Peer-reviewed research on relationships keeps finding the same handful of predictors, and none of them is a personality type. Attachment security, how a couple handles conflict, whether their values line up, and how much they genuinely respect each other outperform every popular compatibility framework — not because the frameworks are worthless, but because they were built to describe people, not to forecast relationships.
What "predicts relationship success" actually means
Two different questions get bundled together under that phrase, and separating them changes what counts as evidence.
The first question is who will two people like. That's an attraction question, and it's answered well by short-term studies: speed dating, first impressions, early dating surveys.
The second question is who will still be satisfied together in ten years. That needs longitudinal research — tracking the same couples repeatedly over time — and it's a much smaller, harder body of evidence to build. Most of the popular compatibility content online quietly answers the first question while claiming to answer the second.
The predictors with the strongest evidence behind them
Across the longitudinal literature, four things keep showing up as genuinely predictive, in roughly this order of weight.
| Predictor | What it looks like day to day |
|---|---|
| Attachment security | Both partners can tolerate distance and closeness without panic or shutdown — the foundation everything else is built on. |
| Conflict repair | Not whether a couple fights, but how fast and how completely they recover afterward. |
| Shared values | Alignment on the big, slow-moving things — money philosophy, family plans, how much risk feels acceptable. |
| Mutual regard | Genuinely liking your partner as a person, independent of romantic feeling — the trait most correlated with lasting satisfaction in older-couple studies. |
None of these show up on a sixteen-type chart. They're behavioral and relational, which means they can be observed in how a couple actually treats each other — and, more usefully, they can be built rather than just discovered.
Relationship success in the research is mostly a story about behavior, not about which two types walked into the room.
Personality's real, smaller role
Personality isn't irrelevant — it's just a weaker predictor than people expect, and it works through a narrower channel than "matching types."
Of the Big Five traits, low neuroticism (emotional stability) has the clearest, most repeated link to relationship satisfaction. Reactive, easily-destabilized partners report more conflict and more dissatisfaction, and pairing two reactive people compounds the effect.
Conscientiousness and agreeableness follow behind it — showing up, following through, and generally cooperating rather than competing. Openness and extraversion barely register as predictors at all; they shape what a relationship feels like far more than whether it survives.
What personality doesn't do, in the research, is function the way a compatibility chart implies — as a code that tells you in advance whether two specific people will work. Trait scores predict tendencies, not outcomes for a given pair.
There's also a pairing question worth separating from an individual-trait question: does it matter whether two partners' traits are similar or different from each other? For most Big Five traits the answer is a modest yes for similarity, but the effect is dwarfed by each partner's absolute level of neuroticism and conscientiousness. A calm, reliable person with a less calm, less reliable partner still predicts better outcomes than two people who are equally reactive, even though the pairing is "mismatched."
Why the popular frameworks fall short of research standards
MBTI, Enneagram, and love languages are useful for self-reflection and for building vocabulary about how you differ from your partner. They were never built, however, to predict relationship outcomes, and that shows up in a few specific ways.
- Weak test-retest reliability. A meaningful fraction of people who retake MBTI within weeks score a different type — a poor foundation for anything meant to forecast years ahead.
- No dose-response relationship with outcomes. Studies looking for a link between MBTI type pairing and relationship satisfaction consistently fail to find one worth reporting.
- Untested pairing rules. The compatibility charts circulating for these frameworks were built by community consensus and theorists, not from outcome data — nobody followed a thousand couples and found that specific pairings lasted longer.
This doesn't make them useless. It makes them descriptive tools that got repurposed as predictive ones, which is a different claim than the one the evidence supports. The deeper breakdown of why MBTI fails at compatibility prediction covers the mechanics of that gap.
What does hold up under scrutiny
Two things in the popular relationship-science space have earned their reputation rather than borrowed it.
The Big Five was built and refined specifically to be reliable and to predict real-world outcomes, and it delivers on relationships about as well as it delivers anywhere else — modestly, but replicably. It measures traits, not preferences, which is the distinction that matters for prediction.
Gottman's conflict-observation research is the other. Watching how couples actually argue, rather than asking them to self-report, produced findings that have replicated across many follow-up studies: contempt, criticism, defensiveness, and stonewalling are reliably corrosive, and repair speed reliably protects against them.
Both approaches share something the popular frameworks don't: they were tested against real outcomes before they were published, not popularized first and validated never.
Why researchers trust behavior over self-report
Ask a couple how they handle conflict and most describe themselves flatteringly. Ask them to actually have a disagreement in a lab while researchers code every facial expression, and a very different picture usually shows up.
