A team of IU researchers says Twitter can better predict elections than traditional polling.
The team of IU researchers collected more than a billion tweets from the months leading up to the 2010 U.S. Congressional Elections.
By analyzing how many times each candidate was mentioned on Twitter, the team was able to correctly predict the winner in 404 out of the 406 cases, or 99.5 percent of the time. In many of the cases, tweets predicted the outcome within 3 percentage points—about the same as traditional polls.
PhD student and researcher on the project Joe Digrazia says Twitter can offer information that polls cannot.
“What it can be useful for is a lot of issues where polling data does not exist, especially in small elections or to get at issues that are difficult to talk about,” he says.
Digrazia says because it is largely objective, using a Twitter analyzes removes the bias seen in polling.
IU Associate Sociology Professor Fabio Rojas, who led the project, says the twitter analysis is easy to use.
“It’s inexpensive, and it’s very accessible. Where in contrast traditional polls cost thousands of dollars and that really favors people who already have a lot of money and who are already well established,” he says.
Rojas says he hopes the team’s findings will translate into a computer program so anyone with basic computer skills can make the same type of predictions.