I know my mood over the course of a day and so, it seems, does Twitter.
Over the years there’s been a lot of work on mood cycles. Much of it has been based on neurochemicals such as serotonin and dopamine.
But the complex biochemistry of this chemical soup is notoriously difficult to relate to mood states.
It has led researchers to ask whether there’s a predictor in other aspects of moods, for instance in the way we talk.
Writing in Science last week, two American sociologists, Scott Golder and Michael Macy, report some intriguing correlations between mood and factors such as time of day, work, sleep and the length of day.
Their results are based on the analysis of 509 million tweets, those ubiquitous digital messages of 140 characters or fewer.
Time of day matters a lot.
According to the data collected, we are happy when we rise in the morning, and again towards midnight. The rest of the day is a depressive slump, with the bottom of the happiness curve hitting when we come back from work at around 6pm.
Weekends were found to be happier times, showing (perhaps unsurprisingly) that work is a negative influence.
In the United Arab Emirates, for instance, the working week is from Sunday to Thursday.
The weekend is Friday and Saturday. And, sure enough, those are the happy days in the UAE. The tweets tell us so.
Golder and Macy also found that, based on their data, it takes seven hours for your average tweeter to get merry, based on the time that elapses between tweets about “beer” and tweets about being “drunk”.
Length of day is also crucial.
It has long been known that long winter nights and short, pallid winter days make people miserable and grumpy. People in Northern Europe and North America are known to suffer from Seasonal Affective Disorder, or SAD.
Long summer days are happy times. The key is relative day length: the relation of day to night. There’s probably a link to circadian rhythms, triggered by the onset of dawn.
And that too is consistent with the tweeting data: we rise and shine with tweets.
This general argument isn’t new.
In 2010, a group at Northeastern and Harvard universities in Boston reported similar results on the moods of Americans, based on 300 million tweets.
What is new with Golder and Macy’s work is that the’re working with a very large database of tweets from 84 countries world-wide.
They used the LIWC, the Linguistic Inquiry and Word Count tool for the computational analysis of text. LIWC is coded for 64 different behavioural dimensions, including PA (positive affect), happy words and NA (negative affect) words.
There are important differences between languages and cultures and their readiness to talk about emotions.
One limitation to Golder and Macy’s work is that they worked only with English-language messages.
And in spite of the fact the LIWC has been validated in psychological testing, there is many a slip twixt the emotion and its vocalisation.
Some stiff-upper-lip Anglo males will use emotion words warily and sparingly. There are other stereotypes of the emotionally profligate in speech, the hyperbolists, the emotionally mendacious. All these types will tend to skew the results.
But with that proviso, the Golder and Macy results confirm the original American data in persuasive ways.
It doesn’t matter where you are. So long as you speak English, at least for the purposes of this experiment, tweets are a reliable and insightful index of moods.
The accessibility of tweets, and the crushing quantities of them in open-access public repositories, means we can track the moods of nations, at least on a daily, weekly and seasonal basis.
We can graph them on maps (cartograms) and plot them by time of day, time of week and time of season.
So if your children or your partner are inexplicably gloomy, your first port of call should be their tweets.
There may be obvious factors you have overlooked.