In 1936 a psychologist at Worcester State Hospital sat down to explain something embarrassing. Every school of therapy claimed a different theory of the mind, a different mechanism of cure, a different account of why people suffer. The theories contradicted each other. They could not all be right. And yet, when he surveyed what actually happened to patients, the treatments all seemed to work about equally well. Saul Rosenzweig reached for a line from Alice in Wonderland, where the Dodo judges a footrace by declaring that everybody has won and all must have prizes. The joke stuck. Forty years later, when Lester Luborsky and colleagues ran the comparative outcome data formally, the verdict held: across modalities, the differences in effectiveness were small to nonexistent (Luborsky, Singer, and Luborsky, 1975). Decades of increasingly sophisticated meta-analysis have not overturned it. Bruce Wampold, whose book The Great Psychotherapy Debate remains the most rigorous accounting of the question, estimates that the specific modality accounts for somewhere between zero and one percent of outcome variance, while the treatment package as a whole, alliance, therapist, technique, allegiance, and placebo combined, explains perhaps thirteen percent (Ahn and Wampold, 2001; Wampold and Imel, 2015).

The field has spent ninety years treating this finding as a scandal to be explained away. I want to argue that it is not a scandal. It is the answer. The dodo bird verdict is what you would expect to find if the mechanism of psychotherapy were never located in the content of any theory in the first place. The modalities are not medicines with different active ingredients. They are lenses. And a lens does not have to be true to work. It only has to make you look.

The Detour

The verdict should have unified the profession. Instead the profession collided with managed care, and the collision produced the strangest chapter in the history of clinical science.

Insurance benefit managers in the 1990s wanted randomized controlled trials before they reimbursed a treatment. This was reasonable on its face. But the gold-standard trial requires a treatment manual, a fixed duration, a single DSM diagnosis, and an outcome defined as symptom reduction on a standardized instrument. Only one family of therapies was built around manuals, and so only one family of therapies accumulated trials. Within a decade the phrase evidence-based therapy had become a synonym for cognitive behavioral treatment, not because the evidence favored it over other approaches, but because the trial architecture could only see it. The verdict said the specifics do not matter. The reimbursement system went looking for studies of specifics anyway, and then mistook the volume of studies for the strength of the finding.

The clinical psychologist who did the most to expose this was Jonathan Shedler. His 2010 paper in American Psychologist assembled the meta-analytic evidence that psychodynamic therapy produces effect sizes as large as those of the actively marketed evidence-based treatments, with one striking difference: the benefits of psychodynamic treatment grow after termination rather than fading. One meta-analysis he reviewed found an effect size of 0.97 for general symptom improvement in short-term psychodynamic therapy, rising to 1.51 at long-term follow-up (Shedler, 2010). His later paper, Where Is the Evidence for Evidence-Based Therapy, made the structural point directly: trials of manualized treatment, run for six to twelve weeks, on patients scrubbed of comorbidity, measuring only symptom checklists, bear little resemblance to what skilled clinicians do with actual patients, and the term evidence-based has come to function as a marketing designation rather than a scientific one (Shedler, 2018).

The trial literature itself began to show cracks. A meta-analysis in Psychological Bulletin covering seventy studies from 1977 to 2014 found that the measured effects of CBT for depression had declined steadily since the publication of the first treatment manuals (Johnsen and Friborg, 2015). Reanalyses disputed the trend, arguing the decline was concentrated in studies before 1995 and driven by methodological artifacts (Ljótsson et al., 2017; Cristea et al., 2017), and the honest summary is that the question remains contested. But the fact that a flagship treatment could plausibly be measured as weakening over time, while its brand only strengthened, tells you something about the relationship between the evidence and the label.

Then there is the question of what the trials were measuring in the first place. In 2024 a team led by Michaela Ladmanová published a qualitative meta-analysis in The Lancet Psychiatry synthesizing 177 studies of what clients themselves report gaining from therapy. Symptom relief appeared, but as a fraction of the picture. What clients described, across diagnoses and across modalities, was deeper self-understanding, an enhanced sense of agency, a changed relationship to themselves, and richer engagement with other people (Ladmanová et al., 2025). The entire clinical trial apparatus had been optimized to measure the one outcome that patients treat as a side effect.

