Big Buzz On Twitter Means Better Chances On Election Day
ROBERT SIEGEL, HOST:
Is the political poll a 20th century dinosaur lumbering toward extinction to the accompaniment of triumphant tweets? Well, a team at Indiana University says you can predict election results just as well by mining Twitter for the names of candidates as you can by polling. They examined a sample of just over a half a billion tweets from August through October 2010. And joining us to talk about what they found, from New York, where the research was presented today to the American Sociological Association, is Fabio Rojas, associate professor of sociology at Indiana University. Welcome.
FABIO ROJAS: Thank you very much. It's a pleasure to be here.
SIEGEL: And you're saying that the candidates with the most buzz in the Twitter-verse also win at the ballot box?
ROJAS: Roughly speaking, yes, more tweets does give you more votes.
SIEGEL: Well, how well would all of those tweets have predicted congressional races in November 2010?
ROJAS: So, for example, if you were to ask the question - does a candidate who gets 50 percent of the tweets about their congressional race get 50 percent more of the vote, then that predicts 404 out of 406 competitive races where you have a Republican running against a Democrat.
SIEGEL: And does it predict the margin of victory pretty well?
ROJAS: In many cases, it does. It tends to do very well when the race is very competitive. For example, there was one in Utah where, you know, somebody was getting about 47 percent of the vote and the Twitter share was about 45 percent of the vote. So a lot of the cases are within the margin of error of a traditional poll.
The cases where the margin of error is big are cases where the race itself is not very competitive, where somebody's running against a very weak competitor.
SIEGEL: That would probably describe about at least 300 of the 435 races for the House of Representatives.
ROJAS: And there, once again, it does accurately predict the winner, but the margin of error tends to be a little bit larger. And what we suspect is happening is that when the race becomes so lopsided, people stop talking about the race and then the mentions of candidates on Twitter becomes very random and not so strongly correlated or tied with how people are going to vote.
SIEGEL: Now, Professor Rojas, explain this. You were only searching for names, so the phrase I love Nancy Pelosi and the phrase don't let Nancy Pelosi destroy America would have counted equally as mentions of Nancy Pelosi.
ROJAS: That is correct. That is one of the remarkable findings of this research, which is that the political sentiment or emotion of the tweet does not matter. It's just simply the raw share of tweets. So in other words, the more that people are paying attention to you, whether they like you or hate you, is the important forecaster of the final vote share.
SIEGEL: Well, you think social media will supplant political polling in our time?
ROJAS: I think political polls are going to remain useful. They've very valuable in specific cases. So, for example, if I'm interviewing a voter and I'd like to know more about them, a poll is a very good place to do that because we can ask the question directly, while we don't have as much information about people from social media unless they reveal it themselves through what they write.
But this is going to transform the polling industry because what this shows is that anybody with a laptop computer can come up with the forecast of an election that may be on par or better as a traditional poll.
SIEGEL: Fabio Rojas, associate professor of sociology at Indiana University and co-author this study of tweets and elections, thanks for talking with us.
ROJAS: Thank you very much.
NPR transcripts are created on a rush deadline by an NPR contractor. This text may not be in its final form and may be updated or revised in the future. Accuracy and availability may vary. The authoritative record of NPR’s programming is the audio record.