The Rocky Science of Good Marriages

Last weekend, Elizabeth Weil's New York Times Magazine story detailed her marriage to Dan Duane, and told of their many year-long adventures with self-improvement gurus in hopes of making their marriage better. During one phase of this venture, the couple attempt to become better communicators. Inevitably, she mentions one of the most famous marriage scientists of all, Dr. John Gottman.

"John Gottman, in his Love Lab in Seattle, claims that he can analyze a conversation between spouses and predict with 94 percent accuracy whether that couple will divorce over the course of six years."

Weil and Duane don't visit the Love Lab – they go to a marriage-education class in Marin, instead. So the story quickly moves on, but not without Weil dropping this little caveat:

"But many academics say that Gottman's powers of prophecy are overblown, that he can't truly predict if a couple will split."

I don't know if Weil is fully-versed in this science or not; she doesn't go into it during her article. But I know the editors at the Times magazine are aware, because in 2007 Ashley and I filled them in on the controversy. At the time, Malcolm Gladwell's Blink topped the bestseller lists, and the book touted Gottman as prime example #1 of blink-thinkers. As a result of the profile, Gottman was more famous than ever, and yet the reading public was completely unaware that many of Gottman's extravagant claims were being contested behind the scenes.

I don't relish pointing this out again now. I think Gottman's advice for couples is tremendously helpful. He has helped thousands of couples with his books, seminars, and science. He's done incredibly important work on gay and lesbian couples. His graduate students have gone on to study incredibly important social science phenomenon, and have become noted scholars at so many universities. He's also one of the great institutions of my hometown, Seattle.

Nevertheless, his money line – the attention getter – has always been his 90%+ accuracy rate for predicting which marriages will end in divorce. All he has to do is see you and your spouse fight. Or more specifically, he'd analyze the first three minutes of a videotape of you and your spouse discussing something you regularly fight over. Gottman's system is a combination of physiology and psychology – measuring heart rates and skin reactivity, analyzing every word and jerk of the head when you argue with your partner.

Not surprisingly, the premise that science can answer the eternal mystical question, "What makes a good marriage?" – with such incredible mathematical certainty – has made Gottman a media darling. Every year, his astonishing ability to predict marriage outcomes creeps higher – 92.7%, then 94%, and lately I've seen reports that he's at 96%.

That number is really his claim to fame, and there's always been a certain braggadocio about it. But it makes other scholars cringe.

They say his "predictions" are post-hoc (he never actually made predictions). His sample sizes are tiny (one of his conclusions is based on only 46 couples; the other on 17 couples). They say he removes couples from his surveys for unexplained reasons. They say he uses a statistical sampling technique that artificially boosts his results. One of Gottman's co-researchers and co-authors, Julie Babcock, subsequently published a paper wherein she failed to reproduce some of his earlier claims.

The language from scholars is blunt. Gottman's findings are unreliable, they say. His findings should not be applied to any clinical setting.

Gottman's biggest problem, though, is that he gives the clear impression his system applies to couples out there in America, not just to small samples drawn from his Love Lab. However, scholars have not been able to reproduce his findings when applying them to other populations. His system's predictive accuracy drops to somewhere between 20% and zero.

For example, in 2007, scholars at the Oregon Social Learning Center (who were trained by Gottman) attempted to apply his system to a population of couples they were studying in Oregon. To their surprise, Gottman's model didn't rate out at 90%, or even 80%, or 70% - it was basically useless and invalid. It failed the test – on all counts. They published their results in the preeminent Journal of Marriage and Family, the same publication that launched Gottman in 1998. Gottman defended himself in the journal's pages, saying the Oregon couples were too different from his Seattle couples (such as the Oregon study included cohabitating couples, not just married couples.) Gottman has some valid points, but so do the Oregon scholars, which you can sample here.

I've studied this some, but not nearly enough to opine on who's right. Ashley and I have long talked about digging into it more deeply. But there's enough disagreement among scientists to be wary of buying the "94 percent" prediction, as every other journalist seems to have done over the years.