Review of population synthesis methodologies

Abstract

In the paper we provide a definition of the term ‘population synthesis’ and discuss it’s important and its challenges from a transport demand perspective. Moreover, we offer a classification of the different methodologies and offer the conclusion that, a major distinction between the different methods, whereas these come from the transport demand literature or the spatial simulation literature, is whether these are based on a deterministic or probabilistic approach. The first approach assumes that the sample, on the basis of which the population is constructed, is a ground truth. Contrarily to this, the probabilistic approach assumes that the original sample is one of many realisations from a ground truth distribution. The paper then moves on to further classify probabilistic methods into data-based and model-based simulation methods. The former refers to the case where proxies for probability are formed by drawing from conditionals and margins, whereas the latter refers to the case where a model-layer is introduced between the original data and the re-sampling of agents. As part of a final discussion the paper considers the different methods from a challenge and opportunity perspective and discusses the role of ‘zeros’, scalability, sample diversity and validation.