Author: Dave Gomboc
Date: 01:02:53 12/12/99
Go up one level in this thread
On December 11, 1999 at 13:36:54, Len Eisner wrote: >On December 11, 1999 at 12:46:46, Christophe Theron wrote: > >>On December 11, 1999 at 11:12:15, Len Eisner wrote: >> >>>On December 11, 1999 at 08:26:16, Chuck wrote: >>> >>>>On December 11, 1999 at 06:13:02, Bertil Eklund wrote: >>>> >>>>> >>>>>Hi! >>>>> >>>>>You have probably mixed up the figures for Mach4 with another program or some >>>>>advertisment for Mach4. January 1993 the rating was 2080 six points down vs >>>>>January 1992. >>>>> >>>>>Regards Bertil >>>> >>>>You are right, Bertil. I realize I made a mistake. Of course, now I am looking >>>>at the SSDF list proper (unadjusted), whereas in 1993 all I had available were >>>>"US-ratings adjusted" copies of the list, such as that published in CCR. >>>>Obviously, they must have added 200 points to the ELO, but on the copy I have >>>>handy it does not specify that any adjustment was made. I apologize for the >>>>confusion. >>>> >>>>Chuck >>> >>>Let's change the question a bit. If the unadjusted ratings for older programs >>>are so low by USCF standards, why shouldn't we add 200 points to the list. And >>>if we do add 200 points, the ratings for the new programs become even more >>>inflated than they currently are. >>> >>>For example, the Super Constellation has a 1731 rating on the SSDF list. Anyone >>>who has played this program can tell you it's either an expert or very close to >>>one, at least by USCF standards. >>> >>>I'm sure there are some folks in this forum who have experience playing the >>>Supper Connie. It was very popular around 1987. In your opinion, is 1731 an >>>accurate rating for it, or is it closer to 2000? Is it just me? >>> >>>Len >> >> >>Once again you are mixing USCF and FIDE ratings I think. >> >> >> >> Christophe > >No, I'm not confusing them. I understand there are three separate lists: USCF >FIDE, and SSDF. > >I'm just pointing out that the older programs on the SSDF list seem to be rated >too low, at least by USCF standards. If you add some number to the SSDF ratings >to get a USCF equivalent, then it would bring the old computers in line with my >expectations, but the newer ones would be too high. If you say it is not >possible to compare the lists, then I have to question the accuracy of the SSDF >ratings, even from a relative perspective. > >You can compare FIDE and USCF ratings by adding a constant, let's say 50 points >for the sake of argument. If you can't do that with the SSDF list, then >something is wrong. > >Len No, nothing is wrong. Even between FIDE and USCF, adding a constant doesn't work so well. AFAIK, a linear formula (ax+b) is used when it's important. But let's proceed with the simplifying assumption that the FIDE and USCF pools are similar enough that adding a constant is sufficient to convert from one to the other. The main problem is that the populations of the SSDF pool and the FIDE pool are not similar. Let there be two versions of a program, A and B, where A is one standard deviation (200 SSDF) less skilled than B in the SSDF pool. (Maybe A is version 2 of a product, and B is version 5... whatever.) Now the big question is, in the very different FIDE player pool, would B still be one standard deviation better than A? This cannot be assumed. B might even be more than one standard deviation (200 FIDE) better than A in the FIDE pool, but I think that most of the people who concern themselves with this sort of thing suspect that B would be less than 200 FIDE better than A. How "inflated" SSDF ratings will appear relative to the FIDE pool depends on how much better B is than A in the FIDE pool. Let's say that B is 150 FIDE better than A in the FIDE pool. Then, to convert ratings from the SSDF pool to the FIDE pool, "a" in the linear formula will have to equal 150/200 = 0.75. Even the above has a ton of simplifying assumptions, but I hope it is good enough to show that it is not reasonable to expect conversion between SSDF and FIDE (or SSDF and USCF) by only adding a constant. Dave
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.