Fantasy Scout RSS Ranking - October 2009
milivella
2. ( =) Enzo 3.40 (+0.40)
3. (-1) Andrea V. 3.32 (+0.32)
4. ( =) Sid 2.50 (+0.50)
5. (+1) Michael H. 2.10 (+0.45)
6. (-1) Abubakr 2.00 (+0.25)
7. ( =) William 1.60 (+0.10)
8. ( =) Jesus 1.00 (+0.00)
9. ( =) Mark 0.44 (+0.22)
10. ( =) Tom two 0.00 (+0.00)
(Comparison with the ranking just before tonight's update.)
Discuss! ;)
--
Cheers
milivella
Enzo
How can this dude Andrea go -1 if the other 2 dudes
above him stay rooted?
milivella
Because he's *very* bad!
Really, the ranking was...
2. Enzo 3.00
2. Andrea V. 3.00
and is...
2. Enzo 3.40
3. Andrea V. 3.32
so I actually lose one place (from tied second to third).
Of course, if your question was made just to reveal that you've
overtook me... I'll remind you that 0.08 points is not the largest
margin, and that I'll kick you out of the second place in exactly one
month! ;)
--
Cheers
milivella
Enzo
Ah, ok. Where is Allesandro? Has he left the game? Not that
I am complaining ! But even if he has, is it fair to totally
like erase the dude?
> Of course, if your question was made just to reveal that you've
> overtook me... I'll remind you that 0.08 points is not the largest
> margin, and that I'll kick you out of the second place in exactly one
> month! ;)
>
Whoa! Is this the uber-humble milivella speaking?
> --
> Cheers
> milivella- Hide quoted text -
>
> - Show quoted text -
futbol...@yahoo.com
> 1. ( =) Daniele 3.87 (+0.13)
> 2. ( =) Enzo 3.40 (+0.40)
> 3. (-1) Andrea V. 3.32 (+0.32)
> 4. ( =) Sid 2.50 (+0.50)
> 5. (+1) Michael H. 2.10 (+0.45)
> 6. (-1) Abubakr 2.00 (+0.25)
> 7. ( =) William 1.60 (+0.10)
> 8. ( =) Jesus 1.00 (+0.00)
> 9. ( =) Mark 0.44 (+0.22)
> 10. ( =) Tom two 0.00 (+0.00)
>
I understand this is the ranking only within the group of regular RSS
contributors. I'm not going to last long at the top, especially if my
point-scoring players keep on getting injured just before
international call-ups (Marchisio and Guiza, just to name a
couple...and to add insult to injury, the guy called up to replace
Guiza was Negredo, and he scored two goals!).
Well, at least Gomis made a return to the NT. Now it's only a matter
of waiting for Thiago Neves's comeback!
By the way, in the Excel file I would add a column to the Players-
Matches worksheet indicating which scout the player belongs to. That
way it becomes easier to check who scored well in each international
match day.
D
milivella
> Ah, ok. Where is Allesandro?
He has never written in RSS, so he's not included in the *RSS*
ranking. ;)
But he still tops the general ranking! See
http://fantasyscout.altervista.org/data.htm
> > Of course, if your question was made just to reveal that you've
> > overtook me... I'll remind you that 0.08 points is not the largest
> > margin, and that I'll kick you out of the second place in exactly one
> > month! ;)
>
> Whoa! Is this the uber-humble milivella speaking?
Yep. Otherwise I would have predicted a jump up to the 1st place for
me!
--
Cheers
milivella
milivella
> On Oct 14, 9:57 pm, milivella <milive...@gmail.com> wrote:
>
> > 1. ( =) Daniele 3.87 (+0.13)
> > 2. ( =) Enzo 3.40 (+0.40)
> > 3. (-1) Andrea V. 3.32 (+0.32)
> > 4. ( =) Sid 2.50 (+0.50)
> > 5. (+1) Michael H. 2.10 (+0.45)
> > 6. (-1) Abubakr 2.00 (+0.25)
> > 7. ( =) William 1.60 (+0.10)
> > 8. ( =) Jesus 1.00 (+0.00)
> > 9. ( =) Mark 0.44 (+0.22)
> > 10. ( =) Tom two 0.00 (+0.00)
>
> I understand this is the ranking only within the group of regular RSS
> contributors.
Right. (to keep the good ol' time spirit when FS meant more or less
RSS... ;) )
> I'm not going to last long at the top
Enzo is coming, eh? (I'll set up a new expected-score ranking... just
to see you kicked off the 1st place!)
> especially if my
> point-scoring players keep on getting injured just before
> international call-ups (Marchisio and Guiza, just to name a
> couple...and to add insult to injury, the guy called up to replace
> Guiza was Negredo, and he scored two goals!).
It was just a matter of time, dude.
> Well, at least Gomis made a return to the NT.
With Benzema out of grace, Gomis could have his chance... but Gignac
won't agree.
> By the way, in the Excel file I would add a column to the Players-
> Matches worksheet indicating which scout the player belongs to.
I'll do it, but I've to found the way to do it... :/ (under the
current set-up, I should add the scout name manually, something that
makes little sense... time to learn a bit more of Excel, or to switch
to a db!)
> That
> way it becomes easier to check who scored well in each international
> match day.
Isn't this already obtained tracking the difference current_score -
previous_score (last column of the ranking I've posted)?
I.e. the better performers in October have been:
1. Sid (+0.50)
2. Michael H. (+0.45)
3. Enzo (+0.40)
4. Andrea V. (+0.32)
5. Abubakr (+0.25)
6. Mark (+0.22)
7. Daniele (+0.13)
8. William (+0.10)
9. Jesus (+0.00)
9. Tom two (+0.00)
[...OK, I'm lazy, and I don't want to learn anything new. :) ]
--
Cheers
milivella
Futbolmetrix
news:72ff9813-6f60-47b9-8502-
>> By the way, in the Excel file I would add a column to the Players-
>> Matches worksheet indicating which scout the player belongs to.
> I'll do it, but I've to found the way to do it... :/ (under the
> current set-up, I should add the scout name manually, something that
> makes little sense... time to learn a bit more of Excel, or to switch
> to a db!)
I can help you with that.
> > That
> > way it becomes easier to check who scored well in each international
> > match day.
> Isn't this already obtained tracking the difference current_score -
> previous_score (last column of the ranking I've posted)?
Yes, but I'd like more detail. For example, knowing on which of your and
Enzo's players I should place an ancient curse that guarantees injuries and
bad form.
D
milivella
> "milivella" <milive...@gmail.com> wrote in message
>
> news:72ff9813-6f60-47b9-8502-
>
> >> By the way, in the Excel file I would add a column to the Players-
> >> Matches worksheet indicating which scout the player belongs to.
> > I'll do it, but I've to found the way to do it... :/ (under the
> > current set-up, I should add the scout name manually, something that
> > makes little sense... time to learn a bit more of Excel, or to switch
> > to a db!)
>
> I can help you with that.
Please! (you know my e-mail address...)
To ingratiate you, here is the ranking based on players' estimated
score... :)
1. Enzo 71
2. Andrea V. 40
3. Jesus 34
4. Michael H. 32
5. Daniele 11
6. Sid 9
7. William 8
8. Abubakr 6
9. Mark 2
10. Tom two 0
It's silly (I don't think that Santon will score 500 points!), but we
could work on the formula, and anyway I bet that this ranking and
these scores will be closer to the final ones than the current
official ranking and scores (i.e. Jesus will be closer to the 3rd
place than to 8th, Enzo's score will be closer to 71 than to 3.40).
> Yes, but I'd like more detail. For example, knowing on which of your and
> Enzo's players I should place an ancient curse that guarantees injuries and
> bad form.
Fair reason!
--
Cheers
milivella
Enzo
> Futbolmetrix:
>
> > "milivella" <milive...@gmail.com> wrote in message
>
> > news:72ff9813-6f60-47b9-8502-
>
> > >> By the way, in the Excel file I would add a column to the Players-
> > >> Matches worksheet indicating which scout the player belongs to.
> > > I'll do it, but I've to found the way to do it... :/ (under the
> > > current set-up, I should add the scout name manually, something that
> > > makes little sense... time to learn a bit more of Excel, or to switch
> > > to a db!)
>
> > I can help you with that.
>
> Please! (you know my e-mail address...)
>
> To ingratiate you, here is the ranking based on players' estimated
> score... :)
>
> 1. Enzo 71
> 2. Andrea V. 40
> 3. Jesus 34
> 4. Michael H. 32
> 5. Daniele 11
> 6. Sid 9
> 7. William 8
> 8. Abubakr 6
> 9. Mark 2
> 10. Tom two 0
Even assuming the formula of projection of career stats
based on current stats and age is applicable to every picked
player based on its applicability to a handful of successful
footballers in history, which is by ***no*** means a given ...
It assumes that my future picks will perform like my present picks,
"performa" meaning performance according to this formula (
the projection of career caps and goals based on current values
and age ).
Since in FS, we are forced to continue picking players every year,
this is a big drawback with the formula. The formula must find
some way, if it is going to be making projections into the future,
to account for the probability of performance of future picks.
milivella
> On Oct 15, 1:23 pm, milivella <milive...@gmail.com> wrote:
>
>
>
> > To ingratiate you, here is the ranking based on players' estimated
> > score... :)
>
> > 1. Enzo 71
> > 2. Andrea V. 40
> > 3. Jesus 34
> > 4. Michael H. 32
> > 5. Daniele 11
> > 6. Sid 9
> > 7. William 8
> > 8. Abubakr 6
> > 9. Mark 2
> > 10. Tom two 0
>
> Even assuming the formula of projection of career stats
> based on current stats and age is applicable to every picked
> player based on its applicability to a handful of successful
> footballers in history, which is by ***no*** means a given ...
It isn't. :)
In fact I'd like to make the formula better. But date of birth and
points are the only data - bar the name, which I guess isn't that
useful in this context... - that we have for each picked player, so
the formula can't use more variables. So it seems to me that progress
can come from the following attempts:
1. Make two separate computations for caps and goals.
2. Feed the formula with more data (from retired players)
3. Consider recent players only (maybe players from the 50s-60s had a
different type of career?)
4. Design a better formula, that e.g. knows that even if you are 18
and already have 5 caps (Santon), you won't be able to gain more than
200 caps, because NTs play 10 matches a year and you won't be able to
play anymore when 40. This is the most interesting attempt, but it's
also the one where I would need help from someone more skilled than
me!
> It assumes that my future picks will perform like my present picks,
> "performa" meaning performance according to this formula (
> the projection of career caps and goals based on current values
> and age ).
>
> Since in FS, we are forced to continue picking players every year,
> this is a big drawback with the formula. The formula must find
> some way, if it is going to be making projections into the future,
> to account for the probability of performance of future picks.
Well, since this picking cycle will end in July 2010 and I guess that
nobody of us will pick more than a couple of players, this shouldn't
be much of an issue... ;)
More: I see this as a "estimated ranking of arrival" _for players
picked so far_, of course things will change with future picks.
But of course you're right: the fact that some more players will be
picked should be taken in account. I don't know how to do it (I can't
think of any predictor of the quality of future picks... other than
the quality of picks made so far!). Any suggestion?
--
Cheers
milivella
milivella
So it seems that the current formula pretty much sucks, and that I'm
pretty clueless when I have to make it better... Still, I defend the
point that such a formula somehow works! :) In fact, I think that
predicted score is a better indicator of the quality of picks made so
far than plain current score. Let's compare the top 10s:
Current score:
1. Gago
2. Cazorla
3. Guiza
4. Quagliarella
5. Pique
6. Rossi
7. Gignac
7. Mandanda
9. Arbeloa
10. Walcott
Predicted score:
1. Santon
2. Walcott
3. Alexandre Pato
4. Marin
5. Busquets
5. Mata
7. Pique
8. Rossi
9. Gago
10. Anderson
--
Cheers
milivella
Futbolmetrix
Enzo:
> In fact I'd like to make the formula better. But date of birth and
> points are the only data -
But there is additional information around. The club the player is at, the
role he plays in, how many top league appearances and goals he has
accumulated so far, possibly even how many CL appearances and goals as well.
The "ideal" formula would be calculated as follows (example for Italy)
1) Collect data on an entire cohort of players. Say, all players born in
1970-1971 who played in the Italian Campionato Primavera in 1989-1990.
2) For each player and every year, calculate:
- cumulative number of NT caps and goals
- cumulative number of Serie A appearances and goals.
