Divorce is best predicted, at the population level, by combinations of relationship-specific processes, accumulated marital problems, personal vulnerabilities, and social circumstances. Longitudinal studies repeatedly implicate lower marital satisfaction, patterns of destructive conflict, serious breaches of trust, and strains arising from finances and life circumstances. Characteristics such as personality, age at marriage, education, and family background sometimes add information. Their meaning depends on how couples and societies organize relationships. None supplies a scientifically reliable verdict about whether a particular marriage will end. Amato, 2010; Karney and Bradbury, 1995.
The crucial distinction is between explaining differences in divorce rates among groups and forecasting the future of a particular couple. A statistical association can be real even when the couples sharing a risk factor have very different futures. Moreover, a stable marriage and a healthy marriage are separate outcomes: a couple can remain legally married amid serious unhappiness or danger, while another may divorce after a thoughtful and protective decision. The most informative science examines marital trajectories, partner interactions, resources, constraints, and changing conditions together. Previti and Amato, 2003; Bradbury, Fincham, and Beach, 2000.
What Does It Mean to Predict Divorce?
A legal outcome, a relationship process, and a forecast
Divorce is the legal dissolution of a marriage under the rules of a particular jurisdiction. It usually follows a relationship process, but the date of an official decree is a distinct measurement point. Marital separation, living apart, filing a petition, discussing divorce, contemplating departure, and finally terminating a marriage are related events that may occur in different sequences. An investigation that counts separations can produce different results from one counting completed legal divorces. The distinction matters when comparing countries, historical periods, and couples who remain legally married after their intimate partnership has effectively ended.
Researchers often study divorce through prospective cohorts: they measure married people at one time and later observe which marriages legally dissolve. A predictor is any measured characteristic associated with subsequent dissolution under the study's specified conditions. A predictor can be a correlate, a precursor, a marker of another process, or part of a causal mechanism. Calling a factor predictive does not establish that changing it would prevent divorce. The classic longitudinal review by Karney and Bradbury summarized 115 studies encompassing more than 45,000 marriages and argued for combining theories of change with better follow-up designs. Karney and Bradbury, 1995.
Risk depends on a population and time horizon
The question “What predicts divorce?” needs a defined reference population. Newlyweds, people married for twenty years, remarried couples, cohabiting partners, and spouses already separated face different circumstances. An association estimated among first marriages in one country or decade may shift when marriage occurs later, divorce is less stigmatized, financial autonomy changes, or family law and social protections differ. A probability for divorce within five years also differs from a probability across an entire lifetime. Data describing the divorces that occurred in one calendar year cannot be converted into the proportion of newly formed marriages destined to end. Amato, 2010.
Individual probabilities require correctly calibrated models, rigorous assessment on people omitted from model development, and attention to missing observations. A model that separates lower-risk from higher-risk groups can still give inaccurate probabilities. Divorce is often a relatively infrequent outcome over short follow-up intervals, so a model can appear accurate by simply predicting that nearly everyone stays married. Balanced accuracy, discrimination, calibration, positive predictive value, and transportability answer different questions. They should all be considered before any numerical score is treated as meaningful in practice.
What Longitudinal Research Establishes
Why prospective studies matter
A cross-sectional comparison of currently married and divorced people cannot establish which differences preceded divorce. Some observed characteristics may be consequences of marital breakdown: depression can intensify under chronic conflict, economic security can change after separation, and social networks may reorganize when a marriage ends. Prospective studies provide temporal ordering, but can still miss early distress, unmeasured economic shocks, coercion, shared causes, and changing expectations. Stronger inference combines repeated measurement with models sensitive to the timing of events. Amato, 2010; Solomon and Jackson, 2014.
A landmark US panel followed reports of marital problems collected in 1980 and examined divorces through 1992. Infidelity, problematic spending, drinking or drug use, jealousy, moodiness, and irritating habits were among the reported problems that predicted later divorce. These findings illustrate why a marital risk profile is usually multidimensional. They also illustrate limits: a problem measure can reflect one spouse's perceptions, several problems may co-occur, and statistical mediation does not alone establish how the marriage would have changed under an intervention. Amato and Rogers, 1997.
Marital satisfaction is informative but not sufficient
Low marital satisfaction has a well-supported association with subsequent deterioration and dissolution. Its predictive significance depends on whether researchers assess starting satisfaction, trajectories, both partners' reports, intentions to leave, or changes over time. Satisfaction is an evaluation of the experienced marriage, while divorce is a later decision and legal outcome subject to constraints and opportunities. A person can feel dissatisfied without seeking divorce; another can leave a partnership whose day-to-day interactions seemed outwardly stable. Satisfaction should be interpreted alongside commitment, safety, social resources, and actual conditions for exit. Karney and Bradbury, 1995; Previti and Amato, 2003.
