H. G. Muller wrote on Mon, Jul 24, 2017 07:42 PM UTC:
First note that the error bars on VR's measurement are rather large. Doing such tests at long TC is very hard. With close to 1/3 of the games a draw, the statistical error in the score is 40%/sqrt(N), where N is the number of games. So for 80 games the error is about 4.5%. To half them would require 4 times as many games.
Kevin is right, in that I would have predicted a somewhat larger difference when two Knights are replaced by two Bishops on 10x8 (assuming that Pawn odds would score about 65%). But it could indeed be an effect of having these in addition to 2 Knights plus 2 Bishops. Perhaps 4 Knights cooperate better than 4 Bishops. Certainly 7 Knights work better against 3 Queens than 7 Bishops; I have extensively tested that. It could also be that with 4 Bishops against 6 minors the chance that they are traded in such a way that you are left with two on the same color is pretty large, and that would also suppress the value of the Bishops.
First note that the error bars on VR's measurement are rather large. Doing such tests at long TC is very hard. With close to 1/3 of the games a draw, the statistical error in the score is 40%/sqrt(N), where N is the number of games. So for 80 games the error is about 4.5%. To half them would require 4 times as many games.
Kevin is right, in that I would have predicted a somewhat larger difference when two Knights are replaced by two Bishops on 10x8 (assuming that Pawn odds would score about 65%). But it could indeed be an effect of having these in addition to 2 Knights plus 2 Bishops. Perhaps 4 Knights cooperate better than 4 Bishops. Certainly 7 Knights work better against 3 Queens than 7 Bishops; I have extensively tested that. It could also be that with 4 Bishops against 6 minors the chance that they are traded in such a way that you are left with two on the same color is pretty large, and that would also suppress the value of the Bishops.