bcw (07/21/82)
From: Bruce C. Wright @ Duke University Re: Chess ratings, Micro's, etc. The recent article about chess ratings for microprocessors mentioned the recent changes in the rating system. In fact, there has been substantial inflation recently in the rating system such that ratings post-1980 are not strictly comparable to pre-1979 ratings (I may have the years slightly off but this is the proper time frame). The USCF delegates decided that there had been a long-term rating deflation and they wanted to bring the ratings back to their original values. They had some statistics to back this up, but the decision was still somewhat controversial - the statistics could be interpreted several ways. The idea makes some sense, since people tend to get better rather than worse and are competing for an essentially fixed "pot" of rating points. The decision was to add some extra points to the ratings of all "active" players. This was done over a fairly long period of time so the rise in ratings is not noticeable from month-to-month but the eventual rating levels are quite different from the pre- inflation levels. The effect is strongest for those whose ratings are below about 1900, but is noticeable up to about 2200 or 2300. The largest differences between old and new ratings would be for those around 1600, where the difference would be in the neighborhood of 100-150 points. It is necessary, in evaluating the strengths of chess programs (or human chess players), to have some idea if you're talking about the old system or the new system. Add to this the fact that there are quite a few strange effects - some players do much better (or worse) than their predicted rating depending on the opposition - whether the opposition tends to play strategically or tactically and what types of positions the player and opponent feel most comfortable with - and assigning exact numerical ratings for chessplayers is not such an easy job. Probably it's not very meaningful for any random encounter if the difference is less than a couple hundred points *except in a statistical sense*. There are cases of individual players numerically much different whose results over time deviate far from the predicted values. Thus, haggling over small differences (1910 vs 1950 for example) has significance mostly for a series of hundreds of games rather than a reliable predictor of the next game to be played. Bruce C. Wright @ Duke University