Can Attachment Styles Change? How Much Moves, How Slowly, and What Moves It

The short answer, and the number behind it

Yes, attachment styles change. That is not the interesting part. The interesting part is how much, how fast, and by what — and on all three the honest answers are smaller, slower and less flattering than the advice around this subject suggests.

Start with the single most informative number. Researchers measure this as rank-order stability: whether the people who were relatively secure at one point are still the relatively secure ones later. Modelling the accumulated longitudinal evidence, R. Chris Fraley and Claudia Brumbaugh estimated the test–retest correlation of attachment security across a ten-year interval at roughly .50 in adulthood, against roughly .30 across ten years in childhood — figures Fraley and Nathan Hudson set out in their chapter on how attachment styles develop.

Half of it is the stable part. The other half is not. A correlation of .50 across a decade is the signature of something real that moves. It is not a trait you are stuck with, and it is not a setting you can flip.

Going further back makes the point harder, not easier. Martin Pinquart, Claudia Feußner and Lieselotte Ahnert pooled 127 papers covering 21,072 attachments and 225 measurement intervals, ranging from half a month to 29 years. Overall stability came out at r = .39. Intervals shorter than two years ran higher, intervals longer than five years lower, and across intervals longer than 15 years there was no significant stability at all.

That last clause is worth reading twice, because it is the opposite of what this genre asserts. The largest synthesis of how well early attachment predicts later attachment finds the link thinning to nothing over the longest spans. Our editorial policy explains why every figure on this page carries a linked source. This one earns it, because most of what gets written about attachment rests on the assumption it undercuts.

What is actually being measured when someone says “my attachment style”

Before asking whether a style can change, it helps to know what kind of thing it is, because the popular version and the measured version are not the same object.

The popular version is four boxes. The measured version is two dimensions. Kelly Brennan, Catherine Clark and Phillip Shaver built the standard questionnaire in 1998 around attachment anxiety — worry about whether a partner will stay available — and attachment avoidance — discomfort with closeness and with depending on anyone. Everyone has a position on both.

The boxes have been tested directly and they are not there. R. Chris Fraley, Nathan Hudson, Marie Heffernan and Nathan Segal ran taxometric analyses on two samples, for general attachment representations and for specific relational contexts such as parents and peers. The structure came out dimensional rather than categorical in every case.

That quietly rewrites the question. It is not “can I switch from one type to another,” which has no coherent answer. It is “how far along a continuum can someone move, and over what period” — and that question has numbers attached to it.

It also explains a finding that otherwise looks like a scandal. Mark Baldwin and Beverley Fehr examined the stability of Hazan and Shaver's single-item attachment measure across a set of data sets, some their own and some gathered by other researchers. About 30% of people changed their self-classification over spans running from one week to several months. Instability was highest among those who had classed themselves anxious-ambivalent: the majority of them picked something different on the next occasion.

Their interpretation is the part worth keeping, and it is not that the measure is broken. They argue the variability may reflect something real — that a rating captures the relational schema active at that moment rather than an enduring disposition — and conclude that stability is therefore neither assumed nor expected.

So if you have taken a quiz twice and got two different results, you have observed the normal behaviour of the instrument, not a transformation. Our attachment style test returns the two scores rather than a verdict, and says on its own page what a score like that cannot tell you. What the anxious end of the scale feels like from inside, and how it differs from love, is in the piece on love versus anxious attachment.

You do not have one attachment style. You have several, and they move separately

This is the single most useful thing in the literature that almost never reaches the reader, and it changes the practical answer more than any finding about therapy.

Attachment representations are hierarchical. Nancy Collins and Stephen Read formalised the idea: people hold general models about whether others can be relied on, and also models specific to particular people — a mother, a best friend, a partner. The general and the specific tend to line up, but nothing requires them to. In the Fraley, Hudson, Heffernan and Segal data, general attachment security correlates between about .30 and .50 with security measured in specific relationship domains. Substantial overlap, and a long way from identity.

