William Watson: The book on happiness... all the way back to 1776

My GDP makes me so happy. My GDP makes me so happy. My GDP makes me so happy.

Actually, I’m not sure whether my GDP makes me all that happy. But I’m always skeptical of studies that find it doesn’t, and there are many of those these days. So I’m web-planting a point of view on the relationship between the two phenomena in case some future researcher scans this website in a Big Data study of how people were feeling about happiness and GDP in the middle of the second decade of the 21st century.

My inspiration for doing so is an ingenious new study by three economists from the University of Warwick that takes a Big Data approach to the evolution of happiness since 1776. Called “Historical Analysis of National Subjective Wellbeing Using Millions of Digitized Books,” it uses the Google Books corpus, “a collection of word frequency data for over 8 million books,” to try to figure out how people living in the societies and eras in which the books were published were feeling about themselves. You might think right off the mark that doing so would tell you how authors were feeling in these times and places but not necessarily everyone else living in them. As it turns out, the research findings suggest books may in fact catch the zeitgeist.

For various purposes, psychologists often ask experimental subjects to score words according to “valence,” a not-very-descriptive word meant to indicate degree of attractiveness. The several thousand words used in this study were scored on a scale of 1 to 9, with 1 indicating low attractiveness and 9 high. It’s not clear to me why “aardvark” has almost as high a valence (6.26) as “ability” (7) but perhaps we’d all agree that “abbey” (5.85) is a more pleasing word than “abortion” (2.58). (The valence coding used in the Warwick study was actually done in an earlier study by researchers at McMaster University.)

Once you know whether a word is, to put it crudely, a happy word or a sad word, you can then calculate the average happiness or sadness of words appearing in any given text. Using today’s phenomenally high-speed computers (high-speed by historical standards at least, though I expect anyone reading this any ways off in the future will find 2016 notions of computer speed quaint) you can examine lots of texts.

Which is exactly what the researchers did. To be precise, they looked at the valences of words used in almost eight million digitized books.

To start with, they looked at the last four decades, correlating average valence with self-reported surveys of happiness or well-being, which are available on a continuous basis from the 1970s on. Doing so for several countries, including the U.S., U.K., France, Germany, Spain and Italy, they found, while controlling econometrically for other possible influences such as wars or civil unrest, that valences generally correlated with the self-evaluations. Greater use of happy words lined up with greater reported well-being. Neither lined up with per capita GDP, however, which, as mentioned, is a big focus of study in this literature.

Next step: look back, look w-a-a-y back. Google Books goes back to the 1500s but because the sample size is small for the first couple of post-printing centuries the researchers decided to start at 1776, which has nice resonance to it, for various reasons, including the publication in that year of the Wealth of Nations, though they don’t mention that. (They do mention Kant’s Critique of Pure Reason, 1781.)

If you look at average valences of the words used in the several countries over these two centuries, do the results make any sense?

The average does rise and fall with obvious social disturbances, such as the American Civil War and the two world wars of the 20th century. There’s also an intriguing negative correlation with the advent of democracy. The authors suggest the explanation may be, not that democracy makes people unhappy, but that it encourages realism and that in realist writing not all words are so happy.

Regarding well-being: other things equal, the average valences were correlated with life expectancy. As life expectancy rose words became happier on average. But they didn’t correlate with rising GDP.
As mentioned, the lack of correlation between GDP and various indicators of well-being is a common though not unanimous finding in the literature. To my mind, it doesn’t jive with what’s probably the most common political complaint these days—that median incomes aren’t rising fast enough.

But the study is ingenious. The authors are acutely aware of its possible pitfalls and discuss them at some length. And in the end it by no means disproves beyond a reasonable doubt that GDP is good for happiness, GDP is good for happiness, GDP is good for happiness…