Fbref.com uses StatsBomb data so they're the most accurate around.
xG limitations
- Thread starter Wibble
- Start date
-
In the latest episode of the United Hour Podcast, Nik and Imran hold a long overdue conversation about United's awful league form with 1 win in 7. They discuss a disappointing draw at Bournemouth, a not so disappointing draw against Liverpool and a crushing defeat at Chelsea - Is time nearly over for Erik Ten Hag?
Then in a special second half of the pod, Nik welcomes Natalie ( https://twitter.com/natalie_17) to discuss the fortunes of Manchester United Women as they booked a place in the FA Cup Final after a historic first ever win vs. Chelsea.
As mentioned before, the Liverpool scenario is commonplace - across all regions of the table, i.e. that a single team maintain an "unsustainable" run over large parts of a season. It happens all the time - happened for us in Jose's second season, for instance ("should have" finished 6th, I think - possibly even worse). Several teams in the last few seasons "should have" been relegated too - but ended up very snugly a fair way up the table.
ETA And Leicester, of course. "Should have" finished 4th the season they won it (which still would have been pretty impressive, one could add).
But as we know, xG is not a perfect prediction tool, and was not designed to be (but it is used as such by many). And statistically speaking, something like that is just an outlier - it's not an indication that the model is seriously flawed. It's a very conspicuous outlier, of course, so there's a story in it.
ETA And Leicester, of course. "Should have" finished 4th the season they won it (which still would have been pretty impressive, one could add).
But as we know, xG is not a perfect prediction tool, and was not designed to be (but it is used as such by many). And statistically speaking, something like that is just an outlier - it's not an indication that the model is seriously flawed. It's a very conspicuous outlier, of course, so there's a story in it.
Last edited:
Useless stats in the long run.
If you have more chances but score none it still means you're shit.
xG is only relevant if you're shit. If you're good, top of the table or actually winning stuff you wont give a damn about xG
It's a trivial stat made like some sort of holy grail when your team was not performing very well
If you have more chances but score none it still means you're shit.
xG is only relevant if you're shit. If you're good, top of the table or actually winning stuff you wont give a damn about xG
It's a trivial stat made like some sort of holy grail when your team was not performing very well
Thanks. Ugly website though.
Had a quick comparison between Understat and Fbref and it seems that Understat has been consistently more lenient in terms of xG (a higher xG for the same shot). It isn't significant though as long as you only compare within the same model.
It is a very limited tool. Since it involves things like blocked shoots too and no not account for chances with no shoots taken at all. Some teams wait or take mainly quality chances for them. Other shoot at everything.
To analyse the flow of a game you need to actually watch the game or look more at depth with the stat. Just looking at a full season stats gives you very little.
To analyse the flow of a game you need to actually watch the game or look more at depth with the stat. Just looking at a full season stats gives you very little.
One of the biggest differences is that the StatsBomb model uses (among many other variables) defensive pressure on the shooter and number of defenders between shot and goalline which Understat's model completely ignores. So in an extreme case a player rounding the goalkeeper outside the box will show up as <0.1 xG on Understat as it gets lumped in with other shots from outside the box. Also you cannot generate more than 1.0 xG from the same chance (if multiple shots are taken) but Understat records every shot individually which can lead to inaccurate numbers.
Yeah you are right, I just checked. Sterling is about on par. Jesus is 7 goals under. Wow!
Or it could just mean that Liverpool have had an exceptional year where everything has gone their way. I think they will drop more points next season. It’s very hard to keep up perfection for 3 seasons in a row.
Last edited:
Two surprises for me are Newcastle, who apparently should have finished 20th this season, and Watford who should have been well clear (and only 1 point behind Spurs).
https://understat.com/league/EPL
https://understat.com/league/EPL
The point of having Data like xG is to highlight anomalies and in doing so highlight teams or players who over-perform or under-perform, generally speaking. The fact that the Premier League table and the xG table don't mirror each other is perfectly normal - it's actually the whole point of doing it in the first place!
