
Re: Sampling From Finite Population with Replacement
Posted:
Sep 25, 2010 8:39 PM


On Sep 25, 5:16 pm, Ray Koopman <koop...@sfu.ca> wrote: > > Dividing by N+1 minimizes the expected squared error > in the estimated variance.
Sorry, I should have been more specific. N+1 is for samples from a mesokurtic distribution such as the normal, with mu4/mu2^2 = 3. In general, the divisor that minimizes the expected squared error is N + k2  (k3)/N, where k = mu4/mu2^2.