- current club
- current role
3) In 2010, once this cohort's career is essentially over,
run the following set of regressions (separately for each age a)
TOTAL_ADDITIONAL_CAPS(a) = B0_a + B1_a*CAPS(a) + B2_a*NTGOALS(a) +
+ B3_a*SERIEA_APPEARANCES(a) + B4_a*SERIEA_goals(a) +
+ B5_a*CLUB(a) + B6_a*ROLE(a)
TOTAL_ADDITIONAL_NTGOALS(a) = B0_a + B1_a*CAPS(a) + B2_a*NTGOALS(a) +
+ B3_a*SERIEA_APPEARANCES(a) + B4_a*SERIEA_goals(a) +
+ B5_a*CLUB(a) + B6_a*ROLE(a)
where TOTAL_ADDITIONAL_CAPS(a) is the total number of additional caps won by
a given player after age a, CAPS(a) is the total number of CAPS up to age a,
and so on.
Then, you have for each age a, a predicted number of additional caps and
goals for each player.
Conceptually, very easy.
Practically, impossible to implement.
D
milivella
> "milivella" <milive...@gmail.com> wrote in message
:)
Yes, it's impossible to implement. Still, if we want to talk theory, I
would not take note of appearances and club, replacing them with
points obtained by the team player X played (or started). (OK, a
quicker - and maybe not less fair - option would be total_team_points/
percentual_of_matches_played_by_X)
Also: "role" is not objective. What about "goals per game ratio"
instead?
But I expect from you the perfect formula to avoid predicting 500 caps
for Santon! :)
I really don't know how to fix it... I'm just doing a step back, and
using this simple formula (for players younger than 37!):
points/((age-17)/19)
E.g. Take a player who (1) is at the middle of his career, i.e. is 26
years and 6 months old, and (2) has 10 points: he is expected to score
in the second part of his career the same amount of points he has
scored so far, i.t. another 10 points:
10/((26.5-17)/19) = 20
Al least the expected scores are more realistic now, the ranking
remaining acceptable:
1. Walcott 104
2. Gago 99
3. Santon 95
4. Pato 86
5. Pique 76
5. Marin 76
7. Rossi 71
8. Cazorla 71
9. Busquets 57
9. Mata 57
And this would be the expected score of each RSSer:
1. Enzo 25
2. Andrea V. 16
3. Michael H. 12
4. Daniele 10
4. Jesus 10
6. Sid 8
7. Abubakr 6
8. William 5
9. Mark 1
10. Tom 0
Where have I to sign?
--
Cheers
milivella
milivella
> I really don't know how to fix it... I'm just doing a step back, and
> using this simple formula (for players younger than 37!):
> points/((age-17)/19)
> Al least the expected scores are more realistic now, the ranking
> remaining acceptable:
> 1. Walcott 104
> 2. Gago 99
> 3. Santon 95
> 4. Pato 86
> 5. Pique 76
> 5. Marin 76
> 7. Rossi 71
> 8. Cazorla 71
> 9. Busquets 57
> 9. Mata 57
>
> And this would be the expected score of each RSSer:
> 1. Enzo 25
> 2. Andrea V. 16
> 3. Michael H. 12
> 4. Daniele 10
> 4. Jesus 10
> 6. Sid 8
> 7. Abubakr 6
> 8. William 5
> 9. Mark 1
> 10. Tom 0
Numbers were wrong, because all the players were treated as 1 year
younger; and anyway players being born in the same year had the same
multiplier, i.e. it didn't matter if they were born in January or
December.
The fixed and definitive numbers (live now: http://fantasyscout.altervista.org/data.htm
) are:
1. Gago 76
2. Cazorla 63
3. Walcott 57
3. Santon 57
5. Pato 55
6. Pique 54
7. Rossi 50
8. Marin 42
9. Busquets 40
10. Mata 38
1. Enzo 16.41
2. Andrea V. 11.45
3. Daniele 9.03
4. Michael H. 8.62
5. Jesus 6.34
6. Sid 6.10
7. Abubakr 4.57
8. William 4.18
9. Mark 1.14
10. Tom two 0.00
Too low?
--
Cheers
milivella
Enzo
>
> - Show quoted text -
Using the same principle as you use now is fine,
too. For each of the 8 NTs, for any pick P from NT-n,
calculate his projected numbers by calculating
the average C or G of all players from NT-n who
have >= current C or G of P, irrespective
of age. This average will naturally be higher for a P
with 20 C already, than for P with 0 C, thus naturally
projecting someone with more C today as likely to have a
higher career C than someone with 1 C just because
he is 17 ( which is where the flaw lies in the
current formula, and which is why I am at the top,
something which made me criticise this formula in
the first place ).
Then, adjust for P's age as follows. When calculating
the average C for all NT-n players with minimum C(P)
above ( and this is easily determined actually, I can
write something which will automatically calculate it
everytime given a database ), also calculate the average
career span of all these players in years. The rest is
easy.
Enzo
>
> - Show quoted text -
I just want to add that any formula that does not displace
me from 1st place will not pass my approval ...
futbol...@yahoo.com
> milivella:
>
>
>
>
> The fixed and definitive numbers (live now:http://fantasyscout.altervista.org/data.htm
> ) are:
>
>
> 2. Andrea V. 11.45
> 3. Daniele 9.03
> 4. Michael H. 8.62
> 5. Jesus 6.34
> 6. Sid 6.10
> 7. Abubakr 4.57
> 8. William 4.18
> 9. Mark 1.14
> 10. Tom two 0.00
>
> Too low?
>
I've now climbed back to third place, so I'm happy with this new
ranking!
Other than that, I'm working on a new regression-based formula that
can be calculated just with the data currently available. I'll keep
you posted.
D
milivella
> Using the same principle as you use now is fine,
> too.
Thanks a lot for the solution! :)
Let's see whether I got it. For example, I want to compute Messi's
projected cup.
> For each of the 8 NTs, for any pick P from NT-n,
> calculate his projected numbers by calculating
> the average C or G of all players from NT-n who
> have >= current C or G of P, irrespective
> of age.
OK, I take Argentina data
http://rsssf.com/miscellaneous/arg-recintlp.html
and put them in Excel. Messi has 39 caps. There are 49 players
(excluding Messi) with 39 or more caps. In average, they have 57.82
caps.
[Question: if a player has 0 caps, there are million of players with
more or same caps. So their average is 0.000...1 (i.e. 0 by any
approximation), isn't it?]
[Problem: I don't know if for all the NTs there is a complete list of
all the players that had at least 1 cap.]
> This average will naturally be higher for a P
> with 20 C already, than for P with 0 C, thus naturally
> projecting someone with more C today as likely to have a
> higher career C than someone with 1 C just because
> he is 17 ( which is where the flaw lies in the
> current formula, and which is why I am at the top,
> something which made me criticise this formula in
> the first place ).
Indeed your score is determined by Busquets (21 y.o., 40 estimated
points) and Otamendi (22 y.o., 18 e.p.), and there are 3 players
(Walcott, Santon, Pato) that are younger than Busquets and have an
higher estimated score, but they don't make their scout (me, Alberto,
Jesus) first! (and anyway, those players have at least 5 times more
caps than 1...)
And I don't find the principle unreasonable: the same player X when
he's 17 has a minor or equal number of caps than when he's 36, despite
having the same number of final career caps. So you can expect that a
17 y.o. has less caps than a 36 y.o. of the same strength. Am I wrong?
Anyway, if you want to lose the first place, just pick some bad
player, or (better!) sell me Busquets. ;)
But back to Messi:
> Then, adjust for P's age as follows. When calculating
> the average C for all NT-n players with minimum C(P)
> above ( and this is easily determined actually, I can
> write something which will automatically calculate it
> everytime given a database ), also calculate the average
> career span of all these players in years.
I.e.? If Zanetti played his first international match in 1994 and his
last in 2009, has he a career span of 16 years?
If so, the average career span is 9.55 (all the players) or 9.48 (only
retired players) years. Which number should I take?
Anyway, in this case they are pretty close, so let's say that the
average career is 9.5 years.
[Question: what if P is the player with most caps or goals in the
history of his NT?]
> The rest is
> easy.
Not for me! :) What should I do now?
Thanks again.
--
Cheers
milivella
milivella
> On Oct 16, 7:17 pm, milivella <milive...@gmail.com> wrote:
>
>
>
> > milivella:
>
> > The fixed and definitive numbers (live now:http://fantasyscout.altervista.org/data.htm
> > ) are:
>
> > 1. Enzo 16.41
> > 2. Andrea V. 11.45
> > 3. Daniele 9.03
> > 4. Michael H. 8.62
> > 5. Jesus 6.34
> > 6. Sid 6.10
> > 7. Abubakr 4.57
> > 8. William 4.18
> > 9. Mark 1.14
> > 10. Tom two 0.00
>
> > Too low?
>
> I've now climbed back to third place, so I'm happy with this new
> ranking!
I'm above you, so I'm happy too. :P
> Other than that, I'm working on a new regression-based formula that
> can be calculated just with the data currently available. I'll keep
> you posted.
Thanks! Can't wait for it. :)
--
Cheers
milivella
milivella
> This average will naturally be higher for a P
> with 20 C already, than for P with 0 C, thus naturally
> projecting someone with more C today as likely to have a
> higher career C than someone with 1 C just because
> he is 17 ( which is where the flaw lies in the
> current formula
Indeed the curren formula has a bias towards 30 y.o. players, as it's
easy (1) to understand (they are projected to keep scoring as many
points per year as in their prime) and (2) to empirically check:
Top 10 historic FS players: highest predicted score, at what age (and
real score)
Matthaeus 122 33-36 (173)
Pele 380 17-18 (169) Exception!
Charlton 185 29 (155)
Klinsmann 173 34 (155)
Rummenigge 193 28 (140)
Zidane 155 34 (139)
Voeller 165 30 (137)
Batistuta 182 29 (134)
Mueller 206 29 (130) He retired from NT when 29!
Lineker 162 31-32 (128)
(But the formula is age-agnostic for the current players: FS picks
average age is 21.91, average age of top 10 players according to
projected score is 21.63.)
---
More: the formula is self-consistent!
Infact, these would have been the top 10 players one year ago (i.e.
picks, age and matches as of 17 October 2008):
Gago 66
Walcott 65
Pato 46
Anderson 38
Cazorla 33
Marin 29
Quaglia 26
Guiza 22
Marcelo 19
Krkic 19
And here's the current top 10 (only players picked before 17 October
2008, of course):
Gago 76
Cazorla 63
Walcott 57
Pato 55
Pique 54
Rossi 50
Marin 42
Quaglia 37
Guiza 36
Anderson 34
Among the 86 players that were picked before October 2008, the top 8
one are still in the top 10 after one year (Marcelo and Bojan out,
Pique and Rossi in).
---
Conclusion: no, IMHO 17 year old are not favored by this formula. :)
--
Cheers
milivella
milivella
[Age-bias]
> More: the formula is self-consistent!
Another feature that a good formula should have is maybe to produce a
power-law-like distribution, i.e. few players should predicted 100
points, many should be predicted 1 point. (current simplistic formula?
quick check... pass this test)
--
Cheers
milivella
milivella
> 4. Design a better formula, that e.g. knows that even if you are 18
> and already have 5 caps (Santon), you won't be able to gain more than
> 200 caps, because NTs play 10 matches a year and you won't be able to
> play anymore when 40.
Random idea.
We know that, among the players born in a given year, the best one
scores in average 120 points, the second 100, the third 90, etc.
(fictional numbers)
So, if we take all the players born e.g. in 1986, we could assign an
expected score of 120 points to the one that currently have most
points, 100 to the second, etc.
Of course, we would need a complete list of all the players who have
at least 1 cap (even the ones not picked in FS): something that
requires some work. So I would classify this attempt as "not
impossible, but certainly takes time".
--
Cheers
milivella
milivella
> More: the formula is self-consistent!
Of course this was only important in the context of replying to Enzo's
criticism. The real positive feature that a prediction formula should
have is to be good at predicting not the future *predicted* ranking,
but the future *real* ranking.
Of course we have to wait 15 or more years before comparing the actual
predicted ranking and the real final ranking. But we can do at least
two things:
1. Checking how good a formula is at predicting the real career of a
set of past players (a different set from the one you eventually used
to design the formula).
2. Checking how good a formula is at predicting the actual FS scores,
given past data.
I think that we could use these benchmarks to compare different
formulas (Enzo, Daniele, I'm challenging you! ;) ).
Here is how the simplistic linear formula rates according to these
benchmarks (abstract: not well!).
