In the last two US presidential elections, polling has come under scrutiny. In 2016, many of the most rigorous polls failed to predict Donald Trump’s win. In 2020, while most polls predicted Joe Biden’s win, most overestimated significantly the margin by which Biden would beat Trump. Can anything be done to improve the accuracy of electoral polling in the future?
In a post-election op-ed for The Conversation SFI Professor Mirta Galesic and Wändi Bruine de Bruin at USC Dornsife describe their polling research with colleagues Henrik Olsson, SFI External Professor, and Drazen Prelec at MIT. The team found that if pollsters start to ask different kinds of questions, their predictions become more accurate. Traditional polls tend to ask directly how individuals intend to vote. Yet polls that ask questions about how people think members of their social circle or state will vote, tend to predict results with far greater accuracy. This is true not only for US elections but for elections around the globe. If pollsters can tap into what researchers call “wisdom of crowds,” they might be able to strengthen the predictive power of future electoral polls—in America and around the globe.
Read the op-ed in The Conversation (November 18, 2020)