RESEARCH DOSSIER
Psychotherapy Mechanisms, Mediators, and Process Research
Abstract. Psychotherapy process research investigates how psychological treatment produces change, which treatment operations precede improvement, and under which conditions these operations carry causal influence. This research dossier reconstructs the historical development of process–outcome research, differentiates processes, mediators, mechanisms, moderators, and outcomes, and critically integrates evidence from therapeutic alliance studies, expectation research, treatment-component experiments, cognitive and behavioral change studies, exposure learning, psychodynamic and relational research, and modern longitudinal causal methodology. The method is a structured critical narrative synthesis of original trials, observational session studies, methodological papers, and peer-reviewed research reviews available through November 2026. The strongest evidence establishes several reproducible process–outcome associations and demonstrates that specific experimentally manipulated treatment packages or components can alter clinical outcomes. Evidence supporting an identified process as the necessary and sufficient causal pathway of improvement is substantially more limited. Temporality, bidirectional feedback, changing treatment dose, measurement error, and confounding by patient and therapist characteristics complicate interpretation. The monograph develops an evidence-sensitive framework that connects manipulations of therapeutic operations, high-frequency observations, within-person modeling, mechanistic tests, patient heterogeneity, adverse outcomes, and implementation contexts. It identifies the design requirements needed to move from plausible therapeutic narratives toward credible explanations of psychological change.
Keywords: psychotherapy process research; mechanisms of change; mediators; causal mediation; moderators; therapeutic alliance; rupture and repair; treatment expectancy; common factors; specific factors; cognitive therapy; behavioral activation; exposure therapy; psychodynamic psychotherapy; longitudinal methods; randomized dismantling; within-person change; process measurement; treatment fidelity; clinical outcomes; psychotherapy safety
Introduction: Explaining Therapeutic Change
Psychotherapy changes the conditions under which people think, feel, act, remember, anticipate, and relate to others. Its clinical record is extensive, but a treatment's ability to improve an outcome and an account of why it works are different scientific propositions. An efficacy trial estimates a contrast between assigned treatment conditions. A process investigation describes what happens within treatment. A mechanistic investigation asks whether a defined change-producing operation participates causally in the observed outcome. Kazdin, A. E. (2007) established this distinction as a central unfinished task of psychotherapy science, arguing that well-supported treatments often lack comparably well-supported explanations of their effects.
The research object of this dossier is the temporal, interpersonal, behavioral, cognitive, and contextual organization of psychological change in treatment. Its basic units include patients, therapists, sessions, interactions, homework episodes, target behaviors, and repeated measures between sessions. The principal questions concern how interventions are delivered, what intermediate events they alter, whether such events precede the outcome, and whether manipulating the intermediate events would alter the outcome under otherwise comparable conditions. These questions extend beyond selecting a favored school of psychotherapy. A cognitive reconstruction of beliefs, an exposure-based change in threat prediction, and a repaired therapeutic disagreement may each be candidates for process research, but each requires a distinct operational test.
A clinically useful explanation must specify a population, treatment contrast, mechanism measure, relevant timescale, endpoint, and counterfactual. Generalizations about all psychotherapy risk concealing important variation: an immediate change in fear responding during a session may predict later approach behavior, while an increase in collaborative engagement may influence whether patients return for another session. In the same treatment the two processes may interact. The central scientific commitment is therefore to multi-level evidence and correctly delimited causal claims, drawing particularly on Kraemer, H. C., Wilson, G. T., Fairburn, C. G., & Agras, W. S. (2002) and Imai, K., Keele, L., & Tingley, D. (2010).
Research Method, Corpus, and Evidentiary Boundaries
This monograph uses a structured critical narrative review. Its source set was selected to represent conceptual milestones, original randomized component studies, session-level and longitudinal observational research, systematic reviews and meta-analyses, and statistical methods for causal mediation. Primary publication records, journal abstracts, accessible full-text research, and official bibliographic metadata were inspected to establish the identity and scope of each cited work. The corpus is purposive and theory-oriented. It is not a preregistered exhaustive systematic review, does not calculate pooled estimates independently, and contains no original participant data. Findings are reported with explicit distinctions between available abstracts, available full texts, and inferential synthesis.
The evidence is organized by the question a design can answer. Random assignment can support a causal contrast between assigned treatment packages or assigned components under trial assumptions. A temporal association between an earlier process measure and a later symptom measure supports a directional predictive relationship but may still reflect time-varying confounding. A post-treatment mediation coefficient ordinarily requires strong unverified assumptions before it can be interpreted as a causal indirect effect. A credible mechanism therefore requires convergent evidence from several designs, not an isolated statistical significance test. Cuijpers, P., Reijnders, M., & Huibers, M. J. H. (2019) systematically articulated this gap between confident treatment effectiveness and uncertain mechanism identification.
The present synthesis evaluates measurement validity, temporal precedence, risk of reverse causation, treatment integrity, multiplicity of candidate mediators, treatment-comparison quality, missingness and dropout, unobserved confounding, transportability, and clinical importance. It also distinguishes theories of treatment from measurements that actually test them. The primary strength of this approach is conceptual integration across psychotherapy traditions; its principal limitation is the selective search procedure, which precludes claims of comprehensive coverage or precise pooled prevalence of a mechanistic finding.
Core Concepts: Process, Mediator, Mechanism, and Moderator
A therapeutic process is an observable or inferable sequence of events within or surrounding treatment. A process measure might record therapist questions, patient self-disclosure, changes in avoidance, shifts in interpersonal collaboration, or practice of an assigned skill. A mediator is an intermediate variable in a statistical or causal model connecting an exposure or treatment assignment to an outcome. A mechanism is a specified change-producing causal operation: a process that, if altered by a suitable intervention, contributes to the clinical endpoint. The concepts overlap, but they correspond to distinct evidentiary demands. Baron, R. M., & Kenny, D. A. (1986) differentiated mediator and moderator variables; MacKinnon, D. P., Fairchild, A. J., & Fritz, M. S. (2007) subsequently situated the analysis of indirect effects in a broader methodological framework.
A moderator determines whether or how strongly the treatment effect varies across a characteristic or context: baseline severity, age, concurrent medication, treatment preference, therapist expertise, or service setting may modify an outcome contrast. Mediation concerns a proposed pathway through which treatment produces effects. A predictor forecasts outcome without necessarily being part of a pathway. A correlate covaries with process or outcome without determining causal direction. A treatment ingredient is an operation included in an intervention protocol; it becomes an empirically demonstrated active ingredient only after an appropriate causal comparison. These categories should be recorded separately in a scientific claim ledger.
