John Gottman (a prominent psychological researcher who has done extensive work on divorce prediction) has some very strong claims about his ability to predict divorce. The claim that led me to start looking into this topic is that he can predict which newlywed couples will divorce with 94% accuracy. He repeats this claim on his website:

But our breakthrough research with thousands of couples has discovered six telltale signs that can predict — with over 94% accuracy — whether a couple will break up within the next four years

In his research FAQ, Gottman makes more claims about his predictive abilities, many of which are very strong:

Dr. Gottman’s ability to predict divorce among newlyweds is more clearly understood by imagining an urn that contains 130 white balls (representing couples that stayed married) and 17 red balls (representing couples that ended up divorcing) for a total of 147 balls. The chances that Dr. Gottman could blindly pick balls out of the urn and guess which were red and which were white with 90% accuracy could only happen by chance 1 x 10-19 times. That is the number point one (0.1) with 18 zeroes in front of the number one. This means it is practically impossible that Dr. Gottman could predict which couples would divorce with much accuracy by chance alone. The factors he used to make his predictions were indeed clearly related to why couples ended up divorced. By looking for those factors, he was able to predict divorce fairly accurately. For the Gottman, Katz and Hooven study, where Gottman et. al. picked out all seven divorced couples out of 56, the probability is approximately .000000000384 or 3.84×10-9.

Gottman's methods have faced criticism on the basis that they are all post-hoc studies. Critics (such as this article) allege that he is merely building a formula to match known outcomes after the fact, rather than truly predicting outcomes. Gottman refutes this claim in his research FAQ:

Altogether, Dr. Gottman has completed seven studies that explored what predicts divorce. [...] Six of the seven studies have been predictive—each began with a hypothesis about factors leading to divorce. Based on these factors, Dr. Gottman predicted who would divorce, then followed the couples for a pre-determined length of time. Finally, he drew conclusions about the accuracy of his predictions. [...] This is true prediction. Prior to his six prediction studies, Dr. Gottman did an initial post-hoc analyses study back in 1980 to help him determine what factors were useful in predicting divorce.

However, I've been unable to find any of these predictive studies. In his research FAQ, he specifically cites two of his own studies: Gottman & Levenson, 2002 and a study by Gottman, Katz and Hooven. The Gottman & Levenson, 2002 study has faced strong criticism (see DeKay, Greeno, and Houck, 2002) for (among many other problems) only studying couples that are already divorced. I've been unable to find a study on divorce by Gottman, Katz and Hooven - on his list of research articles, he only shows four papers by those three authors and all of them deal with parent/child relationships and outcomes, not divorce. I haven't been able to turn up any other articles besides those four in my own research.

Has John Gottman (or the Gottman Institute) actually shown predictive ability with their methods?

  • 11
    ... 1 x 10-19 times. That is the number point one (0.1) with 18 zeroes in front of the number one ... - That is possibly the worst written description of a small standard form I've seen. Also, worth noting that if he just guessed all couples would stay together, in the US he'd have ~60% success rate. If you're actually looking into problems in a marriage, it doesn't seem that outrageous that 94% is possible. Picking balls out of a bag is random, social relation problems really aren't, we're very good at detecting them... Commented Sep 13, 2019 at 10:33
  • 5
    The FAQ seems odd as @Fifth_H0r5eman mentioned, but I also think it sounds plausible to be able to predict who will divorce in four years (not who will divorce EVER). After all, four years is not THAT long, and most issues are probably already present years before the actual divorce.
    – kutschkem
    Commented Sep 13, 2019 at 10:55
  • 3
    "For the Gottman, Katz and Hooven study, where Gottman et. al. picked out all seven divorced couples out of 56, the probability is approximately .000000000384 or 3.84×10-9.": Do they explain that calculation? I mean, if we assume he knew it'd be 7 divorces in 56, then that's (7/56)*(6/55)*(5/54)*(4/53)*(3/52)*(2/51)*(1/50), which is ~4.31*10^-9. Or if we don't assume that he knew it'd actually be 7, then with a binomial distribution and 12.5% odds, it'd be ~15.9% chance of getting 7, for ~6.87*10^-10. So where'd they get "3.84×10-9" from?
    – Nat
    Commented Sep 13, 2019 at 15:17
  • 22
    Without meeting the 147 couples, I could make individual predictions about whether they divorce in four years which would would turn out to be 88.4% accurate when 17 of them had divorced: all I would have to do is say each couple would stay together. It does not make 90% look so impressive
    – Henry
    Commented Sep 13, 2019 at 22:37
  • 1
    One cannot only look at accuracy and evaluate predictions. You should enquire about false positives and false negatives. Look up about precision and recall. Especially when the data is imbalanced ( less number of divorcees ), predicting all remain together gives very good accuracy. Look up Henry's comment. But, the recall will be zero in this case and will mean that the model is no good.
    – naive
    Commented Sep 29, 2019 at 11:41


You must log in to answer this question.

Browse other questions tagged .