MATCHWATCH : Carlisle United (h)

[Snowflake Royal]

wow 29% - really? We're 6 points adrift ffs

There's a number of factors as to why it's dropped so much.

It's first of all worth pointing out that the central forecast is for us to finish 19th - so we've gone from clearly being relegation favourites to relegation isn't an issue. The average points for us is 47.9, compared to 44.8 for (projected) 21st-placed Shrewsbury - so that's only just over a win above the relegation zone by the end of the season.

Net xG is used to project future results:

Net xG per game (sourced from Fotmob) - Team / xGF / xGA / xGD

1. Peterborough / 1.77 / 0.91 / +0.86
2. Portsmouth / 1.71 / 1.04 / +0.67
3. Derby / 1.43 / 0.95 / +0.48 / +0.48
4. Blackpool / 1.43 / 1.00 / +0.43
5. Bolton / 1.40 / 1.05 / +0.35
6. Oxford / 1.23 / 0.88 / +0.34
7. Charlton / 1.42 / 1.16 / +0.27
8. Lincoln / 1.11 / 0.99 / +0.11
9. Stevenage / 1.20 / 1.10 / +0.10
10. Reading / 1.17 / 1.13 / +0.04
11. Leyton Orient / 1.11 / 1.10 / +0.01
12. Bristol Rovers / 1.06 / 1.08 / -0.02
13. Barnsley / 1.31 / 1.37 / -0.06
14. Northampton / 0.96 / 1.08 / -0.12
15. Wycombe / 1.05 / 1.22 / -0.17
16. Port Vale / 0.97 / 1.15 / -0.18
17. Carlisle / 0.87 / 1.13 / -0.26
18. Fleetwood / 1.09 / 1.37 / -0.27
19. Cambridge / 0.92 / 1.21 / -0.29
20. Burton / 0.92 / 1.21 / -0.29
21. Exeter / 1.01 / 1.35 / -0.34
22. Wigan / 0.98 / 1.40 / -0.42
23. Shrewsbury / 0.67 / 1.28 / -0.61
24. Cheltenham / 0.76 / 1.38 / -0.62

Our net xG is noticeably better than most of the sides around us, which means the model I'm using projects us picking up more points over the rest of the season that the sides around us.

The other factor is points on the board. Whilst we've been more likely to pick up points in future matches, based on our xG, actually having points on the board is more important to our chances of survival - actually winning a game is better than having a 2/3rds chances of winning a game. Getting the wins over the past couple of matches means we've converted games we're projected to get points in into actual points (on top of slightly improving our net xG).

Relegation chance

1. Cheltenham / 78% (42% chance of finishing bottom)
2. Cambridge / 55%
3. Shrewsbury / 50%
4. Fleetwood / 48%
5. Carlisle / 41%
6. Exeter / 31%
7. Reading / 29%
8. Port Vale / 22%
9. Burton / 21%
10. Wigan / 13%
11. Leyton Orient / 4%
12. Northampton / 4%
13. Wycombe / 2%
14. Bristol Rovers / 2%

Of these, I'd say anyone with a relegation chance of Wigan or worse should still be concerned about getting relegated. What we've done is the past couple of games is move from a side that looks certain to be relegated to a side that will be in the relegation picture, but more likely to stay up than go down.

Shrewsbury are an interesting flip side to us: currently 12th, but expected to be relegated half the time due to their performances this season. Generally sides drift towards their net xG as the season goes on, so Shrewsbury could be a side that falls down the league in the second half of the season.

Is there any empirical evidence for this model? For example, can you retrospectively view last season in League One (and the previous 2 or 3 seasons) at a similar stage and and see a similar pattern as to the eventual relegation outcomes? If not, then (and i am not knocking it) the model is - at best - an interesting exercise but not one to base too much expectation of accuracy on.

I think expected goals is empirically the most reliable statistical model for predicting results.

The problem with this is that people then conflate that with it being a reliable prediction method. It isn’t. Its inaccurate, just less innacurate than the others.

We've been regularly confounding our expected goals stats for years in extended patches. When we were good we often over performed it. When we were bad we often under performed it.

[tmesis]

Is there any empirical evidence for this model? For example, can you retrospectively view last season in League One (and the previous 2 or 3 seasons) at a similar stage and and see a similar pattern as to the eventual relegation outcomes? If not, then (and i am not knocking it) the model is - at best - an interesting exercise but not one to base too much expectation of accuracy on.

Indeed. I think people put far too much weight into these stats.

I mean, if your xG suggests you have 30 goals by now, but you've only got 15, it might just show that your strikers are a bit crap, and that's not going to change.

