XG?

[glastored]

Please excuse my ignorance, but what is ‘XG’? I keep seeing it mentioned in football analysis, but I’m sure I’m not the only person who hasn’t a clue what it signifies. Can someone enlighten me?

[Super]

1 minute ago, glastored said:

Please excuse my ignorance, but what is ‘XG’? I keep seeing it mentioned in football analysis, but I’m sure I’m not the only person who hasn’t a clue what it signifies. Can someone enlighten me?

The most pointless stat in football expected goals

Edited August 27, 2023 by Super

[Unan]

A useful stat in football that uses more than one isolated variable (such as goals/assists & the Scott comparison) to give a better (far from perfect) understanding of how a teams performing.

Edited August 27, 2023 by Marcus Aurelius

[Ziderarmy]

Can someone explain accurately how it’s calculated as I saw a game last week is was 1-0 at HT and the tens leading had a XG of 0.58

[!james]

https://statsbomb.com/soccer-metrics/expected-goals-xg-explained/

[Unan]

3 minutes ago, Ziderarmy said:

Can someone explain accurately how it’s calculated as I saw a game last week is was 1-0 at HT and the tens leading had a XG of 0.58

Put simply, Expected Goals (xG) is a metric designed to measure the probability of a shot resulting in a goal.

An xG model uses historical information from thousands of shots with similar characteristics to estimate the likelihood of a goal on a scale between 0 and 1.

For example, a shot with an xG value of 0.2 is one that we would generally expect to be converted twice in every 10 attempts.

 

[petehinton]

1 minute ago, Ziderarmy said:

Can someone explain accurately how it’s calculated as I saw a game last week is was 1-0 at HT and the tens leading had a XG of 0.58

Is based on the chances you create, basically. So if you have a penalty, that’ll create an xG of 1 straight away as you’re expected to score in that situation. 
 

Or if a team gets battered, but scores a 35 yard goal, their xG will be tiny as they’ve scored a worldie which wasn’t expected. 
 

People call it pointless, but it’s very very indicative of overall performance and dominance of a game. Give me consistently positive xG over possession any day of the week, as is being shown with a lot of clubs atm as largely useless. 

[bearded_red]

2 minutes ago, petehinton said:

Is based on the chances you create, basically. So if you have a penalty, that’ll create an xG of 1 straight away as you’re expected to score in that situation.

Penalties are roughly 0.8

[Dredd]

It's just another stat pulled out of someone's ass as an indicator of 'good performance' despite losing. 

[glastored]

Thanks for everyone’s replies, but I’m still not sure what I’m reading! Perhaps I should just leave such deliberations to the statisticians among you, and just accept I don’t have that type of mind!

[Davefevs]

18 minutes ago, glastored said:

Please excuse my ignorance, but what is ‘XG’? I keep seeing it mentioned in football analysis, but I’m sure I’m not the only person who hasn’t a clue what it signifies. Can someone enlighten me?

You know when a commentator says:

• that was a half chance, say a one-on-one, xG will give it a probability score between 0 and 1, based on 100s and 1000s of similar chances.  Everyone thinks a one-on-one should be scored everytime.  The reality is, they aren’t.  Aaron Connolly’s chance, on the angle he was on, etc, etc, was a 1 in 9 chance based on similar chances.  It got an xG score of 0.11.
• Nahki’s tap-in goal, a 7 in 10 chance, 0.70 xG.  How was it not a 1?  You saw the miss against Brum!

[bcfc01]

Not a fan, just another subjective metric to analyse a game.

 

[Davefevs]

Just now, bcfc01 said:

Not a fan, just another subjective metric to analyse a game.

 

It’s not subjective, it’s statistical.  It might be used subjectively

[Mr Popodopolous]

It often can be a decent guide over time, limited use in terms of individuals matches or a run of games.

Edited August 27, 2023 by Mr Popodopolous

[bcfc01]

Just now, Davefevs said:

It’s not subjective, it’s statistical.  It might be used subjectively

No, it is subjective.

Who decides what the chance is on a goal attempt ?

It is entirely subjective from its outset.

[spudski]

Funnily enough we had the number 1 stat for XG in the Championship this past week. Unheard of...

[archie andrews]

some super strength cider at the barleymow.......

