Storeytime 2015/16

I have been thinking about this (I need to be getting out more, obv) and I now reckon we need to build some 'tolerance' into these models.

So, divide the sodding League into 2, which looks something like

Group A

Chelsea
City
Arsenal
Utd
Spurs
Liverpool
Swansea
Stoke
Everton
Southampton

Group B
(The rest - 10 teams)

A v A Home 2pts prediction (9 games for A x 2pts, 18/27 for these games
A v B home 2.5pts prediction (10 games for A x 2.5pts 25/30) for these games

42/57 pts = 14 wins from 19 home games, a decent-ish haul if not outstanding

Away

A v A Away 1pt prediction (9 games for A x 1pt, 9/27 for these games, which isn't great but we're building a theoretical model for comparison)
A v B Away 2pt prediction (10 games for A x 2pts, 20/30 for these games, 7 wins from 10 ish - passable form is that.

Which would give us, for an A side, 18 + 25 + 9 + 20 = 72pts as the famous PAR score which is 4th place-ish is it not?

Now, if we track all that, it contains some tolerance for losing silly games because every 3pts is a positive result of some kind (seems fair) and this model also has the requirement of needing points vs. rivals but waters down the 'disaster' of losing to them very slightly, but you still gain from beating them.

It rewards good results (which is any win, innit, basically) but weights fixtures in a moderate fashion.

gives us unsorted Current standings after 1 game of

Liverpool + 2 to finish on 74pts
Utd + 1 to finish on 73pts
Chelsea -1 to finish on 71pts
Arsenal - 2.5 to finish on 69.5 pts (unlikely, obv)
Spurs -1 to finish 71pts
City plus 1 to finish 73pts

Increasing guaranteed accuracy as the season develops...
 
Last edited:
If you want to do the full League, the bottom half (Group B) have that possible pesky point discrepancy which is not worth bothering about imo, just let them have it, none of them is gonna finish in the top 4 anyway which is where our real concern lies.

There we are. Sorted.
 
Eboue said:
It's never going to be good if you divide it into groups. Why is a match against 10th worth the same as 1st but different than 11th? Any dividing into groups messes it up.
Click to expand...

I suppose it defines the fixture imbalance. Or is a guess at doing this.

edit - I've got my fixture numbers wrong anyway but I'm not going through 'em again.
 
Eboue said:
But Stoke should be much closer to West Ham and much farther from Chelsea.
Click to expand...

I know what you mean but it's a trade off between perceived difficulty and easing out the result anomalies, I think...
 
Eboue said:
If someone can find a good site to scrape data from I should be able to do it fairly easily.
Click to expand...
I don't know how anyone else did it, but I copied the 20x20 table from wikipedia, then changes 3 for a home win, 1 for a draw and 0 for a home loss, then let the maths do the work.

Bloody pain.
 
Eboue said:
What are the anomalies in my method? I'm not saying there aren't any, I just want to hear what you think they are.
Click to expand...

I think it was the actual strangeness of the results that threw yours out. (for Utd, anyway)
 
Sied said:
But what's the point? What are we working out here?
Click to expand...
:lol:My view entirely. We can see that United can easily go on a run of beating Liverpool or Arsenal 4-5 games in a row and then lose the next few versus them. I guess if it comes some people off the streets, no harm done.
 
Scrumpet said:
@Damien I think we also have to consider one Middle 8 team a Bottom 6 team for other Bottom 6 teams (as well as considering Liverpool a Middle 8 side for all non-Top 6 teams). I didn't bother last year because I wasn't doing scores for teams that low down the table.

The way it is at the moment, both Top 6 and Middle 8 sides play 5 Top 6 teams, 8 Middle 8 teams and 6 Bottom 6 teams, but Bottom 6 teams play 5 Top 6 teams, 9 Middle 8 teams and 5 Bottom 6 teams, so their par over the course of the season is 2 points lower than the other 14 sides'.
Click to expand...
Easy enough to do with Leicester City. I'll edit the OP later
 
I've done a quick one with top 4, next 8, bottom 8, with United counting as next 8 for all non-top 4 teams, and West Ham counting as bottom 8 for all bottom 8 teams, with a slight adjustment to the points expectations. Top 4 - draw at home, lose away, Next 8 - win at home, draw away, bottom 8 - win home, win away. With this, the GW1 differences are as follows:
  • City remain on 0 rather than having +2 after beating West Brom away
  • Spurs are on -1 rather than 0 after losing to United away
  • West Brom are on -1 rather than -3 after losing to City at home
 
I can correct & update to reflect our fantastic win last night.

Liverpool +2 to finish on 73.5 pts
Utd +2 to finish on 73.5 pts
City plus1 to finish on 72.5 pts
Chelsea -1 to finish on 70.5 pts
Spurs -1 to finish on 70.5 pts
Arsenal -2.5 to finish on 69 pts
 
Peterstorey was definitely a smart guy, think he just trolled the kids a bit too much. Think he secretly was never the same after Plech left. You live on in this thread though buddy.

