Psychotherapy

Abstract

Psychotherapy is the treatment of psychological disorder and distress by systematic conversation and structured psychological procedures rather than by medication. This article treats it as a scientific object: something whose effects can be measured, whose active ingredients can be argued about, and whose mechanisms can be studied. It traces the outcome literature from Eysenck's sceptical challenge, through the first meta-analyses that established a large average benefit, to the enduring debate between those who credit a therapy's specific techniques and those who credit the factors common to all therapies, including the therapeutic alliance. It covers the dose-response relationship between sessions and improvement, the study of mechanisms of change, and current work on process-based and internet-delivered treatment. Three interactive demonstrations model the effect size of treatment, the alliance-outcome correlation, and the dose-response curve.

Keywords: psychotherapy, common factors, therapeutic alliance

Psychotherapy is the deliberate use of a structured relationship and a set of psychological procedures to relieve mental suffering and change unwanted patterns of thought, feeling, and behaviour. What distinguishes it as a subject of cognitive and clinical science is not any single school but the willingness to ask empirical questions about it: does it work, how much, for whom, and by what means. For most of the twentieth century those questions were argued from case reports and allegiance; over the last fifty years they have been answered, and complicated, by controlled trials and their synthesis. This article follows that arc, from the first serious doubt that psychotherapy did anything at all to the current fine-grained dispute over which of its ingredients carry the effect.

What Psychotherapy Is

Psychotherapy is a family of treatments in which a trained practitioner uses psychological means, principally conversation, structured exercises, and a purposeful relationship, to reduce symptoms and foster change in a person seeking help. It is defined by its method rather than its target: the same disorders treated pharmacologically can be treated psychotherapeutically, and the two are often combined. What all its forms share is that the treatment is delivered through interaction and learning rather than through a drug, a device, or surgery.

A working definition due to Bruce Wampold captures the field's own self-understanding: psychotherapy is a primarily interpersonal treatment, based on psychological principles, involving a trained therapist and a client who has a mental disorder, problem, or complaint, that is intended by the therapist to be remedial for the client's difficulties, and that is adapted or individualized for the particular client and problem (Wampold, 2015). Each clause does work. The treatment is interpersonal and psychological, which separates it from medication; it is delivered by a trained agent who intends it to help, which separates it from ordinary supportive conversation; and it is tailored, which separates it from a fixed protocol applied regardless of the person. The scientific questions that follow all concern how much of the resulting benefit flows from the specific psychological principles a school invokes and how much from the interpersonal frame every school shares.

Types of Psychotherapy

Beyond being a subject in its own right, Psychotherapy is a formal category in the National Library of Medicine's Medical Subject Headings, which places it at tree position F04.754, beneath Behavioral Disciplines and Activities, and hangs its recognised narrower kinds beneath it. These subtypes are a classification built to index the literature, not a claim about the mind's natural joints; they cut across one another, so a single course of treatment may belong to several at once, and the categories mix broad traditions with single named techniques. Table 1 lists the direct children of the descriptor; none yet has its own article on this site, so all are shown as plain text.

The list is heterogeneous by design. It places a whole tradition such as Behavior Therapy beside a single technique such as Hypnosis, and an established treatment such as Interpersonal Psychotherapy beside adjuncts of contested evidential status such as Aromatherapy or Color Therapy. Inclusion in the index is a statement that a literature exists to be catalogued, not a warrant of efficacy; the sections below concern the mainstream verbal psychotherapies for which the outcome evidence is strongest.

The Question of Outcome

The modern science of psychotherapy begins with an attack on it. In 1952 Hans Eysenck reviewed the available follow-up data and argued that patients who received psychotherapy improved no more often than those who did not, estimating that roughly two thirds of neurotic patients recovered within two years whether or not they were treated (Eysenck, 1952). The figure was later shown to rest on weak comparisons and mismatched samples, but the challenge was salutary: it forced the field to justify itself with controlled evidence rather than clinical conviction, and it set the outcome question, does psychotherapy produce benefit beyond the passage of time and spontaneous remission, as the one that had to be answered first.