[Slacker]

15 minutes ago, Dredd said:

It's just another stat pulled out of someone's ass as an indicator of 'good performance' despite losing. 

See Lee Johnson thread for more examples. 

[Mr Popodopolous]

It's not exhaustive but if a side are underperforming ie playing well but mediocre or worse the xG table may well reflect this. Usually this turns around.

Likewise if a side are performing better results v performance wise, the xG table will be likely to reflect this and results v performance may align over time.

A good example was when we were clearly better points v results in 2016. It caught up!

[Davefevs]

21 minutes ago, bcfc01 said:

No, it is subjective.

Who decides what the chance is on a goal attempt ?

It is entirely subjective from its outset.

It’s not a person say they’re deciding, it’s based on data.  There is sometimes human checking, but no different to a supervisor sample checking your work.

[Ziderarmy]

So the XG can never be higher than 1.0 then? 

Edited August 27, 2023 by Ziderarmy

[Leveller]

6 minutes ago, Ziderarmy said:

So the XG can never be higher than 1.0 then? 

Presumably for a single chance,no. But for match it could be several. 

[Davefevs]

19 minutes ago, Ziderarmy said:

So the XG can never be higher than 1.0 then? 

Here’s a screenshot of Wyscout xg from the other night.  There were obviously more shots, but I’d have to scroll down.

[Erithacus]

So the XG is based on historical data, but not an actual game in progress? Seems to be a rough indicator of the game or its aspects. I appreciate the ways stats have been integrated into modern football but I feel the application is far from comprehensive. In short, stats are no predictor of a given moment.

[Severn Beach Pigeon]

14 minutes ago, Erithacus said:

So the XG is based on historical data, but not an actual game in progress? Seems to be a rough indicator of the game or its aspects. I appreciate the ways stats have been integrated into modern football but I feel the application is far from comprehensive. In short, stats are no predictor of a given moment.

xG is never a predictor of a given moment

It tells you the strength of a given chance in a game based on past events

 

However it can (with other knowledge) be used to make other predictions.

Let's say a side has an xG of 20 over 10 games, but has only scored 6 times

Meanwhile their xGC (goals conceded) is 10 and they've conceded 20.

 

You could reasonably* suggest they've created a lot of good opportunities and been very wasteful in front of goal, while conceding goals more often than they 'should'.

So you could make a prediction that the poor form those goal numbers would lead to is not really a good representation of the level of performance in the game and that their form will soon turn around. (In fact those arbitrary numbers would be decent even for a team comfortably at the top end of the table)

 

*I say reasonably because in theory you could have 200 shots from 30 yards out with an xG of 0.01 each and get an xG of 2 for a game, but I'm sure we can all agree that's unlikely. Meanwhile a team could have 2 shots all game and an xG of about 1.5 if they were both penalties

Like any stat it can be useful, but doesn't mean much in isolation. It's how you use it with other stats that matters

Edited August 27, 2023 by transfer reader

[And Its Smith]

In theory it’s good. In practise it’s flawed.  One example is if there is a goal mouth scramble with 3 shots a team will get an xG for all 3 shots which doesn’t make sense as if the first shot had gone in then the other two wouldn’t have happened.  It would better if it judged chance creation better as well. Newcastles goal today would have an xG but they haven’t created that chance.  Finally, a shot from the edge of the box with 4 defenders between the shooter and the goal has the same xG as if there are no defenders there.  Hopefully xG will be improved one day as the principle is a good one 

[Mr Popodopolous]

Here is a worked example.. I'm using more basic shot data.

Norwich City. 4-4 draw, 4-0 win. I won't bother to filter out shots on target v penalties. This is their two away games.

24 Shots for, 49 conceded.

Shots on Target- 11 for, 16 against.

Logic dictates that they are outperforming on the road. Vastly. Interesting to see if they will keep it up..not so much the goals but the allowing so many shots and chances on the road.

It does suggest they can be got at, on the road especially. Very clinical at the same time but again that is well above the norm.

All the same, the Sofascore (so probably quite basic) xG..has them winning at Huddersfield but losing the 4-4.

