Doubling down

Commenter adam pointed out that I am not alone in challenging Claire Cain Miller’s NY Times article on divorce.  Professor Steven Ruggles of the University of Minnesota has likewise refuted Miller’s claim that the public’s understanding has failed to keep up with what she asserts is settled science on divorce rates:

The article recycles old research based on bad data. As Sheela Kennedy and I demonstrated in our recent article “Breaking Up is Hard to Count: The Rise of Divorce in the United States, 1980-2010” Demography (2014), available at http://www.hist.umn.edu/~ruggles/Articles/breaking_up.pdf, the much-vaunted decline in divorce is an artifact of bad data and poor measurement. As we show, the only reliable data on current U.S. divorce rates derive from the American Community Survey (ACS). Controlling for the aging of the married population, the ACS data reveal a continuing and dramatic increase in the risk of divorce since 1990. The rise of divorce is especially striking among older adults: among those aged 55 to 64, the divorce rate has quadrupled over the past three decades.

The Miller article appears to be based mainly on a working paper by Betsy Stevenson and Justin Wolfers that first appeared in 2007 and then was published in an obscure British volume of collected essays in 2011. Wolfers is a contributor to the Upshot, but I don’t believe Miller actually interviewed him. I know he is well aware of more recent research based on new data, and I very much doubt that he would endorse the premise of the Miller article.

In response, Professor Justin Wolfers of the University of Michigan followed up with a defense of Miller’s original article, How We Know the Divorce Rate Is Falling.  Wolfers’ defense meanders around the data, and at times he contradicts himself:

There is no question that the data on divorce, like so much social-science data, are imperfect. It’s a good thing that researchers are pointing out these imperfections, because we will get a clearer picture of reality if we can get better data. But the best evidence all points in the same direction: The old claim that one in two marriages ends in divorce is no longer true.

Of course, we don’t know what will happen in the future. Even if recently married couples are divorcing at lower rates, there is some reason to think that divorce rates among long-married couples will rise.

First he says the lifetime divorce rate (a forward looking number) isn’t 50%.  Then he explains that we can’t know what this number will look like today, because it involves guessing the future.  His only evidence that the forward looking number will probably be lower than 50% is a small decline in early divorce rates.  However, he then points out that the divorce rates for older couples is increasing.

It is important to remember that Wolfers’ article is a defense of Miller’s claim that the science is settled, and divorce rates have declined since the early 1980s.  Wolfers opens his article with:

The divorce rate has been falling for more than three decades. That fact is not news, but it still surprises a lot of people. And so when Claire Cain Miller wrote about the trend for The Upshot this week, several readers asked for more detail, with some citing a Huffington Post article questioning the official Census Bureau data on the trends. As one of the researchers who has studied the issue, I thought it worth digging deeper into the data.

In order to back up Miller’s claim, Wolfers would have to prove that the science really is settled here, and that those who don’t believe that divorce rates have declined for 30 years believe in a “myth”.  Instead Wolfers points to problematic and contradictory data, and argues that none of the data sources available can be trusted to accurately represent current divorce rates.  With this in mind, even with the most charitable possible reading Wolfers failed to defend the core claims in Miller’s article, and therefore failed to support his own thesis.

Professor Ruggles responded to Wolfers’ article as well.  Ruggles pointed out specific flaws in Wolfers’ argument, along with the fact that Wolfers had failed to support his own thesis:

Steven Ruggles Minneapolis Yesterday
Dear Justin,

As we noted in the paper, in the 2006 pretest the Census Bureau found that 7.8% of women who reported getting a divorce in the past year did not actually receive a divorce decree in that period. The pretest was not able, however, to determine what percentage of the women who actually did get a divorce in the past 12 months failed to report it on the ACS. Overcount and undercount would cancel, so the pretest implies a net overcount of less than 7.8%.

If we control for the aging of the married population, divorce risk has gone up 40% since 1980. To erase that, you have to assume that the ACS is off by a factor of two. I don’t buy it. It is much more plausible that the discrepancy lies with your alternative source, the SIPP, which has substantially lower divorce rates than the vital statistics, and where nearly half of the divorce dates are imputed in the most recent data.

The number of demographers who believe that overall divorce risk has declined is small. Other than Stevenson and Wolfers, we identified only Heaton (2002) and Ivers and Stevenson (2010). The consensus of most demographers, as Schoen and Canudas-Romo (2006) put it, “it is premature to believe that the probability of divorce has begun to decline.” You are entitled to argue that ACS is wrong and SIPP is right. Nevertheless, I think you should acknowledge that the decline of divorce narrative is a minority viewpoint among professional demographers.

Steve