RESEARCH DOSSIER
Therapeutic Alliance, Rupture, and Repair
Abstract. Therapeutic alliance is the evolving collaborative organization of patient–therapist goals, tasks, and bond. This fundamental research monograph reconstructs its psychoanalytic and transtheoretical origins, analyzes measurement and observational approaches, and distinguishes alliance strength from directly observed rupture, attempted repair, and demonstrated resolution. It synthesizes original instrument validation studies, dyadic and repeated-measures investigations, large meta-analyses, and contemporary evidence through 2026. The adult alliance–outcome relation is reproducible, whereas inferences about causation and the efficacy of repair training require more discriminating designs. Direct rupture research finds an adverse association with outcomes in a small evidence base, and repair studies support a positive association between resolution and outcome while leaving intervention-specific mechanisms unsettled. The analysis further examines developmental, cultural, teletherapy, ethical, and human–artificial comparison contexts. Its central conclusion is that alliance should be studied as a time-varying interactional process through multiple reporters, directly observed events, explicit causal models, and meaningful clinical outcomes.
Keywords: therapeutic alliance; working alliance; psychotherapy process; alliance rupture; rupture repair; confrontation; withdrawal; therapeutic relationship; psychotherapy outcomes; patient agency; alliance measurement; Working Alliance Inventory; Rupture Resolution Rating System; longitudinal process research; common factors; psychotherapy research.
Scientific Overview and Research Questions
Therapeutic alliance is the organized cooperation through which a person seeking psychological treatment and a clinician establish a workable relationship, agree upon therapeutic purposes, and undertake methods intended to produce change. It is simultaneously an experience of being recognized, an agreement about action, and an evolving interpersonal process. The alliance is observed in ordinary clinical encounters through the patient's opportunity to state goals, the therapist's efforts to make treatment intelligible, and their capacity to revise understandings when either participant experiences friction. In the influential formulation of Bordin (1979), goals, tasks, and bond constitute the basic components of a working alliance.
The central questions of this monograph concern its construct validity, origins, measurement, dynamics, and consequences: How does collaboration differ from interpersonal liking? What qualifies as an alliance rupture? Which patterns of withdrawal and confrontation can be observed? Under which circumstances is a repair successful? How does one establish that alliance contributes to clinical improvement when symptomatic relief itself may improve the alliance? What should be concluded from disagreements between patient, clinician, and independent observers? Finally, how do power, cultural context, digital mediation, and different developmental situations modify these relationships? The organization follows these questions from original theory to direct observational studies, longitudinal models, meta-analytic synthesis, and methodological criticism.
This is a critical narrative research monograph based on identified original articles, measurement studies, and secondary evidence syntheses; it does not claim a PRISMA systematic search, unpublished participant data, a new clinical trial, or a quantitative meta-analysis. Empirical statements are tied to identifiable publications. The evidentiary hierarchy distinguishes cross-sectional associations, temporally ordered observational models, experimental manipulation, professional theory, and speculation. The subject belongs primarily to Counseling Psychology and Psychotherapy, intersecting with clinical psychology, relational processes, social cognition, and research methods. The object of explanation remains the therapeutic relationship as enacted between people within a particular treatment setting.
Origins of the Working Alliance Concept
The modern concept of alliance developed from psychoanalytic distinctions concerning the collaborative relationship and the clinical relevance of a patient's capacity to work with a practitioner. Different traditions emphasized the ordinary relationship, transferential phenomena, treatment commitment, and the rational capacity to participate in therapeutic tasks. These antecedents matter because the term alliance inherited several meanings that are sometimes compressed into one rating. A favorable interaction, the experience of trust, a negotiated treatment project, and a patient's attachment to a therapist are related dimensions with different causal implications.
Bordin (1979) reorganized this inheritance into a broadly applicable formulation rather than a technique restricted to a single psychotherapy school. Agreement about goals specifies what improvement or development is sought; agreement about tasks concerns the practical means of therapy; bond denotes the affective connection that sustains cooperation. The account was explicitly proposed for different modalities, with the expectation that alliances would vary according to therapeutic demands. It was a conceptual proposition rather than the result of a large randomized test establishing a universal mechanism.
Subsequent research translated the concept into measures, testable predictions, clinical formulations, and longitudinal designs. Horvath (1989) published a foundational measurement program for the Working Alliance Inventory. Safran (2000) developed a relational approach in which cooperation becomes particularly intelligible when it falters. The growth of meta-analytic outcome research and rupture coding later made it possible to compare the predictive value of alliance ratings with the immediate, visible interactional processes by which strains emerge and are addressed.
The Three-Part Structure of Alliance
Agreement on goals concerns more than a patient's verbal assent to a diagnosis or a professional's treatment proposal. It involves the practical recognition that the work is oriented toward outcomes that the patient can understand and regard as relevant. Goals may include symptom relief, greater functional participation, relational change, or improved self-management, depending on the treatment and the person's circumstances. Goal alignment requires attention to priorities that can evolve over time. A therapist can accurately describe a clinical problem and still be working toward a priority that does not reflect the patient's reasons for seeking help.
Agreement on tasks concerns the procedures used in pursuit of those goals. Depending on the treatment, tasks may involve practicing skills, discussing memories, observing present emotion, exposure exercises, behavior change, or examining interpersonal exchanges. The same task can feel appropriate when its rationale is understood and alienating when it seems imposed. Task agreement is therefore operationally different from mere attendance or compliance. The therapist and patient can disagree about an assignment while preserving a meaningful bond, and such disagreement may be an informative opportunity for collaborative revision.
The bond refers to perceived trustworthiness, mutual respect, emotional connection, and the sense that the relationship can sustain difficult work. It is neither identical to intimate attachment nor reducible to clinician warmth. A bond can support disagreement and can become more credible when conflict is addressed carefully. These dimensions were intended to clarify a single working relationship, although psychometric analyses must test rather than assume their separability. The original Working Alliance Inventory validation already identified high correlations among its subscales as a question for construct differentiation (Horvath (1989)).
Alliance as an Interactional and Institutional Achievement
Alliance is located in a particular dyad, yet its creation depends on the procedures and affordances of the surrounding service. Referral routes, appointment schedules, economic barriers, insurance constraints, access to interpretation, professional authority, clinical record systems, and the duration of treatment influence what patients and therapists can negotiate. A patient's willingness to return may reflect both interpersonal safety and practical affordability. A weak rating may reveal bureaucratic frustration, a mismatch of clinical goals, or a genuinely unhelpful interpersonal pattern. Scientific interpretation requires separating these explanations.
