Dolan, marriage, and saying something interesting about subjective well-being

There was a bit of a scandal recently in the academic twitterverse when Paul Dolan, author of Happily Ever After and Professor of Psychology at the LSE, made a boo boo. The incident was pretty hilarious if you're an academic (especially Dynarski tweeting: "READ THE FUCKING CODEBOOK!"), but leaving that to one side, I think it also reveals a lot about how dismally shallow social science, and my field (subjective well-being; SWB) in particular, is sometimes. 

Paul Dolan said in an interview, quoting from his book, that marriage makes men happier (on average) but women "fucking miserable". He drew on data from the American time-use survey. Advocates of marriage, especially at conservative/traditional values think tanks, quickly responded that data from other (arguably better) social surveys said the exact opposite, and in stark terms. Dolan was ready: he said that women answered that they were happy in their marriages only when their husband was in the room. How intriguing. 

A shame it was a myth - Dolan had misunderstood a variable in the survey, namely "partner absent". What this meant was that the partner had moved out of the household (something you might associate with divorce, which usually precedes a rise in life satisfaction). Dolan interpreted it as meaning that the husband wasn't present at the time of the survey. Everyone had a good laugh at his expense, and he kept on trucking in his almost inimitable, constitutionally-immune-to-criticism, style. You can read all about it here: 

https://www.vox.com/future-perfect/2019/6/4/18650969/married-women-miserable-fake-paul-dolan-happiness

I want to use this incident as a hook to talk about a deeper issue afflicting subjective well-being studies, and social science more broadly. This is the tendency towards shallow theorising when you have big data. Even if you have causation, this sort of thing is not especially constructive. Let me explain. 

Both the "marriage is bad, see the time-use survey" and "marriage is good, see the general social survey" takes are shallow and borderline meaningless. First, they are both based on pure correlations. Married people are happier/sadder, on average. Is it the marriage that causes happiness, or vice versa? Controlling for a bunch of confounders doesn’t get you causation, especially when your R² is trivial (rarely above 15% in SWB research).

But more importantly, marriage is waaaaaaaayyyy more granular than this. As a simple example, consider whether someone should marry just anybody in order to get the causal effects of marriage. The answer is obviously hell no. If you woke up one day, read one of the studies suggesting marriage makes you happy, and then went out and married the first vaguely compatible person you met, I daresay it would not make you happy. And this isn't an average effect in the traditional sense—if you kept marrying strangers at random, the quality of those marriages wouldn't meaningfully average out. Or rather, the causal effect of each marriage wouldn't average out; your compatibility with each person might though. What I mean is, you can't just "marry" a random person off the street. You can certainly go through the legal process of getting married, but I don't think this is what Dolan et al mean when they speak of the effect of marriage on life satisfaction. 

Cavit Guhen and Bruce Chapman did a really straightforward article on this years ago - it was almost a piss take. They analysed marriages using the Household Income and Labour Dynamics of Australia (HILDA) panel, which includes questions about the subjectively assessed quality of a person’s marriage. Unsurprisingly, people in bad marriages were unhappy, while people in good marriages were satisfied. The question then becomes, what makes a good marriage? Or even, what is it about a marriage that can make it have a good/bad effect on life satisfaction?

You quickly get to the need to theorise, which is why this whole episode illustrates how banal a lot of subjective well-being studies are by virtue of the fact that they are unwilling or incapable of doing even a smidgeon of theory. "Does marriage make you happy?" is an important question. But it cannot be answered statistically at that high scale of analysis. You need to break it down in finer pieces. You need to consider things like personality compatibility, children, timing, age differences, cultural compatibility, values, complementarity, the sex etc. Controlling for these things and thereby getting the average effect of marriage ceteris paribus is nonsensical because marriage is a composite of all of these things. You don’t say to someone “you should get married”. You say “you should get married to this detailed sort of person, under these specific circumstances, in this peculiar context, for these personalised reasons”.

This kind of analysis really lends itself to a book where an introduction lays out the broad landscape of marriage and then each follow up chapter examines some particular area of complexity. I often worry that our current obsession with publishing papers really works against this sort of careful, deep work, which sits, at least in my opinion, at the very centre of good social science. Individual elements of a complex system cannot be analysed in isolation no matter how big and rich your data set (talk to a macroeconomist for a bit). Instead, we have people trading inane barbs at an almost abstract scale of analysis with no practical consequences.

I daresay we are getting to the point where the marginal returns to further statistical sophistication are rapidly diminishing while an ability to think in a sophisticated way about social and psychological phenomena has (once again) got high marginal returns.

A final point: without theory, there is no meaningful sense in which you can get at the causal effect of marriage. Why not? Well consider the following question: what even is "marriage"? Are we talking a business partnership like in feudal Europe and many developing countries? Something that begins as a grand amour that finds its ultimate expression in a ring? The ritualistic recognition of a deep love won over a decade of tribulations? A quasi-contractual "commitment" in preparation for child rearing? Or something else? All of these conceptualisations of marriage are going to have implications for the effect of "marriage" on life satisfaction. As soon as you go just a smidge beyond what is the effect of marriage to the much more important question of how does marriage generate this effect, you run into these issues. 

And that's before you get to a critical challenge for the field that I will eventually write a paper on: how do you get at causation when it is fundamentally endogenous? Marriage doesn't happen to people by accident. There is no sense in which you can have random, exogenous variation in marriage and thereby get clean causal analysis of its effects. Same goes for many of the big ticket items in well-being research: divorce, children, job change, migration, goal attainment etc. All of these things are inextricable from endogenous choices. So the identification strategies that dominate mainstream economics, at least at present, at not viable in subjective well-being research. Of course, neither is super lame, high scale regression analysis. I daresay that the only viable strategy, in broad terms, is sophisticated theorising to produce a precise hypothesis that can then be tested with weak data. It is bizarre to me that such a hypothetico-deductive style of research (a fancy way of saying scientific method) is presently out of fashion in economics. Data doesn't make you scientific on its own!