Why Relationships Break Down: Insights from Psychology

In 1972, John Gottman gave a young couple 15 minutes in a room equipped with cameras and physiological sensors. He watched them argue about money. The argument itself was unremarkable - the kind of disagreement that plays out in kitchens and living rooms every day across the world.

But as Gottman and his collaborator Robert Levenson analyzed the recordings, they found something extraordinary: the patterns were consistent.

The way the partners held their faces, the microexpressions that flashed across their features, the precise sequence of claim and counter-claim, the moments when one or the other went physiologically rigid - these patterns recurred with enough regularity across enough couples.

Gottman would eventually claim something almost outrageously confident for a social scientist: he could predict whether a couple would divorce with approximately 90 percent accuracy from a 15-minute conflict discussion.^[1]

That claim has been disputed, replicated, refined, and mythologized in roughly equal measure over the half-century since.

But the core finding - that there are identifiable behavioral signatures of relationship distress that predict dissolution, and that they are measurable from observation - has held up across multiple independent studies and methods.

What Gottman's research did was transform the psychology of relationships from a domain of folk wisdom and therapeutic intuition into something resembling a predictive science.

Understanding why relationships fail requires moving between several levels of analysis: the behavioral patterns that Gottman's cameras captured, the attachment systems that Bowlby and his successors described, the commitment architecture that Rusbult mapped in her Investment Model, and the fundamental question of what partners actually need from each other in order for a relationship to flourish over time.

"In a good relationship, people very often feel very positive about even the irritating things about their partner." - John Gottman, The Seven Principles for Making Marriage Work (1999)

Key Definitions

The Four Horsemen: Gottman's term for four communication patterns that predict relationship dissolution - criticism, contempt, defensiveness, and stonewalling.

Contempt: The expression of moral superiority and disgust toward a partner, conveyed through insults, eye-rolling, sneering, and mockery. The single strongest behavioral predictor of divorce in Gottman's research.

Physiological flooding: A state of physiological arousal (heart rate above approximately 100 bpm) during conflict in which capacity for complex social-emotional processing collapses, producing reactive rather than thoughtful responses.

Attachment style: A pattern of regulating attachment needs and responding to intimacy and separation, derived from early caregiving experiences. Three primary adult patterns: secure, anxious, and avoidant.

Investment Model: Rusbult's framework proposing that relationship commitment is determined by satisfaction, investment size, and quality of alternatives.

Partner responsiveness: The degree to which a partner is perceived to understand, validate, and care about one's experience. The most consistently supported predictor of relationship quality in empirical research.

Bids for connection: Gottman's term for everyday small reaching-out moments - the micro-invitations to emotional contact that partners either turn toward, turn away from, or turn against.

Positive sentiment override: The tendency of satisfied couples to interpret ambiguous acts charitably; distressed couples interpret the same acts negatively.

The Love Lab and the Science of Conflict

John Gottman started working with Robert Levenson at Indiana University in the early 1970s and kept at it for decades, eventually at the University of Washington, where his Relationship Research Institute picked up the informal name everyone still uses: the "Love Lab."

What made the work unusual for psychology at the time was the refusal to take couples at their word. Instead of self-report questionnaires, Gottman and Levenson brought couples into a laboratory and watched them, coding facial expressions, speech patterns, and body posture alongside physiological readings, heart rate and skin conductance and blood velocity.

Doing that systematically required a system. The Specific Affect Coding System (SPAFF) broke conflict down into discrete affective expressions: contempt, disgust, belligerence, domineering behavior, defensiveness, stonewalling, humor, affection, interest. Trained coders could identify each one reliably and count how often it appeared and in what order.

Then came the part that made it predictive. Gottman and Levenson recruited couples, recorded them arguing in the lab, and followed them for years afterward to see who was still together. Their 1992 paper showed that interaction patterns coded during those lab conflicts predicted divorce status many years later, and the power came not from any single variable but from the pattern, specifically the presence and sequencing of the Four Horsemen.

That result landed against a dominant clinical assumption, which held that relationships are too complex, too context-dependent, and too changeable for reliable behavioral prediction. Gottman's data said otherwise.