In a study of 464 newlywed spouses measured repeatedly over the first four years, Lavner and Bradbury identified distinct satisfaction trajectories. Some people remained highly satisfied, while smaller subgroups began at lower satisfaction or declined more sharply. These trajectories differed in later divorce rates. The substantive result is heterogeneity: average downward change in satisfaction conceals stable, improving, and persistently distressed patterns. The study does not yield a universal curve that every marriage follows, and its specific groups should not be exported as diagnoses or personalized forecast categories. Lavner and Bradbury, 2010.
Commitment, Investment, and Reasons for Staying
Satisfaction, alternatives, investments, and commitment
Interdependence and investment approaches explain why people remain in relationships through several linked judgments. Satisfaction represents how positively the present partnership is evaluated. Perceived alternatives include expectations about being single, pursuing another relationship, or living independently. Investments comprise shared resources, time, relationships, history, and commitments that might be disrupted by leaving. Commitment is an orientation toward the relationship's continuation. These dimensions need to be distinguished because the material cost of leaving and the desire to remain emotionally connected can move in different directions. Le and Agnew, 2003.
A meta-analysis of the investment model, covering 52 studies and 60 independent samples, supported associations of satisfaction, perceived alternatives, and investment with commitment. However, this framework was developed across varied close relationships. It helps explain marital persistence but does not by itself provide a divorce-specific actuarial instrument. Married spouses also face legal responsibilities, caregiving arrangements, housing markets, income interdependence, and community norms that can strongly shape the practical decision to remain or leave. Le and Agnew, 2003; Amato, 2010.
Rewards and constraints predict different kinds of stability
Previti and Amato studied respondents' explanations for remaining married and distinguished positive rewards from perceived barriers to divorce. Those reporting barriers alone were often less happy and more likely to contemplate divorce; compared with respondents reporting rewards alone, they also had greater subsequent divorce risk over a long follow-up. The distinction is theoretically important. A marriage maintained through attachment, mutual care, and shared meaning can look identical in legal status to one maintained through financial dependence, fear, or lack of housing options. Research on marital stability must avoid treating legal continuation as a surrogate measure for relational well-being. Previti and Amato, 2003.
Investment and constraints require particular caution in unsafe relationships. Control over money, immigration status, social contacts, or housing can restrict a person's capacity to leave. A predictor of continued marriage under those conditions says nothing about whether staying is desirable. Safety decisions should be informed by coercive control, threats, stalking, and access to specialized support rather than a population model of divorce risk.
Couple Communication and Conflict
What observed behavior adds
Couple researchers distinguish constructive communication, emotional responsiveness, criticism, contempt, hostility, withdrawal, escalation, and repair. Some variables are assessed through questionnaires, while others are coded during observed discussions. Observations help researchers examine interaction rather than relying solely on partners' global judgments. Yet a brief laboratory conversation is a narrow sample from a relationship that unfolds across work, parenting, illness, financial decisions, and daily stress. The credibility of any prediction therefore depends on the time interval, coding reliability, and explanatory contribution beyond information already known about the couple. Bradbury, Fincham, and Beach, 2000.
A major meta-analysis integrated 64 dyadic studies and 1,784 parameter estimates of communication and later relationship outcomes. Positive and negative communication showed small associations with subsequent relationship quality. Positive communication did not show a statistically significant aggregate association with dissolution in that analysis, whereas negative communication was associated with later dissolution with a small-to-moderate estimated effect. These findings give empirical weight to destructive interaction patterns while sharply limiting claims that a single observed argument predicts a marriage's fate. Kanter and colleagues, 2022.
Conflict frequency, conflict severity, and conflict management
Frequency of disagreement is only one dimension. Repeated hostility, humiliation, threats, jealousy, chronic criticism, escalation, and difficulty resolving disputes may carry different implications from disagreements handled with respect. Couples also differ in what they argue about, their surrounding stress, and the resources available to address problems. Some marital problems remain stable while satisfaction changes, which indicates that the mere count of problem topics cannot account for the full trajectory. In a newlywed study of 169 couples, reports of specific problems were often relatively stable across four years, even as average satisfaction declined. Lavner, Karney, and Bradbury, 2014.
Destructive conflict is often embedded in a broader system. A household experiencing precarious employment may enter discussions already exhausted by uncertainty; a couple managing disability or caregiving may have fewer opportunities to recover after conflict. Conversely, supportive exchanges and access to practical help may buffer some stress. Treating interaction style as a detached personality test ignores those environmental influences. Longitudinal work on marital adaptation considers vulnerabilities, stressful events, and adaptive processes jointly. Karney and Bradbury, 1995; Conger, Conger, and Martin, 2010.
Why the popular four-sign checklist is inadequate
Memorable communication labels can prompt discussion, but using a few behaviors as an automatic divorce forecast confuses the identification of potentially harmful dynamics with a validated individualized prediction. A useful prognostic claim would specify the observational protocol, reference sample, follow-up length, absolute probability, error rates, and independent validation. It would also account for contextual circumstances and partners' changes. The contemporary synthesis of communication findings supports modest average effects rather than exceptional certainty for an individual marriage. Kanter and colleagues, 2022.
