Learning Who Will Win the Election

• In Event: Challenges Associated with Media Coverage of Polls and Election Forecasts

Abstract

On average citizens forecast elections more accurately than voter intention polls, betting markets, experts and quantitative models, even before the campaign starts. How do citizens do it? How do they learn who will win the election? And how much do they learn before and during the campaign? To answer these questions, I compare two Bayesian learning models: one assuming a stable informational environment, the other a changing one. The comparison uses the 2010 British General Election Survey. I take advantage of its unique survey design – a rolling cross-section combined with a panel -- to identify both the nature of the informational environment, and the amount of learning before and during the campaign. I find that the informational environment is stable, and that citizens learn the most before the campaign. Taken together both findings explain why citizen forecasting accurately predicts elections well in advance.

Author

• Andreas Erwin Murr, University of Warwick