University of Vermont

Research at The University of Vermont

The Dow Jones of Happiness


Want to measure how financial markets are faring? Check Dow Jones or the S&P 500. Want to measure how happy the world was yesterday? Check… wait a minute. You can't measure global happiness, can you? Yes you can — we've built a tool to do the job, say Peter Dodds, Ph.D., and colleague Chris Danforth, Ph.D.

These UVM scientists, working with others from the MITRE Corporation, have been gaining international attention over the last few years for the creation of what they're calling a hedonometer. It's a happiness sensor — and it made the front page of the Wall Street Journal.

Visit and you'll see a wavering graph that rises and falls like a ticker at the New York Stock Exchange. Except instead of averaging the value of thousands of companies, the hedonometer compiles and averages the emotional state of tens of millions of people.

"What it's doing right now is measuring Twitter, checking the happiness of tweets in English," says Danforth, who co-led the creation of the site with mathematician Dodds. But soon the hedonometer will be drawing in other data streams, like Google Trends, the New York Times, blogs, CNN transcripts, and text captured by the link-shortening service Bitly. And it will be data-mining in twelve languages.

The research team made headlines — including Time magazine and The Atlantic — when they reported on the happiest and saddest cities in America: Napa, Calif., at the top and Beaumont, Texas, at the bottom. In future versions of the hedonometer, the researchers plan to make this kind of geographically linked data available, allowing as-it-happens observation of how a happiness signal varies, say, between Seattle and San Diego.

"Reporters, policymakers, academics — anyone — can come to the site," says Danforth, "and see population-level responses to major events." Like the Boston Marathon bombings, the saddest day measured by the scientists in nearly five years of observations.

The hedonometer draws on what scientists call the "psychological valence" of about 10,000 words. Paid volunteers, using Amazon's Mechanical Turk service, rated these words for their "emotional temperature," says Dodds. The volunteers ranked words they perceived as the happiest near the top of a 1-9 scale; sad words near the bottom. Averaging the volunteers' responses, each word received a score: "happy" itself ranked 8.30, "hahaha" 7.94, "cherry" 7.04, and the more-neutral "pancake" 6.96. Truly neutral words, "and" and "the" scored 5.22 and 4.98. At the bottom, "crash" 2.60, the emoticon ":(" 2.36, "war" 1.80, and "jail" 1.76.

Using these scores, the team collects some 50 million tweets from around the world each day — "then we basically toss all the words into a huge bucket," says Dodds — and calculate the bucket's average happiness score. As the site develops, the scientists anticipate that it will be gathering billions of words and sentences daily. "Our method is only reasonable for large-scale texts, like what's available on the Web," Dodds says. "Any word or expression can be used in different ways. There's too much variability in individual expression" to use this approach to understand small groups or small samples. For example, "sick" may mean something radically different to a 14-yearold skateboarder than it does to his pediatrician. But that's the beauty of big data. Each word is like an atom in the air when you're trying to figure out the temperature. It's the aggregate effect that registers, and no individual tweet or word makes much difference.

Changing which words are used to assess the overall emotional picture, "is like changing the filter on a lens you're using," explains Dodds. "You can take out all the color, or you can turn up the contrast, but you can still see the picture."

Last modified May 15 2014 04:14 PM