I think, you have to really get it, then you can use it to gain some information about a single match as well. For example - if your xG is below 0.5 then either the other team had an absolute field day in terms of defending or you struggled with creating useful chances. If your xG is over 2.0 then it means that you created a fair bit of chances (maybe check for amount of penalties first) which usually should be enough to score at least once.
There's nothing wrong with being a counter based team and growing to get better to control games when you have the players that can control though. Also it doesn't really have a lower ceiling. The best teams just are able to do everything to a high level. You're very much capable of winning everything by focusing on being a direct transition team in the bigger games but who dominates games against the smaller teams.
We have literally been direct constantly since SAF aside from when we had LvG.
Nah, we've generally been a solid top 4 side when we have focused on the counter and when we didn't think about trying to be more than we are.
It really does and we felt that. That style of play was never going to be the way we were going to get back challenging for the league.
Eh I guess it depends. It does have a lower ceiling if you are competing with Pep's peak teams. But we shouldn't base our plan on how to beat Pep Guardiola. City isn't what they used to be, and nobody controls games like they used to.
PSG literally set up to smash the ball long and win second balls from throw ins given to the oppo, and absolutely annihilated Liverpool and City both. There's room for nuance in how we play.
So I guess the question is are we getting better, or is that mainly regression to the mean.
Makes sense, earlier this season we were wasting a lot of good chances. Do the penalty misses factor in that too?
I think the regression is luck factors while performances are getting better.
Must be very, very difficult with headers because so much of it depends on how the ball arrives on the head of the attacking player and also the positioning of the goalkeeper as a result.
I think this isn't something that should be labeled an "xG Problem" since from my understanding, different sources use different models to get to their numbers. The simpler models just look for position of the shot and the more advanced ones take more factors into account. That'll never be perfect but it still remains the most unbiased way of looking at it. Since even when there are some odd things in the model, they are applied to all games the same way.
Fotmob is what I typically use, and I think they use Opta data (and it tends to match up perfectly with the data they show on tv when I'm watching the games vs checking out the scores/data on fotmob. Fbref as well - they have per match data and have Maguire with 0.1xG vs Liverpool while de ligt is at 0.4 for that match.
Thanks, the data on fbref match report is per player though, right? Not per shot as with understat?
I agree
I see your point. I just think that this is just statistical "stuff" in a sense that there isn't somebody sitting thinking "this chance should have an xG of 0.23" or something so there are no "mistakes" at least, thats my take on it. They just take the parameters of a shot, transfer the ones relevant for their model and check their database for shots with similar parameters and then establish how many ended up a goal and thats then the xG for that particular shot.
Yeah its per player, but they had 1 shot each so it's easy to check for that one.
Ah, thats actually smart, didn't think of that. But "only" works for players who won't shoot on goal more than once. Just found the individual shot stats on Futmob, thanks for the info
Fair enough
The problem is headed chances in general. Maguire's header for example should be much higher than 0.09, but fair enough if you think that's accurate. Compare it to Gakpo's just a bit later, at 0.30, or de ligts after that where he was stretching to get a head on at from a set piece, at 0.43. or a Van Dijk one in the first half at 0.21.
It's not a scientific fact at all. You are looking at the value of the xG as if it is because a model compares it based off similar data, when it is clearly failing to accurately read the situation and chance. The model is trying to fit all in as the same but the numbers very clearly look extremely wrong for headed chances, and that is a common theme whenever you just look into it.
That's very strange. The Bruno chance, for instance, was a very good save. I'd expect that to be a goal at least every third time. Welbeck's freekick probably goes in 1:15 or so. Do you have a source for the individual xG of these chances?
Was just looking at fotmob data for individual shots. Understat also does it. But yeah the Bruno one is definitely low, while Welbecks is way too high given it's a free kick and most free kicks from similar positions tend to be far lower (talking like <0.05).
Yep definitely this.
It's not a stat based on fact. It's an analytical model that is trying to quantify a unique situation. How that model collects information is open to interpretation and criticism. A shot is a shot. That is a fact. One model saying Maguire's xG on his chance was 0.25 and another saying 0.09 is not factual. That is 2 different models reading the chance a very different way. Understat has the xG vs Liverpool at 2.61 for Liverpool and 2.53 for United. Fotmob has it completely differently.
Get the same feeling
Those models differ in the number of context parameters they add up. The easiest model might just look only for the position a shot is taken from. Over thousands of matches you'll maybe find a few hundred similar shots and take the average from there. A more sophisticated model will integrate the foot the shooter takes the shot with, the position of the goal keeper and whether the shooter is under pressure or not. The more parameters the lower the number of comparable shots. It isn't different models using wildly different formulas to create a somewhat quasi-subjective set of numbers. It is "just" different parameters from model to model. If you have ideas to improve the current models, I am sure people are up for it. If you find a way to incorporate stretch, player height and level of contestion thats great - the trouble won't be to generate other averages - the trouble will be how to annotate games detailed enough to have that level of data available.
My point on the Maguire one (and headers in general) is that 0.09 is obviously too low, especially as a comparison to other headers within the same model. If your model takes into account keeper position, defender position, the cross details etc - then Maguire's chance should go up from the basic one. He was wide open on the back post and had a relatively simple header to place it on target. I'm using Maguire's header, de ligts header, and multiple other headed chances as pretty jarring examples that xG just has a really hard time quantifying headers. Of course there are reasons for it. But when 20-30% of your set of data is pretty much nonsense numbers, it's kind of hard to take it seriously.
Yeah, no. It's not hard to understand at all. My point is that for our game against Brighton it seems, for whatever reason, that the xG doesn't really accurately reflect quantity and quality of chances in the aggregate. I can't say I came away from that game and thought that us and Brighton had accumulated roughly the same opportunities, but that's what the xG shows. Perhaps it has something to do with quirks related to how each chance is evaluated, like @bosnian_red pointed out.