The cleanest demonstration is a study of parental divorce. R. Chris Fraley and Marie Heffernan examined attachment across two samples and found that parental divorce had selective rather than diffuse implications: it was associated with insecure relationships with parents in adulthood, and more weakly with insecurity toward romantic partners or friends. The association with parental insecurity was strongest when the divorce happened in early childhood.

That is a result the genre cannot easily absorb, because the genre's premise is that a childhood disruption reliably reappears in your dating life. In the one study designed to separate those domains, it mostly stayed in the domain where it happened.

The domains can also move in opposite directions at once. Nathan Hudson, Fraley, William Chopik and Heffernan measured attachment both globally and relationship by relationship across a large age range. With age, adults tended to become more anxious in their relationships with parents and less anxious with romantic partners — the authors read this as two different processes, uncertainty creeping into a relationship you have left home from, and accumulated evidence of availability building up in one that has persisted.

The practical consequence is that “can my attachment style change” is the wrong grain of question. Security with a specific person can be built while the general disposition barely moves, and the specific is what predicts the relationship outcomes anyway. Something similar shows up in the research on separation and reunion covered in what distance does to attraction: the relationship-specific measure and the person-level one come apart.

The two models the data had to choose between, and the one that won

For two decades the field ran an argument with two sides. The revisionist model says working models are continuously rewritten by experience, so what you are now is a product of recent history and the past washes out. The prototype model says a stable underlying factor persists beneath temporary variation, pulling people back toward their own baseline.

It was settled empirically, and with a design good enough to settle it. Fraley, Amanda Vicary, Claudia Brumbaugh and Glenn Roisman ran two intensive longitudinal studies: one sample of 203 adults measured daily for 30 days, another of 388 measured weekly for a year. The patterns of stability matched the prototype model. The Big Five personality traits showed a similar shape of stability over the same intervals, and did not account for the stability observed in attachment.

Joseph Jones, Fraley and colleagues then ran the same test in adolescence — 176 participants, average age 14, followed for five years. The prototype model won again. Stability was lower in adolescence than in adulthood, and exploratory analyses suggested family conflict, parental separation or divorce, minority status and male sex might weaken that prototype-like stability.

Bowlby had a name for the mechanism, borrowed from Conrad Waddington's work on cell development: canalization. A developmental pathway becomes increasingly buffered over time, so an experience that conflicts with a working model nudges a person off course temporarily and the trajectory gradually reasserts itself.

This is the finding to hold on to when a good month makes you think the work is done, and when a bad fortnight makes you think it never started. Both are real and both are mostly the temporary layer. What the prototype model predicts is that the baseline moves slowly and the day-to-day moves a lot, which is exactly what someone trying to change experiences and almost always misreads.

The childhood part is weaker than the genre needs it to be

Almost every article on this subject opens by telling you your adult relationships were set by how you were parented. Two prospective studies, published side by side in the same issue of Child Development and pointing in opposite directions, are a large part of what that claim actually rests on.

Everett Waters and colleagues followed 60 white middle-class infants seen in the Strange Situation at 12 months and recontacted 50 of them 20 years later for the Adult Attachment Interview, with interviewers blind to the infant classifications. 72% received the same secure-versus-insecure classification, κ = .44. Change was not random: 44% (8 of 18) of those whose mothers reported negative life events — loss of a parent, parental divorce, life-threatening illness, parental psychiatric disorder, abuse by a family member — changed classification, against 22% (7 of 32) of those with no such events.

Nancy Weinfield, Alan Sroufe and Byron Egeland ran the equivalent study in a sample originally recruited for poverty and developmental risk: 57 young adults, Strange Situation in infancy, AAI at 19. They found no evidence of significant continuity between infant and adult attachment, with many participants moving to insecurity. The discontinuity was not noise either — the groups were distinguished by child maltreatment, maternal depression and family functioning in early adolescence.

