Xg - is it a load of bo***x

[dc_blade]

It depends on who or what decides on the calibre of chances.
It’s subjective in my opinion and therefore of little more use than all our own views.

They use samples of thousands of game incidents to determine how often a player could be expected to score from a given chance. So it’s not really subjective at all.

 

[YorkBlade]

is anyone else on this forum interested in Baseball sabermetrics? I find it fascinating and it has given me a greater appreciation of football stats and analytics.

 

[Champagneblade]

I don't know why it hasn't been developed to a further level.

It is useful in so much that you weigh up average player scoring from a certain location.

But then someone firing in from a standing ball and someone having to connect from a moving ball at great pace isn't going to have the some conversion rate and I'm not clear this gets taken into account.

Likewise the shooter. If you're constantly playing in your centre forward or the same chance is falling to your full back, they won't have the same likely conversion. In terms of analyzing quality of chances that is fine but not in terms of analyzing the potential outcome i.e. if you got 3 one v ones in a game identically falling to Baldock then the same three falling to Sharp, the likely outcome would be Sharp would score more. There is an expectation on the player it falls to where "I'd expect Billy to bury that one" you might say but you are less likely to attribute the same expectation on a defender. I think an alternative to xG, XGp, based on who the chances are falling to would be useful.

I'd imagine this sort of algorithm of chance conversion multiplied by probability of player conversation is already factored into many football management games already. In that sense it's nothing new. It's probably how some of the Championship manager match simulations work and they have been around for way before xG became a thing.

 

[alcoblade]

We are 5th in the OVERALL xG table with 1.51 and 4th in the HOME xG table with 1.74.

We've played 20 games so 20 x 1.51 should be 30 goals scored. We've actually scored 25. In home games we should have scored 17 goals and have only scored 15. This means we're not converting the clear cut chances as much as we should have.

xGA (Expected Goals Against) - United are 20th (this is good) with an xGA of 1.13 which suggests we should have conceded 22/23 goals. We've conceded 26.

Simply put we're not taking the chances upfront and are conceding when we shouldn't be. We're 5th in xG and 5th best (20th) in xGA yet we're down in 13th position.

Interestingly, Bournemouth's overall xG is 1.36 which suggests they should have scored 27 goals so far. They've actually scored 36 which means they're outperforming.

Here's the xG table:

View attachment 125592

The discrepancy between expected goals and our actual goals is why bookies have us as favourites in most matches. The statistics suggest you will tend towards your expected goal position. That was explained to me by a successful professional gambler.

 

[Scarebus]

The discrepancy between expected goals and our actual goals is why bookies have us as favourites in most matches. The statistics suggest you will tend towards your expected goal position. That was explained to me by a successful professional gambler.

As a fan, I think the best way to use xG, and other expected metrics is over the season rather than for an individual match. Hugely beneficial information for clubs, especially recruitment, for plugging deficiencies, and being used to great effect.

Nearly all of the football metric twitter accounts I have followed over the last few years have been employed by clubs, and their content has really dropped off.

 

[raul]

Dear OP,

Beans is going to kick your head in.

 

[Revolution]

is anyone else on this forum interested in Baseball sabermetrics? I find it fascinating and it has given me a greater appreciation of football stats and analytics.

Yes. Bill James is one of my favourite writers. Learned so much from him and people like him in the baseball field.

Having said that, the state of sabermetrics is such that it bears some responsibility for baseball becoming a poor on field product aesthetically. Long games, too many pitching changes, everyone trying to hit home runs. Not fun.

 

[Stu_Blade]

It's a typical American style stats format that doesn't need to be there. Ultimately doesn't matter what your expected goal rate is its what your actual goal rate and against is. Are the ones that counts, keep one higher than the other and you win the other way round you lose. Its trying to turn the sport in baseball which has so many stats it goes into 0.20 etc

 

[Snowflake]

They use samples of thousands of game incidents to determine how often a player could be expected to score from a given chance. So it’s not really subjective at all.

So who decided that Brewster didn’t touch his chance ?

 

[gordoncharles]

It's highly, highly flawed as a method for determining what a result might have been.

A team which takes a lead, particularly an early lead, and especially away from home, is under no pressure or obligation to keep attacking.

The result of that is they can continue to be more defensive up till the time they may eventually concede. In the meantime, the side behind in the game may have wasted several presentable chances to score. Their xG is higher than it should be merely based on the state of the actual - rather than theoretical - game taking place.

Once they themselves do actually score, or take the lead, the actual requirement to attack changes again.

In other circumstances it can tell more of a story. For instance, yesterday, but that's only because Nigel Pearson's Bristol City were shite.

