There have been several arguments on here about determining the rating
of an absolute beginner at chess, who knows only the moves.

The thread 'Improve your chess in 7 days or less' got its name changed
to 'Rip-off of Reinfeld', then into 'beginner rating' etc.

Rather than change the name again, I thought I'd start a *new* thread.

Values of 0, 500 and 1000 have all been mentioned.

See this thread for example.

http://groups.google.co.uk/group/rec...779ed9a29f868b

Just to be a devils advocate, here is a PROOF, based on arguments like
Heisenberg's uncertainly principal, that the rating of a complete
beginner, who only knows the moves, can NEVER be determined.

1) In order to determine with any useful statistical significance the
ELO rating of a player, you need to establish that with multiple games.
For example, on ICC one gets a provisional rating until one has played
20 games.

2) After playing the number of games necessary to establish the rating
with useful statistical significance, the player is no longer a complete
beginner.

Hence such a rating can NEVER be determined.

Comments?
--
Dave (from the UK)

Please note my email address changes periodically to avoid spam.
It is always of the form:
Hitting reply will work for a few months only - later set it manually.

http://witm.sourceforge.net/ (Web based Mathematica front end)

Dave (from the UK) wrote:
There have been several arguments on here about determining the rating
of an absolute beginner at chess, who knows only the moves.

The thread 'Improve your chess in 7 days or less' got its name changed
to 'Rip-off of Reinfeld', then into 'beginner rating' etc.

Rather than change the name again, I thought I'd start a *new* thread.

Values of 0, 500 and 1000 have all been mentioned.

See this thread for example.

http://groups.google.co.uk/group/rec...779ed9a29f868b

Just to be a devils advocate, here is a PROOF, based on arguments like
Heisenberg's uncertainly principal, that the rating of a complete
beginner, who only knows the moves, can NEVER be determined.

1) In order to determine with any useful statistical significance the
ELO rating of a player, you need to establish that with multiple games.
For example, on ICC one gets a provisional rating until one has played
20 games.

2) After playing the number of games necessary to establish the rating
with useful statistical significance, the player is no longer a complete
beginner.

Hence such a rating can NEVER be determined.

Comments?
--
Dave (from the UK)

Please note my email address changes periodically to avoid spam.
It is always of the form:
Hitting reply will work for a few months only - later set it manually.

http://witm.sourceforge.net/ (Web based Mathematica front end)

It's all well and nice to talk about Heisenberg's uncertainty
principle, but this is the real world. The ELO formula requires that
the player must have SOME number applied to it, although it doesn't
matter WHAT that number is. And since we're talking about "ratings",
we're pretty much talking about ELO (although other formulae exist,
they are much less accepted).

Just because there's a magic rating of 2000 for Masters and 2500 for
GMs, they could easily have been any other numbers. But a player's
rating can never be "undefined", unless the ELO formula is replaced by
something else that can handle such a starting point.

jm

wrote:

Just to be a devils advocate, here is a PROOF, based on arguments like
Heisenberg's uncertainly principal, that the rating of a complete
beginner, who only knows the moves, can NEVER be determined.
http://witm.sourceforge.net/ (Web based Mathematica front end)


It's all well and nice to talk about Heisenberg's uncertainty
principle, but this is the real world.

I would say in the "real world" in the context you mean, the
Heisenberg's Uncertainty Principle has little practical relevance. The
effects of it are too small to worry about.

If someone has played 1000 games of chess, the process of measuring
their performance by getting them to play 20 games against opponents of
a known rating, is probably not going to have much effect on their
performance.

I don't have a copy of Professor Elo's paper on the subject, but to
quote from Wikipedia:

http://en.wikipedia.org/wiki/ELO_rating_system

"Élő's central assumption was that the chess performance of each player
in each game is a normally distributed random variable. Although a
player might perform significantly better or worse from one game to the
next, Élő assumed that the mean value of the performances of any given
player changes only slowly over time."

It is *not* true to say the mean performance of a player changes slowly
over time if they have just leaned the moves. Having played 10 games, I
suspect they are significantly better having played only 1 game. As
such, the process of measurement will have a *very* significant effect
on the quantity you are trying to measure.

