Here’s the previous piece for my Under the Radar column on BBC Future – there will be another column up very shortly. Peter Dodds, tells me that he and his colleagues have now created a “hedometer” site at http://www.hedonometer.org that will “provide a real-time measure of happiness that will be useful for many entities including governments at all scales, journalists, analysts, and citizens.” Peter adds that “initially, we'll be showing an interactive happiness time-series for Twitter but we'll be expanding to geography, social networks, etc., as well as other languages and other emotions.” It sounds rather fabulous, and will be free and open to all users when it goes live tomorrow.
Feeling low? Over-worked, anxious, bored with life? A holiday will do your mood the world of good. Really it will: there’s now scientific proof. A team of researchers at the University of Vermont in the United States has found that tweets contain significantly more happy words the further from home they are sent .
This is the latest dispatch from an emerging discipline in which social-networking media are mined to gauge people’s moods and opinions. Twitter is one of the most fertile sources of information for this kind of study, partly because the comments are less guarded and self-conscious than responses to questionnaires (the social scientist’s traditional means of sampling opinion) but also because huge amounts of data are available, with automatically searchable content. What’s more, Twitter feeds sometimes come accompanied with useful information such as the tweeter’s profile and location.
Previous studies in “twitteromics” have, for example, monitored the spread of news, the demographics of different languages, and the correlations between obesity and expressions of hunger in particular populations. Since public mood changes such as brewing social unrest will show up on Twitter and other social media, governments, police forces and security organizations are showing an increasing interest in twitteromics, raising questions about the right balance between privacy and security. Meanwhile, potential insights into the emergence and propagation of trends are a gift to company marketing departments.
The new study of the link between happiness and geographical location by Christopher Danforth and colleagues at Vermont takes advantage of the “garden hose” public-access feed for Twitter, which makes freely available a random 10 percent of all messages posted. This provided the researchers with four billion tweets for the year 2011 to analyse.
Since Danforth and colleagues were interested in how the mood expressed in the messages correlated with the location from which they were sent, they sifted through this immense data set to pick out those tweets that were accompanied by the precise latitude and longitude of the sender’s mobile phone – a facility optionally available for tweets, which uses the Global Positioning System (GPS) to locate the message’s origin within a 10m radius. About 1% of the messages included this information, giving a data set of 37 million messages sent by more than 180,000 individuals from all over the planet.
But identifying where the sender is situated doesn’t in itself reveal what the researchers wanted to know. They were interested in how the message content varied with distance from home. How could they know where ‘home’ was?
It turns out that positional information disclosed by our mobile phones reveals this pretty clearly. In 2008 a team of researchers in the US used the locations of mobile phones – recorded by phone companies whenever calls are made – to track the trajectories of 100,000 (anonymized) individuals . They found that, as we might imagine, we tend to return over and over again to certain places, especially our homes and workplaces, and only rarely venture very far from these locations.
In much the same way, Danforth and colleagues could figure out the most common locations for each individual in their survey, along with an associated number describing how widely the person tended to roam from those places. They found that people generally have two such preferred locations, just a short distance apart, which they attributed to the home and workplace.
How, then, do the messages differ when individuals are at home, at work, or further away? To assess the ‘happiness’ of a tweet, the Vermont team has developed what they call a ‘hedonometer’ : an algorithm that searches the text for words implying a positive or enjoyable context (such as ‘new’, ‘great’, ‘coffee’ and ‘lunch’) or a negative one (‘no’, ‘not’, ‘hate’, ‘damn’, ‘bored’). On this basis the hedonometer assigns each message a happiness score.
The authors report that “we see a general decline in the use of negative words as individuals travel further from their expected [home] location”. More precisely, the average happiness score first declines slightly for distances of around 1 km – the kind of distance expected for a short commute to work – and then rises steadily with increasing distances of up to several thousand kilometres. What’s more, individuals with a larger typical ‘roaming radius’ use happy words more often – a result that probably reflects the higher socioeconomic status of such jet-setting types.
So it seems we’re least happy at work and most happy when we are farthest from home. At least, that’s the case for the roughly 15% of American adults who use Twitter, or to be even more cautious, for the English-speaking subset of those who chose to ‘geotag’ their tweets. One key question is whether this sample is representative of the population as a whole – Twitter is less used among older people, for example. It’s also an open question whether ‘happy words’ are a true indicator of one’s state of mind – are you less likely to tweet about your holiday when the weather is awful and the family is fractious? But such quibbles aside, you might want to consider that costly flight to Bermuda or Kathmandu after all.
1. M. R. Frank, L. Mitchell, P. S. Dodds & C. M. Danforth, preprint http://www.arxiv.org/abs/1304.1296 (2013).
2. M. C. Gonzalez, C. A. Hidalgo & A. L. Barabasi, Nature 453, 779-782 (2008).
3. P. S. Dodds, K. D. Harris, I. M. Kloumann, C. A. Bliss & C. M. Danforth, PLoS ONE 6(12), e26752 (2011).