Accommodating (non-)monotonically changing hazards in finite mixture models for the inter-episode duration of discretionary activities

Understanding the dynamics behind inter-episode durations of discretionary activities can lay the groundwork for expanding activity-based models beyond single-day representation to encompass multi-day forecasting. Although the field has long recognized the variability in day-to-day activities, many advanced activity-based models still neglect the interdependencies among activities spanning multiple days. This paper examines the intervals between successive participations in grocery shopping, non-grocery shopping and leisure activities specifying different finite mixture models using three parameter Weibull and shifted-log-logistic baseline hazards. Besides accounting for unobserved heterogeneity in a non-parametric fashion, the models accommodate time-varying covariates, household-, individual-, and land use characteristics, and are estimated using over two years of GPS data collected in the Netherlands between 2019 and 2022. Restrictions during COVID-19 have been incorporated in the models as a time-varying covariate since the data collection period encompasses the pandemic period. The significant shape parameters for both the Weibull and log-logistic segments suggest increasing hazards with time, warranting the usage of more flexible models accommodating (non-)monotonically changing hazards. • Explores (non-)monotonicity in hazards for inter-episode durations for grocery shopping, non-grocery shopping and leisure activities • Includes non-parametric heterogeneity in the finite mixture model • Includes time-varying covariates and land use covariates in addition to personal and household characteristics