Trust, Infidelity, Substance Use, and Serious Marital Problems
Infidelity and reciprocal deterioration
Infidelity is frequently reported when divorced people describe why their marriages ended, but retrospective reasons reveal perceptions and narratives after the event. In a national panel analysis of 208 open-ended explanations, infidelity was the most commonly mentioned reason, followed by incompatibility, drinking or drug use, and growing apart. These observations must not be treated as a population-wide ranking of causal mechanisms, because respondents were describing completed divorces rather than testing a causal intervention. Amato and Previti, 2003.
Longitudinal research offers a more nuanced account of temporal order. Previti and Amato studied extramarital sex, marital happiness, divorce proneness, and subsequent divorce over a seventeen-year period. Divorce proneness predicted later infidelity, while infidelity also predicted lower subsequent marital happiness, greater divorce proneness, and greater divorce odds. Relationship deterioration and an affair can therefore form reciprocal processes. Individual responsibility for infidelity remains with the person making that decision; an observed association with earlier relationship problems neither excuses nor proves its occurrence. Previti and Amato, 2004.
Alcohol, drugs, impulsive spending, and recurring instability
Problematic substance use and financial conflict are among the marital problems examined in prospective studies. Their association with divorce may operate through impaired behavior, broken agreements, financial pressure, impaired caregiving, conflict, or wider social adversity. A survey report that a partner “drinks too much” is still not equivalent to a formal substance-use disorder diagnosis. Similarly, spending disputes can represent chronic scarcity, impulsivity, incompatible expectations, deception, or coercive control. A carefully conducted analysis asks what was measured and whether the relationship persisted after controlling for earlier circumstances. Amato and Rogers, 1997.
Personality and Individual Differences
Why traits matter without determining destiny
Personality traits affect how people typically interpret events, regulate emotion, handle uncertainty, seek support, and respond to frustration. Among traits studied in marriage research, emotional instability or neuroticism has attracted sustained attention because it can coincide with more frequent negative affect and perceptions of strain. A nationally representative longitudinal sample of 8,206 individuals supported associations between personality traits and subsequent relationship dissolution, with satisfaction providing part of the explanatory pathway. That study tested enduring relationship dynamics and emerging distress as distinct accounts rather than assuming that traits cause divorce directly. Solomon and Jackson, 2014.
A spouse's score is not a moral diagnosis, a forecast for a particular couple, or proof of responsibility for a divorce. Observed associations can differ by measurement, relationship conditions, and the traits of both people. Personality also interacts with financial stress, mental health, social resources, and the evolving marriage. A more defensible question than “Which personality causes divorce?” is “Under which conditions do individual differences shape the relationship processes that precede dissolution?” Solomon and Jackson, 2014; Karney and Bradbury, 1995.
Attachment and psychological insecurity
Adult attachment orientations may shape expectations about care, sensitivity to rejection, response to distance, and conflict behavior. Their general psychology belongs to attachment research; in divorce prediction their relevance lies in how relationship-specific fears, avoidance, and caregiving dynamics may influence commitment, satisfaction, and behavior over time. The evidence does not justify predicting legal divorce from an “anxious partner” or “avoidant partner” label. These categories are especially vulnerable to retrospective overinterpretation after a breakup, when someone may use a label to construct an overly simple explanation.
Mental health symptoms and causal ambiguity
Depressive symptoms, anxiety, chronic stress, trauma histories, and substance-related problems can influence marital interactions and may also be affected by them. The timing matters. Distress may precede deteriorating marital quality, emerge in response to relationship conditions, or reflect broader life circumstances affecting both people. Psychiatric symptoms and a diagnosed mental disorder are different outcomes, and divorce itself is a legal event rather than a mental-health diagnosis. Research models should adjust for preexisting distress and avoid interpreting every significant symptom association as evidence of an individual disorder. Amato, 2010; Solomon and Jackson, 2014.
Money, Employment, and Social Inequality
Economic pressure is more than a household income score
Economic adversity influences marriages partly through the pressures it generates: insecure housing, debt, inability to pay bills, work instability, unmet needs, and difficult tradeoffs involving health care or childcare. The family stress model describes pathways from material hardship to economic pressure, emotional distress, and relationship processes. Conger, Conger, and Martin's review emphasized interaction between socioeconomic circumstances and family functioning, including the distinction between social selection and social causation. Lower resources can precede strain; marital disruption can also undermine resources, producing reciprocal disadvantage. Conger, Conger, and Martin, 2010.
A single income measure may conceal a great deal. Two households earning the same amount may differ in debt obligations, housing costs, dependents, health expenses, job insecurity, family assistance, and control over money. Researchers must distinguish predictable chronic pressure from a sudden job loss, and household resources from unequal access to those resources within a marriage. Treating economic disadvantage as a defect of the spouses mistakes a structural condition for a character judgment.
