Correlation between USCF Elo and online Elo: post your data here
raylopez99
Elo here--I'm trying to draw a correlation.
I'll start off: Playchess Elo (average): 1600. Estimated USCF Elo:
1900. FICS Elo: 1450 (i think FICS is stronger than Playchess).
Sanny GetClub Elo: Unknown.
Anybody else?
RL
MikeMurray
You won't get a reliable correlation with estimated ratings -- you
gotta use real ones. And asking people to volunteer their ratings
runs a big risk of a biased sample.
Here's what I did a few months ago, just for my own amusement.
(1) Grab a playchess user as randomly as is practical.
(2) See if the user identifies a real name. If not, discard him.
(3) Check the FIDE ratings list for that user.
(4) If no FIDE rating, discard him.
Repeat (1) - (4) until you're satisfied. Didn't take long until I got
tired of it -- I looked at maybe 25 players that seemed to have FIDE
OTB ratings. If you restricted yourself to American-flagged players,
you could use USCF ratings.
Sure, there are still many, many ways the results could be biased.
And players with low PlayChess ratings are unlikely to have FIDE
ratings.
But I found, contrary to my expectations, that PlayChess blitz ratings
were *lower* than FIDE OTB ratings, on the average.
Mark Houlsby
> On Wed, 14 Jul 2010 14:14:30 -0700 (PDT), raylopez99
>
PDFTT
Taylor Kingston
>
> You won't get a reliable correlation with estimated ratings -- you
> gotta use real ones.
Quite.
> But I found, contrary to my expectations, that PlayChess blitz ratings
> were *lower* than FIDE OTB ratings, on the average.
Mike, while you've looked at a larger sample than I have, offhand
this seems implausible based on a comparison of the top 100 for FIDE
vs. Playchess blitz.
Currently the FIDE top 100 ranges from 2639 (Macieja) to 2826
(Carlsen). In contrast, Playchess blitz ranges from 2780 to 3237. A
411-point gap between the two #1s, and you have to go down to
Playchess's #79 before you get anyone lower than Carlsen's FIDE.
While I'm not sure how many in the Playchess blitz 100 are also FIDE-
rated, the two I have checked, just yesterday, Nigel Short and Janos
Farkas, had Playchess ratings higher by 250 and 627 points
respectively. Do you still have your list?
If we're talking about Playchess /slow/ ratings, then what you say
sounds more plausible: the top 100 ranges from just 2034 to 2319.
However, this may just be due to there being few GMs and IMs
interested in slow online chess, so Playchess gets a lower class of
player. Again a statistically significant comparison would be required
to be sure.
raylopez99
wrote:
Good detective work TK and thanks to MikeMurray too.
Also you are comparing, as TK says, apples to oranges to a degree,
since the time controls are different.
But MM is onto something: I think I'm 1900-1950 USCF Elo, based on
triangulation I've done in years past, or (if you believe in the 100
point rule): 1800 to 1850 Fide Elo. That's 200 to 250 points above my
Playchess Blitz rating.
So the formulae for me is: Playchess Blitz Elo + 250 = Fide Slow
(classic chess) Elo.
I'm looking to verify this..or refute it.
Then again, because stronger players play online, you might get
compression of this 250 points.
So, I just logged on to Playchess.com and found something interesting.
In the "Main Playing Hall" you don't have Guests. Let's assume Guests
are patzers (which may not be fair, as I've run across some strong
Guests). So just now I saw 604 non-Guests in the Hall with a rating
of 3032 (!) top, two guys at 2640, lowest guy at 1200 (bunch of them),
some unrated, but if you place the "scroll bar" to 'halfway' on the
screen (I trust you know what I'm talking about), the average rating,
if you sort by Blitz, is 1600 for these 604 players.
So I'm "average" for non-Guests. But--here's the key--if you go into
the "Cafe", and do the same thing, you find the average person is a
unrated "Guest". Keep in mind Guests can only play unrated games in
Cafe. Also the strongest player in Cafe is only 2220, followed by
1954 then 1832. And it seems about 70% (estimating from the
scrollbar) are Guests (unrated). Much weaker players play in Cafe.
And 321 of them.
So do the math: 70% * 321 = 224 Guests (patzers) out of a total of
30%*321 + 604 = 700 non-patzers players, or 224/700+224 = 24%.
