Please log in or register. Registered visitors get fewer ads.
Forum index | Previous Thread | Next thread
Accuracy of Xg, Xp estimates 10:40 - May 11 with 4861 viewsxrayspecs



So for L1 this season, this approach:
- failed to place the champions in the automatics or top 6
- identified 50% of teams getting automatic promotion
- identified 50% of team making play-off (plus Wednesday who they placed in autos)

It was better at predicting the bottom four.

So a little bit like the forecasts that gave SW 99% chance of top 2, these stats are not that reliable or useful What may be missing in the formula is resilience/mental strength, as Plymouth found a way to win despite not playing well on a regular basis.
0
Accuracy of Xg, Xp estimates on 16:08 - May 11 with 3267 viewsHerbivore

I think you misunderstand what xG is for, it's not meant to be predictive and so not sure what you mean when you suggest it's not accurate. Also, with a few major exceptions, most sides are placed within a few places of their finishing position in the xG/xP table which shows that generally teams finish close to their underlying offensive and defensive statistics.

Poll: Should someone on benefits earn more than David Cameron?
Blog: Where Did It All Go Wrong for Paul Hurst?

1
Accuracy of Xg, Xp estimates on 17:10 - May 11 with 3181 viewsNorrisHatter

Accuracy of Xg, Xp estimates on 16:08 - May 11 by Herbivore

I think you misunderstand what xG is for, it's not meant to be predictive and so not sure what you mean when you suggest it's not accurate. Also, with a few major exceptions, most sides are placed within a few places of their finishing position in the xG/xP table which shows that generally teams finish close to their underlying offensive and defensive statistics.


Exactly! xG tells you rather more what and not much why. In the same way as a business might look at its enquiry conversion stats and find they are actually converting less than they had previously the reaction shouldn't be to conclude its not worth tracking enquiry conversion but rather to understand the reasons for under performance.

Not the best sales people on the good opportunities? Too slow to respond to customers so competitors were able to react, get organised and block the chance of getting a sale?

Parallels with what happens on a football pitch?

I believe SmartOdds assess more specific data that identifies not just where a shot occurred but how dangerous the attack was (e.g. slow and laborious vs quick counter attack / turn over of possession). I understand that this company is owned by the same guy who owns Brentford. Brentford's over achievement versus their financial clout a coincidence?
0
Accuracy of Xg, Xp estimates on 18:03 - May 11 with 3103 viewsclive_baker

Accuracy of Xg, Xp estimates on 16:08 - May 11 by Herbivore

I think you misunderstand what xG is for, it's not meant to be predictive and so not sure what you mean when you suggest it's not accurate. Also, with a few major exceptions, most sides are placed within a few places of their finishing position in the xG/xP table which shows that generally teams finish close to their underlying offensive and defensive statistics.


Take your point that they're probably directionally right, but it does beg the question how reliable or meaningful the absolute XG and XP measures are when 21 of the 24 sides have over delivered their XG, some by as many as 20-30 goals. CBA to sum the XG columns on both tables but the actuals must be at least 15% higher than the XG.

When you've got a sample size over the season of 1,100 team performances to pull from it doesn't feel right that there should be such a consistent discrepancy, and would suggest the XG is too stingy.

Poll: Will Boris Johnson be PM this time next week?
Blog: [Blog] Team Spirit Holds the Key

0
Accuracy of Xg, Xp estimates on 18:05 - May 11 with 3100 viewsxrayspecs

Accuracy of Xg, Xp estimates on 16:08 - May 11 by Herbivore

I think you misunderstand what xG is for, it's not meant to be predictive and so not sure what you mean when you suggest it's not accurate. Also, with a few major exceptions, most sides are placed within a few places of their finishing position in the xG/xP table which shows that generally teams finish close to their underlying offensive and defensive statistics.


Ok, thanks for explanation.

If these data are not predictive then why have they created a predicted points table on the left? That is what confused me.