So the field arrived at the present moment holding two findings it did not want. The modality does not matter. And symptom checklists miss the point of the enterprise. What remained was to explain why therapy works at all.

The Mechanism

The most promising answer is coming from an unexpected direction. A neuropsychologist at the University of Cape Town spent his career doing something the field considered impossible, holding a chair in neuroscience while practicing as a psychoanalyst, and his recent book The Only Cure argues that the neuroscience of predictive processing has finally supplied the mechanism Freud was reaching for and could not see (Solms, 2025). The same framework has begun appearing in the clinical literature as a candidate unified theory. Daniel Villiger’s integrative model in the Journal of Contemporary Psychotherapy makes the case explicitly: every effective therapy, whatever its vocabulary, works by disrupting entrenched predictions within a generative model of self and world, under conditions safe enough for the prediction errors to register and the model to update (Villiger, 2025).

The predictive processing account runs like this. The brain has no direct access to reality. It maintains an internal generative model, built from memory, and uses that model to predict incoming experience. When prediction and input match, the prediction becomes what you perceive. When they mismatch, the brain must decide whether to trust the model or the evidence, and under threat it trusts the model, because the model kept you alive. The deepest layers of the model were laid down in early childhood, in relationships on which survival depended, in forms unavailable to conscious recall. This is why a grown professional can perceive a neutral email as an abandonment, why a combat veteran can perceive his wife’s footsteps as an ambush, why the patterns Freud called the repetition compulsion persist after insight. The person is not failing to learn. The person is perceiving, accurately, the output of a model that stopped updating.

On this account therapy needs two things, and the predictive processing literature names them precisely. It needs safety, because a threatened system defends its predictions rather than revising them. And it needs precision, a steady supply of experience targeted at exactly the predictions that are causing trouble, delivered in a way the system can register as error rather than as noise. Safety without precision is pleasant and inert. Precision without safety is an assault the model will armor itself against.

Any clinician will recognize what these two conditions have always been called. Safety is the thing a British pediatrician turned analyst named Donald Winnicott called the holding environment more than fifty years ago, the reliable, confidential, bounded, non-retaliatory presence of another mind. And precision is what every tradition has practiced under its own name: interpretation, exposure, chair work, cognitive restructuring, parts dialogue, dream analysis, behavioral experiment. Every modality is a delivery system for targeted prediction error inside a relationship that makes the error survivable.

Here is where I want to push the argument one step further than the predictive processing writers usually take it, because I think the step is where the clinical truth lives.

The Deconstruction and the Witness

The mechanism is not the witnessing alone, and it is not the deconstruction alone. It is the pairing. And the specific lens used for the deconstruction is arbitrary.

Consider what the vocabularies of the modalities actually are. The id is not a real object. Neither is the schema, the automatic thought, the internal family systems part, the archetype, the ego state, the attachment template, or the maladaptive core belief. Open a skull and you will not find any of them. None of these words points back to a thing. Each one is a notation, a way of chunking the continuous flood of inner experience into discrete, nameable, watchable units. The reason a hundred contradictory theories produce the same outcomes is that they were never competing descriptions of the machinery. They were competing notations for the same activity, which is the activity of noticing.

And noticing is not neutral. This is the piece the field keeps missing. When you give a person any coherent system for deconstructing themselves, you change their relationship to their own experience. The panic stops being weather and becomes a part with a job. The self-criticism stops being the truth and becomes a voice with a history. The rage stops being who I am and becomes something I do when a boundary I never knew I had gets crossed. The specific carving does not matter, because the act of carving converts experience from something you are into something you can observe, and a generative model under observation is a model exposed to its own prediction errors. You cannot watch a pattern run, in the presence of another mind, at the moment it runs, and have it remain fully invisible to you. The notation is scaffolding for attention. Attention is what updates the model.