Edited August 27, 2023 by Mr Popodopolous

[Davefevs]

15 minutes ago, And Its Smith said:

In theory it’s good. In practise it’s flawed.  One example is if there is a goal mouth scramble with 3 shots a team will get an xG for all 3 shots which doesn’t make sense as if the first shot had gone in then the other two wouldn’t have happened.  It would better if it judged chance creation better as well. Newcastles goal today would have an xG but they haven’t created that chance.  Finally, a shot from the edge of the box with 4 defenders between the shooter and the goal has the same xG as if there are no defenders there.  Hopefully xG will be improved one day as the principle is a good one 

….and that is why for ages I’ve been trying to say to you it’s not good for judging the result of a match, and why xg tables based on individual games are pretty futile.  But it is useful for individual player scouting for example.

Worth you looking at Statsbomb’s xG it is far more sophisticated than any others and solved many of the probs of previous models, inc defenders and goalkeeper positions.

[bcfc01]

An algrorithm based on historical subjective data.

I love stats and I use them everyday, but I don't like subjective stats.

 

 

[IAmNick]

1 minute ago, bcfc01 said:

An algrorithm based on historical subjective data.

I love stats and I use them everyday, but I don't like subjective stats.

 

 

How is the historical data subjective?

[Mr Popodopolous]

In defence of tables, Brentford often outperformed their model. Rose up eventually.

Often it catches up. It can, but usually a bit quicker!

[Davefevs]

11 minutes ago, bcfc01 said:

An algrorithm based on historical subjective data.

I love stats and I use them everyday, but I don't like subjective stats.

 

 

Where was the shot taken from, etc, etc….was it a goal?

• yes, or
• no

Repeat, repeat, x thousand times, build model of “similar” shots and score based on probability between 0 and 1, based on actual results.

https://statsbomb.com/news/statsbomb-data-launch-beyond-naive-xg/

 

[IAmNick]

2 hours ago, glastored said:

Thanks for everyone’s replies, but I’m still not sure what I’m reading! Perhaps I should just leave such deliberations to the statisticians among you, and just accept I don’t have that type of mind!

Let's say we have a chance next match, a shot from 10 yards out. Some companies/people have a huge database of every shot taken from 10 yards out from say, the last 20 years, so they know how often they're scored. If 20% are scored, that's an xG of 0.2 for that chance.

Do that for all the chances for each team in the game and you get their overall xG.

Now as fans we know not all chances from 10 yards are equal. Maybe one was at head height in a crowded box, one was booted awkwardly to your centre back at knee height with a defender right in front of him, and one was nicely on the right foot of favourite striker (who'd eaten his Shreddies that morning).

Some models now are including things like the above to make them more accurate, but they'll obviously never be perfect.

We have an xG model running in our heads at all times - it's how we know to say "we should have scored that!" Or "I can't believe we won that..." On the way home.

It's just an objective statistical measure of how good your chances were, a rough "how many goals might we have expected to score given the chances we had". Of course sometimes one team will get an xG of 5.8 and score nothing, while the other team scores from their single 0.04xG attempt from the half way line - that's just football, it doesn't mean xG is 'wrong' any more than flipping heads 10 times in a row means your coin is wrong. It's just unlikely.

It's not predictive, but it is interesting and over many games can show you trends. It's also massively misused imo.

Edited August 27, 2023 by IAmNick

[Severn Beach Pigeon]

1 hour ago, Mr Popodopolous said:

Here is a worked example.. I'm using more basic shot data.

Norwich City. 4-4 draw, 4-0 win. I won't bother to filter out shots on target v penalties. This is their two away games.

24 Shots for, 49 conceded.

Shots on Target- 11 for, 16 against.

Logic dictates that they are outperforming on the road. Vastly. Interesting to see if they will keep it up..not so much the goals but the allowing so many shots and chances on the road.

It does suggest they can be got at, on the road especially. Very clinical at the same time but again that is well above the norm.

All the same, the Sofascore (so probably quite basic) xG..has them winning at Huddersfield but losing the 4-4.

I believe sofascore uses Opta for its data, so in theory should be pretty good for its stats

[Mr Popodopolous]

21 minutes ago, transfer reader said:

I believe sofascore uses Opta for its data, so in theory should be pretty good for its stats

Thanks. Didn't realise that.

[Tom]

There is good evidence in this thread that we should be teaching Critical Thinking in all schools.

