Psychology by BogdanovaThe Encyclopedia of Psychology for the Artificial Era

RESEARCH DOSSIER

Relational Power, Coercive Control, and Intimate Partner Violence

Abstract. Relational power, coercive control, and intimate partner violence form an interconnected research field concerning the capacity to constrain another person's practical choices through intimidation, surveillance, isolation, deprivation, and violence. This fundamental critical narrative monograph analyzes the historical development of coercion theories; relational and institutional mechanisms of entrapment; economic, sexual, postseparation, and technology-facilitated abuse; and population variation in exposure and outcomes. It compares original observational and case-control studies, longitudinal research, psychometric reviews, population surveillance, clinical guidelines, and controlled intervention evidence. The synthesis establishes why discrete incident counts, coercive behavioral patterns, perceived fear, objective constraints, physical injury, and recovery require separate measurements. It evaluates evidence for serious health outcomes, lethality risk markers, prevention, advocacy, and psychological care while identifying gaps in causal inference, instrument validation, survivor safety, and generalizability. The analysis proposes a rigorous measurement and research agenda, including the evolving role of Artificial mediation in interpersonal power. It reports no original human-subject study or new meta-analysis.

Keywords: relational power; coercive control; intimate partner violence; psychological abuse; interpersonal coercion; entrapment; surveillance; economic abuse; technology-facilitated abuse; postseparation abuse; victimization; autonomy; partner violence measurement; lethality risk; longitudinal research; advocacy intervention; trauma; social support; safety; Artificial Era

The Scientific Object: Power, Constraint, and Partner Violence

Relational power denotes a person's capacity to influence another person's feasible choices, the distribution of resources, and the consequences of interaction within an enduring relationship. Intimate partner violence (IPV) encompasses behavior by a current or former intimate partner that causes physical, sexual, or psychological harm, including aggression, sexual coercion, psychological abuse, and controlling behavior. Coercive control identifies an organized course of conduct that constrains the other's practical freedom through intimidation, isolation, surveillance, threats, deprivation, or enforced dependence. These constructs describe related but distinguishable phenomena: power can organize ordinary cooperation, while coercive power makes refusal costly or dangerous. A scientific account must identify whose choices narrow, which mechanisms produce that narrowing, and how behavior changes as the relationship develops WHO (2026) Stark (2023) Dutton & Goodman (2005).

The primary unit of analysis is the relationship-in-context through time. Episodes of assault, interference with work, threats involving children, and restrictions on contact may share a coercive function even when each episode appears unrelated to the others. The analyst therefore needs both event-level information and a longitudinal account of contingencies: what was demanded, what was threatened, what happened after resistance, what the person anticipated, and which realistic alternatives remained. Documenting a victim-survivor's constrained options is different from imputing a perpetrator's internal motives from a checklist. Behavior, credible threat, experienced fear, material dependency, and effects on liberty are separable observables Dutton & Goodman (2005) Hamberger et al. (2017).

This monograph is a critical narrative synthesis of primary empirical studies, theory, population surveillance, systematic reviews, psychometric investigations, clinical guidance, and legal research frameworks published through 2026. It reports neither a new original dataset nor a protocol-based systematic review or meta-analysis. Its questions concern the origins and explanatory scope of coercive-control theory, mechanisms of entrapment and harm, the validity of measures, differentiation of IPV patterns, limits of prediction, and the evidence for prevention and response. Evidence quality varies by question, from controlled interventions to cross-sectional surveys and qualitative case series. The analysis maintains explicit distinctions between historical theories, observed associations, causal estimates, and normative legal rules Capaldi et al. (2012) Coates et al. (2026) WHO (2025, prevalence report).

Interdependence, Autonomy, and the Distribution of Options

Close relationships entail interdependence: partners coordinate housing, finances, care, privacy, time, intimacy, and plans. This coordination can be negotiated with meaningful opportunities to disagree, renegotiate, or leave. A power imbalance becomes coercively relevant when one party controls indispensable resources or sanctions and can use them to suppress the other's independent decisions. A jointly chosen budget, shared password, or location exchange cannot be classified from its surface appearance alone. Researchers must examine consent, reversibility, enforcement, knowledge of consequences, and actual freedom to refuse. The distinction clarifies why behaviorally similar practices differ sharply across contexts Stark (2007) Postmus et al. (2012).

An options-based analysis distinguishes nominal permission from effective autonomy. A person may technically be free to contact relatives while lacking access to transportation, a private device, documents, money, or safe communication. Likewise a person may be permitted to seek employment yet face retaliation whenever an interview is scheduled. These obstacles form a structure of constrained alternatives, which is often more informative than a single report of feeling powerless. The relevant outcomes include freedom of movement, financial agency, bodily autonomy, social participation, safety, self-efficacy, and the ability to make ordinary decisions without fear Stark (2023) Postmus et al. (2012).

Power is relational and institutionally conditioned. Labor markets, housing access, disability accommodations, immigration status, childcare, family-law processes, and social norms influence what threats can accomplish. A threat may become more credible when the threatened person lacks money, legal information, or accessible support. The empirical task is to separate background vulnerability from a partner's active exploitation of that vulnerability. Structural inequality changes exposure and available remedies; it does not determine any individual's behavior or erase agency. Multilevel models should therefore connect material context, interaction histories, behavioral tactics, and resulting restrictions instead of describing abuse solely through personality traits Capaldi et al. (2012) WHO (2025, RESPECT).

Historical Development of the Research Problem

Early research often organized domestic violence around discrete physical acts, injury, or recorded incidents. Conflict-tactics methods made behavior countable across households, while advocacy and feminist analyses emphasized domination, social inequality, and the practical conditions under which violent incidents occurred. These traditions generated different sampling frames and different questions. A population survey asking whether partners pushed or slapped one another does not establish the same phenomenon as a shelter interview exploring prolonged fear, isolation, sexual coercion, and the inability to leave. Understanding this divergence is necessary before comparing historical prevalence claims or gender distributions Johnson & Leone (2005) Myhill (2015) Swan et al. (2008).

Dutton and Goodman (2005) articulated a social-power account in which coercion involves a demand, a credible threatened consequence, monitoring, possible enforcement, and the target's response, all situated in a wider social ecology. The model is especially useful because it makes the mechanism more precise than an undifferentiated label of control. Stark (2007) subsequently organized an influential liberty-centered account of coercive control as entrapment produced through intertwined practices of isolation, intimidation, regulation, and deprivation. Its historical significance lies in shifting the question from how many assaults occurred to how a person's day-to-day liberty came under another's control Dutton & Goodman (2005) Stark (2007).

