Mixing family and politics has always been fraught. I know—my mother was a Democrat, my father a Republican. The night Jimmy Carter won the presidency, dad slept in the guest room. For the U.S., the bitter campaign that ushered in Pres. Donald Trump in 2016 was a lot like that of 1976 in my house. Many families were politically divided, and the calendar forced the issue: The cherished American holiday Thanksgiving came just days after the election.

Anecdotal reports suggest family feasts that year were less festive than usual, with many Americans struggling to sit across the table from relatives whom they knew had voted for a candidate they loathed. Now there is hard data showing political polarization caused quite a few people to skip the pie. A new study published this week in Science reveals families with mixed politics spent 20 to 50 minutes fewer at the table than politically like-minded groups. Even the amount of the difference was partisan: Republicans left earlier than Democrats (some by more than an hour); Democrats were more likely not to go at all. The effect was three times stronger in areas with heavy political advertising. Overall, partisan differences cost Americans 73.6 million person-hours of family time that Thanksgiving, the study says.

Political psychologist John Jost of New York University, who was not involved in the research, finds the study intriguing. “In recent years political scientists have done an admirable job of documenting the political costs of asymmetric polarization,” Jost says. “This new work suggests that there may be social and personal—even familial—costs as well.”

Study co-authors Keith Chen, a behavioral economist at the University of California, Los Angeles, and Ryne Rohla, a graduate student in economics at Washington State University, were initially interested in the nation’s growing geographic polarization. “A dwindling share of Americans live anywhere close to somebody who voted differently,” Chen says. According to the Pew Research Center, Americans were more ideologically divided—with deeper antipathy—in the run-up to the election than at any point in the previous two decades. “We started thinking about Thanksgiving as a prime American civic tradition that might be able to break us out of these bubbles,” Chen says. But the researchers knew from experience and media reports that it had been a rocky holiday for many in 2016, after Trump’s shock win over Democratic contender Hillary Clinton.

To measure the effects of political disagreements on the festivities, Chen and Rohla used communication and location technology. A company called SafeGraph provided smartphone location data, gathered via partner applications such as weather or navigation apps that ask permission to log coordinates. The resulting anonymous, aggregated data sets reported the locations of some 10 million smartphones at various points in time. Chen and Rohla overlaid that data with voting tallies by precinct, each representing about 200 to 1,000 people who used the same polling place.

Then the researchers made some assumptions based on probability. “We take our best guess at what precinct you live in by asking where your smartphone is from 1 A.M. to 4 A.M…most mornings,” Chen says. “Say that tells us that you’re 80 percent likely [to be] a Clinton voter. We can then look at where you spent Thanksgiving dinner—and importantly, how long you spent at Thanksgiving dinner.” Or more specifically, where was your smartphone from 1 P.M. to 5 P.M. on Thursday, November 24, 2016? Only those who made day trips relatively close to home were included. At the most basic level of analysis, Chen and Rohla noted Family A traveled to eat with Family B. Then the researchers computed the likelihood of whether the families agreed or disagreed politically, and asked whether or not the length of dinner seemed to vary with that political accord or discord.

To make sure the effect they identified was political, the researchers took several additional steps: First they compared meal lengths for matched pairs of families who lived within a mile of each other, and had traveled to eat with host families who lived equally close. Of these, families with a higher probability of tearing into each other over the turkey did indeed prove more likely to leave early. Next the scientists compared Thanksgivings in 2016 and 2015. Families who were more likely to be politically distant yet ate together in 2015 were less likely to do so in 2016. Finally, Chen and Rohla considered the effect of political advertising. Voters in states such as New York and California saw almost no political ads. If you lived in Orlando, Fla., on the other hand, more than 20,000 political television ads aired in your media market over course of the campaign. In such saturated areas the diminishment of time together at Thanksgiving tripled. “It’s almost like this drug of political ads is fuel on the fire,” Chen says.

Some critics of the study note a presidential election year is different in many ways from the years on either side—so comparing 2015 to 2016 would be of limited use. Chen and Rohla do not disagree. They plan to analyze the aftermath of the 2018 midterm elections, but they recognize that 2020 and subsequent presidential election years will better clarify what happened in 2016.

Others have expressed concern the use of smartphone data could be considered an invasion of privacy. Chen and Rohla point to the anonymity of the data and the fact the applications involved require permission. “They do have to allow the location,” Rohla says. “My feeling is that math-geolocated data is going to be at the forefront of a lot of social science going forward.”

Methodology aside, the authors say they have captured a trend they would like to see disappear. “I hope the effect is short-lived and that American families find a way past this,” Chen says. Rohla can attest at least one family—his own—did. “Everyone learned their lesson in 2017,” he says. “It was much better.”