---
1. Real career of past players (top 10 historic players by FS points:
http://fantasyscout.altervista.org/history.htm )
Correlation
Age, Correlation
17 0.50
18 0.50
19 0.52
20 0.53
21 0.69
22 0.64
23 0.62
24 0.54
25 0.46
26 0.35
27 0.22
28 0.14
29 0.22
30 0.35
31 0.30
32 0.51
33 0.80
34 0.82
35 0.82
36 0.82
37 0.91
38 0.99
39 1.00
Distance = square root of ((projected_score_player_A-
real_score_player_A)^2 + (projected_score_player_B-real_score_player_B)
^2 + etc.) [I don't know whether this is the right formula, and anyway
what's its technical name]
Age, Distance
17 441
18 490
19 487
20 458
21 395
22 346
23 287
24 222
25 183
26 148
27 140
28 136
29 150
30 147
31 138
32 119
33 96
34 76
35 50
36 31
37 21
38 7
39 0
I.e. if I'm not wrong, the formula is off of an average of 49 points
when the player is 18-19, while the distance is 14 when his career is
at its middle point (27 y.o.).
(I'm sure that there is a better measure than these two. Help!)
---
2. Predicting the current scores
Using matches played up to October 2008 by the 86 players picked up to
that point
Correlation predicted score -> real 2009 score = 0.68
Correlation real 2008 score -> real 2009 score = 0.86 !
Using matches played up to October 2007 by the 18 players picked up to
that point
Correlation predicted score -> real 2009 score = 0.85
Correlation real 2007 score -> real 2009 score = 0.86
--
Cheers
milivella
milivella
Of course I were wrong: distance is not difference.
--
Cheers
milivella
Enzo
> Enzo:
>
> > Using the same principle as you use now is fine,
> > too.
>
> Thanks a lot for the solution! :)
>
> Let's see whether I got it. For example, I want to compute Messi's
> projected cup.
>
> > For each of the 8 NTs, for any pick P from NT-n,
> > calculate his projected numbers by calculating
> > the average C or G of all players from NT-n who
> > have >= current C or G of P, irrespective
> > of age.
>
> and put them in Excel. Messi has 39 caps. There are 49 players
> (excluding Messi) with 39 or more caps. In average, they have 57.82
> caps.
>
Messi will probably get more than 57 caps. But Otamendi will
probably get far fewer than what your formula gives him.
> [Question: if a player has 0 caps, there are million of players with
> more or same caps. So their average is 0.000...1 (i.e. 0 by any
> approximation), isn't it?]
>
0-cap players will be projected to get 0 caps.
I agree that 1-cap players projection will be problematic for
lack of complete data.
> [Problem: I don't know if for all the NTs there is a complete list of
> all the players that had at least 1 cap.]
>
> > This average will naturally be higher for a P
> > with 20 C already, than for P with 0 C, thus naturally
> > projecting someone with more C today as likely to have a
> > higher career C than someone with 1 C just because
> > he is 17 ( which is where the flaw lies in the
> > current formula, and which is why I am at the top,
> > something which made me criticise this formula in
> > the first place ).
>
> Indeed your score is determined by Busquets (21 y.o., 40 estimated
> points) and Otamendi (22 y.o., 18 e.p.), and there are 3 players
> (Walcott, Santon, Pato) that are younger than Busquets and have an
> higher estimated score, but they don't make their scout (me, Alberto,
> Jesus) first! (and anyway, those players have at least 5 times more
> caps than 1...)
They dont make you first because of averaging. You chose more
players. That is not relevant to the discussion of how best
to project, for a given single player, his future marks.
>
> And I don't find the principle unreasonable: the same player X when
> he's 17 has a minor or equal number of caps than when he's 36, despite
> having the same number of final career caps. So you can expect that a
> 17 y.o. has less caps than a 36 y.o. of the same strength. Am I wrong?
>
I am not sure what you mean.
> Anyway, if you want to lose the first place, just pick some bad
> player, or (better!) sell me Busquets. ;)
>
> But back to Messi:
>
> > Then, adjust for P's age as follows. When calculating
> > the average C for all NT-n players with minimum C(P)
> > above ( and this is easily determined actually, I can
> > write something which will automatically calculate it
> > everytime given a database ), also calculate the average
> > career span of all these players in years.
>
> I.e.? If Zanetti played his first international match in 1994 and his
> last in 2009, has he a career span of 16 years?
>
> If so, the average career span is 9.55 (all the players) or 9.48 (only
> retired players) years. Which number should I take?
All retired players, I guess.
>
> Anyway, in this case they are pretty close, so let's say that the
> average career is 9.5 years.
>
> [Question: what if P is the player with most caps or goals in the
> history of his NT?]
>
Then he will be projected to receive zero more caps and goals.
> > The rest is
> > easy.
>
> Not for me! :) What should I do now?
Career span not in NT but total. So,
Average NT caps for all players with Messi's caps is 57?
Messi has 40?
Average 1st class career span for all players with Messi's caps is 15?
Messi 1st class career span is 5? He is a special one! He will
get 120 caps.
36-year old Schiavi has 3 caps. His career span is now 16 years.
Average career span of NT players with 3 caps is 15, lets say.
Schiavi will not get another cap. But, who knows, he may be
a special one too. Average NT caps of all players with 3 or
more caps is 10? Give the sucker 7 more.
milivella
> On Oct 17, 5:43 pm, milivella <milive...@gmail.com> wrote:
>
> > Enzo:
>
> > Let's see whether I got it. For example, I want to compute Messi's
> > projected cup.
>
> > > For each of the 8 NTs, for any pick P from NT-n,
> > > calculate his projected numbers by calculating
> > > the average C or G of all players from NT-n who
> > > have >= current C or G of P, irrespective
> > > of age.
>
> > OK, I take Argentina datahttp://rsssf.com/miscellaneous/arg-recintlp.html
> > and put them in Excel. Messi has 39 caps. There are 49 players
> > (excluding Messi) with 39 or more caps. In average, they have 57.82
> > caps.
>
> Messi will probably get more than 57 caps. But Otamendi will
> probably get far fewer than what your formula gives him.
Far fewer than 21? (he already has 5) Maybe.
But of course this is not the point. The real question is: will the
*average* player who is 22 and has 5 caps get 16 more caps before
retiring? Or (in Messi's case, he too 22): what is he already has 39
caps?
I think that my simple formula (a formula that IMHO needs to be
replaced!) isn't too optimistic when it replies to such questions.
(But of course I think that Messi will have less than 146 caps!)
> > [Question: if a player has 0 caps, there are million of players with
> > more or same caps. So their average is 0.000...1 (i.e. 0 by any
> > approximation), isn't it?]
>
> 0-cap players will be projected to get 0 caps.
OK.
> I agree that 1-cap players projection will be problematic for
> lack of complete data.
If they are not available, we could run a regression to know how many
players could have 29, 28, 27... 3, 2, 1 caps. I've already checked
that regression works in such a context:
http://groups.google.com/group/rec.sport.soccer/browse_thread/thread/8c0eb9b6577c6847/
> > > This average will naturally be higher for a P
> > > with 20 C already, than for P with 0 C, thus naturally
> > > projecting someone with more C today as likely to have a
> > > higher career C than someone with 1 C just because
> > > he is 17 ( which is where the flaw lies in the
> > > current formula, and which is why I am at the top,
> > > something which made me criticise this formula in
> > > the first place ).
>
> > Indeed your score is determined by Busquets (21 y.o., 40 estimated
> > points) and Otamendi (22 y.o., 18 e.p.), and there are 3 players
> > (Walcott, Santon, Pato) that are younger than Busquets and have an
> > higher estimated score, but they don't make their scout (me, Alberto,
> > Jesus) first! (and anyway, those players have at least 5 times more
> > caps than 1...)
>
> They dont make you first because of averaging. You chose more
> players. That is not relevant to the discussion of how best
> to project, for a given single player, his future marks.
So don't criticize the formula just because your average score is the
highest one! ;)
> > And I don't find the principle unreasonable: the same player X when
> > he's 17 has a minor or equal number of caps than when he's 36, despite
> > having the same number of final career caps. So you can expect that a
> > 17 y.o. has less caps than a 36 y.o. of the same strength. Am I wrong?
>
> I am not sure what you mean.
Sorry. Let me try again (twice):
- Find me a player whose caps tally was higher when 17 than when 36.
- Or: let's make a bet. I'm thinking about two real players, and I'll
tell you only some data about them: their age and their caps tally.
Player A is 17 and has 2 caps, while player B is 36 and has 2 caps.
Now, I'll give you 100 RSS-dollars if you'll answer correctly to the
following question: who will have a higher caps tally at the end of
their career? A or B?
> > But back to Messi:
>
> > > Then, adjust for P's age as follows. When calculating
> > > the average C for all NT-n players with minimum C(P)
> > > above ( and this is easily determined actually, I can
> > > write something which will automatically calculate it
> > > everytime given a database ), also calculate the average
> > > career span of all these players in years.
>
> > I.e.? If Zanetti played his first international match in 1994 and his
> > last in 2009, has he a career span of 16 years?
>
> > If so, the average career span is 9.55 (all the players) or 9.48 (only
> > retired players) years. Which number should I take?
>
> All retired players, I guess.
OK.
> > Anyway, in this case they are pretty close, so let's say that the
> > average career is 9.5 years.
>
> > [Question: what if P is the player with most caps or goals in the
> > history of his NT?]
>
> Then he will be projected to receive zero more caps and goals.
Even if he's still in his prime? :/
> > > The rest is
> > > easy.
>
> > Not for me! :) What should I do now?
>
> Career span not in NT but total.
I.e. Romario's career span is 1985-2008?
> So,
>
> Average NT caps for all players with Messi's caps is 57?
> Messi has 40?
>
> Average 1st class career span for all players with Messi's caps is 15?
> Messi 1st class career span is 5? He is a special one! He will
> get 120 caps.
But you aren't using the "57" data anymore, are you? If you don't use
it, you are pretty close to my simple formula... (in fact, 120 is not
so far from 146!)
> 36-year old Schiavi has 3 caps. His career span is now 16 years.
> Average career span of NT players with 3 caps is 15, lets say.
> Schiavi will not get another cap. But, who knows, he may be
> a special one too. Average NT caps of all players with 3 or
> more caps is 10? Give the sucker 7 more.
You should decide... ;)
I am sorry for having replied after a lot of time. Unfortunately, I
have little free time. BTW, this is the reason why I'm not posting
anymore about my FS possible reforms, even if I'm still thinking about
them.
--
Cheers
milivella
milivella
> 2. Predicting the current scores
>
> Using matches played up to October 2008 by the 86 players picked up to
> that point
> Correlation predicted score -> real 2009 score = 0.68
> Correlation real 2008 score -> real 2009 score = 0.86 !
>
> Using matches played up to October 2007 by the 18 players picked up to
> that point
> Correlation predicted score -> real 2009 score = 0.85
> Correlation real 2007 score -> real 2009 score = 0.86
To use the predicted _final_ score was wrong. The right method is to
use tweak the formula to predict just the next (1 or) 2 years:
prediction = points_so_far * ((age-15)/20) / ((age-17)/20)
Using matches played up to October 2008 by the 86 players picked up to
that point
Correlation predicted score -> real 2009 score = 0.85
Correlation real 2008 score -> real 2009 score = 0.86
Using matches played up to October 2007 by the 18 players picked up to
that point
Correlation predicted score -> real 2009 score = 0.86
Correlation real 2007 score -> real 2009 score = 0.86
More, we can now compute the "distance" between the real 2009 score
and the predicted score:
2008: 32.75
2007: 23.17
For the sake of comparison, the distance between real scores is 35.67
(2008) and 26.91 (2007).
--
Cheers
milivella
Enzo
>
>
> - Show quoted text -- Hide quoted text -
>
> - Show quoted text -
The formula use you to predict is no longer on the site.
I clearly remember it, but it seems i may have interpreted
it wrongly. Otamendi more than 21? I dont understand.
If you can put that stuff back on the site, I will be able
to better answer your questions.
Thanks.
milivella
> The formula use you to predict is no longer on the site.
> I clearly remember it, but it seems i may have interpreted
> it wrongly.
I've never published the (current) formula on the site. But of course
it's in the spreadsheet that you can download:
http://fantasyscout.altervista.org/data.xls
But I'd like to write about it, so let me show it.
It's a very simple linear formula that assumes that:
- each player has a 19 year career span, since 18 (because before this
age players score in average 0.01 of their total points) until 37
(because after this age players score in average 0.01 of their total
points.)
- the player will keep until the end of his career the same point
scoring pace that he has kept so far
The formula is:
Prediction = PointsSoFar / ((Age-17)/20)
but you also have to deduct the points that the player had when you
picked him.
The player could be outside the considered age range. In this case...
...if he's less than 18, let's just consider that he's 18:
Prediction = PointsSoFar / (1/20)
...if he's more than 37, let's just consider that he won't score any
more point:
Prediction = PointsSoFar
> Otamendi more than 21? I dont understand.
If you set the format of the column to two decimals, you'll see that
he's actually 21.67, i.e. 21 years and 8 months (Otamendi was born in
February). I used this little trick to make the predicted score change
smoothly, not leaping at the beginning of the year or at the player's
birthday.
