XG is a terrible stat

It's pointless over the course of a game, less pointless over the course of a season.
Over a large chunk of games it has it's uses, but in isolation, which is how the armchair fan uses it these days, it's absolutely pointless.

No manager uses this stat, other than when they feel wronged by a result, it then becomes an excuse to prove that the team deserved better than the result, see Frank at Brentford, he used it once after a loss to describe how he thought Brentford should have won the game, yet several professional players have stated that no manager uses the stat in any format as it's generally pretty useless.
 
acnumber9 said:
It’s factually below the average but if it makes you happier we’ll call it average. Do you believe that Salah is an average finisher? If not, I don’t see how the xG can be considered an average.
Click to expand...
He probably is an average finisher, it's been discussed in other threads but top goalscorers tend to shoot more and create more good chances for themselves but only few are actually better at finishing. If you take Salah this season he scores 22.5% of the shots he takes which isn't bad but isn't remarkable either.
 
acnumber9 said:
I looked at Salah’s stats and he’s underperformed in the last three seasons. I don’t think you’d find anybody who would say he’s a below average finisher.
Click to expand...
In the last 8 seasons, since he joined Liverpool, Salah has underperformed 4 times and overperformed 4 times, and only by a small percentage each year.

https://understat.com/player/1250

Salah's conversion rate this season (goals to shots) is 10%.
 
Last edited:
Champ said:
It's pointless over the course of a game, less pointless over the course of a season.
Over a large chunk of games it has it's uses, but in isolation, which is how the armchair fan uses it these days, it's absolutely pointless.

No manager uses this stat, other than when they feel wronged by a result, it then becomes an excuse to prove that the team deserved better than the result, see Frank at Brentford, he used it once after a loss to describe how he thought Brentford should have won the game, yet several professional players have stated that no manager uses the stat in any format as it's generally pretty useless.
Click to expand...
I mean managers aren't going to talk about xG to players but they'll talk about and train creating better quality chances and train ways to restrict high quality chances, which is literally the same thing. It's just a way to quantify the quality of chances over time that I'm sure they look at as a way to measure how things are going, and it's a known fact that club analysts use (more sophisticated) versions of it.
 
I enjoy watching how xG will continue to develop over time but it has already turned into a feedback loop. It started as a descriptive tool, but when clubs started basing their tactics on it, it shifted into a prescriptive role, changing the very data it was designed to analyze and leading to inflated metrics.

More and more shots come from high xG zones, the dataset becomes biased, potentially making xG appear more predictive than it would be in a less targeted environment.

Thanks to xG, many teams now adopt similar strategies, reducing creativity and variation, instead optimizing for xG-friendly shots. It's getting boring.

Based on the data, clubs oversimplify decision-making. Players are told to shoot from optimal zones, but context (defenders, passing options, game flow) can’t always be quantified by xG alone. Arsenal's set pieces are a good example of killing game flow when playing to the tune of the stats.

xG is turning into a self-reinforcing system. It might improve certain stats but at the sime time it's distorting the game and also the data's objectivity. I'm not sure it won't turn into another VAR situation. We enjoy the novelty and have high hopes for the new angle it adds to the game but in reality it's just the way Six Sigma stifles creativity and innovation in businesses. Not a fan.
 
Champ said:
No manager uses this stat, other than when they feel wronged by a result, it then becomes an excuse to prove that the team deserved better than the result, see Frank at Brentford, he used it once after a loss to describe how he thought Brentford should have won the game, yet several professional players have stated that no manager uses the stat in any format as it's generally pretty useless.
Click to expand...
Interesting, never heard that before. Can you point me to some?
 
horsechoker said:
The problem is no matter how much data you feed it, it is still subjective and based on probability. It shouldn't be used in isolated incidents but across a range of games, that's where it is most effective.
Click to expand...

You read that take pretty oftwn but from a methodological point of view, it is not correct. XG calculates the probability of the shot by looking at past data. You need huge sample sizes in the training and not in the application. The real art is to decide which data you feed the algorithm with.

For instance, you could train a super complex model that not only considers the position of the shot but also nearby defenders, position of the goal keeper, first touch finish or not, distance ran by the shot taker during the match and/or ladt 30 seconds, and much more but the thing is, the more specific you get, the more data you need. So you end up watering down the sample size by including more and more competitions of possibly lower player quality or going back too many seasons (say 10 or so) when football was a different sport.

Most of the time, simplicity is key in such models.
 
Gums said:
I enjoy watching how xG will continue to develop over time but it has already turned into a feedback loop. It started as a descriptive tool, but when clubs started basing their tactics on it, it shifted into a prescriptive role, changing the very data it was designed to analyze and leading to inflated metrics.

