Expected Goals (xG)

[Stevo985]

I thought we already had a topic for this, but I can't find it anywhere. Apologies mods if I've somehow missed it

Anyway thought I'd start this as the discussion around xG crops up in loads of threads and generally people hate those threads being filled with debate on whether xG is or isn't useless. So here's somewhere where we can argue about it to our hearts' content

@MotoMkali @Andy_10 somewhere for us to carry on our discussion

 

As a start, this is a good read for anyone interested in the stat and it's history.

I'll be honest, it's actually quite snobby at the start, so you have to look past that. But it's a good explanation of how it works and where it came from (and interstingly it's connections to Brentford FC!)

[gilbertoAVFC]

I think our game on Tuesday was an excellent example of the limitations of publicly available xG.

We were one pass or touch away from creating all sorts, but it turned out their xG was higher according to most outlets.

The very best xG models will quantify dangerous attack potential which does not necessarily lead to a shot. However, most of these are not public because they make those running it millions through betting.

Bloom, Benham, etc. are the leaders in the field and are miles ahead of most xG models which you see day-to-day, although the gap is closing.

[Czarnikjak]

11 minutes ago, gilbertoAVFC said:

I think our game on Tuesday was an excellent example of the limitations of publicly available xG.

We were one pass or touch away from creating all sorts, but it turned out their xG was higher according to most outlets.

The very best xG models will quantify dangerous attack potential which does not necessarily lead to a shot. However, most of these are not public because they make those running it millions through betting.

Bloom, Benham, etc. are the leaders in the field and are miles ahead of most xG models which you see day-to-day, although the gap is closing.

I think what you described here is called Expected Threat, not xG. There's lots of YouTube videos about it.

[Zatman]

XG is interesting and useful but has its flaws. Main one is that every shot is determined by the percentage only. It doesnt take into account the player taking the shot. If Cash and Watkins have a similar chance they will be ranked the same but we know Watkins should have a higher probability to score

It might have moved on but older XG used to not count the defenders in the way of the shot

[Andy_10]

@Stevo985 , the way I understand xG is that it looks at incidents in a game where a shooting chance is created & compares it to other incidents in other games where shooting chances have been created that are similar in terms of distance, angle, etc, & forms a mathematical chance for the first incident mentioned, based on how many times the other incidents it was compared to, were scored.

I think this stat is pointless because I think to compare any chances, the scenario has to be exactly the same for each goal that is being compared against, & that is impossible in football.

And from a match to match basis, a probability is still only a probability.

Now that is as simple as I understand it (because I am a simple man) & can explain it back. And I apologise that my communication skills probably don't help make it make sense to you, lol, but I don't want to go too deep into my thinking over it, until I have heard why you think that I am wrong.

And I want to point out before we start that Im not being intentionally difficult or argumentative. My opinions come with a smile, a virtual handshake & respect for your opinion. And if you say something that makes sense to me, I am all in for amending my position.

I cant read that book unfortunately as I cant download it on my iPad, but on the other thread, you mentioned that I was incorrect with my assessment that xG is pointless, so could you explain to me why you think that?

Cheers.

 

Edited October 24 by Andy_10

[CarryOnVilla]

32 minutes ago, Zatman said:

XG is interesting and useful but has its flaws. Main one is that every shot is determined by the percentage only. It doesnt take into account the player taking the shot. If Cash and Watkins have a similar chance they will be ranked the same but we know Watkins should have a higher probability to score

It might have moved on but older XG used to not count the defenders in the way of the shot

It doesn’t matter, an xG goal is just a unit of measure. As long as that unit is using the same data to create the measurement, then you get the information you need from it. 

if Ollie on average scores 3 goals per xG goal you get it 

if it takes Matty Cash 10 xG goals per actual goal you’ll be disappointed but you’ll get it. 
 

same with teams

Any difference to the player/team’s average, there’s something worth investigating. Positive or negative 

Edited October 24 by CarryOnVilla

[Stevo985]

1 hour ago, Zatman said:

XG is interesting and useful but has its flaws. Main one is that every shot is determined by the percentage only. It doesnt take into account the player taking the shot. If Cash and Watkins have a similar chance they will be ranked the same but we know Watkins should have a higher probability to score

This is the entire point of xG

if it took into account the player taking the shot it would be useless.

[Stevo985]

1 hour ago, Andy_10 said:

@Stevo985 , the way I understand xG is that it looks at incidents in a game where a shooting chance is created & compares it to other incidents in other games where shooting chances have been created that are similar in terms of distance, angle, etc, & forms a mathematical chance for the first incident mentioned, based on how many times the other incidents it was compared to, were scored.

I think this stat is pointless because I think to compare any chances, the scenario has to be exactly the same for each goal that is being compared against, & that is impossible in football.

And from a match to match basis, a probability is still only a probability.

Now that is as simple as I understand it (because I am a simple man) & can explain it back. And I apologise that my communication skills probably don't help make it make sense to you, lol, but I don't want to go too deep into my thinking over it, until I have heard why you think that I am wrong.

And I want to point out before we start that Im not being intentionally difficult or argumentative. My opinions come with a smile, a virtual handshake & respect for your opinion. And if you say something that makes sense to me, I am all in for amending my position.

I cant read that book unfortunately as I cant download it on my iPad, but on the other thread, you mentioned that I was incorrect with my assessment that xG is pointless, so could you explain to me why you think that?

Cheers.

 

You're understanding of how it is calculated is right, and you're right that it's only a probability. but that's all that it's meant to be.

I think you're misunderstanding comes from what the stat is for.