We live in a world with increasing levels of data, the challenge is to find the insights and intelligence. The SW automatic promotion percentage shows how volatile these data can be.
0
Accuracy of Xg, Xp estimates on 18:07 - May 11 with 3092 viewsTrequartista

Oxford in 6th and Plymouth in 8th tells you all you need to know about this useless stat.

Poll: Who do you blame for our failure to progress?

1
Accuracy of Xg, Xp estimates on 18:23 - May 11 with 3065 viewsNorrisHatter

Accuracy of Xg, Xp estimates on 18:05 - May 11 by xrayspecs

Ok, thanks for explanation.

If these data are not predictive then why have they created a predicted points table on the left? That is what confused me.

We live in a world with increasing levels of data, the challenge is to find the insights and intelligence. The SW automatic promotion percentage shows how volatile these data can be.


From the XG data you can run multiple simulations of how actual games would have turned out - and its not as simple as the team with the higher xG always wins.

Imagine a game were I have a die with 5 sixes and one zero, then my average expected score from rolling the die will be 5. I give you a die with 5 zeros on it and one 6 - so your expected average score is 1. So it looks like I'll just win 5-1 all the time from the "xG" data But it doesn't work out like that. Here's why

We play a game where I roll my die and you then roll yours. Most of the time I win, but occasionally my zero will match with your six and you win and also, very occasionally we finish all square (this is known in the jargon used by experts as a "Plymouth deflection").

From those simulations you can calculate the number of times I get three points from the game and how many times one or even none - and the same for you. Weirdly that means that you can get matches where the total average expected points for each team adds to more than three!

Its from that process that so called "justice tables" can be calculated which shows where teams ought to finish based on their xG performance.

The fact that a side that should finish outside the top six actually wins the title would be described as a (statistical) injustice. I'm sure we can all agree on that
0
Accuracy of Xg, Xp estimates on 18:57 - May 11 with 2977 viewsxrayspecs

Accuracy of Xg, Xp estimates on 18:23 - May 11 by NorrisHatter

From the XG data you can run multiple simulations of how actual games would have turned out - and its not as simple as the team with the higher xG always wins.

Imagine a game were I have a die with 5 sixes and one zero, then my average expected score from rolling the die will be 5. I give you a die with 5 zeros on it and one 6 - so your expected average score is 1. So it looks like I'll just win 5-1 all the time from the "xG" data But it doesn't work out like that. Here's why

We play a game where I roll my die and you then roll yours. Most of the time I win, but occasionally my zero will match with your six and you win and also, very occasionally we finish all square (this is known in the jargon used by experts as a "Plymouth deflection").

From those simulations you can calculate the number of times I get three points from the game and how many times one or even none - and the same for you. Weirdly that means that you can get matches where the total average expected points for each team adds to more than three!

Its from that process that so called "justice tables" can be calculated which shows where teams ought to finish based on their xG performance.

The fact that a side that should finish outside the top six actually wins the title would be described as a (statistical) injustice. I'm sure we can all agree on that


Thanks. I do have a Master's degree in applied statistics, but you were not to know that...

As per my OP, I understand why some of these data are predictors of team performance. I do though believe that these models are misspecified and that there are other factors which impact actual performance.
[Post edited 11 May 2023 18:59]
0
Accuracy of Xg, Xp estimates on 19:21 - May 11 with 2930 viewsHerbivore

Accuracy of Xg, Xp estimates on 18:57 - May 11 by xrayspecs

Thanks. I do have a Master's degree in applied statistics, but you were not to know that...

As per my OP, I understand why some of these data are predictors of team performance. I do though believe that these models are misspecified and that there are other factors which impact actual performance.
[Post edited 11 May 2023 18:59]


They aren't predictors. They are descriptive statistics in that they are reporting on what has happened. The xP is simply what would happen if the xG had played out.

Poll: Should someone on benefits earn more than David Cameron?
Blog: Where Did It All Go Wrong for Paul Hurst?

0
Login to get fewer ads

Accuracy of Xg, Xp estimates on 19:56 - May 11 with 2879 viewsxrayspecs

Accuracy of Xg, Xp estimates on 19:21 - May 11 by Herbivore

They aren't predictors. They are descriptive statistics in that they are reporting on what has happened. The xP is simply what would happen if the xG had played out.


