Using Big Data to Make Political Campaigns Smarter

Nov 16, 2015

For many watchers of political races, it’s a numbers game. Dr. David Sathiaraj, Assistant Professor for Research in Geography and Anthropology at LSU, thinks there’s a better way to crack those numbers.

Traditional polling is good, says Sathiaraj," but in some cases, traditional polling can tend to be particularly small sample sizes, and so they may not be representative of the electorate." 

Credit Dr. David Sathiaraj

Traditional polling is typically done by selecting a group of people, asking them questions - like who they plan on voting for - and using that as an indicator of what the entire electorate may do. 

Sathiaraj wanted to dig deeper. His goal was to capture individual’s feelings of a candidate or an issue. He developed an algorithm that would pull information like voting history, voter registration and polling results.

"The algorithm basically provides scores for people on how they perceive a certain candidate or an issue. You’re able to come up with a more realistic scenario as compared to a broad, statewide poll," he says.

Sathiaraj believes campaigns big or small can use his algorithm to better pinpoint who they should reach out to. By enabling campaigns to see if they're weak in areas like a parish or precinct, says Sathiaraj, "they can start targeting, canvassing, getting the message out in those areas." 

That individualized approach leads to a smaller margin of error. Sathiaraj used the algorithm on last year’s senate runoff election in Louisiana. His model was nearly spot-on, just .2 percent off the actual results. The algorithm correctly predicted how fifty-seven of the sixty-four parishes would vote.

On average, most election polls have a margin of error of three to four percent.

Sathiaraj thinks that granular approach helps. "Polls are much broader, done at a statewide level, and the sample can be randomly selected. You may miss out on certain parishes that may be swaying one way or another. Because it’s so broad-based, a sample size of eight hundred is typically too small for an electorate that’s about three million," he explains.

The next step for Sathiaraj and his team is to scale the new technology to test on larger voting populations.