Why We Can Stay Attached to People Who Hurt Us: What the Research Actually Shows

Evidence at a glance

Some of the most confusing relationship experiences contain two truths at once: someone has hurt you, and you still feel strongly connected to them. A change in feeling may lag far behind a change in judgment. You may remember tenderness accurately, see harmful behavior accurately, and still miss the person. Neither the emotional attachment nor the confusion determines whether the relationship is healthy. The research offers several plausible, partly overlapping mechanisms, but it cannot read a person's private motives or forecast the future of any one couple. [1,3–5]

This article is a critical narrative synthesis of existing studies and meta-analyses. It is not a new quantitative meta-analysis, a prospective test of treatment, or a validated diagnostic instrument. In particular, research about intimate partner violence (IPV) is not interchangeable with research about ordinary disagreements, emotionally unsatisfying relationships, inconsistent texting, or mutually frustrating dynamics. Equating them would inflate both certainty and risk. [3–6]

First distinguish distress, conflict, and coercive abuse

Someone may feel hurt by inconsistency, rejection, unresolved incompatibilities, betrayal, broken promises or different needs without there being evidence of coercive control. Relationships can be painful without involving systematic domination. Conversely, a relationship need not involve visible physical assault to be abusive. The US Department of Justice describes domestic violence in terms of behaviors used to gain or maintain power and control, including psychological, emotional, sexual, economic and technology-facilitated forms. [6]

Ask about patterns and freedom: Is someone controlling where the other person goes, who they can contact, or their access to money? Are there threats, intimidation, surveillance, sexual coercion, violence, or fear of retaliation? Does the person feel able to disagree or leave without punishment? These questions concern observable behavior rather than a purported psychological label. A person with borderline personality disorder, an insecure attachment pattern, PTSD or no diagnosis should not be presumed abusive on that basis. Nor does a diagnostic label excuse abuse. [5,6]

This distinction matters because the scientific explanations below should never be reverse-engineered into accusations: I feel attached after being hurt; therefore my partner is deliberately manipulating me. The premise does not establish the conclusion.

What is traumatic bonding—and what has actually been tested?

The traumatic bonding hypothesis holds that certain conditions, especially pronounced power inequalities and a pattern of intermittent mistreatment, may help explain why an emotional tie persists after an abusive relationship. Popular descriptions often extend the term to any intense, unstable romance. The narrower research definition is more specific. [1,2]

A foundational 1993 study by Dutton and Painter interviewed and assessed 75 women who had recently left abusive relationships, then followed them for approximately six months. It found that measures of intermittent abuse and power differences were associated with post-separation attachment. The original report noted attachment had fallen by about 27% over the interval, while the measured relationship variables accounted for substantial variance in the follow-up attachment measure. These are findings within a selected, relatively small historical sample—not a universal formula. [1]

There are major limitations. People were not randomized to abuse or non-abuse; such an experiment would be unethical. Retrospective accounts can be affected by recall and selection, and a sample of women who had already left abusive relationships cannot be generalized automatically to every gender, age, or kind of relational distress. This study cannot prove that alternating kindness and harm caused attachment, or that every survivor experiences the same dynamic. Its useful contribution is hypothesis support, not diagnostic certainty. [1,2]

Why alternating kindness and hurt might be especially confusing

Repeated reconciliation may preserve hope: a very painful incident is followed by relief, an apology, affection, or a believable promise of change. It becomes tempting to evaluate the relationship by the latest positive event rather than by the full record of behavior. In coercive relationships, intermittency and power inequality are specifically relevant to traumatic-bonding theory. [1]

In less severe situations, the same subjective pattern—anticipation, disappointment, reunion—could involve learned expectations, loneliness, uncertainty, attachment anxiety, shared routines, or simple love for a familiar person. But analogies to laboratory reward schedules do not establish that a partner intentionally designed a behavioral conditioning program. Likewise, claims that a specific dopamine surge or an oxytocin imbalance proves a trauma bond are not supported by the studies summarized here. [1,3–5]

A practical test is not how intense the reunion feels. It is whether behavior changes durably, whether boundaries are respected between reconciliations, and whether the relationship expands rather than narrows a person's agency.

Attachment insecurity: what the meta-analyses say

Adult attachment measures commonly describe anxiety about abandonment or rejection and discomfort with closeness or dependency. These are dimensional measures, not immutable romantic personality types. A person can vary by context and over time; a survey score cannot diagnose the cause of conflict. [3–5]

Cataudella and colleagues' victimization meta-analysis screened 4,343 records and included 34 studies, totaling 1,271 people under its inclusion criteria. Both attachment anxiety and avoidance showed statistically significant associations with IPV victimization across the examined forms, with reported correlations varying by violence category, roughly r = 0.18–0.30. This is a population-level association, not a prediction for the next person someone dates. [3]

A distinct meta-analysis by Knox and colleagues focused on perpetration of partner maltreatment, an outcome not identical to victimization or severe violence. Average weighted correlations were small: approximately r = 0.19 for attachment anxiety and r = 0.14 for avoidance. There was considerable heterogeneity across underlying studies. It would therefore be misleading to say attachment insecurity makes someone abusive or that securely attached people are immune to harming others. [4]

In 2025, Wong and Denson systematically reviewed 63 eligible studies investigating possible mediators of associations between attachment insecurity and IPV. Their models examined jealousy or distrust, dissatisfaction, control, destructive communication, emotion dysregulation and other candidate pathways. Some indirect associations were significant, but statistical mediation in predominantly observational literature does not demonstrate the temporal causal sequence. [5]

Taken together, the reviews make attachment relevant to questions worth discussing and assessing. They do not reduce relationship harm to one person's anxious style, let alone shift responsibility for someone else's abusive actions onto a victim.

