Beyond Reported Trips: A Bayesian Framework for Correcting Household Travel Surveys

Every day, millions of people move through cities, hopping from home to work, stopping for groceries, or taking a quick walk to a friend's house. Transportation planners rely on surveys to understand these movements, asking households to write down every trip they make in a single day. This information helps cities decide where to build new bike lanes, how many buses to run, and whether a new subway line is needed. However, these surveys have a long-standing weakness: people are terrible at remembering everything they did. When asked to recall a day's travel, most people forget the short, spontaneous trips—the quick errand to the corner store, the brief walk to the mailbox, or the stop to drop off a package. They remember the big, structured events like going to work or school, but the small, discretionary movements often vanish from memory. This creates a distorted picture of how people actually live, leading planners to underestimate walking and cycling and overestimate the time people spend in cars.

To fix this, researchers have begun using smartphones to passively track movement, recording exactly where people go without asking them to remember. But these phone-based studies are usually small and lack the detailed personal information found in large household surveys. A new study by researchers at the University of Toronto and the Massachusetts Institute of Technology bridges this gap. They developed a method to combine the broad reach of traditional surveys with the precise detail of smartphone tracking. By using a statistical approach that treats the phone data as a way to correct the memory gaps in the survey, they were able to reconstruct the missing trips that people forgot to report. The result is a more accurate map of daily life that keeps the rich demographic details of the large survey while adding the missing pieces of the travel puzzle.

The researchers focused on the Greater Toronto and Hamilton Area, using data from the 2022 Transportation Tomorrow Survey, a massive household survey that captures travel habits across the region. They paired this with two smaller, more detailed studies conducted in 2023 and 2025, where participants carried smartphones that automatically recorded their movements. The team noticed a clear pattern: the traditional survey showed people making about 2.06 trips per person on a weekday, while the phone data showed significantly more, around 3.16 to 3.83 trips. The difference was not random; it was systematic. People consistently forgot the short, non-essential trips. The researchers separated these trips into two categories: "skeleton" trips, which are the main anchors of the day like commuting to work or school, and "non-skeleton" trips, which are the flexible, discretionary movements like shopping or visiting friends. They found that skeleton trips were reported fairly accurately, but non-skeleton trips were frequently omitted, especially when one family member reported the travel for another.

To recover these missing trips, the team built a framework that acts like a statistical repair kit. They started with the large survey data as a baseline, representing what people said they did. Then, they used the smartphone data to update and refine those numbers. Imagine the survey data as a rough sketch of a day's travel, and the phone data as a high-resolution photograph that reveals the fine details the sketch missed. The researchers used a method that allows them to blend these two sources, treating the survey as a starting guess and the phone data as new evidence that corrects that guess. They did this separately for people who reported their own travel and those who reported for someone else, because the errors in reporting are different for each group. When someone reports for a spouse or child, they often miss even more of the small, intermediate stops than when they report for themselves.

The results showed a significant shift in the numbers once the correction was applied. For people who reported their own travel, the average number of trips per day rose from 3.05 to 3.90. For those who reported for others, the number jumped from 2.64 to 3.63. This increase was almost entirely driven by the recovery of non-skeleton trips. The researchers did not just add random trips; they carefully reconstructed the missing ones based on patterns found in the phone data. They determined what the purpose of the missing trip likely was, when it probably happened, how far it was, and how the person traveled. For instance, they found that the missing trips were disproportionately short and often involved walking or visiting friends, activities that are easily forgotten in a busy day. The corrected data showed a higher share of walking and a more even spread of travel throughout the day, rather than just the morning and evening rush hours.

The study also revealed that the nature of the missing trips depended heavily on who was reporting. Proxy respondents, who report for other family members, were much more likely to miss discretionary trips than those reporting for themselves. This suggests that when a parent fills out a diary for their child, they might remember the school run but forget the child's after-school stop at a park or a friend's house. By modeling these differences separately, the researchers could apply the right amount of correction to each group. The final corrected dataset retained the original structure of the large survey, meaning it still accurately reflected the population's age, income, and location, but it now told a truer story about how those people moved. The distribution of trip purposes shifted to include more shopping and recreation, and the timing of trips showed more activity during the middle of the day and in the evening, filling in the gaps that the traditional survey left empty.

This work offers a practical way to improve the data that cities use to make decisions. Planners often rely on these surveys to set goals for reducing car use or increasing walking, but if the surveys miss the short trips, those goals might be set too low. By integrating the passive data from smartphones, the researchers provided a method to see the full picture of daily travel without needing to replace the large, established surveys entirely. The corrected data suggests that people are more active and walk more than previously thought, particularly for short, local trips. While the reconstructed trips are statistical estimates rather than direct observations of specific individuals, they provide a much more reliable foundation for understanding travel behavior. The study concludes that this approach can help cities better plan for the reality of how people move, ensuring that infrastructure and policies match the actual, often forgotten, rhythms of daily life.