So one study found 72% stability in a low-risk sample and another found none in a high-risk one, and both are telling the truth about their samples. Pooled across everything available, Pinquart's meta-analysis gives the r = .39 from the first section, with security at risk being the hardest thing to maintain.

Fraley's 2019 review in the Annual Review of Psychology states the implication in a sentence the genre never quotes: early experiences do not completely determine whether people become secure or insecure as adults. Many people with supportive caregiving are insecure in adult relationships, and many people with less-than-ideal caregiving are relatively secure.

That is not a licence to dismiss childhood, and the risk-sample result is a reminder of how much damage sustained adversity does. It is a licence to stop treating a childhood as a verdict — which is the same error, run on a different timescale, as the one examined in why intelligent people keep choosing badly.

What has been shown to move it inside a relationship

If relationships change attachment, the obvious question is which part of a relationship. One study was built to answer exactly that, and its result has a twist worth the whole section.

Ximena Arriaga, Madoka Kumashiro, Eli Finkel, Laura VanderDrift and Laura Luchies followed 134 romantically committed couples and separated two things a partner provides: trust, meaning the partner is perceived as available and dependable, and perceived goal validation, meaning the partner is perceived as encouraging your own pursuits.

Measured at the same moment, the pairings were the intuitive ones: trust went with lower attachment anxiety, goal validation with lower avoidance. Measured across time, they swapped. Trust uniquely predicted reduced avoidance later on, and perceived goal validation uniquely predicted reduced anxiety later on.

That inversion is the kind of result that only shows up when someone runs the longitudinal analysis, and it is a quiet argument against the obvious advice. Reassuring an anxious partner addresses the anxiety in the moment; what predicted their anxiety falling over time was being backed in something of their own.

Arriaga, Kumashiro, Jeffry Simpson and Nickola Overall later built that distinction into the Attachment Security Enhancement Model, which separates partner responses that buffer insecurity in the moment from situations that revise the underlying model — and their proposal for the second is explicitly tentative, which is more honesty than this subject usually gets.

The companion caution comes from a study with a longer lens. Lee Kirkpatrick and Cindy Hazan followed 177 adults over four years. Attachment styles were highly stable. Secure respondents were less likely to report a breakup over the interval — and yet, paradoxically, ambivalent respondents were just as likely as secure ones to still be with the partner they had named four years earlier.

Which is to say: “find a secure partner and you will become secure” is a reasonable hypothesis that nobody has demonstrated. What has been measured is narrower and more actionable — trust and being backed, accumulated over time. The behaviours that produce the first of those are set out in what the evidence supports about emotional maturity, and what tends to be visible early in a relationship that lasts is in the signs a new relationship has real potential. What builds over months rather than days is covered in what actually builds attraction over time.

Why it is hard: the loop that keeps the pattern in place

The reason change is slow is not willpower, and framing it that way is both cruel and wrong. It is that insecurity is partly self-maintaining, and there is a study that shows the machinery working.

Meredith McClure and John Lydon ran three studies on what happens at the very start of a relationship. Study 1 was speed-dating. Study 2 videotaped participants introducing themselves to an attractive single confederate who would supposedly decide whether to meet them. Study 3 videotaped a 40-minute semi-structured interaction with an attractive, friendly confederate.

Across all three, attachment anxiety predicted worse interpersonal outcomes, and the mediators were displays rather than attitudes: social disengagement in Study 2, manifest anxiety in Studies 1 and 3. In Study 3 the displays were coded concretely, as verbal disfluencies and interpersonal awkwardness.

The authors' own closing inference is the sentence this whole page is organised around. Those failures reinforce the negative expectations underneath, making it harder for more anxiously attached people to initiate and develop the very relationships that might over time help them overcome the insecurity.

