GW7 GPG predictions

classic Classic list List threaded Threaded
11 messages Options
Reply | Threaded
Open this post in threaded view
|

GW7 GPG predictions

@shots_on_target
Administrator
I've just ran the GW7 fixtures through my latest model which now inclkudes form indicators like creativity, resilience, etc. See post. .  It's ready rough atm so much better buried away in a forum for now.

I'm going to llist SuperGrover's predictions from comments  on this post, then my current model predictions (which are very similar to SuperGrovers) and then those my latest models - which are well out there, but it's time to starting testing these results. I'll come back after the weekend to evalute all this.

ARS 1.3 1.5 0.8
AVL 0.8 0.8 1.6
CHE 2.1 3.0 1.8
EVE 1.7 1.6 2.5
FUL 2.1 2.0 2.9
LIV 1.9 1.4 0.8
MCI 2.3 2.3 1.4
MUN 1.2 1.3 1.7
NEW 1.3 1.8 1.6
NOR 1.0 0.8 0.5
QPR 1.0 0.6 0.9
RDG 1.1 1.3 1.7
SOT 2.3 2.4 1.4
STO 0.8 0.7 1.5
SUN 0.2 0.4 0.9
SWA 2.3 2.4 1.6
TOT 1.9 2.0 1.3
WBA 2.2 2.3 1.9
WHM 1.3 1.4 1.9
WIG 1.6 1.7 0.9
Reply | Threaded
Open this post in threaded view
|

Re: GW7 GPG predictions

AA23
I would just like to add mine into the mix.

Crude results, taking expected GPG to the nearest whole number gives me expected results of -

Manchester City               3 vs 0 Sunderland
Swansea                              2 vs 1 Reading
Chelsea                              2 vs 1 Norwich
West Bromwich Albion 2 vs 1 Queens Park Rangers
Wigan                              1 vs 2 Everton
West Ham               1 vs 1 Arsenal
Southampton               2 vs 3 Fulham
Tottenham               2 vs 1 Aston Villa
Liverpool                              1 vs 1 Stoke
Newcastle United               1 vs 2 Manchester United

Note - The results are different if you make say 1.8 expected goals mean that they will only score 1 as it doesn't reach two. I tend to take 0.97 to mean that they will score, but essentially if the whole number isn't reached then some might argue we can't "expect" that goal.

In this instance the results are -

Manchester City               3 vs 0 Sunderland
Swansea                              2 vs 0 Reading
Chelsea                              1 vs 0 Norwich
West Bromwich Albion 1 vs 1 Queens Park Rangers
Wigan                              1 vs 2 Everton
West Ham               1 vs 1 Arsenal
Southampton               2 vs 2 Fulham
Tottenham               1 vs 0 Aston Villa
Liverpool                              1 vs 1 Stoke
Newcastle United               1 vs 1 Manchester United

My method as applied to Manchester City vs Sunderland.

Man City
Average SoT (home) = 7
Conversion Rate = 1 in 3
Average shots conceded per game by away team = 20.7 (scale factor equates to 1.3 which is over league average).
Expected Goals (Man City) = (7 / 3) *1.3 = 3.03

Sunderland
Average SoT (away) = 1.7
Conversion Rate = 1 in 3
Average shots conceded per game by home team = 10.7 (scale factor equates to 0.8 which is under league average).
Expected Goals (Sunderland) = (1.7/ 3) *0.8 = 0.45

Any thoughts or criticism is welcome.
Reply | Threaded
Open this post in threaded view
|

Re: GW7 GPG predictions

@shots_on_target
Administrator
First thanks so much for getting involved!  We can only do this (conquer to world that is) by working together.   Who are you?  On Twitter?

My thoughts on your analysis are:
1.  Very good & very similar method to how I'm doing mine, and similar results.

2.  How are you avg. a team's SOT and S?  Just this season or combined with last season?

3.  I don't know how much difference it makes but instead of scaling as you do by comparing to league aveage shots conceded and using a multiplication factor (1.3 in your City example) I use standard deviation from the league average.  

