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)

(Comparison with the ranking just before tonight's update.)

Discuss! ;)

--
Cheers
milivella

Futbolmetrix:

"milivella" <milive... at (no spam) 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

"milivella" <milive... at (no spam) gmail.com> wrote in message

news:347fcc5b-bf30-44da-a3ac-713fcaf42083 at (no spam) k17g2000yqb.googlegroups.com...
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.

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

Futbolmetrix:





"milivella" <milive... at (no spam) gmail.com> wrote in message

news:347fcc5b-bf30-44da-a3ac-713fcaf42083 at (no spam) k17g2000yqb.googlegroups.com....
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.

:)

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- Hide quoted text -

- Show quoted text -

On Oct 16, 10:52 pm, milivella <milive... at (no spam) gmail.com> wrote:





Futbolmetrix:

"milivella" <milive... at (no spam) gmail.com> wrote in message

news:347fcc5b-bf30-44da-a3ac-713fcaf42083 at (no spam) k17g2000yqb.googlegroups.com....
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.

:)

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- Hide quoted text -

- 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.- Hide quoted text -

- Show quoted text -

More: the formula is self-consistent!

More: the formula is self-consistent!

On Oct 17, 5:43 pm, milivella <milive... at (no spam) 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.

[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.

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.

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.

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

Enzo:





On Oct 17, 5:43 pm, milivella <milive... at (no spam) 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/...





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- Hide quoted text -

- Show quoted text -- Hide quoted text -

- Show quoted text -- Hide quoted text -

- Show quoted text -

"milivella" <milivella at (no spam) gmail.com> wrote in message
news:9961c7fe-f3df-4b9b-993c-14ce4ac08f11 at (no spam) j24g2000yqa.googlegroups.com...

My predicted final ranking (very partial, algorithm still needs
substantial improvement):

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.