That is a loop, not a character flaw, and it has a structural cousin. Fraley and Roisman describe socialization–selection asymmetries: socialization processes dominate early in life and early in a relationship, while later on person-driven selection takes over and people increasingly seek out contexts that fit the models they already hold. Two predictions follow and both are observed — attachment is more stable in adults than in adolescents, and more stable in long-established relationships than in new ones.

The second prediction has an unexpectedly hopeful edge. If established relationships are the harder place to move an attachment pattern, then a new relationship is the structurally easier one — which is close to the opposite of the usual advice to fix yourself before dating anybody.

It also reframes a familiar complaint. The pattern of ending up in the same position repeatedly, examined in why people keep ending up in the friend zone, is less often a run of bad luck than a loop of this kind with a person inside it.

Therapy has the most direct evidence, and it is still thinner than you would hope

If you want the route with the most published evidence behind it, it is psychotherapy. The evidence is real, it is more encouraging than most things on this page, and it comes with a caveat that is usually dropped.

Peter Taylor, Julia Rietzschel, Adam Danquah and Katherine Berry systematically reviewed studies of change in adult attachment representations during psychological therapy, covering both interview-based and self-report traditions and rating the methodological quality of each. Attachment security increased and attachment anxiety decreased following therapy. Findings for avoidance were unclear. Improvements appeared across different methodologies, patient groups, therapeutic approaches and settings, and held across levels of study quality. Their own conclusion asks for controlled trials before anyone treats it as established.

The strongest single piece of evidence is a randomised one. Kenneth Levy and colleagues randomised 90 patients reliably diagnosed with borderline personality disorder to a year of transference-focused psychotherapy, dialectical behaviour therapy, or a modified psychodynamic supportive therapy. After 12 months, the number classified secure on the Adult Attachment Interview rose significantly in the transference-focused arm and not in the other two. Narrative coherence and reflective function also rose in that arm. Resolution of loss or trauma did not change in any of them.

Diana Diamond and colleagues extended that work with 104 women in a randomised trial comparing transference-focused psychotherapy against experienced community psychotherapists. Patients who shifted from a disorganised, unresolved attachment classification to an organised one showed the predicted improvements in personality organisation; those who did not shift improved in only one domain.

And now the caveat that should travel with all three. These are clinical samples with serious diagnoses, and in the randomised trials the treatment that moved attachment was a specific, manualised, year-long psychodynamic protocol. None of this is evidence about what happens to a person with no diagnosis who takes a quiz, reads about avoidance and books six sessions. It is evidence that the construct is movable under intensive conditions, which is a real and important thing to know, and is not the same claim.

Couple therapy is where “the relationship changed it” has actual numbers

There is one body of work that measured attachment changing inside a relationship while someone deliberately worked on the relationship, and it is worth separating from the rest because it answers a different question.

Emotionally focused couple therapy is built on attachment theory, and Stephanie Wiebe and Susan Johnson's review sets out the case that it meets the published criteria for classification as an evidence-based couple therapy. The interesting work is what happened when researchers checked whether the attachment part was actually moving.

Melissa Burgess Moser, Johnson, Tracy Dalgleish, Marie-France Lafontaine, Wiebe and Giorgio Tasca followed 32 couples session by session, using hierarchical linear modelling and measuring relationship-specific attachment rather than the global kind. Couples significantly decreased in relationship-specific avoidance. Couples who completed a blamer softening — a specific in-session event in this model — significantly decreased in relationship-specific anxiety. Attachment behaviour moved toward security. And session-by-session decreases in both dimensions tracked session-by-session increases in relationship satisfaction.

The follow-up matters more than the main result, because “does it last” is the question nobody answers. Wiebe, Johnson and colleagues modelled the same couples from pre-therapy through 24 months after it ended. Relationship satisfaction and secure-base behaviour kept rising and relationship-specific anxiety kept falling across follow-up, at a decelerated rate.