This is why the honest test of a modality is not whether its theory is true. No modality’s theory is true, in the sense of pointing at real objects. The test is whether the lens is generating curiosity or generating avoidance. A frame that draws the patient deeper into their own system, that makes them lean in, that makes the next session feel like an excavation they cannot wait to continue, is working, whatever it is called and to a large extent whatever it assumes about how the mind operates. A frame that produces compliance, intellectualization, or a new vocabulary for the old avoidance is failing, however impressive its evidence base. The clinical skill is not fidelity to the manual. The clinical skill is reading, moment to moment, whether the deconstruction is opening the person or armoring them, and changing lenses when the current one has become a place to hide.

The witness is the other half, and it cannot be removed. Deconstruction alone is available in any bookstore. People read the psychology shelf for years and stay stuck, because a model does not update its survival-critical predictions in solitude. Those predictions are about other minds. They were formed in relationship and they are tested in relationship. The prediction that says my anger will destroy anyone I show it to cannot be disconfirmed by a worksheet. It can only be disconfirmed by getting angry in front of a person who does not leave. The prediction that says if anyone saw all of me they would be disgusted cannot be reasoned away. It has to fail, live, in the presence of a witness who sees and stays. This is why the common factors research keeps finding that the relationship carries the outcome. The relationship is not the pleasant medium in which the real treatment occurs. The relationship is the laboratory in which the oldest predictions are finally allowed to be wrong.

So the formula, if the field wants one, is this: therapy is witnessed self-deconstruction. Give a person a lens, any lens that produces curiosity rather than avoidance, and a mind that can hold what the lens reveals, and the system will begin to decode itself. Withhold either half and nothing happens.

The Part Nobody Wants

This account would be merely tidy if it did not collide with the most inconvenient finding in the entire outcome literature, which is that therapists are not interchangeable, and the difference between them cannot be explained by anything the training system controls.

The finding is old. In 1974 a researcher named David Ricks followed up adult outcomes of adolescent patients and found that the boys treated by one clinician had fared dramatically better than those treated by another with identical caseloads, and he coined the term supershrink for what he had found. The modern data are blunter. A study of 1,841 clients seen by 91 therapists at a university counseling center found that the clients of the most effective therapists improved at roughly ten times the rate of the average, that the least effective therapists’ caseloads got worse, and that theoretical orientation and type of training explained none of the difference (Okiishi, Lambert, Nielsen, and Ogles, 2003). A large British practice-based study found recovery rates ranging from twenty percent to ninety-six percent depending on which therapist a patient happened to draw (Saxon and Barkham, 2012). Therapist effects reliably account for five to eight percent of outcome variance (Baldwin and Imel, 2013), which does not sound like much until you remember that the modality accounts for roughly zero. And years of experience does not predict effectiveness. In one of the largest longitudinal datasets available, therapist outcomes did not improve with time in practice and showed a slight tendency to erode (Goldberg et al., 2016).

Degree does not predict it. Modality does not predict it. Experience does not predict it. So what does?

Two research programs answered the question, and their answer is the one the profession’s institutions are structurally unable to metabolize. Timothy Anderson and colleagues developed a performance task called Facilitative Interpersonal Skills, in which prospective therapists respond to video recordings of difficult clinical moments, and trained raters score the responses for empathy, warmth, verbal fluency, persuasiveness, emotional expression, and the capacity to work with a rupture rather than flee it. FIS scores predict which therapists get good outcomes (Anderson, Ogles, Patterson, Lambert, and Vermeersch, 2009). The devastating study came later. Anderson’s team administered the task to clinical psychology trainees in the initial weeks of their graduate training, before the training could have taught them anything, and then waited. When those trainees began seeing patients more than a year later, the scores from week one predicted their patients’ outcomes (Anderson, McClintock, Himawan, Song, and Patterson, 2016). A German group led by Henning Schöttke found the same thing over a longer horizon: interpersonal behavior rated during an observed group interaction, assessed prior to formal training, predicted trainees’ patient outcomes across the following five years (Schöttke, Flückiger, Goldberg, Eversmann, and Lange, 2017).

