Conflict Style Is More Predictive Than Shared Interests — Here's How to Test Yours

Two people both put "hiking" in their profile. They match. Six weeks later, they're standing in a kitchen at 11pm having the same fight for the third time — one of them escalating, the other going quiet and flat, both convinced the other one is doing something unforgivable.

Nothing about the hiking was a lie. It just wasn't load-bearing.

The uncomfortable thing about dating profiles is that they inventory the wrong layer of a person. Interests are the top layer: visible, cheap to signal, easy to share, and almost irrelevant to whether a relationship survives its first real disagreement. What sits underneath — how you behave when you feel criticized, what you do when your heart rate spikes, whether you can take a joke mid-argument — is the layer that determines the outcome. Nobody lists it. Nobody's ever been asked.

What Gottman actually measured

The best-known research program on this question spent decades watching couples argue rather than asking them what they liked.

In 1992, John Gottman worked with Kim Buehlman and Lynn Katz on a study of 52 married couples, interviewing them about the history of their relationship — how they met, how they decided to marry, how they handled hard times — and then following up. The researchers reported predicting which couples would separate or stay together three years later with over 94% accuracy.[1] A later University of Washington study of newlyweds using the same Oral History Interview reported 87% accuracy in predicting divorce and marital stability.[2]

Here's the part worth tattooing on the inside of your eyelids: the coding system doesn't score the content of the answers. It scores the delivery. Buehlman, who developed the coding, put it plainly — "How people talk, not the content, is the key."[2] Two couples can tell structurally identical stories about a rough first year. One tells it as a joint achievement; the other tells it as a grievance with a witness. Same facts. Different marriage.

Out of the same research tradition come the four horsemen — criticism, contempt, defensiveness, stonewalling — with contempt typically described as the strongest single predictor of dissolution, and with stable couples maintaining roughly five positive interactions for every negative one during conflict.[3]

The honest caveat

These accuracy numbers deserve an asterisk, and you should distrust anyone who quotes them without one. Critics — most prominently journalist Laurie Abraham, whose analysis was widely discussed by statistician Andrew Gelman — pointed out that the headline percentages were generally obtained by fitting a model to a dataset where the outcomes were already known, rather than by predicting fresh couples in advance. That's model fit, not prophecy. Nobody has demonstrated a 94%-accurate divorce oracle.

But the weaker claim survives intact and is the only one we need: the way a couple handles conflict carries real, measurable information about the relationship's future, and it carries more of it than the couple's shared preferences do.

Why "we like the same things" is a weak signal

There's a broader finding that makes this sting. In 2020, a team led by Samantha Joel pooled 43 longitudinal datasets of couples and threw machine learning at the question of what predicts relationship quality. The strongest predictors were relationship-specific — how you experience this particular partnership. Individual traits added essentially nothing on top of that, partner-reported variables added nothing beyond a person's own reports, and change in relationship quality over time was largely unpredictable from any combination of self-report variables.

Read that carefully, because it cuts both ways. It's bad news for personality-matching maximalists — the idea that your trait profile alone determines your fate doesn't hold up. But it's much worse news for interest-matching, which isn't even a trait; it's a hobby list. If your stable dispositions barely predict outcomes on their own, your mutual enthusiasm for climbing gyms predicts nothing at all.

What does live in that relationship-specific layer? Appreciation, perceived commitment, sexual satisfaction — and conflict. Conflict is the one variable in that cluster that has a recognizable shape before the relationship exists. You bring your conflict style with you. It was installed years ago, and it will show up in week nine whether or not you disclosed it in week one.

We've written before about how matching works at the trait level. Conflict style is the layer beneath it — less about who you are at rest and more about who you become under load.

Translating the horsemen into questions an avatar can ask

The problem with the four horsemen as a dating filter is that they're behaviors, not self-descriptions. Ask someone directly — "are you contemptuous during arguments?" — and you get a 100% denial rate, including from people who roll their eyes so hard they can see their own occipital lobe.

So you don't ask directly. You ask for narrative and score the delivery, exactly as the Oral History coding does. Four translations:

Criticism → the attribution test. Criticism is the move from "you did X" to "you are the kind of person who does X." So: Tell me about the last time a partner disappointed you. Then listen for whether the story stays behavioral or slides into character. "He forgot my birthday" versus "he's fundamentally selfish" are different diagnoses of the same evening. Watch for words like always, never, typical.

