How Twitter could predict elections — in one eye-catching study

August 14, 2013

Want to figure out who is going to win a congressional race? Find out which candidate received the lion's share of tweets in the lead-up to Election Day.

That's the takeaway at the core of a newly-released study conducted by four researchers at Indiana University. The paper stands in stark contrast to other research assessing the usefulness of tweets in assessing public opinion, as well as a number of high-profile whiffs from the Twittersphere.

The study looked at more than 537 million tweets sent between the beginning of August and the start of November 2010, alongside 406 contested congressional elections. "Our results show that the percentage of Republican-candidate name mentions correlates with the Republican vote margin in the subsequent election," the researchers write. "This finding persists even when controlling for incumbency, district partisanship, media coverage of the race, time and demographic variables such as the district's racial and gender composition."

In other words, they found a connection between the number of tweets that included a candidate's name -- relative to how many tweets that candidate's opponent received -- and how the candidate performed on Election Day, as the following chart from the study illustrates. The candidate with the higher percentage of tweets about them overwhelmingly tended to do better.  (Read the entire study here.)

The study doesn't include any data about the context of the tweets -- whether they were good/bad, flattering/unflattering, and otherwise positive/negative. This seems to reflect a long-standing view about the importance of name recognition in politics. Getting people to know you and talk about you -- in whatever context -- is a part of finding success in campaigns.

"The relative share of attention compared to the opponent is all that is needed," write the researchers, Joseph DiGrazia, Karissa McKelvey, Johan Bollen, and Fabio Rojas. "This is evidence for the conventional wisdom that 'all publicity is good publicity.'"

It's an interesting study, especially considering that other data suggest a vast disparity between reactions to major political events on Twitter and the opinions Americans expressed in polls about the events.

In last year’s presidential election, the tone of Twitter’s reaction to campaign events matched the balance of public opinion in random sample surveys only 25 percent of the time, according to a Pew Research Center study.

Some high-profile misses are also illustrative of the challenge of using tweets to reliably project elections. Anthony Weiner’s nearly 250,000 mentions on twitter (according to topsy.com) are unlikely to revive his downward spiral in the New York mayoral race – current front-runner Bill de Blasio has received barely 10,000 mentions in the same period. And while then-Rep. Ron Paul (R-Tex.) received wide recognition on Twitter during the 2012 Republican presidential primaries, he failed to win a single contest.

As Twitter has grown into a dominant means of communication and information-gathering, questions about what it means for politics will continue to get asked. And studies like the one from Indiana University will continue to try to answer those questions.

Scott Clement contributed to this post. Clement is a pollster with Capital Insight, the independent polling group of Washington Post Media. 

Sean Sullivan has covered national politics for The Washington Post since 2012.
Comments
Show Comments

Get the The Fix newsletter

Sign up for daily updates from The Fix.

Most Read Politics
Next Story
Scott Clement · August 14, 2013