Therapeutic alliance provides an instructive example. A high alliance rating may predict improvement because collaboration facilitates therapeutic tasks, because initial improvement leads to a stronger alliance, because both arise from patient optimism, or because therapist competence affects each. Different data-generating processes can produce superficially similar correlations. An alliance–outcome coefficient is thus evidence about association; a treatment-induced alliance change that reliably precedes symptom change adds temporal evidence; randomized influence on the alliance with subsequent symptom effects offers stronger mechanistic leverage. Zilcha-Mano, S. (2017) analyzes precisely these issues.
The Historical Formation of Psychotherapy Process Research
Early psychological treatments developed explicit explanatory languages before the necessary empirical designs became routine. Psychoanalytic accounts emphasized unconscious conflict, insight, and relational repetition; behavior therapies focused on learning contingencies and exposure; client-centered therapy emphasized a relational environment for change; and cognitive therapies proposed the modification of maladaptive appraisals and underlying beliefs. Each tradition contributed candidate processes and observational methods. The explanatory history of psychotherapy is consequently a history of increasingly testable propositions rather than a single convergent theory.
Rogers, C. R. (1957) framed therapeutic personality change through conditions of psychological contact, congruence, empathy, and regard, establishing an influential relational hypothesis. Bordin, E. S. (1979) advanced a cross-orientation account of the working alliance in terms of goals, tasks, and bond. These accounts still shape process coding and alliance measurement. Their historical significance is secure; the sufficiency of any proposed condition for clinical improvement remains an empirical proposition requiring specific tests.
By the later twentieth century, randomized treatment comparisons established the effectiveness of multiple approaches, while process–outcome studies examined whether what therapists did and what patients experienced accounted for variation in results. Baron, R. M., & Kenny, D. A. (1986) influenced the statistical language of mediation, and Howard, K. I., Moras, K., Brill, P. L., Martinovich, Z., & Lutz, W. (1996) emphasized monitoring the actual patient's progress. The field subsequently moved toward repeated session measures, randomized component designs, multilevel models, and counterfactual mediation. Modern approaches add digital trace data, intensive longitudinal measurement, and person-specific predictions, but these increase explanatory validity only when design and causal identification improve alongside measurement density.
Causal Explanation and the Problem of the Indirect Effect
A conventional mediation diagram represents treatment assignment as T, an intermediate change as M, and outcome as Y. A statistical analysis may estimate relationships T→M and M→Y conditional on T and covariates. Even when T is randomized, M is rarely randomized. Patients who change in M may differ in unmeasured ways that independently influence Y. In addition, symptom improvement may occur before, alongside, or as a cause of a measured process change. The estimated indirect association consequently requires more than a significant product of regression coefficients to justify a causal interpretation.
Kraemer, H. C., Wilson, G. T., Fairburn, C. G., & Agras, W. S. (2002) proposed disciplined distinctions between moderators and mediators in randomized clinical trials. Imai, K., Keele, L., & Tingley, D. (2010) formalized counterfactual mediation and sensitivity analysis. Their approaches make the key assumptions visible: consistency of observed and counterfactual variables, appropriate comparison of treatment conditions, and restrictions on unmeasured confounding of mediator–outcome relationships. When treatment itself changes later confounders, standard natural indirect-effect identification becomes especially difficult.
A scientifically informative causal model identifies plausible alternative arrows. Baseline distress can influence both alliance and later symptom improvement. Therapist responsiveness can influence alliance and treatment implementation. Prior symptom change can influence expectations; expectations may alter engagement; engagement may alter exposure dose. A longitudinal directed graph, accompanied by a measurement schedule and recorded treatment decisions, makes these relationships explicit. Claims that a measured mediator is the unique active process should therefore be proportionate to the design, especially in multimodal therapies where several ingredients change together.
Temporality, Bidirectionality, and Treatment Feedback
A mechanism must operate on a timescale compatible with its hypothesized effects. An interpretation can alter immediate appraisal; repeated behavior practice may modify expectancies over days or weeks; shifts in social functioning may depend on opportunities accumulated across months. Measures taken only before and after treatment cannot distinguish many of these possibilities. A mediator assessed at the last session and symptoms assessed at discharge provide little evidence that the intermediate variable changed first. Kazdin, A. E. (2007) made temporality and alternative-mechanism testing central criteria for progress.
Time is also reciprocal. Improvement can strengthen hope and collaboration, thereby increasing subsequent participation. Conversely, intense treatment tasks may initially raise distress while later reducing avoidance. This means a process coefficient may vary in sign according to the assessment interval. Results from daily self-monitoring, weekly questionnaires, and three-month outcome interviews should not be collapsed as if they represent the same mechanism. Sessions differ in content, and intervention dose itself responds to patient progress.
Curran, P. J., & Bauer, D. J. (2011) distinguished within-person from between-person associations in longitudinal models. Hamaker, E. L., Kuiper, R. M., & Grasman, R. P. P. P. (2015) showed how conventional cross-lagged panel analysis can misrepresent within-person dynamics when stable between-person differences remain unseparated. Applied to psychotherapy, patients with generally strong alliance ratings may have better outcomes than other patients, yet it is a different question whether a particular patient's alliance increasing above their own usual level predicts their later improvement.
The Therapeutic Alliance: Replicable Association and Interpretive Limits
Alliance is among the best replicated process correlates in psychotherapy. Flückiger, C., Del Re, A. C., Wampold, B. E., & Horvath, A. O. (2018) synthesized 295 independent studies involving more than 30,000 patients and reported an overall face-to-face alliance–outcome correlation of r = .278 (95% CI .256–.299). The finding spans modalities and settings and gives a quantitative basis for regarding alliance as an important clinical and research construct. A correlation of this magnitude, however, does not establish that changing alliance experimentally produces an equivalent clinical effect.
An earlier observational study by Barber, J. P., Connolly, M. B., Crits-Christoph, P., Gladis, L., & Sique (2000) examined 86 patients receiving supportive-expressive dynamic psychotherapy. Alliance measured at several sessions predicted later change in depression after adjustment for earlier symptom change, though associations with prior improvement also appeared at some time points. The analysis is informative because it separates parts of the temporal sequence; unmeasured relational and patient-state factors remain possible explanations.
The meta-analysis by Flückiger, C., Del Re, A. C., Wlodasch, D., Horvath, A. O., Solomonov, N (2020) examined adjusted alliance–outcome associations in 60 independent samples with 6,061 participants. Associations persisted across adjustments for measured intake and treatment-process characteristics, with reported correlations ranging approximately from .23 to .31. This strengthens the proposition that alliance carries information beyond several assessed covariates. It does not remove all possible common causes, and the optimal clinical policy still requires studies that manipulate or support relevant alliance processes.
