If you are not alone right now, this page will be in your browser history. You can read it in a private window, or clear it when you are done.
Not being able to leave a relationship you already know is bad is produced by the structure of the relationship, not by a weakness in the person standing inside it. The mechanism has a name. Traumatic bonding is the strong emotional attachment that forms when a power imbalance and intermittent, unpredictable maltreatment run in the same relationship at the same time — and it is the reason knowing has not been enough.
Traumatic bonding describes a mechanism. It is not a diagnosis of anyone, not a verdict on any particular marriage, and not an argument that leaving is the right answer for any particular woman. This page explains why the knowing has not moved you. It does not tell you what to do with the explanation, and nothing on it is an assessment of you or of him.
Why doesn’t knowing it’s bad change anything?
Knowing the relationship is bad does not change anything because knowing and leaving run on different systems, and only one of them is made of information. You have already done the part that words can do. You have named it, possibly out loud, possibly only to a browser at one in the morning. The naming was real work and it did not move your feet, and that discrepancy is what you have been treating as evidence about your character.
On a survivors’ forum, one woman asked why she wasn’t strong enough to go. Seventy replies came back, from more than fifty different women. Every one of them recognized her. Not one of them could tell her why. That is the exact shape of the gap this page exists to fill: she was not short of company, and she was not short of sympathy. She was short of a mechanism.
The mechanism was first described by Donald Dutton and Susan Painter, Canadian psychologists who proposed in the early 1980s that two conditions, occurring together, are sufficient to produce powerful attachment to a person who is hurting you: a marked power imbalance, and intermittency — maltreatment that arrives unpredictably, interspersed with periods of ordinary or affectionate treatment. Neither condition on its own does it. It is the combination, and the combination is not rare. It is the ordinary architecture of what coercive control actually is.
Dutton and Painter later tested it. In a study of 75 women who had recently left abusive relationships — 50 of them relationships in which physical violence had occurred — assessed at separation and again six months later, the relationship variables of total abuse, intermittency of abuse and power differential together accounted for 55% of the variance in how attached a woman still felt to her former partner at six months (Dutton & Painter, 1993). More abuse, more unpredictably delivered, in a relationship with a steeper power gradient, produced more felt attachment, not less. That is the finding, and it is the opposite of what everyone around you assumes.
Why do the good stretches make it harder to go, not easier?
The good stretches make it harder to go because a reward that arrives unpredictably produces more persistent behavior than a reward that arrives every time. That is not a metaphor and it is not pop psychology. It is one of the most reliably reproduced findings in the psychology of learning, it has a name, and it was established in laboratories decades before anybody applied it to a marriage.
You have probably described the good weeks apologetically. He was lovely for two weeks straight. There was a holiday, or a Sunday, or one conversation in a car where he was the person you married, and you have been carrying it around ever since as proof of something — either that he is really like that underneath, or that you are a fool. The accurate account is neither. The good stretch is doing work on your behavior, and it does the most work precisely when you cannot predict it.
The construct is the partial reinforcement extinction effect: behaviors that produce only periodic reinforcement are more resistant to extinction than behaviors that have been reinforced every time (Williams, 2019). A response that was always rewarded stops quickly once the reward stops, because the absence is obvious. A response that was rewarded sometimes does not stop, because absence carries no information — you have sat through gaps this long before and the reward came back. There is no point at which the pattern tells you it is over.
Applied to a marriage, the arithmetic is bleak and it is not about you. The behavior analysts Miller, Lund and Weatherly set it out directly: the contingencies in an abusive relationship increase the effectiveness of whatever reinforcement the relationship still delivers, decrease the effectiveness of the abuse as a punishment for staying, and punish the alternative behavior of leaving (Miller, Lund & Weatherly, 2012). That paper is a conceptual analysis rather than a trial, and it should be read as one. What it supplies is the map. Every arrow on it points the same way, and none of them points at your character.
The contingencies in an abusive relationship are why he has been good lately is the hardest moment in the whole thing rather than the easiest.
Why does it get worse when I stop reacting?
Behavior that has been reinforced for years does not stop quietly when the reinforcement stops. It intensifies first. In behavior analysis this is called an extinction burst — a temporary increase in the rate, force or variability of a behavior at the point where it stops producing its usual result. If you have started declining to argue, declining to explain yourself, declining to apologize for things you did not do, and the response has been an escalation rather than a fading, you have not made a mistake. You have produced the textbook effect.