The dyadic unit is particularly important because patient and therapist can experience the same conversation differently. A clinician may experience a firm instruction as appropriately structured, while the patient experiences it as a loss of agency. A patient may describe a meeting as helpful while the clinician notices unspoken resistance. Such discrepancies are data about the interaction rather than automatically evidence of an error by either participant. Zilcha-Mano (2016), using a two-person analytic framework, showed why within-treatment changes in different reporters' alliance ratings carry distinct predictive information.
The institutional level also determines whether clinicians have time, supervision, and authority to respond to strain. Policies that treat disagreement as noncompliance can obscure relational information. Practices that permit structured feedback can make disagreement observable. These propositions are mechanisms for investigation, not proof that any particular service model has superior outcomes. The Wampold (2023) review extends the alliance discussion to mental health care and describes plausible relationship pathways, while emphasizing the breadth of contexts in which collaborative care is formed.
Distinguishing Alliance From Adjoining Constructs
Therapeutic alliance overlaps with empathy, trust, expectations, attachment, treatment satisfaction, adherence, and the real relationship. The overlap complicates measurement. Empathy concerns understanding the patient's experience; trust concerns expectations of another's conduct; alliance includes the collaborative agreement on what will be done. Treatment satisfaction is a retrospective appraisal that can follow improvement. Adherence concerns whether prescribed methods are performed. Attachment can influence the relationship but extends well beyond the treatment contract. Treating all these terms as interchangeable makes mechanism testing circular.
A patient might trust a clinician's competence while rejecting the recommended goal; another may feel personally understood while questioning the relevance of a structured exercise. Conversely, an effective working agreement can exist in a formal setting without high affective expressiveness. These cases explain why a global relational quality score cannot by itself locate the source of strain. They also explain why alliance must be distinguished from treatment outcome: improvement can generate favorable alliance ratings, making contemporaneous correlations vulnerable to reverse explanation. Zilcha-Mano (2017) identifies these causal alternatives as central to the empirical debate.
The term therapeutic relationship should therefore be treated as a broader umbrella containing alliance and additional interpersonal processes. The more precise the research question, the more precise the operationalization required. Investigations of rupture rely on momentary interactional indicators; investigations of treatment engagement may use attendance, homework, or attrition; investigations of perceived bond may use patient report. No one instrument can settle all three without additional validation.
Operationalizing Alliance in Psychotherapy Research
Operationalization turns an interpersonal theory into observations that can be compared. Self-report questionnaires ask patients or therapists about collaboration and emotional connection; observer instruments code speech, nonverbal behavior, or interactional sequences; session-by-session procedures follow change over time. These methods address partially different constructs. Retrospective ratings aggregate experience across a session and can be influenced by perceived symptom change. Observer coding detects local events but depends on a manual, training, observation quality, and cultural competence. Both kinds of data can be valuable when their limitations are retained in analysis.
Time of measurement changes meaning. A rating after the first encounter reflects the first impression and the credibility of the proposed work. A late-treatment rating may reflect accumulated experiences of success, disappointment, or resolution. Average alliance across all sessions conceals disruptions, especially when most ratings are near the top of a scale. Investigators should distinguish early alliance, mean alliance, within-person alliance fluctuations, and the alliance trajectory itself. These are analytically separate predictors with distinct risks of overlap with outcome measurement.
The wider evidence base shows that a positive alliance–outcome association is detectable across many measurement formats (Flückiger (2018)). This result does not imply that the measures are psychometrically equivalent. Correlation stability may coexist with uncertainty about the causal ingredients. Sensitive research should report instrument version, reporter, assessment timing, missingness, clinical population, and precise outcome interval instead of merely stating that alliance was measured.
The Working Alliance Inventory and Its Interpretive Limits
Horvath (1989) introduced and evaluated the Working Alliance Inventory through a program of preliminary validity and reliability studies. Its design made Bordin's proposed goals, tasks, and bond dimensions accessible to standardized comparison. It became important because researchers working in otherwise different psychotherapy traditions could examine a common construct. Measures of the alliance can be completed from patient or therapist perspectives, and shortened versions have supported repeated measurements where lengthy questionnaires would be burdensome.
A questionnaire score is a model-dependent summary. High internal consistency establishes that items respond coherently; it does not demonstrate that subscale scores correspond to three independent processes. A one-time measurement cannot describe a rupture sequence that unfolds during a conversation. A session rating cannot determine whether a client agreed freely, whether a difficult discussion was safe, or whether apparently positive responses conceal reluctance. Construct validity is strengthened when self-reports converge with behavior, change over time, and independent outcome evidence, especially when distinctive predicted patterns can be observed.
There is also a methodological asymmetry between a measure's original validation sample and the diverse populations in which it may subsequently be used. Language, cultural norms about disagreement, age, treatment format, and clinician authority can alter response processes. Contemporary studies should test factor structure, measurement invariance, responsiveness, and ceiling effects instead of treating the presence of an established acronym as sufficient validation for every new context.
Observational and Negotiation-Based Instruments
Observational approaches make the alliance measurable at a finer temporal resolution. The Eubanks (2019), codes recognizable markers of confrontation and withdrawal and strategies through which therapists attempt to address them. Its validation study evaluated early sessions from 42 cognitive behavioral therapy cases. Trained coders showed high agreement for several frequency and resolution indices, and selected ratings predicted premature termination; the results support the instrument's utility while leaving generalizability to other settings open.
The Doran (2016) moves beyond a static perception of positivity toward the capacity of participants to negotiate divergent preferences or experiences. Negotiation matters because the most useful therapeutic response to discord may be to discover why a goal or exercise seems unsuitable. An observational code can document what the therapist attempted, whereas a patient rating can indicate whether that attempt felt satisfactory; ideally a process study integrates both rather than substituting one for the other.
The unit of analysis must remain explicit. A rupture marker is not a diagnosis; a rated repair strategy is not proof of repair; and a strong coder agreement statistic does not establish that a detected event causes subsequent clinical change. Observer ratings can be especially sensitive to training examples and cultural interpretations of silence, assertiveness, deference, or emotional expression. Reliability, criterion validity, ecological validity, and causal evidence are separate research questions.
