The Effects of Different Activity Duration Distributions in PERT Analysis

Abstract

Project Evaluation and Review Technique (PERT) has undergone several extensions over more than six decades, particularly in the choice of probability distributions used to represent activity durations. However, the extent to which alternative duration distributions influence PERT estimates remains debated. This study investigated the effects of different activity-duration distributions on PERT estimates using a shipbuilding project network comprising 18 activities and three-point judgmental estimates (optimistic, most likely, and pessimistic durations). Four approaches—PERT-Burr XII, PERT-Beta, Shankar and Sireesha, and Golenko-Ginzburg—were compared based on estimated activity means, variances, coefficients of variation, and project completion time (PCT). For the activity means, the average absolute percentage errors relative to the standardized Beta reference means were 1.6354%, 1.6224%, 1.0938%, and 1.9166% for PERT-Burr XII, PERT-Beta, Shankar and Sireesha, and Golenko-Ginzburg, respectively. Thus, the Shankar and Sireesha method produced the closest mean estimates to the reference values, whereas Golenko-Ginzburg produced the largest average deviation. PERT-Burr XII generally produced higher variance and coefficient-of-variation estimates than the other approaches. Despite these differences at the activity level, the effect on overall project completion time was small. The estimated PCTs were approximately 149 days for PERT-Burr XII, PERT-Beta, and Golenko-Ginzburg, and 150 days for Shankar and Sireesha, compared with approximately 151 days for the standardized Beta reference case. The findings indicate that alternative activity-duration distributions can produce measurable differences in activity-level parameter estimates but have a relatively limited effect on overall project completion time for the project network examined. Further studies using multiple project networks and independently observed activity-duration data are recommended to establish the generalizability of these findings.