CONCLUSIONS AND RECOMMENDATIONS For small values of c (c < 1), the differences between the two estimators is relatively small. Generally, the confidence interval on thé arithmetic mean provides better coverage at the expense of a wider confidence interval. For larger values of c (1 < c <2), the lognormal estimator becomes very biased for gamma distributed data, and coverage tends to decline with increasing c. If no theoretical reasons are available for selecting one of the distributions simulated, then the arithmetic n-an is to be preferred because it is unbiased for either of the distributions and tends to have reasonable coverage. ACKNOWLEDGMENTS I thank R. O. Gilbert, Battelle Pacific Northwest Laboratories, and R. J. Beckman, Los Alamos Scientific Laboratory, for reviewing drafts of this manuscript. Funding for this work was provided under Contract No. W-7405-ENG.36 between the Department of Energy and Los Alamos Scientific Laboratory. 625