Believe it not, pollsters are always looking for ways to refine our methodologies and make our science more precise. The last few presidential election cycles have presented the challenge of reaching so-called “shy voters”—those who (for whatever reason) are reluctant to either a.) engage with pollsters or, if they do, b.) tell them the truth of who they’re voting for. This phenomenon is heightened whenever Donald Trump is on the ballot.
To be fair, this issue has been a bit overblown—and mostly confined to national election samples. In 2024, for example, 48 of 50 state-based polling averages proved accurate. Still, the public perception is that pollsters “blew it” for the third straight cycle. And, as we know better than most, perception can become reality.
In any event, there is one polling technique that the academic literature suggests is highly accurate in election prediction: Social Circle Polling. Basically, this involves a line of questioning that asks respondents (irrespective of their own choice) who their family, friends and acquaintances are supporting in the election. I’ve always felt that a similar question has helped me in gauging election outcomes—asking voters (irrespective of their own choice) who they think will win the election. The Social Circle method widens this assessment from the personal to the community level—and seems to be even more predictive of outcomes.
This month, a real-world application of Social Circle Polling resulted in one individual cashing-in on the 2024 election. As detailed by Joe Nocera in a recent Free Press article, a person in France bet $85 million on a Trump victory based entirely on using this method in a series of ‘swing state’ polls he commissioned through the polling firm YouGov. Of course, this new spectacle of ‘prediction markets’ and betting on elections is a whole separate issue that needs to be addressed.
Nevertheless, adding this simple battery of “Social Circle” questions to election surveys would seem a prudent step for the polling industry. It may not be the panacea for making election polls more accurate, but both academic and real-world evidence suggests it can certainly help.