xG doesn't "say what they should be having". It's not a prediction. It just quantifies the volume and quality of scoring chances. And it's no surprise that a finisher like Haaland gets a better than statistically average number of goals out of the chances he gets. Nor that teams mostly don't hit that exact number.
xG and finishing under ETH
- Thread starter Samid
- Start date
xG doesn't "say what they should be having". It's not a prediction. It just quantifies the volume and quality of scoring chances. And it's no surprise that a finisher like Haaland gets a better than statistically average number of goals out of the chances he gets. Nor that teams mostly don't hit that exact number.
It has limited value as an actual prediction tool.
Nevertheless, if you consider 12 games a reasonable sample size, I'd imagine it's statistically very improbable that Chelsea will end up in 10th if they continue to create chances on the same scale.
I don't think you can find many examples of this happening, i.e. a team that's consistently in the top four for xG and xP, but that nevertheless finishes 10th.
Such a team would have to be either extremely unlucky - or suffering from consistently fielding genuinely/hopelessly shit finishers. Chelsea's finishers aren't that shit (they've scored 8 more than us so far), as in they're not remarkably, uncommonly shit.
It is a tool to measure the quality of performances. It's not perfect, but it's good enough to give you an idea of how well or badly you are doing in what is a low-scoring team game. It's the very nature of football that can render performances either flattering to deceive or dominant regardless of the scoreline.
Haaland is a freak of a forward who's enjoying an incredible start to his football career. We'll have to wait and see how he will progress. But talking about being clinical, why don't you perform the same experiment with some of the most established attackers in the game, the ones who can offer a larger sample size? Kane, Lewandowski, Salah, Aguero, even the most prolific of them all: Ronaldo. You'll see that their accumulated xG is pretty much even with the actual number of goals they have scored. The statistic's aim isn't to show how clinical they are. In fact, it warns you that, at some point, you'll see regression to the mean. The aim is to show you their individual ability to "see" a higher number of finishes.
The mistake people who dismiss it often make is that they treat it like some kind of crystal ball that is supposed to help you see into the future. What it does is provide you with an image of the recent past and show you whatever trends you seem to be projecting on the pitch. For example, xG suggested that, last season, United earned about 10 more points than we should have. The offence was OK (but could do better), but the warning bell was in the defence, where we're overperforming our stats by a considerable margin. We could have used the summer window and the pre-season to get better. We didn't, and now we're confirming last season's trends. It wasn't unavoidable, and it certainly wasn't a prophecy. It's just the story of a club that refuses to change and reacts too late to almost everything. That's what stats are for, to offer some context
Last edited:
Yeah xG is useful as a basis for evidence and then it should be expected that great strikers like Haaland outperform it, that's their job! Or someone like Maupay in the past few years seasons is probably lagging way behind his xG.
The table on the last page isn't the best representation of the season so far or last year because of how the clubs are ranked on the current month, and also it seems incorrect! So far in Nov we have a total xG GD of 3.4-1.7 from games vs Fulham and Luton, so +1.7, yet it has us as -0.2, which seems excessive the other way even allowing for game difficulty.
We're shite but not that shite, and as others have said you could argue our overall xG is hindered by such a miserly showing against City.
The table on the last page isn't the best representation of the season so far or last year because of how the clubs are ranked on the current month, and also it seems incorrect! So far in Nov we have a total xG GD of 3.4-1.7 from games vs Fulham and Luton, so +1.7, yet it has us as -0.2, which seems excessive the other way even allowing for game difficulty.
We're shite but not that shite, and as others have said you could argue our overall xG is hindered by such a miserly showing against City.
It's a rolling average of the last 12 games excluding penalties. It's a fair reflection of how bad we've been. Our actual points flatter us. Our next fixtures are more difficult so we're very likely to regress unless something changes.
This. IMO we play bad since the day 1 cheat win vs Wolves and based on how bad we actually perform our points/table position is flattering us. This could look way worse in 3,4 weeks, once we are officially out of the CL (probably not even making EL) and fall back to 8/9.
Strictly speaking, it is a tool for judging the volume and quality of scoring chances created. No more, no less.
Not exactly... Someone can be a great striker because he is a clinical finisher, or because he gets himself a lot of chances. The latter is the case for most elite strikers, they don't outscore their xG, but they have a much higher xG than average strikers. There is more than one way to being a great striker.
