How Some Friends and I Feel About Ratings Floors
Cavemanchs
commonly known as "CAVEMAN." Most of my friends and I strongly dislike
ratings floors. We feel they are artificial. They discourage several of
my friends from playing, because they have bottomed out and cannot reach a
competitive level. They say that they have no motivation, because they
know their rating can't go down.
We much preferred bonus points as a way of combatting deflation. Bonus
points reward good results. Rating floors reward bad results. No one
wants to be rewarded for poor chess.
This statement is an oversimplification, but it makes the point. I find
this opinion to be prevalent among chessplayers, yet the ratings
committee/USCF seems to think otherwise. I don't know why.
Most of us with any math background who have read ELO feel that the USCF
has severely bastardized an otherwise pretty good system.
Kenneth Sloan
>I am a USCF master, Kevin Bachler, currently rated about 2220.
2220*X7, to be precise.
>... I am
>commonly known as "CAVEMAN." Most of my friends and I strongly dislike
>ratings floors.
A reasonable opinion.
>... We feel they are artificial.
Yup
>... They discourage several of
>my friends from playing, because they have bottomed out and cannot reach a
>competitive level. They say that they have no motivation, because they
>know their rating can't go down.
No sympathy here. According to my reading, this translates to: "we'd
like to sandbag, but can't". Let's be realistic here - if you are
sitting on a floor, then your rating is at least 101 points below your
(recent) peak. Are you seriously claiming that a player who has peaked
at 2301+ really wants to play in the U2200 section? Much as I
personally dislike the 100-point floors, I can't summon any sympathy for
such a losing attitude.
And - *you* certainly have not "bottomed out". By my estimation, it's
virtually certain that *your* floor is 2100. Can't your "friends" speak
for themselves? Do they exist? Names, please.
>
>We much preferred bonus points as a way of combatting deflation. Bonus
>points reward good results. Rating floors reward bad results. No one
>wants to be rewarded for poor chess.
I maintain that no one is "rewarded" by the 100-point floors. After
all, you still need a 2200+ performance to get your rating off a 2200
floor. Much as I dislike using words like "reward" or "punishment" when
talking about ratings, the worst that can be said (in these terms) about
100-point floors is that they fail to punish poor chess. That's not
quite the same thing as *rewarding* poor chess.
>
>This statement is an oversimplification,
I'll agree with that.
>... but it makes the point.
nope. Try again - what precisely *is* your point?
>... I find
>this opinion to be prevalent among chessplayers,
By way of contrast, I find that most chessplayers don't really care,
because the vast majority of them are well above their floors.
>... yet the ratings
>committee/USCF seems to think otherwise.
I can state with complete confidence that the vast majority of the USCF
Ratings Committee thinks that the 100-point floors ought to be
abolished. Where did you get the idea that anyone on the USCF Ratings
committee supports these floors? Have you read the current
recommendations from that committee, which call for their abolishment?
May I suggest that if you really want the 100-point floors abolished
that you refrain from badmouthing those who agree with you? Quite
frankly, part of the reason that we are having so much trouble
abolishing the floors is that we spend so much time distancing ourselves
from comments such as yours.
>... I don't know why.
>
>Most of us with any math background who have read ELO feel that the USCF
>has severely bastardized an otherwise pretty good system.
a) what's your "math background"
b) please be more concrete. Besides the 100-point floors, what are your
specific examples?
--
Kenneth Sloan Computer and Information Sciences
sl...@cis.uab.edu University of Alabama at Birmingham
(205) 934-2213 115A Campbell Hall, UAB Station
(205) 934-5473 FAX Birmingham, AL 35294-1170
Paul A. Lane
>I am a USCF master, Kevin Bachler, currently rated about 2220. I am
>commonly known as "CAVEMAN." Most of my friends and I strongly dislike
>ratings floors. We feel they are artificial. They discourage several of
>my friends from playing, because they have bottomed out and cannot reach a
>competitive level. They say that they have no motivation, because they
>know their rating can't go down.
This seems a bit odd. Your friends would have had to reach ratings
roughly 150 points above the floor. If they can't play competitive
chess at that level, how'd they get so high?
>We much preferred bonus points as a way of combatting deflation. Bonus
>points reward good results. Rating floors reward bad results. No one
>wants to be rewarded for poor chess.
I think the lifetime norm system could be worked into some kind of safe-
guard. Say, anyone at a0 or higher could not compete in an U1700
section. Of course, this would probably also knock out your friends,
as I would suspect they've already picked up advanced norms.
>Most of us with any math background who have read ELO feel that the USCF
>has severely bastardized an otherwise pretty good system.
They are an attempt to solve one problem, but you're right than they
encourage inflation. I would note that a friend of mine has a case
which is a decent example. He had a spectacular run to bring his rating
over 1800. Upon entering grad. school, he couldn't spend much time
studying chess anymore and slid back to 1700. He's bounced between
1700 and 1750 for several years.
Paul
--
Paul Powell
: No sympathy here. According to my reading, this translates to: "we'd
: like to sandbag, but can't". Let's be realistic here - if you are
: sitting on a floor, then your rating is at least 101 points below your
: (recent) peak. Are you seriously claiming that a player who has peaked
: at 2301+ really wants to play in the U2200 section? Much as I
: personally dislike the 100-point floors, I can't summon any sympathy for
: such a losing attitude.
Very true... but why not mention me by name. :)
My rating max'd at around 2310 and I played well enough to find my floor of
2200 (for awhile at least). The truth is if my rating dropped below 2200
I proably would have considered entering the World Open U2200 section. I
can't disagree that the floors should be abolished but .... I am glad
they were there! To think of playing in the U2200 sections makes me sick!
What stupid things we consider doing for a few extra bucks....
I'd hate myself in the morning...give any winning to charity ....
For the record... I have never played in the US Amateur Individual.
We reward lousy chess by giving awards for it not by allowing rating
floors.....
If the Caveman really thinks players enjoy being at their floor, then I
understand the origin of his name.
R,
Paul Powell
W C Newell Jr
In article <1994Jul10.0...@cis.uab.edu> sl...@cis.uab.edu (Kenneth Sloan) writes:
>In article <2vns39$1...@search01.news.aol.com> cavem...@aol.com (Cavemanchs) writes:
>>[summarizing a master's opinion on the ratings floors]
>>
>>... They discourage several of
>>my friends from playing, because they have bottomed out and cannot reach a
>>competitive level. They say that they have no motivation, because they
>>know their rating can't go down.
>
>No sympathy here. According to my reading, this translates to: "we'd
>like to sandbag, but can't". Let's be realistic here - if you are
>sitting on a floor, then your rating is at least 101 points below your
>(recent) peak. Are you seriously claiming that a player who has peaked
>at 2301+ really wants to play in the U2200 section? Much as I
>personally dislike the 100-point floors, I can't summon any sympathy for
>such a losing attitude.
It seems that anyone who dares to express an opinion against the ratings floors
in this forum has to then put up with your making the ugly insinuation of
"sandbagging". And yet the ratings committee members who post here remain
incredulous that their image is less than perfect...
Let's look at another example. Suppose I'm a lifelong class B player with an
established rating of 1760. Further suppose that I only play in a few
tournaments a year, preferring to play in the larger events in my state with
a specific class B section. Let's say two of these events are scheduled about
a month apart, and I decide to enter both of them in advance. I play well in
the first event and win my section (and a class prize) by going 5-0 against
players rated {1620, 1660, 1700, 1740, 1780}. My performance rating from this
event is -- 2100! From this I would expect to gain approx. 85 rating points.
Now let's say I repeat the performance in the second event against five other
plaeyrs with the exact same set of ratings as in the first event. I can do
this legitimately because it is unlikely the USCF will have published a new
rating for me within the four-week interval. Once again, I have a performance
rating of 2100. I will gain fewer rating points, approx. 60, since my rating
has presumably been recalculated after the first event.
The result of this is that I am left with a rating of approx. 1905 and an
absolute floor of 1800, which means I am stuck in class A, despite the fact
that I have not even >played< a class A opponent, much less defeated one. My
results have done nothing to suggest that I am stronger than high class B.
The only special thing I have done is win two class tournaments in a row,
certainly a believable result if I've had a bit of luck or if I'm better than
most at pressure play. I may proceed to get waxed in my next two tournaments
against class A competition, but the USCF, in its infinite wisdom, doesn't
want me back in class B. And if I complain about this state of affairs, it
is considered prima facie evidence by at least one member of the ratings
committee that I am a sandbagger/cheater/loser. (I might also mention that
the class prizes for which I'm allegedly corrupting myself average about $100.)
>>... I find
>>this opinion to be prevalent among chessplayers,
>
>By way of contrast, I find that most chessplayers don't really care,
>because the vast majority of them are well above their floors.
In other words, the committee is thumbing its collective nose at the 5% who
have experienced something similar to the above scenario and who have
subsequently become stuck at their floor, because they aren't the "majority".
>>... yet the ratings
>>committee/USCF seems to think otherwise.
>
>I can state with complete confidence that the vast majority of the USCF
>Ratings Committee thinks that the 100-point floors ought to be
>abolished. Where did you get the idea that anyone on the USCF Ratings
>committee supports these floors?
In my case it was the fact that their institution had the committee's
favorable recommendation in the first place.
>Have you read the current
>recommendations from that committee, which call for their abolishment?
Perhaps if you were to quote from the "current recommendations", which seem to
change quite frequently, you'd have more credibility on this point.
>May I suggest that if you really want the 100-point floors abolished
>that you refrain from badmouthing those who agree with you? Quite
>frankly, part of the reason that we are having so much trouble
>abolishing the floors is that we spend so much time distancing ourselves
>from comments such as yours.
[Noting that the original comments were by someone else, with which I happen to
agree.]
Absolute nonsense. The floors are with us because major national organizers
(the name Bill Goichberg somehow springs to mind) want them, and because
certain members of the committee (the name Mark Glickman somehow springs to
mind) give more weight to the opinions of the national organizers on the
Policy Board than to those of the lowly members at-large. Dr. Glickman has
admitted as much right here. I seriously doubt that the r.g.c. debate about
the floors has even registered a blip in the minds of the Policy Board,
especially when they are all so absorbed by the noise surrounding the OMOV
proposals.
--
Bill Newell
as always, a USCF Life Member
William B. Wright
(W C Newell Jr) wrote:
>
> The result of this is that I am left with a rating of approx. 1905 and an
> absolute floor of 1800, which means I am stuck in class A, despite the fact
> that I have not even >played< a class A opponent, much less defeated one. My
> results have done nothing to suggest that I am stronger than high class B.
Hardly. The given results, both 2100 performance ratings, have proven
conclusively that the person in your example is quite a bit better than
his/her Class B competition and should really strive to become competitive
in Class A where he/she belongs. Have a more positive attitude in the
games against Class A opposition and he/she will likely see his/her
preformance improve.
--
Pawn grubbing is its own reward!
___________________________
Replies to this e-mail address will bomb.
E-mail replies should be sent to
tob...@bert.lerc.nasa.gov
Bruce_L...@transarc.com
>>my friends from playing, because they have bottomed out and cannot reach a
>>competitive level. They say that they have no motivation, because they
>>know their rating can't go down.
>
>No sympathy here. According to my reading, this translates to: "we'd
>like to sandbag, but can't". Let's be realistic here - if you are
>sitting on a floor, then your rating is at least 101 points below your
>(recent) peak. Are you seriously claiming that a player who has peaked
>at 2301+ really wants to play in the U2200 section? Much as I
>personally dislike the 100-point floors, I can't summon any sympathy for
>such a losing attitude.
>
>And - *you* certainly have not "bottomed out". By my estimation, it's
>virtually certain that *your* floor is 2100. Can't your "friends" speak
>for themselves? Do they exist? Names, please.
It's an ad hominem attack to accuse someone of sandbagging, which is an
unethical practice, at this point in the argument.
Most of the Class A and Expert players that are regulars at Pittsburgh
events have seen variations of more than 100 points in their ratings. I can
think of Castillo, Ambrad, Kownacki, Schreiber, Renk, Lakatos, Geller, Magar,
Nedved, and Martinak. (The last three are Masters.) In some cases there have
been fluctuations of 200 points (Castillo). These fluctuations do not arise
from sandbagging.
Many of these players have been hitting their floors. While the percentage
of people in the rating list that are hitting their floors may be only 4%,
I estimate that those 4% account for a disproportionate share of all tournament
participants. The above named Pittsburgh players are "veterans" and account
for the lion's share of our Tuesday night tournaments as well as quite a lot
of weekend tournament and League games.
Rating fluctuations arise naturally from good and bad performances. My
own observation suggests that there is nothing particularly unusual about a
100+ rating point gain or loss for a player whose general grasp of the game
is not changing. I think it is regrettable if a player cannot compete for
a modest class prize in a Pittsburgh weekend tournament, because natural
fluctuations in his rating have given him an unnaturally high floor.
Having said that, I have to say that I do not know if ratings floors are really
as important a problem as Kevin Bachler has portayed them. Perhaps,
perhaps not. Hope I don't hit mine any time soon :-)
Paul A. Lane
>Itseems that anyone who dares to express an opinion against the ratings floors
>in this forum has to then put up with your making the ugly insinuation of
>"sandbagging". And yet the ratings committee members who post here remain
>incredulous that their image is less than perfect...
The sandbagging charge was loosely thrown (by another poster), though
it is what rating floors are to prevent.
[Example given of someone winning two Class B sectionals with 5-0]
>The result of this is that I am left with a rating of approx. 1905 and an
>absolute floor of 1800, which means I am stuck in class A, despite the fact
>that I have not even >played< a class A opponent, much less defeated one. My
>results have done nothing to suggest that I am stronger than high class B.
>The only special thing I have done is win two class tournaments in a row,
>certainly a believable result if I've had a bit of luck or if I'm better than
>most at pressure play.
Hold on here! This isn't playing well against Class B. This is waxing
10 opponents in a row! There's no magical barrier at 1800 (much like
the barrier at the edge of the galaxy in Star Trek). One of the wins
you used was against a 1760. That might simply be vs. a player who is
normally an 'A', but is down a little. Also, in order to gain around
150 points, the player had to initially on the verge of Class A.
Anyone who can beat Class B's that regularly should easily be a Class A
and is capable of at least 50-50 performances at the higher level.
