Judgment Estimates of the Moments of Pert Type Distributions

This paper considers the problem of estimating the moments of a statistical distribution from judgment estimates of various percentiles of the distribution and its mode. The results of this study indicate that 5 and 95 percentiles are superior to the 0 and 100 percentiles used in classical PERT; they lead to estimates that are robust to variations in the shape of the distribution, and also there is some experimental evidence which indicates that they can be estimated more accurately from one's “experience.” It was also found experimentally that the single time estimate used in CPM gave a slightly biased estimate of the mean, whereas the three time estimates used in PERT gave unbiased estimates of the mean.