Later work subjected the construct to empirical and conceptual revision. Johnson and Leone (2005) distinguished coercively controlling violence from situational couple violence, Myhill (2015) examined the consequences of measuring only acts rather than patterns, and Hamberger and colleagues (2017) documented divergent operational definitions. Stark and Hester (2019) reviewed developments in research and legal recognition. The revised second edition of Stark's book in 2023 extends the analysis to privacy and liberty rights and to experience with criminalization. These works established an active research program, not a universally agreed diagnostic taxonomy; their propositions remain answerable to psychometric evidence and independent tests Johnson & Leone (2005) Myhill (2015) Hamberger et al. (2017) Stark & Hester (2019) Stark (2023).

Theoretical Models and Rival Explanations

The coercion model emphasizes credible negative consequences and compliance; the entrapment model emphasizes the cumulative erosion of liberty and the organization of everyday life. Johnson's typology separates patterns of coercively controlling violence from violence that arises during specific conflicts without a comparable domination structure. Developmental systems research examines how individual histories, partner characteristics, relationship dynamics, and environments jointly contribute to the occurrence of IPV. These approaches differ in their explananda. One seeks the mechanism of a particular demand, another a course of conduct, another heterogeneity across relationships, and another determinants of future victimization and perpetration Dutton & Goodman (2005) Johnson & Leone (2005) Capaldi et al. (2012).

A credible synthesis evaluates explanations by specifying expected observations. If coercion is sustained by sanctions, temporal records should show that resistance predicts retaliation or credible anticipation of retaliation and that restricted behavior persists after threats become implicit. If developmental pathways matter, prospectively measured histories and contextual exposures should predict later risk with the proper controls. If violence types differ meaningfully, clusters defined independently of outcomes should predict harm or persistence in new samples. A typology derived from the same survey items used to demonstrate its consequences invites circularity; strong validation requires temporally distinct measurement, replication, and prediction outside the derivation sample Johnson & Leone (2005) Hamberger et al. (2017) Capaldi et al. (2012).

Attachment anxiety, jealousy, alcohol misuse, financial stress, and personality-related characteristics can correlate with aggression or coercion, but correlations do not define the abuse construct or establish an excuse for conduct. Equally, describing all partner violence as spontaneous conflict leaves the mechanism of systematic domination unexamined. The main scientific principle is to retain explanatory pluralism while protecting conceptual precision: intent, control behavior, resource dependency, threat credibility, injuries, and survivor outcomes are variables to be observed and tested. This architecture supports competing hypotheses instead of assigning explanatory priority to an attractive theory in advance Capaldi et al. (2012) Sam et al. (2026).

Gender, Behavioral Counts, and Contextual Asymmetry

The debate over gender symmetry illustrates why research designs generate apparently incompatible answers. Studies that count selected acts of physical aggression can find similar frequencies by gender in certain community samples. Studies that evaluate sexual violence, coercive domination, fear, injury, and the capacity to escape often identify different distributions. Myhill's analysis of national survey data found that situational violence was more gender-symmetrical whereas coercively controlling abuse was strongly gendered in that dataset. The finding is population- and measure-dependent; it cannot be transferred mechanically to all settings or identities Myhill (2015) Swan et al. (2008).

Swan and colleagues' review of women's use of violence toward male partners cautions against treating raw act counts as equivalent evidence about motivation, context, harm, or control. They discuss defensive violence, mutually violent relationships, and the differential prevalence of sexual coercion, stalking, and injury. At the same time, men and people in sexual- and gender-minority relationships can experience coercive abuse, and their experiences require direct measurement rather than inference from heterosexual surveys. Gender-sensitive analysis and universal eligibility for protection can coexist: each research claim must specify the population, operational definition, sampling frame, and outcomes measured Swan et al. (2008) CDC (2026) Coates et al. (2026).

The ethical meaning of this issue extends to measurement practice. Asking both partners whether they ever threatened one another cannot reveal the direction of control, whether the threat was defensive, whose ordinary activities became restricted, or whether one party could safely disagree. Confidential, separate interviews, incident context, questions about perceived consequences, and time-ordered data are therefore essential for some research questions. A survivor account is significant evidence about lived constraint, while legal culpability and specific causal attribution require their own evidence rules. The methodological obligation is to avoid both false symmetry and premature certainty Myhill (2015) Hamberger et al. (2017) WHO (2014).

Threat, Intimidation, and Learned Compliance

Coercive threats are effective partly through their relationship to prior events. A statement involving harm, eviction, children, immigration sponsorship, employment, private images, or self-harm may have substantial coercive force when a perpetrator has previously imposed penalties or controls the means to do so. Dutton and Goodman's model specifies demands, credible threatened consequences, surveillance, and responses rather than assuming that a harsh statement invariably establishes coercion. A research interview must reconstruct what the person understood the demand to be, which alternatives they perceived, and what followed refusal. Threat credibility is an empirical property of the interaction and its history, not merely the severity of the words Dutton & Goodman (2005) Stark (2023).

Conditioned anticipation can make coercion less visible. After repeated punishment, a person may avoid friends, conceal opinions, change clothing, alter travel, or monitor their own speech without receiving a fresh explicit command. A cross-sectional inventory of recent threats can then undercount control even as the cumulative restriction intensifies. This hypothesis is amenable to longitudinal diary and event-history methods, especially when reports distinguish a direct demand from self-restriction in anticipation of consequences. Safety and privacy procedures are necessary because eliciting information about threatened resistance can itself increase danger if disclosed Hamberger et al. (2017) Stark & Hester (2019).

Coercive regimes may contain periods of affection, reassurance, apology, or apparent calm. These observations cannot be coded automatically as evidence that the threatening structure has dissolved. The analytic question is whether rules, retaliation contingencies, resource restrictions, and actual choices have changed. Claims about intermittent reinforcement must be tested rather than applied as a universal explanatory law to every relationship. A rigorous account separates directly reported behavior from interpretation of reinforcement processes, and measures sustained changes in autonomy and safety rather than immediate relief after an episode Stark (2007) Coates et al. (2026).

Social Isolation, Information Control, and Dependency

Isolation can be produced by direct bans, conflict whenever a person sees friends, confiscation of a device, deliberate reputational damage, or demands that exhaust the time available for social relationships. Social networks provide material assistance, independent interpretation, housing opportunities, employment information, and emotional support. Restricting access to these networks can therefore intensify dependence even when the perpetrator does not physically prevent every contact. The explanatory mechanism is the reduction of practical alternatives and independent sources of information. Research should distinguish network shrinkage caused by control from changes associated with relocation, caregiving, poverty, or voluntary preference Stark (2007) WHO (2026).