>I may proceed to get waxed in my next two tournaments
>against class A competition, but the USCF, in its infinite wisdom, doesn't
>want me back in class B. And if I complain about this state of affairs, it
>is considered prima facie evidence by at least one member of the ratings
>committee that I am a sandbagger/cheater/loser. (I might also mention that
>the class prizes for which I'm allegedly corrupting myself average
>about $100
Average, perhaps. However, the CCA events average somewhere around $1000.
Mind you, it is possible to leap out of one's class. A friend of mine
leapt from C to A before falling back to his floor at 1700. However, as
my rating now approaches his, I can tell you that he is clearly stronger
than I. He simply hasn't had enough consistent tournament play to get
off and stay off of his floor.
>In other words, the committee is thumbing its collective nose at the 5% who
>have experienced something similar to the above scenario and who have
>subsequently become stuck at their floor, because they aren't the "majority".
I think you need to reduce your acetylene gas flow. The fact is, ratings
floors lead to rating inflation. When one player can gain points and the
other not lose any, it inflates the system. I believe a mechanism could
be instituted using norms to do much the same as floors.
I'm less sympathetic to someone who doesn't want to play because they're
on their floor. I would think that being on a floor is at least mildly
embarrassing. If I were on mine (at one time, I approached mine), I
would want to do everything possible to get off it.
Paul
--
Kenneth Sloan
>Note: Distribution changed to 'na'.
>
>In article <1994Jul10.0...@cis.uab.edu> sl...@cis.uab.edu (Kenneth Sloan) writes:
>>In article <2vns39$1...@search01.news.aol.com> cavem...@aol.com (Cavemanchs) writes:
>>>[summarizing a master's opinion on the ratings floors]
>>>
>>>... They discourage several of
>>>my friends from playing, because they have bottomed out and cannot reach a
>>>competitive level. They say that they have no motivation, because they
>>>know their rating can't go down.
>>
>>No sympathy here. According to my reading, this translates to: "we'd
>>like to sandbag, but can't". Let's be realistic here - if you are
>>sitting on a floor, then your rating is at least 101 points below your
>>(recent) peak. Are you seriously claiming that a player who has peaked
>>at 2301+ really wants to play in the U2200 section? Much as I
>>personally dislike the 100-point floors, I can't summon any sympathy for
>>such a losing attitude.
>
>It seems that anyone who dares to express an opinion against the ratings floors
>in this forum has to then put up with your making the ugly insinuation of
>"sandbagging". And yet the ratings committee members who post here remain
>incredulous that their image is less than perfect...
I'm sorry that you read it that way. Let me try again:
First: I oppose the 100-point ratings floors. I have taken every
possible opportunity to work against them.
Second: the vast majority of Ratings Committee Members oppose the
100-point ratings floors.
Third: one obstacle to repealing the 100-point ratings floors is the
argument put forward by the original poster, and you. The reason is
that proponents of the 100-points floors latch onto the opportunity to
refute this argument, and never seem to get around to the more telling
(in my opinion) questions.
Fourth: I stand by my personal opinion (which has nothing to do with
floors, and nothing to do with the Ratings Committee) that any player
capable of raising his rating to 2301+ demonstrates a "losing attitude"
by playing in U2200 sections. It is an exact parallel to my personal
opinion that any squash player who wins a Class C tournament ought to be
immediately elevated to Class B - permanently.
>
>Let's look at another example. Suppose I'm a lifelong class B player with an
>established rating of 1760. Further suppose that I only play in a few
>tournaments a year, preferring to play in the larger events in my state with
>a specific class B section. Let's say two of these events are scheduled about
>a month apart, and I decide to enter both of them in advance. I play well in
>the first event and win my section (and a class prize) by going 5-0 against
>players rated {1620, 1660, 1700, 1740, 1780}. My performance rating from this
>event is -- 2100! From this I would expect to gain approx. 85 rating points.
Thank you for making the discussion concrete with this excellent
example. I completely agree that this situation bears close scrutiny.
So, I'm going to examine it a great (not to say interminable) length.
Let's compute:
Opp 1760-Opp We (from table, p. 149, Elo, 2nd edition)
---- -------- --
1620 140 .69
1660 100 .64
1700 60 .59
1740 20 .53
1780 -20 .47
So, for the tournament, We = .69 + .64 + .59 + .53 + .47 = 2.92
Your new rating is:
Rn = Ro + K * (W - We)
Ro = 1760
K = 32
W = 5.0
We = 2.92
Rn = 1760 + 32 * (5.0 - 2.92)
= 1760 + 32 * (2.08)
= 1760 + 66.56
= 1826.56
which will be rounded up to 1827
>
>Now let's say I repeat the performance in the second event against five other
>plaeyrs with the exact same set of ratings as in the first event. I can do
>this legitimately because it is unlikely the USCF will have published a new
>rating for me within the four-week interval.
Yes. In fact, it is virtually impossible for a new rating to be
published, because there is usually a 4 week "printing and mailing" lead
time, plus a nominal 2 week "reporting and processing" interval.
> Once again, I have a performance
>rating of 2100. I will gain fewer rating points, approx. 60, since my rating
>has presumably been recalculated after the first event.
Once again, your computation does not match mine. The difference in
computation makes a profound difference in the conclusion - so I think
it is worth doing.
In your second effort, I get:
Opp 1827-Opp We (from table, p. 149, Elo, 2nd edition)
---- -------- --
1620 207 .77
1660 167 .72
1700 127 .67
1740 87 .62
1780 47 .57
So, for the tournament, We = .77 + .72 + .67 + .62 + .57 = 3.35
Your new rating is:
Rn = Ro + K * (W - We)
Ro = 1827
K = 32
W = 5.0
We = 3.35
Rn = 1827 + 32 * (5.0 - 3.35)
= 1827 + 32 * (1.65)
= 1827 + 52.8
= 1879.8
which will be rounded up to 1880.
Your 100-point floor is now 1700.
>
>The result of this is that I am left with a rating of approx. 1905
No - your rating will be 1880, assuming, of course, that the wall chart
ratings are correct - always dangerous. For example, in that second
tournament, there was a 1827 player masquerading as a lowly 1760.
>... and an
>absolute floor of 1800,
No - your floor is now 1700
>...which means I am stuck in class A,
No - you can still play with the weakies, if you so choose, once your
rating drops to 1799. The 100-point floor is keeping you out of U1700
events - but you are *not* "stuck in class A".
>...despite the fact
>that I have not even >played< a class A opponent, much less defeated one. My
>results have done nothing to suggest that I am stronger than high class B.
Here, we disagree. I happen to think that you *have* demonstrated that
you are a Class A player. But, my opinion is irrelevant. We were
discussing how the existing system would treat you. Just for grins,
let's see what would have happened to your "Lifetime Title"
In both events, a hypothetical 1800 player would have been expected
to score: .74 + .69 + .64 + .59 + .53 = 3.19. Your scores of 5.0 would
give you a "Certified Class A" norm in each case.
In both events, a hypothetical 1900 player would have been expected
to score: .83 + .80 + .76 + .72 + .67 = 3.78. Your scores of 5.0 would
give you an "Advanced Class A" norm in each case.
In both events, a hypothetical 2000 player would have been expected
to score: .90 + .88 + .85 + .82 + .78 = 4.23. Your scores of 5.0 would
[just barely: 4.23 + 0.6 = 4.83, and you scored 5.0]
give you a "Certified Expert" norm in each case.
[you would also get norms under the most likely table of deltas in the
newly proposes title system - but let's stick with the well understood
current system for now].
So, (assuming that these are career high performance) you would now
be 1880*A4 [translation: you are awarded the "Advanced Class A" title,
and have 4 points (out of the 10 required) to earn your "Certified
Expert" title. According to the present title system, you have
demonstrated that you are a solid 1900 player - but your Rating
continues to remember the bad old days when you were a weak B player.
So, the title system is very eager to mark you as a strong player, your
Rating is a bit more conservative, and the 100-point floor (way back at
1700) is very, very reluctant to push you into the deep end.
Let's step back and consider what you have accomplished, and how the
current system would treat you:
A) your USCF rating*title is 1880*A4
B) your 100-point floor is 1700
In my opinion, you have certainly demonstrated that you are no longer a
B player.
The current title system has awarded you the "1900" title (Advanced
Class A), on the strength of your two "middle-strength Expert"
performance.
The current 100-point floor system has you at 1700 - still eligible for
Class B events, if you can manage to shed 81 rating points.
I leave it as an exercise for the reader to invent a plausible sequence
of events which will drop your rating to 1799. I leave it as an
exercise in fantasy to invent a plausible sequence of events which will
drop you all the way to your floor of 1700 (remember your original
premise that you are a longtime B-player).
>The only special thing I have done is win two class tournaments in a row,
>certainly a believable result if I've had a bit of luck or if I'm better than
>most at pressure play.
Here is the crux of the matter. In my opinion, you are severely
underestimating the difficulty of achieving these two results. If these
happen to be the same 5 players, then the odds improve somewhat (premise
1: things work better in a large, mixed pool). If you have a complaint
about the computations, I suggest that you look at the We table, and
complain *first* about the fact that your rating shot up to 1880 after
these "believable results" for a 1760-strength player. Frankly, I do
not believe that it is reasonable to consider you a "high B" player
after two consecutive results such as these.
But...this has nothing to do with the 100-point floors [remember - I
still oppose them. I'm not arguing in *favor* of keeping them. I'm
trying to concentrate on correct arguments against them, and trying to
eliminate easily refuted arguments. Not surprisingly, the proponents of
the 100-point floors just *love* arguments such as yours - precisely
becaue they are so easily refuted.]
The 100-point floor system is still prepared to treat you like a low-B
player. It *does* refuse to consider you a high-C player after these
performances. I hope you will agree with that.
>... I may proceed to get waxed in my next two tournaments
>against class A competition, but the USCF, in its infinite wisdom, doesn't
>want me back in class B.
Let's continue this marathon. Suppose you go 0-5 against 1850
competition. In your next tournament, you are expected to score .54
against each 1850 opponent, so We = 5 * (0.54) = 2.7. If you are shut
out (not even a draw!), then you lose 32 * 2.7 = 86.4 points, and your
new rating is 1880-87 = 1793. You are back in Class B on rating, and
the 100-point floor of 1700 allows you to go back and play with the
fish.
Where's the beef? All this seems OK to me.
You might complain about the arbitrary 1800 barrier which differentiates
so strongly between ratings of 1799 and 1801. Some might argue that
this is a "feature" of the title system, since it relies on repeated
strong performances and rachets rather than dithering around arbitrary
ratings boundaries.
If I hd my druthers, I would call you (after all of these performances),
an "A4" player. More to the point - I would classify *all* of the 'Ax'
players that way, and move them *all* out of "Class B". The end result
would be that you end up playing against the same competition, with
slightly different labels. Remember - it *you* get moved up on
performances suche as these, so does everyone else!
>... And if I complain about this state of affairs, it
>is considered prima facie evidence by at least one member of the ratings
>committee that I am a sandbagger/cheater/loser.
Sandbagger? no.
Cheater? no.
Loser? yes. In my personal opinion (which has nothing to do with the
Ratings Committee), any player capable of going 10-0 against 1700
competition is wasting his time playing in Class B events. Just an
opinion.
In any event, if my calculations are correct (please do double check
them) on the example that *you* chose, then the current ratings system
does not prevent you from doing so, and the 100-point floor is abolutely
irrelevant to this example.
[Again - recall that I *oppose* the 100-point floors, on other grounds]
>... (I might also mention that
>the class prizes for which I'm allegedly corrupting myself average about $100.)
>
You have me confused with Marcus Roberts.
I did not mention corruption. I did not mention monetary incentives.
I didn't even mention *you*!
I respectfully request a public apology.
W C Newell Jr
>>[regarding my example showing how a class B player can get stuck with a
>>class A ratings floor]
>>
>> The result of this is that I am left with a rating of approx. 1905 and an
>> absolute floor of 1800, which means I am stuck in class A, despite the fact
>> that I have not even >played< a class A opponent, much less defeated one. My
>> results have done nothing to suggest that I am stronger than high class B.
>
>Hardly. The given results, both 2100 performance ratings, have proven
>conclusively that the person in your example is quite a bit better than
>his/her Class B competition and should really strive to become competitive
>in Class A where he/she belongs. Have a more positive attitude in the
>games against Class A opposition and he/she will likely see his/her
>preformance improve.
Re-read Elo's description of the significance of a performance rating. The
+400 used to calculate a performance rating is arbitrary; it comes from
Elo's estimate of the range (+ or -) of playing strength a player must face in
order for the rating to obtain statistical significance. If the player never
faces anyone outside his/her own class, much less +-400 from his/her own
rating, then the performance rating loses significance and is >NOT< a valid
estimate of playing strength. Another way of putting it is that a 10-game
winning streak against players of my own class is not sufficient to prove that
my playing strength is a full class higher. I could just as easily lose 10 in
a row over the next two events. You can press the philosophical argument that
a person who wins two class B tournaments should "try harder" to become a
class A player or else give up competitive chess (I obviously disagree), but
don't try to argue the significance of the Elo system -- the mathematics are
against you.
--
Bill Newell
Joseph Albert
W C Newell Jr <w...@biostat.washington.edu> wrote:
>a month apart, and I decide to enter both of them in advance. I play well in
>the first event and win my section (and a class prize) by going 5-0 against
>players rated {1620, 1660, 1700, 1740, 1780}. My performance rating from this
>event is -- 2100! From this I would expect to gain approx. 85 rating points.
>
>Now let's say I repeat the performance in the second event against five other
>plaeyrs with the exact same set of ratings as in the first event.
>The result of this is that I am left with a rating of approx. 1905 and an
>absolute floor of 1800, which means I am stuck in class A, despite the fact
>that I have not even >played< a class A opponent, much less defeated one.
the odds that a high class B player would win 10 games in a row against
class B players is very small, somewhere in the neighborhood of 1 in 1000.
the possibility that a B player might win, say 100 games in a row against
B players, and in the process attain an expert or master rating is even
much lower of course, but possible in theory.
However, ratings are a statistical measure of performance, one that works
reasonably well, and one that has such flaws, like very small non-zero
odds that a player can perform well against players of their own strength
and get a higher rating than warranted.
I do tend to agree with your point that this is an argument against
rating floors, which tend to rate people in the range of their lifetime
best performance period.
The USCF had the sandbagging problem completely flattenned their first
crack at it-- adorning a rating with a letter that indicates the highest
band of 100 points a player has achieved. this prevents sandbagging,
since the letter can be used to assign categories for prize determinations.
no rating floor is needed.
