XG

[egg]

1 hour ago, Fabrice Fernandes no.1 fan said:

It is also useful for measuring a striker's ability. If a striker is scoring above their expected goals, you can infer they are likely a better finisher, or more in form, than someone who is scoring below their expected goals.

It is less useful when applied to specific circumstances, and better over a wider spread of data. 

However, it does represent Wednesday quite well, our two chances weren't the easiest, and so you would not expect us to score that often. Djenepo misses that chance more often than he scores it, so it has low XG. 

Why do we need stats to tell us that Che Adams can't finish, but that Harry Kane? I'm genuinely confused why anyone needs xG when you could have a stat that says, for example, Che had 5 shots and all were off target. It's still bollox. 

[Weston Super Saint]

27 minutes ago, Saint86 said:

🤦‍♂️ Because yes, if we replayed the chances, then every single time haaland jumps for that header he will miss the goal. And every time Alvarez hits that cross he'll slice it wide 

Sadly, in life, there are no 'do overs'.  The chances don't get to be 'replayed' 

[Weston Super Saint]

1 hour ago, notnowcato said:

You appear to be confusing outcome with probability of outcome.

Any idea why 'Wyscout' (no idea who they are but they seem to think they are the dogs danglies when it comes to xG / xX), give an xG of 0 for all 'wide shots' - i.e. ALL SHOTS NOT ON TARGET!

https://dataglossary.wyscout.com/xg/#:~:text=A penalty xG value is fixed to 0.76.&text=At the moment there are,an xG value of > 1.

Quote

all blocked shots and shots wide automatically have a post-shot xG value of 0

It's almost as if they agree about the probability and outcome being 0  

Also, why are penalties given a fixed xG of 0.76 - does xG not learn based on previous outcomes and haven't we been told it's based on the outcomes of 'average' players?

[Daft Kerplunk]

The classified XGs are in:

Tottingham Hotspuds 0.79 Crystal Vases 0.92

Portsmuff 0.12 Grimbsby Frown 1.36

Sherwood Forest 0.66 Sheffield Thursday 0.81

Ignoramus Academicals 2.04 Caledonian Arse Thistles 1.37

[Whitey Grandad]

4 hours ago, notnowcato said:

No doubt some football data is still recorded by humans but the advances in video analysis are slowly eradicating the human element in data collation. 
 

You’re potentially showing a bias in your interpretation of an event. Maybe you’re seeing the event as either a 0 or 1 outcome.

This is part of the problem. Any video recording is only ever a poorer version of what is seen by an experienced human with their own eyes. It is lower resolution and is only 2D.

I should add here that I spent some time in television equipment research involving studio signal processing and was a UK representative on a European Broadcasting Union committee determining future digital studio standards. I have had technical papers published in relevant publications and presented at international conferences. What you think you see is not what actually happened.

[Convict Colony]

My XG of this thread is:

Nerds XG 1.23 v 2.1 Non Nerds

 

[Whitey Grandad]

2 hours ago, Fabrice Fernandes no.1 fan said:

It is a good metric for the quality of chance a team is producing against the end result, over a spread of games. For example, a team that has 3.5xg per game across half a a season but is only scoring 1 goal a game is making chances but not finishing them, but a team who have 0.5xg per game but score a goal a game are not making enough chances but are finishing those they do have well.

It is also useful for measuring a striker's ability. If a striker is scoring above their expected goals, you can infer they are likely a better finisher, or more in form, than someone who is scoring below their expected goals.

It is less useful when applied to specific circumstances, and better over a wider spread of data. 

However, it does represent Wednesday quite well, our two chances weren't the easiest, and so you would not expect us to score that often. Djenepo misses that chance more often than he scores it, so it has low XG. 

Thanks both of you for your considered replies. In your first paragraph I would argue that the xGs are not accurately computed. If a striker is scoring more than their expected goals would predict then that tells me that their predictions are wrong. As for Wednesday, if we hadn’t scored those two goals then we would have created other opportunities and these of course would not feature in the final evaluation.

