Hardly suprising but xG isn't always that great.
https://www.theguardian.com/football/2020/aug/09/liverpool-xg-jurgen-klopp
https://www.theguardian.com/football/2020/aug/09/liverpool-xg-jurgen-klopp
Last edited:
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.
xG is not great but it is far better than any other statistics we have. For most part, those statistics were number of shots, number of shots on target, number of corners and percentage of ball possession. xG is just another stat in that direction, but that it describes better the game and has more predictive value than them.
Obviously it will be improved. I don’t know if we can get soon into some point of totally describing the game numerically without any human intervention, but these statistical analysis are definitely going to improve a lot and become very important in player selection and transfers.
Obviously it will be improved. I don’t know if we can get soon into some point of totally describing the game numerically without any human intervention, but these statistical analysis are definitely going to improve a lot and become very important in player selection and transfers.
It's not a limitation but more of an incorrect expectation that people have of a statistical tool doing statistical analysis of an environment with millions of variables. It's a measure of performance, and we as football fans should know performance doesn't always equal result in an individual game basis. But over a big enough sample size, the overall result reflects the overall performance with some outliers.
I always thought it was a bit pants because of the Saha example. When we would put him through on goal he wouldnt usually put the ball in the back of the net, but if he had a shooting chance from just inside the box at an angle he'd smash it home. For some players a certain type of attempt or shot from certain areas are easier than what would be generally considered to be an easier chance. I dont think you can generalize a goalscorer like that, it doesnt really matter what the method is and there are many players who have different methods, the important part is that you score.
Jesus Christ, you are using an individual sample to discredit a statistical method.
I always thought that the chances of getting head after tossing a coin are not 50 percent, because once I tossed a coin twice in a row and got tail both times.
This
Weird article really. No one ever thought it was a predictor of the future. That said, I think this seasons data tells you to bet on City next year to win the title, unfortunately.
They already are very important in player recruitment. Liverpool being an early adopter ironically. Brentford are probably the most fundamentalist club in this regard and have reaped big rewards.
Yes an individual sample of a striker who was better at something than another thing... Like all strikers.
You arent better at getting one side of the coin. Its just luck. If you can't understand the difference you should stay away from all analysis
I guess so. And likely it will become even more important in the future, with the stat being better.
To be fair, I have no idea why it took so long to become important in football. Moneyball in the US was a thing since the turn of the century when Brad Pitt built a solid team from farmers using very simple data analysis.
Also ironically, Liverpool fans were one of the early adopters of the alternative league table, so this is going to crush them.
Just read this article (after seeing your post) on Brentford: https://www.google.com/amp/s/talksp...rtest-club-championship-leicester-fa-cup/amp/
Pretty cool what they are doing. Hopefully they’ll get their reward and reach UCL before the rest of the league starts doing the same.
Is there not just a simple issue with aggregation?
Like three fairly difficult chances (e.g. 0.33 probability) is taken to be equal to an absolute stone wall sitter (1.0 probability)?
Liverpool have also scored quite a few from outside the box, and Salah/Mane tend to score their goals from wide angles.
Like three fairly difficult chances (e.g. 0.33 probability) is taken to be equal to an absolute stone wall sitter (1.0 probability)?
Liverpool have also scored quite a few from outside the box, and Salah/Mane tend to score their goals from wide angles.
Or others factors have been at play. Not hard to guess what...., just go watch the first 20 games of the season for City and Liverpool and see if you notice anything.
its not a bug its a feature
interestingly we finished 6th on Xpts in 17/18
https://understat.com/league/EPL/2017
I'm guessing Dea Gea's form was one of the reasons for that
interestingly we finished 6th on Xpts in 17/18
https://understat.com/league/EPL/2017
I'm guessing Dea Gea's form was one of the reasons for that
Last edited:
Carrying Mourinho to 2nd is maybe David de Gea's greatest achievement
Is there a full write up explaining exactly what expected goals against is?
The site above is suggesting that City were positive, almost the same number as us but they were playing a DM as a CB for most of the season and I think we'd generally agree that their defense was their weak point this season, not their midfield or attack.
The site above is suggesting that City were positive, almost the same number as us but they were playing a DM as a CB for most of the season and I think we'd generally agree that their defense was their weak point this season, not their midfield or attack.
Doesn't that highlight the question, how good is our own defence?
City completely dominate most teams and concede few chances which keeps their xGA low, as a team they defend well because of their domination of the ball. No team in the Premier League concedes fewer shots per game than them. You'd have to look at individual games to see if certain individuals let them down at key moments and if that became a theme.
Of course other factors were at play. Like most statistics calculating them is often the easy bit. Yet people love them and think they have to be the whole truth irrespective of interpretation and/or context.
Their defense is weak when they are exposed. But most of the time they dominate the ball up the pitch and are very good at snuffing out counter attacks. This season having Rodri in front of their defense instead of Fernandinho has meant that they haven't been very effective at stopping counters which has been their downfall.
Some good points in this thread. I'd just like to offer a small correction on the bolded part.
