Today, Donald Trump will be inaugurated as US President again—an election outcome that has surprised many, including pollsters. Perhaps a little less so for Mirta Galesic and her team at CSH. And here’s why.
A French investor bet millions on Trump’s victory in the US election and the popular vote in November – ultimately winning $85 million. He told The Wall Street Journal that traditional polls failed to capture the effect of the ‘shy Trump voters’ – meaning Trump supporters either hesitated to tell pollsters they supported him or did not participate in surveys at all. But why was he so confident?
Mirta Galesic and Henrik Olsson who lead CSH’s Collective Minds research group, have been studying (pre-)election surveys for years. They have shown in multiple elections that questions asking whom respondents themselves would vote for yield less accurate predictions than so-called “wisdom of the crowd” questions, where respondents estimate how others will vote. Examples include: “Who do you think will win the election?“, “What percentage of voters in your state will vote for different candidates?”, “Whom do you believe your neighbors will vote for?“, or “What percentage of your friends, family members, and other acquaintances will vote for different candidates?“
OUTPERFORMED IN EIGHT ELECTIONS
“We have shown that the latter – the so-called ‘social circle question’ – averaged across representative national samples, produced better predictions than the question asking those same people who they will personally vote for. We have found this improvement in eight elections so far – the US elections 2016, 2018, 2020, and 2024, the 2017 French and Dutch elections, 2018 Swedish elections, and 2024 Austrian elections,” says Galesic.
In 2024, Mirta and her team conducted a survey on the US election and a survey just before the Austrian election in September (unfortunately, the one in Austria was a last-minute survey, so they couldn’t make predictions before the election). In the Austrian election 2024, in a survey on 990 participants, average absolute error of predictions of popular vote across 5 parties (+ other) was 1.9 for the social circle question, compared to 3 for the own intention question.
“Interestingly, in these two elections (but not the earlier US elections in which we asked that question), another “wisdom of the crowd” question produced even slightly better results: the “population expectation” question. This question asks people to estimate the percentage of voters who will vote for different candidates. This is a slightly refined version of asking people who they think will win the elections that has been shown for decades to produce very good predictions of election winners,” Galesic explains.
“It’s especially important to note that within the same survey, predictions using wisdom-of-crowd questions were always better than those based on the standard question about personal voting intentions,” she adds.
WHY IS IT BETTER TO ASK ABOUT OTHERS?
Galesic attributes the advantage of asking about others rather than personal intentions to three key reasons: First, it provides information about people who might not otherwise participate in surveys. Second, it reveals social influence that could sway undecided voters toward a candidate at the last minute. Third, people may feel more comfortable indicating that others support an unpopular candidate rather than admitting it themselves.
“In the 2024 US election, all three reasons could have played a role,” Galesic says. “If supporters of a particular candidate are less likely to respond to polls at all or to respond honestly – as has been speculated about Trump voters – then the results of standard polls will be biased. Pollsters attempt to correct these biases by weighting their samples by demographic and other variables, but it is hard to account for all biases and weighting increases margins of errors.”
HOW COMMONLY ARE WISDOM-OF-CROWD QUESTIONS USED?
“It is interesting to us that this question is still not more widely used among pollsters. We have presented our results at survey research conferences but have not experienced much interest. We think this is one of the ‘sociology of science’ phenomena where researchers can be reluctant to change their methods, often for valid reasons. They have established methodologies and repeating the same questions every year enables them to track their error rates and fine-tune their methods. Still, it is quite puzzling that these so-called wisdom-of-crowds methods have not been adopted more widely in survey research – not only our social circle question, or the neighbor question, but also the well-established winner expectation question I mentioned earlier, or other variants of those,” says Galesic.
