H. G. Muller wrote on Sat, Sep 17, 2016 11:28 AM UTC:
Indeed, you would have to know how much deletion of a Pawn would change the result. E.g. if the Griffins beat the Aancas by 60% with equal Pawns, but only score 45% with an additional Pawn handicap, you know the Pawn was worth 15%, so that the original superiority of the Griffins was 2/3 of a Pawn (i.e. 33cP per Griffin).
Note that the random error of the average result in N games is 50%/sqrt(N). So in 100 games the error would be 5%. If a Pawn indeed corresponds to 15%, that would mean the random error in the value determination is 33cP. By playing 2 vs 2 that error is divided by 2 in a singlepiecevalue, i.e. you would measure the value to an accuracy of 1/6 of a Pawn. To make the error twice as small you already need 400 games. This is why I suggested to use time controls as fast as possible, If you can do 12 games per hour you can do 100 games overnight. (And if you have a multi-core CPU, you could play several games in paralel.)
In any case it seems that the Griffin - Aanca value difference is not spectacularly large,
Indeed, you would have to know how much deletion of a Pawn would change the result. E.g. if the Griffins beat the Aancas by 60% with equal Pawns, but only score 45% with an additional Pawn handicap, you know the Pawn was worth 15%, so that the original superiority of the Griffins was 2/3 of a Pawn (i.e. 33cP per Griffin).
Note that the random error of the average result in N games is 50%/sqrt(N). So in 100 games the error would be 5%. If a Pawn indeed corresponds to 15%, that would mean the random error in the value determination is 33cP. By playing 2 vs 2 that error is divided by 2 in a singlepiecevalue, i.e. you would measure the value to an accuracy of 1/6 of a Pawn. To make the error twice as small you already need 400 games. This is why I suggested to use time controls as fast as possible, If you can do 12 games per hour you can do 100 games overnight. (And if you have a multi-core CPU, you could play several games in paralel.)
In any case it seems that the Griffin - Aanca value difference is not spectacularly large,