[ExiledAjax]

Essentially, Expected Goals (xG) is a metric that's intended to measure the probability of a shot resulting in a goal. The purpose is to show when a player could be expected to score from a particular opportunity, by basically rating how good of a goal-scoring opportunity it is. What the xG model does is use historical information regarding similar chances and shots to create a picture of how likely a goal is to be scored.

Each shot gets a rating between 0 and 1, which is basically the % likelihood of that shit becoming a goal. For example, about 80% of all penalties are scored, so a penalty generates xG of 0.8 (roughly).

Add up all the ratings for all the shots taken by a team in a match and you get the xG for that team.

The problem though is that:

1. There are multiple models and formulae out there, and they'll each generate slightly different figures; and

2. People wrongly say xG tells you which team "should" have won a game. No, it doesn't, it simply tells you which team had the best chance of scoring.

8 hours ago, Super said:

The most pointless stat in football expected goals

Possession.

[BigTone]

9 hours ago, glastored said:

Please excuse my ignorance, but what is ‘XG’? I keep seeing it mentioned in football analysis, but I’m sure I’m not the only person who hasn’t a clue what it signifies. Can someone enlighten me?

Its just another pointless way of over complicating what is a simple game.  Only 3 things need to be remembered:

1) Score more than the opposition = Win

2) Score the same as the opposition = Draw

3) Score less than the opposition = Lose

[Major Isewater]

I used to drive one of these in the 80’s 

[TV Tom]

It's utter nonsense, sort of thing Lee Johnson would create

[Lorenzos Only Goal]

10 hours ago, glastored said:

Please excuse my ignorance, but what is ‘XG’? I keep seeing it mentioned in football analysis, but I’m sure I’m not the only person who hasn’t a clue what it signifies. Can someone enlighten me?

A complete dispassionate variable that is a retrospective view of the expected goals per game.  

England (0.5) vs Spain (2.2) summed up really well the Spanish dominance and helps punters wade through crap commentary.

[IAmNick]

35 minutes ago, TV Tom said:

It's utter nonsense, sort of thing Lee Johnson would create

Can you explain why?

[TV Tom]

Just now, IAmNick said:

Can you explain why?

Lies, damned lies & statistics

[IAmNick]

Just now, TV Tom said:

Lies, damned lies & statistics

I'll take that as a no then!

[TV Tom]

Just now, IAmNick said:

I'll take that as a no then!

Correct

[Silvio Dante]

Here’s a very interesting example in action from Saturday. Essentially, Brighton had all the ball but created very little clear cut (and tbh you could see that with the naked eye). West Ham winning 3-1 with the xG in place indicated it wasn’t a smash and grab but that they soaked up and then created decent opportunities on the break.

I don’t think it overly shows anything that a seasoned watcher couldn’t see from viewing the game but having evidence to back up opinion helps in any field.

[TV Tom]

2 minutes ago, Silvio Dante said:

Here’s a very interesting example in action from Saturday. Essentially, Brighton had all the ball but created very little clear cut (and tbh you could see that with the naked eye). West Ham winning 3-1 with the xG in place indicated it wasn’t a smash and grab but that they soaked up and then created decent opportunities on the break.

I don’t think it overly shows anything that a seasoned watcher couldn’t see from viewing the game but having evidence to back up opinion helps in any field.

I'm sorry, apart from that XG nonsense Brighton battered them, with those sort of stats under normal circumstances Brighton would of won that game 9 times out of 10 or am i missing something ?, 

[42nite]

What a load of bollox!

[Bazooka Joe]

XG is a recently devised football metric.

It is used to analyse games, performance etc alongside other metrics such as

JG, IG, SG, TG, WG, CG, VG and YG

They couldn't use BG because an Australian trio's lawyers threatened to sue them.

The same with NG which belongs to Cardiff and MG which belongs to a car manufacturer.

Unfortunately, GG was a non-starter.

 

 

[Calculus]

14 minutes ago, Bazooka Joe said:

XG is a recently devised football metric.

It is used to analyse games, performance etc alongside other metrics such as

JG, IG, SG, TG, WG, CG, VG and YG

They couldn't use BG because an Australian trio's lawyers threatened to sue them.

The same with NG which belongs to Cardiff and MG which belongs to a car manufacturer.

