Modeling project schedule uncertainty due to a programmatic risk factor using bayesian networks

The classical method of project risk analysis, for example, the Program Evaluation and Review Technique (PERT), ignores statistical dependence among project activities, which puts limits to its effectiveness as a robust method for probabilistic schedule analysis. Also, some programmatic risk factors have been identified as significant sources of uncertainty in project performance. However, given a project that is progressing within the project life cycle, monitoring of uncertainty in the project completion times resulting from a programmatic risk factor while taking advantage of activity duration statistical dependence, has not been addressed, to the best of our knowledge. In this study, we develop two methods that are based on Bayesian Networks (BN) for evaluating and monitoring uncertainty in project completion times, when an ongoing or progressing project is impacted by a programmatic risk factor. The BN methods developed in this study model statistical dependence in project networks using parametric relationships between nodes which reduce the burden of dependence specification in the BN model. In modeling the relationships of the variables of the BN models developed in this study, concerns about computational complexities and efficient modeling of interactions of variables of a project are enabled by the capacity of the specialized Bayesian Networks software (AgenaRiskRTM) used in the analysis. Other concerns of classical PERT such as: (i) assumption of statistical independence is addressed by using conditional median as a measure of statistical dependence, and (ii) the constant PERT variance assumption is addressed using the Modified PERT Variance. Using the BN models developed in this study, we demonstrate that failure to incorporate statistical dependence grossly underestimates the total uncertainty in project completion times. The graphical dimension of our model, which benefits from the capacities of Bayesian Networks, gives more visibility about the model development and uncertainty analysis process, and which could be helpful to project analysts and managers by providing greater insight and formal mechanisms for interpreting how uncertainties in project performance measures emerge. More so, the faster learning about remaining completion time uncertainty combined with the precision of the BN approach may provide project managers more time to take corrective action to avoid schedule slippage.