Education, resources, and constraints on exit
Education is often associated with marriage stability, but its role depends on economic conditions, selection into marriage, expectations, and barriers to divorce. A study of first marriages in British panel data examined why greater education among women was associated with lower divorce risk. Differences in marital satisfaction explained part of the gradient, while housing and other potential barriers to divorce were also important. Thus, stability can reflect both conditions that make a marriage attractive and conditions that make leaving more costly. A correlation between education and divorce cannot be converted into a claim that schooling itself ensures marital harmony. Boertien and Härkönen, 2018.
The same reasoning applies to work. Employment may supply resources, autonomy, security, and social support, while unpredictable hours, underemployment, and chronic exhaustion can place relationships under pressure. How employment relates to divorce depends on gender arrangements, social services, labor markets, and family policies. Findings anchored in one historical period cannot simply be transported into every contemporary marriage. Conger, Conger, and Martin, 2010.
Structural inequality and social context
Population averages frequently combine people who differ in exposure to discrimination, institutional barriers, neighborhood resources, and family policy. An example comes from research on interracial marriages: Zhang and Van Hook used longitudinal survey panels and found that apparent differences in dissolution risk shifted with adjustment for couple characteristics and comparison groups. The lesson is methodological as well as social: crude group differences can overstate the role of a group label while overlooking selection and material conditions. No race, ethnicity, nationality, or cultural identity is an individual marital prognosis. Zhang and Van Hook, 2009.
Age at Marriage, Family Background, and the Life Course
Timing, maturity, and cohort differences
Age at marriage is commonly investigated as a demographic correlate of divorce, especially in studies comparing very early unions with marriages entered later. Its meaning is conditional. Early marriage may coincide with unfinished education, limited income, major identity change, and earlier childbearing, whereas later marriages are entered by a selected population that has already made different relationship and life decisions. Later age can also be associated with independent expectations, prior relationships, and different economic risks. The observed association should be understood within a generation and setting rather than used to announce a universally ideal age for marrying. Amato, 2010.
The distinction between age, birth cohort, and marriage duration is essential. People married at age twenty in one decade faced different social conditions from people married at the same age decades later. An association observed at marriage year one may change at year fifteen, after childbirth, relocation, retirement, serious illness, or the loss of a parent. Statistical models that merge all periods can hide these effects. Good longitudinal models define when exposure and follow-up begin and consider age, duration, and calendar time separately.
Parental divorce and intergenerational processes
Having parents who divorced may correlate with later marital instability through childhood experiences, economic disruptions, exposure to conflict, attitudes toward marriage, or patterns of selection into partnerships. These pathways cannot be inferred from a person's family history alone. The experience of a parental divorce varies with prior family conflict, subsequent caregiving, resources, supportive adults, and later developmental opportunities. Treating parental marital history as inherited destiny disregards these differences. Studies examining education and marital stability also show that family-background variables can operate through the social and economic conditions of adult unions. Boertien and Härkönen, 2018; Amato, 2010.
Cohabitation, remarriage, and relationship transitions
Premarital cohabitation has been studied as a marker of relationship selection, commitment, and transitions into marriage. A conceptual distinction concerns making deliberate shared decisions about commitment versus moving into more binding arrangements without a corresponding decision. Stanley, Rhoades, and Markman proposed an inertia framework in which shared housing and constraints can increase persistence even when dedication is uncertain. This is a theoretical account rather than proof that cohabitation generally causes divorce. Associations can change across cohorts as cohabitation becomes more common, and any analysis must separate relationship history, timing of engagement, social expectations, and preexisting intentions. Stanley, Rhoades, and Markman, 2006.
Remarriage and later-life marriage involve different social and family configurations: previous children, stepfamily responsibilities, financial agreements, accumulated assets, changing caregiving needs, and earlier experiences of relationship dissolution. A risk estimate drawn from first marriages is inappropriate for those settings without recalibration. The most appropriate predictors must account for the institution and stage under study rather than present one uniform model for all married adults. Amato, 2010.
Children and the changing division of labor
Childbirth and parenting can transform time use, sleep, role expectations, household work, and the balance of paid and unpaid labor. These transitions influence the environment in which satisfaction, conflict, and shared decisions develop. Yet having a child, having several children, being childfree, or having a stepchild cannot on its own be treated as a direct cause of divorce. Cultural norms, economic support, relationship conditions before parenthood, caregiving distribution, and the child's needs can change the association. Work in marital research emphasizes the microcontext of childrearing alongside broader economic conditions. Bradbury, Fincham, and Beach, 2000.