It actually suffers as a tool by adding xG up. Would you rather 10x 0.10 xG attempts or 1x 1.0 xG? Both add up to 1xG but one is clearly more likely to result in a goal.
One of the remarkable things about Haaland is it's both. He'd got both the highest xG in the PL, and a goal output that significantly exceeds that again.
I'd choose the first of those, obviously - as would you, if you'd thought about it more carefully.
There is no such thing as a 1.0 xG. Maybe there is such a thing as a 0.9998 xG, but it's still probabilities against probabilities, not probabilities against certainties.
They add up to exactly the same probability, but with ten 10% opportunities there is a very real possibility you could get more than one goal. With option 2, you can't.
Yeah, you are correct and i should have clarified that. I meant the whole philosophy behind the idea, not just xG in isolation.
This is why context is always important and also why it is a tool that can help professionals and generate discussions for the rest of us.
But, to answer your question: I'd prefer neither. Lots of low probability chances would tell a story of bad decisions in the final third, but it would also indicate that, in terms of rhythm and ball progression, there are things we're doing well. Very few but high probability chances paints a very LVG-esque picture. Different trends, different things to improve on.
I think, it only tells you different stories. I agree, summing it over one game and then looking at one number can create wrong impressions. But if you look at it individually, it tells you something about your teams performances, one team creates many openings with none of them being very threatening while the other one creates next to no openings but when they do, they go for the kill. First could have Bruno in the team and only Maupay as the target, the other could be a very riskaverse team, dominating the ball and then pounce with killer countering.
There is no 1.0 xG chance, so let's take 5x 0.1xG vs 1x 0.5 xG and have a look at what this results in. The latter is easy, it is a 50% chance for a goal and 50% for no goal. But for 5x 0.1xG we get this: 59% no goal. On the other hand you can score more than one goal from those chances. So lots of low xG chances result in a wider spread of results than few high xG chances. Which does prove your point that simply adding xG up can be misleading.
I'd choose the first of those, obviously - as would you, if you'd thought about it more carefully.
They add up to exactly the same probability, but with ten 10% opportunities there is a very real possibility you could get more than one goal. With option 2, you can't.
I remember getting in a debate about this after the Liverpool/Madrid CL final.
They held that Liverpool deserved to win but because their xG was higher. The crux of my argument was that their xG per shot was absolutely pitiful compared to Real only taking their shots when it made sense, with the goal coming from a near dead cert.
Basically, if you want to break a window, are you better off throwing a hundred pebbles at it one at a time, or are you better off using a brick?
While I take the point on the possibility of maybe getting two or more goals from a series of pot shots, as @stefan92 said above, you also increase the probability of failing to score at all with such an approach.
I know there is no 1.0xG chance. I was making a more general point that what you want is high quality chances.
There is likely a model which suggests the best way to maximise xG. For example you could make an argument that trying to deliberately get penalties makes the most sense. I imagine that City’s low crosses and pull backs is data driven as a repeatable way of getting high xG chances.
As a manager I’d largely discourage long range shots because they are such a poor chance.
"Deserved" is a skewed concept to apply to xG, especially in a one off game. Say a team wins 3-0, scoring three world class long-range screamers in the opening ten minutes that kill the game, toy with the opposition in possession for the remaining 80min, before conceding a few chances right at the end when the game is long dead, including a penalty their goalkeeper saves. We'd generally see that as a "deserved" win, even though the xG in that game could be 0.3 versus 1.2 in favour of the team that lost. Winning via the three long-range screamers isn't less deserving, it's just less repeatable. And killing the game after going 3-0 up isn't a fault, it's just a response to game state.
As for multiple low xG chances versus few high xG chances, the latter is preferable. If a team scores 10 goals over 10 games they'd obviously prefer that to be distributed as one goal per game rather than multiple goals in some games and none in others. So a make-up of chances that increases the likelihood of scoring at least one goal per game would be preferred over one that offers greater chance of both blanking and scoring multiple goals.
As for multiple low xG chances versus few high xG chances, the latter is preferable. If a team scores 10 goals over 10 games they'd obviously prefer that to be distributed as one goal per game rather than multiple goals in some games and none in others. So a make-up of chances that increases the likelihood of scoring at least one goal per game would be preferred over one that offers greater chance of both blanking and scoring multiple goals.
