The four horsemen: what the studies did and did not show

By EQTiger Editorial. Review status: needs-expert-review. Qualified review is pending; the review date records an editorial check. This is general education for adults, not therapy or couples counseling, and EQTiger does not diagnose anyone, including your partner.

What the four horsemen are

The phrase comes from the psychologist John Gottman, whose lab spent decades filming couples while they discussed a disagreement, coding every facial expression and remark, and following the couples for years to see who stayed together. Out of that work came four behaviors that the Gottman Institute now teaches as warning signs (The Gottman Institute, 2013):

The suggested alternatives in the right column come from the Institute's own practitioner writing (The Gottman Institute, 2013). They are sensible, and many of them line up with other research, but they are teaching recommendations, not the direct findings of a particular trial.

What the studies actually did

The horsemen grew out of several observational studies rather than one. Three are worth knowing because they are the ones most often quoted.

The 1992 study. Gottman and Robert Levenson brought 73 married couples into the lab in 1983, recorded their conversations and their physiology, and followed up in 1987 (Gottman and Levenson, 1992). Couples whose interactions tilted negative showed more defensiveness, stubbornness and withdrawal, were less satisfied, had poorer later health, and were at greater risk of separating. It was a careful study, but it was one lab and 73 couples, and the authors described their model as only preliminarily supported.

The 1998 newlywed study. This is the source of the most famous claims. Gottman and colleagues observed 130 newlywed couples in conflict discussions and followed them over time (Gottman and colleagues, 1998). Harsh start-up, a husband rejecting his wife's influence, and failing to calm things down predicted later divorce, and the paper reported predicting which couples would divorce with 83 percent accuracy. It also reported that some popular ideas, such as active listening exercises, were not supported by the data.

The 14-year timing study. In 2000, Gottman and Levenson reported on couples followed for 14 years (Gottman and Levenson, 2000). Negativity during conflict predicted divorce early in marriage, while a lack of positive feeling, in arguments and in ordinary conversations, predicted divorce later on. The paper reported 93 percent accuracy.

A related study found that the first three minutes of a 15-minute conflict discussion, coded for emotion, predicted which of 124 newlywed couples divorced over six years (Carrere and Gottman, 1999). It is the research behind the popular idea that how a conversation starts tends to shape how it ends.

What the studies did not show

Three cautions matter more than any single number.

The accuracy figures are not what they sound like. An 83 or 93 percent "prediction" was calculated by fitting a model to the same couples whose outcomes were already known. That tends to overstate how well the model would do with new couples. Richard Heyman and Amy Smith Slep demonstrated the problem with archival data: when divorce-prediction equations were tested on couples they had not been built from, their accuracy dropped sharply (Heyman and Smith Slep, 2001). The 1998 and 2000 papers report their accuracy without that kind of test on new couples, a gap critics pointed out (Gottman and colleagues, 1998) (Gottman and Levenson, 2000). So there is no basis for the claim, common online, that a therapist can watch you argue for fifteen minutes and tell you with 90 percent certainty whether you will divorce.

Correlation is not cause. All of these studies observed couples rather than changing how they fought. Contempt might help cause divorce, or it might be a sign that a relationship is already failing for other reasons, or both. When Scott Stanley, Thomas Bradbury and Howard Markman reviewed the 1998 paper, they argued that its practical recommendations, including dropping active listening and focusing on gentler start-up by wives, went beyond what the data could support (Stanley, Bradbury, and Markman, 2000).

The samples were narrow. These were mostly married, heterosexual couples studied in one research program in the United States. The findings may well apply more widely, but the original studies cannot show it.

Where the idea still holds up

None of that makes the four horsemen useless. The behaviors themselves are well defined and easy to recognize, and the broad direction of the findings is consistent: relationships where fights are full of attack, scorn and shutdown tend to be unhappy ones. A study of conflict styles in the same program proposed that couples can be stable with quite different ways of fighting, whether calm and validating, heated and volatile, or conflict-avoiding, while hostile and hostile-detached patterns were the unstable ones (Gottman, 1993). Later survey research reproduced similar types and found hostile couples the least satisfied (Holman and Jarvis, 2003). That is a more useful way to read the horsemen: the problem is not disagreeing, or even raising your voice, but the slide into treating your partner as an opponent.

Other research also cautions against simply being nicer in every fight. In a year-long study of 61 couples, direct requests for change, even negative ones, felt worse in the moment but predicted more real change than gentle hints (Overall and colleagues, 2009). A "gentle start-up" works best when it is still a clear request, not a softened version that never says what you need.

What to do with this

1. Learn to spot the four behaviors in yourself first. They are much easier to see in a partner.
2. Swap criticism for a complaint about one specific event plus one request. See criticism vs complaint.
3. Treat contempt as the most urgent of the four. Eye-rolling and mockery are worth stopping even if you are right about the issue. See contempt and eye-rolling.
4. When you notice yourself shutting down, name it and take a real break with a return time. See stonewalling.
5. Do not use the horsemen to score your partner or to predict your future. Use them as a short list of habits to change.

If contempt comes with threats, control or fear, it is no longer a conflict habit, and the right next step is the safety resources on this page rather than better phrasing.

Key studies

Start with the free EQTiger assessment (adults 18+), or browse more relationship help.