Contempt → the humor direction. Contempt is criticism plus altitude. It's the belief that you're arguing from a superior position. Probe: What's something a past partner did that you found ridiculous? The tell isn't whether the person is funny. It's whether the joke punches down at someone who was supposed to be on their team, and whether the story is told to an imagined audience — the friend group, the jury, you.

Defensiveness → the accountability inventory. Probe: Describe a fight where you were mostly in the wrong. What did you actually say? The failure mode isn't refusing to answer; it's the counter-charge smuggled inside the confession. "I shouldn't have snapped, but you have to understand what I was dealing with." A person who can produce one clean, unhedged sentence of ownership is signaling something rare.

Stonewalling → the flooding map. Stonewalling is often physiological rather than strategic — the nervous system floods and the person goes silent because they've effectively left the building. Probe: When an argument gets intense, what happens in your body? What do you do next? You're looking for self-awareness and, crucially, for whether the withdrawal comes with a stated return time. "I need twenty minutes" and "I need to not be here" produce very different marriages.

And the fifth question, the one almost nobody in dating asks: the repair attempt. Repair attempts are the small bids to de-escalate mid-fight — a joke, a hand on the arm, "wait, can we start over?" They're the mechanism behind the positive-to-negative ratio that separates stable from unstable couples.[3] The relevant question isn't whether you make them. It's whether you accept them. Plenty of people offer repair beautifully and refuse it flatly, because accepting a repair means the fight ends before you've won.

Why two hikers can be structurally incompatible

Here's the part that a shared-interest algorithm can never see.

Conflict styles aren't just good or bad. They're shaped, and shapes can mismatch even when both are healthy. The classic geometry is pursue–withdraw: one partner needs to resolve now, out loud, in the room; the other needs to cool down before they can think. Neither is broken. But in combination they form a feedback loop where the pursuer's escalation triggers the withdrawer's shutdown, which triggers more pursuit. Both partners are behaving rationally by their own operating logic, and the system fails anyway.

Same with pacing. Some people flood at a 3 out of 10 and need a hard break. Others don't register the argument as serious until it's a 7 and read an early exit as abandonment. Same with repair vocabulary: humor lands as relief for one person and as dismissal for another. A joke offered as a repair attempt and received as contempt is how two decent people end up in a fight about the fight.

None of this is visible from "loves hiking, dogs, and bad sci-fi." All of it is visible in about forty minutes of well-structured conversation — which is precisely the kind of conversation nobody has on a first date, and precisely the kind an avatar can have on your behalf before either human invests an evening. That's the whole design premise: create your avatar, let it run the conversations that are too awkward, too slow, or too revealing for a Tuesday-night drink.

Four prompts you can answer tonight

Open a notes app. Twenty minutes. Be specific — write the actual sentences you said, not summaries.

1. The disappointment story. Describe the last time a partner let you down. Then reread it and highlight every phrase that describes character rather than behavior. High ratio = criticism is your default entry point into conflict.

2. The wrongness story. Describe a fight you were mostly wrong in. Then delete every clause after the word "but." Is anything left? If the ownership evaporates when you remove the justifications, defensiveness is your load-bearing wall.

3. The body log. Recall your last real argument. Where did it live physically — chest, jaw, hands, stomach? At what point did you stop hearing what the other person was saying? That moment is your flooding threshold, and knowing its rough location is worth more than any personality label.

4. The repair audit. Think of the last three times someone tried to de-escalate with you — a joke, an apology, a touch, a subject change. Did you take it? Write down what you said in the ten seconds after. The pattern in those ten seconds is, honestly, more predictive of your next relationship than everything on your dating profile combined.

None of these produce a verdict. Conflict style isn't a fixed trait like height; it's closer to a habit with deep roots, and it shifts with the partner, the stakes, and how much you've slept. That's the good news. Interests are static facts about you. Conflict style is a skill, and skills move.

But you can't move what you haven't measured — and you certainly can't match on it if the only thing your profile says is that you like the outdoors. If you want a structured version of this rather than a notes app, take the tests and see what your own delivery says about you.

The people who love the same things as you are easy to find. The people who fight in a shape that fits yours are the ones worth building a matching system for.

Sources

1. The Research: Predicting Divorce from an Oral History Interview — The Gottman Institute
2. UW researchers can predict newlywed divorce, marital stability with 87 percent accuracy — University of Washington News
3. What Is the Gottman Four Horsemen? A Deep Clinical Guide to Criticism, Contempt, Defensiveness, and Stonewalling — Empathi