From Statsbomb website:

Put simply, Expected Goals (xG) is a metric designed to measure the probability of a shot resulting in a goal.

An xG model uses historical information from thousands of shots with similar characteristics to estimate the likelihood of a goal on a scale between 0 and 1.

For example, a shot with an xG value of 0.2 is one that we would generally expect to be converted twice in every 10 attempts.

https://statsbomb.com/soccer-metrics/expected-goals-xg-explained/

If the historic probabilities are accurate, then you would expect with lots of shots that the actual and predicted goals would be close. No idea how many shots Town had last season but their xG was 74 yet they scored a third more. That is a big margin of difference.
[Post edited 11 May 2023 19:57]
0
Accuracy of Xg, Xp estimates on 20:08 - May 11 with 2839 viewsHerbivore

Accuracy of Xg, Xp estimates on 19:56 - May 11 by xrayspecs

From Statsbomb website:

Put simply, Expected Goals (xG) is a metric designed to measure the probability of a shot resulting in a goal.

An xG model uses historical information from thousands of shots with similar characteristics to estimate the likelihood of a goal on a scale between 0 and 1.

For example, a shot with an xG value of 0.2 is one that we would generally expect to be converted twice in every 10 attempts.

https://statsbomb.com/soccer-metrics/expected-goals-xg-explained/

If the historic probabilities are accurate, then you would expect with lots of shots that the actual and predicted goals would be close. No idea how many shots Town had last season but their xG was 74 yet they scored a third more. That is a big margin of difference.
[Post edited 11 May 2023 19:57]


But your OP was about how 'off' it was on predicting league positions, which it wasn't really and which isn't the point of it. If you're saying it's not a great measure of how many goals you'd actually expect a team to score then that's a different argument and I'm not sure anyone thinks they have it nailed yet on that front. Just speculating but if they are using a lot of data to calculate xG then you'd expect better teams and leagues to outscore their xG, you'd imagine League 1 players will be better at converting chances than teams in the Ridgeons league.

Poll: Should someone on benefits earn more than David Cameron?
Blog: Where Did It All Go Wrong for Paul Hurst?

0
Accuracy of Xg, Xp estimates on 20:23 - May 11 with 2820 viewsxrayspecs

Accuracy of Xg, Xp estimates on 20:08 - May 11 by Herbivore

But your OP was about how 'off' it was on predicting league positions, which it wasn't really and which isn't the point of it. If you're saying it's not a great measure of how many goals you'd actually expect a team to score then that's a different argument and I'm not sure anyone thinks they have it nailed yet on that front. Just speculating but if they are using a lot of data to calculate xG then you'd expect better teams and leagues to outscore their xG, you'd imagine League 1 players will be better at converting chances than teams in the Ridgeons league.


My OP was based on looking at the table that had been created, which I understood to be a prediction, hence the observation that it was not very accurate.

Most recent post is based on the googled xG definition, which is about probabilities. This I do understand, hence comment I would expect actual and predicted to be closer than they are.

I have no idea on what they are basing their probabilities, whether these are league specific or across all leagues. I do think you are right that quality of players will have an impact. We have been much more prolific since Broadhead and Hirst arrived and arguably Burns got second wind after his shortened summer and World Cup.

Cheers.
0
Accuracy of Xg, Xp estimates on 21:10 - May 11 with 2763 viewsHerbivore

Accuracy of Xg, Xp estimates on 20:23 - May 11 by xrayspecs

My OP was based on looking at the table that had been created, which I understood to be a prediction, hence the observation that it was not very accurate.

Most recent post is based on the googled xG definition, which is about probabilities. This I do understand, hence comment I would expect actual and predicted to be closer than they are.

I have no idea on what they are basing their probabilities, whether these are league specific or across all leagues. I do think you are right that quality of players will have an impact. We have been much more prolific since Broadhead and Hirst arrived and arguably Burns got second wind after his shortened summer and World Cup.

Cheers.