So unlike Heisenberg's Uncertainty Principle, the effect would be very
pronounced in the real world.

The ELO formula requires that
the player must have SOME number applied to it, although it doesn't
matter WHAT that number is. And since we're talking about "ratings",
we're pretty much talking about ELO (although other formulae exist,
they are much less accepted).

I believe the assumptions Elo made are simply not valid in the case of
an absolute beginner. As such, attaching an ELO number to something
where the assumptions are very wrong is not sensible.

Just because there's a magic rating of 2000 for Masters and 2500 for
GMs, they could easily have been any other numbers.

Agreed.

But a player's
rating can never be "undefined", unless the ELO formula is replaced by
something else that can handle such a starting point.

I can't see how you can measure it. As such, I can't see how it can
possibly be defined unless the formula is modified to say "By
definition, a beginner has an ELO of 100" or similar.

Since the method has no definition and it can't be measured, I doubt
there is much point attaching a value to it.


--
Dave (from the UK)

Please note my email address changes periodically to avoid spam.
It is always of the form:
Hitting reply will work for a few months only - later set it manually.

http://witm.sourceforge.net/ (Web based Mathematica front end)

"Dave (from the UK)" wrote in
message ...


I believe the assumptions Elo made are simply not valid in the case of an
absolute beginner. As such, attaching an ELO number to something where the
assumptions are very wrong is not sensible.


Of course they aren't. But what would be your
response if I were to claim in Ken Sloan
fashion that an adult of average intelligence
who just knows the rules has a rating of 2700?

The fact that there is uncertainty to ratings doesn't
make the discussion meaningless.

By the way, I think in the USCF system the rating
formulae change at about 8 games. The USCF rating
algorithm assigns beginners a very high K-constant
(80, I think, which can be effectively doubled via
"bonus" points) So the number of games actually
going into a rating is much smaller than the number
which defines a "provisional" rating. Someone can
correct me, but I don't think provisional ratings are
treated differently in the rating algorithm.

David Kane wrote:
"Dave (from the UK)" wrote in
message ...


I believe the assumptions Elo made are simply not valid in the case of an
absolute beginner. As such, attaching an ELO number to something where the
assumptions are very wrong is not sensible.



Of course they aren't.

So we agree!

But what would be your
response if I were to claim in Ken Sloan
fashion that an adult of average intelligence
who just knows the rules has a rating of 2700?

A large number of players who play the game seriously don't have a
rating of 2700, so it's fairly easy to disprove that one. Putting an
upper bound is fairly easy. Anyone who thinks it would be over 1200 is
really got to be mad.

The fact that there is uncertainty to ratings doesn't
make the discussion meaningless.


By the way, I think in the USCF system the rating
formulae change at about 8 games. The USCF rating
algorithm assigns beginners a very high K-constant
(80, I think, which can be effectively doubled via
"bonus" points)

I don't know how the USCF method works, but I assume this K-constant you
talk of is some measure of the standard deviation (uncertainty) of the
rating. For a new member, this will be higher until they are established.

So the number of games actually
going into a rating is much smaller than the number
which defines a "provisional" rating. Someone can
correct me, but I don't think provisional ratings are
treated differently in the rating algorithm.

Anyone who is a member of the USCF would have to be reasonably serious
about the game, so they would have played many times before. I don't
believe someone will learn the rules, then join USCF and play their
first ever game in a rated competition.

As such, it should be possible to determine a rating reasonably quickly,
as they performance is unlikely to change a huge amount from game to game.

In contrast, as absolute beginner who only just learned the rules,
playing his first few games, should learn a lot very quickly. They
should soon learn about the fact knights can fork. So its very likely
the large K-constant you talk of in USCF rating systems is not
sufficiently large. So I don't believe you can rely on the USCF system
in this case. In any case, the USCF system would have been designed for
serious chess players - not those who just learned the rules.

One *possible* way to establish a rating for absolute beginners might be
as follows.

Take 1000 beginners and let them play their *first* ever chess game
against a low rated player (say 1200 for example). Results of any of
their later games are ignored. Each person plays just one game.

