The practitioner parallelsdraft

In psychotherapy, the individual therapist accounts for about one twentieth of the differences between clients' outcomes, and the studies disagree widely

Psychotherapy research calls this share the therapist effect: the proportion of the total variation in clients' outcomes that goes with which therapist the client saw rather than with anything else the study measured. A systematic review published in 2019 reports the earlier pooling and adds its own. The earlier pooling, by Baldwin and Imel in 2013, took 46 studies covering 1,218 therapists and 14,519 patients, and found the therapist accounting for 5% of the variation overall. It differed by setting: in the 17 studies of ordinary practice the share was 7%, and in the 29 trials it was 3%, a difference the 2019 review attributes to trials selecting their patients tightly, following manuals, checking that the manual was followed, supervising closely and using fewer therapists. Across those 46 studies the share ranged from 0% to 55%.

The 2019 review searched the same way for the studies published since and found 20 that qualified, three randomised trials and seventeen studies of ordinary practice. Nineteen of the twenty found a therapist effect. Across all twenty the share ran from 0.2% to 29%, with a weighted average of 5%, the same figure as before. Its two subgroups did not repeat the earlier pattern: the seventeen practice-based studies averaged 5% across a range of 0.2% to 21%, and the three randomised trials averaged 8.2% across a range of 1% to 29%, which is the earlier difference the other way round on a base of three studies. The review's own conclusion is that therapist effects are a robust phenomenon, that considerable heterogeneity exists across studies, and that "using an overall therapist effects statistic may lack precision" so that the effects "might be better reported separately for specific clinical settings". That a practitioners' effect on clients' mental health exists, and the research on medicines against placebo that made it a question worth asking, are set out on the Measurement as the intervention page. The matching medical work, including the author of these Tools' own two systematic reviews with their counts and their shares, is on the State of Research page.

Source Johns, Barkham, Kellett and Saxon, "A systematic review of therapist effects: A critical narrative update and refinement to Baldwin and Imel's (2013) review", Clinical Psychology Review 67, 2019, pages 78 to 93, abstract and introduction. Read on 31 August 2026 from the authors' accepted copy in the White Rose repository, which is published under a Creative Commons licence. The Baldwin and Imel figures above are the 2019 review's own restatement of that work, which is a book chapter whose publisher does not answer a request from this network; the therapist count of 1,218 is the number that review prints, and another restatement of the same chapter gives 1,281.

A share of variation is not a size of benefit, and this share answers only how much of the spread between clients goes with the practitioner. The pooled figure is an average over studies that disagree by more than the figure itself, which is the review's own point. The studies were also not designed to measure this: in the earlier pooling the median number of therapists in a study was nine, with a range of 2 to 581, and in only two of the 46 did the average therapist treat more than 30 patients, so most of those shares were estimated from very few practitioners. The review reports that therapist sample sizes have remained lower than recommended. None of this work is about money, and none of it covers financial advice.

In one large service, the recovery rates of individual therapists ran from about a quarter of their patients to almost all of them

Saxon and Barkham took the records of the United Kingdom National Health Service's primary care counselling and psychological therapy services from September 2000 to July 2008: 119 therapists and 10,786 patients who had at least two sessions, a planned ending, and a completed standard questionnaire before and after treatment. The share of the variation in outcomes going with the individual therapist was 6.6% for a patient of average severity, and it rose with severity, running from 1% to 10% as patients' intake scores rose. The share also moved when the model was changed: leaving out how much risk a therapist's caseload carried put it at 7.8%, so part of what looks like a difference between therapists is a difference in who they were treating.

Recovery rates are the more concrete measure, and they are what the range in this study is. The average therapist's recovery rate was 58.8%, with a standard deviation, which is the usual measure of how far individual figures sit from the average, of 13.7 percentage points. Across the 119 therapists the rates ran from 23.5% to 95.6%, and the paper does not publish the caseload beside each rate, so the two ends are the rates of individual therapists rather than measured effects. The paper's own like-for-like comparison is narrower and can be checked: it identifies 19 therapists as less effective than the rest, and reports that of the 1,704 patients they saw, 786 recovered, where 1,049 would have recovered at the average therapist's rate of 61.7%, a difference of 265 patients. The paper also reports that the deterioration rate among the least effective therapists was around three times that of the others.

Source Saxon and Barkham, "Patterns of therapist variability: Therapist effects and the contribution of patient severity and risk", Journal of Consulting and Clinical Psychology 80(4), 2012, pages 535 to 546, abstract and the comparisons of therapist effectiveness. Read on 31 August 2026 from the authors' pre-publication copy in the White Rose repository.

Patients were not allocated to therapists at random. The study reports differences that remain after adjusting for the patient characteristics it measured, chiefly severity and risk at intake, and adjusting can remove only the differences that were measured, so this is not an estimate of what a therapist caused. The sample is restricted to therapists who returned a before-and-after questionnaire for at least 90% of their patients, which is a selection on record keeping. The service is British, the outcome is a symptom and functioning score over a course of therapy, and none of it is about money.

