Better, More Reliable Transportation Statistics
Transportation researchers and policymakers have an abundance of data to work with, and the volume is growing each day. But simply having data available does not, in itself, lead to meaningful insights—tools are needed to interpret the numbers in ways that can support decision-making.
With the help of a Maryland Transportation Institute Seed Grant, two UMD professors—Cinzia Cirillo of the Department of Civil and Environmental Engineering and Partha Lahiri of the mathematics department and the Joint Program in Survey Methodology—are demonstrating how Bayesian statistics can be used to parse a wide variety of transportation-related data.
Their joint research project, A Bayesian Data Science Methodology for Transportation Statistics at Granular Levels, was among six selected during the first round of MTI Seed Grants, announced in late 2018.
“We’re interested in all kinds of transportation data,” Cirillo explains. “That can include the National Household Transportation Survey, the American Community Survey, census data, or even income data from the IRS. The goal is to develop a tool that researchers can use to harness data from all these sources and apply it to a research problem.”
Using Bayes’ Theorem, named after the 18thcentury English statistician and philosopher, statisticians can systematically update information as new evidence comes in. “Simply put, Bayesian statistical methodology draws its strength from multiple structured and unstructured big data sources, and quantifies the associated uncertainties. It has great potential for providing reliable transportation statistics,” Lahiri said.
By using Bayesian tools, Cirillo said, researchers can better assess the amount and type of information contained in data sets, such as surveys, as well as the kinds of predictions that can be made.
“For example, policymakers grapple with the problem of how best to evacuate people from low-income areas during an emergency or disaster," Cirillo said. “With our method, we’re not only able to pinpoint the locations of low-income individuals very precisely, at the level of census tracts, but also identify which individuals do not own cars and need to be picked up.”
A similar approach, she said, has been applied to issues such as traffic congestion—for example, by producing more accurate estimates of travel times. Cirillo and Lahiri have already received recognition for prior work in this area, having co-authored an award-winning paper that analyzed traffic patterns and used them to project future trends.
The MTI Seed Grant program is expressly designed to foster such collaborations, reflecting a growing awareness that interdisciplinary work is not only desirable, but essential, Cirillo said.
“Increasingly, the issues we’re looking at in transportation—and, indeed, in many other fields, cannot be addressed without strong interdisciplinary collaboration,” she said. “That hasn’t always been the norm in academia—but the paradigm is changing. And MTI is helping to spearhead that change.”
“In an age of evidence-based policy making, demand for reliable statistics at granular levels is steadily increasing,” Lahiri said. “Statistical methodology is increasingly playing an important role in helping scientists design experiments and analyze complex data. Our collaborative work, using Bayesian methods to provide reliable statistics at granular geographical levels, is thus very timely."
“Moreover, it can be applied to other scientific and engineering fields as well.”
Published November 13, 2019