UC Davis Institute of Transportation Studies

In this dissertation, the author estimates effects on travel behavior of two specific technological innovations – emerging shared mobility services and telecommuting – using publicly available travel surveys. These surveys are cross-sectional and observational in nature, which leads to the potential for (1) selection bias due to observed and unobserved differences in characteristics between program participants and non-participants; and (2) reverse causality bias arising because of potential influence of the travel behavior outcome of interest on the propensity to enroll in the program. The methodological framework combines established methods from both statistical and econometric literature to draw causal inferences. The key innovations in this dissertation are the combination of diverse methods to address the joint occurrence of various biases, and their specific empirical applications. The author also compares the results of alternative methods.

In the first study (Parts II & III of the dissertation), the author estimates the effect of carsharing on travel behavior, using data on employed San Francisco Bay Area respondents from the 2011-12 California Household Travel Survey (CHTS). The investigator finds that 80% of the observed difference of 0.9 units in average vehicle holdings between carsharing non-members and members may be explained by self-selection and reverse causality biases. The remaining difference of 0.17 units reflects the estimated effect of carsharing, which is the equivalent of shedding one vehicle by about one out of every six households whose member(s) are enrolled in carsharing. The effect on transit usage and walking and biking frequency is positive, albeit small and statistically non-significant.

In the second study, the researcher estimates the effect of the adoption of telecommuting on travel behavior for full-time employed respondents with a fixed work location outside home, using data from the annual United Kingdom National Travel Surveys for the years 2009 to 2013. On average, telecommuters are observed to travel more than non-telecommuters. However, after accounting for the observed differences in traits and tastes between the two groups using a linear regression model, the differences fade to (nearly-) insignificant levels. Further control of self-selection bias arising from unmeasured differences in “relevant” characteristics leads to the conclusion that telecommuting has a substitution effect on both commute and non-work travel.

The results are broadly consistent with those of earlier studies, which, unlike this study, are based on purpose-built proprietary surveys explicitly designed to evaluate effects of either of the two programs. Although the data collected through those other means are still observational in nature, various biases identified in this dissertation may be addressed by questionnaire design, including retrospective reporting of travel behavior before and after enrollment in the program. By implicitly assuming that the unobserved influencers of both program adoption (either telecom-muting or carsharing as the case may be) as well as travel behavior do not change over the course of the evaluation (an assumption which may or may not be true), those prior studies estimate effect by measuring change in travel behavior before and after program enrollment relative to a control group. Unfortunately, such surveys are expensive, proprietary, and usually one-off studies.

Large regional travel surveys, on the other hand, are publicly available, leading to the potential for replicability and involvement of multiple research teams. Further, these surveys collect information about broader travel behavior patterns and yield samples that are often larger and more representative of the general population. However, the cross-sectional and observational nature of these surveys creates the potential for joint occurrences of various biases identified in this study, which makes it necessary to adopt methodologies that correct and control for these biases when estimating causal effects. The author hopes that the methodological frameworks adopted in this study will provide an example that other researchers can use to analyze various programs in transportation using publicly available travel surveys, and that the causal inferences drawn will offer a sound basis for policymaking.