What Constitutes a Therapeutic Rupture?
An alliance rupture is a meaningful deterioration or strain in the working relationship concerning goals, tasks, or bond. It can be brief and subtle or prolonged and overt. The person seeking treatment may question a method, experience being misunderstood, withdraw from exploration, or contest the therapist's intervention. The therapist may become defensive, overly controlling, avoidant, or inattentive. The defining feature is the disruption of collaboration as enacted in the encounter, not the mere expression of difficult feelings or the presence of a disagreement.
Rupture is therefore a relational and temporal concept. A direct expression of anger can strengthen collaboration when it is recognized and explored; acquiescence can conceal a serious disconnection. Similarly, a demanding clinical task does not inherently constitute a rupture. The relevant question is how the task is presented, understood, negotiated, and experienced by the participants. The empirical rupture literature distinguishes visible confrontation from withdrawal precisely because strains can be manifested through different behaviors (Eubanks (2018)).
The terms rupture, repair, and resolution must also be separated. Rupture names the strain; repair names actions intended to restore or revise collaboration; resolution describes the degree to which the difficulty has been productively addressed. A therapist's acknowledgment is an observable attempt and may be appropriate, yet the patient's sense of understanding and subsequent engagement determine whether the episode is plausibly resolved. Research measures should record these stages independently whenever possible.
Confrontation and Withdrawal as Distinct Rupture Forms
Confrontation rupture markers communicate opposition or dissatisfaction comparatively directly: challenging a therapist's competence, questioning an intervention, expressing anger toward the clinician, or rejecting a treatment rationale. Their clinical meaning depends on context. Such statements may also be expressions of justified boundary-setting or a patient's attempt to make therapy more collaborative. A coding system should capture the event and its relational context without assuming that all criticism is pathology or that opposition is equivalent to therapeutic failure.
Withdrawal rupture markers may include minimal responses, excessive agreement, topic changes, intellectualized detachment, or disengagement from an emotionally important subject. These are potentially ambiguous behaviors. Silence may express thoughtful reflection, language-processing needs, fear, fatigue, or culturally patterned conversational style. It becomes evidence of an alliance rupture only in conjunction with the interpersonal sequence and interpretive information that makes disengagement a defensible inference. Observer ratings therefore require explicit contextual criteria.
The distinction has practical scientific value because a confrontation episode is easier to detect from direct language, whereas withdrawal can be mistaken for cooperation. In the Eubanks (2019), operational categories permitted reliable coding and revealed associations of some rupture indicators with dropout. The data do not warrant inferring that every confrontational act predicts discontinuation or that every instance of withdrawal has the same origin. Accurate analysis must preserve event frequency, intensity, timing, duration, and whether subsequent negotiation followed.
Alliance Development Across Sessions
Alliance is a process distributed over time. Early sessions establish expectations about the therapist's credibility, the work's purposes, and the patient's ability to exercise influence. Middle sessions can reveal discrepancies that were previously invisible, especially when treatment becomes more demanding or the patient's circumstances change. Later sessions bring new questions about progress, termination, and the meaning of the relationship. A single score at one point can predict subsequent functioning without representing the shape of this developmental sequence.
Models of alliance trajectories examine average levels, slopes, abrupt changes, and individual fluctuations. A favorable trajectory can emerge through stable cooperation, renewed agreement after a disruption, or increasing clarity about difficult work. Apparent improvement in alliance can also accompany symptom improvement, making temporal order crucial. Flückiger (2020) analyzed early session-level individual participant data from 17 primary studies in nine countries, comprising 5,350 participants. They found reciprocal within-person relationships between alliance and symptoms in the first seven sessions. The evidence favors a dynamic account over the assumption that an early alliance score acts independently of the treatment's subsequent effects.
Temporal granularity is a methodological decision with theoretical consequences. Momentary rupture coding captures a disturbance as it unfolds, while a questionnaire given after every fifth session may miss brief disagreements and their repairs. A weekly measure can detect slower shifts but may aggregate opposite events. Researchers should define whether the intended process operates over seconds, sessions, treatment phases, or months and choose measurement intervals accordingly.
Repair as Relational Negotiation
Repair is the coordinated process by which therapist and patient recognize an interruption in their working relationship, explore its meanings, and re-establish or revise a collaborative path. The goal of repair is substantive agreement rather than the appearance of interpersonal harmony. A patient may conclude that a method is unsuitable, and an effective repair may involve replacing that method. A therapist may discover that an interpretation felt premature and revise the pace of work. The ensuing collaboration is new because it incorporates information revealed by the rupture.
Safran (2000) articulated a relational framework involving attention to the present interaction and explicit discussion of the therapeutic process. Their model treats the relationship itself as something that can be explored rather than an invisible platform for techniques. The historical significance of this account is its clarification of a clinical sequence that later research could operationalize: noticing the event, identifying mutual contributions, exploring the patient's experience, negotiating meanings and goals, and monitoring the aftermath.
The process is not mechanically linear. An acknowledgment may lead to further criticism; the patient may need time before articulating a concern; an attempted explanation may intensify the strain. An investigator should therefore distinguish attempted repair, patient-perceived repair, observer-coded resolution, and durable recovery in subsequent sessions. Theoretically promising strategies require empirical evaluation of their actual outcomes rather than endorsement solely because they are compatible with a relational model.
Metacommunication and the Process of Naming Discord
Metacommunication refers to talking about what is occurring between therapist and patient as an interaction. When disagreement about an assignment is repeatedly deflected, an explicit inquiry into how the two participants are understanding each other can change the focus of work. This process potentially reveals mismatched assumptions about goals, hierarchy, emotional safety, or the value of a proposed intervention. Metacommunication is particularly useful as a research construct because it can be identified from an actual conversational sequence rather than inferred solely from post-session satisfaction.
Its beneficial potential depends on implementation. A clinician's declaration that the relationship has been ruptured may feel intrusive or invalidating if the patient does not share that description. Similarly, repeatedly directing attention toward the therapy relationship may conflict with a patient's priorities. Repair research should identify the interactional conditions under which discussion of the relationship is invited, accepted, contested, or experienced as unnecessary. It should examine whether metacommunication changes task agreement, emotional safety, or symptom trajectories beyond the mere occurrence of therapist speech about the relationship.
