Why it's so hard to leave: intermittent reinforcement, sunk costs and the 'trauma bond'

People stay in relationships that hurt them for reasons that are rational given their situation, predictable given how attachment and reward work, and invisible from outside. Three research traditions explain most of it:

1. Intermittent reinforcement — warmth that comes unpredictably keeps you working for it.
2. The investment model — what you have put in, and what you believe your alternatives are, hold you in place regardless of how satisfied you feel.
3. Traumatic bonding — attachment grows stronger, not weaker, under power imbalance and alternating kindness and mistreatment.

To these, add the practical realities that no psychological theory should be allowed to obscure: money, a home, children, legal status, physical safety.

In 1939 Lloyd Humphreys published a small experiment on eyelid conditioning: participants who had been reinforced only half the time kept responding far longer, once reinforcement stopped, than those who had been reinforced every time. Ferster and Skinner's 1957 Schedules of Reinforcement mapped the phenomenon systematically. The partial reinforcement extinction effect is now one of the best-replicated findings in behavioural psychology: behaviour rewarded unpredictably is more persistent and far harder to extinguish than behaviour rewarded reliably.

Translate that into a relationship. A partner who is warm every day teaches you that warmth is available; if it stops, you notice quickly and respond. A partner who is warm unpredictably — tender on Tuesday, cold on Wednesday, wonderful again on Saturday for no reason you can identify — teaches you to keep trying, because the next reward may be one more effort away. The uncertainty is not a bug in the attachment; it is the engine.

This is why the "good times" in a difficult relationship are not evidence against leaving. They are the mechanism that makes leaving hard.

Caryl Rusbult's investment model proposes that commitment to a relationship depends on three things:

• Satisfaction — how good it feels.
• Quality of alternatives — what you believe you could have instead, including being alone.
• Investment size — what you have put in that you would lose: time, shared history, money, a home, friends, children, identity.

The model's uncomfortable finding is that satisfaction is only one of the three. People stay in unsatisfying relationships when alternatives look poor and investments are large, and that is not irrational: they are weighing real losses.

In 1995 Rusbult and John Martz applied the model to women arriving at a shelter for domestic abuse. Commitment to the relationship — and the likelihood of returning to it — was higher where economic alternatives were poorer and investments (marriage, shared resources) were larger. Michael Strube's 1988 review of the literature on leaving abusive relationships reached compatible conclusions: employment, relationship length, previous separations and economic dependence were among the strongest predictors of whether someone left.

In 1993 Donald Dutton and Susan Painter tested a theory they called traumatic bonding: that strong emotional attachment forms in relationships marked by a power imbalance and intermittent good–bad treatment. Their study of women leaving abusive relationships found that attachment to the partner was stronger where the abuse had been more intermittent — the same partial-reinforcement logic, now applied to fear and relief rather than warmth and coldness.

Two features of the research deserve emphasis.

First, it concerns abusive relationships with a clear power asymmetry. The now-popular phrase "trauma bond" has drifted a long way from this, and is often applied to any painful attachment. The useful idea survives the drift — unpredictable treatment makes attachment stickier — but the research does not support describing every difficult breakup as a trauma bond.

Second, traumatic bonding is a dynamic, not a diagnosis. It describes what a situation does to a person, not a fault in the person. It is also compatible with the 2023 qualitative work on gaslighting by Klein and colleagues, which found that gaslighting typically occurred within a cycle of affectionate and abusive behaviour. The cycle is the point.

Several further factors, each with its own literature, add to the pull:

• Attachment anxiety. People higher in attachment anxiety experience separation as acute threat and are more vulnerable to the hot-and-cold cycle.
• Isolation. Many controlling partners narrow their partner's world — fewer friends, less family contact, less money of their own — which shrinks the perceived alternatives in the investment model. See The Controlling Partner.
• Eroded self-trust. Where a partner has repeatedly disputed your memory and perception (see gaslighting), the confidence needed to act on your own judgement has been undermined.
• Hope and the "real" partner. The warm phase is experienced as the real person and the cold phase as an aberration — a reading the intermittent schedule reinforces.
• Shame. Having told people it was fine, having gone back before, having stayed this long — each becomes a reason not to admit the situation now.
• Fear. For some, leaving is the most dangerous moment. That is not a psychological mechanism; it is a safety fact, and it should be planned for with specialist help.

The research on leaving points to a few things that make a real difference.

1. Treat leaving as a process, not a test of will. Strube's review and later work consistently find that people typically make several attempts before leaving for good. Each attempt is not a failure; it is a rehearsal that changes the alternatives and the investments.

2. Change the alternatives, concretely. The investment model predicts that improving your alternatives — a job, your own account, a friend's spare room, legal advice, knowing what support exists — changes commitment even before anything else does. This is why specialist services spend so much time on practical planning.

3. Rebuild the outside world. Every friendship resumed and every independent activity reclaimed widens the alternatives and loosens the isolation on which the dynamic depends.

4. Keep a record. Intermittent reinforcement works partly through memory: the warm weekend overwrites the cold week. A private, dated record of what actually happened restores the pattern to view.

5. Name the mechanism. "I'm on a variable reinforcement schedule" sounds clinical, but many people find that recognising the mechanism takes some of its power away. The pull does not vanish; it stops feeling like evidence.

6. Get specialist help if there is control or fear. Domestic abuse services exist precisely because leaving a controlling partner is both hard and, at times, dangerous. They are confidential and they do not require you to have decided anything.

• Love bombing vs genuine affection — the idealisation phase that starts the cycle.
• The Controlling Partner — coercive control as a pattern.
• Gaslighting or an ordinary disagreement? — how eroded self-trust keeps people in place.
• Anxious attachment — why separation feels like danger.