Main
Date: 14 Nov 2006 20:13:47
From: jclaxton
Subject: Endgame Analysis - Help Please
The following position was reached in a USCF rated game recently. We were
going over the position in a coffee shop later and it seemed that whoever
pushed hardest lost. White has the two Bishops - Black is up a Pawn.
Blacks Bishop is (currently) trapped inside his own Pawns. After Blacks
Knight relocates, he has the ...g6, ...f5 break.

Either post results here, or mail to [email protected].

Thanks




 
Date: 15 Nov 2006 09:58:54
From: David Richerby
Subject: Re: Bonus points etc.
Ed Seedhouse <[email protected] > wrote:
> As FIDE rates many more players than it did in, say, the 1960s, the
> number players rated over 2700 these days may simply be because of
> such an effect, but I suppose some fairly sophisticated calculations
> would have to be made to determine if that indeed is the cause of
> the supposed ratings inflation in the FIDE system at the top.

It's possible that rating more players causes inflation and, hence,
means that there are more 2700+ players.

The standard argument for this kind of thing, namely ``Because there
are more rated players, there are more outliers'' doesn't work, of
course. Because FIDE has always rated the top players, all the 2700+
players today would have FIDE ratings (though not necessarily such
high FIDE ratings) if they'd been active players twenty years ago.


Dave.

--
David Richerby Carnivorous Shack (TM): it's like
www.chiark.greenend.org.uk/~davidr/ a house in the woods but it's full
of teeth!


  
Date: 21 Nov 2006 13:55:07
From: bellatori
Subject: Re: Bonus points etc.
This is quite correct. The probability of 2700+ is a direct function of the
number of players under the graph, as it were. The question that arises is
whether the number has increased to such an extent that we should see the
number of players over 2700+ and 2800+ that we now see. My current
calculations suggest that there is rating inflation and it is reasonably
modelled by a truncated normal distribution.



   
Date: 22 Nov 2006 09:15:21
From: David Richerby
Subject: Re: Bonus points etc.
bellatori <[email protected] > wrote:
> David Richerby wrote:
>> It's possible that rating more players causes inflation and, hence,
>> means that there are more 2700+ players.
>>
>> The standard argument for this kind of thing, namely ``Because there
>> are more rated players, there are more outliers'' doesn't work, of
>> course. Because FIDE has always rated the top players, all the 2700+
>> players today would have FIDE ratings (though not necessarily such
>> high FIDE ratings) if they'd been active players twenty years ago.
>
> This is quite correct.

Um. I've reinstated the text you snipped. You say that I'm right and
then explain why the exact opposite is true.


> The probability of 2700+ is a direct function of the number of
> players under the graph, as it were.

Only if the players are chosen randomly. But the players aren't
chosen randomly: FIDE chooses only the strongest players for rating.

Let me try to explain it in a slightly different way. Suppose we
collect all the money on the planet and distribute it randomly back to
people, giving each person an amount drawn from a normal distribution.
Now, consider these two cases.

First, you pick 1000 people at random and I pick 100,000 at random.
The ten richest people that I see are almost certainly far richer than
the ten richest you see. This is because increasing the size of a
random sample increases the probability of outliers.

Second, you pick the thousand richest people in the world and I pick
the hundred thousand richest. Here, we both see exactly the same top
ten and considering more people doesn't make a difference.

The situation of FIDE rating more players is analogous to the second
situation, not the first. It's a little more complicated than that
because the new people in the larger non-random sample interact with
the old people and exchange money between themselves, which might
cause the rich to get richer or, on the other hand, might not. But
that's a second-order effect.


Dave.

--
David Richerby Swiss Sumerian Toy (TM): it's like a
www.chiark.greenend.org.uk/~davidr/ fun child's toy that's really old but
it's made in Switzerland!


  
Date: 15 Nov 2006 12:30:15
From: David Kane
Subject: Re: Bonus points etc.

"David Richerby" <[email protected] > wrote in message
news:EWD*[email protected]...
> Ed Seedhouse <[email protected]> wrote:
>> As FIDE rates many more players than it did in, say, the 1960s, the
>> number players rated over 2700 these days may simply be because of
>> such an effect, but I suppose some fairly sophisticated calculations
>> would have to be made to determine if that indeed is the cause of
>> the supposed ratings inflation in the FIDE system at the top.
>
> It's possible that rating more players causes inflation and, hence,
> means that there are more 2700+ players.


Please explain. Why does rating more players cause inflation?

> The standard argument for this kind of thing, namely ``Because there
> are more rated players, there are more outliers'' doesn't work, of
> course. Because FIDE has always rated the top players, all the 2700+
> players today would have FIDE ratings (though not necessarily such
> high FIDE ratings) if they'd been active players twenty years ago.

An increase in the numbers of a particular rating segment does not by itself
indicate inflation.