Joseph Albert
alb...@cs.wisc.edu
Mike Cormier
In a previous article, w...@biostat.washington.edu (W C Newell Jr) says:
[mucho deleto]
>The result of this is that I am left with a rating of approx. 1905 and an
>absolute floor of 1800, which means I am stuck in class A, despite the fact
>that I have not even >played< a class A opponent, much less defeated one. My
>results have done nothing to suggest that I am stronger than high class B.
>The only special thing I have done is win two class tournaments in a row,
>certainly a believable result if I've had a bit of luck or if I'm better than
>most at pressure play. I may proceed to get waxed in my next two tournaments
>against class A competition, but the USCF, in its infinite wisdom, doesn't
>want me back in class B. And if I complain about this state of affairs, it
>is considered prima facie evidence by at least one member of the ratings
>committee that I am a sandbagger/cheater/loser. (I might also mention that
>the class prizes for which I'm allegedly corrupting myself average about $100.)
>
But you do belong in class A, the way to become an A-Class player is to
beat B-Class players regularly, as you did. The way to become an expert
is to beat A-Class players regularly, etc. What you have done is to go
10-0 vs B-Class players, you belong in A-Class, and if your performance
rating is 2100 don't think at some point after meeting better competition
on a regular basis you will eventually win the A-Class? Are you satisfied
by beating people who don't really challenge you or are you more
interested in the money? You got waxed in the next two tournaments
because you haven't learnt how to beat A-Class players so need growth, by
sending you back to B-Class aren't you stagnating your growth as a chess
player?
MOP,
Chessically yours,
Mike
--
at555@@freenet.carleton.ca
W C Newell Jr
>>Itseems that anyone who dares to express an opinion against the ratings floors
>>in this forum has to then put up with your making the ugly insinuation of
>>"sandbagging". And yet the ratings committee members who post here remain
>>incredulous that their image is less than perfect...
>
>The sandbagging charge was loosely thrown (by another poster), though
>it is what rating floors are to prevent.
Not so loosely thrown when the writer is a member of the Ratings Committee
(IMHO).
>[Example given of someone winning two Class B sectionals with 5-0]
>
>Anyone who can beat Class B's that regularly should easily be a Class A
>and is capable of at least 50-50 performances at the higher level.
I contend that two class B wins is not "regularly". (See my previous post
about the frequent misunderstanding of the statistical significance of
performance ratings when play is restricted to one's own class.) The example
player could just as easily go 0-5 in the next two class B events. My
scenario is not abstract; I know many local players (including myself, once
upon a time) who have followed the pattern. I know players who have been
consistent class B strength for 30 years and yet have made several sojourns
into class A and may have a lifetime best of 1900+, and have also fallen into
class C one or more times after a few bad tournaments, despite their best
efforts to win.
>>In other words, the committee is thumbing its collective nose at the 5% who
>>have experienced something similar to the above scenario and who have
>>subsequently become stuck at their floor, because they aren't the "majority".
>
>I think you need to reduce your acetylene gas flow. The fact is, ratings
>floors lead to rating inflation. When one player can gain points and the
>other not lose any, it inflates the system. I believe a mechanism could
>be instituted using norms to do much the same as floors.
I'm all for the use of norms to award class prizes. I have stated several
times in previous articles that there are a number of ways to address the
sandbagging issue which don't involve disrupting the rating system; norms and
class titles are one such idea. There are people at the other extreme who want
to abandon the rating system outright and use class titles for pairings, but
that's another thread.
As for inflation, I've exhaused myself trying to argue that the USCF could do
a lot more than it's doing now to monitor inflationary effects in the ratings
pool, on the assumption that if we had hard data we might actually stop some
of the arguments. The ratings committee has acknowledged that the floors are
also serving to offset natural deflation in the pool, in lieu of the bonus
point system as originally defined in the Elo model. Whether there is any net
inflation in the pool as things stand today is anyone's guess.
>[...] I'm less sympathetic to someone who doesn't want to play because they're
>on their floor. I would think that being on a floor is at least mildly
>embarrassing. If I were on mine (at one time, I approached mine), I
>would want to do everything possible to get off it.
Some of us have jobs and families and simply do not have the time to pursue
competitive chess with the same vigor we had when we were still in school.
As I have said before, it's quite common for a person to be away from chess
for several years, and in the process lose a full class or more in playing
strength due to inactivity. The person may never have enough time to restudy
the game at the level required to regain his/her lifetime best playing
strength. One of the arguments against the floors is that they are a lifetime
burden, and they discriminate against older, less active players. The fact
that you consider being on a floor "embarrassing" goes to the heart of why I
think they should be abolished, and why I am so sensitive to the assertion that
rating = floor -> player = sandbagger.
--
Bill Newell
W C Newell Jr
>the odds that a high class B player would win 10 games in a row against
>class B players is very small, somewhere in the neighborhood of 1 in 1000.
Not exactly. If I'm high class B and my opponents are all low-mid class B
then the odds of a 10-0 run are closer to 1 in 100. Another factor is that
some players are much better than others in the final rounds when the pressure
is greater. The benefits of experience are independent of chess knowledge.
It's hard to quantify, but I'd say from my own observation (spanning 25 years)
that it reduces the odds to at worst 1 in 50. By the same token, the odds of
losing 10 in a row against my own class are not as long as you might think,
because players who start a class event badly tend to lose concentration and
fish out the rest of the way.
--
Bill Newell
Jeffrey Golds
I think they are a good idea, but need some work in the implementation.
Here is a scenrio which actually DID occur, I don't have all the ratings
or anything, but I can apprimate well enough so you get the point.
Our former club president had a 14 person round-robin game 10 tournament
one night. He is rated about 1900. The other rating varied between
say 2100 to 1600 (I am guessing, I actually was not there that night,
I just saw the crosstable from USCF.) Our club president went 13-0!!
A pretty amazing result! As a result of this his rating went up to
about 1980 (maybe his original rating was somewhat below 1900) and
he received his master certificate in the mail. How did that happen
you may be asking? Well, because of this one brilliant performance,
his lifetime achievement title (or whatever the proper term is) was
something like M2, or maybe m2. Anyhow, because of one tournament,
he was now counted as a master (in quick chess anyway). I think this
is unfair in two ways:
1) He will not be able to play in U2200 sections in quick chess anymore
because most TDs (ones that run large tournaments at least) are
going to see that title and forbid him from playing below his title
2) Do you really think he was a master? I don't think so. I think that
there needs to be a means whereby, it gets harder to make big jumps
such as this. Perhaps a method such as this would work:
Say Player A, scores well enough to earn title points in a tournament
where his delta is d. Why not make it a little harder for A to get
point for titles 2 jumps away from his current lifetime title.
Thus is A were at A5, then delta would be d for A and x but then
would be d+.1 (or someother choice) for X and m. Maybe you could
increment d again if you like for M etc.
I don't know if this has been discussed before, but I know a few local
masters who were dismayed at how "easily" this person got his master
certificate. He only had to do well in one tournament! In order to get
your rating that high from around 1900, would probably take significantly
longer.
I hope I haven't rambled too much. :)
Thanks for your time.
Jeff
Jeffrey Golds
Quinn Hubbard <qhub...@u.washington.edu> wrote:
>
>I haven't followed this thread closely and perhaps this point was
>already made. There are some older gentlemen who play tournament
>chess. As will happen to all of us, it is reasonable to expect that
>playing strength will eventually diminish with age. When one of these
>individuals hits their floor, it seems to me be an unfortunate
>situation.
>
I was under the (erroneous?) impression then your rating floor was
recomputed every two years. Thus if you stay at your floor, or near
enough for two years, you floor would become 1800. Maybe Mr. Sloan
can tell us :)
Jeff
Quinn Hubbard
Quinn Hubbard
sl...@cis.uab.edu (Kenneth Sloan) writes:
> You are back in Class B on rating, and
> the 100-point floor of 1700 allows you to go back and play with the
> fish.
Ouch, as a recent resident of Class B, this was fun to read!
Kenneth Sloan
Strangely, this situation has been used by both sides of the "floors"
debate.
On the one hand: "floors are bad, because it is embarassing when you
land on them".
On the other hand: "floors are good, because it saves you from the
embarrassment of seeing your rating fall even further".
[compare with the rules on publishing ratings of Life Masters - this
rule seems to have been abolished]
titles + current ratings manage to embarass you both ways...Your title
points out how far you are from your all time peak, and your current
rating points out what a rotten player you are.
If 100-point floors had been in effect 15 years ago, I would have landed
on *my* floor about 4 years ago. If memory serves, my "peak letter" was
N, corresponding to my very brief peak at 1902 (June 1983). When I
returned to regular OTB play several years ago, my rating briefly
dropped as low as 1777 (March-June 1992).
Was I embarassed? You bet!
Would a floor have mattered? Not a bit. With or without a floor, I
would have played in precisely the same events, and my current rating
would be exactly what it is now (1827*a4). I suspect that my floor is
still 1700 (I saw an unofficial peak of 1891 in May 1993 - I don't think
I played frequently enough for another, higher peak to sneak by me).
Even if I had managed to get those extra 9 points to 1900, just barely
establishing a floor of 1800, I (personally - not as a representative of
the University of Washington at Seattle) would consider it fair to say
that my strength varied over the range (1800,1900). My famed respect
for current ratings notwithstanding, I would be the first to say that
1777 was never a good estimate of my strength (except perhaps as an
*overestimate* for the three abysmal events which caused it to drop so low).
On the third hand, if I were to establish a floor of 1800 (please? oh
please? just one more flach of brilliance before I die?) I wouldn't feel
too badly about sitting on it, knowing that the points I pumped back
into the system are the ones that I stole from the system by rising from
1273 in the summer of 1975 (actually, that might have been strictly
provisional - so lets' go with 1483 in summer of 1976). Even my
extremely modest rise has deflated the pool by 400 points (spread out
over 70,000 players...) so I ought to give some back.
Now...my son Peter (see this month's Chess Life - use a magnifying
glass) will have hell to pay when he finally sits on his rating floor.
He's now at 422 - but expected to rise somewhat over the next 30 years.
Only then will he start to lose it, at 36...
His floor is 0300 - I think he'd be embarassed if he found himself
sitting on it. In fact, if the lower floors had been in effect earlier,
he might have had a floor of 400, and just barely missed it when he
bottomed out at 416. Of course, he's not an "older gentleman".
Place your bets here - who will get the "Certified Expert" title first?
Me? or Peter?
Paul A. Lane
I wrote:
>>The sandbagging charge was loosely thrown (by another poster), though
>>it is what rating floors are to prevent.
>Not so loosely thrown when the writer is a member of the Ratings Committee
>(IMHO).
Loose is loose. However, it seems that Ken did not mean to specifically
accuse your friends of sandbagging, though it's one of those "fighting
words"
>>[Example given of someone winning two Class B sectionals with 5-0]
>>
>>Anyone who can beat Class B's that regularly should easily be a Class A
>>and is capable of at least 50-50 performances at the higher level.
>I contend that two class B wins is not "regularly".
I disagree. Also, this isn't merely winning two in a row, but 10
consecutive victories. Also, see Ken's post for a more exact cal-
culation of the rating gain.
>(See my previous post about the frequent misunderstanding of the
>statistical significance of performance ratings when play is restricted
>to one's own class.) The example player could just as easily go 0-5 in
>the next two class B events.
Those kinds of fluctuations are statistically rare.
>My scenario is not abstract; I know many local players (including myself, once
>upon a time) who have followed the pattern. I know players who have been
>consistent class B strength for 30 years and yet have made several sojourns
>into class A and may have a lifetime best of 1900+, and have also fallen into
>class C one or more times after a few bad tournaments, despite their best
>efforts to win.
>I'm all for the use of norms to award class prizes. I have stated several
>times in previous articles that there are a number of ways to address the
>sandbagging issue which don't involve disrupting the rating system; norms and
>class titles are one such idea. There are people at the other extreme who want
>to abandon the rating system outright and use class titles for pairings, but
>that's another thread.
>Some of us have jobs and families and simply do not have the time to pursue
>competitive chess with the same vigor we had when we were still in school.
>As I have said before, it's quite common for a person to be away from chess
>for several years, and in the process lose a full class or more in playing
>strength due to inactivity. The person may never have enough time to restudy
>the game at the level required to regain his/her lifetime best playing
>strength. One of the arguments against the floors is that they are a lifetime
>burden, and they discriminate against older, less active players. The fact
>that you consider being on a floor "embarrassing" goes to the heart of why I
>think they should be abolished,and why I am so sensitive to the assertion that
>rating = floor -> player = sandbagger.
We wind up agreeing. I do wonder if there is an equal fraction of
players on their floor who would fight like hell to keep the floor
so that their rating doesn't fall further. Of course, that is a
reason to abolish the floor as it can artificially prop up a rating.
In any event, it seems to me that both you and Ken agree, so you
needn't worry about flaming him.
Paul
--
W C Newell Jr
>possible opportunity to work against them.
>
>Second: the vast majority of Ratings Committee Members oppose the
>100-point ratings floors.
>
>Third: one obstacle to repealing the 100-point ratings floors is the
>argument put forward by the original poster, and you. The reason is
>that proponents of the 100-points floors latch onto the opportunity to
>refute this argument, and never seem to get around to the more telling
>(in my opinion) questions.
For someone who claims to agree with me, you seem to be making a surprisingly
assertive effort to refute me. On the assumption that you are merely playing
devil's advocate, I will suggest that you bear with me a bit longer, because
I believe my argument can be reinforced.
>Fourth: I stand by my personal opinion (which has nothing to do with
>floors, and nothing to do with the Ratings Committee) that any player
>capable of raising his rating to 2301+ demonstrates a "losing attitude"
>by playing in U2200 sections. It is an exact parallel to my personal
>opinion that any squash player who wins a Class C tournament ought to be
>immediately elevated to Class B - permanently.
And I stand by my personal opinion that anyone who fails to recognize the
legitimate circumstances by which a player can drop a class in strength (for
example, by aging or being inactive) isn't facing reality.
Correct me if I'm wrong, but squash (as virtually all other athletic sports)
segregates players by age, providing opportunities for over-40 players to
compete head-to-head without regard to their peak performance at age 20.
