Ange1o DePa1ma wrote:
"Kenneth Sloan" wrote

Sloan may be right that the scale is arbitrary, and we may employ
negative numbers, fractions, or even imaginary numbers if we like. We can
make an average rating zero instead of 1500.
Wherever did you get that the average rating was 1500?
Have you been talking to Dubeck?

Jesus, Ken, I was just picking a number out of my hat. 1200. Whatever.


The point is: the "average rating" is not an interesting concept when
specifying the rating system. It depends more on the demographics (who
chooses to play in your system) than on anything else.

The reason I complained is that there *are* people out there who believe
that the rating system is designed to produce a specific "average
rating" (and surprisingly many of these people think that the magic
number is 1500).


I get the point. I'm not sure if it's mathematically relevant if you simply
define a know-nothing rating as zero.

It may be 'relevant' - but if so, it's flat out wrong!

In any Elo-based system, the only "correct" rating for a player who
truly knows nothing is: -infinity.

I'm not sure what the system gains by
going into negative numbers because as soon as you score 1/2 a point out of
100 games against players rated ELO 100, you have a positive rating. It may
be closer to zero than to one, but it is positive.

that's not correct. If you score 0.5-99.5 against 0100 opposition, your
proper Elo rating is most definitely NOT positive. Do the math.

But, at least it's no longer -infinity!

Remember - what matters is the rating DIFFERENCE between you and all
those 0100 players who are beating up on you. What rating DIFFERENCE
predicts a score of 99.5-0.5 (look it up!). Your new rating after going
0.5-99.5 against 0100 competition is 0100-thatDifference. Which, I
assure you, is a negative number.

Why, exactly, do you think that your rating would be greater than 0000
after that performance? Because USCF has an Absolute Floor of 0100?
That's an administrative twiddle (mostly meant to pacify innumerate folk
who believe that ratings must be positive).




USCF does not assign negative ratings (even though some players earned
them) - but only because the politicians think that innumerate Americans
can't handle the complication of the '-' sign. Alas, they are probably
right. One need only read this thread to see that innumeracy reigns.

Americans' stupidity re math notwithstanding, there is no practical reason
for a negative rating.

Do you know the real reason that the USCF minimum rating is 0100, and
not 0000?

You wouldn't believe me if I told you - it's too silly for words.

In any event, there is indeed a "practical reason for a negative
rating". That reason is: the negative rating more accuratly predicts
the future performance of that player (e.g., that player will score 0.5
out of 100 against 0100 competition).

USCF has decided that this valid reason is counterbalanced by other
reasons (that's how you run a real system - you end up compromising).
But...you'd really laugh if you heard the *real* reason that the minimum
USCF rating is 0100.

HINT: it's closely related to the reason that the minimum USCF rating
used to be (many years ago) 1000.






--
Kenneth Sloan
Computer and Information Sciences +1-205-932-2213
University of Alabama at Birmingham FAX +1-205-934-5473
Birmingham, AL 35294-1170 http://www.cis.uab.edu/sloan/

David Kane wrote:
"Kenneth Sloan" wrote in message
...
A practical rating system can, and does, make SOME (but not much) use
of this very fuzzy estimate. And, as a result, the ratings generated
are just a little bit better than they would be if the system did NOT
make use of this estimate.

True, the rating system imputes initial ratings
to unrated players. In that case it makes very good sense
to give unrateds a starting rating that jibes with the rating
that that class had historically achieved in their first events.

But that is something entirely different than assigning that
rating to random adults who know only the rules of chess.
99+% of people who know the rules have never played
a single USCF rated game. You cannot hope to reliably
extrapolate the performance of the 99% by looking at the
1%.




You're not a statistician, are you?