So 76% of Playchess players are serious enough to sign up, meaning
they are "hard core" probably. from this 68% the average Blitz Elo at
Playchess is 1600. So where (how far) to the right of the mean is this
1600? That is the 64k$ question--figure that out, and we're home
free.
I figure 76% is too high of a number to include many Patzers. I've
studied chess for two decades, can beat Fritz (at less than 30 seconds
a move on a Pentium IV), if I think hard, and I find it hard to
believe I would be a "Class C" in the pool of ALL players. Now mind
you, if I went to the World Open, where Sam Sloan is now, I might find
indeed I'm Class C, only because so many experts, masters and
grandmasters are playing. But that's another question. The issue
here is of ALL chess players, how does the Playchess/FICS/ICC ratings
for blitz correlate with the FIDE/USCF Elo for all these players, if
they played online (which many, being schoolchildren, do not). Tough
question--unsolvable? Methinks not. We need more data and some
massaging of Gaussian curves...Jeff Sonas we need your help.
The quest continues...
RL
MikeMurray
<taylor....@comcast.net> wrote:
>On Jul 14, 5:48 pm, MikeMurray <mikemur...@despammed.com> wrote:
>>
>> You won't get a reliable correlation with estimated ratings -- you
>> gotta use real ones.
>
> Quite.
>
>> But I found, contrary to my expectations, that PlayChess blitz ratings
>> were *lower* than FIDE OTB ratings, on the average.
>
> Mike, while you've looked at a larger sample than I have, offhand
>this seems implausible based on a comparison of the top 100 for FIDE
>vs. Playchess blitz.
> Currently the FIDE top 100 ranges from 2639 (Macieja) to 2826
>(Carlsen). In contrast, Playchess blitz ranges from 2780 to 3237. A
>411-point gap between the two #1s, and you have to go down to
>Playchess's #79 before you get anyone lower than Carlsen's FIDE.
> While I'm not sure how many in the Playchess blitz 100 are also FIDE-
>rated, the two I have checked, just yesterday, Nigel Short and Janos
>Farkas, had Playchess ratings higher by 250 and 627 points
>respectively. Do you still have your list?
I tossed the list months ago. I'd be the first to admit my results
may have been an anomaly. Plus, the players in whom I was most
interested were those roughly in my PlayChess range and a little
higher, which would definitely bias the sample.
However, I think starting from the top of the rating list, either from
the PlayChess or the FIDE side is questionable. It would be better to
start with random selections from the PlayChess side, since it would
be difficult to determine PlayChess handles from a list of FIDE-rated
players. But completely random won't work, since very few players
below, say, 1900 PlayChess will have FIDE ratings.
It's quite possible that for players in certain rating ranges,
PlayChess is higher, and for other ranges, FIDE is higher. There may
also be differences between players who reveal their real names and
those who don't.
Just for grins, I may try another sample. I'll report on the first 10
PlayChess 2100 and above handles that correspond to FIDE ratings. That
shouldn't take long (and is probably worth about that much, since the
online ratings are so volatile nobody can check my results).
Taylor Kingston
> On Wed, 14 Jul 2010 15:40:44 -0700 (PDT), Taylor Kingston
>
> >On Jul 14, 5:48 pm, MikeMurray <mikemur...@despammed.com> wrote:
>
> >> But I found, contrary to my expectations, that PlayChess blitz ratings
> >> were *lower* than FIDE OTB ratings, on the average.
>
> > Mike, while you've looked at a larger sample than I have, offhand
> >this seems implausible based on a comparison of the top 100 for FIDE
> >vs. Playchess blitz.
> > Currently the FIDE top 100 ranges from 2639 (Macieja) to 2826
> >(Carlsen). In contrast, Playchess blitz ranges from 2780 to 3237. A
> >411-point gap between the two #1s, and you have to go down to
> >Playchess's #79 before you get anyone lower than Carlsen's FIDE.
> > While I'm not sure how many in the Playchess blitz 100 are also FIDE-
> >rated, the two I have checked, just yesterday, Nigel Short and Janos
> >Farkas, had Playchess ratings higher by 250 and 627 points
> >respectively. Do you still have your list?
>
> I tossed the list months ago. I'd be the first to admit my results
> may have been an anomaly. Plus, the players in whom I was most
> interested were those roughly in my PlayChess range and a little
> higher, which would definitely bias the sample.
>
> However, I think starting from the top of the rating list, either from
> the PlayChess or the FIDE side is questionable.
