Some practitioners remain wary of microsimulation models because they perceive a difficulty in drawing firm conclusions from their outputs. While these can be remarkably accurate if inputs are properly described, gut feeling derived from graphical output is not enough, and it is the lack of the simple number answer that causes some modellers to seek comfort in deterministic processes. Modellers generally adopt microsimulation because they consider its representation of traffic flow and congestion to be closer to their expectations of real world conditions, and are not surprised when successive runs of the same model produce a range of results. Of course, deterministic models can produce a range too, but, unlike in the real world, their outputs are precisely determined by their inputs, which have difficulty in encompassing detailed descriptions of some common features such as UTC, bus priority systems and pedestrian controlled crossings. Microsimulation models can include these influences, and represent conditions experienced by real transport users, whose journey might not be entirely predictable. Transport users understand the concept of fuzzy modelling, and while some modellers tend not to appreciate this state of affairs, they may be more accepting if in possession of some simple statistical procedures to help interpret the results of models which in most respects deliver what they expect to see. The answer lies in multiple simulation model runs. Deterministic models tend to produce single number answers based on convergent iterative assignment procedures embodied within single model runs. Because it does not exist in reality, the concept of assignment convergence does not exist with microsimulation. Instead, the model must be run enough times to produce a statistically sound conclusion in which there can be confidence in the model's validation. (Author/publisher).
Abstract