The limits are plain and should be stated rather than buried: 32 couples, no control group, self-selected into therapy, and the measure is attachment within that relationship rather than the disposition you carry into the next one. What it supports is a bounded claim — that two people working on a relationship with a structured protocol can measurably change how secure each feels in it, and that the change was still there two years later.

Given what the third section established about relationship-specific representations, that bounded claim may be the more useful one anyway. What the repair research says about the mechanics of those conversations is in how to reconnect after a fight, what the longitudinal work on lasting couples actually measured is in the habits of couples who stay together, and if the recurring problem is the shape of the arguments rather than the closeness, our conflict style test is built on the demand-withdraw literature instead.

Can you do it on your own? One study asked, and here is what it found

The honest state of play: exactly one line of research has tested whether simply wanting to become more secure is associated with becoming more secure, and nobody has run a trial of the self-help version.

Nathan Hudson, William Chopik and Daniel Briley built a measure of people's desires to change their attachment anxiety and avoidance, across three studies with more than 4,000 participants combined. The third study was a 16-wave weekly longitudinal design. Goals to change attachment anxiety and avoidance predicted corresponding growth over time: people who wanted to become less anxious tended to show declines in attachment anxiety across the weeks.

That is genuinely encouraging and it is not an experiment. Nobody was assigned a goal. Everything is self-reported on the same kind of scale. The people who want to change may differ from those who do not in ways that do the work, and sixteen weeks is short against the ten-year figure in the first section. The authors' own framing is appropriately hedged: people may be able to increase in attachment security of their own volition.

The adjacent literature that gets cited as if it were the same thing is security priming — exposing someone to representations of a supportive attachment figure and measuring what shifts. Omri Gillath, Gery Karantzas, Daniel Romano and Katherine Karantzas meta-analysed 120 studies across 97 articles, N = 18,949, and found a large overall effect, d = .51, largest for affect (d = .62) and smaller for behaviour (d = .44) and cognition (d = .45). Subliminal priming effects on behaviour were larger among people higher in trait anxiety and avoidance. Their assessment of publication bias found mixed evidence for possible asymmetry, which is the authors being careful and should be carried along with the number.

The thing to notice is what the outcome variables are. Priming studies measure a state shortly after the prime. No study has shown that priming changes trait attachment, and the meta-analysis does not claim it does. It belongs in the column of evidence that the system is responsive, not in the column of interventions that move a baseline.

If the self-directed route interests you, the closest thing to a mechanism with evidence behind it is not an exercise but a practice: reciprocal, gradually deepening disclosure with people who respond well, which is the subject of how people actually open up.

“Earned security” is a real term with an uncomfortable study behind it

Earned security is the phrase this whole subject hangs its hope on: the person who had a difficult childhood, did the work, and became secure. It is a real construct in the research literature, and the best study of it found something that almost nobody repeats.

The term comes from the Adult Attachment Interview. Retrospectively defined earned-secures are people who describe negative childhood experiences coherently — coherence being what the interview scores. Earlier work had found that they parent as effectively as continuous-secures, who describe positive childhoods coherently, while showing more depressive symptomatology.

Glenn Roisman, Elena Padrón, Alan Sroufe and Byron Egeland tested the premise that was being taken for granted: that these people actually had the insecure history they describe. They used data from a 23-year longitudinal study, so infant attachment and observed parenting were on record rather than recalled.

The premise failed. Retrospectively defined earned-secures were not more likely than continuous-secures to have been anxiously attached in infancy, and in childhood and adolescence they had been observed to receive among the most supportive and structured maternal parenting in a high-risk sample. Meanwhile, prospectively defined earned-secures — people identified by their actual infant classifications rather than their narratives — did go on to succeed in close relationships, many of them without the high internalizing distress the retrospective group showed.

There are two ways to read that and both are worth holding. One is deflationary: a chunk of what gets called earned security is a particular way of narrating a childhood, and the people who narrate theirs darkly may be carrying a depressive style rather than a documented history. The other is the opposite of deflationary, and it is the one the prospective result supports: people who really did start insecure can and do end up functioning well in close relationships.