4. How come you use attempts in total to scale, and not SoT?  Many teams just like Sunerland defend very deep with 2 banks of 4 on their 18yd box.  They are happy to let the oppositon have the ball and have pot shots but they prevent any real good chance (ie SoT, SinBox). Look at Sudnerland's first game against Arsenal - Arsenal had 20+ shots but only 3 on target.  Score: 0-0.

5.  I don't know if you're right or wrong to round up or round down the values you get.  I probaly would not suggeting rounding down just to get an interger.  The predicted figure would be the average over many games played, so a 1.8 predicted game played 5 times over the team would score something like: 1, 2, 1, 3, 2 (Avg. 1.8).  In this case you could take the mode or use poission distribtion to pick the most likley result - chance of 0 goals = 17%, chance of 1 goal = 30%, chance of 2 goals = 27%, chance of 3 goals = 16%, chance of 4 goals = 7%, and so on.  So suprisngly 1 goal is the most likley outcome.   I haven't incorporated poisson distribution into my model yet but I will be soon.

Thanks again for joining the discussion!


Reply | Threaded
Open this post in threaded view
|

Re: GW7 GPG predictions

AA23
Hi Ste,

I'm Grant or Arseshavin from FISO.  I don't use Twitter for FPL purposes so for the moment I'm only going to be kicking around the forums.

Very crude method, done it in 45 minutes based upon what I could find.

2. I just used this seasons at the moment.

3. I was going to use the SD, but thought that by using a scale factor it would do a similar job.  It was late and I didn't know how to work the SD into the formula. :-)

For instance league home average SoTpg is 4.9 and City have 7 on average per home game.  It was simply easier to apply it 7/4.9=1.4 to the model.  Should be very similar either way, as it just scales the expectations up or down based on the teams past performance.

4. I was going to use SoT and had reservations about using overall shots but it was all that was on WhoScored.com :-)  Might be more of a feeling than statisticially sound, but if Sunderland are only conceding 3/4 SoTpg then that's assuming pure discipline and if it's a 1 in 3 conversion then I don't know if I'd be comfortable with Sunderland expecting to only concede 1 a game at most.  They might concede 7 SoTpg and still concede 0 though.

It was more to get on the ladder and get a dataset for myself.  Thanks for the feedback, great forum for discussion.

AA23
Reply | Threaded
Open this post in threaded view
|

Re: GW7 GPG predictions

@shots_on_target
Administrator
Cool, thanks Grant/Arshavin.  The first step in our mission here has to be to predict the team results first, and then look at the players.  GL this weekend.
Reply | Threaded
Open this post in threaded view
|

Re: GW7 GPG predictions

AA23
Thought I was quids in when the City game finished 3-0!

Taking the first ones (rounded expected goals to nearest whole number) I predicted 6 out of 12 teams scoring the correct amount of goals, with one exact correct result and three correct outcomes.

Conclusion - the sample is too small... ;-)
Reply | Threaded
Open this post in threaded view
|

Re: GW7 GPG predictions

@shots_on_target
Administrator
The sample is always too small :)
Reply | Threaded
Open this post in threaded view
|

Re: GW7 GPG predictions

@shots_on_target
Administrator
Right, I'll do mine, although not being whole numbers it may not be as simple.

WHM  1.7 / 1 - 3 / 1.5 ARS  - called this one wrong, underestimated Arsenal
CHE 3.0 / 4 - 1 / 0.8 NOR - I'll take this as a good call
WIG 1.7 / 2 - 2 / 1.6 EVE - also a good call
MCI 2.3 / 3 - 0/ 0.4 SUN - also
WBA 2.3 / 3 - 2/ 0.6 QPR - underestimated QPR's goals but they prob. got lucky.
SWA 2.4 / 2 - 2 /1.3 RDG - same as above, I think scoreline flatters Rdg

Total goals predicted:  19.6  Actual: 25

Overall pretty chuffed, as you should be too.  




Reply | Threaded
Open this post in threaded view
|

Re: GW7 GPG predictions

@shots_on_target
Administrator
SOT 2.4 \ 2 - 2 \ 2.0 FUL  - happy with prediction
LIV 1.4 \ 0 - 0 \ 0.7 STO - need to factor in Stoke's obvious resilience 9can be done with the stats!)
TOT 2.0 \ 2 - 0 \ 0.8 - perfect!
NEW 1.8 \ 0 - 3 \ 1.3 MUN  -WAY OFF. See below.

