Last time we looked at the channel — why the pool of people you meet, the information you get per encounter, and whether anyone knows you both decides who you even get to date. This piece takes up the other half. Once you're actually in something, why does it so often turn into the same conversation, the same fight, the same ending — with a person who looks nothing like the last one?
The popular answer is "you keep picking the same type." That answer contains two separate claims, and they have very different amounts of evidence behind them:
- Selection: you are drawn to a particular kind of person.
- Process: your relationships unfold the same way regardless of who it is.
The second claim is well supported. The first is much shakier than the internet suggests. Most of what "same type" means in your life probably lives in the second one.
Two dials, not four categories
Attachment theory started with Bowlby and Ainsworth studying infants and caregivers, and Hazan and Shaver's 1987 paper extended it to adult romantic love. You've almost certainly met the four-box version — secure, anxious, avoidant, fearful-avoidant. The version researchers actually use came out of factor-analysing all the self-report questionnaires: instead of categories, there are two roughly independent continuous scores.
Attachment anxiety is how strongly you worry about whether a partner is really available to you — whether they'll leave, whether they still want this. High scorers are vigilant: they notice small shifts in tone, latency, and warmth, and their underlying strategy is to turn the signal up. Ask for reassurance. Bring up the problem now. Get closer until it feels settled. Researchers call this hyperactivation.
Attachment avoidance is discomfort with depending on someone. High scorers prize self-sufficiency and read requests for closeness as pressure. Their strategy is to turn the signal down: stay busy, keep an exit route, deal with distress alone, avoid the heavy conversation until it can't be avoided. This is deactivation.
Low on both dials is what "secure" means. High on both is the fearful-avoidant combination. Crucially, these are two different answers to the same question — can I count on this person? One is "find out fast, loudly, repeatedly"; the other is "don't find out, don't need to."
Why the two get stuck together
Because each one's solution is the other's trigger.
Picture a delayed reply at 11pm. High anxiety reads the delay as data, and the strategy says: get more information, make contact, check. So they send a follow-up — maybe a light one, maybe an anxious one. High avoidance reads the follow-up not as a bid for connection but as an obligation, and the strategy says: create space. So the replies get shorter. That's a real cooling now, not an imagined one, and it confirms the anxious partner's original fear. Which raises the volume again. Which increases the pressure. Which lengthens the withdrawal.
This is the pattern couples researchers call demand–withdraw, and it's one of the few things in relationship science that shows up in hormones, behaviour, and self-report at the same time. In a study of newlywed couples discussing an unresolved conflict, spouses in anxious-wife/avoidant-husband pairings showed sharp cortisol spikes in anticipation of the conversation. During it, anxious wives had trouble recognising their avoidant husbands' distress, and avoidant husbands had trouble approaching their anxious wives for support 1. Both people were doing exactly what their strategy taught them, and the strategies were incompatible.
Two useful qualifications. First, this exact pairing isn't uniquely doomed — across studies, what's consistent is that any two insecure partners fare worse than pairs where at least one is relatively secure, and that avoidance is the more corrosive of the two dials. A meta-analysis of 73 studies and 21,602 people found both anxiety and avoidance predicted worse relationship quality, with avoidance more strongly linked to lower satisfaction and connectedness, and anxiety more strongly linked to more conflict 7.
Second, this pattern isn't a moral failing of either person. Both strategies were learned where they made sense. They just don't serve the goal they're aimed at.
Is "the same type" real? Partly, and less than you'd expect
Here's where the popular story gets ahead of the data.
On one side: a longitudinal study of 354 dating couples found that attachment styles were non-randomly paired — there were no anxious–anxious pairs and no avoidant–avoidant pairs in the sample, and anxious-woman/avoidant-man pairings were common 2. The "complementary hypothesis" explains why this could happen: an anxious person's model says partners are distant, and an avoidant partner fits it; an avoidant person's model says partners are clingy, and an anxious partner fits that 1.
On the other side, and this is the part people leave out: when researchers asked people directly, they didn't find the magnetic pull. In one set of three studies, participants were most attracted to partners whose attachment style matched their own; anxious participants were dating anxious partners and rated avoidant partners as the least desirable of the options 3. In another study, people rating profiles of hypothetical partners preferred secure partners above all others regardless of their own style 4.
So the honest summary is: the pairs people end up in are somewhat non-random, but people's stated preferences point toward security, and the evidence for a specific "anxious people are magnetically drawn to avoidants" mechanism is mixed. Selection contributes. It is not the main engine.
The engine is the reading, not the person
Here is where the same story gets reproduced with different casts.
The lens. Highly anxious people run a lower threshold for noticing change in a face. In one series of experiments, participants watched morphed videos in which an emotional expression gradually faded to neutral or emerged from neutral, and had to say when it happened. High-anxiety participants reported the expression appearing and disappearing earlier than everyone else — their detection system fires sooner 6. In one of the studies, that extra vigilance actually made them less accurate at identifying what the expression was.
The interpretation. In a study that deliberately created a relationship threat, anxious daters were more accurate than others at inferring what their partner was thinking and feeling. That accuracy came with more distress, less confidence, a measurable drop in felt closeness — and by the four-month follow-up, those relationships were more likely to have ended 5. Knowing more did not help. In another set of studies spanning 333 couples, both anxious and avoidant partners held a consistent pessimistic bias: they guessed their partners were less satisfied and less committed than the partners actually reported being 11. Avoidant people additionally tend to be poorer readers of a partner's thoughts and to read a partner's negative emotion as more negative than it is 11.
Put those together: the same objectively ambiguous event — a quiet evening, a delayed plan, a flat "fine" — routes through the same lens and lands on the same verdict. Not because the partners are the same person, but because the apparatus producing the verdict is the same.
And then you act on the verdict. You probe, or you withdraw. Which produces a real response in the other person, which is exactly the input your lens was calibrated for.
Why it doesn't break up quickly
You'd think a loop this unpleasant would fail fast. It doesn't always. The 354-couple study found that the relationships of avoidant men and anxious women were surprisingly stable over three years — despite both partners rating them poorly at the start 2.
That's the trap's real signature. It doesn't usually present as chaos. It presents as a familiar rhythm: enough closeness to keep going, enough distance to keep hurting, and a pattern that both people can rationalise as "this is just how relationships are."
None of this is a sentence
Attachment is not fate, and the numbers are worth knowing precisely.
Across studies, roughly 60% of people report the same style eight months later, and about 70% after four years 10. Around 30% of people report changing style across various time windows 10. In one community sample of women, 46% changed over two years 10. A large longitudinal study of over 4,000 people found that about half of the life events it tracked produced immediate changes in attachment — but on average people drifted back toward their pre-event trajectory, and only about a quarter of events produced lasting change 8. A meta-analysis of attachment stability across the first 19 years found security moderately stable and better described as continuous than categorical 9.
Read that together and it says: this is closer to a durable habit of expectation and response than a fixed trait. It's stable enough to feel like your personality. It's changeable enough that the outcome isn't determined.
Two boundaries to keep in view. These measurements are self-report, so they capture how you describe your patterns, not a blood test. And the effects are probabilistic — attachment insecurity shades how a relationship goes; it doesn't dictate it. Partner behaviour genuinely matters too: a securely attached partner can buffer an anxious one's reactivity instead of amplifying it, which is its own topic for another day.
For now the practical point is narrow and specific. If the repetition lives less in who you pick and more in what you do the instant you feel a threat, then the thing worth practising isn't a better type-detector. It's the pause between "they did X" and "so this means Y" — long enough to ask whether Y is what the evidence supports, or what your strategy predicted.