The idea that a calm relationship must be missing something has a long life in popular culture, and the research on adult attachment does not support it. What that research actually measures, though, is narrower and stranger than the version most people have met.

Start with the numbers. A 2019 meta-analysis in Personality and Individual Differences by Octav-Sorin Candel and Maria Nicoleta Turliuc pooled 132 papers containing 148 independent samples and 71,011 people. Attachment anxiety correlated with relationship satisfaction at -0.32. Attachment avoidance correlated at -0.44. The authors describe both as medium effects.

Both point the same way. Neither establishes cause, and the paper is a synthesis of correlational self-report studies rather than a finding about what produces what. The full text is paywalled and we worked from the published abstract and results summary.

What the measures actually measure

The popular version of this research sorts people into four boxes. The research itself does not.

In 2015, Chris Fraley and colleagues published a taxometric analysis in the Journal of Personality and Social Psychology using two online samples of 2,399 and 2,300 people, analyzed separately. Taxometric methods exist to answer one question. Is a trait made of natural categories, or is it a continuum that has been chopped into categories for convenience?

Their conclusion was that individual differences in adult attachment are continuously distributed, both at the level of general orientation and within specific relationships. They recommended that researchers use dimensional models rather than categorical ones.

So there are two dimensions here rather than four types. One is anxiety, roughly how much a person worries about a partner’s availability. The other is avoidance, roughly how uncomfortable they are with closeness and dependence. Everyone sits somewhere on both, and the scores shade into each other rather than clustering into groups.

That matters for reading the original claim honestly. In this measurement framework, “secure” does not identify a separate natural category. It describes people who score lower on both dimensions than others do.

The dimension that mattered more was not anxiety

The claim names anxiety, jealousy and uncertainty. The meta-analysis found the stronger association somewhere else.

Avoidance correlated with satisfaction at -0.44, with a confidence interval running from -0.47 to -0.41. Anxiety correlated at -0.32, interval -0.35 to -0.29. The point estimates differ and their intervals do not overlap, but non-overlap alone is not a formal test of the difference between two pooled, partly dependent estimates.

Discomfort with closeness tracked dissatisfaction more closely than worry about closeness did.

This is a useful correction to the premise rather than a refutation of it. A relationship without anxiety is not automatically the good case. Two people who are not especially invested in each other would score low on anxiety and high on avoidance, and avoidance is the dimension more strongly associated with dissatisfaction. Calm and distant are different states, and only one of them looks good in the data.

The same analysis separated actor effects, meaning how a person’s own scores relate to their own satisfaction, from partner effects, meaning how their scores relate to their partner’s. Actor effects were consistently larger: -0.24 against -0.12 for anxiety, and -0.29 against -0.14 for avoidance. On these measures, your own scores say more about your own satisfaction than about your partner’s.

Where the idea that uncertainty is exciting comes from

There is a real finding behind the folk belief. It is smaller and narrower than its reputation.

In 2011, Erin Whitchurch, Timothy Wilson and Daniel Gilbert published an experiment in Psychological Science with 47 female undergraduates at the University of Virginia. Participants viewed Facebook profiles of four men, having been told either that those men had rated them most highly, or rated them average, or that it was a fifty-fifty mix.

The uncertain group reported the highest attraction, with a mean of 0.57 against 0.12 for the liked-best group and -0.62 for the average group.

This is one study, not settled consensus. Forty-seven participants, all women, rating strangers from profile pages. The authors note themselves that it is unclear whether the effect survives actually meeting the person, and that they cannot speak to men at all.

The two results are not in conflict, because they describe different moments. Uncertainty appeared to sharpen attention toward a stranger. Anxiety inside an ongoing relationship, averaged across 71,011 people whose relationships had lasted 7.5 years on average, went with lower satisfaction.

A 2014 meta-analysis by Benjamin Hadden and colleagues in Personality and Social Psychology Review points the same way. It reports that the negative associations between both insecure dimensions and relationship quality were more negative among longer relationship durations in cross-sectional samples. Whatever charge uncertainty carries at the start, the pattern did not soften with time.

Relationship-specific reports predicted more than either dimension

A large predictor-comparison study found attachment among the leading individual-difference variables, but relationship-specific reports predicted substantially more variance.

Samantha Joel and a large team published a study in the Proceedings of the National Academy of Sciences in 2020 applying machine learning to 11,196 couples across 43 longitudinal datasets from 29 laboratories, drawing on 2,413 self-report measures over an average of four time points.

They divided predictors into what a person brings to a relationship and what they perceive inside it. The second category won clearly. Relationship-specific variables explained 45 percent of the variance in satisfaction at baseline. Individual differences explained 21 percent.

The strongest relationship-specific predictors were perceived partner commitment, appreciation and sexual satisfaction. Among individual differences, the top three were life satisfaction, negative affect and depression. Attachment anxiety and avoidance came fourth and fifth.

The same paper carries a harder result. Change in relationship quality over the following months was largely unpredictable from anything measured at the start.

What none of this settles

Every attachment association here is correlational and self-reported, so the causal direction is genuinely open. Someone in a reliable relationship may report less anxiety because the relationship is reliable, rather than the relationship being reliable because they arrived calm. The measures cannot separate those.

The scope is limited too. The Joel study drew entirely on Western samples, and 99.4 percent of its variables were explicit self-report rather than observed behavior. The Whitchurch experiment used 47 American undergraduates. The Candel meta-analysis had a mean relationship duration of 7.5 years, which describes established couples better than new ones.

What survives is narrower than the sentence that prompted it. Lower scores on attachment anxiety and avoidance went with higher satisfaction; the studies did not directly measure whether a relationship felt calm. The associations are medium rather than overwhelming, the arrow could point either way, and avoidance had the larger pooled correlation.

The open question is the one the Joel team could not answer. Nothing they measured at the start told them which relationships would get better.