bucky
Full Member
- Joined
- Apr 5, 2011
- Messages
- 9,610
Is there an explanation as to what this XG stuff is or what it means anywhere in this thread? I've no idea what you're all going on about.
g = window.googletag || {}; googletag.cmd = googletag.cmd || []; window.googletag = googletag; googletag.cmd.push(function() { var interstitialSlot = googletag.defineOutOfPageSlot('/17085479/redcafe_gam_interstitial', googletag.enums.OutOfPageFormat.INTERSTITIAL); if (interstitialSlot) { interstitialSlot.addService(googletag.pubads()); } });
Is there an explanation as to what this XG stuff is or what it means anywhere in this thread? I've no idea what you're all going on about.
That's a fair question. Personally I'm not in love with statistics and so certainly don't arm myself with them in advance or anything like that, but I play and coach a fair bit and like to get as much out of watching a game as possible. I think when you understand tactics, you get more out of watching, and the same applies to statistics.This is a to each their own debate if there ever was one, but psychologically I am intrigued to know:
What is the primary reason for following or arming yourself with this type of statistic ahead of a match you are about to watch? Is it to make smarter gambling/fantasy football decisions? Is it to pull back emotionally - stop yourself getting too high or too low early in a match? Or is it just that you (this is a general "you" and not you specifically) prefer discussing metrics about an event put on for entertainment?
Again, to each their own, but I am genuinely curious what viewers (and not the data scientists themselves, as this movement is no doubt a jobs creator in that field) are getting out of this. My personal opinion, flawed as it may be, is that the more metrics play a role in sport, the less entertaining they become.
Stop making up what said.I don't even know what to say to that. "Data good. Statistics bad"? My God.
I'm groping here but do you mean that raw data is good and manly, but if you add it up or make a ratio out of it, it's stupid and evil? So, in your mind, if they gather heart rate data it's ok, but if they analyze it (for example by combining it with distance covered data) then it's dumb and irrelevant?
I mean, WTF?
We have every team in the league paying big money for game data and analytics, and people on here saying that "I bet Ole doesn't" use it, and "data is ok but statistics is a step too far". It's fkn hysterical.
That's just not true. For most of our games under Ole, it reflected to what happened in the game. And the games people have problems with in terms of xG figures, we still won. Which is the whole point. Just because a team has a higher xG than another team in a single game, doesn't mean it's supposed to win that game. Ultimately if you look at it in isolation it means as much as a team having more possession. I love when coaches comes out with it: 'we dominated them, we had more possession', but if that doesn't result into chances and if those chances aren't converted, then I'd say, it's fairly obvious, they weren't the better team.I think we are probably on the same page then mate.
I’m not saying the concept is poor, merely the crude excecution of it. It’s miles and miles away from being a statistic you can induce anything from.
Just look at any weekend of xG Scoreline vs Real Scorelines and it’s laughably wrong almost every time.
Now apparently we are supposed to be informed by that which team under or over performed in front of goal in a given match.
The very fact it gets it so wrong in almost every instance shows that the model is fundamentally flawed. If it was a trust worthy stat, the vast majority of matches would fall in line with xG and there would be a few outliers where over or under performance occur.
Maybe you don't realise that "statistic" just means data that has been manipulated in some way. "Heart rate", given that it is an average of individual pulses over time, is a statistic. But really, the explanation is pointless because you're just throwing chaff in place of actually making an argument.Stop making up what said.
The example given wasn't a statistic. I did not say data was good and statistics were bad.
Statistics are tools but the thing with them is that they require good data, a decent sample size and to be used in an appropriate and meaningful way.
The biggest problem with xG (and how people view it) is it doesnt capture the state/flow of the game at all. That Fulham game we dominated and created all the chances and then relaxed at 3-0 up and Babel missed a chance worth pretty close to 1 xg. That gives them an even xG to us even though we hammered them all game, created plenty, scored and relaxed when if needed we could and would have pushed on more. Same with Leicester.That's just not true. For most of our games under Ole, it reflected to what happened in the game. And the games people have problems with in terms of xG figures, we still won. Which is the whole point. Just because a team has a higher xG than another team in a single game, doesn't mean it's supposed to win that game. Ultimately if you look at it in isolation it means as much as a team having more possession. I love when coaches comes out with it: 'we dominated them, we had more possession', but if that doesn't result into chances and if those chances aren't converted, then I'd say, it's fairly obvious, they weren't the better team.
The game against PSG highlights that. We were being dominated in terms of possession, we had the higher xG, we scored more, hence we deserved to win. Even if PSG would have had a higher xG than us, if they don't convert their chances or we are a bit lucky and took our chances or we quite simply have the best goalkeeper in the world, that doesn't mean that we don't deserve to win, if we actually take our chances. A worse team like Palace and Fulham than us having a higher xG than us in a single game quite often will mean that we win, because we are better at taking our chances, hence we are actually the better team and those teams are where they are in the league for a reason.
