I see this XG thing bandied about quite a lot, I assume it means chances at goal but don't really know, can someone please explain and is it something that I should look to include in match stats?
It’s quite a simple idea about how many goals should have been scored given the quality of the chances created. It can be applied to teams or individuals. Luddites and dinosaurs (pretend to) find it confusing.
1. xG stands for "expected goals." It’s a way of measuring how likely a player is to score a goal when they take a shot. For example, if a player has a great chance right in front of the goal, the xG might be 0.8 — meaning there's an 80% chance that shot should result in a goal. It's a tool to see how good the chances were, not just how many goals were scored.
2. To calculate xG, statisticians look at thousands of past shots and consider things like how far the shot was from the goal, the angle, whether it was a header or a kick, and how many defenders were in the way. All these factors help assign a probability to each shot. This lets us judge whether a team was creating high-quality chances or just taking low-probability long shots.
3. xG models use logistic regression or other machine learning algorithms trained on historical shot data to estimate goal probability. Features can include shot distance, shot angle, body part used, defensive pressure, assist type, and preceding events. The models output a continuous probability between 0 and 1, summing up across a match or season to assess team or player efficiency.
4. Model fidelity depends on both granularity and input dimensionality, with higher-resolution tracking data allowing for dynamic xG computation — incorporating not just static shot context but spatiotemporal player positioning and velocities. This leads to more nuanced models such as Post-Shot xG (PSxG) and non-shot expected threat (xT), where pre-shot buildup actions are statistically mapped to future goal likelihoods.
5. From a Bayesian standpoint, xG can be interpreted as the posterior mean of a Bernoulli distribution over shot outcomes, conditioned on covariates and prior distributions derived from league-wide historical data. When modeled hierarchically, individual player or team effects can be captured via partial pooling, improving prediction under data sparsity. However, epistemic uncertainty in xG estimation — due to model specification, input variability, and sampling bias — remains a nontrivial consideration in inferential robustness.
5. From a Bayesian standpoint, xG can be interpreted as the posterior mean of a Bernoulli distribution over shot outcomes, conditioned on covariates and prior distributions derived from league-wide historical data. When modeled hierarchically, individual player or team effects can be captured via partial pooling, improving prediction under data sparsity. However, epistemic uncertainty in xG estimation — due to model specification, input variability, and sampling bias — remains a nontrivial consideration in inferential robustness.
5. From a Bayesian standpoint, xG can be interpreted as the posterior mean of a Bernoulli distribution over shot outcomes, conditioned on covariates and prior distributions derived from league-wide historical data. When modeled hierarchically, individual player or team effects can be captured via partial pooling, improving prediction under data sparsity. However, epistemic uncertainty in xG estimation — due to model specification, input variability, and sampling bias — remains a nontrivial consideration in inferential robustness.
It’s quite a simple idea about how many goals should have been scored given the quality of the chances created. It can be applied to teams or individuals. Luddites and dinosaurs (pretend to) find it confusing.
It’s not confusing. It’s just boring and for the most part wrong .
It’s quite a simple idea about how many goals should have been scored given the quality of the chances created. It can be applied to teams or individuals. Luddites and dinosaurs (pretend to) find it confusing.
It’s not confusing. It’s just boring and for the most part wrong .
The fact that it doesn’t take into account the quality of the goalkeeper is a major flaw in my eyes. Surely it’s much more “expected” that I’ll score the same chance against Yohann Thuram than I would against peak Buffon?
FWIW I think it would be a good inclusion in the statbank @lancashire lad although different stats providers have different xG formulas so I would recommend choosing one and sticking with them for the entire season instead of mixing and matching.
It's the next step in the evolution of data usage to determine how good a player or team is.
As always, you can't trust it in a vacuum, but simply put it is the expected number of goals from any given chance or chances, over the course of a period of play (be that a half, match, run of games, season, or career).
If you have a low xG and a low number of goals, then your team isn't creating much goal threat and they're not capitalising on what they do have. Or, your striker (attacking player) isn't getting in positions to score, and isn't converting them either when he does. Of course, that could be because of numerous factors: the striker could be crap, just as the team could be crap.
If you have a high xG and a low number of goals, then the striker probably IS crap. (Or, he's been monumentally unlucky and a series of worldies have been pulled off by oppo keepers. Again, numerous potential factors here.)
If you have a low xG and a high number of goals, the striker is doing pretty well. (Or, got lucky/keepers have been rubbish.)
