RESEARCH DOSSIER
Treatment Expectancy, Common Factors, and Specific Effects
Abstract. Treatment expectancy, common therapeutic factors, and specific intervention effects are central to explaining outcomes in psychotherapy. This fundamental research monograph distinguishes prospective outcome expectations from credibility, preference, alliance, empathy, and treatment-specific operations; reconstructs the historical debate from Rosenzweig, Rogers, Frank, and Bordin to contemporary contextual and component models; critically examines randomized component studies, comparative trials, individual-participant data syntheses, and process-outcome meta-analyses; and evaluates causal mediation, comparator credibility, heterogeneity, measurement, bias, clinical relevance, and digital delivery. Reviews published through 2026 are synthesized alongside foundational experiments. The evidence supports positive predictive associations for expectancy, alliance, and empathy, domain-specific effects of particular treatment components, and the importance of patient choice for treatment engagement. The scientific conclusion is that observational associations, package effects, component contrasts, and causal mechanisms require distinct forms of evidence. This work is an original critical narrative synthesis of existing scholarship, not a preregistered systematic review, a meta-analysis conducted by the author, or a clinical treatment guideline.
Keywords: treatment expectancy; psychotherapy common factors; specific psychotherapy effects; treatment credibility; therapeutic alliance; empathy; patient preference; psychotherapy mechanisms; treatment dismantling; component network meta-analysis; behavioral activation; exposure therapy; placebo effects; causal mediation; therapist effects; evidence-based psychotherapy; Artificial Era
Scientific Object, Research Questions, and Levels of Explanation
Treatment expectancy, common therapeutic factors, and specific effects form a connected problem in the causal explanation of psychotherapy. Treatment expectancy is a patient's prospective belief about the improvement likely to follow participation in treatment. Common factors are relational, contextual, and procedural conditions shared to varying degrees across legitimate psychotherapies: a credible account of suffering, collaborative goals, an emotionally responsive therapist, participation in a healing ritual, and opportunities for corrective experience. Specific effects are incremental changes attributable to treatment operations hypothesized to distinguish a particular intervention or theoretical family. The central question is how these processes combine to produce clinical improvement, under which conditions they make separable causal contributions, and what evidence can identify those contributions. The object extends beyond a contest among schools of psychotherapy to the structure of evidence required to interpret therapeutic change Cuijpers et al. (2019) Wampold (2015).
A scientifically adequate explanation distinguishes four different estimands. Treatment efficacy asks whether assignment to an intervention changes an outcome relative to a defined counterfactual; comparative effectiveness asks which of two viable approaches produces better outcomes for which people; component efficacy asks whether adding or removing a specified operation alters outcomes; mechanism research asks whether a temporally ordered process causally carries the effect. A statistically significant difference between complete interventions is evidence about packages as delivered. It cannot, by itself, establish the superiority of their named explanatory theories. Conversely, a small average difference between packages cannot establish that their different operations are inert, since different procedures may produce comparable benefit through different processes, benefit different subgroups, or share active components Kazdin (2007) Cuijpers et al. (2019).
The present monograph uses critical narrative synthesis of original theoretical writings, component trials, comparative clinical experiments, psychometric investigations, prospective process studies, meta-analyses, and methodological reviews. It reports published estimates only when identified in source abstracts or accessible original reports. It does not claim an exhaustive database search, preregistered systematic review, individual-patient data analysis, or new pooled effect calculation. Evidence is classified by design, time ordering, comparator, measurement, and risk of confounding. The main empirical frame is adult psychotherapy for mental health problems, with selective evidence from trauma, depression, anxiety, and internet-delivered treatment. The conceptual questions also extend to counseling, therapist training, treatment choice, culturally responsive care, and technologically mediated interaction.
The investigation respects distinct neighboring subjects. Overall treatment efficacy and comparative effectiveness belong to the separate dossier on clinical treatment comparisons; mediation analysis in depth belongs to the dossier on psychotherapy mechanisms; therapist skill, dropout, and psychological-treatment safety each have their own research object. This dossier concentrates on how expectations, cross-cutting interpersonal and contextual processes, and putatively technique-specific components are distinguished, measured, compared, and interpreted within a unified causal account. It asks which conclusions are empirically defensible now, which remain model-dependent, and which experiments would make the distinction more informative.
A Conceptual Taxonomy of Expectations and Therapeutic Factors
Outcome expectancy refers to the judged likelihood or degree of personally meaningful improvement. Treatment credibility refers to the perceived plausibility, coherence, and professional legitimacy of the rationale and procedures. Preference refers to the relative desirability of one treatment over alternatives; anticipated benefit, credibility, and preference are related but separate constructs. Self-efficacy concerns the patient's perceived capacity to perform needed actions. Hope may include valued goals, perceived pathways, and motivational agency, and is broader than a numerical forecast of symptom reduction. Therapeutic alliance describes agreement on goals, agreement on tasks, and an affective bond in the formulation developed by Bordin Bordin (1979). Credibility and expectancy were differentiated empirically in the Credibility/Expectancy Questionnaire by Devilly and Borkovec Devilly & Borkovec (2000).
A common factor is best defined operationally as a process potentially instantiated across multiple treatment schools, rather than an ingredient present in exactly equal quantity everywhere. Empathic understanding, goal agreement, therapist responsiveness, positive expectations, and collaboration can vary in dose, timing, and quality within a school, across clinicians, and among patients. A technique considered distinctive by one tradition may be adopted by another and become widely shared: behavioral experiments, emotion labeling, structured exposure, and activation are familiar examples of procedural borrowing. The distinction between common and specific therefore concerns explanatory role and counterfactual comparison, not a permanently fixed list of school-owned actions Wampold (2015) Cuijpers et al. (2019).
Contextual factors include treatment setting, clinician credentials, professional reputation, cultural legitimacy, interpersonal warmth, access, cost, ritual structure, and cues that a treatment has a coherent rationale. Some influence willingness to enroll and remain in care; others alter expectations after treatment begins; still others change the meaning patients assign to symptoms or therapeutic tasks. An intervention's specific operation may itself affect expectancy by yielding an early experience of mastery. Thus common and specific factors may be sequentially connected: rationale shapes initial engagement, engagement permits exposure to tasks, tasks change behavior, and early improvement reinforces credibility. Additive decompositions of outcome into independent portions for relationship, expectancy, and technique are usually conceptually and statistically misleading because the components can interact Wampold (2015) Kazdin (2007).
The clinical outcome is also multidimensional. Symptom scores, functioning, quality of life, patient-defined recovery, adherence, adverse experiences, and sustained remission need not respond identically. Expectancy may predict self-reported distress differently from observer-rated functioning, especially when demand characteristics or repeated assessments affect the response process. Specific procedures may target domains, such as avoidance, for which an omnibus symptom scale has reduced sensitivity. A valid comparative design specifies both the treatment component and the outcome construct before interpreting the result.
Historical Foundations of the Common-Factors Debate
Saul Rosenzweig's 1936 essay, “Some Implicit Common Factors in Diverse Methods of Psychotherapy,” recognized that distinct therapeutic schools might achieve change through processes their formal theories did not uniquely identify. The influential “Dodo bird” image became shorthand for a question about comparable psychotherapy results; the underlying scientific proposition concerned explanatory ingredients rather than merely the count of therapy brands Rosenzweig (1936). Jerome Frank subsequently advanced a comparative account of psychotherapeutic healing that emphasized a socially sanctioned relationship, a meaningful explanatory framework, a credible therapeutic procedure, and opportunities to overcome demoralization. The original 1961 edition and the revised 1991 account by Jerome and Julia Frank helped establish the study of healing contexts across formal traditions Frank & Frank (1991).
Carl Rogers's 1957 formulation placed the therapeutic relationship at the center of a testable theory of personality change, emphasizing contact, congruence, empathic understanding, positive regard, and communication of these conditions Rogers (1957). Its historical importance resides in specifying candidate interpersonal conditions for scientific investigation. Later research has questioned the stronger claim that such conditions are simultaneously necessary and sufficient for all successful therapy. Edward Bordin's 1979 account reformulated the working alliance across psychotherapy orientations as collaboration concerning goals and tasks within a relational bond. This model gave investigators a construct that could be assessed independently of a therapist's school, while also raising questions about whether alliance causes improvement or reflects a treatment already working Bordin (1979).
As randomized controlled trials expanded, investigators attempted to disentangle school-specific packages, placebo or attention controls, and ingredient-specific components. In the late twentieth century, behavioral and cognitive-behavioral therapies developed increasingly explicit experimental claims about exposure, reinforcement, appraisal, and cognitive restructuring. This made treatment dismantling feasible in principle: remove the postulated active element and test the result. The component analysis by Jacobson and colleagues Jacobson et al. (1996) became prominent because a behavioral-activation condition and more comprehensive cognitive-treatment conditions were experimentally compared within a common therapeutic lineage. Its findings challenged simple assumptions that every theoretically central component was required for improvement in the sampled depressed patients.
Modern controversy has moved from global declarations about all therapies being equivalent toward more precise questions: which therapeutic relationships are associated with improvement; whether expectations have incremental predictive value; when active treatments differ from credible comparators; whether a component contributes an additional clinical effect; and how outcome trajectories and causal mechanisms can be established. Wampold's contextual model Wampold (2015) gives expectation, relationship, and the implementation of meaningful tasks a central explanatory position. Cuijpers, Reijnders, and Huibers Cuijpers et al. (2019) emphasize that correlational process findings and generally similar treatment outcomes cannot decide whether common factors, specific ingredients, or their interaction causes observed improvements. Their competing interpretations define the methodological problem this monograph investigates.
