Adult attachment patterns

Why do people so often reproduce the same relationship pattern with very different partners?

Someone describes three relationships. The partners had nothing in common — different temperaments, different jobs, one a decade older. And yet the shape was identical each time: the same slow withdrawal, or the same escalating demand for reassurance, arriving at roughly the same point and ending the same way.

The obvious inference is that the constant must be the person, since the partners varied. But that inference is doing a lot of work. Something recurred; the question is what, exactly, is carried forward — and whether it lives in one head at all.

Take the folk answer first: you have an attachment style, and you bring it with you. That is not an explanation. It names the regularity and asserts it is a trait; to become a mechanism it must say what the trait consists of and how it produces the pattern with a person it has never met.

The most defensible version says the carried thing is a set of expectations — Bowlby's internal working model, an accumulated estimate of how available and responsive people are when you need them. Expectations do not sit passively. Most of what a partner does is ambiguous: a short reply, a delayed call, an evening elsewhere are each consistent with several readings, and a prior belief is what decides among them. That much is well supported — people high on attachment anxiety do interpret ambiguous partner behaviour more negatively, and remember it that way.

Now push on it. Interpretation only colours events; it cannot make the same events happen. Yet the reports are of the same sequence, not merely the same feelings. So something must be closing the gap between expectation and world.

Here is the candidate. An expectation produces behaviour — seeking reassurance, testing, pre-emptive distancing — and that behaviour lands on a partner who responds to it. Persistent demands for reassurance make withdrawal more likely in almost anyone; pre-emptive distance invites the pursuit that then feels like intrusion. The partner's contribution is not fixed either: it is partly manufactured by the very expectation it goes on to confirm. On this account the pattern belongs to neither person but to a loop that either could in principle break, and neither is well placed to notice.

A third candidate is the one most often missed: selection. The pool is not random. We meet people through a stable set of places and are drawn to those who feel legible, and what feels legible is what is familiar. Pick partners whose responses fit the old shape and the loop is half-built before anything happens. This is distinct from the assortment on background and education that the similar-partners piece in this collection is about: that concerns who is demographically available, this concerns which of them feel like a fit.

And a fourth, which honesty requires: some of the recurrence may not need explaining. Stable circumstances produce stable relationships. A general disposition to experience negative emotion would produce similar-looking difficulties across very different partners without any attachment-specific machinery, and measures of attachment anxiety overlap substantially with ordinary measures of that disposition. The recollection is a reconstruction too — three relationships recalled from the far side of their endings, by someone with a theory about themselves, will be smoothed toward each other.

The four candidates — expectation, evocation, selection, and stable third causes — are not rivals to be settled in favour of one. They compose, and they attach at different points, which is why untangling them experimentally is hard and why the honest summary is a mixture with unknown weights.

But the mixture makes one prediction that the folk version does not, and it turns out to be the useful test. If a person carries a single global style, their pattern should be much the same across relationships at a given time. If instead most of the work is done by a loop built with a particular person, patterns should be substantially relationship-specific. The evidence favours the second: people report different patterns with a partner, a parent, and a close friend, and relationship-specific measures predict what happens in that relationship better than a global score does. This is the same lesson the attachment piece in this collection draws for infants, where the Strange Situation classifies a relationship rather than a child; the fixed point here is that the finding survives into adulthood and is what makes "my style" the wrong unit.

Two further cautions. First, the adult literature is weaker than its popularity suggests. Self-report romantic-attachment questionnaires and interview-based measures of an adult's account of their own childhood correlate only weakly with each other, which is serious, since they supposedly measure the same construct; and continuity from infant classification to adult measures is modest at best. I quote no correlation for any of this deliberately — published estimates vary widely with sample and instrument, and one number would misrepresent the field.

Second, the four-box taxonomy — secure, anxious, avoidant, disorganised — is a convenience laid over what the measures actually show: continuous variation on roughly two dimensions, anxiety and avoidance. Nobody is a box, and treating the label as a diagnosis converts a modest tendency into an identity, which is exactly the sort of thing that then gets defended.

• What "earned security" means, and what evidence there is that the loop can be broken from one side.
• Why anxiety and avoidance behave as two continuous dimensions rather than four types, and what the four-box habit costs.

Every term the collection defines is gathered in the glossary.