Control over information operates through monitoring messages, requiring disclosure, withholding documents, blocking access to health information, or manipulating the account of events presented to relatives and institutions. These practices can undermine privacy, credibility, and the ability to plan safe action. The 2026 systematic review by Coates and colleagues found that many widely used psychological-IPV instruments incompletely represent covert behavior categories including information control and abuse concealment. This is a content-validity problem: an instrument can be internally consistent yet omit crucial aspects of the phenomenon it claims to measure Coates et al. (2026).

The effects of isolation are reciprocal with the material environment. A person who cannot reach coworkers or family may have fewer housing options; restricted finances then make future contact more difficult. Research on these interacting pathways requires measures of networks, communication privacy, material access, and behavior over time. Explanations focusing exclusively on a survivor's attachment or confidence overlook the imposed costs of contact. Survivor coping can be adaptive within a threatening environment while appearing puzzling to observers who do not know the sanctions associated with seemingly minor independent decisions Stark (2023) Postmus et al. (2012).

Economic Abuse, Housing, Work, and Material Entrapment

Economic abuse includes withholding money, controlling accounts, creating coerced debt, sabotaging employment or education, and appropriating earnings or benefits. The underlying mechanism is the conversion of material dependence into power over daily decisions and exit options. Postmus and colleagues (2012) investigated economic abuse among survivors and emphasized its relationship with economic self-sufficiency. The scientific task is to measure the specific action and its consequence separately: restricted access to a bank account, for example, has a different evidential role from the broader outcome of unstable housing or lost income Postmus et al. (2012).

Economic dependence can precede an abusive relationship, arise through jointly chosen family arrangements, or be actively manufactured by a partner. Only careful longitudinal and contextual evidence can discriminate these trajectories. A caregiving parent with limited paid employment may face a structural disadvantage even without partner misconduct; sabotage of work or confiscation of wages changes the explanation. This distinction matters for causal inference and for designing support: increasing income alone may be insufficient when a perpetrator can monitor, confiscate, or retaliate against new resources Capaldi et al. (2012) WHO (2025, RESPECT).

A material-options approach also explains why relationship departure is not a simple behavioral decision. Housing shortages, legal costs, childcare, disability accommodations, transportation, and debt can constrain possibilities. Estrangement can change risk rather than simply terminate it. Campbell and colleagues' multisite case-control investigation identified estrangement, particularly from a controlling partner, among factors associated with intimate partner femicide in the studied population. This observation demands sensitive risk assessment and supports analysis of separation as a transition with its own exposure pathways, not a one-time safety endpoint Campbell et al. (2003).

Sexual Coercion, Reproductive Control, and Bodily Autonomy

Sexual coercion can involve physical force, threats, exploitation of dependency, or persistence under conditions in which refusal carries credible consequences. Within a coercive-control pattern, sexual access may become another domain of regulated autonomy. The scientific unit is the person's opportunity to provide and withdraw freely given consent under the relevant circumstances; formal relationship status cannot establish it. The World Health Organization includes sexual coercion and psychological abuse in its conceptualization of IPV, and research on patterns of partner violence shows that sexual coercion does not track every other form of aggression identically WHO (2026) Swan et al. (2008).

Reproductive interference and coercive conditions involving contraception, pregnancy decisions, or access to reproductive health services require careful definition and culturally appropriate research instruments. A survey question about disagreement over childbearing differs from evidence of threats, contraception sabotage, or denial of medical care. The relevant outcomes may include bodily autonomy, psychological distress, reproductive health, and the practical ability to obtain confidential services. The same measurement principle applies as elsewhere: behavior, coercive context, victim-survivor interpretation, and health outcomes should be captured separately rather than fused into a single unvalidated index WHO (2013) WHO (2014).

Research involving sexual coercion requires stronger privacy protections than ordinary couple surveys. Partner-present screening can compromise disclosure; joint disclosure of results can expose someone to retaliation. Ethical research plans should establish independent contact methods, safe response options, confidentiality limits, and referral procedures before fieldwork. Clinical services have their own obligations concerning consent, injury, trauma-related symptoms, and reproductive health. Their evidence-based protocols should be assessed according to current health guidance rather than inferred from abstract relational theory WHO (2014) WHO (2013).

Digital Monitoring and Technology-Facilitated Coercive Control

Digital platforms expand the instruments through which intimate partners can seek information, require availability, and impose costs on independent action. Location services, messaging platforms, account sharing, social media, connected household devices, and image storage may become channels for surveillance or threats. The mechanism depends on consent, access, persistence, and enforcement rather than on a device category. Requiring location proof after an ordinary trip, using account access to interfere with employment, or threatening release of private content may extend an established pattern of intimidation into digital environments. Kim and Ferraresso's systematic review documents the expanding empirical literature on technology-facilitated IPV Kim & Ferraresso (2023).

Digital monitoring can continue after physical separation, often through shared accounts, cloud backups, device ecosystems, co-parenting communication, or online impersonation. The scientific challenge is to distinguish ordinary digital communication from intrusive monitoring and coercive use of information. Contextual assessment must ask who controls the technical environment, whether participation is freely revocable, whether another person has access to private communications, and what occurs when monitoring is resisted. The extension from an offline pattern to a digitally mediated one should be demonstrated through behavioral evidence and relationship history, rather than presumed from the presence of tracking software Kim & Ferraresso (2023) Stark & Hester (2019).

A 2026 meta-analysis of technology-facilitated IPV by Sam and colleagues synthesized 44 studies and reported modest pooled associations of dispositional vulnerabilities and protective factors with technology-facilitated IPV, alongside heterogeneity and small-study effects. The authors noted predominance of cross-sectional designs and limited technological specificity in existing theories. Consequently, jealousy scores or personality measures should not be treated as tools for diagnosing digital abuse or predicting individual perpetrators. Future work requires validated measures of platform affordances, partner access, coercive consequences, and temporal escalation, alongside privacy-preserving study designs Sam et al. (2026).

Separation, Stalking, and Continuing Patterns of Control

A relationship's legal or social end does not necessarily terminate coercive access. Postseparation stalking, repeated unwanted contact, litigation misuse, financial interference, threats concerning children, and technology-assisted tracking can continue to shape decisions. Research must treat separation as a period of changing interaction opportunities and threats. Campbell and colleagues' case-control data associated estrangement from a controlling partner with increased femicide risk in the sampled cases; the finding identifies a group-level risk factor, not a deterministic prediction about a particular separation Campbell et al. (2003) Stark & Hester (2019).