Session-Level Alliance Dynamics and Within-Patient Change
The longitudinal turn in alliance research addresses the limitations of averaging across people. Falkenström, F., Granström, F., & Holmqvist, R. (2013) studied session-by-session alliance and symptoms in 646 primary-care psychotherapy patients. Their analysis found that alliance predicted subsequent symptom improvement while symptom change also influenced alliance. This reciprocal pattern is theoretically substantial: therapeutic collaboration and relief can coevolve, producing self-reinforcing or destabilizing sequences rather than a single unidirectional chain.
Falkenström, F., Ekeblad, A., & Holmqvist, R. (2016) further examined whether improvements in the working alliance in one session predicted subsequent improvement in depressive symptoms. Such designs use change relative to a patient's own trajectory to address a more clinically actionable question than comparisons of patients who generally rate alliance differently. The appropriate interpretation remains time-indexed prediction with a stronger causal rationale, not complete causal identification.
Session-level studies face further difficulties. Alliance scores can share method variance with symptom self-report, patient characteristics can influence the meaning of scores, and therapy termination can selectively remove patients with weak alliances. Repeated measurement may be beneficial clinically but may itself change disclosure and monitoring. An adequate design measures changes in treatment and symptoms at defensible lags, specifies response-contingent therapist actions, separates stable and momentary variance, and assesses whether results replicate across disorders, cultures, and health systems.
Rupture, Repair, and the Microstructure of Cooperation
An alliance rupture is a clinically meaningful tension in goals, tasks, or relational trust. Withdrawal, compliance without engagement, frustration, or explicit confrontation may signal different rupture configurations. Repair involves recognition, negotiation, clarification, and collaborative reorganization. The potential mechanism is not mere interpersonal warmth: it may include renewed participation in a therapeutic task, increased willingness to reveal conflict, greater accuracy of joint case formulation, and renewed access to corrective interpersonal experiences.
Eubanks, C. F., Muran, J. C., & Safran, J. D. (2018) summarized 11 studies comprising 1,314 patients and reported an association between rupture resolution and positive outcome of r = .29, with a reported 95% confidence interval of .10–.47. Their second synthesis evaluated rupture-resolution training or supervision. These findings make repair research more specific than global alliance ratings because the target is a temporally localized interaction; the observed relationships still reflect heterogeneous methods and do not amount to proof that a single scripted repair operation produces the measured effect in every patient.
Mechanistic investigation can compare therapist behavior before and after an identified rupture, code patient responses independently of therapist judgments, and examine whether repair leads to engagement or symptom changes over appropriate intervals. Culturally conditioned expectations of authority, disclosure, and confrontation alter how an interaction is interpreted. A therapy process model should therefore measure the function of an exchange rather than assume that the same verbal event has a universal meaning.
Expectations, Credibility, and Therapeutic Meaning
Treatment expectations refer to patients' anticipations about the consequences of entering therapy; treatment credibility concerns whether the approach seems coherent and believable. Both can influence willingness to undertake difficult therapeutic tasks, interpretation of short-term discomfort, and continuation after setbacks. Their clinical plausibility makes them attractive mechanism candidates. Yet expectations may also summarize initial severity, prior treatment experiences, perceived therapist competence, and early symptom change.
A meta-analysis of early treatment outcome expectations reported 81 independent samples and 12,722 patients, with a weighted expectation–outcome correlation of r = .18 and substantial heterogeneity, in the update indexed at Constantino, M. J., Vîslă, A., Coyne, A. E., & Boswell, J. F. (2018). The estimate supports a reproducible association; it does not show that artificially increasing expectation will necessarily improve outcomes by the same amount. Constantino, M. J., Coyne, A. E., Boswell, J. F., Iles, B. R., & Vîslă, (2018) separately synthesized treatment credibility, emphasizing an adjacent but distinct construct.
Expectancy manipulation is methodologically delicate. Presenting clear evidence and an understandable rationale is ethically compatible with collaborative care; deceptive promises are not. Randomized comparison of equally credible interventions, measurement of shifting expectations, and controls for early symptom gains help distinguish contextual meaning from specific therapeutic operations. Cuijpers, P., Reijnders, M., & Huibers, M. J. H. (2019) conclude that neither common nor specific mechanisms had yet satisfied all the criteria for full causal validation in their review, a caution that applies directly to strong claims about expectancy.
Therapist Adherence, Competence, and Individual Therapist Effects
Treatment adherence records whether specified procedures are delivered; competence concerns the quality and appropriateness of delivery. These measures can disagree: a therapist can follow a manual closely while applying a technique at an unhelpful moment, or adapt a protocol responsibly while remaining faithful to its therapeutic rationale. The distinction is crucial for evaluating active ingredients because procedural presence, patient receipt, and actual psychological change are three different observations.
Webb, C. A., DeRubeis, R. J., & Barber, J. P. (2010) reviewed 36 studies and reported mean weighted correlations of approximately .02 for adherence–outcome and .07 for competence–outcome, neither statistically different from zero in that synthesis. The null average does not prove therapist implementation irrelevant. The reviewed studies differed in therapies, instruments, variance of adherence, control for alliance, and potentially in range restriction created by training. It instead shows why simple total-session competence scores may be insufficient tests of an active therapeutic operation.
Therapist effects also span stable practitioner characteristics, case mix, clinical decisions and patient–therapist matching. Hierarchical designs must distinguish patients nested within therapists and therapists nested within settings. A therapist effect on outcome cannot automatically identify a mechanism, because the effective behaviors may remain unmeasured. Stronger studies combine independent session coding, clinician randomization where feasible, patient-level outcomes, temporal process models, and appropriately powered therapist-level comparisons.
Common Factors, Specific Operations, and Explanatory Competition
The common-factors research program studies change-relevant conditions shared across different psychotherapies, including credible rationales, collaboration, therapist responsiveness, and expectations. Specific-process research investigates operations distinctive to a protocol, such as exposure to feared cues, a behavioral activation schedule, or a form of interpretation. These programs need not be mutually exclusive, because a credible relationship may enable a patient to perform exposure, while exposure success may strengthen collaborative trust. A serious causal model allows interactions and sequences rather than presuming all treatment effects belong to one category.
Wampold, B. E., & Imel, Z. E. (2015) articulate a contextual account of therapeutic effectiveness in which expectations and the socially situated healing relationship have prominent roles. Cuijpers, P., Reijnders, M., & Huibers, M. J. H. (2019) examine arguments for common factors against alternatives and note that broadly similar average outcomes across different psychotherapies do not logically prove a shared causal mechanism. Comparable outcomes can arise from different mechanisms reaching the same endpoint, insufficient statistical power to distinguish treatments, unequal quality of comparators, or a mixture of shared and distinctive processes.
