Correlation between the predicted scores and the actual scores. It's not hard and it's the fairest assessment. I can help you with the maths (or show you how to set up a spreadsheet to do it).

I would like a rule that says, if a player plays fewer than 60 minutes in a particular gameweek, that observation will be dropped from the competition.

I think that the number of "gut instinct" competitors should be limited to 2-3 "experts".

Furthermore, I think you should add a couple of standard FPL formulae to the mix:

- Value (Season)
- Value (Form)

or some simple variation thereof.

I will up the prize on offer to a standard paperback or CD from Amazon for the winner.

I would like to enter the updated version of my formula (the one that now includes "own team rank" following from your suggestion) under the title Wyld-Hancock's Predictor, if that would meet with your approval.

I'm flattered to have my name linked but my contribution was so tiny I think it should stay as your own.

Fair enough. But thank you anyway for your contribution.

I'm bit late to discover this thread, so excuse me if I seem to be going over old ground.

A 10 gameweek comp seems a bit too long to me. Rather than 25 players over 10 weeks, I think a shorter comp would be just as good (and less likely to lose participants). I would suggest instead that it is played over two game weeks, starting gameweek 11, and involve 100 players. That would give 200 observations (actual scores) which should be enough. It would also provide 10 good gameweeks of historical data this season.

Furthermore, to stop this becoming just another quick-pick or guess-the-score type comp I think that:

a) Only one "gut instinct" team should be allowed to compete (maybe the "gut insticters" could form a committee) and that all other entries should be based on a formula or on a stated algorithm or repeatable method (e.g. "I will predict the score by first doing this, then by doing this...").

b) The formula/algorithm/method should be stated clearly before the first match, so that people don't suddenly switch tactics if their method is cr*p. So that, if they see their intial formula was rubbish, they don't just cherry-pick successful methods from better predictors for the next gameweek.

Why?

Surely the test of whether any system adds value isn't how it compares to another system, but how it compares to someone without a system, in which case the more the better.

But I thought the whole point of this was you declared your system would be more accurate at predicting points than any other systematic or discretionary method, now suddenly you are limiting it to just one discretionary player, why? I don't get what it is you are afraid of, surely if a systematic method is the way to go then you should welcome scores of discretionary entries because they will further your case.

If you're just trying to prove you have the best system then fair enough you should measure only against systems, but if you're trying to prove it's better than any method you should compare it to as many discretionary players as systematic ones otherwise any claim of its superiority is totally without merit.

Surely, if there are three discretionary teams, that "method" has three times the chance of winning. I'm just asking to limit the number of entries where the stated algorithm is "I will predict the score based on my experience" to one.

I've read this post 20 times and I still don't see the logic at all. Maybe we have a different view on how we define systems, (for some context I test and use systems in trading markets, and there is an always healthy debate over systematic traders vs discretionary ones) but to use your logic if you have 3 systematic players and 1 discretionary then the systematic would be three times as likely to win.

I think you are making the mistake of thinking that just because the methodology is the same (predict the score based on my experience) that the results will be the same.

They won't.

Each discretionary participant's 'experience' on which he bases his predictions will be as different as each systematic participant's methodology, therefore each discretionary participant should have as much merit as each systematic participant.

Apologies - I've been away on holiday but keen on getting this back on the tracks - maybe starting GW10 as I can see we're going to need a couple of weeks to iron out what players; points system etc

I'll write up initial list of players tonight for consideration

Here is the first list of players split into 3 sections. Currently only midfielders/forwards with the main point scorers for each team. I'll be adding a few defenders but wanted to get people's thoughts on initial list:

Played Previously in Premiership
Arshavin ARS
Song ARS
A Young AVL
Carew AVL
Fergusson BIR
Bowyer BIR
Pedersen BLR
EH Diouf BLR
Lee BOL
K Davies BOL
Malouda CHE
Anelka CHE
Arteta EVE
Cahill EVE
Dempsey FUL
Murphy FUL
Gerrard LIV
Torres LIV
Tevez MC
Johnson MC
Nani MU
Berbatov MU
Etherington STK
Bent SUN
Bale TOT
Crouch TOT
Parker WHM
N'Zogbia WIG
Jarvis WOL

