Abstract
Various techniques are used to find the first two moments of the statistic T = log (A r /G r ), where A r and G r are, respectively, the Winsorized arithmetic and geometric means based on censored data from a gamma distribution. A two-moment chi-squared approximation is then derived for T and used for testing hypotheses and setting confidence bounds on the shape parameter of a gamma distribution. Power comparisons are made between optimum tests and the corresponding tests based on T.