xG Discussion

[Guest Cujek]

5 hours ago, egg_fried_rice said:

 

@StriderHiryu It may not have felt indicative of the game as a whole because that's not what it measures - if you put all bias to one side and purely look at the quality of the chances where shots were taken, I would suggest Burnley's were superior.

 

@messerschmitt Yep - if you take the Wood chance which Kasper saved as well as his 'shot' which VAR ruled out - they would both have pretty high xG stats because of the position they occurred.

 

 

Yes! Well sort of, but based on a consistent statistical model rather than subjective opinion.

 

 

 

@Foxxed @Cujek No one forces you to take any notice of these stats, nor is this a stat which claims to have boiled down the essence of football. When I'm at the game, the thought of xG never even enters my mind. It just provides interesting context after the fact and can be used as just one of a wide ranging group of metrics for more in-depth analysis on the team's performance

 

@AKCJ It's only bollocks if you expect it to measure those things. It's just a stat which measure one particular metric. The best players consistently outperform xG. It is in fact a great metric for separating the likes of Messi and Low - Messi scores over xG because he's better than average - Low scores less than xG because he's worse than average.

The best xG models do actually account for goalkeeper positioning and you could look at an expected saves metric (and how much a goalkeeper outperformed that) if you wanted to see how the quality of keeper would impact a shot.

Why shouldn't every penalty have the same xG? Given what I've stated above - it's the same quality of chance for everyone and the stat by itself does not claim to account for the quality of the keeper or other factors such as the occasion.

Aah the old "no one forces you" argument. 

 

If you use that then you could argue no one forces you to do anything, no one forced you to tell me no one forces me to do anything. 

[Nalis]

I find it a useful stat as an addition to other stats in football but its not the be all and end all.

[UniFox21]

18 minutes ago, Nalis said:

I find it a useful stat as an addition to other stats in football but its not the be all and end all.

It's a great stat that can add further context to other stats such as possession and shots etc. It can allow a team to work out where they're going right m or where they're losing games. 

 

But as with every stat, when people start taking it above what it is and try to make it mean too much is where it starts to fail.

[Guest Fktf]

Everyone in here is correct. Is xG going to tell you anything about a single game that you can't see with your own eyes? No, probably not. Is xG going to tell you anything about a pattern of play over a season or two, when memories of the individual games have faded. Yes, most likely. 

[BoyJones]

I'm old school, the main stat for me is goals actually scored against goals actually conceded. Tends to be a reliable barometer. 

[Leeds Fox]

2 hours ago, Fktf said:

Everyone in here is correct. Is xG going to tell you anything about a single game that you can't see with your own eyes? No, probably not. Is xG going to tell you anything about a pattern of play over a season or two, when memories of the individual games have faded. Yes, most likely. 

 

It doesn’t tell you anything about a pattern of play. 

[Leeds Fox]

It’s irrelevant for many reasons but it’s extremely flawed too.

 

It doesn’t take into account:

 

- The speed or direction the ball is reaching the player taking the shot. 

 

- If the shot has to be taken with the players stronger/weaker foot.

 

- If the player is sprinting, balanced, well set to take the shot.

 

- If a player is stretching to reach the ball.

 

- The defenders between the ball and the goal, possibly narrowing the angle. 

 

- Importance of scoring, increased pressure on the striker. 

 

- Conditions of the pitch or weather. 

 

Position and previous scoring patterns from such positions are taken into account but all of the above have a massive impact on the successfulness of the shot.

 

Most of the above make every shooting opportunity different than simply what position it was taken from. 

Edited 25 October 2019 by Leeds Fox

[Nalis]

Two questions -

 

Is xG was ever evolving? (I presumed it wa)s)

Does XG take into consideration players and the chance, and also the improvement of players over time?

 

Bit of an out there example, if Gray is one to one with the keeper tonight, is the xG is lower than Vardy's in the same position?

 

But if Gray suddenly turns from winger to 20 goal a season striker a la Henry, then his chances of scoring are seen as much higher, xG of the chance would surely increase, no?

Edited 25 October 2019 by Nalis

[Foxymcoxy]

18 hours ago, KingsX said:

 

That is entirely untrue.  Most (all?) of the models are proprietary.  The events are recorded, and numbers calculated, by companies that sell the results (or by clubs themselves, at our level).  We are simply accessing what end results get published on the Web.

 

Read this carefully.  There will be a quiz:

 

The simple model looks at each attempt (not just on goal) and assigns an Expected Goals figure.  e.g. a shot from that location has proven to result in a goal 25% of the time.  So 0.25 xG.  This may be adjusted for other factors which are trade secrets, e.g. whether there is a defender between the shot and the frame, and/or the historical success rate of the player taking the action.  (So in more advanced models, a Vardy shot would have a higher xG than the same Ndidi shot.)