>>Let's look at another example. Suppose I'm a lifelong class B player with an
>>established rating of 1760. Further suppose that I only play in a few
>>tournaments a year, preferring to play in the larger events in my state with
>>a specific class B section. Let's say two of these events are scheduled about
>>a month apart, and I decide to enter both of them in advance. I play well in
>>the first event and win my section (and a class prize) by going 5-0 against
>>players rated {1620, 1660, 1700, 1740, 1780}. My performance rating from this
>>event is -- 2100! From this I would expect to gain approx. 85 rating points.
>
>Thank you for making the discussion concrete with this excellent
>example. I completely agree that this situation bears close scrutiny.
>So, I'm going to examine it a great (not to say interminable) length.
>
>[details omitted for space]
OK, so my math was a little off, and the numbers in the example above give me
a rating of 1880, not 1905. I can change the example ever so slightly and
make the intended point that I can start as a high class B (let's say I start
at 1780, not 1760, and that the average of my opponents is 1725, not 1700),
go 10-0 in two class B sectionals and finish slightly over 1900, without ever
playing, much less beating, anyone above my class.
>>...despite the fact
>>that I have not even >played< a class A opponent, much less defeated one. My
>>results have done nothing to suggest that I am stronger than high class B.
>
>Here, we disagree. I happen to think that you *have* demonstrated that
>you are a Class A player. But, my opinion is irrelevant. We were
>discussing how the existing system would treat you. Just for grins,
>let's see what would have happened to your "Lifetime Title"
>
>In both events, a hypothetical 1800 player would have been expected
>to score: .74 + .69 + .64 + .59 + .53 = 3.19. Your scores of 5.0 would
>give you a "Certified Class A" norm in each case.
>
>In both events, a hypothetical 1900 player would have been expected
>to score: .83 + .80 + .76 + .72 + .67 = 3.78. Your scores of 5.0 would
>give you an "Advanced Class A" norm in each case.
>
>In both events, a hypothetical 2000 player would have been expected
>to score: .90 + .88 + .85 + .82 + .78 = 4.23. Your scores of 5.0 would
>[just barely: 4.23 + 0.6 = 4.83, and you scored 5.0]
>give you a "Certified Expert" norm in each case.
>[you would also get norms under the most likely table of deltas in the
>newly proposes title system - but let's stick with the well understood
>current system for now].
And I would argue that this data shows how the both the existing and proposed
title systems are fatally flawed -- they allow me to achieve an Expert norm
without playing anyone higher than class B! This is clearly against the
theory of the Elo model. The FIDE system does not allow the awarding of norms
significantly above the top-rated players in the event; in other words, I
cannot get a GM norm (performance 2600, I think) by going 9-0 in a 10-RR event
where my opposition averages only FIDE 2200. I should be required to play
class A opposition before achieving a class A norm, ditto for Expert and above.
>So, (assuming that these are career high performance) you would now
>be 1880*A4 [translation: you are awarded the "Advanced Class A" title,
>and have 4 points (out of the 10 required) to earn your "Certified
>Expert" title. According to the present title system, you have
>demonstrated that you are a solid 1900 player - but your Rating
>continues to remember the bad old days when you were a weak B player.
>So, the title system is very eager to mark you as a strong player, your
>Rating is a bit more conservative, and the 100-point floor (way back at
>1700) is very, very reluctant to push you into the deep end.
Except in my example we are assuming that my initial rating of 1780 is
>accurate<, and that my 10-0 result is due to a combination of luck and
circumstance against other class B players. You use a flawed title system to
"prove" that I am a solid 1900 player, and then turn around and use your
assertion that I'm really a 1900 player to defend the rating system. It's
just circular logic.
>I leave it as an exercise for the reader to invent a plausible sequence
>of events which will drop your rating to 1799. I leave it as an
>exercise in fantasy to invent a plausible sequence of events which will
>drop you all the way to your floor of 1700 (remember your original
>premise that you are a longtime B-player).
Since I'm a longtime class B player burdened with a 1900 rating and forced to
play in class A sections, I then proceed to go 0-10 in my next two events.
If we assume that I broke 1900 to begin with, then my floor is 1800 and I am
now stuck at my floor. If I peaked at say, 1880, then I've fallen to about my
original rating, in the 1760-1780 range. Now let's say I'm discouraged by my
poor results and I ignore chess for six months, no club play, no study, no
nothing. Then I decide to give it another go and enter a class B event, but
since I'm not prepared I fish out and score 0.5-4.5. I'm now at my floor of
1700. Fantasy? No, it happens every weekend, which is why we have 4-5% of
the ratings pool sitting on their floor.
>>The only special thing I have done is win two class tournaments in a row,
>>certainly a believable result if I've had a bit of luck or if I'm better than
>>most at pressure play.
>
>Here is the crux of the matter. In my opinion, you are severely
>underestimating the difficulty of achieving these two results. If these
>happen to be the same 5 players, then the odds improve somewhat (premise
>1: things work better in a large, mixed pool). If you have a complaint
>about the computations, I suggest that you look at the We table, and
>complain *first* about the fact that your rating shot up to 1880 after
>these "believable results" for a 1760-strength player. Frankly, I do
>not believe that it is reasonable to consider you a "high B" player
>after two consecutive results such as these.
I base my assessment on a 25-year history of observing local players in action
(several of whom are personal friends of mine), and I have seen this pattern
over and over again. Swings of 200 points happen much more frequently in
practice than your prediction formulas suggest. Other people have made the
same comments. I don't dispute your computations, but I choose to believe the
reality rather than the formula.
>The 100-point floor system is still prepared to treat you like a low-B
>player. It *does* refuse to consider you a high-C player after these
>performances. I hope you will agree with that.
No, because once I hit 1801, at least temporarily, then my floor is 1700.
I'm not being treaded as low-B, I'm being treated as high-B. Once I hit 1901,
at least temporarily, I'm being treated as class A, and per the example above,
that may not be appropriate.
Once upon a time I asked you why it was decided to use 100-point floors, as
opposed to say, 200, and I don't recall that you ever explained this.
>You might complain about the arbitrary 1800 barrier which differentiates
>so strongly between ratings of 1799 and 1801. Some might argue that
>this is a "feature" of the title system, since it relies on repeated
>strong performances and rachets rather than dithering around arbitrary
>ratings boundaries.
The only reason the class boundaries are significant is because tournament
organizers choose to make it so. If class prizes were awarded on the basis of
achieving a norm rather than simply outscoring lower-rated competition, the
monetary incentive to manipulate one's rating would be eliminated, and then it
would be easier to understand the complaints from people who care more about
level competition than they do about winning a cash prize.
>If I hd my druthers, I would call you (after all of these performances),
>an "A4" player. More to the point - I would classify *all* of the 'Ax'
>players that way, and move them *all* out of "Class B". The end result
>would be that you end up playing against the same competition, with
>slightly different labels. Remember - it *you* get moved up on
>performances suche as these, so does everyone else!
Nonsense. The 1 in 50 class B players who at one time or another have the
same run I do will get moved up, but 98% of the class A pool will be legitimate
class A players, and they'll consistently beat me.
>>... (I might also mention that the class prizes
>>for which I'm allegedly corrupting myself average about $100.)
>>
>You have me confused with Marcus Roberts.
>
>I did not mention corruption. I did not mention monetary incentives.
>I didn't even mention *you*!
>
>I respectfully request a public apology.
Give me a break. Instead of an apology, I'll give you some advice: sit back,
fold your arms, take a deep breath and speak the holy mantra, "It's only
Usenet."
--
Bill Newell
Kenneth Sloan
>I'm kind of glad that Mr. Sloan brought up the lifetime titles.
>I think they are a good idea, but need some work in the implementation.
>Here is a scenrio which actually DID occur, I don't have all the ratings
>or anything, but I can apprimate well enough so you get the point.
>
>Our former club president had a 14 person round-robin game 10 tournament
>one night. He is rated about 1900. The other rating varied between
>say 2100 to 1600 (I am guessing, I actually was not there that night,
>I just saw the crosstable from USCF.) Our club president went 13-0!!
>A pretty amazing result! As a result of this his rating went up to
>about 1980 (maybe his original rating was somewhat below 1900) and
>he received his master certificate in the mail. How did that happen
>you may be asking? Well, because of this one brilliant performance,
>his lifetime achievement title (or whatever the proper term is) was
>something like M2, or maybe m2. Anyhow, because of one tournament,
>he was now counted as a master (in quick chess anyway).
First - this is just the sort of thing that the Ratings Committee is
trying to fix this year. Essentially, the problem is that the current
set of Deltas was picked out of thin air, with much handwaving, and is
seriously incorrect in that norms are very easy to get as the number of
rounds goes up. In a 14 round tournament, getting norms is nearly
trivial.
>...I think this
>is unfair in two ways:
>1) He will not be able to play in U2200 sections in quick chess anymore
> because most TDs (ones that run large tournaments at least) are
> going to see that title and forbid him from playing below his title
>
Not necessarily - first because is't QuickChess and second because not
that many TDs are looking too hard at the titles, yet. Now - if this
were slow chess, and you wanted to play in the World Open...forget it.
>2) Do you really think he was a master? I don't think so. I think that
> there needs to be a means whereby, it gets harder to make big jumps
> such as this. Perhaps a method such as this would work:
> Say Player A, scores well enough to earn title points in a tournament
> where his delta is d. Why not make it a little harder for A to get
> point for titles 2 jumps away from his current lifetime title.
> Thus is A were at A5, then delta would be d for A and x but then
> would be d+.1 (or someother choice) for X and m. Maybe you could
> increment d again if you like for M etc.
I think that the new table of Deltas wil fix this. If would be a great
help if you could locate that cross-table, and send it to me. Even just
the ratings of the opponents will help. I'd like to see what norms were
earned under the current system, and what norms would be earned under
the proposal going to the Policy Board in August.
>
>I don't know if this has been discussed before, but I know a few local
>masters who were dismayed at how "easily" this person got his master
>certificate. He only had to do well in one tournament! In order to get
>your rating that high from around 1900, would probably take significantly
>longer.
>
>I hope I haven't rambled too much. :)
Nope. This is a legitimate problem, which was identified last year, and
is being fixed ASAP.
-Ken "it's Herr Dr. Prof. Sloan - but you can call me Ken" Sloan
Kenneth Sloan
>
>I was under the (erroneous?) impression then your rating floor was
>recomputed every two years. Thus if you stay at your floor, or near
>enough for two years, you floor would become 1800. Maybe Mr. Sloan
>can tell us :)
You probably got that idea because New Windsor seems to change the rules
every 2 years...
The intent of the floors is that they are permanent. But - you can
petition to have your floor lowered. Just collect up your recent
results and write a letter to Al Lawrence. It's worth a shot.
Michael Roca
This is an odd discussion; nobody likes the floors, they just disagree about
what's wrong with them.
There seems to be an assumption that a player's ability will only advance
during their lifetime, or if it doesn't then they're delibrately sandbagging.
Let me given my own case history. This actually happened a bit before rating
floors, but assume it didn't. I played competative chess occasionally for
many years and got about a 1600 rating. Then for one summer I did nothing
(and I mean nothing) but play chess, including rated tournaments most every
weekend. My performance rating for that one summer was 1900 or 2000 and my
rating went over 1800. So far, so good. That was really the level I was
playing at. I've played very little since then and my ability has dropped
drastically. I might not even be at the 1600 level if you dropped me into a
tournament right now. But if you wanted to place me in a tournament based on
my rating I'd get killed. Now I can live with that because it would
presumably be a temporary thing. I would regain my ability and my rating
would settle to match it. But what if I hit a rating floor? Instead of my
rating coming down to my ability my ability would have to come up to my
rating. It would be very discouraging to walk into tournaments knowing I
wasn't in shape to take on my opponents. Who wants to spend a weekend and
entry fee just to get killed? This assumes I could regain my old form. I
don't see why I couldn't, but then I don't see any guarentee I would.
As another example, what about Botvinnik (I think he's alive) and Smyslov and
<fill in old GM here>?
As a third example, what about those youth players who rack up huge ratings
against players their own age but then come back to earth when they encounter
adults?
I think everybody agrees on the solution. Let the rating go where it will and
tag the playes will some sort of high class achieved title so they can't
sandbag - and have something to boast about :-)
Hal Bogner
>
>However, ratings are a statistical measure of performance, one that works
>reasonably well, and one that has such flaws, like very small non-zero
>odds that a player can perform well against players of their own strength
>and get a higher rating than warranted.
>
Perhaps this was mentioned so long ago that it's disappeared from the thread,
but I read somewhere that your peak rating indicated your ability (as
demonstrated through peak performances), and your current rating simply
reflected your current level of performance.
Floors present two problems:
1. They distort (destroy?) the usefulness of the rating as a predictor of
results, which - among other things - undermines the advantages of
ratings-controlled swiss pairings (I haven't seen this mentioned before).
2. For players who are indeed playing at a lower level than before (and this
does not have to result from sandbagging; aging, loss of focus on chess due to
other things going on in the players' life, and other causes may legitamately
exist), their ratings may never catch "up" to their current level of play,
trapping them in a place where they cannot meet the expectations their ratings
create (i.e., no chance at class prizes, few or no results that gain rating
points or otherwise "look good," etc.).
>I do tend to agree with your point that this is an argument against
>rating floors, which tend to rate people in the range of their lifetime
>best performance period.
>
>The USCF had the sandbagging problem completely flattenned their first
>crack at it-- adorning a rating with a letter that indicates the highest
>band of 100 points a player has achieved. this prevents sandbagging,
>since the letter can be used to assign categories for prize determinations.
>no rating floor is needed.
>
Absolutely!
>Joseph Albert
>alb...@cs.wisc.edu
I have seen several posts that misunderstand why Bill Goichberg wanted the
floors. Sandbagging isn't the main reason, as I have heard him tell it.
Rather, it's a fear that as players lose their ability to perform at the level
their rating peaked at, they "become discouraged" and drop out of chess.
If I may resort to aphorisms, it's a chicken and the egg problem. Goichberg,
as other before him did also, pointed to the world of bridge, where one gains
Master Points, and keeps them for life, as a better model. The grass always
being greener on the other side of the hill, he convinced enough people to try
a limited version of that with the floors we have today.
If anyone wants to know where I stand, I would say I side with Bruce Leverett,
out of the posts I have seen so far.
-hal
Mark E. Glickman
<lots of stuff deleted>
>Absolute nonsense. The floors are with us because major national organizers
>(the name Bill Goichberg somehow springs to mind) want them, and because
>certain members of the committee (the name Mark Glickman somehow springs to
>mind) give more weight to the opinions of the national organizers on the
>Policy Board than to those of the lowly members at-large. Dr. Glickman has
>admitted as much right here.