The Eubanks (2019) provides a vocabulary for studying resolution strategies, while Doran (2016) addresses how disagreement is handled. These approaches can be integrated in future multi-method designs. Their different observational targets should remain distinct: naming disagreement is a therapist action, negotiation is a dyadic activity, and resolution is an evaluative conclusion about what subsequently changed.
Therapist Responsiveness and Contribution to Rupture
Therapists contribute to alliance formation through explanations of treatment, emotional attunement, responsiveness to disagreement, and the ability to adapt interventions while maintaining clinical coherence. Therapists also participate in strains through timing, misunderstanding, defensiveness, inflexible use of technique, or failure to acknowledge professional power. A relational study must therefore permit therapist conduct to be an explanatory variable rather than treating rupture exclusively as a patient characteristic.
The Eubanks (2019) validation study included ratings of therapists' contributions to ruptures and found that selected observational indices predicted dropout in the sample investigated. Such findings establish an empirically tractable research question: When is the therapist's response associated with continuing treatment, and when does the therapist's own conduct intensify the episode? The study's modest case count and specific treatment context delimit its generalization; it provides neither a universal taxonomy of misconduct nor a randomized estimate of the benefit of one corrective response.
The question connects to therapist effects, competence, and supervision, which form a related but scientifically separate dossier. In the present analysis, therapist skill matters specifically insofar as it influences the dyad's collaborative processes. Investigators should distinguish interpersonal facilitation from technical adherence, since a clinician can faithfully implement a protocol while responding poorly to an individual's concerns. Treatment quality requires both the appropriate delivery of interventions and defensible attention to the person's experience of them.
Patient Agency, Preference, and Consent
Patient agency is the capacity to articulate purposes, question procedures, and influence the course of treatment within real clinical constraints. Alliance research should register agency because an apparently smooth working relationship can reflect genuine collaboration, acquiescence to authority, or lack of accessible alternatives. An agreement is scientifically meaningful when the patient understands the work sufficiently to participate and can express reservations without being punished for doing so. In voluntary care, withdrawal from an unsuitable relationship can also represent an informed action, rather than a simple failure of alliance.
A longitudinal analysis by Huber (2021) followed 386 patients in psychodynamic outpatient psychotherapy and found that within-person changes in alliance and therapeutic agency were related to later symptom improvement, although their mutual relationships varied across assessment occasions. Theoretical accounts of negotiation connect alliance to epistemic responsibility: each participant has access to different information. The patient knows the felt meaning and situational relevance of an intervention; the clinician contributes specialist knowledge about methods and risk. Productive cooperation integrates these perspectives within professional obligations. The relationship may become stronger through an explicitly managed disagreement because that exchange demonstrates that the patient's voice can modify the treatment project. This possibility is compatible with Bordin (1979), but its effect in any particular setting must be established empirically.
Investigators can measure agency through participation, preference congruence, observed opportunities to disagree, and repeated patient accounts. These constructs require separate instruments from simple liking or warmth. Recording only the clinician's judgment of engagement risks misclassifying disagreement as resistance. Recording only the patient's satisfaction risks missing constraints that discourage candid criticism. The research task is to specify how genuinely negotiated agreement can be observed and whether changes in agency precede changes in alliance or symptoms.
Mechanisms of Change and Competing Models
Several explanatory accounts can accommodate the observed relation between alliance and treatment outcome. A facilitation account proposes that collaboration permits effective engagement with specific techniques. A relational-change account proposes that experiences of being understood and negotiating discord are themselves therapeutically consequential. An expectancy account proposes that trust and a credible treatment rationale influence hope and engagement. A selection account proposes that patients with favorable prognostic characteristics more readily develop strong alliances. A reverse-causality account proposes that early symptom relief produces more positive evaluations of the therapist and the relationship.
These hypotheses have different empirical signatures. If alliance simply enables the use of an effective method, analyses should examine mediation through exposure to or participation in that method. If relational repair is itself a change process, observed repair should precede relevant outcomes after accounting for prior changes. If the association is mainly due to patient characteristics, adjustment for intake factors should markedly attenuate its magnitude. If symptomatic relief drives ratings, lagged symptom measures should predict later alliance independently of prior alliance. These possibilities can operate jointly and may vary across disorders, therapists, and stages.
Wampold (2023) synthesize relationship pathways they describe as CARE, EXPECTANCY, and SPECIFIC, making explicit the importance of human connection, treatment credibility, and intervention processes. Zilcha-Mano (2017) examines what is required before the alliance can be regarded as therapeutic in its own right. Together these works show why theoretical pluralism must be paired with strong causal designs rather than an unsupported choice of one preferred mechanism.
Alliance and Technique as Interdependent Processes
Specific interventions and relational collaboration are often measured separately for analytic convenience. In clinical practice they frequently shape each other. The rationale for a behavioral exercise can strengthen goal agreement; successful completion of a task can improve confidence in the therapist; an unhelpful interpretation can damage the bond while leaving the clinician technically adherent to a model. A scientifically adequate account should study these dependencies rather than force all variance into a binary distinction between common factors and specific techniques.
The evidence for a stable alliance–outcome association across treatment orientations has repeatedly been taken to indicate a broad treatment process. Flückiger (2018) examined multiple approaches and outcome measures; this supports the generality of the association in studied settings. It does not show that therapeutic methods are interchangeable, that alliance effects are constant in every clinical population, or that the effectiveness of specific interventions can be inferred from a strong bond. Treatment-specific mechanisms and alliance processes can have independent, interacting, or sequential influences.
Designs that include technique adherence, therapist competence, patient expectations, alliance, and time-varying symptoms help separate these explanations. Yet statistical adjustment has limitations when variables occur in a causal sequence; controlling a true mediator can understate a total effect, while conditioning on a collider can introduce distortion. Thus even sophisticated models require an explicit causal diagram and transparent timing assumptions.
Meta-Analytic Evidence for Alliance and Treatment Outcome
The quantitative foundation is the meta-analysis by Flückiger (2018). It synthesized 295 independent studies covering more than 30,000 patients and reported an overall face-to-face alliance–outcome correlation of r = .278, with a 95% confidence interval of [.256, .299]. The association was positive across many treatment orientations, countries, measurement practices, and patient categories; heterogeneity was nevertheless present. This estimate is an average relation, not a causal effect of raising one person's alliance score by a specified number of points.
