Author: KarinsDad
Date: 15:02:55 01/29/99
Go up one level in this thread
On January 29, 1999 at 16:46:20, Matt Frank wrote: > >> >>Ah, no. I do not understand your point. Are you saying that since Professor Elo >>said that a 675 point difference should average to a 49 to 1 ratio, that this is >>what should happen. > >No, just that those are the expected results (margin of error is not totally >absent--especially with sample sizes in the dozens or even low hundreds of >games). > >>I do not think that practically, that is the case. I think that the formula >>starts getting more and more off once you get past the 300 to 400 point range. > >This may be so but I would have to be shown the evidence. Sorry, I am a logical >positivist at heart :-) > > >>If has never happened in my experience, nor for anyone I've ever known. I've >>never had a B class player friend say, "Hey, I beat a master". Ever. Not even a >>"Hey, I drew a master." comment. This is not to say that it has never happens, I >>just have never heard of it happening (except in blitz and in one case in my >>local club where a C class player beat an expert giving a simul, but she is also >>an up and coming player and it was a simul) and I know quite a few chess players >>around the country. > >With that anecdote and a dollar I can get you a double-chessburger :-) >> >>The generally used formula is just a rough approximation. The USCF could >>probably analyze it's database and come up with a better formula (however, even >>this would be suspect due to rating deflations due to up and coming players and >>rating inflations due to floors, aging effects, and health effects). > >No measurement is perfect, I agree. However, the data supports that the elo >ratings are reasonably robust over time and across people (machines?), although >the USCF ratings have obviously been subject to some inflation in the last 20 >years. Yeah, it's due to the floors. G*#$#%% stupid system. And even more stupid for the delegates to not get rid of it due to politics. > > >>For computers, I wouldn't have a clue. Maybe some of the other people who test >>them against each other would know something. > >That a good point and should also be answered empirically. > >Matt Frank I agree. Everything I mentioned was anecdotal. It would be good to have empirical data. But sometimes, experience tells you the answer ahead of time. KarinsDad :) PS. Another example. I played a guy who was not rated, but should have been about 100 to 150 points below me at lunch time at G30 for a half a year. I won every game except the last one (where I had a choice of 2 moves and picked the wrong one, moving quickly will get you everytime). So, although he played strong and often had me in tight positions (he was up material about a third of the time), I managed to squeeze out (barely in a lot of cases). According to the formulas, his rating should have been about 850 points lower than mine. But the truth of the matter is that he just didn't have my number. They were almost always close games. I never once blew him out. This was a sample set of over a hundred games which bucked the ratings odds. Look at Anand vs. Karpov. Ratings are a very crude metric, but they tend to mean everything once they are 500 points apart.
This page took 0.01 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.