--
Kenneth Sloan
Computer and Information Sciences +1-205-932-2213
University of Alabama at Birmingham FAX +1-205-934-5473
Birmingham, AL 35294-1170 http://www.cis.uab.edu/sloan/

"Kenneth Sloan" wrote in message
...
David Kane wrote:
"Kenneth Sloan" wrote in message
...
A practical rating system can, and does, make SOME (but not much) use
of this very fuzzy estimate. And, as a result, the ratings generated
are just a little bit better than they would be if the system did NOT
make use of this estimate.

True, the rating system imputes initial ratings
to unrated players. In that case it makes very good sense
to give unrateds a starting rating that jibes with the rating
that that class had historically achieved in their first events.

But that is something entirely different than assigning that
rating to random adults who know only the rules of chess.
99+% of people who know the rules have never played
a single USCF rated game. You cannot hope to reliably
extrapolate the performance of the 99% by looking at the
1%.




You're not a statistician, are you?



You are guilty of what I have heard described the
lazy scientist syndrome. It's the belief that the data you
have on your desk must answer your question.
It's much easier to process data in hand than
to find the correct data. Yet still invalid.

"Kenneth Sloan" wrote in message
...


In any Elo-based system, the only "correct" rating for a player who truly
knows nothing is: -infinity.

This is only true if "knows nothing" equates to "never wins" which you
haven't proven. In fact, if we rated a tournament filled only
with unrated "know-nothings", they'd end up with the rating arbitrarily
assigned to their class by the rating system. No system that I know
of uses -infinity.

David Kane wrote:
"Kenneth Sloan" wrote in message
...

In any Elo-based system, the only "correct" rating for a player who truly
knows nothing is: -infinity.

This is only true if "knows nothing" equates to "never wins" which you
haven't proven. In fact, if we rated a tournament filled only
with unrated "know-nothings", they'd end up with the rating arbitrarily
assigned to their class by the rating system. No system that I know
of uses -infinity.

uses -infinity as *what*? As the initial rating for a first-time
player? Well, of course not - THAT WAS MY POINT!

As the lowest possible rating? USCF does. In *calculating* ratings,
any value (even negative values) are permitted. Only at the end is
an administrative floor imposed.

At one time, the programming assumed a minimum of 0000 - but did not
really impose it (actually, it assumed a minimum of 0001, because 0000
was reserved to mean "UNR". That is, until it was noticed that ratings
for some players were fast approaching 0000, and it was only a matter of
time until a computed rating was actually negative. That's what sparked
the discussion of the Absolute Floor.

Participants in that discussion may recall that I argued for a Floor
of 0400. The end result was a floor at 0100. The official reason
given: "players are embarassed to have double-digit ratings".

The lowest rating ever published (if memory serves) was 0040. There
are now hundreds of players who are being propped up by the floor at
0100 (properly so, in my opinion - for reasons beyond the scope of this
thread), but the basic Elo computation
takes absolutely NO NOTICE of ZERO, for any purpose whatsoever (don't
believe me? read Elo!). There is nothing at all in the current
USCF rating system that prevents a rating from going negative (and,..as
negative as you please, which is a pretty good definition of -infinity).
The ONLY reason you don't see negative ratings reported is the
politically determined Absolute Floor at 0100.

Exercise for statistics journeymen: look at the distribution of USCF
ratings and estimate the lowest "true" USCF rating as of today.


--
Kenneth Sloan
Computer and Information Sciences +1-205-932-2213
University of Alabama at Birmingham FAX +1-205-934-5473
Birmingham, AL 35294-1170 http://www.cis.uab.edu/sloan/

wrote:

Opinion:

Ratings are efficient only within the pool of players you are rated
against. It makes perfect sense that a player who has played no rating
games goes as unrated. You cannot assign an arbitrary value (0, 500,
100, whatever) to a player who has not played any games against the
rating pool, as this would skew the ratings eventually.

Not if you do it right!

And...in some (rare) cases, it turns out to be essential to make
*some* estimate of the rating to use for an UNR player. Once you
find this necessary, it's no longer a question of IF you can
assign a rating, but instead HOW MUCH CONFIDENCE you have
in that assigned rating.