What Would Count as a Specific Effect?
Specificity is a relational causal claim. A specific intervention effect is the incremental benefit attributable to a defined component compared with a counterfactual intervention that differs in the component while controlling, as far as feasible, for the remaining therapeutic conditions. The meaning of “specific” depends on what has been equated: therapist attention, time, participant expectancy, activity, supervision, dose, treatment integrity, and access to alternative effective procedures. If a psychological treatment is compared with a waitlist, superiority establishes an effect of the treatment package plus many contextual changes relative to deferred treatment. It does not identify the characteristic mechanism within the active package Cuijpers et al. (2016) Cuijpers et al. (2019).
An ingredient can be specific in different senses. In a descriptive sense, it may be unusually associated with a named treatment manual. In an operational sense, a trial may isolate a task such as in-vivo exposure, behavioral activation, cognitive restructuring, or memory updating. In a mechanistic sense, evidence must show that the task changed a specified mediator and that change produced improvement. In an effect-modification sense, an operation may be particularly valuable in one clinical or cultural context. These claims have different evidentiary thresholds; some component trials answer only the operational question while the mechanism remains uncertain Kazdin (2007).
The null hypothesis in a dismantling comparison is a difference of zero under the particular study's contrast. Failure to reject it is not proof of clinical equivalence. Confidence intervals must be examined against a prespecified margin for clinically important difference, and a valid equivalence or noninferiority inference requires adequate power, justified margins, and attention to adherence. Similarly, a statistically detectable difference can be too small to alter clinical decisions. Component studies also face dilution when the supposedly deleted ingredient is delivered informally or when participants discover equivalent skills independently. Their strongest contribution is precise local evidence about the necessity or incremental value of an operation within a defined treatment ecology Ahn & Wampold (2001) Jacobson et al. (1996).
A full account treats common factors and specific procedures as potentially interactive causes. A reliable alliance may facilitate challenging exposure exercises; a credible rationale may enhance practice between sessions; targeted skill acquisition may itself improve collaboration; successful behavior change may strengthen trust and increase later expectancy. These pathways create feedback and make ordinary regressions vulnerable to reverse causation. The most informative experiments therefore combine randomized component contrasts, repeated measurement, credible comparator development, long-term outcomes, and process perturbations.
Treatment Expectations: Formation, Updating, and Clinical Meaning
Patients enter treatment with beliefs assembled from previous care, descriptions from trusted people, cultural expectations about clinicians, prior symptom trajectories, and cues supplied during intake. An expectation is an empirical psychological variable, with a referent, forecast horizon, and degree of conviction. Patients may separately expect relief from anxiety, greater capacity for functioning, tolerable difficulty during therapy, or eventual recovery despite temporary worsening. This plurality is important because a generic item asking whether treatment will “help” can conceal disagreement about what help means. The expectancy literature summarized by Constantino and colleagues Constantino et al. (2018b) treats early outcome expectation as a prospective predictor of later clinical outcomes. A valid research design records expectations before the outcome interval of interest and distinguishes stable pretreatment beliefs from session-specific updates.
Formation can be modeled as probabilistic learning under socially meaningful conditions. The therapist provides a rationale; the patient compares it with existing beliefs and experiences; new observations, including early behavioral accomplishments, alter the anticipated probability of success. Expectation then influences willingness to begin treatment, effort during difficult assignments, and interpretation of ambiguous change. Because initial improvement can also strengthen subsequent expectation, an observed association between a later expectation measure and outcome contains feedback. The dynamic model predicts reciprocal, potentially nonlinear relations, but its causal coefficients remain treatment-, patient-, measure-, and design-dependent Constantino et al. (2021) Kazdin (2007).
Credibility is a rationality judgment about the intervention's rationale and procedures. It may coexist with low expected personal benefit when a patient believes that an intervention is generally sound but unlikely to work in their circumstances. Conversely, strong interpersonal trust can produce high anticipated benefit even when the proposed account of the problem is poorly understood. The Credibility/Expectancy Questionnaire includes items intended to distinguish these dimensions, with “thinking” and “feeling” modes of anticipation contributing to its structure Devilly & Borkovec (2000). Researchers should test measurement invariance when applying the instrument across treatment modalities, language groups, clinical populations, or automated delivery environments rather than assuming that an identical score captures the identical construct everywhere.
Expectations have ethical and epistemic importance as well as predictive relevance. Clinicians may strengthen a realistic account of what participation involves, including uncertainty and expected effort, without presenting guaranteed benefit. Deceptive exaggeration can temporarily raise reported confidence while undermining informed consent and later trust. An evidence-aligned formulation gives the person a comprehensible treatment rationale, acknowledges alternative effective options when they exist, and makes outcome monitoring an ordinary collaborative process. Whether any such communication intervention increases long-term benefit over a credible comparison requires direct testing rather than inference from association Wampold (2015) Cuijpers et al. (2019).
Meta-Analytic Evidence for Expectancy and Credibility
The meta-analysis by Constantino, Vîslă, Coyne, and Boswell Constantino et al. (2018b) assembled 81 independent samples comprising 12,722 patients and reported an overall association between patients' early psychotherapy outcome expectation and later posttreatment outcome of approximately r = .18. The estimate is a positive mean predictive association across varied interventions and measures. Its magnitude does not license treating expectation as a sufficient cause of improvement. Heterogeneity was substantial, and individual effect sizes depended on how early expectation and clinical outcome were operationalized. The study demonstrates broad predictive relevance while leaving the causal share attributable to expectancy undetermined.
A complementary review of treatment credibility Constantino et al. (2018a) evaluated the relationship between the plausibility of treatment and improvement. The distinction matters analytically because a credible rationale can encourage adherence and collaborative task engagement, and treatment participation can subsequently change the appraisal of credibility. In the aggregate literature, studies often measure expectancy and credibility close together in time and may rely on similar patient-reported instruments. An association surviving adjustment for initial symptom severity or early therapeutic alliance is evidence of incremental prediction conditional on the chosen covariates, while unmeasured third variables remain possible.
Coyne and colleagues Coyne et al. (2023) synthesized factors associated with patients' psychotherapy expectations. More severe baseline symptoms were modestly associated with lower expectations, whereas aspects of quality of life were positively associated. Such correlates establish that expectation is partly patterned by clinical and personal context. They also warn against interpreting an expectation score as a freely manipulable psychological ingredient: the same patient may express low predicted recovery because of persistent social adversity, prior unsuccessful treatment, economic barriers, or accurate appreciation of illness complexity. Clinical interpretation requires attention to these antecedents rather than treating low expectation as resistance.
A meta-analysis of indirect effects by Constantino and colleagues Constantino et al. (2021) evaluated a pathway linking patient expectation, alliance quality, and improvement. Its mediation model is useful as a candidate explanatory architecture: a credible anticipated path to recovery may support collaboration, which could facilitate clinical gains. However, meta-analytic mediation based on primarily observational relations does not randomize expectations or alliance. Serial confounding, measurement timing, and shared reporter bias remain threats. Mediation coefficients obtained from aggregate correlations should be interpreted as compatible with a proposed pathway, not as a verified proportion of change caused by expectancy.
Expectancy findings also invite denominator discipline. An r of approximately .18 is a correlation between observed constructs, not a statement that expectations account for 18 percent of treatment benefit. Squaring a correlation produces a different quantity and can still fail to represent causal attribution. Observational prediction need not be proportionally translated into a percentage of variance explained by the intervention or into a percentage of patients who improved “because of” belief. The appropriate scientific claim is narrower and more informative: baseline expectations provide modest prognostic information across diverse psychotherapy studies, and experimental manipulation is needed to identify its causal leverage.
Preference, Choice, and Participation as Distinct Pathways
Patient preference introduces a second pathway between treatment belief and outcome. Preference matching can affect whether a person enrolls, attends, completes treatment, or sustains independent practice; any improvement in adherence may subsequently alter measured clinical effects. A meta-analysis of 53 studies including more than 16,000 clients Swift et al. (2018) found that accommodating preferences was associated with lower dropout and more positive outcomes. The reported mean outcome association was d = .28, with an odds-ratio metric for treatment dropout that depended on how groups were coded. Study design, measurement timing, and diagnosis moderated these associations. The result supports the clinical relevance of accommodating choice while preserving the distinction between preference and outcome expectancy.
The review by Windle and colleagues Windle et al. (2020), focused on people with diagnosed mental disorders receiving psychosocial interventions, synthesized 29 randomized trials including 5,294 participants in its meta-analyses. Receipt of a preferred treatment was associated with lower dropout, with relative risk approximately .62, and a stronger therapeutic alliance. It did not find evidence of a statistically significant association with the other prespecified clinical outcomes. This narrower review therefore constrains the interpretation of more expansive preference meta-analyses. The evidence supports improved engagement especially clearly, while the clinical outcome benefit depends on population, design, and which studies enter the synthesis.
Lindhiem and colleagues Lindhiem et al. (2014) considered shared decision making, treatment choice, and receiving one's preferred treatment together. Across the included trials and studies, patient participation or preference accommodation was associated with modest improvements in satisfaction, completion, and clinical outcomes. Their operational category is broader than pure treatment preference matching and may include informational involvement that changes both knowledge and perceived control. Differences across the Lindhiem, Swift, and Windle syntheses therefore arise in part from inclusion criteria and estimands, and cannot be reconciled by comparing effect-size numbers as if each referred to the same intervention contrast.
