I think up to our second goal they totally dominated us really even if they gifted us a goal. Although after that goal we only needed one goal and it really changed things. We didn't create much, but got much better control and you got the feeling we could maybe get the win.
thats my point. One shot from position x goes in, even if its an impossible goal, if someone else ever shoots from that same position, does the stat expect them to score?
The xG value is not assigned by looking up the area of the pitch from which the shot was taken in a database of historical shots. That data is only used to derive a formula through logistic regression. Here are some example formulas: http://crabstats.blogspot.com/2017/08/my-xg-models.html
But when calculating the xG in any given game, isn't there some minimum value that all shots below that threshold get in order to minimise the amount of decimals? For example, if the minimum is 0.02xG, even a shot like this that has 0.001 "real" xG would get 0.02xG on the scoreboard, thereby artificially inflating the score?
Don't think so but they might round to the nearest hundredth.
Ah, but you'd think that rounding to xG=0.00 wouldn't be possible since any shot has a (possibly microscopic) chance to go in, so that makes 0.01 the minimum I suppose? Thanks for the link.
I wonder why it is logistic regression in the end. It is a very simple classifier, but one of the worst performing one. I guess that they want some tradeoff between accuracy and simplicity/interpretability. At the same time, just changing the classifier to xgboost/random forests probably will give a slight increase in accuracy (though you need to calibrate them in order to get probabilities).
I think that some models have a minimal 0.1xG (or was it 0.01xG?) which means that any shot gets at least 0.1xG. In which case a shot from the middle of the pitch despite having 1/1000 chance of being a goal, gets a 0.1xG. Which might explain why some team might have a close to 1 of so xG in matches when they didn't have a clear cut chance.
We are winning the most important competition.
Yeah it shows that people put too much stock in xG. It's a stat like any other. Useful in some ways, but it doesnt show which side has been better.
Doesn't it actually do the opposite? Most will say we should have won with the chances we had and 'xg' will back it up, giving it some sort of credibility.
I read it as saying our 'luck' (form) will run out due to us having a low xG. But today we had a high xG and lost 2-0.
In a nut shell:
I get the impression it's probably more useful as a stat when you have a larger sample - with individual players or teams you're going to have a lot of fluctuations due to the unpredictable nature of football games in general, but if you take a collection teams you'll maybe end up being nearer the mean, if that sort of makes sense. So it's probably useful to consider, and can sometimes highlight teams who're not doing as well as they should be or vice versa, but it's not necessarily something to be obsessed over either.
Geddit.
Are we really going to use this game to mock xG? The one where we missed sitters and most of us feel like we had created enough to get a goal or two?Hey, at least we won the xG today!
Your luck did run out in theory though. You've relied on converting a high % of chances and the opposition failing to convert and today the opposite happened and you lost.
Bizarre, isn’t? Luddites and Brexit campaigning all over again.
Yes - but the angle people have been going with lately is that we aren't "dominant" enough, and that we don't create enough on the whole, i.e. that we've been efficient in capitalizing on too few chances, relatively speaking, to a degree that isn't "sustainable" (buzzword of the past few weeks).
It can tell you about the amount/quality of chances a team gets. In a way better than just looking at the number of shoots a team has.
I mean if you took all those things into account then the xG would be exactly whats on the scoreboard.
Broadly speaking, I agree with this. This can’t be the peak. Ole needs to keep improving us and I hope he does.
agree.
Mourinho this season is going to be a bit skewed. So often we were down a goal in 10 minutes, what we had to spend the entire game throwing the kitchen sink at the opposition to try and get anything.
I think we can perform at the same level as Liverpool though. Probably need a bit more quality on our right hand side. Lingard and Martial is not as good as Mane and Salah.
Had a look at your model. Can you give me an idea of how it performs vs actual outcomes using a hold out sample. (I'm assuming you can do this fairly easily - and probably already have - given that you have two distinct datasets.)