The use of children as communication channels or instruments of pressure complicates the temporal boundary of abuse. Katz, Nikupeteri, and Laitinen (2020) analyzed children's accounts of postseparation coercive control and described constraints affecting their autonomy, school life, social life, and sense of safety. Such qualitative evidence is valuable for specifying mechanisms and generating questions, but cannot independently yield population prevalence or average causal effects. The child may be directly targeted, drawn into surveillance or messaging, or affected through restrictions on a caregiving parent. These roles deserve independent measurement Katz et al. (2020).

Administrative and legal processes may create necessary contact even when personal contact has stopped. The existence of a custody proceeding or complaint is not itself proof of abuse; the question is whether specific conduct repeatedly exploits procedures, resources, or information to intimidate or restrict another person's freedom. Legal categories are jurisdiction-specific. The England and Wales statutory guidance discusses controlling or coercive behavior as a repeated or continuous course of conduct with specified serious effects and states that relevant rules can cover former partners without a cohabitation requirement. Empirical study must keep that legal test separate from psychological construct measurement UK Home Office (2023).

Population Epidemiology and the Denominator Problem

Prevalence estimates depend on who was eligible, what counted as exposure, which time interval was asked about, and whether disclosure could be made safely. Physical or sexual IPV, psychological aggression, and coercive control are overlapping constructs with different denominators. The World Health Organization's 2023 estimates, published in 2025, draw on population-based studies from 168 countries and areas and concern violence against women, including physical or sexual partner violence and nonpartner sexual violence. The report estimates that nearly one in three women aged 15 or older experienced at least one of these categories during their lifetime. That combined figure is not a global prevalence estimate for coercive control WHO (2025, prevalence report) WHO (2026).

The CDC's United States NISVS 2023/2024 data brief separates psychological aggression into expressive aggression and coercive control or entrapment. It reports lifetime coercive control or entrapment for 27.2% of U.S. women and 19.5% of U.S. men under its seven-behavior operationalization. These are survey-defined categories, not observed rates of a continuously organized coercive regime. The same report gives detailed items concerning isolation, resource deprivation, monitoring, threats, and interference with personal decisions. Its value is an explicit sampling and item framework; interpretation must remain faithful to those items and the population CDC (2026).

Survey comparability also depends on cultural interpretation, translation, access to telephones or devices, and the presence of another household member during participation. Disclosure of violence may be suppressed by fear, stigma, or surveillance, causing differential underreporting. Surveys limited to physical events can exclude relationships organized primarily through deprivation and intimidation, whereas broad psychological-abuse scales can capture transient conflict alongside sustained coercion. Reporting several clearly defined estimands is preferable to forcing them into one number. Apparent historical change may reflect changes in instrument coverage, sample composition, recall, or disclosure rather than real incidence Myhill (2015) Coates et al. (2026) WHO (2025, prevalence report).

Injury, Lethality, and Acute Risk

Physical injury remains an indispensable outcome even though coercive control can occur without assault. Campbell and colleagues (2003) compared proxy reports for 220 victims of intimate partner femicide with 343 abused control women across 11 U.S. cities. In multivariable analyses, firearm access, prior threats with a weapon, and estrangement, especially from a controlling partner, were among the identified risk factors. Because the design is case-control, it estimates associations under assumptions about selection and measurement; it cannot establish that these factors cause every homicide. Its strength lies in focusing on severe outcomes and comparing cases with abused controls rather than with all women Campbell et al. (2003).

Glass and colleagues (2008) examined nonfatal strangulation as a risk marker for attempted and completed intimate partner homicide. Strangulation is a medically serious form of assault because injury can occur even when immediately visible signs are limited. The study supports attention to this history during professional risk assessment. It does not establish that every person reporting a single risk marker will experience lethal violence. Clinical examination and referral decisions belong to trained professionals using appropriate protocols, while research should specify whether a measure predicts serious reassault, near-lethal violence, or homicide Glass et al. (2008) Messing et al. (2017).

Risk prediction has practical value and important limits. Messing, Campbell, and Snider's 2017 validation of the Danger Assessment-5 followed women recruited after police domestic-violence calls and assessed severe or near-lethal reassault. The reported discrimination of the original brief measure was moderate, with an area under the receiver operating characteristic curve of 0.68. That value refers to discrimination in the evaluated setting, not a guarantee about an individual. The choice of threshold, consequences of false negatives and false positives, survivor priorities, and available services must shape implementation. A checklist should support an informed safety process rather than stand alone as a decision maker Messing et al. (2017).

Mental Health, Psychological Injury, and Recovery

Partner violence is associated with depression, anxiety, posttraumatic stress symptoms, impaired sleep, and other mental-health burdens. The pathways can involve direct fear and injury, prolonged uncertainty, loss of social support, economic deprivation, sexual violation, and disruption of work and family life. Oram, Khalifeh, and Howard (2017) synthesized the public mental-health implications of violence against women and argued for effective recognition and responses within mental-health services. An explanatory account should distinguish immediate distress, clinically assessed disorders, ongoing threat, and long-term recovery. Psychological harm can be substantial without meeting the criteria for a particular psychiatric diagnosis Oram et al. (2017) WHO (2014).

Longitudinal evidence provides stronger temporal information than a single contemporaneous association. Devries and colleagues (2013) systematically reviewed prospective studies of IPV, depressive symptoms, and suicide attempts. Their synthesis supports an association between exposure and subsequent depression outcomes while also considering bidirectionality and prior vulnerability. Observational longitudinal designs still require careful treatment of unmeasured confounding, differential attrition, and changing exposure. The recognition of earlier depression as a risk correlate should never be converted into an assertion that the survivor caused the abuse. Mechanisms and responsibility are distinct research questions Devries et al. (2013).

Research on recovery must measure both symptom improvement and the conditions supporting it. A decline in anxiety may reflect effective support, temporary cessation of contact, changed housing, or other transitions. Safety, financial autonomy, social connection, functional recovery, and the ability to act without anticipated retaliation are independent outcomes. Standard mental-health instruments can miss a continuing pattern of surveillance or threat, while abuse scales can miss clinical impairment. A multidomain evaluation can better capture the interaction between mental health and material safety Hameed et al. (2020) Rivas et al. (2015) Stark (2023).