Read those studies together and the conclusion is unavoidable. The capacity that makes a therapist effective is substantially present before the training begins. Students do not arrive as blank slates onto which a curriculum writes competence. They arrive already sorted, carrying or not carrying the thing, and the curriculum decorates whatever walked in the door. Training can add technique, and technique matters at the margins. But the core capacity, the ability to sit inside another person’s dysregulation without defending against it, to perceive the emotional subtext of an attack, to stay warm under fire, to be the witness in whose presence an old prediction can safely fail, is not being produced by graduate education. It is being screened for, badly, by accident, or not at all.

This is where the anger belongs, and it belongs at structures rather than at any of the people inside them. A graduate program funded by tuition cannot conclude that admission matters more than curriculum, because the curriculum is the product. A continuing education industry that retails branded acronyms at seminar prices cannot conclude that the acronym is a notation rather than an ingredient, because the notation is the inventory. An accreditation system built to certify the completion of coursework cannot certify a capacity that coursework does not create. And an insurance system that reimburses manualized procedures cannot see a treatment whose active ingredient is a person. So the institutions do what institutions do when a finding threatens their operating model. They act as if the question were unresearchable. If we cannot teach it, the reasoning runs, why would we study it, and the reasoning is exactly backwards, because it has been studied, the studies replicated, and the finding is precise: this is something you select for. The research the field pretends does not exist is not a gap in the literature. It is a fully drawn map of a door the profession keeps declining to walk through, because on the other side of the door the business model changes.

None of this means the capacity is mystical or fixed at birth. Anderson’s own later work suggests some FIS components respond to deliberate, targeted practice, and the deliberate practice movement around researchers like Scott Miller has shown that therapists who systematically study their own failures can improve. But improving a capacity is not the same as installing one, and a profession honest about its own evidence would restructure around the difference. It would screen applicants with performance tasks instead of GRE scores. It would treat modality training as the acquisition of additional lenses rather than as conversion to a faith. It would measure every clinician’s outcomes and feed the data back. It would admit that what it sells, when it works, is not a procedure but a rare pairing: a person who can witness, holding a lens that makes you curious.

What This Means If You Are Looking for a Therapist

The practical translation is short. Stop shopping for modalities. The acronym on the website tells you what notation the clinician was trained in, and the notation accounts for approximately none of your outcome. Shop for the two things the evidence actually supports. First, a person in whose presence you can feel your guard drop, because no model updates behind a wall. Second, a person whose way of taking you apart makes you more curious about yourself rather than more defended, more fluent in excuses, or more skilled at performing insight. If the work is generating genuine curiosity and pulling you deeper into your own system, it is working, whatever it is called. If you notice the frame has become a place to hide, say so, and a good clinician will change the lens rather than defend it. The lens was never the point. You were.

I practice eclectically, and I used to half apologize for that, as if it were a failure to commit. The outcome literature has relieved me of the apology. Every model of the psyche is an impressionist painting of a phenomenon nobody has seen directly. I keep a large bag of them because different patients wake up under different paint. The paint is arbitrary. The waking is not. What a century of research found, under all the institutional noise, is something clinicians of every school have known in the room and been trained to distrust: people heal when they are given a way to watch themselves and someone worth being watched by. The map is not the medicine. The looking, together, is.

References

Ahn, H., and 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.

Anderson, T., Ogles, B. M., Patterson, C. L., Lambert, M. J., and Vermeersch, D. A. (2009). Therapist effects: Facilitative interpersonal skills as a predictor of therapist success. Journal of Clinical Psychology, 65(7), 755–768.

Anderson, T., McClintock, A. S., Himawan, L., Song, X., and Patterson, C. L. (2016). A prospective study of therapist facilitative interpersonal skills as a predictor of treatment outcome. Journal of Consulting and Clinical Psychology, 84(1), 57–66.

Baldwin, S. A., and Imel, Z. E. (2013). Therapist effects: Findings and methods. In M. J. Lambert (Ed.), Bergin and Garfield’s Handbook of Psychotherapy and Behavior Change (6th ed., pp. 258–297). Wiley.

