You are using an outdated browser.
Please upgrade your browser
and improve your visit to our site.
Skip Navigation

The Trouble with Measuring 'Tweets Per Minute'

This past week, Election 2012 found a new favorite metric: Tweets Per Minute (TPM). Apparently Mitt Romney got a lot of them (14,289 during his speech). But Michelle Obama got even more (28,003) and her husband the most of all (52,747). Early last month, Twitter even rolled out a Political Index—the Twindex—that tracks the favorability of Barack Obama and Mitt Romney among its 500 million users. 

The commentariat loved this quantifiable, dissectible, graphable stream of data, citing it alongside conventional polls and Neilsen ratings as just another way to measure public opinion—a new way, perhaps, to measure the psyche of the American people, without even having to ask how they're feeling. Maybe it even meant that conventional polling has been wrong—Romney might not have gotten a bounce in Gallup's estimation, but his Twindex levels were off the charts! 

Twitter analysis, though, has professional takers of the public temperature somewhat puzzled. "It's something no one has studied," says Daniel Kreiss, an assistant professor of media and politics at the University of North Carolina at Chapel Hill. "It's moving so fast, academics have tried to figure out what to do with it. It's been such an amazing challenge, because the data is so huge."

There are a few big reasons to mix that Twitter stream with a few spoonfuls of salt. 

The major one is demographic. The Pew Internet and American Life Project has found that while Twitter usage has doubled over the last two years, with eight percent of online adults tweeting in any given day, they're by no means a representative slice of America. Young people, urban residents, and African Americans make up a disproportionate chunk of the Twitterverse—which makes it a less friendly place for Republicans. There's also a bias towards more politically engaged people generally. 

It's no wonder, then, that Romney has only 1 million followers to Obama's 10 million, or that his campaign would try to gain more visibility by paying Twitter to boost the #RomneyRyan2012 hashtag as a trending topic. Naturally, GOP operatives tend to downplay the value of using the numbers to match the parties up against each other. "I wouldn't make any conclusions about Democrats vs. Republicans here," says Patrick Ruffini, whose communications firm has run analyses of convention applause lines

And of course, the methodology for measuring the meaning of a 140-character tweet is far from precise. The Twindex, for example, reacts not just to how people feel about the candidates, but also how the rest of the Twittering masses feel about everything else in their lives. The dataset for an event like a convention is limited to people who remember to use the right hashtag. And the accuracy of "semantic analysis"—coding each tweet positively or negatively—depends on words that may carry different meanings in different contexts that a computer program can't always comprehend. Under those conditions, a sample size of millions may not be any more precise a gauge of public sentiment than the instant-response dial poll that's scrolled at the bottom of TV screens for years now. 

For now, says Kreiss, professional politickers still view Twitter not so much as a way of taking America's pulse as just another vehicle for their tailoring a message to a narrow audience. 

But outside the land of breathless insta-commentary, a fast-growing field of analysts are eavesdropping on Twitter's side conversations, and finding ways to insert themselves within them. 

Take Stuart Shulman, a political science professor and entrepreneur at the University of Massachusetts. Having pioneered the use of textual analysis to synthesize public comment forms on new government regulations, he transposed techniques like probabilistic mathematics to Twitter, filtering out tweets using the same word that might mean something else in context—Avon beauty products vs. Avon Barksdale or Avon Books, for example. Then, he can give paying corporate clients a profile of public conversation around their brand. Another emerging arena is human resources: Shulman might search Twitter bios for a company's name to find its employees, and what they're saying during work hours (don't post a picture of yourself jet skiing if you've just claimed workers comp!). 

"We're not exactly Big Brother, but some people feel that it creates the tools," Shulman says. "It is a little creepy, if I think about some of the things I've done over the years." 

Twitter is even more useful, however, when used in combination with other streams of internet commentary. A Republican strategic communications shop called Crowdverb is working with the psychological analytics company Behavior Matrix to characterize all public online actions into 55 human emotions. By now, says chief technology officer Chuck Davis, they've built up tens of millions of individual profiles using what people say about themselves across various platforms, sorting them into categories like gender, race, location, and who they're planning to vote for. Pollsters have long used consumer preferences as a way to target political advertising, but with the ability to recognize how people are feeling in real time, the response can be immediate. 

"You can now provide the right message when you know that the person is emotionally receptive to what you're saying," Davis says. "If you're surprised about something, or outraged or something, that might only last a couple of days, or a couple hours."

So who would pay for that? A company that wanted to nudge you into buying a product you've said on Twitter that you liked, for example. Or an advocacy group waging a legislative campaign that wanted you to take an action based on an opinion that you expressed. In other words, beyond taking the temperature of the electorate, really cutting-edge Twitter analysis means that interested parties can talk back—even if Tweeters don't realize who's on the other end of the line.