 

The biggest flaw people have with xG is they think it's meant to predict the score of the game. So when there are small differences in situations, or when the score of a game doesn't match the xG, they think that proves it's wrong.
That's not what it's for. it is not a predictor of scorelines. if it was then I agree it would be pointless

 

The whole point of it is it's mean to give a probability of a goal from a shot in that position by the average player. That way it provides a benchmark for what is expected from that situation.

The best way to think about it is the same way you look at shots on target. It's an indicator of how the game has gone. If you saw a game we had lost 1-0, but we had 20 shots on target and the opposition had 1, then you'd have a good idea that even though we lost, we were probably very unlucky.

xG works in exactly the same way, but far far more accurately

 

When you then look at that data and it's trends over time it becomes very useful at assessing a team or a player's performance. 
If Watkins has scored 10 goals this season from an xG of only 3.5, then he's amazing at finishing. If Haaland has scored 20 goals but has an xG of 38, then despite the number of goals he hasn't been very clinical.

If Aston Villa have lost their last 4 games but have dominated the xG in each of those games, then the likelihood is it's just bad luck, or the strikers are having issues or they've conceded some bizarre goals. The likelihood is that the results will bounce back as the result regress back towards the mean because they're creating enough chances to win games
If they've lost those 4 games AND been dominated on xG then it would indicate results aren't likely to turn around because they're not creating chances

Edited October 24 by Stevo985

[villa89]

I am not a fan of using any statistics (expected or otherwise) to try to measure football as I don't think its a game that can be measured well by statistics. From what I know of xG it essentially just tells you the closer to the goal you are the more likely you are to score and is a relative measure of the quality of chance (i.e. the closer to the goal it is) X the number of chances you had.

The stats in football might be useful to gain information about a game you haven't watched or seen but for games you are watching they're pointless. You're eyes should be telling you everything and more than you get from stats. For scouting maybe you can use them to profile players that might be worth watching.  

Edited October 24 by villa89

[Zatman]

16 minutes ago, Stevo985 said:

This is the entire point of xG

if it took into account the player taking the shot it would be useless.

Yes over a season it probably adds up but if you are analysing an individual game its not helpful

[AndyM3000]

I know some people hate the xG stat but I always wonder do they hate the possession stats, shots on target?

Possession stat on it's own isn't worth much and neither is shots on target. ST/SOT metric paired with xG is useful IMO.

[Stevo985]

Just now, Zatman said:

Yes over a season it probably adds up but if you are analysing an individual game its not helpful

If you were trying to predict the score of an individual game then adjusting the xG to whoever was taking the shot would make sense.

But I don't know why you'd ever want to do that.

 

 

The whole point is that it provides an average benchmark. If you start adjusting it by player you completely lose that benefit

[Zatman]

3 minutes ago, AndyM3000 said:

I know some people hate the xG stat but I always wonder do they hate the possession stats, shots on target?

Possession stat on it's own isn't worth much and neither is shots on target. ST/SOT metric paired with xG is useful IMO.

I personally prefer the Bet365 one which measures Dangerous Attacks when judging a game as think as a stat gives more indication of whose on top

Along with Touches in the box, xg and Corners I think give a better overview of a game than possession. Lambert trying play possession football highlighted the flaws

[MotoMkali]

1 hour ago, Stevo985 said:

This is the entire point of xG

if it took into account the player taking the shot it would be useless.

I'm perfectly fine with using it as a metric for finishing. I just think using it to compare the chances created by teams is utter bollocks. 

[MotoMkali]

52 minutes ago, Zatman said:

I personally prefer the Bet365 one which measures Dangerous Attacks when judging a game as think as a stat gives more indication of whose on top

Along with Touches in the box, xg and Corners I think give a better overview of a game than possession. Lambert trying play possession football highlighted the flaws

Possession is unfortunately just a comparison of the number of passes. Teams that have high passing tempo will have better possession than a team with a lesser tempo even if the ream with the lesser tempo has control of the ball more often. 

[tinker]

Duran breaks Xg, the end.

[AndyM3000]

37 minutes ago, tinker said:

Duran breaks Xg, the end.

[viivvaa66]

Some people have the idea that a model needs to be perfect or else it’s useless, meaning in reality all models are useless.

xG is an improved model on what we had before that were shots and shots on target, and that what is it, nothing more nothing less. Of course actually goal scored is what counts, but xG provides information on the quality and number of chances created during a match, at least more than shots and shots on target does. As long as you understand what xG is and its limitation, I see nothing wrong with it.

 

[villa89]

3 hours ago, AndyM3000 said:

I know some people hate the xG stat but I always wonder do they hate the possession stats, shots on target?

Yeah, they're all pretty meaningless unless you haven't seen any of the game. 

Edited October 24 by villa89

[viivvaa66]

2 hours ago, MotoMkali said:

Possession is unfortunately just a comparison of the number of passes. Teams that have high passing tempo will have better possession than a team with a lesser tempo even if the ream with the lesser tempo has control of the ball more often. 

Number of passes was how Opta was calculating possession before 2017. Now they calculate number of ‘possessions’, a ‘possession’ is when a player is in control of the ball, until he no longer is in control. Then they add up the ‘possessions’ for a team and divide by the total number of ‘possessions’. Obviously teams that play one touch football will get more possession than a team where players hold on to the ball, dribble, run with the ball and so on. But it is an improvement to the passes method, since now if a player takes the ball from the opposition and not passes the ball it will count as possession. 

Calculating possession where time is recorded from a player from one team touches the ball until a player from the opposition touches the ball or the ball goes out of play, and then use that the calculate possession isn’t without flaws either. A route one team would get more possession than what they really should when the boot the ball high and long., the ball will be in the air for many seconds, that would count as possession. Also this method requires more resources, the Opta method is simpler and that’s why it is used.