Next manager

[CrosspoolClarets]

Enjoyed the last page.

I am a stats expert for my sins and feel qualified to consider some stats as vacuous garbage, and these footie ones, in a notoriously hard to predict profession, qualify.

The key in the line chart shown is that correlation on ALL measures is virtually all under 0.5, so only loosely correlated. OK, so xG, SoT etc slightly better predict future points than current points does, but they remain very similarly correlated.

There are loads of holes:
Xg/Xga ratio - doesn’t consider finishing or goalkeeping expertise
SoT ratio - doesn’t reward a “shoot on sight” policy
Neither factor in a future transfer window, money to spend, or players to return (neither does current points, hence poorly correlated with future points)

The Shots on Target ratio is particularly funny - us, West Brom and Boro are bottom 3 on fbRef. Does anyone reasonably expect that to correspond to future points?

No doubts though we have an issue with creativity, we all know this, but our players to return and transfer window power give me more confidence, whatever the stats say. So hopefully the “new manager” debate is moot

[Mattster]

There are loads of holes:
Xg/Xga ratio - doesn’t consider finishing or goalkeeping expertise

The whole point of xG is to measure the quality of a chance, the likelihood that the average player would score that chance. It isn't there to factor in finishing or goalkeeper quality but it can be used to do so - good finishers will consistently overperform xG and good goalkeepers overperform the PSxG (post shot xG, would not be good judging a keeper on the xG since that wouldn't factor in shots that miss, are blocked nor the shot placement). Without the xG as a starting point though you wouldn't be able to quantify either.

So many people seem to get hung up on what it doesn't factor in/out or on the unfounded idea people think it replaces actual goals/points etc. I genuinely don't get what's so controversial/emotive about it.

[ClaretPete001]

Must admit only looked quickly at this but I'm struggling a bit. In a time series linear model you would expect that points accrued would get closer to one as the season progresses and by game 38 it would be 1 because lets be honest and I don't think I'm dealing in rocket science here there is no better predictor of points earned than the points that have been been earned.

Early in the season points earned maybe a less reliable stat than XG. For example, the fixture list being as it is you could get the fourth worst team at the end of the season playing the worst and second worst in the first two games giving them two wins and a false position, which presumably means the impact of points earned is greater than an enhanced XG against these teams therefore reducing the reliability of points earned.

But how can you have earned points as a variable to predict earned points not becoming more reliable by the end of the season?

Someone put me out of my misery so I don't have to look at it again.

[ClaretPete001]

Enjoyed the last page.

I am a stats expert for my sins and feel qualified to consider some stats as vacuous garbage, and these footie ones, in a notoriously hard to predict profession, qualify.

The key in the line chart shown is that correlation on ALL measures is virtually all under 0.5, so only loosely correlated. OK, so xG, SoT etc slightly better predict future points than current points does, but they remain very similarly correlated.

There are loads of holes:
Xg/Xga ratio - doesn’t consider finishing or goalkeeping expertise
SoT ratio - doesn’t reward a “shoot on sight” policy
Neither factor in a future transfer window, money to spend, or players to return (neither does current points, hence poorly correlated with future points)

The Shots on Target ratio is particularly funny - us, West Brom and Boro are bottom 3 on fbRef. Does anyone reasonably expect that to correspond to future points?

No doubts though we have an issue with creativity, we all know this, but our players to return and transfer window power give me more confidence, whatever the stats say. So hopefully the “new manager” debate is moot

How can earned points have no predictive power at the end of the season?

[bumba]

Your done engaging because you've realised you lost yourself your own arguement.
Fosters goal against Blackburn was more than a significant chance, if he missed that are you telling me it wouldn't have gone down as a significant chance or sitter? A striker with a free header central of the goal from less than 10 yards??
Give over, you know that's garbage.

0.8 for an easy chance that ended up a goal
0.7 at Hull when Anthony missed from 2 yards out

[RVclaret]

Your done engaging because you've realised you lost yourself your own arguement.
Fosters goal against Blackburn was more than a significant chance, if he missed that are you telling me it wouldn't have gone down as a significant chance or sitter? A striker with a free header central of the goal from less than 10 yards??
Give over, you know that's garbage.