Bill, you have lost your mind. Perhaps you *wish* I admitted that,
so you can continue your potshots. Not only are you wrong that I
give more weight to nat'l organizers on the policy board than to
"regular" USCF members, but I don't even know if anyone on the policy
board is a national organizer. Bill Goichberg is not on the policy
board.
- Mark
David Goldfarb
)Now...my son Peter['s] floor is 0300 - I think he'd be embarassed if he
)found himself sitting on it.
Correct me if I'm wrong, but there are no floors for ratings
below 1200. After all, once you're in class E, there's nowhere to
sandbag to.
On a larger point related to scholastic chess, I can recall a few
years ago when the high point of the ratings histogram was around 1500.
I look at the ones posted now, and it's down around 1000. We would seem
to be looking at some *massive* ratings deflation over the next ten years
or so as these low-rated juniors improve.
David Goldfarb |"Tom?...I don't get you."
gold...@ocf.berkeley.edu | "Nobody does. I'm the wind, baby."
gold...@UCBOCF.BITNET |
gold...@soda.berkeley.edu | -- Mystery Science Theater 3000
David Goldfarb
)Goichberg,
)as other before him did also, pointed to the world of bridge, where one gains
)Master Points, and keeps them for life, as a better model. The grass always
)being greener on the other side of the hill, he convinced enough people to try
)a limited version of that with the floors we have today.
As a reader of both rgc and rec.games.bridge, this strikes me
as an extremely silly idea. Nobody on rec.games.bridge takes master points
at all seriously ("potzer points" is a common epithet for them). Elo ratings
have a *much* better correlation with chess playing ability than ACBL master
points do with bridge. Surely this is to be desired?
Further, bridge in the US is on the decline while chess is
relatively healthy. Why imitate something that works less well?
David Goldfarb |"You will know pain."
gold...@ocf.berkeley.edu | "You will know fear."
gold...@UCBOCF.BITNET |"And then you will die. Have a pleasant flight."
gold...@soda.berkeley.edu | -- Babylon 5, "The Parliament of Dreams"
Kenneth Sloan
>Kenneth Sloan <sl...@cis.uab.edu> wrote:
>)Now...my son Peter['s] floor is 0300 - I think he'd be embarassed if he
>)found himself sitting on it.
>
> Correct me if I'm wrong, but there are no floors for ratings
>below 1200. After all, once you're in class E, there's nowhere to
>sandbag to.
At a recent Policy Board meeting, the 100-point floors were extended all
the way down to 0000. Of course, there's no guarantee that they have
actually been implemented, yet - but they exist in principle. There is
also an (unadvertised feature) absolute floor at 0000 - which a few
players have bounced off. (so...if you establish a new rating at 0045,
your floor is 0000, and not -100, and it is impossible to have a rating
(provisional or established) lower than 0000 - but a couple of players
have managed to achieve this feat (see below)!
>
> On a larger point related to scholastic chess, I can recall a few
>years ago when the high point of the ratings histogram was around 1500.
>I look at the ones posted now, and it's down around 1000. We would seem
>to be looking at some *massive* ratings deflation over the next ten years
>or so as these low-rated juniors improve.
Absolutely correct. As long as the incremental update formulae remain
zero-sum, and as long as players tend to improve (overall) during their
tournament careers, there will be deflation (unless, of course, it is
balanced by other inflationary effects).
================================================================
[massive aside: it may be best to look an only ESTABLISHED players.
Here's the current (1 June 1994) histogram for ESTABLISHED players only
(this is historical data going back 2-3 years, some ratings may be stale)
ESTABLISHED RATINGS ONLY
Min = 45, Max = 2763, N = 34018, mu = 1495, sigma = 419
< 100 : 1 0.000 0.000
100 - 199 : 2 0.000 0.000
200 - 299 : 13 0.000 0.000
300 - 399 : 43 0.001 0.002
400 - 499 : 131 0.004 0.006
500 - 599 : 264 0.008 0.013
600 - 699 : 532 0.016 0.029
700 - 799 : 821 0.024 0.053
800 - 899 : 1275 0.037 0.091
900 - 999 : 1609 0.047 0.138
1000 - 1099 : 1856 0.055 0.192
1100 - 1199 : 2144 0.063 0.255
1200 - 1299 : 2360 0.069 0.325
1300 - 1399 : 2480 0.073 0.398
1400 - 1499 : 2858 0.084 0.482
1500 - 1599 : 2980 0.088 0.569
1600 - 1699 : 3013 0.089 0.658
1700 - 1799 : 2829 0.083 0.741
1800 - 1899 : 2588 0.076 0.817
1900 - 1999 : 2132 0.063 0.880
2000 - 2099 : 1648 0.048 0.928
2100 - 2199 : 1182 0.035 0.963
2200 - 2299 : 694 0.020 0.983
2300 - 2399 : 294 0.009 0.992
2400 - 2499 : 153 0.004 0.997
2500 - 2599 : 68 0.002 0.999
2600 - 2699 : 44 0.001 1.000
2700 - 2799 : 4 0.000 1.000
2800 - 2899 : 0 0.000 1.000
> 2900 : 0 0.000 1.000
As you can see, the mode is 1600-1699, and the mean is 1495
by way of contrast, here are the PROVISIONAL players - same pool:
PROVISIONAL RATINGS ONLY
Min = 0, Max = 2733, N = 31299, mu = 987, sigma = 359
< 100 : 8 0.000 0.000
100 - 199 : 62 0.002 0.002
200 - 299 : 178 0.006 0.008
300 - 399 : 479 0.015 0.023
400 - 499 : 1137 0.036 0.060
500 - 599 : 2050 0.065 0.125
600 - 699 : 2819 0.090 0.215
700 - 799 : 3584 0.115 0.330
800 - 899 : 3752 0.120 0.450
900 - 999 : 3478 0.111 0.561
1000 - 1099 : 3210 0.103 0.663
1100 - 1199 : 2710 0.087 0.750
1200 - 1299 : 2137 0.068 0.818
1300 - 1399 : 1655 0.053 0.871
1400 - 1499 : 1267 0.040 0.911
1500 - 1599 : 954 0.030 0.942
1600 - 1699 : 630 0.020 0.962
1700 - 1799 : 388 0.012 0.974
1800 - 1899 : 253 0.008 0.982
1900 - 1999 : 195 0.006 0.989
2000 - 2099 : 119 0.004 0.993
2100 - 2199 : 98 0.003 0.996
2200 - 2299 : 50 0.002 0.997
2300 - 2399 : 42 0.001 0.999
2400 - 2499 : 27 0.001 0.999
2500 - 2599 : 13 0.000 1.000
2600 - 2699 : 3 0.000 1.000
2700 - 2799 : 1 0.000 1.000
2800 - 2899 : 0 0.000 1.000
> 2900 : 0 0.000 1.000
Note that the mode has shifted to 800-899, and the mean to 987.
Guess what? Established players are, generally speaking, stronger than
Provisional players!
One can, if necessary, start to split out the Provisional players by
#games. My fuzzy memory tells me that this has been done, and there is
very little effect - the Provisionals based on 4 games look very much
like the Provisionals based on 19 games. But...that's fuzzy recollection.
end of massive aside...]
================================================================
Some amount of inflationary counter-pressure is needed.
Once upon a time, USCF followed Elo's suggestion of bonus points (and
feedback points). Alas, (in my opinion) this got out of hand ("Gee
Wally! Players like it when their ratings go up! Let's make them *all*
go up!"). The result is that the baby was tossed out with the
bathwater.
Now, the only inflationary force I can identify is the 100-point floors
- but I'm not at all sure how much of a force this is.
The most difficult problem is figuring out how to measure and track
"inflation". Looking at the distribution of ratings in the pool at
various points in time doesn't work (the pool changes character).
Looking at the relationship of the upper tail of USCF ratings and the
lower tail of FIDE ratings doesn't work (the tails are inherently
unreliable, and there is no guarantee that any shift can be reliably
assessed to one scale or the other - even if they stay in constant
alignment, they may well be *both* shifting against someone's idea of
"constant ratings).
The easy way out is to recognize that the Elo scales (all of them) have
no fixed anchors. They float. Their only promise is to deliver
relative judgements - *not* absolute ones. This appears to be
unacceptable to many people.
The hard way is to identify some extra-theoretical anchors, and
mechanisms to make sure that the scale stays fixed with respect to those
anchors.
Which leads to: what do you mean by a fixed (over time) "chess
strength"?
Peter Stein
David Goldfarb <gold...@ocf.berkeley.edu> wrote:
>Kenneth Sloan <sl...@cis.uab.edu> wrote:
>)Now...my son Peter['s] floor is 0300 - I think he'd be embarassed if he
>)found himself sitting on it.
>
> Correct me if I'm wrong, but there are no floors for ratings
>below 1200. After all, once you're in class E, there's nowhere to
>sandbag to.
>
> On a larger point related to scholastic chess, I can recall a few
>years ago when the high point of the ratings histogram was around 1500.
>I look at the ones posted now, and it's down around 1000. We would seem
>to be looking at some *massive* ratings deflation over the next ten years
>or so as these low-rated juniors improve.
Better deflation than inflation, but it probably will not happen anyway.
For openers so long as floors are in effect there exists a potent inflationary
source. The number of players hitting their floor is increasing and improving
juniors may help accelerate this trend, but as the number of floor players
increases so does the flow of rating points created. I concur with Bruce
Leverett's observation that floor players tend to be active tournament
participants, that being the case they will continue fueling inflation.
Another important point is that we're doing a miserable job of retaining
juniors beyond junior high and high school where they are apt to make great
strides due to more interaction with adults, access to computers, etc.
I was astounded to discover the numbers of players participating in the grade
school events here in Illinois and from what I've heard this is a nation wide
phenomenon. But their interest wanes in junior high or high school. That being
the case it is sadly true that they will not be contributing to the deflation.
> David Goldfarb |"Tom?...I don't get you."
>gold...@ocf.berkeley.edu | "Nobody does. I'm the wind, baby."
>gold...@UCBOCF.BITNET |
>gold...@soda.berkeley.edu | -- Mystery Science Theater 3000
Peter Stein
pst...@falcon.depaul.edu
Kenneth Sloan
>...
>Better deflation than inflation,
Now this truly puzzles me. Why is either preferable to the other?
>... but it probably will not happen anyway.
>For openers so long as floors are in effect there exists a potent inflationary
>source.
How potent is it?
>... The number of players hitting their floor is increasing
It is? Please send me the data. I *was* going to do that analysis
tonight, but if you've already done it I can save my time.
>Another important point is that we're doing a miserable job of retaining
>juniors
Sadly, this is true. Retaining juiors (and especially retaining girls)
is a significant challenge.
>... That being
>the case it is sadly true that they will not be contributing to the deflation.
>
On the contrary. In Illinois, and in Rochester, NY, we see significant
evidence of the "career effect" on an accellerated scale.
In the beginning, a "tournament" player was already seasoned by club
(unrated) play, and more likely than not to be an adult, or at least a
teenager (on average). Such a player might see a 500-600 point
difference between their first *established* rating and their final
(retiring) rating. Turnover was relatively low.
Now, in some locales, it is not unusual for a First Grader to come into
the system with an established rating of 600 - play for 5 years and quit
with a rating of 1200. Just about the same ort of churning, but with
larger numbers. By the time these kids hit puberty and discover other
interests, they've already deflated the pool. There are other effects
which I'm not yet convinced about - but I firmly believe that the boom
in Scholastics (our great success in getting chess into the schools, and
turning formerly informal unrated play into official USCF-rated events)
is fueling a terrific deflationary pressure.
Here - the floors may help (never mind that I abhor the floors...) stem
the tide. They may even be self-regulating (if the floor adds points to
the pool, everyone goes up. This lifts the players on the floor *off*
their floor, and we go again).
But, in the long run - the right thing to do is to remove the zero-sum
nature of the ratings updating formulae. The rating system must be
fixed to recognize when an individual jumps in strength (and,
conversely, when he declines) and allow that player's rating to rise
(fall) without necessarily making up the difference from his opponent's
rating.
To calibrate - I offer the observation that the K section in San Jose
was roughly half rated and half unrated. My youngest was the second
lowest *previsously rated* player in the group (at about 430). He
finished right were he began, squarely in the middle of this pack. Draw
what conclusions you will. But beware - there was, I think, an 1100
player who finished about 3/4 of the way down.
Anyway - I would not be terribly surprised if his rating were to go up
by 200 points next year, and by 100 points per year for a few years
after that. If that happens, that will equal the deflationary effect of
the "typical" player of ten years ago. [approximately! did I mention
that this was an approximation?]
Oh yeah - the "other effect". If the local scholastic pool doesn't mix well
with the (generally homogeneous) adult pool, then the deflationary
effect is tremendously accellerated. Since the pool is smaller, and the
careers are shorter, the downward spiral goes faster.
At least...that's the theory. No real data, yet.
Kenneth Sloan
>In article <1994Jul13....@hal.depaul.edu> pst...@falcon.depaul.edu (Peter Stein) writes:
>>...
>
>
>It is? Please send me the data. I *was* going to do that analysis
>tonight, but if you've already done it I can save my time.
>
OK, Ken - always glad to provide data (I know - it's not good form to
follow up your own post).
By George, Peter seems to be correct. Here's the trend:
rating list at xx00 increase
----------- ------- --------
December 1993 0.0658
June 1993 0.0761 0.0103
December 1994 0.0827 0.0066
June 1994 0.1004 0.0177
This includes all *established* players with ratings in [1400,2199]
Note that the June lists include players who have had a rated result in
the previous 6 months, while the December lists include players who
have had a rated result in the previous 12 months.
There are other, second-order effects which I think I see in the
detailed data - but Mark Glickman did such a fine job on this point last
year that I think I'll let him do it again this year.
Over to you, Mark...
But first...is this data a warning of impending inflation? or evidence
of recent deflation?
Kenneth Sloan
> rating list at xx00 increase
> ----------- ------- --------
> December 1993 0.0658
> June 1993 0.0761 0.0103
> December 1994 0.0827 0.0066
> June 1994 0.1004 0.0177
oops
will...@delphi.com
> Further, bridge in the US is on the decline while chess is
>relatively healthy. Why imitate something that works less well?
How many people are members in each organization?