A treatment mechanism is established by evidence about the process, not merely by a school's account of that process. Moreover, a shared process may have context-dependent effectiveness and a specific component may operate through processes found in other interventions. The research agenda is to estimate when a proposed ingredient matters, for whom, at what treatment stage, and through what observed change. The resulting synthesis has greater explanatory resolution than an undifferentiated verdict about which entire psychotherapy tradition is superior.
Randomized Dismantling and Component Analysis
A dismantling trial removes a theoretically essential component from a treatment while preserving other elements as much as possible. If the dismantled treatment performs similarly, this challenges the claim that the removed component is necessary under the studied circumstances. It does not demonstrate that the component has no effect in all populations, that the remaining treatment operates through the same mechanisms, or that the omitted procedure was delivered at a sufficient dose in the original package.
Ahn, H.-n., & Wampold, B. E. (2001) synthesized 27 component studies and found no significant average benefit attributable to the theoretically designated components in their analyzed set. The result was influential because it demanded empirical accountability from treatment theories. Its interpretation remains bounded by the component studies actually available, which varied in the quality of dismantling, protocol fidelity, outcome choice, and statistical power. A component may matter in interaction with other techniques even when its isolated removal produces a small average contrast.
More informative component experiments manipulate a tightly defined intervention, measure whether it actually changes the hypothesized mediator, and then evaluate downstream outcomes. They can use factorial designs to identify combinations, micro-randomization to examine proximal effects, or sequential assignment to study adaptations. A component contrast has a meaningful mechanistic interpretation only when the manipulation is sufficiently isolated and the intermediate process is independently measured. Attention-matched comparators, comparable therapist contact, and credible rationales become important design features.
Behavioral Activation and the Depression Dismantling Tradition
The depression literature provides a historically important example of component testing. Jacobson, N. S., Dobson, K. S., Truax, P. A., Addis, M. E., Koerner, K., (1996) randomly assigned 150 outpatients with major depression to behavioral activation alone, behavioral activation with procedures targeting automatic thoughts, or a more complete cognitive therapy package. The design directly tested whether additional cognitive components were required to obtain the clinical benefits attributed to cognitive therapy. Similar performance across the studied conditions challenged simple ingredient-necessity claims while leaving open questions about subgroup effects, different clinical settings, and unmeasured intermediate change.
Dimidjian, S., Hollon, S. D., Dobson, K. S., Schmaling, K. B., Kohlenber (2006) subsequently compared behavioral activation, cognitive therapy, antidepressant medication, and placebo in an acute depression trial. This work expanded the treatment-comparison evidence but did not by itself demonstrate a complete causal chain from scheduled activity through reward learning to symptom relief. Clinical results establish what assignment to protocols can accomplish; the hypothesized path through environmental reinforcement requires proximal measurement of activity, opportunity, perceived reward, avoidance, and mood over time.
Behavioral activation theories specify several related mechanisms. Greater contact with positive reinforcement may strengthen adaptive action; reduced avoidance may permit previously inaccessible experiences; stable activity patterns can alter sleep, social routines, and perceived agency. The same observable increase in activity may be a treatment target, a mediator, and an outcome component depending on the model. Investigators should therefore avoid treating improved functioning as a mediator of depression when their depression instrument already includes the functioning construct. Construct overlap can inflate process–outcome associations and falsely suggest explanation where a common measurement component is responsible.
Cognitive Change, Appraisal, and Depression Recovery
Cognitive therapies propose that modifications to interpretations, predictions, and underlying assumptions can support emotional and behavioral change. Plausible intermediate variables include negative automatic thoughts, dysfunctional attitudes, cognitive flexibility, hopelessness, rumination, and cognitive reappraisal. The fundamental question is whether therapeutic techniques change these processes in ways that precede and contribute causally to outcomes. A fall in negative thinking after depression improves does not by itself show cognitive change initiated recovery.
Lemmens, L. H. J. M., Müller, V. N. L. S., Arntz, A., & Huibers, M. J. H (2016) critically reviewed research on psychological mediators of psychotherapy for depression. The identified literature varied widely in designs, measures, and the temporal ordering of assessments. The authors characterized the state of mechanistic evidence as methodologically unsatisfactory, underscoring the distinction between a promising candidate mediator and an established active causal pathway. Their review is essential because depression affords a large research literature in which competing treatment models frequently improve the same outcomes.
The cognitive and behavioral accounts may describe interacting levels of one change system. Performing a formerly avoided activity provides information that updates beliefs; a revised belief may enable further action; changes in sleep or social contact can reduce distress and therefore alter appraisals. An integrative study should specify whether appraisal change is expected before behavioral engagement, after it, or in a feedback relation. Repeated measurements at defensible intervals, independent assessment of cognitive and behavioral constructs, and controlled manipulation of procedures are needed to separate these explanations.
Exposure Treatment and Inhibitory Learning
Exposure therapies create structured encounters with feared cues, sensations, memories, or situations, usually while reducing the avoidance or safety behavior that maintains threat expectations. Several accounts compete or interact: within-session fear reduction, between-session habituation, expectancy disconfirmation, increased tolerance of distress, and the acquisition and retrieval of inhibitory associations. The mechanisms are experimentally investigable because fear-related predictions, physiological responses, approach behavior, and context changes can be repeatedly measured.
Craske, M. G., Treanor, M., Conway, C. C., Zbozinek, T., & Vervliet, B. (2014) developed an inhibitory-learning framework in which exposure may strengthen nonthreat associations without erasing prior fear learning. Their model emphasizes expectancy violation, contextual variation, retrieval, and attention to safety signals. It yields different predictions from the claim that lower fear at the end of every session is the essential active ingredient. The article proposes clinically relevant optimization strategies supported by learning theory; the individual strategies should not all be treated as independently proven superior clinical procedures.
Mechanistic exposure research is particularly sensitive to the distinction between a process marker and a mediator. Fear activation, subjective distress, extinction retention, threat expectancy, avoidance, and functional participation can change on different trajectories. A participant may continue to experience anxiety while approaching previously avoided situations successfully. Treatment outcome definitions should accommodate meaningful functioning as well as symptom reduction. Laboratory fear-learning tasks afford experimental control but generalization to complex clinical presentations requires validation across ecological conditions and time.
Psychodynamic Interpretation, Insight, and Relational Meaning
Psychodynamic models frequently identify increased awareness of recurring relational patterns, affective conflicts, defenses, and transference configurations as potential routes to change. A transference interpretation invites examination of reactions to the therapist in the context of broader interpersonal patterns. The proposed mediator is sometimes termed insight, though insight itself must be defined operationally rather than assumed from the therapist's preferred interpretation. Awareness, emotional elaboration, behavioral change, and relational flexibility are different constructs.