The evidence deserves reporting with its limits attached, because the phrase is used far more confidently in popular writing than in the literature it comes from. In an analysis of 113 sets of extinction data drawn from the behavioral treatment of severe problem behavior in clinical populations, an increase in responding meeting the burst criterion appeared in 27 of them — about 24% — and was more common when extinction was used on its own than when it was combined with other components (Lerman & Iwata, 1995). That population is not yours; it is clinical treatment data from a completely different field. The burst is a real and well-described phenomenon, it is not universal, and nobody has measured it in marriages.
Two things follow, and the second one matters more than the first. The first is that escalation when you stop complying is a predictable property of the system rather than proof that you provoked something. The second is that a predictable property is still dangerous. Nothing on this page is a recommendation to test it. If withdrawing your responses has produced escalation, that is information for a trained advocate, on the phone, not information to sit on alone.
There is a version of this question that is really a different question — whether the escalation is control or whether it is his condition. That one is answerable, and it is a different question from this one.
If you are in immediate danger, call 911.
National Domestic Violence Hotline — 1-800-799-7233, 24 hours · text START to 88788 · thehotline.org
988 Suicide and Crisis Lifeline — call or text 988
Deaf, DeafBlind and hard-of-hearing callers: National Deaf Domestic Violence Hotline videophone 855-812-1001, 24 hours (a partnership between the National Domestic Violence Hotline and Abused Deaf Women’s Advocacy Services).
Calling leaves the number in your call log and texting leaves the thread in your messages. Both can be deleted afterwards. If that is not something you can do safely, a friend’s phone or a payphone is the safer route.
Why have I gone back before?
Going back is not a failed exit. In the research literature, leaving an abusive relationship is described as a process rather than an event — a sequence of shifts in how a woman understands what is happening to her, frequently including separations that do not hold, before any separation that does.
You already know the going-back part is the least sayable part. It costs you the friend who helped with the boxes, the sister who took the children for two weeks, the manager who was so good about it. Going back does not just return you to the house — it spends the credibility you needed in order to be helped, and it does that publicly, which is why women who have returned once are so much quieter the second time around. That cost is real. It is also imposed by other people’s model of leaving, and their model is wrong.
The research model is different. Deborah Anderson and Daniel Saunders, reviewing the empirical literature, concluded that women typically undergo several shifts in their thinking about the abuse before leaving permanently (Anderson & Saunders, 2003). The returns are inside the process, not evidence against it.
Is there a number of times before it counts?
There is no established number of separations before a woman leaves for good, and you have almost certainly met one anyway — a specific figure for how many times the average woman leaves first. That figure has no study under it. It is printed by large and reputable organizations, and where they cite anything at all for it, the citations lead to a news feature and a patient leaflet rather than to data. It is not on this page for that reason.
What the figure does is the reason for saying so rather than staying quiet. A number like that only ever works as a ruler. If you have left fewer times than it, you are owed some attempts; if you have left more, you are past the point where it worked for other people. Both readings are wrong, because the number was never measuring anything. Leaving does often take more than one separation — that part is documented and it is real. Nobody has counted how many, and there is no quota: not one you are behind on, and not one you have used up.
What has been measured is the movement itself, and one research team measured it because the field had assumed the answer. Bell, Goodman and Dutton open by observing that most writing about abuse assumes a woman has to end the relationship in order to be safer and better, and that very few studies exist to support that assumption. In a longitudinal study of 206 low-income, predominantly Black women who had experienced intimate partner violence, followed for one year, the women who held one relationship state across the whole year — separated the whole year, or together the whole year — were doing better at the end of it than the women who moved in and out repeatedly (Bell, Goodman & Dutton, 2007).
The Bell, Goodman and Dutton finding says neither of the two things it is constantly assumed to say. It does not say that leaving works. It does not say that staying is fine as long as you are consistent about it. The authors themselves note that the causation could run the other way — that changes in the violence across the year may be what moved women in and out, rather than the moving in and out producing the outcomes. One study, one population, one year.
What the study takes away is the idea that a return is the failure. The cost that shows up in the data attaches to the transitions, and the transitions are a property of the pattern rather than of anybody’s resolve.
The women who are hardest on themselves about this are almost always the ones who have already tried. They do not describe the return as a decision. They describe it as the moment the pressure stopped — the calls, the messages, the campaign among the family — and the relief of that is so enormous that it gets filed as love. Then they spend years calling themselves weak for having felt relieved. Relief at the end of a siege is not love and it is not weakness. It is what a nervous system does when a siege ends.
If what stopped you last time was the practical wall rather than the feeling, that wall is a separate problem from this one. If the pressure did not stop when the marriage did, that has a name too: post-separation abuse.
Does this mean there’s something wrong with me?