The table isn't a prediction, it's what the table would look like if games had gone by xG rather than actual goals. How helpful that is will depend on your own point of view but xG is measured the same for everyone so there is some value to it as a guide to quality of chances created, if not for chances taken. How they calculate xG is similarly based on past data and applied to chances created so again it's not really predictive per se.

Poll: Should someone on benefits earn more than David Cameron?
Blog: Where Did It All Go Wrong for Paul Hurst?

0
Accuracy of Xg, Xp estimates on 21:16 - May 11 with 2752 viewsxrayspecs

Accuracy of Xg, Xp estimates on 21:10 - May 11 by Herbivore

The table isn't a prediction, it's what the table would look like if games had gone by xG rather than actual goals. How helpful that is will depend on your own point of view but xG is measured the same for everyone so there is some value to it as a guide to quality of chances created, if not for chances taken. How they calculate xG is similarly based on past data and applied to chances created so again it's not really predictive per se.


Maybe it is a language ting, but expected goals is a prediction imv.

Going to file this under not sure I understand. Thanks for your posts, cheers.
0
Accuracy of Xg, Xp estimates on 22:13 - May 11 with 2686 viewsVegtablue

Accuracy of Xg, Xp estimates on 21:16 - May 11 by xrayspecs

Maybe it is a language ting, but expected goals is a prediction imv.

Going to file this under not sure I understand. Thanks for your posts, cheers.


Do you put forward that it isn't possible to reliably assign probability values onto goalscoring opportunities? Or that the data currently used to generate these values is flawed?

The boffins at xG HQ would argue they're providing accurate measurements: what the scorelines would be if robots took over at every shooting moment. Or at least that their measurements are sufficiently accurate so as to be useful. Rather than predicting goals, they are providing the (cumulative) probability of them, to 0.01 precision for every shot.

I believe they've admitted to not factoring in a striker's preferred foot, nor do they factor in the ability of the player taking (or potentially saving) the shot.
0
Accuracy of Xg, Xp estimates on 23:17 - May 11 with 2631 viewsjayessess

Think both stats are an attempt to measure and standardise something that football fans have tried to talk about intuitively since the beginning of time really. Whenever we talk about whether a team "deserved to win" or whether a team was "flattered" by a big win, we're implicitly making a judgement similar to xG/xP. We're saying from experience we usually expect the team that creates the most good shooting opportunities to win the game, even if we know things don't always work out that way.

Inevitably when you try to standardise something that subjective your model isn't going to capture everything. It's also not designed to just measure how good each team is at accumulating points (that's what the league table is for) but how regularly they created/conceded good shooting opportunities.

Blog: What Now? Taking a Look at Life in League One

1
Accuracy of Xg, Xp estimates on 23:24 - May 11 with 2621 viewsExiled2Surrey

Accuracy of Xg, Xp estimates on 20:23 - May 11 by xrayspecs

My OP was based on looking at the table that had been created, which I understood to be a prediction, hence the observation that it was not very accurate.

Most recent post is based on the googled xG definition, which is about probabilities. This I do understand, hence comment I would expect actual and predicted to be closer than they are.

I have no idea on what they are basing their probabilities, whether these are league specific or across all leagues. I do think you are right that quality of players will have an impact. We have been much more prolific since Broadhead and Hirst arrived and arguably Burns got second wind after his shortened summer and World Cup.

Cheers.


I have asked the question as to whether these XG numbers are league specific previously - I think they must be. However, not sure then what would happen in a cup game between Man City and Crewe. Would love to know the answer to that.

I also asked the question earlier this season whether lots of smaller probability chances was was preferable to a few higher probability chances - if the probabilities are calculated correctly, then there should be no difference, but I doubt that is the case and I think that a few bigger chances tend to win out - or at least they seemed to earlier this season.
0
Accuracy of Xg, Xp estimates on 07:45 - May 12 with 2509 viewsxrayspecs

Accuracy of Xg, Xp estimates on 22:13 - May 11 by Vegtablue

Do you put forward that it isn't possible to reliably assign probability values onto goalscoring opportunities? Or that the data currently used to generate these values is flawed?