I used the Top 100 lists only because that's what was available to
me. I logged into playchess.com only as a guest, and it seems that in
that mode only the top 100 are made available.
> It would be better to
> start with random selections from the PlayChess side, since it would
> be difficult to determine PlayChess handles from a list of FIDE-rated
> players. But completely random won't work, since very few players
> below, say, 1900 PlayChess will have FIDE ratings.
Ideally, one should have a complete Playchess list and a complete
FIDE list, and compare all players they have in common. As things
stand now, it looks like the only way that can be done is a tedious
manual process.
It just seems to me that such a glaring disparity in ratings — such
as we see in the FIDE top 100 vs. the Playchess blitz top 100 —
indicate that one should be very cautious about claiming one means the
same as the other.
> Just for grins, I may try another sample. I'll report on the first 10
> PlayChess 2100 and above handles that correspond to FIDE ratings. That
> shouldn't take long (and is probably worth about that much, since the
> online ratings are so volatile nobody can check my results).
Okee dokee.
MikeMurray
<mikem...@despammed.com> wrote:
>Just for grins, I may try another sample. I'll report on the first 10
>PlayChess 2100 and above handles that correspond to FIDE ratings. That
>shouldn't take long (and is probably worth about that much, since the
>online ratings are so volatile nobody can check my results).
OK. This was quite a boring exercise. Most players don't give their
real names and those with (apparently) real names don't have FIDE
ratings. But this is the first 10 I came up with. One USCF rating
used. Sorry for the ugly formatting.
Handle PlayChess Name country FIDE Rating
-------- ------------ ------- -------- --
Srbislav5 2114 Sirbislav Mincic Serbia 2281
Bica 2131 Richard Bicek Czech 2024
Jokim 2154 Jokim van den Bos Netherlands 1593
Paul octupos 2162 Emad Madi Libya 2112
Jaktkniv 2170 Vilmos Balint Hungary 2259
Marabu 2172 Mike Koganov USA 2179 (USCF)
Mardona 2180 Andro Wagdy Egypt 1905
Amitwit06 2204 Sergio Segura Spain 2071
Sixten 2303 Mikael Sixtensson Sweden 2204
Petko Burek 2411 Aledsandar Serbia 2294
Average PlayChess Rating = 2200
Average FIDE Rating = 2092
So, my earlier claim about FIDE being higher doesn't seem to be
holding up.
Mark Houlsby
> On Wed, 14 Jul 2010 17:28:52 -0700, MikeMurray
>
Right, Mike. Online ratings are basically meaningless.
MikeMurray
<mark_h...@hotmail.com> wrote:
>Right, Mike. Online ratings are basically meaningless.
I wouldn't go quite that far, but
(1) They are less stable
(2) They are less reliable
(3) They are more easily subject to manipulation and cheating
(4) The online blitz ratings reflect a slightly different skill-set
than OTB blitz ratings
(5) They are somewhat dependent on hardware.
Still, I believe they more closely reflect a real pecking order than
(a) computer generated assessments of a player's strength or (b)
somebody's self-generated estimate of their own place in the pecking
order.
Mark Houlsby
> On Wed, 14 Jul 2010 18:32:30 -0700 (PDT), Mark Houlsby
>
Maybe, but you missed out: mouseslips, the different purposes for
which
different people use different servers, the volatility of the servers
themselves,
the whims and propensities of the admins,....
But yeah, a GM is likely to have a higher rating than, say, I have....
MikeMurray
Mouse manipulation is an online skill -- but dexterity in moving
physical pieces is a corresponding skill on the OTB side.
Servers and admins seem to impact both side of an online game equally,
so I'd tend to toss them out as something causing a rating difference.
>But yeah, a GM is likely to have a higher rating than, say, I have....
Somehow, I think the comparisons might be more granular than that, but
we ARE certainly talking fairly big chunks.
Mark Houlsby
Not necessarily corresponding at all. My mouse is really poor quality,
so
I make lots of slips that I wouldn't make in an equivalent OTB game.
You're disregarding such variables at the *opponent's* end of
things...
these variables exert just as great an influence upon rating as your
own circumstances.
Therefore, online ratings are basically meainingless.
> Servers and admins seem to impact both side of an online game equally,
> so I'd tend to toss them out as something causing a rating difference.
>
Again, your tending to toss them out might not be an accurate
reflection
of what *actually goes on*.
Therefore, online ratings are basically meaningless.