Social Networks, Culture, and Institutions
Friends, family, and community support
Couples are embedded in social networks. Approval or opposition from relatives, available childcare, trustworthy friends, access to housing, and support during stressful events can affect both relationship experience and the feasibility of separation. But the same network can offer healthy support in one marriage and exert coercive pressure to stay in another. Social support is therefore best conceptualized according to its content and context rather than scored automatically as a force against divorce. Social norms may change the cost of divorce without resolving the relational problems that preceded it. Bradbury, Fincham, and Beach, 2000.
Religion, norms, and divorce institutions
Religious affiliation and practice, community expectations, and the perceived legitimacy of divorce can correlate with marriage patterns. Their effects are shaped by selection into marriage, shared beliefs, norms about separation, and institutional environments. They may increase relationship investment or create pressure to remain legally married. The distinction between stable marriage and freely chosen, safe marriage remains crucial. Legal rules also define the event being measured: comparisons across jurisdictions must account for differing dissolution procedures, access to divorce, property arrangements, and custody frameworks without presenting any one legal system as universal. Amato, 2010; Previti and Amato, 2003.
Cultural change alters the meaning of predictors
The social meaning of divorce changes with women's employment, educational opportunities, marriage age, housing markets, family policies, stigma, and the availability of nonmarital partnerships. A predictor estimated during an earlier period may lose strength or reverse when people marry for different reasons or when exit becomes more or less feasible. Models should be examined across generations and regions rather than assumed to be culturally invariant. Sociological and psychological explanations are complementary here: social structure shapes exposures and choices, while relational processes shape lived experiences. Amato, 2010; Conger, Conger, and Martin, 2010.
The Difference Between Relationship Quality and Divorce
Why a strong satisfaction model is not a divorce model
Research on relationship quality often appears in discussions of divorce, but the outcomes must be kept distinct. Joel and collaborators combined 43 longitudinal couples datasets involving 11,196 couples. Relationship-specific judgments such as perceived partner commitment, appreciation, sexual satisfaction, perceived partner satisfaction, and conflict were especially useful for estimating current reported relationship quality. Their predictive success weakened for future outcomes, and the investigators could not reliably forecast whose relationship quality would improve or decline. These results concern relationship quality in heterogeneous romantic partnerships, not a direct prediction model of legal divorce. Joel and colleagues, 2020.
The distinction matters because a marriage can maintain legal stability under low satisfaction, and couples can experience improving satisfaction yet later encounter new events or decisions. Conversely, highly satisfied early marriages can change in response to illness, violence, relocation, or discoveries of deception. Studies that use a continuous satisfaction scale should not be cited as though they predicted a binary legal event. Differences in measurement, samples, and observation windows matter as much as statistical significance.
Divorce prediction versus breakup prediction
A 2026 machine-learning study of romantic relationship dissolution analyzed two longitudinal samples and reported, in one North American dating sample of 1,281 people, balanced accuracy around 64%. That result is relevant to the broader promise and limitations of forecasting dissolution, yet it should not be represented as a validated divorce classifier. Dating breakups and legal divorce differ in selection, investment, financial interdependence, parenting, institutional context, and time horizon. Contemporary prediction research invites careful benchmarking of new methods but does not abolish the distinction between these outcomes. Uhlich and colleagues, 2026.
Statistical Prediction, Mechanisms, and Causality
Association does not identify a preventable cause
A predictor may be causally relevant, but longitudinal ordering by itself does not identify causation. Suppose financial strain predicts both greater marital conflict and subsequent divorce. The stress pathway is plausible, but earlier health difficulties, employment opportunities, local housing costs, and other influences may also affect both exposure and outcome. If the model adjusts for a variable caused by financial strain, it may accidentally remove part of the effect the investigator wishes to estimate. If it fails to adjust for a genuine earlier common cause, its estimate can be confounded. Clear theoretical causal diagrams and temporally ordered measurements improve interpretation. Conger, Conger, and Martin, 2010.
Likewise, marital dissatisfaction can act as an early signal, a mediator, or a partial summary of conflict and external strain. Personality may be linked to enduring negative dynamics or contribute to distress that emerges gradually. Infidelity can worsen a struggling marriage and also follow prior deterioration. The empirical task is to distinguish plausible directional pathways from a predictive association that simply signals an elevated risk. Solomon and Jackson, 2014; Previti and Amato, 2004.
Selection, survival, and changing composition
Studies restricted to married individuals are already studying a selected group: some relationships never became marriages, and some couples separated before they could enter a particular dataset. Later study waves may disproportionately lose people who move, experience homelessness, become unreachable, or terminate partnerships. Divorce may be reported inconsistently when partners separate informally or when marriages cross jurisdictions. These selection and attrition processes can distort observed correlations even when the initial sample is nationally representative. Research quality requires explicit attention to sampling, weighting, missingness, and whether both spouses were observed. Amato, 2010.
