Can travel demand models predict cycling?

By Chris McCahill

Try asking a conventional travel demand model about bicycle trips and you might get anything from an educated guess to an error message. A recent study from Sweden, however, shows what it takes to fix them. The short answer is to make the models much bigger. That leaves an important question: is it worth it?

As the authors of this study explain, “large-scale transport models, which serve as the main tools for policy evaluation and cost-benefit analysis, are often designed for modeling of motorised travel modes such as private car and public transport.” For their study, published in Transportation Research Part A, the researchers developed a trip-based model for Stockholm—as opposed to more complex activity-based models—that predicts a 13.5 percent bicycle mode share (compared to 11.9 percent in the regional travel survey) and explains 74 percent of variation in observed bicycle volumes. Those are good results. Their mode also captures important differences in bicycle use among age groups, genders, and income levels.

[Figure 1. Predicted bike volumes throughout Stockholm. Source: Liu et al., 2020]

While their approach could help improve the conventional four-step model that many agencies still rely on, it also requires fundamental changes. The road network is much more detailed, and the traffic analysis zones are smaller than usual (250 by 250 meters, in this case). To predict cycling patterns, the network also needs information about bike facilities and the slope of roads, which the authors explain are critical for assessing bike infrastructure projects. In other words, it takes more data and presumably much more time to run.

This all raises questions about whether building on old travel demand models is the right path forward. First, despite how promising this study is, modeling future demand—the main purpose of these models—is much more difficult than simulating current demand. We recently learned that forecasted traffic volumes are often only marginally better than straight-line projections.

Similarly, most travel models assume future development patterns are fixed, regardless of future transportation investments. In the U.S., the resulting highway construction has chased outward growth while at the same time fueling it. The opposite could be true of cycling: we may underestimate potential bicycle demand if we don’t first imagine the most supportive land uses.

So, while travel demand forecasts were useful in building out regional highway systems across the globe, are they as relevant for stitching together urban bicycle and pedestrian networks? With the potential to understand actual bicycle travel using big data, modeling travel demand may not be as important as methods for identifying critical gaps and evaluating projects in terms of accessibility, which the current research could help with in many ways.

Chris McCahill is the Deputy Director at SSTI.