We have newcastle at home next which we should win but that's followed by a tough three game stretch in swansea and southampton away and liverpool at home. If we're 6/6 by then I think we have a real chance to compete for the title this year.
 
montpelier said:
I can correct & update to reflect our fantastic win last night.

Liverpool +2 to finish on 73.5 pts
Utd +2 to finish on 73.5 pts
City plus1 to finish on 72.5 pts
Chelsea -1 to finish on 70.5 pts
Spurs -1 to finish on 70.5 pts
Arsenal -2.5 to finish on 69 pts
Click to expand...

Surely Villa are down as bottom 6, and therefore we're still on zero?
 
Damien said:
Copied from Scrumpet's thread, if any logic is wrong blame him.

As always the idea is to negate the impact of easy/hard fixture lists to see where teams are really at relative to each other. Par is to win the league with 83 points. To do that you need to:

Win all your home games: 57 points
Lose away against the Top 6: 0 points
Draw away against the Middle 8: 8 points
Win away against the Bottom 6: 18 points

The arbitrarily chosen tiers are:

Top 6: Chelsea, Manchester City, Arsenal, Manchester United, Tottenham Hotspur, Liverpool.*
Bottom 6: Newcastle United, Sunderland, Aston Villa, AFC Bournemouth, Watford, Norwich City.
Middle 8: Southampton, Swansea City, Stoke City, Crystal Palace, Everton, West Ham United, West Bromwich Albion, Leicester City.

*To make sure par is the same for everybody, I considered Liverpool a Middle 8 side for non-Top 6 teams.

After one game the scores for the twenty teams I did are:

+3: West Ham United
+2: Liverpool, Manchester City
+1: Swansea City
0: Aston Villa, Crystal Palace, Leicester City, Manchester United, Tottenham Hotspur, Watford
-1: Sunderland
-2: Chelsea, Everton, Newcastle United, Southampton
-3: AFC Bournemouth, Arsenal, Norwich City, Stoke City, West Bromwich Albion
Click to expand...


this highly hypothetical & unrealistic assumptions make this methodology with very weak prediction power

if you want to segregate the league in three sections: top, mid & bottom, then don't take the home games vs away games into equations. otherwise, further expand the analysis into six parts.

& also you should at least use the past data of the champions to estimate the expected values of the various scores. such a normative assumption basically reduces the validity of the approach to zero

 
the past data of the champions to estimate the expected values of the various scores
Click to expand...

83 pts seems a pretty good estimation of the Champs' points haul, altho altering this anywhere 'sensible' doesn't affect much as it's the relativity you're after tracking, altho (I think) you want it to be in the range of 70-88 to keep the + / - relevant to the top group we're interested in. I don't think a 90pt target figure does this, everyone just proceeds through the season clocking up minus scores, with scope to regularly + / - 6pts a game when the top 6 are playing each other <--- too volatile

So, I would argue that this is too inflexible and the middle 8, bottom 6 split doesn't really matter unless it serves to create the tolerance within the model which I don't really know tbh, and can't be bothered to think about. But the 57 & 0 are definitely WRONG, I mean they don't help - saying this in the nicest possible way, thing is too volatile with this in place.

Win all your home games: 57 points
Lose away against the Top 6: 0 points
Draw away against the Middle 8: 8 points
Win away against the Bottom 6: 18 points
Click to expand...
 
montpelier said:
83 pts seems a pretty good estimation of the Champs' points haul, altho altering this anywhere 'sensible' doesn't affect much as it's the relativity you're after tracking, altho (I think) you want it to be in the range of 70-88 to keep the + / - relevant to the top group we're interested in. I don't think a 90pt target figure does this, everyone just proceeds through the season clocking up minus scores, with scope to regularly + / - 6pts a game when the top 6 are playing each other <--- too volatile

So, I would argue that this is too inflexible and the middle 8, bottom 6 split doesn't really matter unless it serves to create the tolerance within the model which I don't really know tbh, and can't be bothered to think about. But the 57 & 0 are definitely WRONG, I mean they don't help - saying this in the nicest possible way, thing is too volatile with this in place.
Click to expand...

83 is okay is this is basically the average scoreline of the past champions.
the same should do to the score against top/mid/bottom teams of the past champions to estimate the expected value. and to expand it further if home/away games need to be taken into consideration
 
Win all your home games: 57 points
Lose away against the Top 6: 0 points
Draw away against the Middle 8: 8 points
Win away against the Bottom 6: 18 points

this schedule is completely meaningless
 
rover said:
Win all your home games: 57 points
Lose away against the Top 6: 0 points
Draw away against the Middle 8: 8 points
Win away against the Bottom 6: 18 points

this schedule is completely meaningless
Click to expand...
It's like a par in golf. Classic.
 
