Therapeutic Alliance

Abstract

The therapeutic alliance, the collaborative bond between client and therapist that the Medical Subject Headings classify under psychotherapy, is the partnership through which treatment does its work. Edward Bordin recast it as a pantheoretical construct of three parts, an affective bond, agreement on goals, and agreement on tasks, making it measurable through instruments like the Working Alliance Inventory. Four decades of meta-analysis converge on a modest but reliable alliance-outcome correlation near 0.28, and this article traces the construct from its psychoanalytic origin through measurement and the theory of ruptures and their repair. What remains contested is causal: whether a strong alliance produces improvement or an improving patient produces it, a puzzle recent trait-versus-state methods aim to resolve. Three demonstrations let readers compose Bordin's components, watch a 0.28 correlation become outcome differences, and trace a rupture's repair.

Keywords: therapeutic alliance, working alliance, psychotherapy outcome, common factors, rupture repair

The therapeutic alliance is the collaborative, affectively charged relationship between a patient and a therapist through which the work of psychotherapy is carried out, and it is among the most studied constructs in clinical psychology (Horvath, Del Re, Fluckiger, & Symonds, 2011). Its appeal is both practical and theoretical: practically, because clinicians of every orientation recognize that little happens in treatment without a functioning partnership; theoretically, because the alliance is the leading candidate for a common factor, an ingredient shared by therapies that otherwise disagree about technique, that could explain why so many different treatments produce broadly similar benefits (Fluckiger et al., 2018). That rough equivalence is itself an empirical result: a meta-analysis of studies directly comparing bona fide therapies found their outcomes to differ little, the pattern nicknamed the dodo bird verdict, which is precisely what a common factor such as the alliance is invoked to explain (Wampold et al., 1997). For cognitive and clinical psychology the construct matters because it makes the relationship itself an object of measurement and prediction rather than an unexamined background to technique (Fluckiger, Del Re, Wampold, & Horvath, 2018).

What the Therapeutic Alliance Is

The idea has a long lineage in psychoanalysis, where the analyst and the patient's observing ego were said to form a working relationship distinct from the transference. What transformed a school-specific notion into a general construct was Edward Bordin's proposal that the working alliance is pantheoretical, present in every change-oriented relationship and composed of three interdependent parts: a bond of trust and attachment between the participants, agreement on the goals of the enterprise, and agreement on the tasks that will pursue those goals (Bordin, 1979). The reformulation was deliberately generic. By defining the alliance in terms that no single therapy owned, Bordin made it a candidate for the ingredient common to all effective treatments, and made its strength something that could be assessed in behavior therapy and psychoanalysis alike.

The alliance is often traced further back to Carl Rogers, whose account of the necessary and sufficient conditions of therapeutic change, empathy, congruence, and unconditional positive regard, located the engine of improvement in the relationship rather than in technique (Rogers, 1957). Two clarifications separate the modern construct from its caricatures. First, the alliance is not simply the patient liking the therapist; the bond is only one of three components, and warmth without agreement on goals and tasks is not a working alliance (Bordin, 1979). Second, the alliance is not a fixed trait of a therapist or a patient but a property of their particular pairing that develops, strains, and recovers over the course of treatment (Fluckiger et al., 2018).

The Three Components

Bordin's enduring contribution was to decompose a single global impression into three separable elements whose interplay defines the alliance (Bordin, 1979). The bond is the affective tie, the mutual trust, respect, and liking that lets a patient tolerate the demands of treatment. Goal consensus is agreement between patient and therapist about what the treatment is trying to achieve, the outcomes both regard as worth pursuing. Task agreement is agreement about the in-session and between-session activities, the homework, the free association, the exposure exercises, that both accept as relevant means to those goals. The three are interdependent: a strong bond makes it easier to negotiate goals and tasks, while shared goals and credible tasks deepen the bond.

The clinical value of the decomposition is diagnostic. When treatment stalls, the three-part model directs attention to where the alliance has weakened: a warm relationship that goes nowhere may reflect a failure of goal consensus, while a technically well-specified plan that the patient resists may reflect a thin bond or disagreement about tasks. The demonstration below lets the reader vary the three components independently and see how a composite alliance rating responds, illustrating that no single element carries the alliance on its own.

Measuring the Alliance

A construct becomes a research program only when it can be measured, and the instrument that did most to make the alliance measurable is the Working Alliance Inventory, built by Adam Horvath and Leslie Greenberg directly on Bordin's three components (Horvath & Greenberg, 1989). The inventory yields subscales for bond, goals, and tasks, and comes in client, therapist, and observer versions, so that the same relationship can be rated from several vantage points. Its development and validation gave the field a common yardstick, and the accumulation of studies using it made the quantitative synthesis of alliance research possible. A broad empirical review catalogued the several alliance measures that followed and the conceptual choices they embody, noting that different instruments weight Bordin's components differently and do not always converge (Elvins & Green, 2008).