Generally, the teams who create more chances will do better long term. You get streaks in either direction caused by confidence or a bit of luck, but it's rarely sustained. Looking at the opposite scenario, in 20/21 we were massively outperforming xG up until Christmas because Joao was on fire and scored a lot of goals out of nothing, but we couldn't sustain it when he dropped off.

Now it feels like the opposite. We've been underperforming it because of what is mostly confidence and a mental block. But now we have a couple of wins, Smith, Wing, Knibbs and Azeez seem to be coming into form. I would expect a corrective run of form.

Sorry, I don't have data. But teams who create more chances winning more games is logical enough.

I agree, but you could say the same thing about goal difference being an indicator of the quality of a squad, and how well they are stacking up to their potential.

I just don't think working out percentage chances of doing anything from those stats is remotely reliable as there are so many other factors.

[Snowflake Royal]

Indeed. I think people put far too much weight into these stats.

I mean, if your xG suggests you have 30 goals by now, but you've only got 15, it might just show that your strikers are a bit crap, and that's not going to change.

Generally, the teams who create more chances will do better long term. You get streaks in either direction caused by confidence or a bit of luck, but it's rarely sustained. Looking at the opposite scenario, in 20/21 we were massively outperforming xG up until Christmas because Joao was on fire and scored a lot of goals out of nothing, but we couldn't sustain it when he dropped off.

Now it feels like the opposite. We've been underperforming it because of what is mostly confidence and a mental block. But now we have a couple of wins, Smith, Wing, Knibbs and Azeez seem to be coming into form. I would expect a corrective run of form.

Sorry, I don't have data. But teams who create more chances winning more games is logical enough.

I agree, but you could say the same thing about goal difference being an indicator of the quality of a squad, and how well they are stacking up to their potential.

I just don't think working out percentage chances of doing anything from those stats is remotely reliable as there are so many other factors.

It's a trend that will play out fairly often in a large enough sample size over a long enough time, it just doesn't make it good at predicting outcomes for individual clubs over part of a season. As you say there are too many factors at play.

[YorkshireRoyal99]

Generally, the teams who create more chances will do better long term. You get streaks in either direction caused by confidence or a bit of luck, but it's rarely sustained. Looking at the opposite scenario, in 20/21 we were massively outperforming xG up until Christmas because Joao was on fire and scored a lot of goals out of nothing, but we couldn't sustain it when he dropped off.

Now it feels like the opposite. We've been underperforming it because of what is mostly confidence and a mental block. But now we have a couple of wins, Smith, Wing, Knibbs and Azeez seem to be coming into form. I would expect a corrective run of form.

Sorry, I don't have data. But teams who create more chances winning more games is logical enough.

I agree, but you could say the same thing about goal difference being an indicator of the quality of a squad, and how well they are stacking up to their potential.

I just don't think working out percentage chances of doing anything from those stats is remotely reliable as there are so many other factors.

It's a trend that will play out fairly often in a large enough sample size over a long enough time, it just doesn't make it good at predicting outcomes for individual clubs over part of a season. As you say there are too many factors at play.

Yeah they are a fairly good indicator of the "quality" of your side, but it's also just numbers without context, as mentioned you would need the reason why you've only scored 15 goals from an expected of 30, for instance. It could be a poor quality striker, wrong type of chances, bad luck, good goalkeeper, a combination of these etc.

[Clyde1998]

Is there any empirical evidence for this model? For example, can you retrospectively view last season in League One (and the previous 2 or 3 seasons) at a similar stage and and see a similar pattern as to the eventual relegation outcomes? If not, then (and i am not knocking it) the model is - at best - an interesting exercise but not one to base too much expectation of accuracy on.

On this specific model, I'd have to go back to this stage in previous seasons to give any specific figures.

On using xG, however, the Experimental 361 xG league tables for 46-game seasons (ie. Championship, League One and League Two) has shown an average error of 9.5 points since 2019-20 (2020-21 for L1/L2), with a standard deviation of 7.1 (basically, most teams are within 2 and 17 points of their xG by the end of the season) and a positional error of 3.5 with a standard deviation of 3.1 (so a typical error between 0 and 7 places) - most teams end up broadly at their xG position. Of the 240 observations, only eight have been more than ten positions away from their xG position in that time (3.33%).

Of course as others have pointed out, there are other factors than simply the chances created that will determine a season - significantly better/worse players than average, injuries and mid-season transfers changing the strength of the squad, tactical changes, managerial changes, etc. Outliers are always possible: xG shows the likelihood of an average professional player to score a specific chance - players further from that average, in either direction, will convert chances at a different rate.

