A basic statistics question: For a small sample (2 <= n <= 6) least squres
regression, where the variances of the observations are known and the
regression parameters are estimated, I am not sure if I should be using the
gaussian distribution or the t-disribution to calculate the confidence
limits. My understanding is that when the variances are known, I should be
using the gaussian distribution. How does the fact that I am estimating the
regression parameters play into this?
Thanks for any info or pointer to references.
Calvin King
cking@sandia.gov
|