A later Flückiger (2020) examined 125 effect sizes in 60 independent samples comprising 6,061 participants. After accounting for measured intake or simultaneous treatment-process variables, reported correlations remained in the range r = .23 to .31. This finding weakens the claim that the entire observed relationship is reducible to the particular confounders studied. It cannot eliminate unmeasured confounding or prove that an experimentally created improvement in alliance would yield an equivalent improvement in symptoms.
The size and relative consistency of the association justify attention to alliance in psychotherapy process science. However, interpretation must be disciplined: correlations near .28 correspond to a modest share of individual-level outcome variation; many patients improve in treatments whose recorded alliance is less favorable, while some with excellent reported alliance have little improvement. Statistical heterogeneity and the identity of the reporter require consideration before translating a pooled effect into a clinical expectation.
Reciprocal Temporal Effects and Causal Identification
A central advance in psychotherapy process research has been the examination of within-patient temporal relationships. Between-person models show that people who report higher alliance also tend to report better outcomes; within-person models ask whether a person fares differently after a session in which the alliance is better or worse than that person's usual level. The distinction matters because stable differences in motivation, symptom severity, trust, or therapist selection can confound the between-person association.
The Flückiger (2020) found reciprocal early-session effects between alliance and symptoms. This is compatible with the possibility that alliance facilitates improvement and that improvement, in turn, strengthens collaboration. Zilcha-Mano (2016) studied 241 dyads over as many as 30 sessions and identified the importance of within-treatment patient ratings for session outcomes. These designs meaningfully improve temporal explanation but remain vulnerable to processes that change between measurement occasions and to shared-method influences.
A rigorous causal claim would ideally combine experimentally manipulable aspects of relationship practice, carefully timed process measures, validated outcomes, and an account of possible adverse effects. Ethical constraints limit randomization to harmful relational behavior, so natural experiments, training interventions, micro-randomized feedback prompts, and instrumental or quasi-experimental approaches may contribute evidence where their assumptions can be justified. Temporal precedence is necessary for many causal models, but time order alone is insufficient to establish causation.
Direct Evidence on Rupture and Outcome
The study of rupture requires measures of actual strain rather than deductions from a low alliance score. The distinction is decisive: a questionnaire decline could reflect a disagreement, symptom deterioration, fatigue, disillusionment with treatment, or ordinary response variation. Direct methods code an interaction, ask explicitly about rupture, or identify a conflict episode. These methods allow researchers to estimate whether ruptures and later outcomes are associated while maintaining conceptual separation between event and outcome.
A recent contribution is the Babl (2026), which restricted inclusion to directly measured alliance ruptures. Four publications contributed 18 effect sizes from 301 patients. The pooled association was r = −.21, indicating that more severe or frequent measured ruptures tended to accompany worse treatment outcomes in those studies. The analysis used a three-level structure to account for multiple effect sizes nested within studies. Its importance lies in a more focused construct definition and contemporary quantitative synthesis.
The evidentiary ceiling is equally important. Four studies provide a limited basis for examining moderators, population differences, or the consequences of successful repair. The meta-analysis itself did not estimate whether resolution moderated the rupture–outcome association. It cannot demonstrate that experiencing a rupture necessarily worsens treatment or establish that reducing coded rupture frequency would improve clinical outcomes. A trajectory that includes temporary strain and successful recovery may differ substantially from persistent unresolved disengagement.
Empirical Evidence on Repair and Resolution
Repair research examines whether resolving a rupture predicts better outcomes. The Eubanks (2018) meta-analysis synthesized 11 studies with 1,314 patients for the relationship between rupture repair episodes and outcome. The reported pooled association was r = .29, with a 95% confidence interval of [.10, .47]. This result supports the clinical importance of resolution processes while also exhibiting an interval broad enough to caution against a single universal estimate across methods and settings.
Crucially, correlating repair with outcome does not isolate the effect of a repair strategy. Patients who can remain engaged during a difficult period may differ in resilience, treatment suitability, and previous progress from patients whose care ends prematurely. Therapists who repair effectively may have other skills that affect outcome. The very detection and recording of a successful repair can depend on whether a case continued long enough to observe it. These selection and measurement mechanisms should be considered in any interpretation of an apparently beneficial repair association.
The same 2018 research program examined six studies involving 276 trainees or supervisees in which rupture-resolution training or supervision was compared with a reference condition. The pooled association with patients' outcomes was not statistically significant in that evidence set; the Eubanks (2018) reports r = .11 and a confidence interval including zero. This distinction prevents the substitution of one claim for another: repair episodes correlated with outcome, whereas available training comparisons did not establish a consistent patient-level benefit. Clinical plausibility, process association, and intervention efficacy occupy different levels of evidence.
Treatment Retention, Dropout, and Unresolved Strain
Premature termination is a prominent outcome of interest because rupture may interfere with the continuation of care. Yet a dropout label is not synonymous with harm. Some patients cease attending because their goals have been met, they change provider, they face structural barriers, or they reject a treatment they have reason to regard as unsuitable. A scientifically responsible study defines dropout explicitly and explores patient accounts rather than treating every departure as evidence that the patient failed to cooperate.
The Eubanks (2019) found associations between observed rupture indices and treatment discontinuation in a small sample. This supports measuring rupture trajectories as potentially useful predictors of disengagement. It remains necessary to determine whether event coding predicts dropout beyond baseline severity, changes in symptoms, appointment accessibility, therapist effects, and the number of sessions already completed. Temporal selection can be especially strong when sessions of patients who terminate early are compared with sessions of those retained longer.
Future work should treat retention, patient-reported goal attainment, symptom change, adverse experience, and independent functional outcome as complementary endpoints. An alliance-focused intervention could increase retention without improving symptoms, or facilitate a safe referral without preserving the original treatment relationship. A complete evaluation must allow both outcomes. This is especially important where professional power or structural disadvantage makes continuity an imperfect proxy for care quality.
Therapist-Level Variation and Dyadic Specificity
Not all variations in alliance belong to the patient. Some therapists consistently obtain more favorable alliance ratings or more positive outcomes, while individual dyads may differ substantially from a clinician's typical pattern. Separating these effects requires multilevel designs with patients nested within therapists, repeated sessions nested within patients, and sufficiently many providers to estimate between-therapist variance. An analysis that treats every session as independent underestimates uncertainty and can falsely attribute a provider-level effect to a person-level process.