BTW, the formula I'm using to compute a player's age is:
Age = CurrentYear - BirthYear + (CurrentMonth-BirthMonth)/12
> If you can put that stuff back on the site, I will be able
> to better answer your questions.
I've written everything I know about the formulas here. ;) Please ask
me if you've doubts.
Thanks for having cared about the formula.
--
Cheers
milivella
Futbolmetrix
My predicted final ranking (very partial, algorithm still needs substantial
improvement):
+-----------------------+
| scout mean_P~D |
|-----------------------|
1. | Enzo 33.86321 |
6. | Cristian 23.6988 |
12. | Matthias 22.57375 |
17. | Alessandro 17.20138 |
23. | Mattia 17.20103 |
|-----------------------|
32. | Alberto 16.30108 |
37. | Sid 16.13473 |
41. | Michael H. 15.96837 |
61. | Abubakr 15.02278 |
65. | Andrea V. 14.10277 |
|-----------------------|
90. | Daniele 11.35696 |
105. | William 10.73824 |
115. | Benny 9.606962 |
140. | Nigel 7.832734 |
144. | Generoso 6.710965 |
|-----------------------|
156. | Tom one 5.797633 |
168. | Mpfat 5.375605 |
173. | Jesus 5.302326 |
182. | Simone G. 3.980286 |
188. | Mark 3.643015 |
All the other scouts at zero.
The best predicted FS scorers:
+------------------------------+
| player PREDIC~E |
|------------------------------|
1. | Pique 61.3943 |
2. | Cazorla 60.05754 |
3. | Gago 59.21056 |
4. | Rossi 57.48428 |
5. | Busquets 50.08788 |
|------------------------------|
6. | Mata 50.08788 |
7. | Santon 49.65768 |
8. | Gignac 48.54477 |
9. | Walcott 47.78078 |
10. | Alexandre Pato 47.72094 |
|------------------------------|
11. | Ozil 46.14777 |
12. | Guiza 46.08788 |
13. | Marin 45.75086 |
14. | Quagliarella 45.72399 |
15. | Tasci 43.72422 |
|------------------------------|
16. | Mandanda 42.72654 |
17. | Otamendi 42.20765 |
18. | Marcelo 41.29748 |
19. | Di Maria 41.23759 |
20. | Montolivo 39.72654 |
|------------------------------|
21. | Beck 39.72422 |
22. | Zuculini 39.29399 |
23. | Salvio 39.29399 |
24. | De Federico 38.78077 |
25. | Sissoko 38.78077 |
|------------------------------|
26. | Lloris 38.24066 |
27. | Adler 37.75684 |
28. | Elia 37.75421 |
29. | Llorente 37.72654 |
30. | Arbeloa 37.72399 |
|------------------------------|
31. | Andre Santos 37.66089 |
32. | Milner 37.24066 |
33. | Sandro 36.8107 |
34. | Insua 36.8107 |
35. | Higuain 35.78421 |
|------------------------------|
36. | Lucas 34.84419 |
37. | Boateng 34.32743 |
38. | Monzon 33.81419 |
39. | Gomis 32.87802 |
40. | Young 32.84772 |
|------------------------------|
41. | Bolatti 32.78713 |
42. | Negredo 32.78713 |
43. | Diego Tardelli 32.78713 |
44. | Agbonlahor 32.33094 |
45. | Datolo 32.27249 |
|------------------------------|
46. | Lavezzi 31.84772 |
47. | Khedira 31.84419 |
48. | Remy 31.84419 |
49. | Monreal 31.33093 |
50. | Marchisio 31.33093 |
|------------------------------|
51. | Criscito 31.33093 |
52. | Perez 31.33093 |
53. | Bocchetti 31.33093 |
54. | Dossena 29.80747 |
55. | Neuer 29.36103 |
|------------------------------|
56. | Clichy 28.87802 |
57. | Braafheid 27.81862 |
58. | D'Agostino 27.29948 |
59. | Filipe 26.87802 |
60. | Diego Souza 26.87802 |
|------------------------------|
61. | Marchetti 23.88171 |
62. | Loovens 21.91326 |
63. | Dominguez 16.43768 |
64. | Brana 12.01924 |
All others at zero.
Obviously, the first fix that needs to be implemented is to get rid of the
assumption that if a player has a current score of zero, then he will never
score any FS points. But, if I understand correctly, Andrea's formula
suffers from the same problem.
Explanations to follow.
D
Enzo
> "milivella" <milive...@gmail.com> wrote in message
Un tres big problem with fantasy scout is that it rewards
substitute caps equally to starting caps, and friendly caps
equally to competitive caps.
tres big.
Futbolmetrix
"Futbolmetrix" <futbol...@yahoo.com> wrote in message
news:hc8kk8$f34$1...@news.eternal-september.org...
>
> "milivella" <mili...@gmail.com> wrote in message
> news:9961c7fe-f3df-4b9b...@j24g2000yqa.googlegroups.com...
>
> My predicted final ranking (very partial, algorithm still needs
> substantial improvement):
Updated prediction, does not restrict players with zero current score to
have zero score in the future.
+--------------------------------------------------------+
| scout scout_id MEANSCORE TOTSCORE TOTPLAYERS |
|--------------------------------------------------------|
4. | Enzo 12 33.5164 167.582 5 |
9. | Alberto 2 29.9316 149.658 5 |
15. | Alessandro 3 29.759 178.554 6 |
17. | Matthias 21 29.6968 148.484 5 |
27. | Cristian 8 29.027 174.162 6 |
|--------------------------------------------------------|
36. | Giovanni 15 28.31692 368.12 13 |
57. | Michael H. 23 28.2109 564.218 20 |
61. | Generoso 13 27.914 334.968 12 |
73. | Sid 29 27.518 110.072 4 |
82. | Daniel O. 10 27.43833 164.63 6 |
|--------------------------------------------------------|
84. | Nigel 26 27.368 109.472 4 |
97. | Andrea V. 6 27.35192 683.798 25 |
113. | Andrea B. 4 27.22467 81.674 3 |
119. | William 35 27.2088 272.088 10 |
125. | Mago 19 26.97 26.97 1 |
|--------------------------------------------------------|
128. | Tommaso 34 26.926 107.704 4 |
135. | Simone R. 31 26.844 268.44 10 |
141. | Daniel B. 9 26.7796 133.898 5 |
148. | Riccardo 28 26.7252 133.626 5 |
154. | Andrea C. 5 26.69086 186.836 7 |
|--------------------------------------------------------|
157. | Michael M. 24 26.288 52.576 2 |
169. | Benny 7 25.8834 647.085 25 |
193. | Daniele 11 25.78653 386.798 15 |
211. | Pietro 27 25.71487 411.438 16 |
218. | Jacopo 17 25.4015 101.606 4 |
|--------------------------------------------------------|
219. | Giulio 16 25.354 202.832 8 |
227. | Abubakr 1 25.1875 100.75 4 |
232. | Simone G. 30 25.144 150.864 6 |
242. | Gianmarco 14 25.14275 201.142 8 |
253. | Mark 20 25.10222 225.92 9 |
|--------------------------------------------------------|
257. | Jesus 18 23.71578 213.442 9 |
273. | Tom one 32 23.40517 280.862 12 |
279. | Mattia 22 21.29422 191.648 9 |
284. | Tom two 33 20.062 20.062 1 |
288. | Mpfat 25 19.3212 96.606 5 |
+--------------------------------------------------------+
And here is the predicted player rankings:
+------------------------------+
| player mean_P~E |
|------------------------------|
1. | Cazorla 56.376 |
2. | Pique 54.83099 |
3. | Gago 53.774 |
4. | Rossi 49.398 |
5. | Santon 42.864 |
|------------------------------|
6. | Busquets 42.55 |
7. | Walcott 42.384 |
8. | Alexandre Pato 42.07 |
9. | Quagliarella 41.312 |
10. | Guiza 40.994 |
|------------------------------|
11. | Mata 40.452 |
12. | Gignac 39.79 |
13. | Marin 39.436 |
14. | Ozil 39.36 |
15. | Tasci 38.41 |
|------------------------------|
16. | Mandanda 36.936 |
17. | Marcelo 36.288 |
18. | Di Maria 35.824 |
19. | Otamendi 35.578 |
20. | Anderson 34.748 |
|------------------------------|
21. | Beck 33.976 |
22. | Sissoko 33.952 |
23. | Montolivo 33.864 |
24. | Andre Santos 33.824 |
25. | Salvio 33.58 |
|------------------------------|
26. | De Federico 33.328 |
27. | Zuculini 33.236 |
28. | Elia 32.718 |
29. | Sandro 32.54 |
30. | Arbeloa 32.258 |
|------------------------------|
31. | Lloris 32.074 |
32. | Insua 31.994 |
33. | Milner 31.978 |
34. | Adler 31.85 |
35. | Lucas 31.284 |
|------------------------------|
36. | Llorente 30.764 |
37. | Boateng 30.518 |
38. | Higuain 29.832 |
39. | Neymar 29.774 |
40. | Muniain 29.772 |
|------------------------------|
41. | Gomis 29.452 |
42. | Trecarichi 29.336 |
43. | Ozyakup 29.124 |
44. | Caldirola 29.044 |
45. | Monzon 29.036 |
|------------------------------|
46. | Douglas P. 29.008 |
47. | Krkic 28.936 |
48. | Gallinetta 28.912 |
49. | Castaignos 28.91 |
50. | Sneijder 28.882 |
|------------------------------|
51. | Comi 28.808 |
52. | Albertazzi 28.722 |
53. | Wilshere 28.72 |
54. | Fossati 28.688 |
55. | Sukuta-Pasu 28.684 |
|------------------------------|
56. | Macheda 28.678 |
57. | Borini 28.666 |
58. | Zigoni 28.63 |
59. | Pacheco 28.6 |
60. | Bartley 28.566 |
|------------------------------|
61. | Destro 28.552 |
62. | Marrone 28.488 |
63. | Laribi 28.486 |
64. | Luciani 28.476 |
65. | Capel 28.348 |
|------------------------------|
66. | Alcantara 28.342 |
67. | Beretta 28.336 |
68. | Remy 28.284 |
69. | Welbeck 28.284 |
70. | El Sharaawy 28.216 |
|------------------------------|
71. | Douglas C. 28.19 |
72. | Aquino 28.186 |
73. | Agbonlahor 28.148 |
74. | Gosling 28.146 |
75. | Crisetig 28.138 |
|------------------------------|
76. | Sakho 28.126 |
77. | Mattock 28.124 |
78. | Crescenzi 28.08 |
79. | Young 28.054 |
80. | Balotelli 28.042 |
|------------------------------|
81. | Morosini 28.036 |
82. | Lulinha 27.994 |
83. | Diego Tardelli 27.98 |
84. | Zeefuik 27.966 |
85. | Maicon 27.94 |
|------------------------------|
86. | Marcellis 27.94 |
87. | Baxter 27.878 |
88. | Alex Texeira 27.874 |
89. | Munoz 27.852 |
90. | Banega 27.8 |
|------------------------------|
91. | Paloschi 27.792 |
92. | Mannini 27.774 |
93. | Immobile 27.766 |
94. | Lavezzi 27.762 |
95. | Nsue 27.712 |
|------------------------------|
96. | Giuliano 27.69 |
97. | Saivet 27.68 |
98. | Barazite 27.634 |
99. | Kakuta 27.634 |
100. | Holtby 27.628 |
|------------------------------|
101. | Aydilek 27.626 |
102. | Muller 27.624 |
103. | Fabio 27.544 |
104. | Forestieri 27.54 |
105. | Obertan 27.51 |
|------------------------------|
106. | Kroos 27.496 |
107. | Pastore 27.49 |
108. | Rafael 27.476 |
109. | Khedira 27.422 |
110. | Jo 27.404 |
|------------------------------|
111. | Fiorillo 27.364 |
112. | Sciacca 27.35 |
113. | Pasquato 27.344 |
114. | Ciano 27.296 |
115. | Alan Kardec 27.292 |
|------------------------------|
116. | Bellusci 27.264 |
117. | Falque 27.228 |
118. | Angella 27.222 |
119. | Merida 27.174 |
120. | Okaka Chuka 27.166 |
|------------------------------|
121. | Renan Oliveira 27.13 |
122. | Kopplin 27.128 |
123. | Wijnaldum 27.108 |
124. | Asenjo 27.106 |
125. | Monreal 27.072 |
|------------------------------|
126. | Walter 27.018 |
127. | Negredo 26.998 |
128. | Mustacchio 26.982 |
129. | Poli 26.97 |
130. | Marconi 26.966 |
|------------------------------|
131. | Garcia 26.952 |
132. | Marilungo 26.908 |
133. | Rodwell 26.89 |
134. | Camacho 26.882 |
135. | Sidnei 26.872 |
|------------------------------|
136. | Ciro 26.864 |
137. | Di Santo 26.848 |
138. | Piatti 26.836 |
139. | Badstuber 26.798 |
140. | Bocchetti 26.706 |
|------------------------------|
141. | Traore 26.69 |
142. | Conti 26.688 |
143. | Perez 26.678 |
144. | Criscito 26.656 |
145. | Gibbs 26.654 |
|------------------------------|
146. | Breno 26.652 |
147. | N'gog 26.644 |
148. | Sturridge 26.614 |
149. | Ariaudo 26.606 |
150. | Di Gennaro 26.534 |
|------------------------------|
151. | de Jong 26.53 |
152. | Marchisio 26.528 |
153. | Parejo 26.474 |
154. | Niguez 26.456 |
155. | Azpilicueta 26.454 |
|------------------------------|
156. | Bonaventura 26.428 |
157. | Datolo 26.422 |
158. | Bolatti 26.372 |
159. | Cia 26.358 |
160. | Mazzotta 26.358 |
|------------------------------|
161. | Gebhart 26.354 |
162. | Willian 26.3 |
163. | Keirrison 26.294 |
164. | Thuram-Ulien 26.272 |
165. | Marquinhos 26.248 |
|------------------------------|
166. | Ranocchia 26.232 |