Statistically, if there is a rating difference of x between two players,
the probability of wins and losses can be computed. The 1200 players
should beat the beginners in most games, but the percentage might allow
you to determine how much stronger the 1200 players and and so assign an
average rating for those 1000 people after just one game. That would be
VERY doggy, and not very practical to do, but at least it would give you
a result from only one game, so they don't have any chance to improve.

I can't help feeling the concept of a rating for someone who has just
learned the rules is absurd. You can put a lower limit of 0, an upper
limit of 1200, but I don't believe you can say anything else really.

--
Dave (from the UK)

Please note my email address changes periodically to avoid spam.
It is always of the form:
Hitting reply will work for a few months only - later set it manually.

http://witm.sourceforge.net/ (Web based Mathematica front end)

"Dave (from the UK)" wrote in
message ...
David Kane wrote:
"Dave (from the UK)" wrote
in message ...


I believe the assumptions Elo made are simply not valid in the case of an
absolute beginner. As such, attaching an ELO number to something where the
assumptions are very wrong is not sensible.



Of course they aren't.

So we agree!

But what would be your
response if I were to claim in Ken Sloan
fashion that an adult of average intelligence
who just knows the rules has a rating of 2700?

A large number of players who play the game seriously don't have a rating of
2700, so it's fairly easy to disprove that one. Putting an upper bound is
fairly easy. Anyone who thinks it would be over 1200 is really got to be mad.

Similarly, those with experience with players in the
x00's range, also believe that 1000 is mad.

The fact that there is uncertainty to ratings doesn't
make the discussion meaningless.


By the way, I think in the USCF system the rating
formulae change at about 8 games. The USCF rating
algorithm assigns beginners a very high K-constant
(80, I think, which can be effectively doubled via
"bonus" points)

I don't know how the USCF method works, but I assume this K-constant you talk
of is some measure of the standard deviation (uncertainty) of the rating. For
a new member, this will be higher until they are established.

So the number of games actually
going into a rating is much smaller than the number
which defines a "provisional" rating. Someone can
correct me, but I don't think provisional ratings are
treated differently in the rating algorithm.

Anyone who is a member of the USCF would have to be reasonably serious about
the game, so they would have played many times before. I don't believe someone
will learn the rules, then join USCF and play their first ever game in a rated
competition.

Not true. For better of for worse, many kids do start playing rated
games knowing very little.

As such, it should be possible to determine a rating reasonably quickly, as
they performance is unlikely to change a huge amount from game to game.


If your point is that ratings mean more for some than others, then
certainly that is true.

In contrast, as absolute beginner who only just learned the rules, playing his
first few games, should learn a lot very quickly. They should soon learn about
the fact knights can fork. So its very likely the large K-constant you talk of
in USCF rating systems is not sufficiently large. So I don't believe you can
rely on the USCF system in this case. In any case, the USCF system would have
been designed for serious chess players - not those who just learned the
rules.

The value of the K-constant is how much a rating can change from
a single game. I have no idea whether the USCF method
has been proven optimal for all ratings but it covers all ratings.
I only brought it up because the number of games that a
rating is "provisional" doesn't necessarily have anything to
do with the number of games for a rating to equilibrate
in the rating formulae (~800/K).

One *possible* way to establish a rating for absolute beginners might be as
follows.

Take 1000 beginners and let them play their *first* ever chess game against a
low rated player (say 1200 for example). Results of any of their later games
are ignored. Each person plays just one game.

Statistically, if there is a rating difference of x between two players, the
probability of wins and losses can be computed. The 1200 players should beat
the beginners in most games, but the percentage might allow you to determine
how much stronger the 1200 players and and so assign an average rating for
those 1000 people after just one game. That would be VERY doggy, and not very
practical to do, but at least it would give you a result from only one game,
so they don't have any chance to improve.

I can't help feeling the concept of a rating for someone who has just learned
the rules is absurd. You can put a lower limit of 0, an upper limit of 1200,
but I don't believe you can say anything else really.