The Four Horsemen

Gottman's Four Horsemen are not simply four bad communication habits. They represent a progressively more destructive gradient of relational damage, from the manageable (criticism) to the most corrosive (contempt and stonewalling).

Criticism involves attacking a partner's character or personality rather than making a specific complaint about a behavior.

The distinction is precise: a complaint targets a specific act ("You forgot to call, and that hurt me"), while criticism targets the person ("You're always so inconsiderate - you never think about anyone but yourself").

Criticism is widespread in distressed relationships but is not, on its own, lethal - it becomes dangerous in combination with the other horsemen.

Contempt is qualitatively different from the other three. Where criticism attacks character, contempt attacks worth. It communicates not simply that the partner has done something wrong but that they are beneath regard - stupid, inferior, disgusting, unworthy.

It is expressed through insults, name-calling, eye-rolling, sneering, and dismissive mockery. Gottman found contempt to be the single most powerful predictor of relationship dissolution in his longitudinal data, and the most difficult of the four patterns to reverse once established.

The mechanism may be partly immunological: he found that couples in contemptuous relationships had significantly higher rates of infectious illness over time, suggesting that chronic contempt-mediated stress has measurable biological effects.

Defensiveness functions as a barrier to repair. When one partner makes a complaint or bid for change, the defensive partner responds with counter-complaints, denials of responsibility, or victim posturing.

The cumulative message is that the complaining partner's concerns will never be heard or addressed, which drives the escalation toward hopelessness.

Stonewalling is the withdrawal of engagement - the blank face, the monosyllabic responses, the turned back.

Gottman found it more common in men and linked it directly to physiological flooding: when autonomic arousal during conflict exceeds a threshold of approximately 100 beats per minute, people lose access to the neural circuits that support nuanced social-emotional processing.

They cannot think clearly under these conditions. Stonewalling is not, therefore, simply a choice to be cruel - it is often a dysregulated response to intolerable physiological activation.

The treatment implication is significant: structured breaks of at least 20 minutes, allowing physiological de-escalation before returning to the conversation, are more effective than simply urging people to stay engaged when flooded.

What the Predictive Studies Actually Found

Several validation studies have tested Gottman's prediction claims, and the picture needs more nuance than the headline suggests.

Buehlman, Gottman, and Katz looked at oral histories in 1992, meaning the way couples narrate their shared story together, and found that the affective dimensions of those narratives, fondness and contempt and expansiveness versus disappointment, predicted dissolution over three years with substantial accuracy. Carrere and Gottman went shorter in 1999: patterns coded from just three minutes of newlywed conflict discussion predicted divorce over the following six years.

Kim Buehlman and colleagues then separated content from affect in relationship narratives and found that negativity toward the partner and glorifying the relationship's past each contributed to prediction independently.

So what about the "90% accuracy" figure in popular accounts? It is a simplification, flattening findings across multiple studies that used different measurement windows and outcome periods. Accuracy rates in individual studies vary considerably, and the predictions describe group-level probabilities rather than certainties about any particular couple.

What the body of research does robustly support is narrower and more useful: the Four Horsemen are valid predictors of distress and dissolution, contempt is the most powerful of the four, and physiological arousal during conflict is the mechanism carrying the effect.

Attachment Theory in Adult Relationships

Attachment theory started somewhere else entirely. Bowlby built it to explain infant development, arguing that a child's bond with a primary caregiver creates an "internal working model" of relationships, a template for whether other people can be trusted and depended upon.

Cindy Hazan and Philip Shaver made the leap to adult romance in a 1987 paper, proposing that the same attachment behavioral system governing an infant's proximity-seeking switches on in adults too, under threat, illness, and separation. Their initial typology sorted adults into three patterns:

How stable are those patterns over a life? Chris Fraley's 2002 meta-analysis traced them from infancy into adulthood and found moderate stability with significant contextual variation. Across a large number of studies, attachment patterns predict relationship satisfaction, commitment, and how people behave in conflict.

The finding that matters most clinically concerns the anxious-avoidant pairing. Put an anxiously attached person with an avoidant one and their strategies feed each other in a self-escalating spiral. The anxious partner's distress and bids for closeness trip the avoidant partner's defenses, so the avoidant partner withdraws.