The answer arrived a quarter of a century later in a new form. Gene Glass had invented meta-analysis, the statistical synthesis of many studies onto a common scale, precisely to settle the psychotherapy debate, and with Mary Lee Smith he applied it to 375 controlled studies. Expressing every study's result as an effect size, the standardized difference between treated and control groups, they found a mean of about 0.85 standard deviations, which implies that the average treated client ended up better off than roughly eighty percent of untreated controls (Smith & Glass, 1977). Figure 1 shows what an effect of that magnitude looks like as two overlapping distributions, and the first demonstration lets a reader vary the effect size and read off the percentile it implies.

That finding has proved remarkably durable. Successive and more rigorous syntheses, correcting for the methodological faults of the early studies, have continued to find medium-to-large average effects for the major disorders, and the benefit of cognitive behavioural therapy for depression and the anxiety disorders in particular is among the best-established results in clinical science (Cuijpers et al., 2016). The most comprehensive recent appraisal, an umbrella review pooling many meta-analyses across disorders and comparing psychotherapies with medications, concludes that both produce genuine but generally modest effects, that neither is dramatically superior to the other for most conditions, and that the effects are often smaller than the enthusiastic early literature suggested once biases are taken into account (Leichsenring et al., 2022). Even long-term psychodynamic therapy, long criticised as unfalsifiable, has been shown in meta-analysis to outperform shorter or less intensive treatments for complex and chronic disorders (Leichsenring & Rabung, 2008). The outcome question, in short, has been answered in psychotherapy's favour; the harder questions concern why it works.

The Dodo Bird Verdict and the Common Factors

If psychotherapy works, the natural next question is which psychotherapy works best, and here the evidence has produced one of the most contested findings in the field. As early as 1936 Saul Rosenzweig noticed that the wildly different therapies of his day all seemed to help, and proposed that their shared features, rather than their competing theories, did the work; he borrowed the Dodo bird's line from Alice in Wonderland, everybody has won and all must have prizes, to name the puzzle (Rosenzweig, 1936). Four decades later Lester Luborsky and colleagues revived the phrase after a systematic comparison of psychotherapies found few reliable differences in their outcomes, giving the Dodo bird verdict its enduring name and its empirical backing (Luborsky et al., 1975).

The verdict is the central evidence for the common-factors account, which holds that the effects of therapy flow largely from ingredients shared by all credible treatments, the relationship, the rationale, the ritual of a healing practice, the mobilisation of hope, rather than from the specific techniques that distinguish schools from one another. The account was given its most influential early form by Jerome Frank, who argued that all effective psychotherapies, and indeed forms of healing across cultures, share the same core components, an emotionally charged and confiding relationship, a healing setting, a rationale that explains the symptoms, and a ritual that follows from it, and that they work chiefly by countering the demoralization that brings people to treatment (Frank, 1971). Wampold's contextual model develops this into a full theory: a bona fide therapy heals through an emotionally charged bond with a helper, a plausible explanation of the client's problem, and a set of actions consistent with that explanation, and the specific content of the actions matters mainly because it sustains the client's belief and engagement (Wampold, 2015). On this view the demonstrated equivalence of therapies is not a scandal to be explained away but the direct prediction of a model in which the common pathway carries most of the effect. The rival medical or specific-ingredients view holds instead that particular techniques exert particular effects on particular disorders, as a drug acts on a disease, and that better-designed trials will eventually resolve the apparent equivalence into real differences.

The Therapeutic Alliance

Whichever side of the common-factors debate one takes, the single most studied ingredient of therapy is the therapeutic alliance, the quality of the collaborative relationship between client and therapist. Its intellectual root is Carl Rogers' hypothesis that a small set of relationship conditions offered by the therapist, empathic understanding, congruence, and unconditional positive regard, were together the necessary and sufficient conditions of therapeutic personality change, a claim that made the relationship itself the active agent of therapy (Rogers, 1957). Edward Bordin gave the concept its modern, pan-theoretical form in 1979 by decomposing the alliance into three elements that any therapy requires: an affective bond between the two parties, and agreement on the goals and on the tasks of the treatment (Bordin, 1979). Because the definition is stated in terms common to all approaches, it made the alliance measurable across schools and turned it into the workhorse variable of process research.