To reverse this around, I'm gonna tentatively say that ideal points figure to track against is 78pts and then most of the top 6 contenders should be within +8 (finish on 86) or -8 (finish on 70). How you decide to create a 78pt par, with some built in tolerances I'm not quite sure.

10ish home games @ 2.5 pts Teams do have to be 'allowed' to drop home points or gain the slowly accumulating advantage of not doing.
9ish home games @ 2 pts

some away games @ 2 pts
some away games @ 1pt

would still seem pretty good to me

in MY model winning AWAY at a rival was worth a 4pt swing, beating a rival at home was worth a 2pt swing in RELATIVE terms - these seem reasonable to me relative to their value in the real world which is 3, obv.
 
An idea...

Instead of using the top 8, bottom 6 and middle teams as pre determined groupings, why allocate these before each match day based on the table. That way you build into the model a mechanism to cope with teams being above or below expectations.

The other idea I had linked to this could be to allocate the groups based on theit recent form rather than overall position. So if West Ham were away at Chelsea, yet Chelsea had lost 3 and drawn 3 of their last 6 games, Chelsea might have now slipped into the middle grouping and you would expect West Ham to get a draw, assuming West Ham were in form and we're classed as a current top 6 on form.
 
antsmithmk said:
An idea...

Instead of using the top 8, bottom 6 and middle teams as pre determined groupings, why allocate these before each match day based on the table. That way you build into the model a mechanism to cope with teams being above or below expectations.

The other idea I had linked to this could be to allocate the groups based on theit recent form rather than overall position. So if West Ham were away at Chelsea, yet Chelsea had lost 3 and drawn 3 of their last 6 games, Chelsea might have now slipped into the middle grouping and you would expect West Ham to get a draw, assuming West Ham were in form and we're classed as a current top 6 on form.
Click to expand...

That's quite an interesting idea and probably isn't too far away from what actually happens in the real world. But, we want the framework of the model to be set in stone and then track the changes against that, to try and generate a prediction of the way forward afterwards so it doesn't help us. (imo)

If our (let's say MY) model was perfect and the real world followed it, everyone would finish on 78pts. Obv, that doesn't happen which is where all the + / - nonsense comes from. You'd be eliminating that. (I think... and if I'm understanding what you say etc)
 
rover said:
this highly hypothetical & unrealistic assumptions make this methodology with very weak prediction power

if you want to segregate the league in three sections: top, mid & bottom, then don't take the home games vs away games into equations. otherwise, further expand the analysis into six parts.

& also you should at least use the past data of the champions to estimate the expected values of the various scores. such a normative assumption basically reduces the validity of the approach to zero
Click to expand...
All I can do is my best.
 
Damien said:
Copied from Scrumpet's thread, if any logic is wrong blame him.

As always the idea is to negate the impact of easy/hard fixture lists to see where teams are really at relative to each other. Par is to win the league with 83 points. To do that you need to:

Win all your home games: 57 points
Lose away against the Top 6: 0 points
Draw away against the Middle 8: 8 points
Win away against the Bottom 6: 18 points

The arbitrarily chosen tiers are:

Top 6: Chelsea, Manchester City, Arsenal, Manchester United, Tottenham Hotspur, Liverpool.*
Bottom 6: Newcastle United, Sunderland, Aston Villa, AFC Bournemouth, Watford, Norwich City.
Middle 8: Southampton, Swansea City, Stoke City, Crystal Palace, Everton, West Ham United, West Bromwich Albion, Leicester City.*

*To make sure par is the same for everybody, I considered Liverpool a Middle 8 side for non-Top 6 teams and Leicester City a Bottom 6 team for other Bottom 6 teams.

After one game the scores for the twenty teams I did are:

+3: West Ham United
+2: Liverpool, Manchester City
+1: Swansea City
0: Aston Villa, Crystal Palace, Leicester City, Manchester United, Tottenham Hotspur, Watford
-1:
-2: Chelsea, Everton, Newcastle United, Southampton
-3: AFC Bournemouth, Arsenal, Norwich City, Stoke City, Sunderland, West Bromwich Albion
Click to expand...
What if a club in the middle group breaks into the real top 6 at the time you play them do you adjust the table?
 
JustAFan said:
What if a club in the middle group breaks into the real top 6 at the time you play them do you adjust the table?
Click to expand...
I think people have done that before. Two tables, one with the original tiers and one with the actual ones. Might even have been done during Moyes' season.
 
People are getting carried away here with a quest for accuracy. It doesn't matter.

It doesn't matter if 83 points is really what the winners will get, it doesn't matter if the teams stated as bottom 6 are really in the bottom 6. It simply matters that every team is judged against the same criteria, so their relative performance can be gauged.

Putting tons of effort into improving the formula is pointless.
 
You must log in or register to reply here.