
BGonline.org Forums
VRR  Challenge
Posted By: Maik Stiebler
Date: Friday, 30 January 2015, at 8:54 p.m.
In Response To: VRR  Challenge (Rick Janowski)
If VRR was used what would the variance reduce to? 67 perhaps?
Long ago, I tracked my VR rating on FIBS for a while using a simple spreadsheet, and the difference in quality compared to the conventional rating baffled me. So, I think the reduction will be better than that. If one VR match is worth 70 nonVRmatches, as Bob wrote, then a reasonable estimate is that the deviation is reduced to sqrt(1/70) of what it is without VR, just because that's by how much the deviation of the result of a single match is reduced. That's possibly too optimistic for two reasons I can think of: First, Bob's estimate is for botvsbot play, and maybe humanvshuman contains more hardtoreduce variability. Second, the estimate doesn't address that the restoring force to the true rating is stronger simply because the distance from equilibrium is typically larger. Still, I feel safe to bet that the deviation is reduced to 20% or less by using VR.
This assumes that the same formula for updating ratings is used for the VR and the nonVR case. That's not necessarily the best thing to do. For example, instead of using the FIBS match value of 4*sqrt(match length), one could use 30*sqrt(match length) to have the same variance per match but a hugely stronger restoring force.

BGonline.org Forums is maintained by Stick with WebBBS 5.12.