The empirical result is one of the most consistent in the literature: the strength of the alliance predicts the outcome of therapy, moderately but reliably, across thousands of studies and every kind of treatment. The most authoritative meta-analysis, synthesising more than three hundred studies, puts the alliance-outcome correlation at about 0.28 (Fluckiger et al., 2018). That is a modest correlation in absolute terms, accounting for roughly eight percent of the variance in outcome, but it is larger and far more consistent than the effect of any specific technique, which is why reviews of the factors that make therapy work place the relationship near the top (Lambert & Barley, 2001). The evidence-based practice movement has responded by cataloguing the qualities of the therapy relationship that demonstrably improve outcome, including the alliance, empathy, and the collaborative negotiation of goals, and treating them as elements to be cultivated deliberately rather than left to chance (Norcross & Lambert, 2018). A recurring caution attaches to all of this: a correlation between alliance and outcome does not establish that the alliance causes the improvement, since early symptom relief can itself strengthen the alliance, and disentangling the direction of the effect is an active methodological problem.

Cognitive and Behavioural Models

The strongest case for specific ingredients comes from the cognitive and behavioural therapies, which were built from explicit theories of psychopathology and tested as those theories predicted. Aaron Beck's cognitive therapy is the paradigm case. Beck proposed that emotional disorders are maintained by systematically biased thinking, negative automatic thoughts and the dysfunctional schemas beneath them, and that changing those cognitions relieves the disorder; crucially, he framed the approach from the start as continuous with behaviour therapy and as an empirical enterprise whose claims could be tested (Beck, 1970). The resulting cognitive behavioural therapy became the most extensively evaluated of all psychotherapies.

Its evidence base is substantial and specific. Meta-analysis of randomized placebo-controlled trials shows cognitive behavioural therapy to be efficacious across the anxiety and related disorders, with effects that survive comparison against credible placebos rather than only against waiting lists (Carpenter et al., 2018), and the broader synthesis for depression and anxiety confirms medium-to-large effects that place it among the best-supported treatments in mental health (Cuijpers et al., 2016). This is exactly the pattern the specific-ingredients view predicts: a theory of the disorder generating a technique that outperforms an inert control. Yet the same treatments contain a powerful relationship and a compelling rationale, the common factors, so a strong showing for cognitive behavioural therapy does not by itself adjudicate the debate. The two accounts are not mutually exclusive; the live question is how the total effect divides between the specific and the shared, and by what mechanism each acts.

The Dose-Response Relationship

One of the most orderly quantitative regularities in the field concerns not which therapy is given but how much of it. Kenneth Howard and colleagues assembled data across many studies and plotted the probability of measurable improvement against the number of sessions attended, finding a dose-response relationship of a characteristic shape: improvement accumulates rapidly at first and then with diminishing returns, so that the curve is roughly linear in the logarithm of the number of sessions (Howard et al., 1986). In their synthesis about half of patients showed measurable improvement by around the eighth session, and roughly three quarters by the twenty-sixth, with each subsequent block of sessions adding less than the one before.

The practical implications are considerable and have shaped how services are planned. A negatively accelerated dose-response curve means that the first few sessions are the most valuable per session, which argues for ensuring that many people receive at least a short course rather than that a few receive a very long one, and it provides a rational basis for the time-limited treatments that dominate practice. It also supplies a benchmark against which an individual's progress can be judged: a client falling well below the expected recovery curve for their number of sessions can be flagged, and the feedback used to adjust the treatment, an application that has itself been shown to improve outcomes for cases at risk of deterioration. The curve is an average, not a law for any individual, and the second demonstration lets a reader move along it and read the expected proportion improved at each dose.

Mechanisms of Change

Establishing that therapy works, and even which ingredients correlate with its working, still leaves the deepest question open: by what causal mechanism does a given treatment produce its effect. Alan Kazdin has argued forcefully that the field has spent its energy demonstrating that treatments work while neglecting to explain how, and that the study of mechanisms of change, the processes through which therapy actually causes improvement, is the most important unfinished business in psychotherapy research (Kazdin, 2007). The distinction he draws is between a mediator, a variable that statistically carries the effect of treatment on outcome, and a genuine mechanism, the real causal process the mediator only points to; showing that change in a hypothesised process precedes and accounts for change in symptoms is far harder than showing that a treatment beats a control, and far rarer.

