Adult attachment styles: how clinicians actually measure them, and why it isn't a social-media quiz
What "attachment style" actually means
When you type "attachment style test" into a search engine, you land in a thick layer of online quizzes, Instagram cards, and YouTube explainers. Under all of that sits a serious research tradition — it just gets obscured by the pop wrapping. Attachment theory emerged in the 1950s–60s in the work of John Bowlby and Mary Ainsworth on the infant–caregiver bond. In 1987, Cindy Hazan and Phillip Shaver proposed extending the model to adult romantic relationships — and that branch has been a research field of its own ever since.
What an adult attachment-style test actually measures: stable patterns of feeling and behavior in close relationships. Not "character", not "temperament", not "personality at large" — a narrow behavioral domain: what I feel and how I react when closeness becomes tense, when a partner becomes unavailable, or when intimacy demands more than I'm prepared to give. The same person can show different patterns in different relationships. A test measures the "modal" pattern — the one most often and most consistently expressed.
Attachment style is not a personality trait and not a diagnosis. It's a pattern of behavior in close relationships — describable, measurable, and gradually changeable when needed. And it is not Reactive Attachment Disorder (RAD in DSM-5) — RAD refers to severe social deprivation in childhood and is unrelated to the adult typology.
Why four, not three: the Bartholomew model
Hazan and Shaver in 1987 proposed a simple three-category classification: secure, anxious-ambivalent, avoidant. It worked as a starting point — and is still widely used in large epidemiological studies. Mickelson, Kessler, and Shaver (1997) applied it to a US nationally representative sample of 8,098 adults: 59% classified as secure, 25% avoidant, 11% anxious. That's the statistic you see in the hero block above — and the one most overview publications rely on.
But in clinical work, that scheme turned out to be coarse. Inside the "avoidant" category lurked two quite different patterns. So in 1991, Kim Bartholomew and Leonard Horowitz proposed a matrix: model of self (positive/negative) × model of others (positive/negative). The result was four quadrants. It has been the clinical standard ever since. Modern instruments — ECR-R among them — measure not a "type" but two independent dimensions: attachment anxiety and attachment avoidance. The four Bartholomew quadrants are derived from those two numbers — not measured directly.
- Secure: low anxiety + low avoidance. Closeness feels comfortable; distance isn't catastrophic
- Preoccupied (anxious): high anxiety + low avoidance. Strong desire for closeness paired with a persistent fear of losing it
- Dismissive (avoidant): low anxiety + high avoidance. Self-reliant style; emotional distance is preferred over closeness
- Fearful-avoidant: high anxiety + high avoidance. Conflicting pulls — wanting closeness while fearing both rejection and the vulnerability that intimacy creates
"Four prototypic attachment patterns are defined using combinations of a person's self-image (positive or negative) and image of others (positive or negative)."— Bartholomew K. & Horowitz L.M., Journal of Personality and Social Psychology, 1991, 61(2):226–244
How a clinical instrument differs from a social-media quiz
A social-media quiz typically gives you one of 3–4 "types" based on 8–12 questions and a final card with a description. It appeals to recognition: "yes, this is me". That recognition is pleasant, and it has some value: people get curious and read more about the topic. But recognition is not the same as psychometric validity. Astrological descriptions get recognized too, and any attachment-type description tends to fit a person with a different style about 30–40%.
A clinical attachment test works differently. A larger item pool (ECR-R has 36, AAS has 18) — each item passed factor analysis and demonstrably measures the subscale it's assigned to. Some items are reverse-worded — that balance reduces acquiescence bias, the tendency to agree with whatever's asked. The result isn't a "type" but a pair of numbers: a mean anxiety score and a mean avoidance score, each on a 1–7 scale. Those two numbers locate you in a two-dimensional space — and it's that location that maps into one of the four Bartholomew quadrants.
A social-media quiz tells you: "you are type X". A clinical instrument tells you: "your anxiety mean is 4.2 on the 1–7 scale, your avoidance mean is 3.1, which falls into the secure–preoccupied border quadrant". That's a different level of information, and the decisions you can act on with each are different. A quiz is an introduction to the topic. An instrument is an instrument.