> >But yeah, a GM is likely to have a higher rating than, say, I have....
>
> Somehow, I think the comparisons might be more granular than that, but
> we ARE certainly talking fairly big chunks.
Heck, some of us are blowing them....
None
wrote:
>
> Okee dokee.
Surely you jest.
None
>
> In order to draw a *meaningful* correlation you may need a far
> larger
> sample than you are likely to find here. Also note that players who
> respond to your query here are unlikely to be representative. While
> a
> Class A or Expert player may be proud to post his numbers here, a
> weak player may be a bit more reluctant. Sample size and quality is
> crucial when attempting to draw such correlations or any sweeping
> conclusions.
It depends on your R-square
None
> Not necessarily corresponding at all. My mouse is really poor quality,
> so I make lots of slips that I wouldn't make in an equivalent OTB game.
No doubt if you had a better mouse you'd be almost an IM.
Keep banging those rocks together.
Mark Houlsby
Well, you're not often right, but you're wrong again. I have never
suggested
anything other than that I am weak. The point, which anyone except
you,
and maybe Phil Innes, would grasp readily, is that online ratings are
skewed by such phenomena. I suspect it to be a significant part of the
reason why my ICC blitz rating is unfeasibly high. No way am I as good
OTB as that rating suggests I might be.
Now, go fuck yourself, you piece of shit, Stan.
sd
masturbation of thinking their rating must be higher OTB because they
want it to be. Due to be published next week by the Fishi Press.
THE IDIOT'S GUIDE TO KNOWING YOUR OTB STRENGTH
1. Play in OTB tournaments enough times to get an official rating.
2. Wait for the results.
3. Claim there must be something wrong with the rating system,
4. Skulk back to online chess and declare your skill loudly there in
the safety of your living room.
5. Do not lather, rinse, or repeat.
SBD
raylopez99
> On Wed, 14 Jul 2010 17:28:52 -0700, MikeMurray
>
Mike--how did you get this data? I could not reproduce it--in the
sense that I don't know how you find "real" name vs online handle. I
tried Player Info but it only gives the handle. At any rate, can you
please upload more samples?
Here is what I found so far (data below).
Very weak correlation between Online Blitz Elo at Playchess and Fide
Elo: correlation = +0.35 (weak)
Fide Elo (if you believe there is correlation, which there may not be)
is about 108 points less than Playchess.
We need more data: at least 20 more people to get the error rate down
to 5%--right now it's about 35%.
Can you upload another 20 please? (or tell me how to do it).
Name Playchess FIDE
1 Srbislav5 2114 2281
2 Bica 2131 2024
3 Jokim 2154 1593
4 Paul Octopos 2162 2112
5 Jaktkniv 2170 2259
6 Marabu 2172 2179
7 Mardona 2180 1905
8 Amitwit06 2204 2071
9 Sixten 2303 2204
10 Petko Burek 2411 2294
Average 2200.1 2092.2
Correlation 0.351030589
Std.Dev 90.02894596 214.4661589
http://en.wikipedia.org/wiki/Correlation_and_dependence
http://www1.hollins.edu/faculty/clarkjm/Stat140/correlation.htm
Eliminating the one statistical anomaly (#3, Jokim, who may be
cheating by using an engine), gives slightly better correlation at a
cost of even more error than 35%:
Name Playchess FIDE
1 Srbislav5 2114 2281
2 Bica 2131 2024
4 Paul Octopos 2162 2112
5 Jaktkniv 2170 2259
6 Marabu 2172 2179
7 Mardona 2180 1905
8 Amitwit06 2204 2071
9 Sixten 2303 2204
10 Petko Burek 2411 2294
Average 2205.222222 2147.666667
Correlation 0.360192589
Std.Dev 93.93186065 130.896906
We need more data.
RL
raylopez99
>
> We need more data.
>
I found an interesting chart in Arpad Elo's book, p. 113, "The Rating
of Chess Players Past and Present" (1986, 2nd ed.).
In it it shows a chart of Elo rating distributions according to three
pools: (i) USCF All Players, (ii) USCF Established Players (at least
20 games rated), and (iii) Belgian CF Established Players for 1977.
What is surprising is that for group (i), the average rating is
between 1300-1399 (not 1500 as I assumed, which is Class C according
to p. 6 of Elo's book), while the average rating for group (ii) was
about 1500 and for group (iii) about 1400 (there is rating inflation*
between groups (ii) and (iii), even though both are Established
Players.)