A randomized dismantling study by Høglend, P., Bøgwald, K.-P., Amlo, S., Marble, A., Ulberg, R., Sjaastad, (2008) compared dynamic psychotherapy with and without transference interpretation, including longitudinal follow-up. This design offers stronger tests of a particular therapeutic operation than uncontrolled descriptions of interpretive success. However, an effect of an interpretation-containing protocol does not identify insight as the mediator unless insight was measured appropriately and other pathways considered. Subsequent work on insight mediation in this line of research illustrates how randomization and intermediate measurement can be integrated, while the generality of such findings across diagnostic groups, therapists, and treatment durations requires further study.
The psychodynamic case emphasizes the complexity of mechanistic identification when the intervention itself is contextual and responsive. The same verbal interpretation may be experienced as clarifying, premature, intrusive, or reparative depending on preceding interactions. Process coding should record timing, context, patient response, and therapist adaptation. Average counts of interpretations across an entire course of therapy may poorly represent clinically decisive moments.
Humanistic and Experiential Processes
Humanistic and experiential approaches contribute hypotheses about empathic understanding, authenticity, positive regard, emotion articulation, and the transformation of self-experience. Rogers, C. R. (1957) provided a historically influential statement of relational conditions. Research on the alliance and on therapeutic responsiveness subsequently offered partial empirical access to these constructs, although conceptual overlap among therapist empathy, perceived support, and alliance makes distinct causal contributions difficult to identify.
A central methodological challenge is preserving the construct while making it measurable. A patient may report feeling understood because the therapist tracked a previously unexpressed emotion, because symptoms have already improved, or because the setting feels familiar. Observer-rated empathic accuracy, patient experience sampling, transcripts, and behavioral response can triangulate the process. Experimental manipulation is more challenging because genuine responsiveness involves contingent judgment rather than a fixed utterance. Component studies that standardize empathic statements may inadvertently remove the very responsiveness a theory proposes as causally important.
Experiential models also distinguish emotional expression from emotional transformation. High emotional intensity during a session is neither a universal criterion for therapeutic progress nor proof of harmful activation. Stronger designs specify whether the proposed process is access to previously avoided affect, reorganization of meaning, improved regulation, or new action following emotional contact. The hypothesized endpoint should be measured separately from the process itself and assessed over relevant follow-up intervals.
Emotion Regulation, Avoidance, and Transdiagnostic Change
Emotion regulation encompasses the processes by which emotional experience and expression are shaped over time. Relevant therapeutic processes include attentional deployment, reappraisal, acceptance, behavioral approach, interpersonal regulation, and reduced reliance on safety behaviors. Different psychological treatments can recruit overlapping regulatory operations despite using distinct techniques. A transdiagnostic framework is therefore scientifically attractive when a process predicts changes in multiple symptom domains and when manipulating that process changes clinically relevant outcomes.
Mechanistic research should distinguish between regulating the immediate intensity of emotion and altering a person's longer-term response repertoire. A reduction in anxiety during exposure may be neither required nor sufficient for subsequent adaptive behavior; conversely, learning to act while distressed can improve functioning before subjective distress declines. Craske, M. G., Treanor, M., Conway, C. C., Zbozinek, T., & Vervliet, B. (2014) illustrate how competing learning theories generate different predictions about emotional trajectories. Treatment goals such as self-harm reduction, depression relief, and reduced avoidance may require distinct proximal measures and timescales.
The term transdiagnostic should be reserved for evidence extending across more than one diagnostic or symptom configuration. Shared terminology alone does not prove shared causal architecture. Researchers need equivalent process measures, condition-specific tests, baseline severity stratification, and models of interactions among emotional, cognitive, social, and behavioral pathways. A process can be useful in several disorders while explaining different proportions of benefit or operating through different proximal events.
Motivation, Engagement, Dropout, and Treatment Dose
Participation is part of the causal pathway of many treatments. A patient must often attend enough sessions, understand a rationale, practice skills, or approach avoided situations for a proposed intervention to exert its hypothesized effect. Engagement thus can be an outcome in its own right, an intermediate determinant of later intervention dose, and a marker of other processes such as trust and hope. These roles should be distinguished in statistical models.
Treatment dose is especially problematic in observational mediation analyses because it is determined partly by prior symptoms and patient preferences. Early improvement can prompt planned termination; deterioration can increase visit frequency; economic barriers can end therapy regardless of effectiveness. Comparing high-dose and low-dose patients without modeling these pathways confounds the effect of treatment with the reasons treatment continued. A time-varying design can record session attendance, homework completion, allocation of between-session opportunities, and the feedback from improvement to future treatment choices.
Howard, K. I., Moras, K., Brill, P. L., Martinovich, Z., & Lutz, W. (1996) helped establish patient-focused monitoring as a distinct enterprise. Later progress-feedback work, including Lambert, M. J., Whipple, J. L., & Kleinstäuber, M. (2018), evaluates whether supplying clinicians with outcome signals changes the trajectories of patients at risk of poor response. Such interventions offer opportunities to manipulate a feedback process, but the active mechanism might include improved case detection, therapist behavioral change, collaborative discussion, or increased patient engagement rather than the act of measurement alone.
Process-Based Therapy and Personalized Functional Analysis
Process-based therapy proposes selecting and sequencing interventions according to modifiable biopsychosocial processes rather than treating named therapy packages as the sole objects of optimization. Hofmann, S. G., & Hayes, S. C. (2019) outline this program and argue for more direct empirical tests of mediators, moderators, and person-specific change processes. The model is a consequential research proposal; claims that it consistently outperforms established protocols require independent comparative trials and prospective tests of treatment-selection rules.
Personalized functional analysis begins with a hypothesis about an individual's context-sensitive patterns: what precedes an episode of avoidance, how an appraisal changes behavior, what reinforcement follows, and which intervention could alter the sequence. Repeated observations may support individualized predictions. But fitting a unique time-series model to one patient's small data set can overfit noise, especially when there are multiple changing variables and autocorrelation. Reliable clinical decision systems require calibration, external validation, uncertainty estimates, and safeguards when model predictions are wrong.
The relevant scientific unit may be a configuration of co-occurring processes rather than one isolated factor. Improving sleep can increase available attention; reduced avoidance can create reinforcing social experiences; heightened collaboration can permit more demanding exposure. This network language is useful only when the hypothesized interactions are operationalized. A network drawing derived from correlations is an exploratory description until intervention and longitudinal evidence identify causal edges.
Digital, Remote, and Artificially Mediated Psychotherapy Processes
Remote and digital delivery introduce new ways to observe and modify psychotherapy processes. Online forms can gather repeated outcomes, applications can record engagement, and asynchronous communication alters the temporal organization of therapist responsiveness. These changes create opportunities for intensive longitudinal measurement and adaptive prompting. They also produce distinctive selection bias: people who stop using an application may be those who deteriorate, improve quickly, face accessibility problems, or reject the system's assumptions.