The stakes are practical as well as theoretical. If the mechanism of a therapy were known, treatment could be refined to engage it directly, stripped of inert components, and matched to the clients whose difficulty involves that mechanism. The requirements Kazdin sets out for establishing a mechanism, a strong association, a specific temporal order in which the process changes before the outcome, evidence of a gradient, and experimental manipulation of the proposed process, are demanding, and few claimed mechanisms in the literature meet them. This emphasis on process rather than package underlies much of the field's current direction.

Current Directions

The mechanism agenda has given rise to a movement to reorganise therapy around processes rather than named brands. Process-based therapy, advanced by Stefan Hofmann and Steven Hayes, proposes to abandon the proliferation of separately branded treatment packages in favour of identifying the empirically supported processes of change, cognitive, emotional, behavioural, and relational, that cut across therapies, and assembling an individualised treatment from the processes a particular client's problem actually engages (Hofmann & Hayes, 2019). It is an attempt to make Kazdin's programme operational and to reconcile the common-factors and specific-ingredients camps by locating the specificity at the level of process rather than protocol.

A second front is the expansion of delivery. Internet-delivered psychological treatments, especially guided internet-based cognitive behavioural therapy, have accumulated a large trial base showing effects broadly comparable to face-to-face treatment for several common disorders while reaching people who could not otherwise be treated (Andersson, 2016). The finding sharpens the theoretical debate, since a treatment that works with minimal therapist contact strains any account that locates the whole effect in the in-person relationship, while also raising new questions about the form the alliance takes at a distance. Both developments proceed against a backdrop of intensified scrutiny of the evidence base itself, described next.

Criticisms and Open Questions

The outcome literature that underwrites psychotherapy has itself come under critical examination, and two biases in particular temper the headline effect sizes. The first is researcher allegiance: the tendency for a trial's outcome to favour whichever therapy its investigators believe in. An overview of reviews found allegiance to be a substantial and pervasive influence on reported results, large enough to account for a meaningful share of the apparent differences between treatments and to inflate the effects of the favoured one (Munder et al., 2013). The second is publication bias. A review of psychotherapy trials for depression funded by the United States National Institutes of Health, a registry in which unpublished studies could be identified, found that the published literature overstated the treatments' efficacy by around a quarter once the unpublished trials were included (Driessen et al., 2015). Together these findings imply that the true average effect is real but smaller than the enthusiastic literature suggests, which is precisely the conclusion the most careful recent umbrella review reaches (Leichsenring et al., 2022). A further corrective is that the average benefit conceals real harm: a minority of clients deteriorate over the course of treatment, and some interventions have been shown to make people reliably worse, so that psychotherapy, like any potent treatment, carries the capacity to injure as well as to help and cannot be assumed benign (Lilienfeld, 2007).

Beyond the biases lie genuine open questions. The common-factors and specific-ingredients accounts remain unresolved because the two make similar predictions about outcome and are hard to separate with the equivalence designs that dominate the field; distinguishing them requires dismantling studies and mechanism tests that are rarely run (Kazdin, 2007). The alliance-outcome correlation, robust as it is, still lacks a decisive causal interpretation. And the question of for whom a given therapy works, as against the average client, remains largely unanswered, which is the gap the move toward process-based and personalised treatment is meant to close. None of this undoes the central result that psychotherapy helps; it defines the standard of evidence the next generation of research must meet.

Worked Example

The first demonstration makes the meta-analytic effect size concrete. An effect size is the difference between the treated and control means measured in standard deviations, and when the two groups' outcomes are normally distributed with equal spread, the proportion of treated clients who score above the average control equals the area under the standard normal curve up to the effect size. For the Smith and Glass value of 0.85, that area is about 0.80, so the average treated client is better off than roughly eighty percent of untreated controls, the figure the demonstration displays. Reducing the effect size to a more conservative 0.50 lowers that proportion to about 0.69, and raising it to 1.0 lifts it to about 0.84; each value of the effect size maps to a single percentile, which is the number the slider reads out.

The second demonstration turns the alliance-outcome correlation into variance explained. A correlation of 0.28, the meta-analytic estimate for the alliance, explains its square as a fraction of the variance in outcome: 0.28 multiplied by 0.28 is about 0.078, so the alliance accounts for roughly eight percent of the differences in how clients fare. Squaring makes the point that a moderate correlation corresponds to a small share of variance, which is why the alliance is described as reliable rather than large, even as it remains the strongest and most consistent single correlate of outcome that the field has found.

