Twitter and Predicting Elections

Twitter can replace polls in accurately predicting elections, or so says a study getting lots of attention online today.  Why will this excite political campaigners and candidates?

"The point is, it's cheap," [Fabio Rojas] said. "Once you start up software for collecting tweets, it's very cheap. It took one of my Ph.D. students a couple of weeks to set it up."

And thus we set ourselves up for a Literary Digest moment, the infamous 1936 poll of over a million folks that called the election for Alf Landon and proved that size, at least in a sample, doesn't always matter.

Remember President Landon?

Of course not. 

Twitter is used by relatively few people and even fewer actual voters, so how can it work so well?  Sheer numbers, of course.  Brute empiricism.

But what it won't tell you that a poll can, is why someone is voting for a candidate, how soft or strong that commitment is, and what issues appear to be driving that level of commitment.  In other words, Twitter may be good for predictive purposes -- and I'm not convinced yet -- but it doesn't help much in either planning a successful campaign or understanding why a campaign appears to be working, or not.