Examining the Impacts of Residential Self‐Selection on Travel Behaviour: A Focus on Empirical Findings

Abstract

Numerous studies have found that suburban residents drive more and walk less than residents in traditional neighbourhoods. What is less well understood is the extent to which the observed patterns of travel behaviour can be attributed to the residential built environment (BE) itself, as opposed to attitude‐induced residential self‐selection. To date, most studies addressing this self‐selection issue fall into nine methodological categories: direct questioning, statistical control, instrumental variables, sample selection, propensity score, joint discrete choice models, structural equations models, mutually dependent discrete choice models and longitudinal designs. This paper reviews 38 empirical studies using these approaches. Virtually all of the studies reviewed found a statistically significant influence of the BE remaining after self‐selection was accounted for. However, the practical importance of that influence was seldom assessed. Although time and resource limitations are recognized, we recommend usage of longitudinal structural equations modelling with control groups, a design which is strong with respect to all causality requisites.

Acknowledgements

Chandra Bhat, David Brownstone, Hani Mahmassani, David Ory, Deborah Salon, Tim Schwanen and Ed Vytlacil, as well as several anonymous referees, have helped clarify some ideas and improve this work. Any remaining errors of fact or interpretation are, of course, our responsibility.

Notes

1. A report (Cao et al., 2008) contains additional detail related to both papers, which could not be incorporated into journal‐length articles.

2. Of course, these biases are also possible with the design of the self‐administered questionnaires from which the data for quantitative analyses are often collected. However, some scholars (e.g. Dillman, 1978) suggest that all else being equal, the extent of at least the latter three biases could be more severe in the case of direct questioning, where the body language and tone of the interviewer can offer additional cues to the participants, and where (even in the case of a prepared script or set of questions) the interviewer generally has a certain amount of discretion over the spontaneous digressions that the interview might take.

3. They incorporated the IV approach into a technique known as a ‘two‐part model (2PM)’, a variant on the Heckman sample selection model discussed in the next section.

4. Because this approach was not included in Mokhtarian and Cao (2008), we provide more detail on the methodology than we do for the other approaches.

5. The ± sign on w in the equation for RC∗ indicates that a given unobserved factor could affect the sensitivity to a BE trait in one direction for RC and in the opposite direction for TB, depending on the definitions of each measure.

6. Mokhtarian and Cao (2008) only discuss this methodology in passing, since at the time of writing, they were not aware of an empirical application of the approach.