The shorter version of this finding, and what it does to the stories people tell about themselves, is in the hub piece on repeating patterns. What it does not support is the story structure the genre sells, where a bad childhood plus insight equals earned security. The study designed to verify that story could not find the bad childhood in the records. Whatever you are working with, it is more useful to describe it accurately than to narrate it as an origin — and a difficult past is not a prerequisite for a difficult present, nor a disqualification from an ordinary future.

How long it takes, and how to tell a trend from a bad week

Nobody can give you a timeline, and anyone offering one in weeks is inventing it. What the data can give you is a sense of scale and a way to read your own noisy measurements.

The scale, from the first section: about .50 rank-order stability across ten adult years. Rank order is not the only kind of stability, though, and the other kind carries better news. William Chopik, Robin Edelstein and Kevin Grimm pooled three long-running longitudinal studies to follow attachment from age 13 to 72, N = 628: attachment anxiety declined on average with age, particularly through middle and later adulthood, and avoidance declined linearly across the lifespan, with being in a relationship predicting lower levels of both. A cross-sectional snapshot of 86,555 people aged 18 to 70 by Chopik, Edelstein and Fraley points the same way for anxiety, with the caveat that cross-sectional age differences can be cohort effects rather than development. Nobody in those datasets was working on anything. The default drift is gentle and in the direction you would want — which is also why a change you notice over a decade is weak evidence that whatever you tried caused it. The same three findings, sized the same way, are summarised inside the hub piece on repeating patterns.

Underneath that average, individual movement is constant. Fang Zhang and Gisela Labouvie-Vief ran a six-year longitudinal-sequential study from late adolescence to late adulthood and summarised it precisely: attachment style was relatively stable over the period, yet characterised more by fluidity than by stability. Fluctuation in security covaried negatively with defensive coping and depressive symptoms and positively with integrative coping and wellbeing — when people were coping better than their own average, they were also more secure than their own average.

The resolution gets finer still. Joanne Davila and Elizabeth Sargent ran an eight-week daily diary and found that it was not the objective features of events that predicted daily insecurity but the perception of interpersonal loss attached to them, controlling for the events themselves and for mood. Trait security did not moderate it. Everyone's daily reading moves with how the day's events are construed.

The practical rule falls out of that. A single measurement is a day, not a position. If you want to know whether anything is moving, you need months of measurements and you have to compare yourself to your own average rather than to a category.

And here is the warning that keeps this from being a feel-good section. Davila, Dorli Burge and Constance Hammen followed 155 women from just after secondary school through six months and two years, and asked whether attachment change was a reaction to current circumstances or an individual difference in susceptibility to change. The results mainly supported the second. Some women appeared prone to attachment fluctuation because of adverse earlier experiences, and the women who fluctuated resembled the women who were stably insecure.

So “my attachment style keeps changing” is not automatically a sign of progress. Instability can be the symptom. A moving baseline that trends in one direction over a year is change; a reading that swings with the week is the ordinary operation of the system, and in some people a marker in its own right. The same caution applies to what people read into the ups and downs of a single relationship — the piece on texting habits covers how quickly a thread gets mistaken for a trend, and our relationship satisfaction test reports dimensions rather than a verdict for the same reason. The same trap has a well-measured version at the scale of years: why the honeymoon phase ends sets out how much satisfaction normally drops, so that an ordinary slope is not mistaken for a verdict either.

What the field itself says is not known

The most useful thing a page like this can do is mark the edge of the evidence, because the pages competing with it do not have one.

Fraley closed his 2019 review with a list of unresolved questions, and one of them is the subject of this article almost word for word: what causes attachment styles to change; when changes are observed, whether those changes are sustained or transient; and whether there are practical interventions that can facilitate enduring change. That is the field's leading reviewer, in the Annual Review of Psychology, describing the central practical question as open.