Perhaps it's better to think of it as a guide of how teams will do if they carry on playing as they have been throughout the season.

[Mr Angry]

Is there any empirical evidence for this model? For example, can you retrospectively view last season in League One (and the previous 2 or 3 seasons) at a similar stage and and see a similar pattern as to the eventual relegation outcomes? If not, then (and i am not knocking it) the model is - at best - an interesting exercise but not one to base too much expectation of accuracy on.

On this specific model, I'd have to go back to this stage in previous seasons to give any specific figures.

On using xG, however, the Experimental 361 xG league tables for 46-game seasons (ie. Championship, League One and League Two) has shown an average error of 9.5 points since 2019-20 (2020-21 for L1/L2), with a standard deviation of 7.1 (basically, most teams are within 2 and 17 points of their xG by the end of the season) and a positional error of 3.5 with a standard deviation of 3.1 (so a typical error between 0 and 7 places) - most teams end up broadly at their xG position. Of the 240 observations, only eight have been more than ten positions away from their xG position in that time (3.33%).

Of course as others have pointed out, there are other factors than simply the chances created that will determine a season - significantly better/worse players than average, injuries and mid-season transfers changing the strength of the squad, tactical changes, managerial changes, etc. Outliers are always possible: xG shows the likelihood of an average professional player to score a specific chance - players further from that average, in either direction, will convert chances at a different rate.

Perhaps it's better to think of it as a guide of how teams will do if they carry on playing as they have been throughout the season.

Thanks for getting back to me; much appreciated.

That spread of 2 - 17 points is a massive one; last season it could have meant Accrington finishing 11th instead of 23rd for example.

I agree, its a guide as to how teams will possibly do if they carry on as they start, but that is a very simplistically low bar to set and anyone can see without complex models that if Team X carry on not scoring enough goals, they will get relegated. (For example as it stands right now, in the Premier League I can predict with a fair degree of confidence that Burnley, Sheffield Utd and Luton are going to get relegated).

Where this model falls down is when external variables come into play; the most obvious (from our perspective) are points deductions but others could be losing the Manager who has got the team playing so well at the start of the season, having your prime striker come down with a season ending injury or even losing 2 or 3 key players to the African Cup on Nations. And thats what makes football so beguiling - the absolutely random nature of success and failure. It's choas theory on steroids.

[Clyde1998]

Is there any empirical evidence for this model? For example, can you retrospectively view last season in League One (and the previous 2 or 3 seasons) at a similar stage and and see a similar pattern as to the eventual relegation outcomes? If not, then (and i am not knocking it) the model is - at best - an interesting exercise but not one to base too much expectation of accuracy on.

On this specific model, I'd have to go back to this stage in previous seasons to give any specific figures.

On using xG, however, the Experimental 361 xG league tables for 46-game seasons (ie. Championship, League One and League Two) has shown an average error of 9.5 points since 2019-20 (2020-21 for L1/L2), with a standard deviation of 7.1 (basically, most teams are within 2 and 17 points of their xG by the end of the season) and a positional error of 3.5 with a standard deviation of 3.1 (so a typical error between 0 and 7 places) - most teams end up broadly at their xG position. Of the 240 observations, only eight have been more than ten positions away from their xG position in that time (3.33%).

Of course as others have pointed out, there are other factors than simply the chances created that will determine a season - significantly better/worse players than average, injuries and mid-season transfers changing the strength of the squad, tactical changes, managerial changes, etc. Outliers are always possible: xG shows the likelihood of an average professional player to score a specific chance - players further from that average, in either direction, will convert chances at a different rate.

Perhaps it's better to think of it as a guide of how teams will do if they carry on playing as they have been throughout the season.

Thanks for getting back to me; much appreciated.

That spread of 2 - 17 points is a massive one; last season it could have meant Accrington finishing 11th instead of 23rd for example.

I agree, its a guide as to how teams will possibly do if they carry on as they start, but that is a very simplistically low bar to set and anyone can see without complex models that if Team X carry on not scoring enough goals, they will get relegated. (For example as it stands right now, in the Premier League I can predict with a fair degree of confidence that Burnley, Sheffield Utd and Luton are going to get relegated).

Where this model falls down is when external variables come into play; the most obvious (from our perspective) are points deductions but others could be losing the Manager who has got the team playing so well at the start of the season, having your prime striker come down with a season ending injury or even losing 2 or 3 key players to the African Cup on Nations. And thats what makes football so beguiling - the absolutely random nature of success and failure. It's choas theory on steroids.

