The four attachment styles, and where the names came from

The four attachment styles are secure, anxious-ambivalent, avoidant and disorganised. They are classifications of how an infant behaves toward a particular caregiver after a brief separation, developed by Mary Ainsworth in the 1970s with a fourth added by Mary Main and Judith Solomon in 1990. They are not personality types, and the adult questionnaires that borrow the names are a separate measurement tradition that agrees with the infant one only weakly.

• Three styles came from Ainsworth’s Strange Situation: secure, avoidant and anxious-ambivalent. Disorganised was added twenty years later.
• The classification describes one relationship, not one child. An infant can be classified securely with the mother and avoidantly with the father.
• The adult version starts with Hazan and Shaver in 1987, who rewrote Ainsworth’s three infant patterns as descriptions of adult romantic love.
• Adult attachment is better represented as two dimensions, anxiety and avoidance, than as a set of boxes. The boxes are cut points on the dimensions.
• Interview-based and questionnaire-based measures of adult attachment barely agree: a meta-analysis found a mean correlation of .09.
• That is why two online quizzes can give you two different styles and both be reporting their instrument honestly.

What attachment theory says, in one paragraph

John Bowlby’s claim, set out from 1969, was that human infants come equipped with a behavioural system whose function is to keep a caregiver close when the infant is frightened, ill or tired. The system is not about feeding or affection: it is about proximity under threat, and it evolved because proximity kept infants alive. Out of repeated experience with a particular caregiver the child builds an expectation — Bowlby called it an internal working model — of whether that person will be available when needed.

Attachment theory is therefore about a relationship and a history, not about a temperament. That distinction is the source of nearly every misunderstanding downstream of it.

The Strange Situation, and the three styles it produced

Mary Ainsworth turned the theory into a measurement. The Strange Situation is a laboratory procedure of eight short episodes, most of them three minutes long, in which a one-year-old is left with a stranger, left alone, and reunited with the caregiver twice.

The scoring is the part that gets forgotten: what matters is not how much the infant cries during separation but what it does at reunion. Ainsworth and Bell described the procedure and the pattern in 1970; the full classification system arrived with the 1978 book, Patterns of Attachment.

Secure

The secure infant treats the caregiver as a base to explore from and a place to return to. Distress at separation is normal and not the criterion; recovery at reunion is.

Avoidant

The avoidant infant appears unbothered. Spangler and Grossmann measured heart rate and cortisol in 1993 and found these infants physiologically aroused while behaviourally indifferent, which is why the pattern is read as a strategy for keeping a caregiver who does not welcome overt need — not as independence.

Ambivalent

The ambivalent or resistant infant is preoccupied with the caregiver at the cost of exploring, and at reunion mixes contact-seeking with anger. Comfort arrives and does not land.

The fourth style, added later

The three categories left a residue: infants whose behaviour did not fit any of them, and who did contradictory or odd things at reunion — approaching backwards, freezing mid-movement, appearing dazed.

Mary Main and Judith Solomon described these in 1990 and proposed a fourth classification, disorganised or disoriented, coded D. It is not a fourth strategy alongside the other three. It is the breakdown of a strategy — what shows when the person who is supposed to be the solution to fear is also its source.

Two things follow. Disorganised classifications are recorded together with a best-fitting organised category, so a D infant is also A, B or C. And because it is associated with frightening or frightened caregiver behaviour, this is the classification most often misused online, where it is presented as a personality description rather than as a research code.

The infant classification is not an adult personality type

This is the category error the whole subject rests on, and almost nothing on page one of a search result explains it.

An infant classification describes a relationship. Main and Weston showed in 1981 that the same child can be classified differently with mother and with father, which is fatal to reading the label as a property of the child.

An infant classification also describes behaviour in a twenty-minute laboratory procedure at around twelve months. Whether it predicts anything decades later is an empirical question, and Fraley’s 2002 meta-analytic work on stability found continuity that is real, modest, and nothing like the deterministic story the term carries in popular use.

And the adult measures do not descend from the infant ones. They were built separately, which is the next section.

Where the adult version actually comes from

Cindy Hazan and Phillip Shaver published the founding paper in 1987. They took Ainsworth’s three infant patterns, rewrote each as a paragraph describing how somebody feels in romantic relationships, printed the three paragraphs in a newspaper, and asked readers to pick the one that fitted. That is the origin of every three-choice attachment quiz since.

Bartholomew and Horowitz replaced the three with four in 1991, derived from two underlying judgements: whether you see yourself as worth caring for, and whether you see others as available. Crossing them gives secure, preoccupied, dismissing and fearful — which is where the modern popular vocabulary comes from, including fearful-avoidant.

Fraley, Waller and Brennan showed in 2000, applying item response theory to the self-report measures, that adult attachment is better described by two continuous dimensions — attachment-related anxiety and attachment-related avoidance — than by categories. The four styles are quadrants of that space. Nobody sits on a quadrant; people sit at a point.

Running alongside all of this is a third tradition: the Adult Attachment Interview, developed by Mary Main and colleagues in the 1980s, which asks adults about their childhood relationships and scores not what they report but how coherently they can narrate it. Its categories — autonomous, dismissing, preoccupied, unresolved — sound like the self-report ones and are arrived at completely differently.

Why the online quizzes disagree with each other

Three reasons, and none of them is that one quiz is broken.

The first is that the categories are cut points. If attachment is two continuous dimensions, someone near a boundary will flip between styles on two instruments, or on the same instrument in a different month, without anything having changed.

The second is that the traditions do not measure the same thing. Roisman and colleagues meta-analysed the overlap in 2007 across a combined sample of 961 and found a mean correlation of .09 between Adult Attachment Interview security and self-reported attachment style — trivial to small, by the conventional benchmarks. Your interview classification and your questionnaire style are close to independent facts about you.

The third is that attachment is relationship-specific. A quiz asking about your relationships in general is averaging over relationships that the theory expects to differ.

What varies, and what does not

Van IJzendoorn and Kroonenberg’s 1988 meta-analysis pooled 32 studies from eight countries and nearly two thousand Strange Situation classifications. Secure was the majority pattern everywhere, at about 65 per cent overall — but the distribution of the insecure categories moved. In north-German samples, avoidant classifications outnumbered secure ones, while Japanese and Israeli samples produced more ambivalent classifications and almost no avoidant ones.

The finding the paper is best known for is the comparison of variation: differences between samples within a country were larger than differences between countries. That is a caution against reading a national character into a distribution, and it is also a caution against treating the procedure as culture-free — the German result has been discussed since it appeared as possibly reflecting what early independence is taken to mean in different places.

Attachment vocabulary sits next to two other clusters this blog covers, and the confusions run in both directions: the names of the therapies include approaches built on attachment ideas, and who is qualified to use any of them is a separate question from which theory they favour. Four style names, three measurement traditions and two dimensions is a lot of near-identical vocabulary to hold, which is exactly the case for learning the terms against each other rather than one at a time. If you want the plainest possible statement of what this blog is and is not qualified to tell you, it is on the about page.