The boffins at xG HQ would argue they're providing accurate measurements: what the scorelines would be if robots took over at every shooting moment. Or at least that their measurements are sufficiently accurate so as to be useful. Rather than predicting goals, they are providing the (cumulative) probability of them, to 0.01 precision for every shot.

I believe they've admitted to not factoring in a striker's preferred foot, nor do they factor in the ability of the player taking (or potentially saving) the shot.


Good question. I can subscribe to the concept, my observation is that cumulative probabilities and reality are often very different, which to me suggests that the algorithm/model etc. is not sufficiently accurate. Wether you can fix this, I do not know. If you are predicting the champions (on 100+ points) do not get in the top six, then something is clearly amiss.
0
Accuracy of Xg, Xp estimates on 08:03 - May 12 with 2483 viewsDJR

Accuracy of Xg, Xp estimates on 07:45 - May 12 by xrayspecs

Good question. I can subscribe to the concept, my observation is that cumulative probabilities and reality are often very different, which to me suggests that the algorithm/model etc. is not sufficiently accurate. Wether you can fix this, I do not know. If you are predicting the champions (on 100+ points) do not get in the top six, then something is clearly amiss.


I haven't read all the posts in this thread in great detail but what appears to be missing in all this is the likelihood of a player converting a chance.

For example, five identical chances falling to the same player, will have a completely different outcome in terms of goals if the player is Conor Chaplin in contrast to, say, Sam Morsy.

In the case of, say, Plymouth, the impression I got was that they either had a larger number of players with a goal-scoring instinct, or a high percentage of their chances fell to such players in their team.

Another factor that could be in play is whether a chance is on the counter-attack, where defences are less likely to be packed. And there will be other factors, such as whether a team packs its defence throughout the game.

In this day in age, it is not clear to me that such factors could not be built into the system, using for example things like shot accuracy, shot conversion etc. But I assume xG as it stands will continue to be a required piece of data.
[Post edited 12 May 2023 8:06]
0
Accuracy of Xg, Xp estimates on 08:07 - May 12 with 2453 viewsBondiBlue

Accuracy of Xg, Xp estimates on 18:07 - May 11 by Trequartista

Oxford in 6th and Plymouth in 8th tells you all you need to know about this useless stat.


Or it tells you that oxford's strikers are hopeless and plymouth's are brilliant! In all seriousness, i'm sure the recruitment guys probably pay attention to attackers who outperform their XG - means they are scoring half-chances and not missing sitters.

Poll: Which would you accept?

0
Accuracy of Xg, Xp estimates on 08:14 - May 12 with 2430 viewsHerbivore

Accuracy of Xg, Xp estimates on 07:45 - May 12 by xrayspecs

Good question. I can subscribe to the concept, my observation is that cumulative probabilities and reality are often very different, which to me suggests that the algorithm/model etc. is not sufficiently accurate. Wether you can fix this, I do not know. If you are predicting the champions (on 100+ points) do not get in the top six, then something is clearly amiss.


It's not a prediction.

Poll: Should someone on benefits earn more than David Cameron?
Blog: Where Did It All Go Wrong for Paul Hurst?

0
Accuracy of Xg, Xp estimates on 08:46 - May 12 with 2361 viewsrickw

As others have said it's not a prediction, what this tells us is that Plymouth have been extremely clinical (or their opponents wasteful) and the opposite for Oxford

Poll: Of the Non Favourites for ITFC managers job who would you prefer
Blog: Reasons for Relegation

0
Accuracy of Xg, Xp estimates on 09:38 - May 12 with 2284 viewsshady

Do add to this topic there is another rating called SPI: Soccer Power Index

Weekly I've been noticing on
fivethirtyeight that our SPI has been increasing way ahead of say Plymouth's
so much so that we've even powered ahead of all but 4 of the championship teams and we are now at number 151 in their world rankings.
Despite all Plymouth's wins their rating barely changed,
possibly as their wins were by a single goal in close games?
However ours just powered ahead maybe caused by 6-0 wins and emphatic away wins at direct competitors

