What drives active transportation choices among the aging population? Comparing a Bayesian belief network and mixed logit modeling approach

A.D.A.M. Kemperman, H.J.P. Timmermans

Research output: Chapter in Book/Report/Conference proceedingConference contributionAcademicpeer-review

61 Downloads (Pure)

Abstract

As people age, they typically face declining levels of physical ability and mobility. However, walking and bicycling can remain relatively easy ways to be physically active for older adults provided that the built environment facilitates these activities. The aim of this study is to investigate which variables are most effective for urban planners and management to promote the participation in active travel behavior by the older population. To do so we investigate the participation of the aging population in walking and bicycling activities as a function of socio-demographics and physical and social environmental characteristics. Revealed individual travel choice data including all walking and bicycling trips for one day from a random sample of 4396 respondents in the age category of 65 years and over in the Netherlands were analyzed. For each trip, a large number of explanatory variables was available including information on mode choice, purpose of the trip, travel distance, travel duration, weekday, and number of trips per day. In addition, socio-demographics such as gender, age, income, whether the person has a partner, owns a bike, has a drivers license, and car possession were included. The data was fused with social and physical environmental characteristics at the neighborhood level including, urban density level, accessibility of shops, green/recreation areas, and restaurant/cafes, and indicators describing the safety and social cohesion. Mixed logit models (ML) have proven to be a useful tool for predicting active transportation choices and assessing policy measures and planning interventions. However, including such a large number of attributes and covariates in the model and finding meaningful interactions with the mode alternatives is a challenging task. Because variables are often highly correlated and the structure of their relationships is typically not clear (e.g., mediating effects, interaction effects, etc.) model variable selection and defining an appropriate structure for explanatory variables typically is difficult. A Bayesian belief network (BBN) approach can overcome such difficulties by deriving and representing all direct and indirect relations between variables by using a network learning algorithm. The network learning involves two main tasks: first learning the structure of the network and then finding the parameters (Conditional Probability tables) for that structure. However, although BBN is useful in discovering the appropriate data structure among a set of variables it is less well suited to predict outcomes of specific dependent variables. In this paper we therefore analyze the rich available data set using both alternative approaches. By comparing and integrating the outcomes of these analyses we can support more informed decisions about variable and model selection as well as provide guidance for specific urban planning and management interventions to promote active transportation choices by the elderly.
Original languageEnglish
Title of host publicationPaper presented at the International Choice Modelling Conference(ICM2013), 3-5 July 2013, Sidney, Australia
Place of PublicationSydney, Australia
PublisherEindhoven University of Technology
Pages1-15
Publication statusPublished - 2013
Eventconference; International Choice Modelling Conference 2013; 2013-07-03; 2013-07-05 -
Duration: 3 Jul 20135 Jul 2013

Conference

Conferenceconference; International Choice Modelling Conference 2013; 2013-07-03; 2013-07-05
Period3/07/135/07/13
OtherInternational Choice Modelling Conference 2013

Fingerprint Dive into the research topics of 'What drives active transportation choices among the aging population? Comparing a Bayesian belief network and mixed logit modeling approach'. Together they form a unique fingerprint.

Cite this