Reply
 
LinkBack Thread Tools Display Modes
  #1   Report Post  
Old May 25th 04, 02:45 AM
Squark
 
Posts: n/a
Default UPHILL RATING?

What is your "UPHILL RATING"? This is what your rating would be, taking
into account only those games where your opponent was at least 150 points
stronger than you. How do you play against strong opposition? Do you get
intimidated and fall apart? Do you summon every resource and play an extra
sharp game? This is a dimension of your game which can be quantified, and,
in fact, is one of many analytics published by Chess Express Ratings as part
of the CXR System. Take a look for yourself: www.chess-express.com


  #2   Report Post  
Old May 25th 04, 03:32 AM
Kenneth Sloan
 
Posts: n/a
Default UPHILL RATING?

"Squark" writes:

What is your "UPHILL RATING"? This is what your rating would be, taking
into account only those games where your opponent was at least 150 points
stronger than you. How do you play against strong opposition? Do you get
intimidated and fall apart? Do you summon every resource and play an extra
sharp game? This is a dimension of your game which can be quantified, and,
in fact, is one of many analytics published by Chess Express Ratings as part
of the CXR System. Take a look for yourself: www.chess-express.com



This is a great shill.

Theory says that your "uphill rating" should be considerably above your
actual rating.

I predict that this term will become very popular with the players - at
least those who are not statisticians.


--
Kenneth Sloan
Computer and Information Sciences (205) 934-2213
University of Alabama at Birmingham FAX (205) 934-5473
Birmingham, AL 35294-1170
http://www.cis.uab.edu/sloan/
  #3   Report Post  
Old May 25th 04, 04:42 PM
Bruce Leverett
 
Posts: n/a
Default UPHILL RATING?

Kenneth Sloan wrote in message ...
"Squark" writes:

What is your "UPHILL RATING"? This is what your rating would be, taking
into account only those games where your opponent was at least 150 points
stronger than you. How do you play against strong opposition? Do you get
intimidated and fall apart? Do you summon every resource and play an extra
sharp game? This is a dimension of your game which can be quantified, and,
in fact, is one of many analytics published by Chess Express Ratings as part
of the CXR System. Take a look for yourself: www.chess-express.com



This is a great shill.

Theory says that your "uphill rating" should be considerably above your
actual rating.


I was surprised to read this. Of course, my knowledge of the rating
system is completely out of date. But I was under the impression that
I would get about the same rating by playing stronger players as by
playing weaker. I assume that there might be "boundary" effects, due
to rounding or whatever, if my opposition were 400 points stronger or
weaker, but I would think this wouldn't be too important at 150
points. Also there might be some psychological effects ("he beats a
master one round, then loses to a B player the next round"), but that
these likewise would not be too important overall.

I predict that this term will become very popular with the players - at
least those who are not statisticians.

  #4   Report Post  
Old May 25th 04, 11:16 PM
Kenneth Sloan
 
Posts: n/a
Default UPHILL RATING?

(Bruce Leverett) writes:

Kenneth Sloan wrote in message ...
...
Theory says that your "uphill rating" should be considerably above your
actual rating.


I was surprised to read this. Of course, my knowledge of the rating
system is completely out of date. But I was under the impression that
I would get about the same rating by playing stronger players as by
playing weaker. I assume that there might be "boundary" effects, due
to rounding or whatever, if my opposition were 400 points stronger or
weaker, but I would think this wouldn't be too important at 150
points. Also there might be some psychological effects ("he beats a
master one round, then loses to a B player the next round"), but that
these likewise would not be too important overall.


You are assuming that ratings are known perfectly.

First - let's assume that ratings are completely random. What is the
probability that a higher rated player will lose to a lower rated
player?

Answer: 50%.

Now - assume that ratings are known perfectly. Then, the scoring
probability of the higher rated player is known.

Finally - assume that the ratings are some mix of the perfect value with
a random value. Then, the observed results will be a mix of the
predicted result and the 50% results.

Therefore, the observed results will be better for the lower rated
player.

Hence...if you consider only your results against higher rated players,
you will appear to be doing better than expected (from your rating).
The rating that you compute using only these results will be *higher*
than the rating you compute using all of your games.


--
Kenneth Sloan

Computer and Information Sciences (205) 934-2213
University of Alabama at Birmingham FAX (205) 934-5473
Birmingham, AL 35294-1170
http://www.cis.uab.edu/sloan/
  #5   Report Post  
Old May 27th 04, 07:57 AM
James B. Shearer
 
Posts: n/a
Default UPHILL RATING?

Kenneth Sloan wrote in message ...
(Bruce Leverett) writes:

Kenneth Sloan wrote in message ...
...
Theory says that your "uphill rating" should be considerably above your
actual rating.


I was surprised to read this. Of course, my knowledge of the rating
system is completely out of date. But I was under the impression that
I would get about the same rating by playing stronger players as by
playing weaker. I assume that there might be "boundary" effects, due
to rounding or whatever, if my opposition were 400 points stronger or
weaker, but I would think this wouldn't be too important at 150
points. Also there might be some psychological effects ("he beats a
master one round, then loses to a B player the next round"), but that
these likewise would not be too important overall.


You are assuming that ratings are known perfectly.

First - let's assume that ratings are completely random. What is the
probability that a higher rated player will lose to a lower rated
player?

Answer: 50%.

Now - assume that ratings are known perfectly. Then, the scoring
probability of the higher rated player is known.

Finally - assume that the ratings are some mix of the perfect value with
a random value. Then, the observed results will be a mix of the
predicted result and the 50% results.

