Various methods of estimation of the skewness coefficient of hydrologic random variables are compared with respect to bias, variance, mean square error, and robustness. The underlying population distributions are limited to gamma (Pearson type 3) and lognormal. The simulation results show that a new skewness estimator that uses the estimated covariance between the subsample mean and the subsample variance has a lesser bias than the ordinary skewness stimators. It may be used as a surrogate when the sample size is small and population skewness large. The superior estimates of skewness resulting from the maximum likelihood estimation of parameters of the known underlying distribution and the robustness studies point to the importance of the correct inference on the type of underlying probability distribution function in hydrologic applications. Copyright 1984 by the American Geophysical Union.