0.8 for an easy chance that ended up a goal
0.7 at Hull when Anthony missed from 2 yards out

It was actually far less than 0.8, Foster’s chance (goal) was actually around or under 0.3 i think.

Anthony’s chance at Hull didn’t even count as a shot in the official record so therefore wasn’t even given an xG.

Yes I do agree that both these examples in isolation are clear weaknesses of this metric, but over a longer time period (I always argue for larger samples because of this) it can be useful for spotting trends.

[Mattster]

Your done engaging because you've realised you lost yourself your own arguement.
Fosters goal against Blackburn was more than a significant chance, if he missed that are you telling me it wouldn't have gone down as a significant chance or sitter? A striker with a free header central of the goal from less than 10 yards??
Give over, you know that's garbage.

0.8 for an easy chance that ended up a goal
0.7 at Hull when Anthony missed from 2 yards out

Look, mate, you're not intelligent enough to understand probability as a concept. We get it, you don't have to keep putting out post after post reiterating this.

[bumba]

Look, mate, you're not intelligent enough to understand probability as a concept. We get it, you don't have to keep putting out post after post reiterating this.

And your clearly not intelligent enough to understand probability doesn't mean it's going to go that way for certain.
Thank you for confirming that.
Now look at the fact that counts we are sending in the league meaning of this trend continues the probability is we gain automatic promotion.
Enjoy school.

[Mattster]

And your clearly not intelligent enough to understand probability doesn't mean it's going to go that way for certain.
Thank you for confirming that.
Now look at the fact that counts we are sending in the league meaning of this trend continues the probability is we gain automatic promotion.
Enjoy school.

Do you want to try that again? In English this time.

[bumba]

Do you want to try that again? In English this time.

You know what it says when your old enough to receive your first phone you'll learn all about predictive text.
Scotty Parkers claret and blue army UTC

[Tall Paul]

Must admit only looked quickly at this but I'm struggling a bit. In a time series linear model you would expect that points accrued would get closer to one as the season progresses and by game 38 it would be 1 because lets be honest and I don't think I'm dealing in rocket science here there is no better predictor of points earned than the points that have been been earned.

Early in the season points earned maybe a less reliable stat than XG. For example, the fixture list being as it is you could get the fourth worst team at the end of the season playing the worst and second worst in the first two games giving them two wins and a false position, which presumably means the impact of points earned is greater than an enhanced XG against these teams therefore reducing the reliability of points earned.

But how can you have earned points as a variable to predict earned points not becoming more reliable by the end of the season?

Someone put me out of my misery so I don't have to look at it again.

In the time it took you to write this, you could have read the article linked above the graph, which explains it.

[Mattster]

You know what it says when your old enough to receive your first phone you'll learn all about predictive text.
Scotty Parkers claret and blue army UTC

Shame you've not worked out how to use predictive grammar.

Or proof read. Like a grown up.

[Tall Paul]

The Shots on Target ratio is particularly funny - us, West Brom and Boro are bottom 3 on fbRef. Does anyone reasonably expect that to correspond to future points?

You're looking at the wrong stat on FBRef, the shot on target ratio is defined in the article (Shots on Target Ratio: shots on target for / sum (shots on target for + shots on target against). If my calculations are correct it looks like we're 4th by that metric - behind Leeds, Boro and Sheff Utd.

As to your wider point, of course no predictive model is perfect and this points to the huge amount of variance there is in football results (which the vast majority of football fans don't seem to understand). The point is that xG is probably the best we have at the moment and is certainly better than looking at the league table.

Where do you think the line on the graph would be for predictions by "bloke in the pub who watches football with his own eyes" would be? I'm confident it'd be right at the bottom.

[bumba]

Shame you've not worked out how to use predictive grammar.

Or proof read. Like a grown up.

Thank you for going down the grammar route, it shows you can't even argue your own point.

[ClaretPete001]

In the time it took you to write this, you could have read the article linked above the graph, which explains it.