ECR-R, AAS, RSQ: which to choose and how to read the result
Three instruments, each with its niche. ECR-R (Experiences in Close Relationships – Revised; Fraley, Waller & Brennan, 2000) is the gold standard. 36 items, two 18-item subscales: attachment anxiety and attachment avoidance. A 7-point response scale. Internal consistency is reliably high — Cronbach's α ≈ 0.90 for anxiety and ≈ 0.93 for avoidance across validation studies. ECR-R was built on item response theory analysis of 1,085 respondents — which gives it better resolution at the secure end of each dimension than the original ECR.
AAS (Adult Attachment Scale; Collins & Read, 1990) is shorter. 18 items, three subscales (Close, Depend, Anxiety). Used when a brief assessment is enough or when a research protocol needs compatibility with 1990s archives. RSQ (Relationship Scales Questionnaire; Griffin & Bartholomew, 1994) has 30 items, and is unusual in that it produces one of the four Bartholomew categories directly. RSQ is useful when the task is categorical classification rather than profile construction.
Reading the ECR-R result. You get two numbers from 1 to 7 — an anxiety mean and an avoidance mean. A conventional cutoff at mid-scale (3.5) splits each axis into "low" and "high". The point in the 2D plane falls into one of four quadrants: low–low = secure, high–low = preoccupied, low–high = dismissive, high–high = fearful-avoidant. That's the "type" a quiz hands you up front — but a clinical instrument hands it to you as an interpretation of a measurement, not as a label.
Stability of attachment style: for how long is the answer valid
A common expectation: attachment style forms in childhood and stays for life as part of "character". Empirically, it's not quite that. Over short intervals (weeks and months), test-retest stability of ECR-R is high, with strong correlations between "today" and "six weeks later". Over long intervals — years and decades — a noticeable share of people do change classification. Fraley's 2002 meta-analysis of 27 studies found moderate stability from infancy to adulthood — meaning attachment is neither a fixed trait nor random noise.
Davila, Burge, and Hammen's 1997 longitudinal study added important nuance. They followed 155 women for 24 months and found: the tendency to fluctuate in attachment style isn't a reaction to current circumstances — it's a separate characteristic, linked to early adversity. So an "unstable" style doesn't mean an "unstable person" — it's a distinct vulnerability profile. The practical implication: one ECR-R test gives you a reference point for now. After 1–2 years and in new relationships, retesting makes sense — especially after a major life shift or a course of therapy.
"Some women may be prone to attachment fluctuations because of adverse earlier experiences, and women who show attachment fluctuations are similar to women with stably insecure attachments."— Davila J., Burge D. & Hammen C., Journal of Personality and Social Psychology, 1997, 73(4):826–838
What to do with the result: not a diagnosis, a baseline
The main thing the ECR-R result doesn't say: it's not a diagnosis and not a verdict. Sub-secure styles — any style other than secure — are not disorders. By Mickelson's 1997 data, about 41% of US adults fall into non-secure categories, and they work, befriend, partner, parent. Attachment style is how a person experiences closeness, not whether they can experience it. And, to repeat from section one: it's not Reactive Attachment Disorder from DSM — that's a different phenomenon entirely.
What the result does give. A baseline for self-observation and, if you want, for therapy. If ECR-R shows high attachment anxiety — that's useful to know in the moment when a partner is an hour late and your mind is running "they're leaving me". If high avoidance — useful to know when you automatically withdraw after conflict instead of staying in the conversation. Connecting profile to specific situations is the actual clinical work. In that work, the test is a map, not a diagnosis. The same logic we described in the clinical-intuition piece: a validated instrument doesn't replace judgment — it gives judgment something to rest on.
One technical note on Russian-language validation. A full clinical Russian-language validation of ECR-R has not yet been published — there's an adaptation by Kashirsky and Sabelnikova on student samples (≈240) and a 12-item short form by Chistopolskaya and colleagues (2018). That doesn't mean the instrument is unusable in Russian-speaking contexts — for individual work and self-observation it remains the best available. But if you're planning research publication on a Russian-speaking sample, the gap is worth knowing. If you'd like to see your own ECR-R profile with the 2D visualization and four-style description — the instrument is available in Soveria, for free. It's not a quiz. It's a 36-item validated questionnaire with checked Russian and English versions. A reasonable first meeting with the topic if you want to move from a social-media card to a clinical map.