If you think about it even for a second, it's perfectly logical that Liverpool might be able to perform so well without actually posting extraordinary xG numbers. Say they are playing Palace away and it's been a tight game of few chances. The ball bobbles around the Liverpool penalty area at 0-0, it falls to Jordan Ayew (sorry to pick on someone!) who has an 'expected' chance of 75% (to score) but hits his shot slightly too close to Allisson (who now has a 40% chance to make the save) and in-turn Allisson, being a world-class goalkeeper, manages to turn it away for a corner.
Ten minutes later, Salah picks the ball up 25-yards from goal, he lines up a shot from an angle with an xG of 0.10 but strikes it sweetly on his left-foot and it flies straight into the top corner to give Liverpool a 1-0 lead, after which they go on to win the game.
Most football matches play out this way and it's one of my biggest frustrations generally when watching football that fans and pundits can't see the millions of tiny fractions that make up the result of a football match. I've lost track of the number of times a pundit says 'such and such team where totally dominant' or 'such and such team where very poor' when in actual fact what we just watched was a 50:50 game decided by a couple of percentage points going in favour of one team on the day.
Utd vs West Ham was a perfect example of this earlier in the season when we where beaten 0-2 away. We where average but created one great chance to score which Maguire fluffed. They where average and created two very difficult opportunities for Yamalenko and Cresswell, both of which where scored. On another day, Maguire scores the easy chance, you win 1-0 and it's labelled a 'professional performance' or 'the type of performance that wins you titles'. Instead, as usual West Ham where 'absolutely fantastic' and Utd where 'hopeless'.
In short, better players (and of course luck!) make the difference when small fractions are involved...who would have thought it?
It's not surprising the bloke who led on the analysis (brentford fella) was a professional gambler because these guys cut through the 'noise' and look at the hard facts, knowing if they keep making 'value' calls against the grain they will win long term.
If you think about it even for a second, it's perfectly logical that Liverpool might be able to perform so well without actually posting extraordinary xG numbers. Say they are playing Palace away and it's been a tight game of few chances. The ball bobbles around the Liverpool penalty area at 0-0, it falls to Jordan Ayew (sorry to pick on someone!) who has an 'expected' chance of 75% (to score) but hits his shot slightly too close to Allisson (who now has a 40% chance to make the save) and in-turn Allisson, being a world-class goalkeeper, manages to turn it away for a corner.
Ten minutes later, Salah picks the ball up 25-yards from goal, he lines up a shot from an angle with an xG of 0.10 but strikes it sweetly on his left-foot and it flies straight into the top corner to give Liverpool a 1-0 lead, after which they go on to win the game.
Most football matches play out this way and it's one of my biggest frustrations generally when watching football that fans and pundits can't see the millions of tiny fractions that make up the result of a football match. I've lost track of the number of times a pundit says 'such and such team where totally dominant' or 'such and such team where very poor' when in actual fact what we just watched was a 50:50 game decided by a couple of percentage points going in favour of one team on the day.
Utd vs West Ham was a perfect example of this earlier in the season when we where beaten 0-2 away. We where average but created one great chance to score which Maguire fluffed. They where average and created two very difficult opportunities for Yamalenko and Cresswell, both of which where scored. On another day, Maguire scores the easy chance, you win 1-0 and it's labelled a 'professional performance' or 'the type of performance that wins you titles'. Instead, as usual West Ham where 'absolutely fantastic' and Utd where 'hopeless'.
In short, better players (and of course luck!) make the difference when small fractions are involved...who would have thought it?
It's not surprising the bloke who led on the analysis (brentford fella) was a professional gambler because these guys cut through the 'noise' and look at the hard facts, knowing if they keep making 'value' calls against the grain they will win long term.
Yeah, that stands out a bit.
I was browsing the xG tables for the past few seasons not long ago, and it seems to be uncommon for a relegated team to under-perform by that many points (almost 14).
Then again, Norwich under-performed significantly too (but would've gone down regardless).
Yep, it could mean that. It is hard to think they will hit around 100 points again, but their XG numbers were similar the season before so I think they are just an efficient team.