More and more shots come from high xG zones, the dataset becomes biased, potentially making xG appear more predictive than it would be in a less targeted environment.

Thanks to xG, many teams now adopt similar strategies, reducing creativity and variation, instead optimizing for xG-friendly shots. It's getting boring.

Based on the data, clubs oversimplify decision-making. Players are told to shoot from optimal zones, but context (defenders, passing options, game flow) can’t always be quantified by xG alone. Arsenal's set pieces are a good example of killing game flow when playing to the tune of the stats.

xG is turning into a self-reinforcing system. It might improve certain stats but at the sime time it's distorting the game and also the data's objectivity. I'm not sure it won't turn into another VAR situation. We enjoy the novelty and have high hopes for the new angle it adds to the game but in reality it's just the way Six Sigma stifles creativity and innovation in businesses. Not a fan.
Click to expand...

Very good and interesting take
 
He did nick it off the defender and maybe xG is considering that
 
acnumber9 said:
It’s factually below the average but if it makes you happier we’ll call it average. Do you believe that Salah is an average finisher? If not, I don’t see how the xG can be considered an average.
Click to expand...

Looking at Fbref, across the entirety of Salah's Liverpool career (a relatively big sample size) he is an above average finisher, scoring approximately 10 goals above his xG.

But one of the fundamental realities of football that analytics has made evident is that finishing matters a lot less to goalscoring than the ability to get chances. In Salah's case, finishing 0.04 per 90 above his xG matters far, far less than the fact he's expected to score a goal 136 minutes or so even with average finishing.

However, people being people, they tend to think of any player who scores a lot of goals as being an excellent finisher, even when it isn't driven by finishing ability.
 
Gums said:
I enjoy watching how xG will continue to develop over time but it has already turned into a feedback loop. It started as a descriptive tool, but when clubs started basing their tactics on it, it shifted into a prescriptive role, changing the very data it was designed to analyze and leading to inflated metrics.

More and more shots come from high xG zones, the dataset becomes biased, potentially making xG appear more predictive than it would be in a less targeted environment.

Thanks to xG, many teams now adopt similar strategies, reducing creativity and variation, instead optimizing for xG-friendly shots. It's getting boring.

Based on the data, clubs oversimplify decision-making. Players are told to shoot from optimal zones, but context (defenders, passing options, game flow) can’t always be quantified by xG alone. Arsenal's set pieces are a good example of killing game flow when playing to the tune of the stats.

xG is turning into a self-reinforcing system. It might improve certain stats but at the sime time it's distorting the game and also the data's objectivity. I'm not sure it won't turn into another VAR situation. We enjoy the novelty and have high hopes for the new angle it adds to the game but in reality it's just the way Six Sigma stifles creativity and innovation in businesses. Not a fan.
Click to expand...

The few real glimpses I've seen into the data analytics of top clubs suggest that they are like five generations ahead of statistics like XG. They are using much more complicated modeling, taking into account many more game conditions and variables, and often drawing upon data from camera feeds that facilitate the measurement of many other kinds of things beyond what goes into an XG model.

How that influences tactics I have no idea but the data science people at these clubs aren't sitting there with powerpoint presentations showing the average XG in different shot zones.
 
Powderfinger said:
The few real glimpses I've seen into the data analytics of top clubs suggest that they are like five generations ahead of statistics like XG. They are using much more complicated modeling, taking into account many more game conditions and variables, and often drawing upon data from camera feeds that facilitate the measurement of many other kinds of things beyond what goes into an XG model.

How that influences tactics I have no idea but the data science people at these clubs aren't sitting there with powerpoint presentations showing the average XG in different shot zones.
Click to expand...

Yeah, as far as I know they take into account the different surfaces, all the speeds, the moment of the game, players' physical condition because players aren't expected to have the same lucidity at the end of the game or at the end of a long distance sprint. There are models for all these things.
 
Gums said:
I enjoy watching how xG will continue to develop over time but it has already turned into a feedback loop. It started as a descriptive tool, but when clubs started basing their tactics on it, it shifted into a prescriptive role, changing the very data it was designed to analyze and leading to inflated metrics.

More and more shots come from high xG zones, the dataset becomes biased, potentially making xG appear more predictive than it would be in a less targeted environment.

Thanks to xG, many teams now adopt similar strategies, reducing creativity and variation, instead optimizing for xG-friendly shots. It's getting boring.

Based on the data, clubs oversimplify decision-making. Players are told to shoot from optimal zones, but context (defenders, passing options, game flow) can’t always be quantified by xG alone. Arsenal's set pieces are a good example of killing game flow when playing to the tune of the stats.

xG is turning into a self-reinforcing system. It might improve certain stats but at the sime time it's distorting the game and also the data's objectivity. I'm not sure it won't turn into another VAR situation. We enjoy the novelty and have high hopes for the new angle it adds to the game but in reality it's just the way Six Sigma stifles creativity and innovation in businesses. Not a fan.
Click to expand...

