Can Big Data Predict Local Elections?

In the public opinion research office where I work we have several framed bar graphs hanging on the wall.  These bar graphs show how closely our survey results matched the actual voting in several local elections.  We also have a Dilbert comic strip on the wall that succinctly captures the challenges of getting reliable information from telephone polling:

Dilbert is in a meeting with his boss and a consultant.  The consultant is explaining the results of a survey and says that they surveyed 1,000 people who still have landline phones and no caller ID.  When asked for their opinions a new technology, 34 percent replied, “Fiddlesticks” and 23 percent could not hear the question.  The other 43 percent thought that the interviewer was in the room with them and offered hard candy.

The Obama campaign made use of Big Data before the election in 2012.  They were able to precisely predict how the vote would go in each state.  I also learned that the campaign relied on benchmark data that was gathered from conducting 10,000 telephone interviews.  These interviews measured the opinions and demographic characteristics of voters throughout the United States.  The campaign used this information to create models of voters who might be persuaded to change their vote and if so, what might influence them to do so.  This helped the campaign to carefully craft individualized online ads to reach voters even in solidly Republican precincts.

I like to remember the part about the 10,000 telephone interviews when I see posts on LinkedIn warning me that my job may become obsolete.  Big Data and such things as text analytics may automate some of the work done by human analysts, but someone will still need to get on the telephone to gather information that can be used to calibrate such data.  Such calibration will be even more important in predicting local elections. 

We recently completed a survey about a levy issue that will be on the ballot in November of 2014.  We are planning to conduct another in October.  I have taken steps to learn more about Big Data and text analytics to learn if they can be used to predict elections with the same accuracy that we have achieved with telephone surveys.  If our clients perceive that they can get the information they need to influence how likely voters will vote, they will believe that they will not need to spend money on telephone surveys.  It will be important to be ahead of this perception rather than trying to catch up with it.

I doubt that there is enough data on the Internet to provide insight on crafting campaign messages for a local election or accurately predicting such an election.  The Obama campaign had plenty of text to analyze.  People were interested in the presidential election.  They posted blogs and participated in discussions on Facebook about it.  I see almost nothing about local elections.  The posts that I do see are for candidates, not bond or levy issues.

The Dilbert comic strip mentioned above illustrates one of the challenges of polling individuals younger than 65.  I wonder whether Big Data and text analytics can measure the opinions or predict the voting behavior of those older than 65.  Such people are more likely to vote on bond and levy issues than younger people, but are less likely to participate in heated debates on Facebook than younger people.  Calling them on the telephone may be the only way to learn how they plan to vote and why.

John C. Stevens
Saperstein Associates
(614) 261-0065
jstevens@sapersteinassociates.com