High xG and high goals - good team/striker getting into positions and converting them. Or you're playing a load of teams/players that are significantly worse than you, so don't read too much into it.
---------------------
Basically, it's another tool to help determine whether a player or team is any good or not. It shouldn't be used on its own. Saying "he's a good finisher" is one thing; saying "he's got unlucky the last three or four games" is another. xG can help puzzle it out and see if indeed the striker is clinical, or shit, or somewhere in between. But it isn't gospel.
I'd do an illustrative example of e.g. scoring two tap-ins, one worldie, missing a sitter, and a decent finish with a real-life data set but it's late. Maybe I will do that with the thread of goals someone put up for Olaofe to give a very rudimentary idea of how to combine xG / "he should have scored" / "that was a good finish" / "the keeper (or defender) messed that up" and all the rest of it, if I get any time, but I haven't got any right now. But it's one factor to help inform, not the ONLY factor, and that's something people seem to miss. There's nuance.
It’s quite a simple idea about how many goals should have been scored given the quality of the chances created. It can be applied to teams or individuals. Luddites and dinosaurs (pretend to) find it confusing.
It’s not confusing. It’s just boring and for the most part wrong .
The fact that it doesn’t take into account the quality of the goalkeeper is a major flaw in my eyes. Surely it’s much more “expected” that I’ll score the same chance against Yohann Thuram than I would against peak Buffon?
More the case that it doesn't take into account the quality of the player with the chance.
It’s quite a simple idea about how many goals should have been scored given the quality of the chances created. It can be applied to teams or individuals. Luddites and dinosaurs (pretend to) find it confusing.
It’s not confusing. It’s just boring and for the most part wrong .
The fact that it doesn’t take into account the quality of the goalkeeper is a major flaw in my eyes. Surely it’s much more “expected” that I’ll score the same chance against Yohann Thuram than I would against peak Buffon?
More the case that it doesn't take into account the quality of the player with the chance.
Same XG whether it was Sir Clive or Simon Church.
Imo that is where it falls down.
Quality of the, or how well they take their chances, is one of the things it can help measure.
I wonder how many of the 'its bollocks' crowd are just as opposed to shots on/off target data too.
It’s quite a simple idea about how many goals should have been scored given the quality of the chances created. It can be applied to teams or individuals. Luddites and dinosaurs (pretend to) find it confusing.
It’s not confusing. It’s just boring and for the most part wrong .
The fact that it doesn’t take into account the quality of the goalkeeper is a major flaw in my eyes. Surely it’s much more “expected” that I’ll score the same chance against Yohann Thuram than I would against peak Buffon?
More the case that it doesn't take into account the quality of the player with the chance.
Same XG whether it was Sir Clive or Simon Church.
Imo that is where it falls down.
Quality of the, or how well they take their chances, is one of the things it can help measure.
I wonder how many of the 'its bollocks' crowd are just as opposed to shots on/off target data too.
It’s quite a simple idea about how many goals should have been scored given the quality of the chances created. It can be applied to teams or individuals. Luddites and dinosaurs (pretend to) find it confusing.
It’s not confusing. It’s just boring and for the most part wrong .
The fact that it doesn’t take into account the quality of the goalkeeper is a major flaw in my eyes. Surely it’s much more “expected” that I’ll score the same chance against Yohann Thuram than I would against peak Buffon?
More the case that it doesn't take into account the quality of the player with the chance.
Same XG whether it was Sir Clive or Simon Church.
Imo that is where it falls down.
If Clive Mendonca and Simon Church played in the same era and in one season both scored 10 goals but Clive did it from 5 xG while Church did it from 15 xG, would it not be helpful to have something quantitative to prove that Mendonca was the better finisher?
Similarly, imagine two teams who finish 12th and 13th ranked for goals scored across a full season of games.
In this hypothetical, it turns out that the team that finished 13th for goals scored actually produced 6th most xG over the season. The manager and head of recruitment can use that data to understand they are creating a lot of chances but they don’t have a striker to finish them off.
On the other hand, the team with 12th most goal actually had the 6th worst xG. This team realise they need to go into next season with a couple of extra players who can create more chances for their strikers who are doing a great job with the limited service they do get.
Every penalty is ~0.77 xG because every 77 out of 100 penalties are scored.
This shouldn’t be adjusted for the quality of the keeper or the taker because then it loses it’s usefulness as a bar to help understand who actually over performs and under performs in those situations.