Typically the points spread has a greater variation when sides are more even - as there's more potential for other factors to come into play. It's less of trying to be Mystic Meg, more trying to understand how teams are performing and who may be in a false position compared to the current league table.

I'm only really doing this as a personal project to help with my learning and understanding of probability and statistics. I do want to integrate other factors into the mix, especially player ability. I've previously used Football Manager ratings for players to estimate squad strength (although that reckons we've got the fifth strongest squad in League One this season ).

Advice like this does help as people will invariably come up with ideas, suggestions or feedback that I hadn't considered or will help me improve my skills and understanding - so thank you.

There's a degree of uncertainty with the game that makes it interesting - no-one of us would watch football if we all knew what was going to happen each game.

[WestYorksRoyal]

On this specific model, I'd have to go back to this stage in previous seasons to give any specific figures.

On using xG, however, the Experimental 361 xG league tables for 46-game seasons (ie. Championship, League One and League Two) has shown an average error of 9.5 points since 2019-20 (2020-21 for L1/L2), with a standard deviation of 7.1 (basically, most teams are within 2 and 17 points of their xG by the end of the season) and a positional error of 3.5 with a standard deviation of 3.1 (so a typical error between 0 and 7 places) - most teams end up broadly at their xG position. Of the 240 observations, only eight have been more than ten positions away from their xG position in that time (3.33%).

Of course as others have pointed out, there are other factors than simply the chances created that will determine a season - significantly better/worse players than average, injuries and mid-season transfers changing the strength of the squad, tactical changes, managerial changes, etc. Outliers are always possible: xG shows the likelihood of an average professional player to score a specific chance - players further from that average, in either direction, will convert chances at a different rate.

Perhaps it's better to think of it as a guide of how teams will do if they carry on playing as they have been throughout the season.

Thanks for getting back to me; much appreciated.

That spread of 2 - 17 points is a massive one; last season it could have meant Accrington finishing 11th instead of 23rd for example.

I agree, its a guide as to how teams will possibly do if they carry on as they start, but that is a very simplistically low bar to set and anyone can see without complex models that if Team X carry on not scoring enough goals, they will get relegated. (For example as it stands right now, in the Premier League I can predict with a fair degree of confidence that Burnley, Sheffield Utd and Luton are going to get relegated).

Where this model falls down is when external variables come into play; the most obvious (from our perspective) are points deductions but others could be losing the Manager who has got the team playing so well at the start of the season, having your prime striker come down with a season ending injury or even losing 2 or 3 key players to the African Cup on Nations. And thats what makes football so beguiling - the absolutely random nature of success and failure. It's choas theory on steroids.

I've previously used Football Manager ratings for players to estimate squad strength (although that reckons we've got the fifth strongest squad in League One this season ).

In FM24, you spend your first season beating away Championship offers for Tom Holmes, Mola can hold his own in the Championship if you get promoted, Mukairu is amazing and Vickers belongs in the national league.

I think the lower down the pyramid you go, the less accurate it becomes and plays off reputations.

[Clyde1998]

I've previously used Football Manager ratings for players to estimate squad strength (although that reckons we've got the fifth strongest squad in League One this season ).

In FM24, you spend your first season beating away Championship offers for Tom Holmes, Mola can hold his own in the Championship if you get promoted, Mukairu is amazing and Vickers belongs in the national league.

I think the lower down the pyramid you go, the less accurate it becomes and plays off reputations.

Indeed. I found it to be useful for the Premier League and to some extent the Championship, but the further down you go the less accurate it becomes. I understand younger players who haven't played much first team football are hard to judge, but established players should be relatively simple to rate by using real world data.

Still significantly better than FIFA/EAFC's ratings though - League Two players are usually slow and weak simply because they play in League Two.

[Snowflake Royal]

Is there any empirical evidence for this model? For example, can you retrospectively view last season in League One (and the previous 2 or 3 seasons) at a similar stage and and see a similar pattern as to the eventual relegation outcomes? If not, then (and i am not knocking it) the model is - at best - an interesting exercise but not one to base too much expectation of accuracy on.

On this specific model, I'd have to go back to this stage in previous seasons to give any specific figures.

On using xG, however, the Experimental 361 xG league tables for 46-game seasons (ie. Championship, League One and League Two) has shown an average error of 9.5 points since 2019-20 (2020-21 for L1/L2), with a standard deviation of 7.1 (basically, most teams are within 2 and 17 points of their xG by the end of the season) and a positional error of 3.5 with a standard deviation of 3.1 (so a typical error between 0 and 7 places) - most teams end up broadly at their xG position. Of the 240 observations, only eight have been more than ten positions away from their xG position in that time (3.33%).

