Re-imagining Storeytime

Eboue

nasty little twerp with crazy bitter-man opinions
Joined
Jun 6, 2011
Messages
62,627
Location
I'm typing this with my Glock 19 two feet from me
Now with math(s)!

I've always liked the concept but I could never really get on board with it because it had so much subjectivity in determining who went in which group and in how large each group was. I had an idea to apply some math(s) to it and get an idea about how each team is doing according to expectations. I'll just do the top 6, according to my expected points calculations. The expected points adjust for which teams are over or under performing their goal differential using Pythagorean expectation, a concept that originated in baseball and has since been expanded to other sports by people smarter than I.

QFJiVFS.png


Those numbers aren't really relevant to the Storeytime table but I feel they are probably the best numbers to look at when looking at team strength. Pythagorean standings tend to be better predictors than actual standings.

This next table will be used to do the actual Storeytime calculations.

fYXgPJU.png


I'll break down Manchester United's Storeytime calculations to show how I do it.

First match was at Goodison Park, so looking at the table, teams playing Everton away can expect 0.64 points. Manchester United lost 1-0 and received 0 points. So far, -0.64 points from expectation. Following this, Manchester United hosted Fulham. Teams hosting Fulham can expect 1.71 points. Manchester United won 3-2 and received 3 points. So after 2 matches, Manchester United are (1.71+-0.64) 0.53 points ahead of expectation

Here is the table.

xyQhw0z.png


So far, Manchester United are 31 points ahead of what an average team would expect given their fixtures.
 
Conclusions

  • Chelsea are still in quite a good position for a top 4 finish
  • Liverpool and Arsenal will need a major turnaround to make a CL spot
  • Spurs were struggling until a recent run of good form
  • United are far and away above the pack


I also want to point out the Di Matteo and Benitez results

Under Di Matteo:

12 matches - +6.97 - +0.58 per match

Under Benitez:

16 matches - +7.74 - +0.48 per match
 
How does using goal difference tells you how strong a team is on average? Mourinho's Chelsea was comfortably better than the rest of the league and never looked in trouble when they decided to shut things out by keeping games at 1-0s. Won't using your method find them to be overachieving despite the fact that they're comfortably the best team in the league at that time?
 
elmo said:
How does using goal difference tells you how strong a team is on average? Mourinho's Chelsea was comfortably better than the rest of the league and never looked in trouble when they decided to shut things out by keeping games at 1-0s. Won't using your method find them to be overachieving despite the fact that they're comfortably the best team in the league at that time?
Click to expand...

The first table (with expected points) doesn't use pure goal difference. If a team has 80 goals for and 40 allowed, it will be predicted to achieve less points than a team with 60 goals for and 20 goals allowed.

The actual Storeytime predictions don't use that method, I just wanted to post it because it is interesting and isn't worthy of its own thread. The actual story time methods use actual results. So since Manchester United has gone 13-0-1 at home, opponents have averaged 0.21 ppg (3/14).

Edit: Here is the formula I used for the Pythagorean expectation.

http://pena.lt/y/2012/12/03/applying-the-pythagorean-expectation-to-football-part-two/
 
The Neviller was raised by two gay dads. Hope you didn't break them up.

I do like the math. I'm in no condition to comprehend it at the moment.
 
Your method is no less arbitrary than the actual storey time. Just because more maths has gone into it doesn't mean its more accurate.
 
The storey system is actually better, because it is aimed at what it takes to win the league and how far a team is off that. For all the maths, the 'expected points' for united and any other top team is way more than the model. It also appears to account better for the big difference between home and away.
 
Eboue posturing for resident Gooner a single day after Pete's departure. Seems a tad tasteless.

In other news, I'm sure many will appreciate this. I got B at Maths GCSE, so I'm out.
 
I don't see any explanation for how you have decided the 'expected' points against any particular opponent.

Teams playing Everton away can expect an average of 0.64 points. Why? According to what evidence?

Unless you can tell us where these pars come from (unless I've missed it in my relatively quick scan of your posts) then it's absolutely no less arbitrary than Pete's version was. More complicated doesn't equal more accurate, or more credible.
 
How are the expected points averages for each team home and away calculated? Are they continuously upgraded as the season goes on, meaning for instance that our -0.64 against Everton at Goodison will go down in value if Everton wins more matches at Goodison and improve their expected point average at home? Or is it calculated using last season's results? The former would certainly be the better method.

If you use the latter, how are the bottom teams calculated? If you use the former, does this mean that this method, unlike original storeytime, doesnt really tell us anything untill every team has played a good amount of matches to build up reliable point averages?

I like it btw. Good work.
 
kouroux said:
ffs I hate it when people over complicate football with stats and graphs
Click to expand...

ffs I hate it when people post reactionary nonsense like this. It's not as if I'm saying X pass completion percentage + Y tackles made = Z rating for player. The concept is simple. Each team plays the same fixtures of the course of a season but they don't play them in the same order, which means the table will be misleading as some teams have played a harder set of matches at each point in the table. This adjusts for that.

Interval Level said:
Your method is no less arbitrary than the actual storey time. Just because more maths has gone into it doesn't mean its more accurate.
Click to expand...

Of course it is less arbitrary. The actual storey time just said "these 5 teams are bottom" or "you must win all home matches" or whatever. Why are those 5 teams bottom? Why are there 5 teams in the bottom group? Etc.