The alliance meta-analysis by Flückiger, C., Del Re, A. C., Wampold, B. E., & Horvath, A. O. (2018) included Internet-based psychotherapy samples, supporting investigation of alliance beyond conventional in-person settings. Yet patient–clinician therapeutic alliance, a participant's perceived bond with software, and task adherence mediated by an automated system are distinct constructs. Scores designed for human clinician relationships cannot simply be transferred to artificial systems without measurement invariance testing and a clear account of what the respondent is rating.
Within the Artificial Era, digital process research can examine human–artificial collaboration, cognitive offloading, external reminders, perceived social presence, trust calibration, and changes in human access to care. Such proposals are empirical hypotheses rather than established clinical mechanisms. Systems that process personal disclosures require careful privacy design, informed consent, oversight, and monitoring of harmful advice or deterioration. The scientific task is to determine which observable system actions change which human outcomes under controlled conditions, with direct tests rather than anthropomorphic interpretation.
Recent Evidence Synthesis
Luo, X., & Levendosky, A. A. (2025) reviewed 86 quantitative studies of psychotherapy microprocesses. They distinguished affective, behavioral, cognitive, relational, linguistic, and movement domains, and described problems of inconsistent operational definitions, cultural representation, and theory-led measurement. The review maps observational methods, rather than establishing a single necessary mechanism.
Hofmann, S. G., Pahlen, L., Westhoff, M., Hossner, K., Quintarelli, H., (2026) surveyed 778 studies reporting 684 proposed therapeutic-change processes and grouped these into 32 clusters. Their results describe research attention and conceptual overlap. Study frequency does not establish causality.
Gómez Penedo, J. M., & Flückiger, C. (2026) synthesized process–outcome meta-analyses. Associations were positive across several process families; observational associations leave open reverse causation, confounding, and measurement differences. Research that experimentally alters precise process variables is still needed.
Measurement and Causal Identification
A mechanism claim requires a temporally ordered, validly measured process and a defensible intervention contrast. Repeated session observations are valuable but can remain confounded by reciprocal symptom improvement. See Kazdin (2007): https://doi.org/10.1146/annurev.clinpsy.3.022806.091432.
Causal Graphs, Temporal Order, and Mediation Estimands
A treatment mechanism requires an explanation of which event influences which later event. The investigator must specify treatment assignment, mediator measurement, and clinical outcome, along with potential common causes of mediator and outcome. Random assignment to treatment does not randomize post-treatment expectations, therapeutic alliance, or behavioral change. Imai, K., Keele, L., & Tingley, D. (2010) offer counterfactual mediation and sensitivity analysis; Kraemer, H. C., Wilson, G. T., Fairburn, C. G., & Agras, W. S. (2002) explain the implications of moderators and mediators in randomized trials.
The estimand matters: a natural indirect effect, an interventional indirect effect, and a controlled direct effect represent different causal questions. Some require assumptions particularly difficult to satisfy in psychotherapy, where earlier improvement changes later therapist decisions. An observed indirect coefficient is best reported as a statistical pathway when identifying assumptions cannot be defended. Measurement at the beginning and end of treatment alone cannot establish temporal precedence.
Within-Person Modeling and Longitudinal Feedback
Repeated process scores are nested within patients, therapists, and care settings. Curran, P. J., & Bauer, D. J. (2011) distinguish stable between-person associations from dynamic within-person relationships. Hamaker, E. L., Kuiper, R. M., & Grasman, R. P. P. P. (2015) show that traditional cross-lagged models can conflate those relationships. A patient whose average alliance is strong may differ from one with weaker average alliance; neither comparison identifies what happens after a temporary change in the alliance of the same person.
A defensible design measures the proposed process and outcome at intervals consistent with its theory. It models reverse pathways and checks whether the clinical course itself modifies treatment delivery. Cross-lagged associations provide stronger temporal information than simultaneous measures, but causal inference still requires attention to feedback, unobserved confounding, and potential interventions.
Therapist Responsiveness, Competence, and Context
Therapist adherence describes compliance with a planned method; competence describes its appropriate and skillful application. The two are related but distinct. The same verbal intervention may support reflection or disrupt collaboration depending on preceding interactions and patient needs. An aggregated count of interventions does not necessarily register the clinical function of those interventions. Webb, C. A., DeRubeis, R. J., & Barber, J. P. (2010) synthesized a literature in which global adherence and competence scores had small overall correlations with outcomes, exposing limitations of measurement and generalization.
Responsiveness can be modeled as a conditional intervention policy: given a specified patient state, the therapist selects an operation predicted to change the next state. A study can code antecedent patient behavior, therapist action, immediate response, and later outcome, while also measuring treatment fidelity. Appropriate timing, contextual adaptation, and case formulation become variables for empirical study. Therapists are nested within care organizations and may receive unequal training and supervision, so therapist-level variation and institutional constraints belong in the statistical model.
Patient Heterogeneity and Moderated Processes
A process may have different implications across baseline severity, symptom configurations, developmental stage, cultural context, treatment preference, and material circumstances. Høglend, P., Bøgwald, K.-P., Amlo, S., Marble, A., Ulberg, R., Sjaastad, (2008) showed how a randomized comparison of psychodynamic treatment with or without transference interpretations can yield limited overall differences while revealing differential outcomes among patients with different relational characteristics. Their findings illustrate why average treatment effects do not settle a mechanism's relevance to all patients.
Moderated mediation asks whether an indirect pathway varies according to a prespecified characteristic. This requires adequate statistical power, precise measurements, transparent interaction tests, and replication. Small subgroup analyses can generate unstable explanations. Transportability further requires demonstrating that a relation learned in a specialist clinic operates across service settings, languages, and health systems. The credible target is a conditional causal statement whose assumptions are documented.
Dropout, Missingness, and Adverse Outcomes
Session-level research is vulnerable to selective observation. People may discontinue treatment after improvement, frustration, adverse experience, financial strain, or insufficient benefit. Models that analyze only continuing patients can overestimate favorable process–outcome associations if those who remain have both better alliances and more complete outcome records. Investigators should report reasons for attrition, sensitivity to missing-data assumptions, and how treatment decisions respond to earlier progress.
Deterioration and harmful clinical experiences also merit process analysis. A program can have a favorable average outcome while particular participants experience worsening. Research should measure deterioration and functional consequences, record adverse experiences, and consider the conditions in which an otherwise useful therapeutic operation is poorly timed or unsuitable. A causal account of psychotherapy is incomplete if it describes beneficial pathways but systematically excludes unfavorable trajectories.