USCF just (gasp) passed the 70-000 mark. (Someone ocorrect me if I'm wrong.)
The U.S. bridge federation (whatever it is) has had how many members
since when?
Chess "relatively healthy" is an aggressive no-brainer. An organization
boasting a mere 70,000 members, out of a population of 300 million, of
whom a good several million know how to play the game is not doing as
well in any measurement as the bridge competition.
will...@delphi.com
>Rather, it's a fear that as players lose their ability to perform at the level
>their rating peaked at, they "become discouraged" and drop out of chess.
Nevertheless, I'd -LOVE- to have the rating floor on my max 2339 rating.
Play to the hilt, never drop below 2200, get lifetime master tating,
er, rating, all that kind of gravy.
(Currently I'm wallowing in 2169 mires.)
Kenneth Sloan
>In article <1994Jul14.0...@cis.uab.edu>,
>Kenneth Sloan <sl...@cis.uab.edu> wrote:
>>In article <1994Jul13....@hal.depaul.edu> pst...@falcon.depaul.edu (Peter Stein) writes:
>>>...
>>>Better deflation than inflation,
>>
>>Now this truly puzzles me. Why is either preferable to the other?
>...
>If ratings are inflated to begin with then a deflationary policy would
>move them
>in the proper direction. Such a deflationary policy (a relative qualification)
>may simply consists of removing the inflationary components of the current
>implementation such as floors. There doesn't have to be an ultimate policy, as
>I suggested previously if the rating pool is continuously monitored different
>policies can be put into effect when it is determined that ratings have either
>become too inflated or deflated.
>
Excellent points. But the question was: why is deflation better than
inflation?
As near as I can make out, your answer is that "ratings are inflated to
begin with". Which leads us to: why do you believe that ratings are
inflated?
>
>>>... but it probably will not happen anyway.
>>>For openers so long as floors are in effect there exists a potent inflationary
>>>source.
>>
>>How potent is it?
>
>Well if you persist in reductio ad absurdum then my answer will be given
>accordingly: Very! In one of my prior postings I gave some sample calculations
>for situations where a "floor" player who is now overrated to the tune of 200
>rating points is fueling inflation. This effect can be quantified.
But, the question is: how many players are overrated "to the tune of 200
rating points?" I am, in fact, trying hard to quantify the effect.
When you said that the effect was "potent" I thought that you had some
data. My mistake.
Here's the conundrum: If floors are a potent inflationary force, they
will raise the entire pool, and players will float off their floors.
But...if players float off their floors, then the inflation will stop.
The historical evidence is that more and more players are on their
floors (see data, previously posted). Are we seeing a transition from
"no floors" to a steady state? Or, are we seeing a large-scale
deflation which is dragging everyone down?
But, please remember: I'm committed to voting to eliminate all floors.
I think they are poorly motivated, and wrong *in principle*. There are
better methods for dealing with inflation/deflation.
However - that's a separate issues from trying to find out precisely
what their *effect* has been. Clearly, their only effect can be
inflationary. How "potent" this effect is depends on questions such as
how many people are actually *on* their floors, how overrated they are,
and how many games they play.
Which is again separate from observing the number of people *on* their
floors and trying to figure out what that tells us about the behavior of
the system as a whole. Again - the paradox is that the more people you
find on their floors, the more likely it is (not for certain...just more
likely) that the floors are having a good *effect* (notwithstanding
their rotten justifications).
Peter Stein
>>...
>>Better deflation than inflation,
>
>Now this truly puzzles me. Why is either preferable to the other?
First let me ask whether you're having problems with your news reader or are you
randomly sampling articles in a discussion thread? Many of your questions in this
post have already been answered in my prior posts. But I will attend to some of
these issues again.
If ratings are inflated to begin with then a deflationary policy would move them
in the proper direction. Such a deflationary policy (a relative qualification)
may simply consists of removing the inflationary components of the current
implementation such as floors. There doesn't have to be an ultimate policy, as
I suggested previously if the rating pool is continuously monitored different
policies can be put into effect when it is determined that ratings have either
become too inflated or deflated.
Ratings should also present some measure of chess competency. In some of your
previous posts you keep zeroing in on the concept that ratings are relative
for a player and his/her peers. Well, yes and no. Ratings reflect both consistency
and a grasp of the game. It isn't merely enough to have a rating that is correct
relative to your peers. If you and your peers have ratings that represent no
continuity to ratings in the past then what good are these ratings?
>>... but it probably will not happen anyway.
>>For openers so long as floors are in effect there exists a potent inflationary
>>source.
>
>How potent is it?
Well if you persist in reductio ad absurdum then my answer will be given
accordingly: Very! In one of my prior postings I gave some sample calculations
for situations where a "floor" player who is now overrated to the tune of 200
rating points is fueling inflation. This effect can be quantified. What about
deflationary components in the implementation? Can their effect be quantified?
Given a situation where we have known quantifyable inflationary contributors
and very unclear deflationary contributors it seems we ought to focus on the
former. Now if you can give examples of events for which the deflationary effect
can be quantified in a likewise manner and also show that the likelihood of these
events is equal to or greater than their inflationary opposites that would be very
illuminating. Perhaps there is cancellation at work in the present system, but
if we count on it and the amplitudes can get high then I would suggest we have a
volatile system prone to many problems.
>
>>... The number of players hitting their floor is increasing
>
>It is? Please send me the data. I *was* going to do that analysis
You have already done this yourself. In one of your prior posts. My data is
obtained from my observations and those of others. No, I have not undertaken
a systematic study, but that in no way diminishes the significance of the
data.
>tonight, but if you've already done it I can save my time.
>
>>Another important point is that we're doing a miserable job of retaining
>>juniors
>
>Sadly, this is true. Retaining juiors (and especially retaining girls)
>is a significant challenge.
>
>>... That being
>>the case it is sadly true that they will not be contributing to the deflation.
>>
>
>On the contrary. In Illinois, and in Rochester, NY, we see significant
>evidence of the "career effect" on an accellerated scale.
Er, could you please check my post again? What did I say? If they dont play
they can't contribute to either inflation or deflation, can they?
Peter Stein
pst...@falcon.depaul.edu
Mark E. Glickman
>they can't contribute to either inflation or deflation, can they?
This may or may not be relevant to Peter's post, but
I just wanted to mention that inflation (or deflation) is not synonymous
to increase in the avg rating (or decrease). Inflation can occur
if, for example, the majority of players in the rating pool were
getting worse rather than better: because ratings are (roughly)
conserved, the overall avg rating would remain about the same but
level of player ability would be lower, hence ratings would be inflated
relative to where they used to be. The point here is that inflation or
deflation can occur irrespective of player activity.
- Mark
Peter Stein
Kenneth Sloan <sl...@cis.uab.edu> wrote:
>In article <1994Jul14.1...@hal.depaul.edu> pst...@falcon.depaul.edu (Peter Stein) writes:
>>In article <1994Jul14.0...@cis.uab.edu>,
>>Kenneth Sloan <sl...@cis.uab.edu> wrote:
>>>In article <1994Jul13....@hal.depaul.edu> pst...@falcon.depaul.edu (Peter Stein) writes:
>>>>...
>>>>Better deflation than inflation,
>>>
>>>Now this truly puzzles me. Why is either preferable to the other?
>
>>...
>>If ratings are inflated to begin with then a deflationary policy would
>>move them
>>in the proper direction. Such a deflationary policy (a relative qualification)
>>may simply consists of removing the inflationary components of the current
>>implementation such as floors. There doesn't have to be an ultimate policy, as
>>I suggested previously if the rating pool is continuously monitored different
>>policies can be put into effect when it is determined that ratings have either
>>become too inflated or deflated.
>>
>
>Excellent points. But the question was: why is deflation better than
>inflation?
Once again you've chosen not to read one of my postings in its entirety.
Please do so. My discussion of the significance of ratings makes it reasonably
clear.
>As near as I can make out, your answer is that "ratings are inflated to
>begin with". Which leads us to: why do you believe that ratings are
>inflated?
Same answer as above.
>>>>... but it probably will not happen anyway.
>>>>For openers so long as floors are in effect there exists a potent inflationary
>>>>source.
>>>
>>>How potent is it?
>>
>>Well if you persist in reductio ad absurdum then my answer will be given
>>accordingly: Very! In one of my prior postings I gave some sample calculations
>>for situations where a "floor" player who is now overrated to the tune of 200
>>rating points is fueling inflation. This effect can be quantified.
>
>But, the question is: how many players are overrated "to the tune of 200
>rating points?" I am, in fact, trying hard to quantify the effect.
If one is, then that's one too many. I used the value 200 to exaggerate for
effect, but it may be representative of reality. But if everyone in the pool
is already overrated to begin with then the contribution of a floor player
overrated by only 20 points is too much of a contribution.
>When you said that the effect was "potent" I thought that you had some
>data. My mistake.
Does something have to hit you squarely between the eyes to take notice? Do
you find fault with my contention that the floor effect can be quantified?
Can the same be said for deflationary components what if any exist? I thought
it possible to exchange ideas without gnat's assing irrelevant details. My mistake.
>Here's the conundrum: If floors are a potent inflationary force, they
>will raise the entire pool, and players will float off their floors.
How? If they continue to have negative results, how? Are you suggesting that
no matter what their results eventually everyone they'd play would have a rating
of the present day Karpov and therefore they will not lose rating points? Silly
and impossible.
>But...if players float off their floors, then the inflation will stop.
>The historical evidence is that more and more players are on their
>floors (see data, previously posted). Are we seeing a transition from
>"no floors" to a steady state? Or, are we seeing a large-scale
>deflation which is dragging everyone down?
>
>But, please remember: I'm committed to voting to eliminate all floors.
>I think they are poorly motivated, and wrong *in principle*. There are
>better methods for dealing with inflation/deflation.
I'm all ears.
>However - that's a separate issues from trying to find out precisely
>what their *effect* has been. Clearly, their only effect can be
>inflationary. How "potent" this effect is depends on questions such as
>how many people are actually *on* their floors, how overrated they are,
>and how many games they play.
The very fact that this effect can be easily quantified means it's potent.
Now, don't resort to the games being played with Mr. Newell. I will not
go through the calculations. There is no need to. I will leave it to you.
Even if after counting beans we don't agree on the present magnitude of
potent we know a harmful component exists in the present system. That's
reason enough to eliminate it.
>Which is again separate from observing the number of people *on* their
>floors and trying to figure out what that tells us about the behavior of
>the system as a whole. Again - the paradox is that the more people you
>find on their floors, the more likely it is (not for certain...just more
>likely) that the floors are having a good *effect* (notwithstanding
>their rotten justifications).
How having a source of rating points that should never be created could
have "a good *effect*" is beyond me. Please explain. Is there cancellation
at work?
>Kenneth Sloan Computer and Information Sciences
>sl...@cis.uab.edu University of Alabama at Birmingham
>(205) 934-2213 115A Campbell Hall, UAB Station
>(205) 934-5473 FAX Birmingham, AL 35294-1170
Peter Stein
pst...@falcon.depaul.edu
W C Newell Jr
>[...] I just wanted to mention that inflation (or deflation) is not synonymous
>to increase in the avg rating (or decrease). Inflation can occur
>if, for example, the majority of players in the rating pool were
>getting worse rather than better: because ratings are (roughly)
>conserved, the overall avg rating would remain about the same but
>level of player ability would be lower, hence ratings would be inflated
>relative to where they used to be. The point here is that inflation or
>deflation can occur irrespective of player activity.
Ratings are definitely >NOT< conserved. The rating system is a stochastic
process. People enter the pool at level A, interact and exchange ratings
points with an arbitrary number of other people, and then leave the pool at
level B. Sooner or later >everyone< leaves the pool, insofar as death is
inevitable. If A>B, then incremental inflation has occured. If B>A, then
incremental deflation has occured. Net inflation/deflation can only be
measured as an aggregate of these stochastic interactions, plus any additional
points being added to the pool due to a bonus mechanism or to rating floors.
Assuming for the sake of argument that you can see inside a player's mind and
guage his/her strength absolutely, you could say that if that player's strength
has fallen since the initial rating, then it >predicts< an incremental
inflationary effect. The effect doesn't occur, however, until the player
actually scores a negative result, gives up the points, and then leaves the
pool. The only way inflation/deflation can occur "irrespective of player
activity" is if everyone in the world stopped playing at once, and by some
process (aliens?) they were all mentally augmented to simultaneously show
either a mass gain or a mass loss of chess-playing ability. I don't think
that's very likely.
I know I'm going to get flamed for this (again), Mark, but the more you make
statements like the above, the more I become concerned that you do not
understand the Elo model. And yet you're the chair of the ratings committee...
--
Bill Newell
Mike Westbrook
-}
-}There seems to be an assumption that a player's ability will only advance
-}during their lifetime,
Well, it sure won't advance at any other time!!!
> Mike < Jester at the Court of Final Appeal
"Yes, I know what he meant!"
Israel Silverman
KS>Place your bets here - who will get the "Certified Expert" title first?
>Me? or Peter?
What is a "Certified Expert"?
Sincerely,
Israel Silverman
Peak rating: 2172, current 2044+0000
P.S. What does "Titles: x0/Q=0" mean?
* 1st 2.00b #1780s * You've got to have an ace in the hole.
Kenneth Sloan
>
>KS>Place your bets here - who will get the "Certified Expert" title first?
>>Me? or Peter?
>
>
>What is a "Certified Expert"?
You are - I see that USCF lists you as 2044*x0
"Certified Expert" is a USCF title. For details on how you achieved
this grand honor, you will, of course, have to read Chess Life. No USCF
member can be well-informed without it. You can (with some effort)
renounce this title - but no one can take it away from you - it's yours
for life.
RTFCL!
>P.S. What does "Titles: x0/Q=0" mean?
It means that you:
*have the "Certified Expert" title in OTB play (the 'x')
*have 0 (nil, zilch, nada, bupkis) points towards the
"Advanced Expert" title. (the first '0')
*has no title in Quick Chess (in fact, I see that you have
had the good taste to not play in QuickChess competition)
(the "Q=0", HMMM - did you leave out a ' ' somewhere?)
This tells me that (since the title system began) your peak performance
was in the neighborhood of 2100, and you have performed at that level
many times (at least 5). Your current rating of 2044 indicates that you
have mixed these peak performances with several 2000 (or lower)
performances. I would expect you to be a "solid 2050 player". Compared
to other 2050 players, I would guess that you are less of a threat to
beat a 2200+ player, and have less chance of losing to an 1800 player.