The couple, not a single respondent, is the relevant system
Two partners share exposures but do not necessarily share interpretations. One may report low relationship satisfaction, serious conflict, or an intention to leave while the other describes an acceptable marriage. Dyadic models make it possible to examine actor and partner effects and interdependence of reports. Where feasible, repeated measures show when perceptions diverged and whether behavior changed in parallel. Single-person survey data can still be informative, but must be identified as such. A striking statistic derived from one person's questionnaire should not be confused with a complete account of the marriage. Kanter and colleagues, 2022; Lavner and Bradbury, 2010.
Model validation and forecasting performance
Prediction claims require details that are often omitted from popular summaries: the number of marriages and divorces, baseline prevalence, training and test separation, outcome definition, follow-up period, handling of missing data, calibration, and confidence intervals. A result optimized on its development sample can deteriorate when tested in another state, country, or cohort. Continuous measures of marital satisfaction may yield much better predictions of current satisfaction than any model yields for the legal event of divorce. Data collected after one spouse has already decided to leave may also make forecasting appear deceptively impressive. Joel and colleagues, 2020.
Prediction models can inherit inequities from institutions. A model that assigns greater divorce probability to a group may be reproducing differences in access to economic resources, legal remedies, or marriage selection rather than identifying a group-specific psychological trait. A responsible model should document performance across subgroups and refrain from framing elevated probability as personal deficiency. These concerns grow when probabilistic forecasts are incorporated into consumer applications or automated advice systems.
Can Divorce Be Prevented by Changing Predictors?
Modifiable relationship processes
Some relationship processes can change: couples may negotiate responsibilities, develop safer and more respectful communication, address financial stress, seek treatment for substance problems, or obtain professional help for distress. Such changes may improve well-being and relationship functioning. It does not follow that every intervention demonstrably prevents divorce, because improvement in satisfaction and a reduction in legally recorded divorce are different endpoints. A person deciding to leave a harmful relationship may be making an adaptive change even if a risk model would classify the legal dissolution as a negative event.
A 2026 systematic review reassessed evidence for specific couple-therapy models using explicit empirically supported treatment criteria. It found modest support for several approaches under the evaluation framework while highlighting limited independent replication and study size as obstacles to strong model-specific conclusions. The result supports critical assessment of therapy evidence and leaves room for individualized clinical judgment; it does not support claims that a branded method can guarantee that a marriage will survive. Jackson and colleagues, 2026.
Resources and structural prevention
Preventive approaches may also target the context of marital strain: secure housing, economic stability, childcare access, manageable working conditions, and social support. The family stress literature shows why a focus only on a couple's conversational technique misses material hardship and its effects on daily life. However, a plausible mechanism is not automatically proof of a divorce-prevention effect. Evaluating a policy or program requires comparing appropriate groups and measuring outcomes over time, including safety and well-being. Conger, Conger, and Martin, 2010.
When staying together should not be the goal
Threats, coercive control, stalking, physical or sexual violence, and serious intimidation require a safety-oriented response. Joint counseling, direct confrontation, or shared negotiation may increase risk in some abusive situations. A general-purpose article on divorce predictors cannot supply individualized safety planning or determine a safe legal strategy. In threatening circumstances, confidential specialist domestic-violence services, qualified local legal and clinical professionals, and emergency responders when necessary take precedence over calculations about marital stability.
Divorce Prediction in Digital Life and the Artificial Era
Digital footprints and misleading certainty
Digital communication creates new traces of relational life: message frequency, response latency, shared calendars, financial transactions, location services, social-media behavior, and search histories. Such data might provide opportunities for research on changes in communication or daily organization, but using them to infer the likelihood of divorce demands evidence, consent, privacy protection, and independent evaluation. A decline in message frequency can reflect work schedules, illness, changed platforms, or healthy boundaries as readily as relationship deterioration. Surveillance can itself become a harmful feature of a relationship.
Algorithmic forecasts are decision aids only under strict conditions
An algorithm that detects associations in historical data cannot responsibly label an individual marriage as doomed. Deployment would require a clearly defined and relevant population, a verified endpoint, meaningful accuracy and calibration, transparent uncertainty, protections against discrimination, and evidence that the information benefits those who receive it. Research predicting romantic relationship quality or dating dissolution cannot be silently repurposed for divorce prognoses. Even a technically accurate prediction does not determine whether the ethical objective should be to maintain the marriage, improve safety, or support an autonomous decision. Joel and colleagues, 2020; Uhlich and colleagues, 2026.
For Psychology for the Artificial Era, the question also concerns how artificial systems participate in intimate decision-making. Personalized AI dialogue, automated advice, and communication tools may influence how partners interpret events or seek outside perspectives. As of the evidence examined here, no validated AI system can predict the legal future of an individual marriage with warranted clinical certainty. Empirical claims about users' choices must be tested separately from philosophical interpretations of the emerging Artificial Era.
What the Evidence Supports, and What Remains Uncertain
Established findings include the population-level relevance of marital satisfaction trajectories, serious reported marital problems, negative communication, socioeconomic circumstances, and interactions between personal characteristics and relationship functioning. Prospective designs substantially improve temporal interpretation compared with retrospective anecdotes. Distinct predictors often overlap, and estimated associations vary with how marriage and divorce are measured. Amato and Rogers, 1997; Kanter and colleagues, 2022; Solomon and Jackson, 2014.