Preference can also be a source of selection and expectancy confounding in psychotherapy trials. A person randomized to a nonpreferred condition may be disappointed, less adherent, and less willing to rate the treatment as helpful. A person able to choose may experience autonomy and better fit independent of the intervention's specific mechanism. Designs such as preference trials, two-stage randomization, and measurement of baseline strength of preference can address distinct questions. The scientific target is to explain these pathways and estimate the relevant contrasts, not to erase preference-related effects as mere nuisance variance.
Experimental Manipulation and Its Identification Limits
Observational evidence motivates a stronger question: when two individuals receive comparable procedures, does an experimentally induced difference in expectation produce a difference in clinically meaningful outcome? Candidate manipulations include variation in the amount and framing of evidence-based rationale, brief expectancy optimization exercises, clinician communication style, or transparent encouragement to anticipate meaningful but uncertain progress. Random assignment can remove baseline confounding between groups, yet manipulation may simultaneously alter rapport, motivation, perceived professional expertise, and the amount of treatment information received. The treatment of “expectancy” as the unique mediator requires further measurement and, ideally, a factorial or dismantling design.
Manipulation checks should measure expected outcome independently of credibility, treatment satisfaction, affective arousal, and demand characteristics. The intervention must avoid factual misrepresentation and respect informed consent. A successful experimental intervention that changes self-reported expectations but not symptoms still provides evidence that the psychological construct was manipulable; a failed symptom contrast may reflect insufficient statistical precision, low intervention dose, or an expectancy process that matters only for some populations. Conversely, a clinical effect following an encouraging script does not identify expectancy as the mediator unless competing changes have been addressed Kazdin (2007) Cuijpers et al. (2019).
Temporal design is essential. At a minimum, baseline symptom severity, pretreatment expectation, treatment assignment, multiple within-treatment expectations, alliance, adherence, and repeated symptom measures should be recorded. Within-person change scores help distinguish how an individual's fluctuations differ from between-person prognostic differences. Long-term follow-up is necessary because expectations may chiefly influence early engagement or symptom appraisal, whereas stable functional recovery can depend on sustained practice and context. Well-developed experiments can directly perturb expectation between sessions while maintaining delivery of evidence-supported techniques, then examine whether later alliance, task completion, and functioning change in the predicted order.
The newest experimental literature should be evaluated at its actual scope. Small pilot studies that optimize expectations in healthy volunteers performing mindfulness exercises, for example, are informative about feasibility and proximal effects but do not establish efficacy for diagnosed clinical conditions or prove that expectancy explains the effects of a full psychotherapy course. The scientific conclusion is that expectancy manipulation is testable and potentially consequential, while its disorder-specific effectiveness, durability, and clinical effect modification require adequately powered trials with transparent reporting and suitable comparators.
Therapeutic Alliance: Construct, Predictive Evidence, and Causal Ambiguity
Bordin's formulation Bordin (1979) treats the working alliance as a collaboration built around agreement on therapeutic goals, agreement on tasks, and an interpersonal bond. These components explain why a person can trust a clinician personally but struggle to undertake a particular exercise, or can understand the clinical rationale while experiencing the relationship as invalidating. In psychodynamic, behavioral, cognitive, humanistic, and integrative therapies, specific tasks vary, yet the scientific question of negotiated participation remains relevant. The alliance is consequently both a trans-theoretical construct and a treatment-specific achievement: agreement on goals and tasks depends partly on what the actual intervention asks the patient to do.
Flückiger and colleagues Flückiger et al. (2018) synthesized 295 independent studies including over 30,000 patients and reported an alliance–outcome association of approximately r = .278, with a confidence interval near .256 to .299. This is among the most consistent large-scale associations in psychotherapy process research. The pooled coefficient measures association; it cannot be converted into a direct percentage of symptom improvement caused by alliance. Studies vary in disorder, treatment modality, timing, reporter, clinical setting, and analytic adjustment. Even a stable aggregate coefficient does not eliminate the possibility that early improvement leads to more positive alliance ratings.
A later adjusted meta-analytic summary Flückiger et al. (2020a) examined samples that accounted for patient characteristics and treatment processes. Alliance remained associated with outcome, with coefficients in the approximate .23 to .31 range under the assessed adjustments. These comparisons increase confidence that simple baseline severity and selected observable processes do not fully explain the association. They do not achieve random assignment of alliance quality, nor do they establish immunity from unmeasured common causes, therapist selection, changes in treatment adherence, or shared method variance.
The individual participant data meta-analysis by Flückiger and collaborators Flückiger et al. (2020b) analyzed session-by-session relationships between early alliance and symptoms across 17 primary datasets, nine countries, and 5,350 participants. The study's focus on reciprocal lagged effects is a substantial methodological advance over one early alliance score predicting the final symptom score. It frames alliance and symptoms as potentially mutually influencing processes. A process that follows earlier symptom change but predicts later symptom change may be part of a reciprocal clinical system; temporal ordering improves inference while leaving some time-varying confounding unresolved.
A central research priority is thus alliance perturbation. A trial could experimentally train clinicians in specific repair procedures, measure whether alliance trajectories change, and evaluate independent clinical outcomes, while accounting for additional supervision and therapist responsiveness. Because such training may also improve empathy, task clarity, and technical execution, evidence for a unique alliance mechanism requires careful contrasts. The causal object is a modifiable process in a realistic therapeutic ecology, and a precise estimand may be more informative than a declaration that the alliance is either the universal cause or merely an epiphenomenon.
Empathy, Positive Regard, and Relational Responsiveness
Rogers Rogers (1957) presented empathic understanding, genuineness, and unconditional positive regard as core conditions of therapeutic change within his person-centered theory. The empirical questions include whether clients experience these qualities, how clinician behaviors convey them, how experiences differ across settings and cultures, and whether altered responsiveness has a causal effect on symptom and functioning outcomes. An observer's code of therapist empathy, a client's perception of being understood, and the clinician's self-report are related but not equivalent indicators; each may covary differently with treatment success.
An updated meta-analysis by Elliott, Bohart, Watson, and Murphy Elliott et al. (2018) included 82 independent samples with 6,138 clients and found that therapist empathy was positively associated with outcome at approximately r = .28. The association was evident across diverse therapy orientations, supporting the relevance of the construct beyond any one school. Its observational design limits causal interpretation, and its pooled estimate cannot settle whether empathy exerts a direct influence, permits adherence to difficult intervention tasks, or reflects clinicians' responsiveness to patients already making progress. A serious evidence synthesis retains all these models as candidates.
Relational responsiveness is an interactional property rather than a constant amount of warmth. A patient engaged in trauma-focused exposure may benefit from empathic validation of distress together with a firm, collaborative plan to continue work when safe and appropriate. Another patient may need careful renegotiation of goals after an unwanted intervention. Excessive reassurance can sometimes undermine exposure learning, and apparent support can mask avoidance if it repeatedly deflects agreed-upon tasks. These are hypotheses about functional context, not reasons to discount the consistent observed value of empathy. The quality of responsiveness depends on calibration to patient needs, autonomy, treatment purpose, and safety.
Genuineness and positive regard also raise measurement questions. An observational rating may capture the social desirability of clinician behavior; a patient rating may capture felt acceptance but may be affected by symptom improvement or prior relationship experiences. Cross-cultural comparisons must examine whether the same linguistic description and behavioral cue convey the same degree of respect. Relational quality cannot be inferred simply from automated pleasantness, scripted reflection, or lexical sentiment in a transcript. The construct requires a theory linking observable actions to the patient's experience and subsequent therapeutic processes.
Therapist Effects, Competence, and Treatment Fidelity
Therapist effects are systematic differences in patient outcomes associated with the clinician who delivered care, after accounting as far as possible for case mix, setting, and treatment assignment. Therapist effects matter because two clinicians applying the same nominal treatment can produce different results, and clinicians practicing different schools may share effective relational skills or vary in the accuracy and flexibility of technique delivery. Therapist variance is not synonymous with common-factor variance: it can include alliance skills, experience, patient selection, supervision, risk management, and competence with specifically indicated procedures Wampold (2015).
Competence describes the quality with which therapeutic operations are performed, whereas adherence describes the degree to which a clinician delivers prescribed procedures and avoids prohibited deviations. These constructs can diverge. A highly adherent clinician may deliver tasks mechanically and fail to secure collaboration; a responsive clinician may flexibly adapt a protocol while preserving the intervention's functional principles. Research that measures only adherence may classify a treatment as delivered while overlooking whether the theoretically relevant learning conditions were achieved. Conversely, unstructured flexibility may introduce uncontrolled treatment components and compromise the interpretation of a comparison trial Cuijpers et al. (2019).
Multilevel designs are necessary when patients are nested within therapists or clinics. If outcomes of patients treated by the same clinician are correlated, analyses that assume all patient observations are independent may underestimate uncertainty. Therapist random effects, cluster-robust standard errors, sufficient numbers of therapists, and reporting of clinician-level variation improve inference. Small trials with one clinician per arm cannot disentangle treatment effects from therapist differences. Crossing therapists with treatment conditions or using adequate stratification can improve design, although training and allegiance may still affect each intervention's implementation.