Therefore, the observed results will be better for the lower rated
player.

Hence...if you consider only your results against higher rated players,
you will appear to be doing better than expected (from your rating).
The rating that you compute using only these results will be *higher*
than the rating you compute using all of your games.


I think this is correct but that the explanation is not very
clear. Suppose we choose some cutoff like 2000 and look at the games
you play against opponents rated 2000 or higher. Since the rating
system is imperfect some of these opponents will actually be of less
than 2000 strength. So you will score a little better than expected.
James B. Shearer


  #6   Report Post  
Old May 27th 04, 03:59 PM
Bill Smythe
 
Posts: n/a
Default UPHILL RATING?

"James B. Shearer" wrote:
I think this is correct but that the explanation is not very
clear. Suppose we choose some cutoff like 2000 and look at the games
you play against opponents rated 2000 or higher. Since the rating
system is imperfect some of these opponents will actually be of less
than 2000 strength. So you will score a little better than expected.


To carry this a little further, some of the opponents will actually be of
more than 2000 strength, too. But these won't have as much of an effect,
because, if you are rated 1700 for example, the difference in your winning
probabilities against opponents rated 1900 vs 2000 is greater than the
difference in your winning probabilities against opponents rated 2000 vs
2100.

Are we getting close yet?

Bill Smythe



  #8   Report Post  
Old May 28th 04, 03:00 AM
Bruce Leverett
 
Posts: n/a
Default UPHILL RATING?

"Bill Smythe" wrote in message ...
"James B. Shearer" wrote:
I think this is correct but that the explanation is not very
clear. Suppose we choose some cutoff like 2000 and look at the games
you play against opponents rated 2000 or higher. Since the rating
system is imperfect some of these opponents will actually be of less
than 2000 strength. So you will score a little better than expected.


To carry this a little further, some of the opponents will actually be of
more than 2000 strength, too. But these won't have as much of an effect,
because, if you are rated 1700 for example, the difference in your winning
probabilities against opponents rated 1900 vs 2000 is greater than the
difference in your winning probabilities against opponents rated 2000 vs
2100.

Are we getting close yet?


It depends on the audience. I'm the one who originally asked the
question, and I was perfectly OK with Ken Sloan's answer. I suppose,
however, that before this kind of answer feels intuitive to you, your
intuition has to be twisted, like mine, and presumably Ken's.

I find it intriguing that uphill ratings and downhill ratings are
likely to be different. Ken, would you care to estimate the magnitude
of this effect?
  #10   Report Post  
Old May 28th 04, 10:07 AM
James B. Shearer
 
Posts: n/a
Default UPHILL RATING?

Kenneth Sloan wrote in message ...
(James B. Shearer) writes:

...
I think this is correct but that the explanation is not very
clear. Suppose we choose some cutoff like 2000 and look at the games
you play against opponents rated 2000 or higher. Since the rating
system is imperfect some of these opponents will actually be of less
than 2000 strength. So you will score a little better than expected.
James B. Shearer


And some of them will actually be of greater than 2000 strength. So you
will score a little worse than expected.

Hmmm. I like my explanation better.


My claim was the pool of people rated over 2000 at any given
point in time will be weaker than the pool of people with true
strength over 2000 as ratings error will include some weaker players
with true strength under 2000 and exclude some stronger players with
true strength over 2000. Are you disputing this?
However let me try again. Suppose you have pool of players
whose true strength never changes. Let them play a large number of
games and assume the outcome of each game is independently random
depending only on the difference in true strength of the opponents in
accordance with the rating model. Consider a player X. Over time his
rating will randomly fluctuate. Some of the time X will be rated
above 2000 (or any other arbitrary cutoff), some of the time below.
Of course if X's true strength is well under 2000 he will spend little
time rated above 2000 and if X's true strength is well over 2000 he
will spend little time rated below 2000. Now consider your games
against X. Since X's true strength is not changing you will score as
well when X is rated above 2000 as your overall score against X. But
clearly your performance rating will be higher considering just those
games in which X is rated above 2000. This is true for every other
player in the pool. So your performance rating in games against
opponents rated over 2000 will exceed your performance rating in all
games as you are selecting games in which your opponents are over
rated on average.
James B. Shearer
Reply
Thread Tools
Display Modes

Posting Rules

Smilies are On
[IMG] code is Off
HTML code is Off
Trackbacks are On
Pingbacks are On
Refbacks are On


Similar Threads
Thread Thread Starter Forum Replies Last Post
A modest proposal for the electronic rating supplements Bill Smythe rec.games.chess.politics (Chess Politics) 5 May 24th 04 04:31 PM
Prominent TD reports major cheating incident at Foxwoods Tim Hanke rec.games.chess.politics (Chess Politics) 128 May 17th 04 08:09 PM
Annual USCF rating distribution report Dan Heisman rec.games.chess.politics (Chess Politics) 41 December 12th 03 11:32 AM
USCF rating floors... Howard Goldowsky rec.games.chess.analysis (Chess Analysis) 2 August 19th 03 04:49 PM
USCF rating floors... Howard Goldowsky rec.games.chess.politics (Chess Politics) 0 August 19th 03 01:32 PM


All times are GMT +1. The time now is 11:35 PM.

Powered by vBulletin® Copyright ©2000 - 2019, Jelsoft Enterprises Ltd.
Copyright 2004-2019 ChessBanter.
The comments are property of their posters.
 

About Us

"It's about Chess"

 

Copyright © 2017