From: Sean Chasworth AOL COM>
Date: 10 jul 1996
Subject: Re: Theoretical Elo ratings
Be very careful when merging a relative scale (i.e. Elo) with an absolute
scale like the Dan/Kyu scale. The Elo scale is only theoretically valuable
in determine references between two players (A will beat B 7 times out of 10)
If you try to use it to assign a 'grade' such as 1 Dan, problems will occur
if your ratings system is not very secure. A secure rating system usually
includes all rated players having over 30 games against all rated players,
and special measures to avoid the overrating of unrated or provisionally
rated players (<30 games).
Ironically, the main problem is DEflation of ratings, as players almost
always start at a lower rating than they finish. This takes away points from
other players, as the average player rating drops. If you want to establish
such a scale, it is important to have a firm starting point, and maintaining
that point by securing the average rating of all the players in your pool.
For mathematicians, remember that the Elo scale is based on a normal
distibution with a mean of 2000 and a std. deviation of 200. In
non-mathematical terms, this means that the AVERAGE player (of all shogi
players 'worldwide') is rated 2000, and 68 of 100 players are rated
1800-2200, and Your individual shogi club may have a higher mean or a lower
mean, depending on the overall skill of your players.
On the other hand, the dan/kyu scale is more workable with a small group, and
is highly effective when assigning specific ranks to an ongoing ladder
tournament. (For example-the top of the ladder is 3 Dan, rungs 2-5 are 1-2
Dan, and so on)
The book to read is "The Rating of Chessplayers" by Arpad Elo. Publisher is
ARCO, New York. This book may be availiable at a chess club bookstore. My
copy is in English, I do not know if it has been translated.