Heterogeneity in inter-episode intervals for discretionary activities: Covariate-dependent finite mixture models

Even though the importance of considering day-to-day variability in travel demand modeling has long been acknowledged in the field, most state-of-the-art activity-based models still only have a single-day prediction horizon. As such, bias arises from the aggregation to ‘an average’ day. A few which differentiate between days of the week (such as Albatross) still fail to incorporate dependencies between activities conducted in multiple days. Understanding the heterogeneity in (ir)regularity of discretionary activities and the inter-episode durations with which they are conducted, is a stepping stone to extend ABMs to multi-day horizon models. Over two years of GPS data from the Netherlands are used to estimate exponential models to capture irregular activity conductors, while Erlang-k models are estimated to represent the regular activity conductors. A mixture model of the exponential-Erlang-2 model is presented where the extent of activity-regularity is endogenously estimated. The heterogeneity within each group is estimated in a non-parametric fashion and, in certain cases, is shown to outperform the parametric equivalence. The proposed models are applied to grocery shopping, non-grocery shopping and leisure activities.