BGonline.org Forums

standard error reported by gnu

Posted By: Bob Koca
Date: Thursday, 8 December 2016, at 1:24 p.m.

In Response To: standard error reported by gnu (rambiz)

Related to this is the type of statistical test to use. If parallel dice are used doesn't it make more sense to treat the rollouts as matched pairs?

1)" Should't the joint standard deviation rather be: JSD=sqrt(SD1^2+SD2^2-2*COV) ? Where COV is the covariance of the outcomes of the two plays in question."

I think so. This means that the probability of one play being better will be larger than is suggested. This could be empirically tested.

2) " Since all backgammon games end in a race, I think that the covariance is almost always a positive number? Does everybody agree? "

If variance reduction is being used it cuts down the covariance by a lot and the VR is especially powerful for races.

3) "Can you come up with a position, where the outcomes of two plays have a negative covariance (same dice sequence for both plays, of course)?"

A position where you have to choose where to leave a blot and the next dice roll is nearly decisive.

Messages In This Thread

• standard error reported by gnu
  rambiz -- Thursday, 3 February 2011, at 10:11 p.m.
  • standard error reported by gnu
    Tom Keith -- Friday, 4 February 2011, at 3:30 a.m.
    • standard error reported by gnu
      rambiz -- Friday, 4 February 2011, at 9:32 p.m.