Understanding Travel using Anonymized Cell Phone Data
Highlights from NCHRP Research Report 868: Cell Phone Location Data for Travel Behavior Analysis
· TRB recently released a report on how anonymized cell phone signal data can be used to measure observed travel.
· Planning agencies across the country maintain travel forecasting models to quantify the impacts of population and economic growth on regional transportation systems.
· These models have traditionally been developed and updated using surveys of household travel.
· Travel demand models are not updated frequently in part because of the high costs of travel surveys, that are often conducted 10 or even more years apart.
· As locational data from cell phones become more ubiquitous and continuously available, TRB and its partners evaluated whether anonymized location data can be used to assess traveler behavior.
· The study found that the travel estimates based on anonymized cell phone data can be comparable, with some caveats, to results of models developed with household surveys.
· However, the study also found several challenges with using cell phone data, including spatial inaccuracies and a lack of critical demographic information used to conduct equity assessments.
__________________________________________________________________
Since the 1950s , public agencies have used models to forecast travel demand and plan transportation system improvements to address observed and projected travel demand. Historically, agencies develop the models using behavior reported by travelers in household surveys. These surveys are expensive, cumbersome and can be collected 10 to 20 years apart. Since it is not uncommon to observe changes in the underlying travel behavior between surveys reflecting, for example, technology gains that lead to new modes of travel. In these cases, models built using dated input data are likely to result in forecast errors. Over time, surveys have become increasingly expensive to administer and difficult to conduct, due to lower participation rates among possible respondents. At the same time, passive “big” data have become more common with technological advancement and increasing use of cell phones. Such anonymized data have some of the key building blocks necessary to develop, or update, travel models.
TRB selected Cambridge Systematics and MIT to undertake a study on how anonymized cell phone data could be used to update these forecasting models. For the Boston metropolitan area, the study team had access to: detailed anonymized cell phone data; an existing travel model; and data from previous household surveys. The resulting models of travel movements from each of these datasets were compared with each other. The cell phone data included Call Detail Records (CDRs), which includes phone calls, texts, and emails that are sent and received as well as information reflecting a user’s web access activity. CDR data also comes with temporal and locational time-stamps that allow an analyst to track movements over time.
Are the estimates developed with cell phone data comparable to traditional travel model estimates?
There are some intriguing similarities at a higher level of geography and for a limited number of activity measurements. The outputs, travel predictions, and summary results produced based on the cell phone datasets were similar to the same measures produced by models that used household survey datasets.
Ultimately, the value of models is their ability to predict travel in response to different socioeconomic and level of service scenarios. Otherwise, the data are merely descriptive, not predictive. Since cell phone data lack demographic information and other variables that are known predictors of travel, modeling using these data is not easy.
What are the problems with using location data from cell phones to develop models that predict travel?
·Cell phone data can only infer and can’t observe:
- trip ends,
- the purpose of trips, or
- characteristics of the traveler.
· It may be difficult to infer the mode used when traveling; the presence of a cell phone on board a bus may be indistinguishable from a cell phone traveling in a car.
· A cellphone traveling in a car could be associated with the driver or with a passenger.
· The inferred location of a trip end may actually be the location of an intermediate stop.
· Due to spatial inaccuracies, it is difficult to infer detailed activities relying just on stop duration and location. For example, a stop at a multi-tenant retail location could be for coffee, to pick up dry-cleaning, or to pick up a child from daycare — each of which take roughly the same time.
What is the future of using cell phone data?
Technology is changing rapidly; the location-based industry is already switching to cell phone technology that is more advanced than the technology used in this study. The use of Location Based Service (LBS) data is becoming increasingly common, reflecting the improved spatial accuracy in these data and the user approval required to generate and share these data. However, several of the lessons from this study remain valid:
· Passive big data, whether old CDRs, new CDRs or LBS, can only infer, not report, trip ends, unless they are supplemented with travel surveys.
· Parameterized filters are necessary to differentiate between true trip ends and traffic stops in congested urban areas.
· Anonymized LBS datasets lack demographic information, a good feature that protects individual privacy.
· Surveys can produce detailed information about travel purpose, characteristics of the trip end, and traveler attributes. The trade-off can be summarized as the detailed information available in travel surveys versus the volume of data available from CDRs.
· A reasonable approach for future work may be the development of detailed travel models from surveys that can be periodically updated relying on the analysis of cell phone data.
· Many researchers, including our own Data Analytics Team at Cambridge Systematics, are employing advanced machine learning, behavioral econometrics and clustering algorithms to improve the quality of anonymized cell phone data products and develop accurate transportation models.
If you’d like to learn more on this topic, you can find the full report published here or listen to the interview among our research team here. You can also reach out to the authors of the research at the contact information below.
Contributors:
Kimon Proussaloglou, kproussaloglou@camsys.com
Daniel Beagan, dbeagan@camsys.com
Anurag Komanduri, akomanduri@camsys.com
