Multimodal Transportation Choices and Health: Exploratory Analysis Using Data Fusion Techniques

Abstract

This study demonstrates the feasibility of fusing large-scale travel and health surveys and uses the new comprehensive data set generated to model the relationship between health and multimodal (walking, biking, transit, and vehicle usage) long-term (weekly, monthly, and yearly) travel choices. Two measures of health, the body-mass index (BMI) and a self-assessed physical health score (SAPHS), were fused from a health survey onto a travel survey at the disaggregate (individual) level. The probabilistic record linkage software, Link Plus, was used for the data fusion purposes. The methodology was validated by using the eating and health module (EH) of the American Time Use Surveys (ATUS). Subsequently, the algorithm was used to match the health information from the ATUS to the National Household Travel Surveys (NHTS) of 2008 to 2009, and the resulting master data set was used to develop models for multimodal travel choices and health. The statistical analysis indicates that although increasing walking and transit use were associated with better health (relative to nonusers of the mode), those with the highest levels of walking and transit use were also found to be in poor health relative to moderate users of the mode. Similarly, those at the two ends of the vehicle miles traveled spectrum (first and fourth quartiles) had higher BMI compared with those in the middle of the spectrum. There were no statistically significant effects of weekly bike trips on health measures. Overall, this study is envisioned as a proof-of-concept of how data fusion techniques may be used to integrate multiple data sets to facilitate a comprehensive study of multimodal travel choices and health.

A Comprehensive Data Fusion to Evaluate the Impacts of COVID-19 on Passenger Travel Demands: Application of a Core-Satellite Data Collection Paradigm

“…The implicit fusion method has been used in a handful of studies in the transportation survey literature (27,(30)(31)(32)(33). For example, Lugo and Srinivasan (33) used the method to match health information from the American Time Use Survey (donor) to the National Household Travel Surveys of 2008-2009 (receptor).…”

“…This is because the true value of the target variables of the recipient units is rarely known (hence the necessity of the fusion). Some studies, like (33), have tried to circumvent this issue by validating their data fusion method based on a subset of the donor data, which is treated as the pseudo-receptor sample for the validation process. However, such validation should be interpreted cautiously.…”