I don't think it has turned into a feedback loop.

If teams are using xG based models to inform their players on optimal decisions to make, and the numbers show that decisions being made are having a decisive impact on the match, then that isn't a feedback loop, that is simply model validation. Otherwise any decent model trained on recent data would say, "hey, shots outside the box are better now that everyone is spamming cut back goals!" But that's not the case. The data validates the framework.

Now, if teams are solely using xG to inform their tactics prescriptively, that's stupid and short sighted from analytics teams, but that's not a fault of the model. xG is simply predicting the chance a shot will go in the net. That is very far removed from the question: "what decisions do we need to make on the training pitch and the game day pitch to maximize our chances of winning today and throughout the season?" That is a much harder analytical problem of which xG is only a part.

xG gets outsized attention because it's more easily digestible. I'd like to think analytics teams have more predictive and prescriptive models at their fingertips we don't know about.

And yes, everyone doing the same thing is boring but we can't blame xG for that. Because people are naturally copycats of successful systems, and we have seen this dynamic in football even before analytics became a thing. Analytics is capable of answering the question: "how do I yang successfully against the ying?" But that requires courage and conviction from clubs and coaches before the tools come into play.
 
sullydnl said:
Looking at Fbref, across the entirety of Salah's Liverpool career (a relatively big sample size) he is an above average finisher, scoring approximately 10 goals above his xG.

But one of the fundamental realities of football that analytics has made evident is that finishing matters a lot less to goalscoring than the ability to get chances. In Salah's case, finishing 0.04 per 90 above his xG matters far, far less than the fact he's expected to score a goal 136 minutes or so even with average finishing.

However, people being people, they tend to think of any player who scores a lot of goals as being an excellent finisher, even when it isn't driven by finishing ability.
Click to expand...
I agree regarding the ability to get on the end of chances is probably more important than actual finishing but I don’t think there’s many in the game or watching it who would consider Salah to be at best a slightly above average finisher.
 
sullydnl said:
Looking at Fbref, across the entirety of Salah's Liverpool career (a relatively big sample size) he is an above average finisher, scoring approximately 10 goals above his xG.

But one of the fundamental realities of football that analytics has made evident is that finishing matters a lot less to goalscoring than the ability to get chances. In Salah's case, finishing 0.04 per 90 above his xG matters far, far less than the fact he's expected to score a goal 136 minutes or so even with average finishing.

However, people being people, they tend to think of any player who scores a lot of goals as being an excellent finisher, even when it isn't driven by finishing ability.
Click to expand...
Yeah, 10 out of 173 league goals is under 6%, fair to say he's about average as a finisher, but 173 means he's brilliant at finding/making chances.

To add to your last point, the eye test means people remember the goals, not the misses, they're replayed a lot more of course.
 
acnumber9 said:
I agree regarding the ability to get on the end of chances is probably more important than actual finishing but I don’t think there’s many in the game or watching it who would consider Salah to be at best a slightly above average finisher.
Click to expand...

In some cases there's also an element of what @Ekeke alluded to on the previous page, of the overall xG over/under performance not capturing the nuances of players' own finishing strengths and weaknesses.

For example, KDB has statistically been one of the very best "finishers" in terms of xG overperformance in the last decade. But that primarily comes from his long range striking ability specifically. It doesn't necessarily follow that he's the player you most want in a 1v1 situation with the goalkeeper just because he overperforms his xG.

Similarly, you could in theory have some relative weakness in Salah's finishing repetoire (long range shots, headers, whatever) that impacts his overall numbers even while he demonsrates excellent finishing ability in the situations in which he actually tends to score his goals.

But mostly I think it's just what @711 notes above. People remember goals, not misses. And when you get as many chances as Salah you get a lot of goals for people to focus on.
 
Side note, but something that struck me when watching RVN's goals for us back again when he joined as a coach was just how often he used to get what would now be described as very high xG chances.

I used to think it was his finishing that was amazing, but for most of these goals he didn't need particularly good finishing at all. He was ridiculously good at getting shots dead centre and close to goal, in an era before teams were so heavily coached to generate those chances.

 
Understat's model is just obviously shite.

They have all but 3 clubs down as underperforming their xG this season so far.
 
GiveItToGi...nowait said:
I'm no expert, but I think whenever people complain about a statistic it's because they are either evaluating it in isolation or expecting a stat to be 1:1 with reality (fair assumption). xG I think is useful the more you zoom out from individual moments/chances/games and start looking at it over a longer period of time.