Of course as others have pointed out, there are other factors than simply the chances created that will determine a season - significantly better/worse players than average, injuries and mid-season transfers changing the strength of the squad, tactical changes, managerial changes, etc. Outliers are always possible: xG shows the likelihood of an average professional player to score a specific chance - players further from that average, in either direction, will convert chances at a different rate.

Perhaps it's better to think of it as a guide of how teams will do if they carry on playing as they have been throughout the season.

Up to 17 points and 7 places how utterly worthless.

[Snowflake Royal]

I've previously used Football Manager ratings for players to estimate squad strength (although that reckons we've got the fifth strongest squad in League One this season ).

In FM24, you spend your first season beating away Championship offers for Tom Holmes, Mola can hold his own in the Championship if you get promoted, Mukairu is amazing and Vickers belongs in the national league.

I think the lower down the pyramid you go, the less accurate it becomes and plays off reputations.

Indeed. I found it to be useful for the Premier League and to some extent the Championship, but the further down you go the less accurate it becomes. I understand younger players who haven't played much first team football are hard to judge, but established players should be relatively simple to rate by using real world data.

Still significantly better than FIFA/EAFC's ratings though - League Two players are usually slow and weak simply because they play in League Two.

Ever was it thus. In about the 96 version Sheppard and Holsgrove were Premier League quality ffs.

[Clyde1998]

Is there any empirical evidence for this model? For example, can you retrospectively view last season in League One (and the previous 2 or 3 seasons) at a similar stage and and see a similar pattern as to the eventual relegation outcomes? If not, then (and i am not knocking it) the model is - at best - an interesting exercise but not one to base too much expectation of accuracy on.

On this specific model, I'd have to go back to this stage in previous seasons to give any specific figures.

On using xG, however, the Experimental 361 xG league tables for 46-game seasons (ie. Championship, League One and League Two) has shown an average error of 9.5 points since 2019-20 (2020-21 for L1/L2), with a standard deviation of 7.1 (basically, most teams are within 2 and 17 points of their xG by the end of the season) and a positional error of 3.5 with a standard deviation of 3.1 (so a typical error between 0 and 7 places) - most teams end up broadly at their xG position. Of the 240 observations, only eight have been more than ten positions away from their xG position in that time (3.33%).

Of course as others have pointed out, there are other factors than simply the chances created that will determine a season - significantly better/worse players than average, injuries and mid-season transfers changing the strength of the squad, tactical changes, managerial changes, etc. Outliers are always possible: xG shows the likelihood of an average professional player to score a specific chance - players further from that average, in either direction, will convert chances at a different rate.

Perhaps it's better to think of it as a guide of how teams will do if they carry on playing as they have been throughout the season.

Up to 17 points and 7 places how utterly worthless.

90% of sides end up within 7 places and 85% with 17 points. Over half of sides are within two or three places and seven points; over two-thirds of sides are within three of four places and eleven points. Most sides who end up having a huge difference are mid-table sides when the league doesn't have much of gap between sides outside the play-offs and sides outside the relegation zone.

League One last season the difference between xG position and actual position were:

• 0 places - 1 side (Milton Keynes)
• 1 place - 4 sides (Ipswich; Sheff Wed; Portsmouth; Forest Green)
• 2 places - 2 sides (Cheltenham; Morecambe)
• 3 places - 8 sides (Peterborough; Derby; Exeter; Wycombe; Charlton; Bristol Rvrs; Fleetwood; Cambridge)
• 4 places - 3 sides (Plymouth; Barnsley; Lincoln; Accrington)
• 5 places - 1 side (Burton)
• 8 places - 1 side (Port Vale)
• 10 places - 1 side (Shrewsbury)
• 12 places - 1 side (Oxford)

Everyone's mileage will vary.

[Clyde1998]

In FM24, you spend your first season beating away Championship offers for Tom Holmes, Mola can hold his own in the Championship if you get promoted, Mukairu is amazing and Vickers belongs in the national league.

I think the lower down the pyramid you go, the less accurate it becomes and plays off reputations.

Indeed. I found it to be useful for the Premier League and to some extent the Championship, but the further down you go the less accurate it becomes. I understand younger players who haven't played much first team football are hard to judge, but established players should be relatively simple to rate by using real world data.

Still significantly better than FIFA/EAFC's ratings though - League Two players are usually slow and weak simply because they play in League Two.

Ever was it thus. In about the 96 version Sheppard and Holsgrove were Premier League quality ffs.

Are you suggesting Simon Sheppard and Paul Holsgrove weren't Premier League quality players? If only someone had given them a chance.