Critter 1.6 64-bit- Stockfish 2.2.2 64-bit 65.5-34.5(44-13 and 43 draws)
rating difference 114
Critter performance -18
conclusion 65.5-34.5 means 96 elo difference
How is it possible that higher results mean less elo difference?
I am not certain that I truly understand your question.
However, though the score is the same in both cases, the draw rate is higher in the second case. In such situations, Bayeselo predicts a lower Elo difference for the case with the higher draw rate.
I do not understand it and I think that it does not make sense.
I think that the same score should be the same elo difference
I see no logical reason that 20 wins and 80 draws should mean lower elo difference than 60 wins and 40 losses.
That is how the bayeselo model works and it is one of the improvements of bayeselo over Elostat. Elostat gives the same elo difference for a 1-0 and a 10-0 since in both cases the winning percentage is 100%. Bayeselo otoh gives significantly smaller elo difference for a 1-0. Also more draws indicate the players are evenly matched. If A beats B then A could be any elo higher than B so winning or loosing does not give as much information as draws do. Unlike elostat variance has an effect on the mean elo difference in bayeselo. You can find more info in the bayeselo webpage.
All of this is due to the prior used which assumes the elos are unifrom at the beginning. So to show your superiority over other players, you need to assert yourself many times before bayeselo assigns you larger elo difference. In theory if the prior was choosen such that one player has higher rating than the other a 1-0 may not be signficantly different from a 10-0 since it is assumed the player is stronger a priori.
Daniel Shawul wrote:That is how the bayeselo model works and it is one of the improvements of bayeselo over Elostat. Elostat gives the same elo difference for a 1-0 and a 10-0 since in both cases the winning percentage is 100%. Bayeselo otoh gives significantly smaller elo difference for a 1-0. Also more draws indicate the players are evenly matched. If A beats B then A could be any elo higher than B so winning or loosing does not give as much information as draws do. Unlike elostat variance has an effect on the mean elo difference in bayeselo. You can find more info in the bayeselo webpage.
All of this is due to the prior used which assumes the elos are unifrom at the beginning. So to show your superiority over other players, you need to assert yourself many times before bayeselo assigns you larger elo difference. In theory if the prior was choosen such that one player has higher rating than the other a 1-0 may not be signficantly different from a 10-0 since it is assumed the player is stronger a priori.
I agree that 1-0 and 10-0 are different but when we have the same number of games I see no reason to assume smaller elo difference for more draws.
Note that with +20 =80 I can have more confidence about the superiority of the winner relative to +60 -40.
Daniel Shawul wrote:That is how the bayeselo model works and it is one of the improvements of bayeselo over Elostat. Elostat gives the same elo difference for a 1-0 and a 10-0 since in both cases the winning percentage is 100%. Bayeselo otoh gives significantly smaller elo difference for a 1-0. Also more draws indicate the players are evenly matched. If A beats B then A could be any elo higher than B so winning or loosing does not give as much information as draws do. Unlike elostat variance has an effect on the mean elo difference in bayeselo. You can find more info in the bayeselo webpage.
All of this is due to the prior used which assumes the elos are unifrom at the beginning. So to show your superiority over other players, you need to assert yourself many times before bayeselo assigns you larger elo difference. In theory if the prior was choosen such that one player has higher rating than the other a 1-0 may not be signficantly different from a 10-0 since it is assumed the player is stronger a priori.
I agree that 1-0 and 10-0 are different but when we have the same number of games I see no reason to assume smaller elo difference for more draws.
Note that with +20 =80 I can have more confidence about the superiority of the winner relative to +60 -40.
Yes like I said more draws means more confidence in your result. For your example it would be 20(1-0.6)^2+80(0.5-0.6)^2 versus 60(1-0.6)^2+40(0-0.6)^2, so the first one has much smaller variance. The ratings start from equal for uniform prior so I would think the first case will go further to show that the players are have greater strength difference than the latter would.