Measuring Political Buzz with Twitter

Telephone polls have been in their dying throes for years. At about 12% response rates, they shouldn’t even work now. But they do, so far. That said, pollsters expect them to die at any moment.

I preparation for the demise of telephone polls, pollsters have been looking for alternative sampling methods for years. Once Random Digit Dialing no longer produces workable probability samples by phone, we will need reusable representative panel groups, or some other way of gathering survey respondents that will allow us to estimate mass public opinion with statistical significance.

While pollsters look to solve their problem, the online world has been evolving a complement (and maybe alternative) to polling methods: buzz metrics. In 2008, this took the form of Yahoo search statistics for Hillary Clinton and Barack Obama, which proved to be a better predictor of the New Hampshire primary results than the last poll conducted on the Sunday before the Tuesday vote. Even subtracting out searches for Clinton’s “diner cry,” her searches were soaring as Obama’s were flattening. On Tuesday, contrary to the pollsters’ predictions, Clinton won New Hampshire.

Now, Twitter, Topsy, & USA Today have teamed up to create the Twitter Election Meter. It combs through a couple million tweets mentioning Obama and Romney every week and ranks each candidate for the sentiment expressed in those tweets. The score is converted to a 0-100 point scale, where 100 is super awesome, 50 is neutral, and 0 is blech!

Paul Singer, Politics Editor at USA Today (@SingerNews), said of the project that has dominated his life these past few weeks that it made clear how negative the vast majority of tweets are. People are more likely to retweet the negative comments, thus pushing the negative to positive tweet ratio towards its upper limits.

The dominance of negativity isn’t surprising. We know that “no” is easier than “yes.” Ballot initiatives are more likely to get a no vote, it is much easier to scientifically reject an hypothesis than affirm one, and the press is more likely to lede with bad news. Some might say it is human nature to decline.

With all that said, it will be interesting to see how the Twitter Election Meter fares in predicting the election outcome and how it tracks with opinion polls. I’m predicting it will do well.

Check out my interview with Paul Singer on today’s (August 4, 2012) Take Action News with David Shuster social media and politics segment (all TAN archives are on iTunes).