
BGonline.org Forums
Bearing In Lessoneye opener for most
Posted By: Timothy Chow
Date: Friday, 27 June 2014, at 4:05 a.m.
In Response To: Bearing In Lessoneye opener for most (Taper_Mike)
Just to remark again that this is the wellknown Hiawatha debate all over again. If we assume that a long rollout is gospel (a somewhat questionable claim, but widely assumed), then a full rollout is an unbiased estimator of the gospel, whereas XGR++ is potentially biased but with lower variance. Neither unbiasedness nor low variance is what you really care about in the end; what you care about is something like the probability that the verdict you deliver is correct (where the verdict will be something like, "Plays A, B, and C are ranked in that order, with the equity differences to three decimal places being suchandsuch"). Either approach could give you the higher probability depending on how badly biased the biased estimator is and how inefficient (i.e., highvariance) the unbiased estimator is.

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