Children, Caregiving, and Intergenerational Exposure

Children may directly experience coercive rules, be exposed to threats and assault, become intermediaries in communication, or encounter changes in caregiving produced by another adult's control. The distinction between witnessing violence and being subject to coercion is analytically important. Katz and colleagues' study of postseparation fathering documented children's descriptions of restriction and fear in relationships involving coercive control. Such accounts identify outcomes involving school, friendships, personal autonomy, and physical or psychological well-being that adult-focused surveys may omit Katz et al. (2020).

Intergenerational research requires temporal and causal care. Exposure to violence may be associated with later emotional or behavioral difficulties, but adversity is heterogeneous and developmental outcomes depend on many contextual factors, protective relationships, and opportunities for support. A family-systems vocabulary of mutual influence must not imply that all family members share equal power or responsibility. When one member imposes threats or sanctions, a model built exclusively around reciprocal conflict can misidentify the process. Confidential data collection and separate information sources are especially important when caregiver accounts may be influenced by the coercive environment Capaldi et al. (2012) WHO (2025, RESPECT).

Child outcomes should not be reduced to future perpetration or victimization. Developmental research can assess immediate emotional regulation, access to education and social support, physical health, daily routines, parental availability, and sense of safety. Prospective designs can distinguish direct exposure from correlated adversities such as housing instability. Ethical procedures must anticipate that children may have different fears, loyalties, and disclosure needs from adults. Evidence from qualitative accounts is indispensable for identifying constructs, while representative longitudinal research is necessary for estimating their prevalence, developmental course, and effect sizes Katz et al. (2020) Capaldi et al. (2012).

Population Diversity and Unequal Access to Protection

The manifestation and consequences of coercive control vary with the resources, institutions, and social positions available to people. Disability-related care, migration status, language access, financial dependency, rural isolation, and discrimination can alter which threats are credible and what escape options exist. Research must identify these mechanisms without assuming uniform exposure within any demographic category. Valid comparison requires equivalent questions, privacy of disclosure, and explicit attention to sampling exclusions. Estimates from women attending shelters cannot be treated as population prevalence; national surveys may miss people with restricted phone access, precarious housing, or language barriers WHO (2025, prevalence report) Capaldi et al. (2012).

Sexuality- and gender-diverse survivors face additional possibilities for identity-related threats, social exposure, and failures of service recognition. Coates and colleagues (2026) reviewed 157 studies and 77 measures of psychological IPV and found limited development and validation for sexuality- and gender-diverse populations. This gap is not merely a matter of adding demographic covariates. An instrument may overlook tactics specific to outing threats or identity-related institutional vulnerability, while an item about financial restriction may carry different implications depending on material context. Cross-group comparisons require testing construct coverage and measurement invariance Coates et al. (2026).

Gender-inclusive inquiry can preserve the strong evidence for gendered patterns of serious coercive violence while measuring male, transgender, nonbinary, and same-sex survivors on their own terms. The WHO's global estimates of violence against women answer a specific public-health question, whereas the CDC's surveillance reports exposure categories among women and men within the United States. They should not be collapsed into a single claim about all relationships. Comparative science requires explicit denominators, sexual and gender identities where consented and appropriate, relationship structure, and outcome severity WHO (2025, prevalence report) CDC (2026) Swan et al. (2008).

The Evidence Hierarchy and Its Limits

Different research questions require different evidentiary designs. A population survey can estimate exposure under an operational definition but cannot alone identify the causal mechanism of an abusive relationship. A case-control investigation can compare risk markers in serious outcomes but is vulnerable to retrospective reporting and selection bias. A qualitative interview can reveal the sequence and meaning of coercive rules but does not yield a population rate. A randomized support intervention can test a service package in its recruited population while leaving the mechanism of change uncertain. Treating these designs as interchangeable produces artificial consensus Campbell et al. (2003) Katz et al. (2020) Sullivan & Bybee (1999).

Measurement error has directional consequences. Fear of disclosure can understate abuse; broad items can overstate the extent of organized control; attrition may be related to danger or unstable contact; proxy accounts can contain recall differences. Survivor-centered procedures can improve both safety and data quality, yet confidentiality alone does not remove selection and reporting biases. Independent verification of records, time-ordered measurement, preregistration where feasible, validated instruments, and transparent missing-data analysis help strengthen inference. Good studies can disagree because they answer different questions about frequency, severity, context, and harm Hamberger et al. (2017) Myhill (2015) Coates et al. (2026).

Evidence synthesis must also scrutinize whether an effect is directly measured or conceptually inferred. A theory of entrapment can be compelling without establishing the magnitude of each causal pathway. A statistically significant risk factor can have limited individual predictive value. An intervention that reduces depression may have no established effect on renewed assault. These distinctions motivate an explicit outcomes architecture: exposure, pattern organization, restriction, perceived fear, injury, mental health, service access, subsequent violence, and autonomy. Each warrants its own definitions, instruments, and interpretation Messing et al. (2017) Hameed et al. (2020) Stark (2023).

Construct Validity and the Measurement of Coercive Control

Construct validity begins by asking what makes a pattern coercive rather than merely frequent, unpleasant, or unequal. A defensible assessment model distinguishes the perpetrator's actions, the monitoring and sanctioning structure, the target's perceived constraints, actual restrictions in daily functioning, and the outcome of exposure. A measure based solely on a partner's perceived controlling personality can confound traits with conduct; an inventory counting surveillance without assessing consent can confound mutually agreed practices with coerced participation. Dutton and Goodman's account supplies a mechanism-centered starting point, while Hamberger and colleagues document the variable definitions that complicate comparative research Dutton & Goodman (2005) Hamberger et al. (2017).

The taxonomy used to design items determines what a study can detect. Coates and colleagues' 2026 systematic review compared psychological-IPV measures with a taxonomy covering overt and covert abuse and found substantial gaps in the representation of covert and manipulative forms. This means that a measure can exhibit high internal consistency or impressive factor fit while overlooking information control, concealment, and interference in decision making. The problem requires construct coverage studies, cognitive interviewing, expert review with survivor participation, and external criterion validation. Validation should use independent data and report the consequences of item omission, as well as statistical reliability Coates et al. (2026).

Measurement invariance matters whenever scores are compared across genders, sexual orientations, languages, ages, or disability contexts. An item concerning tracking may function differently where a couple routinely exchanges location data, while an item concerning transport deprivation may depend on public infrastructure. Researchers should test configural and metric assumptions before interpreting observed group differences as equivalent latent differences. Assessments also need to capture the time window and chronicity: a one-year exposure checklist cannot directly estimate duration of a coercive regime unless it records onset, intervals, and cessation. Repeated measures and contextual questions are therefore complementary to psychometric scales Myhill (2015) Coates et al. (2026) CDC (2026).