Arpad Elo p. 115 says "In the USCF, since 1977, the Rating Committee
has undertaken an attempt to adjust the U.S. ratings to compensate for
the deflation caused by the great influx of unrated players,
particularly juniors, into the rating pool during the early years of
the 1970 decade.
Elo then shows another chart, p. 115, based on 1983 USCF stats, which
apparently are adjusted to account for the deflation I mentioned
above.
This chart shows Group (i) now is averaged about 1400-1499 (50.17%
cumulative--midpoint), while Group (ii) is closer to 1600 where the
midpoint is.
In fact, here is a more accurate figure for the 1983 list (from Elo's
own text):
Group (i): average = 1505, Std.Dev= 335, Total Sample Size = 48954.
Group (ii): average = 1649, Std. Dev=288, Sample Size = 28988.
So, since my rating at Playchess.com at the moment is 1605, with a 57%
win ratio in 260 stored games (I've played more, but apparently only
the last several hundred are stored at Playchess servers), means I am
above average but not by much. I would argue that the players at
Playchess.com are more like Group (ii), "Established Players", so
there is a 150 point ratings adjustment (inflation or deflation--the
two terms are flip sides of the same coin).
So my "USCF" rating might be, if we assume Established Players play
online as much as they do in tournaments, and that there's a strong
correlation between blitz and OTB play, something close to 150 points
higher (since Established Players are 150 points higher than All
Players, group (ii) vs. (i), see above). So my USCF rating for ALL
players would be about (at the moment) 1605 + 150 = 1755. That
sucks. I want to be closer to 1900. I have 150 points I need to make
up pronto! But I've played better in the past--that's my excuse!
RL
SolomonW
> Please post your estimated or actual USCF Elo and your online average
> Elo here--I'm trying to draw a correlation.
I was remember years ago a guy who was teaching me chess told me that the
chess computers ratings were lower then the actual ratings. He put it down
to advertising trying to make claims
MikeMurray
<raylo...@gmail.com> wrote:
>Mike--how did you get this data? I could not reproduce it--in the
>sense that I don't know how you find "real" name vs online handle. I
>tried Player Info but it only gives the handle.
When you see the list of players, right click on one. Then select
"Picture". If the player has entered what appears to be real first
and last names, you'll see 'em there. Using the name and the country,
you can try to find that player on FIDE's web site
http://ratings.fide.com/ or on the USCF web site.
Of course, you have no guarantee the name or country is correct -- I
could present myself as Bobby Fischer from Uganda if I wanted.
MikeMurray
<raylo...@gmail.com> wrote:
>So, since my rating at Playchess.com at the moment is 1605, with a 57%
>win ratio in 260 stored games (I've played more, but apparently only
>the last several hundred are stored at Playchess servers), means I am
>above average but not by much.
Just as important as the win ratio is the average rating of the
opposition.
I believe that while they *do* keep track of the number of games
played since the handle was created, the rating graph, win ratio and
running average of opponents' ratings reflect only the last 200 games.
> So my USCF rating for ALL
>players would be about (at the moment) 1605 + 150 = 1755.
The most I would infer from a PlayChess rating of 1605 is your USCF
rating is likely to be somewhere in the range 1450 to 1750.
None
>
> Now, go fuck yourself, you piece of shit, Stan.
Unlike you I'm not into gay sex.
Keep banging those rocks together
sd
> When you see the list of players, right click on one. Then select
> "Picture". If the player has entered what appears to be real first
> and last names, you'll see 'em there. Using the name and the country,
> you can try to find that player on FIDE's web site
>
> http://ratings.fide.com/or on the USCF web site.
>
One of the other sites for looking up people that is especially good
for Playchess since there are so many Germans is the Deutscher
Schachbund site (http://www.schachbund.de/ - you don't need to know
much German, on the right side there is a section called DWZ/Elo and
you can enter a name; DWZ are the German national ratings) - in fact
even several years back most countries had their ratings online - for
example, Canada's are at http://www.chess.ca/ratings.shtml.
I used to do this same type of comparison you did Mike. Also useful on
ICC was to check your ranking as well, and occasionally I would look
at the OTB ratings of people rated close to me as a sort of rough
comparison. But there seemed to be as many USCF 2100s at 2500 as
there were FMs with ratings of 1800, and as posters have noted here,
just the very nature of online chess makes a comparison with OTB
difficult.