Del Re (2021) directly examined therapist effects in the alliance–outcome relationship through multilevel meta-analysis. This line of work is important because individual therapeutic practice can influence both the relational climate and the effectiveness of interventions. It motivates studies of trainable interpersonal conduct, but cannot alone establish that a particular training package improves patient health. Therapist allocation is rarely random in ordinary care, and case mix can influence estimated therapist differences.
Dyadic specificity is scientifically meaningful as well. The same practitioner may establish strong working agreement with one patient and persistent friction with another, due to differences in goals, communication, timing, treatment preferences, or the circumstances of referral. Research should avoid converting variability into broad claims about either participant's personality. The relevant unit is often the evolving fit between the people and the therapeutic project.
Cultural Humility, Difference, and Asymmetric Authority
Therapeutic collaboration occurs in culturally situated relationships. Differences in language, identity, social position, disability experience, gendered expectations, and professional status affect how people interpret warmth, questioning, silence, directness, and disagreement. An observer may categorize a form of deference as cooperation, although it arises from fear of contradicting a professional. A client may regard explicit assertiveness as necessary for respect, while a therapist experiences it as confrontation. Measurement and intervention must accommodate these plausible differences.
Hook (2013) developed the construct of therapist cultural humility as an interpersonal orientation characterized by openness and respect toward a client's cultural background. Their study is relevant to alliance research because interpretations of disagreement often depend on the clinician's orientation to difference. The original publication has a documented scoring erratum concerning an additional item included in computing the total; this bibliographic fact is a reminder that psychometric findings should be understood in relation to corrected measurement procedures rather than cited uncritically.
The scientific standard is to investigate experiences and variation within groups instead of relying on stereotypes about cultural conformity or expressiveness. Researchers should report the languages used, translation procedures, relevant identities, clinical setting, and representativeness of samples. They should test whether instruments have comparable meaning across populations. An overall alliance score averaged across many settings cannot establish equity in the repair of a power-related or discriminatory encounter.
Alliance in Child and Adolescent Psychotherapy
Youth psychotherapy complicates the ordinary dyadic model because a child or adolescent may participate in care alongside caregivers, schools, referring services, and legal guardians. Treatment goals may be set by an adult even when the young person sees the presenting concern differently. A youth's agreement with a therapist therefore requires separate consideration from caregiver satisfaction, clinician judgment, and adherence to an institutionally mandated plan. Multiple relationships can affect the opportunity to develop a meaningful working alliance.
The Roest (2023) three-level meta-analysis synthesized 99 studies in child and adolescent psychotherapy. It reported an association between child–therapist alliance and child outcomes of r = .17, and between parent–therapist alliance and child outcomes of r = .13. The work also analyzed alliance changes, agreement among reporters, and moderating relationships. These smaller associations underscore the need to distinguish the participant, informant, and outcome rather than transplant an adult psychotherapy estimate into youth treatment.
A youth may respond to therapy while expressing conflict with a parent, clinician, or institutional demand. Observed confrontation can sometimes represent developmental autonomy. A researcher must ask who defines rupture and whether resolution serves the young person's welfare and expressed priorities. Age-appropriate instruments, repeated patient-reported goals, confidentiality rules, and developmental capacity are necessary design considerations; they cannot be inferred from adult rating scales.
Teletherapy and Mediated Therapeutic Presence
Video and telephone psychotherapy preserve many elements of ordinary collaboration while altering practical conditions for attunement. Delays, connectivity problems, lack of privacy, visual framing, and interruptions can produce experiences that resemble or cause alliance strain. Other patients may find remote contact more accessible and less intimidating. Investigators should distinguish disagreement about treatment from friction generated by a platform and should measure the patient's real control over their environment.
The Aafjes-van Doorn (2024) examined 31 studies comprising 34 independent teletherapy samples and 4,862 adult participants. The weighted alliance–outcome association was r = .15, with a 95% confidence interval of [.07, .24]. This was a smaller association than the broad face-to-face estimate reported by Flückiger and colleagues in 2018, but the samples and methods differ, so the contrast alone does not establish an effect of delivery medium. Timing of alliance measurement and outcome source moderated some findings.
Rupture research in teletherapy additionally needs to specify whether silence follows an emotional intervention, a communication delay, or a technical outage. Recording permissions, confidentiality, access to safe private space, and accessibility features may determine what can be observed. Teletherapy is a setting for investigating alliance rather than a distinct mechanism with a single predictable level of therapeutic intimacy.
Human–Artificial Interaction as a Comparative Research Boundary
The Artificial Era makes relational language increasingly prominent in systems that provide conversational support, simulated counseling, or automated guidance. Users may report perceived empathy, responsiveness, emotional reliance, or a sense of collaboration with an artificial system. These are measurable human experiences, but a formal analogy with human therapeutic alliance requires construct validation. Bordin's account developed in the context of a clinical helping relationship with a human practitioner who bears professional responsibilities. The applicability of its goals, tasks, and bond dimensions to automated systems must be established rather than presumed.
A comparison study should explicitly distinguish human–human therapeutic alliance from user-rated collaboration with a system. It can test whether users understand the system's limitations, whether goals remain stable, how harmful misunderstandings are recognized, whether the system can repair a misunderstanding in subsequent dialogue, and how outcomes are affected by the availability of qualified human oversight. Observational markers developed for clinical sessions cannot be transferred unchanged to machine-generated exchanges without reliability and validity testing. Anthropomorphic questionnaire wording can make unlike processes appear equivalent.
The main empirical subject of this monograph remains psychological treatment among human clinicians and patients. Artificial systems enter as a carefully specified comparative research horizon within Psychology of Artificial and Human–Artificial Relations. Their potential for improving access, tracking user-reported disagreement, and supporting clinician workflow requires direct outcome and safety evidence. The existence of conversational responsiveness alone supplies neither a clinical efficacy estimate nor a professional therapeutic relationship.
Methodological Designs for Future Alliance Science
Robust observational research should follow dyads from the beginning of treatment and predefine measurement occasions. Baseline severity, treatment expectations, relational history, demographic and cultural context, treatment modality, therapist identity, and session attendance should be recorded when relevant. Patient and therapist alliance ratings should be collected independently and connected to temporally ordered, independently measured outcomes. Event-based rupture coding can add microprocess detail while protecting the participant's privacy and limiting interpretive overreach.
