The effect of systemic errors on optimal project buffers
By: TRIETSCH, Dan.
Material type: ArticlePublisher: Amsterdam : Elsevier, May 2005Subject(s): Tempo | CustoInternational Journal of Project Management 23, 4, p. 267-274 Abstract: Existing mathematical models for setting buffers for time or cost in project management assume that project activities are statistically independent. This leads to a highly counterintuitive and damaging conclusion that project buffers should become relatively negligible for projects with long chains of activities. We present a model that considers the statistical dependence between activities caused by estimation bias. We show that if relatively high service levels are desired, this imposes a positive lower bound on the buffer as a data-based fraction of the estimated project duration or budget. We also introduce a new approach for collecting data and estimating the parameters necessary to implement the model. This approach places a smaller burden on decision makers than traditional PERT: they provide single point estimates for means, while variance elements and bias correction are computed electronically using historical data.Existing mathematical models for setting buffers for time or cost in project management assume that project activities are statistically independent. This leads to a highly counterintuitive and damaging conclusion that project buffers should become relatively negligible for projects with long chains of activities. We present a model that considers the statistical dependence between activities caused by estimation bias. We show that if relatively high service levels are desired, this imposes a positive lower bound on the buffer as a data-based fraction of the estimated project duration or budget. We also introduce a new approach for collecting data and estimating the parameters necessary to implement the model. This approach places a smaller burden on decision makers than traditional PERT: they provide single point estimates for means, while variance elements and bias correction are computed electronically using historical data.
There are no comments for this item.