Measurement introduced its own questions. Ratings from client, therapist, and observer agree only moderately, which raises the issue of whose alliance best predicts outcome, and the timing of measurement matters because the alliance changes over sessions (Elvins & Green, 2008). A historical and methodological review by Roberto Ardito and Daniela Rabellino traced how these measurement decisions shaped the field's conclusions, and emphasized that the alliance-outcome relationship is only as clear as the instruments used to establish it (Ardito & Rabellino, 2011). Client-rated early alliance, it turns out, is among the more consistent predictors, a fact that the later meta-analyses exploited.

The Alliance-Outcome Relationship

The central empirical finding is a correlation, modest in size and remarkable in its reliability, between the strength of the alliance and the outcome of therapy. The first meta-analytic estimate, from Adam Horvath and Dianne Symonds, aggregated the early studies to a correlation of about 0.26 (Horvath & Symonds, 1991). A decade later Daniel Martin, John Garske, and Katherine Davis pooled a larger and more varied literature and found a comparable value near 0.22, establishing that the relationship was not an artifact of a few early studies (Martin, Garske, & Davis, 2000). The successive updates commissioned by the field's task force pushed the estimate to roughly 0.275 across some two hundred studies (Horvath et al., 2011), and the most recent synthesis, drawing on more than three hundred studies and thirty thousand patients, settled on approximately 0.28 (Fluckiger et al., 2018). Table 1 records this convergence.

Table 1. Four decades of meta-analytic estimates of the alliance-outcome correlation, converging on a value near 0.28 despite growing and diversifying samples.

A correlation of 0.28 sounds small until it is translated into consequences. Under the binomial effect size display, a correlation of that magnitude corresponds to raising the share of patients who improve from 36 percent to 64 percent, a difference no clinician would dismiss. Figure 1 renders that translation.

An important refinement concerns whose contribution the correlation reflects. Because the alliance is a property of a pairing, a raw alliance-outcome correlation confounds two things: whether therapists who generally form stronger alliances get better results, and whether a given therapist does better with the patients with whom the alliance happens to be stronger. Scott Baldwin, Bruce Wampold, and Zac Imel disentangled these by partitioning the alliance into therapist and patient components, and found that the therapist component predicted outcome while the within-therapist patient component did not, implying that the alliance reflects a therapist skill rather than merely a lucky patient (Baldwin, Wampold, & Imel, 2007). A later meta-analysis of these therapist effects across many datasets confirmed the pattern (Del Re, Fluckiger, Horvath, Symonds, & Wampold, 2012). The demonstration below lets the reader vary an alliance-outcome correlation and watch the implied scatter and improvement rates change.

Ruptures and Their Repair

The alliance is not a quantity that only accumulates; it is strained and sometimes broken, and how those breaches are handled is itself consequential. Jeremy Safran and Christopher Muran developed the concept of the alliance rupture, a deterioration in the collaborative relationship, a tension or breakdown in the bond or in agreement about goals and tasks, and distinguished two forms: withdrawal ruptures, in which the patient retreats into compliance or vagueness, and confrontation ruptures, in which the patient expresses anger or dissatisfaction directly (Safran, Muran, & Eubanks-Carter, 2011). Ruptures are ordinary rather than exceptional; the clinically important variable is not whether they occur but whether they are recognized and repaired.

The theory reframes conflict as opportunity. A rupture successfully repaired can deepen the alliance beyond its pre-rupture level, because the patient learns that disagreement and negative feeling can be voiced and survived, a corrective experience in its own right (Safran et al., 2011). A meta-analysis by Catherine Eubanks, Christopher Muran, and Jeremy Safran examined the evidence and found that the presence of rupture-repair episodes was associated with better treatment outcomes, and that training therapists to attend to ruptures shows promise (Eubanks, Muran, & Safran, 2018). The demonstration below traces two session-by-session alliance trajectories, one in which a rupture is left unrepaired and one in which it is repaired, and contrasts their endpoints.

Worked Example

Consider how an alliance-outcome correlation of 0.28 translates into a concrete difference in patient improvement, following the binomial effect size display used to make small correlations interpretable (Fluckiger et al., 2018). The display maps a correlation r onto two success rates by splitting patients at the median of the alliance and at the median of outcome. The improvement rate for the stronger-alliance half is 0.50 plus half the correlation, and for the weaker-alliance half it is 0.50 minus half the correlation. With r equal to 0.28, half of 0.28 is 0.14, so the stronger-alliance group improves at a rate of 0.50 + 0.14 = 0.64 and the weaker-alliance group at 0.50 − 0.14 = 0.36.