Of course you then have the reliability of the method used by those providing the statistics, which makes xG look even more confusing as it's not like shots on target which is fairly (although not always) straightforward.
Click to expand...

When data is unreliable in isolation I think people tend to underestimate just how big a sample size would be needed to draw any kind of useful results. The more unreliable data gets in isolation, the bigger the required sample size is needed.
 
sullydnl said:
In some cases there's also an element of what @Ekeke alluded to on the previous page, of the overall xG over/under performance not capturing the nuances of players' own finishing strengths and weaknesses.

For example, KDB has statistically been one of the very best "finishers" in terms of xG overperformance in the last decade. But that primarily comes from his long range striking ability specifically. It doesn't necessarily follow that he's the player you most want in a 1v1 situation with the goalkeeper just because he overperforms his xG.

Similarly, you could in theory have some relative weakness in Salah's finishing repetoire (long range shots, headers, whatever) that impacts his overall numbers even while he demonsrates excellent finishing ability in the situations in which he actually tends to score his goals.

But mostly I think it's just what @711 notes above. People remember goals, not misses. And when you get as many chances as Salah you get a lot of goals for people to focus on.
Click to expand...
All fair points.
 
sullydnl said:
Side note, but something that struck me when watching RVN's goals for us back again when he joined as a coach was just how often he used to get what would now be described as very high xG chances.

I used to think it was his finishing that was amazing, but for most of these goals he didn't need particularly good finishing at all. He was ridiculously good at getting shots dead centre and close to goal, in an era before teams were so heavily coached to generate those chances.

Click to expand...


It's very common for the best scorers. In an other thread I hypothesised that the actual best finishers are likely not strikers but some of the high scoring CMs or AMs that are rarely in the best positions but rely heavily on striking technique. Juninho would be an example.
 
sullydnl said:
But mostly I think it's just what @711 notes above. People remember goals, not misses. And when you get as many chances as Salah you get a lot of goals for people to focus on.
Click to expand...
I think that's the key here. If Salah shoots a goal, that's part of the highlight reel (which is what most people will see, more than the full match). If he misses a great chance, that will also make it. But if he scores and in the same match misses a bunch of chances, what will be in the highlights is his goal, nothing else. So getting a lot of chances means that you will get a lot of match highlights that contains goals, not misses (or just a few) of that specific player.
 
stefan92 said:
I think that's the key here. If Salah shoots a goal, that's part of the highlight reel (which is what most people will see, more than the full match). If he misses a great chance, that will also make it. But if he scores and in the same match misses a bunch of chances, what will be in the highlights is his goal, nothing else. So getting a lot of chances means that you will get a lot of match highlights that contains goals, not misses (or just a few) of that specific player.
Click to expand...

Of course, this is the case. acnumber9 is also right in what he's saying. Here's a thought experiment. Try to guess who has missed the most big chances in the PL in the last two seasons. Here's a hint: It's the same player in both seasons, and the number of big chances missed last season was a PL record. Make an honest guess, then check the PL site.
 
KiD MoYeS said:
I wonder do clubs actually use xG to drive decisions, does anyone have any information on this?
Click to expand...
Anyone who can afford it do. Not necessarily to drive decisions, it's just another very useful analytical tool. Mind you most of the time they're hiring genius statisticians and giving them huge money to develop their own highly sophisticated models internally. Or, they subcontract companies like Statsbomb for it.
 
TheRedDevil'sAdvocate said:
Of course, this is the case. acnumber9 is also right in what he's saying. Here's a thought experiment. Try to guess who has missed the most big chances in the PL in the last two seasons. Here's a hint: It's the same player in both seasons, and the number of big chances missed last season was a PL record. Make an honest guess, then check the PL site.
Click to expand...

Guessing Haaland.

Yep.
 
The main purpose of XG is to help dafter players understand and visualise how to find teammates in better scoring positions, i.e. Don't shoot from midfield, don't shoot from an extreme angle, don't shoot if you have 3 defenders blocking, pass to someone who has nobody in front of them etc. In the 90s-00s midfielders used to attempt low XG long-range shots more often than they do today and that's down to players now understanding the concept of XG even if long-range shot attempts are quite useful in terms of forcing rebounds and deflections.

It's pointless in terms of evaluating strikers, but if a decently expensive squad has consistently low XG it's indicative of poor performance or coaching since you'd want to be a side that creates more chances. Also, long before XG became mainstream I once used it on Excel for a Bundesliga Kicktipp game and my overall standings were slightly worse than the previous season.
 
How is the Bournemouth xG only currently 1.73. They should've scored about six today. :lol:
 
You must log in or register to reply here.