Valid points. Good post.The biggest problem with xG (and how people view it) is it doesnt capture the state/flow of the game at all. That Fulham game we dominated and created all the chances and then relaxed at 3-0 up and Babel missed a chance worth pretty close to 1 xg. That gives them an even xG to us even though we hammered them all game, created plenty, scored and relaxed when if needed we could and would have pushed on more. Same with Leicester.
Another example is the Spurs game. When the game was 0-0, United controlled it and took the lead. 2nd half they were desperate so threw every thing forward and created pretty much all their xG because they never took a chance. Loads of people came out with the line of "on another day it couldve been 4-1 Spurs" which is bollocks. On another day they could have equalized. That's it. At 1-1, who is to say the game wouldnt have gone back to how it was at 0-0? Because we definitely wouldn't have been as defensive at 1-1 and they wouldnt be as gung ho.
People use it as a way to represent how a game was, how a team is performing long term, etc when it cant do any of these things well. Timed xG charts (understat does these) work pretty well because you can then see how the game changed with goals, but nobody splits it all up based on xG when leading, when losing or when the game is drawn. If anything, those would all show more information.
Great video.
I am just curious. Do those splits indicate that xG goes up when the score is uneven (i.e. when the game opens up)?The biggest problem with xG (and how people view it) is it doesnt capture the state/flow of the game at all. That Fulham game we dominated and created all the chances and then relaxed at 3-0 up and Babel missed a chance worth pretty close to 1 xg. That gives them an even xG to us even though we hammered them all game, created plenty, scored and relaxed when if needed we could and would have pushed on more. Same with Leicester.
Another example is the Spurs game. When the game was 0-0, United controlled it and took the lead. 2nd half they were desperate so threw every thing forward and created pretty much all their xG because they never took a chance. Loads of people came out with the line of "on another day it couldve been 4-1 Spurs" which is bollocks. On another day they could have equalized. That's it. At 1-1, who is to say the game wouldnt have gone back to how it was at 0-0? Because we definitely wouldn't have been as defensive at 1-1 and they wouldnt be as gung ho.
People use it as a way to represent how a game was, how a team is performing long term, etc when it cant do any of these things well. Timed xG charts (understat does these) work pretty well because you can then see how the game changed with goals, but nobody splits it all up based on xG when leading, when losing or when the game is drawn. If anything, those would all show more information.
https://understat.com/match/9481I am just curious. Do those splits indicate that xG goes up when the score is uneven (i.e. when the game opens up)?
So the biggest problem is that it doesn't capture game states, and you then use it to try to describe three game states?The biggest problem with xG (and how people view it) is it doesnt capture the state/flow of the game at all. [...] People use it as a way to represent how a game was, how a team is performing long term, etc when it cant do any of these things well.
In 10 pages now, I've yet to see a consistent explanation of what its usefulness is. This would imply a lack of understanding among its sales force, or that the value of the product is inherently unknown and therefore shite is just being thrown against the proverbial wall.So the biggest problem is that it doesn't capture game states, and you then use it to try to describe three game states?
It doesn't try to capture game states (though people are working to find statistically valid ways of doing so in future). Why you would use this as a criticism is beyond me.
Really, your criticism rests on misusing xG and then complaining that people misuse it. Nobody who understands it claims that it is predictive or that it describes a game perfectly or how the game flowed. You're the person doing that.
Almost the opposite. Juve's results early in the season were terrible but the underlying stats showed that their performances were actually pretty good. They were basically having bad luck (shots going wide, etc.) Luck plays a big part in football (which is why literally NOBODY who understands xG claims it is predictive for any one game).In the Opta video, yes it is simple to understand that data is collected and percentages of chances are calculated. In the Juventus example at the end, xG did what? Told us that in a string of losses early in the season Juve played poorly, and in a string of wins late in the season they played well?
Literally most people call it a predictor for future performance. And yeah, my problem is mostly with people misusing xG and the people who run those accounts selling it as something it isnt. I dont have much of a problem with xG itself, it's a stat like possession/shots or whatever. The main difference though is there are many things that are pretty much impossible to capture and leads to occasional massive difference between models for one game (see Palace vs United and the xG from understat, Caley and 11tegen11).So the biggest problem is that it doesn't capture game states, and you then use it to try to describe three game states?
It doesn't try to capture game states (though people are working to find statistically valid ways of doing so in future). Why you would use this as a criticism is beyond me.
Really, your criticism rests on misusing xG and then complaining that people misuse it. Nobody who understands it claims that it is predictive or that it describes a game perfectly or how the game flowed. You're the person doing that.
