Compatibility

Compatibility is the most confident idea in the entire relationship market. Two people have properties; the properties fit or they do not; a good enough instrument should be able to compute the fit and report it as a number.

The largest coordinated attempt ever made to find that number could not find it.

This entry is going to be awkward, because this site sells a couples test. It is written the way the evidence reads anyway, and the conclusion is the one at the end of the next section: on the best available evidence, the match between two people barely predicts how happy either of them will be. What predicts it is what each person already perceives about the relationship they are in.

In 2020, Samantha Joel, Paul Eastwick and eighty-four other researchers did something the field had never managed: they pooled the raw data. Forty-three longitudinal datasets from twenty-nine laboratories, 11,196 couples, every measured variable given an equal chance to compete. Then they ran random forests over it — a method that matters here for one specific reason, which the paper spells out. Random forests detect interactions. A "matching effect" is an interaction: your trait times their trait. If compatibility existed in this data in any robust form, this analysis was built to find it.

What it found, in order:

• Your own view of the relationship predicted up to 45% of your satisfaction. The strongest items were perceived partner commitment, appreciation, sexual satisfaction, perceived partner satisfaction, and conflict. All of them are about this relationship, and all of them are things you can be wrong about.
• Individual differences predicted 21%. Life satisfaction, negative affect, depression, attachment avoidance, attachment anxiety. Real, substantial, and — note — these are properties of one person, not of a pair.
• Your partner's own reports predicted two to four times less than yours did.
• And this is the finding: adding individual differences and partner reports on top of a person's own relationship-specific reports improved the prediction by nothing at all. The authors state the implication directly — effects due to moderation by individual differences and by partner reports are likely to be quite small. Moderation by the partner's characteristics is exactly what compatibility means. It was looked for, with the right method, in the largest sample ever assembled, and it was not there in any useful size.

There was one more result, and it is the one that should end the category. Change in relationship quality — whose relationship got better and whose got worse over the course of each study — was largely unpredictable from any combination of these variables. Not weakly predictable. Largely unpredictable.

So: the fit between two people is not what determines this, and neither of you knowing everything about the other at the start would fix it. That is not a gap in the instruments. It is what a hundred-variable, eleven-thousand-couple, twenty-nine-laboratory effort returned.

Joel and colleagues' result did not come out of nowhere. Three earlier findings pointed the same way and were, at the time, ignored by everyone selling a match percentage.

Matching algorithms have never been shown to work. Finkel and colleagues' 2012 review of online dating in Psychological Science in the Public Interest went through the claims systematically and concluded there was no compelling evidence that any matching algorithm produced better outcomes than chance. The algorithms are proprietary, the validating studies are not independent, and the underlying task — predicting a relationship from two questionnaires taken before the people have met — is the task that later failed again with better methods.

Similarity does less than it appears to. Montoya and colleagues meta-analysed 313 studies. Actual similarity predicted attraction strongly when people had not interacted, less after a brief interaction, and not at all in existing relationships. Perceived similarity predicted attraction at every stage. Which is to say: the feeling of being alike does the work, the fact of being alike mostly does not, and the feeling is partly constructed after the fact by people who like each other.

Even attraction between two specific people is unpredictable. In 2017 Joel, Eastwick and Finkel took a speed-dating dataset and tried to predict, from everything both people had reported beforehand, who would be attracted to whom. They could predict how much a person tended to like others, and how much a person tended to be liked. The specific pairing — this one and that one — could not be predicted at all. And that is the single easiest version of the compatibility question: not who will be happy in four years, just who will want a second date on Thursday.

Because it is a good story, and because two true things keep feeding it.

Similar people do end up together. That is real, and it is mostly not compatibility doing the work — it is who was in the room. People meet through school, jobs, neighbourhoods, apps with filters and friends who resemble them. Similarity in a couple is substantially a fact about the population they were drawn from.

Some things genuinely do not work. The evidence says fit barely predicts happiness within the range of ordinary couples; it does not say nothing matters. Wanting children against not wanting children is not a compatibility score, it is an incompatibility, and no amount of relationship skill dissolves it. The same goes for monogamy versus not, and for where you are going to live. These are a short list of structural questions with binary answers — and a percentage is exactly the wrong instrument for them, because a good score on forty items cannot outvote one of these and a bad score cannot create one.

And there is the reason that has nothing to do with evidence: a compatibility score answers the question people are actually asking, which is "may I stop being uncertain now?" Intolerance of uncertainty explains the pull better than any finding in this entry. A number resolves it for about a day.

Here is the part that survives contact with the evidence, and it follows directly from the Joel result rather than around it.

The strongest predictors were perceived partner commitment and perceived partner satisfaction. Not actual commitment. Not actual satisfaction. What you believe about how invested and how happy they are. That gap — between what your partner thinks and what you think your partner thinks — is a real, measurable, movable thing, and it is sitting on top of the largest effect in the literature.

An instrument that shows each of you what the other actually said is working directly on that gap. That is a different job from scoring a match, and it is the one with evidence behind it.

This is why the Couples Compatibility Test does not produce a percentage. It maps the specific patterns between you — who pursues and who withdraws, how repair happens or fails to, where boundaries sit — and gives each of you the other's answers. We would rather say plainly that the category it sits in is built on something the evidence does not support than print a number we would then have to defend.

What no couples test can do, including ours: tell you whether to stay, predict whether this gets better or worse, or locate a compatibility that thirty years of research has not been able to detect. Treat any result as a conversation you had not had yet, and notice that the conversation, not the result, is the intervention.

Two things follow from all of the above, and they point in opposite directions on purpose.

A low score is not a reason to leave. There is no evidence that a computed mismatch predicts anything about your future, and there is quite a lot of evidence that it cannot. If you are looking for permission, the number is not able to give it and should not be used as though it were.

A high score is not a reason to stay. This is the more dangerous one. A good result on a relationship questionnaire has no bearing on whether you are treated well, and people do use reassuring scores to talk themselves out of what they already know. Nothing in this literature outranks your own account of what happens in your house.

The things that genuinely predict — appreciation said out loud, conflict that gets repaired, believing your partner is in this — are behaviours, on both sides, available this week. That is a less satisfying answer than a percentage and it is the one the data supports. The mechanics are in conflict and repair and in relationship styles.

And where this stops being a matter of fit at all: if there is fear, control, isolation from the people who know you, or you are routinely made to doubt what you remember, no compatibility framing applies and none of this entry is the relevant page. Toxic relationships is. In the US, the National Domestic Violence Hotline is 1-800-799-7233; in the UK, the National Domestic Abuse Helpline is 0808 2000 247.