| I still don't understand xG - plus Delap and Enciso 11:45 - Jan 23 with 2851 views | GavTWTD |
Is there one xG calculation? I feel that xG is based on circumstances and therefore opinion on what circumstances to include, so could I come up with my own xG?
I don't understand Delap's figures (from whoscored):
Tournament Apps Mins xG Goals xGDiff xG/90 Shots xG/Shots Rating
Premier League 20(1) 1614 8.43 8 -0.43 0.47 37 0.23 6.71
Ok so his xG per 90 is easily calculated.
His xG is 8.43. What's that mean?
xG Diff is -0.43 What's that mean?
xG/shot I assume is easily worked out if you know how many shots he's made but I've not attempted to work it out.
Enciso
Premier League 2(10) 288 1.16 - -1.16 0.36 20 0.06 6.37
His figures don't look good in comparison but he's not a striker and not played many mins. I just thought it would be interesting to include as he's just signed. |  |
| |  |
| I still don't understand xG - plus Delap and Enciso on 11:49 - Jan 23 with 2627 views | Cotty |
xG is computed using spatio-statistics. They take the positions of the ball and the players when the shot is struck and put it into a big regression model which has lots of data about positions of players and shot outcomes. From that they can estimate what percentage of the time a player in that position would score. Its accuracy depends on the data, and the quality of the regression. This probably hasn't helped at all, has it? |  | | |
| I still don't understand xG - plus Delap and Enciso on 11:50 - Jan 23 with 2616 views | Dennyx4 |
The xg difference is what he has scored v what xg says he should have scored.
I would guess the two big chances v Man Utd were quite high xg.
I believe (may be wrong) that the Xg relates to how many times a goal is scored from the position the shot is taken, never going to be an exact science, as keepers can be positioned differently, time pressures etc.
As a side note, believe penalties are 0.8 xg |  | | |
| I still don't understand xG - plus Delap and Enciso on 11:50 - Jan 23 with 2614 views | Trequartista |
His xG is 8.43. What's that mean? That's his expected goals from 20(1) games.
xG Diff is -0.43 What's that mean? Difference between xG (8.43) and actual goals (8) |  |
| |
| I still don't understand xG - plus Delap and Enciso on 11:50 - Jan 23 with 2610 views | GavTWTD |
| I still don't understand xG - plus Delap and Enciso on 11:49 - Jan 23 by Cotty |
xG is computed using spatio-statistics. They take the positions of the ball and the players when the shot is struck and put it into a big regression model which has lots of data about positions of players and shot outcomes. From that they can estimate what percentage of the time a player in that position would score. Its accuracy depends on the data, and the quality of the regression. This probably hasn't helped at all, has it? |
Who is "they"? I think that's the crux. Are there multiple "theys"?
This post has been edited by an administrator | |
| |
| I still don't understand xG - plus Delap and Enciso on 11:51 - Jan 23 with 2601 views | bluesbrothers |
It analyses all historic shots and whether they were scored or not. Uses data to compare a shot to all that historic data to create a % chance of that shot being a goal, irrelevant of the quality of player who struck it.
For example, historically 77% of penalties have been scored and so the xG of a penalty is 0.77 - i.e. for every penalty that occurs in a match you'd expect 0.77 goals
So Delap "SHOULD" have scored 8.43 goals this season based on the quality of chance where he's taken a shot. However, he's only scored 8 and so has scored 0.43 goals less than he should have.
Some sources will show different xG to others as they might have a different depth of historic data, or be getting the original data from a different source with slight discrepencies.
There's a book called "The xG philosophy" which is good. |  | | |
| I still don't understand xG - plus Delap and Enciso on 11:51 - Jan 23 with 2589 views | Cotty |
| I still don't understand xG - plus Delap and Enciso on 11:50 - Jan 23 by GavTWTD |
Who is "they"? I think that's the crux. Are there multiple "theys"?
This post has been edited by an administrator |
Statisticians with a big computer. | | | |
| I still don't understand xG - plus Delap and Enciso on 11:53 - Jan 23 with 2520 views | bluesbrothers |
| I still don't understand xG - plus Delap and Enciso on 11:51 - Jan 23 by bluesbrothers |
It analyses all historic shots and whether they were scored or not. Uses data to compare a shot to all that historic data to create a % chance of that shot being a goal, irrelevant of the quality of player who struck it.
For example, historically 77% of penalties have been scored and so the xG of a penalty is 0.77 - i.e. for every penalty that occurs in a match you'd expect 0.77 goals
So Delap "SHOULD" have scored 8.43 goals this season based on the quality of chance where he's taken a shot. However, he's only scored 8 and so has scored 0.43 goals less than he should have.
Some sources will show different xG to others as they might have a different depth of historic data, or be getting the original data from a different source with slight discrepencies.