The gap is therefore 64 − 36 = 28 percentage points, exactly the correlation expressed on a 0-to-1 scale. Put differently, moving a patient from the weaker to the stronger half of the alliance distribution is associated with nearly doubling the odds of a good outcome: 0.64 corresponds to odds of 0.64 / 0.36 ≈ 1.78 to 1, against 0.36 / 0.64 ≈ 0.56 to 1, an odds ratio of about 3.2. The squared correlation, 0.28² ≈ 0.078, shows that the alliance accounts for roughly 8 percent of the variance in outcome, which looks negligible and is precisely why the binomial display is the more honest translation: the same relationship that explains under a tenth of the variance also separates a 36 percent from a 64 percent success rate. The lesson the worked example enforces is that a small correlation and a clinically large effect are two descriptions of one number (Martin et al., 2000).

Discussion

The alliance-outcome correlation is one of the most replicated results in psychotherapy research, and its very robustness has fueled the largest debate about the construct: what the correlation means. The straightforward reading is causal, that a strong alliance helps produce improvement, and it underwrites the common-factors view that the relationship, more than the specific technique, drives the benefits of therapy (Fluckiger et al., 2018). A skeptical reading warns that correlation is not cause: patients who are already improving, or who possess the interpersonal capacities that make improvement likely, may simply form stronger alliances, so that the alliance is partly a symptom of progress rather than its source (Zilcha-Mano, 2017). The finding that the therapist component of the alliance predicts outcome while the within-therapist patient component does not weighs against the purely epiphenomenal reading, because it locates predictive power in something the therapist brings (Baldwin et al., 2007; Del Re et al., 2012).

The field's synthesis of this evidence is codified in the periodic reports on psychotherapy relationships that work, which conclude that the alliance is a demonstrably effective element of treatment while cautioning that it operates alongside, not instead of, specific methods (Norcross & Lambert, 2018). The through-line of the modern literature is a shift from establishing that the alliance predicts outcome, now beyond serious doubt, to understanding how and for whom it does so, and how clinicians can build and repair it deliberately rather than hope it arises (Eubanks et al., 2018).

Current Directions

The liveliest current work attacks the causal question head-on with methods designed to separate the alliance's stable from its changing parts. Sigal Zilcha-Mano's influential reframing distinguishes a trait-like component of the alliance, the patient's general capacity to form relationships, which may reflect prognosis rather than active ingredient, from a state-like component, the within-patient fluctuation in alliance around that baseline, which is the part a therapist can actually move and the part most plausibly causal (Zilcha-Mano, 2017). Multilevel and cross-lagged designs now estimate whether session-to-session increases in alliance precede symptom improvement or follow it, turning a decades-old interpretive stalemate into a set of testable within-person hypotheses. The rupture-repair literature is moving in the same direction, from documenting that repairs correlate with outcome toward training studies that ask whether teaching therapists to detect and mend ruptures causally improves results (Eubanks et al., 2018). Open questions remain about how the alliance interacts with specific techniques rather than substituting for them, how it operates in the growing world of internet-delivered and app-based treatment where the traditional relational cues are attenuated, and how measurement can capture its moment-to-moment dynamics rather than a single session-level snapshot (Fluckiger et al., 2018; Norcross & Lambert, 2018).

Common Misconceptions

The therapeutic alliance is just the patient liking the therapist.

The bond is only one of Bordin's three components. A warm relationship without agreement on the goals of treatment and the tasks that pursue them is not a working alliance, and rapport alone does not predict outcome as well as the full construct (Bordin, 1979).

A correlation of 0.28 means the alliance barely matters.

The same correlation that explains under 8 percent of outcome variance also corresponds, on the binomial effect size display, to raising the improvement rate from 36 to 64 percent, a difference of real clinical size (Fluckiger et al., 2018).

A strong alliance simply causes good outcomes.

The direction of causation is genuinely contested. Improving patients may form stronger alliances, so recent work separates a trait-like component that may reflect prognosis from a state-like component that is more plausibly an active ingredient (Zilcha-Mano, 2017).

Ruptures in the alliance are failures to be avoided.

Ruptures are common and, when recognized and repaired, are associated with better outcomes. A repaired rupture can strengthen the alliance beyond its earlier level, making conflict an opportunity rather than only a hazard (Safran et al., 2011; Eubanks et al., 2018).

Glossary

Alliance-outcome correlation.

The statistical association between the strength of the therapeutic alliance and the benefit a patient derives from treatment, estimated by meta-analysis at about 0.28.

Binomial effect size display.

A device that translates a correlation into the difference in success rates it implies, making a small correlation such as 0.28 interpretable as a 36-versus-64 percent split.

Bond.

In Bordin's model, the affective tie of trust, respect, and attachment between patient and therapist; one of the three components of the alliance.

Common factors.

Ingredients shared across differing therapies, such as the alliance, that may account for their broadly comparable effectiveness independent of specific technique.

Congruence.