Last edited:
If we're comparing 3 big chances or 12 smaller ones, probably yes. If we're comparing one big one and ten smaller ones, definitely no.
In any case, it's a question of how the probabilities stack up. One 0.9 xG chance does not in actual fact yield a greater chance of scoring one goal than nine 0.10 ones. That's not how probability works.
Two teams create 10 xG across 10 games. Both are predicted to average the same number of goals over the same number of games.
However, Team A get their 10 xG via 2 shots per game, while Team B gets theirs via 10 shots per game. Team B inherently have more chance of scoring 2+ goals in any given game in that run (as Team A can't score more than their 2 shots in a single game). Conversely, Team A have more chance of scoring at least one goal in any given game in that run.
Because the latter distribution translates more efficiently into results over an extended period, it is preferred. Even though on average both teams are predicted to score one goal per game.
But the latter doesn't translate more efficiently into results. Why would it?
Because on average the first goal a team scores increases their chances of winning more than subsequent goals do.
Reverting to the above example, imagine you were given the choice of:
A) 1 goal every game for 10 games
B) 10 goals randomly assigned across 10 games.
With Option A you immediately have a chance of winning every game. You only draw or lose games in which your opponent themselves scores 1 or more goals. And the goal you've scored has a guaranteed impact on the result unless your opponent scores 2 or more goals.
With Option B you lose the chance of winning in each game your team blanks. You can also fail to win in games in which your opponent blanks. And while you have gained the chance of winning in some games where your opponent scores 1 or more goals, you have no guarantee that those additional goals will impact the result (i.e. they could be the second goal in a 2-0 win).
With football being a fundamentally low scoring game (teams average approx 1.32 goals per game) and with winning providing disproportionate points returns (i.e. one win is worth three draws), it means choosing Option B would be a mistake as you're not maximising the likely impact of the 10 goals. Over a run of games it is likely to translate into fewer points.
No, you misunderstand the objection. It is this: 2 shots each with 0.5 xG does not translate more efficiently than 10 shots each with 0.1 xG. They give exactly the same probability of scoring a goal in the course of any single game (to the extent that xG is a true representation of probability). Hence there is no reason whatsoever why option B would be any different than option A in how goals distribute between games.
Last edited:
Sorry, before I lambast others for their deficient grasp of probability and recalling I'm totally a layman in that area myself: How does 5x 0.1xG result in a 59% chance of no goal?
EDIT: Never mind, I looked it up. And I am clearly wrong above. So, how would that work out if it's two 0.5 xG chances versus ten 0.1 xG chances?
Last edited:
Just no goal, 25% vs 35% - I won't go to the lengths for every result above that as it becomes quite quickly quite complicated.
Anyway the probability distribution always results is more widespread for more bad chances than for few good chances. If you want to score many goals you need many chances, but if you want to be sure you score at all you should have good chances, but this will also make freak scorelines more rare
Thanks for the help!
Was just thinking about this a bit further... so if you want to be sure to score at least one goal with (more than) 75% probability you could have 2x 0.5 xG or 14x 0.1 xG (resulting in 77% probability to score and in a higher total xG).
So the question arises does it make sense to try to get closer to the goal and get a free shot or try to score the low probability one? And this shows us that you have to be able to move more than 1 in 7 attacks "from the edge of the crowded box" to "alone in front of the keeper" (to assign situations to these numbers). If your passing and movement is good enough to beat the defenders that often you should go for it, if you can't beat them to be alone in front of the keeper that often, then go for the low probability shot.
Of course this is very theoretical and massively simplified, but it should give a statistical idea about whether a "shoot at first sight" approach makes sense for a team or not.
We must be getting our wires crossed, because what I'm taking from your posts makes zero sense to me. My point is this:
If a team has 10 shots on target, they can potentially score up to 10 goals in that game. If a team has 2 shots on target, they can potentially score up to 2 goals. If they are both predicted to on average return one goal from the game, it follows that the first team has a greater chance of scoring 3 or more goals, but also a correspondingly greater chance of scoring 0, leading to the predicted average of 1 goal. The other team has no chance of scoring more than 2 goals, but also correspondingly less chance of scoring 0, leading to the same predicted average of 1 goal.
