Intermittent Reinforcement in Relationships: The Mechanism Behind Why You Can’t Leave

RELATIONAL TRAUMA • July 20, 2026

The Good Days Keep Her Hooked

Esperanza is sitting in the parking garage under her hospital at 6:40 on a Tuesday morning, engine off, hands still on the wheel. She has a 7 a.m. leadership huddle in twenty minutes and a phone full of texts from the night before, the good kind, the kind that used to be rare and are rare again now. He called her brilliant three times. He said he didn’t know what he’d do without her. She read the last one twice before she put the phone face down on the passenger seat, the way she always does when she doesn’t trust herself not to read it a third time.

“I know what this is,” she told me the week before, sitting on the same couch she’s been sitting on for eight months. “I know the pattern. I can draw you the whole cycle on a whiteboard. And I still spent forty minutes this morning rereading old texts like they were going to tell me something new.”

Sitting with Esperanza that morning, I felt the particular fatigue that shows up when a client is fighting her own accurate insight. She runs a forty-person department. She has read the books. She knows, cognitively, exactly what is happening to her. None of that knowledge has been enough to make the pull weaker.

In my work with driven women over the past fifteen years, specifically those trying to leave an on-again-off-again relationship, I’ve noticed a pattern that surprises almost none of them once I name it and surprises almost all of them in how little their intelligence has protected them from it. Understanding a mechanism and being free of its grip are two entirely different achievements. This post is about the mechanism, and about what it actually takes to get free of it once you understand it.

What Intermittent Reinforcement Actually Is

Here’s the plain version first, before the clinical one. If someone was unkind to you every single day without exception, you would leave quickly. Consistent harm is legible. Your nervous system knows what to do with a clear, repeated signal: it moves you toward the exit. What confuses the nervous system, and what keeps women sitting in parking garages rereading old texts, is a reward that arrives on no fixed timetable at all.

This is the part I want to be precise about, because precision is what makes this useful rather than just interesting. The relationship doesn’t need to be good most of the time to produce this effect. It needs to be unpredictable. A relationship that is 20 percent wonderful and 80 percent painful can still generate more compulsive attachment than a relationship that is 100 percent pleasant, because the nervous system isn’t grading the average. It’s tracking the uncertainty. The not-knowing is the hook. This is one reason anxious attachment patterns and intermittent reinforcement so often travel together: a nervous system already primed to scan for a partner’s availability is exactly the nervous system most vulnerable to variable reward.

What I see consistently in my practice is that clients describe the good moments with more detail and more feeling than the bad ones, even when the bad moments were objectively worse and more frequent. That isn’t denial. It’s the memory doing exactly what a variable reward schedule trains it to do: highlighting the unpredictable win, because the unpredictable win is the one the brain is organized around chasing.

The Neurobiology of an Unpredictable Reward

The mechanism underneath all of this lives in the brain’s dopamine system, and the research here is old enough and replicated enough that it’s worth naming precisely. Wolfram Schultz, PhD, Professor of Neuroscience at the University of Cambridge, known for discovering the dopamine reward-prediction signal, found something that reframed how an entire field thinks about motivation: dopamine neurons don’t fire most strongly when a reward arrives. They fire most strongly when a reward arrives and wasn’t expected. A fully predicted reward barely moves the dopamine signal at all. An unpredicted one spikes it. This single finding is the biological floor underneath everything this post is about.

Read that again in the context of a relationship that’s warm sometimes and cold sometimes. Every time the good version of him shows up unpredictably, the brain registers something closer to a jackpot than to comfort. Comfort would be a small, steady signal. A jackpot is what happens when the brain didn’t see the reward coming. Research on predictive learning has traced this signal at the level of individual neurons responding to expectation itself, not just to outcome (PMID: 42337376), which is part of why insight alone doesn’t turn the signal off. You can know, cognitively, that the good text is a variable reward. Your dopamine system fires anyway, because it isn’t listening to your cognition. It’s listening to the pattern of surprise.

Kay Tye, PhD, neuroscientist and Wylie Vale Chair at the Salk Institute for Biological Sciences, whose research maps brain circuits underlying motivation, emotion, and social behavior, has spent years mapping how these circuits govern approach and avoidance at the same time, in the same nervous system, often in the same ten minutes. That dual circuitry matters here. It’s not that the fear circuit goes quiet while the reward circuit lights up. Both are active. A woman in an unpredictable relationship is frequently running her fear system and her approach system simultaneously, which is a large part of why the experience feels so physically exhausting. Motivation research on uncertainty itself, separate from the reward’s size, has direct implications for how compulsive pursuit patterns form and persist, including patterns that resemble addiction (PMID: 42137086). Uncertainty is not a side effect of the pull. Uncertainty is the pull.