An Integrative Scientific Framework for Mechanism Evaluation
The central findings can be organized as a sequence of research propositions. First, a treatment protocol specifies opportunities for change, but the presence of an intervention ingredient is separate from the participant's receipt of that operation. Second, receiving an operation may modify a proximal process, which should be measured independently of the clinical endpoint. Third, a measured process can predict an endpoint without yet being its cause. Fourth, the strongest explanation combines credible manipulation, temporal ordering, construct validity, and independent replication. This sequence makes the difference between efficacy, prediction, mediation, and mechanism explicit.
The research literature illustrates each level. Behavioral activation and transference-interpretation trials experimentally contrast intervention components. Alliance and rupture meta-analyses establish reproducible process–outcome associations. Repeated-session studies examine reciprocal temporal relations. Causal mediation methods clarify assumptions that the observational literature often cannot verify. The 2025–2026 syntheses show a research field rich in identified processes but still in need of systematic consolidation and causal tests. No single evidence type can substitute for the others.
A mature explanatory program should link these levels: define the process, specify an intervention able to alter it, measure immediate and delayed consequences, model alternative pathways, and replicate the findings in varied patients and settings. A successful claim also states its limits: the population, therapeutic context, outcome, time horizon, and comparisons to which the mechanism applies. Such a program supports precision without assuming that every patient must follow an identical pathway.
Evidence Assessment Across Psychotherapy Traditions
The strongest empirical basis within the reviewed literature is the demonstrated effectiveness of particular treatment packages and components under specified trial conditions, alongside replicated associations between therapeutic process variables and clinical outcomes. The evidence for alliance as a predictor is strong across many samples, while the evidence that intentionally altering alliance causes an equivalent amount of symptom improvement is more limited. Behavioral activation and exposure-based change models benefit from experimental treatment traditions, but intermediate-variable tests still frequently fail to isolate a complete pathway. Psychodynamic interpretation experiments illustrate the value of controlled component tests and the importance of heterogeneity. Humanistic and experiential frameworks contribute measurable relational and emotional hypotheses whose causal validation requires further work.
A useful evidence statement should report not only whether an association is statistically detectable but what design supports it. A meta-analysis of correlational studies can produce a precise estimate of an association while preserving the same causal uncertainty as the underlying studies. An individual randomized dismantling trial can identify a particular component contrast but may have limited transportability. A session-level study can identify a plausible temporal lag without eliminating feedback or latent confounding. Comparisons across traditions are informative when the respective evidence levels and endpoints are comparable.
The recurring explanatory problem is that several processes can change together, interact, or produce reciprocal effects. Emotional regulation, behavioral approach, belief revision, collaborative alliance, and treatment expectancy can be co-produced by the same intervention. Scientific synthesis therefore benefits from models of multiple interacting processes rather than a search for one universal active ingredient. The key is to maintain operational precision even in multicausal explanations.
Open Research Problems and a Program for Future Studies
Future studies should prioritize measures and designs that discriminate among competing explanations. Longitudinal treatment trials can measure alliance, expectations, avoidant behavior, cognitive appraisal, and symptoms at intervals matched to each hypothesized pathway. Randomized assignment to components or clinically appropriate adaptation strategies can test whether particular operations move the proposed mediator. Independent observers can code the therapeutic interaction and reduce common-method bias. Researchers should preregister primary mediator hypotheses and planned lag intervals, while explicitly labeling exploratory discovery of new relationships.
Several unresolved questions are especially consequential. It remains uncertain which alliance fluctuations are causal drivers of improvement and which are indicators of concurrent change. It remains difficult to distinguish an active cognitive process from a consequence of symptom relief. Exposure-learning predictions must be tested across different clinical environments and long-term outcomes. Therapist responsiveness requires measurement frameworks that capture function and timing without forcing uniform behavior. Heterogeneous treatment pathways must be validated rather than inferred from small subgroups. The reproducibility of fine-grained process models across cultures, ages, clinical services, and remote settings is another major task.
New computational approaches can reveal patterns in language and interaction, but prediction accuracy should not be confused with causal explanation. Models trained on recordings or session notes may capture superficial correlates, therapist identity, or institutional patterns that fail under distribution shift. Human oversight, privacy protection, explicit construct definitions, and prospective evaluation remain necessary. The history of psychotherapy process research shows that progress comes from converting plausible therapeutic narratives into questions that can be answered by observation and intervention.
Conclusion
Psychotherapy mechanism research is the scientific investigation of how treatment operations become psychological change. Its distinctive object is the causal organization of treatment, intermediate processes, reciprocal relationships, patient behavior, therapist decisions, and clinical outcomes across time. The literature has demonstrated reliable treatment effects and many reproducible associations between processes and outcomes. It has also exposed the difficulty of proving that any measured intermediate variable is the active causal pathway through which a particular therapy produces its benefits.
Historical theories of relational change, cognitive revision, learning, emotional transformation, and behavioral activation remain productive sources of hypotheses. Experimental dismantling studies, repeated-session modeling, microprocess observation, and counterfactual mediation now provide a richer methodological repertoire for testing those hypotheses. The 2025–2026 synthesis literature makes the scale of the conceptual and empirical task especially clear: reported processes are numerous, operationalizations are uneven, and much of the evidence continues to be correlational.
An adequate account of psychological change must specify interventions, intermediate states, timing, alternative explanations, endpoints, and conditions of generalization. It must also recognize feedback, treatment adaptation, patient heterogeneity, and adverse outcomes. The principal scientific conclusion is an integrated methodological demand: causal claims about psychotherapy mechanisms should arise from converging experiments, carefully ordered observations, valid measurement, and replication. That approach preserves the explanatory ambitions of psychotherapy theories while making their predictions increasingly accessible to rigorous examination.