On the endpoints of leaving and of getting worse, the practitioner's share was larger still

The same research group measured two endpoints that are not improvement: whether a patient stopped attending, and whether a patient finished worse than they began. Across 85 therapists, with 10,521 patients in the dropout comparison and 6,405 in the deterioration comparison, the individual therapist accounted for 12.6% of the variation in dropout and 10.1% of the variation in deterioration. Both are larger than the pooled share and the single-service share for improvement in the two entries above. Dropout rates for individual therapists ran from 1.2% to 73.2%, and deterioration rates from 0% to 15.4%. The paper also tested whether a therapist's dropout rate went with that therapist's deterioration rate and could not tell the relationship apart from zero at the precision it had, reporting a correlation of 0.07 with a p-value of 0.52, which is the probability of seeing a correlation at least that far from zero if the true correlation were zero. It did not test dropout against improvement, so it says nothing either way about whether the therapists who keep people are the therapists who help them.

Source Saxon, Barkham, Foster and Parry, "The Contribution of Therapist Effects to Patient Dropout and Deterioration in the Psychological Therapies", Clinical Psychology and Psychotherapy 24(3), 2017, pages 575 to 588. The publisher refuses an automated request for this page and answers a person in a browser; the figures here were read from the published abstract, in the publisher's own page and again as indexed by Europe PMC, the European index of life-science literature, on 31 August 2026, and the entry states only what that abstract states.

Same service and same kind of records as the entry above, so the same limits apply: no random allocation, adjustment only for what was measured, British primary care. Stopping attendance in a health service is a different act from ending a paid engagement, and the study does not cover any setting outside psychological therapy.

Where patients did not choose the doctor, which doctor they were sent to went with how long they lived

Norwegian law reallocates patients to a new general practitioner (GP) by a computerised procedure when their own doctor closes a practice or cuts the size of the list, and about 26% of all changes of doctor in the system happen that way. Ginja, Riise, Willage and Willen used that reallocation, with the whole country's patient and doctor register, to compare patients who had shared a doctor and lived in the same municipality and were then sent to different new doctors. The comparison is therefore between doctors the patient did not pick, which is what separates this study from the observational work above and from almost everything published on financial advice.

They measured each doctor by whether the patients allocated to that doctor were alive two years later, and found the spread across doctors large. The standard deviation of that measure was 0.157, which the authors state as a one-standard-deviation change in the quality of a patient's doctor going with a 15.7 percentage point change in the chance of being alive two years after the allocation. Two comparisons in the paper are easier to picture. Moving from a doctor in the middle of the distribution to one at the 75th percentile reduces the two-year risk of dying by 4.9 percentage points, and moving from the middle to the bottom quarter raises it by 6 percentage points. Moving from the bottom quarter of the distribution to the top raises predicted life expectancy by 3.5 months, which the authors set against an average time to death after the reallocation, among those who died, of 5.7 years, so the effect they identify is about a 5% increase in time to death against that average. The authors report that some doctor characteristics and practice styles predict the measure but that most of the variation goes with what they call unobserved ability.

Source Ginja, Riise, Willage and Willen, "Does Your Doctor Matter?", published in the Journal of Political Economy Microeconomics 3(3), 2025, pages 497 to 538. The figures here were read from IZA Discussion Paper 17726 of February 2025, the openly readable version of the same work, on 31 August 2026; its abstract, section 1 and section 4 carry every figure above. The RePEc record for that paper is where the journal publication is recorded.

The measure is a doctor's estimated contribution to whether patients were alive two years later, not a measured effect of any named treatment, and estimated measures of this kind carry estimation error that widens the apparent spread. The authors adjust for that and report an average shrinkage of 94%, which is their own correction rather than a check by anyone else. Everyone reallocated is counted whether or not they stayed with the new doctor. The setting is Norway, with a national health system in which every resident is on a general practitioner's list, and the outcome is death within two years, which is the outcome that can be measured for a whole population and not the outcome most patients would name. The paper also converts its result into money using an assumed value for a year of life; that conversion is not printed here, because the two figures the paper reports for it differ by a factor of about ten depending on which government's value is used.

Whether patients could pick the doctors the measure ranked highly

The Norwegian study above went on to ask whether patients could pick the doctors its measure ranked highly. The authors merged their register data with a proprietary set of patient-written ratings of the same doctors and found no relationship between the two. Their own reading of that, in the body of the paper, is that the result "is consistent with the notion that patients either are unable to identify high-quality GPs, or they value aspects of GPs other than their ability to save lives". They also report that patients who change doctor by choice move towards doctors who are more popular, more experienced, female and holding a specialisation.

The matching Australian measurement is in advice rather than in medicine: ASIC's shadow-shopping study of retirement advice, in which satisfaction with the advice did not vary with the grade the advice received, is treated on the State of Research page.

Source Ginja, Riise, Willage and Willen, "Does Your Doctor Matter?", read from IZA Discussion Paper 17726, section 1 and section 4.5, on 31 August 2026, with the journal publication recorded in the entry above.

The finding is an absence of a relationship rather than a measurement of one, and an absence is only as firm as the numbers behind it. The Norwegian ratings are the ones patients chose to write on a rating platform, which is not a sample of that doctor's patients.

Nothing of this shape has been measured in financial advice

No study was found that reports, for financial advice, the share of the differences between clients' outcomes that goes with which adviser the client had, on any outcome, in any country. The measurements that do exist in advice are about what advisers' clients hold and what those holdings cost, and they are set out with their sources and their limitations on the State of Research page, along with the questions the field could answer and what answering them would take. What the research above has and advice does not is the two things every one of these studies needed: records that group clients under the practitioner who served them, and an agreed measure of the client's position taken more than once.

Source that no such study was found is this practice's own search, of 31 August 2026.