In Rogers's account, the therapist's genuineness or authenticity in the relationship; one of the conditions he held necessary for therapeutic change.

Goal consensus.

Agreement between patient and therapist about the outcomes the treatment is trying to achieve; one of Bordin's three components.

Rupture repair.

The process by which a strained alliance is recognized and mended, often strengthening the relationship beyond its pre-rupture level and associated with better outcomes.

Rupture.

A deterioration or strain in the alliance, taking either a withdrawal form, in which the patient retreats, or a confrontation form, in which the patient expresses dissatisfaction directly.

State-like alliance.

The within-patient fluctuation in the alliance around a personal baseline; the component a therapist can move and the one most plausibly causal for outcome.

Task agreement.

Agreement between patient and therapist about the in-session and between-session activities that will pursue the goals of treatment; one of Bordin's three components.

Therapeutic alliance.

The collaborative, affectively charged partnership between patient and therapist through which the work of psychotherapy is carried out.

Therapist effect.

The portion of the alliance-outcome relationship attributable to differences between therapists rather than between their patients; evidence that the alliance reflects a therapist skill.

Trait-like alliance.

The stable component of the alliance reflecting a patient's general capacity to form relationships, which may index prognosis rather than an active ingredient of change.

Unconditional positive regard.

In Rogers's account, the therapist's nonjudgmental acceptance of the patient; one of the relational conditions he held necessary for change.

Working Alliance Inventory.

The measure built by Horvath and Greenberg on Bordin's three components, yielding bond, goal, and task subscales in client, therapist, and observer versions.

Key Researchers

Edward S. Bordin (University of Michigan). Counseling psychologist who recast the psychoanalytic working alliance as a pantheoretical construct of bond, goals, and tasks, the definition the field still uses. Obituary - Memorial

Christoph Fluckiger (University of Kassel). Psychotherapy researcher who led the successive updates of the definitive alliance-outcome meta-analysis for the field's task force. ORCID - Google Scholar

Adam O. Horvath (Simon Fraser University). Built the Working Alliance Inventory that operationalized Bordin's model and co-authored the field's central alliance-outcome meta-analyses. Obituary - Working Alliance Inventory

Jeremy D. Safran (The New School for Social Research). Relational psychotherapy researcher who developed the theory and measurement of alliance ruptures and their repair. Wikipedia - Wikidata

Bruce E. Wampold (University of Wisconsin-Madison; Modum Bad, Norway). Counseling psychologist whose variance-decomposition work showed the alliance is carried more by the therapist than the patient. ORCID - Faculty Page - Wikipedia

Sigal Zilcha-Mano (University of Haifa). Clinical researcher who reframed the alliance's central causal question by separating its trait-like and state-like components. ORCID - Faculty Page - Google Scholar

Frequently Asked Questions

What is the therapeutic alliance?
It is the collaborative, affectively charged partnership between a patient and a therapist through which psychotherapy does its work, defined by Edward Bordin as an affective bond together with agreement on the goals and the tasks of treatment (Bordin, 1979).

Who defined the modern concept?
Edward Bordin, in 1979, recast an idea borrowed from psychoanalysis as a pantheoretical construct present in every change-oriented relationship, decomposing it into bond, goals, and tasks so that it could be assessed across different schools of therapy (Bordin, 1979).

How is the alliance measured?
Most influentially through the Working Alliance Inventory, built by Adam Horvath and Leslie Greenberg on Bordin's three components, which provides bond, goal, and task subscales in client, therapist, and observer versions (Horvath & Greenberg, 1989).

How strongly does the alliance predict outcome?
Meta-analyses spanning four decades converge on a correlation of about 0.28, one of the most reliable findings in psychotherapy research, robust across disorders, measures, and types of treatment (Fluckiger et al., 2018).

Is a correlation of 0.28 large or small?
Both, depending on the scale. It explains under 8 percent of outcome variance, yet on the binomial effect size display it corresponds to raising the improvement rate from 36 to 64 percent, a clinically substantial difference (Fluckiger et al., 2018).

Does the alliance cause improvement, or the reverse?
The direction is contested. Because improving patients may form stronger alliances, recent methods separate a trait-like component that may reflect prognosis from a state-like component that a therapist can move and that is more plausibly causal (Zilcha-Mano, 2017).

What is an alliance rupture?
A strain or breakdown in the collaborative relationship, taking either a withdrawal form, in which the patient retreats, or a confrontation form, in which the patient voices dissatisfaction; ruptures are common and their repair is associated with better outcomes (Safran et al., 2011).

Does the alliance depend more on the therapist or the patient?
Variance-decomposition studies find that the therapist component of the alliance predicts outcome while the within-therapist patient component does not, implying that forming strong alliances is a therapist skill rather than a matter of favorable patients (Baldwin et al., 2007; Del Re et al., 2012).

References