Taylor noted the large rating inflation at the top. Although I profess
no special knowledge of ratings, I always thought that was due to the
fact that the pool was rather insular at that level, leading to
inflation (think Bloodgood!), which I believe, as you have noted, is
probably less acute at lower ratings. I am sure there is still some
inflation at lower ratings, but probably not the 300 points he found
among the top players.
This sort of thing begs for study rather than our conjectures though.
SBD
raylopez99
> The most I would infer from a PlayChess rating of 1605 is your USCF
> rating is likely to be somewhere in the range 1450 to 1750.
MikeMurray--thanks for your help. Using your help I was able to get
30 more or less randomly selected people from Playchess. The revised
stats below. With a sample size of 30 the error rate drops to 5%--but
that does not change the fact that correlation between Playchess blitz
and Fide OTB regular play is still about 0.33--which is weak.
That said, *if* this correlation is accurate, then the stats below
indicate that your Playchess blitz Elo equals your OTB Fide Elo--for
most people. Some its higher or lower (see below). I am pretty
confident that these people exist--they are not impersonating anybody--
since a lot of them on there Playchess page actually referenced their
Fide Elo and put other stuff like U18 third place finalist (e.g. for
AOut22)--why would they go to such lengths to impersonate somebody?
If you were interested in faking something probably you would be using
a chess engine and not worrying about putting too much personal
information on your profile.
So, *if* I'm one of these people who's online Playchess blitz Elo
equals their Fide Elo, and if Fide Elo is about 100 points lower
(stronger) than USCF Elo, then my current USCF Elo is about 1705.
Keep in mind, just like in Playchess, Elo varies, and goes in cycles.
The standard deviation is about 120, so you can say my USCF Elo--if
the correlation holds--is between 1605 to 1805 at this time.
This was an interesting exercise.
RL
First number to the right of the Playchess handle is the Playchess 5'
blitz Elo, followed by the official Fide Elo from Fide's website for
the person behind the handle. Averages, standard deviations and
correlation between the pools (0.3282) is at the end. Note the average
Elos are about the same for both pools, but correlation, at 0.33, is
not high, so you can be higher or lower, depending on personal
circumstances.
Name Playchess FIDE
1 Srbislav5 2114 2281
2 Bica 2131 2024
3 Master--Sim 1960 2113
4 Paul Octopos 2162 2112
5 Jaktkniv 2170 2259
6 Marabu 2172 2179
7 Mardona 2180 1905
8 Amitwit06 2204 2071
9 Sixten 2303 2204
10 Petko Burek 2411 2294
11 Grande Gila 2018 2121
12 Bullmover 2347 2408
13 Gisman 1943 2063
14 Maciek_92 2649 2247
15 Marian90 2255 1991
16 Biljanic 2216 2301
17 AOut22 1836 1945
18 Benga17 2640 2195
19 Bekbekusv 2300 2244
20 Ds 2037 2124
21 Dutchklatsch 1845 2131
22 Ermal 1979 2097
23 Miki Veliki 2766 2460
24 Kollarz 2206 2296
25 Jan Van Dorp 1967 1977
26 IU008 2070 2162
27 Beauty Prize 2022 2154
28 Bal des laze 1716 1584
29 Avantasia 2271 2150
30 Alfrik 1827 1984
Playchess Fide
Average 2176 2150.409091
Correlation 0.328213765
Std.Dev 117.7125029 124.9158418
MikeMurray
wrote:
>Taylor noted the large rating inflation at the top. Although I profess
>no special knowledge of ratings, I always thought that was due to the
>fact that the pool was rather insular at that level, leading to
>inflation (think Bloodgood!), which I believe, as you have noted, is
>probably less acute at lower ratings. I am sure there is still some
>inflation at lower ratings, but probably not the 300 points he found
>among the top players.
The fact that an online blitz regular can play more rated games in six
months than the average GM plays in a lifetime might magnify any
rating inflationary tendencies.
>This sort of thing begs for study rather than our conjectures though.
Agreed. It seems a fertile field for human performance study.
>SBD
Stephan Bird
7303d1e0-c4dc-4d55...@w31g2000yqb.googlegroups.com,
raylopez99 wrote:
> Also you are comparing, as TK says, apples to oranges to a degree, since
> the time controls are different.
Not just time controls, but the actual ratings calculation is different
too, I believe
raylopez99
>
> raylopez99 wrote:
> > Also you are comparing, as TK says, apples to oranges to a degree, since
> > the time controls are different.
