Can Twitter Predict Elections?

In ancient times, oh about 50 years or so ago, we'd do man-on-the-street interviews to get a sense of who might win an election.  Then polls became the accepted way to predict an election outcome. 

Can Twitter do the job as well?

This study of the 2009 German federal elections suggests yes, it can.  While I unfortunately don't have access to the entire study, and therefore can't judge its methodological merits, according to the abstract the authors analyzed 100,000 tweets to find "the mere number of party mentions accurately reflects the election result."  Wow.  In other words, positive and negative mentions come up to roughly the same as the election outcome.

I've written about this before, in part about elections, in part about using Twitter to make predictions about what movies will succeed at the box office.   Does this mean Twitter will replace expensive polling and surveys?  Nope, not a chance.  A lot more needs to be done to see if this really works, and I can't imagine any serious political consultant going with tweets as the basis of an election strategy.  At least not yet.

But we already see a world in which Twittter is constantly analyzed along with more traditional data from focus groups to surveys to get a sense of the public's response to any number of campaigns, issues, or current events.   It's predictive power remains unproven, but I have a hunch it'll turn out to be fairly accurate, at least at the macro level.  What it won't do, at least yet, is allow you to break people into subgroups (women, Republicans, younger versus older) and explore more deeply what works for some people, what doesn't work for others.  So even if Twitter proves predictive, as of yet it's not particularly good in helping us understand the why of support.