1 hour ago, Saint86 said:

Several examples have been set out above. Put simply, xG adds a nuanced way to interpret team and player performances, such as looking at the effectiveness of a team's attacking play, a player's finishing, or the effectiveness of a tactic / defence at nullifying the attack of an opposition team. For example, we had minimal possession in the city game, but defended well and limited city's entire attack to less expected goals than haaland's average xG for a game this season 👍. That's a significant reduction.

We were then able to use the circa 25-30% possession we had efficiently to generate some decent chances and  had mara (who statistically is a decent finisher) to take advantage of it. It was a good performance where tactically we were very solid and maintained a threat, but the stats show that city still had the chances and should really have scored - e.g. on another night city start haaland, and the Alvarez chance is a goal (as an example), or haaland hasn't just come on to somehow completely miss his header in front of goal. 

If you'd rather count all shots / shooting chances as equal then ofc fee free 😢, but there is definetly value in the newer statistical analysis methods. 

Similarly we should be targeting attacker's that have a high xG conversion rate. 

I don’t know anyone who watched that game who didn’t think that Saints were in any way lucky to have won it or that we were in any way let off somehow by City’s poor finishing. That Alvarez incident was some way off being a serious chance, the ball was played too far in front of him.I must admit that I don’t remember Haaland ever having a sniff.

[SaintlyAnd]

1 hour ago, Weston Super Saint said:

Any idea why 'Wyscout' (no idea who they are but they seem to think they are the dogs danglies when it comes to xG / xX), give an xG of 0 for all 'wide shots' - i.e. ALL SHOTS NOT ON TARGET!

https://dataglossary.wyscout.com/xg/#:~:text=A penalty xG value is fixed to 0.76.&text=At the moment there are,an xG value of > 1.

It's almost as if they agree about the probability and outcome being 0  

Also, why are penalties given a fixed xG of 0.76 - does xG not learn based on previous outcomes and haven't we been told it's based on the outcomes of 'average' players?

Post-Shot xG and xG are different. Post-shot has additional information from after the shot about how likely it was to have been a goal. Given a shot went wide, we know it cannot be a goal, so value is 0. 
In my opinion post-shot xG is a very weird metric. After the shot we know whether it’s a goal or not so it should be black or white, 0 or 1, so pointless.

 

The penalty value is based on the wyscout model for calculating xG. I assume over historic data there’s a 76% chance of scoring a penalty. Other people calculating xG would use different models and get different values. And no models for calculating it are very good because of so many variables that aren’t in the data.

Personally I don’t like using xG in individual games and used in the media or broadcasters to try to tell something insightful about a game. The limitations just can’t add anything more than (or even close to) someone watching the game.

The value from my experience is for clubs in player/match analysis over a series of games and in player recruitment to start identifying players over at least a season. But it’s a tool to help as part of wider processes. Clubs use data alongside video a lot and use tools like wyscout, instat, match tracker and various others to align data/video analysis.

[Turkish]

1 hour ago, Convict Colony said:

My XG of this thread is:

Nerds XG 1.23 v 2.1 Non Nerds

 

Are you suggesting that non nerds are more nerdy than nerds?

[Turkish]

Does anyone just have a few beers and go to a game anymore or do people sit in the stands analysing data, stats, tactics, formations etc? Hope i never sit near most of the posters on this thread next time i'm at a game.

[LGTL]

XG was created by nerds in the hope that it would finally give them a pathway into the national game. Think of all those kids on the playground who had two left feet and never got picked to be on the good team. They all went off, did Statistics at university and created this nonsense to try and ruin it for the rest of us. 

[Give it to Ron]

14 minutes ago, Turkish said:

Does anyone just have a few beers and go to a game anymore or do people sit in the stands analysing data, stats, tactics, formations etc? Hope i never sit near most of the posters on this thread next time i'm at a game.

Only us old men,  rest are buying their 50/50 scarves  and trying to work out the XG on how shit Che is in front of goal from a yard.

My long range specs due to being a decrepid old git that played decent level and coached teams to league wins and cups  can’t do without stats to watch a game and it’s only the young that can understand football 

Edited 13 January, 2023 by Give it to Ron

[SaintlyAnd]

1 hour ago, Turkish said:

Does anyone just have a few beers and go to a game anymore or do people sit in the stands analysing data, stats, tactics, formations etc? Hope i never sit near most of the posters on this thread next time i'm at a game.

