These days I am reading on chessbase about K-factor change proposals, so I guess it was a good thing I did a little exercise on learning about the Elo rating. As it stands now, the K-factor that affects the change in player's rating is not the same for lower and higher rated players. It has lower value for high rated players, and vice versa. Ignoring for the time being objections from statistician Jeff Sonas about the accuracy of Elo rating system, if we assume that the rating distribution has the same variance but different mean for every player, than it would be fair to use the same step size for the rating adjustment regardless of player's strength. Consequently, the K-factor should be the same for any rating.

However, the whole discussion about the “accuracy” of chess rating seems overblown. Above mentioned statistician did the analysis over historical data and came up with “more accurate rating formulas”. While I will by his “bias against the higher-rating player” argument (see http://db.chessmetrics.com/ for details), everything else seems like too much work for nothing. How do you accurately estimate a mean of a random process with a time variable mean if you have limited number of samples? You simply cannot! You can talk about zone of confidence or probability that someone's rating is within a certain range, and that is pretty much it. Everything else is just a matter of convenience and convention. If the same set of rules applies to every player, what you really care about is the relative performance, not some absolute and “accurate” measure.

In my opinion, chess ratings should be described with average rating or a rating interval, plus the probability that your rating is withing a given interval. If you play small number of games than the confidence about the interval is smaller. It might be less exciting, but it would be more accurate. As for the complexity of calculation argument, even cell phones these days have enough processing power do do calculations required for this. But this was just a digression.

So, does it matter if the K-factor is changed? Most likely not. If it gets changed than you will have a larger step size in the rating change, or a faster reacting tracking system, but noisier data. Players like Fisher, Kasparov or Magnus (in the future) would likely get higher rating at the peak of their carrier, and then decline faster. For high rated players it would just mean larger swings and an opportunity to occasionally play in some top-notch tournament, but also they will see their rating drop faster. For everyone else it wouldn't matter anyway since the change would affect only the top rated players.