I hope you'll agree that a rating of 1500, with a variance of 10^10, is
a reasonably accurate rating (for ALL players!). Not very precise - but
reasonably accurate.


I don't pretend to know the math involved, but this seems painfully
obvious to me. This is why you can play 1000 games on 5 different
servers and get 5 different ratings, probably not all that close to
each other. Also, a player can effectively and consistently play
opponents rated below himself and increase (INFLATE) his rating
slowly. You run into that scenario all the time on ICC and Playchess.


In a well designed rating system, you will NOT inflate your rating by
consistently playing low rated players.

In fact, we often hear the opposite complaint - that high rated players
see their ratings DEflate when forced to play low-rated competition.

when you hear both complaints, you are reasonably assured that the world
is nicely balanced.

Interesting thread, however.

http://chess-training.blogspot.com



--
Kenneth Sloan
Computer and Information Sciences +1-205-932-2213
University of Alabama at Birmingham FAX +1-205-934-5473
Birmingham, AL 35294-1170 http://www.cis.uab.edu/sloan/

Martin Brown wrote:


One way of determining the original rating of a complete beginner is to
accept that they will improve with each game played and then model it.
Afterwards you can compute the intercept of their rating vs number of
games played at the 0 games played axis to get a fair estimate.

My instinct is that you would work hard to find someone with an ELO
rating below 500 (who wasn't deliberately playing to lose).

Regards,
Martin Brown


If you are restricting the population to "adults of normal
intelligence", then I would agree. However, if you include
all players, regardless of age or intelligence, I can find at least
10 players with deserved ratings 0000.

and hundreds upon hundreds of players with deserved ratings 0500

The question at the moment is: what does your instinct tell you about
the AVERAGE beginner who knows the rules.


[beware...that might be a trick question...even World Championship
Candidates have demonstrated that they don't know all the rules.
Do *you* know *all* the rules about 0-0-0? Korchnoi didn't!]

--
Kenneth Sloan
Computer and Information Sciences +1-205-932-2213
University of Alabama at Birmingham FAX +1-205-934-5473
Birmingham, AL 35294-1170 http://www.cis.uab.edu/sloan/

Ange1o DePa1ma wrote:


It is so obvious that a rating of 1000 is not a rank beginner. I'm
incredulous that some of you insist it is. Is this some sort of "big lie"
post or what?




That's right. Everyone who disagrees with you is a liar.

You win.

--
Kenneth Sloan
Computer and Information Sciences +1-205-932-2213
University of Alabama at Birmingham FAX +1-205-934-5473
Birmingham, AL 35294-1170 http://www.cis.uab.edu/sloan/

"Kenneth Sloan" wrote

USCF has decided that this valid reason is counterbalanced by other
reasons (that's how you run a real system - you end up compromising).
But...you'd really laugh if you heard the *real* reason that the minimum
USCF rating is 0100.

HINT: it's closely related to the reason that the minimum USCF rating used
to be (many years ago) 1000.

Because there are 100 years in a century?

"Kenneth Sloan" wrote

Ange1o DePa1ma wrote:

It is so obvious that a rating of 1000 is not a rank beginner. I'm
incredulous that some of you insist it is. Is this some sort of "big lie"
post or what?

That's right. Everyone who disagrees with you is a liar.

You win.

I did not mean that you were a liar -- how could your original statement be
a lie when you presented it as an opinion?

What I meant was that it's sometimes easier to convince people of something
outrageous than something that's only slightly off.

If you meant that the rating of one who just knew how pieces moved rules
*could* or *might* be 1000 given other adjustments to the rating scale (e.g.
Kasparov @ 15,000) then I misunderstood you. If you meant under the existing
system, with Kasparov at 2800, then you are clearly wrong.

Note, you did not take the "DePalma Challenge." All those individuals were
rated 1000, and all were "experienced" to some degree.

Angelo