When I started working as a professional football data nerd (not my real job title but for the sake of the thread it sums it up) the enjoyment of watching a game disappeared because I kept seeing everything in a game from an analytics perspective. I think I stopped watching football for about 6 months. Now I manage to switch off the job when I just want to watch what is just a game and enjoy it. I think as a fan the stats can take away the enjoyment of supporting a team, having opinions about who played well etc. I got quite grumpy in the World Cup when some random metrics were flashed up in game because I felt there was no need, no context, told me nothing anyone watching the game couldn’t see and reminded me of work.

[verlaine1979]

Stats like xG are useful for weighing up the strengths of potential new players and future opponents. But I suspect for most fans, they aren't that revealing for analysing your own team, especially not compared to the accumulated evidence of watching games. Adams and Kane being excellent cases in point - the gulf in their xG>Goals performance isn't going to shock anyone who has regularly watched either player over multiple seasons.

But for scouting purposes I'm all for it. Especially if it helps us keep away from dire finishers like Adams and Armstrong in future.

[notnowcato]

13 hours ago, Weston Super Saint said:

You're a weirdo!

If a team has zero shots on target, the probability of scoring is zero.

If a team has zero shots on target, the outcome will be zero goals.

There is no confusion when the answer is zero both times.

Charming. Enjoy your binary life. 

[Miltonaggro]

10 hours ago, Turkish said:

Does anyone just have a few beers and go to a game anymore or do people sit in the stands analysing data, stats, tactics, formations etc? Hope i never sit near most of the posters on this thread next time i'm at a game.

It’s this type of prejudice and short sightedness that is ruining the 2023 matchday experience. 

[Weston Super Saint]

11 hours ago, SaintlyAnd said:

Post-Shot xG and xG are different. Post-shot has additional information from after the shot about how likely it was to have been a goal. Given a shot went wide, we know it cannot be a goal, so value is 0. 
In my opinion post-shot xG is a very weird metric. After the shot we know whether it’s a goal or not so it should be black or white, 0 or 1, so pointless.

 

The penalty value is based on the wyscout model for calculating xG. I assume over historic data there’s a 76% chance of scoring a penalty. Other people calculating xG would use different models and get different values. And no models for calculating it are very good because of so many variables that aren’t in the data.

Personally I don’t like using xG in individual games and used in the media or broadcasters to try to tell something insightful about a game. The limitations just can’t add anything more than (or even close to) someone watching the game.

The value from my experience is for clubs in player/match analysis over a series of games and in player recruitment to start identifying players over at least a season. But it’s a tool to help as part of wider processes. Clubs use data alongside video a lot and use tools like wyscout, instat, match tracker and various others to align data/video analysis.

From the Wyscout link posted it states the following :

Quote

For every shot, the xG model calculates the probability to score based on event parameters:

• Location of the shot
• Location of the assist
• Foot or head
• Assist type
• Was there a dribble of a field player or a goalkeeper immediately before the shot?
• Is it coming from a set piece?
• Was the shot a counterattack or did it happen in a transition?
• Tagger's assessment of the danger of the shot

Therefore a full set of criteria are measured to calculate the xG, many of which have completely binary answers.

However, the part I've highlighted seems very odd.  I'm working on the assumption that the word 'Tagger' is some awful word used in place of what should probably be called the 'assessor' (I thought a 'tagger' was someone who tags local landmarks with their spray cans, ala Banksy?).  Given that assumption (please feel free to correct me if I'm wrong), then the 'Tagger' is potentially a different person each time the xG is calculated.  Given that every person on the planet has a different perception of risk and indeed 'danger', then this final point in the analysis tool is completely subjective.  No idea what weighting is given to 'Tagger's' assessment of the danger but it seems very much 'stick your finger in the air and come up with a figure' to me.

And some people think this is 'good science' 

[Whitey Grandad]

Does anybody know at what stage of an attack the xG calculation is made? Some goal attempts might follow an intricate passing move involving multiple players before the ball arrives at the feet of an attacker. Others might be the result of a long hoof downfield from the goalkeeper. Surely xG cannot be estimated solely on how it leaves the boot of the potential scorer? That would be ridiculous.

