Exploring Best-Fit Hazard Functions and Lifetime Regression Models for Urban Weekend Activities:Case Study

Abstract

Activity-based travel demand forecasting consists of modeling activity type, location, and duration with a view to improving transportation planning and creating effective traffic management systems. Research to date has focused primarily on weekday activity patterns, but given its steady increase, weekend activities and related travel demand also deserve attention. Limited research studied weekend activities, and none of them was found to provide detailed specifications with respect to best-fit hazard functions and lifetime regression models. This study, which took place in Calgary, Alberta (a Canadian city of 1,000,000+), is meant to address that gap. Ten activity patterns of eight demographic groups were assessed and nearly 13,000 observations analyzed. Results affirm that most weekend activities are neither work nor school related and tend to begin mid-day or later; analysis of activity participation by demographic group shows that adults (19-64 years old) are the most active components of our society. Likelihood ratio tests confirm that a two-level modeling exercise is required to handle the heterogeneity within the data: first, analysis by activity type and second, analysis by demographic group. Eleven candidate hazard functions were examined for 10 weekend activities such as shopping or entertainment, then best-fit hazard functions and lifetime regression models were determined. The results show a high degree of fit. It was found that the best-fit parametric models for demographic subgroups are generally consistent with those based on activity type at the aggregate level, a discovery that should simplify future applications. Lifetime regression models show that the starting time of a given activity and personal mobility are the most significant factors influencing activity duration. The applicability of fully parametric, nonparametric, and semiparametric model is discussed and addressed at various points within the paper. The rounding problem of reported durations is also noticed and discussed during the process of identifying best-fit hazard functions and lifetime regression models.