 

[SouthportBlade]

It's a typical American style stats format that doesn't need to be there. Ultimately doesn't matter what your expected goal rate is its what your actual goal rate and against is. Are the ones that counts, keep one higher than the other and you win the other way round you lose. Its trying to turn the sport in baseball which has so many stats it goes into 0.20 etc

Whilst it's that true xG count for fuck all, elite level sport is about marginal gains.

If through 'expected analytics', you can gain small increments in recruitment, tactics, game management etc then of course it matters.

Granted its not everyone's cup of Tea, and chucking out xG at Half Time on Sky Sports is frankly shite. There can also be a degree of smugness from some in the field, who think xG knowledge is the best and only type of insight.

It's like any form of statistics, on there own are meaningless. There are some purveyors of xG analysis who really know there stuff and also 'bring it to life' in articles / charts etc.

Dunno if he posts on here, but always though that Jay/BladeAnalytics was one of the good ones.

 

[Panini]

It's highly, highly flawed as a method for determining what a result might have been.

A team which takes a lead, particularly an early lead, and especially away from home, is under no pressure or obligation to keep attacking.

The result of that is they can continue to be more defensive up till the time they may eventually concede. In the meantime, the side behind in the game may have wasted several presentable chances to score. Their xG is higher than it should be merely based on the state of the actual - rather than theoretical - game taking place.

Once they themselves do actually score, or take the lead, the actual requirement to attack changes again.

In other circumstances it can tell more of a story. For instance, yesterday, but that's only because Nigel Pearson's Bristol City were shite.

No-one is suggesting xG is perfect, or even the number 1 measure.

I'm not sure how those scenarios undermine xG anyway, if a team is winning and chooses to sit back their xG won't be affecting as they're not shooting. Likewise a losing team might attack more, so they will register more shots.

The game state doesn't affect the xG of an attempt on goal. Anyway it's possible to break the total down to xG when winning, drawing or losing, if required.

xG is literally a measure comparing hypothetical with reality, so any number of "might have" situations doesn't skew the comparison.

 

[blade too long]

does this include Wednesday fans expected goals because before every derby match they expect to score 6 against us but are still waiting after 130 years

 

[dc_blade]

So who decided that Brewster didn’t touch his chance ?

Okay, I take your point: there will still be certain situations like that in which a fallible human has to make a judgment call. But by and large the method of comparing a given scoring opportunity to a large sample of other such opportunities removes some of the subjective element.

 

[beatific blade]

I was trying to explain it to my son last night and I noticed that the chance missed by Brewster early doors doesn't get assigned xG because he didn't connect with the ball. He was 2 yards out in the centre of the goal so I'm assuming it would have been similar to the 65% from Sharp's goal or probably even higher.

The in-house team at Brentford uses its own bespoke model for expected goals - and in terms of data-gathering they have teams of people who watch games and place manual values on chances such as Brewster’s, where normally it would get missed as you point out. They’re not stupid at Brentford, they can use xG to spot players that get in great goal scoring positions.

A newer model for calculating a team’s overall attacking play is xT (expected threat) which models the flow of the game and displays value on how much genuine danger a team poses.

 

[beatific blade]

Just looked on a couple of sites and our expecred goals for yesterdays game were 1.7 and 2.1, both equating to a 2-0 victory, as i watched the game i thought we should have had about 5, so is this xg a load of bo***x

My general response to these types of thread is… read this:


The Expected Goals Philosophy: A Game-Changing Way of Analysing Football
https://www.amazon.co.uk/dp/1089883188/ref=cm_sw_r_awdo_navT_a_6A86Z1N8WREF8B4T9ZN9

 

[blade too long]

Sheffield Wednesday players use XW , they expect 1 amount of wages , and bet on if they get it or not

 

[Kenilworth]

What's XG?

It’s a big kiss from a bloke called Gary

 

[raimond]

It’s just one more tool that has to be used correctly.

Not sure if your comment about spontaneity and improvisation suggests that you aren’t a fan of planning and preparing too. Are you?

Anything that gives an edge is worth taking into account.

That's a strange response I'd say mate, to suggest it means anything about me personally. Very organised cheers- that doesn't rely on me being over dependent on data analysis.

 

[raimond]

Elite sport is about fine margins, not big weaknesses.

xG isn't perfect but it's a very decent way of measuring how good a player is in those spontaneous moments you mention, which can then be looked at across a whole team.

It's hated because it proves people's gut instinct wrong. Not many people like being wrong and even fewer like to challenge their own instinct.

I think the name "expected" throws people off. It's nothing to do with prediction. A shot with an xG of 0.5 simply means if it happened 100 times you would "expect" to see a goal scored in 50.