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
- Barber, J. P., Connolly, M. B., Crits-Christoph, P., Gladis, L., & Siqueland, L. (2000). Alliance predicts patients’ outcome beyond in-treatment change in symptoms. Journal of Consulting and Clinical Psychology, 68(6), 1027–1032. https://doi.org/10.1037/0022-006X.68.6.1027
- Baron, R. M., & Kenny, D. A. (1986). The moderator–mediator variable distinction in social psychological research: Conceptual, strategic, and statistical considerations. Journal of Personality and Social Psychology, 51(6), 1173–1182. https://doi.org/10.1037/0022-3514.51.6.1173
- 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
- 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., 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
- Craske, M. G., Treanor, M., Conway, C. C., Zbozinek, T., & Vervliet, B. (2014). Maximizing exposure therapy: An inhibitory learning approach. Behaviour Research and Therapy, 58, 10–23. https://doi.org/10.1016/j.brat.2014.04.006
- 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
- Curran, P. J., & Bauer, D. J. (2011). The disaggregation of within-person and between-person effects in longitudinal models of change. Annual Review of Psychology, 62, 583–619. https://doi.org/10.1146/annurev.psych.093008.100356
- 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
- Eubanks, C. F., Muran, J. C., & Safran, J. D. (2018). Alliance rupture repair: A meta-analysis. Psychotherapy, 55(4), 508–519. https://doi.org/10.1037/pst0000185
- Falkenström, F., Ekeblad, A., & Holmqvist, R. (2016). Improvement of the working alliance in one treatment session predicts improvement of depressive symptoms by the next session. Journal of Consulting and Clinical Psychology, 84(8), 738–751. https://doi.org/10.1037/ccp0000119
- Falkenström, F., Granström, F., & Holmqvist, R. (2013). Therapeutic alliance predicts symptomatic improvement session by session. Journal of Counseling Psychology, 60(3), 317–328. https://doi.org/10.1037/a0032258
- 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
- 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
- Hamaker, E. L., Kuiper, R. M., & Grasman, R. P. P. P. (2015). A critique of the cross-lagged panel model. Psychological Methods, 20(1), 102–116. https://doi.org/10.1037/a0038889
- Hofmann, S. G., & Hayes, S. C. (2019). The future of intervention science: Process-based therapy. Clinical Psychological Science, 7(1), 37–50. https://doi.org/10.1177/2167702618772296
- 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, 102782. https://doi.org/10.1016/j.cpr.2026.102782
- Høglend, P., Bøgwald, K.-P., Amlo, S., Marble, A., Ulberg, R., Sjaastad, M. C., Sørbye, Ø., Heyerdahl, O., & Johansson, P. (2008). Transference interpretations in dynamic psychotherapy: Do they really yield sustained effects? American Journal of Psychiatry, 165(6), 763–771. https://doi.org/10.1176/appi.ajp.2008.07061028
- Howard, K. I., Moras, K., Brill, P. L., Martinovich, Z., & Lutz, W. (1996). Evaluation of psychotherapy: Efficacy, effectiveness, and patient progress. American Psychologist, 51(10), 1059–1064. https://doi.org/10.1037/0003-066X.51.10.1059
- Imai, K., Keele, L., & Tingley, D. (2010). A general approach to causal mediation analysis. Psychological Methods, 15(4), 309–334. https://doi.org/10.1037/a0020761
- 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
- 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
- Kraemer, H. C., Wilson, G. T., Fairburn, C. G., & Agras, W. S. (2002). Mediators and moderators of treatment effects in randomized clinical trials. Archives of General Psychiatry, 59(10), 877–883. https://doi.org/10.1001/archpsyc.59.10.877
- Lambert, M. J., Whipple, J. L., & Kleinstäuber, M. (2018). Collecting and delivering progress feedback: A meta-analysis of routine outcome monitoring. Psychotherapy, 55(4), 520–537. https://doi.org/10.1037/pst0000167
- Lemmens, L. H. J. M., Müller, V. N. L. S., Arntz, A., & Huibers, M. J. H. (2016). Mechanisms of change in psychotherapy for depression: An empirical update and evaluation of research aimed at identifying psychological mediators. Clinical Psychology Review, 50, 95–107. https://doi.org/10.1016/j.cpr.2016.09.004
- Luo, X., & Levendosky, A. A. (2025). A systematic review and integrative framework of psychotherapy microprocess: Linking the science of psychological interventions with the art of moment-to-moment practice. Clinical Psychology Review, 121, 102641. https://doi.org/10.1016/j.cpr.2025.102641
- MacKinnon, D. P., Fairchild, A. J., & Fritz, M. S. (2007). Mediation analysis. Annual Review of Psychology, 58, 593–614. https://doi.org/10.1146/annurev.psych.58.110405.085542
- 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
- Santoft, F., Salomonsson, S., Hesser, H., Lindsäter, E., Ljótsson, B., Lekander, M., Kecklund, G., Öst, L.-G., & Hedman-Lagerlöf, E. (2019). Processes in cognitive behavior therapy for social anxiety disorder: Predicting subsequent symptom change. Journal of Anxiety Disorders, 67, 102118. https://doi.org/10.1016/j.janxdis.2019.102118
- Smits, J. A. J., Rosenfield, D., McDonald, R., & Telch, M. J. (2006). Cognitive mechanisms of social anxiety reduction: An examination of specificity and temporality. Journal of Consulting and Clinical Psychology, 74(6), 1203–1212. https://doi.org/10.1037/0022-006X.74.6.1203
- Wampold, B. E., & Imel, Z. E. (2015). The great psychotherapy debate: The evidence for what makes psychotherapy work (2nd ed.). Routledge. https://www.ncbi.nlm.nih.gov/nlmcatalog/101656107
- Webb, C. A., DeRubeis, R. J., & Barber, J. P. (2010). Therapist adherence/competence and treatment outcome: A meta-analytic review. Journal of Consulting and Clinical Psychology, 78(2), 200–211. https://doi.org/10.1037/a0018912
- Zilcha-Mano, S. (2017). Is the alliance really therapeutic? Revisiting this question in light of recent methodological advances. American Psychologist, 72(4), 311–325. https://doi.org/10.1037/a0040435
Social Anxiety: Temporal Tests of Avoidance and Cognitive Bias
Social anxiety research combines behavioral exposure and cognitive models in an unusually productive mechanism-testing context. Smits, J. A. J., Rosenfield, D., McDonald, R., & Telch, M. J. (2006) examined specificity and temporality in proposed cognitive mechanisms of social anxiety reduction, moving beyond undifferentiated pre/post correlations. Later work by Santoft, F., Salomonsson, S., Hesser, H., Lindsäter, E., Ljótsson, B., L (2019) analyzed weekly within-person changes during guided self-help cognitive behavioral therapy. Its reported temporal modeling highlighted changes in avoidance as predictors of subsequent symptom reduction.
These studies illustrate how symptom and process measures can be collected frequently enough to test rival sequences. If avoidance reduction consistently precedes subsequent symptom decline while earlier symptom decline does not account for the pattern, a behavioral mechanism becomes more plausible. Yet avoidance may share item content with social anxiety outcomes, self-report schedules can miss important exposures, and interventions may alter cognitions and behavior simultaneously. Confirmatory trials that randomize ways of reducing avoidance could further distinguish causal pathways.
Processes are also situated in social environments. A social exposure can be experienced differently according to discrimination risk, power relations, language, or the actual behavior of other people. A model that treats every feared interaction as objectively safe overlooks the reality that some social threats are real. Ecologically responsible exposure research therefore distinguishes distorted threat expectation, adaptive caution, and structural barriers to participation.