How am I doing, so far?
Cavemanchs
writes:
>Most of the Class A and Expert players that are regulars at Pittsburgh
>events have seen variations of more than 100 points in their ratings. I
can
>think of Castillo, Ambrad, Kownacki, Schreiber, Renk, Lakatos, Geller,
Magar,
>Nedved, and Martinak. (The last three are Masters.) In some cases there
have
>been fluctuations of 200 points (Castillo). These fluctuations do not
arise
>from sandbagging.
>Many of these players have been hitting their floors. While the
percentage
>of people in the rating list that are hitting their floors may be only
4%,
>I estimate that those 4% account for a disproportionate share of all
tournament
>participants. The above named Pittsburgh players are "veterans" and
account
>for the lion's share of our Tuesday night tournaments as well as quite a
lot
>of weekend tournament and League games.
By raising this point you made me realize that this may also be a
"rolling" 4%, so that
while only 4% at any one time might be in this situation, maybe 8%-12% are
chronically
in these circumsatnces...?
Cavemanchs
(Kenneth Sloan) writes:
>In article <2vns39$1...@search01.news.aol.com> cavem...@aol.com
>(Cavemanchs) writes:
>>I am a USCF master, Kevin Bachler, currently rated about 2220.
>2220*X7, to be precise.
>>... I am
>>commonly known as "CAVEMAN." Most of my friends and I strongly dislike
>>ratings floors.
>A reasonable opinion.
>>... We feel they are artificial.
>Yup
>>... They discourage several of
>>my friends from playing, because they have bottomed out and cannot reach
a
>>competitive level. They say that they have no motivation, because they
>>know their rating can't go down.
>No sympathy here. According to my reading, this translates to: "we'd
>like to sandbag, but can't".
Actually, the situation is that I have many chess friends and
acquaintances
in classes C, B, and A who peaked several years ago at strengths of 200 to
300 points above their current strengths. They find now that they get
beat
regularly by players rated substantially below themselves. They would
like
to improve but find the current situation VERY demoralizing. Often, under
stress they make very poor moves. Some of these friends are very good
correspondence players who find that they are extremely unsuccessful OTB.
NONE OF THEM WISH TO SANDBAG. They would like to drop to a point
where they could play competitively so that they could improve.
> Let's be realistic here - if you are
>sitting on a floor, then your rating is at least 101 points below your
>(recent) peak. Are you seriously claiming that a player who has peaked
>at 2301+ really wants to play in the U2200 section?
No, I am saying there are CLASS players who are in this situation so that
they can regain a competitive spirit. I can understand that.
> Much as I
>personally dislike the 100-point floors, I can't summon any sympathy for
>such a losing attitude.
Sympathy is a non sequitur. These people don't want sympathy. They want
the opportunity to compete on their own competitive level.
>And - *you* certainly have not "bottomed out". By my estimation, it's
>virtually certain that *your* floor is 2100.
Please read more carefully. I never claimed to have bottomed out. I
stated that I had freiends who had bottomed out and then stated the
feelings that these friends had expressed to me. This is another non
sequitur.
>Can't your "friends" speak
>for themselves? Do they exist? Names, please.
Yes, they can. I am not certain they want to. It is hard to admit this
situation
publically because it is admitting failure. I will say that one such
friend is
a prominent NTD. I do not know if these people have access to INTERNET.
If so, they can speak up for themselves if they choose to do so. Next
time
I see them, I will ask if I mind if I use their names.
>>
>>We much preferred bonus points as a way of combatting deflation. Bonus
>>points reward good results. Rating floors reward bad results. No one
>>wants to be rewarded for poor chess.
>I maintain that no one is "rewarded" by the 100-point floors. After
>all, you still need a 2200+ performance to get your rating off a 2200
>floor. Much as I dislike using words like "reward" or "punishment" when
>talking about ratings, the worst that can be said (in these terms) about
>100-point floors is that they fail to punish poor chess. That's not
>quite the same thing as *rewarding* poor chess.
Logically, I agree with you. Psychologically it isn't necessarily so.
People
relate stories to me about how they were at their floor and tried crazy
ideas
that they knew were of questionable soundness. They described the
of losing the ability to care because they knew that their rating would
be unaffected if they lost.
>>
>>This statement is an oversimplification,
>I'll agree with that.
>>... but it makes the point.
>nope. Try again - what precisely *is* your point?
I think the point is clear.
>>... I find
>>this opinion to be prevalent among chessplayers,
>By way of contrast, I find that most chessplayers don't really care,
>because the vast majority of them are well above their floors.
Could be, but a lack of caring about this issue isn't within my
anecdotal experience.
>>... yet the ratings
>>committee/USCF seems to think otherwise.
>I can state with complete confidence that the vast majority of the USCF
>Ratings Committee thinks that the 100-point floors ought to be
>abolished. Where did you get the idea that anyone on the USCF Ratings
>committee supports these floors? Have you read the current
>recommendations from that committee, which call for their abolishment?
>May I suggest that if you really want the 100-point floors abolished
>that you refrain from badmouthing those who agree with you? Quite
>frankly, part of the reason that we are having so much trouble
>abolishing the floors is that we spend so much time distancing ourselves
>from comments such as yours.
Whoa! I got the idea from the results, from the ratings system that is in
force!
Yes, I have read the current recommendation. Now, I admit, I haven't read
these recommendations historically, and I probably made a rash statement
by basing my conclusions on what the USCF actually prcatices. Please note
the phrase "...seems to think otherwise." And who am I badmouthing? I
said
that the ratings committee seems to think differently than a prevalent
feeling among
chessplayers. This statement may be rash and based on too narrow of a
selection
of anecdotal evidence, but I don't see where its insulting.
>>... I don't know why.
>>
>>Most of us with any math background who have read ELO feel that the USCF
>>has severely bastardized an otherwise pretty good system.
>a) what's your "math background"
1. B.A. Natural Science and Mathematics 1987 from Shimer College, with
Great Distinction
2. 13 years actuarial experience
3. Developer of several financial methods for funding nonqualified benefit
plans
4. Speaker at several different industry groups on funding nonqualified
executive benefits.
5. Published in benefits and compensation journals.
6. Have taught at Shimer College and Elmhurst College
7. Vice President of Consulting and of Technical Services at a major
national benefits consulting firm
8. Currently finishing a masters in mathematics.
etc....
>b) please be more concrete. Besides the 100-point floors, what are your
>specific examples?
Adjusting the k-factor based on quick time limits comes to mind.
Cavemanchs
Powell) writes:
>Kenneth Sloan (sl...@cis.uab.edu) wrote:
>: No sympathy here. According to my reading, this translates to: "we'd
>: like to sandbag, but can't". Let's be realistic here - if you are
>: sitting on a floor, then your rating is at least 101 points below your
>: (recent) peak. Are you seriously claiming that a player who has peaked
>: at 2301+ really wants to play in the U2200 section? Much as I
>: personally dislike the 100-point floors, I can't summon any sympathy
for
>: such a losing attitude.
>Very true... but why not mention me by name. :)
>My rating max'd at around 2310 and I played well enough to find my floor
of
>2200 (for awhile at least). The truth is if my rating dropped below 2200
>I proably would have considered entering the World Open U2200 section. I
>can't disagree that the floors should be abolished but .... I am glad
>they were there! To think of playing in the U2200 sections makes me
sick!
>What stupid things we consider doing for a few extra bucks....
>I'd hate myself in the morning...give any winning to charity ....
>For the record... I have never played in the US Amateur Individual.
>We reward lousy chess by giving awards for it not by allowing rating
>floors.....
>If the Caveman really thinks players enjoy being at their floor, then I
>understand the origin of his name.
Actually, I think my point was that people don't like being at their
floors.
The CLASS players (I probably should have included this point earlier)
of whom I speak would rather drop to a competitive level, in the hope
of rising again, rather than languishing at their floor.
The knickname CAVEMAN was started by Al Chow. A good example
is Informant 53, G316, or Informant 36, combination section. I've
also had some caveman chess games in Chess Life, Fernschach,
The New Chessplayer, and various postal and regional magazines.
They are typically representative of the adage: CAVEMAN chess
is chess without finesse.
R,
Paul Powell
Kenneth Sloan
>
>Actually, the situation is that I have many chess friends and
>acquaintances
>in classes C, B, and A who peaked several years ago at strengths of 200 to
>300 points above their current strengths. They find now that they get
>beat
>regularly by players rated substantially below themselves. They would
>like
>to improve but find the current situation VERY demoralizing. Often, under
>stress they make very poor moves. Some of these friends are very good
>correspondence players who find that they are extremely unsuccessful OTB.
>NONE OF THEM WISH TO SANDBAG. They would like to drop to a point
>where they could play competitively so that they could improve.
All they need to do is write a letter documenting their recent poor
results, and petition to have their floors lowered. It can be done -
but they have to ask.
I'm still trying to figure out how lowering your rating is a
prerequisite for improving your play, but let's leave that alone for
now.
>
>Adjusting the k-factor based on quick time limits comes to mind.
>
I don't understand. Right now there is one rating for "slow" OTB play,
and a completely different rating for "Quick" chess. The whole point of
having a separate rating is so that play at "quick time limits" does not
affect your "Regular" OTB rating.
There *are* still 1/2 K tournaments - but these are fading away for lack
of interest, I think. I just took a quick scan of the TLA's in Chess
Life (what would I do without my trusty Chess Life?), and I didn't spot
a single 1/2-K event (but...I didn't look very hard - there may well be
one lurking there.)
In any event, uniformly changing the K for an event does not do much
damage to the ratings mathematics. It changes the rate of convergence -
but does not in any way change the rating that an individual will
converge to.
Roughly speaking, changing the "K" for both players in a game simply
changes the weight of that game versus the previous histories of both
players. There are many reasons why one game ought to carry less weight
than another. One of these is that we expect the variability of the
players' performance to be different than usual. If we have reason to
believe that a particular game is likely to produce very variable
results (compared to the players' "true" strengths), then lowering the K
factor has a perfectly sound mathematical basis.
With "1/2-K" events, we may have a chicken-and-egg situation, where the
players pay less attention to the games because they know it's a 1/2-K
event, which in turn justifies using a lower K, which in turn....
At least, that's the way it looks to me. How do *you* think that
"adjusting the k-factor based on quick time limits" has harmed the USCF
rating system? Please explain.
Cavemanchs
Sloan) writes:
>How do *you* think that "adjusting the k-factor based on quick time
limits" has >harmed the USCF rating system? Please explain.
In a couple of ways. One is the attitude that seems to have established
itself in the past that the rating system is something to be tweaked
by the Policy Board, delegates, etc., rather than existing as an
"objective" measurement system. This is probably the more important
reason. Not too long ago (although it seems to be a little less now) I
had the feeling that ratings regulations were changing annually. This
creates the impression that the system is poor, and unworkable, or that
those controlling the system don't understand it. Although this seems to
be less the case today, my impression is not so old as to be forgotten...
Why should there be a separate OTB rating system for events at faster
controls? Both players play at the faster rate.
The K factor should probably be adjusted only to reflect that players at
various
strengths or number of games played (perhaps in a recent time period)
may have a more stable rating, and so should have a lower K. I've
heard this arguement proposed before, and believe it was one of the
arguments
for a different K at 2100 and at 2300. Yet, I haven't seen a lot of data
to
support the approach.
Again, a big part of my concern is one of attitude. The rating system to
me
seems to be something that should be tweaked only if the mean, median and
mode are drifting from 1500, or if there are other signs that at specific
levels there is a tendency toward inflation or deflation. Instead, the
system is tweaked (ratings floors for example) with the argument advanced
that players don't like to have their ratings go down, (in addition to a
way to combat deflation) and that players will play more if this is the
case. (At least, this is the argument that I think has been set fourth.)
The original point I was trying to make is that for some, Many!?, players,
the reverse psycholgical effect can occur. Since players are unable to
have their ratings drop to their true level, they continuously have
exaggerated poor results. This is, as described to me by players in this
situation, exceptionally demoralizing. These players feel that they would
be more likely to play if their ratings could drop freely to the
appropriate level, so that they could have competitive positive results,
thus improving their attitudes. The argument is based primarily on
psychological factors, not mathematical ones.
Yes, these players could petition to have their floor decreased. Some,
including the NTD of which I spoke, have considered doing so. However,
this means that a player must actively undertake an act the purpose of
which is to prove that they are a bad player. This is no small task for
any typically egocentric chessplayer. These players find it hard to take
this step.
I realize I have probably rambled a bit with this response. I am writing
this too late in the day, and ask that you forgive the somewhat rambling
nature....
Kenneth Sloan
>
>In article <1994Jul18....@cis.uab.edu>, sl...@cis.uab.edu (Kenneth
>Sloan) writes:
>
>>How do *you* think that "adjusting the k-factor based on quick time
>>limits" has harmed the USCF rating system? Please explain.
>
>In a couple of ways. One is the attitude that seems to have established
>itself in the past that the rating system is something to be tweaked
>by the Policy Board, delegates, etc., rather than existing as an
>"objective" measurement system. This is probably the more important
>reason. Not too long ago (although it seems to be a little less now) I
>had the feeling that ratings regulations were changing annually. This
>creates the impression that the system is poor, and unworkable, or that
>those controlling the system don't understand it. Although this seems to
>be less the case today, my impression is not so old as to be forgotten...
First - I agree with this.
But - my point was that I don't know of an instance of "adjusting the
k-factor based on quick time limits". In other words - I don't remember
this actually occurring. If it didn't occur, then this is not an issue.
There are enough *actual* abominations in the recent history of the
rating system, without worrying about phantoms.
On the general point of "tweaking the system" - I agree completely.
>
>Why should there be a separate OTB rating system for events at faster
>controls? Both players play at the faster rate.
Three reasons. One political, two technical.
Political: the choice was between a separate rating system for Quick
Chess, or no rated fast play.
Technical 1: there is strong opinion (about which reasonable men may
differ) that *very* fast play is a different game than *very* slow play.
Most (but not all) will agree that 5-minute blitz is "different" than
20/1. The problem is: where to draw the line. The current USCF
position (no rated play faster than G/10, one rating for G/10 - G/25,
another for G/30 - ...) is a compromise.
