Elo as a statistical rating measure has this neat property given the Elo rating of two players one can calculate the expected outcome to be somewhere between 0 and 1- loss or win. This is calculated by taking the difference between two Elo's.
This got me thinking. So for two 2800 Elo players the outcome would be 0.5 for both- equal chances of winning and losing. This will be the same for 1100 Elo players. Elo treats both pairings as identical.
However, we know that a 100 games of standard time control between the 2800s would probably end in something like 80 draws 20 decisive games. However, the 1100 pair would probably have less than 10 draws and 90 decisive games.
This is because the Elo model treats a draw as half win and half loss. So draws don't exist inside of Elo.
Is there a method that also accounts for draws explicitly?
Also another side idea, since White has a considerable edge (which massively grows together with the ratings). Would it make any sense to have two separate Elo values for playing as black or white? They would for the most part be skewed in white's favour, however, one could make a more accurate prediction of a specific game where one is playing as white and the other as black. The current Elo's really have little meaning when looking at an individual game, but make more sense when two players play several games where they play half of them as white and half black. It would also be interesting seeing how the two Elo disparity grows as the average rating grows.
This got me thinking. So for two 2800 Elo players the outcome would be 0.5 for both- equal chances of winning and losing. This will be the same for 1100 Elo players. Elo treats both pairings as identical.
However, we know that a 100 games of standard time control between the 2800s would probably end in something like 80 draws 20 decisive games. However, the 1100 pair would probably have less than 10 draws and 90 decisive games.
This is because the Elo model treats a draw as half win and half loss. So draws don't exist inside of Elo.
Is there a method that also accounts for draws explicitly?
Also another side idea, since White has a considerable edge (which massively grows together with the ratings). Would it make any sense to have two separate Elo values for playing as black or white? They would for the most part be skewed in white's favour, however, one could make a more accurate prediction of a specific game where one is playing as white and the other as black. The current Elo's really have little meaning when looking at an individual game, but make more sense when two players play several games where they play half of them as white and half black. It would also be interesting seeing how the two Elo disparity grows as the average rating grows.