An improved parameter estimation procedure has been developed by using
optimization techniques and applied to estimate the parameters of the
log-Pearson type 3 (LP3) distribution. As a result, an improved estim
ation method was found. The new methods estimates the mean and the sta
ndard deviation of the log-transformed data by the method of moments a
nd estimates the coefficient of skewness by minimizing both the relati
ve root average square error (RRASE) and the relative average bias (RA
B). Monte Carlo simulation was conducted for four selected LP3 populat
ions. As compared with the method of moments, larger reductions in sta
ndard root mean square error (SRMSE) and standard bias (SBIAS) for qua
ntile prediction can be achieved by the new method for small sample si
zes and large return periods of quantiles. In addition, the new method
can always fit the observed data better than the method of moments.