There’s a related body of work on the nucleus accumbens, a structure central to reward processing, showing that dopamine signaling there reflects something closer to the brain’s confidence about whether a cue will or won’t be followed by a reward, rather than a simple registration of pleasure (PMID: 41786606). Which means the system tracking your partner’s mixed signals isn’t a pleasure system exactly. It’s a probability system. It’s running odds on him, constantly, in the background, whether you want it to or not.

The last piece of this picture explains why the pattern hardens over time instead of softening with repetition. Trevor Robbins, PhD, Professor of Cognitive Neuroscience at the University of Cambridge, whose work spans reward, compulsion, and the neuroscience of habit, has spent decades tracing how reward-seeking behavior gradually shifts from something goal-directed, done because you want the outcome, into something closer to habit, done because the doing itself has become automatic. Early in an unpredictable relationship, checking the phone is goal-directed: she wants to know if he texted.

Months or years in, the checking can become close to reflexive, running on its own circuitry with only a loose connection to whether she still even wants the outcome. That shift from goal to habit is part of why willpower alone underperforms here. You cannot simply decide your way out of a habit loop any more than you can decide your way out of a reflex.

How It Shows Up in Driven Women

I want to be specific about why this hits differently for the women I work with, because the mechanism is universal but the presentation isn’t. driven women have typically spent years being rewarded, consistently and predictably, for effort. Study harder, get the grade. Work later, get the promotion. Most of the reward schedules that built her adult life have been reasonably reliable: input in, output out. Which means the one domain where the reward schedule is wildly unpredictable, the relationship, becomes almost irresistibly interesting to a brain that has otherwise been trained on cause and effect.

Esperanza runs a hospital department where her decisions are followed by outcomes she can track within a quarter. She is extremely good at systems that respond to input. The relationship doesn’t respond to input in any way she can model, and I’ve watched this become, paradoxically, part of the draw. “If I could just figure out the formula,” she said once, “I could fix this.” There is no formula. That’s the part her particular brilliance keeps refusing to accept. The pull of an outcome she cannot reliably predict maps onto what researchers describe in work on motivation, attention, and uncertainty PMID: 42137086.

What I see in practice, across a caseload skewed heavily toward driven professional women, is that the same traits that build the career, pattern recognition, persistence, an unwillingness to quit before the problem is solved, become liabilities inside an unpredictable relationship. The persistence that gets her promoted is the same persistence that has her checking his location at midnight. Not always. Not for every client. But often enough that I now ask about it directly in intake, because the women least likely to describe themselves as “the type” to stay are frequently the ones most thoroughly caught. This is a pattern I’ve also seen intersect heavily with codependency in driven women, where caretaking and fixing become the primary language of love. It also shows up in people-pleasing as a trauma response, where staying agreeable and available feels like the only lever left to pull on an unpredictable partner.

There’s also a perfectionism layer that I see specifically in this population, and it deserves its own sentence rather than a footnote. A woman who has organized her entire professional identity around getting things right on the first try experiences an unpredictable relationship as an open loop she cannot close, and open loops are precisely what a perfectionist nervous system is worst at tolerating. Perfectionism rooted in earlier trauma often shows up here as an unwillingness to accept that some relationships genuinely cannot be optimized into safety, no matter how much effort or analysis gets thrown at them.

There’s a childhood layer worth naming plainly, without turning it into the whole explanation. Early experiences of unpredictable caregiving, a parent who was warm sometimes and absent or harsh other times, appear to shape how the adult brain later handles delay and reward. Research on early adversity has found that inconsistent caregiving in the form of maternal separation is associated with increased impulsivity and a documented shift toward preferring smaller, sooner rewards over larger, later ones, a pattern known as delay discounting (PMID: 42374171). If your first reward system was unpredictable, a second unpredictable reward system in adulthood doesn’t feel foreign. It can feel, disturbingly, like the shape love is supposed to take. This is part of why childhood emotional neglect so often surfaces in the histories of women describing this exact relational pattern, and part of why attachment theory offers useful language for why familiar, even when familiar is painful, can masquerade as chemistry.

Why This Isn’t Just a Trauma Bond

I want to draw a careful distinction here, because the terms get used interchangeably online and the imprecision costs something clinically. A trauma bond describes the broader attachment that forms specifically inside a cycle of abuse, where fear and attachment become entangled with each other. Intermittent reinforcement is narrower and more mechanical: it’s the reward schedule itself, the specific pattern of unpredictable positive input that trains the pursuit behavior. A trauma bond can exist without much intermittent reinforcement at all, built instead on isolation and fear. And intermittent reinforcement can operate in relationships that never rise to the level of abuse, showing up in ordinary on-and-off dating patterns, in friendships, even in workplace dynamics with an unpredictable boss.