Underneath it sits a measurement problem that is being worked on rather than solved. Sascha Müller and colleagues tested whether two German short forms of the Experiences in Close Relationships–Revised questionnaire can validly measure change, using two independent clinical samples of 493 and 273 people in a pre-post design. Strict longitudinal measurement invariance held, meaning score changes can be read as changes in the underlying dimensions, and both measures were sensitive to change — attachment anxiety was overall reduced after therapy. But the same paper reports consistent problems with structural validity across the samples and says the measures should be re-examined.

Three more gaps are worth naming because their absence is routinely papered over:

Nobody has tested the self-help version. There is no trial of reading about attachment, taking a quiz or following an online programme, measured against a control, with a follow-up long enough to matter.

Nobody has shown that a secure partner makes an insecure person secure. It is a plausible mechanism and the ingredients have been measured piecemeal; the study has not been run.

Nobody knows how much of the change seen in therapy survives the therapy. The two-year couple-therapy follow-up is the longest available and rests on 32 couples.

One number currently circulating is worth naming, because of where it comes from. Search for this question today and the AI summary sitting above the results may tell you that approximately 25% of significant life events produce enduring changes in attachment style. We went looking for the study and could not find it — not in Crossref, not in OpenAlex, not in the pages ranking for the query. What has actually been counted is narrower and comes from Waters and colleagues: 44% of the people whose mothers reported a negative life event changed classification over twenty years, against 22% of those who did not. Those are proportions of people, not of events, and they come from 50 participants.

The pattern is familiar — it is the same thing that happened to the 7-38-55 figure and to Gottman's 69% of perpetual problems, both taken apart elsewhere on this site. What is new is the delivery. A figure with no traceable source is now being assembled by a search engine and placed above the results rather than buried in somebody's blog post, which makes it harder to check and easier to repeat.

Writing that down is not throat-clearing. In a subject where a confident number sells, the absence of one is the finding — the same structural problem described in what can and cannot be predicted about compatibility, where the measurable predictors of a relationship all turn out to be judgments about a relationship that already exists.

When this stops being a question about attachment

One thing on this page can be said with more confidence than anything else on it, so it goes at the end where it will be remembered rather than buried in a disclaimer.

The framing of “working on my attachment style” becomes actively dangerous in one situation: when the insecurity someone is trying to fix in themselves is a reasonable response to how they are being treated. Feeling anxious because a partner monitors your phone, decides who you may see, punishes you with silence or escalates when you say no is not an attachment pattern to be revised. It is an accurate reading of your circumstances, and the literature on control is clear that these behaviours cluster and tend to intensify. Where the feeling and the behaviour have to be told apart, the psychology of jealousy goes through the distinction at length.

If any of that is recognisable, the useful next step is a person, not a framework. Find A Helpline, linked in the sources below, lists free and confidential services by country. In the United States the National Domestic Violence Hotline is 1-800-799-7233, or text START to 88788, 24 hours a day.

Two smaller cautions that follow from the evidence rather than from caution itself.

An attachment score is not a diagnosis and is not a reason to leave or stay. It is a position on two continuous dimensions, measured once, on a day. Our test says so on its own page, and the sections above are an argument for why.

Nothing here is clinical advice. The strongest results on this page came from year-long manualised psychotherapy with clinical populations and from structured couple therapy. If what you are carrying is heavy enough that you are reading articles about it at two in the morning, the thing the evidence actually supports is talking to someone qualified — not harder reading.

And the close, which is the honest version of the question in the title. Attachment styles change, slowly, mostly through relationships and therapy rather than through insight, differently in different relationships, and not always in the direction of security. Over a decade about half of where you sit will still be where you sat. The other half is not fixed, and knowing which half is which is most of what this subject has to offer. More in this vein under Mindset, and if the next question is what moving on from a relationship actually looks like, that is the piece on signs you have moved on.