Transfered between Premiership Teams
Ireland AVL
Petrov BOL
J Cole LIV
Milner MC
Jones STK
Piquionne WHM
Fletcher WOL

New players in Premiership
Chamakh ARS
Zigic BIR
Adam BLP
Dembele FUL
Poulsen LIV
Yaya Toure MC
Silva MC
Carroll NEW
Pennant STK
Elmohamady SUN
Van der Vaart TOT
Odemwingle WBA

Although I'm not participating in this debate,I've been following it with interest nevertheless and agree with Broadsword

Suppose there are a handful of discretionary participants. They will all end up with different final scores: some will be better than others. How then do you measure the success of the discretionary method? Is it the average of all discretionary scores? Or is it just the best of them? How do you account for their variablity? If there are nine discretionary entries vs just one systematic entry, and the systematic entry comes second, who won?

As I said, I think my formula will consistently beat a human being...not the entire human race.

As I said before, a 10 week comp is too long. Rather than 25 players over 10 weeks, I suggest 100 players over 2 weeks....just select players who play week-in-week-out.

You will probably lose a few players to injury/rotation, and some players will play fewer than 60 minutes, so these observations would be dropped. You would still have 80-90 players playing both home and away anyway, which should be ample.

So I suggest that the rules are:

Rules
- The game will run in GW10 and GW11.
- 100 player names will be published at the beginning of GW10. These will be the same 100 players that you need to predict the score for in both GW10 and GW11.
- By Friday 11.59pm, UK time GW10, each participant must publish on this thread:
a) a brief description of their score prediction method e.g. "I will guess the score based on my experience" or "I will use the formula x=y*z-3" AND
b) the FPL points score they believe the 100 players will achieve in GW10 AND in GW11. That is to say you need to predict the scores for both gameweeks BEFORE the GW10 matches are played.
- If a player plays fewer than 60 minutes in a match, that score will be dropped from the analysis.
- The winner will be the participant whose predicted scores have the highest correlation with the actual scores.

Surely being able to predict non-starters is part of the challenge?

Of course that could be taken into account, but why complicate matters? I have a formula for predicting a player's score in his next match. I also have a method for predicting non-starters. I could combine both, but isn't this going to be difficult enough as it is? Let's stick to a formula/method for predicting the scores of players who play this first time around. Perhaps we can have another comp for methods of predicting non-starters later.

How about picking 10 head to head matches in the style of I Know The Score?

It may reflect the decisions of cheap defender at home to poor opposition vs mid priced midfielder away to mid ranking opposition. Who do you bench?

Should be easy to score and understand!

Agree with virtually all of the above except 2 weeks is not long enough. From an analysis point of view, I don't think 2 GWs gives an accurate enough result on whether a formula (however that is defined) works or has "got lucky".

I agree that 10 GWs might be too long, but a min of 4GWs is needed to see how the formulas work out - it can also allow people like yourself who have a mathematical prediction method, the ability to tinker with it over a longer period so you can refine and improve it.

You're still making the mistake of lumping all the discretionary participants together as one method whereas you're treating all systematic participants as individually unique methods. They are all unique. No systematic or discretionary method will be the same. They all have equal merit. Now re-read your post and swap the word discretionary with systematic, or change discretionary and systematic to UK and Rest Of World, it would make no sense to determine the validity of someones method based on where they reside would it? And it doesn't make sense to weight them based on their method either, they are all equally valid.

I agree. I'm no statistician but I'm fairly sure a statistical sample of 2 is not enough to draw a conclusion from, I thought statistical models usually needed a sample of 40 to even become remotely viable, or at least that's the case when testing trading systems, but again I can see that anything beyond 10 would start to get rather tedious for what we are trying to achieve. I fear the whole exercise may just be too impractical.

This is all getting a bit complicated and convoluted, so I'm withdrawing

(well that and I've tested my formula over the past couple of GWs and it's rubbish so I'm saving myself the embaressment of proving that in a public trial )

Except it's not really 2 - it's 2 "trials" x number of players in each trial.

Though I agree more weeks could be better, due to correlation effects.

OK.

If you're short of discretionary numbers,you can include mine if you wish. After all, I don't need to do anything do I ?
You won't be requiring a detailed analysis of my thought processes or anything like that I'm assuming. If so,I'm out