Several mechanisms may operate at the same time

Emotional familiarity. Strong attachment can remain because a partner has been important over years, shared milestones, family rituals and ordinary days. The mind does not simply erase those experiences after a rupture. This explanation is plausible but cannot be assigned a specific measured effect size from the trauma-bonding studies. [1,3]

Hope and intermittent repair. A credible apology or warm period can make change seem possible. A relationship that alternates harm and repair may be difficult to evaluate because recent relief stands out. The existence of some good interactions does not establish that the harmful pattern has ended. [1,7]

Fear and constrained choices. Control of money, housing, documents, children, transportation, or social contact can make departure materially difficult. Threats and surveillance can make it unsafe. These are barriers, not signs of weak will or insufficient insight. Government guidance explicitly considers such constraints when discussing safety plans. [6,7]

Attachment anxiety and abandonment fear. For some people, separation or uncertainty may trigger intense distress and reassurance seeking; the meta-analytic evidence shows associations with broader relationship outcomes but does not confirm any individual's underlying mechanism. [3–5]

Grief, guilt and loyalty. Caring about someone who behaves harmfully is not logically contradictory. Someone can recognize a pattern needs to stop while hoping the other person gets help. Supporting another person and remaining in a particular relationship are different decisions.

Practical and social ties. Shared children, employment, immigration concerns, disability, family pressure, religious or cultural factors and fear of public judgment can all constrain choices. Those realities deserve specific help rather than a one-word psychological explanation. [7]

The evidence does not assign a reliable percentage to how often each reason is decisive. An article pretending to rank these motives precisely would create false certainty.

Trauma bonding, limerence, and relationship OCD are not synonyms

Limerence is used for intense romantic preoccupation, intrusive thoughts about reciprocation and fluctuating emotional responses to perceived romantic signals. Someone could experience limerence without violence, coercion or an abuse history. The term should not be used to label a person's distress a trauma bond by default. See the limerence evidence review.

Traumatic bonding in the original research concerns an attachment pattern described in abusive relationships, especially under intermittent mistreatment and power asymmetry. It is a research hypothesis, not a stand-alone DSM diagnosis or a clinical test. [1,2]

Relationship obsessive-compulsive disorder involves obsessive uncertainty and compulsive responses that may require a different clinical assessment. Persistent rumination alone is not enough to identify OCD. Distinguishing these concepts matters for helpful treatment rather than attaching a dramatic label to every difficult love story.

Does staying mean a person wants the mistreatment?

No. Attachment, consent to abuse, access to options, and responsibility for violence are separate questions. A person may have stayed because of fear, genuine caring, threats, hopes for change, financial dependence, children, or a shifting assessment of what was possible. A return to the relationship does not invalidate a report of harm. The research on attachment and victimization describes associations, not moral responsibility for another person's decisions. [3,6,7]

The opposite oversimplification is also unhelpful: not every person who returns is in a coercive relationship. We need evidence about what actually happened, whether boundaries are respected, and whether anyone feels unsafe or controlled. A broad psychological concept cannot substitute for listening and assessment.

A careful way to evaluate a difficult relationship

One useful approach is to keep a behavior-based timeline rather than an intensity-based verdict. Over weeks or months, record specific events, commitments, boundaries, repairs and recurring patterns—only if doing so is safe and private. What changed after a serious conversation? Did the same conduct recur? Does accountability involve behavior, not only apologies? Can you express needs without threats, intimidation or retaliation?

Next, distinguish what can be negotiated from what requires protection. Scheduling mismatches, unclear expectations and incompatible preferences may be addressed through mutual discussion and boundaries. Threats, coercion, surveillance or physical/sexual harm call for a safety-centered response. Do not frame them as symmetrical communication problems. [6–8]

Finally, notice whether your world is becoming larger or smaller. A safe relationship usually permits friends, interests, independent decisions, lawful access to one's own money and contact with support. This is a practical lens, not a validated clinical score, but it can help shift attention from the immediate relief of reunion to lived autonomy.