The third demonstration traces Howard's dose-response curve. Writing the probability of improvement as the normal-curve value of a linear function of the log number of sessions, with the constants that fit Howard's data, the model returns about a fifty percent chance of measurable improvement by the eighth session, about seventy-five percent by the twenty-sixth, and about eighty-six percent by the fifty-second. The equal steps in outcome come at geometrically growing costs in sessions, the signature of the negatively accelerated curve, and the demonstration draws that curve and marks these three points as the session slider moves.

Discussion

The scientific study of psychotherapy has moved through three questions in sequence, and the field's structure mirrors that history. The first question, whether therapy works at all, was posed sharply by Eysenck and answered by meta-analysis: it does, with a large and durable average effect. The second question, which therapy works best, produced the Dodo bird verdict and the long contest between the common-factors and specific-ingredients accounts, a contest that the evidence for the therapeutic alliance and the parallel evidence for cognitive behavioural technique have deepened rather than settled. The third question, how therapy works, is the current frontier, and it reframes the second: if the mechanisms of change can be identified, the argument over brands dissolves into an inventory of processes. Table 2 sets the two principal accounts of the therapeutic effect side by side.

Read as a whole, the psychotherapy literature is a case study in how a practice becomes a science: by submitting a cherished activity to measurement, discovering that it works, and then finding that the discovery only sharpens the questions of why and how. The average effect is settled and the biases that once inflated it are now measured and discounted. What remains is the harder programme of identifying the processes through which a conversation changes a mind, and of matching those processes to the people they can help, so that the answer to the outcome question can at last be given not for the average client but for each one.

Glossary

Cognitive behavioural therapy.

A family of structured, present-focused treatments derived from Beck's cognitive therapy and from behaviour therapy, which change symptoms by changing maladaptive thoughts and behaviours; the most extensively evaluated of the psychotherapies.

Common factors.

The ingredients shared by all credible therapies, such as the relationship, a persuasive rationale, and the mobilisation of hope, held by the contextual model to carry most of the therapeutic effect.

Contextual model.

Wampold's theory that therapy heals through an emotionally charged bond with a helper, a plausible explanation of the problem, and actions consistent with that explanation, rather than through school-specific techniques.

Dodo bird verdict.

The claim, named from Alice in Wonderland, that different bona fide psychotherapies produce broadly equivalent outcomes, so that everybody has won and all must have prizes.

Dose-response relationship.

The negatively accelerated relation between the number of therapy sessions and the probability of improvement, roughly linear in the logarithm of sessions, so that early sessions yield the largest gains.

Effect size.

A standardized measure of the magnitude of a treatment's benefit, here the difference between treated and control means expressed in standard deviations, allowing results to be pooled across studies in meta-analysis.

Mechanism of change.

The actual causal process through which a therapy produces its effect, distinguished from a mediator, which only statistically carries the effect and points toward the mechanism without establishing it.

Mediator.

A variable that statistically transmits the effect of a treatment on an outcome; a necessary but not sufficient step toward identifying a genuine mechanism of change.

Meta-analysis.

The statistical synthesis of the results of many studies onto a common scale, invented in part to settle the psychotherapy outcome debate and now the standard tool for estimating average effects.

Process-based therapy.

An approach that replaces branded treatment packages with an individualised assembly of the empirically supported processes of change a particular client's problem engages.

Publication bias.

The distortion that arises when studies with positive results are more likely to be published than those without, inflating the apparent efficacy of a treatment in the visible literature.

Researcher allegiance.

The tendency for the outcome of a trial to favour the therapy its investigators believe in, a pervasive bias that can account for part of the apparent differences between treatments.

Specific ingredients.

The techniques unique to a particular therapy, held by the medical model to exert particular effects on particular disorders, as against the common factors shared by all therapies.

Spontaneous remission.

Recovery from a disorder without treatment, the baseline against which any therapy must show added benefit, and the phenomenon on which Eysenck built his challenge to psychotherapy.

Therapeutic alliance.