 

There are non-shot models that estimate how many goals a team “should” have scored based on passes, interceptions, take-ons and tackles around the opposing team’s goal.  e.g. intercepting the ball at the opposing team’s penalty spot results in a goal about 9 percent of the time (0.09 xG), and a completed pass received at the center of the six-yard box leads to a goal about 14 percent of the time (0.14 xG).

 

I’d bet the house that clubs buying and using these stats, use a combination of approaches, plus advanced stats that do require a PhD.  But they are not about to give me a peek behind the curtain.

 

Some of you would fire our new analytics whiz as fast as Ranieri fired the sports psychologist.  Maybe the same ones who would rather see us back in the Championship playing real blood-and-guts football?  You would get your wish, briefly, on our way to League Two.

 

I work in data analysis and I'm telling you anyone with a clue will be cautious using black box metrics like this. Of course it is specifically designed to be predictive of goals but how it works and in what situations it has weaknesses are things you need to know if you are going to discuss it's merits.

[Guest Fktf]

8 hours ago, Leeds Fox said:

It’s irrelevant for many reasons but it’s extremely flawed too.

 

It doesn’t take into account:

 

- The speed or direction the ball is reaching the player taking the shot. 

 

- If the shot has to be taken with the players stronger/weaker foot.

 

- If the player is sprinting, balanced, well set to take the shot.

 

- If a player is stretching to reach the ball.

 

- The defenders between the ball and the goal, possibly narrowing the angle. 

 

- Importance of scoring, increased pressure on the striker. 

 

- Conditions of the pitch or weather. 

 

Position and previous scoring patterns from such positions are taken into account but all of the above have a massive impact on the successfulness of the shot.

 

Most of the above make every shooting opportunity different than simply what position it was taken from. 

Do you know all of these are true? Genuine question, as I've not come across information about how the stat is calculated. I'm inclined to think the first 5 things you list here will be (or at least could be) included in the mathematical model that calculates xG. 

[Guest Fktf]

8 hours ago, Leeds Fox said:

 

It doesn’t tell you anything about a pattern of play. 

Agreed. Wrong terminology on my part. My point was that it gives some sort of record of performance over a long period of time. 

[Captain...]

21 hours ago, messerschmitt said:

After the Burnley game on Saturday I was amazed to see that they had a better XG than we did, having now watched the game twice I still don't see this. When Vardy headed the ball this must have had a higher XG than when Wood scored, he had to jump between two defenders and only had the near corner to put it, even then Kasper got a hand on it? The only other decent chance they had was at the beginning of the second half when Wood turned on a dropping ball and put it wide.

To a fan Wood's chance is harder, but it might have a higher xG because he was closer. My understanding is the position that the shot is taken from is the main factor and proximity of defenders and goal keeper don't really come into it. A free header from 10 yards will have a lower xG then an under pressure header from 5 yards.

 

xG is useful for analysing a strikers finishing ability and is a more useful stat than shots/shots on target for analysing team performance, but is ultimately flawed as it doesn't take into account game management and tactics.

 

Burnley are a classic sit back team, if they score early with a low xG chance and then don't create much they will have a low team xG but still win the game if they defend well.

 

Doesn't mean they are underperforming, likewise a team defending a lead will inevitably concede more chances late on, but are in defence mode so will have more players back defending in numbers making those chances harder in a way xG cannot take into account.

 

It has its place in football analysis, but like any stat it can only tell you part of the story.

[Xen]

10 hours ago, Leeds Fox said:

It’s irrelevant for many reasons but it’s extremely flawed too.

 

It doesn’t take into account:

 

- The speed or direction the ball is reaching the player taking the shot. 

 

- If the shot has to be taken with the players stronger/weaker foot.

 

- If the player is sprinting, balanced, well set to take the shot.

 

- If a player is stretching to reach the ball.

 

- The defenders between the ball and the goal, possibly narrowing the angle. 

 

- Importance of scoring, increased pressure on the striker. 

 

- Conditions of the pitch or weather. 

 

Position and previous scoring patterns from such positions are taken into account but all of the above have a massive impact on the successfulness of the shot.

 

Most of the above make every shooting opportunity different than simply what position it was taken from. 

There are different models which do take into account different combinations of these - along with plenty of other potential stats.

 

 

Besides, statistical modelling is a heavily involved process which may result in certain factors which look like they'll be important actually proving to be statistically insignificant, and being dropped from the final summary during the modelling process. 

[Saxondale]

xG is completely meaningless. 

 

The best indicator of how many goals your team should have scored is...how many they scored.