The way I usually explain xG is starting from goals. Ideally we'd measure all performance based on goals (or points) but they're rare events. An average team will only score 50 goals over a whole league season which is a very limited sample size in statistical terms. Raw goal tallies are highly susceptible to randomness and can often mask underlying performance.
So the logic was to take a step back from goals, losing some of the ultimate clarity but gaining significantly in sample size. That was done by looking at shots. An average team will have 500 shots per season, which gives analysts more to go on.
It was apparent from the start that using raw shot tallies was extremely limited - not all shots are equal. Over the years, more and more complexity has been added into the shot models (xG models) to account for the variability in shot quality. Some models are better than others. Experimental models are looking at going even further back - possessions (passes, phases of play) etc. Again, this increases the sample size but at the cost of losing even more clarity on the ultimate metric.
Liverpool have been fortunate in some ways (high xGA) but valid "over"-performance in other ways (strong set pieces). Regardless of coaching, it seems unlikely that we'll see the same level of over-performance again next season. Admittedly I said the same last year... but I'll hold faith in the numbers one more time.
So the logic was to take a step back from goals, losing some of the ultimate clarity but gaining significantly in sample size. That was done by looking at shots. An average team will have 500 shots per season, which gives analysts more to go on.
It was apparent from the start that using raw shot tallies was extremely limited - not all shots are equal. Over the years, more and more complexity has been added into the shot models (xG models) to account for the variability in shot quality. Some models are better than others. Experimental models are looking at going even further back - possessions (passes, phases of play) etc. Again, this increases the sample size but at the cost of losing even more clarity on the ultimate metric.
Liverpool have been fortunate in some ways (high xGA) but valid "over"-performance in other ways (strong set pieces). Regardless of coaching, it seems unlikely that we'll see the same level of over-performance again next season. Admittedly I said the same last year... but I'll hold faith in the numbers one more time.
Think that's to be expected (pardon the pun).
All the xG values are averages and every team naturally deviates from that. Every shot does if you think about it. There can only be two outcomes: goal or no goal, 1 or 0. But every shot taken has a value that is not 1 or 0 but in between. So every goal in isolation is technically a case of outperfoming the expectation as the expectation is never 1. It makes sense then the winners are likely to have overperformed unless they are so good to be able to 'afford' underperforming. That's not very scientific language but I hope it makes sense.
Last edited:
Well, yes.
And there isn't a single PL champion in recent years that did not significantly out-perform xP.
Liverpool are still extreme though. They outperformed xP by almost 25 points. By comparison, City out-performed xP by 7 the season before.
There are people who disparage xG yet rely on weather forecasts every morning.
Would have been considered very nice in 2001.
Yeah and it is only overperforming in converting chances and maybe defending/goalkeeping. It is natural that quality teams in good momentum and flow can do that.
You can still dominate games without missing lots of chances and that needs to be accounted for.
Still I think Liverpool have not overperformed xG that much though and it is more that they get narrow wins more often than expected.
To my shame, so did I today. Off by 10 degrees! How is that even possible?
Yes...you're right. The under-performance, as such, is no surprise (relegated teams over-performing xP are outliers). Still, Watford are the worst under-performers in recent memory - out of teams that actually ended up going down.
Excellent post.
False narratives are really ingrained throughout the football ecosystem - pundits, supporters, managers too. People love inevitability narratives, the idea of a "professional performance," one side being "always comfortable" against a worse side, tactical substitutions making a decisive difference, etc. They can't wrap their heads around the idea that a huge proportion of matches are pretty even and decided by how the ball happens to bounce on a few critical occasions or whether Player A can take his 30% chance or Player B can take his 20% chance.
Weirdly antagonistic way of looking at it.
xG stats is just a statistical tool to analyze performance. Most if not all clubs use it to good effect. Its just people who argue for it or against it without properly understanding it and having incorrect expectations of it is the biggest problem.
Yes that was a really good post.
The part about the pundits is so true and yet frustrating because they are the ones creating the (often false) talking points.