Cristea, I. A., Stefan, S., Karyotaki, E., David, D., Hollon, S. D., and Cuijpers, P. (2017). The effects of cognitive behavioral therapy are not systematically falling: A revision of Johnsen and Friborg (2015). Psychological Bulletin, 143(3), 326–340.

Goldberg, S. B., Rousmaniere, T., Miller, S. D., Whipple, J., Nielsen, S. L., Hoyt, W. T., and Wampold, B. E. (2016). Do psychotherapists improve with time and experience? A longitudinal analysis of outcomes in a clinical setting. Journal of Counseling Psychology, 63(1), 1–11.

Johnsen, T. J., and Friborg, O. (2015). The effects of cognitive behavioral therapy as an anti-depressive treatment is falling: A meta-analysis. Psychological Bulletin, 141(4), 747–768.

Ladmanová, M., Řiháček, T., Timulak, L., Jonášová, K., Kubantová, B., Mikoška, P., Polakovská, L., and Elliott, R. (2025). Client-identified outcomes of individual psychotherapy: A qualitative meta-analysis. The Lancet Psychiatry, 12(1), 18–31.

Ljótsson, B., Hedman, E., Mattsson, S., and Andersson, E. (2017). The effects of cognitive-behavioral therapy for depression are not falling: A re-analysis of Johnsen and Friborg (2015). Psychological Bulletin, 143(3), 321–325.

Luborsky, L., Singer, B., and Luborsky, L. (1975). Comparative studies of psychotherapies: Is it true that “everyone has won and all must have prizes”? Archives of General Psychiatry, 32(8), 995–1008.

Okiishi, J. C., Lambert, M. J., Nielsen, S. L., and Ogles, B. M. (2003). Waiting for supershrink: An empirical analysis of therapist effects. Clinical Psychology and Psychotherapy, 10(6), 361–373.

Ricks, D. F. (1974). Supershrink: Methods of a therapist judged successful on the basis of adult outcomes of adolescent patients. In D. F. Ricks, M. Roff, and A. Thomas (Eds.), Life History Research in Psychopathology (pp. 231–258). University of Minnesota Press.

Rosenzweig, S. (1936). Some implicit common factors in diverse methods of psychotherapy. American Journal of Orthopsychiatry, 6(3), 412–415.

Saxon, D., and Barkham, M. (2012). Patterns of therapist variability: Therapist effects and the contribution of patient severity and risk. Journal of Consulting and Clinical Psychology, 80(4), 535–546.

Schöttke, H., Flückiger, C., Goldberg, S. B., Eversmann, J., and Lange, J. (2017). Predicting psychotherapy outcome based on therapist interpersonal skills: A five-year longitudinal study of a therapist assessment protocol. Psychotherapy Research, 26(5), 542–551.

Shedler, J. (2010). The efficacy of psychodynamic psychotherapy. American Psychologist, 65(2), 98–109.

Shedler, J. (2018). Where is the evidence for “evidence-based” therapy? Psychiatric Clinics of North America, 41(2), 319–329.

Solms, M. (2025). The Only Cure: Freud and the Neuroscience of Mental Healing. Simon and Schuster.

Villiger, D. (2025). An integrative model of psychotherapeutic interventions based on a predictive processing framework. Journal of Contemporary Psychotherapy, 55, 39–49.

Wampold, B. E., and Imel, Z. E. (2015). The Great Psychotherapy Debate: The Evidence for What Makes Psychotherapy Work (2nd ed.). Routledge.

Winnicott, D. W. (1965). The Maturational Processes and the Facilitating Environment. International Universities Press.

Tags: psychotherapy research, common factors, dodo bird verdict, predictive processing, therapist effects, Jonathan Shedler, Bruce Wampold, Mark Solms, evidence-based practice, psychodynamic therapy, CBT effectiveness, facilitative interpersonal skills, therapeutic alliance, psychotherapy training, depth psychology

Meta description: A century of research says the therapy modality barely matters. What heals is witnessed self-deconstruction, and the capacity to provide it is screened for, not taught.