As mentioned elsewhere, 0 has no special significance in the rating system.
In fact the minimum rating in the USCF system is 100. 1200 (or 1000, or 800)
can be refuted in the same way that 2700 can be refuted - by showing
that players with those ratings can beat average intelligence
adults. (Whether the adults have strictly just learned the rules is
just a detail, unless you are suggesting that player's ratings would
*decline* by playing)

Such complicated thinking on such a simple topic.

A rating is nothing more or less than a prediction about future
performance. It's a single position on a scale.

One way to assign a rating is to observe the player's results.

When you can't do that, there are still useful facts about the
player that can improve the quality of your estimate. For example,
USCF uses the player's AGE to provide an initial rating. One
could conceivably use IQ scores, economic class, gender, eye color...
anything that might be correlated with rating. Of course, some
of these data might be more relevant than others.

Many posters in this thread have demonstrated this by using
the information that the player is playing in a USCF rated event - and
then saying "only players of a certain minimum strength ever play
in USCF rated events, so...". This is a perfect example that makes
my point.

Even if you have ZERO information about the player, it is possible
to come up with a "best estimate" of his rating. Of course, the
variance associated with this estimate will be very high, but the
variance is not infinite.

Someone here demonstrated this by saying: "surely it would be stupid
to claim that a new player is rated 2700".

OK, so...2700 is too high. How about 2000? Still too high...

How about -10000 (yes, that would be a perfectly valid rating). That
seems a bit low, no? (actually, it's too high for my dog - but might
not be too high for a chimpanzee).

So, we now have it bracketed to be somewhere in (-100000, 2000).
Narrowing that range will depend on the data you have at hand.

I still think my estimate is not too far off. Find yourself a random
adult of normal intelligence, teach him the moves and enter him in
5 USCF events of 5 rounds each. I claim that a good estimate of his
performance in those events is that the performance will be that
of a USCF 1000 rated player.

Of course, some individuals will do worse than that - but I'm moderately
confident that many individuals will perform much better than that.
And...some players will continue to perform worse than that, even after
they learn the complete history of chess, memorize 5 useless gimmick
openings, etc - learning much more than the moves, but not being able
to translate that into performance.

If your personal estimate is 0800, I won't argue too much. But, if your
personal estimate is 0100, I think it's clear that that's much, much too
low. By the same token, 1500 is clearly much, much too high.

And, of course, there is NO REASON to select 0000 - certainly not
because of any notion that 0000 means "no rating points".

So...call it (0800, 1200). That's "1000, with a HUGE variance".

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.

That's all there is to it!

--
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:

A large number of players who play the game seriously don't have a rating of
2700, so it's fairly easy to disprove that one. Putting an upper bound is
fairly easy. Anyone who thinks it would be over 1200 is really got to be mad.


Similarly, those with experience with players in the
x00's range, also believe that 1000 is mad.

I don't doubt that. If I had said 1000, there might have been some
argument about it, so 1200 seemed a sensible upper limit.

Anyone who is a member of the USCF would have to be reasonably serious about
the game, so they would have played many times before. I don't believe someone
will learn the rules, then join USCF and play their first ever game in a rated
competition.


Not true. For better of for worse, many kids do start playing rated
games knowing very little.

Having just learned the rules? Which is what I thought we were talking
about.

I could see that someone might learn the rules of chess from a book or
the Internet and not having a human opponent, go onto ICC, FICS or
whatever and play their first game. I doubt they would join USCF and
play their first game there.

I guess there is always the parent who wants his/her child to play
chess, so they pay for membership and get them doing it from the
beginning. I used to teach maths to someone who did not really want to
learn mathematics, but his Dad wanted him to. (I'm not a mathematician
BTW).

As mentioned elsewhere, 0 has no special significance in the rating system.

Sorry, I thought it did, but I see you are right.

In fact the minimum rating in the USCF system is 100. 1200 (or 1000, or 800)
can be refuted in the same way that 2700 can be refuted - by showing
that players with those ratings can beat average intelligence
adults. (Whether the adults have strictly just learned the rules is
just a detail, unless you are suggesting that player's ratings would
*decline* by playing)

I was talking about someone who had JUST LEARNED THE RULES, which is
what I believed the other threads were referring to. In that
circumstance, by playing several games an average (or even somewhat
below average) intelligence human can't fail to learn and so improve.


