With all of that data in hand, the researchers made a pair of inferences about each smartphone user.
First, they estimated where each participant lived, based on the location of their devices between 1 a.m. and 4 a.m. during the three weeks preceding Thanksgiving. In doing so, they identified about six million “homes.”
Then the researchers guessed how each person voted by assigning them a probability score based on the election results for their home precinct. (The assumption they made was that the more heavily a precinct leaned toward a candidate, the more likely one of its residents supported that candidate.)
With an assumed home and a likely political affiliation, Mr. Chen and Mr. Rohla then reviewed how the participants traveled on Thanksgiving.
They took some precautions, though. For one, the researchers limited the analysis only to those who started and ended Thanksgiving Day at home because people who traveled farther most likely had little say over how long dinner would last.
To control for other factors, such as demographics, distance and travel time, they also compared dinner length between people who charted similar paths on Thanksgiving, traveling to and from the same general area.
The results were consistent: At the county, ZIP code and sub-ZIP code levels, people who traveled to an area with opposite political leanings to their own spent less time at Thanksgiving dinner than those who did not.