Four boxes from a newspaper questionnaire

Anyone who writes “he is avoidant, I am anxious” today is using a scheme whose first data came from a daily newspaper. In 1987 the Rocky Mountain News in Denver printed a questionnaire billed as a “love quiz”. Cindy Hazan and Phillip Shaver analysed 620 replies that readers sent in, 205 from men and 415 from women. Everyone who took part had decided to do so — a self-selected sample, neither a clinical survey nor a random draw.

That is not a scandal but research history, and it explains a good deal. The study asked a question for which no instrument existed: do the expectations about closeness, availability and rejection that John Bowlby and Mary Ainsworth had described in infants also operate in the romantic relationships of adults? To test it, Hazan and Shaver first had to translate attachment into sentences that adults could agree with. Those are precisely the sentences now circulating in relationship guides.

How three patterns became four boxes

Hazan and Shaver offered three descriptions carried over from infant research: secure, avoidant, anxious-ambivalent. Respondents read all three and ticked the one that fitted best. Two questionnaire studies produced three results. First, the three styles were distributed among adults roughly as they were among infants. Second, the three groups differed in how they experienced romantic love. Third, style was linked to internal working models of the self and of social life — the term Bowlby had coined to explain why a relational pattern persists over time.

Four years later Kim Bartholomew and Leonard Horowitz proposed the scheme that eventually prevailed. They crossed two judgements: the image of the self, positive or negative, and the image of others, positive or negative. Two by two yielded four cells. Secure means positive on both counts; preoccupied means a negative self-image alongside a positive image of others; dismissing means the reverse; and fearful-avoidant means negative on both. With that, the single avoidance split into two different ones: someone who keeps his distance because he has no need of closeness is not the same as someone who keeps his distance because he fears rejection.

The four boxes are therefore not the finding of a survey but the product of a model hung on two axes. Anyone who meets them online as personality types is seeing the boxes without the axes.

Dimensions instead of boxes

Research has followed that road to its end and dissolved the categories. In 1998 Kelly Brennan, Catherine Clark and Phillip Shaver published the measure that became the standard. It models two dimensions: attachment anxiety — the worry of being abandoned or insufficiently loved — and attachment avoidance — discomfort with dependence and closeness. Categories can be derived from them, but the measurement itself stays dimensional.

The road there was a tidying-up operation. The three of them took the English-language self-report instruments then available, analysed them together by factor analysis, and were left with two largely independent dimensions. Out of the sturdiest statements they built a 36-item questionnaire, the Experiences in Close Relationships. It remains the most widely used instrument in the field.

This is more than a technical nicety. On a dimension there is no inside and outside, only more and less, and most people sit somewhere in the middle. A box, by contrast, implies a boundary that does not exist in the data. To say “I am anxious” means, in the language of the instrument: my score on one of two continuous scales falls above an arbitrarily chosen cut.

Three procedures that do not measure the same thing

There are three established approaches to attachment, and they come out of different traditions. The Strange Situation observes how small children behave during separation and reunion. The Adult Attachment Interview rates not what someone says about their childhood but how coherently they say it. Self-report asks adults how they experience themselves in romantic closeness.

How far these procedures converge was worked out by Glenn Roisman and colleagues in 2007. Their meta-analysis of ten available studies, 961 people in total, found an average association of r = .09 between security as coded in the interview and the dimensions of self-report. By the usual conventions that is negligible to small. A second study with 160 participants confirmed the finding with better instruments; a third, on 50 engaged couples, showed that both approaches predict something, but each predicts something different.

It does not follow that one of the procedures is wrong. It follows that the same words used across different procedures must not be treated as one and the same trait. Someone who counts as secure on the questionnaire may not count as secure in the interview — and the other way round. When a guidebook cites findings from infant observation in order to interpret a questionnaire score, it is transferring results between measures that have little to do with each other.

Stable, but not destiny

Working models grow out of experience and can structure new experience. That means neither that a toddler assessment determines later partnerships nor that a style stays the same for life. People can feel differently secure in different relationships; dependable partnerships, therapy and life events change expectations.

R. Chris Fraley summarised the state of the research in 2019. Attachment security is moderately stable across years, and the same person can be differently secure in different relationships. Relationship-specific expectations and a general style are not the same thing — which alone is reason enough why a single score carries no verdict on a life.

Online guides often turn patterns into motives: withdrawal proves fear of intimacy, frequent messages prove fear of abandonment. Yet the same behaviour can mean stress, conflict style, culture, present uncertainty or simply lack of interest. Attachment language helps when it opens questions: what happens in me when closeness becomes uncertain? Do I seek reassurance, do I withdraw, can I state needs clearly and bear the answer? It does harm when it closes off responsibility. “I ghost because I am avoidant” explains a pattern but does not excuse its consequences.

The relationship is part of the evidence

Attachment behaviour arises between people. The same person can appear open with a dependable partner and watchful with an unpredictable one. Stress, illness, differences in power and earlier injuries also change how closeness is regulated. That is why it is analytically too crude to label every request for reassurance “anxious” and every retreat “avoidant”.

Sequences say more: what sets off the uncertainty? How is it signalled? How does the other side respond, and does repair follow? Security shows itself not in the absence of conflict but often in the fact that needs can be spoken, limits respected and ruptures mended.

What remains

Adult attachment is a serviceable map of recurring regulation and a poor diagnosis. It describes what someone expects in relationships and how they react when closeness becomes uncertain. It says nothing about whether those expectations come from childhood, whether they will last, or whether they will reappear in the next relationship. The newspaper origin of the first data is less of a problem than what happened to them afterwards. Three ticked paragraphs became two scales, two scales became four cells, and four cells became a vocabulary with which people explain one another. Every one of these steps had a methodological rationale. None of them made the categories any harder than they were the first time round.

In everyday use the misuse shows in the verb. “I am avoidant” asserts a property; “I withdraw when I feel put under pressure” describes a sequence and leaves room for a different continuation. Nobody is an attachment style. People bring expectations into relationships — and relationships can teach new expectations.