The withdrawal raises the anxious partner's alarm, so the pursuit intensifies. The intensified pursuit deepens the avoidance. This demand-withdrawal cycle is among the most reliably documented patterns in the couples literature, and among the hardest to interrupt.

None of which makes attachment destiny. "Earned security," the development of secure functioning through a reliably responsive relationship, shows up in longitudinal research. And effective couples therapy aims straight at this level, helping anxious partners build distress tolerance and avoidant partners build the capacity for emotional presence.

Why People Stay: The Investment Model

Relationship science has one insight that reorganizes how you read almost every struggling couple: satisfaction is not what primarily keeps relationships together.

Caryl Rusbult worked that out across a series of studies in the early 1980s, building a three-factor model of commitment from satisfaction, investment size, and quality of alternatives. Satisfaction is the obvious one, how rewarding the relationship feels. The other two carry just as much weight.

Investment size is everything poured in and not recoverable: time, emotional energy, shared friends and family, children, property, an identity as a couple. Quality of alternatives is how good the options outside look.

Which yields a genuinely counterintuitive prediction. People can stay fiercely committed to relationships they find deeply unsatisfying, provided they have invested heavily and believe their alternatives are poor. Brian Le and Christopher Agnew tested this in a 2003 meta-analysis across 52 studies and found extraordinary support: the model's components together explained an enormous share of the variance in commitment, and commitment in turn predicted who actually stayed with high reliability.

Now apply it to the question people ask about abusive relationships, "why does she stay?" The model gives an answer that is not psychological pathology but rational response to constrained choice. Years of shared life constitute massive investment. The abuser's isolation tactics have systematically removed alternatives, social and financial both. And leaving carries real physical risk.

None of that excuses abuse. What it does is explain the decision calculus that traps people inside it, and that explanation has shaped interventions which address the structural barriers to leaving rather than simply telling people to leave.

Positive Sentiment Override and the 5:1 Ratio

Gottman's most quoted finding is about arithmetic. In stable, satisfying relationships he measured the ratio of positive to negative interactions, during conflict and in ordinary life, at roughly 5:1. For every critical remark or dismissal or flash of irritation, about five moments of affection, humor, genuine interest, appreciation, or warmth.

That is not a claim that happy couples do not fight. Gottman identified three types of stable couple and they fight very differently: validating couples run process-focused, emotionally supportive conversations; volatile couples argue passionately and expressively at high energy, with high warmth to match; conflict-avoiding couples sidestep direct confrontation and emphasize what is working. All three held together, as long as the positive-to-negative ratio stayed above about 5:1.

What separated stable couples from dissolving ones was never whether they had conflict. It was whether positive connection outweighed negative interaction across the whole texture of the relationship.

The mechanism Gottman proposes for why the ratio matters is "positive sentiment override." Satisfied couples build up a bank of good feeling, and that balance makes them read ambiguous moments charitably: he was probably just tired. In distressed couples, "negative sentiment override" runs the identical moment through a darker filter: he's always doing this to undermine me. The interpretation drives the response, and the responses accumulate.

One caution belongs here, because a neighbouring claim did not survive scrutiny. The "Losada ratio," which held that teams need a precise 2.9013 positive-to-negative ratio to flourish, has been substantially discredited; Fredrickson and Losada's 2005 paper was partially retracted in 2013 after Nick Brown and colleagues showed the mathematical modeling was invalid. Gottman's 5:1 rests on something different, direct behavioral observation and longitudinal outcomes rather than mathematical modeling.

Partner Responsiveness: What Actually Predicts Success

Eli Finkel and colleagues went through decades of relationship science for a sweeping 2015 review titled "The Suffocation of Marriage," and one variable came out ahead of everything else across studies and methods: partner responsiveness.

It is a more demanding thing than it sounds, and specifically not the same as agreement or approval. Feeling genuinely understood, validated, and cared for requires three separate deliveries from a partner. They have to demonstrate actual understanding of what you are going through. They have to communicate that your experience is valid. And they have to show that your wellbeing matters to them.