My numbers used what actually happened. What has actually happened is that opponents playing at Madejski Stadium have averaged 1.40 points and teams hosting Newcastle have averaged 2.07 points. I'm using what actually happened as the starting point, not an arbitrary group that is decided at the beginning of the season.

sincher said:
The storey system is actually better, because it is aimed at what it takes to win the league and how far a team is off that. For all the maths, the 'expected points' for united and any other top team is way more than the model. It also appears to account better for the big difference between home and away.
Click to expand...

What it takes to win the league is having more points that your rivals. The calculations are for an average opponent so when I calculate a match hosting Wigan, Manchester United and Chelsea both get 1.86 points. Since I have actual data for home and away, I'm not sure how it could be better at accounting for that.

BazzaBear said:
I don't see any explanation for how you have decided the 'expected' points against any particular opponent.

Teams playing Everton away can expect an average of 0.64 points. Why? According to what evidence?

Unless you can tell us where these pars come from (unless I've missed it in my relatively quick scan of your posts) then it's absolutely no less arbitrary than Pete's version was. More complicated doesn't equal more accurate, or more credible.
Click to expand...

This was explained in the opening posts but I'll explain it again.

Everton at home have 7 wins, 6 draws and 1 loss. That's ((7*0)+(6*1)+(1*3))/14 = 0.64. That's the evidence. Teams playing Everton away have averaged 0.64 points so that is the expected points for that fixture.


Brosstan said:
How are the expected points averages for each team home and away calculated? Are they continuously upgraded as the season goes on, meaning for instance that our -0.64 against Everton at Goodison will go down in value if Everton wins more matches at Goodison and improve their expected point average at home? Or is it calculated using last season's results? The former would certainly be the better method.

If you use the latter, how are the bottom teams calculated? If you use the former, does this mean that this method, unlike original storeytime, doesnt really tell us anything untill every team has played a good amount of matches to build up reliable point averages?

I like it btw. Good work.
Click to expand...

My plan was to do it every month or so. At the beginning of the season, I would use last year's results for a month or two until we've got enough data.

You are correct that this would be less informative at the beginning of the season, but after a decent chunk of data, this becomes far more accurate.


ciderman9000000 said:
After the Norwich game City were on +12.67pts, they played Stoke next, expected 2pts and won 3pts, giving them them a +1pt difference; yet their running total as you've recorded it incorrectly remains at +12.67pts.
Click to expand...

Fixed. Moves them up 1.00.
 
What data do you use to build the model? The problem is by definition you can't get enough reliable data to build a good model based on real results without it being significantly out of date.
 
The last ten games clearly is not enough data to work out 'expected points', nor much else.

I think the Pythagorean goal difference measure is pretty good as a measure of relative strength, but I think the expected points data is questionable.
 
sincher said:
The last ten games clearly is not enough data to work out 'expected points', nor much else.

I think the Pythagorean goal difference measure is pretty good as a measure of relative strength, but I think the expected points data is questionable.
Click to expand...

Well I've automated most of it now so I can update it easily after every game. The ten games I was referring to were at the start of the season. Obviously the more data we have, the more accurate it will be. I think you are right that the actual storey time table will be more helpful at the beginning of the year but mine will be more helpful at about the halfway point from then on.
 
Eboue... said:
This was explained in the opening posts but I'll explain it again.

Everton at home have 7 wins, 6 draws and 1 loss. That's ((7*0)+(6*1)+(1*3))/14 = 0.64. That's the evidence. Teams playing Everton away have averaged 0.64 points so that is the expected points for that fixture.
Click to expand...

Righto - I've got you now. So you're using the actual results so far in the premiership games?

By this stage in the season then, I agree - a better (albeit massively more complex) system. But it would be completely useless for the first 10-20 games of each season.

The weighting that you have applied could be more or less achieved much more simply by just using the Storeytime system but using the actual current positions for the weightings instead of a pre-season guess at who was going to end up where.
 
BazzaBear said:
Righto - I've got you now. So you're using the actual results so far in the premiership games?

By this stage in the season then, I agree - a better (albeit massively more complex) system. But it would be completely useless for the first 10-20 games of each season.

The weighting that you have applied could be more or less achieved much more simply by just using the Storeytime system but using the actual current positions for the weightings instead of a pre-season guess at who was going to end up where.
Click to expand...

Then you still have a few problems.

1) Some teams have big variations in performance home and away and others don't. Stoke averaged double the points home compared to away. Spurs average 0.14 more points at home than away. QPR actually do better away.

2) The original storeytime puts teams into arbitrary groups (the size of the groups). What if there is one team far below the rest, like when West Ham were relegated a few years ago? What if there is one team far above the rest like United this year? Should United be weighed the same as Spurs since they are both in the top 4?
 
It is just hard to think that Arsenal have 11 points more than 'expected'...
 
sincher said:
It is just hard to think that Arsenal have 11 points more than 'expected'...
Click to expand...

Well it's more 11 points more than what those fixtures usually yield for the league as a whole. I wouldn't think of it as "points +/- expected" because that would mean we expect QPR to have 8 points.

The Arsenal results are pretty much in line with their competitors.
 
But there is an attempt to work out expected points per team... Which looks good as a relative measure but shit as an absolute one. In this respect the storey system is miles better, as a way to measure how far above or below par a a team is according to how much it takes to win, come 4th etc.
 
I don't think I'm making my point clear. It's expected points per fixture, not per team. It's not designed to be an absolute measure, because after 38 games it will be irrelevant (since everyone will have the same schedule at that point). What it does is neutralize the scheduling differences.
 
Interesting on relegation, suggests Sunderland are not quite safe yet.
 
There must be something wrong with your United graph, no chance we could jump about 12 points in two weeks. It's just not possible.
 
You must log in or register to reply here.