Multilevel models should respect repeated observations nested within patients and therapists, account for unequal treatment length, and examine missingness generated by early termination. Within-person centering can distinguish temporary departures from an individual's usual alliance from stable between-person differences. Longitudinal and dynamic models should include prior symptoms, prior alliance, and prespecified lag structures. Comparisons of findings across different time intervals must consider the interval-dependence emphasized in the reciprocal-process literature (Flückiger (2020)).
For causal claims, a transparent causal graph is essential. Randomized training studies can test whether an intervention changes therapists' observable responses, patients' perceived collaboration, and downstream clinical outcomes. Randomization of ethically permissible feedback procedures may clarify whether greater responsiveness changes alliance trajectories. Mechanism-specific mediation requires testing the path through repair, task engagement, or hope rather than inferring it from a total association. Pre-registration, independent coding, cross-site replication, and open reporting of null results reduce selective interpretation.
Psychometric and Statistical Threats
Alliance studies frequently depend on self-report, which may introduce common-method covariance when symptom change or satisfaction is measured by the same respondent at the same time. Ceiling effects can hide rupture episodes; recall effects can compress a complex session into a single favorable score. Small observational samples are sensitive to coder expectations, therapist clustering, and the selection of sessions available for recording. Coders may be blinded to outcome yet still recognize clinically salient cues that correlate with later ratings. Reliability and validity must be reported separately.
Meta-analysis has its own dependencies. Several effect estimates may come from one clinical sample, one therapist, or one research group. Treating them as independent can exaggerate precision. The 2018 alliance synthesis used multilevel meta-analytic methods; the Babl (2026) rupture analysis likewise modeled effect sizes nested within studies. Substantial between-study heterogeneity calls for careful moderator interpretation, especially when the number of independent studies is small. Selective reporting, publication bias, and inaccessible negative findings can influence pooled estimates.
The concept of repair also invites an outcome-dependent coding problem. An observer may be more likely to rate an episode as resolved after noticing a favorable subsequent session. This can be mitigated through blinded independent coding, event-specific definitions, and outcome measurement at a later and separate time point. Adverse experiences, deterioration, and negative judgments of the clinician should be retained as potential data rather than screened away as anomalies.
Clinical Ethics and Safety of Alliance Interpretation
The therapeutic alliance is embedded in the professional obligations of a clinician. Patient welfare, informed consent, confidentiality, boundaries, competence, and appropriate treatment planning provide conditions under which the relationship is evaluated. Alliance scores never replace risk assessment, diagnostic evaluation, the appropriateness of a clinical method, or consideration of harmful conduct. A positive relational rating can coexist with poor clinical care, and an uncomfortable but necessary discussion can coexist with ethical practice.
The task of studying rupture involves handling sensitive accounts of perceived mistreatment, coercion, exclusion, or professional errors. Research protocols should allow a patient to report these experiences without making access to treatment dependent on a favorable response. Clinical discussions of repair must avoid placing responsibility for professional misconduct on the patient. Studies should identify when consultation, reassignment, escalation, or independent evaluation is appropriate instead of defining the preservation of a particular dyad as the overriding endpoint.
This monograph analyzes psychotherapy process evidence and advances no individual treatment instruction. It does not propose replacing professional clinical judgment with an alliance score or predicting the outcome of a particular person's care. For applied clinical recommendations, independent qualified review and relevant local standards remain necessary. The research value of the alliance construct depends on maintaining this ethical distinction between an observational correlation and a decision about an individual patient.
Integrative Evidence Appraisal
The existing evidence permits several differentiated conclusions. First, alliance is a broadly useful construct with a clear historical theory and multiple measurement traditions. Second, the adult alliance–outcome association is replicated across a large collection of studies, including adjusted and temporally informative analyses, although statistical association alone does not prove intervention causality. Third, direct measures of rupture and measures of resolution describe different parts of a process and should not be combined into one undifferentiated treatment-relationship score.
Fourth, the repair evidence is promising but heterogeneous. The association of observed resolution with outcome in the Eubanks (2018) is materially stronger than the limited evidence that rupture-resolution training reliably improves patient outcomes. The Babl (2026) adds a focused negative rupture–outcome association, but its four underlying publications remain an important limitation. These conclusions can be true simultaneously because event occurrence, event resolution, and training efficacy constitute different causal questions.
Fifth, the evidence base broadens but does not become uniform when youth, digital formats, multiple reporters, and culturally complex care are considered. Roest (2023) and Aafjes-van Doorn (2024) provide domain-specific estimates with their own measurement contexts. These findings support a more precise scientific understanding: alliance is a family of interdependent relational functions that requires population-sensitive and temporally explicit measurement.
Unresolved Questions and Research Priorities
The field needs direct tests of when a rupture is harmful, when disagreement becomes informative, and when an attempted repair provides sustained benefit. A promising design would compare otherwise similar episodes with different observable patterns of patient acknowledgment, therapeutic revision, and later goal agreement. Yet such comparisons must account for the fact that repair strategies are not randomly distributed across cases. Multi-site replication and clinically meaningful endpoints are required before determining which strategies warrant broad implementation.
Another priority is measurement that integrates the patient's own understanding of the event with observer-coded interaction. Culturally varied samples, multilingual research, disability-inclusive methods, youth participation, and studies of involuntary or highly constrained care may reveal where established instruments lack construct equivalence. Researchers should also evaluate whether routine alliance monitoring improves clinical outcomes or merely raises ratings. Demonstrating improved detection is an intermediate result; improved patient functioning and safety are separate endpoints.
Further study should disentangle the pathways of relationship quality, symptom improvement, treatment engagement, and therapist behavior with transparent causal assumptions. Work on teletherapy and human–artificial interaction introduces additional contexts for validation, especially concerning privacy, limits of automated response, and the institutional allocation of responsibility. These research priorities follow from identifiable limitations of current evidence rather than from a claim that a new universal theory has already been established.
Conclusion
Therapeutic alliance organizes a treatment relationship through shared purposes, workable tasks, and a bond that can sustain difficult clinical work. Its scientific significance emerges most clearly when a relationship is examined over time: agreement is formed, sometimes strained, occasionally repaired, and continuously reinterpreted in response to what patient and therapist experience together. Rupture and repair are therefore central objects for process research, including questions about direct observation, patient agency, professional responsibility, and the temporal explanation of change.
