of the top 4 by the end of the season but it wasn't a certainty and I said so earlier.

If you are a 'fan of xg', what's with all the snarky comments?

 

[Fingeredmouse]

In this sense we can't blame Moyes, van Gaal or Mourinho for the decline in recent years. It's just "regression to mean" according to the xG extremists and we can do nothing about it.

I have no idea what you mean by regression to the mean here? xG isn't based on averages.

Given your later post, it seems you fully understand the metric. Nobody has claimed at any point the metric is a panacea nor that the metric has it's limits.

What's the argument?

 

[Paul_Scholes18]

I also think we shouldn't get too hung up on xG itself. It's just one metric to paint us a picture of how teams are performing. A very good and easy-to-use one while still being very accurate at that. I'd be happy just for people to acknowledge that there is very often a discrepancy between performances and results (even over 10, 20 or more games) and the latter are a function of the former and not the other way around. This would go a long way in improving discussions about football in general. I genuinely believe most people would be better informed if they never looked at the scorelines.

I just think using xG without context doesn't fully tell how teams perform though. Plus you got to assume that teams create as many chances in the future as they did in the past.
That is often not the case and it will depend on many variables.
A team in good form will create more high quality chances for themselfes when they need too. A team in poor form will be far more desperate and chase games when they are a goal down etc.

 

[Cheimoon]

I just think using xG without context doesn't fully tell how teams perform though. Plus you got to assume that teams create as many chances in the future as they did in the past.
That is often not the case and it will depend on many variables.
A team in good form will create more high quality chances for themselfes when they need too. A team in poor form will be far more desperate and chase games when they are a goal down etc.

Does anyone claim that there is any single stat that fully (as you say) tell how teams perform? xG is just one stat that tells you something about chance production (and giving up chances through xGA), but no-one sane thinks it's a tell-all. That might present it that way on tv, since discussions have to stay short and hence are often fairly superficial; but no-one who seriously considers football is going to go with xG alone.

 

[Paul_Scholes18]

Does anyone claim that there is any single stat that fully (as you say) tell how teams perform? xG is just one stat that tells you something about chance production (and giving up chances through xGA), but no-one sane thinks it's a tell-all. That might present it that way on tv, since discussions have to stay short and hence are often fairly superficial; but no-one who seriously considers football is going to go with xG alone.

Just the claim you can use it to predict things on it's own. I guess that is a minority.
I think for a quick analysis of a league you never followed it might work to give you a view about chance creation that helps you. Together with looking at the league table.

 

[giorno]

Does anyone claim that there is any single stat that fully (as you say) tell how teams perform? xG is just one stat that tells you something about chance production (and giving up chances through xGA), but no-one sane thinks it's a tell-all. That might present it that way on tv, since discussions have to stay short and hence are often fairly superficial; but no-one who seriously considers football is going to go with xG alone.

It might, in the future if they perfect the model. Hard, but not impossible

Right now it *is* the best tool available to measure games and performances. It's not perfect, but except rare cases it beats the eye test, or any other available tools we have

 

[Cheimoon]

Just the claim you can use it to predict things on it's own. I guess that is a minority.
I think for a quick analysis of a league you never followed it might work to give you a view about chance creation that helps you. Together with looking at the league table.

Even that overall average is messy on its own though. The strength of the opposition compared to your own team should factor into things (a low-ranking team has low overall average xG - no shit!), as well as actual goals scored.

It might, in the future if they perfect the model. Hard, but not impossible

Right now it *is* the best tool available to measure games and performances. It's not perfect, but except rare cases it beats the eye test, or any other available tools we have

It is the best single tool to measure games and performances. But there is no way a proper statistician of football uses a single tool and considers overall average. Or at least, I would be extremely surprised and disappointed if the models they are apparently using at apparently progressive clubs like Liverpool and Brentford don't go much beyond xG.

(To be clear, I do indeed think complex modelling can do a great job in football nowadays.)

 

[hmchan]

You're completely misconstruing what I said. At no point did I say anything about easily predicting anything and I think you may be confusing me with someone else.

We're also not talking about a +10 for the season. It was around +12 with an xg of after 11 games. If that had continued it would be something ridiculous like +40/50 by the end of the season.

I felt at the time that they could drop out of the top 4 by the end of the season but it wasn't a certainty and I said so earlier.

If you are a 'fan of xg', what's with all the snarky comments?

As a matter of fact I do. I admit it and I'm apologizing to you for that.

 

[giorno]

It is the best single tool to measure games and performances. But there is no way a proper statistician of football uses a single tool and considers overall average. Or at least, I would be extremely surprised and disappointed if the models they are apparently using at apparently progressive clubs like Liverpool and Brentford don't go much beyond xG.

Yeah, obviously the more information the better

Think clubs like liverpool and brentford have access to both more advance and more varied models. At the end of the day, knowing their next opponent has a great attack is useful, but knowing how that great attack works and what makes it great is far more important

 

[hmchan]

I have no idea what you mean by regression to the mean here? xG isn't based on averages.