I have read it and if the charts are trying to predict future points at various point in a season (i.e after each game) then the predictive ability of points achieved must improve as more data is added to the dataset because there is clearly a stronger correlation between points achieved after 25 games and points likely achieved after 38 than points achieved after 3 games and those achieved after 38.

Teams at the top of the table in February don't just start losing for fun. So, what exactly is being measured here?

[Stockbrokerbelt]

Think this post is an insult to our present manager, staff, players & owners, we now have a section of fan base who after a draw, being 2nd in the league, with a stack of injuries & group of new players are not satisfied. Perhaps we should run a post on bad fans, Mard arcs & the entitled!

[Blyclaret]

4 pages of pure shite.
Now they are arguing like kids ….. i thought the school holidays were over.

[gandhisflipflop]

Enjoyed the last page.

I am a stats expert for my sins and feel qualified to consider some stats as vacuous garbage, and these footie ones, in a notoriously hard to predict profession, qualify.

The key in the line chart shown is that correlation on ALL measures is virtually all under 0.5, so only loosely correlated. OK, so xG, SoT etc slightly better predict future points than current points does, but they remain very similarly correlated.

There are loads of holes:
Xg/Xga ratio - doesn’t consider finishing or goalkeeping expertise
SoT ratio - doesn’t reward a “shoot on sight” policy
Neither factor in a future transfer window, money to spend, or players to return (neither does current points, hence poorly correlated with future points)

The Shots on Target ratio is particularly funny - us, West Brom and Boro are bottom 3 on fbRef. Does anyone reasonably expect that to correspond to future points?

No doubts though we have an issue with creativity, we all know this, but our players to return and transfer window power give me more confidence, whatever the stats say. So hopefully the “new manager” debate is moot

Excellent post. I’m sure it’s still half term for some on here.

[Tall Paul]

I have read it and if the charts are trying to predict future points at various point in a season (i.e after each game) then the predictive ability of points achieved must improve as more data is added to the dataset because there is clearly a stronger correlation between points achieved after 25 games and points likely achieved after 38 than points achieved after 3 games and those achieved after 38.

Teams at the top of the table in February don't just start losing for fun. So, what exactly is being measured here?

Future point gained, not future points gained plus points already gained.

From the link: For example, after 8 match rounds played, all twelve metrics are computed over match days 1 to 8 and compared to points per game and goal ratio from match round 9 to the end of the season.

[Mattster]

Thank you for going down the grammar route, it shows you can't even argue your own point.

Mate, I've proven my point. Your posts on this are all over the place. Asking how Foster's goal against Rovers wasn't 1 xG whilst in the same post mentioning Anthony's chance against Hull which wasn't scored, essentially answering your own question but to dim to spot it

Then when I've pointed out you clearly struggle with the concept of probability you've posted "probability doesn't mean it's going to go that way for certain". That's the point I honestly can't see how you think you've had any kind of consistency across your posts in this thread.

[bumba]

Mate, I've proven my point. Your posts on this are all over the place. Asking how Foster's goal against Rovers wasn't 1 xG whilst in the same post mentioning Anthony's chance against Hull which wasn't scored, essentially answering your own question but to dim to spot it

Then when I've pointed out you clearly struggle with the concept of probability you've posted "probability doesn't mean it's going to go that way for certain". That's the point I honestly can't see how you think you've had any kind of consistency across your posts in this thread.

You really are beyond help.....
Foster goal should have been 1 XG
Anthony's chance should have been 1 XG
One was scored, one wasn't but both should have been and both should have been 1 XG.
It really isn't difficult to understand, I'm not sure what your struggling with.
Your saying we will basically be dropping down the league as in it is certain due to a 'probability stat', how can you not grasp that it isn't certain to happen?
The only certainty is the fact we are second.

[Mattster]

You really are beyond help.....
Foster goal should have been 1 XG
Anthony's chance should have been 1 XG
One was scored, one wasn't but both should have been and both should have been 1 XG.
It really isn't difficult to understand, I'm not sure what your struggling with.

And there it is ladies and gents. Proof positive you can't teach stupid.

[Tall Paul]

Yeah, only one poster beyond help on this and it isn't Mattster.