How are the possibilities of deflections evaluated? What was the xG when Perraud got his feet confused and scored his own goal?

If football coaches think that it’s a useful metric then that’s up to them but keep it away from the football public please. As for those figures that kept popping up on screen in the World Cup….  Why?

[Hodgey]

For me it’s really simple - I can tell that Che Adams is a poor finisher as he cannot use his left foot, he also tends to just hit the ball as hard as he can with his right. Whatever his xG are according to stats, I know this by watching the game, so whilst his poor xG stat could tell me he isn’t a great finisher, by watching the games I know that anyway, and more I know why.

[Weston Super Saint]

1 hour ago, Whitey Grandad said:

Does anybody know at what stage of an attack the xG calculation is made?

Looking at the bullet points from Wyscout it would appear to only be calculated on the 'assist' and 'shot', but includes if a 'dribble' was also carried out, but I guess only if that dribble immediately preceded the pass that counts as the assist.....

[Whitey Grandad]

1 hour ago, Weston Super Saint said:

Looking at the bullet points from Wyscout it would appear to only be calculated on the 'assist' and 'shot', but includes if a 'dribble' was also carried out, but I guess only if that dribble immediately preceded the pass that counts as the assist.....

That’s what I more or less assumed. So it means that xG after a game is a very limited statistic and only has a place for those who specialise in coaching. As far as a comment on the game is concerned it should be nowhere near the media reports of MOTD.

My annoyance is that anyone who didn’t know how the game went on Wednesday and looked at the xG from the game would rightly assume that Saints were lucky to win.

[SaintlyAnd]

A 'tagger' watches a game and inputs every event that occurs in real-time. Wyscout have a list of event types, like: Pass, Duel, Clearance, Ball Touch, Shot etc, and those events get attributes like the player, the position on the pitch and whether it was free-kick, throw-in, corner, cross etc, so all that needs to be 'tagged'. Each event provider like Opta, Statsbomb, Wyscout have their own definitions and classifications. Taggers are trained but generally low paid, and they have to identify every player immediately and get through games pausing them constantly. It's very prone to human error, and if they're using a video from a tv broadcast then there are gaps in their data (eg. when the camera focuses on a single player or manager or a replay while the game continues). For top leagues where people want fast data, you'll have multiple taggers working a game for different players. The xG value is calculated based on the information available at the time of the shot event, which can include data from the previous event/s.

Their input into the xG calculation is, as you say, really subjective. But I assume it's an attempt to quantify what a human can see and understand, and isn't in the tagged data. If we take the Haaland 'shot' the other night where he flicked the ball with his foot at head height from behind him and it loop over, and if we pause the game at the moment the ball touches his foot (when the 'shot' event happens and xG is calculated from). The data tells us: 1. the position was central and about 10 yards out (this is where lots of goals come from), 2. the shot was with his foot (more goals come from shots than headers, so gets a good score), 3. the assist was from a cross, 4. it was in open play.

All that makes it sound like a great chance, so would get a good xG value. But it doesn't factor in: the ball was behind him, the height of the ball, he was side-on or facing back to the goal, how fast the ball was coming to him, was it deflected before it got to him, has he seen the ball coming clearly, how many opponents are there between him and goal, is the goalkeeper set to try to save it (or is he laying down after just having made a save), how much pressure he's under from defenders, etc. I'm sure we can all think of more details that go into how good we perceive a chance to be. I think most of us would think that Haaland chance is very unlikely to go in, and therefore the xG should be really low. So, like I say, I guess that human input into the xG calculation is to factor in everything else the data can't (which is a lot of things). Just having looked at the xG score from wyscout for that chance - it has calculated 18% of the time that should end in a goal. My assessment would be about 1 or 2%.

Not that I'm defending it as useful. Just trying to point out the limitations and, like I said above, I don't think it's meaningful in the context of individual chances or games.

Out of interest, I just looked at Adam Armstrong's xG for this season in the PL. From one source, I found an xG of 1.1 goals and another 2.1. I think that highlights how it being used in isolation is meaningless and inconsistent calculations are. Whereas if I search for strikers across European leagues, aged less than 27, who have scored over 5 goals so far this season (using a single data provider), and get their xG alongside stats like shots per 90 minutes, % shots on target, number shots blocked (and more stats), then I can start to compare players and see who may be over or under-performing. And from there, start looking at videos and real games to prove or disprove the suggestions from the data, which is probably a very simplified version of what Southampton do with the data-driven recruitment approach.

