Variability in Transport Microsimulations Investigated With the Multi-Agent Transport Simulation MATSim

Abstract

Transport microsimulations are stochastic. Randomness is, for example, introduced by the error terms of discrete choice models, a common component in utility-based microsimulations. This leads to random variability in results at all resolution levels. This paper’s objectives is an analysis of this variability. As a very common and important aggregate measure in transport planning network link volumes are analyzed, based on MATSim simulation experiments. Constrained by modeling and simulation costs, recent large-scale, high-resolution microsimulations are cross-sectional models. When looking at aggregate levels relevant to planning, for common statistics there is relatively little variability over multiple simulation runs. However, these models do not properly account for temporal variability. This is problematic because temporal variability measured in reality is substantial. Thus, considering extension of these cross-sectional microsimulation models to longitudinal models will be necessary in the near future. To support this, the paper also documents first insights about temporal variability and temporal correlations in microsimulations.