[bumba]

And there it is ladies and gents. Proof positive you can't teach stupid.

XG stands for expected goals, yes?
Anthony and Foster are both expected to score in each situation so should get one XG, yes?
But neither got an XG, is it that difficult to then understand the stat is floored?

[ClaretPete001]

Future point gained, not future points gained plus points already gained.

From the link: For example, after 8 match rounds played, all twelve metrics are computed over match days 1 to 8 and compared to points per game and goal ratio from match round 9 to the end of the season.

Then it's a flawed model because at the start of the season you have the randomness of not enough points to make a highly correlated prediction and by the end of the season you have not enough games to make a highly correlated prediction.

That probably explains the low correlation and the shape of the curve.

[bumba]

Yeah, only one poster beyond help on this and it isn't Mattster.

You would say that because you believe it's a good indicator which it isn't, is it taking in to account each teams form who we come up against at that time, new signings, weather, different tactics.
Too many variants I'm afraid for XG to predict how our season will pan out.

[Mattster]

XG stands for expected goals, yes?
Anthony and Foster are both expected to score in each situation so should get one XG, yes?
But neither got an XG, is it that difficult to then understand the stat is floored?

No, mate, you're floored. Because you are just repeatedly drawing everyone's attention to how you're not intelligent enough to understand the concept of probability.

Or maybe you just don't understand xG? But to not even understand the absolute basics of something you are so openly triggered by shows the same lack of intelligence.

[GetIntoEm]

*flawed

carry on

[Mattster]

*flawed

carry on

I was aware.

[Tall Paul]

Then it's a flawed model because at the start of the season you have the randomness of not enough points to make a highly correlated prediction and by the end of the season you have not enough games to make a highly correlated prediction.

That probably explains the low correlation and the shape of the curve.

You would say that because you believe it's a good indicator which it isn't, is it taking in to account each teams form who we come up against at that time, new signings, weather, different tactics.
Too many variants I'm afraid for XG to predict how our season will pan out.

Of course no predictive model is perfect and this points to the huge amount of variance there is in football results (which the vast majority of football fans don't seem to understand). The point is that xG is probably the best we have at the moment and is certainly better than looking at the league table.

Where do you think the line on the graph would be for predictions by "bloke in the pub who watches football with his own eyes" would be? I'm confident it'd be right at the bottom.

[bumba]

Of course no predictive model is perfect and this points to the huge amount of variance there is in football results (which the vast majority of football fans don't seem to understand). The point is that xG is probably the best we have at the moment and is certainly better than looking at the league table.

Where do you think the line on the graph would be for predictions by "bloke in the pub who watches football with his own eyes" would be? I'm confident it'd be right at the bottom.

So XG is the best we have to predict where we will finish the season????
Give it a rest.

[bumba]

No, mate, you're floored. Because you are just repeatedly drawing everyone's attention to how you're not intelligent enough to understand the concept of probability.

Or maybe you just don't understand xG? But to not even understand the absolute basics of something you are so openly triggered by shows the same lack of intelligence.

I've just explained XG to you, are you sure you even understand it?
A lack of intelligence is somebody trying to say how our season will pan out after a few games by a silly XG stat.
Second in the league keep monitoring that stat mate

[ChorltonCharlie]

XG stands for expected goals, yes?
Anthony and Foster are both expected to score in each situation so should get one XG, yes?
But neither got an XG, is it that difficult to then understand the stat is floored?

Not sure why I'm getting into this, but here goes. xG is just a probability expressed as a decimal, but is really a percentage. With a decent xG model they've analysed millions of situations so that for any attempt they can find many similar situations which can give the likelihood of whether a player would score or not. Foster's against Blackburn may seem low, but they've found that in 70% of occasions the player hits the woodwork, is blocked/saved or misses entirely. It's a decent chance, but if you created 3 a game on average you'll score 1 goal.

