Rating calculations and different time controls

[Jouni]

In computer games draw rate increases clearly with more time. In CEGT it is in 40/4m 31% 40/20m 38% and 40/120m 48%. Is this somehow included in calculations with ELO, Ordo or Bayeselo!?

[Adam Hair]

The draw rate is positively correlated with the average strength of the opponents, which increases as the time control increases. However, the rating programs do not take this into account (yet).

[Gurcan Uckardes]

Should they really by the way? Does draw rate increase affect score percentages or not? If you come to let's say 60% with 30% drawn or 10% drawn, the elo is the same. Elo does not care about draw rate.

But... There's the error margin which is different. More draws = narrower margin.

Suppose you have engine A vs B, 100 games.
Scenario-1: 45 games won by each and 10 draws.
Scenario-2: 5 games won by each and 90 draws.

They have same elo in both scenarios but second scenario is more accurate statistically.

[hgm]

The problem is that the score percentage is not the same. If engine A beats B by 60% at fast TC, where the draw rate is still the normal 30%, it will score only ~55% against that same opponent at a long TC where the draw rate is 60%. So the Elo difference will look much smaller at long TC.

The problem can be avoided by testing from sufficiently unbalanced positions.