What both mechanisms share is the outcome that matters most clinically: behavior that persists long after the relationship has stopped making sense on paper. There’s research examining the distinct dimensions of why behavior persists in the face of costs, separate from any single reward’s size, that’s directly relevant here (PMID: 41867846). Persistence isn’t one thing. It’s a cluster of separable tendencies, some about difficulty stopping once started, some about difficulty resisting a cue, some about difficulty updating a plan once it’s failed repeatedly. A woman who has stayed for three years past the point of obvious harm isn’t demonstrating one flaw. She’s demonstrating a specific, nameable configuration of persistence that can be worked with directly once it’s identified, which is very different from being told she simply needs “more willpower” or “more self-esteem” to leave.

In my work with clients navigating fearful-avoidant attachment specifically, this distinction matters because the felt experience of wanting closeness and wanting distance in the same hour can look, from the outside, like ambivalence about the relationship itself. It’s frequently something more precise: a nervous system caught between an approach circuit lit up by unpredictable reward and a threat circuit lit up by real, accumulated harm. Both circuits are telling the truth about something. Neither is lying to her. That’s what makes this so much harder to simply reason your way out of than most people, including well-meaning friends, tend to assume. The brain’s tracking of whether a signal will or won’t arrive appears to be encoded at the level of dopamine itself PMID: 41786606.

Both/And: The Relationship Is Harmful AND Leaving Feels Impossible

Here’s the truth I want you to leave this section holding, because it’s the one that tends to get lost between the friends who say “just leave” and the part of you that knows exactly why you haven’t. The relationship can be genuinely harmful to you AND the difficulty of leaving can be genuinely, neurologically real. These are not competing facts. They are the same situation described from two necessary angles.

Keiko, a litigator I’ve worked with for the better part of a year, put it more precisely than most clinical language manages. “I can build a legal argument for why I should have left eighteen months ago,” she told me, sitting very still, her hands folded the way they fold in court. “I have the argument memorized. I could present it to a jury. And I still spent Sunday hoping he’d text.” She wasn’t confused about the facts. She was describing, with total accuracy, the gap between what she knew and what her reward system was still doing regardless of what she knew. This is consistent with how the brain builds predictions from experience rather than from conscious argument PMID: 42337376.

This is the both/and I ask clients to sit with, because collapsing it in either direction causes harm. If we only say “the relationship is bad, leave,” she walks out feeling stupid for not having left already. If we only say “the pull is neurological, of course you can’t leave,” the mechanism becomes an excuse that keeps her circling. Both things are true at once. The harm is real. The pull is real. Neither cancels the other.

What I’ve come to think of as the credit-and-debit problem shows up here constantly. A woman keeps a mental ledger of the good moments, weighted heavily by their unpredictability, against the bad moments, which she’s learned to normalize because normalizing pain is often what let her function at work the next morning. The ledger isn’t dishonest. It’s the output of a reward system built to detect rare positive surprises and poorly designed for tallying frequent, low-grade harm. Of course the math comes out wrong. It was never built for this problem.

The Systemic Lens: Why We Blame Women for Staying

The pattern I’ve just described isn’t a personal failing, and it isn’t unique to any one woman’s psychology. It’s patterned, and the pattern has a cultural container around it that deserves naming directly. We live inside a culture that still asks, reflexively, “why didn’t she just leave,” as though leaving an unpredictable reward system were a simple act of willpower rather than a fight against one of the most well-documented mechanisms in behavioral science. Early adversity itself can shift how steeply a person discounts delay and tolerates impulsive pursuit PMID: 42374171.

That question, why didn’t she leave, does real work in the world, and none of it favors the woman being asked. It locates the failure in her judgment rather than in the mechanism operating on her nervous system. It ignores that driven professional women in particular are often socialized from early childhood to read inconsistency as a puzzle to solve through more effort rather than a signal to disengage from. A culture that rewards women for not giving up has also quietly trained many to treat “not giving up” on a harmful relationship as a virtue rather than a trap.

Here’s how that inheritance lives in an ordinary week. It’s the friend who says “I would have left after the first time,” delivered without malice but landing as judgment. It’s the therapist, and there are still plenty of them, who frames the staying as a self-esteem problem rather than a reward-schedule problem, sending a driven, insight-oriented woman down a rabbit hole that has nothing to do with what’s actually keeping her stuck. It’s the quiet social calculation every woman makes about whether to tell anyone she’s still in contact with him, because she already knows what the room’s face will do.

You are not weak for having stayed. You are not less intelligent than the women who left sooner. You were operating a nervous system doing exactly what nervous systems, human ones, do under a variable reward schedule, inside a culture that has spent a very long time asking women to solve inconsistency with more patience instead of less exposure to it. That is not a character flaw. That is a structural inheritance layered on top of a biological mechanism, and both layers need to be named before “just leave” can mean anything useful.

