I learned about a new study which correlates aggregate Twitter mentions with election victories from Derek Abma. He contacted me last week to get my opinion on the study for an article he was writing for iPolitics (Forget the polls, Twitter chatter can call elections better: study, is now live).

The study, More Tweets, More Votes: Social Media as a Quantitative Indicator of Political Behavior (co-authored by Joseph DiGrazia, Karissa McKelvey, Johan Bollen, Fabio Rojas) will enter the official record tomorrow, Monday, August 12, when it is presented at the American Sociological Association’s annual meeting in New York. The short version of the study can be summed up in this quote from report:

There exists considerable debate about the validity of data extracted from social media for studying offline behavior. To address this issue, we show that there is a statistically signifi cant correlation between tweets that mention a candidate for the U.S. House of Representatives and their subsequent electoral performance.

In short, this study identified a correlation between the number of tweets, no matter the sentiment, and who won the election. That’s a correlation, not an outcome caused by the volume of relevant tweets.

Derek listened to my opinions and patiently allowed me to work out some connections between the study and Canadian politics out loud . Some of my thoughts are captured in his article. It was during our call that I quickly looked at some data I had nearby for the recent provincial election in B.C. You’ll recall that pollsters called that one completely wrong.

In his article, Dereks quotes Ipsos Public Affairs CEO Darrell Bricker as saying “I always find it ironic when people come up with analyses after the fact about how things are supposed to work but they never actually tested it when they had to put a prediction on the line.” Fair point. I only considered this as a relevant measure because of the report which I learned about last week. It would have been extraordinarily easy for me to have identified the tweet count prediction leading up to election day if I’d considered it meaningful.

Which means, if this study is on to something and I’d been more focused on this specific measure, I could have accurately called the election before voters went to the polls based on the volume of Twitter mentions for each leader. In fact, the share of tweets for each leader is within a striking distance of the share of votes cast for each party (the discrepancy in addition of vote count in the bullets below is due to independents not being considered in my analysis).

  • Liberals/Christy Clark: 44% share of vote, 51% share of tweets
  • NDP/Adrian Dix: 40% share of vote, 36% share of tweets
  • Greens/Jane Sterk: 8% share of vote, 7% share of tweets
  • Conservatives/John Cummins: 5% share of vote, 6% share of tweets

Perhaps the irony is a simple and freely available method of data gathering may have confirmed the business model of the traditional pollster is facing the same kinds of challenges already faced by the music, book and news publishing industries.

I’ll pay closer attention to this stat as future elections unfold.