A proposal on improved procedures for estimating task-time distributions in PERT

The objective of using the well-known PERT task-time estimation formulas has two components: (i) to elicit subjective time estimates from an ‘expert’; and (ii) to translate these estimates into ‘something’ that can be used as inputs to probabilistic network analysis. By incorporating the more current techniques in subjective probability elicitation and distribution theory, this paper proposes and justifies a practical and more logical two-step prodecure for achieving both components of the objective. The first step uses a seven-point fractile-estimation procedure, the second step determines the parameters of a beta and a Ramberg–Schmeiser distribution function that minimize the mean square deviation between the distribution function and the estimated fractiles. The procedure can be easily implemented by practitioners using a canned program.