You are not logged in. LOG IN NOW >

Can Tweets Predict the Vote?

BY Miranda Neubauer | Thursday, April 25 2013

Analyzing nearly 800 competitive races in the 2010 and 2012 congressional elections, researchers have found that the frequency with which a Republican is named correlates with the Republican vote margin the subsequent election, independently of other factors such as incumbency, media coverage, partisanship and demographics.

One of the lead researchers, Joseph DiGrazia, Ph.D candidate in sociology at Indiana University Bloomington, said his team was interested in the offline effects of social networks, rather than connections within an online social network or the type of discourse that takes place online.

The researchers decided to explore whether Twitter could offer additional information about how an election might turn out.

This study focused only on text references to the candidate's names on Twitter, and the researchers have not yet found any such correlation using actual @-mentions of a candidate or hashtags.

"Twitter is tapping into something about public sentiment and public attitude towards these candidates," DiGrazia said. "In any particular race, getting more tweet mentions relative to your opponent means higher margins of victory or higher vote margins."

In doing their calculations, the researchers analyzed all the name mentions of both candidates in races together, and established the metric of "tweet share" to specify proportion mentioning the Republican or the Democrat, he explained. For example, if there were only Republican mentions in a particular race, Republicans would have a tweet share of one. The researchers also used another metric to examine the number of users mentioning a specific name to make sure that outcomes weren't being driven by a small number of users posting spam messages.

DiGrazia emphasized though that the research was not predictive of a candidate winning.

"If you're a Democrat in a very conservative district, getting a lot of tweets would mean you would be expected to get a better than expected result," he said, "but not necessarily mean that you would overcome the disadvantage of being in a conservative district."

DiGrazia said the researchers were already working on another paper to try and examine why other Twitter metrics like @-mentions and hashtags were less predictive.

"We don't believe Twitter is influencing the election very much, we think Twitter is reflecting something about public sentiment and enthusiasm towards those candidates," DiGrazia said.

He noted that the research was aimed at social scientists, and was not done to draw conclusions about how Twitter might be used as a tool for candidates, but suggested it might be able to show some insight in smaller races where polling data is not available.

DiGrazia said the researchers were "kind of surprised" that they saw a correlation without doing sentiment analysis of the Tweets. "We thought we were going to have to look at the sentiment," he said. He speculated that one reason for the correlation could be a so-called Pollyanna Hypothesis, "that people are more likely to gravitate toward subjects that they are positive about and are more likely to talk about candidates that they support."

The paper was a collaboration between researchers in the Department of Sociology and researchers in the School of Informatics and Computing, who had access to the large collection of Twitter data. The researchers analyzed a random sample of over 3 billion tweets and Federal Election Commission data on the voting results.