What supports change—and what the research cannot promise

Support might include confidential conversations with trusted people, trauma-informed professional counseling, practical financial/legal advice, documentation where lawful and safe, and specialist advocacy for coercion or violence. For some, attachment-focused or emotion-regulation work may help examine expectations and regain choice. Yet this article cannot claim any particular therapy specifically cures a "trauma bond," because the cited studies do not directly establish that treatment effect. [1,3–5,7]

Avoid promising that no contact is always appropriate. People may share children, housing, legal obligations or safety concerns. If co-parenting or necessary communication continues, supported limits and safer channels may be more realistic. Unplanned abrupt confrontation or a breakup announcement can increase danger in some abusive situations; the appropriate response is individualized safety planning rather than a universal script. [7,8]

Similarly, couples therapy is not automatically safe when intimidation and coercive control are present. Safety and specialized assessment take precedence over treating both parties as equally able to negotiate. In noncoercive relationships, collaborative communication may be useful, but persistent harm still merits independent judgment. [6–8]

Recent 2024–2026 research on why safety and choice can be constrained

Stulz and colleagues' 2024 integrative review of 12 studies identified recurring barriers experienced by women escaping domestic violence: losses from leaving, ongoing required contact with a former partner, financial insecurity, access to safe housing and services, and responsibility for children. Support, collaboration and providers who listen respectfully were recurring facilitators. These findings describe qualitative and mixed-method experiences, not a randomized trial or a universal psychological reason people stay. [9]

In a 2026 self-report survey of 926 Israeli women, Mehlhausen-Hassoen and Winstok examined positive reasons for staying alongside negative constraints labeled fear of failure, fear of terror and captivity. Recently battered women reported especially high negative barriers. The findings undermine a simplistic either/or claim that people stay entirely from affection or entirely from external force. Because the survey was cross-sectional and culturally specific, it cannot prove why any individual stayed or predict future safety. [10]

Together, these studies reinforce an essential point: psychological theories of attachment and intermittent reinforcement cannot replace resources such as housing, childcare, financial stability and confidential specialist advice. A request to “just leave” without addressing such constraints can be unrealistic or unsafe. When danger or coercion exists, safety planning should be individualized rather than determined from someone's attachment label. [6–10]

What if there are children, financial constraints, or threats?

These are not side notes. They are major parts of the decision environment. The US Office on Women's Health describes planning for documents, housing, medications, phones, trusted contacts, children and confidential support. An advocate can help think through what is safest without requiring a decision to leave during the first conversation. [7]

Digital safety also matters. Shared accounts, location sharing, device access and browsing records can expose searches for assistance. Use a safer device or ask an advocate about confidentiality if monitoring is a concern. Preserve records only where doing so will not increase risk. For urgent immediate danger in the US, call 911. [7,8]

In the US, the National Domestic Violence Hotline provides confidential support at 800-799-SAFE (7233), by text START to 88788, and at thehotline.org. Contacting an advocate is appropriate even when someone is unsure whether their situation qualifies as abuse. [8]

What this evidence cannot tell us

It cannot tell us that a particular partner has borderline personality disorder, narcissism, psychopathy or another diagnosis. It cannot determine intent from a message, social-media activity, a period of silence, an apology, or one emotional argument. It cannot show that insecure attachment inevitably attracts harmful partners. It cannot show that a person who feels attached is incapable of making choices.

It cannot establish that every intermittent relationship is abusive, or that every abusive relationship creates a trauma bond. The foundational trauma-bonding study was small; the newer meta-analyses address broader and partly different constructs. Correlations and statistical mediation are not experiments in human relationships. [1–5]

Nor can it establish that a particular person must leave or remain. Ethical support makes safety and choices clearer; it does not replace the person's decisions.

Evidence limitations and future research

The 1993 post-separation study is unusually relevant to the question in the title, but it is a small historical sample selected from people already leaving abusive partners. It cannot estimate prevalence, isolate all possible mechanisms or demonstrate which intervention improves outcomes. [1]

The 2023–25 meta-analytic literature has the advantage of aggregating many studies, but it also mixes different definitions of attachment, maltreatment and victimization, often relies on self-report, and may suffer residual confounding or uncertain timing. Significant pooled correlations do not settle why individual people stay or return. A correlation from a perpetration study should not be represented as a victimization risk, and vice versa. [3–5]

Future prospective studies could follow diverse people over time, separately measure coercive control, affection, material constraints and attachment, and assess survivor-defined safety and autonomy without pathologizing participants. Intervention research should evaluate benefits and harms of options, including for people who remain in contact because of children or practical necessity. It would be misleading to call the current literature a definitive mechanistic account.

Bottom line

You can understand that a relationship hurts you and still miss the person. The strongest honest explanation is plural: existing attachment, hope after intermittent repair, learned expectations, fear, circumstances and potential coercive control may each matter. The 1993 traumatic-bonding study offers limited observational support for one specific proposed pattern, while modern meta-analyses connect attachment dimensions with broad relationship harm at a population level. None can diagnose an individual from an emotion or prove that longing excuses mistreatment. [1–5]

If the relationship is simply incompatible or repeatedly disappointing, boundaries, clarity and independent support may help. If it involves fear, coercion or danger, use safety-centered advice and specialized advocacy. Either way, judging patterns of behavior and room for genuine choice is more useful than trying to prove a label. [6–8]

Related THS research: Limerence and obsessive love, Limerence and relationships involving BPD, Loving a partner with BPD, BPD vs complex PTSD, and the mental health guides.