"The days of surprise about actual election outcomes in the big data world are likely to be fewer in the years ahead, at least to those who may have access to such data."
Those are the opening words of an article written by a University of South Florida Muma College of Businessprofessor and his research assistant, who broke down nearly a half billion minutes of TV viewing in 2012 to determine that what people watch could show who they'll vote for president.
The paper, "Does Television Viewership Predict Presidential Election Outcomes?” was published in the November 2015 volume of the journal Big Data.
Information Systems and Decision Sciences Professor Balaji Padmanabhan, assisted by USF doctoral student Arash Barfar, used anonymous TV viewership data from Nielsen taken from the four to five weeks before the 2012 presidential election. That data consisted of almost 550 TV programs and nearly 138,000 telecasts.
"And we were constructing variables such as minutes per viewer as well as a percentage of fans to find out which TV shows might predict the outcomes," Padmanabhan said. "Fans" were people who watched a program for a high amount of time.
They looked at each program at the state level (48 states and the District of Columbia) and in 165 individual counties.
The researchers then compared the time spent viewing each program to which party won in each state and county in the 2012 presidential election.
"In that process, we were able to identify quite a few shows that seemed to have predictive power," Padmanabhan said.
Three shows in particular stood out, predicting which way a state would lean more than 79 percent of the time: A&E's "Duck Dynasty," which predicted a Republican victory, Comedy Central's "The Daily Show with Jon Stewart," which leaned Democratic, and HGTV's "Family Renovation," which accurately predicted wins for each party, depending on the state.
The researchers then needed to find out if they could train a predictive model with the data.
To do so, they worked backward, taking information from "safe states" like New York and Texas where the victor could be assumed before the election and used that to predict outcomes in 10 swing states. As a result, "The Daily Show" picked the winner in eight of the 10, including North Carolina, the only swing state that went Republican in 2012.
"What potentially this could do is in a week or two weeks prior to the elections, if you were able to predict in certain states or counties that a certain party is less likely to win, it gives them real time information on what states and counties to invest in first," Padmanabhan said. "Second, they might look at the programs that seem to correlate more with their viewership and seek to very strategically advertise in those programs."
While Barfar has since taken a job as a faculty member at the University of Nevada, Padmanabhan's work is continuing. He'll get real time viewer data from Nielsen during the election season and continuing building his predictive models through the 2016 presidential vote.
"I'm not sure if we'll have the guts to go and call (the election) the night before," Padmanabhan said, smiling. "But it'll be fun trying to play around with this in real time."
