ABSTRACT
From the ahove results it is clear that, as in the case of the LS estimator b, the estimator s2 is unbiased for both RSWOH and HSWR and the variance of/ under HS\1\'0R differs from that under HSWR. But, unlike b, the variance of s2 is affected by nonnormality through Y2· For example when the population is large the variance of s2, compared to the normal case, is larger for allnonnormal distributions with V2 > 0. Also the effect of nonnonnality does not disappear for large n since the ratio of the variance under nonnormality to the variance under normality is 1 + V2(n - l)/n, which converges to 1 + V2 as n -7 00.