The part about the pundits is so true and yet frustrating because they are the ones creating the (often false) talking points.
True - and a major factor there is that both fans and journos tend to use it as a prediction tool, or more precisely as a tool for predicting short-term (relatively speaking) results. And it isn't reliable at all for that.
Is this really the case though? I'm no expert in this aspect and I'm not pretending to be one, but I genuinely think every xG model takes defensive actions into account and it's essentially one of the definitions of xG. I've looked up the situation you mentioned and the closest thing I can find is Harvey Barnes' goal against Villa. The xG according to Understat is 0.45, which is significantly higher than a random shot outside the box (<0.1).
There's no such thing as an expected goal, it's fecking binary
Thank you for your analysis, I never saw it from that perspective.
This is actually one of the reasons why we have a nice xGA stat this season. Many teams, particularly in the first half of the season, scored a goal with a low xG and then decided to sit back for the rest of the match.
I don't know. I used it to predict Vardy's likely goal output back in November after many were claiming he'd hit over 30 and I was pretty cock on.
I could have bodged that in fairness and just got there by accident. It's not an accurate predictor of course but it suggests what you may reasonably expect, as loaded as such a statement is.
Last edited:
I would be a lot more skeptical if it was purported to be a prediction tool. But yeah, it is not.
No problem, happy to offer a different take on this widely debated topic.
On a side note (kind of), I was in the military back in the day, serving at a naval base: one of my duties was to raise the flag every morning. That was a bit of a science, as you couldn't just raise any flag - it had to be the right size and shape according to the wind conditions. My superior officer was an old fecker who had been on boats all his life. He would stick his nose out the window, smell the air for about ten seconds, and then determine precisely how strong the wind was (mps, m/s, whathaveya). He was never wrong. And he could also "smell" whether it would rain or not - and pretty feckin' accurately determine, based on the clouds, WHEN it would start raining too.
Not sure whether he'd be a fan of xG, though. Probably not.
Was about to post that. A lot of clubs use proprietary models or at least buy expensive versions that Joe Public doesnt get to see. It's become a huge part of player scouting.
Understat is the equivalent of whoscored.com, i.e good for a rough idea but otherwise very flawed.
On a side note (kind of), I was in the military back in the day, serving at a naval base: one of my duties was to raise the flag every morning. That was a bit of a science, as you couldn't just raise any flag - it had to be the right size and shape according to the wind conditions. My superior officer was an old fecker who had been on boats all his life. He would stick his nose out the window, smell the air for about ten seconds, and then determine precisely how strong the wind was (mps, m/s, whathaveya). He was never wrong. And he could also "smell" whether it would rain or not - and pretty feckin' accurately determine, based on the clouds, WHEN it would start raining too.
Not sure whether he'd be a fan of xG, though. Probably not.
My grandpa was a farmer and had the same skillset. Knew when rain would come, knew the right day to plant, and so on.Although the quality of a great striker can also be to easily create good chances without being a particularly good finisher, in which case the result would be that the team has a relatively high xG, without necessarily outperforming it.
Because whenever it was brought up vehemently is when your team is performing to the opposite of xG.
Manchester Under Mou : Pts wise vs. xG : very different
Manchester Under Ole : Pts wise vs. xG : very different
For me the only thing that matters in the league is position in the table, anything else is Ifs and buts. It should be taken as nothing but a minor foot note. At the end of the day, you either score, or you don't.
Yep, couldn't give a feck about xG and I tend to tune out as soon as someone brings it up. I don't think football lends itself to those kinds of statistics as well as other sports like baseball or basketball.
Each to their own I guess. If someone's substituting xg stats for points or score then obviously thats wrong. Its just a measure for performance. And as its football, performance doesn't always equal result.
Maybe I'm doing them a disservice or they have updated their model. Not sure. I'm fairly certain they at least don't account for pressure on the ball though (i.e. how close are the defenders to the shooter or are they blocking parts of the goal). The second issue definitely persists. Chelsea had an 1.3 xG chance vs City for example because they took several high-quality shots in the same phase of play.