There's a book called "The xG philosophy" which is good. |
To be clear, there is not one xG stat. Many companies compute their own using variables from Opta or others. Opta also have their own, but it comes at a cost, and so some companies would rather take a more basic set of data feeds and compute their own | | | |
| I still don't understand xG - plus Delap and Enciso on 11:53 - Jan 23 with 2507 views | GavTWTD |
| I still don't understand xG - plus Delap and Enciso on 11:51 - Jan 23 by bluesbrothers |
It analyses all historic shots and whether they were scored or not. Uses data to compare a shot to all that historic data to create a % chance of that shot being a goal, irrelevant of the quality of player who struck it.
For example, historically 77% of penalties have been scored and so the xG of a penalty is 0.77 - i.e. for every penalty that occurs in a match you'd expect 0.77 goals
So Delap "SHOULD" have scored 8.43 goals this season based on the quality of chance where he's taken a shot. However, he's only scored 8 and so has scored 0.43 goals less than he should have.
Some sources will show different xG to others as they might have a different depth of historic data, or be getting the original data from a different source with slight discrepencies.
There's a book called "The xG philosophy" which is good. |
Right, got it.
So xG is a copyrightable stat? | |
| |
Login to get fewer ads
| I still don't understand xG - plus Delap and Enciso on 11:55 - Jan 23 with 2459 views | JakeITFC |
Think of each shot that a footballer takes (and think about position on the pitch, which part of the body they are using, defensive pressure, goalkeeper position etc.) and then imagine every shot that was taken from that position with similar circumstances previously. The xG (expected goals) is taken from assessing how many of those previous chances ended in a goal.
Generally deviation from the xG model on a long-term average basis for a team can decide whether teams are good/bad but lucky/unlucky. For a player it can show which players are on a hot streak or couldn't buy a goal. Drawing too much of an assertion on either of those things is risky though - really good finishers are able to just outperform their expected goals total (which remember is using the average of all footballers) because they are good. I'd say in the case of both Enciso and Delap it is probably too early to tell, but they are both capable of scoring great goals from very low xG positions. |  | | |
| I still don't understand xG - plus Delap and Enciso on 11:56 - Jan 23 with 2447 views | SaleAway |
I think different sites use different calculations for the xG, which makes it all a bit of a lottery, so yeah, circumstances do make it a bit variable to be a massively valid stat.... I remember in our league one promotion, Plymouth were way above their xG.... which may be why they struggled more once they came up....
As I understand it - xG is the goals he would have been expected to score from the shots that he has taken....so he would expect to have 8.43 goals.... he's scored 8, so he is below his xG by 0.43.... this is from 37 shots.... so for each of his shots average xG is 0.23, so he'd expect to score 1 in 4 ( ish).. he's at 8 from 37 so scores 1 goal in every 4.65 shots...
Enciso in comparison, has had 20 shots with an average xG per shot of 0.06.... so he's basically taking pot-shots.... would expect to score 1 in 17, but has yet to score....
Would be more interesting to compare Enciso to someone like Philogene, or Hutchinson or Broadhead, who are probably playing more similar roles to him.... | |
| |
| I still don't understand xG - plus Delap and Enciso on 11:56 - Jan 23 with 2439 views | bluesbrothers |
| I still don't understand xG - plus Delap and Enciso on 11:53 - Jan 23 by GavTWTD |
Right, got it.
So xG is a copyrightable stat? |
Not quite that simple. Different people may include different variables to try and get what they think is the best xG - there is no black and white definition of what should or shouldn't be included in that calculation.
It's also slightly flawed, hence the uprising of other "x" stats, in that if the ball came to me and i had an open goal from 1 yard out but did an airshot and completely missed the ball that would have an xG of 0 as it requires a shot | | | |
| I still don't understand xG - plus Delap and Enciso on 12:13 - Jan 23 with 2241 views | bsw72 |
The main problem with xG is that while it is built around statistical analysis and modelling, there is still no way to factor in all variables in a match, such as player skill, position, tactical decisions, or situational context (such as pressure moments).
Additionally, there are differing models which produce slightly different xG values based on the data and methods used.
In other words there is no single xG.
I think the best way to sum it up is that an xG model could potentially assign the same expected goals (xG) value to a shot taken by an U12 youth player and a full international player if both shots are taken from the same position and under similar conditions (e.g., distance from the goal, angle, type of shot, and pressure from defenders).
So in my mind, it's fundamentally flawed. |  | | |
| I still don't understand xG - plus Delap and Enciso on 12:42 - Jan 23 with 2048 views | TheBlueGnu |
This "XG" stuff is a load of old XXXX(G)
The only statistic that's of interest to me is goals and points - don't give a XXXX(G) how they come about.
