Sensitivity of equilibrium assignment predictions to uncertainty in link service function parameters

This paper focuses on the effects of uncertainty in the parameters of link service functions, or volume delay curves, on the predictions obtained from traffic assignment models. This parameter uncertainty arises when link service functions are estimated empirically using locally collected data. Two effects are identified. The first, a bias effect, is inherent in the conventional network flow predictions which are obtained by ignoring the parameter uncertainty. The second, a variance efect, arises because the uncertainty in the parameters produces an uncertainty in the network flow predictions. These bias and variance effects are measured in a case study. The reuslts highlight a number of important issues which need to be considered when link service functions are estimated in practice.