Supported but conditional mechanisms include stress-related pathways from economic pressure to relationship deterioration, individual-difference pathways through satisfaction, reciprocal relations between infidelity and marital distress, and the role of commitment and exit constraints. Many observational studies remain vulnerable to unmeasured confounding, sample selection, and changing social conditions. Conger, Conger, and Martin, 2010; Previti and Amato, 2004; Previti and Amato, 2003.
Limited or contested claims include universal personality-based divorce formulas, fixed numbers of warning signs, infallible communication tests, a single ideal age for marriage, and promises that a specific intervention will prevent legal dissolution. Forecasts derived from dating samples and relationship-satisfaction studies require fresh evaluation before being applied to marriage. Individual marital futures remain open to decisions, changing circumstances, safety considerations, and events that are outside any fixed baseline measurement. Joel and colleagues, 2020; Uhlich and colleagues, 2026.
Frequently Asked Questions
What is the strongest predictor of divorce?
No single predictor consistently dominates all populations, studies, and time horizons. Lower satisfaction, destructive communication patterns, serious marital problems, and contextual strain repeatedly emerge as relevant, but their importance changes with the outcome measured and variables included. An apparently strong association may decline once other relational and social conditions are considered. The best-supported answer is an interacting profile rather than a solitary sign. Amato, 2010; Kanter and colleagues, 2022.
Can scientists tell whether a particular couple will divorce?
Scientists can estimate patterns of risk in defined groups and evaluate models against future observations. The current literature does not establish a universally reliable individual prognosis applicable across marriages, cultures, and life stages. Trustworthy predictions would need prospective data, time-specific probabilities, external validation, appropriate calibration, and careful handling of the fact that divorce is a legal and social outcome.
Does arguing frequently mean a marriage will end?
Disagreement by itself is insufficient. Researchers distinguish hostile or destructive communication from constructive disagreement and examine broader trajectories of satisfaction and stress. A meta-analysis identified an association between negative communication and later dissolution, with effects too limited to turn any isolated argument into an individual forecast. Kanter and colleagues, 2022.
Are people whose parents divorced more likely to divorce?
Some population studies identify associations between family background and adult marital outcomes. The pathways can include childhood conflict, economic resources, relationship expectations, and selection into partnerships. These associations are not inherited destiny, and the individual effect cannot be calculated from the fact of parental divorce alone. A family history is one contextual variable among many. Amato, 2010.
Does marrying young cause divorce?
Age at marriage is associated with marital stability in some samples, particularly for very early marriages, but the relationship mixes developmental timing, education, financial resources, selection, and historical context. An observed risk difference cannot establish a universal causal cutoff or a single ideal age. The responsible interpretation is population- and cohort-specific. Amato, 2010.
Is infidelity a cause of divorce or a sign of an already failing marriage?
Longitudinal evidence supports bidirectional processes: deteriorating commitment or divorce proneness can precede extramarital involvement, and infidelity can subsequently worsen marital functioning and raise divorce risk. Specific circumstances and responsibility must still be evaluated without assuming that any one pattern applies to every marriage. Previti and Amato, 2004.
Does therapy prevent divorce?
Couple therapy can help many couples address relationship distress, but improvements in distress and legal-divorce prevention are distinct outcomes. A 2026 systematic review found varying and generally modest model-specific evidentiary support under its selected criteria. The goal of care must incorporate participants' autonomy, mental health, and safety; maintaining marriage is not always the appropriate objective. Jackson and colleagues, 2026.
Conclusion
Divorce emerges from processes unfolding between people within economic, developmental, cultural, and institutional environments. The most useful predictors concern evolving marital satisfaction, conflict and communication, commitment and constraints, serious breaches of trust, personality-linked responses, and the resources available to a family. Their relationships are probabilistic, interactive, historically contingent, and sometimes reciprocal. Scientific understanding improves when a model distinguishes actual legal dissolution from dissatisfaction or separation, evaluates changes over time, and tests whether findings generalize beyond the sample in which they were discovered. Karney and Bradbury, 1995; Amato, 2010.
The constructive use of research is to understand mechanisms, identify sources of distress, strengthen support where appropriate, and protect informed and safe decision-making. A marriage is a lived relationship, and its legal continuation is only one possible outcome. Prediction cannot replace careful attention to the circumstances, values, and safety of the people whose lives it concerns.
Related Articles
Cluster Divorce, Separation and Breakup Psychology — The scientific overview of relationship dissolution, attachment disruption, recovery, divorce, and family reorganization.
Psychology of Divorce: Stress, Loss, and Life Transition — Psychological adaptation and life-course transitions during and after legal marriage dissolution.