Also, I like Glynis Barber. | |
| |
| I still don't understand xG - plus Delap and Enciso on 12:47 - Jan 23 with 1994 views | NthQldITFC |
| I still don't understand xG - plus Delap and Enciso on 12:13 - Jan 23 by bsw72 |
The main problem with xG is that while it is built around statistical analysis and modelling, there is still no way to factor in all variables in a match, such as player skill, position, tactical decisions, or situational context (such as pressure moments).
Additionally, there are differing models which produce slightly different xG values based on the data and methods used.
In other words there is no single xG.
I think the best way to sum it up is that an xG model could potentially assign the same expected goals (xG) value to a shot taken by an U12 youth player and a full international player if both shots are taken from the same position and under similar conditions (e.g., distance from the goal, angle, type of shot, and pressure from defenders).
So in my mind, it's fundamentally flawed. |
But in a way that's the point of it. If the dataset says a set of ten chances has a combined xG of 3.0 and Paul Mariner scores 6 of those 10 chances, whilst Gavin Holt scores 1 off his arse, it tells you that Paul Mariner has an xGDiff of +3 and Gavin Holt has an xGDiff of -2 and that one is an ace and the other a donkey. |  |
| |
| I still don't understand xG - plus Delap and Enciso on 12:50 - Jan 23 with 1973 views | Basuco |
It seems a pointless statistic IMO, during a Burnley game on TV they had an XG of 0, then at the end of the first half they produced a wonderful passing move that cut through the opposition midfield and defence to score an easy tap in goal. They were attempting similar moves from the start but the other team defended well and closed down the Burnley attacks. So while the XG was zero you could clearly see one mistake would result in a goal, which it did. So what is the point of the XG stat? |  | | |
| I still don't understand xG - plus Delap and Enciso on 13:37 - Jan 23 with 1783 views | bsw72 |
| I still don't understand xG - plus Delap and Enciso on 12:47 - Jan 23 by NthQldITFC |
But in a way that's the point of it. If the dataset says a set of ten chances has a combined xG of 3.0 and Paul Mariner scores 6 of those 10 chances, whilst Gavin Holt scores 1 off his arse, it tells you that Paul Mariner has an xGDiff of +3 and Gavin Holt has an xGDiff of -2 and that one is an ace and the other a donkey. |
Chuckle, good example and I agree, but there are too many unknown variables relating to player abilty which makes it an ineffective mechanism. | | | |
| I still don't understand xG - plus Delap and Enciso on 13:46 - Jan 23 with 1738 views | Cotty |
| I still don't understand xG - plus Delap and Enciso on 13:37 - Jan 23 by bsw72 |
Chuckle, good example and I agree, but there are too many unknown variables relating to player abilty which makes it an ineffective mechanism. |
What? It's a measure of player ability. | | | |
| Reading a great book on this, XG explainer - How to Win Premier League? on 13:59 - Jan 23 with 1667 views | unstableblue |
From the Liverpool stats guru - who identified all those good players, managers
He's very good - interestingly he sights Brentford and Brighton as the only clubs close to Liverpool's model, not Man U, or Chelsea.
Need to pinch some of these methods.
Simplistically he calls XG - "Weighted Shots" - which is a much better explanation, and brings out the flaws in the metric.
So XG is "Expected goals (xG) measures the quality of a shot based on several variables such as assist type, shot angle and DISTANCE FROM GOAL, whether it was a headed shot and whether it was defined as a big chance" (another source)
What are the flaws of xG?
"Its reliability diminishes significantly when used for single games or individual shots, where context and variance play a much larger role. it often shows unreliable Interpretation with small data. " (another source)
Penalty high XG, shots from edge of box despite being a blaster and on target and really testing keeper do not all get 'clear cut chance' in the stats, hence low XG. We had 4 pretty good efforts first half against City - agree some shoudl be low XG, maybe two higher, assume only Omaris which was flicked over got high XG.
So Gav your own XG would perhaps be more sophisticated in rating a shot for likelihood of a goal, say difficulty of save a shot produced for keeper by some some scale - but this would be too hard to do quickly. So XG is effectively done on all the shots from the D for example how may go in, not the specific shot in question.
That's not the major theme of the book - its brilliant to read the recruitment, players no one in Europe was up for - Matip a good example - he was failing all the measure of other teams. But they measured in a different way, picked him up, got some very important performances out of him. |  |
| |
| I still don't understand xG - plus Delap and Enciso on 14:13 - Jan 23 with 1594 views | ibbleobble |
It's a size above XL. | | | |
| I still don't understand xG - plus Delap and Enciso on 14:21 - Jan 23 with 1548 views | bsw72 |
| I still don't understand xG - plus Delap and Enciso on 13:46 - Jan 23 by Cotty |
What? It's a measure of player ability. |
Not necessarily - Some advanced xG models may incorporate player-specific data, such as historical shooting accuracy or finishing ability, which could lead to different xG values for different players in similar situations. However, this is not standard in all xG calculations. | | | |
| I still don't understand xG - plus Delap and Enciso on 14:43 - Jan 23 with 1446 views | PrrrromotionGiven |
| I still don't understand xG - plus Delap and Enciso on 12:13 - Jan 23 by bsw72 |
The main problem with xG is that while it is built around statistical analysis and modelling, there is still no way to factor in all variables in a match, such as player skill, position, tactical decisions, or situational context (such as pressure moments).