Research Methods: Confidential Interviews, Diaries, and Behavioral Records

Confidential separate-partner interviews are a methodological foundation when intimidation, surveillance, or retaliation may affect disclosure. Joint interviews or paired questionnaires may encourage withholding of information and expose the respondent to direct risks when answers become known. An investigator should conduct a context-specific safety assessment before contacting participants, use agreed private channels, avoid device notifications that could be detected by a controlling partner, and establish procedures for discontinuing contact. Such safeguards are both ethical requirements and determinants of measurement quality because the probability of disclosure depends on the conditions under which information is obtained WHO (2014) WHO (2013).

Life-history interviews can reconstruct the emergence and sequence of rules, sanctions, changes in contact, material restrictions, and separation-related conduct. Event calendars can anchor recollection to verifiable transitions such as changes in housing, employment, schooling, or court processes, while preserving the distinction between an independently verified record and a participant's recollection. Diary studies can provide higher temporal resolution but demand a threat model for device monitoring and data storage. Researchers should never assume that a private application or anonymized identifier makes participation safe. Designs must be tailored to the risks and practical capacities of each population Katz et al. (2020) Stark & Hester (2019).

Administrative records, clinical documentation, and police data can provide independent time stamps and injury information, but each is a selective subset of exposure. Many incidents are never reported; some interventions change whether reports are made. Increases in administrative counts after a service is introduced may therefore reflect greater access and trust rather than rising true incidence. Triangulation improves explanation when each data stream's selection mechanism is explicit. Qualitative material can characterize invisible restrictions and threat meaning, while population data estimate selected exposure categories, and linked longitudinal records test transitions and outcomes Myhill (2015) Campbell et al. (2003) WHO (2025, prevalence report).

Longitudinal Designs, Causal Inference, and Replication

Longitudinal inquiry makes the temporal structure of coercion measurable. A baseline can document resource access, social networks, intimidation history, and previous injury; follow-ups can record changes in imposed rules, employment interference, relocation, social participation, and future violence. However, follow-up loss may be associated with danger, unstable housing, or a partner's surveillance, potentially biasing estimates. Analysis should specify the causal estimand, justify time-varying confounder treatment, and report sensitivity analyses for attrition. The association of particular tactics with later harm is more informative when earlier harm and stable contextual factors are accounted for Capaldi et al. (2012) Devries et al. (2013).

Causal inference is especially difficult when exposure, protective action, and risk affect one another. A survivor may seek a protection order because threats intensify, while subsequent reports may increase because observation improves. Comparing people with and without such orders without modeling selection can create a misleading estimate of harm or effectiveness. Similar problems affect studies of employment, relocation, clinical treatment, and separation. Randomized trials can support causal conclusions about ethically assignable service packages, not about coercive violence itself. Quasi-experimental designs require transparency about their identifying assumptions and the plausibility of unmeasured confounding Sullivan & Bybee (1999) Hegarty et al. (2013) Rivas et al. (2015).

Replication should reproduce operational definitions, inclusion criteria, risk periods, and outcomes rather than merely repeat a questionnaire label. Validation in one police-referred sample does not guarantee discrimination for a general population, a shelter population, or people outside the original gender and jurisdiction categories. Pre-registration and harmonized reporting may improve confidence where feasible, but participant protection can limit open data sharing or procedural transparency. A robust research infrastructure can share de-identified instruments, analysis code, aggregate results, and documented decision rules while restricting access to potentially identifying events and disclosures Messing et al. (2017) Coates et al. (2026) WHO (2014).

Power, Fear, Injury, and Autonomy as Distinct Outcomes

A comprehensive research program should report multiple outcomes without conflating them. Exposure to a behavior, the apparent credibility of a sanction, anticipatory fear, restricted liberty, psychological symptoms, physical injury, and future reassault are analytically distinct. An intervention could increase reporting and improve safety planning without immediately changing mental-health scores; a therapy could reduce depressive symptoms while leaving the perpetrator's behavior unchanged. A single composite endpoint conceals these differences. The conceptualization of coercive control as entrapment particularly demands direct measures of practical options and everyday autonomy alongside conventional incident counts Stark (2007) Stark (2023) Hameed et al. (2020).

Fear is important but requires contextual interpretation. Some people minimize fear while adapting behavior to avoid sanctions; others fear an individual after a serious incident even when no continuous domination structure is observed. Defining coercive control solely by a fear threshold may therefore exclude certain survivors and collapse distinct trajectories. Conversely, universalizing every experience of distress as coercive control can obscure patterns requiring tailored safeguards. A good instrument captures fear, actual changes in behavior, specific threats, and their temporal link, with explicit rules for how a classification is derived Myhill (2015) Dutton & Goodman (2005) Hamberger et al. (2017).

Research should also attend to survivor-defined outcomes, including housing security, uninterrupted employment, meaningful contact with trusted people, access to information, decision-making autonomy, and freedom from coercive monitoring. These outcomes can be measured using longitudinal indicators rather than exclusively retrospective judgments. Their interpretation is affected by external conditions: a shortage of affordable housing is a safety constraint, even though it does not itself prove perpetrator control. A multilevel approach therefore partitions perpetrator tactics, institutional barriers, support availability, and observed functioning to produce actionable but scientifically distinct findings Postmus et al. (2012) WHO (2025, RESPECT) Stark (2023).

Research Ethics, Participation, and Survivor Safety

The obligation to minimize harm shapes recruitment, consent, interviewing, and dissemination. Participants may share a home, account, device, or telephone plan with a person whose behavior is being studied. Confidentiality cannot be assumed merely because names are removed from a dataset; combinations of geography, chronology, household composition, or legally recorded events can reveal identity. Ethical protocols should review safe contact, data encryption, access limitations, mandated-reporting rules, disclosure options, and locally appropriate referrals. Participants must have a meaningful ability to decline questions and discontinue without repercussions WHO (2013) WHO (2014).

Survivor involvement can improve construct specification and identify questions that standard questionnaires miss. However, consultation does not eliminate differences in experience, and a person who safely joins a research committee may not represent those facing the most restrictive surveillance. Ethical participatory design should compensate expertise, protect confidentiality, and create ways to contribute without disclosing sensitive personal histories in group meetings. The research process must also avoid presenting ordinary disagreement, cultural variation in household roles, or consensual exchange of information as abuse without evidence of coercive conditions and consequences Coates et al. (2026) Stark (2023).