[Leeds Fox]

5 hours ago, Fktf said:

Do you know all of these are true? Genuine question, as I've not come across information about how the stat is calculated. I'm inclined to think the first 5 things you list here will be (or at least could be) included in the mathematical model that calculates xG. 

 

No to be fair I don’t, I’ve based that on what I’ve read and the video someone posted explaining it. 

 

The first 5 aren’t mentioned in the calculations but I suppose that doesn’t mean they aren’t taken into account. 

 

For every shot taken in a certain position, there are so many variables that it makes it very difficult to predict the probability of scoring surely. No two opportunities are the same in respect to the received ball dynamics, keeper positioning, defender proximity to the front, need to take the shot quickly under pressure, passing options to alter keeper/defenders decision making. 

[Guest Fktf]

1 hour ago, Leeds Fox said:

 

No to be fair I don’t, I’ve based that on what I’ve read and the video someone posted explaining it. 

 

The first 5 aren’t mentioned in the calculations but I suppose that doesn’t mean they aren’t taken into account. 

 

For every shot taken in a certain position, there are so many variables that it makes it very difficult to predict the probability of scoring surely. No two opportunities are the same in respect to the received ball dynamics, keeper positioning, defender proximity to the front, need to take the shot quickly under pressure, passing options to alter keeper/defenders decision making. 

By no means is it going to be a perfect. As you say, there are too many variables. But if it is accurate enough, then the stat should have some utility. It'll be no more useful than any other stat, but no less useful either. 

[yorkie1999]

What's the score going to be tonight?

[lildave3]

15 hours ago, BoyJones said:

I'm old school, the main stat for me is goals actually scored against goals actually conceded. Tends to be a reliable barometer. 

 

2 hours ago, Saxondale said:

xG is completely meaningless. 

 

The best indicator of how many goals your team should have scored is...how many they scored.

Not though is it. 

[Raw Dykes]

1 hour ago, Leeds Fox said:

 

No to be fair I don’t, I’ve based that on what I’ve read and the video someone posted explaining it. 

 

The first 5 aren’t mentioned in the calculations but I suppose that doesn’t mean they aren’t taken into account. 

 

For every shot taken in a certain position, there are so many variables that it makes it very difficult to predict the probability of scoring surely. No two opportunities are the same in respect to the received ball dynamics, keeper positioning, defender proximity to the front, need to take the shot quickly under pressure, passing options to alter keeper/defenders decision making. 

There are variables, yes. That's why an average has been taken over many of each shot from each area of the pitch. The probability of scoring is based on a large sample for each type of chance.

[Raw Dykes]

2 hours ago, Saxondale said:

xG is completely meaningless. 

 

The best indicator of how many goals your team should have scored is...how many they scored.

It's just not, though. It's a good indication of whether you're creating enough decent chances or not.

 

No. How many you scored isn't the same as how many you should have scored. Obviously.

[Stadt]

Burnley 17/18, Arsenal 18/19, Man Utd 18/19 all over performed massively on xG at times then regressed to the mean. xG isn't the be all and end al but it's the single best predictor of future performance at the minute.

[Les-TA-Jon]

4 hours ago, Saxondale said:

xG is completely meaningless. 

 

The best indicator of how many goals your team should have scored is...how many they scored.

Wait, what!?

 

So if a team misses 5 open goals during a game, but it finishes 0-1, you’re going to confidently say their blank was a true reflection of how many they should have scored? 

[a_k_b]

It’s not a stat it’s an opinion. And it’s rubbish

[Charl91]

11 hours ago, Nalis said:

Two questions -

 

Does XG take into consideration players and the chance, and also the improvement of players over time?

 

No it doesn't individual players into account.  My understanding is that It basically shows what the "average" goal scoring rate from those chances are. For example, we might have an XG of 1.3, but score 3 goals in a game, because Vardy is a very clinicial finisher, so is much more likely to convert chances that the average player won't.

 

It's a statistic that just shows how many goals you can reasonably expect from the average player, based on the chances provided. Over time, good strikers will score more than their XG, while poor players will score less.

 

I like XG, and I'm not quite sure why people get their knickers in a twist about it. Though I do wonder how it determines what the "average" chance of scoring from a certain position is. Is it just using PL data? I suppose my only query is that if it uses data from a range of different leagues, you'd expect the average PL player to be superior to the average player from say the Belgium league - therefore most teams should be out performing their XG.

 

But yeah, still think it's a good. Obviously it's not a be all and end all, but like possession, shots on target, etc. it helps paint a picture of the game.

 

 

Edited 25 October 2019 by Charl91

[Charl91]

1 minute ago, a_k_b said:

It’s not a stat it’s an opinion. And it’s rubbish

Why is it rubbish?