--
Dave (from the UK)

Please note my email address changes periodically to avoid spam.
It is always of the form:
Hitting reply will work for a few months only - later set it manually.

http://witm.sourceforge.net/ (Web based Mathematica front end)

The "Heisenberg" theory is cute, but it only goes as far as the "measurement
affects outcome" idea.

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.

But in the real world that is an unnecessary complication for a very simple
idea. We normally think of achievement in terms of numbers from zero to some
higher number, for example:

* I own six chess trophies (when I was 11 I had none)
* Federer's world tennis association ranking is based on some positive
number of points earned in tournaments (at some point he had no points, and
no ranking)
* I have several hundred ACBL master points (at one time I had zero)

What is the sense of using negative numbers here? I never had minus-three
chess trophies; I never won minus-two tournaments. Why should my rating be
negative?

"Dave (from the UK)"
wrote in message ...
David Kane wrote:

A large number of players who play the game seriously don't have a rating
of
2700, so it's fairly easy to disprove that one. Putting an upper bound is
fairly easy. Anyone who thinks it would be over 1200 is really got to be
mad.


Similarly, those with experience with players in the
x00's range, also believe that 1000 is mad.

I don't doubt that. If I had said 1000, there might have been some
argument about it, so 1200 seemed a sensible upper limit.

Anyone who is a member of the USCF would have to be reasonably serious
about
the game, so they would have played many times before. I don't believe
someone
will learn the rules, then join USCF and play their first ever game in a
rated
competition.


Not true. For better of for worse, many kids do start playing rated
games knowing very little.

Having just learned the rules? Which is what I thought we were talking
about.

I could see that someone might learn the rules of chess from a book or the
Internet and not having a human opponent, go onto ICC, FICS or whatever
and play their first game. I doubt they would join USCF and play their
first game there.

I guess there is always the parent who wants his/her child to play chess,
so they pay for membership and get them doing it from the beginning. I
used to teach maths to someone who did not really want to learn
mathematics, but his Dad wanted him to. (I'm not a mathematician BTW).

As mentioned elsewhere, 0 has no special significance in the rating
system.

Sorry, I thought it did, but I see you are right.

In fact the minimum rating in the USCF system is 100. 1200 (or 1000, or
800)
can be refuted in the same way that 2700 can be refuted - by showing
that players with those ratings can beat average intelligence
adults. (Whether the adults have strictly just learned the rules is
just a detail, unless you are suggesting that player's ratings would
*decline* by playing)

I was talking about someone who had JUST LEARNED THE RULES, which is what
I believed the other threads were referring to. In that circumstance, by
playing several games an average (or even somewhat below average)
intelligence human can't fail to learn and so improve.


--
Dave (from the UK)

Please note my email address changes periodically to avoid spam.
It is always of the form:
Hitting reply will work for a few months only - later set it manually.

http://witm.sourceforge.net/ (Web based Mathematica front end)

Kenneth Sloan wrote:
Such complicated thinking on such a simple topic.

Maybe.

A rating is nothing more or less than a prediction about future
performance. It's a single position on a scale.

Agreed

One way to assign a rating is to observe the player's results.

Yes, obviously.


I still think my estimate is not too far off. Find yourself a random
adult of normal intelligence, teach him the moves and enter him in
5 USCF events of 5 rounds each. I claim that a good estimate of his
performance in those events is that the performance will be that
of a USCF 1000 rated player.

That is after 25 games !!! I thought other threads we were talking about
someone who had JUST learned the rules. Hence they would not have the
experience of playing 25 games.

I taught my wifes grandaugther the rules and played a few games with
her. It was obvious she was better significantly better after the third
game than the first.

What was my point that simply by making the measurement of a rating, you
are affecting the rating.


--
Dave (from the UK)

Please note my email address changes periodically to avoid spam.
It is always of the form:
Hitting reply will work for a few months only - later set it manually.

http://witm.sourceforge.net/ (Web based Mathematica front end)