Challenges in credibly estimating the travel demand effects of mobility services.

Mobility services including carsharing and transportation network company (TNC) services have been growing rapidly in North America and around the world. Measuring the effects of these services on traveler behavior is challenging because the results of any such analysis are sensitive to how (1) outcomes are measured and (2) counterfactuals are constructed. The lack of good control groups or randomization of assignment leaves lingering uncertainty over the contributions of selection bias and treatment effects to reported differences in travel behavior between users and non-users of these services. This paper reports on two approaches for measuring the effects of mobility service adoption on travel rate and car ownership. We first tried a pretest-posttest randomized encouragement experiment to deal with the shortcomings of poor control groups. Then, we turned to the approach of self-reported effects based on hypothetical controls to investigate whether variations in survey question presentation could influence respondents' answers and thus lead to changes in estimated effects. The data to conduct this study came from two sources: a panel survey administered by the authors at the University of Washington (UW), and a survey by Populus Technologies, Inc. (Populus). Various statistical tests were applied to analyze the data, and the results highlight the pivotal role that the research design plays in influencing the outcomes, and manifest the fundamental challenge of establishing credible estimates of the causal effects of adopting mobility services on travel behaviors. • Establishing valid estimates of causal effects of mobility services is challenging. • A randomized encouragement experiment in a pretest-posttest format offers promise. • Question framing is shown to influence self-reported effects of mobility services. • Research design plays a pivotal role in influencing estimated outcomes. [ABSTRACT FROM AUTHOR]