Given your later post, it seems you fully understand the metric. Nobody has claimed at any point the metric is a panacea nor that the metric has it's limits.

What's the argument?

The idea of "regression to the mean" isn't brought up by me actually. It starts from #95 when some argue that a team can't outperform xG sustainably and their performance will somehow inevitably drop to match the xG in the end. They make it sound like xG always predicts the fall coming.

To me football isn't pure statistics, at least for now. Regression may or may not happen depending on the context of the game, which is much more important than xG itself. There are plenty of examples where teams have been outperforming xG over the years.

 

[Withnail]

As a matter of fact I do. I admit it and I'm apologizing to you for that.

That's cool, It's happened me before.

 

[OleBoiii]

I like the idea of xG. It's infinitely more meaningful than "number of shots" at least.

 

[Mb194dc]

Is the xg data for the last couple of nights champions league games available anywhere?

Especially interest in the City game yesterday.

 

[Paul_Scholes18]

Is the xg data for the last couple of nights champions league games available anywhere?

Especially interest in the City game yesterday.

Sterlings chance. 0.99 xG . Apart from that one they didn't miss that many chances though.

 

[hmchan]

Is the xg data for the last couple of nights champions league games available anywhere?

Especially interest in the City game yesterday.

https://fbref.com/en/comps/8/Champions-League-Stats

 

[Mb194dc]

Interesting City and Utd should both have won according to xG.

Don't get the Sevilla Stats, surely the de Jong chance should be more than 0.4, it's a tap in, Suso goal also only registers as 0.2, he'll surely score that about 9/10. What am I missing, does the defending before the chance count as well?

 

[BalanceUnAutreJoint]

Interesting City and Utd should both have won according to xG.

The issue is that there is no context in xG.

Lyon went 1 up ahead and let City come at them until they answered back with a goal
hTey spent close to 50 minutes refusing to attack as to not overcommit and let go of their lead
But immediately after City scored they had no issues going forward again and exploiting City's high line and defensive mistakes.
Individual mistakes are also part of the sport. If you're missing taps in like Sterling did you have no one but yourself to blame and you can't really complain about deserving more.

Now yesterday's game is pretty different as both teams always played with the intention of scoring and Sévilla only went up very late into the game.

 

[Acrobat7]

Interesting City and Utd should both have won according to xG.

Don't get the Sevilla Stats, surely the de Jong chance should be more than 0.4, it's a tap in, Suso goal also only registers as 0.2, he'll surely score that about 9/10. What am I missing, does the defending before the chance count as well?

I think it makes sense that both City and United had a higher xG, so xG is - in these cases - a better representation of the actual game than the naked scores.

 

[FootballHQ]

Man. United one must be off the scale today, 2 shots on goal, two goals. And their only corner of the game lead to the penalty.

 

[Classical Mechanic]

Man. United one must be off the scale today, 2 shots on goal, two goals. And their only corner of the game lead to the penalty.

Not that crazy, the shots that hit the post and bar would have been fairly low xG, that they hit the woodwork doesn’t necessarily mean they were big chances.

Tweet

— Twitter API (@user) date

I guess the Rashford chance is a limitation, it had high xG where he shot from but he worked very hard to make that chance.

 

[hmchan]

Not that crazy, the shots that hit the post and bar would have been fairly low xG, that they hit the woodwork doesn’t necessarily mean they were big chances.

Tweet

— Twitter API (@user) date

I guess the Rashford chance is a limitation, it had high xG where he shot from but he worked very hard to make that chance.

1.8 to 1.0 is already a huge advantage, and Brighton certainly has some big chances in the game apart from those which hit the woodwork.

 

[Classical Mechanic]

1.8 to 1.0 is already a huge advantage, and Brighton certainly has some big chances in the game apart from those which hit the woodwork.

It’s not crazy though but they deserved something from the game. They had a really big chance near the end when he hit it straight at DDG from a few yards out.

 

[bosnian_red]

Not that crazy, the shots that hit the post and bar would have been fairly low xG, that they hit the woodwork doesn’t necessarily mean they were big chances.

Tweet

— Twitter API (@user) date

I guess the Rashford chance is a limitation, it had high xG where he shot from but he worked very hard to make that chance.

The Rashford one gets counted as a huge plus for him though, he's he "increased" the quality of his chance on his own. The best players have high xG/xGa and the actual production tends to match up to that.

 

[Classical Mechanic]

The Rashford one gets counted as a huge plus for him though, he's he "increased" the quality of his chance on his own. The best players have high xG/xGa and the actual production tends to match up to that.

I agree. It was a brilliant piece of play.

 

[FootballHQ]

How does XG judge Newcastle out of interest? Constant matches where they get battered and hardly have any shots yet get points from the games.

 

[Classical Mechanic]

How does XG judge Newcastle out of interest? Constant matches where they get battered and hardly have any shots yet get points from the games.

Depends from game to game but a low block side should restrict the opposition to low quality chance so the xG for both sides would be on the low side.