>
> Not just time controls, but the actual ratings calculation is different
> too, I believe
If so, I doubt it matters much. One thing I notice is that online, at
Playchess, the ratings jump quicker. They are not integrating as much
as I think the traditional formulae do. So beating (or losing) to a
player makes your online Elo jump (or drop) quicker than expected.
But in the long run, it all averages out.
BTW, I looked up in my notes the correlation between Fide Elo and USCF
Elo: it's as follows: USCF = ( FIDE * 0.895) + 367 . For most
people, this is roughly 100 to 120 points or so.
RL
Taylor Kingston
>
> OK. This was quite a boring exercise. Most players don't give their
> real names and those with (apparently) real names don't have FIDE
> ratings. But this is the first 10 I came up with. One USCF rating
> used. Sorry for the ugly formatting.
>
> Handle PlayChess Name country FIDE Rating
> -------- ------------ ------- -------- --
> Srbislav5 2114 Sirbislav Mincic Serbia 2281
> Bica 2131 Richard Bicek Czech 2024
> Jokim 2154 Jokim van den Bos Netherlands 1593
> Paul octupos 2162 Emad Madi Libya 2112
> Jaktkniv 2170 Vilmos Balint Hungary 2259
> Marabu 2172 Mike Koganov USA 2179 (USCF)
> Mardona 2180 Andro Wagdy Egypt 1905
> Amitwit06 2204 Sergio Segura Spain 2071
> Sixten 2303 Mikael Sixtensson Sweden 2204
> Petko Burek 2411 Aledsandar Serbia 2294
>
> Average PlayChess Rating = 2200
> Average FIDE Rating = 2092
>
> So, my earlier claim about FIDE being higher doesn't seem to be
> holding up.
I just did a little more research in this vein, starting at the top
of the Playchess blitz ratings and comparing ratings for the first 10
of them that were also FIDE-rated. Here are the results (apologies if
the columns line up poorly):
Playchess
Name Real Name Country
quenplen Grachev. Boris Russia
gashivmov Gashimov, Vugar Azerbaijan
bobance Bogosavljevic, Boban Serbia
IturrizagaEdu Iturrizaga, Eduardo Venezuela
youngpatzer Grguric, Daria Croatia
powerguru Dorfman, Iosif France
Lord Malshun Markus, Robert Serbia
Tiger Lilov: Lilov, Valeri Bulgaria
brdo Miljkovic, Miroslav D. Serbia
RM Spirin, Oleg Russia
These were the first ten on the Playchess blitz rating list who gave
real names. All of them, it turns out, are FIDE-rated, and all but one
have a FIDE title: six GMs, one IM, and two FMs. Only Grguric is
untitled. Now here are the ratings:
PC FIDE
Rating Rating difference
3237 2668 569
3182 2719 463
3171 2551 620
3107 2614 493
3078 1913 1165
3056 2583 473
3028 2616 412
3025 2402 623
3010 2510 500
3010 2468 542
In every case here the Playchess rating is substantially higher. The
mean difference is 586 points, the median point between 542 and 500.
raylopez99
wrote:
> In every case here the Playchess rating is substantially higher. The
> mean difference is 586 points, the median point between 542 and 500.
TK--I'm glad you did this study, because it forced me to relook at my
data, and i FOUND AN ERROR IN MY DATA!
The error is very significant: in correlation. I found that I was
not properly looking at all the data, just the first 10 or so. A
simple Excel typo.
So below is the revised data, and note, with N=30 players, correlation
climbs to a significant number 0.68, which shows strong correlation
between Playchess blitz and Fide Elos. With your sample of only 10
players, correlation is only 0.33. The more players you add, the
better we can gauge correlation. Can you find the next 20 and post
your data here? I'm curious if the correlation holds for the "top"
players, which is what you did, as opposed to randomly selected
players, which is what I did. I bet it does not (that is, inflation
holds true for top players, due to the rareified nature of those
players, but not for average players, where there is no inflation).
Here below is the new, revised data set of mine, corrected for
correlation, but otherwise the same.
RL
Playchess Handle, Playchess Elo, Fide Elo
All these handles have been verified as being actual people that have
Fide ratings as well.
Correlation 0.675928395 !!! REVISED!!! --note it now shows strong
correlation, unlike before
Std.Dev 117.7125029 124.9158418