Using cluster analysis to identify behaviorally homogeneous groups in multiple stages of the four-stage model

Abstract

Recent events such as the COVID-19 pandemic or rising energy prices have resulted in behavioral changes regarding everyday mobility. Transport models should be able to capture such effects to forecast future behavior and the effects of transport policy measures more precisely. In the status quo, models incorporate “behaviorally homogeneous groups” derived from travel surveys. In previous studies, researchers used a variety of statistical approaches to determine behaviorally homogenous groups. However, variable selection was often only weakly grounded in theory and not always consistent with the four-stage model of transport. The objective of this paper is to present a five-step approach to identify behaviorally homogenous groups according to multiple stages of the four-stage model. We base our analysis on mobility behavior captured in trip diaries within a Germany-wide representative survey. In line with the four-stage model, we perform k-means clustering to group respondents separately according to trip purposes and mode choices. We find seven unique clusters on either stage. We explain cluster assignment with sociodemographic variables using multinomial logit (MNL) models and analyze feedback effects. The measures of fit indicate the necessity to include feedback effects in transport models. Transport modelers can use the identified significant sociodemographic variables to segment the underlying (synthetic) population accordingly.