economic woman

We have a zombie meme on our hands, folks. Betsy Stevenson and Justin Wolfers’ paper, The Paradox of Declining Female Happiness, has been debated online since October 2007, and now it is back and stumbling among us, thanks this Huffington Post piece by Marcus Buckingham, a “leading expert in personal strengths and bestselling author.” (Is that a bio that inspires confidence, or what?) Stevenson and Wolfers’ paper is so obviously relevant to this blog that I am starting to feel silly for not weighing in. I realized that at its core, this debate is inspired by a knotty econometric problem, and there’s nothing I enjoy quite like a knotty econometric problem. So, this afternoon, I read the paper, and a good chunk of the commentary inspired by it.

I have to say, I’ve been pleasantly surprised by the paper, if not all of the commentary. This post mostly summarizes a debate that took place online back in 2007, because I think it is worth understanding. I’ve mixed up some chronology because I assume you are more interested in content than timing. I will make later posts with some more original insights.

Just in case you’ve been living in a cave: Stevenson and Wolfers find that women’s happiness, in total and compared to men, has been declining steadily since the 1970s. Women, on average, used to be slightly happier than men, and now, on average, they are slightly less happy than men. (Slightly is a key word here, and we’ll return to that.) The researchers look at a few different studies, but most of their results are from the General Social Survey, based on a simple question about subjective well-being: “Taken all together, how would you say things are these days, would you say that you are very happy, pretty happy, or not too happy?”

This is not a fluffy trend piece, though it has inspired a lot of fluffy trend pieces. I only have an undergraduate degree in economics, but I’ve read my fair share of econometric papers, in and out of class – I’m reasonably familiar with all the methods used in the paper – and Stevenson and Wolfers’ work does not look like statistical sleight-of-hand to me. The trend is real, but whether it is big enough to worry about is a more complicated question. Here is part of a blog post by Wolfers:

Let’s think of lining up all the men in 1972, in order of their happiness. In 1972, the median woman ranked between the 53rd and 54th man, happier than a slight majority of men.  By 2006, the median woman is somewhat less happy, ranking between the 48th and 49th man.

That does not sound like much. But his next point is a bit more convincing:

We know from prior studies that unemployment lowers average levels of happiness.  Comparing estimates across studies, we can say that the relative decline in women’s happiness that we document is equivalent to the decline in average happiness that would occur in a state if its unemployment rate rose by 8-1/2 percentage points (from, say, today’s 4-1/2% to 13%).

Another concern with the mainstream interpretation of this paper, stated best in this Language Log post (though note that other arguments in this post were later revised) is that the gap between the genders seems even smaller when you compare it to variations in happiness overall. In effect, the distributions of men’s happiness and women’s happiness overlap substantially. In statistical terms, Wolfers notes:

The relative decline in the happiness of women is roughly one-eighth of one standard deviation of the distribution of happiness in the population.  If you think there is a lot of variation in happiness in the population, this is big; if not, it is small.

I think that 1/8 of one standard deviation is pretty small whatever the standard deviation, frankly. Certainly this sort of substantial overlap is not what the Huffington Post et al imply, and that’s important. Language Log has some great suggestions on how to better describe this sort of research:

When we’re looking at some property P of two groups X and Y, and a study shows that the distribution of P in X is different from the distribution of P in Y to an extent that is unlikely to be entirely the result of chance, we should avoid explaining this to the general public by saying “X’s have more P than Y’s”, or “X’s and Y’s differ in P”, or any other form of expression that uses generic plurals to describe a generic difference. This would lead us to avoid statements like “men are happier than women. [...] The reason? Most members of the general public don’t understand statistical-distribution talk, and instead tend to  interpret such statements as expressing general (and essential) properties of the groups involved.

It’s hard to abandon a sentence construction entirely. But there is no doubt that we can do a better job pointing out the subtleties of this sort of statistical analysis. I thought Steven Levitt summed up this debate the best, so I will let him have the last word:

Is this a monumental shift? Maybe not. But compared to how much other factors move happiness metrics, it is pretty large. [Stevenson and Wolfers] are quite honest about the magnitudes in the paper. To the extent their results are being exaggerated, it is by people like me who write blog posts about their paper without being explicit about the size of the effect. The authors can’t reasonably be blamed for that.

Now that we’ve covered some of what the paper actually says, I’m going to move on to what it does not say. Stay tuned for the next post in this series.