A central interpretive problem is therapist allegiance. Researchers and clinicians may be more confident in a favored treatment, communicate stronger rationales, and use the methods with greater fluency. Such differences can change expectancy, alliance, and technique quality at the same time. A fair trial seeks comparably trained clinicians, transparent fidelity measurement, balanced supervision, independent outcome assessors when feasible, and explicit reporting of investigator and therapist allegiance. These safeguards reduce avoidable explanatory ambiguity without implying that conviction in evidence-supported therapy should be eliminated from ordinary clinical care.
Rupture, Repair, Feedback, and the Temporal Ecology of Therapy
A therapeutic relationship changes over time. Goal disagreements, perceived misunderstanding, missed appointments, discouraging assignments, and clinical deterioration can generate ruptures in collaboration. Repair may involve acknowledging the event, clarifying divergent expectations, renegotiating tasks, and restoring workable participation. The theoretical significance is that alliance quality is neither a baseline trait nor a uniform therapeutic dose. Research should identify discrete relational events and test whether specific repair responses predict subsequent participation and recovery beyond spontaneous adaptation or early symptom trends.
Repeated symptom measures can improve care by making progress, plateau, and deterioration observable, but feedback itself becomes an intervention that changes expectations. A clinician who shares a positive trajectory may strengthen confidence; feedback about insufficient progress may prompt revised goals or a new approach. The scientific question is whether structured feedback improves outcomes through better decision-making, greater credibility, changed adherence, or some combination. A feedback package cannot be interpreted as an isolated common factor unless its components and mediating channels are specified.
A rupture-repair mechanism requires at least four temporal observations: a detectable rupture; a repair attempt; change in the targeted relationship process; and subsequent change in a clinical or behavioral outcome. Ideally, variations in repair interventions are randomized or otherwise credibly exogenous. Observations that successful patients retrospectively report helpful repair episodes demonstrate perceived meaning but do not, alone, identify causal contribution. Rigorous relational process research combines transcript coding, patient report, independent assessment, and clinically relevant outcomes.
There is a further distinction between therapeutic relation and therapeutic safety. An alliance can feel strong while an approach remains ineffective or harmful, and an appropriate safety intervention can temporarily reduce satisfaction. Clinical ethics requires monitoring deterioration, adverse reactions, boundary violations, coercion, and inequity even when relational indicators appear positive. The scientific account of common factors must therefore remain connected to independent outcomes and informed consent rather than treating patient affection for a therapist as an adequate measure of successful treatment.
Dismantling Trials and the Logic of Component Removal
Dismantling research experimentally removes a hypothesized active ingredient from an otherwise comparable treatment package. An additive trial reverses the contrast by attaching a component to a base intervention. Both designs seek a more precise causal estimate than a comparison of two complete brands of therapy. Their validity depends on maintaining comparable clinician attention, total treatment dose, expectancy, supervision, participant severity, and access to other procedures. If eliminating a component shortens sessions, changes the coherence of the therapeutic rationale, or reduces therapist confidence, the contrast no longer isolates the theoretical ingredient alone. These difficulties explain why dismantling experiments can provide high-value local evidence without fully settling a universal theory of therapeutic change Ahn & Wampold (2001) Cuijpers et al. (2019).
Ahn and Wampold's component meta-analysis Ahn & Wampold (2001) asked whether removing supposedly specific ingredients reduced the effectiveness of psychotherapy. It provided an early quantitative basis for skepticism about the consistent incremental necessity of named ingredients across the included comparisons. The historical importance of this work is the systematic evaluation of component contrasts. Its limitations include the number and variety of trials, potential differences in design quality, imprecise control of overlap between interventions, and the possibility that the remaining package already contains an effective route to improvement. A null average across heterogeneous components does not establish that no component matters for any indication.
Dismantling also distinguishes necessity from sufficiency. If removing cognitive restructuring does not worsen outcome in a particular depression trial, cognitive restructuring may be unnecessary for that treatment package and sample. That finding does not mean the procedure could never be sufficient to benefit another patient, or that cognitive change played no role during behavioral work. The retained procedures can trigger the very psychological change originally attributed to the deleted component. When two procedures can each recruit a shared learning process, removing one does not necessarily reduce effectiveness; mechanistic redundancy is a plausible explanation.
For clinical decision-making, an adequate dismantling result needs measures of adverse reactions, acceptability, treatment completion, and long-term outcomes. A shorter package may be more accessible yet less durable, or it may achieve equivalent outcomes with less burden. If a study is designed to establish equivalence or noninferiority, it must specify a clinically acceptable margin in advance and maintain adequate statistical precision. Simply observing p greater than .05 for the between-arm difference cannot support a claim of equivalence. The comparison must remain anchored to the intervention, population, and outcome actually studied.
Behavioral Activation and Cognitive Treatment Components in Depression
Jacobson and colleagues Jacobson et al. (1996) randomly allocated 150 adults with major depression to a behavioral-activation component, behavioral activation plus work with automatic thoughts, or a more complete cognitive therapy package including additional procedures. Their component design tested whether adding cognitive elements provided incremental benefit during acute treatment. The findings did not demonstrate superior acute outcomes for the most comprehensive treatment in that sample. The result challenged the assumption that explicit cognitive restructuring was necessary for all benefits of that particular cognitive-behavioral treatment, while leaving several alternative explanations: common behavioral elements may have been sufficient, indirect cognitive change may have occurred, and the study was limited by its specific clinical sample and treatment implementations.
The later randomized trial by Dimidjian and colleagues Dimidjian et al. (2006) compared behavioral activation, cognitive therapy, antidepressant medication, and a medication placebo condition in adults with major depression. It showed that behavioral activation could function as a substantive treatment in its own right and produced clinically important comparisons, including results that varied with baseline depression severity. The study did not reduce all improvement to a single technique. Medication conditions also differ from psychotherapy conditions in patient and clinician expectancy, blinding, adherence, and adverse events, so their contrasts cannot be treated as a pure test of common psychotherapy factors.
Dobson and colleagues Dobson et al. (2008) extended the behavioral activation–cognitive therapy research program to relapse and recurrence after acute depression treatment. Such follow-up addresses a key limit of end-of-treatment component comparisons: two approaches can yield similar acute symptom reduction while differing in sustained benefit or vulnerability to relapse. Long-term contrasts may depend on continuing medication, practice of learned skills, baseline chronicity, residual symptoms, and posttreatment access to services. The mechanism question therefore requires both an acute account of change and a longitudinal account of maintenance.
A later network meta-analysis by Ciharová and colleagues Ciharová et al. (2021) included 45 randomized studies with 3,382 participants examining cognitive restructuring, behavioral activation, and combined cognitive-behavioral therapy for adult depression. It found no clear evidence that one of these approaches was more effective than the others on the primary depression outcomes assessed. All three were superior to waitlist or usual-care comparators in the reported networks. Network comparisons depend on transitivity, consistent definition of treatments, treatment delivery, and the available contrasts; absence of evidence for a between-treatment difference should not be recast as proof that their mechanisms are identical.
The evidence permits a precise synthesis. Behavioral activation can be an effective intervention for depression, and additive cognitive procedures have not uniformly demonstrated incremental benefit in the studied packages. Cognitive mechanisms may nevertheless operate during activation, and cognitive restructuring may contribute in particular contexts or patient subgroups. The explanatory target is the relationship among procedures, learning, expectancy, adherence, and functional improvement, examined with patient-level moderators and repeated mediators. Simplifying a treatment should be based on specific evidence for comparative benefit, feasibility, and durability rather than on an assumed universal irrelevance of one theoretical tradition.
Exposure, Trauma-Focused Treatments, and Mechanism Specificity
Exposure-based treatments are frequently discussed as candidates for relatively specific psychological effects because their procedures deliberately contact feared cues, memories, or situations while facilitating new learning. The scientific construct is more complex than presenting distressing material. Exposure is typically embedded in a rationale, planning, graduated or otherwise structured practice, safety management, consent, and collaboration. Distinct explanations include habituation, extinction and inhibitory learning, expectancy violation, changes in perceived coping ability, and the revision of threat beliefs. These explanations overlap empirically and may operate on different timescales Craske et al. (2008).
Foa and colleagues Foa et al. (1991) compared cognitive-behavioral procedures and counseling approaches in a randomized treatment study with rape survivors experiencing posttraumatic stress. The trial illustrates why choice of comparator and outcome timing matters. Interventions can differ in emotional support, behavioral tasks, intensity, and therapist expectations, and small samples can yield imprecise component estimates. The study supports the possibility of evaluating trauma-focused procedures experimentally while also showing the need to establish which package elements caused observed effects.
A subsequent randomized study by Foa and colleagues Foa et al. (2005) compared prolonged exposure alone with prolonged exposure plus cognitive restructuring, among other conditions, in adults with posttraumatic stress disorder. Adding cognitive restructuring did not show a clear enhancement of the exposure intervention's outcomes within that trial. This is an informative component contrast because a theoretically motivated addition failed to demonstrate an expected incremental clinical benefit. It does not imply that cognitive appraisals are irrelevant; successful exposure can change appraisals without a separate formal cognitive module.
Craske and colleagues Craske et al. (2008) outlined inhibitory-learning principles for optimizing exposure. Their analysis directs attention to how new nonthreat associations are acquired, retrieved, and generalized. It provides a mechanism-focused program in which procedures can be manipulated according to predictions about expectancy violation, variability, and retrieval cues. These mechanisms require rigorous testing through repeated learning measures and clinical endpoints; the existence of a persuasive learning theory does not, by itself, establish that each proposed optimization improves patient outcomes.
