On the use of multivariate regression methods for longest path calculations from earned value management observations

• Multivariate regression methods are tested for top-down project schedule control.
• Principal components remove noise and colinearities from the EVM/ES.
• A kernel transformation handles non-linearities between EVM/ES and activity durations.
• The regression method is combined with longest path calculations to produce signals.
• The control effort is reduced by removing a drill-down of the WBS of the project.

This paper explores the use of multivariate regression methods for project schedule control within a statistical project control framework. These multivariate regression methods monitor the activity level performance of an ongoing project from the earned value management/earned schedule (EVM/ES) observations that are made at a high level of the work breakdown structure (WBS). These estimates can be used to calculate the longest path in the project and to produce warning signals for project schedule control. The effort that is spent by the project manager is thereby reduced, since a drill-down of the WBS is no longer required for every review period. An extensive computational experiment was set up to test and compare four distinct multivariate regression methods on a database of project networks. The kernel principal component regression method, when used with a radial base function kernel, was found to outperform the other presented regression methods.