Enlarge (credit: flickr user: Kelley Minars)
In the run-up to the 2016 US presidential elections, a scuffle broke out about the election forecasts. While FiveThirtyEight predicted a one-in-three chance of a Trump win, other forecasts suggested a near-certain Clinton victory. “[T]he most popular and widely quoted website out there, fivethirtyeight.com, has something tragically wrong with its presidential prediction model,” wrote Evan Cohen at Huffington Post. “If you want to put your faith in the numbers, you can relax. She’s got this,” wrote Ryan Grim, on November 5, in a screed about FiveThirtyEight’s methods.
The widespread failure at predicting the outcome of the election has prompted a lot of analysis. Among the explanations is a simple one: polls are hard. The only way to know for sure how a bunch of people will vote is to hold the actual election. Everything else will involve extrapolating from a sample of people to the whole population, and that’s an imperfect process. A paper in Nature Human Behaviour this week finds some evidence for a method that might improve accuracy: instead of asking people how they plan to vote, ask them how their friends and family will.
You don’t have to lie for your friends
The standard version of election polls involves asking a simple question: which candidate do you plan to vote for? But there are a range of other questions that can be asked. One of these is asking people to express their voting intention as a probability: what is the percent chance that you would vote for a particular candidate?
Read 12 remaining paragraphs | Comments