Trauma-focused research also makes safety, consent, and heterogeneity central. Patients differ in symptom profiles, dissociation, ongoing danger, medical conditions, resources, and treatment preferences. A specific procedure may be supported by a trial yet require adaptation or additional support in a new context. A treatment comparison that ignores these conditions can misattribute failure to technique when access, acceptability, or fidelity is responsible. The relevant evidence consists of disorder-specific trials, credible active comparisons, functional outcomes, and appropriate follow-up, combined with transparent clinical risk assessment.
Component Network Meta-Analysis and Digital Psychotherapy
Component network meta-analysis seeks to estimate the contributions of treatment elements by comparing packages that contain different combinations of components. Whereas a direct dismantling trial holds most of a treatment constant, a component network synthesizes information across many trials. Its interpretation depends on the assumption that component effects combine in an adequately specified manner and that the indirect comparisons are sufficiently exchangeable with direct comparisons. Selection of component taxonomy, trial populations, dose, treatment context, and interactions can materially change results. It is a statistical modeling framework for causal contrasts across a network, not a substitute for randomized isolation of every element.
Furukawa and colleagues Furukawa et al. (2021) conducted an individual-participant-data component network meta-analysis of internet cognitive-behavioral therapy for depression, including 76 randomized trials, with individual-level data from 48 trials and aggregate data from 28. Their reported estimates suggested benefit from behavioral activation and possible poorer average depression outcomes when relaxation components were included. The authors also estimated a contribution for nonspecific treatment effects and found evidence that combining human and automated encouragement improved adherence relative to less supportive configurations. These findings show that specific skills, shared contextual influences, and delivery support can all be quantified in one model, subject to substantial assumptions.
The primary component estimates in the report were expressed in incremental units on a depression scale harmonized to the Patient Health Questionnaire-9. For behavioral activation, the incremental mean difference was approximately −1.83 points (95% credible interval −2.90 to −0.80); for relaxation it was approximately +1.20 points (0.17 to 2.27), where higher scores indicated more depressive symptoms Furukawa et al. (2021). The study describes these as suggestive rather than universal effects. Its limitations included moderate-to-high heterogeneity, high measurement-related risk of bias, and potentially correlated components that were difficult to isolate. A positive component estimate in a statistical network does not mean its removal would improve every individual patient's course.
The same evidence also illustrates how “common” effects become operational model parameters. The nonspecific effects term in the component decomposition pooled contextual and placebo-related influences rather than uniquely identifying expectation, alliance, or empathy. Support messages affected program completion, but encouragement is not identical to a complete therapeutic relationship. Effects observed in digital programs for adult depression should not be generalized mechanically to face-to-face psychotherapy, different psychiatric disorders, or untested artificial conversational systems.
Karyotaki and colleagues Karyotaki et al. (2021) used individual-patient data network meta-analysis to compare guided and unguided internet cognitive-behavioral therapy for depression. The analysis found the relative usefulness of guidance varied with baseline severity, with stronger evidence for guidance benefits among participants with more severe symptoms. This result is particularly relevant to the specific/common distinction because a delivery factor can modify benefit without being the nominal therapeutic technique. Digital treatment research provides opportunities for factorial experiments with reproducible scripts and timing, yet it retains problems of unblinded self-report, selection bias, attrition, and variation in real-world implementation.
Control Conditions, Treatment Comparators, and the Placebo Analogy
A control group defines what causal difference a trial estimates. Waitlist compares immediate access to an intervention with deferred access and typically also changes contact, expectations, monitoring, and the opportunity to practice skills. Usual care is more clinically relevant in many systems but may range from minimal attention to intensive medication or counseling, producing considerable variation in the contrast. Treatment as usual is not a common invariant comparator across hospitals and countries. Supportive counseling, attention controls, structured health education, and other credible active conditions can better match therapist contact while containing their own effective ingredients Cuijpers et al. (2016) Cuijpers et al. (2019).
A credible comparator must be described and measured rather than simply labeled placebo. Psychotherapy cannot usually blind patients or clinicians to the activities undertaken in the way that pharmacological trials can blind the chemical identity of otherwise similar pills. Even when two psychological rationales appear equally credible at baseline, they may yield different experiences of mastery, emotional connection, activity, and symptom feedback. The more thoroughly two interventions are matched, the more likely that the comparator also contains active ingredients. This is both a scientific challenge and a reason that small between-condition differences need careful interpretation.
The medical placebo literature shows that the apparent effect of an inactive intervention depends strongly on outcome type and trial design. The Cochrane review by Hróbjartsson and Gøtzsche Hróbjartsson & Gøtzsche (2010) assessed placebo interventions against no treatment across clinical conditions and found that effects were especially sensitive to subjective and continuously measured outcomes. These results inform expectations about contextual response but do not justify importing pharmacological placebo effect sizes directly into psychotherapy. A psychotherapy control condition can alter relationships, behavior, learning, and support even when it avoids a named treatment technique.
Natural change also matters. Regression toward the mean, spontaneous remission, concurrent services, social changes, repeated measurement, and statistical selection can all contribute to observed symptom improvement in an uncontrolled treatment group. Randomization balances these influences on average for a specified treatment contrast; it does not reveal which ingredients of the assigned treatment package generated the difference. Claims that common factors account for a fixed share of improvement are especially vulnerable when they compare proportions estimated from heterogeneous designs, correlate overlapping measures, or use residual treatment effects as if they were a direct measure of one mechanism.
Time Ordering, Mediation, and Dynamic Confounding
Causal mediation requires a specified treatment contrast, a proposed intervening variable, and an outcome whose timing permits the proposed chain. In simple notation, assignment A changes a process M, which subsequently changes an outcome Y. The quantity of interest is not obtained merely by correlating M and Y. Preexisting severity, motivation, treatment engagement, therapist skill, and life events can affect both mediator and outcome; early response to treatment can cause a change in mediator rating. Standard regression adjustment for observed baseline covariates may not address time-varying confounding created by treatment itself Kazdin (2007).
Temporal ordering is necessary but insufficient. If alliance at session three predicts depression at session eight after controlling depression at session two, the finding is compatible with a causal alliance effect. It is also compatible with a latent change process that improves both alliance and later symptoms, with adherence that changes both, or with measurement errors in earlier symptoms. Repeated-measures cross-lagged models can separate some within-person from between-person dynamics, but estimates depend on lag selection, stability assumptions, distributional form, and measurement invariance. Dense sampling can make causal accounts more plausible, yet only a well-specified intervention or defensible quasi-experiment addresses some sources of confounding.
Randomized mediator interventions are promising when practical and ethical. For expectancy, researchers might manipulate an accurate rational explanation while preserving clinical content; for alliance, therapists might receive randomized collaborative repair training; for a putative specific mechanism, the exposure schedule or behavioral task might be changed systematically. A mediation claim becomes stronger when a manipulation alters the predicted process, the process precedes improvement, competing pathways are measured, and results replicate independently. Experimental manipulation of a complex process can itself affect several constructs, so the design must test alternative mechanisms Cuijpers et al. (2019).
Mechanism research should distinguish a statistical mediator from a causal mechanism. A statistical mediator helps explain an association under assumptions; a causal mechanism is a process whose interventionally relevant change produces a consequent change in the outcome. Therapeutic alliance, expectancy, and skill acquisition can each satisfy the first definition in some datasets while remaining incompletely validated under the second. This distinction permits productive integration of process research with intervention development, without demanding that clinical decisions await a final universal causal account.
Risk of Bias, Researcher Allegiance, and Selective Evidence
Psychotherapy research faces characteristic challenges in blinding. Treatment participants know whether they are asked to perform exposure or complete cognitive worksheets, and therapists know their assigned manuals. Outcomes based entirely on patient symptom reports may therefore reflect both actual experienced improvement and response tendencies shaped by treatment expectations. Masked independent assessments, functional measures, preregistered primary outcomes, treatment fidelity reporting, and transparent missing-data analysis strengthen inference. Each compensates for a particular vulnerability rather than making the whole trial immune to bias.
Missing outcomes often reflect the very processes under investigation. Disappointment, adverse events, financial barriers, insufficient benefit, treatment burden, and early improvement can each cause dropout. Complete-case analysis can overestimate or underestimate treatment benefit when observed participants differ systematically from those whose outcomes are missing. Researchers should report treatment discontinuation separately from loss to research assessment, compare characteristics of completers and noncompleters, and use principled sensitivity analyses. In an expectancy study, treating dropout as a nuisance may remove one of the pathways through which expectation could influence effectiveness.
Investigator allegiance may affect therapy development, training, supervision, and interpretation. A team may compare its well-refined favored intervention with a poorly supported control package, or therapists may convey more enthusiasm for one condition. This is especially consequential when outcomes are reported by participants aware of which treatment they receive. Multisite trials with independent assessors, balanced clinical expertise, independent data analysis, and accessible treatment manuals can improve the credibility of comparisons, even when complete standardization is impossible.
Meta-analysis increases sample size and precision but inherits the designs, measurement errors, and selection mechanisms of the included evidence. Publication bias and small-study effects may inflate average benefits; categorical therapy labels can combine heterogeneous procedures; network models require transitivity and consistency; meta-analytic mediation can summarize observational associations without identifying causes. In the common-factors debate, evidence quality must be assessed at the level of the exact claim. The existence of hundreds of studies does not automatically provide hundreds of independent randomizations of alliance or expectancy.