Short and medium-term, expected goals certainly do predict future performances better than any other metric. And much, much better than actual goals and points as these are infrequent events and there is no point reading too much into a sample of 5-10 results. Now, over time results and performance data will converge and when you have a case such as Liverpool whose results have been way better than their performances data for two seasons now, it makes sense to assume that there is something the data does not capture, as alluded to in the article. Even then, they remain a huge statistical outlier and you'd expect this to level out somewhat. The best reflection of the true performance level of a team will always be the Asian betting markets. And City were a healthy favourite before home advantage when the two sides met at the Etihad a few weeks ago and have continued to be rated the best team in the league throughout the entire season.
Last edited:
Has it though?
Same amount conceded on the counter as us, and thats on the low end. Chelsea and Liverpool both conceded more goals on the counter
Ok, I'm not sure how counter attacks are defined here but I mean more like defensive transitions (how Mourinho puts it) when the defensive shape is not settled when possession is lost.Has it though?
Same amount conceded on the counter as us, and thats on the low end. Chelsea and Liverpool both conceded more goals on the counter
I should have said an exact predictor of the future. I think the data would tell a gambler to bet on City next season. There is something else going on of course. Isn't it mostly individual players outperforming the averages and a some 'luck' with opposition underperforming the averages against them? The 'luck' factor for Liverpool this season certainly seems to have been more pronounced than last season.
How were we on xPenalties?
Of course theres problems. Its always had its flaws. Its the supporters that push it forward like its the be all and end all that lets it down.
It should be supplemental, something to beef up a point. Not just laid out in a post as if its the trump card of an argument like its used so many times on here.
It should be supplemental, something to beef up a point. Not just laid out in a post as if its the trump card of an argument like its used so many times on here.
I mean looking at the league table and the xG table you notice that while Liverpool have certainly outperformed their goals scored and conceded, it is not by an unreasonable margin. That's what a world class goalkeeper and a lethal strikeforce give you. It is their points tally that is through the roof. And when you think about their season and look at individual results, the one thing that stands out is the number of wins by the odd goal, suggesting they scored (and condeded) a lot of their goals at just the right time. I don't want to talk about 'luck' as it is hard to define and just sounds like an excuse when you don't have any other arguments but that alone makes me question whether their (results) form is sustainable.
Last edited:
completed it
I use 'luck' as a term because before xG when people saw games where sides were missing a lot of of 'big' chances in defeats they would call the winning side 'lucky'. Now with xG data that dynamic can be better understood.
Lucien Favre's style has been messing with the xG model since it's been around. Here's a detailled article exploring why xG failed to capture the true quality of his teams, which naturally highlights the model's limitations: https://thefutebolist.wordpress.com...rperform-expected-goals-part-one-the-defence/
Does xG take in to consideration the quality of the player? and if so how? Messi sure outperforms xG nearly every year because his ability to score is higher than everyone not names Ronaldo.
xG model's aim is not to calculate the xPts tbf. It does have a correlation to the actual points, but in fairness that metric is a derivative of the xG metric.
The reason why they are overachieving their xPts model is quite understandable - unlike most big teams, Pool do not go for relentless attacking play till the 90th minute - something City and the likes do. Due to this, their xG is in most cases lower, which results in a lower xPts for such teams.
Now them overachieving xG is an interesting argument, but then again one can argue that they've been really good at set pieces, which is where they've considerably outperformed their xG numbers.
The reason why they are overachieving their xPts model is quite understandable - unlike most big teams, Pool do not go for relentless attacking play till the 90th minute - something City and the likes do. Due to this, their xG is in most cases lower, which results in a lower xPts for such teams.
Now them overachieving xG is an interesting argument, but then again one can argue that they've been really good at set pieces, which is where they've considerably outperformed their xG numbers.
Last edited:
No, it's based on averages, so better teams should have better strikers, therefore be outperforming their XG.
I'd argue comparing Liverpool's vs. City's XG does not show that XG is flawed, it shows that Liverpool's shot takers are more clinical and better at shooting than City's. You could conclude that City therefore need better forwards, but their front line are particularly good at creating chances, and swapping out Sterling with a more clinical forward for example would reduce chance creation, so it's a fine balance.
City = better at chance creating.
Liverpool = better at converting chances.
This is all XG shows, and it does that well. If you think it's a predictor of the future, then yes it is flawed.
Any predictive model isn't 100% accurate, nor are things like human judgement, or things like machine learning in general.
Not surprising at all.
Not surprising at all.
Interesting article but as the models develop they'll be able to account for things like defensive pressure and even more nuance in greater and greater detail. As the article mentions some models already do account for defensive pressure.
I didn't post this to disprove xG as a concept and if you look at the seasons that followed the article you'll see that Favre at least stopped outperforming xGA from then onwards:
Though one the other hand one could argue that this may also be, because his rather passive defending approach didn't work out very well with Dortmund's squad (*cough* Hakimi + Sancho right wing *cough*) and expectations.
Last edited:
Keep in mind there are multiple xG models out there. Often we see the understat one used, because it's free, but the understat model for example is notoriously one of the most basic and not particularly good
To my knowledge they and infogol are also the only ones to use xPoints
xPoints based on xG are not good models in general imho, unless they manage to introduce game states into it
To my knowledge they and infogol are also the only ones to use xPoints
xPoints based on xG are not good models in general imho, unless they manage to introduce game states into it
On top of that the Understat data misses a complete range of xG values, i.e. there are no shots recorded between 0.13 and 0.18 goals or something (don't remember the exact figures) showing how basic their model is.
Sterling is actually clinical in taking the chances. The wasteful one is Jesus.
Are there any better free models?