Additionally, there are differing models which produce slightly different xG values based on the data and methods used.
In other words there is no single xG.
I think the best way to sum it up is that an xG model could potentially assign the same expected goals (xG) value to a shot taken by an U12 youth player and a full international player if both shots are taken from the same position and under similar conditions (e.g., distance from the goal, angle, type of shot, and pressure from defenders).
So in my mind, it's fundamentally flawed. |
One example in particular can display how poor xG is at estimating things like game state and context.
Imagine a player takes a penalty, which the keeper saves, but can only push it back to the penalty taker, who now has a tap-in. Think Harry Kane vs. Denmark.
The penalty has xG of about 0.75, and the follow up shot has xG of about 0.75 as well for the sake of argument.
In these few seconds, according to xG, the striker is expected to score 1.5 goals, and they are seen to have underperformed significantly if they only score 1. This is despite the fact it was obviously impossible for them to score more than 1 goal from these chances!
It also can make some upsets look more ridiculous than they really were. Imagine a team takes a shock 1-0 lead early in the first half, and holds on to win 1-0. They probably have a pathetically low xG of 0.1 or something across the whole game, but that's only because they had no need to attack after going 1-0 up. If instead they scored their winner in injury time, maybe they would have an xG closer to 1 despite playing no better (arguably worse, having come significantly closer to a draw than in the first case) when you consider game state.
So yeah, it's typical Big Data in that it is typically bad for looking at individual shots or even matches, but generally proves itself correct over dozens of games | | | |
| I still don't understand xG - plus Delap and Enciso on 14:47 - Jan 23 with 1436 views | bluesbrothers |
| I still don't understand xG - plus Delap and Enciso on 14:43 - Jan 23 by PrrrromotionGiven |
One example in particular can display how poor xG is at estimating things like game state and context.
Imagine a player takes a penalty, which the keeper saves, but can only push it back to the penalty taker, who now has a tap-in. Think Harry Kane vs. Denmark.
The penalty has xG of about 0.75, and the follow up shot has xG of about 0.75 as well for the sake of argument.
In these few seconds, according to xG, the striker is expected to score 1.5 goals, and they are seen to have underperformed significantly if they only score 1. This is despite the fact it was obviously impossible for them to score more than 1 goal from these chances!
It also can make some upsets look more ridiculous than they really were. Imagine a team takes a shock 1-0 lead early in the first half, and holds on to win 1-0. They probably have a pathetically low xG of 0.1 or something across the whole game, but that's only because they had no need to attack after going 1-0 up. If instead they scored their winner in injury time, maybe they would have an xG closer to 1 despite playing no better (arguably worse, having come significantly closer to a draw than in the first case) when you consider game state.
So yeah, it's typical Big Data in that it is typically bad for looking at individual shots or even matches, but generally proves itself correct over dozens of games |
Most models (the good ones) account for this | | | |
| I still don't understand xG - plus Delap and Enciso on 14:48 - Jan 23 with 1434 views | NthQldITFC |
| I still don't understand xG - plus Delap and Enciso on 14:21 - Jan 23 by bsw72 |
Not necessarily - Some advanced xG models may incorporate player-specific data, such as historical shooting accuracy or finishing ability, which could lead to different xG values for different players in similar situations. However, this is not standard in all xG calculations. |
Ah, I didn't know that. Trying to recalibrate a measurement system for each candidate being measured has a specific purpose I suppose, but not the same purpose as a calibrated general measurement system. Creating oranges from apples in that case. | |
| |
| I still don't understand xG - plus Delap and Enciso on 14:50 - Jan 23 with 1417 views | Marshalls_Mullet |
Based on the quality of chances hes had, he would be expected to score 8.43 goals.
He has scored 8 goals.
8.00 - 8.43 = -0.43.
So his conversion rate is slightly below what would be, based on the quality of the chances that have been created. [Post edited 23 Jan 14:51]
|  |
| |
| I still don't understand xG - plus Delap and Enciso on 14:52 - Jan 23 with 1385 views | Marshalls_Mullet |
I think these stats were basically invented by Tony Bloom to assess football / footballers, and adopted by the rest of the betting industry. | |
| |
| |
By continuing to use the site, you agree to our use of cookies and to abide by our Terms and Conditions.