I don't really see many experts making the claims you describe. Maybe there are and I just haven't noticed them. Lots of fans misunderstand them, of course, as you say and as this thread illustrates magnificently.Literally most people call it a predictor for future performance. And yeah, my problem is mostly with people misusing xG and the people who run those accounts selling it as something it isnt. I dont have much of a problem with xG itself, it's a stat like possession/shots or whatever. The main difference though is there are many things that are pretty much impossible to capture and leads to occasional massive difference between models for one game (see Palace vs United and the xG from understat, Caley and 11tegen11).
I'm giving suggestions on how it can be further improved as a statistic, because it's lacking a lot of context.
The whole purpose of xG in the first place was to get a more accurate representation of a game because shot counts can be skewed by long range attempts vs clear cut chances. Collected over time it brings a decent idea on team level offensively and defensively. Football analysts on twitter then post raw xG totals after the game and try to claim a team is lucky to win essentially. Well if you're breaking down stats like that and trying to assess it like that, then you have to provide further context. Only way to do that is to include the timing and breaking it down between when the game is drawn/winning/losing.
With Leicester they just had confidence and momentum that season. Nothing did seem to stop them.The xG numbers (Understat) for LVG's second season are interesting:
* 8th in the table (behind Chelsea).
* But outperformed both xG and xGA (+4 roughly for both).
* And outperformed xP by ten.
Meanwhile Arsenal win according to the model, followed by City and Spurs (identical to actual position).
Leicester 4th, outperforming the xP by 12 - xG slightly above actual figure, xGA significantly below (meaning they outperformed the model).
And...yeah, no idea where I'm going with this, just found it a bit interesting.
The important thing about that season, we were 10th! in terms of xG overall, which is a testament to how shite we were offensively, while van Gaal expected us to score a lot more from as little as we had. We were already overperforming in terms of finishing and he was still complaining about our finishing, while the obvious issue was that we were creating nowhere near enough.The xG numbers (Understat) for LVG's second season are interesting:
* 8th in the table (behind Chelsea).
* But outperformed both xG and xGA (+4 roughly for both).
* And outperformed xP by ten.
Meanwhile Arsenal win according to the model, followed by City and Spurs (identical to actual position).
Leicester 4th, outperforming the xP by 12 - xG slightly above actual figure, xGA significantly below (meaning they outperformed the model).
And...yeah, no idea where I'm going with this, just found it a bit interesting.
To be fair, we all complained that we weren't creating many chances. We were clinical but we were creating chances as much as a midtable team.The xG numbers (Understat) for LVG's second season are interesting:
* 8th in the table (behind Chelsea).
* But outperformed both xG and xGA (+4 roughly for both).
* And outperformed xP by ten.
Meanwhile Arsenal win according to the model, followed by City and Spurs (identical to actual position).
Leicester 4th, outperforming the xP by 12 - xG slightly above actual figure, xGA significantly below (meaning they outperformed the model).
And...yeah, no idea where I'm going with this, just found it a bit interesting.
OPTAs xG stats are from previous shots from the same position from a database of 300,000 shotsWho decides whether a chance should be an "expected goal". Isn't that subjective.
You could have a chance and 2 different people judging it, one would say he should score, someone else could judge it differently.
Indeed - the numbers do look like they belong to a decent enough (quite efficient, clinical) mid-table team. Add the cost/presumed strength of our squad, though, and they begin to look grim.To be fair, we all complained that we weren't creating many chances. We were clinical but we were creating chances as much as a midtable team.
The biggest problem with xG is that it is very hard to consider the quality of players (sample is so small).
Good post buddy.Who decides whether a chance should be an "expected goal". Isn't that subjective.
You could have a chance and 2 different people judging it, one would say he should score, someone else could judge it differently.
You could ask the same about any stat, though. Before xG we had shots on target which in itself is not a bad indicator for the strength of a team. That has been used for ages is shown after every match on TV without anyone complaining. xG is nothing more than a much more sophisticated version of shots on goal/on target making the latter more or less obsolete. Personally, I use it for betting purposes (by necessity to stay ahead of the market) and because I have a general interest in quantifying things and find the usual clichés and empty phrases that are trotted out by pundits (armchair or professional) really tiresome. Things such as "Spurs have bottled it, they really should beat Burnley away" or "there is NO way City are not winning this weekend" or "we'll definitely make top 4". As if these things are black and white, 0 or 1.This is a to each their own debate if there ever was one, but psychologically I am intrigued to know:
What is the primary reason for following or arming yourself with this type of statistic ahead of a match you are about to watch? Is it to make smarter gambling/fantasy football decisions? Is it to pull back emotionally - stop yourself getting too high or too low early in a match? Or is it just that you (this is a general "you" and not you specifically) prefer discussing metrics about an event put on for entertainment?