Evidence hierarchies should be applied to questions, not merely journal or study labels. A psychometric validation study is central evidence that expectancy can be measured; an observational cohort is informative about natural covariation; a randomized treatment trial evaluates an intervention package; a component trial estimates incremental efficacy; a well-executed factorial experiment can test interactions. Combining their conclusions requires preserving these roles. Descriptive confidence intervals, preregistered outcomes, clinical relevance, replication, and independent verification are essential to scientific synthesis.
Construct Validity and Quantitative Interpretation
A construct is useful only when the measures employed bear an interpretable relation to it. Expectations can be assessed as probabilities, anticipated magnitude of benefit, perceived plausibility, or felt hope. Alliance instruments may combine bond, goals, and tasks into a total score, obscuring dimensions with different predictive functions. Specific treatment mechanisms require measures more proximal to the hypothesized operation, such as the change in avoidance behavior after exposure, altered activity scheduling, or modification of a threat appraisal. A mediator measured on the same questionnaire as the symptom outcome risks criterion overlap.
Construct validity is established through convergent and discriminant evidence, stability or sensitivity over an appropriate interval, invariance across relevant populations, and prediction of external criteria. An alliance measure that is reliable within a treatment school may change meaning when the tasks of therapy are delivered through a digital platform. A credibility questionnaire developed for a particular clinical context may be less valid in a multilingual service or a treatment format where participants never meet a clinician. Investigators should make the referent of each item explicit and report when measures have been adapted Devilly & Borkovec (2000) Bordin (1979).
Effect sizes must be interpreted in their statistical context. A meta-analytic correlation summarizes the association of two observed measures, whereas a standardized mean difference estimates a contrast between treatment groups on a particular outcome scale. An odds ratio or risk ratio depends on the event and group coding, as illustrated by preference and dropout syntheses. Credible intervals in Bayesian component network meta-analysis are not frequentist confidence intervals, and an estimate expressed in PHQ-9 points is not interchangeable with a standardized effect. Apparent disagreements sometimes arise because researchers measure different endpoints, not because the underlying findings are logically inconsistent.
Clinically meaningful change concerns consequences for people, not only statistical detectability. A modest improvement in average depression scores may warrant interest when a low-cost intervention reaches many otherwise untreated people, while an equally sized increase in reported satisfaction may be insufficient if functioning deteriorates. Patient-centered outcomes, work and social functioning, sustained remission, safety, and resource use provide complementary criteria. Claims about universality should be examined against the populations actually enrolled, especially when studies exclude severe risk, comorbidity, or practical barriers.
The 2026 Umbrella Evidence on Psychotherapy Process–Outcome Associations
An umbrella synthesis by Gómez Penedo and Flückiger Gómez Penedo & Flückiger (2026) provides a recent cross-process perspective on the debate. Published in World Psychiatry in 2026, it reviewed 60 meta-analyses covering 24 categories of processes. Forty-eight of the included syntheses furnished primary-study detail from which the authors extracted 1,981 effect sizes representing 993 distinct primary studies. The working alliance was the most frequently meta-analyzed construct, appearing in 25 syntheses; therapist adherence, competence, fidelity, and integrity appeared in four, and expectations in three. This imbalance is substantive: the volume of published evidence about a construct is shaped by its measurement convenience and research tradition, not necessarily its causal importance.
Using multilevel meta-analytic models, the authors reported an overall process–outcome association near r = .238. In their grouped analyses the average associations were approximately r = .102 for technical processes, r = .227 for patient processes, r = .267 for relational processes, and r = .291 for therapist processes Gómez Penedo & Flückiger (2026). These results support the importance of studying a broad spectrum of processes and indicate that relational and therapist categories often show stronger observational associations with outcome than the technical categories sampled in the review. However, the categories differed markedly in how many primary studies and meta-analyses contributed, which technical constructs were included, and how processes were operationalized.
The comparison requires particular causal restraint. A lower correlation for a process labeled technical does not demonstrate that changing a therapeutic technique produces less clinical benefit than changing a relationship process. Treatment techniques may have been measured primarily as adherence or competence, while alliance and empathy were often measured through patient experiences that overlap more directly with outcome reports. The review aggregates associations, including evidence from dependent syntheses handled by multilevel modeling; it does not randomize the process categories. Its findings therefore strengthen the case for process research while preserving the explanatory uncertainty identified in the 2019 review by Cuijpers and colleagues Cuijpers et al. (2019).
The 2026 umbrella review also clarifies the risks of treating common factors as one homogeneous group. Therapeutic relationship, clinician behavior, and patient expectation occupy distinct categories and may form causal chains rather than compete to explain an outcome. The authors explicitly noted that the categories of therapeutic processes operate together during treatment. Future trials can use the evidence map to identify neglected mechanisms and to select process variables for intervention, but every causal attribution still requires an identifiable manipulation, temporally correct measurement, and an appropriate counterfactual.
Recent Evidence on Patient Preference and the Boundary of Generalization
Bennett and colleagues Bennett et al. (2025) published a ten-year update of the meta-analysis on patient preferences, choice, and shared decision making in 2025. The report identified 26 additional randomized trials and synthesized 59 articles describing 60 studies across psychological and medical conditions. The mean reported effects remained modest: standardized effects of about d = .26 for satisfaction and d = .14 for clinical outcome, and an odds ratio around 1.30 for treatment completion. The consistency of the estimates with Lindhiem and colleagues' earlier review Lindhiem et al. (2014) supports a small but replicable association of treatment involvement with several treatment-relevant outcomes.
This newer review also reinforces the need to distinguish study universes. It includes medical and mental-health conditions together and may capture treatment preference, explicit choice, and collaborative decision procedures under one operational umbrella. Windle and colleagues' diagnosis-focused psychotherapy synthesis Windle et al. (2020) reported evidence for reduced dropout and stronger alliance without clear evidence for a significant average symptom benefit. The two reviews can both be correct because they address different populations, definitions, and endpoints. Scientific synthesis should specify the primary estimand and setting before asserting a universal preference effect.
Treatment choice is also an ethical component of care. Preference matching, when feasible among evidence-supported alternatives, can accommodate practical constraints, valued outcomes, and prior experience. In empirical research, however, its mechanism remains open: benefit may arise from autonomy, credibility, improved adherence, greater perceived collaboration, or changes in the actual treatment received. Factorial comparisons of choice opportunities and treatment methods, with repeated assessments of those candidate mediators, would help distinguish these contributions.
A 2026 Map of Mechanism Research and Its Conceptual Limits
Hofmann and colleagues Hofmann et al. (2026) published a scoping review of psychotherapy process research that identified 778 studies reporting 684 named processes since 2007. The authors grouped these processes into 32 clusters using an iterative human-supervised procedure assisted by a language model. Their largest cluster concerned alliance, collaboration, and interpersonal functioning, accounting for about 20.6 percent of the processes classified. The review highlights how extensively the field has elaborated process vocabulary and how important systematic conceptual organization has become.
The mapping exercise and the umbrella meta-analysis answer different questions. The scoping review examines what process constructs researchers have studied and how they can be organized; the umbrella analysis summarizes measured associations between processes and outcomes. Neither demonstrates that 684 independent causal processes exist. Many labels overlap, measures may represent the same psychological operation in different terminologies, and grouping decisions depend on judgment and the chosen taxonomy. The use of machine-assisted classification is a reproducible methodological opportunity when data, prompts, coding rules, and human adjudication are documented; it is not evidence that the classification identifies objective psychological natural kinds.
Together, these recent studies support a more disciplined research program: establish a stable construct vocabulary, map causal hypotheses onto observable variables, identify where evidence is abundant but observational, and prioritize experimental contrasts for neglected candidate mechanisms. This program is especially relevant to an Artificial Era scientific infrastructure in which research retrieval and semantic clustering can scale quickly. Classification systems can aid scientific navigation and hypothesis generation while independent empirical designs remain responsible for verifying therapeutic claims.
Heterogeneity, Cultural Context, and the Meaning of a Credible Treatment
Treatment expectancy is situated in social history. The degree to which an intervention seems plausible can depend on language, prior contact with health institutions, community norms, experiences of discrimination, explanations of suffering, and the social position of the clinician. Cultural adaptation may alter both treatment content and the therapeutic context: changing examples, communication patterns, goals, and accessibility can transform participation as well as credibility. Research comparing adapted interventions should specify these changes separately. An aggregate improvement in an adapted program does not establish whether its effect arose through greater expectancy, more appropriate skill content, improved engagement, or different recruitment Wampold (2015).
Language concordance can affect the accuracy with which clinical goals, emotions, and anticipated benefit are expressed. Translation is not merely replacement of lexical items in a credibility scale or alliance survey. Researchers should examine construct equivalence, response styles, cultural meanings of improvement, and the different social risks of disclosing symptoms. A treatment offered as an evidence-based package in one country may be perceived as foreign, coercive, or irrelevant in another. Its contextual compatibility is a possible moderator of the specific procedure's effectiveness, and its assessment requires qualitative investigation alongside quantitative outcomes.
Patient heterogeneity also arises from severity, chronicity, comorbidity, social resources, disability, and previous treatment experience. A person who can attend regularly and complete between-session work may derive value from a demanding procedure, while someone with unstable housing or unpredictable caregiving demands may face severe barriers that distort exposure to its active elements. Baseline expectations may predict engagement partly because they summarize these unmeasured constraints. Precision-oriented research should distinguish prognostic variables, which predict outcome regardless of treatment, from predictive treatment moderators, which identify relative advantage of one intervention over another.