   
Date: 16 Nov 2006 10:12:55
From: David Richerby
Subject: Re: Bonus points etc.
David Kane <[email protected] > wrote:
> David Richerby <[email protected]> wrote:
>> It's possible that rating more players causes inflation and, hence,
>> means that there are more 2700+ players.
>
> Please explain. Why does rating more players cause inflation?

Where did I say it did?


Dave.

--
David Richerby Poetic Lotion (TM): it's like a
www.chiark.greenend.org.uk/~davidr/ soothing hand lotion but it's in
verse!


    
Date: 21 Nov 2006 14:05:30
From: bellatori
Subject: Re: Bonus points etc.
Rating more players does cause inflation. this is acceptable becuase the
rating expresses a probability of performance. If you have 100 players
then on average you will have 2.5 (ouch for the half!) with a rating of
2400+. If you have 1000 players then you have a probability of a player at
2600+ (I'm guesstimating these as I don't have the tables here)



     
Date: 21 Nov 2006 19:06:23
From: David Kane
Subject: Re: Bonus points etc.

"bellatori" <[email protected] > wrote in message
news:[email protected]...
> Rating more players does cause inflation. this is acceptable becuase the
> rating expresses a probability of performance. If you have 100 players
> then on average you will have 2.5 (ouch for the half!) with a rating of
> 2400+. If you have 1000 players then you have a probability of a player at
> 2600+ (I'm guesstimating these as I don't have the tables here)
>

This is not inflation. If you double the number of people, you'll have twice as
many people with each rating. If the ratings of exisiting players don't change,
you don't say that inflation has occurred.

In any case, ratings are *not* assigned by fitting the results to some
assumed distribution, nor should they be. We could ELO rate the
entire population of the world and wouldn't add a single 2800 player
if we did it correctly. Why not? Because none of those newly-rated
players would consistently defeat existing 2600 players etc.







      
Date: 22 Nov 2006 08:54:51
From: David Richerby
Subject: Re: Bonus points etc.
David Kane <[email protected] > wrote:
> "bellatori" <[email protected]> wrote:
>> Rating more players does cause inflation. this is acceptable
>> becuase the rating expresses a probability of performance. If you
>> have 100 players then on average you will have 2.5 (ouch for the
>> half!) with a rating of 2400+.

That's just two players with a rating of 2400+ and one player with a
rating of 1200+. :-)


>> If you have 1000 players then you have a probability of a player at
>> 2600+ (I'm guesstimating these as I don't have the tables here)
>
> This is not inflation. If you double the number of people, you'll
> have twice as many people with each rating. If the ratings of
> exisiting players don't change, you don't say that inflation has
> occurred.

Exactly.


> We could ELO rate the entire population of the world and wouldn't
> add a single 2800 player if we did it correctly. Why not? Because
> none of those newly-rated players would consistently defeat existing
> 2600 players etc.

Exactly. This is a point I've tried to make several times. For as
long as the FIDE rating system has been in existence, it has always
given a rating to the top players in the world. Simply increasing the
size of the rating pool does not cause you to discover any new world-
class players so does not cause the top end of the rating scale to
increase by increasing the probability of outliers. If it does cause
the top end of the scale to increase, it must be for some other
reason.


Dave.

--
David Richerby Old-Fashioned Solar-Powered Robot
www.chiark.greenend.org.uk/~davidr/ (TM): it's like a high-tech robot but
it doesn't work in the dark and it's
perfect for your grandparents!


    
Date: 16 Nov 2006 10:34:27
From: David Kane
Subject: Re: Bonus points etc.

"David Richerby" <[email protected] > wrote in message
news:NKB*[email protected]...
> David Kane <[email protected]> wrote:
>> David Richerby <[email protected]> wrote:
>>> It's possible that rating more players causes inflation and, hence,
>>> means that there are more 2700+ players.
>>
>> Please explain. Why does rating more players cause inflation?
>
> Where did I say it did?
>

You stated that it was possible. Please explain.

Adding players to a pool doesn't, in principle, change
the ratings of those who are already in it. Of course, real world
implementations of ratings may make various approximations
that have that effect.

For example, imagine a group of new players who have never
played against players with ratings, but who play rated games
against each other many times. Many ratings organizations will
assign these new players ratings, even though they are not
playing in the existing rating pool. The ratings they receive
will be based upon assumptions that the ratings agency makes
for unrated players. If these assumptions were incorrect, their
eventual entrance into the original pool could inflate, or
deflate, the ratings of those originally in the pool.






