Ask someone what makes a relationship work and you’ll typically hear some version of the same answer: find the right person. Someone kind, emotionally intelligent, financially stable, with compatible values and a good sense of humor. The implicit assumption is that relationship quality is largely a function of who your partner is – their traits, their personality, their baggage.

A landmark 2020 study by Samantha Joel, Paul Eastwick, and 84 collaborators1 puts a large dent in that assumption.

The researchers applied machine learning – specifically, a technique called Random Forests – to 43 longitudinal datasets spanning 11,196 couples from 29 laboratories worldwide. The goal was straightforward but ambitious: figure out which variables actually predict relationship quality, and by how much.

What predicts relationship quality?

The study drew a sharp distinction between two categories of predictors: individual differences (traits belonging to one person – their personality, attachment style, depression levels) and relationship-specific variables (things that exist only between two people – how much appreciation they feel, how often they fight, how satisfied they are sexually).

The relationship-specific variables won decisively. They were two to three times more predictive of relationship quality than individual differences, accounting for up to 45% of variance in satisfaction at the start of a study. Individual differences, by contrast, accounted for only around 14%. And as studies progressed over time, relationship-specific variables continued to hold their predictive power better than individual ones.

The top five relationship-specific predictors, in order, were:

  1. Perceived partner commitment – feeling that your partner is invested and “in it”
  2. Appreciation – feeling valued and grateful within the relationship
  3. Sexual satisfaction
  4. Perceived partner satisfaction – sensing that your partner is also happy
  5. Conflict – the only negatively-valenced predictor in the top five; higher conflict is associated with lower satisfaction

These are, notably, levers you can pull. They describe the texture of a relationship as it is lived day-to-day, not fixed properties of either person.

It’s worth dwelling on conflict for a moment, because it is the most commonly misunderstood predictor on this list. The relevant variable isn’t whether conflict occurs – all relationships involve disagreement – but how much of the relationship’s emotional landscape it occupies. John Gottman’s decades of research2 arrive at a similar conclusion from a different direction: couples who stay together and couples who divorce have roughly comparable amounts of conflict. What separates them is the style of engagement. Gottman identified four communication patterns – criticism, contempt, defensiveness, and stonewalling – that are so reliably toxic that he calls them the Four Horsemen. Contempt in particular, which involves treating a partner as beneath you (eye-rolls, mockery, dismissiveness), is the single strongest predictor of relationship dissolution he has found. Joel and Eastwick’s finding that conflict ranks among the top five satisfaction predictors adds large-scale empirical weight to what Gottman established in the lab: the presence and character of conflict is not incidental to relationship quality – it is central to it.

The uncomfortable finding for matchmakers

One of the study’s more striking results concerns what doesn’t predict relationship quality: your partner’s individual characteristics. One partner’s self-reported traits – their personality, their life satisfaction, their attachment style – explained only around 5% of the variance in the other partner’s relationship satisfaction. As Joel put it: “relationship satisfaction is not well-explained by your partner’s own self-reported characteristics.”

This has uncomfortable implications for anyone who approaches dating as a checklist exercise. The qualities that make someone look good on paper – agreeableness, conscientiousness, emotional stability – are largely orthogonal to whether you’ll be happy with them. Eastwick summarized the finding cleanly: "‘Who I am’ doesn’t really matter once I know ‘who I am when I am with you.’"

What to do with this

For those still dating, the implication is to shift attention away from trait auditing and toward relational experience. How do you feel when you’re with this person? Do you feel appreciated? Do you sense that they’re committed? Is the dynamic marked by warmth or friction? These felt experiences are more predictive than any résumé of personal qualities.

For those already in a relationship, the study offers a useful reframe. The biggest predictors of satisfaction – perceived commitment, appreciation, sexual satisfaction, conflict – are things that can be actively cultivated. Rather than wishing for a more agreeable or emotionally stable partner, a more productive question is: which of these levers can we move? Expressing commitment more explicitly, building a regular habit of appreciation and gratitude, and addressing conflict constructively3 are all within reach.

Relationship quality, it turns out, is less about finding the right person and more about becoming the right partner.

  1. Joel, S., Eastwick, P. W., Allison, C. J., Arriaga, X. B., Baker, Z. G., et al. (2020). Machine learning uncovers the most robust self-report predictors of relationship quality across 43 longitudinal couples studies. Proceedings of the National Academy of Sciences, 117(32), 19061–19071. PNAS ↩︎

  2. Gottman, J. M., & Levenson, R. W. (1992). Marital processes predictive of later dissolution: Behavior, physiology, and health. Journal of Personality and Social Psychology, 63(2), 221–233. For an accessible overview of the Four Horsemen, see The Gottman Institute. ↩︎

  3. See How to Apologize and Communication Frameworks for tools on navigating conflict productively. ↩︎