Unfortunately, GG was a non-starter.

 

 

Couldn't we just go back to AG as a performance metric?

(Actual Goals for those not in the statisterati).

[Open End Numb Legs]

To summarise then, the OP asked a good question!

[Silvio Dante]

1 hour ago, TV Tom said:

I'm sorry, apart from that XG nonsense Brighton battered them, with those sort of stats under normal circumstances Brighton would have won that game 9 times out of 10 or am i missing something ?, 

Probably the quality of the chances. Watching the game live, there was one excellent save from Ferguson start of the second half but typically Brighton had a lot of ball but didn’t really test the keeper. You’re totally right to say with those other stats normally you’d expect Brighton to win the game but the point is that they didn’t create good enough chances and the xG is a quantification of that.

I’m no lover of xG - as I said, I’m not sure it shows you anything you don’t see yourself. But again, it’s useful to give meat to the bones of an opinion.

[Alessandro]

10 hours ago, Davefevs said:

Here’s a screenshot of Wyscout xg from the other night.  There were obviously more shots, but I’d have to scroll down.

What I don’t understand about the XG here is how many factors taken into consideration:

Is no.4 the chance Joe missed?

Let’s take that for example - it was 0.20 - so XG is saying only 2/10 times a player will hit the target from 12 yards out?! 
 

That’s surely wrong - so can only presume 0.20 is the 2/10 chance of scoring if the goalkeeper is in the right place - in this case it’s clearly not taken into consideration that the keeper is flat on the floor out of the place and the goal is gaping?!

Also out of interest - if you have time, don’t worry if not - what was the XG on Knights header in the second half from the corner? Thank you! 

[ExiledAjax]

41 minutes ago, Alessandro said:

What I don’t understand about the XG here is how many factors taken into consideration:

Is no.4 the chance Joe missed?

Let’s take that for example - it was 0.20 - so XG is saying only 2/10 times a player will hit the target from 12 yards out?! 
 

That’s surely wrong - so can only presume 0.20 is the 2/10 chance of scoring if the goalkeeper is in the right place - in this case it’s clearly not taken into consideration that the keeper is flat on the floor out of the place and the goal is gaping?!

Also out of interest - if you have time, don’t worry if not - what was the XG on Knights header in the second half from the corner? Thank you! 

No. It's saying that 2/10 times that shot turns into a goal. 

Different models take different factors into account. The most sophisticated adjust for defender and goalkeeper positions, pressure on the player, and for the average ability at the level the player plays. More simple models just use location of the shot, right foot, left foot etc.

This one?

 

Edited August 28, 2023 by ExiledAjax

[Mr Popodopolous]

1 hour ago, Calculus said:

Couldn't we just go back to AG as a performance metric?

(Actual Goals for those not in the statisterati).

It's one thing but if a side are outperforming their results in terms of performance, xG often shows this and it will catch up with a side.

Take 4 years ago for examples, we had 20 pts from 11 games..the rate at which we were gaining points wad outstripping our performances and we regressed to the mean somewhat.

Under Holden in 2020, even more pronounced.

2016 too arguably.

[Henry]

20 minutes ago, ExiledAjax said:

No. It's saying that 2/10 times that shot turns into a goal. 

Different models take different factors into account. The most sophisticated adjust for defender and goalkeeper positions, pressure on the player, and for the average ability at the level the player plays. More simple models just use location of the shot, right foot, left foot etc.

This one?

 

I think this just proves what over complicated crap it is. 
 

Joe Williams shot was a far easier one to score than Knight. 

[Maltshoveller]

XG is a total waste of time

The only thing that matters is what happened in a game

Not what coulda woulda shoulda 

[RoystonFoote'snephew]

I'd rather base the likelihood of scoring on who's taking the shot, for example Bas Savage might of hit the target once in 50 attempts, Oli Gunnar Soljaer 9 out of every 10 

[ExiledAjax]

42 minutes ago, Henry said:

I think this just proves what over complicated crap it is. 
 

Joe Williams shot was a far easier one to score than Knight. 

Yes. That's what xG says.

0.2 for Williams, 0.12 for Knight's header. So xG reckons Williams shot would be a goal 1 in 5 times, and Knight's about 1 in 8. So yes, Williams' chance was quite a bit better.

xG has taught me that we as fans tend to greatly underestimate how hard it is to score a goal.