Harry Reis and Phillip Shaver mapped the causal chain in their 1988 intimacy model: self-disclosure produces partner responsiveness, which produces felt intimacy. Miss any link and the chain breaks. A partner who listens but shows no real understanding gives an incomplete response. So does one who understands perfectly but never communicates care. Neither generates intimacy.

Gottman turned all of this into something countable with "bids for connection." A bid is any reaching-out moment: a question, a comment about something noticed, a touch, a piece of news, a flash of vulnerability. Three things can happen to it. The partner turns toward it and engages, turns away from it with silence or a minimal reply, or turns against it with dismissal or criticism.

Then he counted. Couples who stayed together turned toward each other's bids roughly 86% of the time. Couples who divorced turned toward about 33%. Those missed bids are not individually dramatic, which is exactly the problem: they accumulate into a gradual erosion of connection that leaves conflicts progressively more dangerous and progressively harder to recover from.

Couples Therapy: Evidence and Limits

The evidence here is meaningfully more encouraging than what individual therapy shows for personality disorders, and three approaches account for most of it.

Emotionally Focused Therapy (EFT), developed by Sue Johnson and Les Greenberg, goes straight at attachment-level dynamics and carries the most robust evidence base, with meta-analyses finding clinically significant improvement in roughly 70-73% of couples and gains that tend to hold at follow-up. The Gottman Method works the behavioral layer instead, targeting the Four Horsemen, building emotional friendship and positive connection, and teaching physiological self-soothing for moments of flooding.

Andrew Christensen's Integrative Behavioral Couples Therapy adds something the other two do not: an acceptance component, helping partners live with each other's enduring differences rather than demanding change, which has shown better long-term outcomes than purely change-focused work.

What no modality overcomes is a short list of conditions that predict poor outcomes regardless of method. Waiting too long is the most common, and the numbers are bleak: the average couple waits six years after serious problems surface before seeking therapy, by which point contempt and stonewalling are usually entrenched. Chronic active contempt is another.

So is individual-level pathology left unaddressed alongside the couples work. And domestic violence requires specialized protocols rather than standard couples therapy.

See also: Attachment Theory Explained , What Causes Anxiety

Sources & Further Reading

1. Gottman, J. M., & Levenson, R. W. (1992). Marital processes predictive of later dissolution: Behavior, physiology, and health. Journal of Personality and Social Psychology, 63(2), 221-233. DOI: 10.1037/0022-3514.63.2.221
2. Hazan, C., & Shaver, P. (1987). Romantic love conceptualized as an attachment process. Journal of Personality and Social Psychology, 52(3), 511-524. DOI: 10.1037/0022-3514.52.3.511
3. Rusbult, C. E. (1983). A longitudinal test of the investment model: The development (and deterioration) of satisfaction and commitment in heterosexual involvements. Journal of Personality and Social Psychology, 45(1), 101-117. DOI: 10.1037/0022-3514.45.1.101
4. Le, B., & Agnew, C. R. (2003). Commitment and its theorized determinants: A meta-analysis of the investment model. Personal Relationships, 10(1), 37-57. DOI: 10.1111/1475-6811.00035
5. Fraley, R. C. (2002). Attachment stability from infancy to adulthood: Meta-analysis and dynamic modeling of developmental mechanisms. Personality and Social Psychology Review, 6(2), 123-151. DOI: 10.1207/S15327957PSPR0602_03
6. Buehlman, K. T., Gottman, J. M., & Katz, L. F. (1992). How a couple views their past predicts their future: Predicting divorce from an oral history interview. Journal of Family Psychology, 5(3-4), 295-318. DOI: 10.1037/0893-3200.5.3-4.295

Further Reading

• Amato, P. R., & Previti, D. (2003). People's reasons for divorcing: Gender, social class, the life course, and adjustment. Journal of Family Issues, 24(5), 602-626. DOI: 10.1177/0192513X03254231
• Finkel, E. J., Cheung, E. O., Emery, L. F., Carswell, K. L., & Larson, G. M. (2015). The suffocation of marriage: Climbing Mount Maslow without enough oxygen. Psychological Inquiry, 25(1), 1-41. DOI: 10.1080/1047840X.2014.966154