Harnessing the Power of Social Media to Predict Elections

05 November 2012
Published in Science and Society
Written by  Kirstin Roundy
  • Print

It all began with a tweet that echoed around the political world. In 2011, presidential candidate Jon Huntsman Jr. posted a tweet concerning evolution and global warming. This message went 'viral'; it was re-tweeted 5400 times and led to a spike in Twitter followers for Huntsman.

For the creators of PoliticIt - a startup company of Utah State University students, alumni and faculty - this tweet sparked a discussion about how online digital information can influence political campaigns. Britney Johnson, PoliticIt's chief technology officer and programmer, says, "My dad asked, 'I wonder if we could track this. I wonder if we could track how things go viral and how social media affects politics.'"

To answer this question, the data scientists at PoliticIt collect large amounts of information and 'feed' it to computers to identify trends. It's how weathermen develop their forecasts and how Google uses influenza-related search queries to track flu infections. It's also how PoliticIt is able to uncover links between social media and political campaigns.

"We like to say, in some ways, we're the Big Data of politics, the 'Moneyball' of politics," says John Johnson, chairman of PoliticIt and a professor in the Department of Management Information Systems.

The end result of their analyses is an 'It Score', a number that indicates how well a candidate's message is communicated digitally compared to his or her competitor. The candidate with the highest It Score 'wins'.

"Ultimately, it's a measure of how well I'm [a politician] going to do...since the It Score actually moves the way the polls do and since the polls move the way voters do," comments John Johnson. The It Score is posted in real time in PoliticIt Campaign, a software program that helps candidates manage their campaigns. For local candidates working with small budgets, PoliticIt Campaign provides the same information, without the expense, that can be gained from a standard poll.

The secret to PoliticIt's approach is big data. PoliticIt uses application-programming interfaces (APIs) to access data already collected by online sources - Facebook, Twitter, YouTube, Wikipedia, news media sites, online newspapers – and uses it for different purposes.

"For instance, Britney wrote a program that goes out to the Twitter API and she can say, 'Give me every single tweet that mentions Mitt Romney," says John Johnson. "And it will deliver back to me a formatted response that's actually data that my machine can understand."

"Machines [computers] are very good at looking at data. They're very good at seeing patterns," says Britney Johnson. Once they have the data, PoliticIt uses computer-learning algorithms to identify correlations between all of the data. "That's the beauty of having a machine learning algorithm on that It Score, is that it gets better as it [the computer] gets more information. It evolves across time," John Johnson adds.

PoliticIt has moved beyond analyzing how social media impacts politics to using the It Score to predict election results "We actually forecast 100 percent of all of the convention results, both Democrat and Republican, in Utah," says John Johnson. "We forecasted that Mia Love would win... People thought that Wimmer was going to beat Love...and, come to find out, she won."

Yet using social media data to predict election results has become a controversial topic. Research indicates that analyzing information from just one data source (e.g. total number of tweets about a candidate) produces inconsistent election predictions. For instance, President Obama has more Facebook and Twitter followers than Mitt Romney. If an election forecast were based only on the number of followers, these trends would wrongfully indicate that Obama would win.

Instead, based on the It Score, the candidates are tied. John Johnson says what sets PoliticIt apart from other social media-based predictors is that they analyze large datasets from several data sources. He continues, "Data exists there and there are stories that can be told of what's going on. We just need to unlock these stories."

According to PoliticIt, the presidential race is too close to call. As for the other races, only the candidates themselves, who are privy to their It Score, know if they will win or lose.

Additional Info

comments powered by Disqus