Again, to each their own, but I am genuinely curious what viewers (and not the data scientists themselves, as this movement is no doubt a jobs creator in that field) are getting out of this. My personal opinion, flawed as it may be, is that the more metrics play a role in sport, the less entertaining they become.
You also have different xG models, so every model will produce a different number for the same shot. Yet people will use xG without referencing which model they're using, which renders the whole thing kind of pointless to me. It's like saying "damn, I really love that song I Love Rock ‘N’ Roll!". You would have no idea if it was referencing the Joan Jett version or the Britney Spears version.Who decides whether a chance should be an "expected goal". Isn't that subjective.
You could have a chance and 2 different people judging it, one would say he should score, someone else could judge it differently.
Thats a better way than evaluating it than say a guy behind a computer saying that should be a goal.OPTAs xG stats are from previous shots from the same position from a database of 300,000 shots
There are different models though, so depending on where you look you might find different values for the same game, don't know where the others get their data from.
Well, pretty much no shot ever is "an expected goal", ie has an xG value of 1.0. A shot from the halfway line would also be assigned a value, but an incredibly low one, like 0.01xG or whatever the minimum value is for that particular model.Thats a better way than evaluating it than say a guy behind a computer saying that should be a goal.
But say someone has a shot from inside the semi circle of the half way line in their own half, is that an expected goal because Beckham scored from there?
Whoever has said that's a good performance? That was most definitely not a good performance, understandably so but still doesn't change the fact it was not a good performance. It was clever on our part to play the way we did but I doubt anybody truly believes it was a good performance. If I'm not mistaken our xG was higher but that again speaks for looking at the flow of the game and not looking at xG for one game to decide who had the better performance.Good post buddy.
They are just a stat like possession that doesn't give a true indication of the game. (Bearing in mind Leicester won the PL)
Millions of people see the PSG game as a good performance, but just because we don't have a huge xG, people who think they know it all like Reddit bloggers will complain.
Ok maybe it's just them, the ones i have looked at tend to include penalties such as understatCaley for example shows open play xG and then in brackets has (1 pen). The main xG total is pretty much always not including pens. I went game by game to double check, they aren't included. Obviously so they can try and drive their agenda.
This post is insane. You're ignoring all context that's involved.Whoever has said that's a good performance? That was most definitely not a good performance, understandably so but still doesn't change the fact it was not a good performance. It was clever on our part to play the way we did but I doubt anybody truly believes it was a good performance. If I'm not mistaken our xG was higher but that again speaks for looking at the flow of the game and not looking at xG for one game to decide who had the better performance.
Garbage in, garbage out.538 considers also the quality of a player. Same chance to Messi and Welbeck gets weighted differently. In the end it is a model, there is no truth to give generalised opinions like 'xG are bad, good, etc'. xG is as good as the model and data from that company. Can be great and useful, can be bad and terrible.
The xG for it would be number_of_goals_scored_from_that_postion / number_of_shots_attempted_from_that_position, so likely an attempt from there it would have a very low xG.Thats a better way than evaluating it than say a guy behind a computer saying that should be a goal.
But say someone has a shot from inside the semi circle of the half way line in their own half, is that an expected goal because Beckham scored from there?
What if hardly anyone attempts it because its so unlikely so that the xG actually becomes higher than expected?The xG for it would be number_of_goals_scored_from_that_postion / number_of_shots_attempted_from_that_position, so likely an attempt from there it would have a very low xG.
The xG concept it is actually very simple. It takes like 2 mins to understand it.
Depends on how the model deals with outliers. Possibly it will give those types of shots a very low xG. I guess that they try these things and test the models to make them better.What if hardly anyone attempts it because its so unlikely so that the xG actually becomes higher than expected?
I am not ignoring any context. I said our not so great performance was understandable...what I left unsaid and that was obvious I thought was due to our injuries and the fact we were away. And I also said we were clever to go with the strategy we did. None of that changes the fact our performance wasn't great. Which I don't care cuz we were not looking for a performance, we were looking for a result.This post is insane. You're ignoring all context that's involved.
Our performance was great though? We couldn't play any better with the players we had.I am not ignoring any context. I said our not so great performance was understandable...what I left unsaid and that was obvious I thought was due to our injuries and the fact we were away. And I also said we were clever to go with the strategy we did. None of that changes the fact our performance wasn't great. Which I don't care cuz we were not looking for a performance, we were looking for a result.
Yeah, most likely, but if it's purely goals from position/shots from position it'd stand to reason that nobody would attempt that kind of shot unless the keeper is way out of position and the shot is on. But in reality, yeah it'd probably be assigned the minimum value (which is probably still higher than the possibility of any given shot going in from that position).The xG for it would be number_of_goals_scored_from_that_postion / number_of_shots_attempted_from_that_position, so likely an attempt from there it would have a very low xG.
The xG concept it is actually very simple. It takes like 2 mins to understand it.