Are our digital moods biological?

Social data and the social sciences are a powerful match. New research from Cornell University suggests that our moods are partly driven by a shared underlying biological rhythm that transcends culture and environment. The sample population studied? Twitter users.

This is the first cross-cultural study of mood rhythms using text analysis of Twitter, and it’s just one example of the sort of insights that can be extracted from social data. Commenting in The New York Times on the emerging use of social data, University of Vermont Mathematics and Statistics researcher Peter Sheridan Dodds said:

“There’s just a torrent of new digital data coming into the field, and it’s transforming the social sciences, creating new lenses to look at all sorts of behaviors.”

So, what did the Cornell mood analysis uncover?

Happy to bed, happy to rise

Over two years, the team analyzed 400 tweets each from 2.4 million English-speaking Twitter users in 84 countries. Their program associated certain words like “awesome” and “agree” with positive moods, and others like “annoy” and “afraid” with negative ones.

For the average Twitter user, positive tweets crested between 6-9am, then fell to a low between 3-4pm before rising again in the evening. The findings complement what’s known about mood fluctuation in general, as well as another study by the University of Vermont researcher quoted above, which showed that the use of swear words on Twitter correlates to negative sentiment.

We’re happier on weekends—but it’s not that simple

Not surprisingly, positive moods peaked on the weekends and moods were lowest at the beginning of the work week. This trend was seen even in countries where Saturday and Sunday are not considered the weekend.

Sentiment on weekends shifted a few hours later, peaking close to 9am and after 9pm, but following the same general shape as weekdays. Study author Scott A. Golder explains why this matters:

“This is a significant finding because one explanation out there for the pattern was just that people hate going to work. But if that were the case, the pattern should be different on the weekends, and it’s not. That suggests that something more fundamental is driving this – that it’s due to biological or circadian factors.”

Caution: the current limits of text analysis 

While the findings are interesting, it’s important to remember that text analysis alone can’t gauge sentiment with total accuracy, as we explored in our last post. Sarcasm and slang are common on Twitter, and the software isn’t sophisticated enough to catch these human subtleties. On top of that, Harvard psychologist Dan Gilbert explains to The New York Times why tweets shouldn’t be considered accurate sentiment indicators by themselves:

“Tweets may tell us more about what the tweeter thinks the follower wants to hear than about what the tweeter is actually feeling. In short, tweets are not a simple reflection of a person’s current affective state and should not be taken at face value.”

This study represents both the promise and the current challenges of social data analysis. As word of mouth is digitally archived for the first time in history, the benefit for businesses will come from the insights we’re able to extract, the context we’re able to place them into, and the decisions they’re able to support.