[ExiledAjax]

11 hours ago, IAmNick said:

We have an xG model running in our heads at all times - it's how we know to say "we should have scored that!" Or "I can't believe we won that..." On the way home.

Love this example.

To add, our internal xG is compromised by our biases, alcohol levels, peer pressure, whether we like the player or not, Alan Shearer's mediocre opinion, is it a City player or the opponent? All that subjective crap that makes us wonderfully fallible and human.

xG models strip all that away and gives you the objective facts.

Have all the inconsequential chat on the way home, argue it out, and enjoy the debate. But.  Before you actually decide if the manager should be sacked or not...check the actual xG figures.

[Mr Popodopolous]

1 hour ago, Maltshoveller said:

XG is a total waste of time

The only thing that matters is what happened in a game

Not what coulda woulda shoulda 

It predicted our decline in 2016, relative decline in 2019 and decline in 2020 from a strong start however.

Take the below. 30 pts from 17 is quite impressive. However we we know we were lucky in a number of those games..the XG reflects!

Lo and behold we did decline. Quite sharply!

Edited August 28, 2023 by Mr Popodopolous

[TV Tom]

In Brightons case they had

80% possession,

25 shots to 12,

736 passes to 199 (not that i'm a fan of that stat')

17 corners to four

 

These stats tell me that in all likelihood Brighton should of won and probably would do 4 times out of 5 yet XG says differently by quite a considerable margin all though in fairness it says that WHU should of won 3-1 which remarkably is exactly what they did !!!!!

 

[Maltshoveller]

3 minutes ago, TV Tom said:

In Brightons case they had

80% possession,

25 shots to 12,

736 passes to 199 (not that i'm a fan of that stat')

17 corners to four

 

These stats tell me that in all likelihood Brighton should of won and probably would do 4 times out of 5 yet XG says differently by quite a considerable margin all though in fairness it says that WHU should of won 3-1 which remarkably is exactly what they did !!!!!

 

So what it tells you is

That if you score 3 and only let in 1 

you will win 100% of the time

[downendcity]

Only one stat that matters - the final score.

As they say in golf, at the end of the round it's not the how that counts, just the how many

[ExiledAjax]

2 minutes ago, downendcity said:

Only one stat that matters - the final score.

As they say in golf, at the end of the round it's not the how that counts, just the how many

How do you score goals?

How do you make a birdie on the golf course?

[Mr Popodopolous]

2019. While this model is a bit questionable, it also shows some patterns that were borne out ie our form was vastly exceeding our performance level. I don't agree with the 14 Point difference but it was quite prescient for a few sides.

At the other end, Leeds were comfortably the best side under that model..and so it proved.

Brentford were 17th after 11 games in real time yet the model fairly accurately forecast that they would be up nearr the top if they kept playing as they were.

Charlton also fell away majorly by the end of the season. Their 18 pts from 11 games did not reflect and was unsustainable on that Performance level.

Otoh Stoke there was a huge mismatch!

Edited August 28, 2023 by Mr Popodopolous

[johnbytheriver]

14 hours ago, !james said:

https://statsbomb.com/soccer-metrics/expected-goals-xg-explained/

I have just tried to read this! I must be too old! What utter Bollocks!

 

[bearded_red]

Almost all professional clubs use it in analysis of games and scouting of players.

Professional gamblers made a lot of money using it before it got into the mainstream.

Must be bollocks though, otib said so.

[Alessandro]

43 minutes ago, ExiledAjax said:

Yes. That's what xG says.

0.2 for Williams, 0.12 for Knight's header. So xG reckons Williams shot would be a goal 1 in 5 times, and Knight's about 1 in 8. So yes, Williams' chance was quite a bit better.

xG has taught me that we as fans tend to greatly underestimate how hard it is to score a goal.

Really surprised by those stats (an open goal at 12 yards and a free header at 5 yards) but there we go, who am I to argue with the computers!! 

Edited August 28, 2023 by Alessandro

[cidercity1987]

Said it before , il say it again.

Don't trust XG and haven't since Diedhiou tap in from a barely trickling cross on the goal line at Fulham was only about 0.6XG. Great team move though