Choice of probability distributions for activity durations in project networks with limited sample size

Accurate estimation of activity durations is a critical project scheduling step. If abundant data are available for activity durations, then distribution fitting is straightforward. However, if sample sizes are small, then choosing an appropriate theoretical distribution for activity duration is challenging. In this study, we propose a methodology for experimenting with the impact of sample size on the accuracy of the estimated duration of a stochastic project network. We synthesized a total of 3,264 stochastic project networks and estimated the true project duration distribution for each network using large-sample Monte Carlo simulation. We varied the sample size per activity and investigated how the choice of theoretical distribution (normal, beta, triangular, and lognormal) for estimated activity durations impacted the accuracy of the estimated project duration distribution. We find that the best choice for a theoretical activity duration distribution highly depends on the sample size. For sample sizes below 50, we find that not much is gained from the common practice of putting extensive effort into selecting best-fit theoretical distributions compared to simply using normal distributions.