John Gottman's Research

I recently read The Seven Principles for Making Marriage Work (my review here), a marriage counseling book written by John Gottman. In it, he makes a lot of claims about the research, experiments, and validity of his curriculum. I set off to determine whether Gottman’s research holds up to scrutiny, and boy, did I fall down the rabbit hole.

Gottman’s Research

Right on the front cover of his book, you see “from the Country’s Foremost Relationship Expert.” And in the first few pages, you’ll see this claim:

Plenty of people consider themselves to be experts on marriage— and are more than happy to give them their opinion… but that’s the key word— opinion.

It’s pretty clear that Gottman sees himself as a researcher, and has a high level of confidence in his results.

Divorce Prediction

Gottman’s most controversial research is what led him to the 7 principles: his experiments watching couples fight, then analyzing their behavior to make predictions about whether they’ll divorce.

Over seven separate studies, my accuracy rate in making predictions has averaged 91 percent. In other words, in 91% of the cases where I predicted that a couple’s marriage would eventually either fail or succeed, time proved me right.

— John Gottman, T7PfMMW

Statements about the 94% accuracy rate of divorce prediction have become a source of confusion.

— The Gottman Institute FAQ

Yeah, you could say that again! I spent an absurd amount of time absorbing research, Wikipedia talk pages, book reviews, and several scientific studies, and I’m starting to feel like I have some understanding of Gottman’s research story.

Prediction vs Postdiction

Okay, first you need to understand the difference between prediction and postdiction.

Prediction is pretty straight forward, and you probably already understand it. You start with a hypothesis, like “people who eat carrots will get better eyesight.” Then, you test your hypothesis by observing whether your prediction occurs.

Postdiction is less intuitive. This process aims to explain an event AFTER it happens. For example, if it rains Monday, then you walk outside on Tuesday and see that your yard is wet, you might theorize “my yard is wet because it rained yesterday.” You would be correct, but it would be a dangerous generalization to say “anytime my yard it wet, it rained the day before,” because that’s obviously not always true. It could be morning dew, or the sprinkler system.

See the difference? In prediction, you’re guessing about the future, and you can easily test your theory. In postdiction, you’re looking at something that’s already happened, and theorizing why it happened.

Postdiction is dangerous, because it’s really easy to look at your data and manufacture a pattern that “solves” your problem, but the solution might only fit that specific case and not generalize well to other situations. Postdiction can give you a false sense of certainty because you’re working backward from a known outcome, making it tempting to overlook other potential causes or overfit your explanation to the data. In contrast, prediction requires testing a hypothesis in advance, giving you a better sense of whether your theory holds up across different cases, not just in hindsight.

Okay, Back to Gottman

It all begins in 1980, when Gottman does a study to “help him determine what factors were useful in predicting divorce.” This study was definitely post-hoc analysis, meaning Gottman was postdicting divorce. His own FAQ says so!

Dr. Gottman did an initial post-hoc analyses study back in 1980 to help him determine what factors were useful in predicting divorce.

— The Gottman Institute FAQ

But, after this initial study, Gottman claims to have done 6 studies consisting of “true prediction” (emphasis mine):

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.

— The Gottman Institute FAQ

Okay, the way this FAQ describes it, that sounds truly predictive!

The first major criticism of Gottman’s research comes out in 2001, when Richard Heyman publishes a paper calling out Gottman’s studies, titled The Hazards of Predicting Divorce Without Crossvalidation. Heyman makes two big claims in here. First, Heyman calls Gottman’s “prediction” more like postdiction:

Discriminant function analyses and logistic regression can be used to predict a categorical outcome (e.g., divorce) in any given sample. More accurately, however, we should say that the analyst asks the software to reconstruct, rather than predict, because the computer develops an equation to optimally reconstruct an already-known group status. This is not a trivial, semantic distinction.

The second big claim is that Gottman’s studies didn’t include crossvalidation, which basically means that Gottman didn’t test his divorce-predicting-equations on any data outside of the training data he used to come up with the equations. In the study, Richard Heyman comes up with a new, very accurate divorce-predicting equation (90% accurate), then crossvalidates it. The results show that Heyman’s “accurate” equation was actually terrible (29% accurate). So, without crossvalidation, we don’t really know if Gottman’s equations are actually accurate.

And just to be sure Richard Heyman’s claim of Gottman’s postdiction is accurate, I wanted to check one of Gottman’s “true prediction” studies. Unfortunately, the current FAQ does not list any references, but an older version of the page does.

I pulled up one of Gottman’s “true prediction” studies from 1998 titled “Predicting Marital Happiness and Stability from Newlywed Interactions”. It contains this important paragraph in Results:

For all sequences, this procedure was followed: A sequence first was examined that indexed a particular process model. If the results turned out to be statistically significant, a covariance analysis also was conducted, using the frequency of the consequent code as the covariate.

