Estimating Heterogeneity in Travel Mode Choice Shifts with Causal Forests
This study applies causal forests to National Household Travel Survey data to quantify the heterogeneous causal effects of the COVID-19 pandemic on travel mode choices, revealing a significant shift toward car use that varies by trip distance, income, and gender.
Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Imagine you are a detective trying to solve a mystery, but instead of looking for a stolen jewel, you are trying to figure out why people are suddenly driving their cars more and walking less. In the world of transportation science, researchers have spent decades building maps to predict how people move. Traditionally, they used tools that were great at spotting patterns—like noticing that people with big houses tend to drive big cars. But spotting a pattern isn't the same as proving cause and effect. It's like seeing that people carry umbrellas when it rains and concluding that the umbrellas caused the rain. To really understand why something happens, you need to know what would have occurred if the situation were different. This is the tricky part of "causality." Usually, to prove cause and effect, scientists run experiments where they change one thing and keep everything else the same, like a chef testing a new spice. But you can't easily run a giant experiment on a whole city to see what happens if you suddenly make everyone stay home for a year. That's where a new kind of detective tool comes in: a method called "causal forests." Think of these forests not as trees with leaves, but as a massive, digital forest of decision trees that can look at millions of past trips, split them into tiny groups based on who the travelers are and where they went, and ask, "If this specific person had lived in a different time, would they have chosen a different way to travel?"
This paper takes that digital forest and plants it in the soil of a very real, very messy event: the COVID-19 pandemic. The pandemic was a giant, unplanned experiment that changed how the world moved. The authors, Rishabh S. Chauhan, Mahdi Ghadimi, and Lishun Liu Gaara, wanted to use their causal forest to figure out exactly how the pandemic shifted people's travel habits, not just on average, but for specific types of people. They didn't just ask, "Did people drive more?" They asked, "Did rich people drive more? Did women drive more? Did people taking short trips drive more?" By analyzing over 800,000 trip records from before the pandemic (in 2017) and after it had settled into a new normal (in 2022), they used this advanced machine learning method to isolate the true "pandemic effect" from all the other changes happening at the same time.
Here is what their digital forest found. When they looked at the average person, the pandemic caused a shift where the chance of driving a car went up by 1.86 percentage points, while the chance of taking public transit dropped by 0.38 percentage points, and the chance of walking fell by 1.57 percentage points. But the real story is in the details, because the forest revealed that these changes weren't the same for everyone. The biggest jump in car use happened for people taking very short trips (one mile or less), for households earning over $200,000 a year, and for female travelers. In fact, women shifted toward driving and away from walking much more dramatically than men did. On the flip side, the people who saw the biggest drop in public transit use were those with no cars at all, and the elderly (65 and older) barely changed their habits at all, showing a kind of "behavioral inertia."
The authors also used their tool to look for the most extreme groups. They found that a specific combination of factors—living in the Western US, earning between $25,000 and $49,999, and taking a trip of one mile or less—led to a massive 16.37 percentage point increase in car use compared to everyone else. Conversely, people with no household vehicles taking medium-length non-work trips saw their car use drop significantly. The study suggests that while the overall trend was a move toward cars, the "why" and "who" were incredibly complex. The researchers are careful to note that this is a snapshot in time and that their method relies on the data they had, but the results offer a much clearer, more nuanced picture than simple averages ever could. By using causal forests, they moved beyond just saying "people drove more" to understanding exactly which pockets of the population changed their minds and by how much, providing a powerful new way for city planners to see the hidden layers of human behavior.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.