VOTING WITH YOUR TWEET: An experiment in political forecasting

The “Voting with your tweet” project is a window into an emerging field in political research — the automated, computational mining of personal data. At this stage it is very much a work in progress – a draft for comment if you will, in the tradition of all serious academic research.

So why publish intermediate results on a news site?

Because the mining of social media data is becoming increasingly important in politics. Our goal is to help journalists and the public better understand new political prediction techniques and to help them become critical consumers of this kind of research.

For journalistic purposes, the “story” is not the prediction of any individual race. If anything, the current project suggests what political science has been telling us for decades – that incumbency, particularly at the congressional level, is a powerful predictor of electoral success.

The “story” is the attempt to catalog and parse how Twitter users express their political preferences and/or pass on information about candidates. Is Twitter actually a medium for novel communication? Or do people basically just talk about what we already know — who’s up in the polls, etc. How much activity is there? Can we effectively detect sentiment?

For example, the current “Voting with your tweet” project collects Twitter messages (social data), processes them, compares them to Twitter messages collected in the 2010 election (to improve accuracy) and creates daily predictions of congressional election outcomes and each candidate’s vote share.

But that doesn’t mean all predictions are accurate. We suspect there may be Republican bias in some areas and that outliers may surface as the experiment progresses. We’ll post our findings in the following weeks to illustrate what’s working and what isn’t.  We hope you find them interesting and that they provide insight into a process that can be difficult to report.

Our FAQ attempts to document the collection and filtering process in plain language. If you have questions, please ask. We’ll do our best to help you explain our results.