(Apologies for the long reply)

[The Kraken]

1 hour ago, Hodgey said:

For me it’s really simple - I can tell that Che Adams is a poor finisher as he cannot use his left foot, he also tends to just hit the ball as hard as he can with his right. Whatever his xG are according to stats, I know this by watching the game, so whilst his poor xG stat could tell me he isn’t a great finisher, by watching the games I know that anyway, and more I know why.

Which is why XG starts to become useful when you’re not able to watch the game. If you’re watching a game you’re able to get a feel for it. If not, XG is as useful a stat as shots on target or possession, no more no less than that really.

[egg]

4 minutes ago, The Kraken said:

Which is why XG starts to become useful when you’re not able to watch the game. If you’re watching a game you’re able to get a feel for it. If not, XG is as useful a stat as shots on target or possession, no more no less than that really.

There's the myth about xG. It's not a statistic. It's a number based on someone or something's opinion. Shots on target, possession, passes, tackles, etc, are based on reality. 

[The Kraken]

16 minutes ago, egg said:

There's the myth about xG. It's not a statistic. It's a number based on someone or something's opinion. Shots on target, possession, passes, tackles, etc, are based on reality. 

Call it what you want, it’s just another number that might be useful in quickly summarising a game if you didn’t get to see it.

[Weston Super Saint]

15 minutes ago, egg said:

There's the myth about xG. It's not a statistic. It's a number based on someone or something's opinion. Shots on target, possession, passes, tackles, etc, are based on reality. 

Therefore, as we've all pointed out, it's made up bullshit to make the 'data' try and fit the 'facts' 😉

For the record, I'm not angry or upset about that!  It seems there's an entire industry that's been created which is probably superfluous for the overwhelming majority of fans, but if it makes others happy to knock one out over the figures in the privacy of their own bedroom, I'm very happy for them  

[egg]

22 minutes ago, Weston Super Saint said:

Therefore, as we've all pointed out, it's made up bullshit to make the 'data' try and fit the 'facts' 😉

For the record, I'm not angry or upset about that!  It seems there's an entire industry that's been created which is probably superfluous for the overwhelming majority of fans, but if it makes others happy to knock one out over the figures in the privacy of their own bedroom, I'm very happy for them  

Indeed. Take actual stats, then apply an opinion based calculation, you get opinion. Each to their own, but I'm not persuaded that computer generated opinion is better than simple stats plus what's visible. 

[Holmes_and_Watson]

On 14/01/2023 at 12:20, Whitey Grandad said:

That’s what I more or less assumed. So it means that xG after a game is a very limited statistic and only has a place for those who specialise in coaching. As far as a comment on the game is concerned it should be nowhere near the media reports of MOTD.

It is a limited stat. The "very" varying over time as depth is added. The use of taggers and the gaps outlined in SaintlyAnd's posts show just how limited it is.

And I'd agree it does have a place when used by those specialisng in coaching. Even using xG in isolation is to ignore all of the other stats being generated now.

All of which, across a period of time (season/ seasons) add to the information available to the coaching staff specialising in it.

But they do have a place. From simple things like determining that your forward keeps thumping it from a position unlikely to score, or that you concede from a lot of corners aimed at a certain area in the box, to using the data to develop styles of play to maximise your chances of using attributes that suit your team.

Football hackers mentioned a German ex-player/journalist who trotted out “packing” as the new big thing. He went on about it so much, that it set the whole thing back.

And we see that with xG and its numerous fellow figures. Trotted out to fill a graphic, and some air time by commentators and hacks. We point out their useless analysis on everything else, so it’s no surprise this wouldn’t be used well either.

But there’s no doubt while the press were going on about Potter’s chances at Brighton, the analysis that they were doing was telling them just how much they were getting right/ to their plans. That helped, not all by itself, lead to long term consistency and a sustainable Premier League team (along with lots of other metrics they also work away at. Because they know it’s not just about one thing, even if some broadcasters/ bandwagon jumpers clearly don’t.)