Dear all,
I am trying to construct a confidence interval for a variance estimator of a variable that is chi-square distributed itself. The bounds of [(n-1)*var.hat(x)/quant.chisquare(1-(p/2), n) ; (n-1)*var.hat(x)/quant.chisquare(p/2, n)] are too narrow, as the variance estimator's S.E. of a chi-squared distributed variable is obviously larger than the the corresponding S.E. of a variance estimator of normally distributed observations.
I am working with simulated data sets of size n=200, so I wondered if there is a way to use a modification of the approximation to the normal distribution of log(s^2.hat)?
Thanks,
Florian
|