Previous studies testing the association between the built environment and walking behavior have been largely cross-sectional and have yielded mixed results. This study reports on a natural experiment in which changes to the built environment were implemented at a university campus in Hong Kong. Longitudinal data on walking behaviors were collected using surveys, one before and one after changes to the built environment, to test the influence of changes to the built environment on walking behavior.

Hong Kong is a densely populated and transit-oriented Chinese city, which provides an ideal urban environment with which to study the various successful facets of land use policy as a model for potential replication to curb increasing car use in other Chinese cities. We examine the commuting mode choice of 203,900 households living in 4768 street blocks in Hong Kong from 2011 census. A street block is the smallest planning unit, made up of one or more housing estates with a homogenous built environment and socioeconomic status. The built environment is measured using the five Ds framework, an international dimensioning framework for classifying and measuring attributes of the built environment for physical activity and travel behaviors. Generalized, multi-level mixed models were applied to detect the associations between travel choice and built environment characteristics, while adjusting for socioeconomic status. Design and destination accessibility had greater effects on the choices to walk and take public transport than on the choice to drive. Density and diversity had only marginal effects on mode choice. Unexpectedly, distance to the urban center had the opposite effect on automobile use to that found in Western studies. Hong Kong residents living close to the urban center were more likely to drive for commuting trips. The contrasting findings between our study and Western studies suggest that the associations between a high-density built environment and travel choice vary with urban context.

We used the first metro system in a developing city as a natural experiment to investigate the causal inference in the new metro's impact on modal shift and active travel. The treatment group was formed by residents from neighbourhoods located within the 800-m walking distance to new metro stations. The first control group was formed by residents lived 1.6 km away from and outside of walking distance to the nearest station, and the second was 5 km away and outside of cycling distance. The groups were determined by local transit-oriented planning practice and empirical studies on active travel. Of the 5627 participants who had finished a baseline travel behaviour survey before new metro launched, 1770 returned and completed the follow-up survey a year after the metro's operation, which consists of 833 cohort participants in the treatment group and 937 in the two types of control groups. We used a difference-in-difference method to make before and after comparisons of travel behaviour changes between treatment and control groups. Our longitudinal data analyses revealed diverse travel behaviour changes. In general, people who used to take bus have adopted metro. The average metro usage was 30.9 (28.8-33.3) minutes daily for work trips and 16.6 (14.9-18.7) minutes daily for non-work trips. Walking time decreased 19.7 minutes at most (p < 0.001), and cycling decreased 22.1 minutes daily (p < 0.001). Car and e-bike usages remained largely unchanged before and after new metro, without difference between treatment and control groups. The natural experiment study provided the first empirical evidence in a developing city context on causal inference in new metro's impact on active travel. A new metro does not necessarily promote active travel increase or car use reduction, calling for caution in making general assumptions about the effects of urban rail transit investments. We suggest local urban and transport planning knowledge could be useful in designing and explaining the complex natural experiments in transport and health.