How to Cope With Divorce: Adjustment, Support, and Daily Functioning — Research-informed approaches to adult adjustment, support, and everyday functioning during divorce.
Gray Divorce: Psychological Adjustment After Age 50
Divorce and Mental Health: Depression, Anxiety, and Adjustment
Divorce and Physical Health: Stress, Sleep, and Health Outcomes
References
Amato, P. R. (2010). Research on divorce: Continuing trends and new developments. Journal of Marriage and Family, 72(3), 650–666. https://doi.org/10.1111/j.1741-3737.2010.00723.x
Amato, P. R., and Previti, D. (2003). People's reasons for divorcing: Gender, social class, the life course, and adjustment. Journal of Family Issues, 24(5), 602–626. https://doi.org/10.1177/0192513X03254507
Amato, P. R., and Rogers, S. J. (1997). A longitudinal study of marital problems and subsequent divorce. Journal of Marriage and Family, 59(3), 612–624. https://doi.org/10.2307/353949
Boertien, D., and Härkönen, J. (2018). Why does women's education stabilize marriages? The role of marital attraction and barriers to divorce. Demographic Research, 38, 1241–1276. https://doi.org/10.4054/DemRes.2018.38.41
Bradbury, T. N., Fincham, F. D., and Beach, S. R. H. (2000). Research on the nature and determinants of marital satisfaction: A decade in review. Journal of Marriage and Family, 62(4), 964–980. https://doi.org/10.1111/j.1741-3737.2000.00964.x
Conger, R. D., Conger, K. J., and Martin, M. J. (2010). Socioeconomic status, family processes, and individual development. Journal of Marriage and Family, 72(3), 685–704. https://doi.org/10.1111/j.1741-3737.2010.00725.x
Jackson, J. B., Codecá, L., Miller, R. B., Simpson, J. E., O'Leary, A. M., and Renner, S. M. (2026). Empirically supported couple therapy models for couple relationship distress: A replication and update. Journal of Marital and Family Therapy, 52(4), e70162. https://doi.org/10.1111/jmft.70162
Joel, S., et al. (2020). Machine learning uncovers the most robust self-report predictors of relationship quality across 43 longitudinal couples studies. Proceedings of the National Academy of Sciences, 117(32), 19061–19071. https://doi.org/10.1073/pnas.1917036117
Kanter, J. B., Lavner, J. A., Lannin, D. G., Hilgard, J., and Monk, J. K. (2022). Does couple communication predict later relationship quality and dissolution? A meta-analysis. Journal of Marriage and Family, 84(2), 533–551. https://doi.org/10.1111/jomf.12804
Karney, B. R., and Bradbury, T. N. (1995). The longitudinal course of marital quality and stability: A review of theory, method, and research. Psychological Bulletin, 118(1), 3–34. https://doi.org/10.1037/0033-2909.118.1.3
Lavner, J. A., and Bradbury, T. N. (2010). Patterns of change in marital satisfaction over the newlywed years. Journal of Marriage and Family, 72(5), 1171–1187. https://doi.org/10.1111/j.1741-3737.2010.00757.x
Lavner, J. A., Karney, B. R., and Bradbury, T. N. (2014). Relationship problems over the early years of marriage: Stability or change? Journal of Family Psychology, 28(6), 979–985. https://doi.org/10.1037/a0037752
Le, B., and Agnew, C. R. (2003). Commitment and its theorized determinants: A meta-analysis of the investment model. Personal Relationships, 10(1), 37–57. https://doi.org/10.1111/1475-6811.00035
Previti, D., and Amato, P. R. (2003). Why stay married? Rewards, barriers, and marital stability. Journal of Marriage and Family, 65(3), 561–573. https://doi.org/10.1111/j.1741-3737.2003.00561.x
Previti, D., and Amato, P. R. (2004). Is infidelity a cause or a consequence of poor marital quality? Journal of Social and Personal Relationships, 21(2), 217–230. https://doi.org/10.1177/0265407504041384
Solomon, B. C., and Jackson, J. J. (2014). Why do personality traits predict divorce? Multiple pathways through satisfaction. Journal of Personality and Social Psychology, 106(6), 978–996. https://doi.org/10.1037/a0036190
Stanley, S. M., Rhoades, G. K., and Markman, H. J. (2006). Sliding versus deciding: Inertia and the premarital cohabitation effect. Family Relations, 55(4), 499–509. https://doi.org/10.1111/j.1741-3729.2006.00418.x
Uhlich, M., Impett, E. A., Spielmann, S. S., and Bojar, D. (2026). Forecasting relationship dissolution: Machine learning identifies key predictors across two longitudinal samples. European Journal of Personality, advance online publication. https://doi.org/10.1177/08902070261455278
Zhang, Y., and Van Hook, J. (2009). Marital dissolution among interracial couples. Journal of Marriage and Family, 71(1), 95–107. https://doi.org/10.1111/j.1741-3737.2008.00582.x