The established literature supports a consistent positive association between alliance and psychotherapy outcomes and an association between successful rupture resolution and better outcomes. More recent directly measured rupture evidence finds an adverse association with outcome, while intervention research still requires stronger proof that particular training or repair methods cause improved clinical results. The reciprocal relationship between symptoms and alliance, differences among raters, the nested organization of therapists and patients, and variation across populations give these findings their proper scientific context.
The resulting synthesis positions alliance as a dynamic, measurable, and ethically situated process. A rigorous future science will identify when collaborative negotiation causally contributes to improvement, how relational strains are distinguished from legitimate disagreement, and what forms of repair are beneficial for whom and under which treatment conditions. The central research imperative is to bring observational precision, transparent methods, and the patient's perspective into sustained alignment with clinical outcomes.
References
- Aafjes-van Doorn, K., Spina, D. S., Horne, S. J., & Békés, V. (2024). The association between quality of therapeutic alliance and treatment outcomes in teletherapy: A systematic review and meta-analysis. Clinical Psychology Review, 110, 102430. DOI https://doi.org/10.1016/j.cpr.2024.102430
- Babl, A., Gómez Penedo, J. M., Nimphy, C., Keßel, M., Crainic, C., Boendermaker, N., & Eubanks, C. F. (2026). Alliance ruptures and psychotherapy outcomes: A multilevel meta-analysis of their association. Journal of Clinical Psychology, 82(10), 1683–1691. DOI https://doi.org/10.1002/jclp.70192
- Bordin, E. S. (1979). The generalizability of the psychoanalytic concept of the working alliance. Psychotherapy: Theory, Research & Practice, 16(3), 252–260. DOI https://doi.org/10.1037/h0085885
- Del Re, A. C., Flückiger, C., Horvath, A. O., & Wampold, B. E. (2021). Examining therapist effects in the alliance-outcome relationship: A multilevel meta-analysis. Journal of Consulting and Clinical Psychology, 89(5), 371–378. DOI https://doi.org/10.1037/ccp0000637
- Doran, J. M., Safran, J. D., & Muran, J. C. (2016). The Alliance Negotiation Scale: A psychometric investigation. Psychological Assessment, 28(8), 885–897. DOI https://doi.org/10.1037/pas0000222
- Eubanks, C. F., Lubitz, J., Muran, J. C., & Safran, J. D. (2019). Rupture Resolution Rating System (3RS): Development and validation. Psychotherapy Research, 29(3), 306–319. DOI https://doi.org/10.1080/10503307.2018.1552034
- Eubanks, C. F., Muran, J. C., & Safran, J. D. (2018). Alliance rupture repair: A meta-analysis. Psychotherapy, 55(4), 508–519. DOI https://doi.org/10.1037/pst0000185
- Flückiger, C., Del Re, A. C., Wampold, B. E., & Horvath, A. O. (2018). The alliance in adult psychotherapy: A meta-analytic synthesis. Psychotherapy, 55(4), 316–340. DOI https://doi.org/10.1037/pst0000172
- Flückiger, C., Del Re, A. C., Wlodasch, D., Horvath, A. O., Solomonov, N., & Wampold, B. E. (2020). Assessing the alliance-outcome association adjusted for patient characteristics and treatment processes: A meta-analytic summary of direct comparisons. Journal of Counseling Psychology, 67(6), 706–711. DOI https://doi.org/10.1037/cou0000424
- Flückiger, C., Rubel, J., Del Re, A. C., Horvath, A. O., Wampold, B. E., Crits-Christoph, P., Atzil-Slonim, D., Compare, A., Falkenström, F., Ekeblad, A., Errázuriz, P., Fisher, H., Hoffart, A., Huppert, J. D., Kivity, Y., Kumar, M., Lutz, W., Muran, J. C., Strunk, D. R., … Barber, J. P. (2020). The reciprocal relationship between alliance and early treatment symptoms: A two-stage individual participant data meta-analysis. Journal of Consulting and Clinical Psychology, 88(9), 829–843. DOI https://doi.org/10.1037/ccp0000594
- Hook, J. N., Davis, D. E., Owen, J., Worthington, E. L., Jr., & Utsey, S. O. (2013). Cultural humility: Measuring openness to culturally diverse clients. Journal of Counseling Psychology, 60(3), 353–366. DOI https://doi.org/10.1037/a0032595
- Horvath, A. O., & Greenberg, L. S. (1989). Development and validation of the Working Alliance Inventory. Journal of Counseling Psychology, 36(2), 223–233. DOI https://doi.org/10.1037/0022-0167.36.2.223
- Huber, J., Jennissen, S., Nikendei, C., Schauenburg, H., & Dinger, U. (2021). Agency and alliance as change factors in psychotherapy. Journal of Consulting and Clinical Psychology, 89(3), 214–226. DOI https://doi.org/10.1037/ccp0000628
- Roest, J. J., Welmers-Van de Poll, M. J., Van der Helm, G. H. P., Stams, G. J. J. M., & Hoeve, M. (2023). A three-level meta-analysis on the alliance-outcome association in child and adolescent psychotherapy. Research on Child and Adolescent Psychopathology, 51(3), 275–293. DOI https://doi.org/10.1007/s10802-022-00986-2
- Safran, J. D., & Muran, J. C. (2000). Negotiating the therapeutic alliance: A relational treatment guide. Guilford Press. Publisher record https://www.guilford.com/books/Negotiating-the-Therapeutic-Alliance/Safran-Muran/9781572308695
- Wampold, B. E., & Flückiger, C. (2023). The alliance in mental health care: Conceptualization, evidence and clinical applications. World Psychiatry, 22(1), 25–41. DOI https://doi.org/10.1002/wps.21035
- Zilcha-Mano, S. (2017). Is the alliance really therapeutic? Revisiting this question in light of recent methodological advances. American Psychologist, 72(4), 311–325. DOI https://doi.org/10.1037/a0040435
- Zilcha-Mano, S., Muran, J. C., Hungr, C., Eubanks, C. F., Safran, J. D., & Winston, A. (2016). The relationship between alliance and outcome: Analysis of a two-person perspective on alliance and session outcome. Journal of Consulting and Clinical Psychology, 84(6), 484–496. DOI https://doi.org/10.1037/ccp0000058