A bad shootout keeper will let in more than 77% of penalties faced and a good one will allow less.
In my opinion it can be useful information in a game where statistics play an increasing part. Does it tell you more than you see, no, but it provides a record of what you have seen and can help, help only, in identifying issues over a period. It shouldn't be over emphasised in importance.
Data analysis in football has been used for many years. There was of course the late, much derided Charles Hughes, who is credited with the introduction of the long ball game. Many years ago, he did painstaking analysis of how goals were scored and concluded that most came from three passes or fewer. What his data didn't factor in was the ability of those players so it was flawed.
Having said that, it is still IMO a template for how a team with less talent can beat a team with more. When I managed a youth team, I employed the philosophy with success. He should be acclaimed as a pioneer and not castigated. Yes, the data took the british game in the wrong direction but it was the beginnings of something that is now computer driven and complex. The top teams now look at contribution that lead to goals from a player they are considering buying. Not direct contributions but even short passes in their own half! All computer driven.
It’s quite a simple idea about how many goals should have been scored given the quality of the chances created. It can be applied to teams or individuals. Luddites and dinosaurs (pretend to) find it confusing.
It’s not confusing. It’s just boring and for the most part wrong .
The fact that it doesn’t take into account the quality of the goalkeeper is a major flaw in my eyes. Surely it’s much more “expected” that I’ll score the same chance against Yohann Thuram than I would against peak Buffon?
More the case that it doesn't take into account the quality of the player with the chance.
Same XG whether it was Sir Clive or Simon Church.
Imo that is where it falls down.
I don’t massively want to support xG but isn’t it the opposite of what you’ve just said? If Clive Mendonca and Simon Church had the same chance then the “expected goal” stat would be the same but the “goals” stat (which nobody questions) would be different. Thus you would be able to statistically prove that Clive Mendonca, with the same or similar opportunities, is more likely to score.
For me it’s two main flaws are:
1) the lack of consideration for the quality of the goalkeeper, as explained above, and, 2) fundamentally, it’s not objective which feels pretty important to me if we’re going to describe something as a “statistic”. As per @chizz XG is awarded based on the opinion of a statistician using historical precedent. As per @Callumcafc different statisticians use different formulas to calculate XG.
For what it’s worth I don’t think XG is “bollocks” or nonsense. There’s some merit to it, and it’s an interesting concept, I just think it’s flawed and proponents of it should be honest about that. I don’t think of it as a statistic in the way that goals, assists etc are clearly and objectively calculated. But it is a useful guide around chance creation and scoring efficiency.
It’s quite a simple idea about how many goals should have been scored given the quality of the chances created. It can be applied to teams or individuals. Luddites and dinosaurs (pretend to) find it confusing.
It’s not confusing. It’s just boring and for the most part wrong .
The fact that it doesn’t take into account the quality of the goalkeeper is a major flaw in my eyes. Surely it’s much more “expected” that I’ll score the same chance against Yohann Thuram than I would against peak Buffon?
More the case that it doesn't take into account the quality of the player with the chance.
Same XG whether it was Sir Clive or Simon Church.
Imo that is where it falls down.
If Clive Mendonca and Simon Church played in the same era and in one season both scored 10 goals but Clive did it from 5 xG while Church did it from 15 xG, would it not be helpful to have something quantitative to prove that Mendonca was the better finisher?
You don't need xG to work out who is the better player. Just use your eyes and watch them play.
XG is a modern phenomenon aimed at reducing the soul, spirit and passion of one of the most beautiful art forms in human history into a sterile, statistical borefest. It is often used to justify that whilst losing 3 nil for third week in a row your donkey striker got into enough good positions in the 75th minute to demonstrate that you really won on paper.
Whilst it's origin is unknown the leading school of thought dictates that it was likely invented by Crystal Palace fans to pass the time between making banners and stave off the frustration of life long virginity.
Proponents of xg also tend to vehemently argue the virtues of VAR, traffic wardens and are credited with creating needless terminology like 'inverted false tricefta vanguard' to describe a right footed left back.
It doesn’t interest me - but can see how it could be useful.
To be really useful the ‘Expected’ number has to be consistent in order to be measured against.
It never is consistent. Thats why it’s massively flawed . I’ll use my own knowledge for betting purposes rather than AI or a computer telling me that each team has a 0.41 XG and the score ends up being 2-2 .