The ethics of dissemination includes calibrated language about group risk. Published risk factors should inform professional judgment and safety-oriented service design without being used to predict one person's future with certainty. Findings about survivor responses must be contextualized by constrained options, and therapeutic recommendations must reflect evidence about the particular outcome measured. Responsible scientific communication reports what a study sampled, how it defined coercion, whether it measured perpetration or victimization, the direction of observed associations, and what limits remain. This precision is a practical component of harm prevention Messing et al. (2017) Campbell et al. (2003) Hameed et al. (2020).

Prevention Across Relationships, Communities, and Institutions

Prevention is a multilevel question because coercive violence occurs through behavior enacted within relationships and sustained or constrained by social conditions. The WHO-led RESPECT women framework emphasizes coordinated strategies addressing relationship skills, women's empowerment, services, poverty, safer environments, child and adolescent abuse prevention, and transformation of unequal attitudes and norms. The framework organizes evidence and priorities; it is not itself a single randomized trial. Its applicability depends on population, resources, implementation fidelity, and the outcome being evaluated. Effective prevention research therefore specifies whether it targets perpetration, victimization, coercive control, injury, service access, or broader conditions that make coercion effective WHO (2025, RESPECT) WHO (2025, prevalence report).

Educational programs may improve recognition of threats, consent, autonomy, and help-seeking, but changes in knowledge cannot be treated automatically as reductions in violence. A participant can identify abusive tactics on a questionnaire while still facing resource barriers or retaliation for seeking help. Evaluation requires appropriately timed outcomes and comparison conditions. When communities implement multiple coordinated measures, researchers must distinguish an intervention's components, exposure intensity, implementation context, and plausible mechanisms. Population-level prevention also calls for attention to housing, economic security, legal access, and institutional trust as possible modifiers of risk WHO (2025, RESPECT) Capaldi et al. (2012).

Prevention also requires theories of how harmful behavior changes. Approaches emphasizing attitudes, self-regulation, or conflict-management skills may address some risk pathways but may be insufficient for a relationship in which intimidation and material restrictions serve an ongoing strategy of domination. A conceptual mistake arises when a service designed for reciprocal conflict is assumed appropriate for ongoing coercive threat. Mechanism-focused research should therefore document baseline coercive patterns, track specific tactics, and separately evaluate survivor safety and perpetrator behavior. The use of a program label or completion certificate is not proof of reduced harm Johnson & Leone (2005) Dutton & Goodman (2005) Stark (2023).

Advocacy, Material Support, and Controlled Intervention Evidence

Advocacy interventions attempt to increase access to services, legal information, housing, financial resources, and supportive networks. Sullivan and Bybee's randomized study of community-based advocacy followed women after shelter residence and found that sustained advocacy could improve resource access and reduce violence exposure in the studied conditions. Its contribution is evidence about a concrete support model with a comparison group. The result should be interpreted in its historical service context and population rather than generalized to every survivor, modality, or legal system Sullivan & Bybee (1999).

The Cochrane review by Rivas and colleagues (2015) evaluated advocacy interventions and judged effects to vary across settings and outcomes, with many comparisons affected by uncertainty. Advocacy may improve some psychosocial and safety-related outcomes while evidence for sustained violence reduction can be less clear. Studies also differ in intervention intensity, comparator, participant populations, and length of follow-up. A scientifically useful review states which endpoint improves, the certainty of that finding, and the likelihood that implementation constraints would alter it. Linking advocacy to resource access is theoretically coherent, but causal pathway claims require appropriate mediation evidence Rivas et al. (2015) Postmus et al. (2012).

Material assistance can affect the feasible choices available to someone under control. Safe housing, flexible financial help, transportation, childcare, and communication privacy may change practical alternatives even before symptoms improve. Evaluation should record access to these resources and whether a partner can interfere with them. The impact of an intervention is likely modified by local housing markets, employment conditions, disability services, and the availability of confidential support. Economic autonomy and reduction in coercive surveillance are independently meaningful endpoints, and should be followed after program completion WHO (2025, RESPECT) Postmus et al. (2012) Stark (2007).

Health-System Response and Psychological Therapies

Health services encounter the consequences of IPV through injury, reproductive health concerns, sleep disturbance, depression, anxiety, and trauma-related symptoms. WHO clinical guidance advocates confidential, person-centered responses, attention to immediate safety and health needs, and knowledge of appropriate referral pathways. A clinical encounter should not treat a disclosure of coercion solely as a symptom inventory or assume that a psychological diagnosis explains the behavior of the abusive partner. The scientific objectives include accurate recognition of exposure, compassionate clinical care, access to services, and evaluation of the patient's subsequent health and safety outcomes WHO (2013) WHO (2014).

Hameed and colleagues' Cochrane review (2020) found that psychological therapies may reduce depression in women exposed to IPV, while evidence for other outcomes, including anxiety, PTSD symptoms, and continuing violence, was more limited or uncertain across the evaluated studies. The distinction matters. A therapy can address trauma-related suffering without altering the perpetrator's conduct or external threats. Trials should report diagnostic and symptom outcomes separately from exposure to abuse, resource access, and autonomy. The effectiveness of particular therapies must be assessed within the populations and comparators actually studied Hameed et al. (2020) Oram et al. (2017).

Primary-care screening and counseling are separate intervention questions. The WEAVE cluster randomized trial investigated screening and counseling for women experiencing IPV in primary care; its findings caution against assuming that simply adding screening yields a broad reduction in violence or better quality of life. Implementation should be evaluated as a package of safe identification, response quality, support availability, and follow-up. Universalized slogans about screening cannot replace an assessment of privacy, care pathways, resources, and evidence for the actual outcome sought Hegarty et al. (2013) WHO (2014) Rivas et al. (2015).

Responses to Perpetration, Joint Interventions, and Safety Boundaries

Programs addressing perpetrators vary in theoretical orientation, target population, delivery format, monitoring, and outcome definitions. Changes in attitudes or attendance can be measured reliably without establishing change in violent behavior. The research design must specify whether outcomes are self-reported, partner-reported, police-recorded, or independently assessed, and whether outcomes include coercive restrictions as well as physical assault. Attrition, mandated participation, selective reporting, and differences in follow-up create obstacles to causal interpretation. A mechanism-focused program should address intimidation, surveillance, entitlement to control, and material interference when these are part of the presenting pattern Dutton & Goodman (2005) Johnson & Leone (2005) Hamberger et al. (2017).