167. | Dentinho 26.206 |
168. | Kerlon 26.11 |
169. | Ogbonna 26.092 |
170. | Dossena 25.988 |
|------------------------------|
171. | Raggio Garibaldi 25.924 |
172. | Taison 25.878 |
173. | De Silvestri 25.862 |
174. | Martinez 25.86 |
175. | Denilson 25.822 |
|------------------------------|
176. | Donald 25.738 |
177. | Nijland 25.724 |
178. | Renato Augusto 25.692 |
179. | Muamba 25.68 |
180. | Buonanotte 25.64 |
|------------------------------|
181. | Barill� 25.568 |
182. | Mancienne 25.536 |
183. | Biabiany 25.464 |
184. | Guilherme 25.454 |
185. | Aissati 25.422 |
|------------------------------|
186. | Enves 25.248 |
187. | Garay 25.17 |
188. | Cattermole 25.168 |
189. | Neuer 25.128 |
190. | Mannone 25.044 |
|------------------------------|
191. | Vaughan 25.014 |
192. | Felipe Mattioni 24.996 |
193. | Clichy 24.942 |
194. | Bonucci 24.9 |
195. | Johnson 24.896 |
|------------------------------|
196. | Wheater 24.89 |
197. | Diaby 24.868 |
198. | Chantome 24.722 |
199. | de Guzman 24.72 |
200. | Fernando 24.694 |
|------------------------------|
201. | Philippe Coutinho 24.668 |
202. | Rodriguez 24.658 |
203. | Howedes 24.574 |
204. | Carlos Eduardo 24.544 |
205. | Granero 24.472 |
|------------------------------|
206. | Danilinho 24.35 |
207. | Matuidi 24.286 |
208. | Fazio 24.282 |
209. | Hart 24.276 |
210. | de la Red 24.25 |
|------------------------------|
211. | Bruins 24.204 |
212. | Campbell 24.12 |
213. | Bentley 24.11 |
214. | Cissokho 24.108 |
215. | Menez 23.988 |
|------------------------------|
216. | Rubin 23.942 |
217. | Candreva 23.888 |
218. | Giovinco 23.878 |
219. | Zarate 23.856 |
220. | Drenthe 23.838 |
|------------------------------|
221. | Baumjohann 23.758 |
222. | Aogo 23.586 |
223. | Noble 23.528 |
224. | Acquafresca 23.52 |
225. | Dessena 23.504 |
|------------------------------|
226. | Sirigu 23.472 |
227. | Susaeta 23.244 |
228. | Filipe 23.096 |
229. | Cabaye 23.046 |
230. | Diego Souza 23.008 |
|------------------------------|
231. | Paletta 22.92 |
232. | Marzoratti 22.792 |
233. | Vermeer 22.72 |
234. | Polanski 22.702 |
235. | Abate 22.622 |
|------------------------------|
236. | Gouffran 22.596 |
237. | Valeri 22.578 |
238. | D'Agostino 22.514 |
239. | Bertolo 22.458 |
240. | Motta M. 22.39 |
|------------------------------|
241. | Cigarini 22.296 |
242. | Braafheid 22.144 |
243. | Ustari 22.11 |
244. | Osvaldo 22.046 |
245. | De Ceglie 22.044 |
|------------------------------|
246. | Rafinha 21.822 |
247. | Santacroce 21.662 |
248. | Nocerino 21.534 |
249. | Kaboul 21.492 |
250. | Schumacher 21.322 |
|------------------------------|
251. | Tissone 21.256 |
252. | Thiago Neves 20.872 |
253. | Guarente 20.836 |
254. | Navas 20.616 |
255. | Compper 20.59 |
|------------------------------|
256. | Renan 20.58 |
257. | Viviano 20.524 |
258. | Flamini 20.45 |
259. | Marchetti 20.396 |
260. | Hernanes 20.234 |
|------------------------------|
261. | Coda 20.106 |
262. | Rami 20.062 |
263. | Zapater 20.048 |
264. | Tremoulinas 20.002 |
265. | Cahill 19.976 |
|------------------------------|
266. | Geromel 19.86 |
267. | Maidana 19.804 |
268. | Leandro Lima 19.638 |
269. | Loovens 18.644 |
270. | Cavenaghi 18.472 |
|------------------------------|
271. | Rosina 18.372 |
272. | Lell 18.06 |
273. | Hoarau 17.608 |
274. | Valbuena 17.306 |
275. | Baines 17.038 |
|------------------------------|
276. | Raggi 16.982 |
277. | Michel Bastos 15.53 |
278. | Turner 15.338 |
279. | Borriello 15.194 |
280. | Pennant 14.72 |
|------------------------------|
281. | Bianchi 14.562 |
282. | Dominguez 13.884 |
283. | Maggio C. 12.77 |
284. | Pelletieri 12.482 |
285. | Motta T. 12.04 |
|------------------------------|
286. | Brana 10.236 |
287. | Frey 7.102 |
288. | Amauri 4.618 |
289. | Ceara 4.266 |
+------------------------------+
Even though I find this theoretically more appealing, I'm not entirely
satisfied with the outcome (and not just because I tank completely under
this new ranking). Players who currently have zero caps are predicted to
score a bit too much for my taste.
Gory statistical details to follow.
D
milivella
> Even though I find this theoretically more appealing, I'm not entirely
> satisfied with the outcome (and not just because I tank completely under
> this new ranking). Players who currently have zero caps are predicted to
> score a bit too much for my taste.
>
> Gory statistical details to follow.
Feed us with them!
A *big warm thank* for the prediction formula! As soon as you'll tell
us what the formula is, I'll put it on the website.
Unless you want to check how it rates against other formulas (i.e.
which one is better at predicting past players' careers - I guess this
is the only thing that counts -), of course.
The scores that your formula predicts seems a bit too close to me
(e.g. 5th = 43 points, 15th = 38 points), but it's just my ignorant
instinct.
Have you checked what 2009 scores your formula would have predicted
one and/or two years ago?
--
Cheers
milivella
milivella
> Un tres big problem with fantasy scout is that it rewards
> substitute caps equally to starting caps, and friendly caps
> equally to competitive caps.
>
> tres big.
...but easy to fix. ;)
My thoughts about the problems you're pointing at:
- They have been around since some time:
http://groups.google.com/group/rec.sport.soccer/browse_thread/thread/dc7fc41a42615ee1/
http://groups.google.com/group/rec.sport.soccer/msg/86c872f512068dcf
- They are still in my agenda of things to think about, but there are
"tres bigger" problems to solve first IMHO! ;) It goes without saying:
anyone can propose a reform about this (or any other) point in any
moment.
- I wouldn't vote against a reform that gives different weights to
starting/sub caps, official/friendlies caps.
- I'd like to see some data: how would the score of past players
change with such weights? A complete table would be great, but we
could check at least some players. E.g. the players who top the
following table
http://www.fifa.com/worldfootball/statisticsandrecords/tournaments/worldcup/players/mostmatches.html
would probably have a better score if WC matches weight more.
--
Cheers
milivella
Futbolmetrix
> Futbolmetrix:
>
>> Even though I find this theoretically more appealing, I'm not entirely
>> satisfied with the outcome (and not just because I tank completely under
>> this new ranking). Players who currently have zero caps are predicted to
>> score a bit too much for my taste.
>>
>> Gory statistical details to follow.
>
> Feed us with them!
>
OK, you asked for it.
Let's start from the observation that it may be too ambitious trying to
predict a player's total career score. But we could start from something a
lot easier, like predicting a player's score in 2010. So, we could start
from a model like
y(i,t) = b0 + b1*y(i,t-1) + u(i,t)
where y(i,t) is the score of player i in year t and u(i,t) is an error term.
b0 and b1 can be estimated from the existing set of players. But then, it's
easy to forecast y(i,t+1), y(i,t+2), etc. Assume that a player's NT career
is over at age 35, and you have the very first prediction I came up with.
The formula for the remaining score in a player's career can actually be
written down in closed form as a function of b0, b1 and the player's age.
Very preliminary, and not that satisfactory for a number of reasons:
1) the formula was estimated based only on those players who currently have
positive caps, it fails to recognize that even an uncapped player has some
positive probability of breaking into the NT, and then he will start to
accumulate caps.
2) The regression model above fails to recognize that FS score has two
important properties: a) it's an integer number, and b) it's non-negative.
So, if you try to forecast a single's player individual career, you may end
up with negative score (and you will surely end up with non-integer score).
But there are statistical models used to estimate exactly these type of
data. The most popular is the Poisson model, and it has several variants.
After some tinkering, I ended up with the following model:
a) If caps(t-1)>0, predict caps(t) using a zero-inflated Poisson model based
on caps(t-1). [I can give you more gory details if you want, but the basic
idea is that a zero-inflated model recognizes that there are more zeros in
the data than a typical Poisson distribution]
b) If caps(t-1)=0, predict caps(t) using a zero-inflated Poisson model based
on age and age^2. It captures the fact that players can break into the NT,
but the probability of accumulating caps initially increases, and then
decreases with age.
c) Given caps(t), predict the number of goals as: 0, if caps(t)=0; Poisson
depending on goals and caps in year t-1, and the average goal per game rate
among the players currently picked (at the moment it's 1/9)
Given the above model, one can estimate the model parameters (how caps(t-1)
and age enter the formula), and then generate N simulated careers for reach
of the 289 players currently picked in FS. Average over the N simulations,
and you get the ranking.
I've tinkered with the model a bit since my last post, and the current
ranking is slightly different (it appears posted at the bottom of this post)
> A *big warm thank* for the prediction formula! As soon as you'll tell
> us what the formula is, I'll put it on the website.
There is no exact formula, but a reasonable approximation is:
PREDICTED SCORE = CURRENT SCORE + ADDITIONAL SCORE
ADDITIONAL SCORE = 255.1718 + 0.0223868*CURRENT_SCORE - 50.16514*AGE +
3.98231*AGE^2 - 0.1351248*AGE^3 + 0.001628*AGE^4 +
2.193766*CAPS2009 -0.0978368*CAPS2009^2 + 0.0263713*GOALS2009
Things that I don't like about this method:
a) Despite the attempts to separate between players who have already broken
into the NT and those who have not, I think the model is still to generous
with the latter. I believe that the difference between Marchisio (age 23, 2
caps, predicted score 25.75) and Abate (age 23, 0 caps, predicted score
19.8) should be greater than what it is. And there are a bit too many caps
predicted for the 18-year olds who haven't even ever played a professional
match yet.
b) It should really use more information in making predictions.
Things that I like about this method.
a) it's really quite flexible, and it's very easy to incorporate additional
information. For example, that strikers are expected to have a higher goal
per game rate than defenders. Or that a 22-year old who plays regularly for
a CL team in one of the top leagues (Giovinco) is expected to do better than
a 22-year old who is loaned around all the time to smaller teams
(Acquafresca).
b) It generates a lot of interesting information: for example, you want to
know how Criscito's career will evolve in terms of number of caps? Here it
is:
2010 (24): 2.27
2011 (25): 2.23
2012 (26): 2.21
2013 (27): 2.19
2014 (28): 2.12
2015 (29): 1.98
2016 (30): 1.86
2017 (31): 1.61
2018 (32): 1.46
2019 (33): 1.21
2020 (34): 1.02
2021 (35): 0.84
2022 (36): 0 (by assumption)
The decline in the number of caps occurs because if Criscito happens to lose
his spot in the NT (zero caps in any given year), it becomes much more
difficult to reenter the NT as he becomes older.
