While an ABM is superior to conventional aggregated models in numerous areas, it has one clear weakness: it is unstable, and the results always include some degree of white noise. The uncertainty arises from the simulation nature: each choice corresponds to a Monte Carlo simulation, which is random to some extent.
ABM Nested Demand does not have this instability. This is achieved by modeling individual preferences (see Snethlage 2023a). The resulting model is still a nested logit model.
An individual preference is that share of the utility that is often called the ε (epsilon) or error term. It is interpreted as an individual (i.e., per person different) utility of an alternative, which appears random only because it cannot be observed. Explicit modeling of ε allows for the preservation of this individual utility across scenarios. Chance is not eliminated but held constant in some sense.
The result is a transport model whose responses to change are consistent and not obscured by any white noise. The effect of measures, even very small-scale ones, can thus be determined precisely.
