grenv:

The results of the new system still don't look right to me, as the win/lose differential is very large. This might be appropriate in a game like chess, where the better player should usually win. However, in backgammon, even against a weak opponent, a strong player will lose a significant number of games. For example, I would probably lose 10-15% of my games against a beginning player just because of the luck of the dice. But even against opponents close to my skill level, I currently stand to lose much more than I stand to gain. For example, in a Nackgammon game I am currently playing, the ratings are 1925 and 2208. The win/lose differential is 3-13 or -10 against me. This implies the odds of my winning are something like 13/16, which I think overstates the probabilities.

It would be interesting to see if one could devise a ratings system that more closely reflected the actual likelihood of winning based on a given rating differential.