   
Date: 16 Nov 2006 02:18:20
From: Ed Seedhouse
Subject: Re: Bonus points etc.
On Wed, 15 Nov 2006 12:30:15 -0800, "David Kane"
<[email protected] > wrote:

>
>"David Richerby" <[email protected]> wrote in message
>news:EWD*[email protected]...
>> Ed Seedhouse <[email protected]> wrote:
>>> As FIDE rates many more players than it did in, say, the 1960s, the
>>> number players rated over 2700 these days may simply be because of
>>> such an effect, but I suppose some fairly sophisticated calculations
>>> would have to be made to determine if that indeed is the cause of
>>> the supposed ratings inflation in the FIDE system at the top.
>>
>> It's possible that rating more players causes inflation and, hence,
>> means that there are more 2700+ players.
>
>
>Please explain. Why does rating more players cause inflation?

It doesn't. But it can give the appearance of inflation to the casual
observer who is only noticing the top ratings.



  
Date: 15 Nov 2006 14:27:19
From: Ed Seedhouse
Subject: Re: Bonus points etc.
On 15 Nov 2006 09:58:54 +0000 (GMT), David Richerby
<[email protected] > wrote:


>It's possible that rating more players causes inflation and, hence,
>means that there are more 2700+ players.

Merely increasing the ratings base cannot by itself cause inflation no
matter what the distribution of ratings is. But as I tried to show it
can lead to the appearance that inflation is happening even when it
isn't.

>The standard argument for this kind of thing, namely ``Because there
>are more rated players, there are more outliers'' doesn't work, of
>course.

If the underlying distribution of ratings stays the same then it is
necessarily true.

>Because FIDE has always rated the top players, all the 2700+
>players today would have FIDE ratings (though not necessarily such
>high FIDE ratings) if they'd been active players twenty years ago.

So far as I can see that doesn't matter. The same players may still be
on top but they will also be at a higher top. Their ratings may
increase despite no change in skill if the rating pool size increases
without any underlying change of distribution (and we were discussing a
proposal which would ensure this). The weakest players might also see
their ratings go down despite no decrease in skill. But since an
expanding rating pool means new players coming in then it's likely that
some new players may also occupy the tails as well. And of course all
the slots in between but that probably wouldn't be noticed.







   
Date: 15 Nov 2006 16:35:59
From: David Richerby
Subject: Re: Bonus points etc.
Ed Seedhouse <[email protected] > wrote:
> David Richerby <[email protected]> wrote:
>> The standard argument for this kind of thing, namely ``Because
>> there are more rated players, there are more outliers'' doesn't
>> work, of course.
>
> If the underlying distribution of ratings stays the same then it is
> necessarily true.

But the underlying distribution of ratings *doesn't* stay the same!
That's the whole point. FIDE isn't expanding the rating pool by just
adding more players (who could be expected to have the same rating
distribution as the previously-rated players): FIDE is expanding the
rating pool by extending it downwards.

FIDE used to only rate players above 2000, as I recall. Therefore,
the ratings curve used to be flat zero below 2000, by construction.
Now, FIDE rates down to 1800 so the curve is no longer flat zero in
the range 1800--2000. If FIDE later start to rate down to 1600, the
distribution will change again.

(I may have got the numbers 2000 and 1800 wrong but the exact values
don't matter for these purposes.)


Dave.

--
David Richerby Mouldy Hungry Priest (TM): it's like
www.chiark.greenend.org.uk/~davidr/ a man of the cloth but it'll eat you
and it's starting to grow mushrooms!


    
Date: 16 Nov 2006 02:17:15
From: Ed Seedhouse
Subject: Re: Bonus points etc.
On 15 Nov 2006 16:35:59 +0000 (GMT), David Richerby
<[email protected] > wrote:

>Ed Seedhouse <[email protected]> wrote:
>> David Richerby <[email protected]> wrote:
>>> The standard argument for this kind of thing, namely ``Because
>>> there are more rated players, there are more outliers'' doesn't
>>> work, of course.
>>
>> If the underlying distribution of ratings stays the same then it is
>> necessarily true.
>
>But the underlying distribution of ratings *doesn't* stay the same!
>That's the whole point.

No it's not the whole point. It's not the point at all. I was
responding to a proposal that the rating system be set up so that the
underlying distribution does not change. That it be normalized to a
gaussian distribution with standard deviations of 200 points and a
predetermined mean rating.

Of course at present it does not stay unchanged (or at least nothing is
done to prevent it from changing) but that isn't what I was talking
about. I was suggesting that even given a rating system that is forced
to normal and a fixed mean and standard deviation the ratings of the top
players would go up if the rating pool expanded significantly, and that
this would probably be interpreted by the average player as proof that
inflation was still happening.



     
Date: 21 Nov 2006 13:58:37
From: bellatori
Subject: Re: Bonus points etc.
I agree entirely. The top ratings are a function of the number of players
within the pool. What I object to is that the ratings are skewed by the
acceleration of new players and the deceleration of very old ones together
with the truncating effect all of which skew the top higher. If you are
the big fish in a bigger pool then I have no objection to the rating
increasing.