An equitable scientific program measures who entered a trial and who was excluded, rather than extrapolating an average result to all potential patients. It reports transportation and financial burdens, interpreter access, treatment cost, therapist availability, and whether participants selected their preferred modality. These variables do not diminish the scientific importance of psychological processes. They specify the conditions under which expectancy can be formed, alliance maintained, and procedures implemented.
Clinical Relevance, Harms, and Responsible Interpretation
A scientific account of treatment effects should evaluate more than mean symptom improvement. For individual clinical decisions, treatment burden, acceptability, early deterioration, sustained recovery, functioning, and access to alternatives are relevant endpoints. A therapy can be persuasive and relationally satisfying while producing limited clinical benefit for a particular disorder; a demonstrably effective procedure can also be difficult, distressing, and poorly tolerated by some patients. Expectations should therefore be calibrated by outcome data and informed patient choice rather than used to substitute confidence for clinical evaluation.
Ethically sound expectancy enhancement gives accurate information about plausible benefits, uncertainty, side effects or short-term distress, and the patient's role in therapeutic work. When a protocol is disorder-specific, the benefit claims must reflect relevant trials and guidelines. Attempts to optimize expectations through deception introduce a separate risk to consent and trust. Researchers can investigate transparent rationale framing, normalization of uncertainty, and collaborative goal setting without misrepresenting treatment evidence. Such interventions should include both process outcomes and independent clinical endpoints.
A relationship-centered account similarly needs a method for detecting poor care. Strong reported alliance could reflect social pressure, idealization, or fear of challenging a clinician; low ratings can sometimes reflect necessary boundary setting or initially unwelcome but appropriate safety decisions. Independent outcome measures, adverse-event assessment, mechanisms for complaints and second opinions, and continuous attention to patient autonomy are part of a defensible interpretation of relational quality. This monograph concerns explanatory science and offers no individual treatment prescription; qualified clinicians must interpret evidence in relation to real clinical indications and patient circumstances.
The risk of harm also appears in research inference. If common factors are described as sufficient for every condition, readers may infer that specific validated procedures are dispensable; if a technique is presented as uniquely causal without reliable component evidence, readers may be led to accept a rigid treatment irrespective of preference or context. The scientifically grounded alternative is to integrate the strongest disorder-specific outcome evidence with attention to credibility, alliance, safety, and delivery. Both the confidence of the clinician and the therapeutic actions of the patient deserve study, but neither replaces independent evaluation of clinical effects.
Artificial Era: Digital Delivery and Human–Artificial Therapeutic Contexts
Digital mental health care separates the content of an intervention from some traditional features of therapist contact. A structured online program can present a credible rationale, guide behavioral exercises, measure progress, and provide reminders; a clinician may supply guidance asynchronously, or no human guidance may be provided. As a research setting, digital delivery makes some procedures reproducible and supports randomization of interface descriptions, task schedules, and encouragement. It also changes the social setting in which trust, expectation, and adherence arise. The distinction among contextual, relational, and specific operations becomes especially consequential Furukawa et al. (2021) Karyotaki et al. (2021).
The 2021 component analysis of internet-based CBT Furukawa et al. (2021) estimated that combining human and automated encouragement reduced dropout, while some therapeutic components had different estimated outcome contributions. Karyotaki and colleagues Karyotaki et al. (2021) found that the advantage of therapeutic guidance for internet CBT could vary with initial depression severity. These findings demonstrate that delivery support can function as a candidate ingredient and potential moderator in digital psychotherapy. They do not establish that a text-generating conversational system can provide equivalent clinical care, diagnose disorders, or independently deliver safe treatment.
Artificial systems can be studied as sources of procedural instruction, conversational support, behavioral monitoring, and expectation-shaping messages. A research program should operationalize whether perceived responsiveness is measured by user judgments, independently coded linguistic contingencies, actual task responsiveness, or stable outcomes over time. Anthropomorphic attribution to a system may increase reported credibility, yet it may also obscure limits in clinical competence. Human–Artificial research should explicitly compare perceived relationship, psychological reliance, actual task performance, and safe escalation to qualified human care.
The Artificial Era framework asks how therapeutic functions are distributed among patient, clinician, institution, and artificial systems. Distribution is a hypothesis about the organization of care and evidence-producing interactions. It does not, by itself, establish psychological properties or clinical authority in an artificial agent. The relevant experimental questions are whether transparent human–artificial delivery improves adherence, makes specific skills accessible, changes informed expectations, or introduces new risks, and whether these effects replicate across populations. The scholarly priority remains verifiable clinical outcomes and explicit distinctions between operational observation, causal explanation, and philosophical interpretation.
Integrative Synthesis and an Experimental Research Agenda
The cumulative evidence supports a layered conception of therapeutic change. Treatment expectations modestly predict later outcomes across diverse studies; working alliance and therapist empathy show robust associations with outcome; accommodating preferences can improve engagement and may affect clinical benefit under some study definitions; and several specific intervention components have demonstrated clinical utility or informative incremental contrasts. None of these patterns licenses a universal additive partition of “how much of psychotherapy works” into fixed percentages for relationship, expectation, and technique. The processes may interact, substitute for each other, and operate at different temporal scales Constantino et al. (2018b) Flückiger et al. (2018) Elliott et al. (2018) Furukawa et al. (2021).
An integrative causal model can be specified as follows. At intake, prior illness history, contextual resources, explanatory beliefs, and patient preference shape initial expectation. The therapeutic rationale and early interpersonal experience alter credibility and task agreement. Active procedures then create opportunities for learning, mastery, and changes in behavior. Early outcome feedback revises expectation and alliance, while adherence mediates the amount of active practice received. Therapist skill and access constraints influence all these transitions. The model is a structured research hypothesis whose component pathways require separate tests; its scientific value lies in generating discriminating predictions and transparent measurement plans.
A rigorous program would begin with preregistered multiarm trials that compare evidence-supported treatment packages and credible active controls; add factorial manipulation of rationale and one candidate technique; measure expectancy, credibility, alliance, task exposure, and independent outcomes repeatedly; and follow participants beyond acute treatment. It would model therapist clustering and treatment-by-patient interactions, quantify missing-data sensitivity, and report clinically interpretable differences. Sequential randomization of process interventions could probe whether changing a mediator at a particular session alters later functioning. Replication at independent sites would assess transportability beyond one therapist team or national health system.
Scientific progress will also depend on comparing models by predictions rather than terminological loyalty. A common-factors model should predict which changes occur when credibility or relationship is experimentally perturbed while techniques are held comparable. A specific-mechanism model should predict how altering the hypothesized learning operation affects proximal mediators and distal outcomes. An interaction model should predict, for example, that the effect of a difficult exposure task depends on collaboration or that early mastery changes the subsequent benefit of encouragement. Each can be tested with explicit time ordering, adequate comparator integrity, and appropriately scoped generalization.
The final explanatory object is the person undergoing treatment in a changing context, not a label assigned to a school. Meaningful clinical understanding requires connecting what the therapist and patient do, what they expect, how they relate, what they learn, and how their psychological and functional trajectories change. Evidence is strongest when these components are measured independently, experimentally varied when possible, and interpreted according to the actual contrast.
Conclusion
Treatment expectancy, common factors, and specific effects are interdependent constructs in psychotherapy research. The evidence assembled in this monograph establishes consistent predictive associations for expectations, alliance, and empathy, evidence for the engagement value of accommodating preferences, and experimental findings that distinguish some intervention components from others in specific clinical contexts. Credible rationales, collaborative relationships, and active procedures are all plausible contributors to improvement. Their respective causal roles remain contingent on measurement, time ordering, comparator conditions, and the nature of the clinical problem.
The historical debate has matured from broad claims about whether all therapy schools succeed equally to a set of more discriminating questions. Which expectations can be changed accurately and ethically? Which aspects of alliance precede symptom improvement and respond to targeted interventions? Which treatment components produce incremental effects under credible active comparisons? How do therapist expertise, patient preference, severity, accessibility, and culture modify these relationships? Contemporary component trials and individual-patient data syntheses begin to answer particular instances of these questions while demonstrating that the effects of procedures and contexts cannot be assigned universal constant weights.
The principal methodological conclusion follows directly: associations identify candidates for explanation, randomized package contrasts identify comparative effects, component manipulations identify more localized contributions, and repeated mediator perturbations are needed for stronger causal mechanism claims. The strongest future evidence will combine these levels within transparent, replicated clinical experiments and longitudinal observation. Such research can support treatment choice that respects scientific specificity, patient agency, relational quality, and the practical conditions of care.
In the Artificial Era, digital and human–artificial forms of delivery create additional opportunities to separate instruction, encouragement, monitoring, credible explanation, and interpersonal contact. These possibilities expand the field of experimental inquiry while reinforcing the need for valid constructs, independent outcomes, and rigorous safety oversight. The enduring scientific task is to explain therapeutic improvement through testable causal organization rather than attributing it in advance to a single favored ingredient.