Couple-focused interventions raise a distinct design and safety issue. Dyadic conflict therapies often presume that participants can disagree and disclose relevant information without retaliation. A relationship defined by systematic coercion may violate that assumption. Research and practice therefore need confidential separate assessment for coercive threats and power asymmetry before interpreting a couple as a unit of mutually responsible conflict. The presence of strong feelings, reconciliation, or stated willingness to attend together does not establish conditions for safe joint disclosure. The appropriate professional response depends on individual safety assessments and established clinical protocols WHO (2014) Stark & Hester (2019) Johnson & Leone (2005).

Comparative intervention research should stratify by coercive pattern, prior injury, severity, and study context rather than merely use the undifferentiated outcome 'domestic violence.' Differences between situational conflict and coercive domination affect mechanisms, consent to participation, and the interpretation of improvement. Ultimately the relevant endpoint is a durable reduction in imposed restrictions and harm, with an increase in safety and effective autonomy. Attribution to an intervention requires a defensible comparison and time frame; reductions in reported incidents can instead reflect displacement into less observable tactics or diminished capacity to report Myhill (2015) Coates et al. (2026) Rivas et al. (2015).

Human–Artificial Mediation and Emerging Digital Risks

Artificial systems increasingly mediate communication, household organization, location services, search, and access to information. They may function as resources that support independent decisions or as instruments incorporated into coercive monitoring. The scientific question concerns how access rights, logging, notification policies, account recovery, and automated inference redistribute practical control among relationship participants. A shared assistant, connected-device platform, or predictive monitoring system can expose private behavior even without a traditionally defined spyware application. These scenarios warrant evaluation of the sociotechnical arrangement, consent, and enforceable constraints rather than treating the presence of artificial intelligence as a psychological cause of abuse Kim & Ferraresso (2023) Sam et al. (2026).

Within the Artificial Era, a rigorous inquiry into human–artificial relations examines the effects of Artificial mediation on autonomy, privacy, informational asymmetry, and safety. Claims concerning a system's agency or psychological identity belong to a distinct conceptual and empirical inquiry. The directly testable questions here are whether particular affordances enable coercive surveillance, whether users can withdraw permissions without retaliation or loss of essential services, and how automated recommendations affect access to independent advice. Studies require careful threat modeling, secure datasets, informed consent, and protection against the involuntary exposure of survivors' communications Kim & Ferraresso (2023) Stark (2023).

A design principle follows from the evidence: the unit of evaluation is the person's feasible control over data and decisions. Private mode, shared-account separation, selective notification, and meaningful revocation may be relevant technical safeguards, but their effectiveness is an empirical question. Interfaces that advertise consent can still create power asymmetry when one party possesses administrator privileges or physical access to devices. Interdisciplinary research should combine IPV scholarship, privacy engineering, human-computer interaction, and independent safety evaluation, and explicitly state whether a finding concerns usability, disclosure, actual victimization, or long-term autonomy Sam et al. (2026) Coates et al. (2026) Dutton & Goodman (2005).

Open Scientific Problems and a Cumulative Research Agenda

The field still lacks one uniformly validated measure that captures behavioral tactics, credible sanctions, sustained entrapment, and restrictions on autonomy across populations. Important tasks are developing item banks with transparent construct maps; validating duration and pattern indicators; testing measurement invariance; comparing partner-reported, survivor-reported, and independently documented outcomes; and establishing how much information is lost when questionnaires count isolated incidents. Qualitative work can clarify covert tactics, while prospective studies test whether revised measures predict future harm better than simpler scales. External validation must be reported by population and purpose Coates et al. (2026) Hamberger et al. (2017) Myhill (2015).

Causal questions concern sequences and feedback. Which economic restrictions produce durable dependency, which social network changes mediate intimidation, and how do digital surveillance opportunities alter the anticipation of sanctions? How do institutional interventions interact with the behavior of perpetrators and the options available to survivors? Answering these questions requires time-resolved, privacy-preserving methods and transparent assumptions about selection, confounding, and missing data. Injury-based and homicide-based risk prediction should continue to be distinguished from research on psychological coercion and reduced liberty because their endpoints and prevalence differ greatly Campbell et al. (2003) Messing et al. (2017) Postmus et al. (2012).

Intervention science must also distinguish symptom relief, resource access, prevention of reassault, reduction in coercive conduct, and sustained restoration of autonomy. Trials need sufficiently long observation, prespecified outcomes, safeguards against retaliation, appropriate comparison conditions, and reporting of harmful as well as beneficial effects. Population coverage should extend beyond samples most easily reached by shelters or telephone surveys, including people whose device access, language, disability, legal status, or housing circumstances affect recruitment. Transparency must be balanced with the confidentiality necessary for scientific participation Sullivan & Bybee (1999) Rivas et al. (2015) Hameed et al. (2020) WHO (2014).

Conclusion

Relational power becomes a scientific problem of coercion when behavior, sanctions, and dependency systematically organize another person's choices. Coercive control is best analyzed as a temporally extended pattern whose observable components include demanded compliance, monitoring, restriction, credible threats, and their consequences for liberty. Intimate partner violence encompasses overlapping physical, sexual, and psychological behaviors; the relationship between an individual incident and an enduring coercive regime must be established rather than presumed. The history of research demonstrates why incident counts, coercion models, control typologies, and longitudinal developmental explanations answer partially different questions Dutton & Goodman (2005) Johnson & Leone (2005) Stark (2023).

The most persuasive evidence emerges from combining carefully defined population surveillance, contextual interviewing, longitudinal observation, risk studies, psychometric validation, and controlled evaluation of interventions. The evidence establishes serious risks to physical health and mental well-being, substantial heterogeneity among relationships, and important benefits and limitations of advocacy and psychological support. It also reveals persistent gaps in measurement, subgroup coverage, causal mechanism testing, and evaluation of safety and autonomy as distinct outcomes. Numerical estimates and predictive tools have meaning only within their original operational definitions and populations WHO (2025, prevalence report) CDC (2026) Campbell et al. (2003) Coates et al. (2026) Hameed et al. (2020).

The research program that follows is explicit: describe the conduct, document the conditions under which it constrains real alternatives, evaluate its consequences, test competing explanations, and investigate interventions against outcomes that matter to the people affected. Digital and Artificial mediation add new configurations of information and access control to this existing scientific problem and invite the same methodological discipline. A mature psychology of relational power explains how coercion is organized and how safety and autonomy can be measured, protected, and recovered under concrete social and technological conditions Stark & Hester (2019) Kim & Ferraresso (2023) WHO (2025, RESPECT).

References