> The scores that your formula predicts seems a bit too close to me
> (e.g. 5th = 43 points, 15th = 38 points), but it's just my ignorant
> instinct.
Yes, see above. But I think that it's a reflection of the fact that there's
still a *lot* of uncertainty about these players. Gago has lost his starting
spot at Real Madrid. How long will he hold onto his starting spot for the
Argentina NT?
> Have you checked what 2009 scores your formula would have predicted
> one and/or two years ago?
And take away all the fun from you?
Cheers,
D
-----------------------------------------------------------------------
Scouts' ranking:
+--------------------------------------------------------+
| scout scout_id MEANSC~t TOTSCO~t TOTPLA~t |
|--------------------------------------------------------|
4. | Enzo 12 33.10788 165.5394 5 |
6. | Matthias 21 28.56644 142.8322 5 |
14. | Alessandro 3 27.9415 167.649 6 |
21. | Alberto 2 27.831 139.155 5 |
24. | Cristian 8 26.96667 161.8 6 |
|--------------------------------------------------------|
34. | Michael H. 23 26.32805 526.561 20 |
62. | Andrea V. 6 25.30708 632.677 25 |
77. | William 35 25.17866 251.7866 10 |
84. | Giovanni 15 25.16305 327.1196 13 |
96. | Generoso 13 25.15723 301.8868 12 |
|--------------------------------------------------------|
108. | Nigel 26 25.1314 100.5256 4 |
114. | Sid 29 24.90645 99.6258 4 |
120. | Daniel O. 10 24.71287 148.2772 6 |
122. | Mago 19 24.5072 24.5072 1 |
124. | Daniel B. 9 24.234 121.17 5 |
|--------------------------------------------------------|
129. | Tommaso 34 23.95485 95.8194 4 |
139. | Simone R. 31 23.91886 239.1886 10 |
147. | Andrea C. 5 23.65223 165.5656 7 |
150. | Andrea B. 4 23.62527 70.8758 3 |
159. | Benny 7 23.55681 588.9202 25 |
|--------------------------------------------------------|
188. | Daniele 11 23.54408 353.1612 15 |
192. | Michael M. 24 23.5378 47.0756 2 |
195. | Riccardo 28 23.48384 117.4192 5 |
201. | Abubakr 1 23.369 93.476 4 |
205. | Pietro 27 22.97808 367.6492 16 |
|--------------------------------------------------------|
226. | Giulio 16 22.7372 181.8976 8 |
227. | Gianmarco 14 22.44852 179.5882 8 |
237. | Jacopo 17 22.4246 89.6984 4 |
239. | Mark 20 22.41424 201.7282 9 |
252. | Simone G. 30 22.34843 134.0906 6 |
|--------------------------------------------------------|
258. | Jesus 18 21.43409 192.9068 9 |
264. | Tom one 32 21.0961 253.1532 12 |
275. | Mattia 22 18.89049 170.0144 9 |
284. | Tom two 33 17.6108 17.6108 1 |
287. | Mpfat 25 16.77908 83.8954 5 |
+--------------------------------------------------------+
Player ranking:
+------------------------------+
| player mean_P~E |
|------------------------------|
1. | Gago 54.3216 |
2. | Cazorla 53.6002 |
3. | Pique 51.2836 |
4. | Rossi 49.2484 |
5. | Busquets 44.2524 |
|------------------------------|
6. | Santon 42.611 |
7. | Mata 41.9566 |
8. | Quagliarella 41.7794 |
9. | Alexandre Pato 41.716 |
10. | Guiza 41.4876 |
|------------------------------|
11. | Walcott 40.9746 |
12. | Ozil 39.6612 |
13. | Gignac 39.6306 |
14. | Marin 39.0826 |
15. | Tasci 38.3746 |
|------------------------------|
16. | Mandanda 37.8138 |
17. | Otamendi 36.5812 |
18. | Marcelo 35.1188 |
19. | Di Maria 34.7918 |
20. | Montolivo 34.71 |
|------------------------------|
21. | Beck 34.6954 |
22. | Lloris 33.4626 |
23. | Arbeloa 33.3934 |
24. | Adler 32.9152 |
25. | Andre Santos 32.8672 |
|------------------------------|
26. | Milner 32.3202 |
27. | Anderson 31.6292 |
28. | Sissoko 31.225 |
29. | Elia 31.1996 |
30. | Zuculini 30.8038 |
|------------------------------|
31. | Llorente 30.7956 |
32. | Salvio 30.4432 |
33. | De Federico 29.7104 |
34. | Insua 29.324 |
35. | Sandro 29.3116 |
|------------------------------|
36. | Higuain 28.8316 |
37. | Lucas 28.6256 |
38. | Monzon 28.1088 |
39. | Diego Tardelli 27.5252 |
40. | Young 27.2236 |
|------------------------------|
41. | Boateng 26.779 |
42. | Gomis 26.4176 |
43. | Dossena 26.3022 |
44. | Agbonlahor 26.1234 |
45. | Bolatti 26.1058 |
|------------------------------|
46. | Krkic 26.0006 |
47. | Baxter 25.8682 |
48. | Perez 25.7952 |
49. | Marchisio 25.7502 |
50. | Bartley 25.6378 |
|------------------------------|
51. | El Sharaawy 25.6132 |
52. | Wilshere 25.5976 |
53. | Lavezzi 25.58 |
54. | Ozyakup 25.578 |
55. | Crisetig 25.4842 |
|------------------------------|
56. | Marrone 25.455 |
57. | Morosini 25.441 |
58. | Pacheco 25.4086 |
59. | Comi 25.3848 |
60. | Giuliano 25.3742 |
|------------------------------|
61. | Criscito 25.372 |
62. | Khedira 25.37 |
63. | Banega 25.354 |
64. | Fossati 25.3514 |
65. | Monreal 25.3472 |
|------------------------------|
66. | Laribi 25.3298 |
67. | Macheda 25.2904 |
68. | Kakuta 25.26 |
69. | Saivet 25.2442 |
70. | Zigoni 25.2396 |
|------------------------------|
71. | Marcellis 25.2388 |
72. | Muniain 25.227 |
73. | Neymar 25.1936 |
74. | Capel 25.1784 |
75. | Alcantara 25.1614 |
|------------------------------|
76. | Bocchetti 25.133 |
77. | Castaignos 25.1256 |
78. | Mannini 25.1248 |
79. | Rodwell 25.1236 |
80. | Gallinetta 25.093 |
|------------------------------|
81. | Albertazzi 25.0872 |
82. | Negredo 25.0678 |
83. | Munoz 25.0626 |
84. | Destro 25.0502 |
85. | Datolo 25.045 |
|------------------------------|
86. | Sakho 24.9866 |
87. | Remy 24.9736 |
88. | Douglas P. 24.966 |
89. | Merida 24.955 |
90. | Gosling 24.9442 |
|------------------------------|
91. | Beretta 24.9434 |
92. | Trecarichi 24.9408 |
93. | Maicon 24.9336 |
94. | Aquino 24.927 |
95. | Camacho 24.9186 |
|------------------------------|
96. | Rafael 24.8974 |
97. | Falque 24.8956 |
98. | Sukuta-Pasu 24.8844 |
99. | Alex Texeira 24.8806 |
100. | Luciani 24.8736 |
|------------------------------|
101. | Kroos 24.8534 |
102. | Ciano 24.824 |
103. | Fiorillo 24.8112 |
104. | Caldirola 24.8102 |
105. | Forestieri 24.7596 |
|------------------------------|
106. | Holtby 24.733 |
107. | Mattock 24.7004 |
108. | Wijnaldum 24.6938 |
109. | Immobile 24.6824 |
110. | Sturridge 24.6784 |
|------------------------------|
111. | Douglas C. 24.664 |
112. | Balotelli 24.6534 |
113. | Okaka Chuka 24.633 |
114. | Paloschi 24.5912 |
115. | Borini 24.575 |
|------------------------------|
116. | Lulinha 24.5268 |
117. | Sneijder 24.5184 |
118. | Poli 24.5072 |
119. | Welbeck 24.4848 |
120. | Barazite 24.4654 |
|------------------------------|
121. | Asenjo 24.4576 |
122. | Gebhart 24.4424 |
123. | Bonaventura 24.4124 |
124. | Jo 24.384 |
125. | Zeefuik 24.3726 |
|------------------------------|
126. | Kopplin 24.3602 |
127. | Crescenzi 24.357 |
128. | Fabio 24.299 |
129. | Marquinhos 24.2902 |
130. | Walter 24.2358 |
|------------------------------|
131. | Azpilicueta 24.2128 |
132. | Raggio Garibaldi 24.204 |
133. | Obertan 24.184 |
134. | Marilungo 24.1624 |
135. | Niguez 24.143 |
|------------------------------|
136. | Mustacchio 24.1056 |
137. | Gibbs 24.0734 |
138. | Alan Kardec 24.0468 |
139. | Nsue 24.042 |
140. | Ciro 24.0354 |
|------------------------------|
141. | Di Santo 24.0144 |
142. | Mazzotta 24.0136 |
143. | Parejo 24.0116 |
144. | Piatti 23.9906 |
145. | Marconi 23.9844 |
|------------------------------|
146. | Garcia 23.9798 |
147. | Bellusci 23.9532 |
148. | Badstuber 23.9042 |
149. | Angella 23.8844 |
150. | Traore 23.8768 |
|------------------------------|
151. | Pastore 23.8668 |
152. | Renan Oliveira 23.8606 |
153. | Dentinho 23.8422 |
154. | Pasquato 23.8386 |
155. | Sidnei 23.8098 |
|------------------------------|
156. | Muller 23.7998 |
157. | N'gog 23.782 |
158. | de Jong 23.7696 |
159. | Sciacca 23.745 |
160. | Conti 23.7258 |
|------------------------------|
161. | Aydilek 23.6976 |
162. | Breno 23.6872 |
163. | Ranocchia 23.6842 |
164. | Ariaudo 23.6578 |
165. | D'Agostino 23.5248 |
|------------------------------|
166. | Cattermole 23.452 |
167. | Donald 23.3662 |
168. | Di Gennaro 23.3606 |
169. | Keirrison 23.3364 |
170. | Taison 23.2706 |
|------------------------------|
171. | Ogbonna 23.266 |
172. | Muamba 23.2628 |
173. | Kerlon 23.2566 |
174. | Thuram-Ulien 23.216 |
175. | Aissati 23.1452 |
|------------------------------|
176. | Biabiany 23.1446 |
177. | Cia 23.1186 |
178. | Guilherme 23.1064 |
179. | Denilson 23.0736 |
180. | Vaughan 23.0634 |
|------------------------------|
181. | Howedes 23.0266 |
182. | Enves 23.026 |
183. | Renato Augusto 22.9982 |
184. | Willian 22.9832 |
185. | Felipe Mattioni 22.9728 |
|------------------------------|
186. | Neuer 22.9544 |
187. | Mancienne 22.9368 |
188. | Barill� 22.927 |
189. | Braafheid 22.8942 |
190. | Garay 22.8866 |
|------------------------------|
191. | Martinez 22.8646 |
192. | Nijland 22.77 |
193. | Mannone 22.7202 |
194. | Hart 22.5926 |
195. | Buonanotte 22.5628 |
|------------------------------|
196. | De Silvestri 22.4064 |
197. | Clichy 22.1634 |
198. | Bentley 22.1034 |
199. | Rodriguez 22.0912 |
200. | Carlos Eduardo 21.9412 |
|------------------------------|
201. | Fazio 21.9156 |
202. | Matuidi 21.914 |
203. | Diaby 21.8754 |
204. | Campbell 21.857 |
205. | Johnson 21.8314 |
|------------------------------|
206. | Susaeta 21.7738 |
207. | Candreva 21.7454 |
208. | de la Red 21.7336 |
209. | Dessena 21.6938 |
210. | Giovinco 21.678 |
|------------------------------|
211. | Acquafresca 21.676 |
212. | Baumjohann 21.676 |
213. | Fernando 21.67 |
214. | Menez 21.664 |
215. | Philippe Coutinho 21.657 |
|------------------------------|
216. | Granero 21.6214 |
217. | Cissokho 21.582 |
218. | Rubin 21.5056 |
219. | de Guzman 21.4746 |
220. | Chantome 21.4728 |
|------------------------------|
221. | Zarate 21.4318 |
222. | Drenthe 21.3952 |
223. | Noble 21.3308 |
224. | Wheater 21.3266 |
225. | Danilinho 21.319 |
|------------------------------|
226. | Bonucci 21.2372 |
227. | Aogo 21.202 |
228. | Bruins 21.124 |
229. | Sirigu 21.0692 |
230. | Diego Souza 20.4058 |
|------------------------------|
231. | Filipe 20.3984 |
232. | Ustari 20.347 |
233. | Vermeer 20.2454 |
234. | Marzoratti 19.939 |
235. | Schumacher 19.899 |
|------------------------------|
236. | Santacroce 19.873 |
237. | Abate 19.8068 |
238. | De Ceglie 19.7936 |
239. | Polanski 19.7846 |
240. | Valeri 19.735 |
|------------------------------|
241. | Cabaye 19.7272 |
242. | Bertolo 19.6344 |
243. | Cigarini 19.6094 |
244. | Gouffran 19.6024 |
245. | Osvaldo 19.5108 |
|------------------------------|
246. | Paletta 19.455 |
247. | Kaboul 19.2956 |
248. | Motta M. 19.2806 |
249. | Tissone 19.1218 |
250. | Hernanes 18.9992 |
|------------------------------|
251. | Rafinha 18.644 |
252. | Compper 18.604 |
253. | Nocerino 18.5608 |
254. | Marchetti 18.3352 |
255. | Thiago Neves 18.1756 |
|------------------------------|
256. | Flamini 17.9828 |
257. | Navas 17.7086 |
258. | Rami 17.6108 |
259. | Maidana 17.526 |
260. | Viviano 17.4218 |
|------------------------------|
261. | Zapater 17.3856 |
262. | Tremoulinas 17.37 |
263. | Cahill 17.3532 |
264. | Guarente 17.3118 |
265. | Geromel 17.2784 |
|------------------------------|
266. | Leandro Lima 17.1996 |
267. | Renan 17.1402 |
268. | Coda 16.902 |
269. | Cavenaghi 16.2662 |
270. | Rosina 15.8144 |
|------------------------------|
271. | Loovens 15.3388 |
272. | Baines 14.9104 |
273. | Raggi 14.9082 |
274. | Valbuena 14.8672 |
275. | Hoarau 14.843 |
|------------------------------|
276. | Lell 14.8056 |
277. | Dominguez 12.3568 |
278. | Michel Bastos 12.3126 |
279. | Pennant 12.2902 |
280. | Borriello 12.275 |
|------------------------------|
281. | Turner 11.8858 |
282. | Bianchi 11.8322 |
283. | Maggio C. 10.4634 |
284. | Motta T. 9.5722 |
285. | Pelletieri 9.011 |
|------------------------------|
286. | Brana 8.0838 |
287. | Frey 6.465 |
288. | Amauri 4.4896 |
289. | Ceara 4.3814 |
+------------------------------+
milivella
> I am sorry for having replied after a lot of time. Unfortunately, I
> have little free time.