Nothing described on this page requires there to be anything wrong with you, and the mechanisms only work because there is nothing wrong with you. Traumatic bonding needs an ordinary capacity for attachment. The partial reinforcement effect needs an ordinary nervous system that learns from what happens. An extinction burst needs a behavior that was previously working on somebody who was previously responding. Every one of these mechanisms runs on standard equipment. That is precisely why they are so hard to beat.
The question underneath why am I not strong enough is usually a different question, and it is worth separating out. It is rarely about strength. It is about whether you are allowed to describe what has happened to you without first proving that you did everything correctly — and you have almost certainly been in at least one conversation where the price of being believed was having already left.
Coercive control specifically is engineered to remove the resources that leaving requires, which means the situation itself has been shrinking the very capacities you are now grading yourself against. Money, transport, the sister who stopped calling, a job you turned down, the friend he decided against in 2019, and your own confidence in your account of events. A person is not weak for finding it hard to walk out of a room whose exits have been closed one at a time over eleven years.
The women who arrive most convinced of their own weakness are usually the ones running the most complicated operation. They are forecasting a mood from the sound of a car door, managing the children’s timing, keeping a separate mental ledger of what was actually said, deciding which version of the day is safe to tell, and doing all of it while holding down work. That is not passivity. It is continuous, skilled, exhausting labor performed under threat, and nobody has ever counted it as anything.
What this page can tell you, and what it cannot
This page can tell you what traumatic bonding is, why an unpredictable reward schedule produces the most persistent behavior anybody has been able to produce in a laboratory, why escalation follows the withdrawal of your responses, and why the research treats leaving as a process containing returns rather than as a single act with a pass mark. This page cannot tell you whether this describes your relationship, whether you should leave it, or when. Those need a person, and the first two need a person who is not a website.
One thing said plainly, because it is the thing the mechanism actually establishes. Understanding why you have not left is not a step toward leaving. It is not a stage, it is not a softening-up, and it does not commit you to anything. A woman who reads this page, recognizes every mechanism on it, and stays exactly where she is has not failed the page. She has got an accurate account of her own situation, which is more than she had this morning, and it is hers to do nothing with.
A next step, if you want one, does not have to be a decision. It can be working out what else there is if you are not leaving — a question that belongs to the woman who is staying and does not pretend otherwise. If you would rather talk to somebody now, the National Domestic Violence Hotline is 1-800-799-7233, 24 hours, or text START to 88788. Advocates there do safety planning, which is skilled work, done free, with information a website cannot see, and you do not have to have decided anything to call. For anything legal, WomensLaw.org is free and organized state by state.
If you would rather take this to a person than a page, ask enough to tell, before booking, whether somebody understands the mechanism this page describes or will treat it as a decision you keep failing to make.
You do not have to have decided anything to start. Not leaving. Not staying. Nothing at all.
SOURCES
- Dutton, D.G., & Painter, S.L. (1981). Traumatic bonding: The development of emotional attachments in battered women and other relationships of intermittent abuse. Victimology: An International Journal, 6(1– 4), 139–155.
- Dutton, D.G., & Painter, S. (1993). Emotional attachments in abusive relationships: A test of traumatic bonding theory. Violence and Victims, 8(2), 105–120. PubMed 8193053
- Williams, W.A. (2019). Partial reinforcement extinction effect. In J. Vonk & T. Shackelford (Eds.), Encyclopedia of Animal Cognition and Behavior. Cham: Springer. doi:10.1007/978-3-319-47829-6_809-1
- Miller, K.B., Lund, E., & Weatherly, J. (2012). Applying operant learning to the stay-leave decision in domestic violence. Behavior and Social Issues, 21, 135–151. doi:10.5210/bsi.v21i0.4015
- Lerman, D.C., & Iwata, B.A. (1995). Prevalence of the extinction burst and its attenuation during treatment. Journal of Applied Behavior Analysis, 28(1), 93–94. doi:10.1901/jaba.1995.28-93
- Anderson, D.K., & Saunders, D.G. (2003). Leaving an abusive partner: An empirical review of predictors, the process of leaving, and psychological well-being. Trauma, Violence, & Abuse, 4(2), 163–191. doi:10.1177/1524838002250769
- Bell, M.E., Goodman, L.A., & Dutton, M.A. (2007). The dynamics of staying and leaving: Implications for battered women’s emotional well-being and experiences of violence at the end of a year. Journal of Family Violence, 22(6), 413–428. doi:10.1007/s10896-007-9096-9
- Stark, E. (2007). Coercive Control: How Men Entrap Women in Personal Life. New York: Oxford University Press.
- National Domestic Violence Hotline — 1-800-799-7233, 24 hours. thehotline.org
Educational content. Not therapy, legal, or financial advice. Nothing on this page is an assessment of any person or any relationship.