What Actually Helps

Understanding the mechanism is the first step, and it’s genuinely useful; most of my clients report real relief just from having an accurate name for what’s happening to them instead of a vague sense of personal failure. Why a behavior persists at all, long after it stops serving the person, is itself a well-studied question PMID: 41867846. But understanding alone rarely breaks a variable reward pattern, because the pattern doesn’t operate at the level of understanding. It operates at the level of exposure. The single most reliable lever I’ve seen work, across hundreds of cases, is reducing or eliminating contact long enough for the reward schedule itself to stop being reinforced. This is also where working with a trauma-informed therapist who understands driven women specifically tends to matter most, because generic advice to simply have more willpower badly misreads what’s actually happening in the nervous system.

Shoshana, a finance director who came to me describing the unpredictability itself as “the thing I miss, not him,” gave me one of the clearest descriptions of an extinction burst I’ve heard in fifteen years of doing this work. “The third day of no contact was worse than any fight we ever had,” she said, arms crossed, sitting forward on the edge of the couch. “My whole body was waiting for something that wasn’t coming. I kept checking my phone like it owed me an answer.” That’s not weakness. That’s a reward system, correctly, running out its old programming with nothing new to replace it yet.

What helps, concretely, in the order I tend to see it help: first, structured no-contact or drastically reduced contact, long enough to let the extinction burst pass, usually longer than clients expect and shorter than they fear. Second, deliberately building new, reliable sources of positive feeling, work wins, friendships, physical movement, so the nervous system has somewhere else to register reward while the old pattern goes quiet. Third, and this is the part clients resist most, learning to tolerate the discomfort of predictability itself, because a nervous system trained on variable reward can experience calm, steady affection as boring or even suspicious at first. Nervous system regulation work matters here, not as an abstract wellness practice but as the literal retraining of a system that has spent years on high alert for the next unpredictable cue.

I also want to name something practical about why setting boundaries feels impossible after trauma like this: a boundary against a person who trained your reward system to chase him is not the same skill as a boundary against someone who never had that pull on you. It needs more scaffolding, more support, and often more time, and expecting yourself to do it with the ease you’d use on a difficult coworker sets you up to feel like you’re failing at something that was never simple.

Rebuilding this capacity is closely tied to what I write about under the window of tolerance more broadly, and to the slower work of recognizing the signs you are healing from trauma, which rarely look like triumph and much more often look like a quieter phone and a longer stretch of ordinary, uneventful Tuesdays.

Esperanza, six months into structured no-contact when we last spoke, still keeps her phone face down on the passenger seat out of habit. She doesn’t check it in the parking garage anymore. Most mornings. Not every morning. “I still notice the pull,” she told me. “I just don’t follow it into the parking lot the way I used to.” The mechanism hasn’t disappeared entirely. It’s gotten quieter, and quieter is not nothing. Quieter is most of what recovery actually looks like, at least at first.

If you recognize yourself in any part of this, in the parking garage, in the ledger that never balances, in the friend’s face you already know how to predict, I want you to know that the pull you’ve been fighting has a name and a mechanism, and neither of those things makes you weak for having been caught in it. You’re allowed to still be figuring this out. Understanding the wiring doesn’t mean the untangling happens overnight, and it doesn’t need to, for the untangling to be real and already underway.

Warmly, Annie.

Related Reading

Schultz, Wolfram. “Dopamine Reward Prediction Error Signaling.” Nature Reviews Neuroscience, various years. https://en.wikipedia.org/wiki/Wolfram_Schultz.

Tye, Kay M. “Neural Circuit Motifs Underlying Emotion and Motivation.” Salk Institute for Biological Sciences. https://en.wikipedia.org/wiki/Kay_Tye.

Robbins, Trevor W. “Neurobehavioural Mechanisms of Reward and Motivation.” University of Cambridge, Department of Psychology. https://en.wikipedia.org/wiki/Trevor_Robbins.

Shire, Warsan. “For Women Who Are ‘Difficult’ to Love.” Teaching My Mother How to Give Birth, 2011.

About the Author

Annie Wright, LMFT

LMFT · Relational Trauma Specialist · W.W. Norton Author

Helping driven women finally feel as good as their resume looks.

Annie Wright is a licensed psychotherapist (LMFT #95719) and trauma-informed executive coach with over 15,000 clinical hours. She works with driven women, including Silicon Valley leaders, physicians, and entrepreneurs, in repairing the psychological foundations beneath their impressive lives. Annie is the founder and former CEO of Evergreen Counseling, a multimillion-dollar trauma-informed therapy center she built, scaled, and successfully exited. She is licensed across multiple jurisdictions, from California to Maine, and brings the same rigor to her writing that she brings to every session in her practice. A regular contributor to Psychology Today, her expert commentary has appeared in Forbes, Business Insider, Inc., NBC, and The Information. She is currently writing her first book with W.W. Norton.

Work With Annie

Medical Disclaimer