What Millions of Books Reveal About 200 Years of Happiness
Researchers analyzed eight million texts to gauge how lifespan, warfare and the economy affect national well-being
A new study published in the journal Nature Human Behavior draws on 200 years of literature to assess the validity of an old adage: You are what you read.
Researchers from the University of Glasgow, the University of Warwick and the Alan Turing Institute surveyed more than eight million digitized texts available on Google Books to determine how well literature reflects its writers’ and readers’ subjective well-being. As Natasha Frost reports for Quartz, the team hypothesized that works published during the so-called “good old days” would be more uplifting than those penned during times of hardship.
According to the study, scientists led by Thomas T. Hills, a psychologist at Warwick, created an index of words based on their valence, or how “good” versus “bad” survey participants deemed them to be. Using this list, the team then created an algorithm that analyzed texts published in the United Kingdom, United States, Germany and Italy between 1820 and 2009.
By tracking changes over time, Hills and his colleagues were able to juxtapose shifts in subjective well-being (as represented by what the researchers call a “National Valence Index”) with factors including gross domestic product, average lifespan, war and democratization.
Perhaps unsurprisingly, the team found that money can’t buy very much happiness. Although increases in GDP tended to improve overall well-being, only large upticks in income had a noticeable effect on national happiness levels.
Life expectancy had a much stronger impact on people’s well-being: Per the paper, living one year longer made people as happy as a 4.3 percent rise in GDP. Most strikingly, one fewer year of war had the same impact on happiness as a 30 percent increase in GDP.
The U.S.’ national happiness post-World War II reached its lowest point during the mid-1970s, a period punctuated by the U.S. failure in Vietnam. The U.K., meanwhile, experienced its strongest sense of well-being during the late 19th century—when the country was at the peak of its colonial prowess—but faltered during the Winter of Discontent, an industrial crisis that took place during the late 1970s.
Data shows that events like the Great Depression and the rise of Italian fascism impacted well-being in the short term but did not scar people for very long.
“What’s remarkable is that national subjective well-being is incredibly resilient to wars,” lead author Hills says in a statement. “Even temporary economic booms and busts have little long-term effect.”
Hills adds, “Our national happiness is like an adjustable spanner that we open and close to calibrate our experiences against our recent past, with little lasting memory for the triumphs and tragedies of our age.”
As Vox’s Sigal Samuel reports, the researchers checked their findings against the Eurobarometer survey and the World Database of Happiness, both of which draw on data dating back several decades. To gauge the sentiments of people who lived centuries ago, however, the team had to rely largely on the NVI.
According to Samuel, the study does not measure objective well-being, as determined by physiological factors including stress hormone levels. Instead, the scientists used subjective reasoning to determine whether a word conveyed happiness or discontent.
It’s worth noting that cultural differences in how people express emotions and define certain words likely distorted the researchers’ results. Predictive algorithms’ poor understanding of social context also could have influenced the findings.
“At this point, what we have is really, really crappy software,” Meredith Broussard, a data journalism expert at New York University, tells Vox. “Computers can’t understand nuance or jokes.”
The new algorithm simply counts the frequency of certain words. Humans, on the other hand, understand language in a broader context and often derive meaning beyond the literal definition of words on a page. Aware of these limitations, the authors tried to study words that retained a stable meaning over time or use measures that accounted for changing definitions over time. The word “gay,” for instance, does not necessarily mean the same thing now as it did 200 years ago.
To compile their data, the researchers turned to Google Books, which hosts more than eight million digitized volumes—more than 6 percent of all books ever published. The authors drew on a wide array of texts, including books, newspapers and magazines, to diversify the information inputted into the algorithm.
Still, the data used may exclude important writings from marginalized communities systemically underrepresented in published works. It’s also worth noting that the texts used for the study were not filtered by potential censorship.
“As our data are drawn from public text, it may be subject to censorship,” the authors write. “… Germany in the 1940s, when negative portrayals of the Nazi regime were censored, is a case in point.”
There are definite challenges associated with measuring qualitative data using quantitative methods, but as the researchers note, the approach described in their paper has practical implications: In addition to helping scientists better understand the past, this method could be used to assess such varied issues as political candidates’ popularity, the societal impact of celebrity deaths and earthquakes’ economic aftershocks.