Because of the way football works, the latter probability distribution is more favourable over an extended run of games. Because there's more value in an increased chance of scoring at least one goal in a game than an increased chance of scoring a second.
Rewriting this as feel it is interesting but a bit unclear. Is this what you are talking about?
Team 1. Shots on target 10. xG 1.
Team 2. Shots on target 2. xG 1.
You can confidently say Team 2 had at least one very high quality goalscoring chance (lowest quality distribution possible is two 0.5 xG chances which is still very high) and it is unlikely Team 1 had a chance of a similar quality, but not impossible. It is more probable Team 1 had a lot of poor chances. Therefore Team 2, if this form held up for the season, have a more favourable xG if you go behind the numbers.
I assume the logic is, scoring exactly 1 goal, 10 games in a row, is likely to result in more points accumulated than scoring a more random/inconsistent distribution of those same 10 goals - let's say 0 goals in 4 games, 1 goal in 3 games, 2 goals in 2 games, 3 goals in 1 game?
I agree but I think there's definitely some people who treat it as a prediction. On a few podcasts I've heard some say that a team in bad form is still doing well in xG so they'll "return to the mean" or something like that as if the finishing will naturally improve. I definitely think there's merit to it but sometimes it's taken as more important than the score in judging performance.
Grim reading
We're 12th best in the EPL for expected goals this season.
14th best for expected goals conceded.
14th best for expected goals conceded.
Indeed.
If we imagine that pattern held over an extended run, we would say both teams are expected to score the same number of goals as they are generating the same xG.
But you'd also expect those the goals to be distributed across those games differently. You would likely see Team 1 score 3+ goals in some of their games, which Team 2 can't do from their 2 shots. But, because we know both teams ultimately average the same number of goals, that in turn means Team 2 would have more games in which they score at least 1 or 2 goals.
And because football is a low scoring game, the latter distribution tends to translate into points more efficiently.
It depends on wether you have good finishers in your team or not.
A 1,0xG chance means 1 goal, and exactly that.
10 x 0,1xG chances could be anywhere between 0-10 goals and the better finishers you have the more likely it means more than one goal.
It's a very good tool to measure your chance creation, your players finishing abilities (or hot/cold streaks), your defensive qualities when it comes to conceding chances and your goalkeeper's ability to keep them out of the net. It's by far the best tool we have to give indication regarding this. It is of course not absolute or 100% accurate in the short or long run, but it gets better and more precise the bigger the sample size gets.
This season we've been missing more chances than we should have, with our two main guys up front, Rashford and Hojlund being the main source for that with 1 goal scored from 7,32xG total between the two. This is unsustainably bad (no one is this bad at finishing in the long run). None of them are especially good finishers, but their numbers from earlier season suggest they are around or just below average, which suggest that this goal drought will not last forever for any of them. Having good finishers in good positions is really important though. Right now our two main guys is not. For comparison has notoriously good finishers like Son scored 8 goals from 4,91xG so far this season and Kane scored 14 non penalty goals from 8,95npxG.
The rest of the our team has been slightly overperforming their combined xG with 12 goals from 11,15xG.
I hope Marcus and Rasmus are putting in some extra hours on the training ground finishing, because they need it.
We can also read from the numbers that Onana has gained some shotstopping form the last 7 games where we have won at least 4 flukey one goal wins in bad performances (Not counting the Luton game as that was very one sided): We have conceded 6 goals from 10,73xGA (Expected goals against) in the last 7 in the PL compared to 10 goals from 9,04xGA in the first 5 games.
Of course it doesn't make sense to you. That's because what I wrote reflected an idiotic brain fart concerning a basic aspect of how probability is calculated. So I'm bowing out of this discussion, hanging my head in shame.
Can't mods remove any thread with a reference to XG?!
It's the most ridiculous creation, along with VAR in recent years.
It's the most ridiculous creation, along with VAR in recent years.
Or just don't comment/visit a thread that is clearly going to be about xG/stats in general?
Yeah Stalin, lets make sure people do what you think is correct to make the world a better place
Apparently our xG at half time was better than any other team in the league has managed in any fixture this season. I guess that particular stat was far too positive to get a mention by the thread starter.