The collaborative relationship between client and therapist, decomposed by Bordin into an affective bond and agreement on the goals and tasks of treatment; the most consistent process correlate of outcome.

Working alliance.

Bordin's term for the therapeutic alliance conceived as the collaboration required for the work of therapy, framed in terms general enough to apply across every school of treatment.

Key Researchers

Gerhard Andersson. Professor of clinical psychology at Linkoping University and Karolinska Institutet; a pioneer of internet-delivered cognitive behavioural therapy and its evidence base. ORCID - Faculty Page - Google Scholar - Wikipedia

Aaron T. Beck (1921-2021). Professor at the University of Pennsylvania; founder of cognitive therapy and, through it, of the cognitive behavioural tradition that became the most evaluated of the psychotherapies. Faculty Page - Wikipedia

Pim Cuijpers. Professor of clinical psychology at Vrije Universiteit Amsterdam; the leading meta-analyst of psychotherapy for depression and the anxiety disorders. ORCID - Faculty Page - Google Scholar

Hans J. Eysenck (1916-1997). Professor at the Institute of Psychiatry, King's College London; his 1952 critique of psychotherapy's effectiveness catalysed the whole enterprise of controlled outcome research. Google Scholar - Wikipedia

Stefan G. Hofmann. Alexander von Humboldt Professor at the Philipps-University of Marburg; a CBT meta-analyst and, with Steven Hayes, co-architect of process-based therapy. ORCID - Faculty Page - Google Scholar - Wikipedia

Alan E. Kazdin. Sterling Professor of Psychology Emeritus at Yale University; a leading voice for the study of mediators and mechanisms of change and for closing the treatment-delivery gap. ORCID - Faculty Page - Google Scholar - Wikipedia

John C. Norcross. Professor of psychology at the University of Scranton; leader of the evidence-based programme on the therapy relationships that work and on psychotherapy integration. ORCID - Faculty Page - Google Scholar - Wikipedia

Carl R. Rogers (1902-1987). Founder of client-centred therapy; his necessary-and-sufficient-conditions hypothesis put the therapeutic relationship at the centre of the change process. Google Scholar - Wikipedia

Bruce E. Wampold. Emeritus Professor at the University of Wisconsin-Madison and at the Modum Bad Psychiatric Center; architect of the contextual model and leading proponent of the common-factors account. ORCID - Google Scholar - Wikipedia

Frequently Asked Questions

What is psychotherapy?
It is the treatment of psychological disorder and distress through structured conversation, purposeful relationship, and psychological procedures delivered by a trained practitioner, rather than through medication (Wampold, 2015).

Does psychotherapy actually work?
Yes. Meta-analysis of hundreds of controlled studies finds a large average benefit, with the typical treated person ending up better off than about eighty percent of comparable untreated people, a result that has held up across decades of more rigorous synthesis (Smith & Glass, 1977).

What is the Dodo bird verdict?
It is the finding that different established therapies tend to produce broadly similar outcomes, named from the Dodo bird in Alice in Wonderland who declares that everybody has won and all must have prizes (Luborsky et al., 1975).

What are common factors?
They are the ingredients shared by all credible therapies, such as the therapeutic relationship, a persuasive rationale, and the mobilisation of hope, which the contextual model holds to carry most of the effect of treatment (Wampold, 2015).

What is the therapeutic alliance and why does it matter?
It is the collaborative bond between client and therapist together with their agreement on the goals and tasks of treatment; its strength correlates with outcome at about 0.28 across every kind of therapy, making it the most consistent single process predictor of benefit (Fluckiger et al., 2018).

Is one type of therapy better than the others?
For most disorders the differences between established therapies are small, though cognitive behavioural therapy has the largest and most specific evidence base, particularly for depression and the anxiety disorders (Cuijpers et al., 2016).

How many sessions of therapy are needed?
Improvement follows a dose-response curve with diminishing returns: about half of clients show measurable improvement by the eighth session and about three quarters by the twenty-sixth, so the earliest sessions are the most valuable (Howard et al., 1986).

Why do experts still debate how psychotherapy works?
Because showing that a treatment works does not reveal the causal mechanism behind it, and separating the effect of specific techniques from that of shared factors requires mechanism studies that are demanding and rarely conducted (Kazdin, 2007).

References