— Gottman et al., 1998

What does that even mean? I’m not a scientist, but my current understanding is this: after they gathered the data on which couples divorced, they sequenced several “process models” (AKA divorce prediction theories), and then re-analyzed the best ones to see which variables predicted the divorce. This is true postdiction.

If you trust Richard Heyman’s paper, then the other “true prediction” studies are likely to show similar results. It’s my understanding that when Gottman uses the phrase “true prediction”, it means that the variables are chosen in advance (e.g. the “models” in the above study), but the exact prediction equation is determined post-hoc.

I think this comment, which I found on a blog discussing Gottman’s Wikipedia page, sums it up pretty well:

Gottman’s real error seems to be he doesn’t seem to have attempted to propose one unified model that can be validated with multiple independent samples. He is able to generate significant predictive models within a single study (which is equivalent to finding a correlation between divorce occurrence and a given set of predictors); but use of the word “prediction” in a larger context seems to presuppose more rigorous attempts at validation.

— Hamdan Azhar, “A Wikipedia whitewash” comment

Does Gottman’s Curriculum Work?

It’s interesting to analyze and critique Gottman’s divorce prediction claims, but it’s really secondary to this book. To me, the more important question is “does it work?” I think Gottman agrees:

Although the ability to predict divorce through laboratory research has made for great advances in the study of long-term relationships, I don’t think it is this project’s most significant contribution. Instead, I think my most rewarding discovery has been the Seven Principles, which aren’t just about predicting divorce but also about preventing it. Unlike so many other approaches to helping couples, ours is based on knowing what makes marriages succeed rather than on what makes them fail.

— John Gottman, T7PfMMW

Gottman cites at least one study:

a randomized clinical study… found that married couples who simply read The Seven Principles and worked through the quizzes and exercises on their own (but received no additional professional aid) were significantly happier in their relationship, and these effects lasted when assessed a year later.

— John Gottman, T7PfMMW

Gottman’s Study

I did find this study: “A component analysis of a brief psycho-educational couples’ workshop: one-year follow-up results”. In the study, there are 4 groups, and one group is simply assigned “bibliotherapy”. Tragically, there is no control group.

TL;DR: Small sample size (16 couples), they didn’t check if the couples who got the book actually read the book, but still, of the couples that were assigned the book, there was improvement: 75% of husbands and 56% of wives improved (more than std error), and 19% of husbands and 6% of wives deteriorated. Not too shabby, especially considering some of those deteriorating couples may not have used the book at all.

Study 1

There is an independent study published in the Iranian Journal of Psychiatry titled “Examining the Effectiveness of Gottman Couple Therapy on Improving Marital Adjustment and Couples’ Intimacy”. The study done on a small sample size of 16 couples (8 control) but found the following:

The results of the present study confirmed the effectiveness of Gottman’s couple therapy on improving couples’ intimacy

Study 2

Another study created their own curriculum based on Gottman’s “Sound Relationship House Theory (SRHT)” which appears to align with his 7 principles, then delivered the training online and measured specific outcomes:

Evaluation of an Online Gottman’s Psychoeducational Intervention to Improve Marital Communication among Iranian Couples

Despite the high effectiveness of Gottman’s psychoeducational intervention, there is a lack of study on the online Gottman’s psychoeducation intervention (O-GPI) to improve marital communication and dyadic adjustments. The aim of this study was to evaluate the effectiveness of O-GPI on the improvement of marital communication patterns among Iranian couples. Method: The study followed a single-blind parallel group in a randomized controlled trial using an experimental longitudinal design, comprising 72 heterosexual couples living in Shiraz, Iran, with a 1–7-year marital age and no severe marital problems. The experimental group received eight consecutive O-GPIs via the Zoom platform, while the control group received information related to parenting skills via email.

The protocol of the online Gottman’s psychoeducational intervention (O-GPI) was based on the Sound Relationship House Theory (SRHT) by Gottman

Hence, this study concluded that the online Gottman’s psychoeducational intervention is an effective intervention with standard effects without gender approaches to improve marital communication among 72 Iranian couples. The results suggested that couples who participated in Gottman’s intervention had significantly greater enhancements in their constructive communication than those who did not. Additionally, they meaningfully decreased their demand–withdraw communication and mutual avoidance.

The Bottom Line

In my opinion, Gottman’s early “divorce prediction” research was important for leading him to his 7 principles, but it was done poorly and communicated poorly to this day. His claim that he can “predict divorce” with high accuracy in only a few minutes is shaky at best. That research is a scar on his reputation that may never fully heal. However, based on multiple studies I’ve seen, there is validity and effectiveness in Gottman’s marital curriculum— he’s correct about the success of his training.

So yes, read The Seven Principles for Making Marriage Work!