Sign In

Communications of the ACM

ACM TechNews

Election Polling: When 'who Will Win?' Beats 'who Do You Like?'

View as: Print Mobile App Share:
A line of voting booths in Cuyahoga County, OH.

Surveys of voters on who they actually expect to win an election correlate more strongly with election results than surveys asking which candidate they like the most.

Credit: Marvin Fong/The Plain Dealer

Whereas recent poll numbers for U.S. presidential candidates are based on gauges of voter preferences, statisticians and political scientists are more reliant on the question, "Who do you think will actually win?" when it comes to predictive markets.  

Microsoft Research economist David Rothschild says asking voters' expectations of the actual winner "grabs a much larger slice of people's experience and knowledge, including a whole range of idiosyncratic facts," which otherwise cannot be measured.  

Microsoft's PredictWise project employs an aggregate of several Internet betting exchanges and prediction markets to create a running tabulation of the likelihood of something happening.  PredictWise established a 37-percent chance that Jeb Bush will be the Republican presidential nominee, while Donald Trump is currently the leading candidate with 28 percent in a running average of national preference polls.  

In a 2012 study, Rothschild and economist Justin Wolfers argued voter expectation is the more accurate data point, while a 2013 report on U.S. presidential elections found voter expectation polls failed to pick the winner "in only 8 percent of 214 surveys conducted from 1932 to 2012."  

Last year, Rothschild calculated the expectation question is 10 times more potent than the preference survey.

From The Wall Street Journal
View Full Article


Abstracts Copyright © 2015 Information Inc., Bethesda, Maryland, USA


No entries found