Technical 2: Please refer to Elo's discussion of the choice of K (not
because it is necessarily technically perfect - but just to set
context). The choice of K is primarily motivated by the desire to
reconcile the latest *performance* with the new *rating*. It is helpful
to think about:
*the variability in the measurement of the performance
*the variability of performances, givena particular rating
If you believe (for whatever reason) that the current performance is
likely to have a very wide spread - then you can justify a smaller K.
If you believe (for whatever reason) that the pre-event rating is very
uncertain - then you can justify a smaller K.
Reverse either of these, and you can justify a larger K.
So - the rough argument for "fast play -> small K" is: we believe that
perfomances at fast play exhibit a larger variability than performances
at slow play. This justifies a smaller K.
>...
>Again, a big part of my concern is one of attitude. The rating system to
>me
>seems to be something that should be tweaked only if the mean, median and
>mode are drifting from 1500,
Question: Why should any of the mean, median, or mode be fixed at 1500?
For reference: looking only at ESTABLISHED ratings, the current values
are:
mean: 1495
median: ~1430
mode: 1600-1699
I can find nothing in Elo which treats "1500" as a magic value. I have
seen many people make the claim that the distribution of all USCF
ratings 'ought to be' a Normal distribution with the mean at 1500 - but
I can't find any theoretical reason for that to be true. Looking at the
actual distribution, I doubt that it is Normal.
Finally - the distribution certainly has another interesting statistic -
the variance.
Question: if you believe that the system should be tweaked to maintain a
mean of 1500 - what value should be enforced for the variance?
It can be difficult to tease apart the social effects of the pool
changing it's nature from the mathematical effects of the pool drifting
with respect to some arbitrary landmark. It makes no sense to specify
any particular parameters (mean, variance, ...) for the distribution in
the presence of so much self-selection of the population.
I happen to think that it is possible to identify a particular
sub-population which we can (by disputation, but not by mathematics)
convince ourselves is a reasonable standard - but I do not believe that
eh entire rated pool is sufficiently stable to serve this purpose.
>... or if there are other signs that at specific
>levels there is a tendency toward inflation or deflation. Instead, the
>system is tweaked (ratings floors for example) with the argument advanced
>that players don't like to have their ratings go down, (in addition to a
>way to combat deflation) and that players will play more if this is the
>case. (At least, this is the argument that I think has been set fourth.)
Goichberg publicly espouses the claim that floors (which he invented, in
their current form) are meant to prevent players from being discouraged
when their ratings fall. I agree with you that this is foolish.
Other people claim that Goichberg's true motive was to prevent
sandbagging. I tend to agree with this point of view - but I have no
special access to the inside of Bill's head. In my opinion, if this is
the goal, floors are a silly way to achieve it. They are too visible,
too controversial, and ineffective to boot.
I have pointed out that one practical *effect* (not a justification -
just a provable effect) of floors is a self-regulating hedge against
deflation. {briefly - if the pool is already inflating, then floors
have no effect; if the pool is already deflating, then floors tend to
counteract it}. But - I have never claimed that this is the *purpose*
of the floors, and I would never *defend* the floors on this basis.
Bottom line (I think):
* we agree that 100-point floors should be abolished.
* we disagree on how to measure drift in ratings wrt chess skill
* we disagree on the effect of tweaking K.
Have I got *that* right?
Paul A. Lane
>I'm still trying to figure out how lowering your rating is a
>prerequisite for improving your play, but let's leave that alone for
>now.
If you've something for which to play, it increases the incentive.
Also, I'm sure it's not fun being upset bait.
Paul
--
Cavemanchs
Sloan) writes:
In article cavem...@aol.com (Cavemanchs) writes:
>>The rating system to me seems to be something that should be
>>tweaked only if the mean, median and mode are drifting from 1500.
>Question: Why should any of the mean, median, or mode be fixed at 1500?
>For reference: looking only at ESTABLISHED ratings, the current values
>are:
> mean: 1495
> median: ~1430
> mode: 1600-1699
>I can find nothing in Elo which treats "1500" as a magic value. I have
>seen many people make the claim that the distribution of all USCF
>ratings 'ought to be' a Normal distribution with the mean at 1500 - but
>I can't find any theoretical reason for that to be true. Looking at the
>actual distribution, I doubt that it is Normal.
>Finally - the distribution certainly has another interesting statistic -
>the variance.
Elo believed that performances (and so ratings) were normally distributed.
See "The Rating of Chessplayers Past and Present", 1.31, as well as many
other references.
I expected to find the mean/median/mode response there as well, but
didn't. Elo states only that the initial rating Ri must be large enough
to "prevent" negative ratings.
I do recall reading an article several years ago that stated that the
mean, median, and mode were supposed to be 1500 (as implemented) and that
the variance was supposed to be under 50 points (although as I receall at
the time of the article it was 56.) I think this was in Chess Life. Am I
recalling the article incorrectly? Can anyone quote the issue this was
in?
Your point on the variance is of course good and valid.
I am not sure that we disagree on the selection of K in general. I would
like to see statistics used to validate the selection of K, however.
Kenneth Sloan
E...>
>Elo believed that performances (and so ratings) were normally distributed.
> See "The Rating of Chessplayers Past and Present", 1.31, as well as many
>other references.
"7.31 Unlike the distribution of player performances, the distribution of
player proficiencies is not described by the normal probability
function" The Rating of Chess Players Past & Present, 2nd
edition, p. 112.
"7.35 The chi-square test,...was applied... The test very definitely
rejects the normal distribution hypthesis." ibid, p. 114
"7.36 Nonetheless, it is a distinct possibility that the distribution of
proficiencies of the entire chessplaying population is normal. The
rating system is applied under controlled conditions only to a
special portion of the entire population...
... Therefore the distribution proficien-
cies of established players cannot in all likelihood be expected to
be normal." p. 114
"1.31 ...The many performances of an individual will be normally
distributed." p. 7
According to my reading, Elo believed that individual performances were
likely to be normally distributed, but that the distribution of observed
ratings were, in fact, *not* normally distributed. Have I misread the
above? If so, please point out my error.
>I expected to find the mean/median/mode response there as well, but
>didn't.
Because it's not there. Consider:
"3.36 The Elo system determines only differences in the ratings, or the
relative ratings, of the individuals in the pool. These relative
ratings may be expressed on any arbitrary scale with any divisions,
...", p. 44
"8.21 The Rating Scale
An interval scale was selected for the Elo system,...
... The sub-division C = 200E, the scale
midpoint at 2000, and the use of four-digit numberw were adapted from
usage and are entirely arbitrary." p. 136
"1.26 The rating scale itself -- its range of numbers -- is, likd any scale
without reproducible fixed points, necessarily an open-ended
floating scale.... the present range originally took 2000 as the
upper level for the strong amateur of club player and arrange the
other categories above and below...", p. 6
There is another passage which speaks of supplying enough "ballast" to
prevent ratings from going negative - but I can't find it right now.
Christopher Yep
Cavemanchs <cavem...@aol.com> wrote:
|>In article <1994Jul19....@cis.uab.edu>, sl...@cis.uab.edu (Kenneth
|>Sloan) writes:
|>
|>In article cavem...@aol.com (Cavemanchs) writes:
|>
|> >>The rating system to me seems to be something that should be
|>>>tweaked only if the mean, median and mode are drifting from 1500.
|>
|>>Question: Why should any of the mean, median, or mode be fixed at 1500?
|>
|>>For reference: looking only at ESTABLISHED ratings, the current values
|>>are:
|>
|>> mean: 1495
|>> median: ~1430
|>> mode: 1600-1699
|>
|>>I can find nothing in Elo which treats "1500" as a magic value. I have
|>>seen many people make the claim that the distribution of all USCF
|>>ratings 'ought to be' a Normal distribution with the mean at 1500 - but
|>>I can't find any theoretical reason for that to be true. Looking at the
|>>actual distribution, I doubt that it is Normal.
|>
|>>Finally - the distribution certainly has another interesting statistic -
|>>the variance.
|>
|>Elo believed that performances (and so ratings) were normally distributed.
^^^^^^^^^^^^^^^^^^^^^
A *particular* (i.e. specific, given) player's performances may be normally
distributed. However, if we make this same assumption for everyone in
the rating pool, this does NOT imply that the ratings distribution (of
all players) should necessarily also be normally distributed.
|> See "The Rating of Chessplayers Past and Present", 1.31, as well as many
|>other references.
I don't think that Elo ever claimed that the rating distribution was
normal.
Something to think about: currently there are quite a lot of low rated
scholastic players. Think about how this skews the rating distribution.
(I am almost positive that one can see, just by a quick glance at a
histogram of the ratings, that they are far from being normally distributed.)
|>
|>I expected to find the mean/median/mode response there as well, but
|>didn't. Elo states only that the initial rating Ri must be large enough
|>to "prevent" negative ratings.
|>
|>I do recall reading an article several years ago that stated that the
|>mean, median, and mode were supposed to be 1500 (as implemented) and that
|>the variance was supposed to be under 50 points (although as I receall at
|>the time of the article it was 56.) I think this was in Chess Life. Am I
|>recalling the article incorrectly? Can anyone quote the issue this was
|>in?
Do you know what variance is? I would estimate the current variance of
USCF ratings is on the order of 200,000, or about 4000 * 50.
USCF initially set up its rating system so that all 3 of
{mean,median,mode} were supposed to be 1500?? This seems hard to
believe. Can you try to find the article?
|>
|>Your point on the variance is of course good and valid.
|>
|>I am not sure that we disagree on the selection of K in general. I would
|>like to see statistics used to validate the selection of K, however.
Despite arguments to the contrary, the choice of K has very little to do with
the general accuracy of the rating system, up to a point, as long as
reasonable K's are used. Other factors, which have been brought up in
previous articles, affect the rating system's accuracy much more than
the choice of K does. What's "reasonable,"? Well, K=32 is about as
high a value that I would suggest using. I would suggest using a lower
value of K, even for the general pool, but that is another issue. There
is no real difference in terms of accuracy, between K=3 and K=10 for
instance.
Mark E. Glickman
Caveman first writes:
>|>I am not sure that we disagree on the selection of K in general. I would
>|>like to see statistics used to validate the selection of K, however.
To which Christopher Yep responds:
>Despite arguments to the contrary, the choice of K has very little to do with
>the general accuracy of the rating system, up to a point, as long as
>reasonable K's are used. Other factors, which have been brought up in
>previous articles, affect the rating system's accuracy much more than
>the choice of K does. What's "reasonable,"? Well, K=32 is about as
>high a value that I would suggest using. I would suggest using a lower
>value of K, even for the general pool, but that is another issue. There
>is no real difference in terms of accuracy, between K=3 and K=10 for
>instance.
Actually, I believe the choice of K can improve the accuracy of the
system. It is basically a measure of how much weight the performance
from a new event should be given relative to a pre-event rating.
Large values of K indicate the belief that a performance should
have a large impact on one's rating, whereas a small value of K
corresponds to a performance having a small impact on one's rating.
A situation that K ought to be large is when a player hasn't competed in
a long time. The reason I believe that a choice of K improves accuracy
is that players' abilities do change over time, and K essentially tracks
the magnitude of these changes.
The USCF Ratings Committee is proposing a major change in the
provisional rating system whereby updates are performed using the
established rating formula with a value of K that depends on the number
of games played (it starts out large, and slowly decreases to 32).
By the way, I don't find Elo's derivation of the choice of K very
compelling. For example, his derivation involves computing ratios of
sums of standard deviations (a no-no in statistics), and he also
isn't careful on how he incorporates the number of games in an event
(it can be shown that K should be roughly proportional to 1/n, where
n is the number of games in an event being updated).
- Mark
Steven Rix
In article <1994Jul21.1...@Princeton.EDU>, glic...@phoenix.Princeton.EDU (Mark E. Glickman) writes:
->
->Actually, I believe the choice of K can improve the accuracy of the
->system. It is basically a measure of how much weight the performance
->from a new event should be given relative to a pre-event rating.
->
->The USCF Ratings Committee is proposing a major change in the
->provisional rating system whereby updates are performed using the
->established rating formula with a value of K that depends on the number
->of games played (it starts out large, and slowly decreases to 32).
->
->By the way, I don't find Elo's derivation of the choice of K very
->compelling. For example, his derivation involves computing ratios of
->sums of standard deviations (a no-no in statistics), and he also
->isn't careful on how he incorporates the number of games in an event
->(it can be shown that K should be roughly proportional to 1/n, where
->n is the number of games in an event being updated).
The way the Scottish Chess Association treat K is as follows. Ratings
are updated ANNUALLY, every August. Thus some players play many games
in the rating period (150 was the maximum, last year). If you use a
fixed K, there can be excessive rating swings.
Imagine a 2200 who plays 100 games at 2300. With K=15, he will gain
about 200 points and gain about twice as much as he "should". The SCA
use a variable K, given by K=800/(MAX(n,30)); using this, players with
a large number of games rated all at once are unlikely to jump to much
over their performance rating for the period. The 2200 player above
would have K=800/100=8.
An alternative way to think of the effect of K is to consider the old,
established rating (R_old) and the performance rating over the update
period (R_p). The new rating should lie between R_old and R_p, eg
R_new = R_old + lambda (R_p - R_old), with lambda a weighting factor:
lambda approx 0 corresponds to low K, low change, effectively ignoring
the new results data ("we do not fully trust R_p, because it was based
on too few games..."). This is overdamping: R_new will reach R_P only
if play continues at the same standard for several seasons.
Lambda = 1 corresponds to a maximum "sensible" value of K, and
effectively throws away the previous rating (R_new=R_p).
Lambda > 1 seems "underdamped". The 2200 above has lambda=2, approximately,
and if he continues to play at 2300 standard at K=15, his rating will
oscillate something like 2200, 2390, 2220, 2380, only converging slowly.
Where there is a large update period, such as in Britain (annual) or
with FIDE (six months), and some players might have more than 30 games
rated all at once, it seems sensible to have K <= 800/(MAX(30,n)). Some
players in the FIDE list have 50 or 60 games rated at once, which suggests
K=13 for "critical damping"; K=15 (established FIDE rating < 2400) will
cause a jump to above the performance rating for the period, whilst K=10
(established FIDE rating, at one time > 2400) is slightly cautious.
Of course, if you update ratings monthly, as the mega-keen USCF do,
then larger values of K can be used because no player is likely to have
much more than 10 games rated at once. K*n will probably be less than
800, because n is low.
--
Steve Rix (ste...@chemeng.ed.ac.uk)
"A morbid, Edinburgh-based Chemical Engineer" - and no misprint!