Also it showed that Oxford about 13th in Division 1 listing and Plymouth 5th.

https://projects.fivethirtyeight.com/soccer-predictions/league-one/

Tis all somewhat confusing and shows how reliance on statistics to predict
results is still in its infancy.
0
Accuracy of Xg, Xp estimates on 10:31 - May 12 with 2229 viewsjayessess

Accuracy of Xg, Xp estimates on 07:45 - May 12 by xrayspecs

Good question. I can subscribe to the concept, my observation is that cumulative probabilities and reality are often very different, which to me suggests that the algorithm/model etc. is not sufficiently accurate. Wether you can fix this, I do not know. If you are predicting the champions (on 100+ points) do not get in the top six, then something is clearly amiss.


Well, it would tell you that Plymouth are either (a) doing something that escapes your analysis, meaning the opportunities they create are more valuable than they appear or (b) Plymouth are doing things that you aren't trying to measure with an xG stat (finishing more efficiently, stopping more shots, late game mentality).

Blog: What Now? Taking a Look at Life in League One

0
Accuracy of Xg, Xp estimates on 10:35 - May 12 with 2212 viewsTalkingBlues

Accuracy of Xg, Xp estimates on 21:16 - May 11 by xrayspecs

Maybe it is a language ting, but expected goals is a prediction imv.

Going to file this under not sure I understand. Thanks for your posts, cheers.


Nail on head, the major obstacle for most of us to get over is the what the jargon actually means. The "expected" bit, which in most minds (including my own) suggests some sort of expectation of a future outcome, but that's not the case here and in fact, the "G" shouldn't be referred to as "Goals" at all, because it doesn't relate to goals, but chances, nutty, but true.

All "xG" actually measures, is the quality of a chance, it in no way is intended to predict the number of goals a team will score, or predict an outcome. The team with the higher "xG" prior to kick off, is purely "expected" to make better quality chances in the game based off previous data, but in no way does that suggest an outcome, it in no way is intended to infer they will actually capitalise on those chances and score goals. The "expected" bit, is derived from the mathematical concept of "expected value" to measure the likelihood of an outcome occurring (which in itself, in my mind, suggests some sort or prediction, but apparently that's not the intention at all, !?). The whole concept of this stat appears to be that over time (by the end of the season) there will be a reversion to mean, so you can (and do) get wildly below, or above "xG" from one game to the next, but by the end of the season, this should have smoothed out, which begs the question itself, what's the point of the metric at all during the season?

What's the upshot of all this? The metric would be far better labelled completely differently, as its name implies in our minds, something that the metric is in no way designed to deliver on, neither "expected" or "goals", but merely expected "quality of chances" based on previous data, that has no bearing whatsoever on future outcome.

Basically, it's a nonsense stat that is widely misinterpreted and misused and is utterly pointless IMO, other than possibly to chuck in a spreadsheet to your boss at the end of the season to make it look like you know your job, if you work in data analytics at a football club that is.
[Post edited 12 May 2023 10:51]

Poll: Is it wrong to take more pleasure from seeing Germany lose, than England win

0
Accuracy of Xg, Xp estimates on 10:41 - May 12 with 2204 viewsHerbivore

Accuracy of Xg, Xp estimates on 10:31 - May 12 by jayessess

Well, it would tell you that Plymouth are either (a) doing something that escapes your analysis, meaning the opportunities they create are more valuable than they appear or (b) Plymouth are doing things that you aren't trying to measure with an xG stat (finishing more efficiently, stopping more shots, late game mentality).


Think Plymouth have also just scored from quite a lot of low percentage chances. Against us they scored three times, all from outside the box and two with the aid of a deflection. Fair play to them, they've outperformed their underlying statistics consistently all season and the xG table posted shows that they are something of an outlier in that regard.

Poll: Should someone on benefits earn more than David Cameron?
Blog: Where Did It All Go Wrong for Paul Hurst?

0
About Us Contact Us Terms & Conditions Privacy Cookies Advertising
© TWTD 1995-2024