×

Twitter predicted the post-debate polls

Donald Trump and Ben Carson at the CNBC GOP Debate in Boulder, Colorado.
David A. Grogan | CNBC
Donald Trump and Ben Carson at the CNBC GOP Debate in Boulder, Colorado.

It seems our Twitter-based poll predictions turned out right.

Ben Carson, Marco Rubio, and Ted Cruz were among the three winners in our debate night tracking system.

And they were among the three big winners in the latest polls that just came out.

Last week during the Republican debate, we tracked in real time the number of Twitter followers each candidate gained during the debate.

Our hypothesis: An increase in Twitter followers would translate into more poll support.

You can see below the red line dominating the entire debate — that's Carson. And he is now the leader (or closing in on the lead) across a variety of national polls.

This strategy in worked in a past debate, where Carson and Carly Fiorina each gained 22,000 Twitter followers — and then saw their poll numbers rise the most.

The idea here is a potential voter clicking on the FOLLOW button for a candidate is a real transaction in favor of the candidate, and could tie directly into "following" them to the polls.

It's more specific and actionable than simply measuring conversations or mentions on social media. Those type of data can be cluttered with people who already support a candidate, rather than new supporters. It might also be cluttered with certain voters who just talk too much on social media, outweighing the opinions of many more quiet voters. As anybody knows in real life, it's one thing to talk about somebody — or a lot of somebodies — and another to actually voting for one.

And of course there are also a lot of mentions that are negative and don't translate to any kind of vote. Several attempts have been made at understanding "sentiment analysis" but this field of research still has some room to grow.

A lot of data experts think this could be the election cycle where social media plays a significant and accurate role in predicting the results. Traditional polling techniques have come into question in recent months, due to the changing nature of people's communication techniques — for example, who still answers a landline phones? This is why people have been looking for additional layers of data to help tell the right story.