From: "Daniel M. Toebbens" HMI DE> Date: 16 dec 2002 Subject: Re: Correlation between ratings and grades On Sat, 14 Dec 2002, Pieter Stouten wrote: > On Thu 12 Dec 2002 14:16 +0100, Daniel Toebbens" HMI DE> wrote: > > >>Looking at the Elo list, you will notice significant > >>correlation between ratings and grades. In fact, I hope > >>that Asle can determine the statistical correlation when > >>he gets back from vacation. > >> > >I think he meant that there is no strict relationship > >between the current Elo and the grade. Of course there is > >a strong correlation between grade and Elo (though it is > >not linear, more like exponential growth). > > > The progressive widening of grade brackets is an arbitrary, but deliberate > choice. It was done to reflect the Japanese situation. However, when > projecting players' current Elo rating back onto a grade and then comparing > that projected grade with their nominal grade, I would expect a linear > relationship (although that may be hard to decide with only 20 or so grades). > > >However, if one looks at i.e. the 2 dan grade, the > >corresponding elo points reach from 1546 to 2118! > > > One should do that only for E-players (i.e., ones with an established > rating). I also believe (but am not sure) that the lowest rated 2-dan > obtained his grade in Japan, and not through our system. Other grades may > have been awarded prior to implementation of the current (Asle's) or even > the previous (Eric's) ratings systems. Having said that, I still think it is > of interest to investigate the correlation between grades and ratings for > all E-players. Just out of curiosity I have done this. You can find the corresponding curve based on the newest FESA Elo List at http://www.hmi.de/people/toebbens/elo.jpg The best fit considering all players with a grade in the list fits nearly perfect the grade/Elo relation given by the midpoint values. Only below 1000 Elo points the curves deviate; this is mainly due to unestablished players. Considering established players only gives a better fit at the low grades, but nothing significant changes. It seems that the various effects leading to deviations counterbalance each other nearly perfect. The distribution of Elo points is for each grade something like two times the LP-HP width; a little less for established players only. Just because it was the trigger of this discussion I have put into the grafik the position of Gert Schnider's 5 Dan. To me it does not look out of position, but that is strictly imho and not for me to decide but for Gert and the FESA's officials. Yours, Daniel Toebbens