Spotting a player who outperforms their xG is a huge part of scouting an attacking player. If they do it consistently, they're world class.

It can also be used for other purposes, ie if a player keeps shooting from low xG opportunities he shouldn't be shooting as often.

Possession was the new trendy stat in the 90s, but it doesn't mean everything. Shots on target is another indicator of how a game panned out. xG is merely another. Others you'll start to see more: PPDA - passes per defensive action. This is an indicator of how long a team allows their opponents to keep the ball, ie how many passes they allow to happen before trying to tackle or intercept. That can also indicate how high a team chooses to press their opponents.

Anyone who has ever said "he should have scored from there" or "we should've won 5-0" is interested in xG whether they realise it or not.

CSK just won the IPL largely moving away from data analysis. They still used it for targeting matchups, areas etc but tried to move away from their players over thinking into reacting. I think this will prove to be a recurring theme across sports as overthought tends to hinder performance.

 

[jt64]

It's highly, highly flawed as a method for determining what a result might have been.

Which is fine, as it's not intended to do that in the slightest

CSK just won the IPL largely moving away from data analysis. They still used it for targeting matchups, areas etc but tried to move away from their players over thinking into reacting. I think this will prove to be a recurring theme across sports as overthought tends to hinder performance.

I give zero fucks about T20 cricket but it would be hilarious if this was just down to variance

 

[Snowflake]

Okay, I take your point: there will still be certain situations like that in which a fallible human has to make a judgment call. But by and large the method of comparing a given scoring opportunity to a large sample of other such opportunities removes some of the subjective element.

Thanks. I understand it much better now.

 

[FallowfieldBlade]

Just looked on a couple of sites and our expecred goals for yesterdays game were 1.7 and 2.1, both equating to a 2-0 victory, as i watched the game i thought we should have had about 5, so is this xg a load of bo***x

"Just looked on a couple of sites and our expected goals for yesterdays game were 1.7 and 2.1, both equating to a 2-0 victory"

So basically XG was bang on with its prediction

 

[raimond]

Which is fine, as it's not intended to do that in the slightest



I give zero fucks about T20 cricket but it would be hilarious if this was just down to variance

Oh. Whether you give a fuck or not, no problem. I just think it's probably significant that football in many ways has come a bit later to this sort of analysis than cricket, US sports, even rugby to some extent (yes maybe all more formulaic and predictable than football).

They're already (and I bloody hate this phrase) starting to push back and use less, relying more on coaching and talent to be creative, rather than the over-formulaic stats analysis based stuff. Probably fair enough to think that'll be a pattern.

 

[A Storm of Blades]

New idea for a TV show “Expected Girls” whereby I rank the likelihood of positive interactions with opposite sex through the medium of dance.

 

[Midland Blade]

I know very little about xG and even less about how clubs use analytics, but surely these kind of stats are used more to identify trends in strengths and weakness rather than to judge individual games in a silo? I can't imagine a coaching team will walk off after a 2-0 loss happy with the game as the xG actually indicated it should've been a 2-1 win, but if there is a stretch of 5/10 games where the side have underperformed compared to their xG then it is clear that finishing is an issue and I am sure they will then dig into the chances that are being missed, analyse these and work on whatever the weakness is in training rather than it being specific instructions during a game. In a game with such fine margins, if used correctly then I can see how much of a difference using these types of stats could make over a season if used correctly.

 

[PokerBlade]

Given my username people might guess I have a bit of a bias when it comes to analysing games mathematically.

I guess the thing I'd appeal to is when someone says "We should've scored five" what exactly are you basing it on? It's going to be you remembering the chances in the game, thinking about the quality of them, and assessing how many of those you think would typically be scored, right? And then the question is if you can eyeball that and make estimates on how many goals should have been scored, why wouldn't you be able to start measuring the things you're basing your estimate on to come to a better estimate?

That's all XG is, really. When you see a chance and think "You really should be scoring that", XG is calculated by taking a bunch of similar chances and seeing how many are actually scored. If someone wants to say you can't do that, then what are you basing your opinion that they "really should be scoring" on in the first place? XG is doing the same thing you are for the most part, it's just refining the process.

As a general point, we're really bad at doing things like in our heads. Our memories are biased and our brains are inaccurate. We're flawed thinkers in many ways and often we make big mistakes when we're spotting patterns in data.

One famous example is how Apple had to artificially make the Ipod shuffle features LESS random because people were frustrated that it would play the same song back to back, play three tracks off the same album, etc. All things you'd expect from a random system, but not things people think are random. We're just bad at this kind of estimation of what should happen. And since we're bad at it we use statistics to give ourselves an accurate model.

 

[Scarebus]

Expected stats are shit.....unless they make you look better