TREC Friday Seminar: Travel Demand Analysis on Transport Disadvantaged Communities

I attended the October 30th seminar at which Dr. Tierra Bills, professor at Wayne State University, presented case study findings on attempts to include survey data from hard-to-reach, disadvantaged communities that would better inform travel demand modeling.

Dr. Bills identified 4 different types of new transportation technologies that are "big data" generators being used to inform travel demand modeling: Navigation tools (e.g., Google Maps, Waze), Car Share Services (e.g., Car2Go, Zipcar), Ride Hailing Services (e.g., Uber, Lyft), and autonomous vehicle programs being tested in Michigan specifically. Dr. Bills began with the question of where "big data" informed Travel Demand Analysis fits in and how we know about its quality? These questions were tested in 2 case studies: Benton Harbor, MI Travel Needs Survey from 2019, and a Detroit Microtransit Survey which is currently being conducted.

Dr. Bills explained that travel demand models are tools used to understand travel behaviors regarding destinations, routes, mode choices, and how they change with revisions to the built environment. The nagging question with these models is whether or not they contain data that is adequately representative data from disadvantage communities, identified as people with low incomes, un/underemployed people, transit dependent people, the disabled, and the elderly). At every stage of the TDM process (Data collection, data cleaning, model estimation, and analysis/decision-making), the presence or absence of representative data from disadvantaged populations might impact the quality of the process's ultimate findings.

Travel surveys have several challenges (response biases and sampling errors) that are typically mitigated by increasing sample sizes, stratified sampling, and anticipatory sampling relative to expected response rates from certain demographics. Often, "big data" is used to help ameliorate issues with low sample sizes, so Dr. Bills asked the question: How well are certain disadvantaged groups represented compared to others in big data samples? Her conclusion was that "big data" sources appear to have similar problems with relative representation - that is, underrepresentation of disadvantaged demographic groups - as traditional survey data methods.

The main case study that was relevant to determining the answer to this question was Dr. Bills's Benton Harbor, MI Travel Needs Survey. Benton Harbor is the headquarters of Whirlpool, an appliance manufacturer. It's a city around 10,000 in population, with about 50% of its population living at or below the poverty line. Employment density is on the decline, and the city is in close proximity to significant agricultural industry jobs. Residents were reportedly struggling to access employment, education, healthcare, and shopping services. The local transit authority was looking for data that could help inform them on how to improve transportation services in order to improve access to employment, education, shopping, etc.

Dr. Bills tailored the methods of surveying the population with the expectation that disadvantaged groups would be harder to reach and would require additional measures to compile a more representative sample. The campaign involved using both paper and online surveys. Notices about the survey were posted in Dial-A-Ride shuttles, in local newspaper and radio announcements, and on Facebook. In-person workshops were also held.

The survey methods got conclusive results. Relative to prior Michigan DOT surveys that had been supplemented with "big data" sources to increase sample size, Dr. Bills's survey reached a population with a much smaller share of auto users and a much higher share of transit users. Paper surveys, in particular, had a significant effect, with paper survey respondents being much less likely to travel by personal vehicle or carpool, and were much more likely to be elderly or disabled compared to online survey respondents. The survey also asked specific questions to identify clear accessibility needs. What were respondents' access levels to local grocery stores, general shopping centers, and their workplaces? Had any lost jobs because of their lack of reliable transportation? In each case, 20-25% of respondents answered affirmatively and in ways that enabled the team to spatially map poor-access areas and begin analyzing the reasons for their poor transit accessibility.

To summarize, "big data" was not able to fill in representation gaps for disadvantaged groups in the studied transportation surveys. Specifically targeted outreach strategies directed at disadvantaged groups were, however, able to more helpfully represent Benton Harbor's transportation issues relative to the needs of those for whom access was inequitable.

Seminar Link: www.youtube.com/watch?v=X0nztBtaygc