This is the core method behind Gottman's work, and it's a large part of why it holds up better than survey-based relationship research. Couples were recorded arguing about a real, current disagreement, and their behavior — not their self-description — was coded and tracked forward in time.
That method surfaced something self-report studies had mostly missed: it isn't the topic of a fight that predicts trouble, it's the specific moves made during it. Contempt in particular — mockery, eye-rolling, sarcasm aimed at the person rather than the problem — turned out to be far more corrosive than the volume or frequency of arguing.
The gap between self-report and observed behavior is worth remembering any time a relationship claim is built entirely on questionnaires. People are decent judges of how satisfied they feel and poor judges of what they actually do when they're upset.
The caveats researchers themselves flag
The predictors above are well replicated, but the field is honest about their limits in ways popular summaries usually aren't.
- Correlation, mostly. Secure attachment correlates with satisfaction, but satisfied relationships also make people feel more secure over time — the arrow plausibly runs both directions.
- Sample skew. A large share of the foundational studies were run on couples from Western, educated, higher-income populations who were willing to be observed arguing on camera. That's not everyone.
- Survivorship in longitudinal samples. Couples who divorce early sometimes drop out of a study before the follow-up wave, which can quietly flatter whatever pattern the remaining couples share.
- Effect sizes are moderate, not deterministic. Every predictor here shifts the odds. None of them guarantees an outcome for a specific two people, no matter how confidently a headline states it.
None of this erases the findings — it's the reason they're trustworthy in the first place. A field that reports its own limitations is doing something a compatibility-chart Pinterest post never has to do.
Reading a compatibility claim before you believe it
Most claims about what makes relationships work never specify their evidence. A short checklist separates the load-bearing ones from the recycled ones.
- Is it longitudinal or cross-sectional? A single survey of currently-happy couples tells you what happy couples currently look like, not what causes happiness.
- Is it behavioral or self-report? People are unreliable narrators of their own conflict style. Observed behavior (like Gottman's lab work) is harder evidence than a questionnaire about how you think you argue.
- Does it report an effect size, or just a direction? "Correlates with satisfaction" can mean anything from trivial to substantial — the size is the part that tells you whether to act on it.
- Was the pairing rule tested, or just theorized? Most type-compatibility rules were never checked against real couples' outcomes at all.
Applying that checklist to most viral relationship content thins it out considerably. It also explains why the same four or five predictors keep reappearing across otherwise very different studies, in different decades, run by different research groups — they're simply the ones that survive the checklist.
Sorting frameworks by what they were actually built for
Laid out side by side, the gap between description and prediction becomes easier to see.
| Framework | What it measures | Outcome evidence |
|---|---|---|
| MBTI | Cognitive preference (how you like to think) | No reliable link between type pairing and relationship outcomes |
| Enneagram | Core motivation and fear | Largely untested against longitudinal outcomes |
| Love Languages | Preferred way to receive affection | Weak and mixed; useful as a conversation starter, not a predictor |
| Big Five | Behavioral trait tendencies | Replicated links to satisfaction and stability, moderate effect sizes |
| Attachment style | Security-seeking behavior under stress | Strong, repeated replication across decades of research |
| Gottman conflict coding | Observed behavior during real disagreements | Strong, replicated in independent follow-up studies |
The pattern in that table isn't about which framework is more fun or more insightful for self-understanding — several of the weaker-evidence ones are genuinely great for that. It's specifically about which ones have been checked against what actually happens to real couples over real years, and which haven't.
What this means for how you evaluate your own relationship
If you're trying to gauge a relationship's prospects, type-matching is the wrong tool for the job, even though it's the most fun one. The instruments worth trusting measure what a person actually does under stress and what they actually believe about the shape of a shared life.
The Big Five assessment gets at the trait side of that — emotional stability, conscientiousness, agreeableness — in a form that's been checked against outcomes rather than assembled from forum consensus. For the type-based comparison and where exactly it breaks down against trait measures, MBTI versus the Big Five puts the two frameworks side by side.
None of this makes the fun frameworks worthless for what they're actually good at: giving two people shared language for their differences. It just means the question "will this work" and the question "what type are we" are answered by two different bodies of evidence, and only one of them has been tested against real couples over real years.
If there's one habit worth taking from this literature into your own relationship, it's the shift Gottman's lab made decades ago: stop asking what type you are, and start noticing what you actually do the next time something small goes wrong between you. That single observation carries more predictive weight than any chart.