The Voting with your Tweet experiment

Voting with your Tweet is an experiment in election prediction. It’s a work in progress and one step in a  broader research process that seeks to build a greater understanding of how to use social media to forecast elections.

Between now and Election Day on November 6, 2012, we will test whether, using only data from the Twitter feed, we can successfully predict who will win each of the 435 seats in the US House of Representatives. We have built a number of visualizations that are updated daily so that you can follow along.

At the same time, we hope to learn a lot more about how candidates and voters use social media during elections.

It’s already clear that there has been growth in Twitter’s importance to politics and political communication. What remains to be seen is if Twitter can be a reliable predictor of election results. We hope this experiment provides part of an answer to that question.

This experiment also marks one of the first attempts at predicting elections across multiple cycles with Twitter. In the two months prior to the 2010 Congressional election, we gathered about 250,000 Tweets about the candidates running in 356 Congressional races. This year, we’re using prediction methods built on those races to forecast who will win the 2012 election.

Whether this works depends on how successful our 2010 models are at tracking the actual development of the 2012 Congressional races. We have already noticed that we are predicting more Republican victories than national polls like those shown at the New York Times. There are several reasons that this could be true. For instance, 2010 was a highly successful year for Republicans, and a bad year for incumbents. If our models embed these biases, they may predict that Republicans or incumbents will do better than in reality. We will have to wait for the election itself to see how this experiment turns out.

We’re also seeing another big change relative to 2010: attention to the Congressional elections on Twitter appears to have grown substantially. We gathered about 250,000 Tweets over the course of the entire 2010 election. But for 2012, with just a few weeks yet to go before the election, we’ve already more than 340,000 Tweets in about 350 House races.

We’ll update our predictions each day and retain previous predictions so that you can look for changes and trends over time. This is most easily done in our vote share projection map and win-loss prediction map.

You should also expect to see additional graphics as the race goes on. We’ll have data on Tweet frequency, Tweets per candidate and analyses of what Twitter users are talking about when they Tweet about these races. We also plan some ad-hoc analyses before the election, and a big post-hoc discussion of how successful (or unsuccessful) we were at predicting who will win.

For more information about the methodology behind the experiment, please see the FAQ.

Finally, you can follow us on Twitter, @tweetvote2012, and contact us at

Comments are closed.