Election forecasting and the Facebookpocalypse

in Uncategorized by

This week, at least two academic reports have been released on Facebook… With very, very different messages.

On one hand, Princeton researchers forecasted that Facebook will be 80 percent obsolete by 2017, using epidemiological models of infectious diseases and anecdotes on MySpace to arrive at this unlikely conclusion.

Then, UMass researchers suggested that Facebook itself can be used to predict political election outcomes — circumventing more costly methods of forecasting such as polling or econometric analysis.

This analysis, published Tuesday in POLITICO, begins with the premise that in the 2012 House elections, “20 of the 33 most competitive races across the country were won by the candidate with a measurable Facebook fan engagement advantage.”

From these observations, Matthew MacWilliams and Edward Erikson developed a Facebook Forecasting Model based on statistics tracked by the social media site that peg a candidate’s level of engagement with a fan base as well as past electoral history.

Using this model, MacWilliams and Erikson were able to accurately predict election returns in eight of the nine Senate toss-up races of 2012. They also found that this model did just as well, if not better, than polling.

“Our model is based on a set of simple assumptions that combine certain statistics tracked by Facebook, which we believe measure campaign effectiveness, with basic electoral fundamentals, such as organizational and communication capacity, to forecast the results of congressional campaigns,” write MacWilliams and Erikson.

In short, this implies that a campaign’s level of social media engagement reflects its organizational strength and appeal to voters, and in turn contributes to the campaign’s ability to reach or mobilize new and greater support.

Beginning in April, eight months out from the November 2014 midterm elections, MacWilliams and Erikson will put their Facebook Forecasting Model to the test on Senate general election contests. So far, their model predicts Republicans potentially picking up one seat, but the real work will begin in the spring with predictions and updated reports published weekly.

The use of Facebook as a proxy for other types of forecasting data is fascinating. I think it is safe to say that such data will be just as relevant, if not more relevant, after Princeton’s predicted “Facebookpocalypse” of 2017.