I think this is going to be a major challenge for us.  Teams like Utd suddenly hit form and when the do - boom!  It may require looking at their players available and trying to factor in the extra class that someone like Rooney adds to the team, and the impact it has when he doesn't play.  At the very least our model, or whatever we use, should be able to react quickly to this kind of form explosion, and not take several weeks to "catch-up".

One final point - it's ludicroud to expect we can actually predict the scores - we'd be on a beach in the bahamas by now ;) but - I think it's important to predict the results relative to one and another, and I think it's vital that predictions are very accurate over a larger number of games, e.g. 3-4 weeks.

What do you guys think?
Reply | Threaded
Open this post in threaded view
|

Re: GW7 GPG predictions

SuperGrover
How did I miss this thread?  Too much beer I think!

Since you guys did yours, I guess I should do mine:

Pred / Actual

ARS 1.34 (3...they really looked impressive here)
AVL 0.75 (0...no surprise)
CHE 2.13 (4...NOR is struggling right now.  They aren't as bad as they've been the past two weeks.)
EVE 1.70 (2)
FUL 2.14 (2)
LIV 1.87 (0...Liverpool continues to baffle me.  How a team can have 18 shots, 10 in the box and only 2 on target is baffling)
MCI 2.30 (3)
MUN 1.22 (3.  Probably underestimated Man U's offensive presence a bit)
NEW 1.28 (0.  Should have had 1 probably.  14 shots but only 2 on target)
NOR 0.96 (1)
QPR 0.98 (2...on 3 SOT.  Feel good about this one)
RDG 1.05 (2...on 4 SOT.  See above)
SOT 2.27 (2)
STO 0.84 (0)
SUN 0.22 (0...would have bet my house MCI would get a CS Saturday)
SWA 2.29 (2)
TOT 1.87 (2)
WBA 2.16 (2)
WHM 1.26 (1)
WIG 1.56 (2)

Not too bad.  Liverpool continues and will continue to break any model I construct.  You simply cannot predict a team will have that many shots and that many shots in the box yet continuously miss the target.  They did it last year too.  I wonder if it is Suarez?  

BTW, welcome AA23.  BTW, I show up as John Doe 2008 when posting via Google blogger.

Let's do this again next week!  This was a hoot!
Reply | Threaded
Open this post in threaded view
|

Re: GW7 GPG predictions

SuperGrover
In reply to this post by @shots_on_target
@shots_on_target wrote
I think this is going to be a major challenge for us.  Teams like Utd suddenly hit form and when the do - boom!  It may require looking at their players available and trying to factor in the extra class that someone like Rooney adds to the team, and the impact it has when he doesn't play.  At the very least our model, or whatever we use, should be able to react quickly to this kind of form explosion, and not take several weeks to "catch-up".
I feel that is the version 2 (or more likely 3) of the model.  The first is to get really good at understanding what underlying activities a team needs to score.  SOT is a simple means to an end (and fairly accurate as well) but we need to dig deeper as they are simply too variable. Something like hitting the WW more often than one should may skew SOT substantially (e.g. Liverpool all last season).  How can we eliminate that noise to get to the heart of team quality?

Once that is done, the next step is to try and ascertain how players add to those fundamental activities.  Adding Wayne Rooney certainly helps, but why exactly?  He creates more chances certainly, but is that because he is a better passer or simply better able to put himself in position in which making the pass is easier?  Once we answer that, we need to figure out how opposition impacts those behaviors.  A team like Stoke might be very good at cutting off passing lanes even if space is allowed.  That might impact someone who is a mediocre passer but always finds himself in open space.  On the other hand, that same aspect might not impact someone who generates chances by making accurate passes in tight spaces.  He is used to threading the needle in his passing lanes and won't be impacted by Stoke's defense at all.

I will say that the sport of soccer is extremely complex.  As an American, I can tell you the complexity is similar to that of American football.  Sabermetrics in that sport is still on it's beginnings, but folks have a pretty good understanding of how teams score (and prevent scoring) and are beginning to try and determine how individual player characteristics affect team performance.  I think, in time, football/soccer will get there as well.  Hopefully, folks like us can be part of the movement that achieves it!