For me what xG does is take the emotion out of looking at a game in hindsight. Sometimes that emotion can be positive, and sometimes it can be negative. It's why so many managers and teams respect it. In the cold light of day it gives them a metric that shows how well they're performing against their competitors. It avoids excuses and helps see where you're under performing. It helps you see where you're getting the rub of the green and know that you'll need to change, or see a downturn in results. It can identify where you're on the right track and just need a better striker, or even a turn of luck. I doubt any manager is a slave to xG and similar stats, but I also very much doubt there's any managers at top clubs who are dismissive of it.

As a final point, usually across a season/division the xG will be pretty much bang on with goals scored. Some teams will over perform, some will under perform. You want your finishers to over perform, but a team looking to be successful will know it's even more important to generate chances which give a high overall xG.

[quoonbeatz]

Refs should start using xF and xHB, might see some consistency then.

[Darnhill Claret]

I've only ever used xG and similar stats to find a value bet, especially useful for clubs placed high in the league, not playing well, compared to a team in the bottom half, but showing good recent form.

[houseboy]

The only thing tedious is this thread. It's all that's so wrong in football, wanting a manager, who was dealt a shocking hand, gone after twelve games.

And a manager who has:

Led us to currently second.
Salvaged a great deal out of the debacle after two games.
Got us defensively sound (5 conceded all season).
Only one defeat to top of league away when we had hardly a team to speak of. Only 1-0 at that.
One of the best goal differences in the division.
Created some semblance of team ethic that seemed shockingly missing last season.

All this criticism of Parker seems to be coming from a place of entitlement. We strolled the Championship last time round under a manager who became the devil incarnate a few months later according to some. Because we have a recent record of immediate promotion after dropping from the PL some seem to think it is a given right that we do so. It is not. We have to work at it and these criticisms of Parker seem groundless to me.
Are we playing pretty football? No.
Did we play pretty football under Dyche? No.
Is what we are doing effective? Yes.
If the season ended tomorrow I don’t think we would have too many complaints.

[gandhisflipflop]

The only thing tedious is this thread. It's all that's so wrong in football, wanting a manager, who was dealt a shocking hand, gone after twelve games.

What about wanting a manager, who led us to a title winning season, playing some of the most enjoyable football we have ever seen, culminating in winning the league at our biggest rivals which will never be repeated, gone after less than 12 games?

[boyyanno]

Of course no predictive model is perfect and this points to the huge amount of variance there is in football results (which the vast majority of football fans don't seem to understand). The point is that xG is probably the best we have at the moment and is certainly better than looking at the league table.

Where do you think the line on the graph would be for predictions by "bloke in the pub who watches football with his own eyes" would be? I'm confident it'd be right at the bottom.

I don't think either the league table or XG are accurate at this stage. Both are as flawed as each other.

You could have an XG of 3 but an expected goals against of 4. You can't simply say one statistic viewed in isolation is any more relevant than anything else.

If anyone tries to say that using the data from 12 games they can predict anything then I call bullshit.

And to easily demonstrate my point I'll ask you the same question as I asked Mattster, tell me the result on Saturday? Tell me where we will finish in the league table?

The stats have us as 14th for xg or somthing so I read on here , somehow I don't think you or Mattster are going to be willing to bet we finish outside the top 10?

Apparently XG is more relevant than the league table so I'll predict top 6 and you go for bottom half and let's see who is closer? No I didn't think so.

[bumba]

Inter Milan currently 0-0 against 10 men with an XG of 0.17 and 70% possession, sat 2nd in the league Inzaghi out they'll get relegated!!!!!

[Tall Paul]

I don't think either the league table or XG are accurate at this stage. Both are as flawed as each other.

You could have an XG of 3 but an expected goals against of 4. You can't simply say one statistic viewed in isolation is any more relevant than anything else.

If anyone tries to say that using the data from 12 games they can predict anything then I call bullshit.

And to easily demonstrate my point I'll ask you the same question as I asked Mattster, tell me the result on Saturday? Tell me where we will finish in the league table?

The stats have us as 14th for xg or somthing so I read on here , somehow I don't think you or Mattster are going to be willing to bet we finish outside the top 10?

Apparently XG is more relevant than the league table so I'll predict top 6 and you go for bottom half and let's see who is closer? No I didn't think so.

