On causality and the returns to late marriage

Miss Havisham's vast wealth is undoubtedly due to her failure to marry.

Miss Havisham’s vast wealth is undoubtedly due to her failure to marry.

Over the last week or two there has been a fair bit of chatter about a report released by the University of Virgina’s National Marriage Project. While the results pertain primarily to marriage outcomes in the United States, the general interpretation of those results is a textbook case of making strongly-causal statements using methods which are insufficient for making these claims. These two graphs, documenting the earnings of women and men by age of marriage, provide a starting point for the discussion:

knotyet_income

knotyet_incomemen

 

These descriptive stats paint a pretty clear picture, allowing us to make the following statements:

  • On average, women who have married at a later age also tend to have higher incomes. 
  • On average, men who have married at a later age mostly have lower incomes (there is a bit of an inverse relationship here, especially at higher levels of education)

These statements are not causal: I can easily say “women who have higher incomes tend to marry at a later age,” which is an equivalent point to the one above. It is just a descriptive statement. Contrast these statements with quotes from these articles on the study, including one from the chief author, Brad Wilcox:

These highly educated adults have embraced a “capstone” model of marriage that typically leads them to put off marriage until they have had a chance to establish themselves professionally, personally, and relationship-wise. This capstone model is paying big dividends to the college-educated: Their divorce rate is low, and their income is high. We find, for instance, that college-educated women who postpone marriage to their 30s earn about $10,000 more than their college-educated sisters who marry in their mid-20s

From Ross Douthat in the New York Times:

Upper-class women reap a large wage premium from delaying marriage — a college-educated woman who marries in her 30s earns over $15,000 more annually than a woman who marries in her early 20s, and when you look at household income, the premium for marrying later rises to more than $20,000. Women without 4-year degrees also enjoy a wage premium when they delay marriage, albeit a smaller one (and a very small one when you look at household income). Men, meanwhile, reap a wage premium from marrying earlier, so late marriage tends to hurt their economic prospects.

From Eleanor Barkhorn in the Atlantic:

Financially, college-educated women benefit the most from marrying later. Women who marry later make more money per year than women who marry young.

Using the above data as a basis for their arguments, all of these authors, are, to varying degrees, are making implicit or explicit causal statements: delaying marriage is good for women and bad for men. Yet, given that the Wilcox et al. study is strictly observational, with (as far as I can tell) little effort being made to discern a causal relationship between age of marriage and labour market outcomes, we’re really far more limited in what we can say. Take, for instance, a model of the marriage `market’ where women want to be picky and marry late and high income acts a bargaining chip in the matching process. Richer men will inevitably be able to secure a bride at a much earlier age and richer women will inevitably be able to stave off marriage and find a good husband at a later age. Suddenly, it’s income affecting the age of marriage, not the other way around.

I am not claiming that this model represents “the truth” and that the prevailing explanation doesn’t – far from it, but we can come up with a million different explanations for the correlation observed above which do not involve a direct causal relationship between delaying marriage and income. In general, be cautious when you’re presented with simple stories based on descriptive statistics, both in work like this as well as development research.

One thought on “On causality and the returns to late marriage

  1. Michael

    March 25, 2013 at 3:50pm

    While, I’ll agree that these statements are not justified based on the pure analtyic methods, it’s also true that we’ll unlikely ever have rigorous causal evidence to prove (or disprove) this theory (perhaps at best, some interesting IVs).

    A significant question therefore becomes… does this evidence update our priors about what the correct theory is?

    http://www.slate.com/articles/health_and_science/science/2012/10/correlation_does_not_imply_causation_how_the_internet_fell_in_love_with_a_stats_class_clich_.html

Comments are closed.