This is still true. :( So, Daniele, I'll reply to your insightful post
(thanks!) about your prediction formula, but I'll do it later.
> BTW, this is the reason why I'm not posting
> anymore about my FS possible reforms, even if I'm still thinking about
> them.
And this is getting worse, because lately I've not enough mental
energy to think about a good reform. I hope that someone else will do
milivella
> Enzo:
>
> > Un tres big problem with fantasy scout is that it rewards
> > substitute caps equally to starting caps, and friendly caps
> > equally to competitive caps.
>
> > tres big.
>
> ...but easy to fix. ;)
>
> My thoughts about the problems you're pointing at:
>
> - They have been around since some time:http://groups.google.com/group/rec.sport.soccer/browse_thread/thread/...http://groups.google.com/group/rec.sport.soccer/msg/86c872f512068dcf
>
> - They are still in my agenda of things to think about, but there are
> "tres bigger" problems to solve first IMHO! ;) It goes without saying:
> anyone can propose a reform about this (or any other) point in any
> moment.
>
> - I wouldn't vote against a reform that gives different weights to
> starting/sub caps, official/friendlies caps.
Little argument against: to compute the score of just a scout, or to
compare the scores of just two scouts, you would have to check all the
matches played by the eight NTs since 2007. Under the current rules,
you should just check the data of the players picked by that scout /
those scouts.
--
Cheers
milivella
milivella
> milivella:
>
> > Enzo:
>
> > > Un tres big problem with fantasy scout is that it rewards
> > > substitute caps equally to starting caps, and friendly caps
> > > equally to competitive caps.
>
> > > tres big.
>
> > ...but easy to fix. ;)
>
> > My thoughts about the problems you're pointing at:
>
> > - They have been around since some time:http://groups.google.com/group/rec.sport.soccer/browse_thread/thread/...
>
> > - They are still in my agenda of things to think about, but there are
> > "tres bigger" problems to solve first IMHO! ;) It goes without saying:
> > anyone can propose a reform about this (or any other) point in any
> > moment.
>
> > - I wouldn't vote against a reform that gives different weights to
> > starting/sub caps, official/friendlies caps.
>
> Little argument against: to compute the score of just a scout, or to
> compare the scores of just two scouts, you would have to check all the
> matches played by the eight NTs since 2007.
Or (assuming that for each picked player you have a list of all the
matches he played in NT, something that right now you can only find,
AFAIK, in footballdatabase.ey) you should check each match played by
each player picked by that scout / those scouts.
--
Cheers
milivella
milivella
> "milivella" <milive...@gmail.com> wrote in message
>
> news:043be24a-9c78-44ce...@r24g2000yqd.googlegroups.com...
>
> > Futbolmetrix:
>
> >> Gory statistical details to follow.
>
> > Feed us with them!
>
> OK, you asked for it.
Thanks. A lot of them.
Sorry for the delay of this reply.
> Let's start from the observation that it may be too ambitious trying to
> predict a player's total career score.
Yep.
> But we could start from something a
> lot easier, like predicting a player's score in 2010. So, we could start
> from a model like
>
> y(i,t) = b0 + b1*y(i,t-1) + u(i,t)
>
> where y(i,t) is the score of player i in year t and u(i,t) is an error term.
> b0 and b1 can be estimated from the existing set of players.
[I know that we're entering a specialists' field, so I can just try to
understand something, and ask some questions. Hoping that they are not
too stupid. Or at least that I'll give you the chance to start a
"Futbolmetrix's prediction formula for dummies".]
I.e. you are supposing that there is a base level for all the players
picked in FS? (something like: if a player has been picked picked, it
means that he has some chances to win some caps; it's not like
randomly picking a player among a list of all the existing players;
etc.)
> 1) the formula was estimated based only on those players who currently have
> positive caps, it fails to recognize that even an uncapped player has some
> positive probability of breaking into the NT, and then he will start to
> accumulate caps.
Why? Isn't this the function of b0?
> 2) The regression model above fails to recognize that FS score has two
> important properties: a) it's an integer number, and b) it's non-negative.
> So, if you try to forecast a single's player individual career, you may end
> up with negative score (and you will surely end up with non-integer score).
How could you end up with a negative score? Aren't b0 and b1 always
positive?
> a) If caps(t-1)>0, predict caps(t) using a zero-inflated Poisson model based
> on caps(t-1). [I can give you more gory details if you want, but the basic
> idea is that a zero-inflated model recognizes that there are more zeros in
> the data than a typical Poisson distribution]
> b) If caps(t-1)=0, predict caps(t) using a zero-inflated Poisson model based
> on age and age^2. It captures the fact that players can break into the NT,
> but the probability of accumulating caps initially increases, and then
> decreases with age.
Great. Why don't you use a similar formula for players that have
caps>0? Or do you?
> c) Given caps(t), predict the number of goals as: 0, if caps(t)=0; Poisson
> depending on goals and caps in year t-1, and the average goal per game rate
> among the players currently picked (at the moment it's 1/9)
You can't use the goal per game ratio of that player because it could
be a distorted sample? (e.g. 1 cap, 1 goal = 1 goal per game)
Couldn't you use a Bayesian estimate? We don't know what will be
Santon's goal per game ratio (too few caps), but we're pretty sure
that 1/9 is too much for Cannavaro, too low for Eto'o.
> > A *big warm thank* for the prediction formula! As soon as you'll tell
> > us what the formula is, I'll put it on the website.
>
> There is no exact formula, but a reasonable approximation is:
:/ So we should change it every year? (or month?)
> PREDICTED SCORE = CURRENT SCORE + ADDITIONAL SCORE
>
> ADDITIONAL SCORE = 255.1718 + 0.0223868*CURRENT_SCORE - 50.16514*AGE +
> 3.98231*AGE^2 - 0.1351248*AGE^3 + 0.001628*AGE^4 +
> 2.193766*CAPS2009 -0.0978368*CAPS2009^2 + 0.0263713*GOALS2009
I'll add it to the site. Is it OK if instead of caps2009 I use
caps_in_the_last_12_months?
Do all that age^x assure that a 45 year old player (with 0 caps in
2009!) won't be predicted any cap?
> Things that I don't like about this method:
>
> a) Despite the attempts to separate between players who have already broken
> into the NT and those who have not, I think the model is still to generous
> with the latter. I believe that the difference between Marchisio (age 23, 2
> caps, predicted score 25.75) and Abate (age 23, 0 caps, predicted score
> 19.8) should be greater than what it is. And there are a bit too many caps
> predicted for the 18-year olds who haven't even ever played a professional
> match yet.
Yep. Probably because they've been picked in a different context (more
competition) than early picks, that spread chances of a bright future
to them.
> b) It should really use more information in making predictions.
An additional information that we have and that we're not using is
caps and goals two/three/etc. years ago.
But I know that you mean external data (club football, etc.).
> a) it's really quite flexible, and it's very easy to incorporate additional
> information. For example, that strikers are expected to have a higher goal
> per game rate than defenders.
OK (see above).
> > The scores that your formula predicts seems a bit too close to me
> > (e.g. 5th = 43 points, 15th = 38 points), but it's just my ignorant
> > instinct.
>
> Yes, see above. But I think that it's a reflection of the fact that there's
> still a *lot* of uncertainty about these players. Gago has lost his starting
> spot at Real Madrid. How long will he hold onto his starting spot for the
> Argentina NT?
Right.
> > Have you checked what 2009 scores your formula would have predicted
> > one and/or two years ago?
>
> And take away all the fun from you?
:)
I could do it in time... but you could tell me the exact formula. ;)
--
Cheers
milivella
milivella
> Futbolmetrix:
>
> > a) If caps(t-1)>0, predict caps(t) using a zero-inflated Poisson model based
> > on caps(t-1). [I can give you more gory details if you want, but the basic
> > idea is that a zero-inflated model recognizes that there are more zeros in
> > the data than a typical Poisson distribution]
> > b) If caps(t-1)=0, predict caps(t) using a zero-inflated Poisson model based
> > on age and age^2. It captures the fact that players can break into the NT,
> > but the probability of accumulating caps initially increases, and then
> > decreases with age.
>
> Great. Why don't you use a similar formula for players that have
> caps>0? Or do you?
_Maybe_ now I got it a bit more.
For each player, you predict his possible career many times, and then
you compute the average among these possible careers. You compute each
possible career year by year, using formula a if the player had (in
that simulation) at least 1 cap in the previous year, otherwise
formula b.
This is why there is no exact formula. The one you gave me...
> > ADDITIONAL SCORE = 255.1718 + 0.0223868*CURRENT_SCORE - 50.16514*AGE +
> > 3.98231*AGE^2 - 0.1351248*AGE^3 + 0.001628*AGE^4 +
> > 2.193766*CAPS2009 -0.0978368*CAPS2009^2 + 0.0263713*GOALS2009
...is the result of a polynomial regression.
Have I understood something right? :)
---
I have four more questions for you:
- How could we post the prediction in the site? Should you run the
simulations every time? (e.g. after every international match)
- If I'm not wrong, if Messi has a feud with Maradona and has 0 caps
in 2010, in 2011 he will be treated as any other 24 year old player
with 0 caps in the previous year. Wouldn't it be better to count (with
less weight, of course) previous years as well? I mean: if a player
has 10 caps each year for 4 years and then 0 caps in a given year, the
causes could be many, from death to disciplinary suspensions, and,
while some of them mean that he won't have a future in the NT, others
don't. So I think that such a player should be predicted less caps
than if he had 10 caps in 2010, but more caps than a player who has
never been in the NT.
- Your method, being based also on the career of the other picked
players, can't be used to predict the career of a player that is out
of the game context, can it?
- If the answer to the previous question is "no it can't": so your
method can't be evaluated against other eventual methods asking them
to predict e.g. Maradona's career based on his data when he was 21...
What comparison method would you use?
Of course (I should have said it earlier, I forgot to do it) you can
skip my questions: it's not important that I understand your work. The
important thing is that you did it! thanks again. :)
--
Cheers
milivella
futbol...@yahoo.com
>
Skipped first questions because it looks as if you got the answer in
your next post.