Did you even read the article or what what I wrote about it not being perfect?

No I didn't think so.

[boyyanno]

Did you even read the article or what what I wrote about it not being perfect?

No I didn't think so.

Of course I did, but I disagree that your metric in isolation is any more relevant than where we are in the table after only 12 games- which is what you said.

But the reason you've tried to deflect is because you can't really argue with the point I've made.

If XG is a better indicator right now than points on the table why don't you think we will finish where the XG has us ranked?

[Mattster]

I don't think either the league table or XG are accurate at this stage. Both are as flawed as each other.

You could have an XG of 3 but an expected goals against of 4. You can't simply say one statistic viewed in isolation is any more relevant than anything else.

If anyone tries to say that using the data from 12 games they can predict anything then I call bullshit.

And to easily demonstrate my point I'll ask you the same question as I asked Mattster, tell me the result on Saturday? Tell me where we will finish in the league table?

The stats have us as 14th for xg or somthing so I read on here , somehow I don't think you or Mattster are going to be willing to bet we finish outside the top 10?

Apparently XG is more relevant than the league table so I'll predict top 6 and you go for bottom half and let's see who is closer? No I didn't think so.

No it doesn't. The xGD has us somewhere in the 4th - 9th range depending on which model and median/mean you opt for.

For xG as a standalone we're down in the 19th because we are clueless in the final third, but that doesn't factor in our defence (xGA) which is very solid. Which is why it's better to look at xGD if you're predicting as that considers both.

I'd say we're more likely to finish in that 4th-9th range than we where we are currently, based off how we're performing. Obviously, we could improve with time, players returning from injury etc.

[Conroy92]

xGA averaged 0.8 after the Leeds game. 0.78 for the 7 games (all against bottom half sides) since. So a slight improvement rather than a decline. I'm not hiding from stats, I just figure it's so small a change as to be irrelevant.

So perhaps when you are less creative then stats show you concede less? Is that a fair assessment.

Conceding fewer should get your more points, right.

So could I not say that goals against is a better indicator of points gained.

Or does it also consist of how many you've scored too?

Can you see why your isolated stats don't work.

[Mattster]

So perhaps when you are less creative then stats show you concede less? Is that a fair assessment.

Conceding fewer should get your more points, right.

So could I not say that goals against is a better indicator of points gained.

Or does it also consist of how many you've scored too?

Can you see why your isolated stats don't work.

Where did I say to / use stats in isolation?

[ClaretPete001]

At the end of the day, points achieved are the best way to predict the points that are going to be achieved.

The problem with the model is it has numerous flaws the most important of which it is never measuring a specific outcome. If the model had used the whole seasons results then points gained would have eventually become a correlation of 1 and at each point you could get a sense of the accuracy of the various variables as the season progresses. It doesn't do that it tries to predict an ever changing number of games as the season progresses so at the start of the season you have not enough points data to make a prediction and at the end not enough game data.

Another is the variables themselves. XG is an aggregated statistic based upon assumption whereas Points achieved and XGD are based upon real world outcomes.

What the chart is essentially doing is telling the designer that his model is not a good fit for the data. If mid-season had a correlation of 0.75 or 0.8 then it would be interesting because it would suggest that XG was a reliable indicator of points achieved mid season but it doesn't.

What it likely does show is that points achieved does not become a reliable stat until towards the end of the season and at various points within it you cannot predict using either XG or points achieved with any reliability.

Unless of course you are a City fan and let's be honest they don't need models.

[daveisaclaret]

I have to say, as someone who isn't too interested in xG and similar stats, the arrogance, ignorance and willful misunderstanding of the stats from those who are actively against them is quite illuminating.

[bumba]

Sunderland draw 0-0 against QPR with only 41% possession, 1 on target and an XG of 0.87.....wonder who their fans will suggest on Sunderland's 'next manager' thread now they are heading on a downward spiral

[burnley007]

Is Ruud available yet?

[Burnley1989]

Is Ruud available yet?

You've got issues, it's the first defeat in 10

[Blyclaret]

Is Ruud available yet?

Get off this board … you are not a Burnley supporter