References
- Ahn, H.-n., & Wampold, B. E. (2001). Where oh where are the specific ingredients? A meta-analysis of component studies in counseling and psychotherapy. Journal of Counseling Psychology, 48(3), 251–257. https://doi.org/10.1037/0022-0167.48.3.251
- Bennett, C. B., Toevs, E. K., Angus, A., Krawczak, R., Trentacosta, C. J., Jagadeesh, D., Tomlinson, C. S., & Lindhiem, O. (2025). Meta-analysis on the effects of client preferences on treatment satisfaction, completion, and clinical outcome: A 10-year update. Clinical Psychology Review, 121, Article 102644. https://doi.org/10.1016/j.cpr.2025.102644
- Bordin, E. S. (1979). The generalizability of the psychoanalytic concept of the working alliance. Psychotherapy: Theory, Research & Practice, 16(3), 252–260. https://doi.org/10.1037/h0085885
- Ciharová, M., Furukawa, T. A., Efthimiou, O., Karyotaki, E., Miguel, C., Noma, H., Cipriani, A., Riper, H., & Cuijpers, P. (2021). Cognitive restructuring, behavioral activation and cognitive-behavioral therapy in the treatment of adult depression: A network meta-analysis. Journal of Consulting and Clinical Psychology, 89(6), 563–574. https://doi.org/10.1037/ccp0000654
- Constantino, M. J., Coyne, A. E., Boswell, J. F., Iles, B. R., & Vîslă, A. (2018). A meta-analysis of the association between patients’ early perception of treatment credibility and their posttreatment outcomes. Psychotherapy, 55(4), 486–495. https://doi.org/10.1037/pst0000168
- Constantino, M. J., Coyne, A. E., Goodwin, B. J., Vîslă, A., Flückiger, C., Muir, H. J., & Gaines, A. N. (2021). Indirect effect of patient outcome expectation on improvement through alliance quality: A meta-analysis. Psychotherapy Research, 31(6), 711–725. https://doi.org/10.1080/10503307.2020.1851058
- Constantino, M. J., Vîslă, A., Coyne, A. E., & Boswell, J. F. (2018). A meta-analysis of the association between patients’ early treatment outcome expectation and their posttreatment outcomes. Psychotherapy, 55(4), 473–485. https://doi.org/10.1037/pst0000169
- Coyne, A. E., Constantino, M. J., Muir, H. J., Gaines, A. N., & Vîslă, A. (2023). Participant factors as correlates of patients’ psychotherapy outcome expectation: A meta-analytic and box-count review. Psychotherapy Research, 33(7), 974–988. https://doi.org/10.1080/10503307.2023.2197629
- Craske, M. G., Kircanski, K., Zelikowsky, M., Mystkowski, J., Chowdhury, N., & Baker, A. (2008). Optimizing inhibitory learning during exposure therapy. Behaviour Research and Therapy, 46(1), 5–27. https://doi.org/10.1016/j.brat.2007.10.003
- Cuijpers, P., Cristea, I. A., Karyotaki, E., Reijnders, M., & Huibers, M. J. H. (2016). How effective are cognitive behavior therapies for major depression and anxiety disorders? A meta-analytic update of the evidence. World Psychiatry, 15(3), 245–258. https://doi.org/10.1002/wps.20346
- Cuijpers, P., Reijnders, M., & Huibers, M. J. H. (2019). The role of common factors in psychotherapy outcomes. Annual Review of Clinical Psychology, 15, 207–231. https://doi.org/10.1146/annurev-clinpsy-050718-095424
- Devilly, G. J., & Borkovec, T. D. (2000). Psychometric properties of the credibility/expectancy questionnaire. Journal of Behavior Therapy and Experimental Psychiatry, 31(2), 73–86. https://doi.org/10.1016/S0005-7916(00)00012-4
- Dimidjian, S., Hollon, S. D., Dobson, K. S., Schmaling, K. B., Kohlenberg, R. J., Addis, M. E., Gallop, R., McGlinchey, J. B., Markley, D. K., Gollan, J. K., Atkins, D. C., Dunner, D. L., & Jacobson, N. S. (2006). Randomized trial of behavioral activation, cognitive therapy, and antidepressant medication in the acute treatment of adults with major depression. Journal of Consulting and Clinical Psychology, 74(4), 658–670. https://doi.org/10.1037/0022-006X.74.4.658
- Dobson, K. S., Hollon, S. D., Dimidjian, S., Schmaling, K. B., Kohlenberg, R. J., Gallop, R., Rizvi, S. L., Gollan, J. K., Dunner, D. L., & Jacobson, N. S. (2008). Randomized trial of behavioral activation, cognitive therapy, and antidepressant medication in the prevention of relapse and recurrence in major depression. Journal of Consulting and Clinical Psychology, 76(3), 468–477. https://doi.org/10.1037/0022-006X.76.3.468
- Elliott, R., Bohart, A. C., Watson, J. C., & Murphy, D. (2018). Therapist empathy and client outcome: An updated meta-analysis. Psychotherapy, 55(4), 399–410. https://doi.org/10.1037/pst0000175
- 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. 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. 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. https://doi.org/10.1037/ccp0000594
- Foa, E. B., Hembree, E. A., Cahill, S. P., Rauch, S. A. M., Riggs, D. S., Feeny, N. C., & Yadin, E. (2005). Randomized trial of prolonged exposure for posttraumatic stress disorder with and without cognitive restructuring: Outcome at academic and community clinics. Journal of Consulting and Clinical Psychology, 73(5), 953–964. https://doi.org/10.1037/0022-006X.73.5.953
- Foa, E. B., Rothbaum, B. O., Riggs, D. S., & Murdock, T. B. (1991). Treatment of posttraumatic stress disorder in rape victims: A comparison between cognitive-behavioral procedures and counseling. Journal of Consulting and Clinical Psychology, 59(5), 715–723. https://doi.org/10.1037/0022-006X.59.5.715
- Frank, J. D., & Frank, J. B. (1991). Persuasion and healing: A comparative study of psychotherapy (3rd ed.). Johns Hopkins University Press. https://books.google.com/books?id=-S8pAAAAYAAJ
- Furukawa, T. A., Suganuma, A., Ostinelli, E. G., Andersson, G., Beevers, C. G., Shumake, J., Berger, T., Boele, F. W., Buntrock, C., Carlbring, P., Choi, I., Christensen, H., Mackinnon, A., Dahne, J., Huibers, M. J. H., Ebert, D. D., et al. (2021). Dismantling, optimising, and personalising internet cognitive behavioural therapy for depression: A systematic review and component network meta-analysis using individual participant data. The Lancet Psychiatry, 8(6), 500–511. https://doi.org/10.1016/S2215-0366(21)00077-8
- Gómez Penedo, J. M., & Flückiger, C. (2026). Process-outcome effects in psychotherapy research: An umbrella systematic review and meta-analysis. World Psychiatry, 25(1), 151–152. https://doi.org/10.1002/wps.70018
- Hofmann, S. G., Pahlen, L., Westhoff, M., Hossner, K., Quintarelli, H., Usée, F., Stenzel, K., & Stangier, U. (2026). Processes in psychotherapy: A scoping review with LLM-assisted clustering. Clinical Psychology Review, 128, Article 102782. https://doi.org/10.1016/j.cpr.2026.102782
- Hróbjartsson, A., & Gøtzsche, P. C. (2010). Placebo interventions for all clinical conditions. Cochrane Database of Systematic Reviews, 2010(1), CD003974. https://doi.org/10.1002/14651858.CD003974.pub3
- Jacobson, N. S., Dobson, K. S., Truax, P. A., Addis, M. E., Koerner, K., Gollan, J. K., Gortner, E., & Prince, S. E. (1996). A component analysis of cognitive-behavioral treatment for depression. Journal of Consulting and Clinical Psychology, 64(2), 295–304. https://doi.org/10.1037/0022-006X.64.2.295
- Karyotaki, E., Efthimiou, O., Miguel, C., Maas genannt Bermpohl, F., Furukawa, T. A., Cuijpers, P., et al. (2021). Internet-based cognitive behavioral therapy for depression: A systematic review and individual patient data network meta-analysis. JAMA Psychiatry, 78(4), 361–371. https://doi.org/10.1001/jamapsychiatry.2020.4364
- Kazdin, A. E. (2007). Mediators and mechanisms of change in psychotherapy research. Annual Review of Clinical Psychology, 3, 1–27. https://doi.org/10.1146/annurev.clinpsy.3.022806.091432
- Lindhiem, O., Bennett, C. B., Trentacosta, C. J., & McLear, C. (2014). Client preferences affect treatment satisfaction, completion, and clinical outcome: A meta-analysis. Clinical Psychology Review, 34(6), 506–517. https://doi.org/10.1016/j.cpr.2014.06.002
- Rogers, C. R. (1957). The necessary and sufficient conditions of therapeutic personality change. Journal of Consulting Psychology, 21(2), 95–103. https://doi.org/10.1037/h0045357
- Rosenzweig, S. (1936). Some implicit common factors in diverse methods of psychotherapy. American Journal of Orthopsychiatry, 6(3), 412–415. https://doi.org/10.1111/j.1939-0025.1936.tb05248.x
- Swift, J. K., Callahan, J. L., Cooper, M., & Parkin, S. R. (2018). The impact of accommodating client preference in psychotherapy: A meta-analysis. Journal of Clinical Psychology, 74(11), 1924–1937. https://doi.org/10.1002/jclp.22680
- Wampold, B. E. (2015). How important are the common factors in psychotherapy? An update. World Psychiatry, 14(3), 270–277. https://doi.org/10.1002/wps.20238
- Windle, E., Tee, H., Sabitova, A., Jovanovic, N., Priebe, S., & Carr, C. (2020). Association of patient treatment preference with dropout and clinical outcomes in adult psychosocial mental health interventions: A systematic review and meta-analysis. JAMA Psychiatry, 77(3), 294–302. https://doi.org/10.1001/jamapsychiatry.2019.3750