>
> > c) Given caps(t), predict the number of goals as: 0, if caps(t)=0; Poisson
> > depending on goals and caps in year t-1, and the average goal per game rate
> > among the players currently picked (at the moment it's 1/9)
>
> You can't use the goal per game ratio of that player because it could
> be a distorted sample? (e.g. 1 cap, 1 goal = 1 goal per game)
Exactly.
> Couldn't you use a Bayesian estimate? We don't know what will be
> Santon's goal per game ratio (too few caps), but we're pretty sure
> that 1/9 is too much for Cannavaro, too low for Eto'o.
In fact, the current formula is already Bayesian: it assigns some
weight to the average gpg of all players, and some weight to the gpg
rate of player i in year t-1. The latter weight is larger, the more
caps obtained by player i.
>
> > There is no exact formula, but a reasonable approximation is:
>
> :/ So we should change it every year? (or month?)
Well, as more data becomes available, we should tend to get more
accurate estimates, and at some point the formula will stabilize.
Hopefully. I'm pretty sure it's not stable now.
>
> > PREDICTED SCORE = CURRENT SCORE + ADDITIONAL SCORE
>
> > ADDITIONAL SCORE = 255.1718 + 0.0223868*CURRENT_SCORE - 50.16514*AGE +
> > 3.98231*AGE^2 - 0.1351248*AGE^3 + 0.001628*AGE^4 +
> > 2.193766*CAPS2009 -0.0978368*CAPS2009^2 + 0.0263713*GOALS2009
>
> I'll add it to the site. Is it OK if instead of caps2009 I use
> caps_in_the_last_12_months?
I guess it won't matter much.
> Do all that age^x assure that a 45 year old player (with 0 caps in
> 2009!) won't be predicted any cap?
Yeah, pertty much. In fact, in generating the simulations I assumed
that nobody gets any caps beyond age 36, which of course can be
modified.
> Yep. Probably because they've been picked in a different context (more
> competition) than early picks, that spread chances of a bright future
> to them.
This can be adjusted for. Include in the regression the number of
scouts present at the time a player was picked.
> > b) It should really use more information in making predictions.
>
> An additional information that we have and that we're not using is
> caps and goals two/three/etc. years ago.
>
> But I know that you mean external data (club football, etc.).
We could start by including a player's role (GK/DF/MF/FW) and I agree
with Benny that in the vast majority of cases it's something clearly
defined.
futbol...@yahoo.com
Yes, you got it exactly.
>
> This is why there is no exact formula. The one you gave me...
>
> > > ADDITIONAL SCORE = 255.1718 + 0.0223868*CURRENT_SCORE - 50.16514*AGE +
> > > 3.98231*AGE^2 - 0.1351248*AGE^3 + 0.001628*AGE^4 +
> > > 2.193766*CAPS2009 -0.0978368*CAPS2009^2 + 0.0263713*GOALS2009
>
> ...is the result of a polynomial regression.
Bingo!
> I have four more questions for you:
>
> - How could we post the prediction in the site? Should you run the
> simulations every time? (e.g. after every international match)
Once every few months sounds more reasonable.
> - If I'm not wrong, if Messi has a feud with Maradona and has 0 caps
> in 2010, in 2011 he will be treated as any other 24 year old player
> with 0 caps in the previous year. Wouldn't it be better to count (with
> less weight, of course) previous years as well? I mean: if a player
> has 10 caps each year for 4 years and then 0 caps in a given year, the
> causes could be many, from death to disciplinary suspensions, and,
> while some of them mean that he won't have a future in the NT, others
> don't. So I think that such a player should be predicted less caps
> than if he had 10 caps in 2010, but more caps than a player who has
> never been in the NT.
Yes, of course. But you never have to assign arbitrary weights. Just
include in the Poisson regression the cumulative number of caps up to
that point. By the way, I don't think this will matter much in terms
of the prediction: there haven't been that many interrupted careers up
to now.
> - Your method, being based also on the career of the other picked
> players, can't be used to predict the career of a player that is out
> of the game context, can it?
>
> - If the answer to the previous question is "no it can't": so your
> method can't be evaluated against other eventual methods asking them
> to predict e.g. Maradona's career based on his data when he was 21...
> What comparison method would you use?
The main problem with applying this method to Maradona is that the
number of international matches has increased substantially relative
to the 1980s. But could this method be applied to the careers of
current Portuguese or Russian players? Yes, why not? You're making the
assumption that the career path of a Portuguese international is
governed by the same parameters as the career of one of the Big 8.
That may be debatable, but it's no more outlandish than many other
assumptions that are in this model.
milivella
[I'll reply later.]
> > > PREDICTED SCORE = CURRENT SCORE + ADDITIONAL SCORE
>
> > > ADDITIONAL SCORE = 255.1718 + 0.0223868*CURRENT_SCORE - 50.16514*AGE +
> > > 3.98231*AGE^2 - 0.1351248*AGE^3 + 0.001628*AGE^4 +
> > > 2.193766*CAPS2009 -0.0978368*CAPS2009^2 + 0.0263713*GOALS2009
>
> > I'll add it to the site. Is it OK if instead of caps2009 I use
> > caps_in_the_last_12_months?
>
> I guess it won't matter much.
It's live now:
http://fantasyscout.altervista.org/data.htm
Please tell me if I got to change something (i.e. the formula, or its
description in the last tab).
milivella
> Skipped first questions because it looks as if you got the answer in
> your next post.
Again, I'll fully reply later. Now I just have a question that I could
forget before replying, so here it is: I guess that, using your
method, if there are more 0 cap players, the ones with some caps have
a better prediction (than in a situation where there are less 0 cap
players). Am I wrong?
--
Cheers
milivella
Futbolmetrix
>
> Again, I'll fully reply later. Now I just have a question that I could
> forget before replying, so here it is: I guess that, using your
> method, if there are more 0 cap players, the ones with some caps have
> a better prediction (than in a situation where there are less 0 cap
> players). Am I wrong?
>
Not sure I undersrtand the question. What do you mean by "better
prediction"? More accurate prediction? Predicted to have higher scores?
D
milivella
> "milivella" <milive...@gmail.com> wrote in message
Sorry. "Predicted to have high scores".
--
Cheers
milivella
milivella
Indeed I was (and still am...) confused. I meant: let's assume that
you have a group G of 50 players, some with caps, some still at 0. Run
a prediction about their final score, and let's call P the predicted
score of each player. Now let's fill the group with 50 0-cappers: this
is the group G'. A new prediction is computed, so each player is
predicted a final score P'.
My questions are:
- Take a 0-capper (already present in G). Is P' higher than, equal to,
or lower than P? (I guess: lower)
- Same question, for a player with some caps. (I guess: lower?)
--
Cheers
milivella
Futbolmetrix
>
> Indeed I was (and still am...) confused. I meant: let's assume that
> you have a group G of 50 players, some with caps, some still at 0. Run
> a prediction about their final score, and let's call P the predicted
> score of each player. Now let's fill the group with 50 0-cappers: this
> is the group G'. A new prediction is computed, so each player is
> predicted a final score P'.
> My questions are:
> - Take a 0-capper (already present in G). Is P' higher than, equal to,
> or lower than P? (I guess: lower)
The new zero-cappers have zero-caps in 2008 and 2009. Remember that the key
part of my formula involves estimating the number of caps in year t for
players that had zero caps in year t-1. I'll assume that when you talk about
zero-cappers, you mean the players who had zero caps in 2008. The average
number of caps of these players in 2009 will be greater than zero. However,
when you add the new zero-cappers (who also have zero caps) in 2009, it
lowers the average numbers of 2009 caps for all zero-cappers in 2008.
Therefore, the predicted number of caps in the new group of zero cappers is
indeed lower, as per your intuition. (if this is what you meant).
> - Same question, for a player with some caps. (I guess: lower?)
The prediction for the positive cappers (i.e., those who already had
positive caps in 2008) is not affected at all by the inclusion of the new
zero-cappers.
D
milivella
> On Nov 9, 4:27 pm, milivella <milive...@gmail.com> wrote:
>
>
>
> Skipped first questions because it looks as if you got the answer in
> your next post.
Yoo-hoo! :)
> > Couldn't you use a Bayesian estimate? We don't know what will be
> > Santon's goal per game ratio (too few caps), but we're pretty sure
> > that 1/9 is too much for Cannavaro, too low for Eto'o.
>
> In fact, the current formula is already Bayesian: it assigns some
> weight to the average gpg of all players, and some weight to the gpg
> rate of player i in year t-1. The latter weight is larger, the more
> caps obtained by player i.
Perfect. I hope that you hadn't specified this earlier, because it
would mean that I have no clue!
> > > There is no exact formula, but a reasonable approximation is:
>
> > :/ So we should change it every year? (or month?)
>
> Well, as more data becomes available, we should tend to get more
> accurate estimates, and at some point the formula will stabilize.
> Hopefully. I'm pretty sure it's not stable now.
OK. Just send me the updated formula when you want, and I'll put it on
the site.
> > Yep. Probably because they've been picked in a different context (more
> > competition) than early picks, that spread chances of a bright future
> > to them.
>
> This can be adjusted for. Include in the regression the number of
> scouts present at the time a player was picked.
This is what I had in mind.
--
Cheers
milivella
milivella
> > - If I'm not wrong, if Messi has a feud with Maradona and has 0 caps
> > in 2010, in 2011 he will be treated as any other 24 year old player
> > with 0 caps in the previous year. Wouldn't it be better to count (with
> > less weight, of course) previous years as well? I mean: if a player
> > has 10 caps each year for 4 years and then 0 caps in a given year, the
> > causes could be many, from death to disciplinary suspensions, and,
> > while some of them mean that he won't have a future in the NT, others
> > don't. So I think that such a player should be predicted less caps
> > than if he had 10 caps in 2010, but more caps than a player who has
> > never been in the NT.
>
> Yes, of course. But you never have to assign arbitrary weights. Just
> include in the Poisson regression the cumulative number of caps up to
> that point.
Of course.
> By the way, I don't think this will matter much in terms
> of the prediction: there haven't been that many interrupted careers up
> to now.
Right.
> > - Your method, being based also on the career of the other picked
> > players, can't be used to predict the career of a player that is out
> > of the game context, can it?
>
> > - If the answer to the previous question is "no it can't": so your
> > method can't be evaluated against other eventual methods asking them
> > to predict e.g. Maradona's career based on his data when he was 21...
> > What comparison method would you use?
>
> The main problem with applying this method to Maradona is that the
> number of international matches has increased substantially relative
> to the 1980s.
I thought that the main problem was that the prediction is based on
the career of the other picked players, but you have explained me that
it isn't so:
http://groups.google.com/group/rec.sport.soccer/msg/621d5687a800c525
But then we can compare your method with others, because they all
share the same problem; cant' we? In your opinion, how many retired
players would be enough for a benchmark? 100?
--
Cheers
milivella
milivella
> "milivella" <milive...@gmail.com> wrote in message
>
> news:d302fd29-8790-49f7...@g23g2000yqh.googlegroups.com...
>
>
>
> > Indeed I was (and still am...) confused. I meant: let's assume that
> > you have a group G of 50 players, some with caps, some still at 0. Run
> > a prediction about their final score, and let's call P the predicted
> > score of each player. Now let's fill the group with 50 0-cappers: this
> > is the group G'. A new prediction is computed, so each player is
> > predicted a final score P'.
> > My questions are:
> > - Take a 0-capper (already present in G). Is P' higher than, equal to,
> > or lower than P? (I guess: lower)
>
> The new zero-cappers have zero-caps in 2008 and 2009. Remember that the key
> part of my formula involves estimating the number of caps in year t for
> players that had zero caps in year t-1. I'll assume that when you talk about
> zero-cappers, you mean the players who had zero caps in 2008. The average
> number of caps of these players in 2009 will be greater than zero. However,
> when you add the new zero-cappers (who also have zero caps) in 2009, it
> lowers the average numbers of 2009 caps for all zero-cappers in 2008.
> Therefore, the predicted number of caps in the new group of zero cappers is
> indeed lower, as per your intuition. (if this is what you meant).
Yes it is.
> > - Same question, for a player with some caps. (I guess: lower?)
>
> The prediction for the positive cappers (i.e., those who already had
> positive caps in 2008) is not affected at all by the inclusion of the new
> zero-cappers.
Sorry, I had forgotten what you have written explaining your method:
http://groups.google.com/group/rec.sport.soccer/msg/381211388ed5771f
"If caps(t-1)>0, predict caps(t) using a zero-inflated Poisson model
based on caps(t-1)."
--
Cheers
milivella