Comments
It's bollox
1. xG stands for "expected goals." It’s a way of measuring how likely a player is to score a goal when they take a shot. For example, if a player has a great chance right in front of the goal, the xG might be 0.8 — meaning there's an 80% chance that shot should result in a goal. It's a tool to see how good the chances were, not just how many goals were scored.
/thread
As always, you can't trust it in a vacuum, but simply put it is the expected number of goals from any given chance or chances, over the course of a period of play (be that a half, match, run of games, season, or career).
If you have a low xG and a low number of goals, then your team isn't creating much goal threat and they're not capitalising on what they do have. Or, your striker (attacking player) isn't getting in positions to score, and isn't converting them either when he does. Of course, that could be because of numerous factors: the striker could be crap, just as the team could be crap.
If you have a high xG and a low number of goals, then the striker probably IS crap. (Or, he's been monumentally unlucky and a series of worldies have been pulled off by oppo keepers. Again, numerous potential factors here.)
If you have a low xG and a high number of goals, the striker is doing pretty well. (Or, got lucky/keepers have been rubbish.)
High xG and high goals - good team/striker getting into positions and converting them. Or you're playing a load of teams/players that are significantly worse than you, so don't read too much into it.
---------------------
Basically, it's another tool to help determine whether a player or team is any good or not. It shouldn't be used on its own. Saying "he's a good finisher" is one thing; saying "he's got unlucky the last three or four games" is another. xG can help puzzle it out and see if indeed the striker is clinical, or shit, or somewhere in between. But it isn't gospel.
I'd do an illustrative example of e.g. scoring two tap-ins, one worldie, missing a sitter, and a decent finish with a real-life data set but it's late. Maybe I will do that with the thread of goals someone put up for Olaofe to give a very rudimentary idea of how to combine xG / "he should have scored" / "that was a good finish" / "the keeper (or defender) messed that up" and all the rest of it, if I get any time, but I haven't got any right now. But it's one factor to help inform, not the ONLY factor, and that's something people seem to miss. There's nuance.
Same XG whether it was Sir Clive or Simon Church.
Imo that is where it falls down.
I wonder how many of the 'its bollocks' crowd are just as opposed to shots on/off target data too.
In this hypothetical, it turns out that the team that finished 13th for goals scored actually produced 6th most xG over the season. The manager and head of recruitment can use that data to understand they are creating a lot of chances but they don’t have a striker to finish them off.
On the other hand, the team with 12th most goal actually had the 6th worst xG. This team realise they need to go into next season with a couple of extra players who can create more chances for their strikers who are doing a great job with the limited service they do get.
A bad shootout keeper will let in more than 77% of penalties faced and a good one will allow less.
Data analysis in football has been used for many years. There was of course the late, much derided Charles Hughes, who is credited with the introduction of the long ball game. Many years ago, he did painstaking analysis of how goals were scored and concluded that most came from three passes or fewer. What his data didn't factor in was the ability of those players so it was flawed.
Having said that, it is still IMO a template for how a team with less talent can beat a team with more. When I managed a youth team, I employed the philosophy with success. He should be acclaimed as a pioneer and not castigated. Yes, the data took the british game in the wrong direction but it was the beginnings of something that is now computer driven and complex. The top teams now look at contribution that lead to goals from a player they are considering buying. Not direct contributions but even short passes in their own half! All computer driven.
For me it’s two main flaws are:
1) the lack of consideration for the quality of the goalkeeper, as explained above, and,
2) fundamentally, it’s not objective which feels pretty important to me if we’re going to describe something as a “statistic”. As per @chizz XG is awarded based on the opinion of a statistician using historical precedent. As per @Callumcafc different statisticians use different formulas to calculate XG.
Just use your eyes and watch them play.
XG is a modern phenomenon aimed at reducing the soul, spirit and passion of one of the most beautiful art forms in human history into a sterile, statistical borefest. It is often used to justify that whilst losing 3 nil for third week in a row your donkey striker got into enough good positions in the 75th minute to demonstrate that you really won on paper.
Whilst it's origin is unknown the leading school of thought dictates that it was likely invented by Crystal Palace fans to pass the time between making banners and stave off the frustration of life long virginity.
Proponents of xg also tend to vehemently argue the virtues of VAR, traffic wardens and are credited with creating needless terminology like 'inverted false tricefta vanguard' to describe a right footed left back.