Improved moment-estimation formulas using more than three subjective fractiles

Abstract

PERT-type subjective estimations are used in many stochastic decision models to estimate the random variables' mean and standard deviation (s.d.). The approach is based on the beta-distribution assumption; also, most PERT-type formulas use only three estimated fractiles. We point out that: (i) it is desirable to consider a substantially richer set of distributions than the beta in developing PERT-type formulas; (ii) it may be beneficial to use more than three fractile-estimates in PERT-type formulas. We then develop formulas for estimating the mean and s.d. that are based on a substantially richer set of distributions than the beta and that use more than three estimated fractiles. These formulas perform better than the best currently-available formulas when the subjective distribution is not restricted to be beta.